[PATCH] drm/amdgpu: potential shift wrapping bug

[PATCH] drm/amdgpu: potential shift wrapping bug

[Dan Carpenter]

"frag_align" is a u64, so presumably we want to use the high bits as
well instead of shift wrapping.

Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for Vega10 v2")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index ba0407d12525..d9a8e942ac3b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1459,7 +1459,7 @@ static int amdgpu_vm_frag_ptes(struct amdgpu_pte_update_params	*params,
 	/* SI and newer are optimized for 64KB */
 	unsigned pages_per_frag = AMDGPU_LOG2_PAGES_PER_FRAG(params->adev);
 	uint64_t frag_flags = AMDGPU_PTE_FRAG(pages_per_frag);
-	uint64_t frag_align = 1 << pages_per_frag;
+	uint64_t frag_align = 1ULL << pages_per_frag;
 
 	uint64_t frag_start = ALIGN(start, frag_align);
 	uint64_t frag_end = end & ~(frag_align - 1);

Re: [PATCH] drm/amdgpu: potential shift wrapping bug

[Christian König]

Am 10.08.2017 um 14:16 schrieb Dan Carpenter:
> "frag_align" is a u64, so presumably we want to use the high bits as
> well instead of shift wrapping.
>
> Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for Vega10 v2")
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

The fragment field has only 5bits in hardware and can never be more than 
31, so the correct fix would actually be using uint32_t here instead.

Christian.

>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index ba0407d12525..d9a8e942ac3b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -1459,7 +1459,7 @@ static int amdgpu_vm_frag_ptes(struct amdgpu_pte_update_params	*params,
>   	/* SI and newer are optimized for 64KB */
>   	unsigned pages_per_frag = AMDGPU_LOG2_PAGES_PER_FRAG(params->adev);
>   	uint64_t frag_flags = AMDGPU_PTE_FRAG(pages_per_frag);
> -	uint64_t frag_align = 1 << pages_per_frag;
> +	uint64_t frag_align = 1ULL << pages_per_frag;
>   
>   	uint64_t frag_start = ALIGN(start, frag_align);
>   	uint64_t frag_end = end & ~(frag_align - 1);
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH] drm/amdgpu: potential shift wrapping bug

[Dan Carpenter]

On Thu, Aug 10, 2017 at 02:30:15PM +0200, Christian König wrote:
> Am 10.08.2017 um 14:16 schrieb Dan Carpenter:
> > "frag_align" is a u64, so presumably we want to use the high bits as
> > well instead of shift wrapping.
> > 
> > Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for Vega10 v2")
> > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> 
> The fragment field has only 5bits in hardware and can never be more than 31,
> so the correct fix would actually be using uint32_t here instead.
> 

Alright.  Thanks.  I'll resend.

regards,
dan carpenter

--
To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] drm/amdgpu: potential shift wrapping bug

[Dan Carpenter]

On Thu, Aug 10, 2017 at 02:30:15PM +0200, Christian König wrote:
> Am 10.08.2017 um 14:16 schrieb Dan Carpenter:
> > "frag_align" is a u64, so presumably we want to use the high bits as
> > well instead of shift wrapping.
> > 
> > Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for Vega10 v2")
> > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> 
> The fragment field has only 5bits in hardware and can never be more than 31,
> so the correct fix would actually be using uint32_t here instead.
> 

Changing it to uint32_t introduces a new static checker warning:

    drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c:1465 amdgpu_vm_frag_ptes()
    warn: was expecting a 64 bit value instead of '~(frag_align - 1)'

Unfortunately, I get so many thousands of those I can't normally even
review that sort of bug...

Let me resend the original patch but with a modified changelog to say
that the bug is a false positive.

regards,
dan carpenter

--
To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] drm/amdgpu: potential shift wrapping bug

[Christian König]

Am 10.08.2017 um 14:53 schrieb Dan Carpenter:
> On Thu, Aug 10, 2017 at 02:30:15PM +0200, Christian König wrote:
>> Am 10.08.2017 um 14:16 schrieb Dan Carpenter:
>>> "frag_align" is a u64, so presumably we want to use the high bits as
>>> well instead of shift wrapping.
>>>
>>> Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for Vega10 v2")
>>> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
>> The fragment field has only 5bits in hardware and can never be more than 31,
>> so the correct fix would actually be using uint32_t here instead.
>>
> Changing it to uint32_t introduces a new static checker warning:
>
>      drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c:1465 amdgpu_vm_frag_ptes()
>      warn: was expecting a 64 bit value instead of '~(frag_align - 1)'
>
> Unfortunately, I get so many thousands of those I can't normally even
> review that sort of bug...
>
> Let me resend the original patch but with a modified changelog to say
> that the bug is a false positive.

Ah, yes of course that's why I made it a 64bit value in the first place.

Mhm, could we use something like (u32)(1 << pages_per_frag) instead to 
silence the static checker warning?

It doesn't make much sense to use a 64bit shift here.

Christian.

>
> regards,
> dan carpenter
>

--
To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] drm/amdgpu: potential shift wrapping bug

[Dan Carpenter]

On Thu, Aug 10, 2017 at 03:02:53PM +0200, Christian König wrote:
> Am 10.08.2017 um 14:53 schrieb Dan Carpenter:
> > On Thu, Aug 10, 2017 at 02:30:15PM +0200, Christian König wrote:
> > > Am 10.08.2017 um 14:16 schrieb Dan Carpenter:
> > > > "frag_align" is a u64, so presumably we want to use the high bits as
> > > > well instead of shift wrapping.
> > > > 
> > > > Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for Vega10 v2")
> > > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> > > The fragment field has only 5bits in hardware and can never be more than 31,
> > > so the correct fix would actually be using uint32_t here instead.
> > > 
> > Changing it to uint32_t introduces a new static checker warning:
> > 
> >      drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c:1465 amdgpu_vm_frag_ptes()
> >      warn: was expecting a 64 bit value instead of '~(frag_align - 1)'
> > 
> > Unfortunately, I get so many thousands of those I can't normally even
> > review that sort of bug...
> > 
> > Let me resend the original patch but with a modified changelog to say
> > that the bug is a false positive.
> 
> Ah, yes of course that's why I made it a 64bit value in the first place.
> 
> Mhm, could we use something like (u32)(1 << pages_per_frag) instead to
> silence the static checker warning?

That wouldn't silence it and I think that's not super pretty either.

> 
> It doesn't make much sense to use a 64bit shift here.
> 

I'm just going to ignore the warning.  This driver isn't part of my
.config so I'm not really compiling it the way it was designed which
means I don't have the cross function database enabled.  Probably if I
compiled this normally, I wouldn't even get the warning.

regards,
dan carpenter

--
To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] drm/amdgpu: potential shift wrapping bug

[walter harms]

Am 10.08.2017 15:02, schrieb Christian König:
> Am 10.08.2017 um 14:53 schrieb Dan Carpenter:
>> On Thu, Aug 10, 2017 at 02:30:15PM +0200, Christian König wrote:
>>> Am 10.08.2017 um 14:16 schrieb Dan Carpenter:
>>>> "frag_align" is a u64, so presumably we want to use the high bits as
>>>> well instead of shift wrapping.
>>>>
>>>> Fixes: 6be7adb37d9b ("drm/amdgpu: increase fragmentation size for
>>>> Vega10 v2")
>>>> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
>>> The fragment field has only 5bits in hardware and can never be more
>>> than 31,
>>> so the correct fix would actually be using uint32_t here instead.
>>>
>> Changing it to uint32_t introduces a new static checker warning:
>>
>>      drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c:1465 amdgpu_vm_frag_ptes()
>>      warn: was expecting a 64 bit value instead of '~(frag_align - 1)'
>>
>> Unfortunately, I get so many thousands of those I can't normally even
>> review that sort of bug...
>>
>> Let me resend the original patch but with a modified changelog to say
>> that the bug is a false positive.
> 
> Ah, yes of course that's why I made it a 64bit value in the first place.
> 
> Mhm, could we use something like (u32)(1 << pages_per_frag) instead to
> silence the static checker warning?
> 
> It doesn't make much sense to use a 64bit shift here.
> 
> Christian.
> 



Why not keeping Dan 1. patch and add a comment that pages_per_frag is always >31 ?

Using 32bit in a 64bit is not forbidden, and changing it causes more problems than
it solves. But doing so should be done in a clean way.

just my 2 cents,
re,
 wh