mlx4: Fix chunk sg list overflow in mlx4_alloc_icm()

If the number of sg entries in the ICM chunk reaches MLX4_ICM_CHUNK_LEN,
we must set chunk = NULL __even__ for coherent mappings (as is done for mthca). Otherwise we may overflow the sg list.

Signed-off-by Sebastien Dugue <sebastien.dugue-6ktuUTfB/[email protected]>

---
 drivers/net/mlx4/icm.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

drivers/net/mlx4/icm.c +2 -1

index 57288ca..ef62f17 100644
@@ -175,9 +175,10 @@ struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages,
 
 				if (chunk->nsg <= 0)
 					goto fail;
+			}
 
+			if (chunk->npages == MLX4_ICM_CHUNK_LEN)
 				chunk = NULL;
-			}
 
 			npages -= 1 << cur_order;
 		} else {

Signature · 6 lines

1.6.0.4

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

good catch, applied. Did this hit you in practice? I guess it would take a big coherent ICM allocation, were you getting those?

Also what do you think of this independent cleanup on top of your patch? It handles the error case for allocation and then avoids having the common case inside a deeper nested block:

 drivers/net/mlx4/icm.c |   37 +++++++++++++++++++------------------
 1 files changed, 19 insertions(+), 18 deletions(-)

drivers/net/mlx4/icm.c +15 -19

index ef62f17..b07e4de 100644
@@ -163,29 +163,30 @@ struct mlx4_icm *mlx4_alloc_icm(struct mlx4_dev *dev, int npages,
 			ret = mlx4_alloc_icm_pages(&chunk->mem[chunk->npages],
 						   cur_order, gfp_mask);
 
-		if (!ret) {
-			++chunk->npages;
-
-			if (coherent)
-				++chunk->nsg;
-			else if (chunk->npages == MLX4_ICM_CHUNK_LEN) {
-				chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
-							chunk->npages,
-							PCI_DMA_BIDIRECTIONAL);
-
-				if (chunk->nsg <= 0)
-					goto fail;
-			}
+		if (ret) {
+			if (--cur_order < 0)
+				goto fail;
+			else
+				continue;
+		}
 
-			if (chunk->npages == MLX4_ICM_CHUNK_LEN)
-				chunk = NULL;
+		++chunk->npages;
 
-			npages -= 1 << cur_order;
-		} else {
-			--cur_order;
-			if (cur_order < 0)
+		if (coherent)
+			++chunk->nsg;
+		else if (chunk->npages == MLX4_ICM_CHUNK_LEN) {
+			chunk->nsg = pci_map_sg(dev->pdev, chunk->mem,
+						chunk->npages,
+						PCI_DMA_BIDIRECTIONAL);
+
+			if (chunk->nsg <= 0)

Yes, some customer got hit by this, which ended up corrupting memory.