hwmon: (fam15h_power) Add compute unit accumulated power
[deliverable/linux.git] / drivers / hwmon / fam15h_power.c
CommitLineData
512d1027
AH
1/*
2 * fam15h_power.c - AMD Family 15h processor power monitoring
3 *
4 * Copyright (c) 2011 Advanced Micro Devices, Inc.
d034fbf0 5 * Author: Andreas Herrmann <herrmann.der.user@googlemail.com>
512d1027
AH
6 *
7 *
8 * This driver is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License; either
10 * version 2 of the License, or (at your option) any later version.
11 *
12 * This driver is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
15 * See the GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this driver; if not, see <http://www.gnu.org/licenses/>.
19 */
20
21#include <linux/err.h>
22#include <linux/hwmon.h>
23#include <linux/hwmon-sysfs.h>
24#include <linux/init.h>
25#include <linux/module.h>
26#include <linux/pci.h>
27#include <linux/bitops.h>
fa794344
HR
28#include <linux/cpu.h>
29#include <linux/cpumask.h>
512d1027 30#include <asm/processor.h>
3b5ea47d 31#include <asm/msr.h>
512d1027
AH
32
33MODULE_DESCRIPTION("AMD Family 15h CPU processor power monitor");
d034fbf0 34MODULE_AUTHOR("Andreas Herrmann <herrmann.der.user@googlemail.com>");
512d1027
AH
35MODULE_LICENSE("GPL");
36
37/* D18F3 */
38#define REG_NORTHBRIDGE_CAP 0xe8
39
40/* D18F4 */
41#define REG_PROCESSOR_TDP 0x1b8
42
43/* D18F5 */
44#define REG_TDP_RUNNING_AVERAGE 0xe0
45#define REG_TDP_LIMIT3 0xe8
46
7deb14b1
HR
47#define FAM15H_MIN_NUM_ATTRS 2
48#define FAM15H_NUM_GROUPS 2
fa794344 49#define MAX_CUS 8
7deb14b1 50
fa794344 51#define MSR_F15H_CU_PWR_ACCUMULATOR 0xc001007a
3b5ea47d
HR
52#define MSR_F15H_CU_MAX_PWR_ACCUMULATOR 0xc001007b
53
eff2a945
HR
54#define PCI_DEVICE_ID_AMD_15H_M70H_NB_F4 0x15b4
55
512d1027 56struct fam15h_power_data {
562dc973 57 struct pci_dev *pdev;
512d1027
AH
58 unsigned int tdp_to_watts;
59 unsigned int base_tdp;
60 unsigned int processor_pwr_watts;
1ed32160 61 unsigned int cpu_pwr_sample_ratio;
7deb14b1
HR
62 const struct attribute_group *groups[FAM15H_NUM_GROUPS];
63 struct attribute_group group;
3b5ea47d
HR
64 /* maximum accumulated power of a compute unit */
65 u64 max_cu_acc_power;
fa794344
HR
66 /* accumulated power of the compute units */
67 u64 cu_acc_power[MAX_CUS];
512d1027
AH
68};
69
70static ssize_t show_power(struct device *dev,
71 struct device_attribute *attr, char *buf)
72{
73 u32 val, tdp_limit, running_avg_range;
74 s32 running_avg_capture;
75 u64 curr_pwr_watts;
512d1027 76 struct fam15h_power_data *data = dev_get_drvdata(dev);
562dc973 77 struct pci_dev *f4 = data->pdev;
512d1027
AH
78
79 pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
80 REG_TDP_RUNNING_AVERAGE, &val);
e9cd4d55
HR
81
82 /*
83 * On Carrizo and later platforms, TdpRunAvgAccCap bit field
84 * is extended to 4:31 from 4:25.
85 */
86 if (boot_cpu_data.x86 == 0x15 && boot_cpu_data.x86_model >= 0x60) {
87 running_avg_capture = val >> 4;
88 running_avg_capture = sign_extend32(running_avg_capture, 27);
89 } else {
90 running_avg_capture = (val >> 4) & 0x3fffff;
91 running_avg_capture = sign_extend32(running_avg_capture, 21);
92 }
93
941a956b 94 running_avg_range = (val & 0xf) + 1;
512d1027
AH
95
96 pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
97 REG_TDP_LIMIT3, &val);
98
60dee3ca
GK
99 /*
100 * On Carrizo and later platforms, ApmTdpLimit bit field
101 * is extended to 16:31 from 16:28.
102 */
103 if (boot_cpu_data.x86 == 0x15 && boot_cpu_data.x86_model >= 0x60)
104 tdp_limit = val >> 16;
105 else
106 tdp_limit = (val >> 16) & 0x1fff;
107
62867d49
GR
108 curr_pwr_watts = ((u64)(tdp_limit +
109 data->base_tdp)) << running_avg_range;
941a956b 110 curr_pwr_watts -= running_avg_capture;
512d1027
AH
111 curr_pwr_watts *= data->tdp_to_watts;
112
113 /*
114 * Convert to microWatt
115 *
116 * power is in Watt provided as fixed point integer with
117 * scaling factor 1/(2^16). For conversion we use
118 * (10^6)/(2^16) = 15625/(2^10)
119 */
941a956b 120 curr_pwr_watts = (curr_pwr_watts * 15625) >> (10 + running_avg_range);
512d1027
AH
121 return sprintf(buf, "%u\n", (unsigned int) curr_pwr_watts);
122}
123static DEVICE_ATTR(power1_input, S_IRUGO, show_power, NULL);
124
125static ssize_t show_power_crit(struct device *dev,
126 struct device_attribute *attr, char *buf)
127{
128 struct fam15h_power_data *data = dev_get_drvdata(dev);
129
130 return sprintf(buf, "%u\n", data->processor_pwr_watts);
131}
132static DEVICE_ATTR(power1_crit, S_IRUGO, show_power_crit, NULL);
133
fa794344
HR
134static void do_read_registers_on_cu(void *_data)
135{
136 struct fam15h_power_data *data = _data;
137 int cpu, cu;
138
139 cpu = smp_processor_id();
140
141 /*
142 * With the new x86 topology modelling, cpu core id actually
143 * is compute unit id.
144 */
145 cu = cpu_data(cpu).cpu_core_id;
146
147 rdmsrl_safe(MSR_F15H_CU_PWR_ACCUMULATOR, &data->cu_acc_power[cu]);
148}
149
150/*
151 * This function is only able to be called when CPUID
152 * Fn8000_0007:EDX[12] is set.
153 */
154static int read_registers(struct fam15h_power_data *data)
155{
156 int this_cpu, ret, cpu;
157 int core, this_core;
158 cpumask_var_t mask;
159
160 ret = zalloc_cpumask_var(&mask, GFP_KERNEL);
161 if (!ret)
162 return -ENOMEM;
163
164 get_online_cpus();
165 this_cpu = smp_processor_id();
166
167 /*
168 * Choose the first online core of each compute unit, and then
169 * read their MSR value of power and ptsc in a single IPI,
170 * because the MSR value of CPU core represent the compute
171 * unit's.
172 */
173 core = -1;
174
175 for_each_online_cpu(cpu) {
176 this_core = topology_core_id(cpu);
177
178 if (this_core == core)
179 continue;
180
181 core = this_core;
182
183 /* get any CPU on this compute unit */
184 cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(cpu)), mask);
185 }
186
187 if (cpumask_test_cpu(this_cpu, mask))
188 do_read_registers_on_cu(data);
189
190 smp_call_function_many(mask, do_read_registers_on_cu, data, true);
191 put_online_cpus();
192
193 free_cpumask_var(mask);
194
195 return 0;
196}
197
7deb14b1
HR
198static int fam15h_power_init_attrs(struct pci_dev *pdev,
199 struct fam15h_power_data *data)
961a2378 200{
7deb14b1
HR
201 int n = FAM15H_MIN_NUM_ATTRS;
202 struct attribute **fam15h_power_attrs;
46f29c2b 203 struct cpuinfo_x86 *c = &boot_cpu_data;
961a2378 204
46f29c2b
HR
205 if (c->x86 == 0x15 &&
206 (c->x86_model <= 0xf ||
eff2a945 207 (c->x86_model >= 0x60 && c->x86_model <= 0x7f)))
7deb14b1 208 n += 1;
961a2378 209
7deb14b1
HR
210 fam15h_power_attrs = devm_kcalloc(&pdev->dev, n,
211 sizeof(*fam15h_power_attrs),
212 GFP_KERNEL);
512d1027 213
7deb14b1
HR
214 if (!fam15h_power_attrs)
215 return -ENOMEM;
216
217 n = 0;
218 fam15h_power_attrs[n++] = &dev_attr_power1_crit.attr;
46f29c2b
HR
219 if (c->x86 == 0x15 &&
220 (c->x86_model <= 0xf ||
eff2a945 221 (c->x86_model >= 0x60 && c->x86_model <= 0x7f)))
7deb14b1
HR
222 fam15h_power_attrs[n++] = &dev_attr_power1_input.attr;
223
224 data->group.attrs = fam15h_power_attrs;
225
226 return 0;
227}
512d1027 228
d83e92b3 229static bool should_load_on_this_node(struct pci_dev *f4)
512d1027
AH
230{
231 u32 val;
232
233 pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 3),
234 REG_NORTHBRIDGE_CAP, &val);
235 if ((val & BIT(29)) && ((val >> 30) & 3))
236 return false;
237
238 return true;
239}
240
00250ec9
AP
241/*
242 * Newer BKDG versions have an updated recommendation on how to properly
243 * initialize the running average range (was: 0xE, now: 0x9). This avoids
244 * counter saturations resulting in bogus power readings.
245 * We correct this value ourselves to cope with older BIOSes.
246 */
5f0ecb90 247static const struct pci_device_id affected_device[] = {
c3e40a99
GR
248 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F4) },
249 { 0 }
250};
251
5f0ecb90 252static void tweak_runavg_range(struct pci_dev *pdev)
00250ec9
AP
253{
254 u32 val;
00250ec9
AP
255
256 /*
257 * let this quirk apply only to the current version of the
258 * northbridge, since future versions may change the behavior
259 */
c3e40a99 260 if (!pci_match_id(affected_device, pdev))
00250ec9
AP
261 return;
262
263 pci_bus_read_config_dword(pdev->bus,
264 PCI_DEVFN(PCI_SLOT(pdev->devfn), 5),
265 REG_TDP_RUNNING_AVERAGE, &val);
266 if ((val & 0xf) != 0xe)
267 return;
268
269 val &= ~0xf;
270 val |= 0x9;
271 pci_bus_write_config_dword(pdev->bus,
272 PCI_DEVFN(PCI_SLOT(pdev->devfn), 5),
273 REG_TDP_RUNNING_AVERAGE, val);
274}
275
5f0ecb90
AH
276#ifdef CONFIG_PM
277static int fam15h_power_resume(struct pci_dev *pdev)
278{
279 tweak_runavg_range(pdev);
280 return 0;
281}
282#else
283#define fam15h_power_resume NULL
284#endif
285
7deb14b1
HR
286static int fam15h_power_init_data(struct pci_dev *f4,
287 struct fam15h_power_data *data)
512d1027 288{
1ed32160 289 u32 val, eax, ebx, ecx, edx;
512d1027 290 u64 tmp;
7deb14b1 291 int ret;
512d1027
AH
292
293 pci_read_config_dword(f4, REG_PROCESSOR_TDP, &val);
294 data->base_tdp = val >> 16;
295 tmp = val & 0xffff;
296
297 pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
298 REG_TDP_LIMIT3, &val);
299
300 data->tdp_to_watts = ((val & 0x3ff) << 6) | ((val >> 10) & 0x3f);
301 tmp *= data->tdp_to_watts;
302
303 /* result not allowed to be >= 256W */
304 if ((tmp >> 16) >= 256)
b55f3757
GR
305 dev_warn(&f4->dev,
306 "Bogus value for ProcessorPwrWatts (processor_pwr_watts>=%u)\n",
512d1027
AH
307 (unsigned int) (tmp >> 16));
308
309 /* convert to microWatt */
310 data->processor_pwr_watts = (tmp * 15625) >> 10;
1ed32160 311
7deb14b1
HR
312 ret = fam15h_power_init_attrs(f4, data);
313 if (ret)
314 return ret;
315
1ed32160
HR
316 cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
317
318 /* CPUID Fn8000_0007:EDX[12] indicates to support accumulated power */
319 if (!(edx & BIT(12)))
7deb14b1 320 return 0;
1ed32160
HR
321
322 /*
323 * determine the ratio of the compute unit power accumulator
324 * sample period to the PTSC counter period by executing CPUID
325 * Fn8000_0007:ECX
326 */
327 data->cpu_pwr_sample_ratio = ecx;
7deb14b1 328
3b5ea47d
HR
329 if (rdmsrl_safe(MSR_F15H_CU_MAX_PWR_ACCUMULATOR, &tmp)) {
330 pr_err("Failed to read max compute unit power accumulator MSR\n");
331 return -ENODEV;
332 }
333
334 data->max_cu_acc_power = tmp;
335
fa794344 336 return read_registers(data);
512d1027
AH
337}
338
6c931ae1 339static int fam15h_power_probe(struct pci_dev *pdev,
7deb14b1 340 const struct pci_device_id *id)
512d1027
AH
341{
342 struct fam15h_power_data *data;
87432a2e 343 struct device *dev = &pdev->dev;
562dc973 344 struct device *hwmon_dev;
7deb14b1 345 int ret;
512d1027 346
00250ec9
AP
347 /*
348 * though we ignore every other northbridge, we still have to
349 * do the tweaking on _each_ node in MCM processors as the counters
350 * are working hand-in-hand
351 */
352 tweak_runavg_range(pdev);
353
d83e92b3 354 if (!should_load_on_this_node(pdev))
87432a2e
GR
355 return -ENODEV;
356
357 data = devm_kzalloc(dev, sizeof(struct fam15h_power_data), GFP_KERNEL);
358 if (!data)
359 return -ENOMEM;
512d1027 360
7deb14b1
HR
361 ret = fam15h_power_init_data(pdev, data);
362 if (ret)
363 return ret;
364
562dc973 365 data->pdev = pdev;
512d1027 366
7deb14b1
HR
367 data->groups[0] = &data->group;
368
562dc973
AL
369 hwmon_dev = devm_hwmon_device_register_with_groups(dev, "fam15h_power",
370 data,
7deb14b1 371 &data->groups[0]);
562dc973 372 return PTR_ERR_OR_ZERO(hwmon_dev);
512d1027
AH
373}
374
cd9bb056 375static const struct pci_device_id fam15h_power_id_table[] = {
512d1027 376 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F4) },
0a0039ad 377 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M30H_NB_F4) },
5dc08725 378 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M60H_NB_F4) },
eff2a945 379 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M70H_NB_F4) },
22e32f4f 380 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_NB_F4) },
0bd52941 381 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_M30H_NB_F4) },
512d1027
AH
382 {}
383};
384MODULE_DEVICE_TABLE(pci, fam15h_power_id_table);
385
386static struct pci_driver fam15h_power_driver = {
387 .name = "fam15h_power",
388 .id_table = fam15h_power_id_table,
389 .probe = fam15h_power_probe,
5f0ecb90 390 .resume = fam15h_power_resume,
512d1027
AH
391};
392
f71f5a55 393module_pci_driver(fam15h_power_driver);
This page took 0.485569 seconds and 5 git commands to generate.