1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2003-2008 Joseph Koshy
5 * Copyright (c) 2007 The FreeBSD Foundation
6 * All rights reserved.
7 *
8 * Portions of this software were developed by A. Joseph Koshy under
9 * sponsorship from the FreeBSD Foundation and Google, Inc.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 */
32
33#include <sys/cdefs.h>
34__FBSDID("$FreeBSD$");
35
36#include "opt_hwpmc_hooks.h"
37
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>
44#include <sys/lock.h>
45#include <sys/mutex.h>
46#include <sys/pmc.h>
47#include <sys/pmckern.h>
48#include <sys/smp.h>
49#include <sys/sysctl.h>
50#include <sys/systm.h>
51
52#include <vm/vm.h>
53#include <vm/vm_extern.h>
54#include <vm/vm_kern.h>
55
56#ifdef	HWPMC_HOOKS
57FEATURE(hwpmc_hooks, "Kernel support for HW PMC");
58#define	PMC_KERNEL_VERSION	PMC_VERSION
59#else
60#define	PMC_KERNEL_VERSION	0
61#endif
62
63MALLOC_DECLARE(M_PMCHOOKS);
64MALLOC_DEFINE(M_PMCHOOKS, "pmchooks", "Memory space for PMC hooks");
65
66/* memory pool */
67MALLOC_DEFINE(M_PMC, "pmc", "Memory space for the PMC module");
68
69const int pmc_kernel_version = PMC_KERNEL_VERSION;
70
71/* Hook variable. */
72int __read_mostly (*pmc_hook)(struct thread *td, int function, void *arg) = NULL;
73
74/* Interrupt handler */
75int __read_mostly (*pmc_intr)(struct trapframe *tf) = NULL;
76
77DPCPU_DEFINE(uint8_t, pmc_sampled);
78
79/*
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.
82 */
83volatile int pmc_ss_count;
84
85/*
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.
89 *
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.
93 */
94struct sx pmc_sx;
95SX_SYSINIT(pmcsx, &pmc_sx, "pmc-sx");
96
97/*
98 * PMC Soft per cpu trapframe.
99 */
100struct trapframe pmc_tf[MAXCPU];
101
102/*
103 * Per domain list of buffer headers
104 */
105__read_mostly struct pmc_domain_buffer_header *pmc_dom_hdrs[MAXMEMDOM];
106
107/*
108 * PMC Soft use a global table to store registered events.
109 */
110
111SYSCTL_NODE(_kern, OID_AUTO, hwpmc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
112    "HWPMC parameters");
113
114static int pmc_softevents = 16;
115SYSCTL_INT(_kern_hwpmc, OID_AUTO, softevents, CTLFLAG_RDTUN,
116    &pmc_softevents, 0, "maximum number of soft events");
117
118int pmc_softs_count;
119struct pmc_soft **pmc_softs;
120
121struct mtx pmc_softs_mtx;
122MTX_SYSINIT(pmc_soft_mtx, &pmc_softs_mtx, "pmc-softs", MTX_SPIN);
123
124/*
125 * Helper functions.
126 */
127
128/*
129 * A note on the CPU numbering scheme used by the hwpmc(4) driver.
130 *
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.
135 *
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.
141 *
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.
147 */
148
149int
150pmc_cpu_is_active(int cpu)
151{
152#ifdef	SMP
153	return (pmc_cpu_is_present(cpu) &&
154	    !CPU_ISSET(cpu, &hlt_cpus_mask));
155#else
156	return (1);
157#endif
158}
159
160/* Deprecated. */
161int
162pmc_cpu_is_disabled(int cpu)
163{
164	return (!pmc_cpu_is_active(cpu));
165}
166
167int
168pmc_cpu_is_present(int cpu)
169{
170#ifdef	SMP
171	return (!CPU_ABSENT(cpu));
172#else
173	return (1);
174#endif
175}
176
177int
178pmc_cpu_is_primary(int cpu)
179{
180#ifdef	SMP
181	return (!CPU_ISSET(cpu, &logical_cpus_mask));
182#else
183	return (1);
184#endif
185}
186
187/*
188 * Return the maximum CPU number supported by the system.  The return
189 * value is used for scaling internal data structures and for runtime
190 * checks.
191 */
192unsigned int
193pmc_cpu_max(void)
194{
195#ifdef	SMP
196	return (mp_maxid+1);
197#else
198	return (1);
199#endif
200}
201
202#ifdef	INVARIANTS
203
204/*
205 * Return the count of CPUs in the `active' state in the system.
206 */
207int
208pmc_cpu_max_active(void)
209{
210#ifdef	SMP
211	/*
212	 * When support for CPU hot-plugging is added to the kernel,
213	 * this function would change to return the current number
214	 * of "active" CPUs.
215	 */
216	return (mp_ncpus);
217#else
218	return (1);
219#endif
220}
221
222#endif
223
224/*
225 * Cleanup event name:
226 * - remove duplicate '_'
227 * - all uppercase
228 */
229static void
230pmc_soft_namecleanup(char *name)
231{
232	char *p, *q;
233
234	p = q = name;
235
236	for ( ; *p == '_' ; p++)
237		;
238	for ( ; *p ; p++) {
239		if (*p == '_' && (*(p + 1) == '_' || *(p + 1) == '\0'))
240			continue;
241		else
242			*q++ = toupper(*p);
243	}
244	*q = '\0';
245}
246
247void
248pmc_soft_ev_register(struct pmc_soft *ps)
249{
250	static int warned = 0;
251	int n;
252
253	ps->ps_running  = 0;
254	ps->ps_ev.pm_ev_code = 0; /* invalid */
255	pmc_soft_namecleanup(ps->ps_ev.pm_ev_name);
256
257	mtx_lock_spin(&pmc_softs_mtx);
258
259	if (pmc_softs_count >= pmc_softevents) {
260		/*
261		 * XXX Reusing events can enter a race condition where
262		 * new allocated event will be used as an old one.
263		 */
264		for (n = 0; n < pmc_softevents; n++)
265			if (pmc_softs[n] == NULL)
266				break;
267		if (n == pmc_softevents) {
268			mtx_unlock_spin(&pmc_softs_mtx);
269			if (!warned) {
270				printf("hwpmc: too many soft events, "
271				    "increase kern.hwpmc.softevents tunable\n");
272				warned = 1;
273			}
274			return;
275		}
276
277		ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + n;
278		pmc_softs[n] = ps;
279	} else {
280		ps->ps_ev.pm_ev_code = PMC_EV_SOFT_FIRST + pmc_softs_count;
281		pmc_softs[pmc_softs_count++] = ps;
282	}
283
284	mtx_unlock_spin(&pmc_softs_mtx);
285}
286
287void
288pmc_soft_ev_deregister(struct pmc_soft *ps)
289{
290
291	KASSERT(ps != NULL, ("pmc_soft_deregister: called with NULL"));
292
293	mtx_lock_spin(&pmc_softs_mtx);
294
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;
301	}
302
303	mtx_unlock_spin(&pmc_softs_mtx);
304}
305
306struct pmc_soft *
307pmc_soft_ev_acquire(enum pmc_event ev)
308{
309	struct pmc_soft *ps;
310
311	if (ev == 0 || (ev - PMC_EV_SOFT_FIRST) >= pmc_softevents)
312		return NULL;
313
314	KASSERT((int)ev >= PMC_EV_SOFT_FIRST &&
315	    (int)ev <= PMC_EV_SOFT_LAST,
316	    ("event out of range"));
317
318	mtx_lock_spin(&pmc_softs_mtx);
319
320	ps = pmc_softs[ev - PMC_EV_SOFT_FIRST];
321	if (ps == NULL)
322		mtx_unlock_spin(&pmc_softs_mtx);
323
324	return ps;
325}
326
327void
328pmc_soft_ev_release(struct pmc_soft *ps)
329{
330
331	mtx_unlock_spin(&pmc_softs_mtx);
332}
333
334/*
335 *  Initialise hwpmc.
336 */
337static void
338init_hwpmc(void *dummy __unused)
339{
340	int domain, cpu;
341
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;
347	}
348	pmc_softs = malloc(pmc_softevents * sizeof(*pmc_softs), M_PMCHOOKS,
349	    M_WAITOK | M_ZERO);
350
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);
357	}
358	CPU_FOREACH(cpu) {
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++;
362	}
363
364}
365
366SYSINIT(hwpmc, SI_SUB_KDTRACE, SI_ORDER_FIRST, init_hwpmc, NULL);
367