Message-ID: <CAJuCfpHG5U5Radv+_D3nD3OxAgqzBa_Za28+RS=NTG53Y=xpsg@mail.gmail.com>
Date: Fri, 5 Apr 2024 06:47:17 -0700
From: Suren Baghdasaryan <surenb@...gle.com>
To: Jan Kara <jack@...e.cz>
Cc: Kent Overstreet <kent.overstreet@...ux.dev>, Andrew Morton <akpm@...ux-foundation.org>, 
	Matthew Wilcox <willy@...radead.org>, joro@...tes.org, will@...nel.org, 
	trond.myklebust@...merspace.com, anna@...nel.org, arnd@...db.de, 
	herbert@...dor.apana.org.au, davem@...emloft.net, jikos@...nel.org, 
	benjamin.tissoires@...hat.com, tytso@....edu, jack@...e.com, 
	dennis@...nel.org, tj@...nel.org, cl@...ux.com, jakub@...udflare.com, 
	penberg@...nel.org, rientjes@...gle.com, iamjoonsoo.kim@....com, 
	vbabka@...e.cz, edumazet@...gle.com, kuba@...nel.org, pabeni@...hat.com, 
	iommu@...ts.linux.dev, linux-kernel@...r.kernel.org, 
	linux-nfs@...r.kernel.org, linux-acpi@...r.kernel.org, 
	acpica-devel@...ts.linux.dev, linux-arch@...r.kernel.org, 
	linux-crypto@...r.kernel.org, bpf@...r.kernel.org, 
	linux-input@...r.kernel.org, linux-ext4@...r.kernel.org, linux-mm@...ck.org, 
	netdev@...r.kernel.org, linux-security-module@...r.kernel.org
Subject: Re: [PATCH 1/1] mm: change inlined allocation helpers to account at
 the call site

On Fri, Apr 5, 2024 at 2:53 AM Jan Kara <jack@...e.cz> wrote:
>
> On Thu 04-04-24 16:16:15, Suren Baghdasaryan wrote:
> > On Thu, Apr 4, 2024 at 4:01 PM Kent Overstreet
> > <kent.overstreet@...ux.dev> wrote:
> > >
> > > On Thu, Apr 04, 2024 at 03:41:50PM -0700, Andrew Morton wrote:
> > > > On Thu, 4 Apr 2024 18:38:39 -0400 Kent Overstreet <kent.overstreet@...ux.dev> wrote:
> > > >
> > > > > On Thu, Apr 04, 2024 at 11:33:22PM +0100, Matthew Wilcox wrote:
> > > > > > On Thu, Apr 04, 2024 at 03:17:43PM -0700, Suren Baghdasaryan wrote:
> > > > > > > Ironically, checkpatch generates warnings for these type casts:
> > > > > > >
> > > > > > > WARNING: unnecessary cast may hide bugs, see
> > > > > > > http://c-faq.com/malloc/mallocnocast.html
> > > > > > > #425: FILE: include/linux/dma-fence-chain.h:90:
> > > > > > > + ((struct dma_fence_chain *)kmalloc(sizeof(struct dma_fence_chain),
> > > > > > > GFP_KERNEL))
> > > > > > >
> > > > > > > I guess I can safely ignore them in this case (since we cast to the
> > > > > > > expected type)?
> > > > > >
> > > > > > I find ignoring checkpatch to be a solid move 99% of the time.
> > > > > >
> > > > > > I really don't like the codetags.  This is so much churn, and it could
> > > > > > all be avoided by just passing in _RET_IP_ or _THIS_IP_ depending on
> > > > > > whether we wanted to profile this function or its caller.  vmalloc
> > > > > > has done it this way since 2008 (OK, using __builtin_return_address())
> > > > > > and lockdep has used _THIS_IP_ / _RET_IP_ since 2006.
> > > > >
> > > > > Except you can't. We've been over this; using that approach for tracing
> > > > > is one thing, using it for actual accounting isn't workable.
> > > >
> > > > I missed that.  There have been many emails.  Please remind us of the
> > > > reasoning here.
> > >
> > > I think it's on the other people claiming 'oh this would be so easy if
> > > you just do it this other way' to put up some code - or at least more
> > > than hot takes.
> > >
> > > But, since you asked - one of the main goals of this patchset was to be
> > > fast enough to run in production, and if you do it by return address
> > > then you've added at minimum a hash table lookup to every allocate and
> > > free; if you do that, running it in production is completely out of the
> > > question.
> > >
> > > Besides that - the issues with annotating and tracking the correct
> > > callsite really don't go away, they just shift around a bit. It's true
> > > that the return address approach would be easier initially, but that's
> > > not all we're concerned with; we're concerned with making sure
> > > allocations get accounted to the _correct_ callsite so that we're giving
> > > numbers that you can trust, and by making things less explicit you make
> > > that harder.
> > >
> > > Additionally: the alloc_hooks() macro is for more than this. It's also
> > > for more usable fault injection - remember every thread we have where
> > > people are begging for every allocation to be __GFP_NOFAIL - "oh, error
> > > paths are hard to test, let's just get rid of them" - never mind that
> > > actually do have to have error paths - but _per callsite_ selectable
> > > fault injection will actually make it practical to test memory error
> > > paths.
> > >
> > > And Kees working on stuff that'll make use of the alloc_hooks() macro
> > > for segregating kmem_caches.
> >
> > Yeah, that pretty much summarizes it. Note that we don't have to make
> > the conversions in this patch and accounting will still work but then
> > all allocations from different callers will be accounted to the helper
> > function and that's less useful than accounting at the call site.
> > It's a sizable churn but the conversions are straight-forward and we
> > do get accurate, performant and easy to use memory accounting.
>
> OK, fair enough. I guess I can live with the allocation macros in jbd2 if
> type safety is preserved. But please provide a short summary of why we need
> these macros (e.g. instead of RET_IP approach) in the changelog (or at
> least a link to some email explaining this if the explanation would get too
> long). Because I was wondering about the same as Andrew (and yes, this is
> because I wasn't really following the huge discussion last time).

Ack. I'll write up the explanation or if there is a good one already
in our previous discussion will add a link to it. Thanks!

>
>                                                                 Honza
> --
> Jan Kara <jack@...e.com>
> SUSE Labs, CR

Powered by blists - more mailing lists