Re: [PATCH v3] mm: Avoid unnecessary page fault retires on shared memory types

From: Ingo Molnar <mingo@kernel.org>
To: Peter Xu <peterx@redhat.com>
Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	Richard Henderson <rth@twiddle.net>,
	David Hildenbrand <david@redhat.com>,
	Matt Turner <mattst88@gmail.com>,
	Albert Ou <aou@eecs.berkeley.edu>,
	Michal Simek <monstr@monstr.eu>,
	Russell King <linux@armlinux.org.uk>,
	Ivan Kokshaysky <ink@jurassic.park.msu.ru>,
	linux-riscv@lists.infradead.org,
	Alexander Gordeev <agordeev@linux.ibm.com>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Jonas Bonn <jonas@southpole.se>, Will Deacon <will@kernel.org>,
	"James E . J . Bottomley" <James.Bottomley@hansenpartnership.com>,
	"H . Peter Anvin" <hpa@zytor.com>,
	Andrea Arcangeli <aarcange@redhat.com>,
	openrisc@lists.librecores.org, linux-s390@vger.kernel.org,
	Ingo Molnar <mingo@redhat.com>,
	linux-m68k@lists.linux-m68k.org,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Heiko Carstens <hca@linux.ibm.com>,
	Chris Zankel <chris@zankel.net>,
	Peter Zijlstra <peterz@infradead.org>,
	Alistair Popple <apopple@nvidia.com>,
	linux-csky@vger.kernel.org, linux-hexagon@vger.kernel.org,
	Vlastimil Babka <vbabka@suse.cz>,
	Thomas Gleixner <tglx@linutronix.de>,
	sparclinux@vger.kernel.org,
	Christian Borntraeger <borntraeger@linux.ibm.com>,
	Stafford Horne <shorne@gmail.com>,
	Michael Ellerman <mpe@ellerman.id.au>,
	x86@kernel.org, Thomas Bogendoerfer <tsbogend@alpha.franken.de>,
	Paul Mackerras <paulus@samba.org>,
	linux-arm-kernel@lists.infradead.org,
	Sven Schnelle <svens@linux.ibm.com>,
	Benjamin Herrenschmidt <benh@kernel.crashing.org>,
	linux-xtensa@linux-xtensa.org,
	Nicholas Piggin <npiggin@gmail.com>,
	linux-sh@vger.kernel.org, Vasily Gorbik <gor@linux.ibm.com>,
	Borislav Petkov <bp@alien8.de>,
	linux-mips@vger.kernel.org, Max Filippov <jcmvbkbc@gmail.com>,
	Helge Deller <deller@gmx.de>, Vineet Gupta <vgupta@kernel.org>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Paul Walmsley <paul.walmsley@sifive.com>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Anton Ivanov <anton.ivanov@cambridgegreys.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	linux-um@lists.infradead.org, linux-alpha@vger.kernel.org,
	Johannes Berg <johannes@sipsolutions.net>,
	linux-ia64@vger.kernel.org,
	Geert Uytterhoeven <geert@linux-m68k.org>,
	Dinh Nguyen <dinguyen@kernel.org>, Guo Ren <guoren@kernel.org>,
	linux-snps-arc@lists.infradead.org,
	Hugh Dickins <hughd@google.com>, Rich Felker <dalias@libc.org>,
	Andy Lutomirski <luto@kernel.org>,
	Richard Weinberger <richard@nod.at>,
	linuxppc-dev@lists.ozlabs.org, Brian Cain <bcain@quicinc.com>,
	Yoshinori Sato <ysato@users.sourceforge.jp>,
	Andrew Morton <akpm@linux-foundation.org>,
	Stefan Kristiansson <stefan.kristiansson@saunalahti.fi>,
	linux-parisc@vger.kernel.org,
	"David S . Miller" <davem@davemloft.net>
Subject: Re: [PATCH v3] mm: Avoid unnecessary page fault retires on shared memory types
Date: Fri, 27 May 2022 12:46:31 +0200	[thread overview]
Message-ID: <YpCsBwFArieTpvg2@gmail.com> (raw)
In-Reply-To: <20220524234531.1949-1-peterx@redhat.com>

* Peter Xu <peterx@redhat.com> wrote:

> This patch provides a ~12% perf boost on my aarch64 test VM with a simple
> program sequentially dirtying 400MB shmem file being mmap()ed and these are
> the time it needs:
>
>   Before: 650.980 ms (+-1.94%)
>   After:  569.396 ms (+-1.38%)

Nice!

>  arch/x86/mm/fault.c           |  4 ++++

Reviewed-by: Ingo Molnar <mingo@kernel.org>

Minor comment typo:

> +		/*
> +		 * We should do the same as VM_FAULT_RETRY, but let's not
> +		 * return -EBUSY since that's not reflecting the reality on
> +		 * what has happened - we've just fully completed a page
> +		 * fault, with the mmap lock released.  Use -EAGAIN to show
> +		 * that we want to take the mmap lock _again_.
> +		 */

s/reflecting the reality on what has happened
 /reflecting the reality of what has happened

>  	ret = handle_mm_fault(vma, address, fault_flags, NULL);
> +
> +	if (ret & VM_FAULT_COMPLETED) {
> +		/*
> +		 * NOTE: it's a pity that we need to retake the lock here
> +		 * to pair with the unlock() in the callers. Ideally we
> +		 * could tell the callers so they do not need to unlock.
> +		 */
> +		mmap_read_lock(mm);
> +		*unlocked = true;
> +		return 0;

Indeed that's a pity - I guess more performance could be gained here, 
especially in highly parallel threaded workloads?

Thanks,

	Ingo