habanalabs: handle MMU cache invalidation timeout

MMU cache invalidation timeout indicates that the device is unstable and
therefore unusable.
Hence in such case do hard reset and return an error to the user if was
called from ioctl.
In addition, change the print to error level and rephrase its text.

Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>

1 files changed, 28 insertions, 7 deletions

diff --git a/drivers/misc/habanalabs/memory.c b/drivers/misc/habanalabs/memory.c
index a72f766ca470..4b8eed1ca513 100644
--- a/drivers/misc/habanalabs/memory.c
+++ b/drivers/misc/habanalabs/memory.c
@@ -886,6 +886,7 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 
 		vm_type = (enum vm_type_t *) userptr;
 		hint_addr = args->map_host.hint_addr;
+		handle = phys_pg_pack->handle;
 	} else {
 		handle = lower_32_bits(args->map_device.handle);
 
@@ -954,10 +955,17 @@ static int map_device_va(struct hl_ctx *ctx, struct hl_mem_in *args,
 		goto map_err;
 	}
 
-	hdev->asic_funcs->mmu_invalidate_cache(hdev, false, *vm_type);
+	rc = hdev->asic_funcs->mmu_invalidate_cache(hdev, false, *vm_type);
 
 	mutex_unlock(&ctx->mmu_lock);
 
+	if (rc) {
+		dev_err(hdev->dev,
+			"mapping handle %u failed due to MMU cache invalidation\n",
+			handle);
+		goto map_err;
+	}
+
 	ret_vaddr += phys_pg_pack->offset;
 
 	hnode->ptr = vm_type;
@@ -1083,21 +1091,34 @@ static int unmap_device_va(struct hl_ctx *ctx, u64 vaddr, bool ctx_free)
 	 * at the loop end rather than for each iteration
 	 */
 	if (!ctx_free)
-		hdev->asic_funcs->mmu_invalidate_cache(hdev, true, *vm_type);
+		rc = hdev->asic_funcs->mmu_invalidate_cache(hdev, true,
+								*vm_type);
 
 	mutex_unlock(&ctx->mmu_lock);
 
 	/*
-	 * No point in maintaining the free VA block list if the context is
-	 * closing as the list will be freed anyway
+	 * If the context is closing we don't need to check for the MMU cache
+	 * invalidation return code and update the VA free list as in this flow
+	 * we invalidate the MMU cache outside of this unmap function and the VA
+	 * free list will be freed anyway.
 	 */
 	if (!ctx_free) {
-		rc = add_va_block(hdev, va_range, vaddr,
-					vaddr + phys_pg_pack->total_size - 1);
+		int tmp_rc;
+
 		if (rc)
+			dev_err(hdev->dev,
+				"unmapping vaddr 0x%llx failed due to MMU cache invalidation\n",
+				vaddr);
+
+		tmp_rc = add_va_block(hdev, va_range, vaddr,
+					vaddr + phys_pg_pack->total_size - 1);
+		if (tmp_rc) {
 			dev_warn(hdev->dev,
 					"add va block failed for vaddr: 0x%llx\n",
 					vaddr);
+			if (!rc)
+				rc = tmp_rc;
+		}
 	}
 
 	atomic_dec(&phys_pg_pack->mapping_cnt);
@@ -1108,7 +1129,7 @@ static int unmap_device_va(struct hl_ctx *ctx, u64 vaddr, bool ctx_free)
 		dma_unmap_host_va(hdev, userptr);
 	}
 
-	return 0;
+	return rc;
 
 mapping_cnt_err:
 	if (is_userptr)