[PATCH] amd/amdgpu: Reduce unnecessary repetitive GPU resets

Zhang, Hawking Hawking.Zhang at amd.com
Fri Sep 20 08:01:04 UTC 2024


[AMD Official Use Only - AMD Internal Distribution Only]

Reviewed-by: Hawking Zhang <Hawking.Zhang at amd.com>

Regards,
Hawking
-----Original Message-----
From: Chai, Thomas <YiPeng.Chai at amd.com>
Sent: Friday, September 20, 2024 15:36
To: amd-gfx at lists.freedesktop.org
Cc: Zhang, Hawking <Hawking.Zhang at amd.com>; Zhou1, Tao <Tao.Zhou1 at amd.com>; Li, Candice <Candice.Li at amd.com>; Wang, Yang(Kevin) <KevinYang.Wang at amd.com>; Yang, Stanley <Stanley.Yang at amd.com>; Chai, Thomas <YiPeng.Chai at amd.com>
Subject: [PATCH] amd/amdgpu: Reduce unnecessary repetitive GPU resets

In multiple GPUs case, after a GPU has started resetting all GPUs on hive, other GPUs do not need to trigger GPU reset again.

Signed-off-by: YiPeng Chai <YiPeng.Chai at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index dbfc41ddc3c7..7d48541b03d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -4306,8 +4306,27 @@ int amdgpu_ras_reset_gpu(struct amdgpu_device *adev)
                ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE1_RESET;
        }

-       if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0)
+       if (atomic_cmpxchg(&ras->in_recovery, 0, 1) == 0) {
+               struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);
+               int hive_ras_recovery = 0;
+
+               if (hive) {
+                       hive_ras_recovery = atomic_read(&hive->ras_recovery);
+                       amdgpu_put_xgmi_hive(hive);
+               }
+               /* In the case of multiple GPUs, after a GPU has started
+                * resetting all GPUs on hive, other GPUs do not need to
+                * trigger GPU reset again.
+                */
+               if (!hive_ras_recovery)
+                       amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work);
+               else
+                       atomic_set(&ras->in_recovery, 0);
+       } else {
+               flush_work(&ras->recovery_work);
                amdgpu_reset_domain_schedule(ras->adev->reset_domain, &ras->recovery_work);
+       }
+
        return 0;
 }

--
2.34.1



More information about the amd-gfx mailing list