On 2/10/21 11:17 PM, Jason Gunthorpe wrote:

On Fri, Feb 05, 2021 at 08:41:27PM +0000, Joao Martins wrote:

quoted

Use the newly added unpin_user_page_range_dirty_lock()
for more quickly unpinning a consecutive range of pages
represented as compound pages. This will also calculate
number of pages to unpin (for the tail pages which matching
head page) and thus batch the refcount update.

Running a test program which calls mr reg/unreg on a 1G in size
and measures cost of both operations together (in a guest using rxe)
with THP and hugetlbfs:

Before:
590 rounds in 5.003 sec: 8480.335 usec / round
6898 rounds in 60.001 sec: 8698.367 usec / round

After:
2688 rounds in 5.002 sec: 1860.786 usec / round
32517 rounds in 60.001 sec: 1845.225 usec / round

Signed-off-by: Joao Martins <redacted>
---
 drivers/infiniband/core/umem.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

Would best for this to go through Andrew's tree

Acked-by: Jason Gunthorpe <jgg@nvidia.com>

4x improvement is pretty good!

It would only be half of that improvement if it wasn't for your
unpin_user_page_range_dirty_lock() suggestion, so thanks for all
the input :)

	Joao