Re: [patch 02/14] tmpfs: fix regressions from wider use of ZERO_PAGE

From: Borislav Petkov <bp@alien8.de>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mark Hemment <markhemm@googlemail.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	the arch/x86 maintainers <x86@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	patrice.chotard@foss.st.com,
	Mikulas Patocka <mpatocka@redhat.com>,
	Lukas Czerner <lczerner@redhat.com>,
	Christoph Hellwig <hch@lst.de>,
	"Darrick J. Wong" <djwong@kernel.org>,
	Chuck Lever <chuck.lever@oracle.com>,
	Hugh Dickins <hughd@google.com>,
	patches@lists.linux.dev, Linux-MM <linux-mm@kvack.org>,
	mm-commits@vger.kernel.org, Mel Gorman <mgorman@suse.de>
Subject: Re: [patch 02/14] tmpfs: fix regressions from wider use of ZERO_PAGE
Date: Wed, 4 May 2022 23:01:08 +0200	[thread overview]
Message-ID: <YnLplKy0Y66SsvQw@zn.tnic> (raw)
In-Reply-To: <CAHk-=wiGbrJMim6EWncZUQBzguqy-vtNd+grfNizm5L8Vcmu+w@mail.gmail.com>

On Wed, May 04, 2022 at 01:40:07PM -0700, Linus Torvalds wrote:
> Sadly, I'm on Zen2 with my 3970X, and the Zen 3 threadrippers seem to
> be basically impossible to get.

Yeah, what we did is get a gigabyte board:

[    0.000000] DMI: GIGABYTE MZ32-AR0-00/MZ32-AR0-00, BIOS M06 07/10/2021

and stick a server CPU in it:

[    2.352371] smpboot: CPU0: AMD EPYC 7313 16-Core Processor (family: 0x19, model: 0x1, stepping: 0x1)

so that we can have the memory encryption stuff. 32 threads is fairly
decent and kernel builds are fast enough, at least for me. If you need
to do a lot of allmodconfigs, you probably need something bigger though.

> Oh, the clear_user() in the ELF loader? I wouldn't have expected that
> to be noticeable.

Yah, that guy in load_elf_binary(). At least it did hit my breakpoint
fairly often so I thought, what is a benchmark that does create a lot of
processes...

> Now, clear_page() is *very* noticeable, but that has its own special
> magic and doesn't use clear_user().

Yeah, that's with the alternative CALL thing. Could be useful to try to
see how much the inlined rep; movsb would bring...

> Maybe there is some other clear_user() case I never thought of. My
> dim memories of profiles definitely had copy_to_user, clear_page and
> copy_page being some of the top copy loops.

I could try to do a perf probe or whatever fancy new thing we do now on
clear_user to get some numbers of how many times it gets called during
the benchmark run. Or do you wanna know the callers too?

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette