When encoder validation of a display mode fails, retry with less bandwidth
heavy YCbCr420 color mode, if available. This enables some HDMI 1.4 setups
to support 4k60Hz output, which previously failed silently.
On some setups, while the monitor and the gpu support display modes with
pixel clocks of up to 600MHz, the link encoder might not. This prevents
YCbCr444 and RGB encoding for 4k60Hz, but YCbCr420 encoding might still be
possible. However, which color mode is used is decided before the link
encoder capabilities are checked. This patch fixes the problem by retrying
to find a display mode with YCbCr420 enforced and using it, if it is
valid.
Reviewed-by: Harry Wentland <harry.wentland@amd.com> Signed-off-by: Werner Sembach <wse@tuxedocomputers.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
clang points out that the %hu format string does not match the type
of the variables here:
drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c:263:7: warning: format specifies type 'unsigned short' but the argument has type 'unsigned int' [-Wformat]
version_major, version_minor);
^~~~~~~~~~~~~
include/drm/drm_print.h:498:19: note: expanded from macro 'DRM_ERROR'
__drm_err(fmt, ##__VA_ARGS__)
~~~ ^~~~~~~~~~~
Change it to a regular %u, the same way a previous patch did for
another instance of the same warning.
Reviewed-by: Christian König <christian.koenig@amd.com> Reviewed-by: Tom Rix <trix@redhat.com> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Reviewed-by: Christian König <christian.koenig@amd.com> Suggested-by: Christian König <christian.koenig@amd.com> Signed-off-by: Tong Zhang <ztong0001@gmail.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why]
On NAVI14 CONFIG_UBSAN reported shift-out-of-bounds at
display_rq_dlg_calc_20v2.c:304:38
rq_param->misc.rq_c.blk256_height is 0 when chroma(*_c) is invalid.
dml_log2 returns -1023 for log2(0), although log2(0) is undefined.
Which ended up as:
rq_param->dlg.rq_c.swath_height = 1 << -1023
[How]
Fix applied on all dml versions.
1. Ensure dml_log2 is only called if the argument is greater than 0.
2. Subtract req128_l/req128_c from log2_swath_height_l/log2_swath_height_c
only when it is greater than 0.
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Anson Jacob <Anson.Jacob@amd.com> Reviewed-by: Dmytro Laktyushkin <Dmytro.Laktyushkin@amd.com> Reviewed-by: Jun Lei <Jun.Lei@amd.com> Acked-by: Solomon Chiu <solomon.chiu@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
1. Catch invalid link_rate and link_count settings
2. Call dc interface to overwrite preferred link settings, and wait
until next stream update to apply the new settings.
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com> Reviewed-by: Nicholas Kazlauskas <Nicholas.Kazlauskas@amd.com> Acked-by: Solomon Chiu <solomon.chiu@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
If userptr pages have been pinned but not bounded,
they remain uncleared.
Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Daniel Gomez <daniel@qtec.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
If userptr pages have been pinned but not bounded,
they remain uncleared.
Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Daniel Gomez <daniel@qtec.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Neither vtotal nor drm_mode_vrefresh contain a value that is
premultiplied by 100 making the x100 variable name incorrect and
resulting in vclks_line to become 100 times larger than it is supposed
to be. The hardware counts 100 clockticks too many before tearcheck,
leading to severe panel issues on at least the Sony Xperia lineup.
This is likely an artifact from the original MDSS DSI panel driver where
the calculation [1] corrected for a premultiplied reference framerate by
100 [2]. It does not appear that the above values were ever
premultiplied in the history of the DRM MDP5 driver.
With this change applied the value written to the SYNC_CONFIG_VSYNC
register is now identical to downstream kernels.
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@somainline.org> Signed-off-by: Marijn Suijten <marijn.suijten@somainline.org> Link: https://lore.kernel.org/r/20210406214726.131534-3-marijn.suijten@somainline.org Signed-off-by: Rob Clark <robdclark@chromium.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Leaving this at a close-to-maximum register value 0xFFF0 means it takes
very long for the MDSS to generate a software vsync interrupt when the
hardware TE interrupt doesn't arrive. Configuring this to double the
vtotal (like some downstream kernels) leads to a frame to take at most
twice before the vsync signal, until hardware TE comes up.
In this case the hardware interrupt responsible for providing this
signal - "disp-te" gpio - is not hooked up to the mdp5 vsync/pp logic at
all. This solves severe panel update issues observed on at least the
Xperia Loire and Tone series, until said gpio is properly hooked up to
an irq.
Suggested-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@somainline.org> Signed-off-by: Marijn Suijten <marijn.suijten@somainline.org> Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@somainline.org> Link: https://lore.kernel.org/r/20210406214726.131534-2-marijn.suijten@somainline.org Signed-off-by: Rob Clark <robdclark@chromium.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
During load balance, LBF_SOME_PINNED will be set if any candidate task
cannot be detached due to CPU affinity constraints. This can result in
setting env->sd->parent->sgc->group_imbalance, which can lead to a group
being classified as group_imbalanced (rather than any of the other, lower
group_type) when balancing at a higher level.
In workloads involving a single task per CPU, LBF_SOME_PINNED can often be
set due to per-CPU kthreads being the only other runnable tasks on any
given rq. This results in changing the group classification during
load-balance at higher levels when in reality there is nothing that can be
done for this affinity constraint: per-CPU kthreads, as the name implies,
don't get to move around (modulo hotplug shenanigans).
It's not as clear for userspace tasks - a task could be in an N-CPU cpuset
with N-1 offline CPUs, making it an "accidental" per-CPU task rather than
an intended one. KTHREAD_IS_PER_CPU gives us an indisputable signal which
we can leverage here to not set LBF_SOME_PINNED.
Note that the aforementioned classification to group_imbalance (when
nothing can be done) is especially problematic on big.LITTLE systems, which
have a topology the likes of:
Here, setting LBF_SOME_PINNED due to a per-CPU kthread when balancing at MC
level on CPUs [0-1] will subsequently prevent CPUs [2-3] from classifying
the [0-1] group as group_misfit_task when balancing at DIE level. Thus, if
CPUs [0-1] are running CPU-bound (misfit) tasks, ill-timed per-CPU kthreads
can significantly delay the upgmigration of said misfit tasks. Systems
relying on ASYM_PACKING are likely to face similar issues.
Signed-off-by: Lingutla Chandrasekhar <clingutla@codeaurora.org>
[Use kthread_is_per_cpu() rather than p->nr_cpus_allowed]
[Reword changelog] Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Link: https://lkml.kernel.org/r/20210407220628.3798191-2-valentin.schneider@arm.com Signed-off-by: Sasha Levin <sashal@kernel.org>
Mel Gorman did some nice work in 9fe1f127b913 ("sched/fair: Merge
select_idle_core/cpu()"), resulting in the kernel being more efficient
at finding an idle CPU, and in tasks spending less time waiting to be
run, both according to the schedstats run_delay numbers, and according
to measured application latencies. Yay.
The flip side of this is that we see more task migrations (about 30%
more), higher cache misses, higher memory bandwidth utilization, and
higher CPU use, for the same number of requests/second.
This is most pronounced on a memcache type workload, which saw a
consistent 1-3% increase in total CPU use on the system, due to those
increased task migrations leading to higher L2 cache miss numbers, and
higher memory utilization. The exclusive L3 cache on Skylake does us
no favors there.
On our web serving workload, that effect is usually negligible.
It appears that the increased number of CPU migrations is generally a
good thing, since it leads to lower cpu_delay numbers, reflecting the
fact that tasks get to run faster. However, the reduced locality and
the corresponding increase in L2 cache misses hurts a little.
The patch below appears to fix the regression, while keeping the
benefit of the lower cpu_delay numbers, by reintroducing
select_idle_smt with a twist: when a socket has no idle cores, check
to see if the sibling of "prev" is idle, before searching all the
other CPUs.
This fixes both the occasional 9% regression on the web serving
workload, and the continuous 2% CPU use regression on the memcache
type workload.
With Mel's patches and this patch together, task migrations are still
high, but L2 cache misses, memory bandwidth, and CPU time used are
back down to what they were before. The p95 and p99 response times for
the memcache type application improve by about 10% over what they were
before Mel's patches got merged.
Signed-off-by: Rik van Riel <riel@surriel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vincent Guittot <vincent.guittot@linaro.org> Link: https://lkml.kernel.org/r/20210326151932.2c187840@imladris.surriel.com Signed-off-by: Sasha Levin <sashal@kernel.org>
For two of the supported sensors the stv06xx driver allocates memory which
is stored in sd->sensor_priv. This memory is freed on a disconnect, but if
the probe() fails, then it isn't freed and so this leaks memory.
Add a new probe_error() op that drivers can use to free any allocated
memory in case there was a probe failure.
Thanks to Pavel Skripkin <paskripkin@gmail.com> for discovering the cause
of the memory leak.
In case of error in dvb_usb_adapter_dvb_init() or
dvb_usb_adapter_dvb_init() d->num_adapters_initialized won't be
incremented, but dvb_usb_adapter_exit() relies on it:
for (n = 0; n < d->num_adapters_initialized; n++)
So, allocated objects won't be freed.
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com> Reported-by: syzbot+3c2be7424cea3b932b0e@syzkaller.appspotmail.com Signed-off-by: Sean Young <sean@mess.org> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
pm_runtime_get_sync() will increase the runtime PM counter
even it returns an error. Thus a pairing decrement is needed
to prevent refcount leak. Fix this by replacing this API with
pm_runtime_resume_and_get(), which will not change the runtime
PM counter on error.
Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
pm_runtime_get_sync() will increase the runtime PM counter
even it returns an error. Thus a pairing decrement is needed
to prevent refcount leak. Fix this by replacing this API with
pm_runtime_resume_and_get(), which will not change the runtime
PM counter on error.
Signed-off-by: Dinghao Liu <dinghao.liu@zju.edu.cn> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com> Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
This driver's remove path calls cancel_delayed_work(). However, that
function does not wait until the work function finishes. This means
that the callback function may still be running after the driver's
remove function has finished, which would result in a use-after-free.
Fix by calling cancel_delayed_work_sync(), which ensures that
the work is properly cancelled, no longer running, and unable
to re-schedule itself.
Reported-by: Hulk Robot <hulkci@huawei.com> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com> Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
We were not programing the correct bit while clearing the perfcounter oob.
So, clear it correctly using the new 'clear' bit. This fixes the below
error:
[drm:a6xx_gmu_set_oob] *ERROR* Timeout waiting for GMU OOB set PERFCOUNTER: 0x80000000
Some devices reference an output terminal as the source of extension
units. This is incorrect, as output terminals only have an input pin,
and thus can't be connected to any entity in the forward direction. The
resulting topology would cause issues when registering the media
controller graph. To avoid this problem, connect the extension unit to
the source of the output terminal instead.
While at it, and while no device has been reported to be affected by
this issue, also handle forward scans where two output terminals would
be connected together, and skip the terminals found through such an
invalid connection.
Reported-and-tested-by: John Nealy <jnealy3@yahoo.com> Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Signed-off-by: Hans de Goede <hdegoede@redhat.com> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The EDID had a few mistakes as reported by edid-decode:
Block 1, CTA-861 Extension Block:
Video Data Block: For improved preferred timing interoperability, set 'Native detailed modes' to 1.
Video Capability Data Block: S_PT is equal to S_IT and S_CE, so should be set to 0 instead.
If some error occurs, URB buffers should also be freed. If they aren't
freed with the dvb here, the em28xx_dvb_fini call doesn't frees the URB
buffers as dvb is set to NULL. The function in which error occurs should
do all the cleanup for the allocations it had done.
Tested the patch with the reproducer provided by syzbot. This patch
fixes the memleak.
Reported-by: syzbot+889397c820fa56adf25d@syzkaller.appspotmail.com Signed-off-by: Muhammad Usama Anjum <musamaanjum@gmail.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Some arrays return ILLEGAL_REQUEST with ASC 00h if they don't support the
RTPG extended header so remove the check for INVALID FIELD IN CDB.
Link: https://lore.kernel.org/r/20210331201154.20348-1-emilne@redhat.com Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ewan D. Milne <emilne@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Link: https://lore.kernel.org/r/161549386882.25025.2594251735886014958.stgit@brunhilda Reviewed-by: Scott Benesh <scott.benesh@microchip.com> Reviewed-by: Scott Teel <scott.teel@microchip.com> Acked-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Kevin Barnett <kevin.barnett@microchip.com> Signed-off-by: Don Brace <don.brace@microchip.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
While failing queued I/Os in TMF path, there was a request leak and hence
stale entries in request pool with ref count being non-zero. In shutdown
path we have a BUG_ON to catch stuck I/O either in firmware or in the
driver. The stale requests caused a system crash. The I/O request pool
leakage also lead to a significant performance drop.
Link: https://lore.kernel.org/r/161549370379.25025.12793264112620796062.stgit@brunhilda Reviewed-by: Scott Teel <scott.teel@microchip.com> Reviewed-by: Scott Benesh <scott.benesh@microchip.com> Reviewed-by: Kevin Barnett <kevin.barnett@microchip.com> Signed-off-by: Murthy Bhat <Murthy.Bhat@microchip.com> Signed-off-by: Don Brace <don.brace@microchip.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Correct SCSI midlayer sending more requests than exposed host queue depth
causing firmware ASSERT and lockup issues by enabling host-wide tags.
Note: This also results in better performance.
Link: https://lore.kernel.org/r/161549369787.25025.8975999483518581619.stgit@brunhilda Suggested-by: Ming Lei <ming.lei@redhat.com> Suggested-by: John Garry <john.garry@huawei.com> Reviewed-by: Scott Benesh <scott.benesh@microchip.com> Reviewed-by: Scott Teel <scott.teel@microchip.com> Reviewed-by: Mike McGowen <mike.mcgowen@microchip.com> Reviewed-by: Kevin Barnett <kevin.barnett@microchip.com> Signed-off-by: Don Brace <don.brace@microchip.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Adds 80000 us sleep when the usb cable is plugged in to hopefully avoid
bouncing contacts.
Upon pluging in the usb cable vbus will bounce for some time, causing cpcap to
dissconnect charging due to detecting an undervoltage condition. This is a
scope of vbus on xt894 while quickly inserting the usb cable with firm force,
probed at the far side of the usb socket and vbus loaded with approx 1k:
http://uvos.xyz/maserati/usbplug.jpg.
As can clearly be seen, vbus is all over the place for the first 15 ms or so
with a small blip at ~40 ms this causes the cpcap to trip up and disable
charging again.
The delay helps cpcap_usb_detect avoid the worst of this. It is, however, still
not ideal as strong vibrations can cause the issue to reapear any time during
charging. I have however not been able to cause the device to stop charging due
to this in practice as it is hard to vibrate the device such that the vbus pins
start bouncing again but cpcap_usb_detect is not called again due to a detected
disconnect/reconnect event.
Signed-off-by: Carl Philipp Klemm <philipp@uvos.xyz> Tested-by: Tony Lindgren <tony@atomide.com> Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Carl Philipp Klemm <philipp@uvos.xyz> Acked-by: Tony Lindgren <tony@atomide.com> Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
These warnings are reported because
1. measure_vals() declares 'bw_imc' and 'bw_resc_end' as unsigned long
variables
2. Return value of get_mem_bw_imc() and get_mem_bw_resctrl() are assigned
to 'bw_imc' and 'bw_resc_end' respectively
3. The returned values are checked for <= 0 to see if the calls failed
Checking for < 0 for an unsigned value doesn't make any sense.
Fix this issue by changing the implementation of get_mem_bw_imc() and
get_mem_bw_resctrl() such that they now accept reference to a variable
and set the variable appropriately upon success and return 0, else return
< 0 on error.
iMC (Integrated Memory Controller) counters are usually at
"/sys/bus/event_source/devices/" and are named as "uncore_imc_<n>".
num_of_imcs() function tries to count number of such iMC counters so that
it could appropriately initialize required number of perf_attr structures
that could be used to read these iMC counters.
num_of_imcs() function assumes that all the directories under this path
that start with "uncore_imc" are iMC counters. But, on some systems there
could be directories named as "uncore_imc_free_running" which aren't iMC
counters. Trying to read from such directories will result in "not found
file" errors and MBM/MBA tests will fail.
Hence, fix the logic in num_of_imcs() such that it looks at the first
character after "uncore_imc_" to check if it's a numerical digit or not. If
it's a digit then the directory represents an iMC counter, else, skip the
directory.
Resctrl test suite before running any unit test (like cmt, cat, mbm and
mba) should first check if the feature is enabled (by kernel and not just
supported by H/W) on the platform or not.
validate_resctrl_feature_request() is supposed to do that. This function
intends to grep for relevant flags in /proc/cpuinfo but there are several
issues here
1. validate_resctrl_feature_request() calls fgrep() to get flags from
/proc/cpuinfo. But, fgrep() can only return a string with maximum of 255
characters and hence the complete cpu flags are never returned.
2. The substring search logic is also busted. If strstr() finds requested
resctrl feature in the cpu flags, it returns pointer to the first
occurrence. But, the logic negates the return value of strstr() and
hence validate_resctrl_feature_request() returns false if the feature is
present in the cpu flags and returns true if the feature is not present.
3. validate_resctrl_feature_request() checks if a resctrl feature is
reported in /proc/cpuinfo flags or not. Having a cpu flag means that the
H/W supports the feature, but it doesn't mean that the kernel enabled
it. A user could selectively enable only a subset of resctrl features
using kernel command line arguments. Hence, /proc/cpuinfo isn't a
reliable source to check if a feature is enabled or not.
The 3rd issue being the major one and fixing it requires changing the way
validate_resctrl_feature_request() works. Since, /proc/cpuinfo isn't the
right place to check if a resctrl feature is enabled or not, a more
appropriate place is /sys/fs/resctrl/info directory. Change
validate_resctrl_feature_request() such that,
1. For cat, check if /sys/fs/resctrl/info/L3 directory is present or not
2. For mba, check if /sys/fs/resctrl/info/MB directory is present or not
3. For cmt, check if /sys/fs/resctrl/info/L3_MON directory is present and
check if /sys/fs/resctrl/info/L3_MON/mon_features has llc_occupancy
4. For mbm, check if /sys/fs/resctrl/info/L3_MON directory is present and
check if /sys/fs/resctrl/info/L3_MON/mon_features has
mbm_<total/local>_bytes
Please note that only L3_CAT, L3_CMT, MBA and MBM are supported. CDP and L2
variants can be added later.
resctrl test suite accepts command line arguments (like -b, -t, -n and -p)
as documented in the help. But passing -n and -p throws an invalid option
error. This happens because -n and -p are missing in the list of
characters that getopt() recognizes as valid arguments. Hence, they are
treated as invalid options.
Fix this by adding them to the list of characters that getopt() recognizes
as valid arguments. Please note that the main() function already has the
logic to deal with the values passed as part of these arguments and hence
no changes are needed there.
Checking resctrl features call strcmp() to compare feature strings
(e.g. "mba", "cat" etc). The checkings are error prone and don't have
good coding style. Define the constant strings in macros and call
strncmp() to solve the potential issues.
Reinette reported following compilation issue on Fedora 32, gcc version
10.1.1
/usr/bin/ld: resctrl_tests.o:<src_dir>/resctrl.h:65: multiple definition
of `bm_pid'; cache.o:<src_dir>/resctrl.h:65: first defined here
Other variables are ppid, tests_run, llc_occup_path, is_amd. Compiler
isn't happy because these variables are defined globally in two .c files
but are not declared as extern.
To fix issues for the global variables, declare them as extern.
Chang Log:
- Split this patch from v4's patch 1 (Shuah).
Reinette reported following compilation issue on Fedora 32, gcc version
10.1.1
/usr/bin/ld: cqm_test.o:<src_dir>/cqm_test.c:22: multiple definition of
`cache_size'; cat_test.o:<src_dir>/cat_test.c:23: first defined here
The same issue is reported for long_mask, cbm_mask, count_of_bits etc
variables as well. Compiler isn't happy because these variables are
defined globally in two .c files namely cqm_test.c and cat_test.c and
the compiler during compilation finds that the variable is already
defined (multiple definition error).
Taking a closer look at the usage of these variables reveals that these
variables are used only locally in functions such as cqm_resctrl_val()
(defined in cqm_test.c) and cat_perf_miss_val() (defined in cat_test.c).
These variables are not shared between those functions. So, there is no
need for these variables to be global. Hence, fix this issue by making
them static variables.
David reported a buffer overflow error in the check_results() function of
the cmt unit test and he suggested enabling _FORTIFY_SOURCE gcc compiler
option to automatically detect any such errors.
Feature Test Macros man page describes_FORTIFY_SOURCE as below
"Defining this macro causes some lightweight checks to be performed to
detect some buffer overflow errors when employing various string and memory
manipulation functions (for example, memcpy, memset, stpcpy, strcpy,
strncpy, strcat, strncat, sprintf, snprintf, vsprintf, vsnprintf, gets, and
wide character variants thereof). For some functions, argument consistency
is checked; for example, a check is made that open has been supplied with a
mode argument when the specified flags include O_CREAT. Not all problems
are detected, just some common cases.
If _FORTIFY_SOURCE is set to 1, with compiler optimization level 1 (gcc
-O1) and above, checks that shouldn't change the behavior of conforming
programs are performed.
With _FORTIFY_SOURCE set to 2, some more checking is added, but some
conforming programs might fail.
Some of the checks can be performed at compile time (via macros logic
implemented in header files), and result in compiler warnings; other checks
take place at run time, and result in a run-time error if the check fails.
Use of this macro requires compiler support, available with gcc since
version 4.0."
Fix the buffer overflow error in the check_results() function of the cmt
unit test and enable _FORTIFY_SOURCE gcc check to catch any future buffer
overflow errors.
Another issue found by KASAN. The bit finding is buried inside the
dp_for_each_set_bit() macro (that passes on to for_each_set_bit() that
calls the bit stuff. These bit functions want an unsigned long pointer
as input and just dumbly casting leads to out-of-bounds accesses.
This fixes that.
On Hisilicon Kunpeng920, ESP is set to 1 by default for all ports of
SATA controller. In some scenarios, some ports are not external SATA ports,
and it cause disks connected to these ports to be identified as removable
disks. So disable the SXS capability on the software side to prevent users
from mistakenly considering non-removable disks as removable disks and
performing related operations.
Signed-off-by: Xingui Yang <yangxingui@huawei.com> Signed-off-by: Luo Jiaxing <luojiaxing@huawei.com> Reviewed-by: John Garry <john.garry@huawei.com> Link: https://lore.kernel.org/r/1615544676-61926-1-git-send-email-luojiaxing@huawei.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Sasha Levin <sashal@kernel.org>
Remove the CQHCI_QUIRK_SHORT_TXFR_DESC_SZ quirk because the
latest chips have this fixed and earlier chips have other
CQE problems that prevent the feature from being enabled.
On bsg command completion, bsg_job_done() was called while qla driver
continued to access the bsg_job buffer. bsg_job_done() would free up
resources that ended up being reused by other task while the driver
continued to access the buffers. As a result, driver was reading garbage
data.
When the io_latency heuristic is off, bfq_queues must not start to be
weight-raised. Unfortunately, by mistake, this may happen when the
state of a previously weight-raised bfq_queue is resumed after a queue
split. This commit fixes this error.
Tested-by: Jan Kara <jack@suse.cz> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Link: https://lore.kernel.org/r/20210304174627.161-5-paolo.valente@linaro.org Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Sasha Levin <sashal@kernel.org>
vkms_vblank_simulate() uses WARN_ON for timing-dependent condition
(timer overrun). This is a mis-use of WARN_ON, WARN_ON must be used
to denote kernel bugs. Use pr_warn() instead.
CID 361199 (#1 of 1): Unchecked return value (CHECKED_RETURN)
3. check_return: Calling qla24xx_get_isp_stats without checking return
value (as is done elsewhere 4 out of 5 times).
Link: https://lore.kernel.org/r/20210320232359.941-7-bvanassche@acm.org Cc: Quinn Tran <qutran@marvell.com> Cc: Mike Christie <michael.christie@oracle.com> Cc: Himanshu Madhani <himanshu.madhani@oracle.com> Cc: Daniel Wagner <dwagner@suse.de> Cc: Lee Duncan <lduncan@suse.com> Reviewed-by: Daniel Wagner <dwagner@suse.de> Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com> Signed-off-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why]
dc_cursor_position do not initialise position.translate_by_source when
crtc or plane->state->fb is NULL. UBSAN caught this error in
dce110_set_cursor_position, as the value was garbage.
[How]
Initialise dc_cursor_position structure elements to 0 in handle_cursor_update
before calling get_cursor_position.
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1471 Reported-by: Lyude Paul <lyude@redhat.com> Signed-off-by: Anson Jacob <Anson.Jacob@amd.com> Reviewed-by: Aurabindo Jayamohanan Pillai <Aurabindo.Pillai@amd.com> Acked-by: Solomon Chiu <solomon.chiu@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
This recent change introduce SDMA interrupt info printing with irq->process function.
These functions do not require a set function to enable/disable the irq
[Why]
The registers for the address of the cursor are aligned to 2KB, so all
cursor surfaces also need to be aligned to 2KB. Currently, the
provided cursor cache surface is not aligned, so we need a workaround
until alignment is enforced by the surface provider.
[How]
- round up surface address to nearest multiple of 2048
- current policy is to provide a much bigger cache size than
necessary,so this operation is safe
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Joshua Aberback <joshua.aberback@amd.com> Reviewed-by: Jun Lei <Jun.Lei@amd.com> Acked-by: Eryk Brol <eryk.brol@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
If get_num_sdma_queues or get_num_xgmi_sdma_queues is 0, we end up
doing a shift operation where the number of bits shifted equals
number of bits in the operand. This behaviour is undefined.
Set num_sdma_queues or num_xgmi_sdma_queues to ULLONG_MAX, if the
count is >= number of bits in the operand.
Reported-by: Lyude Paul <lyude@redhat.com> Signed-off-by: Anson Jacob <Anson.Jacob@amd.com> Reviewed-by: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Tested-by: Lyude Paul <lyude@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The psp supplies the link type in the upper 2 bits of the psp xgmi node
information num_hops field. With a new link type, Aldebaran has these
bits set to a non-zero value (1 = xGMI3) so the KFD topology will report
the incorrect IO link weights without proper masking.
The actual number of hops is located in the 3 least significant bits of
this field so mask if off accordingly before passing it to the KFD.
Signed-off-by: Jonathan Kim <jonathan.kim@amd.com> Reviewed-by: Amber Lin <amber.lin@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
By default this timestamp is 32 bit counter. It gets
overflowed in around 10 minutes.
Signed-off-by: Alex Sierra <alex.sierra@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
If xnack is on, VM retry fault interrupt send to IH ring1, and ring1
will be full quickly. IH cannot receive other interrupts, this causes
deadlock if migrating buffer using sdma and waiting for sdma done while
handling retry fault.
Remove VMC from IH storm client, enable ring1 write pointer overflow,
then IH will drop retry fault interrupts and be able to receive other
interrupts while driver is handling retry fault.
IH ring1 write pointer doesn't writeback to memory by IH, and ring1
write pointer recorded by self-irq is not updated, so always read
the latest ring1 write pointer from register.
Signed-off-by: Philip Yang <Philip.Yang@amd.com> Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
WLED3_SINK_REG_SYNC is, as the name implies, a sink register offset.
Therefore, use the sink address as base instead of the ctrl address.
This fixes the sync toggle on wled4, which can be observed by the fact
that adjusting brightness now works.
It has no effect on wled3 because sink and ctrl base addresses are the
same. This allows adjusting the brightness without having to disable
then reenable the module.
Signed-off-by: Obeida Shamoun <oshmoun100@googlemail.com> Signed-off-by: Konrad Dybcio <konrad.dybcio@somainline.org> Signed-off-by: Marijn Suijten <marijn.suijten@somainline.org> Reviewed-by: Daniel Thompson <daniel.thompson@linaro.org> Acked-by: Kiran Gunda <kgunda@codeaurora.org> Signed-off-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The V4L2_CID_STATELESS_FWHT_PARAMS compound control was missing a
proper initialization of the flags field, so after loading the vicodec
module for the first time, running v4l2-compliance for the stateless
decoder would fail on this control because the initial control value
was considered invalid by the vicodec driver.
Initializing the flags field to sane values fixes this.
Add a fix for the memory leak bugs that can occur when the
saa7164_encoder_register() function fails.
The function allocates memory without explicitly freeing
it when errors occur.
Add a better error handling that deallocate the unused buffers before the
function exits during a fail.
Signed-off-by: Daniel Niv <danielniv3@gmail.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
We are spending way too much effort on qdio-internal bookkeeping for
QAOB management & caching, and it's still not robust. Once qdio's
TX path has detached the QAOB from a PENDING buffer, we lost all
track of it until it shows up in a CQ notification again. So if the
device is torn down before that notification arrives, we leak the QAOB.
Just have the driver take care of it, and simply pass down a QAOB if
they want a TX with async-completion capability. For a buffer in PENDING
state that requires the QAOB for final completion, qeth can now also try
to recycle the buffer's QAOB rather than unconditionally freeing it.
This also eliminates the qdio_outbuf_state array, which was only needed
to transfer the aob->user1 tag from the driver to the qdio layer.
Commit aaaa93eda64b ("media] media: venus: venc: add video encoder files")
is the last in a series of three commits to add core.c vdec.c and venc.c
adding core, encoder and decoder.
The encoder and decoder check for core drvdata as set and return -EPROBE_DEFER
if it has not been set, however both the encoder and decoder rely on
core.v4l2_dev as valid.
core.v4l2_dev will not be valid until v4l2_device_register() has completed
in core.c's probe().
Normally this is never seen however, Dmitry reported the following
backtrace when compiling drivers and firmware directly into a kernel image.
- Only setting drvdata after v4l2_device_register() completes
- Moving v4l2_device_register() so that suspend/reume in core::probe()
stays as-is
- Changes pm_ops->core_function() to take struct venus_core not struct
device
- Minimal rework of v4l2_device_*register in probe/remove
We must free/disable all interrupts and cancel all pending works
before doing further cleanup.
Before this commit arizona_extcon_remove() was doing several
register writes to shut things down before disabling the IRQs
and it was cancelling only 1 of the 3 different works used.
Move all the register-writes shutting things down to after
the disabling of the IRQs and add the 2 missing
cancel_delayed_work_sync() calls.
This fixes various possible races on driver unbind. One of which
would always trigger on devices using the mic-clamp feature for
jack detection. The ARIZONA_MICD_CLAMP_MODE_MASK update was
done before disabling the IRQs, causing:
1. arizona_jackdet() to run
2. detect a jack being inserted (clamp disabled means jack inserted)
3. call arizona_start_mic() which:
3.1 Enables the MICVDD regulator
3.2 takes a pm_runtime_reference
And this was all happening after the ARIZONA_MICD_ENA bit clearing,
which would undo 3.1 and 3.2 because the ARIZONA_MICD_CLAMP_MODE_MASK
update was being done after the ARIZONA_MICD_ENA bit clearing.
So this means that arizona_extcon_remove() would exit with
1. MICVDD enabled and 2. The pm_runtime_reference being unbalanced.
MICVDD still being enabled caused the following oops when the
regulator is released by the devm framework:
Note this oops is just one of various theoretically possible races caused
by the wrong ordering inside arizona_extcon_remove(), this fixes the
ordering fixing all possible races, including the reported oops.
Signed-off-by: Hans de Goede <hdegoede@redhat.com> Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com> Acked-by: Charles Keepax <ckeepax@opensource.cirrus.com> Tested-by: Charles Keepax <ckeepax@opensource.cirrus.com> Acked-by: Chanwoo Choi <cw00.choi@samsung.com> Signed-off-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
When the jack is partially inserted and then removed again it may be
removed while the hpdet code is running. In this case the following
may happen:
1. The "JACKDET rise" or ""JACKDET fall" IRQ triggers
2. arizona_jackdet runs and takes info->lock
3. The "HPDET" IRQ triggers
4. arizona_hpdet_irq runs, blocks on info->lock
5. arizona_jackdet calls arizona_stop_mic() and clears info->hpdet_done
6. arizona_jackdet releases info->lock
7. arizona_hpdet_irq now can continue running and:
7.1 Calls arizona_start_mic() (if a mic was detected)
7.2 sets info->hpdet_done
Step 7 is undesirable / a bug:
7.1 causes the device to stay in a high power-state (with MICVDD enabled)
7.2 causes hpdet to not run on the next jack insertion, which in turn
causes the EXTCON_JACK_HEADPHONE state to never get set
This fixes both issues by skipping these 2 steps when arizona_hpdet_irq
runs after the jack has been unplugged.
Signed-off-by: Hans de Goede <hdegoede@redhat.com> Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com> Acked-by: Charles Keepax <ckeepax@opensource.cirrus.com> Tested-by: Charles Keepax <ckeepax@opensource.cirrus.com> Acked-by: Chanwoo Choi <cw00.choi@samsung.com> Signed-off-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
On all newer bq27xxx ICs, the AveragePower register contains a signed
value; in addition to handling the raw value as unsigned, the driver
code also didn't convert it to µW as expected.
At least for the BQ28Z610, the reference manual incorrectly states that
the value is in units of 1mW and not 10mW. I have no way of knowing
whether the manuals of other supported ICs contain the same error, or if
there are models that actually use 1mW. At least, the new code shouldn't
be *less* correct than the old version for any device.
power_avg is removed from the cache structure, se we don't have to
extend it to store both a signed value and an error code. Always getting
an up-to-date value may be desirable anyways, as it avoids inconsistent
current and power readings when switching between charging and
discharging.
In practice, IA_CSS_PIPE_ID_NUM should never be used when
calling atomisp_q_video_buffers_to_css(), as the driver should
discover the right pipe before calling it.
Yet, if some pipe parsing issue happens, it could end using
it.
The correct return code to report an invalid pipeline configuration is
-EPIPE. Return it instead of -EINVAL from __capture_legacy_try_fmt()
when the capture format doesn't match the media bus format of the
connected subdev.
The folowing AMD IOMMU are affected by the RiSC engine stall, requiring a
reset to maintain continual operation. After being added to the
broken_dev_id list the systems are functional long term.
0x1481 is the PCI ID for the IOMMU found on Starship/Matisse
0x1419 is the PCI ID for the IOMMU found on 15h (Models 10h-1fh) family
0x5a23 is the PCI ID for the IOMMU found on RD890S/RD990
Signed-off-by: Brad Love <brad@nextdimension.cc> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
syzbot reported memory leak in zr364xx_probe()[1].
The problem was in invalid error handling order.
All error conditions rigth after v4l2_ctrl_handler_init()
must call v4l2_ctrl_handler_free().
Reported-by: syzbot+efe9aefc31ae1e6f7675@syzkaller.appspotmail.com Signed-off-by: Pavel Skripkin <paskripkin@gmail.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This fixes a compilation warning in pscsi_complete_cmd():
drivers/target/target_core_pscsi.c: In function ‘pscsi_complete_cmd’:
drivers/target/target_core_pscsi.c:624:5: warning: suggest braces around empty body in an ‘if’ statement [-Wempty-body]
; /* XXX: TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE */
Link: https://lore.kernel.org/r/20210228055645.22253-5-chaitanya.kulkarni@wdc.com Reviewed-by: Mike Christie <michael.christie@oracle.com> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Running an rcuscale stress-suite can lead to "Out of memory" of a
system. This can happen under high memory pressure with a small amount
of physical memory.
For example, a KVM test configuration with 64 CPUs and 512 megabytes
can result in OOM when running rcuscale with below parameters:
Because kvfree_rcu() has a fallback path, memory allocation failure is
not the end of the world. Furthermore, the added overhead of aggressive
GFP settings must be balanced against the overhead of the fallback path,
which is a cache miss for double-argument kvfree_rcu() and a call to
synchronize_rcu() for single-argument kvfree_rcu(). The current choice
of GFP_KERNEL|__GFP_NOWARN can result in longer latencies than a call
to synchronize_rcu(), so less-tenacious GFP flags would be helpful.
Here is the tradeoff that must be balanced:
a) Minimize use of the fallback path,
b) Avoid pushing the system into OOM,
c) Bound allocation latency to that of synchronize_rcu(), and
d) Leave the emergency reserves to use cases lacking fallbacks.
This commit therefore changes GFP flags from GFP_KERNEL|__GFP_NOWARN to
GFP_KERNEL|__GFP_NORETRY|__GFP_NOMEMALLOC|__GFP_NOWARN. This combination
leaves the emergency reserves alone and can initiate reclaim, but will
not invoke the OOM killer.
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com> Signed-off-by: Paul E. McKenney <paulmck@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
As long as NUMA diameter > 2, building sched_domain by sibling's child
domain will definitely create a sched_domain with sched_group which will
span out of the sched_domain:
when node2 is added into the domain2 of node0, kernel is using the child
domain of node2's domain2, which is domain1(node2+3). Node 3 is outside
the span of the domain including node0+1+2.
This will make load_balance() run based on screwed avg_load and group_type
in the sched_group spanning out of the sched_domain, and it also makes
select_task_rq_fair() pick an idle CPU outside the sched_domain.
Real servers which suffer from this problem include Kunpeng920 and 8-node
Sun Fire X4600-M2, at least.
Here we move to use the *child* domain of the *child* domain of node2's
domain2 as the new added sched_group. At the same, we re-use the lower
level sgc directly.
+------+ +------+ +-------+ +------+
| node | 12 |node | 20 | node | 12 |node |
| 0 +---------+1 +--------+ 2 +-------+3 |
+------+ +------+ +-------+ +------+
While the lower level sgc is re-used, this patch only changes the remote
sched_groups for those sched_domains playing grandchild trick, therefore,
sgc->next_update is still safe since it's only touched by CPUs that have
the group span as local group. And sgc->imbalance is also safe because
sd_parent remains the same in load_balance and LB only tries other CPUs
from the local group.
Moreover, since local groups are not touched, they are still getting
roughly equal size in a TL. And should_we_balance() only matters with
local groups, so the pull probability of those groups are still roughly
equal.
Being called for each dequeue, util_est reduces the number of its updates
by filtering out when the EWMA signal is different from the task util_avg
by less than 1%. It is a problem for a sudden util_avg ramp-up. Due to the
decay from a previous high util_avg, EWMA might now be close enough to
the new util_avg. No update would then happen while it would leave
ue.enqueued with an out-of-date value.
Taking into consideration the two util_est members, EWMA and enqueued for
the filtering, ensures, for both, an up-to-date value.
This is for now an issue only for the trace probe that might return the
stale value. Functional-wise, it isn't a problem, as the value is always
accessed through max(enqueued, ewma).
This problem has been observed using LISA's UtilConvergence:test_means on
the sd845c board.
No regression observed with Hackbench on sd845c and Perf-bench sched pipe
on hikey/hikey960.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Link: https://lkml.kernel.org/r/20210225165820.1377125-1-vincent.donnefort@arm.com Signed-off-by: Sasha Levin <sashal@kernel.org>
The task contribution to the energy delta can then be either:
(1) _task_util_est, on a mostly idle CPU, where cpu_util is close to 0
and _task_util_est > cpu_util.
(2) task_util, on a mostly busy CPU, where cpu_util > _task_util_est.
(cpu_util_est doesn't appear here. It is 0 when a CPU is idle and
otherwise must be small enough so that feec() takes the CPU as a
potential target for the task placement)
This is problematic for feec(), as cpu_util_next() might give an unfair
advantage to a CPU which is mostly busy (2) compared to one which is
mostly idle (1). _task_util_est being always bigger than task_util in
feec() (as the task is waking up), the task contribution to the energy
might look smaller on certain CPUs (2) and this breaks the energy
comparison.
This issue is, moreover, not sporadic. By starving idle CPUs, it keeps
their cpu_util < _task_util_est (1) while others will maintain cpu_util >
_task_util_est (2).
Fix this problem by always using max(task_util, _task_util_est) as a task
contribution to the energy (ENERGY_UTIL). The new estimated CPU
utilization for the energy would then be:
compute_energy() still needs to know which OPP would be selected if the
task would be migrated in the perf_domain (FREQUENCY_UTIL). Hence,
cpu_util_next() is still used to estimate the maximum util within the pd.
When unloading driver after killing some applications, it will hit sdma
flush tlb job timeout which is called by ttm_bo_delay_delete. So
to avoid the job submit after fence driver fini, call ttm_bo_lock_delayed_workqueue
before fence driver fini. And also put drm_sched_fini before waiting fence.
Signed-off-by: Emily Deng <Emily.Deng@amd.com> Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
While testing target port swap test with ADISC enabled, several nodes
remain in UNUSED state. These nodes are never freed and rmmod hangs for
long time before finising with "0233 Nodelist not empty" error.
During PLOGI completion lpfc_plogi_confirm_nport() looks for existing nodes
with same WWPN. If found, the existing node is used to continue discovery.
The node on which plogi was performed is freed. When ADISC is enabled, an
ADISC els request is triggered in response to an RSCN. It's possible that
the ADISC may be rejected by the remote port causing the ADISC completion
handler to clear the port and node name in the node. If this occurs, if a
PLOGI is received it causes a node lookup based on wwpn to now fail,
causing the port swap logic to kick in which allocates a new node and swaps
to it. This effectively orphans the original node structure.
Fix the situation by detecting when the lookup fails and forgo the node
swap and node allocation by using the node on which the PLOGI was issued.
Link: https://lore.kernel.org/r/20210301171821.3427-15-jsmart2021@gmail.com Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The driver is seeing a scenario where PLOGI response was issued and traffic
is arriving while the adapter is still setting up the login context. This
is resulting in errors handling the traffic.
Change the driver so that PLOGI response is sent after the login context
has been setup to avoid the situation.
Link: https://lore.kernel.org/r/20210301171821.3427-14-jsmart2021@gmail.com Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
An unlikely error exit path from lpfc_els_retry() returns incorrect status
to a caller, erroneously indicating that a retry has been successfully
issued or scheduled.
Change error exit path to indicate no retry.
Link: https://lore.kernel.org/r/20210301171821.3427-12-jsmart2021@gmail.com Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
On a pt2pt setup, between 2 initiators, if one side issues a a LOGO, there
is no relogin attempt. The FC specs are grey in this area on which port
(higher wwn or not) is to re-login.
As there is no spec guidance, unconditionally re-PLOGI after the logout to
ensure a login is re-established.
Link: https://lore.kernel.org/r/20210301171821.3427-8-jsmart2021@gmail.com Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The wqe_dbde field indicates whether a Data BDE is present in Words 0:2 and
should therefore should be clear in the abts request wqe. By setting the
bit we can be misleading fw into error cases.
Clear the wqe_dbde field.
Link: https://lore.kernel.org/r/20210301171821.3427-2-jsmart2021@gmail.com Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com> Signed-off-by: James Smart <jsmart2021@gmail.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
In file included from drivers/gpu/drm/amd/amdgpu/../display/dc/dce112/dce112_resource.c:59:
drivers/gpu/drm/amd/amdgpu/../include/asic_reg/dce/dce_11_2_sh_mask.h:10014:58: warning: initialized field overwritten [-Woverride-init]
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:214:16: note: in expansion of macro ‘AUX_SW_DATA__AUX_SW_AUTOINCREMENT_DISABLE__SHIFT’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:127:2: note: in expansion of macro ‘AUX_SF’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce112/dce112_resource.c:177:2: note: in expansion of macro ‘DCE_AUX_MASK_SH_LIST’
drivers/gpu/drm/amd/amdgpu/../include/asic_reg/dce/dce_11_2_sh_mask.h:10014:58: note: (near initialization for ‘aux_shift.AUX_SW_AUTOINCREMENT_DISABLE’)
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:214:16: note: in expansion of macro ‘AUX_SW_DATA__AUX_SW_AUTOINCREMENT_DISABLE__SHIFT’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:127:2: note: in expansion of macro ‘AUX_SF’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce112/dce112_resource.c:177:2: note: in expansion of macro ‘DCE_AUX_MASK_SH_LIST’
drivers/gpu/drm/amd/amdgpu/../include/asic_reg/dce/dce_11_2_sh_mask.h:10013:56: warning: initialized field overwritten [-Woverride-init]
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:214:16: note: in expansion of macro ‘AUX_SW_DATA__AUX_SW_AUTOINCREMENT_DISABLE_MASK’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:127:2: note: in expansion of macro ‘AUX_SF’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce112/dce112_resource.c:181:2: note: in expansion of macro ‘DCE_AUX_MASK_SH_LIST’
drivers/gpu/drm/amd/amdgpu/../include/asic_reg/dce/dce_11_2_sh_mask.h:10013:56: note: (near initialization for ‘aux_mask.AUX_SW_AUTOINCREMENT_DISABLE’)
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:214:16: note: in expansion of macro ‘AUX_SW_DATA__AUX_SW_AUTOINCREMENT_DISABLE_MASK’
drivers/gpu/drm/amd/amdgpu/../display/dc/dce/dce_aux.h:127:2: note: in expansion of macro ‘AUX_SF’
Cc: Harry Wentland <harry.wentland@amd.com> Cc: Leo Li <sunpeng.li@amd.com> Cc: Alex Deucher <alexander.deucher@amd.com> Cc: "Christian König" <christian.koenig@amd.com> Cc: David Airlie <airlied@linux.ie> Cc: Daniel Vetter <daniel@ffwll.ch> Cc: amd-gfx@lists.freedesktop.org Cc: dri-devel@lists.freedesktop.org Signed-off-by: Lee Jones <lee.jones@linaro.org> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
amdgpu DM handles INTERRUPT_LOW_IRQ_CONTEXT interrupt(hpd, hpd_rx) by using work
queue and uses single work_struct. If new interrupt is recevied before the
previous handler finished, new interrupts(same type) will be discarded and
driver just sends "amdgpu_dm_irq_schedule_work FAILED" message out. If some
important hpd, hpd_rx related interrupts are missed by driver the hot (un)plug
devices may cause system hang or instability, such as issues with system
resume from S3 sleep with mst device connected.
This patch dynamically allocates new amdgpu_dm_irq_handler_data for new
interrupts if previous INTERRUPT_LOW_IRQ_CONTEXT interrupt work has not been
handled. So the new interrupt works can be queued to the same workqueue_struct,
instead of discard the new interrupts. All allocated amdgpu_dm_irq_handler_data
are put into a single linked list and will be reused after.
[Why]
GPINT timeout is causing PSR_STATE_0 to be returned when it shouldn't.
We must guarantee that PSR is fully disabled before doing hw programming
on driver-side.
[How]
Return invalid state if GPINT command times out. Let existing retry
logic send the GPINT until successful.
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com> Signed-off-by: Wyatt Wood <wyatt.wood@amd.com> Reviewed-by: Anthony Koo <Anthony.Koo@amd.com> Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
projects / thirdparty / kernel / stable.git / log
