Alexandre Courbot
2015-Oct-26  05:54 UTC
[Nouveau] [PATCH] 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 <acourbot at nvidia.com>
---
 drm/nouveau/nvkm/subdev/instmem/gk20a.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)
diff --git a/drm/nouveau/nvkm/subdev/instmem/gk20a.c
b/drm/nouveau/nvkm/subdev/instmem/gk20a.c
index fc419bb8eab7..d015633b8edd 100644
--- a/drm/nouveau/nvkm/subdev/instmem/gk20a.c
+++ b/drm/nouveau/nvkm/subdev/instmem/gk20a.c
@@ -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;
-
-	spin_lock_irqsave(&imem->lock, flags);
 
 	/* add ourselves to the LRU list so our CPU mapping can be freed */
 	list_add_tail(&node->vaddr_node, &imem->vaddr_lru);
 
-	spin_unlock_irqrestore(&imem->lock, flags);
+	spin_unlock_irqrestore(&imem->lock, imem->flags);
 
 	wmb();
 	nvkm_ltc_invalidate(ltc);
-- 
2.6.1
Ben Skeggs
2015-Nov-04  21:19 UTC
[Nouveau] [PATCH] instmem/gk20a: exclusively acquire instobjs
On 10/26/2015 03:54 PM, Alexandre Courbot wrote:> 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. Ben.> > Signed-off-by: Alexandre Courbot <acourbot at nvidia.com> > --- > drm/nouveau/nvkm/subdev/instmem/gk20a.c | 15 ++++++--------- > 1 file changed, 6 insertions(+), 9 deletions(-) > > diff --git a/drm/nouveau/nvkm/subdev/instmem/gk20a.c b/drm/nouveau/nvkm/subdev/instmem/gk20a.c > index fc419bb8eab7..d015633b8edd 100644 > --- a/drm/nouveau/nvkm/subdev/instmem/gk20a.c > +++ b/drm/nouveau/nvkm/subdev/instmem/gk20a.c > @@ -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; > - > - spin_lock_irqsave(&imem->lock, flags); > > /* add ourselves to the LRU list so our CPU mapping can be freed */ > list_add_tail(&node->vaddr_node, &imem->vaddr_lru); > > - spin_unlock_irqrestore(&imem->lock, flags); > + spin_unlock_irqrestore(&imem->lock, imem->flags); > > wmb(); > nvkm_ltc_invalidate(ltc); >-------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 819 bytes Desc: OpenPGP digital signature URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20151105/4f513a82/attachment.sig>
Alexandre Courbot
2015-Nov-05  06:44 UTC
[Nouveau] [PATCH] instmem/gk20a: exclusively acquire instobjs
On Thu, Nov 5, 2015 at 6:19 AM, Ben Skeggs <skeggsb at gmail.com> wrote:> On 10/26/2015 03:54 PM, Alexandre Courbot wrote: >> 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.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...