2.6.30-9 sata softreset failed on AMD SB600

[Pitabred]

This seems to be a regression, as this problem went away on the 2.6.28 kernels of Jaunty, but now that I've upgraded to Karmic (2.6.30 kernels, all versions I've tested) I again have to disable the write caching in order for the drives to function under heavy data loads. This happens both on the "root" single disk and my raid5 array of 3 disks, all connected to the ATI SB600 chip.

These are excerpts from my dmesg:

[ 1.864515] ata2: softreset failed (device not ready)
[ 1.864557] ata2: failed due to HW bug, retry pmp=0
[ 1.864574] ata4: softreset failed (device not ready)
[ 1.864613] ata4: failed due to HW bug, retry pmp=0
[ 1.864629] ata3: softreset failed (device not ready)
[ 1.864667] ata3: failed due to HW bug, retry pmp=0
[ 2.436905] ata1.00: ATA-7: SAMSUNG SP2504C, VT100-33, max UDMA7
[39056.481268] ata1.00: SB600 AHCI: limiting to 255 sectors per cmd
[39056.486260] ata1.00: SB600 AHCI: limiting to 255 sectors per cmd
[39056.486268] ata1.00: configured for UDMA/133
[39056.486275] ata1: EH complete
[58596.804070] ata3.00: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen
[58596.804090] ata3.00: cmd ea/00:00:00:00:00/00:00:00:00:00/a0 tag 0
[58596.804099] ata3.00: status: { DRDY }
[58596.804109] ata3: hard resetting link
[58597.288535] ata3: softreset failed (device not ready)
[58597.288545] ata3: failed due to HW bug, retry pmp=0
[58597.452056] ata3: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[58597.457033] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd
[58597.462059] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd
[58597.462068] ata3.00: configured for UDMA/133
[58597.462155] ata3: EH complete

I couldn't find this reported linked to the latest Karmic kernels, so I want to make sure it's known that it still exists, and I can reproduce it fairly regularly by simply doing some heavy file operations. Filesystem is ext4 on both the array and the root drive, default options when creating it.

[Pitabred (ubuntu-pitabred)]

I reported this in the kernel, it may also be part of libata. I don't know for sure.

[Pitabred (ubuntu-pitabred)]

Update: Even with write-caching off, the drives still give the softreset error. This is not the kind of stability I've come to expect from Linux.

[oheohe (ohefan)]

I only tested 2.6.30 to find out if this softreset-error is fix or not. In contrary to the reporter, my AMD SB600 also leads to these errors under 2.6.28 kernels.
I have no idea if there is any connection at all, but since these error messages pop up, I'm not able to use any stand-by mode at all. Under 8.04 all worked well!
Hopefully there is a fix any time soon - errors related to this AMD Southbrigde are reported in Ubuntu-forums worldwide, with no apparent solution anywhere.

[Pitabred (ubuntu-pitabred)]

I suppose I was premature in saying that 2.6.28 worked. It simply worked better, needed more data moved before the softreset bug hits.

I did just find this in the kernel list:
http://kerneltrap.org/mailarchive/git-commits-head/2008/6/14/2122314

It seems as if it may be relevant to this bug.

[oheohe (ohefan)]

Yes, found that patch, too. I had a look at the sources for 2.6.30 - basically those changes are in already. Didn't look closer, though.
Perhaps the guy who submitted the patch might be the right contact person for those problems, as he seems to work for AMD/ATI.

[oheohe (ohefan)]

Did you try Ubuntu 9.10 Alpha2? I didn't experience any visible errors during boot time.

[Pitabred (ubuntu-pitabred)]

I'm running 9.10 with all the updates. I update it daily. I just installed the 2.6.31RC kernel, and it has an SB600 fix in it for 64bit, so I will see what happens.

I get the softreset errors when booting, but that happens on the SB700/800 chipset, too, and that runs fine. The softreset errors I get are when I do a lot of heavy writing to the drive(s) (RAID5 array).

[Pitabred (ubuntu-pitabred)]

Updated: Just tried to copy a lot of data to it under 2.6.31, no dice. Array blew itself apart.