On Tue, Jun 28, 2022 at 2:18 AM Ivan Malov <ivan.malov@xxxxxxxxxxxx> wrote:
>
> When a XSK pool gets mapped, xp_check_dma_contiguity() adds bit 0x1
> to pages' DMA addresses that go in ascending order and at 4K stride.
> The problem is that the bit does not get cleared before doing unmap.
> As a result, a lot of warnings from iommu_dma_unmap_page() are seen
> suggesting mapping lookup failures at drivers/iommu/dma-iommu.c:848.
Thanks Ivan for spotting this. Please add if this patch is for bpf or
net in your subject line. E.g., [PATCH net].
Also, I cannot find a warning at drivers/iommu/dma-iommu.c:848. For
net and bpf I have a WARN() at line 679 in __iommu_dma_unmap(). Maybe
it would be better to just refer to __iommu_dma_unmap() and the
warning in that function. Line numbers tend to change.
> Fixes: 2b43470add8c ("xsk: Introduce AF_XDP buffer allocation API")
> Signed-off-by: Ivan Malov <ivan.malov@xxxxxxxxxxxx>
> ---
> v1 -> v2: minor adjustments to dispose of the "Fixes:" tag warning
>
> net/xdp/xsk_buff_pool.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
> index 87bdd71c7bb6..f70112176b7c 100644
> --- a/net/xdp/xsk_buff_pool.c
> +++ b/net/xdp/xsk_buff_pool.c
> @@ -332,6 +332,7 @@ static void __xp_dma_unmap(struct xsk_dma_map *dma_map, unsigned long attrs)
> for (i = 0; i < dma_map->dma_pages_cnt; i++) {
> dma = &dma_map->dma_pages[i];
> if (*dma) {
> + *dma &= ~XSK_NEXT_PG_CONTIG_MASK;
> dma_unmap_page_attrs(dma_map->dev, *dma, PAGE_SIZE,
> DMA_BIDIRECTIONAL, attrs);
> *dma = 0;
> --
> 2.30.2
>
