Re: Instability issues with NetBSD-9, xen-4.11 and the xbdb backend driver

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Nov 15, 2019 at 09:35:14AM -0800, Brian Buhrow wrote:
> 	Hello.  The error handling patch I provided doesn't fix the underlying
> problem, I agree, but it does bring the error handling code into alignment
> across the file.  If you look at:
> sys/arch/xen/xen/xbdback_xenbus.c:1491, you'll see that I simply made the
> behavior the same in the two places where xbdback_io_error is called.

And this is wrong. In xbdback_co_io_gotfrag2() we certainly want to got out of
xbdback_co_io_loop() and this won't have hill effects.
But doing the same in xbdback_co_map_io() may get is out of cache flush
processing, leaving it unprocessed forever.


> 	The problem with error handling as it is now is that two requests with
> the same data get passed to the domu, which is what causes the biodone2

But I don't understand *why* the code, as is, has this behavior.
xbdback_iodone() shouldn't push the reply to the guest unless all segments
have been handled, in which case calling xbdback_co_io_loop() again should
be harmless.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--