Currently the log backend prints the process id of QEMU at the start
of each output line, but since threads share the same PID there is no
clear distinction between their outputs.
Having the thread id present in the log makes it easier to see when
output comes from different threads. E.g.:
12423@1538597569.672527:qemu_mutex_lock waiting on mutex 0x1103ee60 (/root/qemu/util/main-loop.c:236)
...
12430@1538597569.503928:qemu_mutex_unlock released mutex 0x1103ee60 (/root/qemu/cpus.c:1238)
12431@1538597569.503937:qemu_mutex_locked taken mutex 0x1103ee60 (/root/qemu/cpus.c:1257)
^here
In the above, 12423 is the main process id and 12430 & 12431 are the
two vcpu threads.
(qemu) info cpus
* CPU #0: thread_id=12430
CPU #1: thread_id=12431
Suggested-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Signed-off-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This fixes when using GCC with -Wformat-signedness:
migration/trace.h: In function ‘_nocheck__trace_dirty_bitmap_load_success’:
migration/trace.h:6368:24: error: format ‘%zd’ expects argument of type ‘signed size_t’, but argument 3 has type ‘long unsigned int’ [-Werror=format=]
qemu_log("%d@%zd.%06zd:dirty_bitmap_load_success " "" "\n",
~~^
%ld
migration/trace.h:6370:18:
(size_t)_now.tv_sec, (size_t)_now.tv_usec
~~~~~~~~~~~~~~~~~~~
migration/trace.h:6368:30: error: format ‘%zd’ expects argument of type ‘signed size_t’, but argument 4 has type ‘long unsigned int’ [-Werror=format=]
qemu_log("%d@%zd.%06zd:dirty_bitmap_load_success " "" "\n",
~~~~^
%06ld
migration/trace.h:6370:39:
(size_t)_now.tv_sec, (size_t)_now.tv_usec
~~~~~~~~~~~~~~~~~~~~
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
A persistent build problem we see is where a source file
accidentally omits the #include of log.h. This slips through
local developer testing because if you configure with the
default (log) trace backend trace.h will pull in log.h for you.
Compilation fails only if some other backend is selected.
To make this error cause a compile failure regardless of
the configured trace backend, split out the parts of log.h
that trace.h requires into a new log-for-trace.h header.
Since almost all manual uses of the log.h functions will
use constants or functions which aren't in log-for-trace.h,
this will let us catch missing #include "qemu/log.h" more
consistently.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20180213140029.8308-1-peter.maydell@linaro.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
QEMU keeps track of trace event enabled/disabled state and provides
monitor commands to inspect and modify the "dstate". SystemTap and
LTTng UST maintain independent enabled/disabled states for each trace
event, the other backends rely on QEMU dstate.
Introduce a new per-event macro that combines backend-specific dstate
like this:
#define TRACE_MY_EVENT_BACKEND_DSTATE() ( \
QEMU_MY_EVENT_ENABLED() || /* SystemTap */ \
tracepoint_enabled(qemu, my_event) /* LTTng UST */ || \
false)
This will be used to extend trace_event_get_state() in the next patch.
[Daniel Berrange pointed out that QEMU_MY_EVENT_ENABLED() must be true
by default, not false. This way events will fire even if the DTrace
implementation does not implement the SystemTap semaphores feature.
Ubuntu Precise uses lttng-ust-dev 2.0.2 which does not have
tracepoint_enabled(), so we need a compatibility wrapper to keep Travis
builds passing.
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20170731140718.22010-2-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
fixup! trace: add TRACE_<event>_BACKEND_DSTATE()
Last patch removed a nesting level in generated code. Re-align all code
generated by backends to be 4-column aligned.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-id: 149915824586.6295.17820926011082409033.stgit@frigg.lan
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Introduce rules in the top level Makefile that are able to generate
trace.[ch] files in every subdirectory which has a trace-events file.
The top level directory is handled specially, so instead of creating
trace.h, it creates trace-root.h. This allows sub-directories to
include the top level trace-root.h file, without ambiguity wrt to
the trace.g file in the current sub-dir.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-id: 20170125161417.31949-7-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Currently, the generated function body will do "strlen(arg)" but the
argument could be 'char **' or 'char * const *'. Avoid that by excluding
such cases in is_string check.
Reported by patchew's "make docker-test-mingw@fedora".
Suggested-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1477453806-21097-1-git-send-email-famz@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The declarations in the generated-tracers.h file are
assuming there's only ever going to be one instance
of this header, as they are not namespaced. When we
have one header per event group, if a single source
file needs to include multiple sets of trace events,
the symbols will all clash.
This change thus introduces a '--group NAME' arg to the
'tracetool' program. This will cause all the symbols in
the generated header files to be given a unique namespace.
If no group is given, the group name 'common' is used,
which is suitable for the current usage where there is
only one global trace-events file used for code generation.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>
Message-id: 1475588159-30598-21-git-send-email-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The TraceEventID and TraceEventVCPUID enums constants are
no longer actually used for anything critical.
The TRACE_EVENT_COUNT limit is used to determine the size
of the TraceEvents array, and can be removed if we just
NULL terminate the array instead.
The TRACE_VCPU_EVENT_COUNT limit is used as a magic value
for marking non-vCPU events, and also for declaring the
size of the trace dstate mask in the CPUState struct.
The former usage can be replaced by a dedicated constant
TRACE_EVENT_VCPU_NONE, defined as (uint32_t)-1. For the
latter usage, we can simply define a constant for the
number of VCPUs, avoiding the need for the full enum.
The only other usages of the enum values can be replaced
by accesing the id/vcpu_id fields via the named TraceEvent
structs.
Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-id: 1475588159-30598-11-git-send-email-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The format/h.py file adds an include for control.h to
generated-tracers.h. ftrace, log and syslog, then
add more duplicate includes for control.h.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-id: 1475588159-30598-8-git-send-email-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch adds a tracing backend which sends output using syslog().
The syslog backend is limited to POSIX compliant systems.
openlog() is called with facility set to LOG_DAEMON, with the LOG_PID
option. Trace events are logged at level LOG_INFO.
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Message-id: 1470318254-29989-1-git-send-email-paul.durrant@citrix.com
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Events with the 'vcpu' property are conditionally emitted according to
their per-vCPU state. Other events are emitted normally based on their
global tracing state.
Note that the per-vCPU condition check applies to all tracing backends.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Manually drop redundant includes that scripts/clean-includes misses,
e.g. because they're hidden in generator programs, or they use the
wrong kind of delimiter.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Include qemu/osdep.h as the first include in generated .c files,
so they don't implicitly rely on some other included header
to pull it in.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
When debugging migration it's useful to know the PID of
each trace message so you can figure out if it came from the source
or the destination.
Printing the time makes it easy to do latency measurements or timings
between trace points.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-id: 1421746875-9962-1-git-send-email-dgilbert@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Red Hat Enterprise Linux 5 ships Python 2.4.3. The all() function was
added in Python 2.5 so we cannot use it.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Benoît Canet <benoit.canet@nodalink.com>
Adds support to compile QEMU with multiple tracing backends at the same time.
For example, you can compile QEMU with:
$ ./configure --enable-trace-backends=ftrace,dtrace
Where 'ftrace' can be handy for having an in-flight record of events, and 'dtrace' can be later used to extract more information from the system.
This patch allows having both available without recompiling QEMU.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Backends now only contain the essential backend-specific code, and most of the work is moved to frontend code.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Makes it easier to ensure proper naming across the different frontends and backends.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
* A new format is required to generate definitions for ust tracepoints.
Files ust_events_h.py and ust_events_c.py define common macros, while
new function ust_events_h in events.py does the actual definition of
each tracepoint.
* ust.py generates the new interface for calling userspace tracepoints
with LTTng 2.x, replacing trace_name(args) to tracepoint(name, args).
* As explained in ust_events_c.py, -Wredundant-decls gives a warning
when compiling with gcc 4.7 or older. This is specific to lttng-ust so
for now use a pragma clause to avoid getting a warning.
Signed-off-by: Mohamad Gebai <mohamad.gebai@polymtl.ca>
Reviewed-by: Alex Bennée <alex@bennee.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The trace-events "disable" keyword turns an event into a nop at
compile-time. This is important for high-frequency events that can
impact performance.
The "disable" keyword is currently broken in the simple trace backend.
This patch fixes the problem as follows:
Trace events are identified by their TraceEventID number. When events
are disabled there are two options for assigning TraceEventID numbers:
1. Skip disabled events and don't assign them a number.
2. Assign numbers for all events regardless of the disabled keyword.
The simple trace backend and its binary file format uses approach #1.
The tracetool infrastructure has been using approach #2 for a while.
The result is that the numbers used in simple trace files do not
correspond with TraceEventIDs. In trace/simple.c we assumed that they
are identical and therefore emitted bogus numbers.
This patch fixes the bug by using TraceEventID for trace_event_id()
while sticking to approach #1 for simple trace file numbers. This
preserves simple trace file format compatibility.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The header is not necessary, given that the simple backend does not define any
inlined tracing routines.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch adds a ftrace tracing backend which sends trace event to
ftrace marker file. You can effectively compare qemu trace data and
kernel(especially, kvm.ko when using KVM) trace data.
The ftrace backend is restricted to Linux only.
To try out the ftrace backend:
$ ./configure --trace-backend=ftrace
$ make
if you use KVM, enable kvm events in ftrace:
# sudo echo 1 > /sys/kernel/debug/tracing/events/kvm/enable
After running qemu by root user, you can get the trace:
# cat /sys/kernel/debug/tracing/trace
Signed-off-by: Eiichi Tsukata <eiichi.tsukata.xh@hitachi.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The backend is forced to dump event numbers using 64 bits, as TraceEventID is
an enum.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Uses tracetool to generate a backend-independent tracing event description
(struct TraceEvent).
The values for such structure are generated with the non-public "events"
backend ("events-c" frontend).
The generation of the defines to check if an event is statically enabled is also
moved to the "events" backend ("events-h" frontend).
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Public backends are those printed by "--list-backends" and thus considered valid
by the configure script.
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
--
Changes in v2:
* Do not depend on "qemu-timer-common.o".
* Use "$(obj)" in rules to refer to the build sub-directory.
* Remove dependencies against "$(GENERATED_HEADERS)".
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Over time various systemtap reserved words have been blacklisted
in the trace backend generator. The list is not complete though,
so there is continued risk of problems in the future. Preempt
such problems by specifying the full list of systemtap keywords
listed in its parser as identified here:
http://sourceware.org/ml/systemtap/2012-q4/msg00157.html
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
gcc complains when a 32 bit pointer is casted to a 64 bit integer.
Cc: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Existing simpletrace backend allows to trace at max 6 args and does not
support strings. This newer tracelog format gets rid of fixed size records
and therefore allows to trace variable number of args including strings.
Sample trace with strings:
v9fs_version 0.000 tag=0xffff id=0x64 msize=0x2000 version=9P2000.L
v9fs_version_return 6.705 tag=0xffff id=0x64 msize=0x2000 version=9P2000.L
Signed-off-by: Harsh Prateek Bora <harsh@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
The pkgutil.iter_modules() function provides a way to enumerate child
modules. Unfortunately it's missing in Python <2.7 so we must implement
similar behavior ourselves.
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>
Language keywords cannot be used as argument names. The DTrace backend
appends an underscore to the argument name in order to make the argument
name legal.
This patch adds 'in', 'next', and 'self' keywords to dtrace.py.
Also drop the unnecessary argument name lstrip() call. The
Arguments.build() method already ensures there is no space around
argument names. Furthermore it is misleading to do the lstrip() *after*
checking against keywords because the keyword check would not match if
spaces were in the name.
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Reviewed-by: Alon Levy <alevy@redhat.com>
Reviewed-by: Lluís Vilanova <vilanova@ac.upc.edu>
The tracetool script is written in shell and has hit several portability
problems due to shell quirks or external tools across host platforms.
Additionally the amount of string processing and lack of real data
structures makes it tough to implement code generator backends for
tracers that are more complex.
This patch replaces the shell version of tracetool with a Python
version. The new tracetool design is:
scripts/tracetool.py - top-level script
scripts/tracetool/backend/ - tracer backends live here (simple, ust)
scripts/tracetool/format/ - output formats live here (.c, .h)
There is common code for trace-events definition parsing so that
backends can focus on generating code rather than parsing input.
Support for all existing backends (nop, stderr, simple, ust,
and dtrace) is added back in follow-up patches.
[Commit description written by Stefan Hajnoczi]
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Fabiano Rosas