freebsd-skq/sys/kern/kern_pmc.c

367 lines
8.9 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2003-2008 Joseph Koshy
* Copyright (c) 2007 The FreeBSD Foundation
* All rights reserved.
*
* Portions of this software were developed by A. Joseph Koshy under
* sponsorship from the FreeBSD Foundation and Google, Inc.
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHORS 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 AUTHORS 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.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_hwpmc_hooks.h"
#include <sys/param.h>
#include <sys/ctype.h>
#include <sys/domainset.h>
#include <sys/param.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/pmc.h>
#include <sys/pmckern.h>
#include <sys/smp.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
#include <vm/vm_kern.h>
2005-12-04 02:12:43 +00:00
#ifdef HWPMC_HOOKS
FEATURE(hwpmc_hooks, "Kernel support for HW PMC");
#define PMC_KERNEL_VERSION PMC_VERSION
#else
#define PMC_KERNEL_VERSION 0
#endif
MALLOC_DECLARE(M_PMCHOOKS);
MALLOC_DEFINE(M_PMCHOOKS, "pmchooks", "Memory space for PMC hooks");
/* memory pool */
MALLOC_DEFINE(M_PMC, "pmc", "Memory space for the PMC module");
const int pmc_kernel_version = PMC_KERNEL_VERSION;
/* Hook variable. */
int __read_mostly (*pmc_hook)(struct thread *td, int function, void *arg) = NULL;
/* Interrupt handler */
int __read_mostly (*pmc_intr)(struct trapframe *tf) = NULL;
hwpmc(9): Make pmclog buffer pcpu and update constants On non-trivial SMP systems the contention on the pmc_owner mutex leads to a substantial number of samples captured being from the pmc process itself. This change a) makes buffers larger to avoid contention on the global list b) makes the working sample buffer per cpu. Run pmcstat in the background (default event rate of 64k): pmcstat -S UNHALTED_CORE_CYCLES -O /dev/null sleep 600 & Before: make -j96 buildkernel -s >&/dev/null 3336.68s user 24684.10s system 7442% cpu 6:16.50 total After: make -j96 buildkernel -s >&/dev/null 2697.82s user 1347.35s system 6058% cpu 1:06.77 total For more realistic overhead measurement set the sample rate for ~2khz on a 2.1Ghz processor: pmcstat -n 1050000 -S UNHALTED_CORE_CYCLES -O /dev/null sleep 6000 & Collecting 10 samples of `make -j96 buildkernel` from each: x before + after real time: N Min Max Median Avg Stddev x 10 76.4 127.62 84.845 88.577 15.100031 + 10 59.71 60.79 60.135 60.179 0.29957192 Difference at 95.0% confidence -28.398 +/- 10.0344 -32.0602% +/- 7.69825% (Student's t, pooled s = 10.6794) system time: N Min Max Median Avg Stddev x 10 2277.96 6948.53 2949.47 3341.492 1385.2677 + 10 1038.7 1081.06 1070.555 1064.017 15.85404 Difference at 95.0% confidence -2277.47 +/- 920.425 -68.1574% +/- 8.77623% (Student's t, pooled s = 979.596) x no pmc + pmc running real time: HEAD: N Min Max Median Avg Stddev x 10 58.38 59.15 58.86 58.847 0.22504567 + 10 76.4 127.62 84.845 88.577 15.100031 Difference at 95.0% confidence 29.73 +/- 10.0335 50.5208% +/- 17.0525% (Student's t, pooled s = 10.6785) patched: N Min Max Median Avg Stddev x 10 58.38 59.15 58.86 58.847 0.22504567 + 10 59.71 60.79 60.135 60.179 0.29957192 Difference at 95.0% confidence 1.332 +/- 0.248939 2.2635% +/- 0.426506% (Student's t, pooled s = 0.264942) system time: HEAD: N Min Max Median Avg Stddev x 10 1010.15 1073.31 1025.465 1031.524 18.135705 + 10 2277.96 6948.53 2949.47 3341.492 1385.2677 Difference at 95.0% confidence 2309.97 +/- 920.443 223.937% +/- 89.3039% (Student's t, pooled s = 979.616) patched: N Min Max Median Avg Stddev x 10 1010.15 1073.31 1025.465 1031.524 18.135705 + 10 1038.7 1081.06 1070.555 1064.017 15.85404 Difference at 95.0% confidence 32.493 +/- 16.0042 3.15% +/- 1.5794% (Student's t, pooled s = 17.0331) Reviewed by: jeff@ Approved by: sbruno@ Differential Revision: https://reviews.freebsd.org/D15155
2018-05-12 01:26:34 +00:00
DPCPU_DEFINE(uint8_t, pmc_sampled);
/*
* A global count of SS mode PMCs. When non-zero, this means that
* we have processes that are sampling the system as a whole.
*/
volatile int pmc_ss_count;
/*
* Since PMC(4) may not be loaded in the current kernel, the
* convention followed is that a non-NULL value of 'pmc_hook' implies
* the presence of this kernel module.
*
* This requires us to protect 'pmc_hook' with a
* shared (sx) lock -- thus making the process of calling into PMC(4)
* somewhat more expensive than a simple 'if' check and indirect call.
*/
struct sx pmc_sx;
SX_SYSINIT(pmcsx, &pmc_sx, "pmc-sx");
/*
* PMC Soft per cpu trapframe.
*/
struct trapframe pmc_tf[MAXCPU];
/*
* Per domain list of buffer headers
*/
__read_mostly struct pmc_domain_buffer_header *pmc_dom_hdrs[MAXMEMDOM];
/*
* PMC Soft use a global table to store registered events.
*/
SYSCTL_NODE(_kern, OID_AUTO, hwpmc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
"HWPMC parameters");
static int pmc_softevents = 16;
SYSCTL_INT(_kern_hwpmc, OID_AUTO, softevents, CTLFLAG_RDTUN,
&pmc_softevents, 0, "maximum number of soft events");
int pmc_softs_count;
struct pmc_soft **pmc_softs;
struct mtx pmc_softs_mtx;
MTX_SYSINIT(pmc_soft_mtx, &pmc_softs_mtx, "pmc-softs", MTX_SPIN);
/*
* Helper functions.
*/
/*
* A note on the CPU numbering scheme used by the hwpmc(4) driver.
*
* CPUs are denoted using numbers in the range 0..[pmc_cpu_max()-1].
* CPUs could be numbered "sparsely" in this range; the predicate
* `pmc_cpu_is_present()' is used to test whether a given CPU is
* physically present.
*
* Further, a CPU that is physically present may be administratively
* disabled or otherwise unavailable for use by hwpmc(4). The
* `pmc_cpu_is_active()' predicate tests for CPU usability. An
* "active" CPU participates in thread scheduling and can field
* interrupts raised by PMC hardware.
*
* On systems with hyperthreaded CPUs, multiple logical CPUs may share
* PMC hardware resources. For such processors one logical CPU is
* denoted as the primary owner of the in-CPU PMC resources. The
* pmc_cpu_is_primary() predicate is used to distinguish this primary
* CPU from the others.
*/
int
pmc_cpu_is_active(int cpu)
{
#ifdef SMP
return (pmc_cpu_is_present(cpu) &&
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
!CPU_ISSET(cpu, &hlt_cpus_mask));
#else
return (1);
#endif
}
/* Deprecated. */
int
pmc_cpu_is_disabled(int cpu)
{
return (!pmc_cpu_is_active(cpu));
}
int
pmc_cpu_is_present(int cpu)
{
#ifdef SMP
return (!CPU_ABSENT(cpu));
#else
return (1);
#endif
}
int
pmc_cpu_is_primary(int cpu)
{
#ifdef SMP
Commit the support for removing cpumask_t and replacing it directly with cpuset_t objects. That is going to offer the underlying support for a simple bump of MAXCPU and then support for number of cpus > 32 (as it is today). Right now, cpumask_t is an int, 32 bits on all our supported architecture. cpumask_t on the other side is implemented as an array of longs, and easilly extendible by definition. The architectures touched by this commit are the following: - amd64 - i386 - pc98 - arm - ia64 - XEN while the others are still missing. Userland is believed to be fully converted with the changes contained here. Some technical notes: - This commit may be considered an ABI nop for all the architectures different from amd64 and ia64 (and sparc64 in the future) - per-cpu members, which are now converted to cpuset_t, needs to be accessed avoiding migration, because the size of cpuset_t should be considered unknown - size of cpuset_t objects is different from kernel and userland (this is primirally done in order to leave some more space in userland to cope with KBI extensions). If you need to access kernel cpuset_t from the userland please refer to example in this patch on how to do that correctly (kgdb may be a good source, for example). - Support for other architectures is going to be added soon - Only MAXCPU for amd64 is bumped now The patch has been tested by sbruno and Nicholas Esborn on opteron 4 x 12 pack CPUs. More testing on big SMP is expected to came soon. pluknet tested the patch with his 8-ways on both amd64 and i386. Tested by: pluknet, sbruno, gianni, Nicholas Esborn Reviewed by: jeff, jhb, sbruno
2011-05-05 14:39:14 +00:00
return (!CPU_ISSET(cpu, &logical_cpus_mask));
#else
return (1);
#endif
}
/*
* Return the maximum CPU number supported by the system. The return
* value is used for scaling internal data structures and for runtime
* checks.
*/
unsigned int
pmc_cpu_max(void)
{
#ifdef SMP
return (mp_maxid+1);
#else
return (1);
#endif
}
#ifdef INVARIANTS
/*
* Return the count of CPUs in the `active' state in the system.
*/
int
pmc_cpu_max_active(void)
{
#ifdef SMP
/*
* When support for CPU hot-plugging is added to the kernel,
* this function would change to return the current number
* of "active" CPUs.
*/
return (mp_ncpus);
#else
return (1);
#endif
}
#endif
/*
* Cleanup event name:
* - remove duplicate '_'
* - all uppercase
*/
static void
pmc_soft_namecleanup(char *name)
{
char *p, *q;
p = q = name;
for ( ; *p == '_' ; p++)
;
for ( ; *p ; p++) {
if (*p == '_' && (*(p + 1) == '_' || *(p + 1) == '\0'))
continue;
else
*q++ = toupper(*p);
}
*q = '\0';
}
void
pmc_soft_ev_register(struct pmc_soft *ps)
{
static int warned = 0;
int n;
ps->ps_running = 0;
ps->ps_ev.pm_ev_code = 0; /* invalid */
pmc_soft_namecleanup(ps->ps_ev.pm_ev_name);
mtx_lock_spin(&pmc_softs_mtx);
if (pmc_softs_count >= pmc_softevents) {
/*
* XXX Reusing events can enter a race condition where
* new allocated event will be used as an old one.
*/
for (n = 0; n < pmc_softevents; n++)
if (pmc_softs[n] == NULL)
break;
if (n == pmc_softevents) {
mtx_unlock_spin(&pmc_softs_mtx);
if (!warned) {
printf("hwpmc: too many soft events, "
"increase kern.hwpmc.softevents tunable\n");
warned = 1;
}
return;
}
ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + n;
pmc_softs[n] = ps;
} else {
ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + pmc_softs_count;
pmc_softs[pmc_softs_count++] = ps;
}
mtx_unlock_spin(&pmc_softs_mtx);
}
void
pmc_soft_ev_deregister(struct pmc_soft *ps)
{
KASSERT(ps != NULL, ("pmc_soft_deregister: called with NULL"));
mtx_lock_spin(&pmc_softs_mtx);
if (ps->ps_ev.pm_ev_code != 0 &&
(ps->ps_ev.pm_ev_code - PMC_EV_SOFT_FIRST) < pmc_softevents) {
KASSERT((int)ps->ps_ev.pm_ev_code >= PMC_EV_SOFT_FIRST &&
(int)ps->ps_ev.pm_ev_code <= PMC_EV_SOFT_LAST,
("pmc_soft_deregister: invalid event value"));
pmc_softs[ps->ps_ev.pm_ev_code - PMC_EV_SOFT_FIRST] = NULL;
}
mtx_unlock_spin(&pmc_softs_mtx);
}
struct pmc_soft *
pmc_soft_ev_acquire(enum pmc_event ev)
{
struct pmc_soft *ps;
if (ev == 0 || (ev - PMC_EV_SOFT_FIRST) >= pmc_softevents)
return NULL;
KASSERT((int)ev >= PMC_EV_SOFT_FIRST &&
(int)ev <= PMC_EV_SOFT_LAST,
("event out of range"));
mtx_lock_spin(&pmc_softs_mtx);
ps = pmc_softs[ev - PMC_EV_SOFT_FIRST];
if (ps == NULL)
mtx_unlock_spin(&pmc_softs_mtx);
return ps;
}
void
pmc_soft_ev_release(struct pmc_soft *ps)
{
mtx_unlock_spin(&pmc_softs_mtx);
}
/*
* Initialise hwpmc.
*/
static void
init_hwpmc(void *dummy __unused)
{
int domain, cpu;
if (pmc_softevents <= 0 ||
pmc_softevents > PMC_EV_DYN_COUNT) {
(void) printf("hwpmc: tunable \"softevents\"=%d out of "
"range.\n", pmc_softevents);
pmc_softevents = PMC_EV_DYN_COUNT;
}
pmc_softs = malloc(pmc_softevents * sizeof(*pmc_softs), M_PMCHOOKS,
M_WAITOK | M_ZERO);
for (domain = 0; domain < vm_ndomains; domain++) {
pmc_dom_hdrs[domain] = malloc_domainset(
sizeof(struct pmc_domain_buffer_header), M_PMC,
DOMAINSET_PREF(domain), M_WAITOK | M_ZERO);
mtx_init(&pmc_dom_hdrs[domain]->pdbh_mtx, "pmc_bufferlist_mtx", "pmc-leaf", MTX_SPIN);
TAILQ_INIT(&pmc_dom_hdrs[domain]->pdbh_head);
}
CPU_FOREACH(cpu) {
domain = pcpu_find(cpu)->pc_domain;
KASSERT(pmc_dom_hdrs[domain] != NULL, ("no mem allocated for domain: %d", domain));
pmc_dom_hdrs[domain]->pdbh_ncpus++;
}
}
SYSINIT(hwpmc, SI_SUB_KDTRACE, SI_ORDER_FIRST, init_hwpmc, NULL);