NetBSD/sys/kern/subr_prof.c

606 lines
17 KiB
C

/* $NetBSD: subr_prof.c,v 1.50 2021/08/14 17:51:20 ryo Exp $ */
/*-
* Copyright (c) 1982, 1986, 1993
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)subr_prof.c 8.4 (Berkeley) 2/14/95
*/
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: subr_prof.c,v 1.50 2021/08/14 17:51:20 ryo Exp $");
#ifdef _KERNEL_OPT
#include "opt_gprof.h"
#include "opt_multiprocessor.h"
#endif
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/mount.h>
#include <sys/syscallargs.h>
#include <sys/sysctl.h>
#include <sys/cpu.h>
#ifdef GPROF
#include <sys/malloc.h>
#include <sys/gmon.h>
#include <sys/xcall.h>
MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer");
static int sysctl_kern_profiling(SYSCTLFN_ARGS);
#ifdef MULTIPROCESSOR
void _gmonparam_merge(struct gmonparam *, struct gmonparam *);
#endif
/*
* Froms is actually a bunch of unsigned shorts indexing tos
*/
struct gmonparam _gmonparam = { .state = GMON_PROF_OFF };
/* Actual start of the kernel text segment. */
extern char kernel_text[];
extern char etext[];
void
kmstartup(void)
{
char *cp;
struct gmonparam *p = &_gmonparam;
unsigned long size;
/*
* Round lowpc and highpc to multiples of the density we're using
* so the rest of the scaling (here and in gprof) stays in ints.
*/
p->lowpc = rounddown(((u_long)kernel_text),
HISTFRACTION * sizeof(HISTCOUNTER));
p->highpc = roundup((u_long)etext,
HISTFRACTION * sizeof(HISTCOUNTER));
p->textsize = p->highpc - p->lowpc;
printf("Profiling kernel, textsize=%ld [%lx..%lx]\n",
p->textsize, p->lowpc, p->highpc);
p->kcountsize = p->textsize / HISTFRACTION;
p->hashfraction = HASHFRACTION;
p->fromssize = p->textsize / HASHFRACTION;
p->tolimit = p->textsize * ARCDENSITY / 100;
if (p->tolimit < MINARCS)
p->tolimit = MINARCS;
else if (p->tolimit > MAXARCS)
p->tolimit = MAXARCS;
p->tossize = p->tolimit * sizeof(struct tostruct);
size = p->kcountsize + p->fromssize + p->tossize;
#ifdef MULTIPROCESSOR
CPU_INFO_ITERATOR cii;
struct cpu_info *ci;
for (CPU_INFO_FOREACH(cii, ci)) {
p = malloc(sizeof(struct gmonparam) + size, M_GPROF,
M_NOWAIT | M_ZERO);
if (p == NULL) {
printf("No memory for profiling on %s\n",
cpu_name(ci));
/* cannot profile on this cpu */
continue;
}
memcpy(p, &_gmonparam, sizeof(_gmonparam));
ci->ci_gmon = p;
/*
* To allow profiling to be controlled only by the global
* _gmonparam.state, set the default value for each CPU to
* GMON_PROF_ON. If _gmonparam.state is not ON, mcount will
* not be executed.
* This is For compatibility of the kgmon(8) kmem interface.
*/
p->state = GMON_PROF_ON;
cp = (char *)(p + 1);
p->tos = (struct tostruct *)cp;
p->kcount = (u_short *)(cp + p->tossize);
p->froms = (u_short *)(cp + p->tossize + p->kcountsize);
}
sysctl_createv(NULL, 0, NULL, NULL,
0, CTLTYPE_NODE, "percpu",
SYSCTL_DESCR("per cpu profiling information"),
NULL, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, CTL_EOL);
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
sysctl_createv(NULL, 0, NULL, NULL,
0, CTLTYPE_NODE, cpu_name(ci),
NULL,
NULL, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, cpu_index(ci), CTL_EOL);
sysctl_createv(NULL, 0, NULL, NULL,
CTLFLAG_READWRITE, CTLTYPE_INT, "state",
SYSCTL_DESCR("Profiling state"),
sysctl_kern_profiling, 0, (void *)ci, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, cpu_index(ci),
GPROF_STATE, CTL_EOL);
sysctl_createv(NULL, 0, NULL, NULL,
CTLFLAG_READWRITE, CTLTYPE_STRUCT, "count",
SYSCTL_DESCR("Array of statistical program counters"),
sysctl_kern_profiling, 0, (void *)ci, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, cpu_index(ci),
GPROF_COUNT, CTL_EOL);
sysctl_createv(NULL, 0, NULL, NULL,
CTLFLAG_READWRITE, CTLTYPE_STRUCT, "froms",
SYSCTL_DESCR("Array indexed by program counter of "
"call-from points"),
sysctl_kern_profiling, 0, (void *)ci, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, cpu_index(ci),
GPROF_FROMS, CTL_EOL);
sysctl_createv(NULL, 0, NULL, NULL,
CTLFLAG_READWRITE, CTLTYPE_STRUCT, "tos",
SYSCTL_DESCR("Array of structures describing "
"destination of calls and their counts"),
sysctl_kern_profiling, 0, (void *)ci, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, cpu_index(ci),
GPROF_TOS, CTL_EOL);
sysctl_createv(NULL, 0, NULL, NULL,
CTLFLAG_READWRITE, CTLTYPE_STRUCT, "gmonparam",
SYSCTL_DESCR("Structure giving the sizes of the above "
"arrays"),
sysctl_kern_profiling, 0, (void *)ci, 0,
CTL_KERN, KERN_PROF, GPROF_PERCPU, cpu_index(ci),
GPROF_GMONPARAM, CTL_EOL);
}
/*
* For minimal compatibility of the kgmon(8) kmem interface,
* the _gmonparam and cpu0:ci_gmon share buffers.
*/
p = curcpu()->ci_gmon;
if (p != NULL) {
_gmonparam.tos = p->tos;
_gmonparam.kcount = p->kcount;
_gmonparam.froms = p->froms;
}
#else /* MULTIPROCESSOR */
cp = malloc(size, M_GPROF, M_NOWAIT | M_ZERO);
if (cp == 0) {
printf("No memory for profiling.\n");
return;
}
p->tos = (struct tostruct *)cp;
cp += p->tossize;
p->kcount = (u_short *)cp;
cp += p->kcountsize;
p->froms = (u_short *)cp;
#endif /* MULTIPROCESSOR */
}
#ifdef MULTIPROCESSOR
static void
prof_set_state_xc(void *arg1, void *arg2 __unused)
{
int state = PTRTOUINT64(arg1);
struct gmonparam *gp = curcpu()->ci_gmon;
if (gp != NULL)
gp->state = state;
}
#endif /* MULTIPROCESSOR */
/*
* Return kernel profiling information.
*/
/*
* sysctl helper routine for kern.profiling subtree. enables/disables
* kernel profiling and gives out copies of the profiling data.
*/
static int
sysctl_kern_profiling(SYSCTLFN_ARGS)
{
struct sysctlnode node = *rnode;
struct gmonparam *gp;
int error;
#ifdef MULTIPROCESSOR
CPU_INFO_ITERATOR cii;
struct cpu_info *ci, *target_ci;
uint64_t where;
int state;
bool prof_on, do_merge;
target_ci = (struct cpu_info *)rnode->sysctl_data;
do_merge = (oldp != NULL) && (target_ci == NULL) &&
((node.sysctl_num == GPROF_COUNT) ||
(node.sysctl_num == GPROF_FROMS) ||
(node.sysctl_num == GPROF_TOS));
if (do_merge) {
/* kern.profiling.{count,froms,tos} */
unsigned long size;
char *cp;
/* allocate temporary gmonparam, and merge results of all CPU */
size = _gmonparam.kcountsize + _gmonparam.fromssize +
_gmonparam.tossize;
gp = malloc(sizeof(struct gmonparam) + size, M_GPROF,
M_NOWAIT | M_ZERO);
if (gp == NULL)
return ENOMEM;
memcpy(gp, &_gmonparam, sizeof(_gmonparam));
cp = (char *)(gp + 1);
gp->tos = (struct tostruct *)cp;
gp->kcount = (u_short *)(cp + gp->tossize);
gp->froms = (u_short *)(cp + gp->tossize + gp->kcountsize);
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
_gmonparam_merge(gp, ci->ci_gmon);
}
} else if (target_ci != NULL) {
/* kern.profiling.percpu.* */
gp = target_ci->ci_gmon;
} else {
/* kern.profiling.{state,gmonparam} */
gp = &_gmonparam;
}
#else /* MULTIPROCESSOR */
gp = &_gmonparam;
#endif
switch (node.sysctl_num) {
case GPROF_STATE:
#ifdef MULTIPROCESSOR
/*
* if _gmonparam.state is OFF, the state of each CPU is
* considered to be OFF, even if it is actually ON.
*/
if (_gmonparam.state == GMON_PROF_OFF ||
gp->state == GMON_PROF_OFF)
state = GMON_PROF_OFF;
else
state = GMON_PROF_ON;
node.sysctl_data = &state;
#else
node.sysctl_data = &gp->state;
#endif
break;
case GPROF_COUNT:
node.sysctl_data = gp->kcount;
node.sysctl_size = gp->kcountsize;
break;
case GPROF_FROMS:
node.sysctl_data = gp->froms;
node.sysctl_size = gp->fromssize;
break;
case GPROF_TOS:
node.sysctl_data = gp->tos;
node.sysctl_size = gp->tossize;
break;
case GPROF_GMONPARAM:
node.sysctl_data = gp;
node.sysctl_size = sizeof(*gp);
break;
default:
return (EOPNOTSUPP);
}
error = sysctl_lookup(SYSCTLFN_CALL(&node));
if (error || newp == NULL)
goto done;
#ifdef MULTIPROCESSOR
switch (node.sysctl_num) {
case GPROF_STATE:
if (target_ci != NULL) {
where = xc_unicast(0, prof_set_state_xc,
UINT64TOPTR(state), NULL, target_ci);
xc_wait(where);
/* if even one CPU being profiled, enable perfclock. */
prof_on = false;
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
if (ci->ci_gmon->state != GMON_PROF_OFF) {
prof_on = true;
break;
}
}
mutex_spin_enter(&proc0.p_stmutex);
if (prof_on)
startprofclock(&proc0);
else
stopprofclock(&proc0);
mutex_spin_exit(&proc0.p_stmutex);
if (prof_on) {
_gmonparam.state = GMON_PROF_ON;
} else {
_gmonparam.state = GMON_PROF_OFF;
/*
* when _gmonparam.state and all CPU gmon state
* are OFF, all CPU states should be ON so that
* the entire CPUs profiling can be controlled
* by _gmonparam.state only.
*/
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
ci->ci_gmon->state = GMON_PROF_ON;
}
}
} else {
_gmonparam.state = state;
where = xc_broadcast(0, prof_set_state_xc,
UINT64TOPTR(state), NULL);
xc_wait(where);
mutex_spin_enter(&proc0.p_stmutex);
if (state == GMON_PROF_OFF)
stopprofclock(&proc0);
else
startprofclock(&proc0);
mutex_spin_exit(&proc0.p_stmutex);
}
break;
case GPROF_COUNT:
/*
* if 'kern.profiling.{count,froms,tos}' is written, the same
* data will be written to 'kern.profiling.percpu.cpuN.xxx'
*/
if (target_ci == NULL) {
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
memmove(ci->ci_gmon->kcount, gp->kcount,
newlen);
}
}
break;
case GPROF_FROMS:
if (target_ci == NULL) {
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
memmove(ci->ci_gmon->froms, gp->froms, newlen);
}
}
break;
case GPROF_TOS:
if (target_ci == NULL) {
for (CPU_INFO_FOREACH(cii, ci)) {
if (ci->ci_gmon == NULL)
continue;
memmove(ci->ci_gmon->tos, gp->tos, newlen);
}
}
break;
}
#else
if (node.sysctl_num == GPROF_STATE) {
mutex_spin_enter(&proc0.p_stmutex);
if (gp->state == GMON_PROF_OFF)
stopprofclock(&proc0);
else
startprofclock(&proc0);
mutex_spin_exit(&proc0.p_stmutex);
}
#endif
done:
#ifdef MULTIPROCESSOR
if (do_merge)
free(gp, M_GPROF);
#endif
return error;
}
SYSCTL_SETUP(sysctl_kern_gprof_setup, "sysctl kern.profiling subtree setup")
{
sysctl_createv(clog, 0, NULL, NULL,
CTLFLAG_PERMANENT,
CTLTYPE_NODE, "profiling",
SYSCTL_DESCR("Profiling information (available)"),
NULL, 0, NULL, 0,
CTL_KERN, KERN_PROF, CTL_EOL);
sysctl_createv(clog, 0, NULL, NULL,
CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
CTLTYPE_INT, "state",
SYSCTL_DESCR("Profiling state"),
sysctl_kern_profiling, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_STATE, CTL_EOL);
sysctl_createv(clog, 0, NULL, NULL,
CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
CTLTYPE_STRUCT, "count",
SYSCTL_DESCR("Array of statistical program counters"),
sysctl_kern_profiling, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_COUNT, CTL_EOL);
sysctl_createv(clog, 0, NULL, NULL,
CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
CTLTYPE_STRUCT, "froms",
SYSCTL_DESCR("Array indexed by program counter of "
"call-from points"),
sysctl_kern_profiling, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_FROMS, CTL_EOL);
sysctl_createv(clog, 0, NULL, NULL,
CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
CTLTYPE_STRUCT, "tos",
SYSCTL_DESCR("Array of structures describing "
"destination of calls and their counts"),
sysctl_kern_profiling, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_TOS, CTL_EOL);
sysctl_createv(clog, 0, NULL, NULL,
CTLFLAG_PERMANENT,
CTLTYPE_STRUCT, "gmonparam",
SYSCTL_DESCR("Structure giving the sizes of the above "
"arrays"),
sysctl_kern_profiling, 0, NULL, 0,
CTL_KERN, KERN_PROF, GPROF_GMONPARAM, CTL_EOL);
}
#endif /* GPROF */
/*
* Profiling system call.
*
* The scale factor is a fixed point number with 16 bits of fraction, so that
* 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling.
*/
/* ARGSUSED */
int
sys_profil(struct lwp *l, const struct sys_profil_args *uap, register_t *retval)
{
/* {
syscallarg(char *) samples;
syscallarg(size_t) size;
syscallarg(u_long) offset;
syscallarg(u_int) scale;
} */
struct proc *p = l->l_proc;
struct uprof *upp;
if (SCARG(uap, scale) > (1 << 16))
return (EINVAL);
if (SCARG(uap, scale) == 0) {
mutex_spin_enter(&p->p_stmutex);
stopprofclock(p);
mutex_spin_exit(&p->p_stmutex);
return (0);
}
upp = &p->p_stats->p_prof;
/* Block profile interrupts while changing state. */
mutex_spin_enter(&p->p_stmutex);
upp->pr_off = SCARG(uap, offset);
upp->pr_scale = SCARG(uap, scale);
upp->pr_base = SCARG(uap, samples);
upp->pr_size = SCARG(uap, size);
startprofclock(p);
mutex_spin_exit(&p->p_stmutex);
return (0);
}
/*
* Scale is a fixed-point number with the binary point 16 bits
* into the value, and is <= 1.0. pc is at most 32 bits, so the
* intermediate result is at most 48 bits.
*/
#define PC_TO_INDEX(pc, prof) \
((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
(u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
/*
* Collect user-level profiling statistics; called on a profiling tick,
* when a process is running in user-mode. This routine may be called
* from an interrupt context. We schedule an AST that will vector us
* to trap() with a context in which copyin and copyout will work.
* Trap will then call addupc_task().
*
* XXX We could use ufetch/ustore here if the profile buffers were
* wired.
*
* Note that we may (rarely) not get around to the AST soon enough, and
* lose profile ticks when the next tick overwrites this one, but in this
* case the system is overloaded and the profile is probably already
* inaccurate.
*/
void
addupc_intr(struct lwp *l, u_long pc)
{
struct uprof *prof;
struct proc *p;
u_int i;
p = l->l_proc;
KASSERT(mutex_owned(&p->p_stmutex));
prof = &p->p_stats->p_prof;
if (pc < prof->pr_off ||
(i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
return; /* out of range; ignore */
mutex_spin_exit(&p->p_stmutex);
/* XXXSMP */
prof->pr_addr = pc;
prof->pr_ticks++;
cpu_need_proftick(l);
mutex_spin_enter(&p->p_stmutex);
}
/*
* Much like before, but we can afford to take faults here. If the
* update fails, we simply turn off profiling.
*/
void
addupc_task(struct lwp *l, u_long pc, u_int ticks)
{
struct uprof *prof;
struct proc *p;
void *addr;
int error;
u_int i;
u_short v;
p = l->l_proc;
if (ticks == 0)
return;
mutex_spin_enter(&p->p_stmutex);
prof = &p->p_stats->p_prof;
/* Testing P_PROFIL may be unnecessary, but is certainly safe. */
if ((p->p_stflag & PST_PROFIL) == 0 || pc < prof->pr_off ||
(i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) {
mutex_spin_exit(&p->p_stmutex);
return;
}
addr = prof->pr_base + i;
mutex_spin_exit(&p->p_stmutex);
if ((error = copyin(addr, (void *)&v, sizeof(v))) == 0) {
v += ticks;
error = copyout((void *)&v, addr, sizeof(v));
}
if (error != 0) {
mutex_spin_enter(&p->p_stmutex);
stopprofclock(p);
mutex_spin_exit(&p->p_stmutex);
}
}