[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZeBoTNe2cjb4BeXb@alley>
Date: Thu, 29 Feb 2024 12:19:40 +0100
From: Petr Mladek <pmladek@...e.com>
To: "John B. Wyatt IV" <jwyatt@...hat.com>
Cc: John Ogness <john.ogness@...utronix.de>,
Clark Williams <williams@...hat.com>, jlelli@...hat.com,
Derek Barbosa <debarbos@...hat.com>,
"John B. Wyatt IV" <sageofredondo@...il.com>,
LKML <linux-kernel@...r.kernel.org>,
linux-rt-users <linux-rt-users@...r.kernel.org>
Subject: Re: NMI Reported with console_blast.sh
On Thu 2024-02-29 12:15:30, Petr Mladek wrote:
> On Thu 2024-02-22 00:21:19, John B. Wyatt IV wrote:
> > On Mon, Feb 12, 2024 at 04:23:04PM -0500, John B. Wyatt IV wrote:
> > >
> > > Red Hat's conservers are having an issue with the machine I was testing
> > > on. It may take me a while to get back to you with more test results.
> > >
> >
> > Found a work-around with conserver. I was able to follow up on the printk-caller
> > info you requested.
> >
> > I found 2 additional NMIs for a total of 3. Number 2 is very
> > large-please feel free to let me know what specific information you
> > wanted if it was unnecessary.
[...]
> > Compared to the two NMIs with throughput-performance (no preemption)
> >
> > <NMI>
> > cpus=0
> > .runnable_avg : 3072
> > kthread (kernel/kthread.c:388)
> > .util_est_enqueued : 0
> > stack:0 pid:1733 tgid:1733 ppid:2 flags:0x00004000
> > .min_vruntime : 2084315.290254
> > .removed.load_avg : 0
> > .avg_vruntime : 2084315.290254
> > console_blast.s 3497 34770.405603 N 34773.405603 3.000000 34764.898340 4 120
> > .util_avg : 1024
> > .util_avg : 1024
>
> It looks like messages from more (many) CPUs are mixed. I guess that they
> are printed by print_cfs_rq(). But the order looks random.
>
> Also I wonder why it is printed from NMI context. Maybe, it is from
> some perf event, similar to hardlockup detector?
I have realized that we most likely see only small part of the mixed
output. I wonder if it is because it is printed from the emergency
context. Here the messages are flushed when leaving the context
and many might be lost.
Best Regards,
Petr
Powered by blists - more mailing lists