[PATCH v3] drm/amdgpu: Check fence emitted count to identify bad jobs

[Fan, Shikang]

[AMD Official Use Only - AMD Internal Distribution Only]

ABANDON this patch. Need further modification.

Regards,
Shikang

________________________________
From: Shikang Fan <shikang.fan at amd.com>
Sent: Thursday, November 21, 2024 11:43 AM
To: amd-gfx at lists.freedesktop.org <amd-gfx at lists.freedesktop.org>
Cc: Fan, Shikang <Shikang.Fan at amd.com>; Deng, Emily <Emily.Deng at amd.com>
Subject: [PATCH v3] drm/amdgpu: Check fence emitted count to identify bad jobs

In SRIOV, when host driver performs MODE 1 reset and notifies FLR to
guest driver, there is a small chance that there is no job running on hw
but the driver has not updated the pending list yet, causing the driver
not respond the FLR request. Modify the has_job_running function to
make sure if there is still running job.

v2: Use amdgpu_fence_count_emitted to determine job running status.
v3: Remove the timeout wait in has_job_running

Signed-off-by: Emily Deng <Emily.Deng at amd.com>
Signed-off-by: Shikang Fan <shikang.fan at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index b3ca911e55d6..ff9995c0f764 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -5222,15 +5222,19 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
 }

 /**
- * amdgpu_device_has_job_running - check if there is any job in mirror list
+ * amdgpu_device_has_job_running - check if there is any unfinished job
  *
  * @adev: amdgpu_device pointer
  *
- * check if there is any job in mirror list
+ * check if there is any job running on the device when guest driver receives
+ * FLR notification from host driver. If there are still jobs running and not
+ * signaled after 1s, the hardware is most likely hung already, then the guest
+ * driver will not respond the FLR reset. Instead, let the job hit the timeout
+ * and guest driver then issue the reset request.
  */
 bool amdgpu_device_has_job_running(struct amdgpu_device *adev)
 {
-       int i;
+       int i, j;
         struct drm_sched_job *job;

         for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
@@ -5239,11 +5243,7 @@ bool amdgpu_device_has_job_running(struct amdgpu_device *adev)
                 if (!amdgpu_ring_sched_ready(ring))
                         continue;

-               spin_lock(&ring->sched.job_list_lock);
-               job = list_first_entry_or_null(&ring->sched.pending_list,
-                                              struct drm_sched_job, list);
-               spin_unlock(&ring->sched.job_list_lock);
-               if (job)
+               if (amdgpu_fence_count_emitted(ring))
                         return true;
         }
         return false;
--
2.34.1

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20241121/9e126533/attachment.htm>