On 2020/04/29 23:21, Michal Hocko wrote: > I am trying to say that KERN_NO_CONSOLES resembles more a policy than a > priority. Because I as a developer have no idea whether the message is > good enough for console or not.
Right, KERN_NO_CONSOLES is not a priority. > I believe we are free to change kernel log levels as we find a fit. I > was not aware that KERN_DEBUG messages are automatically filtered out. Below is the default rules for rsyslog-8.24.0-52.el7 (userspace syslog daemon). Of course administrators can modify as needed, but notice that KERN_INFO is saved to /var/log/messages but KERN_DEBUG is saved to nowhere. ---------- # Log all kernel messages to the console. # Logging much else clutters up the screen. #kern.* /dev/console # Log anything (except mail) of level info or higher. # Don't log private authentication messages! *.info;mail.none;authpriv.none;cron.none /var/log/messages # The authpriv file has restricted access. authpriv.* /var/log/secure # Log all the mail messages in one place. mail.* -/var/log/maillog # Log cron stuff cron.* /var/log/cron # Everybody gets emergency messages *.emerg :omusrmsg:* # Save news errors of level crit and higher in a special file. uucp,news.crit /var/log/spooler # Save boot messages also to boot.log local7.* /var/log/boot.log ---------- > Even if this is the case then this doesn't really disallow admins to > allow KERN_DEBUG into log files. Dump of the oom eligible tasks is > arguably a debugging output anyway. So I disagree with your statement. If dump_tasks() were changed to use KERN_DEBUG, administrators have to add "kern.debug" rule (at the same time endure a lot of noise from KERN_DEBUG) in order to record OOM victim candidates for later analysis. > >> If the kernel allows the former to use KERN_NO_CONSOLES in addition to >> KERN_INFO, the administrator can >> select from two choices: printing "both the former and the latter" or "only >> the latter" to consoles. > > I am not really familiar with all the possibilities admins have when > setting filtering for different consoles but KERN_NO_CONSOLES sounds > rather alien to the existing priority based approach. KERN_NO_CONSOLES is not a priority based approach. KERN_NO_CONSOLES resembles CONSOLE_LOGLEVEL_SILENT. > You can fine tune > priorities and that is all right because they should be reflecting > importance. Developer's importance and administrator's interests are different. Any printk() user is randomly selecting KERN_$LEVEL. Administrators are swayed by having to record the lowest priority from all interested messages. > But global no-consoles doesn't really fit in here because > each console might require a different policy but the marking is > unconditional and largely unaware of existing consoles. Why unconditional? I'm saying that users of KERN_NO_CONSOLES marking (in other words, "whether the message is good enough for console or not") should be configurable via e.g. sysctl. If administrators want to use per-console loglevel setting, they can tell the kernel not to mark KERN_NO_CONSOLES via e.g. sysctl.
