Re: notify_page_fault() problem

[Quentin Barnes <qbarnes at urbana dot css dot mot dot com>]

On Tue, May 01, 2007 at 11:40:29AM -0400, Frank Ch. Eigler wrote:
> Quentin Barnes <qbarnes@urbana.css.mot.com> writes:
>
> > [...]  That doesn't matter.  Just having the kprobes and kretprobes
> > data structures being in module memory is all that matters.  [...]
>
> I recall thinking of extending the systemtap startup code sequence to
> force mapping of the entire module data/text memory, hoping to prevent
> vmalloc faults during execution.  Does this sound like a plausible
> hack around such problems?

That would be a workaround for the bug for Systemtap modules, but it's
a Kprobe implementation problem, not a Systemtap problem.  I'd rather
not hide the problem, but to see it get addressed directly like it has
on x86_64 and i386.

> - FChE

Quentin