This reverts commit 0efd2d2f68.
It's still causing problems for V3D.
v2: keep rearming the timeout.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
A bad job is the one triggered TDR(In the current amdgpu's
implementation, actually all the jobs in the current joq-queue will
be treated as bad jobs). In the recovery process, its fence
will be fake signaled and as a result, the work behind will be scheduled
to delete it from the mirror list, but if the TDR process is invoked
before the work's execution, then this bad job might be processed again
and the call dma_fence_set_error to its fence in TDR process will lead to
kernel warning trace:
[ 143.033605] WARNING: CPU: 2 PID: 53 at ./include/linux/dma-fence.h:437 amddrm_sched_job_recovery+0x1af/0x1c0 [amd_sched]
kernel: [ 143.033606] Modules linked in: amdgpu(OE) amdchash(OE) amdttm(OE) amd_sched(OE) amdkcl(OE) amd_iommu_v2 drm_kms_helper drm i2c_algo_bit fb_sys_fops syscopyarea sysfillrect sysimgblt kvm_intel kvm irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel aes_x86_64 snd_hda_codec_generic crypto_simd glue_helper cryptd snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq joydev snd_seq_device snd_timer snd soundcore binfmt_misc input_leds mac_hid serio_raw nfsd auth_rpcgss nfs_acl lockd grace sunrpc sch_fq_codel parport_pc ppdev lp parport ip_tables x_tables autofs4 8139too floppy psmouse 8139cp mii i2c_piix4 pata_acpi
[ 143.033649] CPU: 2 PID: 53 Comm: kworker/2:1 Tainted: G OE 4.15.0-20-generic #21-Ubuntu
[ 143.033650] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014
[ 143.033653] Workqueue: events drm_sched_job_timedout [amd_sched]
[ 143.033656] RIP: 0010:amddrm_sched_job_recovery+0x1af/0x1c0 [amd_sched]
[ 143.033657] RSP: 0018:ffffa9f880fe7d48 EFLAGS: 00010202
[ 143.033659] RAX: 0000000000000007 RBX: ffff9b98f2b24c00 RCX: ffff9b98efef4f08
[ 143.033660] RDX: ffff9b98f2b27400 RSI: ffff9b98f2b24c50 RDI: ffff9b98efef4f18
[ 143.033660] RBP: ffffa9f880fe7d98 R08: 0000000000000001 R09: 00000000000002b6
[ 143.033661] R10: 0000000000000000 R11: 0000000000000000 R12: ffff9b98efef3430
[ 143.033662] R13: ffff9b98efef4d80 R14: ffff9b98efef4e98 R15: ffff9b98eaf91c00
[ 143.033663] FS: 0000000000000000(0000) GS:ffff9b98ffd00000(0000) knlGS:0000000000000000
[ 143.033664] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 143.033665] CR2: 00007fc49c96d470 CR3: 000000001400a005 CR4: 00000000003606e0
[ 143.033669] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 143.033669] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 143.033670] Call Trace:
[ 143.033744] amdgpu_device_gpu_recover+0x144/0x820 [amdgpu]
[ 143.033788] amdgpu_job_timedout+0x9b/0xa0 [amdgpu]
[ 143.033791] drm_sched_job_timedout+0xcc/0x150 [amd_sched]
[ 143.033795] process_one_work+0x1de/0x410
[ 143.033797] worker_thread+0x32/0x410
[ 143.033799] kthread+0x121/0x140
[ 143.033801] ? process_one_work+0x410/0x410
[ 143.033803] ? kthread_create_worker_on_cpu+0x70/0x70
[ 143.033806] ret_from_fork+0x35/0x40
So just delete the bad job from mirror list directly
Changes in v3:
- Add a helper function to delete the bad jobs from mirror list and call
it directly *before* the job's fence is signaled
Changes in v2:
- delete the useless list node check
- also delete bad jobs in drm_sched_main because:
kthread_unpark(ring->sched.thread) will be invoked very early before
amdgpu_device_gpu_recover's return, then drm_sched_main will have
chance to pick up a new job from the job queue. This new job will be
added into the mirror list and processed by amdgpu_job_run, but may
not be deleted from the mirror list on time due to the same reason.
And finally re-processed by drm_sched_job_recovery
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Reviewed-by: Christian König <chrstian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds a new API to clean up the scheduler job resources. This
is primarliy needed in cases the job was created but was not queued to
the scheduler queue. Additionally with this change, the layer which
creates the scheduler job also gets to free up the job's resources and
this entails moving the dma_fence_put(finished_fence) to the drivers
ops free handler routines.
Signed-off-by: Sharat Masetty <smasetty@codeaurora.org>
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Problem:
A particular scheduler may become unsuable (underlying HW) after
some event (e.g. GPU reset). If it's later chosen by
the get free sched. policy a command will fail to be
submitted.
Fix:
Add a driver specific callback to report the sched status so
rq with bad sched can be avoided in favor of working one or
none in which case job init will fail.
v2: Switch from driver callback to flag in scheduler.
v3: rebase
v4: Remove ready paramter from drm_sched_init, set
uncoditionally to true once init done.
v5: fix missed change in v3d in v4 (Alex)
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add a helper to immediately start timeout handling in case of a hardware
fault.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Make sure we always restart the timer after a timeout and remove the
device specific workarounds.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We need to make sure that we don't race between job completion and
timeout.
v2: put revert label after calling the handling manually
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cleanup starting the timeout a bit.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clang generates a warning when it sees a logical not followed by a
conditional operator like ==, >, or <.
drivers/gpu/drm/scheduler/sched_entity.c:470:6: warning: logical not is
only applied to the left hand side of this comparison
[-Wlogical-not-parentheses]
if (!spsc_queue_count(&entity->job_queue) == 0 ||
^ ~~
drivers/gpu/drm/scheduler/sched_entity.c:470:6: note: add parentheses
after the '!' to evaluate the comparison first
if (!spsc_queue_count(&entity->job_queue) == 0 ||
^
( )
drivers/gpu/drm/scheduler/sched_entity.c:470:6: note: add parentheses
around left hand side expression to silence this warning
if (!spsc_queue_count(&entity->job_queue) == 0 ||
^
( )
1 warning generated.
It assumes the author might have made a mistake in their logic:
if (!a == b) -> if (!(a == b))
Sometimes that is the case; other times, it's just a super convoluted
way of saying 'if (a)' when b = 0:
if (!1 == 0) -> if (0 == 0) -> if (true)
Alternatively:
if (!1 == 0) -> if (!!1) -> if (1)
Simplify this comparison so that Clang doesn't complain.
Fixes: 35e160e781 ("drm/scheduler: change entities rq even earlier")
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
having a delayed work item per job is redundant as we only need one
per scheduler to track the time out the currently executing job.
v2: the first element of the ring mirror list is the currently
executing job so we don't need a additional variable for it
v3: squash in fixes for v3d and etnaviv
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
do not remove entity from the rq if the current rq is from
the least loaded scheduler.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The flag will prevent another thread from same process to
reinsert the entity queue into scheduler's rq after it was already
removewd from there by another thread during drm_sched_entity_flush.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Better match the naming of the other components.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cleanup coding style in sched_entity.c
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This is complex enough on it's own. Move it into a separate C file.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Return -ENOMEM when allocating the rq_list fails.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Looks like for correct debugging we need to know the scheduler even
earlier. So move picking a rq for an entity into job creation.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Make sure we access last_scheduled only after checking that there are no
more jobs on the entity.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
That is superflous now.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Update job earlier with the scheduler it is supposed to be scheduled on.
Otherwise we could incorrectly optimize dependencies when moving an
entity from one scheduler to another.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Since we now deal with multiple rq we need to update all of them, not
just the current one.
v2: Trivial: Removed unused variable (Alex)
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
During debug sessions I encountered a need to trace
back a job dependecy a few steps back to the first failing
job. This trace helpped me a lot.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This is the first attempt to move entities between schedulers to
have dynamic load balancing. We just move entities with no jobs for
now as moving the ones with jobs will lead to other compilcations
like ensuring that the other scheduler does not remove a job from
the current entity while we are moving.
v2: remove unused variable and an unecessary check
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The function selects the run queue from the rq_list with the
least load. The load is decided by the number of jobs in a
scheduler.
v2: avoid using atomic read twice consecutively, instead store
it locally
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
To keep track of the scheduler load.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
These are the potential run queues on which the jobs from this
entity can be scheduled. We will use this to do load balancing.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We no longer have sched parameter so remove its description
as well
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drm_sched_job_finish() is a work item scheduled for each finished job on
a unbound system workqueue. This means the workers can execute out of order
with regard to the real hardware job completions.
If this happens queueing a timeout worker for the first job on the ring
mirror list is wrong, as this may be a job which has already finished
executing. Fix this by reorganizing the code to always queue the worker
for the next job on the list, if this job hasn't finished yet. This is
robust against a potential reordering of the finish workers.
Also move out the timeout worker cancelling, so that we don't need to
take the job list lock twice. As a small optimization list_del is used
to remove the job from the ring mirror list, as there is no need to
reinit the list head in the job we are about to free.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We removed the redundancy of having an extra scheduler field, so we
can't set the rq to NULL any more or otherwise won't know which
scheduler to use for the cleanup.
Just remove the entity from the scheduling list instead.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Bug: https://bugs.freedesktop.org/show_bug.cgi?id=107367
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Note which task is using the entity and only kill it if the last user of
the entity is killed. This should prevent problems when entities are leaked to
child processes.
v2: add missing kernel doc
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
I refactored the include directives under include/drm/ some time ago.
This flag is unneeded.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The scheduler of the entity is decided by the run queue on which
it is queued. This patch avoids us the effort required to maintain
a sync between rq and sched field when we start shifting entites
among different rqs.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
entity has a scheduler field and we don't need the sched argument
in any of the functions where entity is provided.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
To check rq pointer before adding entity into it.
That avoids NULL pointer access in some case.
v2: move the check to caller
Suggested-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
replace run queue by a list of run queues and remove the
sched arg as that is part of run queue itself
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch is in preparation for a better load balancing in
scheduler. It allows us to associate entities with the
run queues instead of binding them to a scheduler.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
More features for 4.19:
- Use core pcie functionality rather than duplicating our own for pcie
gens and lanes
- Scheduler function naming cleanups
- More documentation
- Reworked DC/Powerplay interfaces to improve power savings
- Initial stutter mode support for RV (power feature)
- Vega12 powerplay updates
- GFXOFF fixes
- Misc fixes
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180705221447.2807-1-alexander.deucher@amd.com
Everything in the flush code path (i.e. waiting for SW queue
to become empty) names with *_flush()
and everything in the release code path names *_fini()
This patch also effect the amdgpu and etnaviv drivers which
use those functions.
v2:
Also pplay the change to vd3.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Acked-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
dma_fence_default_wait is the default now, same for the trivial
enable_signaling implementation.
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Sean Paul <seanpaul@chromium.org>
Cc: David Airlie <airlied@linux.ie>
Link: https://patchwork.freedesktop.org/patch/msgid/20180503142603.28513-7-daniel.vetter@ffwll.ch
Dying process might be blocked from receiving any more signals
so avoid using it.
Also retire enity->fini_status and just check the SW queue,
if it's not empty do the fallback cleanup.
Also handle entity->last_scheduled == NULL use case which
happens when HW ring is already hangged whem a new entity
tried to enqeue jobs.
v2:
Return the remaining timeout and use that as parameter for the next call.
This way when we need to cleanup multiple queues we don't wait for the
entire TO period for each queue but rather in total.
Styling comments.
Rebase.
v3:
Update types from unsigned to long.
Work with jiffies instead of ms.
Return 0 when TO expires.
Rebase.
v4:
Remove unnecessary timeout calculation.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
convert existing raw comments into kernel-doc format as well
as add new documentation
v2: reword the overview
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel@ffwll.ch>
When checking for a dependency fence for belonging to the same entity
compare it with scheduled as well finished fence. Earlier we were only
comparing it with the scheduled fence.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
To free the fence from the amdgpu_fence_slab, need twice call_rcu, to avoid
the amdgpu_fence_slab_fini call kmem_cache_destroy(amdgpu_fence_slab) before
kmem_cache_free(amdgpu_fence_slab, fence), add rcu_barrier after drm_sched_entity_fini.
The kmem_cache_free(amdgpu_fence_slab, fence)'s call trace as below:
1.drm_sched_entity_fini ->
drm_sched_entity_cleanup ->
dma_fence_put(entity->last_scheduled) ->
drm_sched_fence_release_finished ->
drm_sched_fence_release_scheduled ->
call_rcu(&fence->finished.rcu, drm_sched_fence_free)
2.drm_sched_fence_free ->
dma_fence_put(fence->parent) ->
amdgpu_fence_release ->
call_rcu(&f->rcu, amdgpu_fence_free) ->
kmem_cache_free(amdgpu_fence_slab, fence);
v2:put the barrier before the kmem_cache_destroy
v3:put the dma_fence_put(fence->parent) before call_rcu in
drm_sched_fence_release_scheduled
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>
That got missed while moving the files outside of amdgpu.
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This spinlock is superfluous, any call to drm_sched_entity_push_job
should already be under a lock together with matching drm_sched_job_init
to match the order of insertion into queue with job's fence seqence
number.
v2:
Improve patch description.
Add functions documentation describing the locking considerations
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
this patch also effect the amdgpu and etnaviv drivers which
use the function drm_sched_entity_init
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Acked-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The current sequence in scheduler thread is:
1. update last sched fence
2. job begin (adding to mirror list)
3. job finish (remove from mirror list)
4. back to 1
Since we update last sched prior to joining mirror list, the jobs
in mirror list already pass the last sched fence. TDR just run
the jobs in mirror list, so we should not update the last sched
fences in TDR.
Signed-off-by: Pixel Ding <Pixel.Ding@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Make sure main thread won't update last_sched fence when entity
is cleanup.
Fix a racing issue which is caused by putting last_sched fence
twice. Running vulkaninfo in tight loop can produce this issue
as seeing wild fence pointer.
v2: squash in build fix (Christian)
Signed-off-by: Pixel Ding <Pixel.Ding@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Monk Liu <Monk.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the potential memleak since scheduler main thread always
hold one last_sched fence.
Signed-off-by: Pixel Ding <Pixel.Ding@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
issue:
there are VMC page fault occurred if force APP kill during
3dmark test, the cause is in entity_fini we manually signal
all those jobs in entity's queue which confuse the sync/dep
mechanism:
1)page fault occurred in sdma's clear job which operate on
shadow buffer, and shadow buffer's Gart table is cleaned by
ttm_bo_release since the fence in its reservation was fake signaled
by entity_fini() under the case of SIGKILL received.
2)page fault occurred in gfx' job because during the lifetime
of gfx job we manually fake signal all jobs from its entity
in entity_fini(), thus the unmapping/clear PTE job depend on those
result fence is satisfied and sdma start clearing the PTE and lead
to GFX page fault.
fix:
1)should at least wait all jobs already scheduled complete in entity_fini()
if SIGKILL is the case.
2)if a fence signaled and try to clear some entity's dependency, should
set this entity guilty to prevent its job really run since the dependency
is fake signaled.
v2:
splitting drm_sched_entity_fini() into two functions:
1)The first one is does the waiting, removes the entity from the
runqueue and returns an error when the process was killed.
2)The second one then goes over the entity, install it as
completion signal for the remaining jobs and signals all jobs
with an error code.
v3:
1)Replace the fini1 and fini2 with better name
2)Call the first part before the VM teardown in
amdgpu_driver_postclose_kms() and the second part
after the VM teardown
3)Keep the original function drm_sched_entity_fini to
refine the code.
v4:
1)Rename entity->finished to entity->last_scheduled;
2)Rename drm_sched_entity_fini_job_cb() to
drm_sched_entity_kill_jobs_cb();
3)Pass NULL to drm_sched_entity_fini_job_cb() if -ENOENT;
4)Replace the type of entity->fini_status with "int";
5)Remove the check about entity->finished.
Signed-off-by: Monk Liu <Monk.Liu@amd.com>
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>
Christian König