Probelm:
Singlaning one sched fence from within another's sched
fence singal callback generates lockdep splat because
the both have same lockdep class of their fence->lock
Fix:
Fix bellow stack by rescheduling to irq work of
signaling and killing of jobs that left when entity is killed.
[11176.741181] dump_stack+0x10/0x12
[11176.741186] __lock_acquire.cold+0x208/0x2df
[11176.741197] lock_acquire+0xc6/0x2d0
[11176.741204] ? dma_fence_signal+0x28/0x80
[11176.741212] _raw_spin_lock_irqsave+0x4d/0x70
[11176.741219] ? dma_fence_signal+0x28/0x80
[11176.741225] dma_fence_signal+0x28/0x80
[11176.741230] drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[11176.741240] drm_sched_entity_kill_jobs_cb+0x1c/0x50 [gpu_sched]
[11176.741248] dma_fence_signal_timestamp_locked+0xac/0x1a0
[11176.741254] dma_fence_signal+0x3b/0x80
[11176.741260] drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[11176.741268] drm_sched_job_done.isra.0+0x7f/0x1a0 [gpu_sched]
[11176.741277] drm_sched_job_done_cb+0x12/0x20 [gpu_sched]
[11176.741284] dma_fence_signal_timestamp_locked+0xac/0x1a0
[11176.741290] dma_fence_signal+0x3b/0x80
[11176.741296] amdgpu_fence_process+0xd1/0x140 [amdgpu]
[11176.741504] sdma_v4_0_process_trap_irq+0x8c/0xb0 [amdgpu]
[11176.741731] amdgpu_irq_dispatch+0xce/0x250 [amdgpu]
[11176.741954] amdgpu_ih_process+0x81/0x100 [amdgpu]
[11176.742174] amdgpu_irq_handler+0x26/0xa0 [amdgpu]
[11176.742393] __handle_irq_event_percpu+0x4f/0x2c0
[11176.742402] handle_irq_event_percpu+0x33/0x80
[11176.742408] handle_irq_event+0x39/0x60
[11176.742414] handle_edge_irq+0x93/0x1d0
[11176.742419] __common_interrupt+0x50/0xe0
[11176.742426] common_interrupt+0x80/0x90
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Suggested-by: Christian König <christian.koenig@amd.com>
Tested-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://www.spinics.net/lists/dri-devel/msg321250.html
drm_sched_job_cleanup() will pass an uninitialized fence to
drm_sched_fence_free(), which will cause to_drm_sched_fence() to return
a NULL fence object, causing a NULL pointer deref when this NULL object
is passed to kmem_cache_free().
Let's create a new drm_sched_fence_free() function that takes a
drm_sched_fence pointer and suffix the old function with _rcu. While at
it, complain if drm_sched_fence_free() is passed an initialized fence
or if drm_sched_fence_free_rcu() is passed an uninitialized fence.
Fixes: dbe48d030b ("drm/sched: Split drm_sched_job_init")
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210903120554.444101-1-boris.brezillon@collabora.com
[Why]
Previous tdr design treats the first job in job_timeout as the bad job.
But sometimes a later bad compute job can block a good gfx job and
cause an unexpected gfx job timeout because gfx and compute ring share
internal GC HW mutually.
[How]
This patch implements an advanced tdr mode.It involves an additinal
synchronous pre-resubmit step(Step0 Resubmit) before normal resubmit
step in order to find the real bad job.
1. At Step0 Resubmit stage, it synchronously submits and pends for the
first job being signaled. If it gets timeout, we identify it as guilty
and do hw reset. After that, we would do the normal resubmit step to
resubmit left jobs.
2. For whole gpu reset(vram lost), do resubmit as the old way.
v2: squash in build fix (Alex)
Signed-off-by: Jack Zhang <Jack.Zhang1@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Rename "ring_mirror_list" to "pending_list",
to describe what something is, not what it does,
how it's used, or how the hardware implements it.
This also abstracts the actual hardware
implementation, i.e. how the low-level driver
communicates with the device it drives, ring, CAM,
etc., shouldn't be exposed to DRM.
The pending_list keeps jobs submitted, which are
out of our control. Usually this means they are
pending execution status in hardware, but the
latter definition is a more general (inclusive)
definition.
Signed-off-by: Luben Tuikov <luben.tuikov@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/405573/
Cc: Alexander Deucher <Alexander.Deucher@amd.com>
Cc: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Christian König <christian.koenig@amd.com>
Remove DRM_SCHED_PRIORITY_INVALID. We no longer
carry around an invalid priority and cut it off
at the source.
Backwards compatibility behaviour of AMDGPU CTX
IOCTL passing in garbage for context priority
from user space and then mapping that to
DRM_SCHED_PRIORITY_NORMAL is preserved.
v2: Revert "res" --> "r" and
"prio" --> "priority".
Signed-off-by: Luben Tuikov <luben.tuikov@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Remove DRM_SCHED_PRIORITY_LOW, as it was used
in only one place.
Rename and separate by a line
DRM_SCHED_PRIORITY_MAX to DRM_SCHED_PRIORITY_COUNT
as it represents a (total) count of said
priorities and it is used as such in loops
throughout the code. (0-based indexing is the
the count number.)
Remove redundant word HIGH in priority names,
and rename *KERNEL* to *HIGH*, as it really
means that, high.
v2: Add back KERNEL and remove SW and HW,
in lieu of a single HIGH between NORMAL and KERNEL.
Signed-off-by: Luben Tuikov <luben.tuikov@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch uses score to select a new drm scheduler for better
loadbalance between multiple drm schedulers instead of num_jobs.
Below are test results after running amdgpu_test for ~10 times.
Before this patch:
sched_name num of many times it got schedule
========= ==================================
sdma0 1463
sdma1 198
comp_1.0.1 280
After this patch:
sched_name num of many times it got schedule
========= ==================================
sdma0 925
sdma1 928
comp_1.0.1 177
comp_1.1.1 44
comp_1.2.1 43
comp_1.3.1 44
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://patchwork.freedesktop.org/patch/373000/
Signed-off-by: Christian König <christian.koenig@amd.com>
Remove drm_sched_entity_get_free_sched() and use the logic of picking
the least loaded drm scheduler from a drm scheduler list to implement
drm_sched_pick_best(). This patch also exports drm_sched_pick_best() so
that it can be utilized by other drm drivers.
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Implement drm_sched_entity_modify_sched() which modifies existing
sched_list with a different one. This is going to be helpful when
userspace changes priority of a ctx/entity then the driver can switch
to the corresponding HW scheduler list for that priority.
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch uses score based logic to select a new rq for better
loadbalance between multiple rq/scheds instead of num_jobs.
Below are test results after running amdgpu_test from mesa drm
Before this patch:
sched_name num of many times it got scheduled
========= ==================================
sdma0 314
sdma1 32
comp_1.0.0 56
comp_1.0.1 0
comp_1.1.0 0
comp_1.1.1 0
comp_1.2.0 0
comp_1.2.1 0
comp_1.3.0 0
comp_1.3.1 0
After this patch:
sched_name num of many times it got scheduled
========= ==================================
sdma0 216
sdma1 185
comp_1.0.0 39
comp_1.0.1 9
comp_1.1.0 12
comp_1.1.1 0
comp_1.2.0 12
comp_1.2.1 0
comp_1.3.0 12
comp_1.3.1 0
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a drm scheduler.
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Removes thread park/unpark hack from drm_sched_entity_fini and
by this fixes reactivation of scheduler thread while the thread
is supposed to be stopped.
v2: Per sched entity completion.
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.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>
Expedite job deletion from ring mirror list to the HW fence signal
callback instead from finish_work, together with waiting for all
such fences to signal in drm_sched_stop we garantee that
already signaled job will not be processed twice.
Remove the sched finish fence callback and just submit finish_work
directly from the HW fence callback.
v2: Fix comments.
v3: Attach hw fence cb to sched_job
v5: Rebase
Suggested-by: Christian Koenig <Christian.Koenig@amd.com>
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>
Decauple sched threads stop and start and ring mirror
list handling from the policy of what to do about the
guilty jobs.
When stoppping the sched thread and detaching sched fences
from non signaled HW fenes wait for all signaled HW fences
to complete before rerunning the jobs.
v2: Fix resubmission of guilty job into HW after refactoring.
v4:
Full restart for all the jobs, not only from guilty ring.
Extract karma increase into standalone function.
v5:
Rework waiting for signaled jobs without relying on the job
struct itself as those might already be freed for non 'guilty'
job's schedulers.
Expose karma increase to drivers.
v6:
Use list_for_each_entry_safe_continue and drm_sched_process_job
in case fence already signaled.
Call drm_sched_increase_karma only once for amdgpu and add documentation.
v7:
Wait only for the latest job's fence.
Suggested-by: Christian Koenig <Christian.Koenig@amd.com>
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 patch adds two new functions to help client drivers suspend and
resume the scheduler job timeout. This can be useful in cases where the
hardware has preemption support enabled. Using this, it is possible to have
the timeout active only for the ring which is active on the ringbuffer.
This patch also makes the job_list_lock IRQ safe.
Suggested-by: Christian Koenig <Christian.Koenig@amd.com>
Signed-off-by: Sharat Masetty <smasetty@codeaurora.org>
Reviewed-by: Christian König <christian.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>
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>
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>
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>
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>
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>
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>
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>
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>
