linux-kernel - Re: [PATCH 0/2] optimise sqe-to-req flags

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <fa7b4b9a-b0a4-7c1f-d3e9-c5bfa8fff74b@gmail.com>
Date:   Sat, 18 Jan 2020 02:14:56 +0300
From:   Pavel Begunkov <asml.silence@...il.com>
To:     Jens Axboe <axboe@...nel.dk>, io-uring@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 0/2] optimise sqe-to-req flags

On 18/01/2020 01:49, Jens Axboe wrote:
> On 1/17/20 3:41 PM, Pavel Begunkov wrote:
>> *lost the cover-letter, but here we go*
>>
>> The main idea is to optimise code like the following by directly
>> copying sqe flags:
>>
>> if (sqe_flags & IOSQE_IO_HARDLINK)
>> 	req->flags |= REQ_F_HARDLINK;
>>
>> The first patch is a minor cleanup, and the second one do the
>> trick. No functional changes.
>>
>> The other thing to consider is whether to use such flags as 
>> REQ_F_LINK = IOSQE_IO_LINK, or directly use IOSQE_IO_LINK instead.
> 
> I think we should keep the names separate. I think it looks fine, though
> I do wish that we could just have both in an enum and not have to do
> weird naming. We sometimes do:
> 
> enum {
> 	__REQ_F_FOO
> };
> 
> #define REQ_F_FOO	(1U << __REQ_F_FOO)
>

I thought it will be too bulky as it needs retyping the same name many times.
Though, it solves numbering problem and is less error-prone indeed. Let me to
play with it a bit.

BTW, there is another issue from development perspective -- it's harder to find
from where a flag is came. E.g. search for REQ_F_FOO won't give you a place,
where it was set. SQE_INHERITED_FLAGS is close in the code to its usage exactly
for this reason.


> and with that we could have things Just Work in terms of numbering, if
> we keep the overlapped values at the start. Would need IOSQE_* to also
> use an enum, ala:
> 
> enum {
> 	__IOSQE_FIXED_FILE,
> 	__IOSQE_IO_DRAIN,
> 	...
> };
> 

I tried to not modify the userspace header. Wouldn't it be better to add _BIT
postfix instead of the underscores?

> and then do
> 
> #define IOSQE_FIXED_FILE	(1U << __IOSQE_FIXED_FILE)
> 
> and have the __REQ enum start with
> 
> enum {
> 	__REQ_F_FIXED_FILE = __IOSQE_FIXED_FILE,
> 	__REQ_F_IO_DRAIN = __IOSQE_IO_DRAIN,
> 	...
> 	__REQ_F_LINK_NEXT,
> 	__REQ_F_FAIL_LINK,
> 	...
> };
> 

-- 
Pavel Begunkov



Download attachment "signature.asc" of type "application/pgp-signature" (834 bytes)