On Fri, Feb 03, 2006 at 02:46:33PM +0100, Paul Slootman wrote: > On Fri 03 Feb 2006, Steve Langasek wrote: > > On Fri, Feb 03, 2006 at 01:03:07PM +0100, Paul Slootman wrote:
> > > I did an upgrade yesterday, and since then I'm being bombarded with > > > "Segfault in Samba" email messages. > > Did those emails not include a usable backtrace? (Possibly not, since I'm > > told gdb doesn't handle -fPIE very well, which is now a default upstream. > > :/) > No, "process already gone" or something like that. Ah, hmm. Yes, that wouldn't be an -fPIE problem then; you could probably track it down by running samba under gdb, though you'd still have the -fPIE problem on top of this. > > > On sparc, unfortunately, I don't know how to track these down :-( > > > I'll try to reproduce it on an alpha (the kernel there is helpful enough > > > to report such accesses and to fix them up, although there's a > > > significant speed penalty). > > Eh, it's far easier to debug a sigbus on sparc than it is on alpha, since > > you have to go to so much effort just to get the signal to userspace on > > alpha... (which, completely randomly, I'm working on fixing by porting prctl > > to alpha, but that'll require a new kernel anyway) > Well, on alpha you get a kernel message with the address of the > offending instruction, with a suitably compiled executable (-g) and gdb > you can list the source line number where it occurs. Well, true. I don't think that's easier than trapping the SIGBUS with gdb itself when it happens, though. > > I remember seeing some comments about Sparc SIGBUS issues recently on the > > upstream mailing list; so with luck this is already fixed upstream. > Ah, that would be good. > I'm getting a bit frustrated with the SIGBUSes on sparc, 'aptitude > install non-existent-package' also suffers from it. And mysql, if it > tries to resolve a hostname :-( :-) Yes, a lot of this is probably due to a behavior change in gcc4, which seems to now frequently put char arrays on the stack at 16-bit-aligned addresses instead of 32-bit-aligned addresses. Strange, perfectly valid, and breaks all kinds of code on sparc that assumes casting to int* is going to work... Anyway, we should see if this bug goes away with samba 3.0.21b before spending too much painful effort trying to track it down. -- Steve Langasek Give me a lever long enough and a Free OS Debian Developer to set it on, and I can move the world. [EMAIL PROTECTED] http://www.debian.org/
signature.asc
Description: Digital signature
