On Fri, 2007-02-02 at 13:58 +0100, M. Koehrer wrote: > Hi Philippe, > > I can. Which of the many values in /proc/vmstat is the relevant one to check? >
$ vmstat 1 or $ watch -n 1 free The value in the free space column should decrease slowly but constantly while the read loop runs. > Regards > > Mathias > > > On Fri, 2007-02-02 at 12:23 +0100, Philippe Gerum wrote: > > > On Fri, 2007-02-02 at 10:51 +0100, Gilles Chanteperdrix wrote: > > > > Philippe Gerum wrote: > > > > > On Fri, 2007-02-02 at 10:33 +0100, Philippe Gerum wrote: > > > > > > > > > >>On Fri, 2007-02-02 at 09:43 +0100, M. Koehrer wrote: > > > > >> > > > > >>>Hi everybody, > > > > >>> > > > > >>>I have a strange issue frequently when reading /proc/xenomai/stat. > > > > >>>I do a cat /proc/xenomai/stat and I get an "Cannot allocate memory" > > error. > > > > >>>This happens typcially after I have started and terminated by > > application that uses a number of tasks. > > > > >>>I am able to reproduce it deterministically... > > > > >>>While the application is running, I can see the stat values > > perfectly. > > > > >>>Here is the head output of all /proc/xenomai/* files: > > > > >> > > > > >>At first sight looking at the code, there seems to be a memory leak > > > > >>caused by an unfree kmalloc() block in the sched & stat sequence > > > > >>routines. Will confirm and fix. Thanks, > > > > >> > > > > > > > > > > > > > > > Not confirmed for the leakage part regarding kmalloc(). We do have a > > > > > seq_release_private() freeing the allocated block, so back to square > > #1. > > > > > This said, some kernel memory is not released somewhere, as running > > > > > vmstat tells me here. > > > > > > > > > > Btw, is the rest of your system functional after the error occurs, or > > is > > > > > a reboot needed afterwise? > > > > > > > > Do you observe the same leakage without the nocow patch ? > > > > > > > > > > Yes, 2.6.19-1.6-04 (without nowcow) has the same problem. But the > > > interesting things is that reading /proc/interrupts in loop seems to > > > silently eat kernel memory by small chunks too. I'm now trying on a > > > vanilla kernel. > > > > > > > Confirmed. A vanilla 2.6.19 silently eats memory when > > reading /proc/interrupts in loop. Mathias, could you please confirm this > > on your box too? TIA, > > > > -- > > Philippe. > > > > > > > -- Philippe. _______________________________________________ Xenomai-help mailing list [email protected] https://mail.gna.org/listinfo/xenomai-help
