drm/amdgpu: delay ttm buffer func enablement on xgmi
authorPierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Mon, 15 Jun 2026 11:48:23 +0000 (13:48 +0200)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 19 Aug 2026 14:09:46 +0000 (10:09 -0400)
When amdgpu_init_minimal_xgmi is used, SDMA engines init
is delayed so amdgpu_ttm_enable_buffer_funcs must be
called later.

Without this, the check for num_buffer_funcs_scheds will
fail and using ttm buffer funcs later will fail.

Given that amdgpu_ttm_enable_buffer_funcs is a no-op if
amdgpu_in_reset() returns true, the call has to occur
after the reset lock is dropped.

Cc: stable@vger.kernel.org
Fixes: e4029f7a9474 ("drm/amdgpu: only use working sdma schedulers for ttm")
Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c

index ca385ed..b61c641 100644 (file)
@@ -2529,7 +2529,11 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
        if (r)
                goto init_failed;
 
-       amdgpu_ttm_enable_buffer_funcs(adev);
+       /* If SDMA is not brought up during hwini, the ttm buffer funcs enablement
+        * is delayed after reset-on-init completes.
+        */
+       if (amdgpu_ip_member_of_hwini(adev, AMD_IP_BLOCK_TYPE_SDMA))
+               amdgpu_ttm_enable_buffer_funcs(adev);
 
        /* Don't init kfd if whole hive need to be reset during init */
        if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
@@ -5174,8 +5178,6 @@ int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context)
                                if (r)
                                        goto out;
 
-                               amdgpu_ttm_enable_buffer_funcs(tmp_adev);
-
                                r = amdgpu_device_ip_resume_phase3(tmp_adev);
                                if (r)
                                        goto out;
index 2725230..45e31b3 100644 (file)
@@ -1380,6 +1380,9 @@ static void amdgpu_xgmi_reset_on_init_work(struct work_struct *work)
        amdgpu_device_unlock_reset_domain(tmp_adev->reset_domain);
 
        list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) {
+               /* Enable ttm buffers funcs after the reset lock has been dropped. */
+               amdgpu_ttm_enable_buffer_funcs(tmp_adev);
+
                r = amdgpu_ras_init_badpage_info(tmp_adev);
                if (r && r != -EHWPOISON)
                        dev_err(tmp_adev->dev,