port-sparc64/46461: Sun E450 drops to DDB on boot of 6.0_BETA

[he%NetBSD.org@localhost]

>Number:         46461
>Category:       port-sparc64
>Synopsis:       Sun E450 drops to DDB on boot of 6.0_BETA
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-sparc64-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu May 17 21:30:00 +0000 2012
>Originator:     Havard Eidnes
>Release:        NetBSD 6.0_BETA
>Organization:
        None
>Environment:
System: NetBSD torbidinsu.urc.uninett.no 6.0_BETA NetBSD 6.0_BETA (TORBIDINSU) 
#0: Thu May 17 17:55:31 CEST 2012 
he%torbidinsu.urc.uninett.no@localhost:/usr/obj/sys/arch/sparc64/compile/TORBIDINSU
 sparc64
Architecture: sparc64
Machine: sparc64
>Description:
        Attempting to upgrade from 5.99.38 to 6.0_BETA, I notice that
        this E450 now drops to DDB during the autoconf phase.

        In particular, it gets this far:

psycho4 at mainbus0 addr 0xfff36000
psycho4: SUNW,psycho: impl 0, version 4: ign 180 bus range 0 to 0; PCI bus 0
pci4 at psycho4
pm2fb0 at pci4 dev 4 function 0: 3D Labs GLINT Permedia 2V (rev. 0x01)
pm2fb0: 8 MB aperture at 0x00800000
text_access_fault: pc=0 va=0
kernel trap 64: +fast instruction access MMU miss
Stopped in pid 0.1 (system) at  0:      undefined
db{0}: tra
config_attach_loc(6d3dc40, 1c080e8, 1, 2005428, 0, 1387c00) at 
netbsd:config_attach_loc+0x1a0
pci_probe_device(6dbc000, f009e59000002000, 0, 93d3d, ff00, 2005490) at 
netbsd:pci_probe_device+0x448
sparc64_pci_enumerate_bus(6dbc000, 174cdd0, 0, 21, 1c75c00, 179d800) at 
netbsd:sparc64_pci_enumerate_bus+0x2e8
pcirescan(6dbc000, 1791f18, 174cdd0, 17f6ee8, 490, 6d30137) at 
netbsd:pcirescan+0x4c
pciattach(1, 6d30100, 20058a0, 1cb2950, 1791c00, 5ddb680) at 
netbsd:pciattach+0x15c
config_attach_loc(6d30100, 1c07848, 1, 20058a0, 0, 13a82a0) at 
netbsd:config_attach_loc+0x1a0
psycho_attach(5b41200, 6d30400, 2005ab0, 1cb2950, 1791c00, 200592c) at 
netbsd:psycho_attach+0x4ec
config_attach_loc(6d30400, 1c077b8, 1, 2005ab0, 0, 1093020) at 
netbsd:config_attach_loc+0x1a0
mainbus_attach(0, 5b41200, 0, 1cb2950, 179d800, 1093000) at 
netbsd:mainbus_attach+0x4a8
config_attach_loc(5b41200, 1c07308, 1, 0, 0, 7) at 
netbsd:config_attach_loc+0x1a0
cpu_configure(17bfc00, 12a7420, 0, 5b2df80, 35, 1c09c00) at 
netbsd:cpu_configure+0x20c
main(0, 1226320, fff7fa80, f0075a4c, fffffffff00874a0, ff00) at 
netbsd:main+0x314
cpu_initialize(f0075a4c, 0, 10fc00, fff7fd00, 10e9d8, fff7fe00) at 
netbsd:cpu_initialize+0xe8
db{0}: 
        
        NetBSD 5.99.38 prints this at this point, and continues:

psycho4 at mainbus0 addr 0xfff36000
psycho4: SUNW,psycho: impl 0, version 4: ign 180 bus range 0 to 0; PCI bus 0
pci4 at psycho4
pm2fb0 at pci4 dev 4 function 0: 3D Labs GLINT Permedia 2V
pm2fb0: no width property
psycho5 at mainbus0 addr 0xfff48000

        This tends to point to the pm2fb0 driver, 

        Why it said "no width property" in 5.99.38 and then suddenly
        it finds it in 6.0_BETA will excercise the latter parts of
        pm2fb_attach().

        I should perhaps note that the serial port is the console, so
        the pm2fb0 isn't, this determines the branch in the is_console
        test.  The o7 register points to that, at least:

db{0}: show reg
...
pc          0
...
o7          1397648     pm2fb_attach+0x748
...

        That's the only register value which decodes to a symbol+offset.

        A dmesg from 5.99.38 will be submitted as a comment after this
        initial submission.

>How-To-Repeat:
        Try to boot a Sun E450 with graphics card but not console on
        graphics, watch it drop to DDB.

>Fix:
        Sorry, don't know.
        The 5.99.38 kernel is dated Sep 1 2010, at the time built from
        pretty fresh sources, and quite a few changes have gone into
        the pm2fb driver since then, among them i2c code.