instmem/gk20a: exclusively acquire instobjs

Although I would not have expected this to happen, we seem to run into race conditions if instobjs are accessed concurrently. Use a global lock for safety.

Signed-off-by Alexandre Courbot <[email protected]>

---
 drm/nouveau/nvkm/subdev/instmem/gk20a.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

drm/nouveau/nvkm/subdev/instmem/gk20a.c +5 -7

index fc419bb8eab7..d015633b8edd 100644
@@ -92,6 +92,7 @@ struct gk20a_instmem {
 
 	/* protects vaddr_* and gk20a_instobj::vaddr* */
 	spinlock_t lock;
+	unsigned long flags;
 
 	/* CPU mappings LRU */
 	unsigned int vaddr_use;

@@ -188,12 +189,11 @@ gk20a_instobj_acquire(struct nvkm_memory *memory)
 	struct gk20a_instobj *node = gk20a_instobj(memory);
 	struct gk20a_instmem *imem = node->imem;
 	struct nvkm_ltc *ltc = imem->base.subdev.device->ltc;
-	const u64 size = nvkm_memory_size(memory);
-	unsigned long flags;
+	u64 size;
 
 	nvkm_ltc_flush(ltc);
 
-	spin_lock_irqsave(&imem->lock, flags);
+	spin_lock_irqsave(&imem->lock, imem->flags);
 
 	if (node->vaddr) {
 		/* remove us from the LRU list since we cannot be unmapped */

@@ -202,6 +202,8 @@ gk20a_instobj_acquire(struct nvkm_memory *memory)
 		goto out;
 	}
 
+	size = nvkm_memory_size(memory);
+
 	/* try to free some address space if we reached the limit */
 	gk20a_instmem_vaddr_gc(imem, size);
 

@@ -218,8 +220,6 @@ gk20a_instobj_acquire(struct nvkm_memory *memory)
 		   imem->vaddr_use, imem->vaddr_max);
 
 out:
-	spin_unlock_irqrestore(&imem->lock, flags);
-
 	return node->vaddr;
 }
 

@@ -229,14 +229,11 @@ gk20a_instobj_release(struct nvkm_memory *memory)
 	struct gk20a_instobj *node = gk20a_instobj(memory);
 	struct gk20a_instmem *imem = node->imem;
 	struct nvkm_ltc *ltc = imem->base.subdev.device->ltc;
-	unsigned long flags;
-

Quoted 3 lines · Alexandre Courbot

Although I would not have expected this to happen, we seem to run into race conditions if instobjs are accessed concurrently. Use a global lock for safety.

I wouldn't expect this to be an issue either.

Before merging such a large hammer of a fix, I'd strongly prefer to see at least a better justification for why this is happening rather than potentially papering over a larger issue.

Quoted 6 lines · Ben Skeggs

Quoted 3 lines · Alexandre Courbot

Although I would not have expected this to happen, we seem to run into race conditions if instobjs are accessed concurrently. Use a global lock for safety.

I wouldn't expect this to be an issue either.

Quoted 3 lines

Before merging such a large hammer of a fix, I'd strongly prefer to see at least a better justification for why this is happening rather than potentially papering over a larger issue.

I was afraid you would say that. ;)

But you're right. I am really busy with lots of stuff but will try to rootcause this more precisely...