linux-kernel - Re: [RFC] tracing: Add support for critical section event tracing

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]

Date:   Fri, 30 Jun 2017 14:24:47 -0700
From:   Joel Fernandes <joelaf@...gle.com>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     LKML <linux-kernel@...r.kernel.org>
Subject: Re: [RFC] tracing: Add support for critical section event tracing

On Fri, Jun 30, 2017 at 2:07 PM, Steven Rostedt <rostedt@...dmis.org> wrote:
> On Fri, 30 Jun 2017 14:03:14 -0700
> Joel Fernandes <joelaf@...gle.com> wrote:
>
>> Hi Steven,
>>
>> Thanks a lot for the comments, I agree with all of them and had a
>> comment about one of them:
>>
>> On Fri, Jun 30, 2017 at 5:51 AM, Steven Rostedt <rostedt@...dmis.org> wrote:
>> [..]
>> > Are you not worried about recursion here? There's no protection.
>> > Wouldn't it be better to have:
>> >
>> >         if (!this_cpu_read(tracing_events_cpu))
>> >                 return;
>> >
>> >         trace_critical_end(ip, parent_ip);
>> >
>> >         this_cpu_write(tracing_events_cpu, 0);
>> >
>> > ?
>> >
>>
>> I tried to go over some scenarios and I think it shouldn't be a
>> problem because we start the critical event only when either
>> interrupts are turned off while preemption is turned on, or preempt is
>> turned off while interrupts are turned on, and the fact that we call
>> the tracer while still in the critical section. Let me know if you had
>> a scenario in mind that can cause problems with this.
>
> Then may I ask what is tracing_events_cpu actually protecting?

Yes actually I think its not needed considering the above. When I was
developing the code I wanted to do something similar
start_critical_timing which has a tracing_cpu per-cpu variable, and at
the time wasn't sure if it was needed so I just added it in anyway.

Since the case of this patch is more specific  (both preempt and irq
turned back on to mark end of critical section), it may not be needed
and I can drop it. I will think some more about it as well let you
know if I find a scenario that says otherwise.

Thanks so much for making the patch better!!

Regards,
Joel