Re: Regression in v4.19.106 breaking waking up of readers of /proc/kmsg and /dev/kmsg

[John Ogness]

On 2020-02-28, Sergey Senozhatsky <sergey.senozhatsky.work@xxxxxxxxx> wrote:
> Cc-ing Petr, Steven, John

Thanks.

> https://lore.kernel.org/lkml/e9358218-98c9-2866-8f40-5955d093dc1b@xxxxxxxxxxxxxxxxxxxxxx
>
> On (20/02/27 14:08), Lech Perczak wrote:
>> >>> My test scenario for bisecting was:
>> >>> 1. run 'dmesg --follow' as root
>> >>> 2. run 'echo t > /proc/sysrq-trigger'
>> >>> 3. If trace appears in dmesg output -> good, otherwise, bad. If trace doesn't appear in output of 'dmesg --follow', re-running it will show the trace.
>> >>>
>> >>> I ran my tests on Debian 10.3 with configuration based directly on one from 4.19.0-8-amd64 (4.19.98-1) in Qemu.
>> >>> I could reproduce the same issue on several boards with x86 and ARMv7 CPUs alike, with 100% reproducibility.
>
> This is very-very odd... Hmm.
> Just out of curiosity, what happens if you comment out that
> printk() entirely?
>
> printk_deferred() should not affect the PRINTK_PENDING_WAKEUP path.

It is the printk_deferred() causing the issue. This is relatively early,
so perhaps something is not yet properly initialized.

> Either we never queue wakeup irq_work(), e.g. because
> waitqueue_active() never lets us to do so or because `(curr_log_seq !=
> log_next_seq)' is always zero

wake_up_klogd() is called and the waitqueue (@log_wait) is
active. irq_work_queue() is called, but the work function,
wake_up_klogd_work_func(), is never called.

Perhaps @wake_up_klogd_work gets broken somehow. I'm looking into it.

John Ogness