ata problems @ r189170

[Richard Todd]

Alexander Motin <mav at FreeBSD.org> writes:

> Looks like while fixing small problems I have found bigger ones. Try
> please this patch against HEAD, it reverts r189166, r189091 and
> partially r188903:
>
> http://people.freebsd.org/~mav/ata.rollback.patch
>
> If it works, I will stop for some time to cool down and search for
> better solution.

I too found problems with the "new" ATA code you'd tried -- not any
obvious error messages or timeouts from the ATA system, but it was
obviously sometimes returning corrupt data, as evidenced by ZFS giving
checksum errors.  (Yay for ZFS for catching this!) Moving to the
current (post-r189195) code that has your aforementioned rollback patch seems
to have fixed the problem, and I've been "zfs scrubbing" the disks without any
sign of errors.  So something in the patches you reverted apparently can 
cause silent data corruption that could go unnoticed without a higher-level
checksum like ZFS has.   Something you might want to watch out for when 
reworking those patches....

If it matters, this is on a Core2Duo system in amd64 mode, with a
Intel DP965LT motherboard; the SATA controller is, if I recall right,
ICH8:

atapci1: <Intel AHCI controller> port 0x3108-0x310f,0x3114-0x3117,0x3100-0x3107,0x3110-0x3113,0x3020-0x303f mem 0xe8325000-0xe83257ff irq 19 at device 31.2 on pci0

atapci1 at pci0:0:31:2:    class=0x010601 card=0x514d8086 chip=0x28248086 rev=0x02 hdr=0x00