If there are gced dirty pages and normal dirty pages in the mapping
of one inode, we might writeback them alternately with discontinuous
block address, resulting in low performance.
This patch introduces f2fs_write_cache_pages with codes copied from
write_cache_pages in mm/page-writeback.c.
In this function, we refactor flow with two steps:
1) writeback all cold type pages.
2) writeback all non-cold type pages.
By using this method, f2fs will writeback dirty pages with the same
temperature in bunch mode, it makes writeouted block being with
more continuous address, so they can be merged as much as possible
in f2fs bio cache, and also it will reduce the chance of submiting
small IO from block layer.
Test environment: 8g nokia sd card (very old sd card, but it shows
better effect when testing with this patch, and with a 32g kingston
sd card, I didn't see much more improvement).
Test step:
1. touch testfile;
2. truncate -s 512K testfile;
3. write all pages with odd index;
4. trigger gc by ioctl;
5. write all pages with even index;
6. time fsync testfile.
before:
real 0m0.402s
user 0m0.000s
sys 0m0.000s
after:
real 0m0.143s
user 0m0.004s
sys 0m0.004s
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes to return correct error number of ->setxattr, which
is reported by xfstest tests/generic/026 as below:
generic/026 - output mismatch
--- tests/generic/026.out
+++ results/generic/026.out.bad
@@ -4,6 +4,6 @@
1 below acl max
acl max
1 above acl max
-chacl: cannot set access acl on "largeaclfile": Argument list too long
+chacl: cannot set access acl on "largeaclfile": Numerical result out of range
use 16 aces
use 17 aces
...
Ran: generic/026
Failures: generic/026
Failed 1 of 1 tests
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, since 'commit 4531929e39 ("f2fs: move grabing orphan
pages out of protection region")' was committed, in write_orphan_inodes(),
we will grab all meta page in a batch before we use them under spinlock,
so that we can avoid large time delay of grabbing meta pages under
spinlock.
Now, 'commit d6c67a4fee ("f2fs: revmove spin_lock for
write_orphan_inodes")' remove the spinlock in write_orphan_inodes,
so there is no issue we describe above, we'd better recover to move
the grab operation to original place for readability.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
With cost-benifit method, background gc will consider old section with
fewer valid blocks as candidate victim, these old blocks in section will
be treated as cold data, and laterly will be moved into cold segment.
But if the gcing page is attached by user through buffered or mmaped
write, we should reset the page as non-cold one, because this page may
have more opportunity for further updating.
So fix to add clearing code for the missed 'mmap' case.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When background gc is off, the only way to trigger gc is executing
a force gc in some operations who wants to grab space in disk.
The executing condition is limited: to execute force gc, we should
wait for the time when there is almost no more free section for LFS
allocation. This seems not reasonable for our user who wants to
control triggering gc by himself.
This patch introduces F2FS_IOC_GARBAGE_COLLECT interface for
triggering garbage collection by using ioctl. It provides our users
one more option to trigger gc.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch moves extent cache related code from data.c into extent_cache.c
since extent cache is independent feature, and its codes are not relate to
others in data.c, it's better for us to maintain them in separated place.
There is no functionality change, but several small coding style fixes
including:
* rename __drop_largest_extent to f2fs_drop_largest_extent for exporting;
* rename misspelled word 'untill' to 'until';
* remove unneeded 'return' in the end of f2fs_destroy_extent_tree().
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Since only parts of extents longer than F2FS_MIN_EXTENT_LEN will
be kept in extent cache after split, extents already shorter than
F2FS_MIN_EXTENT_LEN don't need to try split at all.
Signed-off-by: Fan Li <fanofcode.li@samsung.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes to update page flag (e.g. Uptodate/cold flag) in
->write_begin.
Otherwise, page will be non-uptodate when we try to write entire
page, and cold data flag in page will not be clean when gced page
is being rewritten.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If an extent_tree entry has a zero reference count, we can drop it from the
cache in higher priority rather than currently referencing entries.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In ->writepages, we use writepages mutex lock to serialize all block
address allocation and page submitting pairs from different inodes.
This method makes our delayed dirty pages of one inode being written
continously as many as possible.
But there is one problem that we did not submit current cached bio in
protection region of writepages mutex lock, so there is a small chance
that we submit the one of other thread's as below, resulting in
splitting more bios.
thread 1 thread 2
->writepages
lock(writepages)
->write_cache_pages
unlock(writepages)
lock(writepages)
->write_cache_pages
->f2fs_submit_merged_bio
->writepage
unlock(writepages)
fs_mark-6535 [002] .... 2242.270230: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, DATA, sector = 5766152, size = 524288
fs_mark-6536 [000] .... 2242.270361: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, DATA, sector = 5767176, size = 4096
fs_mark-6536 [000] .... 2242.270370: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, NODE, sector = 8138112, size = 4096
fs_mark-6535 [002] .... 2242.270776: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, DATA, sector = 5767184, size = 516096
This may really increase time of block layer works, and may cause
larger IO lantency.
This patch moves the submitting operation into region of writepages
mutex lock to avoid bio splits when concurrently writebacking is
intensive.
my test environment: virtual machine,
intel cpu i5 2500, 8GB size memory, 4GB size ramdisk
time fs_mark -t 16 -L 1 -s 524288 -S 1 -d /mnt/f2fs/
before:
real 0m4.244s
user 0m0.088s
sys 0m12.336s
after:
real 0m3.822s
user 0m0.072s
sys 0m10.760s
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When testing with fs_mark, some blocks were written out as cold
data which were mixed with warm data, resulting in splitting more
bios.
This is because fs_mark will create file with random filename as
below:
559551ee~~~~~~~~15Z29OCC05JCKQP60JQ42MKV
559551ee~~~~~~~~NZAZ6X8OA8LHIIP6XD0L58RM
559551ef~~~~~~~~B15YDSWAK789HPSDZKYTW6WM
559551f1~~~~~~~~2DAE5DPS79785BUNTFWBEMP3
559551f1~~~~~~~~1MYDY0BKSQCJPI32Q8C514RM
559551f1~~~~~~~~YQOTMAOMN5CVRFOUNI026MP4
559551f3~~~~~~~~1WF42LPRTQJNPPGR3EINKMPE
559551f3~~~~~~~~8Y2NRK7CEPPAA02LY936PJPG
They are regarded as cold file since their filename are ended with
multimedia files' extension, but this should be wrong as we only
match the extension of filename, not the whole one.
In this patch, we try to fix the format of multimedia filename to:
"filename + '.' + extension", then we set cold file only its
filename matches the format.
So after this change, it will reduce the probability we set the
wrong cold file, also it helps a little for fs_mark's performance
on f2fs.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds missed trace file in maintainer-ship of f2fs,
so it completes the description of files maintained in f2fs,
and also it allows people to find correct mailing list by using
get_maintainer.pl when only patching the trace file of f2fs.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This makes the function check_dnode have a return type of bool
due to this particular function only ever returning either one
or zero as its return value and changes the name of the function
to is_alive in order to better explain this function's intended
work of checking if a dnode is still in use by the filesystem.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
[Jaegeuk Kim: change the return value check for the renamed function]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Because of the extent shrinker or other -ENOMEM scenarios, it cannot guarantee
that the largest extent would be cached in the tree all the time.
Instead of relying on extent_tree, we can simply check the cached one in extent
tree accordingly.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We don't need to handle the duplicate extent information.
The integrated rule is:
- update on-disk extent with largest one tracked by in-memory extent_cache
- destroy extent_tree for the truncation case
- drop per-inode extent_cache by shrinker
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces a shrinker targeting to reduce memory footprint consumed
by a number of in-memory f2fs data structures.
In addition, it newly adds:
- sbi->umount_mutex to avoid data races on shrinker and put_super
- sbi->shruinker_run_no to not revisit objects
Note that the basic implementation was copied from fs/ubifs/shrinker.c
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch relocates cached_en not only to be covered by spin_lock, but also
to set once after checking out completely.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, f2fs_update_extent_cache() updates in-memory extent_cache all the
time, and then finally preserves its up-to-date extent into on-disk one during
f2fs_evict_inode.
But, in the following scenario:
1. mount
2. open & write an extent X
3. f2fs_evict_inode; on-disk extent is X
4. open & update the extent X with Y
5. sync; trigger checkpoint
6. power-cut
after power-on, f2fs should serve extent Y, but we have an on-disk extent X.
This causes a failure on xfstests/311.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes wrong calculation on block address field when an extent is
split.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
For newly added fallocate types, it should convert inline_data before handling
block swapping.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Before iput is called, the inode number used by a bad inode can be reassigned
to other new inode, resulting in any abnormal behaviors on the new inode.
This should not happen for the new inode.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The write_checkpoint can update stat information, so we should destroy the stat
structure after it.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Dirty page can be exist in mapping of newly created symlink, but previously
we did not maintain the counting of dirty page for symlink like we maintained
for regular/directory, so the counting we lookuped should be wrong.
This patch adds missed dirty page counting for symlink to fix this issue.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The key_put() function tests whether its argument is NULL and then
returns immediately. Thus the test around the call is not needed.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Traditionally, each cgroup controller implemented whatever interface
it wanted leading to interfaces which are widely inconsistent.
Examining the requirements of the controllers readily yield that there
are only a few control schemes shared among all.
Two major controllers already had to implement new interface for the
unified hierarchy due to significant structural changes. Let's take
the chance to establish common conventions throughout all controllers.
This patch defines CGROUP_WEIGHT_MIN/DFL/MAX to be used on all weight
based control knobs and documents the conventions that controllers
should follow on the unified hierarchy. Except for io.weight knob,
all existing unified hierarchy knobs are already compliant. A
follow-up patch will update io.weight.
v2: Added descriptions of min, low and high knobs.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Peter Zijlstra <peterz@infradead.org>
The GPIO subsystem provides dummy GPIO consumer functions if GPIOLIB is
not enabled. Hence drivers that depend on GPIOLIB, but use GPIO consumer
functionality only, can still be compiled if GPIOLIB is not enabled.
Relax the dependency on GPIOLIB if COMPILE_TEST is enabled, where
appropriate.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
When building a Tx Command for management frames, we are lacking
a check for action frames, for which we should set a different
pm_timeout. This cause the fw to stay awake for 100TU after each
such frame is transmitted, resulting an excessive power consumption.
Signed-off-by: Avri Altman <avri.altman@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Family 8000 products has 2 embedded processors, the first
known as LMAC (lower MAC) and implements the functionality from
previous products, the second one is known as UMAC (upper MAC)
and is used mainly for driver offloads as well as new features.
The UMAC is typically “less” real-time than the LMAC and is used
for higher level controls.
The UMAC's code/data size is estimated to be in the mega-byte arena,
taking into account the code it needs to replace in the driver and
the set of new features.
In order to allow the UMAC to execute code that is bigger than its code
memory, we allow the UMAC embedded processor to page out code pages on
DRAM.
When the device is slave on the bus(SDIO) the driver saves the UMAC's
image pages in blocks of 32K in the DRAM and sends the layout of the
pages to the FW. When the FW wants load / unload pages, it creates an
interrupt, and the driver uploads / downloads the page to an address in
the a specific address on the device's memory.
The driver can support up to 1 MB of pages.
Add paging mechanism for the UMAC on SDIO in order to allow the program to
use a larger virtual space while using less physical memory on the device
itself.
Signed-off-by: Matti Gottlieb <matti.gottlieb@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
As a preperation for multiple RX queues change the RBD
allocation model.
The new model includes a background allocator. The allocator is
called by the interrupt handler when there are two released
buffers by the queue, and the allocator starts allocating eight
pages per request.
When the queue has released 8 pages it tries claiming the
request. If the pages are not ready - it keeps claiming.
This new model should make sure that RBDs are always available
across the multiple queues.
The RBDs are transferred between the allocator and the queue.
The queue moves the free RBDs upon freeing them to the allocator.
The allocator moves them back to the queue's possession when the
request is claimed.
The allocator has an initial pool to make sure there are always RBDs
available for the request completion.
Release of the buffers at exit is done per pools - the allocator
frees its own initial pool and the queue frees its own pool.
Existing code refactor -
-Queue's initial pool is the size of the queue only as the allocation
of the new buffers no longer uses this pool.
-Removal of replenish background work, and replenish calls in the
interrupt handler and restock().
-The replenish() and the rxq used_list are used only during
initialization.
-Moved page allocation to a new function for code reuse.
New code -
Allocator code - new structure and functions.
Interrupt handler uses the allocator functions for replenishing buffers.
Reuse of the restock() method.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Change the FW debug trigger tlv to include a monitor only
option. Setting this option to true will cause fw dump triggers
to only collect monitor data and skip other dumps such as
SMEM, SRAM, CSR, PRPH, etc.
This option is used when accessing the different parts of the
firmware memory is not wanted and can cause unwanted behavior
like when debugging TX latency.
Signed-off-by: Oren Givon <oren.givon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Allow frag SKBs in PCIe and advertise the maximum number of frags
to the opmode. As a fallback. linearize the SKB if it exceeds the
maximum number of fragments. This allows using the hardware better
(filling more TBs) and should improve performance when used by the
opmode.
Also adjust tracing to be able to deal with this.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Gscan is a scan feature which is supported on certain devices only,
hence the need for a TLV flag for it. For devices that support gscan
store the gscan capabilities advertised by the FW so the driver can
report it to upper layers.
Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Set max TX aggregation size for 8260 SDIO devices series to 40 frames.
Fine tune max RX aggregation size - change it to 21.
Signed-off-by: Alexander Bondar <alexander.bondar@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Our firmware scheduler used to suffer from false wake-up on 500 time units.
We had to came up with a formula to address this buggy behavior.
Now that our firmware is fixed, we can go back to our old policy.
Signed-off-by: Avri Altman <avri.altman@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Add support for dumping all the RBs in the RX queue
when FW error occurs.
This will assist debugging.
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Print the CPU1 and CPU2 secured boot status registers from the NIC
to indicate a SYSASSERT during secured engine unlocking process
on init/protocol image.
Signed-off-by: Dor Shaish <dor.shaish@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Family 8000 products has 2 embedded processors, the first
known as LMAC (lower MAC) and implements the functionality from
previous products, the second one is known as UMAC (upper MAC)
and is used mainly for driver offloads as well as new features.
The UMAC is typically “less” real-time than the LMAC and is used
for higher level controls.
The UMAC's code/data size is estimated to be in the mega-byte arena,
taking into account the code it needs to replace in the driver and
the set of new features.
In order to allow the UMAC to execute code that is bigger than its code
memory, we allow the UMAC embedded processor to page out code pages on
DRAM.
When the device is master on the bus(PCI) the driver saves the UMAC's
image pages in blocks of 32K in the DRAM and sends the layout of the
pages to the FW. The FW can load / unload the pages on its own.
The driver can support up to 1 MB of pages.
Add paging mechanism for the UMAC on PCI in order to allow the program
to use a larger virtual space while using less physical memory on the
device.
Signed-off-by: Eran Harary <eran.harary@intel.com>
Signed-off-by: Matti Gottlieb <matti.gottlieb@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Add support for extended command id in triggers handling.
Extended command id header contains group id in addition to command id.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Add support for extended firmware event header that contains
a group id as well as the command id.
Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Add support for extended command id in notification system.
Extended command id header contains group id in addition to command id.
Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
This makes various functions in the file rs.c void due to these
functions never returning a error code to signal to their callers
if and how they have failed to complete their intended work.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Move the TX PN assignment (for CCMP only) to the driver. This prepares
the driver for future DSO (driver segmentation offload) where it will
split an SKB into multiple MPDUs by itself.
For TDLS, split out the CCMP TX command handling so that it won't get
a PN assigned, the firmware assigns the PN in that case.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Commit 37b1ef31a5 ("workqueue: move
flush_scheduled_work() to workqueue.h") moved the exported non GPL
flush_scheduled_work() from a function to an inline wrapper.
Unfortunately, it directly calls flush_workqueue() which is a GPL function.
This has the effect of changing the licensing requirement for this function
and makes it unavailable to non GPL modules.
See commit ad7b1f841f ("workqueue: Make
schedule_work() available again to non GPL modules") for precedent.
Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
In order to guarantee that the patched instruction stream is visible to
a CPU, that CPU must execute an isb instruction after any related cache
maintenance has completed.
The instruction patching routines in kernel/insn.c get this right for
things like jump labels and ftrace, but the alternatives patching omits
it entirely leaving secondary cores in a potential limbo between the old
and the new code.
This patch adds an isb following the secondary polling loop in the
altenatives patching.
Signed-off-by: Will Deacon <will.deacon@arm.com>
