Re: [RFC] what to do with IOCB_DSYNC?

From: Jens Axboe <axboe@kernel.dk>
To: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>,
	Matthew Wilcox <willy@infradead.org>,
	linux-fsdevel@vger.kernel.org
Subject: Re: [RFC] what to do with IOCB_DSYNC?
Date: Sun, 22 May 2022 20:43:26 -0600	[thread overview]
Message-ID: <6ea33ba8-c5a3-a1e7-92d2-da8744662ed9@kernel.dk> (raw)
In-Reply-To: <9c3a6ad4-cdb5-8e0d-9b01-c2825ea891ad@kernel.dk>

On 5/22/22 7:50 PM, Jens Axboe wrote:
> On 5/22/22 7:28 PM, Jens Axboe wrote:
>> On 5/22/22 7:22 PM, Jens Axboe wrote:
>>> On 5/22/22 6:42 PM, Al Viro wrote:
>>>> On Sun, May 22, 2022 at 02:03:35PM -0600, Jens Axboe wrote:
>>>>
>>>>> Right, I'm saying it's not _immediately_ clear which cases are what when
>>>>> reading the code.
>>>>>
>>>>>> up a while ago.  And no, turning that into indirect calls ended up with
>>>>>> arseloads of overhead, more's the pity...
>>>>>
>>>>> It's a shame, since indirect calls make for nicer code, but it's always
>>>>> been slower and these days even more so.
>>>>>
>>>>>> Anyway, at the moment I have something that builds; hadn't tried to
>>>>>> boot it yet.
>>>>>
>>>>> Nice!
>>>>
>>>> Boots and survives LTP and xfstests...  Current variant is in
>>>> vfs.git#work.iov_iter (head should be at 27fa77a9829c).  I have *not*
>>>> looked into the code generation in primitives; the likely/unlikely on
>>>> those cascades of ifs need rethinking.
>>>
>>> I noticed too. Haven't fiddled much in iov_iter.c, but for uio.h I had
>>> the below. iov_iter.c is a worse "offender" though, with 53 unlikely and
>>> 22 likely annotations...
>>
>> Here it is...
> 
> Few more, most notably making sure that dio dirties reads even if they
> are not of the iovec type.
> 
> Last two just add a helper for import_ubuf() and then adopts it for
> io_uring send/recv which is also a hot path. The single range read/write
> can be converted too, but that needs a bit more work...

Branch here:

https://git.kernel.dk/cgit/linux-block/log/?h=iov-iter

First 5 are generic ones, and some of them should just be folded with
your changes.

Last 2 are just converting io_uring to use it where appropriate.

We can also use it for vectored readv/writev and recvmsg/sendmsg with
one segment. The latter is mostly single segment in the real world
anyway, former probably too. Though not sure it's worth it when we're
copying a single iovec first anyway? Something to test...

-- 
Jens Axboe