Re: [PATCH] iommu/dma: Fix IOVA reserve dma ranges

[Robin Murphy]

On 2020-09-09 06:32, Srinath Mannam wrote:
> Fix IOVA reserve failure for memory regions listed in dma-ranges in the
> following cases.
>
> - start address of memory region is 0x0.

That's fair enough, and in fact generalises to the case of zero-sized 
gaps between regions, which is indeed an oversight.

> - end address of a memory region is equal to start address of next memory
>    region.

This part doesn't make much sense, however - if the regions described in 
bridge->dma_ranges overlap, that's a bug in whoever created a malformed 
list to begin with. Possibly it's just poor wording, and you're using 
"memory regions" to refer to any or all of the dma_ranges, the reserved 
IOVA ranges, and what "start" and "end" in this function represent which 
isn't quite either of those.

> Fixes: aadad097cd46f ("iommu/dma: Reserve IOVA for PCIe inaccessible DMA address")
> Signed-off-by: Srinath Mannam <srinath.mannam@xxxxxxxxxxxx>
> ---
>   drivers/iommu/dma-iommu.c | 15 +++++++++++----
>   1 file changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 5141d49a046b..0a3f67a4f9ae 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -213,14 +213,21 @@ static int iova_reserve_pci_windows(struct pci_dev *dev,
>   	resource_list_for_each_entry(window, &bridge->dma_ranges) {
>   		end = window->res->start - window->offset;
>   resv_iova:
> +		if (end < start) {
> +			/* dma_ranges list should be sorted */
> +			dev_err(&dev->dev, "Failed to reserve IOVA\n");
> +			return -EINVAL;
> +		}
> +		/*
> +		 * Skip the cases when start address of first memory region is
> +		 * 0x0 and end address of one memory region and start address
> +		 * of next memory region are equal. Reserve IOVA for rest of
> +		 * addresses fall in between given memory ranges.
> +		 */
>   		if (end > start) {
>   			lo = iova_pfn(iovad, start);
>   			hi = iova_pfn(iovad, end);
>   			reserve_iova(iovad, lo, hi);
> -		} else {

Surely this only needs to be a one-liner?

-		} else {
+		} else if (end < start) {

(or possibly "end != start"; I can't quite decide which expresses the 
semantic intent better)

The rest just looks like unnecessary churn - I don't think it needs 
commenting that a sorted list may simply not have gaps between entries, 
and as above I think the wording of that comment is actively misleading.

Robin.

> -			/* dma_ranges list should be sorted */
> -			dev_err(&dev->dev, "Failed to reserve IOVA\n");
> -			return -EINVAL;
>   		}
>   
>   		start = window->res->end - window->offset + 1;