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/types.h>
39 #include <sys/ctype.h>
40 #include <sys/param.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h>
44 #include <sys/mutex.h>
46 #include <sys/pmckern.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_kern.h>
56 FEATURE(hwpmc_hooks, "Kernel support for HW PMC");
57 #define PMC_KERNEL_VERSION PMC_VERSION
59 #define PMC_KERNEL_VERSION 0
62 MALLOC_DECLARE(M_PMCHOOKS);
63 MALLOC_DEFINE(M_PMCHOOKS, "pmchooks", "Memory space for PMC hooks");
66 MALLOC_DEFINE(M_PMC, "pmc", "Memory space for the PMC module");
68 const int pmc_kernel_version = PMC_KERNEL_VERSION;
71 int __read_mostly (*pmc_hook)(struct thread *td, int function, void *arg) = NULL;
73 /* Interrupt handler */
74 int __read_mostly (*pmc_intr)(struct trapframe *tf) = NULL;
76 DPCPU_DEFINE(uint8_t, pmc_sampled);
79 * A global count of SS mode PMCs. When non-zero, this means that
80 * we have processes that are sampling the system as a whole.
82 volatile int pmc_ss_count;
85 * Since PMC(4) may not be loaded in the current kernel, the
86 * convention followed is that a non-NULL value of 'pmc_hook' implies
87 * the presence of this kernel module.
89 * This requires us to protect 'pmc_hook' with a
90 * shared (sx) lock -- thus making the process of calling into PMC(4)
91 * somewhat more expensive than a simple 'if' check and indirect call.
94 SX_SYSINIT(pmcsx, &pmc_sx, "pmc-sx");
97 * PMC Soft per cpu trapframe.
99 struct trapframe pmc_tf[MAXCPU];
102 * Per domain list of buffer headers
104 __read_mostly struct pmc_domain_buffer_header *pmc_dom_hdrs[MAXMEMDOM];
107 * PMC Soft use a global table to store registered events.
110 SYSCTL_NODE(_kern, OID_AUTO, hwpmc, CTLFLAG_RW, 0, "HWPMC parameters");
112 static int pmc_softevents = 16;
113 SYSCTL_INT(_kern_hwpmc, OID_AUTO, softevents, CTLFLAG_RDTUN,
114 &pmc_softevents, 0, "maximum number of soft events");
117 struct pmc_soft **pmc_softs;
119 struct mtx pmc_softs_mtx;
120 MTX_SYSINIT(pmc_soft_mtx, &pmc_softs_mtx, "pmc-softs", MTX_SPIN);
127 * A note on the CPU numbering scheme used by the hwpmc(4) driver.
129 * CPUs are denoted using numbers in the range 0..[pmc_cpu_max()-1].
130 * CPUs could be numbered "sparsely" in this range; the predicate
131 * `pmc_cpu_is_present()' is used to test whether a given CPU is
132 * physically present.
134 * Further, a CPU that is physically present may be administratively
135 * disabled or otherwise unavailable for use by hwpmc(4). The
136 * `pmc_cpu_is_active()' predicate tests for CPU usability. An
137 * "active" CPU participates in thread scheduling and can field
138 * interrupts raised by PMC hardware.
140 * On systems with hyperthreaded CPUs, multiple logical CPUs may share
141 * PMC hardware resources. For such processors one logical CPU is
142 * denoted as the primary owner of the in-CPU PMC resources. The
143 * pmc_cpu_is_primary() predicate is used to distinguish this primary
144 * CPU from the others.
148 pmc_cpu_is_active(int cpu)
151 return (pmc_cpu_is_present(cpu) &&
152 !CPU_ISSET(cpu, &hlt_cpus_mask));
160 pmc_cpu_is_disabled(int cpu)
162 return (!pmc_cpu_is_active(cpu));
166 pmc_cpu_is_present(int cpu)
169 return (!CPU_ABSENT(cpu));
176 pmc_cpu_is_primary(int cpu)
179 return (!CPU_ISSET(cpu, &logical_cpus_mask));
187 * Return the maximum CPU number supported by the system. The return
188 * value is used for scaling internal data structures and for runtime
204 * Return the count of CPUs in the `active' state in the system.
207 pmc_cpu_max_active(void)
211 * When support for CPU hot-plugging is added to the kernel,
212 * this function would change to return the current number
224 * Cleanup event name:
225 * - remove duplicate '_'
229 pmc_soft_namecleanup(char *name)
235 for ( ; *p == '_' ; p++)
238 if (*p == '_' && (*(p + 1) == '_' || *(p + 1) == '\0'))
247 pmc_soft_ev_register(struct pmc_soft *ps)
249 static int warned = 0;
253 ps->ps_ev.pm_ev_code = 0; /* invalid */
254 pmc_soft_namecleanup(ps->ps_ev.pm_ev_name);
256 mtx_lock_spin(&pmc_softs_mtx);
258 if (pmc_softs_count >= pmc_softevents) {
260 * XXX Reusing events can enter a race condition where
261 * new allocated event will be used as an old one.
263 for (n = 0; n < pmc_softevents; n++)
264 if (pmc_softs[n] == NULL)
266 if (n == pmc_softevents) {
267 mtx_unlock_spin(&pmc_softs_mtx);
269 printf("hwpmc: too many soft events, "
270 "increase kern.hwpmc.softevents tunable\n");
276 ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + n;
279 ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + pmc_softs_count;
280 pmc_softs[pmc_softs_count++] = ps;
283 mtx_unlock_spin(&pmc_softs_mtx);
287 pmc_soft_ev_deregister(struct pmc_soft *ps)
290 KASSERT(ps != NULL, ("pmc_soft_deregister: called with NULL"));
292 mtx_lock_spin(&pmc_softs_mtx);
294 if (ps->ps_ev.pm_ev_code != 0 &&
295 (ps->ps_ev.pm_ev_code - PMC_EV_SOFT_FIRST) < pmc_softevents) {
296 KASSERT((int)ps->ps_ev.pm_ev_code >= PMC_EV_SOFT_FIRST &&
297 (int)ps->ps_ev.pm_ev_code <= PMC_EV_SOFT_LAST,
298 ("pmc_soft_deregister: invalid event value"));
299 pmc_softs[ps->ps_ev.pm_ev_code - PMC_EV_SOFT_FIRST] = NULL;
302 mtx_unlock_spin(&pmc_softs_mtx);
306 pmc_soft_ev_acquire(enum pmc_event ev)
310 if (ev == 0 || (ev - PMC_EV_SOFT_FIRST) >= pmc_softevents)
313 KASSERT((int)ev >= PMC_EV_SOFT_FIRST &&
314 (int)ev <= PMC_EV_SOFT_LAST,
315 ("event out of range"));
317 mtx_lock_spin(&pmc_softs_mtx);
319 ps = pmc_softs[ev - PMC_EV_SOFT_FIRST];
321 mtx_unlock_spin(&pmc_softs_mtx);
327 pmc_soft_ev_release(struct pmc_soft *ps)
330 mtx_unlock_spin(&pmc_softs_mtx);
334 #define NDOMAINS vm_ndomains
342 return (pc->pc_domain);
346 #define malloc_domain(size, type, domain, flags) malloc((size), (type), (flags))
347 #define getdomain(cpu) 0
353 init_hwpmc(void *dummy __unused)
357 if (pmc_softevents <= 0 ||
358 pmc_softevents > PMC_EV_DYN_COUNT) {
359 (void) printf("hwpmc: tunable \"softevents\"=%d out of "
360 "range.\n", pmc_softevents);
361 pmc_softevents = PMC_EV_DYN_COUNT;
363 pmc_softs = malloc(pmc_softevents * sizeof(struct pmc_soft *), M_PMCHOOKS, M_NOWAIT|M_ZERO);
364 KASSERT(pmc_softs != NULL, ("cannot allocate soft events table"));
366 for (domain = 0; domain < NDOMAINS; domain++) {
367 pmc_dom_hdrs[domain] = malloc_domain(sizeof(struct pmc_domain_buffer_header), M_PMC, domain,
369 mtx_init(&pmc_dom_hdrs[domain]->pdbh_mtx, "pmc_bufferlist_mtx", "pmc-leaf", MTX_SPIN);
370 TAILQ_INIT(&pmc_dom_hdrs[domain]->pdbh_head);
373 domain = getdomain(cpu);
374 KASSERT(pmc_dom_hdrs[domain] != NULL, ("no mem allocated for domain: %d", domain));
375 pmc_dom_hdrs[domain]->pdbh_ncpus++;
380 SYSINIT(hwpmc, SI_SUB_KDTRACE, SI_ORDER_FIRST, init_hwpmc, NULL);