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;
From:   laniel_francis@privacyrequired.com
To:     linux-kernel@vger.kernel.org
Cc:     akpm@linux-foundation.org,
        Francis Laniel <laniel_francis@privacyrequired.com>
Subject: [RFC PATCH v2 0/1] Create header for fortified string functions.
Date:   Mon, 11 Jan 2021 10:21:40 +0100
Message-Id: <20210111092141.22946-1-laniel_francis@privacyrequired.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Francis Laniel <laniel_francis@privacyrequired.com>

Hi.


First, I hope your families, friends and yourself are fine.

In a recent mail about the merge of a new fortified string function, Linus
Torvalds suggested the creation of a dedicated header file for these functions:
https://marc.info/?l=linux-mm-commits&m=160810366111244
This will make the code cleaner and also improve compile time for people who do
not set CONFIG_FORTIFY_SOURCE.

So, this patch creates fortify-string.h which is a new header which contains
all the fortified versions of functions declared in string.h.
Since code was moved, I might as well correct the warnings raised by
checkpatch.pl.

I benchmarked the code compilation with and without CONFIG_FORTIFY_SOURCE.
To do this, I compiled 10 times a x86_64_defconfig'ured kernel using make -j4
and cleaning after each compilation.
These compilations were first done without CONFIG_FORTIFY_SOURCE defined,
then with this option defined.
The results were collected using the time bash builtin and are the following
(in seconds, rounded to 10^-3):
|     |   min   |   max   |   mean  | std. dev. |  median | 99th percentile |
| --- | ------- | ------- | ------- | --------- | ------- | --------------- |
| w/o | 524.488 | 526.982 | 525.111 |   0.722   | 524.901 |     526.848     |
| w/  | 529.502 | 531.795 | 529.939 |   0.671   | 529.783 |     531.633     |
First, the results are quite stable as shown by the standard deviation
(less than 1 second).
On average, compile time without CONFIG_FORTIFY_SOURCE is 0.919% faster.
For the median case, compiling without setting this option is  0.930% faster.
Finally, with the 99th percentile, not using CONFIG_FORTIFY_SOURCE is 0.908%
faster.

Globally, using a different header seems to provide a roughly 1% faster compile
time for people who do not set CONFIG_FORTIFY_SOURCE.
This is not a huge gain... but still a gain!
Especially on compilation which is an operation kernel developers do a lot.

So, I await your opinions and reviews on this patch.

Version history:
* v2: Correct misguarded instructions which follow if.


Best regards.

Francis Laniel (1):
  string.h: Move fortified functions definitions in a dedicated header.

 include/linux/fortify-string.h | 302 +++++++++++++++++++++++++++++++++
 include/linux/string.h         | 282 +-----------------------------
 2 files changed, 303 insertions(+), 281 deletions(-)
 create mode 100644 include/linux/fortify-string.h

-- 
2.20.1