drm/msm: Add VM logging for VM_BIND updates

When userspace opts in to VM_BIND, the submit no longer holds references
keeping the VMA alive.  This makes it difficult to distinguish between
UMD/KMD/app bugs.  So add a debug option for logging the most recent VM
updates and capturing these in GPU devcoredumps.

The submitqueue id is also captured, a value of zero means the operation
did not go via a submitqueue (ie. comes from msm_gem_vm_close() tearing
down the remaining mappings when the device file is closed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
Tested-by: Antonino Maniscalco <antomani103@gmail.com>
Reviewed-by: Antonino Maniscalco <antomani103@gmail.com>
Patchwork: https://patchwork.freedesktop.org/patch/661518/

1 files changed, 47 insertions, 5 deletions

diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c
index ccd9ebfc5c7c..c317b25a8162 100644
--- a/drivers/gpu/drm/msm/msm_gpu.c
+++ b/drivers/gpu/drm/msm/msm_gpu.c
@@ -259,9 +259,6 @@ static void crashstate_get_bos(struct msm_gpu_state *state, struct msm_gem_submi
 {
 	extern bool rd_full;
 
-	if (!submit)
-		return;
-
 	if (msm_context_is_vmbind(submit->queue->ctx)) {
 		struct drm_exec exec;
 		struct drm_gpuva *vma;
@@ -318,6 +315,48 @@ static void crashstate_get_bos(struct msm_gpu_state *state, struct msm_gem_submi
 	}
 }
 
+static void crashstate_get_vm_logs(struct msm_gpu_state *state, struct msm_gem_vm *vm)
+{
+	uint32_t vm_log_len = (1 << vm->log_shift);
+	uint32_t vm_log_mask = vm_log_len - 1;
+	int first;
+
+	/* Bail if no log, or empty log: */
+	if (!vm->log || !vm->log[0].op)
+		return;
+
+	mutex_lock(&vm->mmu_lock);
+
+	/*
+	 * log_idx is the next entry to overwrite, meaning it is the oldest, or
+	 * first, entry (other than the special case handled below where the
+	 * log hasn't wrapped around yet)
+	 */
+	first = vm->log_idx;
+
+	if (!vm->log[first].op) {
+		/*
+		 * If the next log entry has not been written yet, then only
+		 * entries 0 to idx-1 are valid (ie. we haven't wrapped around
+		 * yet)
+		 */
+		state->nr_vm_logs = MAX(0, first - 1);
+		first = 0;
+	} else {
+		state->nr_vm_logs = vm_log_len;
+	}
+
+	state->vm_logs = kmalloc_array(
+		state->nr_vm_logs, sizeof(vm->log[0]), GFP_KERNEL);
+	for (int i = 0; i < state->nr_vm_logs; i++) {
+		int idx = (i + first) & vm_log_mask;
+
+		state->vm_logs[i] = vm->log[idx];
+	}
+
+	mutex_unlock(&vm->mmu_lock);
+}
+
 static void msm_gpu_crashstate_capture(struct msm_gpu *gpu,
 		struct msm_gem_submit *submit, struct msm_gpu_fault_info *fault_info,
 		char *comm, char *cmd)
@@ -351,7 +390,10 @@ static void msm_gpu_crashstate_capture(struct msm_gpu *gpu,
 		msm_iommu_pagetable_walk(mmu, info->iova, info->ptes);
 	}
 
-	crashstate_get_bos(state, submit);
+	if (submit) {
+		crashstate_get_vm_logs(state, to_msm_vm(submit->vm));
+		crashstate_get_bos(state, submit);
+	}
 
 	/* Set the active crash state to be dumped on failure */
 	gpu->crashstate = state;
@@ -452,7 +494,7 @@ static void recover_worker(struct kthread_work *work)
 		 * VM_BIND)
 		 */
 		if (!vm->managed)
-			vm->unusable = true;
+			msm_gem_vm_unusable(submit->vm);
 	}
 
 	get_comm_cmdline(submit, &comm, &cmd);