Re: Re: [PATCH] xen block backend driver.

[Christoph Hellwig <hch@infradead.org>]

On Thu, Apr 21, 2011 at 12:03:12PM -0700, Daniel Stodden wrote:
> Yes, everybody is aware that the semantics were broken. But note it's
> not even a consistency issue at this point, because there's currently no
> frontend which relies on the original ordering semantics either. Take
> xen-blkfront, since blk_flush it uses the barrier op for a flush, being
> just a superset when ordering is enforced.

There is a huge userbase of guests out there that does rely on it.

> But before we just enumerate a new command, a potentially more viable
> option would be FLUSH+FUA flags on the WRITE operation. As if mapping
> bio bits.
> 
> The advantage is that it avoids the extra round trip implied by having
> the frontend driving writes through FSEQ_PREFLUSH on their own. I'd
> expect that to make much more of a performance difference. Somewhat
> differentiating PV from the low physical layer.
> 
> Would you, maybe did you, consider that? I think it sounds interesting
> enough to gather performance data, just asking beforehand.

You will need a pure flush anyway.  Once you actually have a correct
implementation you can look into optimizing it.  Note that at least
the Solaris Xen coded added a cache flush to the protocol.