[1/2] drm/msm: gpu: don't abuse dma_alloc for non-DMA allocations

Submitted by Arnd Bergmann on June 20, 2017, 8:16 p.m.

Details

Message ID 20170620201720.225593-1-arnd@arndb.de
State New
Headers show
Series "Series without cover letter" ( rev: 1 ) in DRI devel

Not browsing as part of any series.

Commit Message

Arnd Bergmann June 20, 2017, 8:16 p.m.
In zap_shader_load_mdt(), we pass a pointer to a phys_addr_t
into dmam_alloc_coherent, which the compiler warns about:

drivers/gpu/drm/msm/adreno/a5xx_gpu.c: In function 'zap_shader_load_mdt':
drivers/gpu/drm/msm/adreno/a5xx_gpu.c:54:50: error: passing argument 3 of 'dmam_alloc_coherent' from incompatible pointer type [-Werror=incompatible-pointer-types]

The returned DMA address is later passed on to a function that
takes a phys_addr_t, so it's clearly wrong to use the DMA
mapping interface here: the memory may be uncached, or the
address may be completely wrong if there is an IOMMU connected
to the device.

My interpretation is that using dmam_alloc_coherent() had two
purposes:

 a) get a chunk of consecutive memory that may be larger than
    the limit for kmalloc()

 b) use the devres infrastructure to simplify the unwinding
    in the error case.

I think ideally we'd use a devres-based version of
alloc_pages_exact() here, but since that doesn't exist,
let's use devm_get_free_pages() instead. This wastes a little
memory as the size gets rounded up to a power of two, but
is otherwise harmless. If we want to save memory here, calling
devm_free_pages() to release the memory once it is no longer
needed is probably better anyway.

Fixes: 7c65817e6d38 ("drm/msm: gpu: Enable zap shader for A5XX")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
index b4b54f1c24bc..eee9ac81aaa1 100644
--- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
@@ -51,11 +51,13 @@  static int zap_shader_load_mdt(struct device *dev, const char *fwname)
 	}
 
 	/* Allocate memory for the firmware image */
-	mem_region = dmam_alloc_coherent(dev, mem_size, &mem_phys, GFP_KERNEL);
+	mem_region = (void *)devm_get_free_pages(dev, GFP_KERNEL,
+			get_order(mem_size));
 	if (!mem_region) {
 		ret = -ENOMEM;
 		goto out;
 	}
+	mem_phys = virt_to_phys(mem_region);
 
 	/* Load the rest of the MDT */
 	ret = qcom_mdt_load(dev, fw, fwname, GPU_PAS_ID, mem_region, mem_phys,

Comments

On Tue, Jun 20, 2017 at 10:16:50PM +0200, Arnd Bergmann wrote:
> In zap_shader_load_mdt(), we pass a pointer to a phys_addr_t
> into dmam_alloc_coherent, which the compiler warns about:
> 
> drivers/gpu/drm/msm/adreno/a5xx_gpu.c: In function 'zap_shader_load_mdt':
> drivers/gpu/drm/msm/adreno/a5xx_gpu.c:54:50: error: passing argument 3 of 'dmam_alloc_coherent' from incompatible pointer type [-Werror=incompatible-pointer-types]
> 
> The returned DMA address is later passed on to a function that
> takes a phys_addr_t, so it's clearly wrong to use the DMA
> mapping interface here: the memory may be uncached, or the
> address may be completely wrong if there is an IOMMU connected
> to the device.
> 
> My interpretation is that using dmam_alloc_coherent() had two
> purposes:
> 
>  a) get a chunk of consecutive memory that may be larger than
>     the limit for kmalloc()
> 
>  b) use the devres infrastructure to simplify the unwinding
>     in the error case.
> 
> I think ideally we'd use a devres-based version of
> alloc_pages_exact() here, but since that doesn't exist,
> let's use devm_get_free_pages() instead. This wastes a little
> memory as the size gets rounded up to a power of two, but
> is otherwise harmless. If we want to save memory here, calling
> devm_free_pages() to release the memory once it is no longer
> needed is probably better anyway.
> 
> Fixes: 7c65817e6d38 ("drm/msm: gpu: Enable zap shader for A5XX")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Seems reasonable to me if Bjorn agrees.

Acked-by: Jordan Crouse <jcrouse@codeaurora.org>

> ---
>  drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> index b4b54f1c24bc..eee9ac81aaa1 100644
> --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> @@ -51,11 +51,13 @@ static int zap_shader_load_mdt(struct device *dev, const char *fwname)
>  	}
>  
>  	/* Allocate memory for the firmware image */
> -	mem_region = dmam_alloc_coherent(dev, mem_size, &mem_phys, GFP_KERNEL);
> +	mem_region = (void *)devm_get_free_pages(dev, GFP_KERNEL,
> +			get_order(mem_size));
>  	if (!mem_region) {
>  		ret = -ENOMEM;
>  		goto out;
>  	}
> +	mem_phys = virt_to_phys(mem_region);
>  
>  	/* Load the rest of the MDT */
>  	ret = qcom_mdt_load(dev, fw, fwname, GPU_PAS_ID, mem_region, mem_phys,
> -- 
> 2.9.0
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
On Tue 20 Jun 13:16 PDT 2017, Arnd Bergmann wrote:

> In zap_shader_load_mdt(), we pass a pointer to a phys_addr_t
> into dmam_alloc_coherent, which the compiler warns about:
> 
> drivers/gpu/drm/msm/adreno/a5xx_gpu.c: In function 'zap_shader_load_mdt':
> drivers/gpu/drm/msm/adreno/a5xx_gpu.c:54:50: error: passing argument 3 of 'dmam_alloc_coherent' from incompatible pointer type [-Werror=incompatible-pointer-types]
> 
> The returned DMA address is later passed on to a function that
> takes a phys_addr_t, so it's clearly wrong to use the DMA
> mapping interface here: the memory may be uncached, or the
> address may be completely wrong if there is an IOMMU connected
> to the device.
> 
> My interpretation is that using dmam_alloc_coherent() had two
> purposes:
> 
>  a) get a chunk of consecutive memory that may be larger than
>     the limit for kmalloc()
> 
>  b) use the devres infrastructure to simplify the unwinding
>     in the error case.
> 
> I think ideally we'd use a devres-based version of
> alloc_pages_exact() here, but since that doesn't exist,
> let's use devm_get_free_pages() instead. This wastes a little
> memory as the size gets rounded up to a power of two, but
> is otherwise harmless. If we want to save memory here, calling
> devm_free_pages() to release the memory once it is no longer
> needed is probably better anyway.
> 
> Fixes: 7c65817e6d38 ("drm/msm: gpu: Enable zap shader for A5XX")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Acked-by: Bjorn Andersson <bjorn.andersson@linaro.org>

Regards,
Bjorn

> ---
>  drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> index b4b54f1c24bc..eee9ac81aaa1 100644
> --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> @@ -51,11 +51,13 @@ static int zap_shader_load_mdt(struct device *dev, const char *fwname)
>  	}
>  
>  	/* Allocate memory for the firmware image */
> -	mem_region = dmam_alloc_coherent(dev, mem_size, &mem_phys, GFP_KERNEL);
> +	mem_region = (void *)devm_get_free_pages(dev, GFP_KERNEL,
> +			get_order(mem_size));
>  	if (!mem_region) {
>  		ret = -ENOMEM;
>  		goto out;
>  	}
> +	mem_phys = virt_to_phys(mem_region);
>  
>  	/* Load the rest of the MDT */
>  	ret = qcom_mdt_load(dev, fw, fwname, GPU_PAS_ID, mem_region, mem_phys,
> -- 
> 2.9.0
>