Currently the tracing 'log' back emits special code to add timestamps
to trace points sent via qemu_log(). This current impl is a bad design
for a number of reasons.
* It changes the QEMU headers, such that 'error-report.h' content
is visible to all files using tracing, but only when the 'log'
backend is enabled. This has led to build failure bugs as devs
rarely test without the (default) 'log' backend enabled, and
CI can't cover every scenario for every trace backend.
* It bloats the trace points definitions which are inlined into
every probe location due to repeated inlining of timestamp
formatting code, adding MBs of overhead to QEMU.
* The tracing subsystem should not be treated any differently
from other users of qemu_log. They all would benefit from
having timestamps present.
* The timestamp emitted with the tracepoints is in a needlessly
different format to that used by error_report() in response
to '-msg timestamp=on'.
This fixes all these issues simply by moving timestamp formatting
into qemu_log, using the same approach as for error_report.
The code before:
static inline void _nocheck__trace_qcrypto_tls_creds_get_path(void * creds, const char * filename, const char * path)
{
if (trace_event_get_state(TRACE_QCRYPTO_TLS_CREDS_GET_PATH) && qemu_loglevel_mask(LOG_TRACE)) {
if (message_with_timestamp) {
struct timeval _now;
gettimeofday(&_now, NULL);
qemu_log("%d@%zu.%06zu:qcrypto_tls_creds_get_path " "TLS creds path creds=%p filename=%s path=%s" "\n",
qemu_get_thread_id(),
(size_t)_now.tv_sec, (size_t)_now.tv_usec
, creds, filename, path);
} else {
qemu_log("qcrypto_tls_creds_get_path " "TLS creds path creds=%p filename=%s path=%s" "\n", creds, filename, path);
}
}
}
and after:
static inline void _nocheck__trace_qcrypto_tls_creds_get_path(void * creds, const char * filename, const char * path)
{
if (trace_event_get_state(TRACE_QCRYPTO_TLS_CREDS_GET_PATH) && qemu_loglevel_mask(LOG_TRACE)) {
qemu_log("qcrypto_tls_creds_get_path " "TLS creds path creds=%p filename=%s path=%s" "\n", creds, filename, path);
}
}
The log and error messages before:
$ qemu-system-x86_64 -trace qcrypto* -object tls-creds-x509,id=tls0,dir=$HOME/tls -msg timestamp=on
2986097@1753122905.917608:qcrypto_tls_creds_x509_load TLS creds x509 load creds=0x55d925bd9490 dir=/var/home/berrange/tls
2986097@1753122905.917621:qcrypto_tls_creds_get_path TLS creds path creds=0x55d925bd9490 filename=ca-cert.pem path=<none>
2025-07-21T18:35:05.917626Z qemu-system-x86_64: Unable to access credentials /var/home/berrange/tls/ca-cert.pem: No such file or directory
and after:
$ qemu-system-x86_64 -trace qcrypto* -object tls-creds-x509,id=tls0,dir=$HOME/tls -msg timestamp=on
2025-07-21T18:43:28.089797Z qcrypto_tls_creds_x509_load TLS creds x509 load creds=0x55bf5bf12380 dir=/var/home/berrange/tls
2025-07-21T18:43:28.089815Z qcrypto_tls_creds_get_path TLS creds path creds=0x55bf5bf12380 filename=ca-cert.pem path=<none>
2025-07-21T18:43:28.089819Z qemu-system-x86_64: Unable to access credentials /var/home/berrange/tls/ca-cert.pem: No such file or directory
The binary size before:
$ ls -alh qemu-system-x86_64
-rwxr-xr-x. 1 berrange berrange 87M Jul 21 19:39 qemu-system-x86_64
$ strip qemu-system-x86_64
$ ls -alh qemu-system-x86_64
-rwxr-xr-x. 1 berrange berrange 30M Jul 21 19:39 qemu-system-x86_64
and after:
$ ls -alh qemu-system-x86_64
-rwxr-xr-x. 1 berrange berrange 85M Jul 21 19:41 qemu-system-x86_64
$ strip qemu-system-x86_64
$ ls -alh qemu-system-x86_64
-rwxr-xr-x. 1 berrange berrange 29M Jul 21 19:41 qemu-system-x86_64
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-id: 20250721185452.3016488-1-berrange@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The build fails on Windows. Replace calls to Unix programs like ´cat´,
´sed´ and ´true´ with calls to ´python´ and wrap calls to
´os.path.relpath´ in try-except because it can fail when the two paths
are on different drives. Make sure to convert the Windows paths to Unix
paths to prevent warnings in generated files.
Signed-off-by: oltolm <oleg.tolmatcev@gmail.com>
Message-id: 20250612221521.1109-2-oleg.tolmatcev@gmail.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This change improves performance by moving the hot path of the trace_vhost_commit()(or any other trace function) logic to the header file.
Previously, even when the trace event was disabled, the function call chain:-
trace_vhost_commit()(Or any other trace function) → _nocheck__trace_vhost_commit() → _simple_trace_vhost_commit()
incurred a significant function prologue overhead before checking the trace state.
Disassembly of _simple_trace_vhost_commit() (from the .c file) showed that 11 out of the first 14 instructions were prologue-related, including:
0x10 stp x29, x30, [sp, #-64]! Prologue: allocates 64-byte frame and saves old FP (x29) & LR (x30)
0x14 adrp x3, trace_events_enabled_count Prologue: computes page-base of the trace-enable counter
0x18 adrp x2, __stack_chk_guard Important (maybe prolog don't know?)(stack-protector): starts up the stack-canary load
0x1c mov x29, sp Prologue: sets new frame pointer
0x20 ldr x3, [x3] Prologue: loads the actual trace-enabled count
0x24 stp x19, x20, [sp, #16] Prologue: spills callee-saved regs used by this function (x19, x20)
0x28 and w20, w0, #0xff Tracepoint setup: extracts the low-8 bits of arg0 as the “event boolean”
0x2c ldr x2, [x2] Prologue (cont’d): completes loading of the stack-canary value
0x30 and w19, w1, #0xff Tracepoint setup: extracts low-8 bits of arg1
0x34 ldr w0, [x3] Important: loads the current trace-enabled flag from memory
0x38 ldr x1, [x2] Prologue (cont’d): reads the canary
0x3c str x1, [sp, #56] Prologue (cont’d): writes the canary into the new frame
0x40 mov x1, #0 Prologue (cont’d): zeroes out x1 for the upcoming branch test
0x44 cbnz w0, 0x88 Important: if tracing is disabled (w0==0) skip the heavy path entirely
The trace-enabled check happens after the prologue. This is wasteful when tracing is disabled, which is often the case in production.
To optimize this:
_nocheck__trace_vhost_commit() is now fully inlined in the .h file with
the hot path.It checks trace_event_get_state() before calling into _simple_trace_vhost_commit(), which remains in .c.
This avoids calling into the .c function altogether when the tracepoint is disabled, thereby skipping unnecessary prologue instructions.
This results in better performance by removing redundant instructions in the tracing fast path.
Signed-off-by: Tanish Desai <tanishdesai37@gmail.com>
Message-id: 20250528192528.3968-1-tanishdesai37@gmail.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This file, and a couple of uses, got left behind when the
tcg stuff was removed from tracetool.
Fixes: 126d4123c5 ("tracing: excise the tcg related from tracetool")
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
On Fedora 36, with lttng-ust 2.13.1, compilation fails with:
In file included from trace/trace-ust-all.h:49085,
from trace/trace-ust-all.c:13:
/usr/include/lttng/tracepoint-event.h:67:10: error: #include expects "FILENAME" or <FILENAME>
67 | #include LTTNG_UST_TRACEPOINT_INCLUDE
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
In lttng-ust commit 41858e2b6e8 ("Fix: don't do macro expansion in
tracepoint file name") from 2012, starting from lttng-ust 2.1, the API
was changed to expect TRACEPOINT_INCLUDE to be defined as a string.
In lttng-ust commit d2966b4b0b2 ("Remove TRACEPOINT_INCLUDE_FILE
macro"), in 2021, the compatibility macro was removed.
Use the "new" API from 2012, and bump the version requirement to 2.1 to
fix compilation with >= 2.13.
According to repology, all distributions we support have >= 2.1 (centos
8 has oldest with 2.8.1 afaict)
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220328084717.367993-2-marcandre.lureau@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
macro is not reset after use, so the format decoded is always the
one of the first "PRI" in the format string.
For instance:
vhost_vdpa_set_config(void *dev, uint32_t offset, uint32_t size, \
uint32_t flags) "dev: %p offset: %"PRIu32" \
size: %"PRIu32" flags: 0x%"PRIx32
generates:
printf("%d@%d vhost_vdpa_set_config dev: %p offset: %u size: %u \
flags: 0x%u\n", pid(), gettimeofday_ns(), dev, offset, \
size, flags)
for the "flags" parameter, we can see a "0x%u" rather than a "0x%x"
because the first macro was "PRIu32" (for offset).
In the loop, macro becomes "PRIu32PRIu32PRIx32", and c_macro_to_format()
returns always macro[3] ('u' in this case). This patch resets macro after
the format has been decoded.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-id: 20210105191721.120463-3-lvivier@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The compiler encounters trace event format strings in generated code.
Format strings are error-prone and therefore clear compiler errors are
important.
Use the #line directive to show the trace-events filename and line
number in format string errors:
https://gcc.gnu.org/onlinedocs/gcc-10.2.0/cpp/Line-Control.html
For example, if the cpu_in trace event's %u is changed to %p the
following error is reported:
trace-events:29:18: error: format ‘%p’ expects argument of type ‘void *’, but argument 7 has type ‘unsigned int’ [-Werror=format=]
Line 29 in trace-events is where cpu_in is defined. This works for any
trace-events file in the QEMU source tree and the correct path is
displayed.
Unfortunately there does not seem to be a way to set the column, so "18"
is not the right character on that line.
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20200827142915.108730-5-stefanha@redhat.com>
The tracetool.py script writes to stdout. This means the output filename
is not available to the script. Add the output filename to the
command-line so that the script has access to the filename.
This also simplifies the tracetool.py invocation. It's no longer
necessary to use meson's custom_build(capture : true) to save output.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20200827142915.108730-2-stefanha@redhat.com>
SystemTap's dtrace(1) prints the following warning when it encounters
long long arguments:
Warning: /usr/bin/dtrace:trace/trace-dtrace-hw_virtio.dtrace:76: syntax error near:
probe vhost_vdpa_dev_start
Warning: Proceeding as if --no-pyparsing was given.
Use the uint64_t and int64_t types, respectively. This works with all
host CPU 32- and 64-bit data models (ILP32, LP64, and LLP64) that QEMU
supports.
Reported-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20201020094043.159935-1-stefanha@redhat.com
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
dtrace on macOS complains that CPUState * is used for a few probes:
dtrace: failed to compile script trace-dtrace-root.dtrace: line 130: syntax error near "CPUState"
A comment in scripts/tracetool/__init__.py mentions that:
We only want to allow standard C types or fixed sized
integer types. We don't want QEMU specific types
as we can't assume trace backends can resolve all the
typedefs
Fixes: 3d211d9f4d ("trace: Add 'vcpu' event property to trace guest vCPU")
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Message-id: 20200717093517.73397-3-r.bolshakov@yadro.com
Cc: Cameron Esfahani <dirty@apple.com>
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
