cvs commit: src/sys/dev/pci pci.c pci_if.m pci_private.h pcivar.h src/sys/dev/sk if_sk.c if_skreg.h

[Oleg Bulyzhin]

On Mon, Oct 16, 2006 at 10:05:17AM -0700, John-Mark Gurney wrote:
> Oleg Bulyzhin wrote this message on Mon, Oct 16, 2006 at 12:14 +0400:
> > On Mon, Oct 09, 2006 at 04:15:56PM +0000, John-Mark Gurney wrote:
> > > jmg         2006-10-09 16:15:56 UTC
> > > 
> > >   FreeBSD src repository
> > > 
> > >   Modified files:
> > >     sys/dev/pci          pci.c pci_if.m pci_private.h pcivar.h 
> > >     sys/dev/sk           if_sk.c if_skreg.h 
> > >   Log:
> > >   provide routines to access VPD data at the PCI layer...
> > >   
> > >   remove sk's own implementation, and use the new calls to get the data...
> > >   
> > >   Reviewed by:    -arch
> > >   
> > >   Revision  Changes    Path
> > >   1.315     +339 -3    src/sys/dev/pci/pci.c
> > >   1.9       +13 -0     src/sys/dev/pci/pci_if.m
> > >   1.18      +4 -0      src/sys/dev/pci/pci_private.h
> > >   1.71      +34 -0     src/sys/dev/pci/pcivar.h
> > >   1.131     +7 -148    src/sys/dev/sk/if_sk.c
> > >   1.39      +0 -31     src/sys/dev/sk/if_skreg.h
> > 
> > I have problem with my test machine since this commit:
> > kernel is panicing on boot if i have my pci bge(4) NIC plugged in.
> > 
> > Last kernel messages are:
> > pci1: physical bus=1
> > pci1:2:0: bad VPD cksum, remain 244
> > 
> > Invoking ddb after panic gives this backtrace:
> > [skipped]
> > pci_read_vpd
> > pci_read_extcap
> > pci_read_device
> > pci_add_children
> > acpi_pci_attach
> > device_attach
> > [skipped]
> > 
> > (i'm unable to get crashdump)
> > 
> > If i unplug bge, kernel boots just fine.
> > 
> > P.S. i can provide any additional info needed and can test patches.
> 
> Can you get a line number from pci_read_vpd?  Even if you can't get a
> crash dump, you can use addr2line (or kgdb) w/ the ip of the panic...
> That would help..
> 
> Looks like some manufacturers aren't following the PCI standard.. :(
> 
> -- 
>   John-Mark Gurney				Voice: +1 415 225 5579
> 
>      "All that I will do, has been done, All that I have, has not."

Okay, there is some details:

When kernel boots with bge plugged in:
1) message about bad VPD cksum
2) then kernel seems to hang for about 30 seconds (i'm not able to break into
DDB)
3) then i get panic.

Panic message is about trap inside 'swapper' process (backtrace shows 2 traps
first inside pci_read_vpd, next one in swapper).

1st trap is at pci_read_vpd+0x2c6 it is /usr/src/sys/dev/pci/pci.c:665
I've added debug printf there:
		case 3: /* VPD-R Keyword Value */
			printf("off: %d i: %d\n", off, i);
			cfg->vpd.vpd_ros[off].value[i++] = byte;

It seems that 30 seconds delay is memory rewriting (off is constant, while i
is incrementing).  I get panic message after:
off:5 i:15407

-- 
Oleg.