msm: kgsl: Avoid dynamically allocating small command buffers

Most command buffers here are rather small (fewer than 256 words); it's
a waste of time to dynamically allocate memory for such a small buffer
when it could easily fit on the stack.

Conditionally using an on-stack command buffer when the size is small
enough eliminates the need for using a dynamically-allocated buffer most
of the time, reducing GPU command submission latency.

drivers/gpu/msm/adreno_ringbuffer.c

@@ -869,6 +869,7 @@ int adreno_ringbuffer_submitcmd(struct adreno_device *adreno_dev,
 	struct kgsl_memobj_node *ib;
 	unsigned int numibs = 0;
 	unsigned int *link;
+	unsigned int link_onstack[SZ_256] __aligned(sizeof(long));
 	unsigned int *cmds;
 	struct kgsl_context *context;
 	struct adreno_context *drawctxt;
@@ -996,10 +997,14 @@ int adreno_ringbuffer_submitcmd(struct adreno_device *adreno_dev,
 	if (gpudev->ccu_invalidate)
 		dwords += 4;
 
-	link = kvcalloc(dwords, sizeof(unsigned int), GFP_KERNEL);
-	if (!link) {
-		ret = -ENOMEM;
-		goto done;
+	if (dwords <= ARRAY_SIZE(link_onstack)) {
+		link = link_onstack;
+	} else {
+		link = kvcalloc(dwords, sizeof(unsigned int), GFP_KERNEL);
+		if (!link) {
+			ret = -ENOMEM;
+			goto done;
+		}
 	}
 
 	cmds = link;
@@ -1129,7 +1134,9 @@ int adreno_ringbuffer_submitcmd(struct adreno_device *adreno_dev,
 	trace_kgsl_issueibcmds(device, context->id, numibs, drawobj->timestamp,
 			drawobj->flags, ret, drawctxt->type);
 
-	kvfree(link);
+	if (link != link_onstack)
+		kvfree(link);
+
 	return ret;
 }