Message-ID: <aLDTBSuM3xVbUok2@devbig569.cln6.facebook.com>
Date: Thu, 28 Aug 2025 15:07:01 -0700
From: Yueyang Pan <pyyjason@...il.com>
To: "Vishal Moola (Oracle)" <vishal.moola@...il.com>
Cc: Suren Baghdasaryan <surenb@...gle.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Vlastimil Babka <vbabka@...e.cz>, Michal Hocko <mhocko@...e.com>,
	Brendan Jackman <jackmanb@...gle.com>,
	Johannes Weiner <hannes@...xchg.org>, Zi Yan <ziy@...dia.com>,
	Usama Arif <usamaarif642@...il.com>, linux-mm@...ck.org,
	kernel-team@...a.com, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1 0/2] mm/show_mem: Bug fix for print mem alloc info

On Thu, Aug 28, 2025 at 10:05:18AM -0700, Vishal Moola (Oracle) wrote:
> On Thu, Aug 28, 2025 at 01:29:08AM -0700, Yueyang Pan wrote:
> > On Wed, Aug 27, 2025 at 12:51:17PM -0700, Vishal Moola (Oracle) wrote:
> > > On Wed, Aug 27, 2025 at 11:34:21AM -0700, Yueyang Pan wrote:
> > > > This patch set fixes two issues we saw in production rollout. 
> > > > 
> > > > The first issue is that we saw all zero output of memory allocation 
> > > > profiling information from show_mem() if CONFIG_MEM_ALLOC_PROFILING 
> > > > is set and sysctl.vm.mem_profiling=0. In this case, the behaviour 
> > > > should be the same as when CONFIG_MEM_ALLOC_PROFILING is unset, 
> > > 
> > > Did you mean to say when sysctl.vm.mem_profiling=never?
> > > 
> > > My understanding is that setting the sysctl=0 Pauses memory allocation
> > > profiling, while 1 Resumes it. When the sysctl=never should be the same
> > > as when the config is unset, but I suspect we might still want the info
> > > when set to 0.
> > 
> > Thanks for your feedback Vishal. Here I mean for both =0 and =never. 
> > In both cases, now __show_mem() will print all 0s, which both is redundant 
> > and also makes differentiate hard. IMO when __show_mem() prints something 
> > the output should be useful at least. 
> 
> If differentiating between 0 allocations vs disabled is the primary
> concern, I think prefacing the dump with the status of the tool is
> better than treating =0 and =never as the same.
> 
> The way I see it, the {0,1,never} tristate offers a level of versatility
> that I'm not sure we need to eliminate.
> 
> I'm thinking about cases where we may temporarily set =1 to track some
> allocations, then back to =0 'pause' on that exact period of time. Memory
> allocation profiling still has those allocations tracked while set to =0
> (we can still see them in /proc/allocinfo at least). If a user decided to
> do that just before an oom, could they see something useful from
> show_mem() even when =0?

This is a good point. I agree with your suggestion about adding the state 
to print. I am still unsure about if we want to print it when =0. The first 
reason is that memory allocation profiler does not support runtime enabling 
now. We have to set it via boot cmdline. It will make more sense if we 
have this feature. Second is because memory allocation profiling is quite 
light-weighted, I would assume user really don't need this feature when 
they set =0.

The original reason why I tried to disable this is because in our production 
table we see a lot of 0Bs coming from the machines where the changes in boot 
cmdline have not been pushed to. If we have state info, we could possibly 
filter this info out before sending it to the table. So I agree upon adding 
the state to print.

Maybe others also have thoughts about this?

> 
> > > 
> > > > where show_mem prints nothing about the information. This will make 
> > > > further parse easier as we don't have to differentiate what a all 
> > > > zero line actually means (Does it mean  0 bytes are allocated 
> > > > or simply memory allocation profiling is disabled).
> > > > 
> > > > The second issue is that multiple entities can call show_mem() 
> > > > which messed up the allocation info in dmesg. We saw outputs like this:  
> > > > ```
> > > >     327 MiB    83635 mm/compaction.c:1880 func:compaction_alloc
> > > >    48.4 GiB 12684937 mm/memory.c:1061 func:folio_prealloc
> > > >    7.48 GiB    10899 mm/huge_memory.c:1159 func:vma_alloc_anon_folio_pmd
> > > >     298 MiB    95216 kernel/fork.c:318 func:alloc_thread_stack_node
> > > >     250 MiB    63901 mm/zsmalloc.c:987 func:alloc_zspage
> > > >     1.42 GiB   372527 mm/memory.c:1063 func:folio_prealloc
> > > >     1.17 GiB    95693 mm/slub.c:2424 func:alloc_slab_page
> > > >      651 MiB   166732 mm/readahead.c:270 func:page_cache_ra_unbounded
> > > >      419 MiB   107261 net/core/page_pool.c:572 func:__page_pool_alloc_pages_slow
> > > >      404 MiB   103425 arch/x86/mm/pgtable.c:25 func:pte_alloc_one
> > > > ```
> > > > The above example is because one kthread invokes show_mem() 
> > > > from __alloc_pages_slowpath while kernel itself calls 
> > > > oom_kill_process()
> > > 
> > > I'm not familiar with show_mem(). Could you spell out what's wrong with
> > > the output above?
> > 
> > So here in the normal case, the output should be sorted by size. Here 
> > two print happen at the same time so they interleave with each other, 
> > making further parse harder (need to sort again and dedup).
> 
> Gotcha.
> 
> > > 
> > > > Yueyang Pan (2):
> > > >   mm/show_mem: No print when not mem_alloc_profiling_enabled()
> > > >   mm/show_mem: Add trylock while printing alloc info
> > > > 
> > > >  mm/show_mem.c | 5 ++++-
> > > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > > 
> > > > -- 
> > > > 2.47.3
> > > > 
> > 
> > Thanks,
> > Pan

Powered by blists - more mailing lists