[PATCH v2 1/4] drm/amdgpu: Set fatal errror detected flag earlier

[Kamal, Asad]

[AMD Official Use Only - General]

Series is
Reviewed-by: Asad Kamal <asad.kamal at amd.com>

Thanks & Regards
Asad


-----Original Message-----
From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Lijo Lazar
Sent: Thursday, March 28, 2024 8:06 AM
To: amd-gfx at lists.freedesktop.org
Cc: Zhang, Hawking <Hawking.Zhang at amd.com>; Deucher, Alexander <Alexander.Deucher at amd.com>; Wang, Yang(Kevin) <KevinYang.Wang at amd.com>
Subject: [PATCH v2 1/4] drm/amdgpu: Set fatal errror detected flag earlier

In case of fatal errors, set FED status when interrupt is received. Set the flag on other devices in the hive before RAS recovery work.

Signed-off-by: Lijo Lazar <lijo.lazar at amd.com>
---
v2: Avoid accessing hive in interrupt handler as it may take mutex path (Kevin)

 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 41 +++++++++++++++++--------
 1 file changed, 28 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index b8c7d0bf8fb1..352ce16a0963 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -2399,6 +2399,19 @@ static int amdgpu_ras_badpages_read(struct amdgpu_device *adev,
        return ret;
 }

+static void amdgpu_ras_set_fed_all(struct amdgpu_device *adev,
+                                  struct amdgpu_hive_info *hive, bool status) {
+       struct amdgpu_device *tmp_adev;
+
+       if (hive) {
+               list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head)
+                       amdgpu_ras_set_fed(tmp_adev, status);
+       } else {
+               amdgpu_ras_set_fed(adev, status);
+       }
+}
+
 static void amdgpu_ras_do_recovery(struct work_struct *work)  {
        struct amdgpu_ras *ras =
@@ -2408,8 +2421,21 @@ static void amdgpu_ras_do_recovery(struct work_struct *work)
        struct list_head device_list, *device_list_handle =  NULL;
        struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev);

-       if (hive)
+       if (hive) {
                atomic_set(&hive->ras_recovery, 1);
+
+               /* If any device which is part of the hive received RAS fatal
+                * error interrupt, set fatal error status on all. This
+                * condition will need a recovery, and flag will be cleared
+                * as part of recovery.
+                */
+               list_for_each_entry(remote_adev, &hive->device_list,
+                                   gmc.xgmi.head)
+                       if (amdgpu_ras_get_fed_status(remote_adev)) {
+                               amdgpu_ras_set_fed_all(adev, hive, true);
+                               break;
+                       }
+       }
        if (!ras->disable_ras_err_cnt_harvest) {

                /* Build list of devices to query RAS related errors */ @@ -2454,18 +2480,6 @@ static void amdgpu_ras_do_recovery(struct work_struct *work)
                                ras->gpu_reset_flags &= ~AMDGPU_RAS_GPU_RESET_MODE1_RESET;
                                set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);

-                               /* For any RAS error that needs a full reset to
-                                * recover, set the fatal error status
-                                */
-                               if (hive) {
-                                       list_for_each_entry(remote_adev,
-                                                           &hive->device_list,
-                                                           gmc.xgmi.head)
-                                               amdgpu_ras_set_fed(remote_adev,
-                                                                  true);
-                               } else {
-                                       amdgpu_ras_set_fed(adev, true);
-                               }
                                psp_fatal_error_recovery_quirk(&adev->psp);
                        }
                }
@@ -3550,6 +3564,7 @@ void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev)
                RAS_EVENT_LOG(adev, event_id, "uncorrectable hardware error"
                              "(ERREVENT_ATHUB_INTERRUPT) detected!\n");

+               amdgpu_ras_set_fed(adev, true);
                ras->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE1_RESET;
                amdgpu_ras_reset_gpu(adev);
        }
--
2.25.1