Re: how to log reiser and raid0 crash? 2.6.0-t4

Re: how to log reiser and raid0 crash? 2.6.0-t4

[Resident Boxholder]

http://www.promise.com/product/product_detail_eng.asp?productId=87&familyId=3
promise bios flash page--
http://www.promise.com/support/download/download2_eng.asp?productId=87&category=bios&os=100
wget this bios flash if today--(will try later tonight)
http://www.promise.com/support/file/bios/ultra133tx2b220015.zip

Mudama, Eric wrote:

>LBA48 shouldn't affect 60GB drives.
>
>To my understanding, it is only an issue with drives >137GB (128GiB) that
>are moved to some controllers (promise?) after being used on other systems.
>
>--eric
>
True, but I think Alan Cox mentioned something about new code setting 
pio4 and udma6
at the same time so I was hoping new code might relate to system hangs 
on mkfs, fsck,
cp to raid on commodity-promise-controlled drives. I might try the lba48 
patch on a
null-modem serial setup in case there's other code in there. I could use 
a promise
sx-6000 according to Oliver Pitzeier oliver@linux-kernel.at (see below) 
but that's
an expensive onboard raid card, not commodity booty using linux software 
raid.
Lost time moots the expense issue but commodity servers need a cheap 
reliable controller
card, a tulip equivalent in the controller card category, to make a 
cheap net-speed
file-server.
-Bob D

>2.6.0-t4 amd 3000+ 1G four maxtor 60G drives on two 
>> controllers(mb's and promise)
>
As I always tell my customers: Promise means trouble; At least within Linux.

The only Promise controller that works fine for me is a SX-6000 with 6 drives (raid-5, 1 spare).

Best regards,
 Oliver








>
>-----Original Message-----
>From: Resident Boxholder [mailto:resid@boxho.com]
>Sent: Thursday, August 28, 2003 1:29 AM
>To: Oleg Drokin
>Subject: Re: how to log reiser and raid0 crash? 2.6.0-t4
>
>
>2.6.0-t4 amd 3000+ 1G four maxtor 60G drives on two controllers(mb's and 
>promise)
>
>problem isolated to promise card(have to verify the other is error-free 
>longer though)
>
>Oleg Drokin wrote:
>
>  
>
>>Hello!
>>
>>On Tue, Aug 26, 2003 at 03:57:39PM -0400, Resident Boxholder wrote:
>>
>> 
>>
>>    
>>
>>>I cause a lock up by doing a cp -aR /usr/src /mnt/usr...
>>>   
>>>
>>>      
>>>
>>Is there any chance of using sirial console to see if you can capture
>>    
>>
>something on that?
>  
>
>>Bye,
>>   Oleg
>>
>>    
>>
>I have a second linux pc and I can google whether to use null or regular 
>serial cable,
>howto put a console on serial, then log to that, I guess that's all. As 
>a bonus I won't
>need a kvm switch anymore.
>
>Here's some repeatable "high-level(monkey-level)" info.
>
>I have four drives, two on mboard controller and two on a 133 Promise card.
>If I make a four-drive four-partition raid zero md device, I get enough 
>lockups
>to draw my attention, so then I make two-drive md devices to see which
>controller works, if any. Without really flogging it I got ext2 and 
>reiserfs on
>the mboard's two drives to work for mkfs, fsck, and copy /usr/src/ /tmp
>so maybe this is a Promise problem. The mboard controller handles udma6
>no problem for that two-drive md.
>
>With only two drives on the promise card forming a raid zero md device,
>crashes happen on mkfs and fscks and even though no md's are mounted
>on boot, the boot won't happen sometimes after a crash.
>
>I have no errors logging anymore. The only one I got was by switching to
>vc/5 and seeing reiserfs info but I don't see that anymore since fixing one
>thing, it was because of mdadm and debian config conflict, forget that.
>The irq error storm was a false lead as well, no more such errors, no
>errors at all logged, since removing cd's and second promise card and
>turning usb off and turning apic off in bios and letting linux turn apic on.
>ACPI is smooth, no errors, just sudden death.
>
>Things I've tried are no fastrak in promise kernel config, hdparm no dma,
>hdparm udma4 instead of udma6. I'll try pio4 pio3 on the promise md
>and really flog the mboard-controlled md to verify that there is no
>problem except on the promise card. I'm reading what Alan Cox is
>saying about "LBA48 pio and udma" so I will try turning dma off
>and going down in pio modes.
>
>Swap is working with four drives, so two on the promise card, but with 1G of
>ram swap on the promise card may never be used.
>
>-Bob D
>
>-
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>Please read the FAQ at  http://www.tux.org/lkml/
>
>
>  
>

Re: how to log reiser and raid0 crash? 2.6.0-t4

[Resident Boxholder]

2.6.0-t4 amd 3000+ 1G four maxtor 60G drives on two controllers(mb's and 
promise)

problem isolated to promise card(have to verify the other is error-free 
longer though)

Oleg Drokin wrote:

>Hello!
>
>On Tue, Aug 26, 2003 at 03:57:39PM -0400, Resident Boxholder wrote:
>
>  
>
>>I cause a lock up by doing a cp -aR /usr/src /mnt/usr...
>>    
>>
>
>Is there any chance of using sirial console to see if you can capture something on that?
>
>Bye,
>    Oleg
>
I have a second linux pc and I can google whether to use null or regular 
serial cable,
howto put a console on serial, then log to that, I guess that's all. As 
a bonus I won't
need a kvm switch anymore.

Here's some repeatable "high-level(monkey-level)" info.

I have four drives, two on mboard controller and two on a 133 Promise card.
If I make a four-drive four-partition raid zero md device, I get enough 
lockups
to draw my attention, so then I make two-drive md devices to see which
controller works, if any. Without really flogging it I got ext2 and 
reiserfs on
the mboard's two drives to work for mkfs, fsck, and copy /usr/src/ /tmp
so maybe this is a Promise problem. The mboard controller handles udma6
no problem for that two-drive md.

With only two drives on the promise card forming a raid zero md device,
crashes happen on mkfs and fscks and even though no md's are mounted
on boot, the boot won't happen sometimes after a crash.

I have no errors logging anymore. The only one I got was by switching to
vc/5 and seeing reiserfs info but I don't see that anymore since fixing one
thing, it was because of mdadm and debian config conflict, forget that.
The irq error storm was a false lead as well, no more such errors, no
errors at all logged, since removing cd's and second promise card and
turning usb off and turning apic off in bios and letting linux turn apic on.
ACPI is smooth, no errors, just sudden death.

Things I've tried are no fastrak in promise kernel config, hdparm no dma,
hdparm udma4 instead of udma6. I'll try pio4 pio3 on the promise md
and really flog the mboard-controlled md to verify that there is no
problem except on the promise card. I'm reading what Alan Cox is
saying about "LBA48 pio and udma" so I will try turning dma off
and going down in pio modes.

Swap is working with four drives, so two on the promise card, but with 1G of
ram swap on the promise card may never be used.

-Bob D

Re: how to log reiser and raid0 crash? 2.6.0-t4

[Oleg Drokin]

Hello!

On Tue, Aug 26, 2003 at 03:57:39PM -0400, Resident Boxholder wrote:

> I cause a lock up by doing a cp -aR /usr/src /mnt/usr which moves data 
> larger
> than total hard buffer cache, to raid0 reiserfs or ext2 ( NOT reiser4!) 
> I'm wondering what to send in. Maybe I could send a log from successful
> copy with swap off, showing reiser logging, and config, in case a stress
> condition or misconfig shows up even when catastrophic failure doesn't
> occur. With swap on the fail is sudden and no error logging is coming
> through.

Is there any chance of using sirial console to see if you can capture something on that?

Bye,
    Oleg

Re: how to log reiser and raid0 crash? 2.6.0-t4

[Resident Boxholder]

I had syslog send just *.info to vc/5 and though the *.info messages 
never make
it into syslog file, I started cp -aR /usr/src/kernel-source-2.6.0 /tmp 
with md/2 mounted
to /tmp and saw this *.info on vc/5 and then it locked up.  How about 
those zeros, skip,
assign id anyway?

journal_1037: journal_read_transaction, offset 3623, len 0 mount_id 0
journal_1039: journal_read_trans skipping because 3623 is too old
journal_1299: Setting newest_mount_id to 12

Anything?

-Bob D

Resident Boxholder wrote:

> 2.6.0-test4 amd xp 3000+ msi mbo nforce2  four Maxtor 60G 8mbcache raid0
>
> I cause a lock up by doing a cp -aR /usr/src /mnt/usr which moves data 
> larger
> than total hard buffer cache, to raid0 reiserfs or ext2 ( NOT 
> reiser4!) Copy ops
> smaller than buffer cache(8mb x 4 = 32mb) don't fail. Nothing fails on 
> a single
> drive, compiles or copies, just copy to a mounted raid0 device. What 
> should I
> try, test, or dump?
>
> No irq error storm. No cd drives installed. Smaller copy ops work. 
> Turning
> swap off first slows things down enough to work, but swap itself is 
> probably
> OK. I have bios turn apic off, then linux turns it on, which is good 
> until the
> turn apic off before turn apic on patch gets into test5 or whatever.
>
> hdparm sets all four drives the same, udma6 but have tried down to udma4
> and pio4 and dma turned off, unmask on or off.
>
> If two drives are on the mboard controller and two on a promise pci card,
> does /proc/ide/amd74xx refer to the two drives on the motherboard only?
> It seems to mention four drives but the two slow ones might just be a
> ref to unoccupied slave drive positions.
>
> I'm wondering what to send in. Maybe I could send a log from successful
> copy with swap off, showing reiser logging, and config, in case a stress
> condition or misconfig shows up even when catastrophic failure doesn't
> occur. With swap on the fail is sudden and no error logging is coming
> through.
>
> I could incrementally copy /usr/src to one raid, then do a copy from that
> raid to another raid. All that would do is test copying from one balanced
> set of four drives/partitions to another balanced set, versus copying 
> from
> one drive's /usr/src to that and three other drives' raid set, which is
> unbalanced, dragging on one drive.
>
> -Bob
>
> -
> To unsubscribe from this list: send the line "unsubscribe 
> linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>
>

how to log reiser and raid0 crash? 2.6.0-t4

[Resident Boxholder]

2.6.0-test4 amd xp 3000+ msi mbo nforce2  four Maxtor 60G 8mbcache raid0

I cause a lock up by doing a cp -aR /usr/src /mnt/usr which moves data 
larger
than total hard buffer cache, to raid0 reiserfs or ext2 ( NOT reiser4!) 
Copy ops
smaller than buffer cache(8mb x 4 = 32mb) don't fail. Nothing fails on a 
single
drive, compiles or copies, just copy to a mounted raid0 device. What 
should I
try, test, or dump?

No irq error storm. No cd drives installed. Smaller copy ops work. Turning
swap off first slows things down enough to work, but swap itself is probably
OK. I have bios turn apic off, then linux turns it on, which is good 
until the
turn apic off before turn apic on patch gets into test5 or whatever.

hdparm sets all four drives the same, udma6 but have tried down to udma4
and pio4 and dma turned off, unmask on or off.

If two drives are on the mboard controller and two on a promise pci card,
does /proc/ide/amd74xx refer to the two drives on the motherboard only?
It seems to mention four drives but the two slow ones might just be a
ref to unoccupied slave drive positions.

I'm wondering what to send in. Maybe I could send a log from successful
copy with swap off, showing reiser logging, and config, in case a stress
condition or misconfig shows up even when catastrophic failure doesn't
occur. With swap on the fail is sudden and no error logging is coming
through.

I could incrementally copy /usr/src to one raid, then do a copy from that
raid to another raid. All that would do is test copying from one balanced
set of four drives/partitions to another balanced set, versus copying from
one drive's /usr/src to that and three other drives' raid set, which is
unbalanced, dragging on one drive.

-Bob