I understand there is a bug in earlier revisions of Schizo PCI bridge
(that I have in my V480) and we do not have a workaround for that. Now
when I happen to hit it, I get a crash in the handler and that seems a
bug whether or not we have the workaround.
In particular, it seems that pbm->pci_bus->self is NULL and that causes
a NULL pointer dereference in schizo_pcierr_intr_other:
pci_read_config_word(pbm->pci_bus->self, PCI_STATUS, &stat);
Since the error happens during PCI scanning, maybe we just do not have
this field initialized yet - in pci_scan_one_pbm() or in something it
calls?
I am looking into this now. Any chance I can get a prtconf dump?
Sent privately.
This schizo error handler definitely needs to be adjusted. The
bus->self value is not necessarily going to be there. In fact
often it won't be.
The PCI controller's themselves don't show up in the device tree as
bonafide PCI devices, so we don't instantiate them with the PCI
subsystem, and therefore bus->self never takes on a non-NULL value.
If you look at arch/sparc/kernel/psycho_common.c:psycho_pcierr_intr_other()
it addresses this by doing the PCI config space access by hand, thus
avoiding the bus->self dereference.
Meanwhile, this patch might allow things to get further:
It seems to work - no crash but this in dmesg, and it continues booting
fine:
[ 69.137735] /pci@9,600000: PCI Error, primary error type[Master Abort]
[ 69.144196] /pci@9,600000: bytemask[0003] was_block(0) space(Memory)
[ 69.150520] /pci@9,600000: PCI AFAR [00000000004000c0]
[ 69.155632] /pci@9,600000: PCI Secondary errors [(none)]
[ 69.160984] /pci@9,600000: PCI bus error, PCI_STATUS[22a0]
So at least this instance of the crash is gone. I seem to remember getting
some crashes later too occassionally, so that I had some first lines of
the error printed and then crash, but I do not seem to have preserved the
console logs from them. I will keep an eye open.
Thank you!
--
Meelis Roos (mroos@xxxxxxxx)
--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
