[AMD Official Use Only]
This patch looks ok to me .
Patch 2 is actually add the PSP xgmi init not the whole XGMI init, can you change the description according to this ?
Patch 3, You take the hive lock inside the reset sriov function , but the hive lock already be took before this function is called in gpu_recovery function, so is it real necessary to get hive inside the reset sriov function , can you try remove the code to check hive ? Or maybe pass the hive as a parameter into this function if the hive is needed?
Patch 4 looks ok to me , but may need SRDC engineer confirm it won't have side effect on other AI asic .
Regards
Shaoyun.liu
-----Original Message-----
From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of Zhigang Luo
Sent: Tuesday, December 7, 2021 11:57 AM
To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Cc: Luo, Zhigang <Zhigang.Luo@xxxxxxx>
Subject: [PATCH 1/4] drm/amdgpu: skip reset other device in the same hive if it's SRIOV VF
On SRIOV, host driver can support FLR(function level reset) on individual VF within the hive which might bring the individual device back to normal without the necessary to execute the hive reset. If the FLR failed , host driver will trigger the hive reset, each guest VF will get reset notification before the real hive reset been executed. The VF device can handle the reset request individually in it's reset work handler.
This change updated gpu recover sequence to skip reset other device in the same hive for SRIOV VF.
Signed-off-by: Zhigang Luo <zhigang.luo@xxxxxxx>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 3c5afa45173c..474f8ea58aa5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4746,7 +4746,7 @@ static int amdgpu_device_lock_hive_adev(struct amdgpu_device *adev, struct amdgp {
struct amdgpu_device *tmp_adev = NULL;
- if (adev->gmc.xgmi.num_physical_nodes > 1) {
+ if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1))
+{
if (!hive) {
dev_err(adev->dev, "Hive is NULL while device has multiple xgmi nodes");
return -ENODEV;
@@ -4958,7 +4958,8 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
* We always reset all schedulers for device and all devices for XGMI
* hive so that should take care of them too.
*/
- hive = amdgpu_get_xgmi_hive(adev);
+ if (!amdgpu_sriov_vf(adev))
+ hive = amdgpu_get_xgmi_hive(adev);
if (hive) {
if (atomic_cmpxchg(&hive->in_reset, 0, 1) != 0) {
DRM_INFO("Bailing on TDR for s_job:%llx, hive: %llx as another already in progress", @@ -4999,7 +5000,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
* to put adev in the 1st position.
*/
INIT_LIST_HEAD(&device_list);
- if (adev->gmc.xgmi.num_physical_nodes > 1) {
+ if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1))
+{
list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head)
list_add_tail(&tmp_adev->reset_list, &device_list);
if (!list_is_first(&adev->reset_list, &device_list))
--
2.17.1
