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;
Subject: Re: [PATCH 18/25] btrfs: Use readahead_batch_length
To:     "Matthew Wilcox (Oracle)" <willy@infradead.org>,
        <linux-fsdevel@vger.kernel.org>, <linux-mm@kvack.org>
CC:     <linux-kernel@vger.kernel.org>
References: <20201216182335.27227-1-willy@infradead.org>
 <20201216182335.27227-19-willy@infradead.org>
From:   John Hubbard <jhubbard@nvidia.com>
Message-ID: <a5b979d7-1086-fe6c-6e82-f20ecb56d24c@nvidia.com>
Date:   Thu, 17 Dec 2020 01:15:10 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:84.0) Gecko/20100101
 Thunderbird/84.0
MIME-Version: 1.0
In-Reply-To: <20201216182335.27227-19-willy@infradead.org>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 12/16/20 10:23 AM, Matthew Wilcox (Oracle) wrote:
> Implement readahead_batch_length() to determine the number of bytes in
> the current batch of readahead pages and use it in btrfs.
> 
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
>   fs/btrfs/extent_io.c    | 6 ++----
>   include/linux/pagemap.h | 9 +++++++++
>   2 files changed, 11 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index 6e3b72e63e42..42936a83a91b 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -4436,10 +4436,8 @@ void extent_readahead(struct readahead_control *rac)
>   	int nr;
>   
>   	while ((nr = readahead_page_batch(rac, pagepool))) {
> -		u64 contig_start = page_offset(pagepool[0]);
> -		u64 contig_end = page_offset(pagepool[nr - 1]) + PAGE_SIZE - 1;
> -
> -		ASSERT(contig_start + nr * PAGE_SIZE - 1 == contig_end);
> +		u64 contig_start = readahead_pos(rac);
> +		u64 contig_end = contig_start + readahead_batch_length(rac);

Something in this tiny change is breaking btrfs: it hangs my Fedora 33 test
system (which changed over to btrfs) on boot. I haven't quite figured out
what's really wrong, but git bisect lands here, *and* turning the whole
extent_readahead() function into a no-op (on top of the whole series)
allows everything to work once again.

Sorry for not actually solving the root cause, but I figured you'd be able
to jump straight to the answer, with the above information, so I'm sending
it out early.


thanks,
-- 
John Hubbard
NVIDIA