drm/amdgpu: move kfd post_reset out of reset_sriov function
authorshaoyunl <shaoyun.liu@amd.com>
Thu, 18 Nov 2021 16:43:16 +0000 (11:43 -0500)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 24 Nov 2021 19:06:52 +0000 (14:06 -0500)
Fixes: 9f4f2c1a3524 ("drm/amd/amdgpu: fix the kfd pre_reset sequence in sriov")

For sriov XGMI  configuration, the host driver will handle the hive reset,
so in guest side, the reset_sriov only be called once on one device. This will
make kfd post_reset unblanced with kfd pre_reset since kfd pre_reset already
been moved out of reset_sriov function. Move kfd post_reset out of reset_sriov
function to make them balance .

Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

index 5b95665..6acdf7c 100644 (file)
@@ -4312,7 +4312,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
 
        amdgpu_irq_gpu_reset_resume_helper(adev);
        r = amdgpu_ib_ring_tests(adev);
-       amdgpu_amdkfd_post_reset(adev);
 
 error:
        if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) {
@@ -5085,7 +5084,7 @@ retry:    /* Rest of adevs pre asic reset from XGMI hive. */
 
        tmp_vram_lost_counter = atomic_read(&((adev)->vram_lost_counter));
        /* Actual ASIC resets if needed.*/
-       /* TODO Implement XGMI hive reset logic for SRIOV */
+       /* Host driver will handle XGMI hive reset for SRIOV */
        if (amdgpu_sriov_vf(adev)) {
                r = amdgpu_device_reset_sriov(adev, job ? false : true);
                if (r)
@@ -5145,8 +5144,8 @@ skip_hw_reset:
 
 skip_sched_resume:
        list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
-               /* unlock kfd: SRIOV would do it separately */
-               if (!need_emergency_restart && !amdgpu_sriov_vf(tmp_adev))
+               /* unlock kfd */
+               if (!need_emergency_restart)
                        amdgpu_amdkfd_post_reset(tmp_adev);
 
                /* kfd_post_reset will do nothing if kfd device is not initialized,