Re: [PATCHv2] printk: queue wake_up_klogd irq_work only if per-CPU areas are ready

[Sergey Senozhatsky]

On (20/03/04 16:21), Petr Mladek wrote:
[..]
> > Fix printk_deferred() and do not queue per-CPU irq_work
> > before per-CPU areas are initialized.
> >
> > [0] https://lore.kernel.org/lkml/aa0732c6-5c4e-8a8b-a1c1-75ebe3dca05b@xxxxxxxxxxxxxxxxxxxxxx/
> >
> > Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@xxxxxxxxx>
> > Reported-by: Lech Perczak <l.perczak@xxxxxxxxxxxxxxxxxxxxxx>
> > Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> > Cc: Theodore Ts'o <tytso@xxxxxxx>
> > Cc: John Ogness <john.ogness@xxxxxxxxxxxxx>
>
> Reviewed-by: Petr Mladek <pmladek@xxxxxxxx>

Thanks!

> Now, the question is whether to hurry this fix into 5.6 or if
> it could wait for 5.7.
>
> I think that it could wait because 5.6 is not affected by
> the particular printk_deferred(). This patch fixes a long-term
> generic problem. But I am open for other opinions.

Good question. My 5 cents, I would _probably_ push it now. Not
because it fixes any known issues on 5.6, but because we have
a number of LTS kernel (4.19, 4.14, 4.9, 4.4, 3.16) that can be
affected should 1b710b1b10eff9d4 be backported to those kernels.
Which is quite likely, I suspect. The sooner we fix printk_deferred(),
the sooner -stable/LTS picks up the fix, so that we don't have same
regression reports in the future. The regression in question is
pretty hard to track down, git-bisect, perhaps, is the only reasonably
fast way.

	-ss