xref: /openbmc/linux/drivers/hwmon/fam15h_power.c (revision 275876e2)
1 /*
2  * fam15h_power.c - AMD Family 15h processor power monitoring
3  *
4  * Copyright (c) 2011 Advanced Micro Devices, Inc.
5  * Author: Andreas Herrmann <herrmann.der.user@googlemail.com>
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>
28 #include <asm/processor.h>
29 
30 MODULE_DESCRIPTION("AMD Family 15h CPU processor power monitor");
31 MODULE_AUTHOR("Andreas Herrmann <herrmann.der.user@googlemail.com>");
32 MODULE_LICENSE("GPL");
33 
34 /* D18F3 */
35 #define REG_NORTHBRIDGE_CAP		0xe8
36 
37 /* D18F4 */
38 #define REG_PROCESSOR_TDP		0x1b8
39 
40 /* D18F5 */
41 #define REG_TDP_RUNNING_AVERAGE		0xe0
42 #define REG_TDP_LIMIT3			0xe8
43 
44 struct fam15h_power_data {
45 	struct pci_dev *pdev;
46 	unsigned int tdp_to_watts;
47 	unsigned int base_tdp;
48 	unsigned int processor_pwr_watts;
49 };
50 
51 static ssize_t show_power(struct device *dev,
52 			  struct device_attribute *attr, char *buf)
53 {
54 	u32 val, tdp_limit, running_avg_range;
55 	s32 running_avg_capture;
56 	u64 curr_pwr_watts;
57 	struct fam15h_power_data *data = dev_get_drvdata(dev);
58 	struct pci_dev *f4 = data->pdev;
59 
60 	pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
61 				  REG_TDP_RUNNING_AVERAGE, &val);
62 	running_avg_capture = (val >> 4) & 0x3fffff;
63 	running_avg_capture = sign_extend32(running_avg_capture, 21);
64 	running_avg_range = (val & 0xf) + 1;
65 
66 	pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
67 				  REG_TDP_LIMIT3, &val);
68 
69 	tdp_limit = val >> 16;
70 	curr_pwr_watts = ((u64)(tdp_limit +
71 				data->base_tdp)) << running_avg_range;
72 	curr_pwr_watts -= running_avg_capture;
73 	curr_pwr_watts *= data->tdp_to_watts;
74 
75 	/*
76 	 * Convert to microWatt
77 	 *
78 	 * power is in Watt provided as fixed point integer with
79 	 * scaling factor 1/(2^16).  For conversion we use
80 	 * (10^6)/(2^16) = 15625/(2^10)
81 	 */
82 	curr_pwr_watts = (curr_pwr_watts * 15625) >> (10 + running_avg_range);
83 	return sprintf(buf, "%u\n", (unsigned int) curr_pwr_watts);
84 }
85 static DEVICE_ATTR(power1_input, S_IRUGO, show_power, NULL);
86 
87 static ssize_t show_power_crit(struct device *dev,
88 			       struct device_attribute *attr, char *buf)
89 {
90 	struct fam15h_power_data *data = dev_get_drvdata(dev);
91 
92 	return sprintf(buf, "%u\n", data->processor_pwr_watts);
93 }
94 static DEVICE_ATTR(power1_crit, S_IRUGO, show_power_crit, NULL);
95 
96 static struct attribute *fam15h_power_attrs[] = {
97 	&dev_attr_power1_input.attr,
98 	&dev_attr_power1_crit.attr,
99 	NULL
100 };
101 
102 ATTRIBUTE_GROUPS(fam15h_power);
103 
104 static bool fam15h_power_is_internal_node0(struct pci_dev *f4)
105 {
106 	u32 val;
107 
108 	pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 3),
109 				  REG_NORTHBRIDGE_CAP, &val);
110 	if ((val & BIT(29)) && ((val >> 30) & 3))
111 		return false;
112 
113 	return true;
114 }
115 
116 /*
117  * Newer BKDG versions have an updated recommendation on how to properly
118  * initialize the running average range (was: 0xE, now: 0x9). This avoids
119  * counter saturations resulting in bogus power readings.
120  * We correct this value ourselves to cope with older BIOSes.
121  */
122 static const struct pci_device_id affected_device[] = {
123 	{ PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F4) },
124 	{ 0 }
125 };
126 
127 static void tweak_runavg_range(struct pci_dev *pdev)
128 {
129 	u32 val;
130 
131 	/*
132 	 * let this quirk apply only to the current version of the
133 	 * northbridge, since future versions may change the behavior
134 	 */
135 	if (!pci_match_id(affected_device, pdev))
136 		return;
137 
138 	pci_bus_read_config_dword(pdev->bus,
139 		PCI_DEVFN(PCI_SLOT(pdev->devfn), 5),
140 		REG_TDP_RUNNING_AVERAGE, &val);
141 	if ((val & 0xf) != 0xe)
142 		return;
143 
144 	val &= ~0xf;
145 	val |=  0x9;
146 	pci_bus_write_config_dword(pdev->bus,
147 		PCI_DEVFN(PCI_SLOT(pdev->devfn), 5),
148 		REG_TDP_RUNNING_AVERAGE, val);
149 }
150 
151 #ifdef CONFIG_PM
152 static int fam15h_power_resume(struct pci_dev *pdev)
153 {
154 	tweak_runavg_range(pdev);
155 	return 0;
156 }
157 #else
158 #define fam15h_power_resume NULL
159 #endif
160 
161 static void fam15h_power_init_data(struct pci_dev *f4,
162 					     struct fam15h_power_data *data)
163 {
164 	u32 val;
165 	u64 tmp;
166 
167 	pci_read_config_dword(f4, REG_PROCESSOR_TDP, &val);
168 	data->base_tdp = val >> 16;
169 	tmp = val & 0xffff;
170 
171 	pci_bus_read_config_dword(f4->bus, PCI_DEVFN(PCI_SLOT(f4->devfn), 5),
172 				  REG_TDP_LIMIT3, &val);
173 
174 	data->tdp_to_watts = ((val & 0x3ff) << 6) | ((val >> 10) & 0x3f);
175 	tmp *= data->tdp_to_watts;
176 
177 	/* result not allowed to be >= 256W */
178 	if ((tmp >> 16) >= 256)
179 		dev_warn(&f4->dev,
180 			 "Bogus value for ProcessorPwrWatts (processor_pwr_watts>=%u)\n",
181 			 (unsigned int) (tmp >> 16));
182 
183 	/* convert to microWatt */
184 	data->processor_pwr_watts = (tmp * 15625) >> 10;
185 }
186 
187 static int fam15h_power_probe(struct pci_dev *pdev,
188 					const struct pci_device_id *id)
189 {
190 	struct fam15h_power_data *data;
191 	struct device *dev = &pdev->dev;
192 	struct device *hwmon_dev;
193 
194 	/*
195 	 * though we ignore every other northbridge, we still have to
196 	 * do the tweaking on _each_ node in MCM processors as the counters
197 	 * are working hand-in-hand
198 	 */
199 	tweak_runavg_range(pdev);
200 
201 	if (!fam15h_power_is_internal_node0(pdev))
202 		return -ENODEV;
203 
204 	data = devm_kzalloc(dev, sizeof(struct fam15h_power_data), GFP_KERNEL);
205 	if (!data)
206 		return -ENOMEM;
207 
208 	fam15h_power_init_data(pdev, data);
209 	data->pdev = pdev;
210 
211 	hwmon_dev = devm_hwmon_device_register_with_groups(dev, "fam15h_power",
212 							   data,
213 							   fam15h_power_groups);
214 	return PTR_ERR_OR_ZERO(hwmon_dev);
215 }
216 
217 static const struct pci_device_id fam15h_power_id_table[] = {
218 	{ PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F4) },
219 	{ PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_NB_F4) },
220 	{}
221 };
222 MODULE_DEVICE_TABLE(pci, fam15h_power_id_table);
223 
224 static struct pci_driver fam15h_power_driver = {
225 	.name = "fam15h_power",
226 	.id_table = fam15h_power_id_table,
227 	.probe = fam15h_power_probe,
228 	.resume = fam15h_power_resume,
229 };
230 
231 module_pci_driver(fam15h_power_driver);
232