Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 30 Apr 2020 14:40:18 -0700
From:   Andrew Morton <akpm@linux-foundation.org>
To:     fdmanana@kernel.org
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        dennis@kernel.org, tj@kernel.org, cl@linux.com,
        linux-btrfs@vger.kernel.org, Filipe Manana <fdmanana@suse.com>
Subject: Re: [PATCH] percpu: make pcpu_alloc() aware of current gfp context
Message-Id: <20200430144018.c855f031b321d68e5c89b30c@linux-foundation.org>
In-Reply-To: <20200430164356.15543-1-fdmanana@kernel.org>
References: <20200430164356.15543-1-fdmanana@kernel.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 30 Apr 2020 17:43:56 +0100 fdmanana@kernel.org wrote:

> From: Filipe Manana <fdmanana@suse.com>
> 
> Since 5.7-rc1, on btrfs we have a percpu counter initialization for which
> we always pass a GFP_KERNEL gfp_t argument (this happens since commit
> 2992df73268f78 ("btrfs: Implement DREW lock")).  That is safe in some
> contextes but not on others where allowing fs reclaim could lead to a
> deadlock because we are either holding some btrfs lock needed for a
> transaction commit or holding a btrfs transaction handle open.  Because
> of that we surround the call to the function that initializes the percpu
> counter with a NOFS context using memalloc_nofs_save() (this is done at
> btrfs_init_fs_root()).
> 
> However it turns out that this is not enough to prevent a possible
> deadlock because percpu_alloc() determines if it is in an atomic context
> by looking exclusively at the gfp flags passed to it (GFP_KERNEL in this
> case) and it is not aware that a NOFS context is set.  Because it thinks
> it is in a non atomic context it locks the pcpu_alloc_mutex, which can
> result in a btrfs deadlock when pcpu_balance_workfn() is running, has
> acquired that mutex and is waiting for reclaim, while the btrfs task that
> called percpu_counter_init() (and therefore percpu_alloc()) is holding
> either the btrfs commit_root semaphore or a transaction handle (done at
> fs/btrfs/backref.c:iterate_extent_inodes()), which prevents reclaim from
> finishing as an attempt to commit the current btrfs transaction will
> deadlock.
> 

Patch looks good and seems sensible, thanks.

But why did btrfs use memalloc_nofs_save()/restore() rather than
s/GFP_KERNEL/GFP_NOFS/?