Re: Can printk() sleep at runtime?

[Jia-Ju Bai]

On 2018/6/1 5:13, Steven Rostedt wrote:
> On Thu, 31 May 2018 18:42:48 +0200
> Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>
> > Hi Stephen,
> >
> > On Thu, May 31, 2018 at 5:19 PM, Stephen Boyd <sboyd@xxxxxxxxxx> wrote:
> > > Quoting Linus Torvalds (2018-05-31 07:32:10)
> > > > On Thu, May 31, 2018 at 5:05 AM Petr Mladek <pmladek@xxxxxxxx> wrote:
> > > > > Anyway, we need to fix or remove this format. vsprintf-like functions
> > > > > are called in any context and nobody expect that they might sleep.
> > > > Ack. I guess the argument is that "%pCr" is rare, and none of *those*
> > > > users may care, but I do think that doing things wrong as-is.
> > > >
> > > > It's too subtle to have to know you're in a particular locking context
> > > > when you use a particular %p modifier.
> > > Agreed. Removing the format seems to be the best approach. It looks like
> > > only Geert has used it in the last few years and it hasn't been used
> > > much otherwise.
> > Indeed, just 3 users (the broadcom one isn't mine):
> >      drivers/clk/renesas/renesas-cpg-mssr.c
> >      drivers/thermal/broadcom/bcm2835_thermal.c
> >      drivers/tty/serial/sh-sci.c
> >
> > Alternatively, can we have a special version __clk_get_rate() that just
> > returns clk->core->rate?
> > Or would that be too inaccurate in the presence of CLK_GET_RATE_NOCACHE?
> > The function could still return 0 in case the flag is set.
> If it's only used in three locations, I think it would be better to
> simply remove it from vsprintf() and have the three callers call
> clk_get_rate() directly.

Agreed.


Best wishes,
Jia-Ju Bai