linux-kernel - [PATCH v8 0/2] tracing: Add trace events for preemption and irq disable/enable

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-next>] [day] [month] [year] [list]

Date:   Thu,  5 Oct 2017 17:54:30 -0700
From:   Joel Fernandes <joelaf@...gle.com>
To:     linux-kernel@...r.kernel.org
Cc:     kernel-team@...roid.com, Joel Fernandes <joelaf@...gle.com>,
        Steven Rostedt <rostedt@...dmis.org>,
        Peter Zilstra <peterz@...radead.org>
Subject: [PATCH v8 0/2] tracing: Add trace events for preemption and irq disable/enable

These patches add trace events support for preempt and irq disable/enable
events.

Here's an example of how Android's systrace will be using it to show atomic
sections as a gantt chart: http://imgur.com/download/TZplEVp
Other advantages of this initial work could be rewriting of preemptirqs off
tracer to use trace events, and replacing kprobes with tracepoint hooks for
these events in BPF samples (see samples/bpf/lathist_kern.c).

Changes since v7:
* Defining the tracepoints when they are unused gives a false impression to the
user that the tracepoints are present but unused, for this reason:
- preempt toggle tracepoints are unused when CONFIG_PREEMPT_DEBUG is off, lets
  not define them.
- irq toggle tracepoints are unused when CONFIG_PROVE_LOCKING is on, lets not
  define them.
I think in future patches, we should also do the same to the irqsoff tracer
since it can appear that the tracer is not working when the issue is hooks
aren't called.
Also in future patches, we should unify the trace_hardirqs* paths of
trace_irqsoff.c and lockdep.c and avoid duplication of the 2 paths - but I
leave that for a future patch set as I don't want to disrupt the functionality
of irqsoff tracer or lockdep in this patchset and increase the risk of breaking
something else. I am focusing on getting the tracepoints in first in this set.
Hope that's Ok.

* Clarified comment about per-CPU variable used for protection in
trace_hardirqs_*

Joel Fernandes (2):
  tracing: Prepare to add preempt and irq trace events
  tracing: Add support for preempt and irq enable/disable events

 include/linux/ftrace.h            |   3 +-
 include/trace/events/preemptirq.h |  70 ++++++++++++++++++++
 kernel/trace/Kconfig              |  11 ++++
 kernel/trace/Makefile             |   1 +
 kernel/trace/trace_irqsoff.c      | 133 ++++++++++++++++++++++++++++++--------
 5 files changed, 191 insertions(+), 27 deletions(-)
 create mode 100644 include/trace/events/preemptirq.h

Cc: Steven Rostedt <rostedt@...dmis.org>
Cc: Peter Zilstra <peterz@...radead.org>
Cc: kernel-team@...roid.com
-- 
2.14.2.920.gcf0c67979c-goog