1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Intel Running Average Power Limit (RAPL) Driver via MSR interface
4  * Copyright (c) 2019, Intel Corporation.
5  */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/list.h>
11 #include <linux/types.h>
12 #include <linux/device.h>
13 #include <linux/slab.h>
14 #include <linux/log2.h>
15 #include <linux/bitmap.h>
16 #include <linux/delay.h>
17 #include <linux/sysfs.h>
18 #include <linux/cpu.h>
19 #include <linux/powercap.h>
20 #include <linux/suspend.h>
21 #include <linux/intel_rapl.h>
22 #include <linux/processor.h>
23 #include <linux/platform_device.h>
24 
25 #include <asm/cpu_device_id.h>
26 #include <asm/intel-family.h>
27 
28 /* Local defines */
29 #define MSR_PLATFORM_POWER_LIMIT	0x0000065C
30 #define MSR_VR_CURRENT_CONFIG		0x00000601
31 
32 /* private data for RAPL MSR Interface */
33 static struct rapl_if_priv *rapl_msr_priv;
34 
35 static struct rapl_if_priv rapl_msr_priv_intel = {
36 	.type = RAPL_IF_MSR,
37 	.reg_unit.msr = MSR_RAPL_POWER_UNIT,
38 	.regs[RAPL_DOMAIN_PACKAGE][RAPL_DOMAIN_REG_LIMIT].msr	= MSR_PKG_POWER_LIMIT,
39 	.regs[RAPL_DOMAIN_PACKAGE][RAPL_DOMAIN_REG_STATUS].msr	= MSR_PKG_ENERGY_STATUS,
40 	.regs[RAPL_DOMAIN_PACKAGE][RAPL_DOMAIN_REG_PERF].msr	= MSR_PKG_PERF_STATUS,
41 	.regs[RAPL_DOMAIN_PACKAGE][RAPL_DOMAIN_REG_INFO].msr	= MSR_PKG_POWER_INFO,
42 	.regs[RAPL_DOMAIN_PP0][RAPL_DOMAIN_REG_LIMIT].msr	= MSR_PP0_POWER_LIMIT,
43 	.regs[RAPL_DOMAIN_PP0][RAPL_DOMAIN_REG_STATUS].msr	= MSR_PP0_ENERGY_STATUS,
44 	.regs[RAPL_DOMAIN_PP0][RAPL_DOMAIN_REG_POLICY].msr	= MSR_PP0_POLICY,
45 	.regs[RAPL_DOMAIN_PP1][RAPL_DOMAIN_REG_LIMIT].msr	= MSR_PP1_POWER_LIMIT,
46 	.regs[RAPL_DOMAIN_PP1][RAPL_DOMAIN_REG_STATUS].msr	= MSR_PP1_ENERGY_STATUS,
47 	.regs[RAPL_DOMAIN_PP1][RAPL_DOMAIN_REG_POLICY].msr	= MSR_PP1_POLICY,
48 	.regs[RAPL_DOMAIN_DRAM][RAPL_DOMAIN_REG_LIMIT].msr	= MSR_DRAM_POWER_LIMIT,
49 	.regs[RAPL_DOMAIN_DRAM][RAPL_DOMAIN_REG_STATUS].msr	= MSR_DRAM_ENERGY_STATUS,
50 	.regs[RAPL_DOMAIN_DRAM][RAPL_DOMAIN_REG_PERF].msr	= MSR_DRAM_PERF_STATUS,
51 	.regs[RAPL_DOMAIN_DRAM][RAPL_DOMAIN_REG_INFO].msr	= MSR_DRAM_POWER_INFO,
52 	.regs[RAPL_DOMAIN_PLATFORM][RAPL_DOMAIN_REG_LIMIT].msr	= MSR_PLATFORM_POWER_LIMIT,
53 	.regs[RAPL_DOMAIN_PLATFORM][RAPL_DOMAIN_REG_STATUS].msr	= MSR_PLATFORM_ENERGY_STATUS,
54 	.limits[RAPL_DOMAIN_PACKAGE] = BIT(POWER_LIMIT2),
55 	.limits[RAPL_DOMAIN_PLATFORM] = BIT(POWER_LIMIT2),
56 };
57 
58 static struct rapl_if_priv rapl_msr_priv_amd = {
59 	.type = RAPL_IF_MSR,
60 	.reg_unit.msr = MSR_AMD_RAPL_POWER_UNIT,
61 	.regs[RAPL_DOMAIN_PACKAGE][RAPL_DOMAIN_REG_STATUS].msr	= MSR_AMD_PKG_ENERGY_STATUS,
62 	.regs[RAPL_DOMAIN_PP0][RAPL_DOMAIN_REG_STATUS].msr	= MSR_AMD_CORE_ENERGY_STATUS,
63 };
64 
65 /* Handles CPU hotplug on multi-socket systems.
66  * If a CPU goes online as the first CPU of the physical package
67  * we add the RAPL package to the system. Similarly, when the last
68  * CPU of the package is removed, we remove the RAPL package and its
69  * associated domains. Cooling devices are handled accordingly at
70  * per-domain level.
71  */
72 static int rapl_cpu_online(unsigned int cpu)
73 {
74 	struct rapl_package *rp;
75 
76 	rp = rapl_find_package_domain(cpu, rapl_msr_priv, true);
77 	if (!rp) {
78 		rp = rapl_add_package(cpu, rapl_msr_priv, true);
79 		if (IS_ERR(rp))
80 			return PTR_ERR(rp);
81 	}
82 	cpumask_set_cpu(cpu, &rp->cpumask);
83 	return 0;
84 }
85 
86 static int rapl_cpu_down_prep(unsigned int cpu)
87 {
88 	struct rapl_package *rp;
89 	int lead_cpu;
90 
91 	rp = rapl_find_package_domain(cpu, rapl_msr_priv, true);
92 	if (!rp)
93 		return 0;
94 
95 	cpumask_clear_cpu(cpu, &rp->cpumask);
96 	lead_cpu = cpumask_first(&rp->cpumask);
97 	if (lead_cpu >= nr_cpu_ids)
98 		rapl_remove_package(rp);
99 	else if (rp->lead_cpu == cpu)
100 		rp->lead_cpu = lead_cpu;
101 	return 0;
102 }
103 
104 static int rapl_msr_read_raw(int cpu, struct reg_action *ra)
105 {
106 	if (rdmsrl_safe_on_cpu(cpu, ra->reg.msr, &ra->value)) {
107 		pr_debug("failed to read msr 0x%x on cpu %d\n", ra->reg.msr, cpu);
108 		return -EIO;
109 	}
110 	ra->value &= ra->mask;
111 	return 0;
112 }
113 
114 static void rapl_msr_update_func(void *info)
115 {
116 	struct reg_action *ra = info;
117 	u64 val;
118 
119 	ra->err = rdmsrl_safe(ra->reg.msr, &val);
120 	if (ra->err)
121 		return;
122 
123 	val &= ~ra->mask;
124 	val |= ra->value;
125 
126 	ra->err = wrmsrl_safe(ra->reg.msr, val);
127 }
128 
129 static int rapl_msr_write_raw(int cpu, struct reg_action *ra)
130 {
131 	int ret;
132 
133 	ret = smp_call_function_single(cpu, rapl_msr_update_func, ra, 1);
134 	if (WARN_ON_ONCE(ret))
135 		return ret;
136 
137 	return ra->err;
138 }
139 
140 /* List of verified CPUs. */
141 static const struct x86_cpu_id pl4_support_ids[] = {
142 	X86_MATCH_INTEL_FAM6_MODEL(TIGERLAKE_L, NULL),
143 	X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE, NULL),
144 	X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_L, NULL),
145 	X86_MATCH_INTEL_FAM6_MODEL(ATOM_GRACEMONT, NULL),
146 	X86_MATCH_INTEL_FAM6_MODEL(RAPTORLAKE, NULL),
147 	X86_MATCH_INTEL_FAM6_MODEL(RAPTORLAKE_P, NULL),
148 	X86_MATCH_INTEL_FAM6_MODEL(METEORLAKE, NULL),
149 	X86_MATCH_INTEL_FAM6_MODEL(METEORLAKE_L, NULL),
150 	{}
151 };
152 
153 static int rapl_msr_probe(struct platform_device *pdev)
154 {
155 	const struct x86_cpu_id *id = x86_match_cpu(pl4_support_ids);
156 	int ret;
157 
158 	switch (boot_cpu_data.x86_vendor) {
159 	case X86_VENDOR_INTEL:
160 		rapl_msr_priv = &rapl_msr_priv_intel;
161 		break;
162 	case X86_VENDOR_HYGON:
163 	case X86_VENDOR_AMD:
164 		rapl_msr_priv = &rapl_msr_priv_amd;
165 		break;
166 	default:
167 		pr_err("intel-rapl does not support CPU vendor %d\n", boot_cpu_data.x86_vendor);
168 		return -ENODEV;
169 	}
170 	rapl_msr_priv->read_raw = rapl_msr_read_raw;
171 	rapl_msr_priv->write_raw = rapl_msr_write_raw;
172 
173 	if (id) {
174 		rapl_msr_priv->limits[RAPL_DOMAIN_PACKAGE] |= BIT(POWER_LIMIT4);
175 		rapl_msr_priv->regs[RAPL_DOMAIN_PACKAGE][RAPL_DOMAIN_REG_PL4].msr =
176 			MSR_VR_CURRENT_CONFIG;
177 		pr_info("PL4 support detected.\n");
178 	}
179 
180 	rapl_msr_priv->control_type = powercap_register_control_type(NULL, "intel-rapl", NULL);
181 	if (IS_ERR(rapl_msr_priv->control_type)) {
182 		pr_debug("failed to register powercap control_type.\n");
183 		return PTR_ERR(rapl_msr_priv->control_type);
184 	}
185 
186 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powercap/rapl:online",
187 				rapl_cpu_online, rapl_cpu_down_prep);
188 	if (ret < 0)
189 		goto out;
190 	rapl_msr_priv->pcap_rapl_online = ret;
191 
192 	return 0;
193 
194 out:
195 	if (ret)
196 		powercap_unregister_control_type(rapl_msr_priv->control_type);
197 	return ret;
198 }
199 
200 static int rapl_msr_remove(struct platform_device *pdev)
201 {
202 	cpuhp_remove_state(rapl_msr_priv->pcap_rapl_online);
203 	powercap_unregister_control_type(rapl_msr_priv->control_type);
204 	return 0;
205 }
206 
207 static const struct platform_device_id rapl_msr_ids[] = {
208 	{ .name = "intel_rapl_msr", },
209 	{}
210 };
211 MODULE_DEVICE_TABLE(platform, rapl_msr_ids);
212 
213 static struct platform_driver intel_rapl_msr_driver = {
214 	.probe = rapl_msr_probe,
215 	.remove = rapl_msr_remove,
216 	.id_table = rapl_msr_ids,
217 	.driver = {
218 		.name = "intel_rapl_msr",
219 	},
220 };
221 
222 module_platform_driver(intel_rapl_msr_driver);
223 
224 MODULE_DESCRIPTION("Driver for Intel RAPL (Running Average Power Limit) control via MSR interface");
225 MODULE_AUTHOR("Zhang Rui <rui.zhang@intel.com>");
226 MODULE_LICENSE("GPL v2");
227