Re: [PATCH v1 1/8] drm/etnaviv: Using the size_t variable to store the number of pages

[Lucas Stach]

Am Montag, dem 17.07.2023 um 18:12 +0800 schrieb Sui Jingfeng:
> Hi
> 
> On 2023/7/17 17:43, Lucas Stach wrote:
> > Hi Jingfeng,
> > 
> > Am Freitag, dem 23.06.2023 um 18:08 +0800 schrieb Sui Jingfeng:
> > > From: Sui Jingfeng <suijingfeng@xxxxxxxxxxx>
> > > 
> > > Because the etnaviv_gem_new_private() function receives the size_t argument
> > > for the number of pages. And the number of pages should be unsigned.
> > > 
> > > Note that Most 32-bit architectures use "unsigned int" size_t,
> > > and all 64-bit architectures use "unsigned long" size_t.
> > > So, let's keep the argument and parameter consistent.
> > > 
> > This explanation doesn't add up. npages is just that: a number of
> > pages. Why would it make sense to use size_t here?
> 
> Because the 'size' variable in the etnaviv_gem_prime_import_sg_table() 
> function is declared
> 
> as size_t type. On 64-bit machine, size_t is actually is 64-bit wide and 
> it is *unsigned*.
> 
> While 'int' is actually 32-bit wide(at both 32-bit system and 64-bit 
> system) and it is *signed*,
> 
> So, my point (argument) is that
> 
> 
> 1) This patch help to avoid the unnecessary 64 bit to 32 bit conversion.
> 
> 2) The kvmalloc_array() function also accept  size_t type (see the 
> prototype of  kvmalloc_array function include/linux/slab.h)
> 
> 
> So my patch do helps to keep the code style consistent.
> 
But then we go on to call drm_prime_sq_to_page_array(), which takes a
integer as the number of pages parameter, so the parameter types are
inconsistent before and after your patch, it just switches which
function call has to do some conversion.

> 
> > If you want to be consistent I would have expected this change to
> > switch things to unsigned int,
> 
> This may introduce a truncate operation (from a 64-bit to 32-bit), which 
> is necessary.
> 
If this truncation happens in the real world then something else is
already badly broken. All Vivante GPUs to date can only deal with 32bit
virtual addresses, so a buffer exhausting 31 bits of pages is way
larger than we could ever fit into the GPU VM.

Regards,
Lucas

> And when you pass the 'npages' parameter to kvmalloc_array() function,
> 
> The compiler probably has to do the integer promotion (from a 32-bit to 
> 64-bit) for you.
> 
> 
> > as you did in the second patch of this
> > series.
> > 
> > Regards,
> > Lucas
> > 
> > > Signed-off-by: Sui Jingfeng <suijingfeng@xxxxxxxxxxx>
> > > ---
> > >   drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c | 5 ++---
> > >   1 file changed, 2 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c b/drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c
> > > index 3524b5811682..b003481adc2b 100644
> > > --- a/drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c
> > > +++ b/drivers/gpu/drm/etnaviv/etnaviv_gem_prime.c
> > > @@ -114,7 +114,8 @@ struct drm_gem_object *etnaviv_gem_prime_import_sg_table(struct drm_device *dev,
> > >   {
> > >   	struct etnaviv_gem_object *etnaviv_obj;
> > >   	size_t size = PAGE_ALIGN(attach->dmabuf->size);
> > > -	int ret, npages;
> > > +	size_t npages = size / PAGE_SIZE;
> > > +	int ret;
> > >   
> > >   	ret = etnaviv_gem_new_private(dev, size, ETNA_BO_WC,
> > >   				      &etnaviv_gem_prime_ops, &etnaviv_obj);
> > > @@ -123,8 +124,6 @@ struct drm_gem_object *etnaviv_gem_prime_import_sg_table(struct drm_device *dev,
> > >   
> > >   	lockdep_set_class(&etnaviv_obj->lock, &etnaviv_prime_lock_class);
> > >   
> > > -	npages = size / PAGE_SIZE;
> > > -
> > >   	etnaviv_obj->sgt = sgt;
> > >   	etnaviv_obj->pages = kvmalloc_array(npages, sizeof(struct page *), GFP_KERNEL);
> > >   	if (!etnaviv_obj->pages) {