linux-kernel - Re: [PATCH v7 2/2] tracing: Add support for preempt and irq enable/disable events

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 for Android: free password hash cracker in your pocket

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

Message-ID: <20171006070711.czekvhnwty6bouxi@hirez.programming.kicks-ass.net>
Date:   Fri, 6 Oct 2017 09:07:11 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Joel Fernandes <joelaf@...gle.com>
Cc:     LKML <linux-kernel@...r.kernel.org>, kernel-team@...roid.com,
        Steven Rostedt <rostedt@...dmis.org>,
        Ingo Molnar <mingo@...nel.org>
Subject: Re: [PATCH v7 2/2] tracing: Add support for preempt and irq
 enable/disable events

On Thu, Oct 05, 2017 at 04:28:10PM -0700, Joel Fernandes wrote:
> > lockdep implements the trace_hardirq_*() in terms of *_caller(). Would
> > that make sense here?
> 
> In lockdep code, when trace_hardirqs_off is called,
> trace_hardirqs_off_caller would pass CALLER_ADDR0 as
> trace_hardirqs_off.
> 
> Because of this, the first argument passed to time_hardirqs_off would
> always be an offset within trace_hardirqs_off:
> time_hardirqs_off(CALLER_ADDR0, ip);
> 
> Is that intended? Seems to me that in the lockdep implementation of
> trace_hardirqs_* in terms of *_caller(), we would completely miss the
> second-last return address (CALLER_ADDR1) of trace_hardirqs_off().
> Also for the above reasons, I don't think it doesn't make sense to use
> this reuse logic for the tracer. Atleast I feel it might change the
> current behavior of the preempt/irqsoff tracer which I don't intend to
> change with my current patch set.

Hurm.. so I've no clue. I never looked at any of this. I think Ingo
wrote that before I came joined the fun. He might (although unlikely,
because its been _many_ years) still have some memories.

But given that lockdep enabled kernels never got those arguments right,
how useful are they? I mean, nobody seems to have noticed much.