xref: /illumos-gate/usr/src/lib/commpage/common/cp_main.c (revision 1a065e93eee983124652c3eb0cfdcb4776cd89ab)
1 /*
2  * This file and its contents are supplied under the terms of the
3  * Common Development and Distribution License ("CDDL"), version 1.0.
4  * You may only use this file in accordance with the terms of version
5  * 1.0 of the CDDL.
6  *
7  * A full copy of the text of the CDDL should have accompanied this
8  * source.  A copy of the CDDL is also available via the Internet at
9  * http://www.illumos.org/license/CDDL.
10  */
11 
12 /*
13  * Copyright 2019 Joyent, Inc.
14  */
15 
16 #include <sys/comm_page.h>
17 #include <sys/tsc.h>
18 
19 
20 /*
21  * Interrogate if querying the clock via the comm page is possible.
22  */
23 int
24 __cp_can_gettime(comm_page_t *cp)
25 {
26 	switch (cp->cp_tsc_type) {
27 	case TSC_TSCP:
28 	case TSC_RDTSC_LFENCE:
29 	case TSC_RDTSC_CPUID:
30 		return (1);
31 	default:
32 		break;
33 	}
34 	return (0);
35 }
36 
37 #ifdef __amd64
38 
39 /*
40  * The functions used for calculating time (both monotonic and wall-clock) are
41  * implemented in assembly on amd64.  This is primarily for stack conservation.
42  */
43 
44 #else /* i386 below */
45 
46 /*
47  * ASM-defined functions.
48  */
49 extern hrtime_t __cp_tsc_read(comm_page_t *);
50 extern hrtime_t __cp_gethrtime_fasttrap();
51 
52 /*
53  * These are cloned from TSC and time related code in the kernel.  The should
54  * be kept in sync in the case that the source values are changed.
55  */
56 #define	NSEC_SHIFT	5
57 #define	ADJ_SHIFT	4
58 #define	NANOSEC		1000000000LL
59 
60 #define	TSC_CONVERT_AND_ADD(tsc, hrt, scale) do {		\
61 	uint32_t *_l = (uint32_t *)&(tsc);			\
62 	uint64_t sc = (uint32_t)(scale);			\
63 	(hrt) += (uint64_t)(_l[1] * sc) << NSEC_SHIFT;		\
64 	(hrt) += (uint64_t)(_l[0] * sc) >> (32 - NSEC_SHIFT);	\
65 } while (0)
66 
67 /*
68  * Userspace version of tsc_gethrtime.
69  * See: uts/i86pc/os/timestamp.c
70  */
71 hrtime_t
72 __cp_gethrtime(comm_page_t *cp)
73 {
74 	uint32_t old_hres_lock;
75 	hrtime_t tsc, hrt, tsc_last;
76 
77 	/*
78 	 * Several precautions must be taken when collecting the data necessary
79 	 * to perform an accurate gethrtime calculation.
80 	 *
81 	 * While much of the TSC state stored in the comm page is unchanging
82 	 * after boot, portions of it are periodically updated during OS ticks.
83 	 * Changes to hres_lock during the course of the copy indicates a
84 	 * potentially inconsistent snapshot, necessitating a loop.
85 	 *
86 	 * Even more complicated is the handling for TSCs which require sync
87 	 * offsets between different CPUs.  Since userspace lacks the luxury of
88 	 * disabling interrupts, a validation loop checking for CPU migrations
89 	 * is used.  Pathological scheduling could, in theory, "outwit"
90 	 * this check.  Such a possibility is considered an acceptable risk.
91 	 *
92 	 */
93 	do {
94 		old_hres_lock = cp->cp_hres_lock;
95 		tsc_last = cp->cp_tsc_last;
96 		hrt = cp->cp_tsc_hrtime_base;
97 		tsc = __cp_tsc_read(cp);
98 
99 		/*
100 		 * A TSC reading of 0 indicates the special case of an error
101 		 * bail-out.  Rely on the fasttrap to supply an hrtime value.
102 		 */
103 		if (tsc == 0) {
104 			return (__cp_gethrtime_fasttrap());
105 		}
106 	} while ((old_hres_lock & ~1) != cp->cp_hres_lock);
107 
108 	if (tsc >= tsc_last) {
109 		tsc -= tsc_last;
110 	} else if (tsc >= tsc_last - (2 * cp->cp_tsc_max_delta)) {
111 		tsc = 0;
112 	} else if (tsc > cp->cp_tsc_resume_cap) {
113 		tsc = cp->cp_tsc_resume_cap;
114 	}
115 	TSC_CONVERT_AND_ADD(tsc, hrt, cp->cp_nsec_scale);
116 
117 	return (hrt);
118 }
119 
120 /*
121  * Userspace version of pc_gethrestime.
122  * See: uts/i86pc/os/machdep.c
123  */
124 int
125 __cp_clock_gettime_realtime(comm_page_t *cp, timespec_t *tsp)
126 {
127 	int lock_prev, nslt;
128 	timespec_t now;
129 	int64_t hres_adj;
130 
131 loop:
132 	lock_prev = cp->cp_hres_lock;
133 	now.tv_sec = cp->cp_hrestime[0];
134 	now.tv_nsec = cp->cp_hrestime[1];
135 	nslt = (int)(__cp_gethrtime(cp) - cp->cp_hres_last_tick);
136 	hres_adj = cp->cp_hrestime_adj;
137 	if (nslt < 0) {
138 		/*
139 		 * Tick came between sampling hrtime and hres_last_tick;
140 		 */
141 		goto loop;
142 	}
143 	now.tv_nsec += nslt;
144 
145 	/*
146 	 * Apply hres_adj skew, if needed.
147 	 */
148 	if (hres_adj > 0) {
149 		nslt = (nslt >> ADJ_SHIFT);
150 		if (nslt > hres_adj)
151 			nslt = (int)hres_adj;
152 		now.tv_nsec += nslt;
153 	} else if (hres_adj < 0) {
154 		nslt = -(nslt >> ADJ_SHIFT);
155 		if (nslt < hres_adj)
156 			nslt = (int)hres_adj;
157 		now.tv_nsec += nslt;
158 	}
159 
160 	/*
161 	 * Rope in tv_nsec from any excessive adjustments.
162 	 */
163 	while ((unsigned long)now.tv_nsec >= NANOSEC) {
164 		now.tv_nsec -= NANOSEC;
165 		now.tv_sec++;
166 	}
167 
168 	if ((cp->cp_hres_lock & ~1) != lock_prev)
169 		goto loop;
170 
171 	*tsp = now;
172 	return (0);
173 }
174 
175 /*
176  * The __cp_clock_gettime_monotonic function expects that hrt2ts be present
177  * when the code is finally linked.
178  * (The amd64 version has no such requirement.)
179  */
180 extern void hrt2ts(hrtime_t, timespec_t *);
181 
182 int
183 __cp_clock_gettime_monotonic(comm_page_t *cp, timespec_t *tsp)
184 {
185 	hrtime_t hrt;
186 
187 	hrt = __cp_gethrtime(cp);
188 	hrt2ts(hrt, tsp);
189 	return (0);
190 }
191 
192 #endif /* __amd64 */
193