[PATCH 2/3] gpu: drm: dead code elimination

[PATCH 2/3] gpu: drm: dead code elimination

[Pan Zhang]

this set adds support for removal of gpu drm dead code.

patch2:
`num_entries` is a data of unsigned type(compilers always treat as unsigned int) and SIZE_MAX == ~0,

so there is a impossible condition:
'(num_entries > ((~0) - 56) / 72) => (max(0-u32) > 256204778801521549)'.

Signed-off-by: Pan Zhang <zhangpan26@huawei.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
index 85b0515..10a7f30 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
@@ -71,10 +71,6 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
 	unsigned i;
 	int r;
 
-	if (num_entries > (SIZE_MAX - sizeof(struct amdgpu_bo_list))
-				/ sizeof(struct amdgpu_bo_list_entry))
-		return -EINVAL;
-
 	size = sizeof(struct amdgpu_bo_list);
 	size += num_entries * sizeof(struct amdgpu_bo_list_entry);
 	list = kvmalloc(size, GFP_KERNEL);
-- 
2.7.4

Re: [PATCH 2/3] gpu: drm: dead code elimination

[Christian König]

Am 18.12.19 um 04:50 schrieb Pan Zhang:
> this set adds support for removal of gpu drm dead code.
>
> patch2:
> `num_entries` is a data of unsigned type(compilers always treat as unsigned int) and SIZE_MAX == ~0,
>
> so there is a impossible condition:
> '(num_entries > ((~0) - 56) / 72) => (max(0-u32) > 256204778801521549)'.

NAK, that calculation is not correct on 32-bit systems.

Christian.

>
> Signed-off-by: Pan Zhang <zhangpan26@huawei.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 4 ----
>   1 file changed, 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> index 85b0515..10a7f30 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> @@ -71,10 +71,6 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
>   	unsigned i;
>   	int r;
>   
> -	if (num_entries > (SIZE_MAX - sizeof(struct amdgpu_bo_list))
> -				/ sizeof(struct amdgpu_bo_list_entry))
> -		return -EINVAL;
> -
>   	size = sizeof(struct amdgpu_bo_list);
>   	size += num_entries * sizeof(struct amdgpu_bo_list_entry);
>   	list = kvmalloc(size, GFP_KERNEL);

Re: [PATCH 2/3] gpu: drm: dead code elimination

[Michel Dänzer]

On 2019-12-18 4:50 a.m., Pan Zhang wrote:
> this set adds support for removal of gpu drm dead code.
> 
> patch2:
> `num_entries` is a data of unsigned type(compilers always treat as unsigned int) and SIZE_MAX == ~0,
> 
> so there is a impossible condition:
> '(num_entries > ((~0) - 56) / 72) => (max(0-u32) > 256204778801521549)'.

While this looks correct for 64-bit, where size_t is unsigned long, on
32-bit it's unsigned int, in which case this isn't dead code.


> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> index 85b0515..10a7f30 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> @@ -71,10 +71,6 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
>  	unsigned i;
>  	int r;
>  
> -	if (num_entries > (SIZE_MAX - sizeof(struct amdgpu_bo_list))
> -				/ sizeof(struct amdgpu_bo_list_entry))
> -		return -EINVAL;
> -
>  	size = sizeof(struct amdgpu_bo_list);
>  	size += num_entries * sizeof(struct amdgpu_bo_list_entry);
>  	list = kvmalloc(size, GFP_KERNEL);
> 


-- 
Earthling Michel Dänzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer

Re: Re: [PATCH 2/3] gpu: drm: dead code elimination

[Pan Zhang]

On 2019-12-18 5:52 p.m., Michel Dänzer <michel@daenzer.net> wrote:

>> this set adds support for removal of gpu drm dead code.
>> 
>> patch2:
>> `num_entries` is a data of unsigned type(compilers always treat as 
>> unsigned int) and SIZE_MAX == ~0,
>> 
>> so there is a impossible condition:
>> '(num_entries > ((~0) - 56) / 72) => (max(0-u32) > 256204778801521549)'.

>While this looks correct for 64-bit, where size_t is unsigned long, on 32-bit it's unsigned int, in which case this isn't dead code.

and

On 2019-12-18 5:52 p.m., Christian König <christian.koenig@xxxxxxx> wrote:

>NAK, that calculation is not correct on 32-bit systems.

>Christian.

Yes, you are right. I just figured this out (because our Server usually uses 64 bits).

thanks.


>>Signed-off-by: Pan Zhang <zhangpan26@xxxxxxxxxx>
>>---
>>  drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 4 ----
>>  1 file changed, 4 deletions(-)
>>diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
>>index 85b0515..10a7f30 100644
>>--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
>>+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
>>@@ -71,10 +71,6 @@ int amdgpu_bo_list_create(struct amdgpu_device *adev, struct drm_file *filp,
>> 	unsigned i;
>> 	int r;
>> 
>>-	if (num_entries > (SIZE_MAX - sizeof(struct amdgpu_bo_list))
>>-				/ sizeof(struct amdgpu_bo_list_entry))
>>-		return -EINVAL;
>>-
>> 	size = sizeof(struct amdgpu_bo_list);
>> 	size += num_entries * sizeof(struct amdgpu_bo_list_entry);
>> 	list = kvmalloc(size, GFP_KERNEL);
>>