SX8 stability issue

SX8 stability issue

[Alexander Sbitnev]

  Working with Promise SX8 card for a while we still can't get NCQ
working stable for both original promise and vanilla kernel drivers.
We using latest firmware 1.00.0.37 and kernels from 2.6.8 up to 2.6.15.
Problems occured at least on two different hardware platforms.
We don't played with max_queue option (or it's header file analog value) yet,
keeping it on default value.
Controller work almost stable while NCQ option disabled in BIOS.
Once feature enabled in BIOS the system become hanging on IO with next
error messages appearing in the kernel log:

syslog.0:Jan 10 23:30:57 cell kernel: sx8(0000:01:06.0): unhandled event type 16
syslog.0:Jan 10 23:31:18 cell kernel: sx8(0000:01:06.0): unhandled event type 16
syslog.0:Jan 10 23:31:40 cell kernel: sx8(0000:01:06.0): unhandled event type 16
syslog.0:Jan 10 23:32:01 cell kernel: sx8(0000:01:06.0): unhandled event type 16
syslog.0:Jan 10 23:32:18 cell kernel: end_request: I/O error, dev sx8/0, sector 220064
syslog.0:Jan 10 23:32:18 cell kernel: Buffer I/O error on device sx8/0, logical block 27508
syslog.0:Jan 10 23:32:18 cell kernel: lost page write due to I/O error on sx8/0
syslog.0:Jan 10 23:32:18 cell kernel: Buffer I/O error on device sx8/0, logical block 27509
syslog.0:Jan 10 23:32:18 cell kernel: lost page write due to I/O error on sx8/0
syslog.0:Jan 10 23:32:18 cell kernel: Buffer I/O error on device sx8/0, logical block 27510

  We don't really sure this problem is a 100% linux related, but it is
still persist after big hardware upgrade (From dual Pentium III to
Dual Opteron based system recomended for this controller).
  Maybe it is not right way at all to turn NCQ in BIOS? Мауве we must
just increase max_queue parameter while keeping NCQ in BIOS disabled?

Re: SX8 stability issue

[Jeff Garzik]

Alexander Sbitnev wrote:
>   Working with Promise SX8 card for a while we still can't get NCQ
> working stable for both original promise and vanilla kernel drivers.
> We using latest firmware 1.00.0.37 and kernels from 2.6.8 up to 2.6.15.
> Problems occured at least on two different hardware platforms.
> We don't played with max_queue option (or it's header file analog value) yet,
> keeping it on default value.
> Controller work almost stable while NCQ option disabled in BIOS.
> Once feature enabled in BIOS the system become hanging on IO with next
> error messages appearing in the kernel log:
> 
> syslog.0:Jan 10 23:30:57 cell kernel: sx8(0000:01:06.0): unhandled event type 16
> syslog.0:Jan 10 23:31:18 cell kernel: sx8(0000:01:06.0): unhandled event type 16
> syslog.0:Jan 10 23:31:40 cell kernel: sx8(0000:01:06.0): unhandled event type 16
> syslog.0:Jan 10 23:32:01 cell kernel: sx8(0000:01:06.0): unhandled event type 16
> syslog.0:Jan 10 23:32:18 cell kernel: end_request: I/O error, dev sx8/0, sector 220064
> syslog.0:Jan 10 23:32:18 cell kernel: Buffer I/O error on device sx8/0, logical block 27508
> syslog.0:Jan 10 23:32:18 cell kernel: lost page write due to I/O error on sx8/0
> syslog.0:Jan 10 23:32:18 cell kernel: Buffer I/O error on device sx8/0, logical block 27509
> syslog.0:Jan 10 23:32:18 cell kernel: lost page write due to I/O error on sx8/0
> syslog.0:Jan 10 23:32:18 cell kernel: Buffer I/O error on device sx8/0, logical block 27510
> 
>   We don't really sure this problem is a 100% linux related, but it is
> still persist after big hardware upgrade (From dual Pentium III to
> Dual Opteron based system recomended for this controller).

Well, the in-kernel sx8 driver does not use NCQ at all, so you would 
have to have modified the driver to turn it on.  Presumably this means 
there is a bug in your modifications?


>   Maybe it is not right way at all to turn NCQ in BIOS? Мауве we must
> just increase max_queue parameter while keeping NCQ in BIOS disabled?

This is the best first step...

	Jeff

Re[2]: SX8 stability issue

[Alexander Sbitnev]

> Well, the in-kernel sx8 driver does not use NCQ at all, so you would
> have to have modified the driver to turn it on.  Presumably this means 
> there is a bug in your modifications?

Hmmm... Ok, driver doesn't care about NCQ at all, as i understood all NCQ work
performed by it's firmware. Then we have just a stability issue, not
directly linked to NCQ. Right?
              Shuras

Re: SX8 stability issue

[Jeff Garzik]

Alexander Sbitnev wrote:
>>Well, the in-kernel sx8 driver does not use NCQ at all, so you would
>>have to have modified the driver to turn it on.  Presumably this means 
>>there is a bug in your modifications?
> 
> 
> Hmmm... Ok, driver doesn't care about NCQ at all, as i understood all NCQ work
> performed by it's firmware. Then we have just a stability issue, not
> directly linked to NCQ. Right?

I cannot say "correct", I can only guess "possibly yes."

	Jeff