kern/38698: NMI during boot on Fujitsu-Siemens Lifebook S-4546

[gson%gson.org@localhost (Andreas Gustafsson)]

>Number:         38698
>Category:       kern
>Synopsis:       NMI during boot on Fujitsu-Siemens Lifebook S-4546
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon May 19 12:15:00 +0000 2008
>Originator:     Andreas Gustafsson
>Release:        NetBSD 4.99.62 as of 20080513-1503 EET
>Organization:
>Environment:
System: NetBSD 4.99.62
Architecture: i386
Machine: i386
>Description:

When I boot a NetBSD-current installation CD on a Fujitsu-Siemens Lifebook
S-4546 laptop, it drops into DDB with an NMI during autoconfiguration.

The final messages (transcribed by hand) are:

  cbb0 at pci0 dev 19 function 0: vendor 0x1217 product 0x6933 (rev. 0x02)
  cbb1 at pci0 dev 19 function 1: vendor 0x1217 product 0x6933 (rev. 0x02)
  vga1 at pci0 dev 20 function 0: vendor 0x1002 product 0x4c52 (rev. 0x64)
  wsdisplay0 at vga1 kbdmux 1: console (80x25, vt100 emulation)
  drm at vga1 not configured
  NMI ... going to debugger
  Stopped in pid 0.1 (system) at  netbsd:outl+0x9:        ret
  db{0}> 

The first few locations shown in the stack trace are

  outl+0x9
  pccbb_chipinint+0x3b5
  pccbb_pci_callback+0xde
  config_process_deferred+0x44

The instruction at outl+0x8 is "outl %eax,%dx".  According to "show
registers", eax is 0x0, and edx is 0xcf8.  I believe this outl() is
part of the pci_conf_write() to set the CardBus latency timer at line
759 of pccbb.c.

Booting a 4.0 release CD works.

>How-To-Repeat:

>Fix:

The following change may be more of a workaround than a fix, but
it make the machine boot successfully.

Index: pccbb.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/pccbb.c,v
retrieving revision 1.167
diff -u -r1.167 pccbb.c
--- pccbb.c     4 Mar 2008 22:15:16 -0000       1.167
+++ pccbb.c     19 May 2008 12:04:56 -0000
@@ -746,7 +746,9 @@
        /* I believe it is harmless. */
        csr |= (PCI_COMMAND_IO_ENABLE | PCI_COMMAND_MEM_ENABLE |
            PCI_COMMAND_MASTER_ENABLE);
+#if 0
        csr |= (PCI_COMMAND_PARITY_ENABLE|PCI_COMMAND_SERR_ENABLE);
+#endif
        pci_conf_write(pc, tag, PCI_COMMAND_STATUS_REG, csr);
 
        /*