Re: [PATCH, RFC/T] Fix handling of write failures to swap devices

[Nick Piggin <nickpiggin@yahoo.com.au>]

Richard Purdie wrote:
> On Sat, 2006-10-28 at 22:10 +1000, Nick Piggin wrote: 

>>That isn't your only problem though, and we simply don't want to do
>>this (potentially expensive) unusing from interrupt context. Noting
>>the error and dealing with it in process context I think is the best
>>way to do it.
> 
> 
> The reasoning was that this circumstance should be extremely rare. If it
> happens, we have a hardware problem. Recovering from that hardware
> problem gracefully is more important than a slightly longer interrupt.
> But yes, process context would be nicer, *if* we can find a way to do
> it. 

Also note that the current code *should* "gracefully" handle the failure.
In that, it will not reclaim the page on a write error, so it isn't going
to cause a data loss...

It's just that it currently results in unswappable pages.

Handling it more gracefully by allowing the page to be retried with another
swap entry is OK I guess, but given the added complexity, I'm not even sure
it is worthwhile.

Perhaps we should just do the ClearPageError in the try_to_unuse path,
because the sysadmin should take down that swap device on failure. So if a
new device is added, we want to be able to unpin the failed pages.

-- 
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com