iommu/iova: Avoid false sharing on fq_timer_on
authorEric Dumazet <edumazet@google.com>
Wed, 28 Aug 2019 13:13:38 +0000 (06:13 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 5 Oct 2019 11:09:53 +0000 (13:09 +0200)
commitc55659cd145904b4386c0600e9131b5350bac9f3
tree7025c96044193b84c70078237801352889dbefcc
parent223b0481ed4b95485e0b749c1a6e5c50f1ff1209
iommu/iova: Avoid false sharing on fq_timer_on

[ Upstream commit 0d87308cca2c124f9bce02383f1d9632c9be89c4 ]

In commit 14bd9a607f90 ("iommu/iova: Separate atomic variables
to improve performance") Jinyu Qi identified that the atomic_cmpxchg()
in queue_iova() was causing a performance loss and moved critical fields
so that the false sharing would not impact them.

However, avoiding the false sharing in the first place seems easy.
We should attempt the atomic_cmpxchg() no more than 100 times
per second. Adding an atomic_read() will keep the cache
line mostly shared.

This false sharing came with commit 9a005a800ae8
("iommu/iova: Add flush timer").

Signed-off-by: Eric Dumazet <edumazet@google.com>
Fixes: 9a005a800ae8 ('iommu/iova: Add flush timer')
Cc: Jinyu Qi <jinyuqi@huawei.com>
Cc: Joerg Roedel <jroedel@suse.de>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/iommu/iova.c