Implementation of kernel tracepoints. Inspired from the Linux Kernel
Markers. Allows complete typing verification by declaring both tracing
statement inline functions and probe registration/unregistration static
inline functions within the same macro "DEFINE_TRACE". No format string
is required. See the tracepoint Documentation and Samples patches for
usage examples.
Taken from the documentation patch :
"A tracepoint placed in code provides a hook to call a function (probe)
that you can provide at runtime. A tracepoint can be "on" (a probe is
connected to it) or "off" (no probe is attached). When a tracepoint is
"off" it has no effect, except for adding a tiny time penalty (checking
a condition for a branch) and space penalty (adding a few bytes for the
function call at the end of the instrumented function and adds a data
structure in a separate section). When a tracepoint is "on", the
function you provide is called each time the tracepoint is executed, in
the execution context of the caller. When the function provided ends its
execution, it returns to the caller (continuing from the tracepoint
site).
You can put tracepoints at important locations in the code. They are
lightweight hooks that can pass an arbitrary number of parameters, which
prototypes are described in a tracepoint declaration placed in a header
file."
Addition and removal of tracepoints is synchronized by RCU using the
scheduler (and preempt_disable) as guarantees to find a quiescent state
(this is really RCU "classic"). The update side uses rcu_barrier_sched()
with call_rcu_sched() and the read/execute side uses
"preempt_disable()/preempt_enable()".
We make sure the previous array containing probes, which has been
scheduled for deletion by the rcu callback, is indeed freed before we
proceed to the next update. It therefore limits the rate of modification
of a single tracepoint to one update per RCU period. The objective here
is to permit fast batch add/removal of probes on _different_
tracepoints.
Changelog :
- Use #name ":" #proto as string to identify the tracepoint in the
tracepoint table. This will make sure not type mismatch happens due to
connexion of a probe with the wrong type to a tracepoint declared with
the same name in a different header.
- Add tracepoint_entry_free_old.
- Change __TO_TRACE to get rid of the 'i' iterator.
Masami Hiramatsu <mhiramat@redhat.com> :
Tested on x86-64.
Performance impact of a tracepoint : same as markers, except that it
adds about 70 bytes of instructions in an unlikely branch of each
instrumented function (the for loop, the stack setup and the function
call). It currently adds a memory read, a test and a conditional branch
at the instrumentation site (in the hot path). Immediate values will
eventually change this into a load immediate, test and branch, which
removes the memory read which will make the i-cache impact smaller
(changing the memory read for a load immediate removes 3-4 bytes per
site on x86_32 (depending on mov prefixes), or 7-8 bytes on x86_64, it
also saves the d-cache hit).
About the performance impact of tracepoints (which is comparable to
markers), even without immediate values optimizations, tests done by
Hideo Aoki on ia64 show no regression. His test case was using hackbench
on a kernel where scheduler instrumentation (about 5 events in code
scheduler code) was added.
Quoting Hideo Aoki about Markers :
I evaluated overhead of kernel marker using linux-2.6-sched-fixes git
tree, which includes several markers for LTTng, using an ia64 server.
While the immediate trace mark feature isn't implemented on ia64, there
is no major performance regression. So, I think that we don't have any
issues to propose merging marker point patches into Linus's tree from
the viewpoint of performance impact.
I prepared two kernels to evaluate. The first one was compiled without
CONFIG_MARKERS. The second one was enabled CONFIG_MARKERS.
I downloaded the original hackbench from the following URL:
http://devresources.linux-foundation.org/craiger/hackbench/src/hackbench.c
I ran hackbench 5 times in each condition and calculated the average and
difference between the kernels.
The parameter of hackbench: every 50 from 50 to 800
The number of CPUs of the server: 2, 4, and 8
Below is the results. As you can see, major performance regression
wasn't found in any case. Even if number of processes increases,
differences between marker-enabled kernel and marker- disabled kernel
doesn't increase. Moreover, if number of CPUs increases, the differences
doesn't increase either.
Curiously, marker-enabled kernel is better than marker-disabled kernel
in more than half cases, although I guess it comes from the difference
of memory access pattern.
* 2 CPUs
Number of | without | with | diff | diff |
processes | Marker [Sec] | Marker [Sec] | [Sec] | [%] |
--------------------------------------------------------------
50 | 4.811 | 4.872 | +0.061 | +1.27 |
100 | 9.854 | 10.309 | +0.454 | +4.61 |
150 | 15.602 | 15.040 | -0.562 | -3.6 |
200 | 20.489 | 20.380 | -0.109 | -0.53 |
250 | 25.798 | 25.652 | -0.146 | -0.56 |
300 | 31.260 | 30.797 | -0.463 | -1.48 |
350 | 36.121 | 35.770 | -0.351 | -0.97 |
400 | 42.288 | 42.102 | -0.186 | -0.44 |
450 | 47.778 | 47.253 | -0.526 | -1.1 |
500 | 51.953 | 52.278 | +0.325 | +0.63 |
550 | 58.401 | 57.700 | -0.701 | -1.2 |
600 | 63.334 | 63.222 | -0.112 | -0.18 |
650 | 68.816 | 68.511 | -0.306 | -0.44 |
700 | 74.667 | 74.088 | -0.579 | -0.78 |
750 | 78.612 | 79.582 | +0.970 | +1.23 |
800 | 85.431 | 85.263 | -0.168 | -0.2 |
--------------------------------------------------------------
* 4 CPUs
Number of | without | with | diff | diff |
processes | Marker [Sec] | Marker [Sec] | [Sec] | [%] |
--------------------------------------------------------------
50 | 2.586 | 2.584 | -0.003 | -0.1 |
100 | 5.254 | 5.283 | +0.030 | +0.56 |
150 | 8.012 | 8.074 | +0.061 | +0.76 |
200 | 11.172 | 11.000 | -0.172 | -1.54 |
250 | 13.917 | 14.036 | +0.119 | +0.86 |
300 | 16.905 | 16.543 | -0.362 | -2.14 |
350 | 19.901 | 20.036 | +0.135 | +0.68 |
400 | 22.908 | 23.094 | +0.186 | +0.81 |
450 | 26.273 | 26.101 | -0.172 | -0.66 |
500 | 29.554 | 29.092 | -0.461 | -1.56 |
550 | 32.377 | 32.274 | -0.103 | -0.32 |
600 | 35.855 | 35.322 | -0.533 | -1.49 |
650 | 39.192 | 38.388 | -0.804 | -2.05 |
700 | 41.744 | 41.719 | -0.025 | -0.06 |
750 | 45.016 | 44.496 | -0.520 | -1.16 |
800 | 48.212 | 47.603 | -0.609 | -1.26 |
--------------------------------------------------------------
* 8 CPUs
Number of | without | with | diff | diff |
processes | Marker [Sec] | Marker [Sec] | [Sec] | [%] |
--------------------------------------------------------------
50 | 2.094 | 2.072 | -0.022 | -1.07 |
100 | 4.162 | 4.273 | +0.111 | +2.66 |
150 | 6.485 | 6.540 | +0.055 | +0.84 |
200 | 8.556 | 8.478 | -0.078 | -0.91 |
250 | 10.458 | 10.258 | -0.200 | -1.91 |
300 | 12.425 | 12.750 | +0.325 | +2.62 |
350 | 14.807 | 14.839 | +0.032 | +0.22 |
400 | 16.801 | 16.959 | +0.158 | +0.94 |
450 | 19.478 | 19.009 | -0.470 | -2.41 |
500 | 21.296 | 21.504 | +0.208 | +0.98 |
550 | 23.842 | 23.979 | +0.137 | +0.57 |
600 | 26.309 | 26.111 | -0.198 | -0.75 |
650 | 28.705 | 28.446 | -0.259 | -0.9 |
700 | 31.233 | 31.394 | +0.161 | +0.52 |
750 | 34.064 | 33.720 | -0.344 | -1.01 |
800 | 36.320 | 36.114 | -0.206 | -0.57 |
--------------------------------------------------------------
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Acked-by: Masami Hiramatsu <mhiramat@redhat.com>
Acked-by: 'Peter Zijlstra' <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Mathieu Desnoyers