After lots of debugging on an occasional DMA ERR issue, I realized
that the desc structures which we point the dma hardware are being
allocated out of regular memory. This means when we fill the desc
structures, that data doesn't always get flushed out to memory by
the time we start the dma transfer, resulting in the dma engine getting
some null values, resulting in a DMA ERR on the first irq.
Thus, this patch adopts mechanism similar to the zx296702_dma of
allocating the desc structures from a dma pool, so the memory caching
rules are properly set to avoid this issue.
Cc: Zhangfei Gao <zhangfei.gao@linaro.org>
Cc: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Cc: Maxime Ripard <maxime.ripard@free-electrons.com>
Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Andy Green <andy@warmcat.com>
Acked-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Signed-off-by: John Stutlz <john.stultz@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
As it was before, as soon as the DMAC IP felt there was an error
he would return IRQ_NONE since no actual transfer had completed.
After spinning on that for 100K interrupts, Linux yanks the IRQ with
a "nobody cared" error.
This patch lets it handle the interrupt and keep the IRQ alive.
Cc: Zhangfei Gao <zhangfei.gao@linaro.org>
Cc: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Cc: Maxime Ripard <maxime.ripard@free-electrons.com>
Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Andy Green <andy@warmcat.com>
Acked-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Signed-off-by: Andy Green <andy.green@linaro.org>
[jstultz: Forward ported to mainline]
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
The offsets for ERR1 and ERR2 are wrong actually.
That's why you can never clear an error.
Cc: Zhangfei Gao <zhangfei.gao@linaro.org>
Cc: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Cc: Maxime Ripard <maxime.ripard@free-electrons.com>
Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Andy Green <andy@warmcat.com>
Acked-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Signed-off-by: Andy Green <andy.green@linaro.org>
[jstultz: Forward ported to mainline]
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Max burst len is a 4-bit field, but at the moment it's clipped with
a 5-bit constant... reduce it to that which can be expressed
Cc: Zhangfei Gao <zhangfei.gao@linaro.org>
Cc: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Cc: Maxime Ripard <maxime.ripard@free-electrons.com>
Cc: Vinod Koul <vinod.koul@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Mark Brown <broonie@kernel.org>
Cc: Andy Green <andy@warmcat.com>
Acked-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Signed-off-by: Andy Green <andy.green@linaro.org>
[jstultz: Forward ported to mainline]
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
* revert a recent wext patch, which Ben Hutchings noticed was
wrong, and it turns out not to be necessary for any driver
* fix an infinite loop that can occur under certain conditions
in mac80211's TDLS code (depending on regulatory information)
* add a cfg80211_get_station() static inline when cfg80211 isn't
built, to allow other modules to not have to depend on it for it
-----BEGIN PGP SIGNATURE-----
iQIcBAABCgAGBQJXxSR4AAoJEGt7eEactAAdl3EP/AiUwqrYqbLnnFy6C7obFS3p
eBBMxQAZbT+q+fFlZvqRrt5tPdkYriPLhm/0sAzuapnyS+Q6seNJ/vPoo91uC1jU
ZI/j97v9NwUtRLfNCq+0Jwvs7ma0U1VEcPV9wDdV5JgnKk0Z1CUIcsErYr1+v0YQ
EpRwxczhzJNTULW36UP7RvVQpxwIGldPhxSZ0t1uHWaYTFliaTlnJUAk0ql44Lmm
WLvoMSjFgX99P11ToCe81MPEzF2IXILvxPwtNZmn5tldEN2xknKEoEmmbN65fYDf
OIJIJ3s1CijQvnkgXtU0RWWCMnyOoJjsLckgSDdy0euhbS5xRIfxBN2n+kqaI9WV
a/aIvWNNhvAy2vNdWUJk0FrVBnDjlTtG1afIEAgJyP7uxTQqepQfyaRENLtH+kKe
lWbOITUZztyagGIn8Bv1pDrrqwO+fSjiEsVEVAQMMmNKBpUWf8urhDQmabCLYGDB
Nxh2e3wjv5ZQ+55uJIGRDCcPIrddh86FVtQBqTID+86r4a1RwPaWfhzFZYVj84fg
504UzwtYlw1ITUhGbdMwribLVkwtBMVuEvpPrh6avwzS8wAH4upkhp4GGl/tfd/Z
De0LqpxCKbDiI+VmmDo8FD4nx4wu4nTYIaecLjNoUSXhbjbPyI6V8/hIXqKiQUA3
ObkKlGicZJmhMa4zna0q
=dFzv
-----END PGP SIGNATURE-----
Merge tag 'mac80211-for-davem-2016-08-30' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211
Johannes Berg says:
====================
Three little fixes:
* revert a recent wext patch, which Ben Hutchings noticed was
wrong, and it turns out not to be necessary for any driver
* fix an infinite loop that can occur under certain conditions
in mac80211's TDLS code (depending on regulatory information)
* add a cfg80211_get_station() static inline when cfg80211 isn't
built, to allow other modules to not have to depend on it for it
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
We get 2 warnings about global functions without a declaration in the
scsi driver when building with W=1:
drivers/scsi/scsi_lib.c:467:6: warning: no previous prototype for 'scsi_requeue_run_queue' [-Wmissing-prototypes]
drivers/scsi/scsi_lib.c:2609:6: warning: no previous prototype for 'scsi_evt_thread' [-Wmissing-prototypes]
In fact, both functions are declared in drivers/scsi/scsi_scan.c but
need to move them into scsi_priv.h.
Signed-off-by: Baoyou Xie <baoyou.xie@linaro.org>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Trivial fix to spelling mistakes in ql_dbg messages.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Himanshu Madhani <himanshu.madhani@qlogic.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Allow i.MX7 to work with the imx-sdma driver.
Signed-off-by: Fabio Estevam <fabio.estevam@nxp.com>
Tested-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
With commit 3a0f64bfa9 ("mac_scsi: Fix pseudo DMA implementation")
some versions of gcc now warn:
In file included from drivers/scsi/mac_scsi.c:335:
drivers/scsi/NCR5380.h:295: warning: `NCR5380_poll_politely' declared inline after being called
drivers/scsi/NCR5380.h:295: warning: previous declaration of `NCR5380_poll_politely' was here
Avoid this by defining NCR5380_poll_politely() in NCR5380.h.
[mkp: checkpatch warnings]
Suggested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The driver that used the 'nodisconnect' parameter was removed in commit
565bae6a4a ("[SCSI] 53c7xx: kill driver"). Related documentation was
cleaned up in commit f37a7238d3 ("[SCSI] 53c7xx: fix removal
fallout"), except for the remaining two mentions that are removed here.
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Using list_move_tail() instead of list_del() + list_add_tail().
Signed-off-by: Wei Yongjun <weiyj.lk@gmail.com>
Reviewed-by: Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Trivial fix to spelling mistake in dev_warn message.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Julia pointed out beiscsi_boot_get_sinfo does not unlock mbox_lock on
nonemb_cmd memory allocation failure.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
alloc_wrb_handle already does memset zero of iscsi_wrb descriptor so
remove redundant memset in WRB submission paths.
Add pwrb_handle NULL check before memsett'ing pwrb.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
We know that 'ret' is not an error code because it has been tested a few
lines above. So, if one of these function fails, 0 will be returned
instead of an error code. Return -ENOMEM instead.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
'ret' needs to be set with error code if hba_setup_cid_tbls fails.
Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The 2nd parameter of 'find_first_bit' is the number of bits to search.
In this case, we are passing 'sizeof(unsigned long)' which is likely to
be 4.
It is likely that the number of bits in a long was expected here, so use
BITS_PER_LONG instead.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Acked-by: Sumit Saxena <sumit.saxena@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The MULDIV macro is essentially a duplicate of the more standard
mult_frac macro. Replace use of MULDIV with mult_frac & drop the
duplication.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Calculating the maximum timeout that a user can set via the
SG_SET_TIMEOUT ioctl involves multiplying INT_MAX by USER_HZ/HZ. If
USER_HZ is larger than HZ then this results in an overflow when
performed as a 32 bit integer calculation, resulting in compiler
warnings such as the following:
drivers/scsi/sg.c: In function 'sg_ioctl':
drivers/scsi/sg.c:91:67: warning: integer overflow in expression [-Woverflow]
#define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
^
drivers/scsi/sg.c:887:14: note: in expansion of macro 'MULDIV'
if (val >= MULDIV (INT_MAX, USER_HZ, HZ))
^
drivers/scsi/sg.c:91:67: warning: integer overflow in expression [-Woverflow]
#define MULDIV(X,MUL,DIV) ((((X % DIV) * MUL) / DIV) + ((X / DIV) * MUL))
^
drivers/scsi/sg.c:888:13: note: in expansion of macro 'MULDIV'
val = MULDIV (INT_MAX, USER_HZ, HZ);
^
Avoid this overflow by performing the (constant) arithmetic on 64 bit
integers, which ensures that overflow from multiplying the 32 bit values
cannot occur. When converting the result back to a 32 bit integer use
min_t to ensure that we don't simply truncate a value beyond INT_MAX to
a 32 bit integer, but instead use INT_MAX where the result was larger
than it. As the values are all compile time constant the 64 bit
arithmetic should have no runtime cost.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This commit appends a few _rcuidle suffixes to fix the following
RCU-used-from-idle bug:
> ===============================
> [ INFO: suspicious RCU usage. ]
> 4.6.0-rc5-next-20160426+ #1116 Not tainted
> -------------------------------
> include/trace/events/rpm.h:95 suspicious rcu_dereference_check() usage!
>
> other info that might help us debug this:
>
>
> RCU used illegally from idle CPU!
> rcu_scheduler_active = 1, debug_locks = 0
> RCU used illegally from extended quiescent state!
> 1 lock held by swapper/0/0:
> #0: (&(&dev->power.lock)->rlock){-.-...}, at: [<c052cc2c>] __rpm_callback+0x58/0x60
>
> stack backtrace:
> CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.6.0-rc5-next-20160426+ #1116
> Hardware name: Generic OMAP36xx (Flattened Device Tree)
> [<c0110290>] (unwind_backtrace) from [<c010c3a8>] (show_stack+0x10/0x14)
> [<c010c3a8>] (show_stack) from [<c047fd68>] (dump_stack+0xb0/0xe4)
> [<c047fd68>] (dump_stack) from [<c052d5d0>] (rpm_suspend+0x580/0x768)
> [<c052d5d0>] (rpm_suspend) from [<c052ec58>] (__pm_runtime_suspend+0x64/0x84)
> [<c052ec58>] (__pm_runtime_suspend) from [<c04bf25c>] (omap2_gpio_prepare_for_idle+0x5c/0x70)
> [<c04bf25c>] (omap2_gpio_prepare_for_idle) from [<c0125568>] (omap_sram_idle+0x140/0x244)
> [<c0125568>] (omap_sram_idle) from [<c01269dc>] (omap3_enter_idle_bm+0xfc/0x1ec)
> [<c01269dc>] (omap3_enter_idle_bm) from [<c0601bdc>] (cpuidle_enter_state+0x80/0x3d4)
> [<c0601bdc>] (cpuidle_enter_state) from [<c0183b08>] (cpu_startup_entry+0x198/0x3a0)
> [<c0183b08>] (cpu_startup_entry) from [<c0b00c0c>] (start_kernel+0x354/0x3c8)
> [<c0b00c0c>] (start_kernel) from [<8000807c>] (0x8000807c)
In the immortal words of Steven Rostedt, "*Whack* *Whack* *Whack*!!!"
Reported-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Tested-by: Guenter Roeck <linux@roeck-us.net>
WhACKED-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The workqueue "dm_bufio_wq" queues a single work item &dm_bufio_work so
it doesn't require execution ordering. Hence, alloc_workqueue() has
been used to replace the deprecated create_singlethread_workqueue().
The WQ_MEM_RECLAIM flag has been set since DM requires forward progress
under memory pressure.
Since there are fixed number of work items, explicit concurrency
limit is unnecessary here.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
If crypt_alloc_tfms() had to allocate multiple tfms and it failed before
the last allocation, then it would call crypt_free_tfms() and could free
pointers from uninitialized memory -- due to the crypt_free_tfms() check
for non-zero cc->tfms[i]. Fix by allocating zeroed memory.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
When dm-crypt processes writes, it allocates a new bio in
crypt_alloc_buffer(). The bio is allocated from a bio set and it can
have at most BIO_MAX_PAGES vector entries, however the incoming bio can be
larger (e.g. if it was allocated by bcache). If the incoming bio is
larger, bio_alloc_bioset() fails and an error is returned.
To avoid the error, we test for a too large bio in the function
crypt_map() and use dm_accept_partial_bio() to split the bio.
dm_accept_partial_bio() trims the current bio to the desired size and
asks DM core to send another bio with the rest of the data.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # v3.16+
The kthread_run() function returns either a valid task_struct or
ERR_PTR() value, check for NULL is invalid. This change fixes potential
for oops, e.g. in OOM situation.
Signed-off-by: Vladimir Zapolskiy <vz@mleia.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
The function acpi_parse_entries_array() has a limiting parameter,
max_entries, which tells the function to stop looking at subtables
once that limit has been reached. If the limit is reached, it is
reported. However, the logic is incorrect in that the loop to
examine all subtables will always report that zero subtables have
been ignored since it does not continue once the max_entries have
been reached.
One approach to fixing this would be to correct the logic so that
all subtables are examined, even if we have hit the max_entries, but
without executing all the callback functions. This could be risky
since we cannot guarantee that no callback will ever have side effects
that another callback depends on to work correctly.
So, the simplest approach is to just remove the part of the error
message that will always be incorrect.
Signed-off-by: Al Stone <ahs3@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The acpi_parse_entries_array() function currently returns the very first
time there is any error found by one of the callback functions, or if one
of the callbacks returns a non-zero value. However, the ACPI subtables
being traversed could still have valid entries that could be used by one
of the callback functions. And, if the comments are correct, that is
what should happen -- always traverse all of the subtables, calling as
many of the callbacks as possible.
This patch makes the function consistent with its description so that it
will properly invoke all callbacks for all matching entries, for all
subtables, instead of stopping abruptly as it does today.
This does change the semantics of using acpi_parse_entries_array(). In
examining all users of the function, none of them rely on the current
behavior; that is, there appears to be no assumption that either all
subtables are traversed and all callbacks invoked, or that the function
will return immediately on any error from a callback. Each callback
operates independently. Hence, there should be no functional change
due to this change in semantics.
Future patches being prepared will rely on this new behavior; indeed,
they were written assuming the acpi_parse_entries_array() function
operated as its comments describe. For example, a callback that
counts the number of subtables of a specific type can now be assured
that as many subtables as possible have been enumerated.
Signed-off-by: Al Stone <ahs3@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The static function acpi_parse_entries_array() is provided an array of
type struct acpi_subtable_proc that has a callback function and a count.
The count should reflect how many times the callback has been called.
However, the current code only increments the 0th element of the array,
regardless of the number of entries in the array, or which callback has
been invoked. The result is that we know the total number of callbacks
made but we cannot determine which callbacks were made, nor how often.
The fix is to index into the array of structs and increment the proper
counts.
There is one place in the x86 code for acpi_parse_madt_lapic_entries()
where the counts for each callback are used. If no LAPICs *and* no
X2APICs are found, an ENODEV is supposed to be returned; as it stands,
the count of X2APICs will always be zero, regardless of what is in the
MADT. Should there be no LAPICs, ENODEV will be returned in error, if
there are X2APICs in the MADT.
Otherwise, there are no other functional consequences of the count being
done as it currently is; all other uses simply check that the return value
from acpi_parse_entries_array() or passed back via its callers is either
non-zero, an error, or in one case just ignored.
In future patches, I will also need these counts to be correct; I need
to count the number of instances of subtables of certain types within
the MADT to determine whether or not an ACPI IORT is required or not,
and report when it is not present when it should be.
Signed-off-by: Al Stone <ahs3@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The UART driver, dw8250.c, needs some details regarding the
Designware UART. For ACPI enumerated devices the values are
hard-coded, but since the driver also reads the values from
device properties, providing them with build-in properties.
This allows us to later remove the hard-coded values from
the driver.
Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The UART driver, dw8250.c, needs some details regarding the
Designware UART. For ACPI enumerated devices the values are
hard-coded, but since the driver also reads the values from
device properties, providing them with build-in properties.
This allows us to later remove the hard-coded values from
the driver.
Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This fixes a ptrace vs fatal pending signals bug as manifested in
seccomp now that seccomp was reordered to happen after ptrace. The
short version is that seccomp should not attempt to call do_exit()
while fatal signals are pending under a tracer. The existing code was
trying to be as defensively paranoid as possible, but it now ends up
confusing ptrace. Instead, the syscall can just be skipped (which solves
the original concern that the do_exit() was addressing) and normal signal
handling, tracer notification, and process death can happen.
Paraphrasing from the original bug report:
If a tracee task is in a PTRACE_EVENT_SECCOMP trap, or has been resumed
after such a trap but not yet been scheduled, and another task in the
thread-group calls exit_group(), then the tracee task exits without the
ptracer receiving a PTRACE_EVENT_EXIT notification. Test case here:
https://gist.github.com/khuey/3c43ac247c72cef8c956ca73281c9be7
The bug happens because when __seccomp_filter() detects
fatal_signal_pending(), it calls do_exit() without dequeuing the fatal
signal. When do_exit() sends the PTRACE_EVENT_EXIT notification and
that task is descheduled, __schedule() notices that there is a fatal
signal pending and changes its state from TASK_TRACED to TASK_RUNNING.
That prevents the ptracer's waitpid() from returning the ptrace event.
A more detailed analysis is here:
https://github.com/mozilla/rr/issues/1762#issuecomment-237396255.
Reported-by: Robert O'Callahan <robert@ocallahan.org>
Reported-by: Kyle Huey <khuey@kylehuey.com>
Tested-by: Kyle Huey <khuey@kylehuey.com>
Fixes: 93e35efb8d ("x86/ptrace: run seccomp after ptrace")
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: James Morris <james.l.morris@oracle.com>
This patch adds documentation for the usage model of the control method lid
device on some buggy platforms.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Acked-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
On most platforms, _LID returning value, lid open/close events are all
reliable, but there are exceptions. Some AML tables report wrong initial
lid state [1], and some of them never report lid open state [2].
The usage model on such buggy platforms is:
1. The initial lid state returned from _LID is not reliable;
2. The lid open event is not reliable;
3. The lid close event is always reliable, used by the platform firmware to
trigger OSPM power saving operations.
This usage model is not compliant to the Linux SW_LID model as the Linux
userspace is very strict to the reliability of the open events.
In order not to trigger issues on such buggy platforms, the ACPI button
driver currently implements a lid_init_state=open quirk to send additional
"open" event after resuming. However, this is still not sufficient because:
1. Some special usage models (e.x., the dark resume scenario) cannot be
supported by this mode.
2. If a "close" event is not used to trigger "suspend", then the subsequent
"close" events cannot be seen by the userspace.
So we need to stop sending the additional "open" event and switch the
driver to lid_init_state=ignore mode and make sure the platform triggered
events can be reliably delivered to the userspace. The userspace programs
then can be changed to not to be strict to the "open" events on such buggy
platforms.
Why will the subsequent "close" events be lost? This is because the input
layer automatically filters redundant events for switch events. Thus given
that the buggy AML tables do not guarantee paired "open"/"close" events,
the ACPI button driver currently is not able to guarantee that the platform
triggered reliable events can be always be seen by the userspace via
SW_LID.
This patch adds a mechanism to insert lid events as a compensation for the
platform triggered ones to form a complete event switches in order to make
sure that the platform triggered events can always be reliably delivered
to the userspace. This essentially guarantees that the platform triggered
reliable "close" events will always be relibly delivered to the userspace.
However this mechanism is not suitable for lid_init_state=open/method as
it should not send the complement switch event for the unreliable initial
lid state notification. 2 unreliable events can trigger unexpected
behavior. Thus this patch only implements this mechanism for
lid_init_state=ignore.
Known issues:
1. Possible alternative approach
This approach is based on the fact that Linux requires a switch event
type for LID events. Another approach is to use key event type to
implement ACPI lid events.
With SW event type, since ACPI button driver inserts wrong lid events,
there could be a potential issue that an "open" event issued from some
AML update methods could result in a wrong "close" event to be delivered
to the userspace. While using KEY event type, there is no such problem.
However there may not be such a kind of real case, and if there is such
a case, it is worked around in this patch as the complement switch event
is only generated for "close" event in order to deliver the reliable
"close" event to the userspace.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=89211 # [1]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=106151 # [1]
Link: https://bugzilla.kernel.org/show_bug.cgi?id=106941 # [2]
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Suggested-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
PCC status field exposes an error bit(2) to indicate any errors during
the execution of last comamnd. This patch checks the error bit before
notifying success/failure to the cpufreq driver.
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There are several global variables in cppc driver that are related
to PCC channel used for CPPC. This patch collects all such
information into a single consolidated structure(cppc_pcc_data).
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The CPPC tables contain entries for per CPU feedback counters which
allows us to compute the delivered performance over a given interval
of time.
The math for delivered performance per the CPPCv5.0+ spec is:
reference perf * delta(delivered perf ctr)/delta(ref perf ctr)
Maintaining deltas of the counters in the kernel is messy, as it
depends on when the reads are triggered. (e.g. via the cpufreq
->get() interface). Also the ->get() interace only returns one
value, so cant return raw values. So instead, leave it to userspace
to keep track of raw values and do its math for CPUs it cares about.
delivered and reference perf counters are exposed via the same
sysfs file to avoid the potential "skid", if these values are read
individually from userspace.
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Ashwin Chaugule <ashwin.chaugule@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Compute the expected transition latency for frequency transitions
using the values from the PCCT tables when the desired perf
register is in PCC.
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Reviewed-by: Alexey Klimov <alexey.klimov@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CPPC defined in section 8.4.7 of ACPI 6.0 specification suggests
"To amortize the cost of PCC transactions, OSPM should read or write
all PCC registers via a single read or write command when possible"
This patch enables opportunistic batching of frequency transition
requests whenever the request happen to overlap in time.
Currently the access to pcc is serialized by a spin lock which does
not scale well as we increase the number of cores in the system. This
patch improves the scalability by allowing the differnt CPU cores to
update PCC subspace in parallel and by batching requests which will
reduce the certain types of operation(checking command completion bit,
ringing doorbell) by a significant margin.
Profiling shows significant improvement in the overall effeciency
to service freq. transition requests. With this patch we observe close
to 30% of the frequency transition requests being batched with other
requests while running apache bench on a ARM platform with 6
independent domains(or sets of related cpus).
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
We need to acquire pcc_lock only when we are accessing registers
that are in the PCC subspsace.
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
For cases where sys mapped CPC registers need to be accessed
frequently, it helps immensly to pre-map them rather than map
and unmap for each operation. e.g. case where feedback counters
are sys mem map registers.
Restructure cpc_read/write and the cpc_regs structure to allow
pre-mapping the system addresses and unmap them when the CPU exits.
Signed-off-by: Ashwin Chaugule <ashwin.chaugule@linaro.org>
Signed-off-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
ACPI 6.1 has a PCC HW-Reduced Communication Subspace type 2 intended for
use on HW-Reduce ACPI Platform, which requires read-modify-write sequence
to acknowledge doorbell interrupt. This patch provides the implementation
for the Communication Subspace Type 2.
Signed-off-by: Hoan Tran <hotran@apm.com>
Reviewed-by: Prashanth Prakash <pprakash@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Thus move sysfs_add_battery() after acpi_battery_get_state(), which doesn't
require the power_supply. Prevents possible hanged tasks if
acpi_battery_get_state() fails consistently (and takes a long time in doing
so) when called inside acpi_battery_add().
In this situation the battery module first calls sysfs_add_battery(),
which creates a power_supply, which spawns an async
power_supply_deferred_register_work() task, which shall try to hold the
parent battery device mutex (being already held) so this register work
is set up after device initialization. If initialization takes long enough
the thread will be eventually run and try to hold the mutex before
acpi_battery_add() had the chance to finish.
Eventually the 5 retries in acpi_battery_update_retry() fail, the error
state is propagated, and results in sysfs_remove_battery() being called
within the error handling paths of acpi_battery_add(), and the power_supply
tear down too.
This triggers a cancel_delayed_work_sync() of the deferred_register_work
task, which ends up in schedule(). The end result is that the deferred
task is blocked trying to acquire the parent device mutex, which is not
released because the thread doing initialization (and failure handling)
went to sleep awaiting for the deferred task to be cancelled.
The hanged tasks look like this:
INFO: task kworker/u8:0:6 blocked for more than 120 seconds.
...
Call Trace:
[<ffffffff815daec5>] schedule+0x35/0x80
[<ffffffff815dda3c>] schedule_timeout+0x1ec/0x250
[<ffffffff810a0572>] ? check_preempt_curr+0x52/0x90
[<ffffffff810a05c9>] ? ttwu_do_wakeup+0x19/0xe0
[<ffffffff815db915>] wait_for_common+0xc5/0x190
[<ffffffff810a1500>] ? wake_up_q+0x70/0x70
[<ffffffff815db9fd>] wait_for_completion+0x1d/0x20
[<ffffffff8108ffb1>] flush_work+0x111/0x1c0
[<ffffffff8108dfe0>] ? flush_workqueue_prep_pwqs+0x1a0/0x1a0
[<ffffffff810909af>] __cancel_work_timer+0x9f/0x1d0
[<ffffffff81090b13>] cancel_delayed_work_sync+0x13/0x20
[<ffffffff8147ac67>] power_supply_unregister+0x37/0xc0
[<ffffffffa058b03d>] sysfs_remove_battery+0x3d/0x52 [battery]
[<ffffffffa058bf3a>] acpi_battery_add+0x112/0x181 [battery]
[<ffffffff81366db6>] acpi_device_probe+0x54/0x19b
[<ffffffff81427e9c>] driver_probe_device+0x22c/0x440
[<ffffffff81428181>] __driver_attach+0xd1/0xf0
[<ffffffff814280b0>] ? driver_probe_device+0x440/0x440
[<ffffffff8142591c>] bus_for_each_dev+0x6c/0xc0
[<ffffffff8142758e>] driver_attach+0x1e/0x20
[<ffffffff81426fc3>] bus_add_driver+0x1c3/0x280
[<ffffffff81428b00>] driver_register+0x60/0xe0
[<ffffffff81366c80>] acpi_bus_register_driver+0x3b/0x43
[<ffffffffa0591040>] acpi_battery_init_async+0x1c/0x1e [battery]
[<ffffffff81099268>] async_run_entry_fn+0x48/0x150
[<ffffffff81090d09>] process_one_work+0x1e9/0x440
[<ffffffff81090fab>] worker_thread+0x4b/0x4f0
[<ffffffff81090f60>] ? process_one_work+0x440/0x440
[<ffffffff81096b58>] kthread+0xd8/0xf0
[<ffffffff815de97f>] ret_from_fork+0x1f/0x40
[<ffffffff81096a80>] ? kthread_worker_fn+0x180/0x180
INFO: task kworker/u8:4:282 blocked for more than 120 seconds.
...
Call Trace:
[<ffffffff810ad745>] ? put_prev_entity+0x35/0x8b0
[<ffffffff815daec5>] schedule+0x35/0x80
[<ffffffff815db14e>] schedule_preempt_disabled+0xe/0x10
[<ffffffff815dc533>] __mutex_lock_slowpath+0xb3/0x120
[<ffffffff815dc5bf>] mutex_lock+0x1f/0x30
[<ffffffff8147a59b>] power_supply_deferred_register_work+0x2b/0x50
[<ffffffff81090d09>] process_one_work+0x1e9/0x440
[<ffffffff81090fab>] worker_thread+0x4b/0x4f0
[<ffffffff81090f60>] ? process_one_work+0x440/0x440
[<ffffffff81090f60>] ? process_one_work+0x440/0x440
[<ffffffff81096b58>] kthread+0xd8/0xf0
[<ffffffff815de97f>] ret_from_fork+0x1f/0x40
[<ffffffff81096a80>] ? kthread_worker_fn+0x180/0x180
Making sysfs_add_battery() the last operation here means that the
power_supply won't be created yet when the acpi_add_battery() failure
handling happens, the deferred task won't even spawn, and
sysfs_remove_battery will just skip over the NULL battery->bat.
Signed-off-by: Carlos Garnacho <carlosg@gnome.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This patch enables the event freeze mode, flushing the EC event handling in
.suspend() callback. This feature is experimental, if it is bisected out to
be the cause of the real issues, please report the issues to the kernel
bugzilla for further root causing and improvement.
This mode eliminates useless _Qxx handling during the power saving
operations, thus can help to tune the power saving operations faster. Tests
show that this mode can efficiently block flooding _Qxx during the suspend
process and tune the speed of the suspend faster.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Tested-by: Todd E Brandt <todd.e.brandt@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
In the original EC driver, though the event handling is not explicitly
stopped, the EC driver is actually not able to handle events during the
noirq stage as the EC driver is not prepared to handle the EC events in the
polling mode. So if there is no advance_transaction() triggered, the EC
driver couldn't notice the EC events.
However, do we actually need to handle EC events during suspend/resume
stage? EC events are mostly useless for the suspend/resume period (key
strokes and battery/thermal updates, etc.,), and the useful ones (lid
close, power/sleep button press) should have already been delivered to the
OSPM to trigger the power saving operations.
Thus this patch implements acpi_ec_disable_event() to be a reverse call of
acpi_ec_enable_event(), with which, the EC driver is able to stop handling
the EC events in a position before entering the noirq stage.
Since there are actually 2 choices for us:
1. implement event handling in polling mode;
2. stop event handling before entering noirq stage.
And this patch only implements the second choice using .suspend() callback.
Thus this is experimental (first choice is better? or different hook
position is better?). This patch finally keeps the old behavior by default
and prepares a boot parameter to enable this feature.
The differences of the event handling availability between the old behavior
(this patch is not applied) and the new behavior (this patch is applied)
are as follows:
!FreezeEvents FreezeEvents
before suspend Y Y
suspend before EC Y Y
suspend after EC Y N
suspend_late Y N
suspend_noirq Y (actually N) N
resume_noirq Y (actually N) N
resume_late Y (actually N) N
resume before EC Y (actually N) N
resume after EC Y Y
after resume Y Y
Where "actually N" means if there is no EC transactions, the EC driver
is actually not able to notice the pending events.
We can see that FreezeEvents is the only approach now can actually flush
the EC event handling with both query commands and _Qxx evaluations
flushed, other modes can only flush the EC event handling with only query
commands flushed, _Qxx evaluations occurred after stopping the EC driver
may end up failure due to the failure of the EC transaction carried out in
the _Qxx control methods.
We also can see that this feature should be able to trigger some platform
notifications later than resuming other drivers.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Tested-by: Todd E Brandt <todd.e.brandt@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This patch makes 2 changes:
1. Restore old behavior
Originally, EC driver stops handling both events and transactions in
acpi_ec_block_transactions(), and restarts to handle transactions in
acpi_ec_unblock_transactions_early(), restarts to handle both events and
transactions in acpi_ec_unblock_transactions().
While currently, EC driver still stops handling both events and
transactions in acpi_ec_block_transactions(), but restarts to handle both
events and transactions in acpi_ec_unblock_transactions_early().
This patch tries to restore the old behavior by dropping
__acpi_ec_enable_event() from acpi_unblock_transactions_early().
2. Improve old behavior
However this still cannot fix the real issue as both of the
acpi_ec_unblock_xxx() functions are invoked in the noirq stage. Since the
EC driver actually doesn't implement the event handling in the polling
mode, re-enabling the event handling too early in the noirq stage could
result in the problem that if there is no triggering source causing
advance_transaction() to be invoked, pending SCI_EVT cannot be detected by
the EC driver and _Qxx cannot be triggered.
It actually makes sense to restart the event handling in any point during
resuming after the noirq stage. Just like the boot stage where the event
handling is enabled in .add(), this patch further moves
acpi_ec_enable_event() to .resume(). After doing that, the following 2
functions can be combined:
acpi_ec_unblock_transactions_early()/acpi_ec_unblock_transactions().
The differences of the event handling availability between the old behavior
(this patch isn't applied) and the new behavior (this patch is applied) are
as follows:
!Applied Applied
before suspend Y Y
suspend before EC Y Y
suspend after EC Y Y
suspend_late Y Y
suspend_noirq Y (actually N) Y (actually N)
resume_noirq Y (actually N) Y (actually N)
resume_late Y (actually N) Y (actually N)
resume before EC Y (actually N) Y (actually N)
resume after EC Y (actually N) Y
after resume Y (actually N) Y
Where "actually N" means if there is no triggering source, the EC driver
is actually not able to notice the pending SCI_EVT occurred in the noirq
stage. So we can clearly see that this patch has improved the situation.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Tested-by: Todd E Brandt <todd.e.brandt@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
After enabling the EC event handling, Linux is still in the noirq stage, if
there is no triggering source (EC transaction, GPE STS status),
advance_transaction() will not be invoked and SCI_EVT cannot be detected.
This patch adds one more triggering source after enabling the EC event
handling to poll the pending SCI_EVT.
Known issues:
1. Still no SCI_EVT triggering source
There could still be no SCI_EVT triggering source after handling the
first SCI_EVT (polled by this patch if any). Because after handling the
first SCI_EVT, Linux could still be in noirq stage and there could still
be no further triggering source in this stage. Then the second SCI_EVT
indicated during this stage still cannot be detected by the EC driver.
With this improvement applied, it is then possible to move
acpi_ec_enable_event() out of the noirq stage to fix this issue (if the
first SCI_EVT is handled out of the noirq stage, the follow-up SCI_EVTs
should be able to trigger IRQs).
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Tested-by: Todd E Brandt <todd.e.brandt@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There is a hidden logic in the EC driver:
1. During boot, EC_FLAGS_QUERY_PENDING is responsible for blocking event
handling;
2. During suspend, EC_FLAGS_STARTED is responsible for blocking event
handling.
This patch uses a new EC_FLAGS_QUERY_ENABLED flag to make this hidden
logic explicit and have code cleaned up. No functional change.
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Tested-by: Todd E Brandt <todd.e.brandt@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If device_add_property_set() is called for a device, a secondary fwnode
is allocated and assigned to the device but currently not freed once the
device is removed.
This can be triggered on Apple Macs if a Thunderbolt device is plugged
in on boot since Apple's NHI EFI driver sets a number of properties for
that device which are leaked on unplug.
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
