xfs
[Top] [All Lists]

Re: [PATCH 2/2] xfs_io: configure tests for preadv/pwritev & sync_file_r

To: Eric Sandeen <sandeen@xxxxxxxxxxx>
Subject: Re: [PATCH 2/2] xfs_io: configure tests for preadv/pwritev & sync_file_range
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Wed, 10 Oct 2012 10:05:36 +1100
Cc: xfs-oss <xfs@xxxxxxxxxxx>
In-reply-to: <5074ABF4.2000807@xxxxxxxxxxx>
References: <50749EE2.6070408@xxxxxxxxxx> <5074A05A.6000709@xxxxxxxxxxx> <20121009222806.GS23644@dastard> <5074A5FE.50101@xxxxxxxxxxx> <20121009225620.GU23644@dastard> <5074ABF4.2000807@xxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Tue, Oct 09, 2012 at 05:57:56PM -0500, Eric Sandeen wrote:
> On 10/9/12 5:56 PM, Dave Chinner wrote:
> > On Tue, Oct 09, 2012 at 05:32:30PM -0500, Eric Sandeen wrote:
> >> On 10/9/12 5:28 PM, Dave Chinner wrote:
> >>> On Tue, Oct 09, 2012 at 05:08:26PM -0500, Eric Sandeen wrote:
> >>>> On older systems we may not have preadv/pwritev and/or
> >>>> sync_file_range.
> >>>>
> >>>> Add the configure magic, and stub out the code
> >>>> where needed.
> >>>>
> >>>> (sync_file_range just needed a better test; preadv/pwritev
> >>>> took a little more rearranging)
> >>>>
> >>>> And fix a couple typos ("numberic") while we're at it.
> >>>>
> >>>> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>
> >>>
> >>> Looks fine, though the "HAVE_PREADV" in the pwritev case is a bit
> >>> wierd. Perhaps doing this:
> >>>
> >>>> diff --git a/include/builddefs.in b/include/builddefs.in
> >>>> index 81ebfcd..4a0e910 100644
> >>>> --- a/include/builddefs.in
> >>>> +++ b/include/builddefs.in
> >>>> @@ -101,6 +101,8 @@ HAVE_GETMNTENT = @have_getmntent@
> >>>>  HAVE_GETMNTINFO = @have_getmntinfo@
> >>>>  HAVE_FALLOCATE = @have_fallocate@
> >>>>  HAVE_FIEMAP = @have_fiemap@
> >>>> +HAVE_PREADV = @have_preadv@
> >>>
> >>> HAVE_PWRITEV = @have_preadv@
> >>>
> >>> Would reduce that wierdness a bit when looking at the pwritev
> >>> code....
> >>
> >> Yeah I was a little unhappy w/ that.  But I think then we'd also need:
> >>
> >> +ifeq ($(HAVE_PWRITEV),yes)
> >> +LCFLAGS += -DHAVE_PWRITEV
> >> +endif
> > 
> > True.
> > 
> >> and at some point we might run out of room on the cmdline ;)
> > 
> > Unlikely ;)
> > 
> >> Could change it to HAVE_PREADWRITEV or something ...
> > 
> > HAVE_VECTORED_IO?
> 
> Thought about that but we already have readv/writev, just not *P*readv/writev.

xfs_io doesn't use readv/writev, so I don't think it' a major
concern.

Red, green, blue, the colour of the bike shed  really doesn't matter
that much....

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

<Prev in Thread] Current Thread [Next in Thread>