net: sched: add barrier to ensure correct ordering for lockless qdisc
authorYunsheng Lin <linyunsheng@huawei.com>
Thu, 17 Jun 2021 01:04:14 +0000 (09:04 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 14 Jul 2021 14:56:24 +0000 (16:56 +0200)
commite7c3ae47978f97f528d95b0c86de51896e78d9f0
treeb88ffda017907aeb8fe7bd56da615ab3db21b7d1
parentca9b5ab7910cdc4dd02219ebfe7e468709bbafac
net: sched: add barrier to ensure correct ordering for lockless qdisc

[ Upstream commit 89837eb4b2463c556a123437f242d6c2bc62ce81 ]

The spin_trylock() was assumed to contain the implicit
barrier needed to ensure the correct ordering between
STATE_MISSED setting/clearing and STATE_MISSED checking
in commit a90c57f2cedd ("net: sched: fix packet stuck
problem for lockless qdisc").

But it turns out that spin_trylock() only has load-acquire
semantic, for strongly-ordered system(like x86), the compiler
barrier implicitly contained in spin_trylock() seems enough
to ensure the correct ordering. But for weakly-orderly system
(like arm64), the store-release semantic is needed to ensure
the correct ordering as clear_bit() and test_bit() is store
operation, see queued_spin_lock().

So add the explicit barrier to ensure the correct ordering
for the above case.

Fixes: a90c57f2cedd ("net: sched: fix packet stuck problem for lockless qdisc")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
include/net/sch_generic.h