Re: [PATCH] gpu: host1x: Fix compiler errors

[Thierry Reding]

On Mon, Mar 26, 2018 at 04:44:14PM +0200, Emil Goode wrote:
> The compiler is complaining with the following errors:
> 
> drivers/gpu/host1x/cdma.c:94:48: error:
> 	passing argument 3 of âdma_alloc_wcâ from incompatible pointer type
> 	[-Werror=incompatible-pointer-types]
> 
> drivers/gpu/host1x/cdma.c:113:48: error:
> 	passing argument 3 of âdma_alloc_wcâ from incompatible pointer type
> 	[-Werror=incompatible-pointer-types]
> 
> The expected pointer type of the third argument to dma_alloc_wc() is
> dma_addr_t but phys_addr_t is passed. Fix this by adding casts to the
> expected pointer type.
> 
> Signed-off-by: Emil Goode <emil.fsw@xxxxxxxx>
> ---
>  drivers/gpu/host1x/cdma.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/host1x/cdma.c b/drivers/gpu/host1x/cdma.c
> index 28541b280739..5e8b321a751e 100644
> --- a/drivers/gpu/host1x/cdma.c
> +++ b/drivers/gpu/host1x/cdma.c
> @@ -91,8 +91,8 @@ static int host1x_pushbuffer_init(struct push_buffer *pb)
>  
>  		size = iova_align(&host1x->iova, size);
>  
> -		pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
> -					  GFP_KERNEL);
> +		pb->mapped = dma_alloc_wc(host1x->dev, size,
> +					  (dma_addr_t *)&pb->phys, GFP_KERNEL);
>  		if (!pb->mapped)
>  			return -ENOMEM;
>  
> @@ -110,8 +110,8 @@ static int host1x_pushbuffer_init(struct push_buffer *pb)
>  		if (err)
>  			goto iommu_free_iova;
>  	} else {
> -		pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
> -					  GFP_KERNEL);
> +		pb->mapped = dma_alloc_wc(host1x->dev, size,
> +					  (dma_addr_t *)&pb->phys, GFP_KERNEL);
>  		if (!pb->mapped)
>  			return -ENOMEM;
>  

This doesn't seem right. There's no guarantee that phys_addr_t and
dma_addr_t will be compatible, so the above isn't always correct. Also,
I don't see a need for pb->phys to ever be phys_addr_t. It's allocated
through dma_alloc_wc() exclusively, so it should just be dma_addr_t.

Note that the !pb->phys check in host1x_pushbuffer_destroy() becomes
technically wrong if pb->phys is dma_addr_t (0 is a perfectly valid
value for dma_addr_t), so make sure to flip that to !pb->mapped instead.
pb->mapped and pb->phys are always set in tandem, and checking mapped
for non-NULL is the right check to test whether the pair is valid or
not.

Thierry

Attachment: signature.asc
Description: PGP signature