[lkml]   [2020]   [Mar]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Subject[RFC PATCH 0/3] Fix quiet console in pre-panic scenarios
Dear community,

The motivation behind this seris is to save days/weeks, if not months,
of debugging efforts for users who:

* experience an issue like softlockup/hardlockup/hung task/oom, whose
reproduction is not clear and whose occurrence rate is very low
* are constrained to use a low loglevel value (1,2,3) in production
* mostly rely on console logs to debug the issue post-mortem
(e.g. saved to persistent storage via e.g. pstore)

As pointed out in the last patch from this series, under the above
circumstances, users might simply lack any relevant logs during
post-mortem analysis.

Why this series is marked as RFC is because:
* There are several possible approaches to turn console verbosity on
and off. Current series employs the 'ignore_loglevel' functionality,
but an alternative way is to use the 'console_loglevel' variable. The
latter is more intrusive, hence the former has been chosen as v1.
* Manipulating 'ignore_loglevel' might be seen as an abuse, especially
because it breaks the expectation of users who assume the system to
be dead silent after passing loglevel=0 on kernel command line.

Thank you for your comments!

Eugeniu Rosca (3):
printk: convert ignore_loglevel to atomic_t
printk: add console_verbose_{start,end}
watchdog: Turn console verbosity on when reporting softlockup

include/linux/printk.h | 10 ++++++++++
kernel/printk/printk.c | 30 ++++++++++++++++++++++++++----
kernel/watchdog.c | 4 ++++
3 files changed, 40 insertions(+), 4 deletions(-)


 \ /
  Last update: 2020-03-15 18:10    [W:0.191 / U:13.232 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site