2019-05-29 07:18:02 -07:00
|
|
|
|
2015-02-16 15:58:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-09-09 10:29:40 -06:00
|
|
|
|
2015-02-16 15:58:56 -08:00
|
|
|
|
2015-02-16 15:59:02 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-02-16 15:58:56 -08:00
|
|
|
|
2015-02-16 15:58:59 -08:00
|
|
|
|
2017-02-03 23:47:37 +01:00
|
|
|
|
2015-02-16 15:58:56 -08:00
|
|
|
|
2015-02-16 15:59:02 -08:00
|
|
|
|
2016-01-15 16:55:53 -08:00
|
|
|
|
2016-12-14 15:07:53 -08:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2025-10-24 20:30:47 +09:00
|
|
|
|
2015-02-16 15:58:56 -08:00
|
|
|
|
2017-02-22 15:39:50 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:58 -07:00
|
|
|
|
|
|
|
|
|
2018-01-31 16:17:36 -08:00
|
|
|
|
2017-09-06 16:18:58 -07:00
|
|
|
|
2016-11-08 11:31:44 +11:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2017-09-06 16:18:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-16 15:07:31 -05:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2018-11-16 15:07:31 -05:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
2018-03-28 11:01:43 -04:00
|
|
|
|
2016-11-08 11:32:20 +11:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-20 12:06:13 +02:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:12 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
2016-11-08 11:32:20 +11:00
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2016-11-08 11:32:20 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2017-09-06 16:18:58 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2016-11-08 11:32:20 +11:00
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2016-11-08 11:32:20 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:01:43 -04:00
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:01:43 -04:00
|
|
|
|
2016-11-08 11:32:20 +11:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
2018-04-10 16:36:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
2021-04-28 15:03:12 -04:00
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-10 16:36:56 -07:00
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:12 -04:00
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-16 15:19:13 -05:00
|
|
|
|
2019-10-21 09:29:20 -07:00
|
|
|
|
|
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
|
|
|
|
|
2019-10-21 09:29:20 -07:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-23 14:37:49 +10:00
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-27 13:16:34 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-21 11:35:53 -08:00
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
2018-12-21 11:35:53 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-27 13:16:34 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:13 -04:00
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2019-07-29 13:57:49 +02:00
|
|
|
|
2021-04-28 15:03:13 -04:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 20:09:37 -05:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:12 -04:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:03 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-12-16 15:53:56 +00:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:03 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-02-28 14:31:03 +11:00
|
|
|
|
2022-06-03 13:37:31 +08:00
|
|
|
|
2025-02-28 14:31:03 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2022-06-03 13:37:31 +08:00
|
|
|
|
|
|
|
|
|
2026-03-27 21:04:22 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-06-15 16:06:24 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-27 21:04:22 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-27 21:04:35 +00:00
|
|
|
|
2026-03-27 21:04:22 +00:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2022-06-03 13:37:31 +08:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-27 21:04:22 +00:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2026-03-27 21:04:22 +00:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-27 21:04:22 +00:00
|
|
|
|
2022-06-03 13:37:31 +08:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
fsdax: introduce page->share for fsdax in reflink mode
Patch series "fsdax,xfs: fix warning messages", v2.
Many testcases failed in dax+reflink mode with warning message in dmesg.
Such as generic/051,075,127. The warning message is like this:
[ 775.509337] ------------[ cut here ]------------
[ 775.509636] WARNING: CPU: 1 PID: 16815 at fs/dax.c:386 dax_insert_entry.cold+0x2e/0x69
[ 775.510151] Modules linked in: auth_rpcgss oid_registry nfsv4 algif_hash af_alg af_packet nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip_set nf_tables nfnetlink ip6table_filter ip6_tables iptable_filter ip_tables x_tables dax_pmem nd_pmem nd_btt sch_fq_codel configfs xfs libcrc32c fuse
[ 775.524288] CPU: 1 PID: 16815 Comm: fsx Kdump: loaded Tainted: G W 6.1.0-rc4+ #164 eb34e4ee4200c7cbbb47de2b1892c5a3e027fd6d
[ 775.524904] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS Arch Linux 1.16.0-3-3 04/01/2014
[ 775.525460] RIP: 0010:dax_insert_entry.cold+0x2e/0x69
[ 775.525797] Code: c7 c7 18 eb e0 81 48 89 4c 24 20 48 89 54 24 10 e8 73 6d ff ff 48 83 7d 18 00 48 8b 54 24 10 48 8b 4c 24 20 0f 84 e3 e9 b9 ff <0f> 0b e9 dc e9 b9 ff 48 c7 c6 a0 20 c3 81 48 c7 c7 f0 ea e0 81 48
[ 775.526708] RSP: 0000:ffffc90001d57b30 EFLAGS: 00010082
[ 775.527042] RAX: 000000000000002a RBX: 0000000000000000 RCX: 0000000000000042
[ 775.527396] RDX: ffffea000a0f6c80 RSI: ffffffff81dfab1b RDI: 00000000ffffffff
[ 775.527819] RBP: ffffea000a0f6c40 R08: 0000000000000000 R09: ffffffff820625e0
[ 775.528241] R10: ffffc90001d579d8 R11: ffffffff820d2628 R12: ffff88815fc98320
[ 775.528598] R13: ffffc90001d57c18 R14: 0000000000000000 R15: 0000000000000001
[ 775.528997] FS: 00007f39fc75d740(0000) GS:ffff88817bc80000(0000) knlGS:0000000000000000
[ 775.529474] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 775.529800] CR2: 00007f39fc772040 CR3: 0000000107eb6001 CR4: 00000000003706e0
[ 775.530214] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 775.530592] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 775.531002] Call Trace:
[ 775.531230] <TASK>
[ 775.531444] dax_fault_iter+0x267/0x6c0
[ 775.531719] dax_iomap_pte_fault+0x198/0x3d0
[ 775.532002] __xfs_filemap_fault+0x24a/0x2d0 [xfs aa8d25411432b306d9554da38096f4ebb86bdfe7]
[ 775.532603] __do_fault+0x30/0x1e0
[ 775.532903] do_fault+0x314/0x6c0
[ 775.533166] __handle_mm_fault+0x646/0x1250
[ 775.533480] handle_mm_fault+0xc1/0x230
[ 775.533810] do_user_addr_fault+0x1ac/0x610
[ 775.534110] exc_page_fault+0x63/0x140
[ 775.534389] asm_exc_page_fault+0x22/0x30
[ 775.534678] RIP: 0033:0x7f39fc55820a
[ 775.534950] Code: 00 01 00 00 00 74 99 83 f9 c0 0f 87 7b fe ff ff c5 fe 6f 4e 20 48 29 fe 48 83 c7 3f 49 8d 0c 10 48 83 e7 c0 48 01 fe 48 29 f9 <f3> a4 c4 c1 7e 7f 00 c4 c1 7e 7f 48 20 c5 f8 77 c3 0f 1f 44 00 00
[ 775.535839] RSP: 002b:00007ffc66a08118 EFLAGS: 00010202
[ 775.536157] RAX: 00007f39fc772001 RBX: 0000000000042001 RCX: 00000000000063c1
[ 775.536537] RDX: 0000000000006400 RSI: 00007f39fac42050 RDI: 00007f39fc772040
[ 775.536919] RBP: 0000000000006400 R08: 00007f39fc772001 R09: 0000000000042000
[ 775.537304] R10: 0000000000000001 R11: 0000000000000246 R12: 0000000000000001
[ 775.537694] R13: 00007f39fc772000 R14: 0000000000006401 R15: 0000000000000003
[ 775.538086] </TASK>
[ 775.538333] ---[ end trace 0000000000000000 ]---
This also affects dax+noreflink mode if we run the test after a
dax+reflink test. So, the most urgent thing is solving the warning
messages.
With these fixes, most warning messages in dax_associate_entry() are gone.
But honestly, generic/388 will randomly failed with the warning. The
case shutdown the xfs when fsstress is running, and do it for many times.
I think the reason is that dax pages in use are not able to be invalidated
in time when fs is shutdown. The next time dax page to be associated, it
still remains the mapping value set last time. I'll keep on solving it.
The warning message in dax_writeback_one() can also be fixed because of
the dax unshare.
This patch (of 8):
fsdax page is used not only when CoW, but also mapread. To make the it
easily understood, use 'share' to indicate that the dax page is shared by
more than one extent. And add helper functions to use it.
Also, the flag needs to be renamed to PAGE_MAPPING_DAX_SHARED.
[ruansy.fnst@fujitsu.com: rename several functions]
Link: https://lkml.kernel.org/r/1669972991-246-1-git-send-email-ruansy.fnst@fujitsu.com
[ruansy.fnst@fujitsu.com: v2.2]
Link: https://lkml.kernel.org/r/1670381359-53-1-git-send-email-ruansy.fnst@fujitsu.com
Link: https://lkml.kernel.org/r/1669908538-55-1-git-send-email-ruansy.fnst@fujitsu.com
Link: https://lkml.kernel.org/r/1669908538-55-2-git-send-email-ruansy.fnst@fujitsu.com
Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-12-01 15:28:51 +00:00
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
fsdax: introduce page->share for fsdax in reflink mode
Patch series "fsdax,xfs: fix warning messages", v2.
Many testcases failed in dax+reflink mode with warning message in dmesg.
Such as generic/051,075,127. The warning message is like this:
[ 775.509337] ------------[ cut here ]------------
[ 775.509636] WARNING: CPU: 1 PID: 16815 at fs/dax.c:386 dax_insert_entry.cold+0x2e/0x69
[ 775.510151] Modules linked in: auth_rpcgss oid_registry nfsv4 algif_hash af_alg af_packet nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nft_chain_nat iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip_set nf_tables nfnetlink ip6table_filter ip6_tables iptable_filter ip_tables x_tables dax_pmem nd_pmem nd_btt sch_fq_codel configfs xfs libcrc32c fuse
[ 775.524288] CPU: 1 PID: 16815 Comm: fsx Kdump: loaded Tainted: G W 6.1.0-rc4+ #164 eb34e4ee4200c7cbbb47de2b1892c5a3e027fd6d
[ 775.524904] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS Arch Linux 1.16.0-3-3 04/01/2014
[ 775.525460] RIP: 0010:dax_insert_entry.cold+0x2e/0x69
[ 775.525797] Code: c7 c7 18 eb e0 81 48 89 4c 24 20 48 89 54 24 10 e8 73 6d ff ff 48 83 7d 18 00 48 8b 54 24 10 48 8b 4c 24 20 0f 84 e3 e9 b9 ff <0f> 0b e9 dc e9 b9 ff 48 c7 c6 a0 20 c3 81 48 c7 c7 f0 ea e0 81 48
[ 775.526708] RSP: 0000:ffffc90001d57b30 EFLAGS: 00010082
[ 775.527042] RAX: 000000000000002a RBX: 0000000000000000 RCX: 0000000000000042
[ 775.527396] RDX: ffffea000a0f6c80 RSI: ffffffff81dfab1b RDI: 00000000ffffffff
[ 775.527819] RBP: ffffea000a0f6c40 R08: 0000000000000000 R09: ffffffff820625e0
[ 775.528241] R10: ffffc90001d579d8 R11: ffffffff820d2628 R12: ffff88815fc98320
[ 775.528598] R13: ffffc90001d57c18 R14: 0000000000000000 R15: 0000000000000001
[ 775.528997] FS: 00007f39fc75d740(0000) GS:ffff88817bc80000(0000) knlGS:0000000000000000
[ 775.529474] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 775.529800] CR2: 00007f39fc772040 CR3: 0000000107eb6001 CR4: 00000000003706e0
[ 775.530214] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 775.530592] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 775.531002] Call Trace:
[ 775.531230] <TASK>
[ 775.531444] dax_fault_iter+0x267/0x6c0
[ 775.531719] dax_iomap_pte_fault+0x198/0x3d0
[ 775.532002] __xfs_filemap_fault+0x24a/0x2d0 [xfs aa8d25411432b306d9554da38096f4ebb86bdfe7]
[ 775.532603] __do_fault+0x30/0x1e0
[ 775.532903] do_fault+0x314/0x6c0
[ 775.533166] __handle_mm_fault+0x646/0x1250
[ 775.533480] handle_mm_fault+0xc1/0x230
[ 775.533810] do_user_addr_fault+0x1ac/0x610
[ 775.534110] exc_page_fault+0x63/0x140
[ 775.534389] asm_exc_page_fault+0x22/0x30
[ 775.534678] RIP: 0033:0x7f39fc55820a
[ 775.534950] Code: 00 01 00 00 00 74 99 83 f9 c0 0f 87 7b fe ff ff c5 fe 6f 4e 20 48 29 fe 48 83 c7 3f 49 8d 0c 10 48 83 e7 c0 48 01 fe 48 29 f9 <f3> a4 c4 c1 7e 7f 00 c4 c1 7e 7f 48 20 c5 f8 77 c3 0f 1f 44 00 00
[ 775.535839] RSP: 002b:00007ffc66a08118 EFLAGS: 00010202
[ 775.536157] RAX: 00007f39fc772001 RBX: 0000000000042001 RCX: 00000000000063c1
[ 775.536537] RDX: 0000000000006400 RSI: 00007f39fac42050 RDI: 00007f39fc772040
[ 775.536919] RBP: 0000000000006400 R08: 00007f39fc772001 R09: 0000000000042000
[ 775.537304] R10: 0000000000000001 R11: 0000000000000246 R12: 0000000000000001
[ 775.537694] R13: 00007f39fc772000 R14: 0000000000006401 R15: 0000000000000003
[ 775.538086] </TASK>
[ 775.538333] ---[ end trace 0000000000000000 ]---
This also affects dax+noreflink mode if we run the test after a
dax+reflink test. So, the most urgent thing is solving the warning
messages.
With these fixes, most warning messages in dax_associate_entry() are gone.
But honestly, generic/388 will randomly failed with the warning. The
case shutdown the xfs when fsstress is running, and do it for many times.
I think the reason is that dax pages in use are not able to be invalidated
in time when fs is shutdown. The next time dax page to be associated, it
still remains the mapping value set last time. I'll keep on solving it.
The warning message in dax_writeback_one() can also be fixed because of
the dax unshare.
This patch (of 8):
fsdax page is used not only when CoW, but also mapread. To make the it
easily understood, use 'share' to indicate that the dax page is shared by
more than one extent. And add helper functions to use it.
Also, the flag needs to be renamed to PAGE_MAPPING_DAX_SHARED.
[ruansy.fnst@fujitsu.com: rename several functions]
Link: https://lkml.kernel.org/r/1669972991-246-1-git-send-email-ruansy.fnst@fujitsu.com
[ruansy.fnst@fujitsu.com: v2.2]
Link: https://lkml.kernel.org/r/1670381359-53-1-git-send-email-ruansy.fnst@fujitsu.com
Link: https://lkml.kernel.org/r/1669908538-55-1-git-send-email-ruansy.fnst@fujitsu.com
Link: https://lkml.kernel.org/r/1669908538-55-2-git-send-email-ruansy.fnst@fujitsu.com
Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-12-01 15:28:51 +00:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:31 +08:00
|
|
|
|
|
|
|
|
|
2018-07-13 21:49:50 -07:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
fs/dax: check zero or empty entry before converting xarray entry
Calling dax_to_folio() with empty entry causes kernel panic below when
booting a VM with DAX enabled storage.
This patch checks empty entry before calling dax_to_folio() on
dax_associate_entry(), dax_disassociate_entry(), and dax_busy_page().
Commit 98c183a4fccf ("fs/dax: don't disassociate zero page entries") added
guards in the associate and disassociate paths, but the guards still come
after dax_to_folio(), and dax_busy_page() still has the same problem.
[ 0.737679] EXT4-fs (pmem0p1): mounted filesystem 79676804-7c8b-491a-b2a6-9bae3c72af70 ro with ordered data mode. Quota mode: disabled.
[ 0.737891] VFS: Mounted root (ext4 filesystem) readonly on device 259:1.
[ 0.739119] devtmpfs: mounted
[ 0.739476] Freeing unused kernel memory: 1920K
[ 0.740156] Run /sbin/init as init process
[ 0.740229] with arguments:
[ 0.740286] /sbin/init
[ 0.740321] with environment:
[ 0.740369] HOME=/
[ 0.740400] TERM=linux
[ 0.743162] Unable to handle kernel paging request at virtual address fffffdffbf000008
[ 0.743285] Mem abort info:
[ 0.743316] ESR = 0x0000000096000006
[ 0.743371] EC = 0x25: DABT (current EL), IL = 32 bits
[ 0.743444] SET = 0, FnV = 0
[ 0.743489] EA = 0, S1PTW = 0
[ 0.743545] FSC = 0x06: level 2 translation fault
[ 0.743610] Data abort info:
[ 0.743656] ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[ 0.743720] CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 0.743785] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 0.743848] swapper pgtable: 4k pages, 48-bit VAs, pgdp=00000000b9d17000
[ 0.743931] [fffffdffbf000008] pgd=10000000bfa3d403, p4d=10000000bfa3d403, pud=1000000040bfe403, pmd=0000000000000000
[ 0.744070] Internal error: Oops: 0000000096000006 [#1] SMP
[ 0.748888] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted 6.18.4 #1 NONE
[ 0.749421] pstate: 004000c5 (nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 0.749969] pc : dax_disassociate_entry.constprop.0+0x20/0x50
[ 0.750444] lr : dax_insert_entry+0xcc/0x408
[ 0.750802] sp : ffff80008000b9e0
[ 0.751083] x29: ffff80008000b9e0 x28: 0000000000000000 x27: 0000000000000000
[ 0.751682] x26: 0000000001963d01 x25: ffff0000004f7d90 x24: 0000000000000000
[ 0.752264] x23: 0000000000000000 x22: ffff80008000bcc8 x21: 0000000000000011
[ 0.752836] x20: ffff80008000ba90 x19: 0000000001963d01 x18: 0000000000000000
[ 0.753407] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[ 0.753970] x14: ffffbf3154b9ae70 x13: 0000000000000000 x12: ffffbf3154b9ae70
[ 0.754548] x11: ffffffffffffffff x10: 0000000000000000 x9 : 0000000000000000
[ 0.755122] x8 : 000000000000000d x7 : 000000000000001f x6 : 0000000000000000
[ 0.755707] x5 : 0000000000000000 x4 : 0000000000000000 x3 : fffffdffc0000000
[ 0.756287] x2 : 0000000000000008 x1 : 0000000040000000 x0 : fffffdffbf000000
[ 0.756871] Call trace:
[ 0.757107] dax_disassociate_entry.constprop.0+0x20/0x50 (P)
[ 0.757592] dax_iomap_pte_fault+0x4fc/0x808
[ 0.757951] dax_iomap_fault+0x28/0x30
[ 0.758258] ext4_dax_huge_fault+0x80/0x2dc
[ 0.758594] ext4_dax_fault+0x10/0x3c
[ 0.758892] __do_fault+0x38/0x12c
[ 0.759175] __handle_mm_fault+0x530/0xcf0
[ 0.759518] handle_mm_fault+0xe4/0x230
[ 0.759833] do_page_fault+0x17c/0x4dc
[ 0.760144] do_translation_fault+0x30/0x38
[ 0.760483] do_mem_abort+0x40/0x8c
[ 0.760771] el0_ia+0x4c/0x170
[ 0.761032] el0t_64_sync_handler+0xd8/0xdc
[ 0.761371] el0t_64_sync+0x168/0x16c
[ 0.761677] Code: f9453021 f2dfbfe3 cb813080 8b001860 (f9400401)
[ 0.762168] ---[ end trace 0000000000000000 ]---
[ 0.762550] note: init[1] exited with irqs disabled
[ 0.762631] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Link: https://lore.kernel.org/m2y0enxtzk.fsf@arm.com
Fixes: 38607c62b34b ("fs/dax: properly refcount fs dax pages")
Signed-off-by: Seunguk Shin <seunguk.shin@arm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Kiara Grouwstra <cinereal@riseup.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-08-03 13:34:55 +01:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-03-18 18:33:01 -07:00
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
fs/dax: check zero or empty entry before converting xarray entry
Calling dax_to_folio() with empty entry causes kernel panic below when
booting a VM with DAX enabled storage.
This patch checks empty entry before calling dax_to_folio() on
dax_associate_entry(), dax_disassociate_entry(), and dax_busy_page().
Commit 98c183a4fccf ("fs/dax: don't disassociate zero page entries") added
guards in the associate and disassociate paths, but the guards still come
after dax_to_folio(), and dax_busy_page() still has the same problem.
[ 0.737679] EXT4-fs (pmem0p1): mounted filesystem 79676804-7c8b-491a-b2a6-9bae3c72af70 ro with ordered data mode. Quota mode: disabled.
[ 0.737891] VFS: Mounted root (ext4 filesystem) readonly on device 259:1.
[ 0.739119] devtmpfs: mounted
[ 0.739476] Freeing unused kernel memory: 1920K
[ 0.740156] Run /sbin/init as init process
[ 0.740229] with arguments:
[ 0.740286] /sbin/init
[ 0.740321] with environment:
[ 0.740369] HOME=/
[ 0.740400] TERM=linux
[ 0.743162] Unable to handle kernel paging request at virtual address fffffdffbf000008
[ 0.743285] Mem abort info:
[ 0.743316] ESR = 0x0000000096000006
[ 0.743371] EC = 0x25: DABT (current EL), IL = 32 bits
[ 0.743444] SET = 0, FnV = 0
[ 0.743489] EA = 0, S1PTW = 0
[ 0.743545] FSC = 0x06: level 2 translation fault
[ 0.743610] Data abort info:
[ 0.743656] ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[ 0.743720] CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 0.743785] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 0.743848] swapper pgtable: 4k pages, 48-bit VAs, pgdp=00000000b9d17000
[ 0.743931] [fffffdffbf000008] pgd=10000000bfa3d403, p4d=10000000bfa3d403, pud=1000000040bfe403, pmd=0000000000000000
[ 0.744070] Internal error: Oops: 0000000096000006 [#1] SMP
[ 0.748888] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted 6.18.4 #1 NONE
[ 0.749421] pstate: 004000c5 (nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 0.749969] pc : dax_disassociate_entry.constprop.0+0x20/0x50
[ 0.750444] lr : dax_insert_entry+0xcc/0x408
[ 0.750802] sp : ffff80008000b9e0
[ 0.751083] x29: ffff80008000b9e0 x28: 0000000000000000 x27: 0000000000000000
[ 0.751682] x26: 0000000001963d01 x25: ffff0000004f7d90 x24: 0000000000000000
[ 0.752264] x23: 0000000000000000 x22: ffff80008000bcc8 x21: 0000000000000011
[ 0.752836] x20: ffff80008000ba90 x19: 0000000001963d01 x18: 0000000000000000
[ 0.753407] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[ 0.753970] x14: ffffbf3154b9ae70 x13: 0000000000000000 x12: ffffbf3154b9ae70
[ 0.754548] x11: ffffffffffffffff x10: 0000000000000000 x9 : 0000000000000000
[ 0.755122] x8 : 000000000000000d x7 : 000000000000001f x6 : 0000000000000000
[ 0.755707] x5 : 0000000000000000 x4 : 0000000000000000 x3 : fffffdffc0000000
[ 0.756287] x2 : 0000000000000008 x1 : 0000000040000000 x0 : fffffdffbf000000
[ 0.756871] Call trace:
[ 0.757107] dax_disassociate_entry.constprop.0+0x20/0x50 (P)
[ 0.757592] dax_iomap_pte_fault+0x4fc/0x808
[ 0.757951] dax_iomap_fault+0x28/0x30
[ 0.758258] ext4_dax_huge_fault+0x80/0x2dc
[ 0.758594] ext4_dax_fault+0x10/0x3c
[ 0.758892] __do_fault+0x38/0x12c
[ 0.759175] __handle_mm_fault+0x530/0xcf0
[ 0.759518] handle_mm_fault+0xe4/0x230
[ 0.759833] do_page_fault+0x17c/0x4dc
[ 0.760144] do_translation_fault+0x30/0x38
[ 0.760483] do_mem_abort+0x40/0x8c
[ 0.760771] el0_ia+0x4c/0x170
[ 0.761032] el0t_64_sync_handler+0xd8/0xdc
[ 0.761371] el0t_64_sync+0x168/0x16c
[ 0.761677] Code: f9453021 f2dfbfe3 cb813080 8b001860 (f9400401)
[ 0.762168] ---[ end trace 0000000000000000 ]---
[ 0.762550] note: init[1] exited with irqs disabled
[ 0.762631] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Link: https://lore.kernel.org/m2y0enxtzk.fsf@arm.com
Fixes: 38607c62b34b ("fs/dax: properly refcount fs dax pages")
Signed-off-by: Seunguk Shin <seunguk.shin@arm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Kiara Grouwstra <cinereal@riseup.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-08-03 13:34:55 +01:00
|
|
|
|
2018-07-13 21:49:50 -07:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
fs/dax: check zero or empty entry before converting xarray entry
Calling dax_to_folio() with empty entry causes kernel panic below when
booting a VM with DAX enabled storage.
This patch checks empty entry before calling dax_to_folio() on
dax_associate_entry(), dax_disassociate_entry(), and dax_busy_page().
Commit 98c183a4fccf ("fs/dax: don't disassociate zero page entries") added
guards in the associate and disassociate paths, but the guards still come
after dax_to_folio(), and dax_busy_page() still has the same problem.
[ 0.737679] EXT4-fs (pmem0p1): mounted filesystem 79676804-7c8b-491a-b2a6-9bae3c72af70 ro with ordered data mode. Quota mode: disabled.
[ 0.737891] VFS: Mounted root (ext4 filesystem) readonly on device 259:1.
[ 0.739119] devtmpfs: mounted
[ 0.739476] Freeing unused kernel memory: 1920K
[ 0.740156] Run /sbin/init as init process
[ 0.740229] with arguments:
[ 0.740286] /sbin/init
[ 0.740321] with environment:
[ 0.740369] HOME=/
[ 0.740400] TERM=linux
[ 0.743162] Unable to handle kernel paging request at virtual address fffffdffbf000008
[ 0.743285] Mem abort info:
[ 0.743316] ESR = 0x0000000096000006
[ 0.743371] EC = 0x25: DABT (current EL), IL = 32 bits
[ 0.743444] SET = 0, FnV = 0
[ 0.743489] EA = 0, S1PTW = 0
[ 0.743545] FSC = 0x06: level 2 translation fault
[ 0.743610] Data abort info:
[ 0.743656] ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[ 0.743720] CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 0.743785] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 0.743848] swapper pgtable: 4k pages, 48-bit VAs, pgdp=00000000b9d17000
[ 0.743931] [fffffdffbf000008] pgd=10000000bfa3d403, p4d=10000000bfa3d403, pud=1000000040bfe403, pmd=0000000000000000
[ 0.744070] Internal error: Oops: 0000000096000006 [#1] SMP
[ 0.748888] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted 6.18.4 #1 NONE
[ 0.749421] pstate: 004000c5 (nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 0.749969] pc : dax_disassociate_entry.constprop.0+0x20/0x50
[ 0.750444] lr : dax_insert_entry+0xcc/0x408
[ 0.750802] sp : ffff80008000b9e0
[ 0.751083] x29: ffff80008000b9e0 x28: 0000000000000000 x27: 0000000000000000
[ 0.751682] x26: 0000000001963d01 x25: ffff0000004f7d90 x24: 0000000000000000
[ 0.752264] x23: 0000000000000000 x22: ffff80008000bcc8 x21: 0000000000000011
[ 0.752836] x20: ffff80008000ba90 x19: 0000000001963d01 x18: 0000000000000000
[ 0.753407] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[ 0.753970] x14: ffffbf3154b9ae70 x13: 0000000000000000 x12: ffffbf3154b9ae70
[ 0.754548] x11: ffffffffffffffff x10: 0000000000000000 x9 : 0000000000000000
[ 0.755122] x8 : 000000000000000d x7 : 000000000000001f x6 : 0000000000000000
[ 0.755707] x5 : 0000000000000000 x4 : 0000000000000000 x3 : fffffdffc0000000
[ 0.756287] x2 : 0000000000000008 x1 : 0000000040000000 x0 : fffffdffbf000000
[ 0.756871] Call trace:
[ 0.757107] dax_disassociate_entry.constprop.0+0x20/0x50 (P)
[ 0.757592] dax_iomap_pte_fault+0x4fc/0x808
[ 0.757951] dax_iomap_fault+0x28/0x30
[ 0.758258] ext4_dax_huge_fault+0x80/0x2dc
[ 0.758594] ext4_dax_fault+0x10/0x3c
[ 0.758892] __do_fault+0x38/0x12c
[ 0.759175] __handle_mm_fault+0x530/0xcf0
[ 0.759518] handle_mm_fault+0xe4/0x230
[ 0.759833] do_page_fault+0x17c/0x4dc
[ 0.760144] do_translation_fault+0x30/0x38
[ 0.760483] do_mem_abort+0x40/0x8c
[ 0.760771] el0_ia+0x4c/0x170
[ 0.761032] el0t_64_sync_handler+0xd8/0xdc
[ 0.761371] el0t_64_sync+0x168/0x16c
[ 0.761677] Code: f9453021 f2dfbfe3 cb813080 8b001860 (f9400401)
[ 0.762168] ---[ end trace 0000000000000000 ]---
[ 0.762550] note: init[1] exited with irqs disabled
[ 0.762631] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Link: https://lore.kernel.org/m2y0enxtzk.fsf@arm.com
Fixes: 38607c62b34b ("fs/dax: properly refcount fs dax pages")
Signed-off-by: Seunguk Shin <seunguk.shin@arm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Kiara Grouwstra <cinereal@riseup.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-08-03 13:34:55 +01:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-03-18 18:33:01 -07:00
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
fs/dax: check zero or empty entry before converting xarray entry
Calling dax_to_folio() with empty entry causes kernel panic below when
booting a VM with DAX enabled storage.
This patch checks empty entry before calling dax_to_folio() on
dax_associate_entry(), dax_disassociate_entry(), and dax_busy_page().
Commit 98c183a4fccf ("fs/dax: don't disassociate zero page entries") added
guards in the associate and disassociate paths, but the guards still come
after dax_to_folio(), and dax_busy_page() still has the same problem.
[ 0.737679] EXT4-fs (pmem0p1): mounted filesystem 79676804-7c8b-491a-b2a6-9bae3c72af70 ro with ordered data mode. Quota mode: disabled.
[ 0.737891] VFS: Mounted root (ext4 filesystem) readonly on device 259:1.
[ 0.739119] devtmpfs: mounted
[ 0.739476] Freeing unused kernel memory: 1920K
[ 0.740156] Run /sbin/init as init process
[ 0.740229] with arguments:
[ 0.740286] /sbin/init
[ 0.740321] with environment:
[ 0.740369] HOME=/
[ 0.740400] TERM=linux
[ 0.743162] Unable to handle kernel paging request at virtual address fffffdffbf000008
[ 0.743285] Mem abort info:
[ 0.743316] ESR = 0x0000000096000006
[ 0.743371] EC = 0x25: DABT (current EL), IL = 32 bits
[ 0.743444] SET = 0, FnV = 0
[ 0.743489] EA = 0, S1PTW = 0
[ 0.743545] FSC = 0x06: level 2 translation fault
[ 0.743610] Data abort info:
[ 0.743656] ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[ 0.743720] CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 0.743785] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 0.743848] swapper pgtable: 4k pages, 48-bit VAs, pgdp=00000000b9d17000
[ 0.743931] [fffffdffbf000008] pgd=10000000bfa3d403, p4d=10000000bfa3d403, pud=1000000040bfe403, pmd=0000000000000000
[ 0.744070] Internal error: Oops: 0000000096000006 [#1] SMP
[ 0.748888] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted 6.18.4 #1 NONE
[ 0.749421] pstate: 004000c5 (nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 0.749969] pc : dax_disassociate_entry.constprop.0+0x20/0x50
[ 0.750444] lr : dax_insert_entry+0xcc/0x408
[ 0.750802] sp : ffff80008000b9e0
[ 0.751083] x29: ffff80008000b9e0 x28: 0000000000000000 x27: 0000000000000000
[ 0.751682] x26: 0000000001963d01 x25: ffff0000004f7d90 x24: 0000000000000000
[ 0.752264] x23: 0000000000000000 x22: ffff80008000bcc8 x21: 0000000000000011
[ 0.752836] x20: ffff80008000ba90 x19: 0000000001963d01 x18: 0000000000000000
[ 0.753407] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[ 0.753970] x14: ffffbf3154b9ae70 x13: 0000000000000000 x12: ffffbf3154b9ae70
[ 0.754548] x11: ffffffffffffffff x10: 0000000000000000 x9 : 0000000000000000
[ 0.755122] x8 : 000000000000000d x7 : 000000000000001f x6 : 0000000000000000
[ 0.755707] x5 : 0000000000000000 x4 : 0000000000000000 x3 : fffffdffc0000000
[ 0.756287] x2 : 0000000000000008 x1 : 0000000040000000 x0 : fffffdffbf000000
[ 0.756871] Call trace:
[ 0.757107] dax_disassociate_entry.constprop.0+0x20/0x50 (P)
[ 0.757592] dax_iomap_pte_fault+0x4fc/0x808
[ 0.757951] dax_iomap_fault+0x28/0x30
[ 0.758258] ext4_dax_huge_fault+0x80/0x2dc
[ 0.758594] ext4_dax_fault+0x10/0x3c
[ 0.758892] __do_fault+0x38/0x12c
[ 0.759175] __handle_mm_fault+0x530/0xcf0
[ 0.759518] handle_mm_fault+0xe4/0x230
[ 0.759833] do_page_fault+0x17c/0x4dc
[ 0.760144] do_translation_fault+0x30/0x38
[ 0.760483] do_mem_abort+0x40/0x8c
[ 0.760771] el0_ia+0x4c/0x170
[ 0.761032] el0t_64_sync_handler+0xd8/0xdc
[ 0.761371] el0t_64_sync+0x168/0x16c
[ 0.761677] Code: f9453021 f2dfbfe3 cb813080 8b001860 (f9400401)
[ 0.762168] ---[ end trace 0000000000000000 ]---
[ 0.762550] note: init[1] exited with irqs disabled
[ 0.762631] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Link: https://lore.kernel.org/m2y0enxtzk.fsf@arm.com
Fixes: 38607c62b34b ("fs/dax: properly refcount fs dax pages")
Signed-off-by: Seunguk Shin <seunguk.shin@arm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Kiara Grouwstra <cinereal@riseup.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-08-03 13:34:55 +01:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
fs/dax: check zero or empty entry before converting xarray entry
Calling dax_to_folio() with empty entry causes kernel panic below when
booting a VM with DAX enabled storage.
This patch checks empty entry before calling dax_to_folio() on
dax_associate_entry(), dax_disassociate_entry(), and dax_busy_page().
Commit 98c183a4fccf ("fs/dax: don't disassociate zero page entries") added
guards in the associate and disassociate paths, but the guards still come
after dax_to_folio(), and dax_busy_page() still has the same problem.
[ 0.737679] EXT4-fs (pmem0p1): mounted filesystem 79676804-7c8b-491a-b2a6-9bae3c72af70 ro with ordered data mode. Quota mode: disabled.
[ 0.737891] VFS: Mounted root (ext4 filesystem) readonly on device 259:1.
[ 0.739119] devtmpfs: mounted
[ 0.739476] Freeing unused kernel memory: 1920K
[ 0.740156] Run /sbin/init as init process
[ 0.740229] with arguments:
[ 0.740286] /sbin/init
[ 0.740321] with environment:
[ 0.740369] HOME=/
[ 0.740400] TERM=linux
[ 0.743162] Unable to handle kernel paging request at virtual address fffffdffbf000008
[ 0.743285] Mem abort info:
[ 0.743316] ESR = 0x0000000096000006
[ 0.743371] EC = 0x25: DABT (current EL), IL = 32 bits
[ 0.743444] SET = 0, FnV = 0
[ 0.743489] EA = 0, S1PTW = 0
[ 0.743545] FSC = 0x06: level 2 translation fault
[ 0.743610] Data abort info:
[ 0.743656] ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[ 0.743720] CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 0.743785] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 0.743848] swapper pgtable: 4k pages, 48-bit VAs, pgdp=00000000b9d17000
[ 0.743931] [fffffdffbf000008] pgd=10000000bfa3d403, p4d=10000000bfa3d403, pud=1000000040bfe403, pmd=0000000000000000
[ 0.744070] Internal error: Oops: 0000000096000006 [#1] SMP
[ 0.748888] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted 6.18.4 #1 NONE
[ 0.749421] pstate: 004000c5 (nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 0.749969] pc : dax_disassociate_entry.constprop.0+0x20/0x50
[ 0.750444] lr : dax_insert_entry+0xcc/0x408
[ 0.750802] sp : ffff80008000b9e0
[ 0.751083] x29: ffff80008000b9e0 x28: 0000000000000000 x27: 0000000000000000
[ 0.751682] x26: 0000000001963d01 x25: ffff0000004f7d90 x24: 0000000000000000
[ 0.752264] x23: 0000000000000000 x22: ffff80008000bcc8 x21: 0000000000000011
[ 0.752836] x20: ffff80008000ba90 x19: 0000000001963d01 x18: 0000000000000000
[ 0.753407] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[ 0.753970] x14: ffffbf3154b9ae70 x13: 0000000000000000 x12: ffffbf3154b9ae70
[ 0.754548] x11: ffffffffffffffff x10: 0000000000000000 x9 : 0000000000000000
[ 0.755122] x8 : 000000000000000d x7 : 000000000000001f x6 : 0000000000000000
[ 0.755707] x5 : 0000000000000000 x4 : 0000000000000000 x3 : fffffdffc0000000
[ 0.756287] x2 : 0000000000000008 x1 : 0000000040000000 x0 : fffffdffbf000000
[ 0.756871] Call trace:
[ 0.757107] dax_disassociate_entry.constprop.0+0x20/0x50 (P)
[ 0.757592] dax_iomap_pte_fault+0x4fc/0x808
[ 0.757951] dax_iomap_fault+0x28/0x30
[ 0.758258] ext4_dax_huge_fault+0x80/0x2dc
[ 0.758594] ext4_dax_fault+0x10/0x3c
[ 0.758892] __do_fault+0x38/0x12c
[ 0.759175] __handle_mm_fault+0x530/0xcf0
[ 0.759518] handle_mm_fault+0xe4/0x230
[ 0.759833] do_page_fault+0x17c/0x4dc
[ 0.760144] do_translation_fault+0x30/0x38
[ 0.760483] do_mem_abort+0x40/0x8c
[ 0.760771] el0_ia+0x4c/0x170
[ 0.761032] el0t_64_sync_handler+0xd8/0xdc
[ 0.761371] el0t_64_sync+0x168/0x16c
[ 0.761677] Code: f9453021 f2dfbfe3 cb813080 8b001860 (f9400401)
[ 0.762168] ---[ end trace 0000000000000000 ]---
[ 0.762550] note: init[1] exited with irqs disabled
[ 0.762631] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Link: https://lore.kernel.org/m2y0enxtzk.fsf@arm.com
Fixes: 38607c62b34b ("fs/dax: properly refcount fs dax pages")
Signed-off-by: Seunguk Shin <seunguk.shin@arm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Kiara Grouwstra <cinereal@riseup.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-08-03 13:34:55 +01:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
fs/dax: check zero or empty entry before converting xarray entry
Calling dax_to_folio() with empty entry causes kernel panic below when
booting a VM with DAX enabled storage.
This patch checks empty entry before calling dax_to_folio() on
dax_associate_entry(), dax_disassociate_entry(), and dax_busy_page().
Commit 98c183a4fccf ("fs/dax: don't disassociate zero page entries") added
guards in the associate and disassociate paths, but the guards still come
after dax_to_folio(), and dax_busy_page() still has the same problem.
[ 0.737679] EXT4-fs (pmem0p1): mounted filesystem 79676804-7c8b-491a-b2a6-9bae3c72af70 ro with ordered data mode. Quota mode: disabled.
[ 0.737891] VFS: Mounted root (ext4 filesystem) readonly on device 259:1.
[ 0.739119] devtmpfs: mounted
[ 0.739476] Freeing unused kernel memory: 1920K
[ 0.740156] Run /sbin/init as init process
[ 0.740229] with arguments:
[ 0.740286] /sbin/init
[ 0.740321] with environment:
[ 0.740369] HOME=/
[ 0.740400] TERM=linux
[ 0.743162] Unable to handle kernel paging request at virtual address fffffdffbf000008
[ 0.743285] Mem abort info:
[ 0.743316] ESR = 0x0000000096000006
[ 0.743371] EC = 0x25: DABT (current EL), IL = 32 bits
[ 0.743444] SET = 0, FnV = 0
[ 0.743489] EA = 0, S1PTW = 0
[ 0.743545] FSC = 0x06: level 2 translation fault
[ 0.743610] Data abort info:
[ 0.743656] ISV = 0, ISS = 0x00000006, ISS2 = 0x00000000
[ 0.743720] CM = 0, WnR = 0, TnD = 0, TagAccess = 0
[ 0.743785] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
[ 0.743848] swapper pgtable: 4k pages, 48-bit VAs, pgdp=00000000b9d17000
[ 0.743931] [fffffdffbf000008] pgd=10000000bfa3d403, p4d=10000000bfa3d403, pud=1000000040bfe403, pmd=0000000000000000
[ 0.744070] Internal error: Oops: 0000000096000006 [#1] SMP
[ 0.748888] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted 6.18.4 #1 NONE
[ 0.749421] pstate: 004000c5 (nzcv daIF +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[ 0.749969] pc : dax_disassociate_entry.constprop.0+0x20/0x50
[ 0.750444] lr : dax_insert_entry+0xcc/0x408
[ 0.750802] sp : ffff80008000b9e0
[ 0.751083] x29: ffff80008000b9e0 x28: 0000000000000000 x27: 0000000000000000
[ 0.751682] x26: 0000000001963d01 x25: ffff0000004f7d90 x24: 0000000000000000
[ 0.752264] x23: 0000000000000000 x22: ffff80008000bcc8 x21: 0000000000000011
[ 0.752836] x20: ffff80008000ba90 x19: 0000000001963d01 x18: 0000000000000000
[ 0.753407] x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000
[ 0.753970] x14: ffffbf3154b9ae70 x13: 0000000000000000 x12: ffffbf3154b9ae70
[ 0.754548] x11: ffffffffffffffff x10: 0000000000000000 x9 : 0000000000000000
[ 0.755122] x8 : 000000000000000d x7 : 000000000000001f x6 : 0000000000000000
[ 0.755707] x5 : 0000000000000000 x4 : 0000000000000000 x3 : fffffdffc0000000
[ 0.756287] x2 : 0000000000000008 x1 : 0000000040000000 x0 : fffffdffbf000000
[ 0.756871] Call trace:
[ 0.757107] dax_disassociate_entry.constprop.0+0x20/0x50 (P)
[ 0.757592] dax_iomap_pte_fault+0x4fc/0x808
[ 0.757951] dax_iomap_fault+0x28/0x30
[ 0.758258] ext4_dax_huge_fault+0x80/0x2dc
[ 0.758594] ext4_dax_fault+0x10/0x3c
[ 0.758892] __do_fault+0x38/0x12c
[ 0.759175] __handle_mm_fault+0x530/0xcf0
[ 0.759518] handle_mm_fault+0xe4/0x230
[ 0.759833] do_page_fault+0x17c/0x4dc
[ 0.760144] do_translation_fault+0x30/0x38
[ 0.760483] do_mem_abort+0x40/0x8c
[ 0.760771] el0_ia+0x4c/0x170
[ 0.761032] el0t_64_sync_handler+0xd8/0xdc
[ 0.761371] el0t_64_sync+0x168/0x16c
[ 0.761677] Code: f9453021 f2dfbfe3 cb813080 8b001860 (f9400401)
[ 0.762168] ---[ end trace 0000000000000000 ]---
[ 0.762550] note: init[1] exited with irqs disabled
[ 0.762631] Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Link: https://lore.kernel.org/m2y0enxtzk.fsf@arm.com
Fixes: 38607c62b34b ("fs/dax: properly refcount fs dax pages")
Signed-off-by: Seunguk Shin <seunguk.shin@arm.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reported-by: Kiara Grouwstra <cinereal@riseup.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-08-03 13:34:55 +01:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
|
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-11-30 11:05:06 -05:00
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
|
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2018-11-30 11:05:06 -05:00
|
|
|
|
2018-11-27 13:16:33 -08:00
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-30 11:05:06 -05:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
|
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
|
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
|
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-11-27 13:16:34 -08:00
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-11-06 13:11:57 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
|
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2018-11-16 14:37:06 -05:00
|
|
|
|
2018-11-30 11:05:06 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
|
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2023-08-23 00:13:14 +01:00
|
|
|
|
|
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
2018-06-12 09:46:30 -04:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
|
|
|
|
|
2018-11-30 11:05:06 -05:00
|
|
|
|
2018-07-13 21:50:16 -07:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:28 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-06-28 19:35:04 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-06-28 19:35:04 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
|
|
|
|
|
2018-11-16 15:19:13 -05:00
|
|
|
|
2020-07-29 11:44:36 +08:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-04-07 16:04:57 -07:00
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:12 -04:00
|
|
|
|
2021-05-04 18:32:51 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-05-04 18:32:51 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-08 11:32:12 +11:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2025-02-28 14:30:57 +11:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
|
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:13 -04:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
2018-05-17 13:03:48 -04:00
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
2020-03-03 14:58:21 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm, fs, dax: handle layout changes to pinned dax mappings
Background:
get_user_pages() in the filesystem pins file backed memory pages for
access by devices performing dma. However, it only pins the memory pages
not the page-to-file offset association. If a file is truncated the
pages are mapped out of the file and dma may continue indefinitely into
a page that is owned by a device driver. This breaks coherency of the
file vs dma, but the assumption is that if userspace wants the
file-space truncated it does not matter what data is inbound from the
device, it is not relevant anymore. The only expectation is that dma can
safely continue while the filesystem reallocates the block(s).
Problem:
This expectation that dma can safely continue while the filesystem
changes the block map is broken by dax. With dax the target dma page
*is* the filesystem block. The model of leaving the page pinned for dma,
but truncating the file block out of the file, means that the filesytem
is free to reallocate a block under active dma to another file and now
the expected data-incoherency situation has turned into active
data-corruption.
Solution:
Defer all filesystem operations (fallocate(), truncate()) on a dax mode
file while any page/block in the file is under active dma. This solution
assumes that dma is transient. Cases where dma operations are known to
not be transient, like RDMA, have been explicitly disabled via
commits like 5f1d43de5416 "IB/core: disable memory registration of
filesystem-dax vmas".
The dax_layout_busy_page() routine is called by filesystems with a lock
held against mm faults (i_mmap_lock) to find pinned / busy dax pages.
The process of looking up a busy page invalidates all mappings
to trigger any subsequent get_user_pages() to block on i_mmap_lock.
The filesystem continues to call dax_layout_busy_page() until it finally
returns no more active pages. This approach assumes that the page
pinning is transient, if that assumption is violated the system would
have likely hung from the uncompleted I/O.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reported-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2018-03-09 17:44:31 -08:00
|
|
|
|
|
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
2018-03-28 15:40:41 -04:00
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 15:40:41 -04:00
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
|
|
|
|
|
2018-03-28 15:40:41 -04:00
|
|
|
|
|
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2018-03-28 15:40:41 -04:00
|
|
|
|
2021-05-04 18:32:51 -07:00
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
|
|
|
|
|
dax: Wake up all waiters after invalidating dax entry
I am seeing missed wakeups which ultimately lead to a deadlock when I am
using virtiofs with DAX enabled and running "make -j". I had to mount
virtiofs as rootfs and also reduce to dax window size to 256M to reproduce
the problem consistently.
So here is the problem. put_unlocked_entry() wakes up waiters only
if entry is not null as well as !dax_is_conflict(entry). But if I
call multiple instances of invalidate_inode_pages2() in parallel,
then I can run into a situation where there are waiters on
this index but nobody will wake these waiters.
invalidate_inode_pages2()
invalidate_inode_pages2_range()
invalidate_exceptional_entry2()
dax_invalidate_mapping_entry_sync()
__dax_invalidate_entry() {
xas_lock_irq(&xas);
entry = get_unlocked_entry(&xas, 0);
...
...
dax_disassociate_entry(entry, mapping, trunc);
xas_store(&xas, NULL);
...
...
put_unlocked_entry(&xas, entry);
xas_unlock_irq(&xas);
}
Say a fault in in progress and it has locked entry at offset say "0x1c".
Now say three instances of invalidate_inode_pages2() are in progress
(A, B, C) and they all try to invalidate entry at offset "0x1c". Given
dax entry is locked, all tree instances A, B, C will wait in wait queue.
When dax fault finishes, say A is woken up. It will store NULL entry
at index "0x1c" and wake up B. When B comes along it will find "entry=0"
at page offset 0x1c and it will call put_unlocked_entry(&xas, 0). And
this means put_unlocked_entry() will not wake up next waiter, given
the current code. And that means C continues to wait and is not woken
up.
This patch fixes the issue by waking up all waiters when a dax entry
has been invalidated. This seems to fix the deadlock I am facing
and I can make forward progress.
Reported-by: Sergio Lopez <slp@redhat.com>
Fixes: ac401cc78242 ("dax: New fault locking")
Reviewed-by: Jan Kara <jack@suse.cz>
Suggested-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Link: https://lore.kernel.org/r/20210428190314.1865312-4-vgoyal@redhat.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
2021-04-28 15:03:14 -04:00
|
|
|
|
2018-03-28 15:40:41 -04:00
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
|
|
|
|
|
2018-03-28 15:40:41 -04:00
|
|
|
|
2023-03-24 10:28:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-03-24 10:28:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
2025-02-28 14:31:01 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:00 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:02 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:00 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:01 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:00 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:01 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:00 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:01 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:00 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:02 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:02 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2016-08-10 17:22:44 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2016-05-12 18:29:18 +02:00
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:48 +01:00
|
|
|
|
2015-02-16 15:59:02 -08:00
|
|
|
|
2021-11-29 11:22:00 +01:00
|
|
|
|
2021-11-29 11:21:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:48 +01:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-18 13:55:41 -06:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2022-05-13 15:10:58 -07:00
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:47 +01:00
|
|
|
|
|
|
|
|
|
2015-02-16 15:59:02 -08:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2015-02-16 15:59:02 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2022-12-01 15:28:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2017-11-01 16:36:40 +01:00
|
|
|
|
2016-02-05 15:36:55 -08:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2022-12-01 15:28:54 +00:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
|
|
|
|
|
2018-01-31 16:17:36 -08:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:54 +00:00
|
|
|
|
2019-06-06 11:10:28 +02:00
|
|
|
|
|
|
|
|
|
2017-12-22 22:02:48 -08:00
|
|
|
|
2025-03-18 18:33:01 -07:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-06 11:10:28 +02:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
2017-11-01 16:36:40 +01:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
2022-12-01 15:28:54 +00:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
|
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2017-10-14 17:13:45 -07:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-10 16:36:56 -07:00
|
|
|
|
|
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2019-03-01 11:12:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-14 17:13:45 -07:00
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2019-03-01 11:12:41 -08:00
|
|
|
|
|
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 21:16:49 +01:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2019-03-01 11:12:41 -08:00
|
|
|
|
2016-12-14 15:07:53 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:12 -04:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
2019-03-01 11:12:41 -08:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
2016-12-14 15:07:47 -08:00
|
|
|
|
2021-04-28 15:03:13 -04:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-02-26 15:19:55 -08:00
|
|
|
|
2020-01-03 13:33:07 -05:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-05-04 18:32:45 -07:00
|
|
|
|
2016-02-26 15:19:55 -08:00
|
|
|
|
|
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
2017-05-08 16:00:10 -07:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
2018-03-28 16:03:45 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-22 15:10:47 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:32 +08:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
2021-11-29 11:21:48 +01:00
|
|
|
|
2022-06-03 13:37:32 +08:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
2022-06-03 13:37:32 +08:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2022-06-03 13:37:32 +08:00
|
|
|
|
|
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
2022-06-03 13:37:32 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2017-11-01 16:36:33 +01:00
|
|
|
|
2015-02-16 15:59:02 -08:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-11-08 10:15:50 +05:30
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-15 12:13:25 -06:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
|
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
dax: use common 4k zero page for dax mmap reads
When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.
This has three major drawbacks:
1) It consumes memory unnecessarily. For every 4k page that is read via
a DAX mmap() over a hole, we allocate a new page cache page. This
means that if you read 1GiB worth of pages, you end up using 1GiB of
zeroed memory. This is easily visible by looking at the overall
memory consumption of the system or by looking at /proc/[pid]/smaps:
7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 1048576 kB
Pss: 1048576 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 1048576 kB
Private_Dirty: 0 kB
Referenced: 1048576 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
2) It is slower than using a common zero page because each page fault
has more work to do. Instead of just inserting a common zero page we
have to allocate a page cache page, zero it, and then insert it. Here
are the average latencies of dax_load_hole() as measured by ftrace on
a random test box:
Old method, using zeroed page cache pages: 3.4 us
New method, using the common 4k zero page: 0.8 us
This was the average latency over 1 GiB of sequential reads done by
this simple fio script:
[global]
size=1G
filename=/root/dax/data
fallocate=none
[io]
rw=read
ioengine=mmap
3) The fact that we had to check for both DAX exceptional entries and
for page cache pages in the radix tree made the DAX code more
complex.
Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead. As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.
Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page. If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.
This solution also removes the extra memory consumption. Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:
7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12 /root/dax/data
Size: 1048576 kB
Rss: 0 kB
Pss: 0 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 0 kB
Referenced: 0 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Locked: 0 kB
Overall system memory consumption is similarly improved.
Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable. The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:
"To be able to use the common 4k zero page in DAX we need to have our
PTE fault path look more like our PMD fault path where a PTE entry
can be marked as dirty and writeable as it is first inserted rather
than waiting for a follow-up dax_pfn_mkwrite() =>
finish_mkwrite_fault() call.
Right now we can rely on having a dax_pfn_mkwrite() call because we
can distinguish between these two cases in do_wp_page():
case 1: 4k zero page => writable DAX storage
case 2: read-only DAX storage => writeable DAX storage
This distinction is made by via vm_normal_page(). vm_normal_page()
returns false for the common 4k zero page, though, just as it does
for DAX ptes. Instead of special casing the DAX + 4k zero page case
we will simplify our DAX PTE page fault sequence so that it matches
our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
We will instead use dax_iomap_fault() to handle write-protection
faults.
This means that insert_pfn() needs to follow the lead of
insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
'mkwrite' is set insert_pfn() will do the work that was previously
done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"
Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-06 16:18:43 -07:00
|
|
|
|
2026-02-11 12:31:39 +02:00
|
|
|
|
2018-09-11 21:27:44 -07:00
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2017-11-03 13:30:42 -04:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2017-09-06 16:18:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-26 20:28:27 +00:00
|
|
|
|
fs/dax: use vmf_insert_folio_pmd() to insert the huge zero folio
Let's convert to vmf_insert_folio_pmd().
There is a theoretical change in behavior: in the unlikely case there is
already something mapped, we'll now still call trace_dax_pmd_load_hole()
and return VM_FAULT_NOPAGE.
Previously, we would have returned VM_FAULT_FALLBACK, and the caller would
have zapped the PMD to try a PTE fault.
However, that behavior was different to other PTE+PMD faults, when there
would already be something mapped, and it's not even clear if it could be
triggered.
Assuming the huge zero folio is already mapped, all good, no need to
fallback to PTEs.
Assuming there is already a leaf page table ... the behavior would be
just like when trying to insert a PMD mapping a folio through
dax_fault_iter()->vmf_insert_folio_pmd().
Assuming there is already something else mapped as PMD? It sounds like a
BUG, and the behavior would be just like when trying to insert a PMD
mapping a folio through dax_fault_iter()->vmf_insert_folio_pmd().
So, it sounds reasonable to not handle huge zero folios differently to
inserting PMDs mapping folios when there already is something mapped.
Link: https://lkml.kernel.org/r/20250811112631.759341-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Juegren Gross <jgross@suse.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Mariano Pache <npache@redhat.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-08-11 13:26:24 +02:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2024-03-26 20:28:27 +00:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
fs/dax: use vmf_insert_folio_pmd() to insert the huge zero folio
Let's convert to vmf_insert_folio_pmd().
There is a theoretical change in behavior: in the unlikely case there is
already something mapped, we'll now still call trace_dax_pmd_load_hole()
and return VM_FAULT_NOPAGE.
Previously, we would have returned VM_FAULT_FALLBACK, and the caller would
have zapped the PMD to try a PTE fault.
However, that behavior was different to other PTE+PMD faults, when there
would already be something mapped, and it's not even clear if it could be
triggered.
Assuming the huge zero folio is already mapped, all good, no need to
fallback to PTEs.
Assuming there is already a leaf page table ... the behavior would be
just like when trying to insert a PMD mapping a folio through
dax_fault_iter()->vmf_insert_folio_pmd().
Assuming there is already something else mapped as PMD? It sounds like a
BUG, and the behavior would be just like when trying to insert a PMD
mapping a folio through dax_fault_iter()->vmf_insert_folio_pmd().
So, it sounds reasonable to not handle huge zero folios differently to
inserting PMDs mapping folios when there already is something mapped.
Link: https://lkml.kernel.org/r/20250811112631.759341-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Juegren Gross <jgross@suse.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Mariano Pache <npache@redhat.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-08-11 13:26:24 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
fs/dax: use vmf_insert_folio_pmd() to insert the huge zero folio
Let's convert to vmf_insert_folio_pmd().
There is a theoretical change in behavior: in the unlikely case there is
already something mapped, we'll now still call trace_dax_pmd_load_hole()
and return VM_FAULT_NOPAGE.
Previously, we would have returned VM_FAULT_FALLBACK, and the caller would
have zapped the PMD to try a PTE fault.
However, that behavior was different to other PTE+PMD faults, when there
would already be something mapped, and it's not even clear if it could be
triggered.
Assuming the huge zero folio is already mapped, all good, no need to
fallback to PTEs.
Assuming there is already a leaf page table ... the behavior would be
just like when trying to insert a PMD mapping a folio through
dax_fault_iter()->vmf_insert_folio_pmd().
Assuming there is already something else mapped as PMD? It sounds like a
BUG, and the behavior would be just like when trying to insert a PMD
mapping a folio through dax_fault_iter()->vmf_insert_folio_pmd().
So, it sounds reasonable to not handle huge zero folios differently to
inserting PMDs mapping folios when there already is something mapped.
Link: https://lkml.kernel.org/r/20250811112631.759341-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Juegren Gross <jgross@suse.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Mariano Pache <npache@redhat.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-08-11 13:26:24 +02:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
fs/dax: use vmf_insert_folio_pmd() to insert the huge zero folio
Let's convert to vmf_insert_folio_pmd().
There is a theoretical change in behavior: in the unlikely case there is
already something mapped, we'll now still call trace_dax_pmd_load_hole()
and return VM_FAULT_NOPAGE.
Previously, we would have returned VM_FAULT_FALLBACK, and the caller would
have zapped the PMD to try a PTE fault.
However, that behavior was different to other PTE+PMD faults, when there
would already be something mapped, and it's not even clear if it could be
triggered.
Assuming the huge zero folio is already mapped, all good, no need to
fallback to PTEs.
Assuming there is already a leaf page table ... the behavior would be
just like when trying to insert a PMD mapping a folio through
dax_fault_iter()->vmf_insert_folio_pmd().
Assuming there is already something else mapped as PMD? It sounds like a
BUG, and the behavior would be just like when trying to insert a PMD
mapping a folio through dax_fault_iter()->vmf_insert_folio_pmd().
So, it sounds reasonable to not handle huge zero folios differently to
inserting PMDs mapping folios when there already is something mapped.
Link: https://lkml.kernel.org/r/20250811112631.759341-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Juegren Gross <jgross@suse.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Mariano Pache <npache@redhat.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-08-11 13:26:24 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
fsdax: dax_unshare_iter needs to copy entire blocks
The code that copies data from srcmap to iomap in dax_unshare_iter is
very very broken, which bfoster's recent fsx changes have exposed.
If the pos and len passed to dax_file_unshare are not aligned to an
fsblock boundary, the iter pos and length in the _iter function will
reflect this unalignment.
dax_iomap_direct_access always returns a pointer to the start of the
kmapped fsdax page, even if its pos argument is in the middle of that
page. This is catastrophic for data integrity when iter->pos is not
aligned to a page, because daddr/saddr do not point to the same byte in
the file as iter->pos. Hence we corrupt user data by copying it to the
wrong place.
If iter->pos + iomap_length() in the _iter function not aligned to a
page, then we fail to copy a full block, and only partially populate the
destination block. This is catastrophic for data confidentiality
because we expose stale pmem contents.
Fix both of these issues by aligning copy_pos/copy_len to a page
boundary (remember, this is fsdax so 1 fsblock == 1 base page) so that
we always copy full blocks.
We're not done yet -- there's no call to invalidate_inode_pages2_range,
so programs that have the file range mmap'd will continue accessing the
old memory mapping after the file metadata updates have completed.
Be careful with the return value -- if the unshare succeeds, we still
need to return the number of bytes that the iomap iter thinks we're
operating on.
Cc: ruansy.fnst@fujitsu.com
Fixes: d984648e428b ("fsdax,xfs: port unshare to fsdax")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Link: https://lore.kernel.org/r/172796813328.1131942.16777025316348797355.stgit@frogsfrogsfrogs
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <brauner@kernel.org>
2024-10-03 08:09:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
2025-02-24 09:47:57 -05:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
2024-10-03 08:09:16 -07:00
|
|
|
|
2025-02-24 09:47:57 -05:00
|
|
|
|
fsdax: dax_unshare_iter needs to copy entire blocks
The code that copies data from srcmap to iomap in dax_unshare_iter is
very very broken, which bfoster's recent fsx changes have exposed.
If the pos and len passed to dax_file_unshare are not aligned to an
fsblock boundary, the iter pos and length in the _iter function will
reflect this unalignment.
dax_iomap_direct_access always returns a pointer to the start of the
kmapped fsdax page, even if its pos argument is in the middle of that
page. This is catastrophic for data integrity when iter->pos is not
aligned to a page, because daddr/saddr do not point to the same byte in
the file as iter->pos. Hence we corrupt user data by copying it to the
wrong place.
If iter->pos + iomap_length() in the _iter function not aligned to a
page, then we fail to copy a full block, and only partially populate the
destination block. This is catastrophic for data confidentiality
because we expose stale pmem contents.
Fix both of these issues by aligning copy_pos/copy_len to a page
boundary (remember, this is fsdax so 1 fsblock == 1 base page) so that
we always copy full blocks.
We're not done yet -- there's no call to invalidate_inode_pages2_range,
so programs that have the file range mmap'd will continue accessing the
old memory mapping after the file metadata updates have completed.
Be careful with the return value -- if the unshare succeeds, we still
need to return the number of bytes that the iomap iter thinks we're
operating on.
Cc: ruansy.fnst@fujitsu.com
Fixes: d984648e428b ("fsdax,xfs: port unshare to fsdax")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Link: https://lore.kernel.org/r/172796813328.1131942.16777025316348797355.stgit@frogsfrogsfrogs
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <brauner@kernel.org>
2024-10-03 08:09:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
fsdax: dax_unshare_iter needs to copy entire blocks
The code that copies data from srcmap to iomap in dax_unshare_iter is
very very broken, which bfoster's recent fsx changes have exposed.
If the pos and len passed to dax_file_unshare are not aligned to an
fsblock boundary, the iter pos and length in the _iter function will
reflect this unalignment.
dax_iomap_direct_access always returns a pointer to the start of the
kmapped fsdax page, even if its pos argument is in the middle of that
page. This is catastrophic for data integrity when iter->pos is not
aligned to a page, because daddr/saddr do not point to the same byte in
the file as iter->pos. Hence we corrupt user data by copying it to the
wrong place.
If iter->pos + iomap_length() in the _iter function not aligned to a
page, then we fail to copy a full block, and only partially populate the
destination block. This is catastrophic for data confidentiality
because we expose stale pmem contents.
Fix both of these issues by aligning copy_pos/copy_len to a page
boundary (remember, this is fsdax so 1 fsblock == 1 base page) so that
we always copy full blocks.
We're not done yet -- there's no call to invalidate_inode_pages2_range,
so programs that have the file range mmap'd will continue accessing the
old memory mapping after the file metadata updates have completed.
Be careful with the return value -- if the unshare succeeds, we still
need to return the number of bytes that the iomap iter thinks we're
operating on.
Cc: ruansy.fnst@fujitsu.com
Fixes: d984648e428b ("fsdax,xfs: port unshare to fsdax")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Link: https://lore.kernel.org/r/172796813328.1131942.16777025316348797355.stgit@frogsfrogsfrogs
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <brauner@kernel.org>
2024-10-03 08:09:48 -07:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
fsdax: dax_unshare_iter needs to copy entire blocks
The code that copies data from srcmap to iomap in dax_unshare_iter is
very very broken, which bfoster's recent fsx changes have exposed.
If the pos and len passed to dax_file_unshare are not aligned to an
fsblock boundary, the iter pos and length in the _iter function will
reflect this unalignment.
dax_iomap_direct_access always returns a pointer to the start of the
kmapped fsdax page, even if its pos argument is in the middle of that
page. This is catastrophic for data integrity when iter->pos is not
aligned to a page, because daddr/saddr do not point to the same byte in
the file as iter->pos. Hence we corrupt user data by copying it to the
wrong place.
If iter->pos + iomap_length() in the _iter function not aligned to a
page, then we fail to copy a full block, and only partially populate the
destination block. This is catastrophic for data confidentiality
because we expose stale pmem contents.
Fix both of these issues by aligning copy_pos/copy_len to a page
boundary (remember, this is fsdax so 1 fsblock == 1 base page) so that
we always copy full blocks.
We're not done yet -- there's no call to invalidate_inode_pages2_range,
so programs that have the file range mmap'd will continue accessing the
old memory mapping after the file metadata updates have completed.
Be careful with the return value -- if the unshare succeeds, we still
need to return the number of bytes that the iomap iter thinks we're
operating on.
Cc: ruansy.fnst@fujitsu.com
Fixes: d984648e428b ("fsdax,xfs: port unshare to fsdax")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Link: https://lore.kernel.org/r/172796813328.1131942.16777025316348797355.stgit@frogsfrogsfrogs
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <brauner@kernel.org>
2024-10-03 08:09:48 -07:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:57 -05:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:52 -05:00
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:57 -05:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-02 08:02:13 -07:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
2024-10-02 08:02:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2022-12-01 15:32:33 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:35 +08:00
|
|
|
|
2021-11-29 11:21:51 +01:00
|
|
|
|
2022-06-03 13:37:35 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:51 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:35 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-15 12:13:25 -06:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:35 +08:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:35 +08:00
|
|
|
|
2021-11-29 11:21:51 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2016-05-09 10:47:04 +02:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2021-11-29 11:21:51 +01:00
|
|
|
|
2022-12-01 15:28:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:52 +00:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
2022-06-30 10:04:18 -07:00
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:51 -05:00
|
|
|
|
2016-05-09 10:47:04 +02:00
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:58 +01:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2021-11-29 11:21:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
2022-12-01 15:28:52 +00:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2022-12-01 15:28:52 +00:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2018-03-16 17:36:44 -07:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
2016-08-10 17:10:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:52 +00:00
|
|
|
|
2023-03-24 10:28:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
2016-08-10 17:10:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2021-11-29 11:21:48 +01:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2022-04-22 16:45:06 -06:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2017-02-03 13:13:26 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2022-05-13 15:10:58 -07:00
|
|
|
|
2023-06-15 12:13:25 -06:00
|
|
|
|
2022-04-22 16:45:06 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2023-06-15 12:13:25 -06:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-01 15:28:52 +00:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-04-22 16:45:06 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2018-03-16 17:36:44 -07:00
|
|
|
|
2017-05-29 21:56:49 -07:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2018-03-16 17:36:44 -07:00
|
|
|
|
2018-05-02 06:46:33 -07:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
2018-03-16 17:36:44 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2017-01-27 13:31:42 -08:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-08 11:32:46 +11:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-08 11:32:46 +11:00
|
|
|
|
2017-01-27 23:20:26 -08:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:58 +01:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2025-07-24 08:12:13 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-25 11:20:50 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-08 14:43:13 -05:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
|
|
|
|
|
2025-09-30 13:42:57 +08:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
2017-02-08 14:43:13 -05:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2020-02-05 14:15:58 -05:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
2020-02-05 14:15:58 -05:00
|
|
|
|
2025-02-24 09:47:50 -05:00
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
2021-08-10 18:33:13 -07:00
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:49 +10:00
|
|
|
|
|
|
|
|
|
2016-11-08 11:32:46 +11:00
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
2016-10-19 14:34:31 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-01-05 00:54:11 +05:30
|
|
|
|
2016-10-19 14:34:31 +02:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:47 +01:00
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:47 +01:00
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2022-12-01 15:28:53 +00:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2022-06-03 13:37:34 +08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2018-01-07 16:38:43 -05:00
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:58 +01:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
dax: add tracepoints to dax_iomap_pte_fault()
Patch series "second round of tracepoints for DAX".
This second round of DAX tracepoint patches adds tracing to the PTE
fault path (dax_iomap_pte_fault(), dax_pfn_mkwrite(), dax_load_hole(),
dax_insert_mapping()) and to the writeback path
(dax_writeback_mapping_range(), dax_writeback_one()).
The purpose of this tracing is to give us a high level view of what DAX
is doing, whether faults are being serviced by PMDs or PTEs, and by real
storage or by zero pages covering holes.
I do have some patches nearly ready which also add tracing to
grab_mapping_entry() and dax_insert_mapping_entry(). These are more
targeted at logging how we are interacting with the radix tree, how we
use empty entries for locking, whether we "downgrade" huge zero pages to
4k PTE sized allocations, etc. In the end it seemed to me that this
might be too detailed to have as constantly present tracepoints, but if
anyone sees value in having tracepoints like this in the DAX code
permanently (Jan?), please let me know and I'll add those last two
patches.
All these tracepoints were done to be consistent with the style of the
XFS tracepoints and with the existing DAX PMD tracepoints.
This patch (of 6):
Add tracepoints to dax_iomap_pte_fault(), following the same logging
conventions as the rest of DAX.
Here is an example fault that initially tries to be serviced by the PMD
fault handler but which falls back to PTEs because the VMA isn't large
enough to hold a PMD:
small-1086 [005] ....
71.140014: xfs_filemap_huge_fault: dev 259:0 ino 0x1003
small-1086 [005] ....
71.140027: dax_pmd_fault: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 vm_start 0x10200000 vm_end 0x10500000 pgoff 0x220 max_pgoff 0x1400
small-1086 [005] ....
71.140028: dax_pmd_fault_done: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 vm_start 0x10200000 vm_end 0x10500000 pgoff 0x220 max_pgoff 0x1400 FALLBACK
small-1086 [005] ....
71.140035: dax_pte_fault: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 pgoff 0x220
small-1086 [005] ....
71.140396: dax_pte_fault_done: dev 259:0 ino 0x1003 shared WRITE|ALLOW_RETRY|KILLABLE|USER address 0x10420000 pgoff 0x220 MAJOR|NOPAGE
Link: http://lkml.kernel.org/r/20170221195116.13278-2-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-05-08 16:00:00 -07:00
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-05-12 15:46:57 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-02 14:46:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:57:59 +10:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
2017-06-02 14:46:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2018-01-07 16:38:43 -05:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2016-09-19 11:24:50 +10:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2017-02-22 15:40:03 -08:00
|
|
|
|
|
|
|
|
|
2017-02-22 15:39:50 -08:00
|
|
|
|
dax: fix deadlock due to misaligned PMD faults
In DAX there are two separate places where the 2MiB range of a PMD is
defined.
The first is in the page tables, where a PMD mapping inserted for a
given address spans from (vmf->address & PMD_MASK) to ((vmf->address &
PMD_MASK) + PMD_SIZE - 1). That is, from the 2MiB boundary below the
address to the 2MiB boundary above the address.
So, for example, a fault at address 3MiB (0x30 0000) falls within the
PMD that ranges from 2MiB (0x20 0000) to 4MiB (0x40 0000).
The second PMD range is in the mapping->page_tree, where a given file
offset is covered by a radix tree entry that spans from one 2MiB aligned
file offset to another 2MiB aligned file offset.
So, for example, the file offset for 3MiB (pgoff 768) falls within the
PMD range for the order 9 radix tree entry that ranges from 2MiB (pgoff
512) to 4MiB (pgoff 1024).
This system works so long as the addresses and file offsets for a given
mapping both have the same offsets relative to the start of each PMD.
Consider the case where the starting address for a given file isn't 2MiB
aligned - say our faulting address is 3 MiB (0x30 0000), but that
corresponds to the beginning of our file (pgoff 0). Now all the PMDs in
the mapping are misaligned so that the 2MiB range defined in the page
tables never matches up with the 2MiB range defined in the radix tree.
The current code notices this case for DAX faults to storage with the
following test in dax_pmd_insert_mapping():
if (pfn_t_to_pfn(pfn) & PG_PMD_COLOUR)
goto unlock_fallback;
This test makes sure that the pfn we get from the driver is 2MiB
aligned, and relies on the assumption that the 2MiB alignment of the pfn
we get back from the driver matches the 2MiB alignment of the faulting
address.
However, faults to holes were not checked and we could hit the problem
described above.
This was reported in response to the NVML nvml/src/test/pmempool_sync
TEST5:
$ cd nvml/src/test/pmempool_sync
$ make TEST5
You can grab NVML here:
https://github.com/pmem/nvml/
The dmesg warning you see when you hit this error is:
WARNING: CPU: 13 PID: 2900 at fs/dax.c:641 dax_insert_mapping_entry+0x2df/0x310
Where we notice in dax_insert_mapping_entry() that the radix tree entry
we are about to replace doesn't match the locked entry that we had
previously inserted into the tree. This happens because the initial
insertion was done in grab_mapping_entry() using a pgoff calculated from
the faulting address (vmf->address), and the replacement in
dax_pmd_load_hole() => dax_insert_mapping_entry() is done using
vmf->pgoff.
In our failure case those two page offsets (one calculated from
vmf->address, one using vmf->pgoff) point to different order 9 radix
tree entries.
This failure case can result in a deadlock because the radix tree unlock
also happens on the pgoff calculated from vmf->address. This means that
the locked radix tree entry that we swapped in to the tree in
dax_insert_mapping_entry() using vmf->pgoff is never unlocked, so all
future faults to that 2MiB range will block forever.
Fix this by validating that the faulting address's PMD offset matches
the PMD offset from the start of the file. This check is done at the
very beginning of the fault and covers faults that would have mapped to
storage as well as faults to holes. I left the COLOUR check in
dax_pmd_insert_mapping() in place in case we ever hit the insanity
condition where the alignment of the pfn we get from the driver doesn't
match the alignment of the userspace address.
Link: http://lkml.kernel.org/r/20170822222436.18926-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: "Slusarz, Marcin" <marcin.slusarz@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-25 15:55:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
dax: fix deadlock due to misaligned PMD faults
In DAX there are two separate places where the 2MiB range of a PMD is
defined.
The first is in the page tables, where a PMD mapping inserted for a
given address spans from (vmf->address & PMD_MASK) to ((vmf->address &
PMD_MASK) + PMD_SIZE - 1). That is, from the 2MiB boundary below the
address to the 2MiB boundary above the address.
So, for example, a fault at address 3MiB (0x30 0000) falls within the
PMD that ranges from 2MiB (0x20 0000) to 4MiB (0x40 0000).
The second PMD range is in the mapping->page_tree, where a given file
offset is covered by a radix tree entry that spans from one 2MiB aligned
file offset to another 2MiB aligned file offset.
So, for example, the file offset for 3MiB (pgoff 768) falls within the
PMD range for the order 9 radix tree entry that ranges from 2MiB (pgoff
512) to 4MiB (pgoff 1024).
This system works so long as the addresses and file offsets for a given
mapping both have the same offsets relative to the start of each PMD.
Consider the case where the starting address for a given file isn't 2MiB
aligned - say our faulting address is 3 MiB (0x30 0000), but that
corresponds to the beginning of our file (pgoff 0). Now all the PMDs in
the mapping are misaligned so that the 2MiB range defined in the page
tables never matches up with the 2MiB range defined in the radix tree.
The current code notices this case for DAX faults to storage with the
following test in dax_pmd_insert_mapping():
if (pfn_t_to_pfn(pfn) & PG_PMD_COLOUR)
goto unlock_fallback;
This test makes sure that the pfn we get from the driver is 2MiB
aligned, and relies on the assumption that the 2MiB alignment of the pfn
we get back from the driver matches the 2MiB alignment of the faulting
address.
However, faults to holes were not checked and we could hit the problem
described above.
This was reported in response to the NVML nvml/src/test/pmempool_sync
TEST5:
$ cd nvml/src/test/pmempool_sync
$ make TEST5
You can grab NVML here:
https://github.com/pmem/nvml/
The dmesg warning you see when you hit this error is:
WARNING: CPU: 13 PID: 2900 at fs/dax.c:641 dax_insert_mapping_entry+0x2df/0x310
Where we notice in dax_insert_mapping_entry() that the radix tree entry
we are about to replace doesn't match the locked entry that we had
previously inserted into the tree. This happens because the initial
insertion was done in grab_mapping_entry() using a pgoff calculated from
the faulting address (vmf->address), and the replacement in
dax_pmd_load_hole() => dax_insert_mapping_entry() is done using
vmf->pgoff.
In our failure case those two page offsets (one calculated from
vmf->address, one using vmf->pgoff) point to different order 9 radix
tree entries.
This failure case can result in a deadlock because the radix tree unlock
also happens on the pgoff calculated from vmf->address. This means that
the locked radix tree entry that we swapped in to the tree in
dax_insert_mapping_entry() using vmf->pgoff is never unlocked, so all
future faults to that 2MiB range will block forever.
Fix this by validating that the faulting address's PMD offset matches
the PMD offset from the start of the file. This check is done at the
very beginning of the fault and covers faults that would have mapped to
storage as well as faults to holes. I left the COLOUR check in
dax_pmd_insert_mapping() in place in case we ever hit the insanity
condition where the alignment of the pfn we get from the driver doesn't
match the alignment of the userspace address.
Link: http://lkml.kernel.org/r/20170822222436.18926-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: "Slusarz, Marcin" <marcin.slusarz@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-25 15:55:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
dax: fix deadlock due to misaligned PMD faults
In DAX there are two separate places where the 2MiB range of a PMD is
defined.
The first is in the page tables, where a PMD mapping inserted for a
given address spans from (vmf->address & PMD_MASK) to ((vmf->address &
PMD_MASK) + PMD_SIZE - 1). That is, from the 2MiB boundary below the
address to the 2MiB boundary above the address.
So, for example, a fault at address 3MiB (0x30 0000) falls within the
PMD that ranges from 2MiB (0x20 0000) to 4MiB (0x40 0000).
The second PMD range is in the mapping->page_tree, where a given file
offset is covered by a radix tree entry that spans from one 2MiB aligned
file offset to another 2MiB aligned file offset.
So, for example, the file offset for 3MiB (pgoff 768) falls within the
PMD range for the order 9 radix tree entry that ranges from 2MiB (pgoff
512) to 4MiB (pgoff 1024).
This system works so long as the addresses and file offsets for a given
mapping both have the same offsets relative to the start of each PMD.
Consider the case where the starting address for a given file isn't 2MiB
aligned - say our faulting address is 3 MiB (0x30 0000), but that
corresponds to the beginning of our file (pgoff 0). Now all the PMDs in
the mapping are misaligned so that the 2MiB range defined in the page
tables never matches up with the 2MiB range defined in the radix tree.
The current code notices this case for DAX faults to storage with the
following test in dax_pmd_insert_mapping():
if (pfn_t_to_pfn(pfn) & PG_PMD_COLOUR)
goto unlock_fallback;
This test makes sure that the pfn we get from the driver is 2MiB
aligned, and relies on the assumption that the 2MiB alignment of the pfn
we get back from the driver matches the 2MiB alignment of the faulting
address.
However, faults to holes were not checked and we could hit the problem
described above.
This was reported in response to the NVML nvml/src/test/pmempool_sync
TEST5:
$ cd nvml/src/test/pmempool_sync
$ make TEST5
You can grab NVML here:
https://github.com/pmem/nvml/
The dmesg warning you see when you hit this error is:
WARNING: CPU: 13 PID: 2900 at fs/dax.c:641 dax_insert_mapping_entry+0x2df/0x310
Where we notice in dax_insert_mapping_entry() that the radix tree entry
we are about to replace doesn't match the locked entry that we had
previously inserted into the tree. This happens because the initial
insertion was done in grab_mapping_entry() using a pgoff calculated from
the faulting address (vmf->address), and the replacement in
dax_pmd_load_hole() => dax_insert_mapping_entry() is done using
vmf->pgoff.
In our failure case those two page offsets (one calculated from
vmf->address, one using vmf->pgoff) point to different order 9 radix
tree entries.
This failure case can result in a deadlock because the radix tree unlock
also happens on the pgoff calculated from vmf->address. This means that
the locked radix tree entry that we swapped in to the tree in
dax_insert_mapping_entry() using vmf->pgoff is never unlocked, so all
future faults to that 2MiB range will block forever.
Fix this by validating that the faulting address's PMD offset matches
the PMD offset from the start of the file. This check is done at the
very beginning of the fault and covers faults that would have mapped to
storage as well as faults to holes. I left the COLOUR check in
dax_pmd_insert_mapping() in place in case we ever hit the insanity
condition where the alignment of the pfn we get from the driver doesn't
match the alignment of the userspace address.
Link: http://lkml.kernel.org/r/20170822222436.18926-1-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: "Slusarz, Marcin" <marcin.slusarz@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-25 15:55:36 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-29 11:21:58 +01:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-22 15:39:50 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2017-02-22 15:39:50 -08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2017-02-22 15:39:50 -08:00
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
2021-08-10 18:33:14 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2016-10-19 14:34:31 +02:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-10-19 14:34:31 +02:00
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2017-05-12 15:47:00 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2017-05-12 15:47:00 -07:00
|
|
|
|
2017-06-02 14:46:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:57:59 +10:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2017-06-02 14:46:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2023-06-21 14:02:56 +01:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2017-05-12 15:47:00 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2018-03-29 22:58:27 -04:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
|
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
|
|
|
|
|
2017-02-22 15:39:50 -08:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
2021-08-10 18:33:15 -07:00
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2017-02-27 14:26:44 -08:00
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
dax: add struct iomap based DAX PMD support
DAX PMDs have been disabled since Jan Kara introduced DAX radix tree based
locking. This patch allows DAX PMDs to participate in the DAX radix tree
based locking scheme so that they can be re-enabled using the new struct
iomap based fault handlers.
There are currently three types of DAX 4k entries: 4k zero pages, 4k DAX
mappings that have an associated block allocation, and 4k DAX empty
entries. The empty entries exist to provide locking for the duration of a
given page fault.
This patch adds three equivalent 2MiB DAX entries: Huge Zero Page (HZP)
entries, PMD DAX entries that have associated block allocations, and 2 MiB
DAX empty entries.
Unlike the 4k case where we insert a struct page* into the radix tree for
4k zero pages, for HZP we insert a DAX exceptional entry with the new
RADIX_DAX_HZP flag set. This is because we use a single 2 MiB zero page in
every 2MiB hole mapping, and it doesn't make sense to have that same struct
page* with multiple entries in multiple trees. This would cause contention
on the single page lock for the one Huge Zero Page, and it would break the
page->index and page->mapping associations that are assumed to be valid in
many other places in the kernel.
One difficult use case is when one thread is trying to use 4k entries in
radix tree for a given offset, and another thread is using 2 MiB entries
for that same offset. The current code handles this by making the 2 MiB
user fall back to 4k entries for most cases. This was done because it is
the simplest solution, and because the use of 2MiB pages is already
opportunistic.
If we were to try to upgrade from 4k pages to 2MiB pages for a given range,
we run into the problem of how we lock out 4k page faults for the entire
2MiB range while we clean out the radix tree so we can insert the 2MiB
entry. We can solve this problem if we need to, but I think that the cases
where both 2MiB entries and 4K entries are being used for the same range
will be rare enough and the gain small enough that it probably won't be
worth the complexity.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-11-08 11:34:45 +11:00
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
2017-11-01 16:36:39 +01:00
|
|
|
|
2018-01-07 16:38:43 -05:00
|
|
|
|
2017-11-01 16:36:38 +01:00
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
|
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
2018-01-07 16:38:43 -05:00
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
2017-11-01 16:36:39 +01:00
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
mm,fs,dax: change ->pmd_fault to ->huge_fault
Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-02-24 14:56:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2018-03-27 13:39:38 -04:00
|
|
|
|
|
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
|
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2025-02-28 14:30:58 +11:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2019-07-03 23:21:25 -04:00
|
|
|
|
|
|
|
|
|
2021-04-28 15:03:13 -04:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
2025-02-28 14:31:14 +11:00
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2018-06-07 17:04:29 -07:00
|
|
|
|
|
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-18 21:23:35 +01:00
|
|
|
|
2025-06-19 18:58:05 +10:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-28 11:48:03 -04:00
|
|
|
|
2017-11-01 16:36:43 +01:00
|
|
|
|
|
|
|
|
|
2022-06-03 13:37:36 +08:00
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2022-06-03 13:37:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2022-06-03 13:37:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
|
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2025-09-19 14:42:50 -07:00
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2022-06-03 13:37:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2022-06-03 13:37:36 +08:00
|
|
|
|
2022-12-01 15:31:41 +00:00
|
|
|
|
|
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2023-03-22 07:25:58 +00:00
|
|
|
|
2025-02-24 09:47:53 -05:00
|
|
|
|
2022-12-01 15:31:41 +00:00
|
|
|
|
2025-02-24 09:47:56 -05:00
|
|
|
|
2022-06-03 13:37:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
