xref: /freebsd/sys/x86/x86/tsc.c (revision 16808549)
1dd7d207dSJung-uk Kim /*-
2dd7d207dSJung-uk Kim  * Copyright (c) 1998-2003 Poul-Henning Kamp
3dd7d207dSJung-uk Kim  * All rights reserved.
4dd7d207dSJung-uk Kim  *
5dd7d207dSJung-uk Kim  * Redistribution and use in source and binary forms, with or without
6dd7d207dSJung-uk Kim  * modification, are permitted provided that the following conditions
7dd7d207dSJung-uk Kim  * are met:
8dd7d207dSJung-uk Kim  * 1. Redistributions of source code must retain the above copyright
9dd7d207dSJung-uk Kim  *    notice, this list of conditions and the following disclaimer.
10dd7d207dSJung-uk Kim  * 2. Redistributions in binary form must reproduce the above copyright
11dd7d207dSJung-uk Kim  *    notice, this list of conditions and the following disclaimer in the
12dd7d207dSJung-uk Kim  *    documentation and/or other materials provided with the distribution.
13dd7d207dSJung-uk Kim  *
14dd7d207dSJung-uk Kim  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15dd7d207dSJung-uk Kim  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16dd7d207dSJung-uk Kim  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17dd7d207dSJung-uk Kim  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18dd7d207dSJung-uk Kim  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19dd7d207dSJung-uk Kim  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20dd7d207dSJung-uk Kim  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21dd7d207dSJung-uk Kim  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22dd7d207dSJung-uk Kim  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23dd7d207dSJung-uk Kim  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24dd7d207dSJung-uk Kim  * SUCH DAMAGE.
25dd7d207dSJung-uk Kim  */
26dd7d207dSJung-uk Kim 
27dd7d207dSJung-uk Kim #include <sys/cdefs.h>
28dd7d207dSJung-uk Kim __FBSDID("$FreeBSD$");
29dd7d207dSJung-uk Kim 
30aea81038SKonstantin Belousov #include "opt_compat.h"
31dd7d207dSJung-uk Kim #include "opt_clock.h"
32dd7d207dSJung-uk Kim 
33dd7d207dSJung-uk Kim #include <sys/param.h>
34dd7d207dSJung-uk Kim #include <sys/bus.h>
35dd7d207dSJung-uk Kim #include <sys/cpu.h>
365da5812bSJung-uk Kim #include <sys/limits.h>
37dd7d207dSJung-uk Kim #include <sys/malloc.h>
38dd7d207dSJung-uk Kim #include <sys/systm.h>
39dd7d207dSJung-uk Kim #include <sys/sysctl.h>
40dd7d207dSJung-uk Kim #include <sys/time.h>
41dd7d207dSJung-uk Kim #include <sys/timetc.h>
42dd7d207dSJung-uk Kim #include <sys/kernel.h>
43dd7d207dSJung-uk Kim #include <sys/power.h>
44dd7d207dSJung-uk Kim #include <sys/smp.h>
45aea81038SKonstantin Belousov #include <sys/vdso.h>
46dd7d207dSJung-uk Kim #include <machine/clock.h>
47dd7d207dSJung-uk Kim #include <machine/cputypes.h>
48dd7d207dSJung-uk Kim #include <machine/md_var.h>
49dd7d207dSJung-uk Kim #include <machine/specialreg.h>
5001e1933dSJohn Baldwin #include <x86/vmware.h>
5116808549SKonstantin Belousov #include <dev/acpica/acpi_hpet.h>
52dd7d207dSJung-uk Kim 
53dd7d207dSJung-uk Kim #include "cpufreq_if.h"
54dd7d207dSJung-uk Kim 
55dd7d207dSJung-uk Kim uint64_t	tsc_freq;
56dd7d207dSJung-uk Kim int		tsc_is_invariant;
57155094d7SJung-uk Kim int		tsc_perf_stat;
58155094d7SJung-uk Kim 
59dd7d207dSJung-uk Kim static eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag;
60dd7d207dSJung-uk Kim 
61dd7d207dSJung-uk Kim SYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN,
62dd7d207dSJung-uk Kim     &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant");
63dd7d207dSJung-uk Kim 
64dd7d207dSJung-uk Kim #ifdef SMP
651472b87fSNeel Natu int	smp_tsc;
66dd7d207dSJung-uk Kim SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0,
67dd7d207dSJung-uk Kim     "Indicates whether the TSC is safe to use in SMP mode");
68b2c63698SAlexander Motin 
69b2c63698SAlexander Motin int	smp_tsc_adjust = 0;
70b2c63698SAlexander Motin SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc_adjust, CTLFLAG_RDTUN,
71b2c63698SAlexander Motin     &smp_tsc_adjust, 0, "Try to adjust TSC on APs to match BSP");
72dd7d207dSJung-uk Kim #endif
73dd7d207dSJung-uk Kim 
74e7f1427dSKonstantin Belousov static int	tsc_shift = 1;
75e7f1427dSKonstantin Belousov SYSCTL_INT(_kern_timecounter, OID_AUTO, tsc_shift, CTLFLAG_RDTUN,
76e7f1427dSKonstantin Belousov     &tsc_shift, 0, "Shift to pre-apply for the maximum TSC frequency");
77e7f1427dSKonstantin Belousov 
7879422085SJung-uk Kim static int	tsc_disabled;
7979422085SJung-uk Kim SYSCTL_INT(_machdep, OID_AUTO, disable_tsc, CTLFLAG_RDTUN, &tsc_disabled, 0,
8079422085SJung-uk Kim     "Disable x86 Time Stamp Counter");
8179422085SJung-uk Kim 
82a4e4127fSJung-uk Kim static int	tsc_skip_calibration;
83a4e4127fSJung-uk Kim SYSCTL_INT(_machdep, OID_AUTO, disable_tsc_calibration, CTLFLAG_RDTUN,
84a4e4127fSJung-uk Kim     &tsc_skip_calibration, 0, "Disable TSC frequency calibration");
85a4e4127fSJung-uk Kim 
86dd7d207dSJung-uk Kim static void tsc_freq_changed(void *arg, const struct cf_level *level,
87dd7d207dSJung-uk Kim     int status);
88dd7d207dSJung-uk Kim static void tsc_freq_changing(void *arg, const struct cf_level *level,
89dd7d207dSJung-uk Kim     int *status);
90dd7d207dSJung-uk Kim static unsigned tsc_get_timecount(struct timecounter *tc);
91814124c3SKonstantin Belousov static inline unsigned tsc_get_timecount_low(struct timecounter *tc);
92814124c3SKonstantin Belousov static unsigned tsc_get_timecount_lfence(struct timecounter *tc);
93814124c3SKonstantin Belousov static unsigned tsc_get_timecount_low_lfence(struct timecounter *tc);
94814124c3SKonstantin Belousov static unsigned tsc_get_timecount_mfence(struct timecounter *tc);
95814124c3SKonstantin Belousov static unsigned tsc_get_timecount_low_mfence(struct timecounter *tc);
96dd7d207dSJung-uk Kim static void tsc_levels_changed(void *arg, int unit);
9716808549SKonstantin Belousov static uint32_t x86_tsc_vdso_timehands(struct vdso_timehands *vdso_th,
9816808549SKonstantin Belousov     struct timecounter *tc);
9916808549SKonstantin Belousov #ifdef COMPAT_FREEBSD32
10016808549SKonstantin Belousov static uint32_t x86_tsc_vdso_timehands32(struct vdso_timehands32 *vdso_th32,
10116808549SKonstantin Belousov     struct timecounter *tc);
10216808549SKonstantin Belousov #endif
103dd7d207dSJung-uk Kim 
104dd7d207dSJung-uk Kim static struct timecounter tsc_timecounter = {
10516808549SKonstantin Belousov 	.tc_get_timecount =		tsc_get_timecount,
10616808549SKonstantin Belousov 	.tc_counter_mask =		~0u,
10716808549SKonstantin Belousov 	.tc_name =			"TSC",
10816808549SKonstantin Belousov 	.tc_quality =			800,	/* adjusted in code */
10916808549SKonstantin Belousov 	.tc_fill_vdso_timehands = 	x86_tsc_vdso_timehands,
11016808549SKonstantin Belousov #ifdef COMPAT_FREEBSD32
11116808549SKonstantin Belousov 	.tc_fill_vdso_timehands32 = 	x86_tsc_vdso_timehands32,
11216808549SKonstantin Belousov #endif
113dd7d207dSJung-uk Kim };
114dd7d207dSJung-uk Kim 
11501e1933dSJohn Baldwin static void
1165da5812bSJung-uk Kim tsc_freq_vmware(void)
1175da5812bSJung-uk Kim {
1185da5812bSJung-uk Kim 	u_int regs[4];
1195da5812bSJung-uk Kim 
1205da5812bSJung-uk Kim 	if (hv_high >= 0x40000010) {
1215da5812bSJung-uk Kim 		do_cpuid(0x40000010, regs);
1225da5812bSJung-uk Kim 		tsc_freq = regs[0] * 1000;
1235da5812bSJung-uk Kim 	} else {
1245da5812bSJung-uk Kim 		vmware_hvcall(VMW_HVCMD_GETHZ, regs);
1255da5812bSJung-uk Kim 		if (regs[1] != UINT_MAX)
1265da5812bSJung-uk Kim 			tsc_freq = regs[0] | ((uint64_t)regs[1] << 32);
1275da5812bSJung-uk Kim 	}
1285da5812bSJung-uk Kim 	tsc_is_invariant = 1;
1295da5812bSJung-uk Kim }
1305da5812bSJung-uk Kim 
131a4e4127fSJung-uk Kim static void
132a4e4127fSJung-uk Kim tsc_freq_intel(void)
133dd7d207dSJung-uk Kim {
134a4e4127fSJung-uk Kim 	char brand[48];
135a4e4127fSJung-uk Kim 	u_int regs[4];
136a4e4127fSJung-uk Kim 	uint64_t freq;
137a4e4127fSJung-uk Kim 	char *p;
138a4e4127fSJung-uk Kim 	u_int i;
139dd7d207dSJung-uk Kim 
140a4e4127fSJung-uk Kim 	/*
141a4e4127fSJung-uk Kim 	 * Intel Processor Identification and the CPUID Instruction
142a4e4127fSJung-uk Kim 	 * Application Note 485.
143a4e4127fSJung-uk Kim 	 * http://www.intel.com/assets/pdf/appnote/241618.pdf
144a4e4127fSJung-uk Kim 	 */
145a4e4127fSJung-uk Kim 	if (cpu_exthigh >= 0x80000004) {
146a4e4127fSJung-uk Kim 		p = brand;
147a4e4127fSJung-uk Kim 		for (i = 0x80000002; i < 0x80000005; i++) {
148a4e4127fSJung-uk Kim 			do_cpuid(i, regs);
149a4e4127fSJung-uk Kim 			memcpy(p, regs, sizeof(regs));
150a4e4127fSJung-uk Kim 			p += sizeof(regs);
151a4e4127fSJung-uk Kim 		}
152a4e4127fSJung-uk Kim 		p = NULL;
153a4e4127fSJung-uk Kim 		for (i = 0; i < sizeof(brand) - 1; i++)
154a4e4127fSJung-uk Kim 			if (brand[i] == 'H' && brand[i + 1] == 'z')
155a4e4127fSJung-uk Kim 				p = brand + i;
156a4e4127fSJung-uk Kim 		if (p != NULL) {
157a4e4127fSJung-uk Kim 			p -= 5;
158a4e4127fSJung-uk Kim 			switch (p[4]) {
159a4e4127fSJung-uk Kim 			case 'M':
160a4e4127fSJung-uk Kim 				i = 1;
161a4e4127fSJung-uk Kim 				break;
162a4e4127fSJung-uk Kim 			case 'G':
163a4e4127fSJung-uk Kim 				i = 1000;
164a4e4127fSJung-uk Kim 				break;
165a4e4127fSJung-uk Kim 			case 'T':
166a4e4127fSJung-uk Kim 				i = 1000000;
167a4e4127fSJung-uk Kim 				break;
168a4e4127fSJung-uk Kim 			default:
169dd7d207dSJung-uk Kim 				return;
170a4e4127fSJung-uk Kim 			}
171a4e4127fSJung-uk Kim #define	C2D(c)	((c) - '0')
172a4e4127fSJung-uk Kim 			if (p[1] == '.') {
173a4e4127fSJung-uk Kim 				freq = C2D(p[0]) * 1000;
174a4e4127fSJung-uk Kim 				freq += C2D(p[2]) * 100;
175a4e4127fSJung-uk Kim 				freq += C2D(p[3]) * 10;
176a4e4127fSJung-uk Kim 				freq *= i * 1000;
177a4e4127fSJung-uk Kim 			} else {
178a4e4127fSJung-uk Kim 				freq = C2D(p[0]) * 1000;
179a4e4127fSJung-uk Kim 				freq += C2D(p[1]) * 100;
180a4e4127fSJung-uk Kim 				freq += C2D(p[2]) * 10;
181a4e4127fSJung-uk Kim 				freq += C2D(p[3]);
182a4e4127fSJung-uk Kim 				freq *= i * 1000000;
183a4e4127fSJung-uk Kim 			}
184a4e4127fSJung-uk Kim #undef C2D
185a4e4127fSJung-uk Kim 			tsc_freq = freq;
186a4e4127fSJung-uk Kim 		}
187a4e4127fSJung-uk Kim 	}
188a4e4127fSJung-uk Kim }
189dd7d207dSJung-uk Kim 
190a4e4127fSJung-uk Kim static void
191a4e4127fSJung-uk Kim probe_tsc_freq(void)
192a4e4127fSJung-uk Kim {
193155094d7SJung-uk Kim 	u_int regs[4];
194a4e4127fSJung-uk Kim 	uint64_t tsc1, tsc2;
195dd7d207dSJung-uk Kim 
1965da5812bSJung-uk Kim 	if (cpu_high >= 6) {
1975da5812bSJung-uk Kim 		do_cpuid(6, regs);
1985da5812bSJung-uk Kim 		if ((regs[2] & CPUID_PERF_STAT) != 0) {
1995da5812bSJung-uk Kim 			/*
2005da5812bSJung-uk Kim 			 * XXX Some emulators expose host CPUID without actual
2015da5812bSJung-uk Kim 			 * support for these MSRs.  We must test whether they
2025da5812bSJung-uk Kim 			 * really work.
2035da5812bSJung-uk Kim 			 */
2045da5812bSJung-uk Kim 			wrmsr(MSR_MPERF, 0);
2055da5812bSJung-uk Kim 			wrmsr(MSR_APERF, 0);
2065da5812bSJung-uk Kim 			DELAY(10);
2075da5812bSJung-uk Kim 			if (rdmsr(MSR_MPERF) > 0 && rdmsr(MSR_APERF) > 0)
2085da5812bSJung-uk Kim 				tsc_perf_stat = 1;
2095da5812bSJung-uk Kim 		}
2105da5812bSJung-uk Kim 	}
2115da5812bSJung-uk Kim 
21201e1933dSJohn Baldwin 	if (vm_guest == VM_GUEST_VMWARE) {
21301e1933dSJohn Baldwin 		tsc_freq_vmware();
2145da5812bSJung-uk Kim 		return;
21501e1933dSJohn Baldwin 	}
2165da5812bSJung-uk Kim 
217dd7d207dSJung-uk Kim 	switch (cpu_vendor_id) {
218dd7d207dSJung-uk Kim 	case CPU_VENDOR_AMD:
219a106a27cSJung-uk Kim 		if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 ||
220a106a27cSJung-uk Kim 		    (vm_guest == VM_GUEST_NO &&
221a106a27cSJung-uk Kim 		    CPUID_TO_FAMILY(cpu_id) >= 0x10))
222dd7d207dSJung-uk Kim 			tsc_is_invariant = 1;
223814124c3SKonstantin Belousov 		if (cpu_feature & CPUID_SSE2) {
224814124c3SKonstantin Belousov 			tsc_timecounter.tc_get_timecount =
225814124c3SKonstantin Belousov 			    tsc_get_timecount_mfence;
226814124c3SKonstantin Belousov 		}
227dd7d207dSJung-uk Kim 		break;
228dd7d207dSJung-uk Kim 	case CPU_VENDOR_INTEL:
229a106a27cSJung-uk Kim 		if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 ||
230a106a27cSJung-uk Kim 		    (vm_guest == VM_GUEST_NO &&
231a106a27cSJung-uk Kim 		    ((CPUID_TO_FAMILY(cpu_id) == 0x6 &&
232dd7d207dSJung-uk Kim 		    CPUID_TO_MODEL(cpu_id) >= 0xe) ||
233dd7d207dSJung-uk Kim 		    (CPUID_TO_FAMILY(cpu_id) == 0xf &&
234a106a27cSJung-uk Kim 		    CPUID_TO_MODEL(cpu_id) >= 0x3))))
235dd7d207dSJung-uk Kim 			tsc_is_invariant = 1;
236814124c3SKonstantin Belousov 		if (cpu_feature & CPUID_SSE2) {
237814124c3SKonstantin Belousov 			tsc_timecounter.tc_get_timecount =
238814124c3SKonstantin Belousov 			    tsc_get_timecount_lfence;
239814124c3SKonstantin Belousov 		}
240dd7d207dSJung-uk Kim 		break;
241dd7d207dSJung-uk Kim 	case CPU_VENDOR_CENTAUR:
242a106a27cSJung-uk Kim 		if (vm_guest == VM_GUEST_NO &&
243a106a27cSJung-uk Kim 		    CPUID_TO_FAMILY(cpu_id) == 0x6 &&
244dd7d207dSJung-uk Kim 		    CPUID_TO_MODEL(cpu_id) >= 0xf &&
245dd7d207dSJung-uk Kim 		    (rdmsr(0x1203) & 0x100000000ULL) == 0)
246dd7d207dSJung-uk Kim 			tsc_is_invariant = 1;
247814124c3SKonstantin Belousov 		if (cpu_feature & CPUID_SSE2) {
248814124c3SKonstantin Belousov 			tsc_timecounter.tc_get_timecount =
249814124c3SKonstantin Belousov 			    tsc_get_timecount_lfence;
250814124c3SKonstantin Belousov 		}
251dd7d207dSJung-uk Kim 		break;
252dd7d207dSJung-uk Kim 	}
253dd7d207dSJung-uk Kim 
254a4e4127fSJung-uk Kim 	if (tsc_skip_calibration) {
255a4e4127fSJung-uk Kim 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
256a4e4127fSJung-uk Kim 			tsc_freq_intel();
257a4e4127fSJung-uk Kim 		return;
258a4e4127fSJung-uk Kim 	}
259a4e4127fSJung-uk Kim 
260a4e4127fSJung-uk Kim 	if (bootverbose)
261a4e4127fSJung-uk Kim 	        printf("Calibrating TSC clock ... ");
262a4e4127fSJung-uk Kim 	tsc1 = rdtsc();
263a4e4127fSJung-uk Kim 	DELAY(1000000);
264a4e4127fSJung-uk Kim 	tsc2 = rdtsc();
265a4e4127fSJung-uk Kim 	tsc_freq = tsc2 - tsc1;
266a4e4127fSJung-uk Kim 	if (bootverbose)
267a4e4127fSJung-uk Kim 		printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq);
268a4e4127fSJung-uk Kim }
269a4e4127fSJung-uk Kim 
270a4e4127fSJung-uk Kim void
271a4e4127fSJung-uk Kim init_TSC(void)
272a4e4127fSJung-uk Kim {
273a4e4127fSJung-uk Kim 
274a4e4127fSJung-uk Kim 	if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled)
275a4e4127fSJung-uk Kim 		return;
276a4e4127fSJung-uk Kim 
277fe760cfaSJohn Baldwin #ifdef __i386__
278fe760cfaSJohn Baldwin 	/* The TSC is known to be broken on certain CPUs. */
279fe760cfaSJohn Baldwin 	switch (cpu_vendor_id) {
280fe760cfaSJohn Baldwin 	case CPU_VENDOR_AMD:
281fe760cfaSJohn Baldwin 		switch (cpu_id & 0xFF0) {
282fe760cfaSJohn Baldwin 		case 0x500:
283fe760cfaSJohn Baldwin 			/* K5 Model 0 */
284fe760cfaSJohn Baldwin 			return;
285fe760cfaSJohn Baldwin 		}
286fe760cfaSJohn Baldwin 		break;
287fe760cfaSJohn Baldwin 	case CPU_VENDOR_CENTAUR:
288fe760cfaSJohn Baldwin 		switch (cpu_id & 0xff0) {
289fe760cfaSJohn Baldwin 		case 0x540:
290fe760cfaSJohn Baldwin 			/*
291fe760cfaSJohn Baldwin 			 * http://www.centtech.com/c6_data_sheet.pdf
292fe760cfaSJohn Baldwin 			 *
293fe760cfaSJohn Baldwin 			 * I-12 RDTSC may return incoherent values in EDX:EAX
294fe760cfaSJohn Baldwin 			 * I-13 RDTSC hangs when certain event counters are used
295fe760cfaSJohn Baldwin 			 */
296fe760cfaSJohn Baldwin 			return;
297fe760cfaSJohn Baldwin 		}
298fe760cfaSJohn Baldwin 		break;
299fe760cfaSJohn Baldwin 	case CPU_VENDOR_NSC:
300fe760cfaSJohn Baldwin 		switch (cpu_id & 0xff0) {
301fe760cfaSJohn Baldwin 		case 0x540:
302fe760cfaSJohn Baldwin 			if ((cpu_id & CPUID_STEPPING) == 0)
303fe760cfaSJohn Baldwin 				return;
304fe760cfaSJohn Baldwin 			break;
305fe760cfaSJohn Baldwin 		}
306fe760cfaSJohn Baldwin 		break;
307fe760cfaSJohn Baldwin 	}
308fe760cfaSJohn Baldwin #endif
309fe760cfaSJohn Baldwin 
310a4e4127fSJung-uk Kim 	probe_tsc_freq();
311a4e4127fSJung-uk Kim 
312dd7d207dSJung-uk Kim 	/*
313dd7d207dSJung-uk Kim 	 * Inform CPU accounting about our boot-time clock rate.  This will
314dd7d207dSJung-uk Kim 	 * be updated if someone loads a cpufreq driver after boot that
315dd7d207dSJung-uk Kim 	 * discovers a new max frequency.
316dd7d207dSJung-uk Kim 	 */
317a4e4127fSJung-uk Kim 	if (tsc_freq != 0)
3185ac44f72SJung-uk Kim 		set_cputicker(rdtsc, tsc_freq, !tsc_is_invariant);
319dd7d207dSJung-uk Kim 
320dd7d207dSJung-uk Kim 	if (tsc_is_invariant)
321dd7d207dSJung-uk Kim 		return;
322dd7d207dSJung-uk Kim 
323dd7d207dSJung-uk Kim 	/* Register to find out about changes in CPU frequency. */
324dd7d207dSJung-uk Kim 	tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change,
325dd7d207dSJung-uk Kim 	    tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST);
326dd7d207dSJung-uk Kim 	tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change,
327dd7d207dSJung-uk Kim 	    tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST);
328dd7d207dSJung-uk Kim 	tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed,
329dd7d207dSJung-uk Kim 	    tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY);
330dd7d207dSJung-uk Kim }
331dd7d207dSJung-uk Kim 
33265e7d70bSJung-uk Kim #ifdef SMP
33365e7d70bSJung-uk Kim 
334814124c3SKonstantin Belousov /*
335814124c3SKonstantin Belousov  * RDTSC is not a serializing instruction, and does not drain
336814124c3SKonstantin Belousov  * instruction stream, so we need to drain the stream before executing
337814124c3SKonstantin Belousov  * it.  It could be fixed by use of RDTSCP, except the instruction is
338814124c3SKonstantin Belousov  * not available everywhere.
339814124c3SKonstantin Belousov  *
340814124c3SKonstantin Belousov  * Use CPUID for draining in the boot-time SMP constistency test.  The
341814124c3SKonstantin Belousov  * timecounters use MFENCE for AMD CPUs, and LFENCE for others (Intel
342814124c3SKonstantin Belousov  * and VIA) when SSE2 is present, and nothing on older machines which
343814124c3SKonstantin Belousov  * also do not issue RDTSC prematurely.  There, testing for SSE2 and
344e1a18e46SKonstantin Belousov  * vendor is too cumbersome, and we learn about TSC presence from CPUID.
345814124c3SKonstantin Belousov  *
346814124c3SKonstantin Belousov  * Do not use do_cpuid(), since we do not need CPUID results, which
347814124c3SKonstantin Belousov  * have to be written into memory with do_cpuid().
348814124c3SKonstantin Belousov  */
34965e7d70bSJung-uk Kim #define	TSC_READ(x)							\
35065e7d70bSJung-uk Kim static void								\
35165e7d70bSJung-uk Kim tsc_read_##x(void *arg)							\
35265e7d70bSJung-uk Kim {									\
3537bfcb3bbSJim Harris 	uint64_t *tsc = arg;						\
35465e7d70bSJung-uk Kim 	u_int cpu = PCPU_GET(cpuid);					\
35565e7d70bSJung-uk Kim 									\
356814124c3SKonstantin Belousov 	__asm __volatile("cpuid" : : : "eax", "ebx", "ecx", "edx");	\
3577bfcb3bbSJim Harris 	tsc[cpu * 3 + x] = rdtsc();					\
35865e7d70bSJung-uk Kim }
35965e7d70bSJung-uk Kim TSC_READ(0)
36065e7d70bSJung-uk Kim TSC_READ(1)
36165e7d70bSJung-uk Kim TSC_READ(2)
36265e7d70bSJung-uk Kim #undef TSC_READ
36365e7d70bSJung-uk Kim 
36465e7d70bSJung-uk Kim #define	N	1000
36565e7d70bSJung-uk Kim 
36665e7d70bSJung-uk Kim static void
36765e7d70bSJung-uk Kim comp_smp_tsc(void *arg)
36865e7d70bSJung-uk Kim {
3697bfcb3bbSJim Harris 	uint64_t *tsc;
3707bfcb3bbSJim Harris 	int64_t d1, d2;
37165e7d70bSJung-uk Kim 	u_int cpu = PCPU_GET(cpuid);
37265e7d70bSJung-uk Kim 	u_int i, j, size;
37365e7d70bSJung-uk Kim 
37465e7d70bSJung-uk Kim 	size = (mp_maxid + 1) * 3;
37565e7d70bSJung-uk Kim 	for (i = 0, tsc = arg; i < N; i++, tsc += size)
37665e7d70bSJung-uk Kim 		CPU_FOREACH(j) {
37765e7d70bSJung-uk Kim 			if (j == cpu)
37865e7d70bSJung-uk Kim 				continue;
37965e7d70bSJung-uk Kim 			d1 = tsc[cpu * 3 + 1] - tsc[j * 3];
38065e7d70bSJung-uk Kim 			d2 = tsc[cpu * 3 + 2] - tsc[j * 3 + 1];
38165e7d70bSJung-uk Kim 			if (d1 <= 0 || d2 <= 0) {
38265e7d70bSJung-uk Kim 				smp_tsc = 0;
38365e7d70bSJung-uk Kim 				return;
38465e7d70bSJung-uk Kim 			}
38565e7d70bSJung-uk Kim 		}
38665e7d70bSJung-uk Kim }
38765e7d70bSJung-uk Kim 
388b2c63698SAlexander Motin static void
389b2c63698SAlexander Motin adj_smp_tsc(void *arg)
390b2c63698SAlexander Motin {
391b2c63698SAlexander Motin 	uint64_t *tsc;
392b2c63698SAlexander Motin 	int64_t d, min, max;
393b2c63698SAlexander Motin 	u_int cpu = PCPU_GET(cpuid);
394b2c63698SAlexander Motin 	u_int first, i, size;
395b2c63698SAlexander Motin 
396b2c63698SAlexander Motin 	first = CPU_FIRST();
397b2c63698SAlexander Motin 	if (cpu == first)
398b2c63698SAlexander Motin 		return;
399b2c63698SAlexander Motin 	min = INT64_MIN;
400b2c63698SAlexander Motin 	max = INT64_MAX;
401b2c63698SAlexander Motin 	size = (mp_maxid + 1) * 3;
402b2c63698SAlexander Motin 	for (i = 0, tsc = arg; i < N; i++, tsc += size) {
403b2c63698SAlexander Motin 		d = tsc[first * 3] - tsc[cpu * 3 + 1];
404b2c63698SAlexander Motin 		if (d > min)
405b2c63698SAlexander Motin 			min = d;
406b2c63698SAlexander Motin 		d = tsc[first * 3 + 1] - tsc[cpu * 3 + 2];
407b2c63698SAlexander Motin 		if (d > min)
408b2c63698SAlexander Motin 			min = d;
409b2c63698SAlexander Motin 		d = tsc[first * 3 + 1] - tsc[cpu * 3];
410b2c63698SAlexander Motin 		if (d < max)
411b2c63698SAlexander Motin 			max = d;
412b2c63698SAlexander Motin 		d = tsc[first * 3 + 2] - tsc[cpu * 3 + 1];
413b2c63698SAlexander Motin 		if (d < max)
414b2c63698SAlexander Motin 			max = d;
415b2c63698SAlexander Motin 	}
416b2c63698SAlexander Motin 	if (min > max)
417b2c63698SAlexander Motin 		return;
418b2c63698SAlexander Motin 	d = min / 2 + max / 2;
419b2c63698SAlexander Motin 	__asm __volatile (
420b2c63698SAlexander Motin 		"movl $0x10, %%ecx\n\t"
421b2c63698SAlexander Motin 		"rdmsr\n\t"
422b2c63698SAlexander Motin 		"addl %%edi, %%eax\n\t"
423b2c63698SAlexander Motin 		"adcl %%esi, %%edx\n\t"
424b2c63698SAlexander Motin 		"wrmsr\n"
425b2c63698SAlexander Motin 		: /* No output */
426b2c63698SAlexander Motin 		: "D" ((uint32_t)d), "S" ((uint32_t)(d >> 32))
427b2c63698SAlexander Motin 		: "ax", "cx", "dx", "cc"
428b2c63698SAlexander Motin 	);
429b2c63698SAlexander Motin }
430b2c63698SAlexander Motin 
43165e7d70bSJung-uk Kim static int
432e7f1427dSKonstantin Belousov test_tsc(void)
43365e7d70bSJung-uk Kim {
4347bfcb3bbSJim Harris 	uint64_t *data, *tsc;
435b2c63698SAlexander Motin 	u_int i, size, adj;
43665e7d70bSJung-uk Kim 
437e7f1427dSKonstantin Belousov 	if ((!smp_tsc && !tsc_is_invariant) || vm_guest)
43865e7d70bSJung-uk Kim 		return (-100);
43965e7d70bSJung-uk Kim 	size = (mp_maxid + 1) * 3;
44065e7d70bSJung-uk Kim 	data = malloc(sizeof(*data) * size * N, M_TEMP, M_WAITOK);
441b2c63698SAlexander Motin 	adj = 0;
442b2c63698SAlexander Motin retry:
44365e7d70bSJung-uk Kim 	for (i = 0, tsc = data; i < N; i++, tsc += size)
44465e7d70bSJung-uk Kim 		smp_rendezvous(tsc_read_0, tsc_read_1, tsc_read_2, tsc);
44565e7d70bSJung-uk Kim 	smp_tsc = 1;	/* XXX */
44665e7d70bSJung-uk Kim 	smp_rendezvous(smp_no_rendevous_barrier, comp_smp_tsc,
44765e7d70bSJung-uk Kim 	    smp_no_rendevous_barrier, data);
448b2c63698SAlexander Motin 	if (!smp_tsc && adj < smp_tsc_adjust) {
449b2c63698SAlexander Motin 		adj++;
450b2c63698SAlexander Motin 		smp_rendezvous(smp_no_rendevous_barrier, adj_smp_tsc,
451b2c63698SAlexander Motin 		    smp_no_rendevous_barrier, data);
452b2c63698SAlexander Motin 		goto retry;
453b2c63698SAlexander Motin 	}
45465e7d70bSJung-uk Kim 	free(data, M_TEMP);
45565e7d70bSJung-uk Kim 	if (bootverbose)
456b2c63698SAlexander Motin 		printf("SMP: %sed TSC synchronization test%s\n",
457b2c63698SAlexander Motin 		    smp_tsc ? "pass" : "fail",
458b2c63698SAlexander Motin 		    adj > 0 ? " after adjustment" : "");
45926e6537aSJung-uk Kim 	if (smp_tsc && tsc_is_invariant) {
46026e6537aSJung-uk Kim 		switch (cpu_vendor_id) {
46126e6537aSJung-uk Kim 		case CPU_VENDOR_AMD:
46226e6537aSJung-uk Kim 			/*
46326e6537aSJung-uk Kim 			 * Starting with Family 15h processors, TSC clock
46426e6537aSJung-uk Kim 			 * source is in the north bridge.  Check whether
46526e6537aSJung-uk Kim 			 * we have a single-socket/multi-core platform.
46626e6537aSJung-uk Kim 			 * XXX Need more work for complex cases.
46726e6537aSJung-uk Kim 			 */
46826e6537aSJung-uk Kim 			if (CPUID_TO_FAMILY(cpu_id) < 0x15 ||
46926e6537aSJung-uk Kim 			    (amd_feature2 & AMDID2_CMP) == 0 ||
47026e6537aSJung-uk Kim 			    smp_cpus > (cpu_procinfo2 & AMDID_CMP_CORES) + 1)
47126e6537aSJung-uk Kim 				break;
47226e6537aSJung-uk Kim 			return (1000);
47326e6537aSJung-uk Kim 		case CPU_VENDOR_INTEL:
47426e6537aSJung-uk Kim 			/*
47526e6537aSJung-uk Kim 			 * XXX Assume Intel platforms have synchronized TSCs.
47626e6537aSJung-uk Kim 			 */
47726e6537aSJung-uk Kim 			return (1000);
47826e6537aSJung-uk Kim 		}
47926e6537aSJung-uk Kim 		return (800);
48026e6537aSJung-uk Kim 	}
48126e6537aSJung-uk Kim 	return (-100);
48265e7d70bSJung-uk Kim }
48365e7d70bSJung-uk Kim 
48465e7d70bSJung-uk Kim #undef N
48565e7d70bSJung-uk Kim 
486e7f1427dSKonstantin Belousov #else
487e7f1427dSKonstantin Belousov 
488e7f1427dSKonstantin Belousov /*
489e7f1427dSKonstantin Belousov  * The function is not called, it is provided to avoid linking failure
490e7f1427dSKonstantin Belousov  * on uniprocessor kernel.
491e7f1427dSKonstantin Belousov  */
492e7f1427dSKonstantin Belousov static int
493e7f1427dSKonstantin Belousov test_tsc(void)
494e7f1427dSKonstantin Belousov {
495e7f1427dSKonstantin Belousov 
496e7f1427dSKonstantin Belousov 	return (0);
497e7f1427dSKonstantin Belousov }
498e7f1427dSKonstantin Belousov 
49965e7d70bSJung-uk Kim #endif /* SMP */
50065e7d70bSJung-uk Kim 
50165e7d70bSJung-uk Kim static void
502dd7d207dSJung-uk Kim init_TSC_tc(void)
503dd7d207dSJung-uk Kim {
50495f2f098SJung-uk Kim 	uint64_t max_freq;
50595f2f098SJung-uk Kim 	int shift;
506dd7d207dSJung-uk Kim 
50738b8542cSJung-uk Kim 	if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled)
508dd7d207dSJung-uk Kim 		return;
509dd7d207dSJung-uk Kim 
510dd7d207dSJung-uk Kim 	/*
51195f2f098SJung-uk Kim 	 * Limit timecounter frequency to fit in an int and prevent it from
51295f2f098SJung-uk Kim 	 * overflowing too fast.
51395f2f098SJung-uk Kim 	 */
51495f2f098SJung-uk Kim 	max_freq = UINT_MAX;
51595f2f098SJung-uk Kim 
51695f2f098SJung-uk Kim 	/*
517dd7d207dSJung-uk Kim 	 * We can not use the TSC if we support APM.  Precise timekeeping
518dd7d207dSJung-uk Kim 	 * on an APM'ed machine is at best a fools pursuit, since
519dd7d207dSJung-uk Kim 	 * any and all of the time spent in various SMM code can't
520dd7d207dSJung-uk Kim 	 * be reliably accounted for.  Reading the RTC is your only
521dd7d207dSJung-uk Kim 	 * source of reliable time info.  The i8254 loses too, of course,
522dd7d207dSJung-uk Kim 	 * but we need to have some kind of time...
523dd7d207dSJung-uk Kim 	 * We don't know at this point whether APM is going to be used
524dd7d207dSJung-uk Kim 	 * or not, nor when it might be activated.  Play it safe.
525dd7d207dSJung-uk Kim 	 */
526dd7d207dSJung-uk Kim 	if (power_pm_get_type() == POWER_PM_TYPE_APM) {
527dd7d207dSJung-uk Kim 		tsc_timecounter.tc_quality = -1000;
528dd7d207dSJung-uk Kim 		if (bootverbose)
529dd7d207dSJung-uk Kim 			printf("TSC timecounter disabled: APM enabled.\n");
53065e7d70bSJung-uk Kim 		goto init;
531dd7d207dSJung-uk Kim 	}
532dd7d207dSJung-uk Kim 
533a49399a9SJung-uk Kim 	/*
53492597e06SJohn Baldwin 	 * Intel CPUs without a C-state invariant TSC can stop the TSC
535d1411416SJohn Baldwin 	 * in either C2 or C3.  Disable use of C2 and C3 while using
536d1411416SJohn Baldwin 	 * the TSC as the timecounter.  The timecounter can be changed
537d1411416SJohn Baldwin 	 * to enable C2 and C3.
538d1411416SJohn Baldwin 	 *
539d1411416SJohn Baldwin 	 * Note that the TSC is used as the cputicker for computing
540d1411416SJohn Baldwin 	 * thread runtime regardless of the timecounter setting, so
541d1411416SJohn Baldwin 	 * using an alternate timecounter and enabling C2 or C3 can
542d1411416SJohn Baldwin 	 * result incorrect runtimes for kernel idle threads (but not
543d1411416SJohn Baldwin 	 * for any non-idle threads).
544a49399a9SJung-uk Kim 	 */
54592597e06SJohn Baldwin 	if (cpu_deepest_sleep >= 2 && cpu_vendor_id == CPU_VENDOR_INTEL &&
546a49399a9SJung-uk Kim 	    (amd_pminfo & AMDPM_TSC_INVARIANT) == 0) {
54792597e06SJohn Baldwin 		tsc_timecounter.tc_flags |= TC_FLAGS_C2STOP;
548a49399a9SJung-uk Kim 		if (bootverbose)
549d1411416SJohn Baldwin 			printf("TSC timecounter disables C2 and C3.\n");
550a49399a9SJung-uk Kim 	}
551a49399a9SJung-uk Kim 
552dd7d207dSJung-uk Kim 	/*
553e7f1427dSKonstantin Belousov 	 * We can not use the TSC in SMP mode unless the TSCs on all CPUs
554e7f1427dSKonstantin Belousov 	 * are synchronized.  If the user is sure that the system has
555e7f1427dSKonstantin Belousov 	 * synchronized TSCs, set kern.timecounter.smp_tsc tunable to a
556e7f1427dSKonstantin Belousov 	 * non-zero value.  The TSC seems unreliable in virtualized SMP
5575cf8ac1bSMike Silbersack 	 * environments, so it is set to a negative quality in those cases.
558dd7d207dSJung-uk Kim 	 */
559e7f1427dSKonstantin Belousov 	if (mp_ncpus > 1)
560e7f1427dSKonstantin Belousov 		tsc_timecounter.tc_quality = test_tsc();
561e7f1427dSKonstantin Belousov 	else if (tsc_is_invariant)
56226e6537aSJung-uk Kim 		tsc_timecounter.tc_quality = 1000;
563e7f1427dSKonstantin Belousov 	max_freq >>= tsc_shift;
56426e6537aSJung-uk Kim 
56565e7d70bSJung-uk Kim init:
566e7f1427dSKonstantin Belousov 	for (shift = 0; shift <= 31 && (tsc_freq >> shift) > max_freq; shift++)
56795f2f098SJung-uk Kim 		;
568e7f1427dSKonstantin Belousov 	if ((cpu_feature & CPUID_SSE2) != 0 && mp_ncpus > 1) {
569814124c3SKonstantin Belousov 		if (cpu_vendor_id == CPU_VENDOR_AMD) {
570e7f1427dSKonstantin Belousov 			tsc_timecounter.tc_get_timecount = shift > 0 ?
571e7f1427dSKonstantin Belousov 			    tsc_get_timecount_low_mfence :
572e7f1427dSKonstantin Belousov 			    tsc_get_timecount_mfence;
573814124c3SKonstantin Belousov 		} else {
574e7f1427dSKonstantin Belousov 			tsc_timecounter.tc_get_timecount = shift > 0 ?
575e7f1427dSKonstantin Belousov 			    tsc_get_timecount_low_lfence :
576e7f1427dSKonstantin Belousov 			    tsc_get_timecount_lfence;
577814124c3SKonstantin Belousov 		}
578e7f1427dSKonstantin Belousov 	} else {
579e7f1427dSKonstantin Belousov 		tsc_timecounter.tc_get_timecount = shift > 0 ?
580e7f1427dSKonstantin Belousov 		    tsc_get_timecount_low : tsc_get_timecount;
581e7f1427dSKonstantin Belousov 	}
582e7f1427dSKonstantin Belousov 	if (shift > 0) {
58395f2f098SJung-uk Kim 		tsc_timecounter.tc_name = "TSC-low";
58495f2f098SJung-uk Kim 		if (bootverbose)
585bc8e4ad2SJung-uk Kim 			printf("TSC timecounter discards lower %d bit(s)\n",
58695f2f098SJung-uk Kim 			    shift);
58795f2f098SJung-uk Kim 	}
588bc34c87eSJung-uk Kim 	if (tsc_freq != 0) {
58995f2f098SJung-uk Kim 		tsc_timecounter.tc_frequency = tsc_freq >> shift;
59095f2f098SJung-uk Kim 		tsc_timecounter.tc_priv = (void *)(intptr_t)shift;
591dd7d207dSJung-uk Kim 		tc_init(&tsc_timecounter);
592dd7d207dSJung-uk Kim 	}
593dd7d207dSJung-uk Kim }
59465e7d70bSJung-uk Kim SYSINIT(tsc_tc, SI_SUB_SMP, SI_ORDER_ANY, init_TSC_tc, NULL);
595dd7d207dSJung-uk Kim 
596dd7d207dSJung-uk Kim /*
597dd7d207dSJung-uk Kim  * When cpufreq levels change, find out about the (new) max frequency.  We
598dd7d207dSJung-uk Kim  * use this to update CPU accounting in case it got a lower estimate at boot.
599dd7d207dSJung-uk Kim  */
600dd7d207dSJung-uk Kim static void
601dd7d207dSJung-uk Kim tsc_levels_changed(void *arg, int unit)
602dd7d207dSJung-uk Kim {
603dd7d207dSJung-uk Kim 	device_t cf_dev;
604dd7d207dSJung-uk Kim 	struct cf_level *levels;
605dd7d207dSJung-uk Kim 	int count, error;
606dd7d207dSJung-uk Kim 	uint64_t max_freq;
607dd7d207dSJung-uk Kim 
608dd7d207dSJung-uk Kim 	/* Only use values from the first CPU, assuming all are equal. */
609dd7d207dSJung-uk Kim 	if (unit != 0)
610dd7d207dSJung-uk Kim 		return;
611dd7d207dSJung-uk Kim 
612dd7d207dSJung-uk Kim 	/* Find the appropriate cpufreq device instance. */
613dd7d207dSJung-uk Kim 	cf_dev = devclass_get_device(devclass_find("cpufreq"), unit);
614dd7d207dSJung-uk Kim 	if (cf_dev == NULL) {
615dd7d207dSJung-uk Kim 		printf("tsc_levels_changed() called but no cpufreq device?\n");
616dd7d207dSJung-uk Kim 		return;
617dd7d207dSJung-uk Kim 	}
618dd7d207dSJung-uk Kim 
619dd7d207dSJung-uk Kim 	/* Get settings from the device and find the max frequency. */
620dd7d207dSJung-uk Kim 	count = 64;
621dd7d207dSJung-uk Kim 	levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT);
622dd7d207dSJung-uk Kim 	if (levels == NULL)
623dd7d207dSJung-uk Kim 		return;
624dd7d207dSJung-uk Kim 	error = CPUFREQ_LEVELS(cf_dev, levels, &count);
625dd7d207dSJung-uk Kim 	if (error == 0 && count != 0) {
626dd7d207dSJung-uk Kim 		max_freq = (uint64_t)levels[0].total_set.freq * 1000000;
627dd7d207dSJung-uk Kim 		set_cputicker(rdtsc, max_freq, 1);
628dd7d207dSJung-uk Kim 	} else
629dd7d207dSJung-uk Kim 		printf("tsc_levels_changed: no max freq found\n");
630dd7d207dSJung-uk Kim 	free(levels, M_TEMP);
631dd7d207dSJung-uk Kim }
632dd7d207dSJung-uk Kim 
633dd7d207dSJung-uk Kim /*
634dd7d207dSJung-uk Kim  * If the TSC timecounter is in use, veto the pending change.  It may be
635dd7d207dSJung-uk Kim  * possible in the future to handle a dynamically-changing timecounter rate.
636dd7d207dSJung-uk Kim  */
637dd7d207dSJung-uk Kim static void
638dd7d207dSJung-uk Kim tsc_freq_changing(void *arg, const struct cf_level *level, int *status)
639dd7d207dSJung-uk Kim {
640dd7d207dSJung-uk Kim 
641dd7d207dSJung-uk Kim 	if (*status != 0 || timecounter != &tsc_timecounter)
642dd7d207dSJung-uk Kim 		return;
643dd7d207dSJung-uk Kim 
644dd7d207dSJung-uk Kim 	printf("timecounter TSC must not be in use when "
645dd7d207dSJung-uk Kim 	    "changing frequencies; change denied\n");
646dd7d207dSJung-uk Kim 	*status = EBUSY;
647dd7d207dSJung-uk Kim }
648dd7d207dSJung-uk Kim 
649dd7d207dSJung-uk Kim /* Update TSC freq with the value indicated by the caller. */
650dd7d207dSJung-uk Kim static void
651dd7d207dSJung-uk Kim tsc_freq_changed(void *arg, const struct cf_level *level, int status)
652dd7d207dSJung-uk Kim {
6533453537fSJung-uk Kim 	uint64_t freq;
654dd7d207dSJung-uk Kim 
655dd7d207dSJung-uk Kim 	/* If there was an error during the transition, don't do anything. */
65679422085SJung-uk Kim 	if (tsc_disabled || status != 0)
657dd7d207dSJung-uk Kim 		return;
658dd7d207dSJung-uk Kim 
659dd7d207dSJung-uk Kim 	/* Total setting for this level gives the new frequency in MHz. */
6603453537fSJung-uk Kim 	freq = (uint64_t)level->total_set.freq * 1000000;
6613453537fSJung-uk Kim 	atomic_store_rel_64(&tsc_freq, freq);
66295f2f098SJung-uk Kim 	tsc_timecounter.tc_frequency =
66395f2f098SJung-uk Kim 	    freq >> (int)(intptr_t)tsc_timecounter.tc_priv;
664dd7d207dSJung-uk Kim }
665dd7d207dSJung-uk Kim 
666dd7d207dSJung-uk Kim static int
667dd7d207dSJung-uk Kim sysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS)
668dd7d207dSJung-uk Kim {
669dd7d207dSJung-uk Kim 	int error;
670dd7d207dSJung-uk Kim 	uint64_t freq;
671dd7d207dSJung-uk Kim 
6723453537fSJung-uk Kim 	freq = atomic_load_acq_64(&tsc_freq);
6733453537fSJung-uk Kim 	if (freq == 0)
674dd7d207dSJung-uk Kim 		return (EOPNOTSUPP);
675cbc134adSMatthew D Fleming 	error = sysctl_handle_64(oidp, &freq, 0, req);
6767ebbcb21SJung-uk Kim 	if (error == 0 && req->newptr != NULL) {
6773453537fSJung-uk Kim 		atomic_store_rel_64(&tsc_freq, freq);
678bc8e4ad2SJung-uk Kim 		atomic_store_rel_64(&tsc_timecounter.tc_frequency,
679bc8e4ad2SJung-uk Kim 		    freq >> (int)(intptr_t)tsc_timecounter.tc_priv);
6807ebbcb21SJung-uk Kim 	}
681dd7d207dSJung-uk Kim 	return (error);
682dd7d207dSJung-uk Kim }
683dd7d207dSJung-uk Kim 
684cbc134adSMatthew D Fleming SYSCTL_PROC(_machdep, OID_AUTO, tsc_freq, CTLTYPE_U64 | CTLFLAG_RW,
6855331d61dSJung-uk Kim     0, 0, sysctl_machdep_tsc_freq, "QU", "Time Stamp Counter frequency");
686dd7d207dSJung-uk Kim 
687727c7b2dSJung-uk Kim static u_int
68895f2f098SJung-uk Kim tsc_get_timecount(struct timecounter *tc __unused)
689dd7d207dSJung-uk Kim {
690727c7b2dSJung-uk Kim 
691727c7b2dSJung-uk Kim 	return (rdtsc32());
692dd7d207dSJung-uk Kim }
69395f2f098SJung-uk Kim 
694814124c3SKonstantin Belousov static inline u_int
695bc8e4ad2SJung-uk Kim tsc_get_timecount_low(struct timecounter *tc)
69695f2f098SJung-uk Kim {
6975df88f46SJung-uk Kim 	uint32_t rv;
69895f2f098SJung-uk Kim 
6995df88f46SJung-uk Kim 	__asm __volatile("rdtsc; shrd %%cl, %%edx, %0"
7005df88f46SJung-uk Kim 	    : "=a" (rv) : "c" ((int)(intptr_t)tc->tc_priv) : "edx");
7015df88f46SJung-uk Kim 	return (rv);
70295f2f098SJung-uk Kim }
703aea81038SKonstantin Belousov 
704814124c3SKonstantin Belousov static u_int
705814124c3SKonstantin Belousov tsc_get_timecount_lfence(struct timecounter *tc __unused)
706814124c3SKonstantin Belousov {
707814124c3SKonstantin Belousov 
708814124c3SKonstantin Belousov 	lfence();
709814124c3SKonstantin Belousov 	return (rdtsc32());
710814124c3SKonstantin Belousov }
711814124c3SKonstantin Belousov 
712814124c3SKonstantin Belousov static u_int
713814124c3SKonstantin Belousov tsc_get_timecount_low_lfence(struct timecounter *tc)
714814124c3SKonstantin Belousov {
715814124c3SKonstantin Belousov 
716814124c3SKonstantin Belousov 	lfence();
717814124c3SKonstantin Belousov 	return (tsc_get_timecount_low(tc));
718814124c3SKonstantin Belousov }
719814124c3SKonstantin Belousov 
720814124c3SKonstantin Belousov static u_int
721814124c3SKonstantin Belousov tsc_get_timecount_mfence(struct timecounter *tc __unused)
722814124c3SKonstantin Belousov {
723814124c3SKonstantin Belousov 
724814124c3SKonstantin Belousov 	mfence();
725814124c3SKonstantin Belousov 	return (rdtsc32());
726814124c3SKonstantin Belousov }
727814124c3SKonstantin Belousov 
728814124c3SKonstantin Belousov static u_int
729814124c3SKonstantin Belousov tsc_get_timecount_low_mfence(struct timecounter *tc)
730814124c3SKonstantin Belousov {
731814124c3SKonstantin Belousov 
732814124c3SKonstantin Belousov 	mfence();
733814124c3SKonstantin Belousov 	return (tsc_get_timecount_low(tc));
734814124c3SKonstantin Belousov }
735814124c3SKonstantin Belousov 
73616808549SKonstantin Belousov static uint32_t
73716808549SKonstantin Belousov x86_tsc_vdso_timehands(struct vdso_timehands *vdso_th, struct timecounter *tc)
738aea81038SKonstantin Belousov {
739aea81038SKonstantin Belousov 
74016808549SKonstantin Belousov 	vdso_th->th_algo = VDSO_TH_ALGO_X86_TSC;
741d1b1b600SNeel Natu 	vdso_th->th_x86_shift = (int)(intptr_t)tc->tc_priv;
74216808549SKonstantin Belousov 	vdso_th->th_x86_hpet_idx = 0xffffffff;
743aea81038SKonstantin Belousov 	bzero(vdso_th->th_res, sizeof(vdso_th->th_res));
74416808549SKonstantin Belousov 	return (1);
745aea81038SKonstantin Belousov }
746aea81038SKonstantin Belousov 
747aea81038SKonstantin Belousov #ifdef COMPAT_FREEBSD32
74816808549SKonstantin Belousov static uint32_t
74916808549SKonstantin Belousov x86_tsc_vdso_timehands32(struct vdso_timehands32 *vdso_th32,
750d1b1b600SNeel Natu     struct timecounter *tc)
751aea81038SKonstantin Belousov {
752aea81038SKonstantin Belousov 
75316808549SKonstantin Belousov 	vdso_th32->th_algo = VDSO_TH_ALGO_X86_TSC;
754d1b1b600SNeel Natu 	vdso_th32->th_x86_shift = (int)(intptr_t)tc->tc_priv;
75516808549SKonstantin Belousov 	vdso_th32->th_x86_hpet_idx = 0xffffffff;
756aea81038SKonstantin Belousov 	bzero(vdso_th32->th_res, sizeof(vdso_th32->th_res));
75716808549SKonstantin Belousov 	return (1);
758aea81038SKonstantin Belousov }
759aea81038SKonstantin Belousov #endif
760