Merge tag 'amd-drm-next-6.11-2024-06-22' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-6.11-2024-06-22:

amdgpu:
- HPD fixes
- PSR fixes
- DCC updates
- DCN 4.0.1 fixes
- FAMS fixes
- Misc code cleanups
- SR-IOV fixes
- GPUVM TLB flush cleanups
- Make VCN less verbose
- ACPI backlight fixes
- MES fixes
- Firmware loading cleanups
- Replay fixes
- LTTPR fixes
- Trap handler fixes
- Cursor and overlay fixes
- Primary plane zpos fixes
- DML 2.1 fixes
- RAS updates
- USB4 fixes
- MALL fixes
- Reserved VMID fix
- Silence UBSAN warnings

amdkfd:
- Misc code cleanups

From: Alex Deucher <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
Signed-off-by: Dave Airlie <[email protected]>

1 files changed, 27 insertions, 9 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
index 3d7fcdeaf8cf..b5b9d4f40f53 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c
@@ -290,18 +290,36 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm,
 	     !dma_fence_is_signaled((*id)->last_flush))) {
 		struct dma_fence *tmp;
 
-		/* Don't use per engine and per process VMID at the same time */
-		if (adev->vm_manager.concurrent_flush)
-			ring = NULL;
-
-		/* to prevent one context starved by another context */
-		(*id)->pd_gpu_addr = 0;
-		tmp = amdgpu_sync_peek_fence(&(*id)->active, ring);
-		if (tmp) {
+		/* Wait for the gang to be assembled before using a
+		 * reserved VMID or otherwise the gang could deadlock.
+		 */
+		tmp = amdgpu_device_get_gang(adev);
+		if (!dma_fence_is_signaled(tmp) && tmp != job->gang_submit) {
 			*id = NULL;
-			*fence = dma_fence_get(tmp);
+			*fence = tmp;
 			return 0;
 		}
+		dma_fence_put(tmp);
+
+		/* Make sure the id is owned by the gang before proceeding */
+		if (!job->gang_submit ||
+		    (*id)->owner != vm->immediate.fence_context) {
+
+			/* Don't use per engine and per process VMID at the
+			 * same time
+			 */
+			if (adev->vm_manager.concurrent_flush)
+				ring = NULL;
+
+			/* to prevent one context starved by another context */
+			(*id)->pd_gpu_addr = 0;
+			tmp = amdgpu_sync_peek_fence(&(*id)->active, ring);
+			if (tmp) {
+				*id = NULL;
+				*fence = dma_fence_get(tmp);
+				return 0;
+			}
+		}
 		needs_flush = true;
 	}