Re: [BUG] Kprobes fails on 2.6.25-rc3-mm1 (x86) systems, if CONFIG_DEBUG_RODATA is set.

[Srinivasa DS <srinivasa@in.ibm.com>]

On Thursday 06 March 2008 06:25:51 pm Mathieu Desnoyers wrote:
> * Srinivasa DS (srinivasa@in.ibm.com) wrote:
> > Insertion of kprobe module causes oops on 2.6.25-rc3-mm1 (x86) systems if
> > we enable CONFIG_DEBUG_RODATA.  While registering the probe, memcpy()
> > fails to copy breakpoint instruction to the instruction address and
> > generates the oops.
> >
> > BUG: unable to handle kernel paging request at ffffffff8047d1a7
> > IP: [<ffffffff8047ff62>] text_poke+0xa/0x10
> > PGD 203067 PUD 207063 PMD 7e191163 PTE 47d161
> > Oops: 0003 [1] PREEMPT SMP
> > ................................................
> >
> > This is because, Mathieu's patch (http://lkml.org/lkml/2008/2/2/226)
> > makes entire text segment as READONLY, if we enable CONFIG_DEBUG_RODATA.
> >
> > So reverting this patch or new patch, which considers kprobes while
> > deciding boundary for setting readonly pages might solve the probelm.
>
> Hi,
>
> Are you running under Xen, lguest, kvm.. ?

No, on a bare x86_64 system.   That shouldn't matter for this bug, A quick 
look at the call path shows me this.
     start_kernel()->rest_init()->kernel_init()->init_post()->mark_rodata_ro()

Srinivasa DS
>
> Mathieu
>
> > Thanks
> >  Srinivasa DS
> >  LTC-IBM