Re: [RFC][PATCHv2 2/4] printk: move printk_safe macros to printk header

[Sergey Senozhatsky <sergey.senozhatsky@gmail.com>]

On (10/17/18 09:57), Peter Zijlstra wrote:
> On Wed, Oct 17, 2018 at 01:32:51PM +0900, Sergey Senozhatsky wrote:
> > This probably will be a bit more hairy. logbuf is written to by many
> > sources and is read from by many sides, including user-space [both read()
> > and write()]. So we will need more flags/magic around memcpy(). A simple,
> > "grab the logbuf entry, set the proper offset to point to the next available
> > logbuf record and then do memcpy()" won't suffice. We need a flag for
> > "memcpy() complete, we can read this entry". Otherwise:
> 
> Sure, but lockless buffers mostly have reserve and commit stages anyway.
> Exactly to avoid that problem.

Right. So it should be a lockless ringbuffer supporting multiple
parallel readers and multiple parallel writers from multiple contexts
(task, sortirq, irq, NMI); with sane wrap around, etc. And we probably
need to somehow keep the existing format of logbuf entries to avoid
breakages in crash tool, and so on. It will take a lot of time, I'm
afraid.

> > All right. OK. So we are on the same page here:
> 
> > - Have more opinions on this. People please speak out.
> > - Have clear "let's do it" from Cc-ed people.
> > 
> > 
> > If we are really doing this, then let's split it and have
> > incremental changes. Namely, what I suggest is:
> 
> I'd start by replacing logbuf with the lockless buffer and ripping out
> the current nmi/safe/etc.. bollocks.
> 
> There is absolutely no point what so ever in doing anything until that
> is sorted.

Peter, you have your point, however, I think I have a slightly different
priority list. On the top of my list - deadlocks in printk(). That's the
whole reason I sent this series. Deadlocks are real. We have real reports;
in the best case in form of lockdep splats; in the worst case - nothing,
simply because the system deadlocks in printk(), and even more - printk()
can deadlock in panic(). So addressing deadlocks in printk() will fix a
rather big, real problem. Then we can have fun with the rest of the things
you have mentioned. What do you think?

	-ss