Re: [PATCH] ALSA: pcm: Convert multiple {get/put}_user to user_access_begin/user_access_end()

[Christophe Leroy]

Le 10/06/2025 à 21:53, David Laight a écrit :
> On Sat,  7 Jun 2025 13:37:42 +0200
> Christophe Leroy <christophe.leroy@xxxxxxxxxx> wrote:
>
> > With user access protection (Called SMAP on x86 or KUAP on powerpc)
> > each and every call to get_user() or put_user() performs heavy
> > operations to unlock and lock kernel access to userspace.
> >
> > To avoid that, perform user accesses by blocks using
> > user_access_begin/user_access_end() and unsafe_get_user()/
> > unsafe_put_user() and alike.
>
> Did you consider using masked_user_access_begin() ?
> It removes a conditional branch and lfence as well.

Thanks, was not aware of that new function, allthought I remember some 
discussion about masked user access.

Looks like this is specific to x86 at the time being. I would have 
expected that to be transparent to the consumer. Allthought looking at 
strncpy_from_user() I understand the benefit of keeping it separate.

However is it worth the effort and the ugliness of having to do (copied 
from fs/select.c):

		if (can_do_masked_user_access())
			from = masked_user_access_begin(from);
		else if (!user_read_access_begin(from, sizeof(*from)))
			return -EFAULT;

In addition I would expect a masked_user_read_access_begin() and a 
masked_write_access_begin(). It looks odd (and would be wrong on 
powerpc) to not be able to differentiate between read and write in the 
begin yet using user_read_access_end() at the end, ref get_sigset_argpack()

Christophe