Received-SPF: pass (google.com: domain of linux-kernel+bounces-182554-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Precedence: bulk
MIME-Version: 1.0
References: <87o79cjjik.fsf@kernel.org> <20240511184847.GCZj-9j2sh1Akpt9iS@fat_crate.local>
 <20240511184945.GDZj-9yaOEWqf1ng8u@fat_crate.local> <87h6f4jdrq.fsf@kernel.org>
 <878r0djxgc.fsf@kernel.org> <874jb0jzx5.fsf@kernel.org> <20240514160555.GCZkOL41oB3hBt45eO@fat_crate.local>
 <87msoofjg1.fsf@kernel.org> <35086bb6-ee11-4ac6-b8ba-5fab20065b54@intel.com>
 <871q60ffnr.fsf@kernel.org> <7813dff5-b140-48c4-bc15-ed25c7a07591@intel.com> <87eda0cljg.fsf@kernel.org>
In-Reply-To: <87eda0cljg.fsf@kernel.org>
From: "Rafael J. Wysocki" <rafael@kernel.org>
Date: Fri, 17 May 2024 21:08:40 +0200
Message-ID: <CAJZ5v0hJnXPMq5cjEs=yPrSU1hpJ3H3cMce8fOGQCripYfbeGQ@mail.gmail.com>
Subject: Re: [regression] suspend stress test stalls within 30 minutes
To: Kalle Valo <kvalo@kernel.org>
Cc: Dave Hansen <dave.hansen@intel.com>, Borislav Petkov <bp@alien8.de>, 
	Pawan Gupta <pawan.kumar.gupta@linux.intel.com>, Thomas Gleixner <tglx@linutronix.de>, 
	Ingo Molnar <mingo@redhat.com>, Dave Hansen <dave.hansen@linux.intel.com>, 
	"Rafael J. Wysocki" <rafael@kernel.org>, x86@kernel.org, linux-pm@vger.kernel.org, 
	linux-kernel@vger.kernel.org, regressions@lists.linux.dev, 
	Jeff Johnson <quic_jjohnson@quicinc.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Fri, May 17, 2024 at 8:59=E2=80=AFPM Kalle Valo <kvalo@kernel.org> wrote=
:
>
> Dave Hansen <dave.hansen@intel.com> writes:
>
> > On 5/17/24 11:37, Kalle Valo wrote:
> >> While writing this email I found another way to continue the suspend
> >> after a stall: terminate rtcwake with CTRL-C in the ssh session runnin=
g
> >> the for loop. That explains why 'sudo shutdown -h now' makes the suspe=
nd
> >> go forward, it most likely kills the stalled rtcwake process.
> >
> > Could we try and figure out what rtcwake is doing during its stall?  A
> > couple of ideas:
> >
> > You could strace it to see if it's hung in the kernel:
> >
> >       strace -o strace.log rtcwake ... <args here>
> >
> > You could look at its stack in /proc, like this:
> >
> > # cat /proc/`pidof sleep`/stack
> > [<0>] hrtimer_nanosleep+0xb5/0x190
> > [<0>] common_nsleep+0x44/0x50
> > [<0>] __x64_sys_clock_nanosleep+0xcb/0x140
> > [<0>] do_syscall_64+0x65/0x140
> > [<0>] entry_SYSCALL_64_after_hwframe+0x6e/0x76
> >
> > Or you can use sysrq:
> >
> >       echo t > /proc/sysrq-trigger
> >
> > to get *all* tasks' stacks dumped out to dmesg.
> >
> > I'd probably do all three in that order.
> >
> > Getting a function-graph trace of rtcwake during the stall would also b=
e
> > nice, but that's a lot of data so let's try the easier things first.
>
> I can do all that but most probably not this week. Luckily it's quite
> easy to reproduce the bug, one time I even saw it in the first iteration
> and usually within 15 minutes or so.
>
> And do let me know if there's anything else I should try.

My somewhat educated guess is that pm_notifier_call_chain_robust()
blocks for you, so you can add debug printk()s around the call to this
in suspend_prepare().

It is also possible that pm_prepare_console() does something weird and
your description of the problem indicates that it doesn't get to user
space freezing.