[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5DB7DD91-7213-4A30-B89E-FB28AD96F8C5@fb.com>
Date: Wed, 17 May 2023 04:38:52 +0000
From: Song Liu <songliubraving@...a.com>
To: "Li, Xin3" <xin3.li@...el.com>
CC: Song Liu <song@...nel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Kernel Team <kernel-team@...a.com>,
Andrew Morton <akpm@...ux-foundation.org>,
Peter Zijlstra <peterz@...radead.org>
Subject: Re: [PATCH v2] watchdog: Prefer use "ref-cycles" for NMI watchdog
> On May 16, 2023, at 6:23 PM, Li, Xin3 <xin3.li@...el.com> wrote:
>
>> NMI watchdog permanently consumes one hardware counters per CPU on the
>> system. For systems that use many hardware counters, this causes more
>> aggressive time multiplexing of perf events.
>>
>> OTOH, some CPUs (mostly Intel) support "ref-cycles" event, which is rarely
>> used. Try use "ref-cycles" for the watchdog, so that one more hardware
>> counter is available to the user. If the CPU doesn't support "ref-cycles",
>> fall back to "cycles".
>>
>> The downside of this change is that users of "ref-cycles" need to disable
>> nmi_watchdog.
>
> From the discussion in v1, the users don't have to disable the NMI watchdog
> *permanently*, right?
The users need to disable NMI watchdog when using ref-cycles. For example:
# disable nmi_watchdog
sysctl kernel.nmi_watchdog=0
# use ref-cycles
perf stat/record -e ref-cycles ...
# reenable nmi_watchdog
sysctl kernel.nmi_watchdog=1
Thanks,
Song
Powered by blists - more mailing lists