Re: 2.4.x and DAC960 issues

Re: 2.4.x and DAC960 issues

[Albert Max Lai]

This is the summary of the off-list discussion and solution to my
DAC960 problem.

-Albert

On Saturday, 6 April 2002, Marc A. Volovic wrote:

> Quoth Leonard N. Zubkoff:
> 
> >   From: Albert Max Lai <amlai@bitsorcery.com>
> > 
> >   I moved the card into the slot closest to the CPU that I could, and
> >   voila!  everything works correctly; no lockups, ext3 works, even
> > 
> > Excellent news.  That's not a fix I've heard of before.
> 
> Hi,
> 
> Alas, it is very simple. Many (most? in my experience - Tyan, MSI, ASUS)
> motherboards leave their outer (farthest from the CPU area) PCI slots 
> ___NON-bus mastering___. In most cases, this is the outermost slot, but
> sometimes it is more than one slot, but again, the outermost, leftmost.
> 
> Sometimes, the masterlessness is dynamic - i.e. based on the number of
> populated slots, counting from the CPU. This is EXTREMELY rare. I saw
> it only once, I think, and on a board I did not trust even as far as 
> I could toss it. (Well, I could toss it some reasonable distance, which
> I did ;-)...
> 
> In some rare cases (errrr... ummmm... SOME Tyan board, I cannot
> currently remember the model) ran in the reverse direction.
> 
> Populating these masterless slots with anything but a sound card (and in
> many cases even by a sound card) leads to loss of stability. 
> 
> Moving a board INWARD (i.e. toward the CPU) in many cases solves the
> problem.
> 
> -- 
> ---MAV
>                        Linguists Do It Cunningly
> Marc A. Volovic                                          marc@bard.org.il

2.4.x and DAC960 issues

[Albert Max Lai]

I have had problems using the DAC960 driver under any 2.4.x kernel
(currently using 2.4.18).  I have not experienced these problems under
2.2.x.

1.  Under reasonably low loads the driver will hang.  It is very
    reproducible when using the benchmark program "Bonnie."  This
    problem is exacerbated when using the ext3 filing system, locking
    up almost immediately.

2.  I am not sure if this is related, but I believe that it is.  I see
    the message "spurious 8259A interrupt: IRQ7."  Most of the
    examples of this that I have read about involve having APIC
    enabled.  But, for me, this is not the case.

3.  If I boot without appending the "noapic" option, the driver hangs
    after scanning the bus for cards, but before getting to
    configuring the card.  This is minor.  I can live with this
    workaround.

The machine is a Tyan S1836DLUAN-BX, dual PIII 600Mhz.  The controller
is a DAC1164P w/ Firmware Version: 5.08-0-87.  The problem exists with
prior versions of the firmware and either RAID 0 or RAID 5 setups.
Kernels are compiled w/ egcs-2.91.66.

Please let me know if any additional information is needed.  Any help
debugging these problems would be appreciated.  Thanks in advance.

-Albert

Re: 2.4.x and DAC960 issues

[Marc A. Volovic]

Albert Max Lai said:

> I have had problems using the DAC960 driver under any 2.4.x kernel
> (currently using 2.4.18).  I have not experienced these problems under
> 
> 1.  Under reasonably low loads the driver will hang.  It is very
>     reproducible when using the benchmark program "Bonnie."  This
[snip]
>     problem is exacerbated when using the ext3 filing system, locking

I am using a Mylex 170LP with no problem, running on a dual 550MHz 
MSI 6120S under quite a few 2.4.x kernels, lately 2.4.18 and 2.4.19pre5,
all under reiserfs.  The firmware is 6.00-15, carrying 6 9GB disks
(5 RAID5 + 1 spare).

There have been NO lockups under any version of the kernel, not under 
multiple bonnie runs.

What is your interrupt breakdown? Could your machine be doing something
naughty with the interrupts?

---MAV
                       Linguists Do It Cunningly
Marc A. Volovic                                          marc@bard.org.il

Re: 2.4.x and DAC960 issues

[Albert Max Lai]

On Friday, 5 April 2002, Marc A. Volovic wrote:

> I am using a Mylex 170LP with no problem, running on a dual 550MHz 
> MSI 6120S under quite a few 2.4.x kernels, lately 2.4.18 and 2.4.19pre5,
> all under reiserfs.  The firmware is 6.00-15, carrying 6 9GB disks
> (5 RAID5 + 1 spare).
> 
> There have been NO lockups under any version of the kernel, not under 
> multiple bonnie runs.
> 
> What is your interrupt breakdown? Could your machine be doing something
> naughty with the interrupts?

This what /proc/interrupts says:
           CPU0       CPU1       
  0:    3874524          0          XT-PIC  timer
  1:      18836          0          XT-PIC  keyboard
  2:          0          0          XT-PIC  cascade
  4:          4          0          XT-PIC  serial
  5:      46479          0          XT-PIC  soundblaster
  8:     218807          0          XT-PIC  rtc
  9:     128424          0          XT-PIC  aic7xxx, aic7xxx
 10:      52035          0          XT-PIC  Mylex DAC1164P
 12:     342261          0          XT-PIC  PS/2 Mouse
 14:     209669          0          XT-PIC  eth0
 15:      44772          0          XT-PIC  eth1, usb-uhci
NMI:          0          0 
LOC:    3874766    3874768 
ERR:         16
MIS:          0

-Albert

Re: 2.4.x and DAC960 issues

[Marc A. Volovic]

Albert Max Lai said:

> On Friday, 5 April 2002, Marc A. Volovic wrote:
> > What is your interrupt breakdown? Could your machine be doing something
> > naughty with the interrupts?
> 
>            CPU0       CPU1       
>   9:     128424          0          XT-PIC  aic7xxx, aic7xxx
>  10:      52035          0          XT-PIC  Mylex DAC1164P
[snips]

Ouch, this does not look healthy. You're at 'noapic'? Seems so.

Mine is:
           CPU0       CPU1       
 17:         22         17   IO-APIC-level  BusLogic BT-958
 18:      83988      84319   IO-APIC-level  Mylex AcceleRAID 160

However, I can see no indication for misbehaviour. Let's take it off 
the list. Can you send me a driver startup dmesg?


---MAV
                       Linguists Do It Cunningly
Marc A. Volovic                                          marc@bard.org.il

Re: 2.4.x and DAC960 issues

[Marcelo Tosatti]

On Fri, 5 Apr 2002, Albert Max Lai wrote:

> On Friday, 5 April 2002, Marc A. Volovic wrote:
> 
> > I am using a Mylex 170LP with no problem, running on a dual 550MHz 
> > MSI 6120S under quite a few 2.4.x kernels, lately 2.4.18 and 2.4.19pre5,
> > all under reiserfs.  The firmware is 6.00-15, carrying 6 9GB disks
> > (5 RAID5 + 1 spare).
> > 
> > There have been NO lockups under any version of the kernel, not under 
> > multiple bonnie runs.
> > 
> > What is your interrupt breakdown? Could your machine be doing something
> > naughty with the interrupts?
> 
> This what /proc/interrupts says:
>            CPU0       CPU1       
>   0:    3874524          0          XT-PIC  timer
>   1:      18836          0          XT-PIC  keyboard
>   2:          0          0          XT-PIC  cascade
>   4:          4          0          XT-PIC  serial
>   5:      46479          0          XT-PIC  soundblaster
>   8:     218807          0          XT-PIC  rtc
>   9:     128424          0          XT-PIC  aic7xxx, aic7xxx
>  10:      52035          0          XT-PIC  Mylex DAC1164P
>  12:     342261          0          XT-PIC  PS/2 Mouse
>  14:     209669          0          XT-PIC  eth0
>  15:      44772          0          XT-PIC  eth1, usb-uhci
> NMI:          0          0 
> LOC:    3874766    3874768 
> ERR:         16
> MIS:          0

I've forwarded your first message to Leonard (the driver author)... well
probably get some feedback soon.