linux/drivers/gpu/drm/i915/gt
Chris Wilson b82a8b93b4 drm/i915: Be wary of data races when reading the active execlists
To implement preempt-to-busy (and so efficient timeslicing and best utilization
of the hardware submission ports) we let the GPU run asynchronously in respect
to the ELSP submission queue. This created challenges in keeping and accessing
the driver state mirroring the asynchronous GPU execution.

The latest occurence of this was spotted by KCSAN:

[ 1413.563200] BUG: KCSAN: data-race in __await_execution+0x217/0x370 [i915]
[ 1413.563221]
[ 1413.563236] race at unknown origin, with read to 0xffff88885bb6c478 of 8 bytes by task 9654 on cpu 1:
[ 1413.563548]  __await_execution+0x217/0x370 [i915]
[ 1413.563891]  i915_request_await_dma_fence+0x4eb/0x6a0 [i915]
[ 1413.564235]  i915_request_await_object+0x421/0x490 [i915]
[ 1413.564577]  i915_gem_do_execbuffer+0x29b7/0x3c40 [i915]
[ 1413.564967]  i915_gem_execbuffer2_ioctl+0x22f/0x5c0 [i915]
[ 1413.564998]  drm_ioctl_kernel+0x156/0x1b0
[ 1413.565022]  drm_ioctl+0x2ff/0x480
[ 1413.565046]  __x64_sys_ioctl+0x87/0xd0
[ 1413.565069]  do_syscall_64+0x4d/0x80
[ 1413.565094]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

To complicate matters, we have to both avoid the read tearing of *active and
avoid any write tearing as perform the pending[] -> inflight[] promotion of the
execlists.

This is because we cannot rely on the memcpy doing u64 aligned copies on all
kernels/platforms and so we opt to open-code it with explicit WRITE_ONCE
annotations to satisfy KCSAN.

v2: When in doubt, write the same comment again.
v3: Expanded commit message.

Fixes: b55230e5e8 ("drm/i915: Check for awaits on still currently executing requests")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200716142207.13003-1-chris@chris-wilson.co.uk
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[Joonas: Rebased and reordered into drm-intel-gt-next branch]
[Joonas: Added expanded commit message from Tvrtko and Chris]
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
(cherry picked from commit b4d9145b01)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
2020-09-16 11:09:51 +03:00
..
selftests
shaders drm/i915: Include asm sources for {ivb, hsw}_clear_kernel.c 2020-06-29 11:29:12 +03:00
uc drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
debugfs_engines.c drm/i915/gt: allow setting generic data pointer 2020-03-26 21:20:51 +00:00
debugfs_engines.h drm/i915/gt: Move pm debug files into a gt aware debugfs 2019-12-22 15:25:10 +00:00
debugfs_gt_pm.c drm/i915/gt: Move rps.enabled/active to flags 2020-04-30 00:57:35 +01:00
debugfs_gt_pm.h drm/i915/gt: Move pm debug files into a gt aware debugfs 2019-12-22 15:25:10 +00:00
debugfs_gt.c drm/i915: Move sseu debugfs under gt/ 2020-07-08 21:40:15 +01:00
debugfs_gt.h drm/i915/gt: allow setting generic data pointer 2020-03-26 21:20:51 +00:00
gen2_engine_cs.c drm/i915/gt: Flush gen3 relocs harder, again 2020-06-13 10:30:01 +01:00
gen2_engine_cs.h drm/i915/gt: Split low level gen2-7 CS emitters 2020-06-02 10:42:00 +01:00
gen6_engine_cs.c drm/i915/gt: Split low level gen2-7 CS emitters 2020-06-02 10:42:00 +01:00
gen6_engine_cs.h drm/i915/gt: Split low level gen2-7 CS emitters 2020-06-02 10:42:00 +01:00
gen6_ppgtt.c drm/i915/gt: Optimise aliasing-ppgtt allocations 2020-07-09 20:40:49 +01:00
gen6_ppgtt.h drm/i915/gtt: split up i915_gem_gtt 2020-01-07 19:27:36 +00:00
gen6_renderstate.c
gen7_renderclear.c drm/i915: Release shortlived maps of longlived objects 2020-07-08 22:05:50 +01:00
gen7_renderclear.h drm/i915/gen7: Clear all EU/L3 residual contexts 2020-03-06 08:59:06 +00:00
gen7_renderstate.c
gen8_ppgtt.c drm/i915: Remove duplicate inline specifier on write_pte 2020-05-13 20:01:28 +01:00
gen8_ppgtt.h drm/i915/gtt: split up i915_gem_gtt 2020-01-07 19:27:36 +00:00
gen8_renderstate.c
gen9_renderstate.c
hsw_clear_kernel.c drm/i915/gen7: Clear all EU/L3 residual contexts 2020-03-06 08:59:06 +00:00
intel_breadcrumbs.c drm/i915: Skip signaling a signaled request 2020-07-13 17:57:54 +01:00
intel_context_param.c drm/i915: Allow userspace to specify ringsize on construction 2020-02-25 19:23:19 +00:00
intel_context_param.h drm/i915: Allow userspace to specify ringsize on construction 2020-02-25 19:23:19 +00:00
intel_context_sseu.c drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
intel_context_types.h drm/i915/execlists: Avoid reusing the same logical CCID 2020-04-28 22:17:36 +01:00
intel_context.c drm/i915/gt: Pin the rings before marking active 2020-07-07 18:00:15 -07:00
intel_context.h drm/i915: Use explicit flag to mark unreachable intel_context 2020-03-26 10:21:04 -07:00
intel_engine_cs.c treewide: Use fallthrough pseudo-keyword 2020-08-23 17:36:59 -05:00
intel_engine_heartbeat.c drm/i915/gt: Move the heartbeat into the high priority system wq 2020-07-02 12:30:24 +01:00
intel_engine_heartbeat.h
intel_engine_pm.c drm/i915/gt: Assert the kernel context is using the HWSP 2020-07-15 10:18:29 +01:00
intel_engine_pm.h drm/i915: Extend intel_wakeref to support delayed puts 2020-03-23 12:51:05 +00:00
intel_engine_types.h drm/i915/gt: Be defensive in the face of false CS events 2020-07-10 15:24:17 +01:00
intel_engine_user.c drm/i915: Reboot CI if we get wedged during driver init 2020-07-06 19:21:07 +01:00
intel_engine_user.h
intel_engine.h drm/i915/gt: Always report the sample time for busy-stats 2020-06-18 09:26:54 +01:00
intel_ggtt_fencing.c drm/i915/gt: Make fence revocation unequivocal 2020-04-01 23:34:17 +01:00
intel_ggtt_fencing.h drm/i915/gt: Store the fence details on the fence 2020-04-01 23:34:16 +01:00
intel_ggtt.c treewide: Use fallthrough pseudo-keyword 2020-08-23 17:36:59 -05:00
intel_gpu_commands.h drm/i915/gen12: Fix HDC pipeline flush 2020-05-07 07:44:41 +01:00
intel_gt_buffer_pool_types.h drm/i915/gt: Move the batch buffer pool from the engine to the gt 2020-04-30 19:12:02 +01:00
intel_gt_buffer_pool.c drm/i915/gt: Cancel the flush worker more thoroughly 2020-05-25 16:45:17 +01:00
intel_gt_buffer_pool.h drm/i915/gt: Move the batch buffer pool from the engine to the gt 2020-04-30 19:12:02 +01:00
intel_gt_clock_utils.c drm/i915/gt: Fix up clock frequency 2020-04-27 17:34:33 +01:00
intel_gt_clock_utils.h drm/i915/gt: Use the RPM config register to determine clk frequencies 2020-04-24 19:10:17 +01:00
intel_gt_irq.c drm/i915/gt: Be defensive in the face of false CS events 2020-07-10 15:24:17 +01:00
intel_gt_irq.h drm/i915/gt: Reorganise gen8+ interrupt handler 2020-01-28 12:30:00 +00:00
intel_gt_pm_irq.c
intel_gt_pm_irq.h
intel_gt_pm.c drm/i915: Reboot CI if we get wedged during driver init 2020-07-06 19:21:07 +01:00
intel_gt_pm.h drm/i915/gt: Apply sanitiization just before resume 2019-12-26 12:37:30 +00:00
intel_gt_requests.c drm/i915/gem: Unpin idle contexts from kswapd reclaim 2020-07-08 22:05:49 +01:00
intel_gt_requests.h
intel_gt_types.h drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
intel_gt.c drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
intel_gt.h drm/i915: Introduce gt_init_mmio 2020-07-08 21:07:13 +01:00
intel_gtt.c drm/i915/gt: Pull marking vm as closed underneath the vm->mutex 2020-02-28 12:33:07 +00:00
intel_gtt.h drm/i915: Export ppgtt_bind_vma 2020-07-03 15:14:35 +01:00
intel_llc_types.h
intel_llc.c drm/i915/gt: Do not attempt to reprogram IA/ring frequencies for dgfx 2020-02-19 22:27:10 +00:00
intel_llc.h
intel_lrc_reg.h drm/i915/selftests: Add context batchbuffers registers to live_lrc_fixed 2020-04-24 00:36:13 +01:00
intel_lrc.c drm/i915: Be wary of data races when reading the active execlists 2020-09-16 11:09:51 +03:00
intel_lrc.h drm/i915: Make define for lrc state offset 2020-04-24 00:52:14 +01:00
intel_mocs.c drm/i915/gt: Refactor l3cc/mocs availability 2020-02-19 14:09:18 +00:00
intel_mocs.h
intel_ppgtt.c drm/i915: Export ppgtt_bind_vma 2020-07-03 15:14:35 +01:00
intel_rc6_types.h drm/i915/gt: Only ignore rc6 parking for PCU on byt/bsw 2019-12-12 08:35:48 +00:00
intel_rc6.c drm/i915: Remove cnl pre-prod workarounds 2020-05-04 18:44:52 +03:00
intel_rc6.h
intel_renderstate.c drm/i915: Release shortlived maps of longlived objects 2020-07-08 22:05:50 +01:00
intel_renderstate.h
intel_reset_types.h drm/i915: Reboot CI if we get wedged during driver init 2020-07-06 19:21:07 +01:00
intel_reset.c drm/i915: Move the engine mask to intel_gt_info 2020-07-08 21:07:11 +01:00
intel_reset.h drm/i915: Reboot CI if we get wedged during driver init 2020-07-06 19:21:07 +01:00
intel_ring_submission.c treewide: Use fallthrough pseudo-keyword 2020-08-23 17:36:59 -05:00
intel_ring_types.h drm/i915/gt: Avoid resetting ring->head outside of its timeline mutex 2020-02-11 12:03:22 +00:00
intel_ring.c drm/i915/gt: Incrementally check for rewinding 2020-06-16 11:34:23 +03:00
intel_ring.h drm/i915/gt: Mark up racy read of intel_ring.head 2020-04-08 13:40:07 +01:00
intel_rps_types.h drm/i915/gt: Switch to manual evaluation of RPS 2020-04-30 00:57:37 +01:00
intel_rps.c drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
intel_rps.h drm/i915/gt: Sanitize RPS interrupts upon resume 2020-05-03 08:24:36 +01:00
intel_sseu_debugfs.c drm/i915: Move sseu debugfs under gt/ 2020-07-08 21:40:15 +01:00
intel_sseu_debugfs.h drm/i915: Move sseu debugfs under gt/ 2020-07-08 21:40:15 +01:00
intel_sseu.c drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
intel_sseu.h drm/i915/sseu: Move sseu_info under gt_info 2020-07-08 21:13:09 +01:00
intel_timeline_types.h drm/i915/gt: Pull GT initialisation under intel_gt_init() 2019-12-22 12:51:32 +00:00
intel_timeline.c drm/i915/gt: Trace placement of timeline HWSP 2020-07-15 10:17:44 +01:00
intel_timeline.h drm/i915: Make intel_timeline_init static 2020-05-11 12:52:08 +01:00
intel_workarounds_types.h drm/i915/gt: Skip rmw for masked registers 2020-02-01 09:21:57 +00:00
intel_workarounds.c drm/i915/tgl: Implement WAs 18011464164 and 22010931296 2020-07-09 10:05:07 -07:00
intel_workarounds.h
ivb_clear_kernel.c drm/i915/gen7: Clear all EU/L3 residual contexts 2020-03-06 08:59:06 +00:00
mock_engine.c drm/i915/gt: Move the batch buffer pool from the engine to the gt 2020-04-30 19:12:02 +01:00
mock_engine.h
selftest_context.c drm/i915: Drop no-semaphore boosting 2020-05-14 06:14:33 +01:00
selftest_engine_cs.c drm/i915: Drop i915_request.i915 backpointer 2020-06-03 13:53:39 +01:00
selftest_engine_heartbeat.c drm/i915/params: switch to device specific parameters 2020-06-22 23:26:40 +03:00
selftest_engine_heartbeat.h drm/i915/selftests: Enable selftesting of busy-stats 2020-06-18 09:26:53 +01:00
selftest_engine_pm.c drm/i915/gt: Always report the sample time for busy-stats 2020-06-18 09:26:54 +01:00
selftest_engine.c
selftest_engine.h
selftest_gt_pm.c drm/i915/selftests: Measure CS_TIMESTAMP 2020-05-21 09:30:16 +01:00
selftest_hangcheck.c drm/i915/selftests: Enable selftesting of busy-stats 2020-06-18 09:26:53 +01:00
selftest_llc.c drm/i915/gt: Do not attempt to reprogram IA/ring frequencies for dgfx 2020-02-19 22:27:10 +00:00
selftest_llc.h
selftest_lrc.c drm/i915: Move the engine mask to intel_gt_info 2020-07-08 21:07:11 +01:00
selftest_mocs.c drm/i915/gt: Incrementally check for rewinding 2020-06-16 11:34:23 +03:00
selftest_rc6.c drm/i915: Print caller when tainting for CI 2020-07-06 19:21:07 +01:00
selftest_rc6.h
selftest_reset.c drm/i915/guc: Kill USES_GUC_SUBMISSION macro 2020-02-20 17:48:03 +00:00
selftest_ring_submission.c drm/i915/selftests: Always flush before unpining after writing 2020-05-11 16:50:04 +01:00
selftest_ring.c drm/i915/gt: Incrementally check for rewinding 2020-06-16 11:34:23 +03:00
selftest_rps.c drm/i915/selftests: Fix compare functions provided for sorting 2020-07-14 20:30:58 +03:00
selftest_rps.h drm/i915/gt: Fix up clock frequency 2020-04-27 17:34:33 +01:00
selftest_timeline.c drm/i915/gt: Trace placement of timeline HWSP 2020-07-15 10:17:44 +01:00
selftest_workarounds.c drm/i915: work around false-positive maybe-uninitialized warning 2020-06-15 12:36:07 +03:00
shmem_utils.c drm/i915/gt: Keep a no-frills swappable copy of the default context state 2020-04-29 19:02:37 +01:00
shmem_utils.h drm/i915/gt: Keep a no-frills swappable copy of the default context state 2020-04-29 19:02:37 +01:00
st_shmem_utils.c drm/i915/gt: Keep a no-frills swappable copy of the default context state 2020-04-29 19:02:37 +01:00
sysfs_engines.c drm/i915: Show per-engine default property values in sysfs 2020-05-14 08:25:05 +01:00
sysfs_engines.h drm/i915/gt: Expose engine properties via sysfs 2020-02-28 22:03:19 +00:00