[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAPjX3FfCw3E68XA+XT7rfW+fdUnfaTDMT2sH3_EjNep7hYYD5Q@mail.gmail.com>
Date: Tue, 18 Mar 2025 18:08:55 +0100
From: Daniel Vacek <neelx@...e.com>
To: dsterba@...e.cz
Cc: Chris Mason <clm@...com>, Josef Bacik <josef@...icpanda.com>, David Sterba <dsterba@...e.com>,
linux-btrfs@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] btrfs: remove EXTENT_BUFFER_IN_TREE flag
On Tue, 18 Mar 2025 at 17:04, David Sterba <dsterba@...e.cz> wrote:
>
> On Tue, Mar 18, 2025 at 05:00:17PM +0100, David Sterba wrote:
> > On Tue, Mar 18, 2025 at 10:54:38AM +0100, Daniel Vacek wrote:
> > > This flag is set after inserting the eb to the buffer tree and cleared on
> > > it's removal. But it does not bring any added value. Just kill it for good.
> >
> > I na similar way the flag EXTENT_BUFFER_READ_ERR is unused (was removed
> > in eb/io rework in 046b562b20a5cf ("btrfs: use a separate end_io handler
> > for read_extent_buffer").
>
> And EXTENT_BUFFER_READAHEAD, removed by f26c9238602856 ("btrfs: remove
> reada infrastructure").
I see. I can have a look into further cleanup later. I was not
checking any other flags yet as that's unrelated to my issue.
The full story is I'm chasing an eb refcount issue on downstream v6.4
running on aarch64 with 64k pages. First I suspected it's due to this
flag racing (only setting the flag after inserting a newly allocated
eb into the tree but without holding the eb->refs_lock) but after a
bit of hammering I managed to reproduce the soft lockup even with this
patch applied.
That will need to be solved downstream. Still this looks like a worthy cleanup.
Powered by blists - more mailing lists