Re: [PATCH v4 08/13] drm/v3d: Fix builds with CONFIG_TRANSPARENT_HUGEPAGE=n

["Loïc Molinari" <loic.molinari@collabora.com>]

Hi Boris,

On 16/10/2025 07:56, Boris Brezillon wrote:
> On Wed, 15 Oct 2025 22:41:59 +0200
> Loïc Molinari <loic.molinari@collabora.com> wrote:
> 
>> On 15/10/2025 20:17, Boris Brezillon wrote:
>>> On Wed, 15 Oct 2025 17:30:12 +0200
>>> Loïc Molinari <loic.molinari@collabora.com> wrote:
>>>    
>>>> Don't declare "super_pages" on builds with CONFIG_TRANSPARENT_HUGEPAGE
>>>> disabled to prevent build error:
>>>>
>>>> ERROR: modpost: "super_pages" [drivers/gpu/drm/v3d/v3d.ko] undefined!
>>>
>>> I believe this is a bug introduced by the previous commit: the
>>> compiler probably drops any code between the
>>> IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) check and the err label
>>> because IS_ENABLED() evaluates to false at compile time. So I'd squash
>>> those changes in the previous commit.
>>
>> Right, it's been introduced in previous commit.
>>
>>>   
>>>>
>>>> Signed-off-by: Loïc Molinari <loic.molinari@collabora.com>
>>>> ---
>>>>    drivers/gpu/drm/v3d/v3d_drv.h | 2 ++
>>>>    drivers/gpu/drm/v3d/v3d_gem.c | 2 ++
>>>>    2 files changed, 4 insertions(+)
>>>>
>>>> diff --git a/drivers/gpu/drm/v3d/v3d_drv.h b/drivers/gpu/drm/v3d/v3d_drv.h
>>>> index 99a39329bb85..481502104391 100644
>>>> --- a/drivers/gpu/drm/v3d/v3d_drv.h
>>>> +++ b/drivers/gpu/drm/v3d/v3d_drv.h
>>>> @@ -564,7 +564,9 @@ extern const struct dma_fence_ops v3d_fence_ops;
>>>>    struct dma_fence *v3d_fence_create(struct v3d_dev *v3d, enum v3d_queue q);
>>>>    
>>>>    /* v3d_gem.c */
>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>    extern bool super_pages;
>>>> +#endif
>>>>    int v3d_gem_init(struct drm_device *dev);
>>>>    void v3d_gem_destroy(struct drm_device *dev);
>>>>    void v3d_reset_sms(struct v3d_dev *v3d);
>>>> diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
>>>> index 635ff0fabe7e..0039063eb8b2 100644
>>>> --- a/drivers/gpu/drm/v3d/v3d_gem.c
>>>> +++ b/drivers/gpu/drm/v3d/v3d_gem.c
>>>> @@ -269,7 +269,9 @@ v3d_huge_mnt_init(struct v3d_dev *v3d)
>>>>    	 * match our usecase.
>>>>    	 */
>>>>    
>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>    	if (super_pages)
>>>> +#endif
>>>>    		err = drm_gem_huge_mnt_create(&v3d->drm, "within_size");
>>>
>>> Why not
>>>
>>> #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>     	if (super_pages)
>>>     		err = drm_gem_huge_mnt_create(&v3d->drm, "within_size");
>>> #endif
>>>
>>> I guess
>>>
>>> 	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && super_pages)
>>> 		err = drm_gem_huge_mnt_create(&v3d->drm, "within_size");
>>>
>>> would also do, since it's likely to rely on the same optimization the
>>> previous v3d_gemfs_init() implementation was relying on, but it's
>>> fragile (not sure what happens when compiled with -O0).
>>
>> I'll remove the #ifdef/#endif around the super_pages declaration in
>> v3d_drv.h because it isn't necessary if super_pages is compiled out in
>> v3d_huge_mnt_init().
>>
>> In v3d_huge_mnt_init(), I'd add the #ifdef before the ret variable
>> declaration and the #endif right after the last else so that it's clear
>> drm_notice("THP is recommended...") is called unconditionally when
>> CONFIG_TRANSPARENT_HUGEPAGE=n, whatever the optim level. What do you think?
> 
> First off, I'm not a huge fan of the following pattern
> 
> #if foo
> 	if (xxxx)
> #endif
> 		do_something
> 
> which also applies to
> 
> #if foo
> 	if (xxxx)
> 		do_xxx
> 	else if (yyy)
> 		do_yyy
> 	else
> #endif
> 		do_something
> 
> I'd rather have do_something duplicated in an #else section
> like that:
> 
> #if foo
> 	if (xxxx)
> 		do_xxx
> 	else if (yyy)
> 		do_yyy
> 	else
> 		do_something
> #else
> 	do_something
> #endif
> 
> But I'm not even seeing what the problem is here. If you do:
> 
> 	int err = 0;
> 
> #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>      	if (super_pages)
>      		err = drm_gem_huge_mnt_create(&v3d->drm, "within_size");
> #endif
> 
> 	if (v3d->drm.huge_mnt)
> 		drm_info(&v3d->drm, "Using Transparent Hugepages\n");
> 	else if (err)
> 		drm_warn(&v3d->drm, "Can't use Transparent Hugepages (%d)\n", err);
> 	else
> 		drm_notice(&v3d->drm,
> 			   "Transparent Hugepage support is recommended for optimal performance on this platform!\n");
> 
> You're guaranteed that err=0 and v3d->drm.huge_mnt=NULL when
> CONFIG_TRANSPARENT_HUGEPAGE=n, so the "THP recommended"
> message should be displayed unconditionally. Am I missing
> something?

It doesn't really matter here but I just thought it would be cleaner to 
explicitly let just the drm_notice() because the compiler doesn't know 
v3d->drm.huge_mnt is always NULL here and would emit a branch in 
CONFIG_TRANSPARENT_HUGEPAGE=n builds. I know your dislike for this 
pattern now, so I will stick to the suggestion :)

Loïc