linux-kernel - Re: [PATCH] kernel/entry: Remove some redundancy checks on syscall works

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] [day] [month] [year] [list]

Message-ID: <878qlvipxb.ffs@tglx>
Date: Fri, 13 Jun 2025 18:09:36 +0200
From: Thomas Gleixner <tglx@...utronix.de>
To: Khalid Ali <khaliidcaliy@...il.com>, peterz@...radead.org, luto@...nel.org
Cc: linux-kernel@...r.kernel.org, Khalid Ali <khaliidcaliy@...il.com>
Subject: Re: [PATCH] kernel/entry: Remove some redundancy checks on syscall
 works

On Wed, Jun 11 2025 at 11:43, Khalid Ali wrote:
> There is a redundant checks of thread syscall work.

Not really.

> After we read thread syscall work we are checking the work bits using

We are doing nothing. Please write your changelogs in imperative mood
and do not try to impersonate code.

> SYSCALL_WORK_ENTER and SYSCALL_WORK_EXIT on syscall entry and exit
> respectively, and at the same time syscall_trace_enter() and
> syscall_exit_work() checking bits one by one, the bits we already checked.
> This is redundancy. So either we need to check the work bits one by one as I
> did, or check as whole. On my prespective, i think the way code is
> implemented now checking work bits one by one is simpler and gives us
> more granular control.

That's just wrong and absolutely not redundant. Care to look at the
definition of SYSCALL_WORK_ENTER:

#define SYSCALL_WORK_ENTER	(SYSCALL_WORK_SECCOMP |			\
				 SYSCALL_WORK_SYSCALL_TRACEPOINT |	\
				 SYSCALL_WORK_SYSCALL_TRACE |		\
				 SYSCALL_WORK_SYSCALL_EMU |		\
				 SYSCALL_WORK_SYSCALL_AUDIT |		\
				 SYSCALL_WORK_SYSCALL_USER_DISPATCH |	\
				 ARCH_SYSCALL_WORK_ENTER)

So this initial check avoids:

    1) Doing an unconditional out of line call

    2) Checking bit for bit to figure out that there is none set.

Same applies for SYSCALL_WORK_EXIT.

Your change neither makes anything simpler nor provides more granular
control.

All it does is adding overhead and therefore guaranteed to introduce a
performance regression.

Not going to happen.

Thanks,

        tglx