Re: Suspend to {mem,disk} broken in 2.6.15-rc6/rc7 on my T42

[Jules Villard]

Le mar, 27 déc 2005 12:27:08 +1100, Benjamin Herrenschmidt a écrit :
> 
> > > Also, does it work if you don't use radeonfb ? radeonfb shouldn't touch
> > > MC_AGP_LOCATION and the DRM change only affects that, so I'm a bit
> > > surprised.
> > > 
> > > Ben.
> > > 
> > 
> > Do you still want me to try that now that reverting the two patches
> > made the job?
> 
> Definitely, and we need to figure out why the patch cause a regression.
> Those patches fixes a serious issues with a number of machines.

Removing radeonfb from the kernel only makes things worse: the box
gets completly frozen when reproducing the bug (no more ssh access nor
sysrq).

> 
> The problem is very nasty as all the various parties involved (radeonfb,
> X radeon driver, radeon DRM, etc...) all try to reconfigure the card
> memory map in differently bogus ways...
> 
> Can you add printk's to the kernel to check the values in
> CONFIG_MEMSIZE, CONFIG_APER_SIZE, priv->fb_location and the values
> calculated for gart_vm_start ? Then tell me what that printk gets on X
> start and when switching consoles.

I get these figures when I first start X:
[  104.399101] ### fb_location is now e0000000
[  104.399104] ### mem_size is 2000000
[  104.399107] ### aper_size is 4000000
[  104.399109] ### gart_vm_start is e2000000

The sad thing is that it looks like the crash occurs *before* entering
the radeon_do_init_cp function, assuming it should enter it again when
I switch back from a tty to X (I've put some printk's at the
beginning of the function but didn't see them in dmesg although other
things showed up), so I don't know where to put the printk's in order
to get other figures...

Thanks,

Jules

Attachment: pgp00001.pgp
Description: PGP signature