Serverworks CSB5 IDE-DMA Problem (2.4 and 2.6)

Hello!

we have a bunch of IBM x305 here which are entrylevel 1HE servers based on a Serverworks CSB5 chipset.
One of those has 2 120GB IDE disks in a software RAID1 and the main userspace-application is a heavly (mostly insert/update) used postgresql-database. The database generates a lot of sustained IO-traffic and after some minutes (depends on the load - sometimes it even works for one or two hours) the kernel generates the following messages(2.4.22 and 2.6.0-test6 behave almost identically - error-messages are from 2.6.0-test6):

hdc: dma_timer_expiry: dma status == 0x20
hdc: DMA timeout retry
hdc: timeout waiting for DMA
hdc: status timeout: status=0xd0 { Busy }

hdc: drive not ready for command
ide1: reset: success
hdc: dma_timer_expiry: dma status == 0x20
hdc: DMA timeout retry
hdc: timeout waiting for DMA
hdc: status timeout: status=0xd0 { Busy }

hdc: drive not ready for command
ide1: reset: success
hdc: dma_timer_expiry: dma status == 0x20
hdc: DMA timeout retry
hdc: timeout waiting for DMA
hdc: status timeout: status=0xd0 { Busy }

hdc: drive not ready for command
ide1: reset: success
hdc: dma_timer_expiry: dma status == 0x20
hdc: DMA timeout retry
hdc: timeout waiting for DMA
hdc: status timeout: status=0xd0 { Busy }

hdc: drive not ready for command
ide1: reset: success
hda: dma_timer_expiry: dma status == 0x60
hda: DMA timeout retry
hda: timeout waiting for DMA
hda: status timeout: status=0xd0 { Busy }

hdb: DMA disabled
hda: drive not ready for command
ide0: reset: success
blk: queue dfdee200, I/O limit 4095Mb (mask 0xffffffff)
hda: dma_timer_expiry: dma status == 0x20
hda: DMA timeout retry
hda: timeout waiting for DMA
hda: status timeout: status=0xd0 { Busy }

hda: drive not ready for command
ide0: reset: success
hda: dma_timer_expiry: dma status == 0x20
hda: DMA timeout retry
hda: timeout waiting for DMA
hda: status timeout: status=0xd0 { Busy }

hda: drive not ready for command
ide0: reset: success

after one of this events DMA on one of the disks (either hdc or hda) gets disabled and the maschine is heavily overloaded and the database cannot keep up any more with the incoming load of database-updates. It's also worth mentioning that the kernel reports a "DMA disabled" only for hdb which is the internal cd-drive and completely unused.

I do know that Serverworks IDE has been flaky (especially with the CSB4) in the past but I thought this had been fixed in newer chipset-revisions
- is there anything I can do to solve this problem?

dmesg of the machine in question can be found at
http://www.kaltenbrunner.cc/files/dmesg.txt

many thanks

Stefan Kaltenbrunner

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Bart,

Do you have any idea ?

Wild guess: Disable APIC?

Sorry, I screwed up, again.

Thats me asking, not Bart. Duh.

Quoted 5 lines

Bart,

Do you have any idea ?

Wild guess: Disable APIC?

APIC problem should be fixed, but yes it's better to disable ACPI.

These "timeout due to drive busy" needs to be resolved.

Quoted 1 line · Bartlomiej Zolnierkiewicz

APIC problem should be fixed, but yes it's better to disable ACPI.

Not sure if I understand this one right - the dmesg was from the 2.6.0-test6 kernel which did have ACPI HT-enum-only compiled in but no "local APIC support".
The 2.4.22 one that has the same problem does neither have ACPI nor APIC support compiled in - so no this doesn't seem to be the problem.

Quoted 1 line

These "timeout due to drive busy" needs to be resolved.

Yes - I really hope this will be fixed soon. I was forced to add a fiberchannel HBA into this maschine today to integrate it into our SAN to get the database up to speed again.
However I'm willing to move the database to the local disks again if you want me to test a patch or something along that line.

Quoted 7 lines

Quoted 1 line

These "timeout due to drive busy" needs to be resolved.

Yes - I really hope this will be fixed soon. I was forced to add a fiberchannel HBA into this maschine today to integrate it into our SAN to get the database up to speed again.
However I'm willing to move the database to the local disks again if you want me to test a patch or something along that line.

Did some kernel worked okay or this is new system?

Quoted 9 lines · Bartlomiej Zolnierkiewicz

Quoted 7 lines

Quoted 1 line

These "timeout due to drive busy" needs to be resolved.

Yes - I really hope this will be fixed soon. I was forced to add a fiberchannel HBA into this maschine today to integrate it into our SAN to get the database up to speed again.
However I'm willing to move the database to the local disks again if you want me to test a patch or something along that line.

Did some kernel worked okay or this is new system?

This is a new system - I can try an older kernel if you can give me some hints about how old it should be :-)

Stefan

2.4.18, 2.4.19 w/o APIC and ACPI

Quoted 1 line · Bartlomiej Zolnierkiewicz

2.4.18, 2.4.19 w/o APIC and ACPI

ok 2.4.18 (dmesg at http://www.kaltenbrunner.cc/files/dmesg2418.txt)
seems to work better(although not as fast as I would like to have it) but I suspect that:

ide1: Speed warnings UDMA 3/4/5 is not functional.
ide0: Speed warnings UDMA 3/4/5 is not functional.

is quite interesting - if these UDMA-modes do not work reliable - why do they get enabled with later kernels(not that I would have a problem with getting UDMA > 2 working *g*) ?

Quoted 11 lines · Stefan Kaltenbrunner

Quoted 1 line · Bartlomiej Zolnierkiewicz

2.4.18, 2.4.19 w/o APIC and ACPI

ok 2.4.18 (dmesg at http://www.kaltenbrunner.cc/files/dmesg2418.txt)
seems to work better(although not as fast as I would like to have it) but I suspect that:

ide1: Speed warnings UDMA 3/4/5 is not functional.
ide0: Speed warnings UDMA 3/4/5 is not functional.

Quoted 3 lines

is quite interesting - if these UDMA-modes do not work reliable - why do they get enabled with later kernels(not that I would have a problem with getting UDMA > 2 working *g*) ?

2.4.22 has 80-pin cable dedetecion for more vendors. You can try passing "ide0=ata66 ide1=ata66" boot options.