Messages in this thread | | | From | "Valdis Klētnieks" <> | Subject | Re: [RFC PATCH 3/3] watchdog: Turn console verbosity on when reporting softlockup | Date | Thu, 19 Mar 2020 03:38:26 -0400 |
| |
On Thu, 19 Mar 2020 15:48:36 +0900, Sergey Senozhatsky said:
> So the issue is that when we bump `console_loglevel' or `ignore_loglevel' > we lift restrictions globally. For all CPUs, for all contexts which call > printk(). This may have negative impact. Fuzzing is one example. It > usually generates a lot of printk() noise, so lifting global printk() > log_level restrictions does cause problems. I recall that fuzzing people > really disliked the > old_level = console_loglevel > console_loglevel = new_level > .... > console_loglevel = old_level > > approach. Because if lets all CPUs and tasks to pollute serial logs. > While what we want is to print messages from a particular CPU/task.
Well... how does this sound for a RFC idea? We already have CONFIG_PRINTK_TIME and CONFIG_PRINTK_CALLER. How about adding an option to allow printing the calling CPU as well, or just extend CALLER to print [pid/cpu] rather than just [pid]?
(And yes, I know that the concept of "this CPU" can change quickly, which is why per_cpu data may not be accurate for anything more than statistical aggregation purposes. It's only worth pondering how to deal with that if it's worth developing the patch at all... :)
[unhandled content-type:application/pgp-signature] | |