From: Björn Stenberg <bjorn_at_haxx.se>
Date: Wed, 30 Apr 2003 12:43:20 +0200

Mike Holden wrote:
> You may not have seen it happen, but I have. With appropriate debug
> information (I have added some extra WPS elements so I can mointor things
> like this) I can conclusively see that the disk spins down before the read
> operation has completed.

If you say so, I believe you. But we need to find out why/where it happens. The only way I can see this happen is if a single read command takes longer than your spindown timeout, without any read retries.

> I am convinced that there is at least one more occasion (maybe more) where
> this is still happening, despite out best endeavours. If we can find out
> where this is, then there is no need for this patch, but this should sort
> that particular issue out once and for all, both now and in the future.

No, that's just painting over broken code with new code. If the current code doesn't work as intended, we should fix it.

The design of last_disk_activitity is sound. Spindown should be handled by the driver, not the upper layers. I think you are exaggerating the complexity of it.

-- 
Björn

Received on 2003-04-30