2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2003-2008 Joseph Koshy
5 * Copyright (c) 2007 The FreeBSD Foundation
8 * Portions of this software were developed by A. Joseph Koshy under
9 * sponsorship from the FreeBSD Foundation and Google, Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include "opt_hwpmc_hooks.h"
38 #include <sys/param.h>
39 #include <sys/ctype.h>
40 #include <sys/domainset.h>
41 #include <sys/param.h>
42 #include <sys/malloc.h>
43 #include <sys/kernel.h>
45 #include <sys/mutex.h>
47 #include <sys/pmckern.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_kern.h>
57 FEATURE(hwpmc_hooks, "Kernel support for HW PMC");
58 #define PMC_KERNEL_VERSION PMC_VERSION
60 #define PMC_KERNEL_VERSION 0
63 MALLOC_DECLARE(M_PMCHOOKS);
64 MALLOC_DEFINE(M_PMCHOOKS, "pmchooks", "Memory space for PMC hooks");
67 MALLOC_DEFINE(M_PMC, "pmc", "Memory space for the PMC module");
69 const int pmc_kernel_version = PMC_KERNEL_VERSION;
72 int __read_mostly (*pmc_hook)(struct thread *td, int function, void *arg) = NULL;
74 /* Interrupt handler */
75 int __read_mostly (*pmc_intr)(struct trapframe *tf) = NULL;
77 DPCPU_DEFINE(uint8_t, pmc_sampled);
80 * A global count of SS mode PMCs. When non-zero, this means that
81 * we have processes that are sampling the system as a whole.
83 volatile int pmc_ss_count;
86 * Since PMC(4) may not be loaded in the current kernel, the
87 * convention followed is that a non-NULL value of 'pmc_hook' implies
88 * the presence of this kernel module.
90 * This requires us to protect 'pmc_hook' with a
91 * shared (sx) lock -- thus making the process of calling into PMC(4)
92 * somewhat more expensive than a simple 'if' check and indirect call.
95 SX_SYSINIT(pmcsx, &pmc_sx, "pmc-sx");
98 * PMC Soft per cpu trapframe.
100 struct trapframe pmc_tf[MAXCPU];
103 * Per domain list of buffer headers
105 __read_mostly struct pmc_domain_buffer_header *pmc_dom_hdrs[MAXMEMDOM];
108 * PMC Soft use a global table to store registered events.
111 SYSCTL_NODE(_kern, OID_AUTO, hwpmc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
114 static int pmc_softevents = 16;
115 SYSCTL_INT(_kern_hwpmc, OID_AUTO, softevents, CTLFLAG_RDTUN,
116 &pmc_softevents, 0, "maximum number of soft events");
119 struct pmc_soft **pmc_softs;
121 struct mtx pmc_softs_mtx;
122 MTX_SYSINIT(pmc_soft_mtx, &pmc_softs_mtx, "pmc-softs", MTX_SPIN);
129 * A note on the CPU numbering scheme used by the hwpmc(4) driver.
131 * CPUs are denoted using numbers in the range 0..[pmc_cpu_max()-1].
132 * CPUs could be numbered "sparsely" in this range; the predicate
133 * `pmc_cpu_is_present()' is used to test whether a given CPU is
134 * physically present.
136 * Further, a CPU that is physically present may be administratively
137 * disabled or otherwise unavailable for use by hwpmc(4). The
138 * `pmc_cpu_is_active()' predicate tests for CPU usability. An
139 * "active" CPU participates in thread scheduling and can field
140 * interrupts raised by PMC hardware.
142 * On systems with hyperthreaded CPUs, multiple logical CPUs may share
143 * PMC hardware resources. For such processors one logical CPU is
144 * denoted as the primary owner of the in-CPU PMC resources. The
145 * pmc_cpu_is_primary() predicate is used to distinguish this primary
146 * CPU from the others.
150 pmc_cpu_is_active(int cpu)
153 return (pmc_cpu_is_present(cpu) &&
154 !CPU_ISSET(cpu, &hlt_cpus_mask));
162 pmc_cpu_is_disabled(int cpu)
164 return (!pmc_cpu_is_active(cpu));
168 pmc_cpu_is_present(int cpu)
171 return (!CPU_ABSENT(cpu));
178 pmc_cpu_is_primary(int cpu)
181 return (!CPU_ISSET(cpu, &logical_cpus_mask));
188 * Return the maximum CPU number supported by the system. The return
189 * value is used for scaling internal data structures and for runtime
205 * Return the count of CPUs in the `active' state in the system.
208 pmc_cpu_max_active(void)
212 * When support for CPU hot-plugging is added to the kernel,
213 * this function would change to return the current number
225 * Cleanup event name:
226 * - remove duplicate '_'
230 pmc_soft_namecleanup(char *name)
236 for ( ; *p == '_' ; p++)
239 if (*p == '_' && (*(p + 1) == '_' || *(p + 1) == '\0'))
248 pmc_soft_ev_register(struct pmc_soft *ps)
250 static int warned = 0;
254 ps->ps_ev.pm_ev_code = 0; /* invalid */
255 pmc_soft_namecleanup(ps->ps_ev.pm_ev_name);
257 mtx_lock_spin(&pmc_softs_mtx);
259 if (pmc_softs_count >= pmc_softevents) {
261 * XXX Reusing events can enter a race condition where
262 * new allocated event will be used as an old one.
264 for (n = 0; n < pmc_softevents; n++)
265 if (pmc_softs[n] == NULL)
267 if (n == pmc_softevents) {
268 mtx_unlock_spin(&pmc_softs_mtx);
270 printf("hwpmc: too many soft events, "
271 "increase kern.hwpmc.softevents tunable\n");
277 ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + n;
280 ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + pmc_softs_count;
281 pmc_softs[pmc_softs_count++] = ps;
284 mtx_unlock_spin(&pmc_softs_mtx);
288 pmc_soft_ev_deregister(struct pmc_soft *ps)
291 KASSERT(ps != NULL, ("pmc_soft_deregister: called with NULL"));
293 mtx_lock_spin(&pmc_softs_mtx);
295 if (ps->ps_ev.pm_ev_code != 0 &&
296 (ps->ps_ev.pm_ev_code - PMC_EV_SOFT_FIRST) < pmc_softevents) {
297 KASSERT((int)ps->ps_ev.pm_ev_code >= PMC_EV_SOFT_FIRST &&
298 (int)ps->ps_ev.pm_ev_code <= PMC_EV_SOFT_LAST,
299 ("pmc_soft_deregister: invalid event value"));
300 pmc_softs[ps->ps_ev.pm_ev_code - PMC_EV_SOFT_FIRST] = NULL;
303 mtx_unlock_spin(&pmc_softs_mtx);
307 pmc_soft_ev_acquire(enum pmc_event ev)
311 if (ev == 0 || (ev - PMC_EV_SOFT_FIRST) >= pmc_softevents)
314 KASSERT((int)ev >= PMC_EV_SOFT_FIRST &&
315 (int)ev <= PMC_EV_SOFT_LAST,
316 ("event out of range"));
318 mtx_lock_spin(&pmc_softs_mtx);
320 ps = pmc_softs[ev - PMC_EV_SOFT_FIRST];
322 mtx_unlock_spin(&pmc_softs_mtx);
328 pmc_soft_ev_release(struct pmc_soft *ps)
331 mtx_unlock_spin(&pmc_softs_mtx);
338 init_hwpmc(void *dummy __unused)
342 if (pmc_softevents <= 0 ||
343 pmc_softevents > PMC_EV_DYN_COUNT) {
344 (void) printf("hwpmc: tunable \"softevents\"=%d out of "
345 "range.\n", pmc_softevents);
346 pmc_softevents = PMC_EV_DYN_COUNT;
348 pmc_softs = malloc(pmc_softevents * sizeof(*pmc_softs), M_PMCHOOKS,
351 for (domain = 0; domain < vm_ndomains; domain++) {
352 pmc_dom_hdrs[domain] = malloc_domainset(
353 sizeof(struct pmc_domain_buffer_header), M_PMC,
354 DOMAINSET_PREF(domain), M_WAITOK | M_ZERO);
355 mtx_init(&pmc_dom_hdrs[domain]->pdbh_mtx, "pmc_bufferlist_mtx", "pmc-leaf", MTX_SPIN);
356 TAILQ_INIT(&pmc_dom_hdrs[domain]->pdbh_head);
359 domain = pcpu_find(cpu)->pc_domain;
360 KASSERT(pmc_dom_hdrs[domain] != NULL, ("no mem allocated for domain: %d", domain));
361 pmc_dom_hdrs[domain]->pdbh_ncpus++;
366 SYSINIT(hwpmc, SI_SUB_KDTRACE, SI_ORDER_FIRST, init_hwpmc, NULL);