Re: [PATCH v2] x86: bring back rep movsq for user access on CPUs without ERMS

[Mateusz Guzik]

On 9/4/23, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Sun, 3 Sept 2023 at 20:07, Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>>
>> Try it and you'll see it is not even *remotely* as easy as you claim.
>> Not when you have to deal with random sizes and padding of 20+
>> different architectures.
>
> Perhaps more importantly, nobody actually seems to have the energy to care.
>

Perhaps in this context I should have explicitly mentioned that only
64-bit archs would be considered, as in placing the func you wrote in
fs/stat.c and it is explicitly 64-bit. This should whack all archs
which would never bother with their implementation anyway.

Worst case if the 64 bit structs differ one can settle for
user-accessing INIT_STRUCT_STAT_PADDING.

My point though was that code filling the struct escaping to
arch-dependent file is rather regrettable, even if x86-64 would be the
only case. And it definitely can be avoided, even if would require
tolerating that some of the stores are reordered. They are already
reordered for dodging memset.

-- 
Mateusz Guzik <mjguzik gmail.com>