[Gcl-devel] Re: siginfo fault addresses now apprently working on hppa

[Carlos O'Donell]

Cam,

> > a. siginfo_t is properly delivered to 32-bit userspace from a 32-bit
> >    kernel. If it doesn't, please show me the example code, kernel config,
> >    and hardware used to run the test. This has always worked.
> > 
> 
> Well, I'm not sure how long is meant by always, but lets forget about
> the past for the moment.  Right now on paer I can get the address in
>  ((siginfo_t *)code)->si_ptr
> where the handler is defined as
> void
> memprotect_handler(int sig, long code, void *scp, char *addr)
> and setup with SA_RESTART and SA_SIGINFO.  On most machines its under 
>  ((siginfo_t *)code)->si_addr
> but not on paer.  I can show you a gdb trace if you need.

si_ptr is only valid for POSIX real-time signals, don't use that.
paer is a 64-bit box running 2.4.23-pa5, and thus does not have the
fixes required. The fact that si_ptr works is a *complete* fluke, the
kernel has shifted all values in the structure. You are really reading a
64-bit version of the structure.

Please apply pressure to have paer updated to a new 2.6.x kernel :)
 
> Thanks for this jog.  This has been on my todo list for sometime now.
> I've just implemented a runtime test for the fault recovery
> mechanism.  I'm not checking kernel versions per se, just the compile
> time determined code snippet for recovering the fault address on the
> runtime kernel.  Here's the outline:

It looks fine, you just can't run that code on a 64-bit box without a
2.6.1 or newer kernel from cvs.parisc-linux.org.

c.