Re: [Qemu-devel] [PATCH v2] xen_disk: support "direct-io-safe" backend option

From: Alex Bligh <alex@alex.org.uk>
To: Ian Jackson <Ian.Jackson@eu.citrix.com>
Cc: xen-devel@lists.xensource.com,
	Anthony Liguori <anthony@codemonkey.ws>,
	Stefano Stabellini <stefano.stabellini@eu.citrix.com>,
	qemu-devel@nongnu.org, Alex Bligh <alex@alex.org.uk>,
	Paolo Bonzini <pbonzini@redhat.com>
Subject: Re: [Qemu-devel] [PATCH v2] xen_disk: support "direct-io-safe" backend	option
Date: Fri, 28 Jun 2013 17:17:37 +0100	[thread overview]
Message-ID: <624B402D40D846D55A77B826@nimrod.local> (raw)
In-Reply-To: <20941.26905.519774.239644@mariner.uk.xensource.com>

--On 28 June 2013 11:44:41 +0100 Ian Jackson <Ian.Jackson@eu.citrix.com> 
wrote:

>> Looks useful. Are you planning to do this for both emulated and pv
>> disks?
>
> Emulated disks don't have the same problem because they don't try to
> use O_DIRECT on pages shared with the guest via the Xen grant table
> mechanism.

I should have been more specific. The original thread maintained
emulated disks always had O_DIRECT turned off, despite the fact
the rationale for using O_DIRECT for PV disks was that not using
O_DIRECT in some circumstances might be unsafe, because it was
the only way to get any decent performance out of them. I think
we ran the 'no O_DIRECT might be unsafe' argument to ground, but
if the rationale for Stefano's patch is not just speed but additional
safety (for instance against the host dying and losing the page
cache for file systems that have barriers switched off), then
there is an argument to use it for emulated disks too.

But as Stefano says:

--On 28 June 2013 11:56:29 +0100 Stefano Stabellini 
<stefano.stabellini@eu.citrix.com> wrote:

> This is PV only, at least for the moment: emulated disks always use
> writeback caching.
> From the performance point of view, making this change for IDE disks is
> not very important (because IDE is slow anyway).

... perhaps 'who cares'.

-- 
Alex Bligh