[PATCH] nvme/pci: Fix HMB size calculation

[PATCH] nvme/pci: Fix HMB size calculation

[Keith Busch]

It's possible the preferred HMB size may not be a multiple of the
chunk_size. This patch will set the chunk_size to the descriptor's actual
length so the last iteration doesn't cause the total size to exceed the
allocated HMB size.

Signed-off-by: Keith Busch <keith.busch at intel.com>
---
 drivers/nvme/host/pci.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index d321597..e22ffda 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1641,6 +1641,7 @@ static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred)
 		if (!bufs[i])
 			break;
 
+		chunk_size = len;
 		descs[i].addr = cpu_to_le64(dma_addr);
 		descs[i].size = cpu_to_le32(len / dev->ctrl.page_size);
 		i++;
-- 
2.5.5

[PATCH] nvme/pci: Fix HMB size calculation

[Keith Busch]

On Mon, Jul 17, 2017@07:45:39PM -0400, Keith Busch wrote:
> It's possible the preferred HMB size may not be a multiple of the
> chunk_size. This patch will set the chunk_size to the descriptor's actual
> length so the last iteration doesn't cause the total size to exceed the
> allocated HMB size.
> 
> Signed-off-by: Keith Busch <keith.busch at intel.com>

Reported-by: Dan Carpenter <dan.carpenter at oracle.com>

> ---
>  drivers/nvme/host/pci.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> index d321597..e22ffda 100644
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -1641,6 +1641,7 @@ static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred)
>  		if (!bufs[i])
>  			break;
>  
> +		chunk_size = len;
>  		descs[i].addr = cpu_to_le64(dma_addr);
>  		descs[i].size = cpu_to_le32(len / dev->ctrl.page_size);
>  		i++;
> -- 
> 2.5.5
> 

[PATCH] nvme/pci: Fix HMB size calculation

[Christoph Hellwig]

On Mon, Jul 17, 2017@07:45:39PM -0400, Keith Busch wrote:
> It's possible the preferred HMB size may not be a multiple of the
> chunk_size. This patch will set the chunk_size to the descriptor's actual
> length so the last iteration doesn't cause the total size to exceed the
> allocated HMB size.

This looks a little weird, and might do the wrong thing
if we hit the retry loop after reducing chunk_size.

I suspect we just need to change to

	for (size = 0; size < preferred; size += len) {

and move len to function scope.

[PATCH] nvme/pci: Fix HMB size calculation

[Keith Busch]

On Tue, Jul 18, 2017@07:39:54PM +0200, Christoph Hellwig wrote:
> On Mon, Jul 17, 2017@07:45:39PM -0400, Keith Busch wrote:
> > It's possible the preferred HMB size may not be a multiple of the
> > chunk_size. This patch will set the chunk_size to the descriptor's actual
> > length so the last iteration doesn't cause the total size to exceed the
> > allocated HMB size.
> 
> This looks a little weird, and might do the wrong thing
> if we hit the retry loop after reducing chunk_size.

It may look weird when you consider the retry, but I'm pretty sure it
isn't wrong. The only way you can get to the retry is if the for-loop
breaks before the last interation, which means chunk_size will be
unchanged. That's not immediately obvious, though, so maybe this
implementation was trying to be too clever.
 
> I suspect we just need to change to
> 
> 	for (size = 0; size < preferred; size += len) {
> 
> and move len to function scope.

That's fine too.