Messages in this thread |  | | | Subject | Re: kmemleak suggestion (long message) | | From | Catalin Marinas <> | | Date | Fri, 26 Jun 2009 09:25:40 +0100 |
| |
On Fri, 2009-06-26 at 08:59 +0200, Ingo Molnar wrote: > * Sergey Senozhatsky <sergey.senozhatsky@mail.by> wrote: > > Currently kmemleak prints info about all objects. I guess > > sometimes kmemleak gives you more than you actually need. > > It prints _a lot_ of info and spams the syslog. I lost crash info a > few days ago due to that: by the time i inspected a crashed machine > the tons of kmemleak output scrolled out the crash from the dmesg > buffer. > > This is not acceptable. > > Instead it should perhaps print _at most_ a single line every few > minutes, printing a summary about _how many_ leaked entries it > suspects, and should offer a /debug/mm/kmemleak style of file where > the entries can be read out from.
I agree as well. It already provides the /sys/kernel/debug/kmemleak which triggers a scan and shows possible leaks. That's easily fixable.
BTW, this was questioned in the past as well - do we still need the automatic scanning from a kernel thread? Can a user cron job just read the kmemleak file?
-- Catalin
|  |