svn commit: r212160 - in head/sys: cam/ata cam/scsi cddl/contrib/opensolaris/uts/common/fs/zfs geom geom/sched kern sys

[Alexander Motin]

Pawel Jakub Dawidek wrote:
> On Thu, Sep 02, 2010 at 04:07:36PM -0600, Justin T. Gibbs wrote:
>>  On 9/2/2010 3:39 PM, Pawel Jakub Dawidek wrote:
>>> Don't you think it would be better to set the flag from within
>>> g_io_request()? This way every BIO_FLUSH consumer doesn't have to
>>> remember to set it. Or am I missing something?
>> I don't feel strongly one way or the other, but I thought that
>> g_io_request()'s job was to execute the request and to test invariants,
>> not to set policy.  Perhaps I misinterpreted it's role.
> 
> Does BIO_FLUSH make sense without BIO_ORDERED? My understanding is that
> it doesn't. 

IMHO it does. If caller manages ordering by waiting for all required
writes to complete before submitting BIO_FLUSH. Such technics probably
can be more difficult, but IMHO should give more flexibility to disk
schedulers under parallel load.

> But we still want BIO_ORDERED for use with BIO_WRITE for
> write barriers without cache flushing.

Sure.

-- 
Alexander Motin