CTR0, CTR1, CTR2, CTR3, CTR4, CTR5 -- kernel tracing facility
extern int ktr_cpumask;
extern int ktr_entries;
extern int ktr_extend;
extern int ktr_mask;
extern int ktr_verbose;
extern struct ktr_entry ktr_buf;
CTR0(u_int mask, char *format);
CTR1(u_int mask, char *format, arg1);
CTR2(u_int mask, char *format, arg1, arg2);
CTR3(u_int mask, char *format, arg1, arg2, arg3);
CTR4(u_int mask, char *format, arg1, arg2, arg3, arg4);
CTR5(u_int mask, char *format, arg1, arg2, arg3, arg4, arg5);
KTR provides a circular buffer of events that can be logged in a
printf(9) style fashion. These events can then be dumped either via
ddb(4) or gdb(1).
Events are created and logged in the kernel via the CTRx macros. The
first parameter is a mask of event types (KTR_*) defined in <sys/ktr.h>.
The event will be logged only if any of the event types specified in mask
are enabled in the global event mask stored in ktr_mask. The format
argument is a printf(9) style format string used to build the text of the
event log message. Following the format string are zero to five arguments
referenced by format. Note that the different macros differ only
in the number of arguments each one takes, as indicated by its name.
Each event is logged with a timestamp in addition to the log message.
The ktr_entries variable contains the number of entries in the ktr_buf
array. These variables are mostly useful for post-mortem crash dump
tools to locate the base of the circular trace buffer and its length.
The ktr_mask variable contains the run time mask of events to log.
The CPU event mask is stored in the ktr_cpumask variable.
The ktr_verbose variable stores the verbose flag that controls whether
events are logged to the console in addition to the event buffer.
This example demonstrates the use of tracepoints at the KTR_PROC logging
* Pick a new current process and record its start time.
CTR3(KTR_PROC, "mi_switch: old proc %p (pid %d, %s)", p, p->p_pid,
CTR3(KTR_PROC, "mi_switch: new proc %p (pid %d, %s)", p, p->p_pid,
The KTR kernel tracing facility first appeared in BSD/OS 3.0 and was
imported into FreeBSD 5.0.
Currently there is one global buffer shared among all CPUs. It might be
profitable at some point in time to use per-CPU buffers instead so that
if one CPU halts or starts spinning, then the log messages it emitted
just prior to halting or spinning will not be drowned out by events from
the other CPUs.
FreeBSD 5.2.1 February 15, 2001 FreeBSD 5.2.1 [ Back ]