Re: possible mmap() problems

[Christopher Faylor <cgf at redhat dot com>]

On Wed, Apr 16, 2003 at 04:02:22PM -0400, Joe Buehler wrote:
>Christopher Faylor wrote:
>
>>If it's predictable where the page is, then I wonder if you could set a
>>hardware watchpoint to catch when it is happening?  Or maybe even strace
>>would tell you.
>
>It's a hard one.  I have a build that fails randomly.  On one run,
>out of 51 errors (handle_exceptions) caught by strace, 49 of them
>happened right after a pipe close.  I believe that they all did happen
>at the same address.  Adding strace printfs for Virtual* calls caused
>the problem to (apparently) go away (sigh).
>
>I assume I can't use hardware watchpoints because there are lots and lots
>of processes coming and going while the build is going on, and no apparent
>rhyme or reason to the problem.

Yow.  I hate those.  Do you think it could be the pipe thread thing in a
different form?  If so, I could easily hack together a patch which
reinstates the previous polling method for dealing with signals.

cgf