Re: [RFC PATCH 3/3] ALSA: pcm: Convert snd_pcm_sync_ptr() to user_access_begin/user_access_end()

[Takashi Iwai <tiwai@suse.de>]

On Fri, 13 Jun 2025 14:46:46 +0200,
Christophe Leroy wrote:
> 
> 
> 
> Le 13/06/2025 à 14:37, Takashi Iwai a écrit :
> > On Fri, 13 Jun 2025 13:03:04 +0200,
> > Christophe Leroy wrote:
> >> 
> >> 
> >> 
> >> Le 13/06/2025 à 11:29, Takashi Iwai a écrit :
> >>> On Thu, 12 Jun 2025 12:51:05 +0200,
> >>> Christophe Leroy wrote:
> >>>> 
> >>>> Now that snd_pcm_sync_ptr_get_user() and snd_pcm_sync_ptr_put_user()
> >>>> are converted to user_access_begin/user_access_end(),
> >>>> snd_pcm_sync_ptr_get_user() is more efficient than a raw get_user()
> >>>> followed by a copy_from_user(). And because copy_{to/from}_user() are
> >>>> generic functions focussed on transfer of big data blocks to/from user,
> >>>> snd_pcm_sync_ptr_put_user() is also more efficient for small amont of
> >>>> data.
> >>>> 
> >>>> So use snd_pcm_sync_ptr_get_user() and snd_pcm_sync_ptr_put_user() in
> >>>> snd_pcm_sync_ptr() too.
> >>>> 
> >>>> In order to have snd_pcm_mmap_status32 similar to snd_pcm_mmap_status,
> >>>> replace to tsamp_{sec/nsec} and audio_tstamp_{sec/nsec} by equivalent
> >>>> struct __snd_timespec.
> >>>> 
> >>>> snd_pcm_ioctl_sync_ptr_buggy() is left as it is because the conversion
> >>>> wouldn't be straigh-forward do to the workaround it provides.
> >>>> 
> >>>> Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
> >>> 
> >>> Through a quick glance, all patches look almost fine, but one favor to
> >>> ask: this patch contains the convert from s32/s32 pair to struct
> >>> __snd_timespec.  It should be factored out to a prerequisite patch
> >>> instead of burying in a big change.
> >> 
> >> Shall I understand you prefer this series over the more simple "ALSA:
> >> pcm: Convert snd_pcm_ioctl_sync_ptr_{compat/x32} to
> >> user_access_begin/user_access_end()" patch ?
> > 
> > Err, no, sorry for ambiguity.
> 
> Then I'm lost.
> 
> I sent two alternative proposals:
> A/ Single patch, simple, handling only two fonctions
> snd_pcm_ioctl_sync_ptr_{compat/x32} , without refactoring. [1]
> B/ This RFC series, more elaborate, refactoring and putting user copy
> into helper macros. [2]
> 
> So the question was to be sure you prefer alternative B over
> alternative A. I guess the answer is YES as you asking me improve it.

Right, let's go with the RFC series with refactoring.


thanks,

Takashi

> 
> [1]
> https://patchwork.ozlabs.org/project/linuxppc-dev/patch/8df11af98033e4cb4d9b0f16d6e9d5b69110b036.1749724057.git.christophe.leroy@csgroup.eu/
> [2]
> https://patchwork.ozlabs.org/project/linuxppc-dev/list/?state=*&series=460665
> 
> 
> > I wanted to move the replacement of tstamp_sec/nsec with struct
> > __snd_timespec as a small preliminary patch from patch#3.
> > That is,
> 
> Yes that's what I understood.
> 
> Thanks
> Christophe
> 
> 
> > --- a/sound/core/pcm_native.c
> > +++ b/sound/core/pcm_native.c
> > @@ -3103,11 +3103,9 @@ struct snd_pcm_mmap_status32 {
> >   	snd_pcm_state_t state;
> >   	s32 pad1;
> >   	u32 hw_ptr;
> > -	s32 tstamp_sec;
> > -	s32 tstamp_nsec;
> > +	struct __snd_timespec tstamp;
> >   	snd_pcm_state_t suspended_state;
> > -	s32 audio_tstamp_sec;
> > -	s32 audio_tstamp_nsec;
> > +	struct __snd_timespec audio_tstamp;
> >   } __packed;
> > etc.  By factoring this out, it becomes clear that the timespec
> > compatibility is fully cared.
> > 
> > __snd_timespec may be defined in different ways on user-space, but in
> > the kernel code, it's a single definition of s32/s32 pair.  This needs
> > to be emphasized.
> > 
> > 
> > thanks,
> > 
> > Takashi
>