On Thursday 15 June 2006 09:40, Piet Delaney wrote: > On Thu, 2006-06-15 at 09:19 +0530, Amit S. Kale wrote: > > Steve, > > > > You are using tg3 driver which runs in NAPI mode. KGDB won't work with > > NAPI. > > Why is that?
This has to do with the fact that kgdb as well as NAPI use tasklets, which are serialized by definition. > > A few months ago I was using the e1000 for kgdb; now I'm using the e100. > I've got E1000 enabled with NAPI so I likely was then. So the e1000 > likely didn't have this limitation. That could be the case. We found some problems recently on e1000 running in NAPI mode. I don't know exactly which kernel or kgdb release this started with. > > Since the tg3 is a very commonly used ethernet interface, shouldn't > this limitation be mentioned in the release notes? It's worth mentioning in the next set of release notes. -Amit > > I think I recall the author of the kgdboe code mentioning ethernet > drive limitations when he first submitted it to the 2.6 mm patch > series. > > -piet > > > Unfortunately tg3 driver can't run in a non-NAPI (interrupt driven) mode, > > so you might have to buy a new card. > > -Amit > > > > On Thursday 15 June 2006 06:57, Andi Kleen wrote: > > > On Wed, Jun 14, 2006 at 04:47:58PM -0700, Tom Rini wrote: > > > > > Then all of our stuff should be #ifdef KGDB; while merging from > > > > > 2.6.12 to 2.6.13 I recall it noticing a few of the #ifdef KGDB's > > > > > had been removed. > > > > > > > > Yes, there really shouldn't be any #ifdef KGDB's in the kernel tree > > > > and we're down to very few[1] and I think it's really just one in the > > > > end we'll need. > > > > > > That was __might_sleep()? I'm still dubious about that one. It might > > > be possible to just change the caller. Do you have a full backtrace > > > for it? > > > > > > I suppose you're doing a *_user in kgdb somewhere, right? > > > > > > -Andi _______________________________________________ Kgdb-bugreport mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/kgdb-bugreport
