linux-kernel - Re: [patch 10/41] x86/apic: Remove the duplicated tracing versions of interrupts

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: <20170828075041.afwjlxflk4nod4aa@hirez.programming.kicks-ass.net>
Date:   Mon, 28 Aug 2017 09:50:41 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        LKML <linux-kernel@...r.kernel.org>,
        Ingo Molnar <mingo@...nel.org>, Peter Anvin <hpa@...or.com>,
        Andy Lutomirski <luto@...nel.org>,
        Borislav Petkov <bp@...en8.de>
Subject: Re: [patch 10/41] x86/apic: Remove the duplicated tracing versions
 of interrupts

On Fri, Aug 25, 2017 at 11:49:47AM -0400, Steven Rostedt wrote:
> On Fri, 25 Aug 2017 12:31:13 +0200
> Thomas Gleixner <tglx@...utronix.de> wrote:
> 
> > The error and the spurious interrupt are really rare events and not at all
> > so performance sensitive that two NOP5s can not be tolerated when tracing
> > is disabled.
> 
> Just a note. I'm sure if we disassembled it, it may be a little more
> work done than just two NOPs, as parameter passing to the tracepoints
> sometimes leak out of the static jump block. It's moot on this patch,
> but other irqs with fast paths may need to be looked at.

Is that something we can fix with the trace macros?

They have a general shape of:

#define trace_foo(args...)
	if (static_branch_unlikely(&foo_enabled)) {
		__trace_foo(args...);
	}

Right? And I suppose I see why the compiler would want to sometimes lift
stuff out of the branch block, but we'd really like it not to do that.
Would putting a barrier() in front of __trace_foo() help?