[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200211222754.0185b9b3@rorschach.local.home>
Date: Tue, 11 Feb 2020 22:27:54 -0500
From: Steven Rostedt <rostedt@...dmis.org>
To: Sergey Senozhatsky <sergey.senozhatsky.work@...il.com>
Cc: Petr Mladek <pmladek@...e.com>,
Peter Zijlstra <peterz@...radead.org>,
linux-kernel@...r.kernel.org,
Sergey Senozhatsky <sergey.senozhatsky@...il.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Dmitry Monakhov <dmtrmonakhov@...dex-team.ru>,
Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
Subject: Re: [PATCH] kernel/watchdog: flush all printk nmi buffers when
hardlockup detected
On Wed, 12 Feb 2020 12:04:03 +0900
Sergey Senozhatsky <sergey.senozhatsky.work@...il.com> wrote:
> Oh, yes, ftrace printks a lot. But I sort of forgot why don't we do
> the same for "regular" NMIs. So NMIs use per-cpu buffers, expect for
> NMIs which involve ftrace dump. I'm missing something here.
Well, ftrace_dump() is called from NMI context when the system is about
to crash (ftrace_dump_on_oops). And at that moment, we care more about
the trace than other output (it's most like to contain the information
that caused the bug).
But for things like sysrq-l, that does a printk in NMI context for
normal operations, we don't want strange races to occur and affect
other printk operations. Having them in buffers controls the output a
bit better.
But with the new printk ring buffer work, perhaps that's no longer
necessary.
-- Steve
Powered by blists - more mailing lists