After patch ddbd89deb7d3 ("swiotlb: fix info leak with DMA_FROM_DEVICE"),
swiotlb_bounce will be called in swiotlb_tbl_map_single unconditionally.
This requires that the physical address must be valid, which is not always
true on stable-4.19 or earlier version.
On stable-4.19, swiotlb_alloc_buffer will call swiotlb_tbl_map_single with
orig_addr equal to zero, which cause such a panic:

Unable to handle kernel paging request at virtual address ffffb77a40000000
...
pc : __memcpy+0x100/0x180
lr : swiotlb_bounce+0x74/0x88
...
Call trace:
 __memcpy+0x100/0x180
 swiotlb_tbl_map_single+0x2c8/0x338
 swiotlb_alloc+0xb4/0x198
 __dma_alloc+0x84/0x1d8
 ...

On stable-4.9 and stable-4.14, swiotlb_alloc_coherent wille call map_single
with orig_addr equal to zero, which can cause same panic.

Fix this by skipping swiotlb_bounce when orig_addr is zero.

Fixes: ddbd89deb7d3 ("swiotlb: fix info leak with DMA_FROM_DEVICE")
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
---
 kernel/dma/swiotlb.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
index 8b1360772fc5..b1e2ce2f9c2d 100644
--- a/kernel/dma/swiotlb.c
+++ b/kernel/dma/swiotlb.c
@@ -594,7 +594,8 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
 	 * unconditional bounce may prevent leaking swiotlb content (i.e.
 	 * kernel memory) to user-space.
 	 */
-	swiotlb_bounce(orig_addr, tlb_addr, size, DMA_TO_DEVICE);
+	if (orig_addr)
+		swiotlb_bounce(orig_addr, tlb_addr, size, DMA_TO_DEVICE);
 	return tlb_addr;
 }
 
-- 
2.25.1

On Thu, 2022-06-30 at 19:33 +0800, Liu Shixin wrote:
> After patch ddbd89deb7d3 ("swiotlb: fix info leak with DMA_FROM_DEVICE"),
> swiotlb_bounce will be called in swiotlb_tbl_map_single unconditionally.
> This requires that the physical address must be valid, which is not always
> true on stable-4.19 or earlier version.
> On stable-4.19, swiotlb_alloc_buffer will call swiotlb_tbl_map_single with
> orig_addr equal to zero, which cause such a panic:
> 
> Unable to handle kernel paging request at virtual address ffffb77a40000000
> ...
> pc : __memcpy+0x100/0x180
> lr : swiotlb_bounce+0x74/0x88
> ...
> Call trace:
>  __memcpy+0x100/0x180
>  swiotlb_tbl_map_single+0x2c8/0x338
>  swiotlb_alloc+0xb4/0x198
>  __dma_alloc+0x84/0x1d8
>  ...
> 
> On stable-4.9 and stable-4.14, swiotlb_alloc_coherent wille call map_single
> with orig_addr equal to zero, which can cause same panic.
> 
> Fix this by skipping swiotlb_bounce when orig_addr is zero.

Thanks for fixing this.  I tried to test the backports by forcing use
of swiotlb, but apparently it still didn't get used.

Ben.

> Fixes: ddbd89deb7d3 ("swiotlb: fix info leak with DMA_FROM_DEVICE")
> Signed-off-by: Liu Shixin <liushixin2@huawei.com>
> ---
>  kernel/dma/swiotlb.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> index 8b1360772fc5..b1e2ce2f9c2d 100644
> --- a/kernel/dma/swiotlb.c
> +++ b/kernel/dma/swiotlb.c
> @@ -594,7 +594,8 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev,
>  	 * unconditional bounce may prevent leaking swiotlb content (i.e.
>  	 * kernel memory) to user-space.
>  	 */
> -	swiotlb_bounce(orig_addr, tlb_addr, size, DMA_TO_DEVICE);
> +	if (orig_addr)
> +		swiotlb_bounce(orig_addr, tlb_addr, size, DMA_TO_DEVICE);
>  	return tlb_addr;
>  }
>  

-- 
Ben Hutchings
Nothing is ever a complete failure;
it can always serve as a bad example.