kernel_5.10 no rt

hc

2023-12-11 6778948f9de86c3cfaf36725a7c87dcff9ba247f

 kernel/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c
..
..
@@ -1,7 +1,6 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
1
2
 /*
2
3
  *  (C) 2010,2011       Thomas Renninger <trenn@suse.de>, Novell Inc.
3
- *
4
- *  Licensed under the terms of the GNU GPL License version 2.
5
4
  */
6
5
 
7
6
 #if defined(__i386__) || defined(__x86_64__)
..
..
@@ -19,6 +18,10 @@
19
18
 
20
19
 #define MSR_APERF	0xE8
21
20
 #define MSR_MPERF	0xE7
21
+
22
+#define RDPRU ".byte 0x0f, 0x01, 0xfd"
23
+#define RDPRU_ECX_MPERF	0
24
+#define RDPRU_ECX_APERF	1
22
25
 
23
26
 #define MSR_TSC	0x10
24
27
 
..
..
@@ -87,15 +90,51 @@
87
90
 	return ret;
88
91
 }
89
92
 
90
-static int mperf_init_stats(unsigned int cpu)
93
+static int get_aperf_mperf(int cpu, unsigned long long *aval,
94
+				    unsigned long long *mval)
91
95
 {
92
-	unsigned long long val;
96
+	unsigned long low_a, high_a;
97
+	unsigned long low_m, high_m;
93
98
 	int ret;
94
99
 
95
-	ret = read_msr(cpu, MSR_APERF, &val);
96
-	aperf_previous_count[cpu] = val;
97
-	ret |= read_msr(cpu, MSR_MPERF, &val);
98
-	mperf_previous_count[cpu] = val;
100
+	/*
101
+	 * Running on the cpu from which we read the registers will
102
+	 * prevent APERF/MPERF from going out of sync because of IPI
103
+	 * latency introduced by read_msr()s.
104
+	 */
105
+	if (mperf_monitor.flags.per_cpu_schedule) {
106
+		if (bind_cpu(cpu))
107
+			return 1;
108
+	}
109
+
110
+	if (cpupower_cpu_info.caps & CPUPOWER_CAP_AMD_RDPRU) {
111
+		asm volatile(RDPRU
112
+			     : "=a" (low_a), "=d" (high_a)
113
+			     : "c" (RDPRU_ECX_APERF));
114
+		asm volatile(RDPRU
115
+			     : "=a" (low_m), "=d" (high_m)
116
+			     : "c" (RDPRU_ECX_MPERF));
117
+
118
+		*aval = ((low_a) | (high_a) << 32);
119
+		*mval = ((low_m) | (high_m) << 32);
120
+
121
+		return 0;
122
+	}
123
+
124
+	ret  = read_msr(cpu, MSR_APERF, aval);
125
+	ret |= read_msr(cpu, MSR_MPERF, mval);
126
+
127
+	return ret;
128
+}
129
+
130
+static int mperf_init_stats(unsigned int cpu)
131
+{
132
+	unsigned long long aval, mval;
133
+	int ret;
134
+
135
+	ret = get_aperf_mperf(cpu, &aval, &mval);
136
+	aperf_previous_count[cpu] = aval;
137
+	mperf_previous_count[cpu] = mval;
99
138
 	is_valid[cpu] = !ret;
100
139
 
101
140
 	return 0;
..
..
@@ -103,13 +142,12 @@
103
142
 
104
143
 static int mperf_measure_stats(unsigned int cpu)
105
144
 {
106
-	unsigned long long val;
145
+	unsigned long long aval, mval;
107
146
 	int ret;
108
147
 
109
-	ret = read_msr(cpu, MSR_APERF, &val);
110
-	aperf_current_count[cpu] = val;
111
-	ret |= read_msr(cpu, MSR_MPERF, &val);
112
-	mperf_current_count[cpu] = val;
148
+	ret = get_aperf_mperf(cpu, &aval, &mval);
149
+	aperf_current_count[cpu] = aval;
150
+	mperf_current_count[cpu] = mval;
113
151
 	is_valid[cpu] = !ret;
114
152
 
115
153
 	return 0;
..
..
@@ -241,7 +279,8 @@
241
279
 	if (!(cpupower_cpu_info.caps & CPUPOWER_CAP_INV_TSC))
242
280
 		goto use_sysfs;
243
281
 
244
-	if (cpupower_cpu_info.vendor == X86_VENDOR_AMD) {
282
+	if (cpupower_cpu_info.vendor == X86_VENDOR_AMD ||
283
+	    cpupower_cpu_info.vendor == X86_VENDOR_HYGON) {
245
284
 		/* MSR_AMD_HWCR tells us whether TSC runs at P0/mperf
246
285
 		 * freq.
247
286
 		 * A test whether hwcr is accessable/available would be:
..
..
@@ -305,6 +344,9 @@
305
344
 	if (init_maxfreq_mode())
306
345
 		return NULL;
307
346
 
347
+	if (cpupower_cpu_info.vendor == X86_VENDOR_AMD)
348
+		mperf_monitor.flags.per_cpu_schedule = 1;
349
+
308
350
 	/* Free this at program termination */
309
351
 	is_valid = calloc(cpu_count, sizeof(int));
310
352
 	mperf_previous_count = calloc(cpu_count, sizeof(unsigned long long));
..
..
@@ -333,7 +375,7 @@
333
375
 	.stop			= mperf_stop,
334
376
 	.do_register		= mperf_register,
335
377
 	.unregister		= mperf_unregister,
336
-	.needs_root		= 1,
378
+	.flags.needs_root	= 1,
337
379
 	.overflow_s		= 922000000 /* 922337203 seconds TSC overflow
338
380
 					       at 20GHz */
339
381
 };