On 11/17/2016 05:50 AM, Christoph Hellwig wrote:

On Wed, Nov 16, 2016 at 09:43:57PM -0700, Jens Axboe wrote:

quoted

OK, I'm getting reasonably happy with it now:

http://git.kernel.dk/cgit/linux-block/log/?h=for-4.10/hch-dio

Sharing the FUA setting to avoid the manual flush between the two, and
removing the remnants of SYNC support from __blkdev_direct_IO().

I'll fire off some overnight testing now and leave it for the day.

This looks reasonable, a couple questions and comments:

 - why the 8 * BIO_MAX_PAGES in the call to iov_iter_npages,
   that probably needs a comment

Looks like I forgot to push out the latest changes, that is gone. We're
just using + 1 now to see if we should use the multibio function or not.

 - I'll probably need to write up a proper changelog and fold
   in at least the multi_bio patch.

That's fine.

 - I'd just return the op from dio_bio_set_write_op instead of
   passing the bio, that reads a lot easier.

Fine with me.

 - I should probably get rid of blkdev_dio_pool in favour
   of bio_kmalloc

Agree, though not a requirement imho.

I can look into doing these fixups this afternoon and prepare a tre
for you.

Sounds good.

-- 
Jens Axboe