Re: dmesg -w regression in v5.4.22, bisected, was: Re: [PATCH] char/random: silence a lockdep splat with printk()

[Sergey Senozhatsky]

On (20/03/24 11:13), Zygo Blaxell wrote:
> On Wed, Nov 13, 2019 at 04:16:25PM -0500, Qian Cai wrote:
> > From: Sergey Senozhatsky <sergey.senozhatsky.work@xxxxxxxxx>
> > 
> > Sergey didn't like the locking order,
> > 
> > uart_port->lock  ->  tty_port->lock
> > 
> > uart_write (uart_port->lock)
> >   __uart_start
> >     pl011_start_tx
> >       pl011_tx_chars
> >         uart_write_wakeup
> >           tty_port_tty_wakeup
> >             tty_port_default
> >               tty_port_tty_get (tty_port->lock)
> > 
> > but those code is so old, and I have no clue how to de-couple it after
> > checking other locks in the splat. There is an onging effort to make all
> > printk() as deferred, so until that happens, workaround it for now as a
> > short-term fix.
> 
> Starting with v5.4.22 I noticed 'dmesg -w' stopped working on some
> machines.  dmesg will follow console output for a few seconds, then it
> stops.  strace indicates dmesg is blocked in read() on the /dev/kmsg fd.
> If a new dmesg process starts, it gives messages for a few seconds,
> then also stops.  rsyslog's kernel logging is similarly affected.
> 
> Bisection points to this patch (now known as
> 1b710b1b10eff9d46666064ea25f079f70bc67a8 upstream).  I can't reproduce
> the problem on a test VM, and some machines are running v5.4.22..v5.4.26
> with no dmesg problems.  It seems there is some magic in the startup
> sequence of affected machines.  This code isn't executed after RNG is
> seeded, so it would have to get its bad stuff done before that happens.
> 
> Reverting commit 1b710b1b10eff9d46666064ea25f079f70bc67a8 fixes the
> dmesg regression on 5.4.26.  It might put the original lockdep bug back,
> but on machines running stable kernels, I prefer randomly broken lockdep
> over repeatably broken dmesg.

This should fix the problem

https://lore.kernel.org/lkml/20200303113002.63089-1-sergey.senozhatsky@xxxxxxxxx

	-ss