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 v3 4/4] io_uring: add support for zone-append
To:     Matthew Wilcox <willy@infradead.org>
Cc:     Kanchan Joshi <joshi.k@samsung.com>, viro@zeniv.linux.org.uk,
        bcrl@kvack.org, hch@infradead.org, Damien.LeMoal@wdc.com,
        asml.silence@gmail.com, linux-fsdevel@vger.kernel.org,
        mb@lightnvm.io, linux-kernel@vger.kernel.org, linux-aio@kvack.org,
        io-uring@vger.kernel.org, linux-block@vger.kernel.org,
        Selvakumar S <selvakuma.s1@samsung.com>,
        Nitesh Shetty <nj.shetty@samsung.com>,
        Javier Gonzalez <javier.gonz@samsung.com>
References: <fe0066b7-5380-43ee-20b2-c9b17ba18e4f@kernel.dk>
 <20200705210947.GW25523@casper.infradead.org>
 <239ee322-9c38-c838-a5b2-216787ad2197@kernel.dk>
 <20200706141002.GZ25523@casper.infradead.org>
 <4a9bf73e-f3ee-4f06-7fad-b8f8861b0bc1@kernel.dk>
 <20200706143208.GA25523@casper.infradead.org>
 <20200707151105.GA23395@test-zns>
 <20200707155237.GM25523@casper.infradead.org>
 <20200707202342.GA28364@test-zns>
 <7a44d9c6-bf7d-0666-fc29-32c3cba9d1d8@kernel.dk>
 <20200707221812.GN25523@casper.infradead.org>
From:   Jens Axboe <axboe@kernel.dk>
Message-ID: <145cc0ad-af86-2d6a-78b3-9ade007aae52@kernel.dk>
Date:   Tue, 7 Jul 2020 16:37:55 -0600
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.8.0
MIME-Version: 1.0
In-Reply-To: <20200707221812.GN25523@casper.infradead.org>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 7/7/20 4:18 PM, Matthew Wilcox wrote:
> On Tue, Jul 07, 2020 at 02:40:06PM -0600, Jens Axboe wrote:
>>>> so we have another 24 bytes before io_kiocb takes up another cacheline.
>>>> If that's a serious problem, I have an idea about how to shrink struct
>>>> kiocb by 8 bytes so struct io_rw would have space to store another
>>>> pointer.
>>> Yes, io_kiocb has room. Cache-locality wise whether that is fine or
>>> it must be placed within io_rw - I'll come to know once I get to
>>> implement this. Please share the idea you have, it can come handy.
>>
>> Except it doesn't, I'm not interested in adding per-request type fields
>> to the generic part of it. Before we know it, we'll blow past the next
>> cacheline.
>>
>> If we can find space in the kiocb, that'd be much better. Note that once
>> the async buffered bits go in for 5.9, then there's no longer a 4-byte
>> hole in struct kiocb.
> 
> Well, poot, I was planning on using that.  OK, how about this:

Figured you might have had your sights set on that one, which is why I
wanted to bring it up upfront :-)

> +#define IOCB_NO_CMPL		(15 << 28)
> 
>  struct kiocb {
> [...]
> -	void (*ki_complete)(struct kiocb *iocb, long ret, long ret2);
> +	loff_t __user *ki_uposp;
> -	int			ki_flags;
> +	unsigned int		ki_flags;
> 
> +typedef void ki_cmpl(struct kiocb *, long ret, long ret2);
> +static ki_cmpl * const ki_cmpls[15];
> 
> +void ki_complete(struct kiocb *iocb, long ret, long ret2)
> +{
> +	unsigned int id = iocb->ki_flags >> 28;
> +
> +	if (id < 15)
> +		ki_cmpls[id](iocb, ret, ret2);
> +}
> 
> +int kiocb_cmpl_register(void (*cb)(struct kiocb *, long, long))
> +{
> +	for (i = 0; i < 15; i++) {
> +		if (ki_cmpls[id])
> +			continue;
> +		ki_cmpls[id] = cb;
> +		return id;
> +	}
> +	WARN();
> +	return -1;
> +}

That could work, we don't really have a lot of different completion
types in the kernel.

-- 
Jens Axboe