Re: SGI O2 R12 NetBSD5.1 - good but some problems

On 11 Jan 2011, at 20:13, Stephen M. Rumble wrote:

> On Jan 11, 2011, at 12:02 PM, [email protected] wrote:
> 
>> I'll have to cross compile a kernel on my PC, enabling some of the
>> debugging options.  My friend who has more experience in these matters
>> tells me the kind of fault doesn't sound like one we're going to get
>> much out of the debugger.  The fault is a freeze, a tight loop rather
>> than a kernel fault or oops.  But we can try.
> 
> It's probably worth debugging on the matt-nb5-mips64 branch, or -current if it gets merged soon. The MIPS code is set to significantly change very soon.

Thanks for pointing out this branch.  I successfully compiled two
trees, One of -current and one of matt-nb5-mips64.

I have three reports here of the three kernels (stock, 5.99.43 and
matts).  All on the standard 5.1 software base.

It's really cool there is some really good work here.  The
stock 5.1 experience is still the best, with working Xrender.  All
kernels are hit and miss when the system will hard lock.

Running ntpd seems to cause problems, so I've stopped running it.

Hopefully this admittedly poor reporting will be helpful to
someone.  As I get better with this I'll try and get more empirical.
I did try each of these kernels a few times after they failed.
Overall failure is random :(


5.1 GENERIC32_IP3x
==================
 * my mouse problem fixed with a different mouse
 * kernel has reasonable stability, with occasional hard lock
 * SCSI bus resets pausing the card, preventing further disk I/O.
 * CPU error 4 not limited to network transfer
 * CPU error 4 more frequent when under network transfer

Also noticed wsmoused(1) and X do not get on, denoted by screen 
corruption with mouse movement, and dmesg listing hundreds of
crime: memory error address 4294958976 status 33591312
crime: memory error address 4294958976 status 33591312
...

Disabling rc,conf wsmoused=NO solves this problem. 

System feels nice and snappy, running window maker until the 
SCSI bus stopped.

If this is useful:

(the following quote is recreated by hand) <<EOF
sg[0] - Addr 0x603207 : Length 524416
ach0:BDR message in message buffer
sd0(ahc0:0:2:0) ach0: no longer in timeout, status = 0
ahc0:A:2: Missed bussfree. Lastphase = 0xe0, Curphase = 0xb6
ahc0:SCB 0x4 - timed out
>>>>> Dump Card State Begins <<<<<
ahc0: Dumping Card State while idle, as SEQADDR 0x18
Card was paused
ACCUM = 0x4, SINDEX = 0x20, DINDEX = 0xc0, ARG_2 = 0x1
HCNT = 0x0 SCBPTR = 0xa
SCSISIGI[0xa6] ERROR[0x0] SCSIBUSL[0xb] LASTPHASE[0x1]
SCSISEQ[0x1a] SBLKCTL[0x2] SCSIRATE[0x0] SEQCTL[0x10]
SEQ_FLAGS[0xc0] SSTAT0[0x5] SSTAT1[0x13] SSTAT2[0x0]
SSTAT3[0x0] SIMODE0[0x0] SIMODE1[0xa4] SXFRCTL0[0x80]
DFCNTRL[0x4] DFSTATUS[0x6d]
STACK: 0x45 0x0 0x19c 0x17
SCB count = 32
Kernel NEXTQSCB = 5
Card NEXTQSCB = 5
QINFIFO entries:
Waiting Queue entries: 10:4
Disconnected Queue entries: 7:11 1:7 4:12 13:2 14:13 3:6
 15:9 12:10 2:15 8:14 9:1 11:31 5:0 0:8
EOF

5.1 stock with -current kernel 5.99.43
======================================
  * as GENERIC32_IP3x config, with DIAGNOSTIC, DEBUG
  * died immediately on X start, Xorg.log is zero bytes
  * rebooted, window make graphics and text are corrupted.
  * TWM is ok, but xclock is corrupted.
  * reboot exits to debugger with following:

fpemul_trapsignals(4,44c8f800,0x439b54)
pid 253(ntpd): trap: TLB miss (store) in kernel mode
status=0xfc03, cause=0xc, epc=0x8006952c, vaddr=0xcc417014 tf=0xcc416fa0 ksp=0xcc417000 ra=0x80069528
Stopped in pid 251.1 (ntpd) as netbsd:MachFPInterrupt+0xc0:   sw   ra,20(sp
)
db>

For X it looks like this kernel doesn't get on with the X 
acceleration work (xrender?).
Also I wonder if the system doesn't like ntpd updating the clock?

As if to exercise this I played around with ntpd.
So maybe I have the start of some empirical data.
 

5.1 stock with matt-nb5-mips64 from yesterday
=============================================
  * system boots, lots of CPU error 4
  * X with TWM is ok, xclock is corrupted
  * moving windows sometimes causes X server to exit 11
  * windowmaker graphics/text corrupted
  * top(1) consumes high cpu usage
  * reboot(1) drops out to kdb with ntp fault


Hope this helps,