Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220408181523.92322-1-schspa@gmail.com> <20220408184449.GB15609@twin.jikos.cz>
In-Reply-To: <20220408184449.GB15609@twin.jikos.cz>
From:   Schspa Shi <schspa@gmail.com>
Date:   Sat, 9 Apr 2022 15:36:54 +0800
Message-ID: <CAMA88Tp26+AWtcgU5yN=3Q5B8MSxqWMt=BpigQ3XADRJdOrpiA@mail.gmail.com>
Subject: Re: [PATCH] btrfs: zstd: use spin_lock in timer function
To:     dsterba@suse.cz
Cc:     clm@fb.com, Josef Bacik <josef@toxicpanda.com>, dsterba@suse.com,
        terrelln@fb.com, linux-btrfs@vger.kernel.org,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

David Sterba <dsterba@suse.cz> writes:

> On Sat, Apr 09, 2022 at 02:15:23AM +0800, Schspa Shi wrote:
>> timer callback was running on bh, and there is no need to disable bh again.
>
> Why do you think so? There was a specific fix fee13fe96529 ("btrfs:
> correct zstd workspace manager lock to use spin_lock_bh()") that
> actually added the _bh, so either you need to explain why exactly it's
> not needed anymore and verify that the reported lockdep warning from the
> fix does not happen.

Yes, I've seen this fix, and wsm.lru_list is protected by wsm.lock.
This patch will not remove all changes that were fixed. Just a little
improvement
to remove the unnecessary bh disabling. Like
static inline void red_adaptative_timer(struct timer_list *t)
in net/sched/sch_red.c.

Because the critical section is only used by the process context and
the softirq context,
it is safe to remove bh_disable in the softirq context since it will
not be preempted by the softirq.

BRs
Schspa Shi