Received-SPF: pass (google.com: domain of linux-kernel+bounces-103313-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Precedence: bulk
MIME-Version: 1.0
References: <20240313033417.447216-1-pasha.tatashin@soleen.com> <7af73776-06f9-42e6-9bfc-fabe8f8b002e@csgroup.eu>
In-Reply-To: <7af73776-06f9-42e6-9bfc-fabe8f8b002e@csgroup.eu>
From: Pasha Tatashin <pasha.tatashin@soleen.com>
Date: Thu, 14 Mar 2024 09:42:05 -0400
Message-ID: <CA+CK2bBdJ6RcSdqfE+EkmxgnVnzpRvE+Vz3PUbvYHEHbAQs2NQ@mail.gmail.com>
Subject: Re: [PATCH] vmstat: Keep count of the maximum page reached by the
 kernel stack
To: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: "akpm@linux-foundation.org" <akpm@linux-foundation.org>, "jpoimboe@kernel.org" <jpoimboe@kernel.org>, 
	"kent.overstreet@linux.dev" <kent.overstreet@linux.dev>, "peterz@infradead.org" <peterz@infradead.org>, 
	"nphamcs@gmail.com" <nphamcs@gmail.com>, 
	"cerasuolodomenico@gmail.com" <cerasuolodomenico@gmail.com>, "surenb@google.com" <surenb@google.com>, 
	"lizhijian@fujitsu.com" <lizhijian@fujitsu.com>, "willy@infradead.org" <willy@infradead.org>, 
	"shakeel.butt@linux.dev" <shakeel.butt@linux.dev>, "vbabka@suse.cz" <vbabka@suse.cz>, 
	"ziy@nvidia.com" <ziy@nvidia.com>, 
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>, "linux-mm@kvack.org" <linux-mm@kvack.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Thu, Mar 14, 2024 at 4:19=E2=80=AFAM Christophe Leroy
<christophe.leroy@csgroup.eu> wrote:
>
>
>
> Le 13/03/2024 =C3=A0 04:34, Pasha Tatashin a =C3=A9crit :
> > CONFIG_DEBUG_STACK_USAGE provides a mechanism to know the minimum amoun=
t
> > of memory that was left in stack. Every time the new anti-record is
> > reached a message is printed to the console.
> >
> > However, this is not useful to know how much each page within stack was
> > actually used. Provide a mechanism to count the number of time each
> > stack page was reached throughout the live of the stack:
>
> by "this is not useful to know ", you mean "this does not allow us to
> know" ?

Yes, bad wording from my side, I will change it to you suggestion in
the next version.

>
> >
> >       $
> >       kstack_page_1 19974
> >       kstack_page_2 94
> >       kstack_page_3 0
> >       kstack_page_4 0
>
> That's probably only usefull when THREAD_SIZE is larger than PAGE_SIZE.

That is right, if THREAD_SIZE <=3D PAGE_SIZE, only "kstack_page_1" would
be filled.

>
> On powerpc 8xx, THREAD_SIZE is 8k by default and PAGE_SIZE can be either
> 4k or 16k.

With THREAD_SIZE =3D=3D 8K, and  PAGE_SIZE =3D 4K
There will be  two counters in /proc/vmstat, something like this:

kstack_page_1 XXX
kstack_page_2 YYY

With THREAD_SIZE=3D16K, and PAGE_SIZE =3D 16K
There will be two counters, but one will always be zero:

kstack_page_1 XXX
kstack_page_2 0

Thanks,
Pasha