[ltp] Re: Make disk-protect work on 2.6.24

[Elias Oltmanns]

Rolf Kutz <rk-2RCiGu8/rP8@public.gmane.org> wrote:
> On 13/02/08 12:39 +0100, Elias Oltmanns wrote:
>>
>>Anyway, there is a temporary solution to this problem. If you apply the
>>little patch attached to this email, the usual disk-protect patch will
>>work on 2.6.24. This change is supposed to fix a generic problem in
>>libata and may eventually be merged upstream, but Tejun Heo (libata
>>developer) said that some testing was required first.
>
> Would it be possible to include it in a bugfix
> release (2.6.24.x)?

That's not for me to say, I'm afraid. Tejun wanted to look into the
matter and do some testing before submitting any changes. It is not
really clear even to him why the scheduler wouldn't serve I/O threads
under certain circumstances. So, we'll basically just have to wait and
see.

>
>>This change is absolutely safe for everyone using a non-NCQ capable
>>system. In particular, ata_piix doesn't support NCQ. If you are using
>>ahci and /sys/class/scsi_disk/.../device/queue_depth reports a value
>>greater than 1, i.e. NCQ is enabled, then the patch shouldn't cause any
>>harm either but it may or may not affect I/O performance.
>
> I will try it, though I'm not affected by the
> freezes. Should I run some benchmarks to test
> performance with and without the patch? Which
> ones?

Benchmarks and the like would only make sense on systems using NCQ. I
hardly know anything about NCQ and I don't have access to an NCQ capable
system. The libata people, on the other hand, certainly know about NCQ
and will most likely have easy access to test systems, so I leave that
to them.

Regards,

Elias