Messages in this thread | | | Date | Tue, 23 Oct 2007 19:49:45 +0300 | From | Sami Farin <> | Subject | Re: [patch 1/2] oProfile: oops when profile_pc() return ~0LU |
| |
On Tue, Oct 23, 2007 at 18:13:21 +0200, Philippe Elie wrote: > On Tue, 23 Oct 2007 at 13:10 +0000, Sami Farin wrote: > > > On Mon, Oct 22, 2007 at 19:38:10 -0700, Linus Torvalds wrote: > > > > > > This set of two patches look ok by me, but I'd like sign-offs.. Also, were > > > they tested and found to fix the problem by Sami? > > > > > > Linus > > For the signed-offs I thought the From: was an implicit Signed-offs. > > Test was done privately, Sami helped to narrow down the trouble, but > he didn't test the last patch, nothing bad on Sami side, I was too > confident the fix was obvious after narrowing it. > > > > > The previous patch I tested by Philippe, oprof-fix-profile_pc-use.patch, > > worked ok, but with this latest patch oprofiled aborts. > > But kernel does not oops or print msgs. > > argh, I just moved the wrong eip from kernel to user space where the same > problem occur too, *sighs*, since I can't reproduce Sami problem, my own > test obviously worked... > > Sami, can you test this new patch. After testing can you report > the contents of /dev/oprofile/stats/cpu*/sample_invalid_eip ?
cat /dev/oprofile/stats/cpu?/sample_invalid_eip; sleep 10; cat /dev/oprofile/stats/cpu?/sample_invalid_eip 834 835 0 0 906 911 0 0
For some reason there are four directories, but I have only two CPUs in reality.
And oprofiled survives the test OK.
> Linus, there is two way to fix this problem, the attached patch fix it > by sanitizing the sampled eip, the other is to replace the use of > profile_pc(); by instruction_pointer(); in cpu_buffer.c, that one was > tested by Sami but 1) it'll break the 'use oprofile as a sort of lockometer' > 2) I think sanitizing the eip will be necessary anyway as I'm not really > confident than instruction_pointer() can never return weird eip on some > weird arch and/or some weird circumstances.
-- Do what you love because life is too short for anything else.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |