printk: do cond_resched() between lines while outputting to consoles
authorTejun Heo <[email protected]>
Sat, 16 Jan 2016 00:58:24 +0000 (16:58 -0800)
committerLinus Torvalds <[email protected]>
Sat, 16 Jan 2016 19:17:25 +0000 (11:17 -0800)
commit8d91f8b15361dfb438ab6eb3b319e2ded43458ff
tree2462bccec78ae730a5e0801dbb636cb0710a6c88
parent81cc26f2bd11ba4421a17a2d5cebe4bba206c239
printk: do cond_resched() between lines while outputting to consoles

@console_may_schedule tracks whether console_sem was acquired through
lock or trylock.  If the former, we're inside a sleepable context and
console_conditional_schedule() performs cond_resched().  This allows
console drivers which use console_lock for synchronization to yield
while performing time-consuming operations such as scrolling.

However, the actual console outputting is performed while holding
irq-safe logbuf_lock, so console_unlock() clears @console_may_schedule
before starting outputting lines.  Also, only a few drivers call
console_conditional_schedule() to begin with.  This means that when a
lot of lines need to be output by console_unlock(), for example on a
console registration, the task doing console_unlock() may not yield for
a long time on a non-preemptible kernel.

If this happens with a slow console devices, for example a serial
console, the outputting task may occupy the cpu for a very long time.
Long enough to trigger softlockup and/or RCU stall warnings, which in
turn pile more messages, sometimes enough to trigger the next cycle of
warnings incapacitating the system.

Fix it by making console_unlock() insert cond_resched() between lines if
@console_may_schedule.

Signed-off-by: Tejun Heo <[email protected]>
Reported-by: Calvin Owens <[email protected]>
Acked-by: Jan Kara <[email protected]>
Cc: Dave Jones <[email protected]>
Cc: Kyle McMartin <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
include/linux/console.h
kernel/panic.c
kernel/printk/printk.c