Re: [PATCH] [PARISC] Clean up syslog output of crashing userspace apps.

Helge Deller <[email protected]> Sun, 15 Jul 2007 19:35:27 +0200
Newsgroups gmane.linux.ports.hppa
Message-ID <[email protected]>
Hi Randolph,

On Sunday 15 July 2007, Randolph Chung wrote:
> In reference to a message from Helge Deller, dated Jul 15:
> > Clean up syslog output of crashing userspace apps.
> > With this patch the syslog will print:
> >   portmap(1480): invalid access to 0x8232b578 at ip=0x41192af3 (segfault code 15)
> > which is consistent with outher output like:
> >   hald-probe-stor(1672): unaligned access to 0xfafcbd26 at ip=0x00014013
> 
> The verbose dump we have now is actually quite useful for debugging.
> Whereas unaligned accesses are usually very straightforward to debug
> once you know where they are coming from (thus only need ip), segfaults
> are not always that way. The full register dump comes in handy...

I do not disagree, that dumps are useful in general.
Nevertheless, is this specific dump really that useful ?
Can you really fix bugs with the dump without use of gdb ?
On i386 userspace crashes aren't even shown in the syslog.
 
> Can we make this controllable by prctl, so that we can have both verbose
> and terse output options, with the verbose one being the default? My
> rationale is that you don't usually know if an app is going to crash, so
> it's better to have the debug info you need if it does crash, and you
> can turn it off if your app runs in a way that the verbose output is
> somehow adversely affecting it (slowing it down too much, etc).

Good idea.
But again, on i386 there isn't shown anything.
Do we really want the overhead to add a new prctl option ?

> Also, you might want to remove "segfault" from the output string, since
> the crash can be from SIGSEGV, SIGILL, SIGBUS, etc...

Ok.

Helge