xref: /linux/drivers/gpu/drm/xe/xe_device_sysfs.c (revision 89748acdf226fd1a8775ff6fa2703f8412b286c8)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2023 Intel Corporation
4  */
5 
6 #include <linux/device.h>
7 #include <linux/kobject.h>
8 #include <linux/pci.h>
9 #include <linux/sysfs.h>
10 
11 #include "xe_device.h"
12 #include "xe_device_sysfs.h"
13 #include "xe_mmio.h"
14 #include "xe_pcode_api.h"
15 #include "xe_pcode.h"
16 #include "xe_pm.h"
17 
18 /**
19  * DOC: Xe device sysfs
20  * Xe driver requires exposing certain tunable knobs controlled by user space for
21  * each graphics device. Considering this, we need to add sysfs attributes at device
22  * level granularity.
23  * These sysfs attributes will be available under pci device kobj directory.
24  *
25  * vram_d3cold_threshold - Report/change vram used threshold(in MB) below
26  * which vram save/restore is permissible during runtime D3cold entry/exit.
27  *
28  * lb_fan_control_version - Fan control version provisioned by late binding.
29  * Exposed only if supported by the device.
30  *
31  * lb_voltage_regulator_version - Voltage regulator version provisioned by late
32  * binding. Exposed only if supported by the device.
33  */
34 
35 static ssize_t
vram_d3cold_threshold_show(struct device * dev,struct device_attribute * attr,char * buf)36 vram_d3cold_threshold_show(struct device *dev,
37 			   struct device_attribute *attr, char *buf)
38 {
39 	struct pci_dev *pdev = to_pci_dev(dev);
40 	struct xe_device *xe = pdev_to_xe_device(pdev);
41 	int ret;
42 
43 	xe_pm_runtime_get(xe);
44 	ret = sysfs_emit(buf, "%d\n", xe->d3cold.vram_threshold);
45 	xe_pm_runtime_put(xe);
46 
47 	return ret;
48 }
49 
50 static ssize_t
vram_d3cold_threshold_store(struct device * dev,struct device_attribute * attr,const char * buff,size_t count)51 vram_d3cold_threshold_store(struct device *dev, struct device_attribute *attr,
52 			    const char *buff, size_t count)
53 {
54 	struct pci_dev *pdev = to_pci_dev(dev);
55 	struct xe_device *xe = pdev_to_xe_device(pdev);
56 	u32 vram_d3cold_threshold;
57 	int ret;
58 
59 	ret = kstrtou32(buff, 0, &vram_d3cold_threshold);
60 	if (ret)
61 		return ret;
62 
63 	drm_dbg(&xe->drm, "vram_d3cold_threshold: %u\n", vram_d3cold_threshold);
64 
65 	xe_pm_runtime_get(xe);
66 	ret = xe_pm_set_vram_threshold(xe, vram_d3cold_threshold);
67 	xe_pm_runtime_put(xe);
68 
69 	return ret ?: count;
70 }
71 
72 static DEVICE_ATTR_RW(vram_d3cold_threshold);
73 
74 static ssize_t
lb_fan_control_version_show(struct device * dev,struct device_attribute * attr,char * buf)75 lb_fan_control_version_show(struct device *dev, struct device_attribute *attr, char *buf)
76 {
77 	struct xe_device *xe = pdev_to_xe_device(to_pci_dev(dev));
78 	struct xe_tile *root = xe_device_get_root_tile(xe);
79 	u32 cap, ver_low = FAN_TABLE, ver_high = FAN_TABLE;
80 	u16 major = 0, minor = 0, hotfix = 0, build = 0;
81 	int ret;
82 
83 	xe_pm_runtime_get(xe);
84 
85 	ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_CAPABILITY_STATUS, 0),
86 			    &cap, NULL);
87 	if (ret)
88 		goto out;
89 
90 	if (REG_FIELD_GET(V1_FAN_PROVISIONED, cap)) {
91 		ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_VERSION_LOW, 0),
92 				    &ver_low, NULL);
93 		if (ret)
94 			goto out;
95 
96 		ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_VERSION_HIGH, 0),
97 				    &ver_high, NULL);
98 		if (ret)
99 			goto out;
100 
101 		major = REG_FIELD_GET(MAJOR_VERSION_MASK, ver_low);
102 		minor = REG_FIELD_GET(MINOR_VERSION_MASK, ver_low);
103 		hotfix = REG_FIELD_GET(HOTFIX_VERSION_MASK, ver_high);
104 		build = REG_FIELD_GET(BUILD_VERSION_MASK, ver_high);
105 	}
106 out:
107 	xe_pm_runtime_put(xe);
108 
109 	return ret ?: sysfs_emit(buf, "%u.%u.%u.%u\n", major, minor, hotfix, build);
110 }
111 static DEVICE_ATTR_ADMIN_RO(lb_fan_control_version);
112 
113 static ssize_t
lb_voltage_regulator_version_show(struct device * dev,struct device_attribute * attr,char * buf)114 lb_voltage_regulator_version_show(struct device *dev, struct device_attribute *attr, char *buf)
115 {
116 	struct xe_device *xe = pdev_to_xe_device(to_pci_dev(dev));
117 	struct xe_tile *root = xe_device_get_root_tile(xe);
118 	u32 cap, ver_low = VR_CONFIG, ver_high = VR_CONFIG;
119 	u16 major = 0, minor = 0, hotfix = 0, build = 0;
120 	int ret;
121 
122 	xe_pm_runtime_get(xe);
123 
124 	ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_CAPABILITY_STATUS, 0),
125 			    &cap, NULL);
126 	if (ret)
127 		goto out;
128 
129 	if (REG_FIELD_GET(VR_PARAMS_PROVISIONED, cap)) {
130 		ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_VERSION_LOW, 0),
131 				    &ver_low, NULL);
132 		if (ret)
133 			goto out;
134 
135 		ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_VERSION_HIGH, 0),
136 				    &ver_high, NULL);
137 		if (ret)
138 			goto out;
139 
140 		major = REG_FIELD_GET(MAJOR_VERSION_MASK, ver_low);
141 		minor = REG_FIELD_GET(MINOR_VERSION_MASK, ver_low);
142 		hotfix = REG_FIELD_GET(HOTFIX_VERSION_MASK, ver_high);
143 		build = REG_FIELD_GET(BUILD_VERSION_MASK, ver_high);
144 	}
145 out:
146 	xe_pm_runtime_put(xe);
147 
148 	return ret ?: sysfs_emit(buf, "%u.%u.%u.%u\n", major, minor, hotfix, build);
149 }
150 static DEVICE_ATTR_ADMIN_RO(lb_voltage_regulator_version);
151 
late_bind_create_files(struct device * dev)152 static int late_bind_create_files(struct device *dev)
153 {
154 	struct xe_device *xe = pdev_to_xe_device(to_pci_dev(dev));
155 	struct xe_tile *root = xe_device_get_root_tile(xe);
156 	u32 cap;
157 	int ret;
158 
159 	xe_pm_runtime_get(xe);
160 
161 	ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_CAPABILITY_STATUS, 0),
162 			    &cap, NULL);
163 	if (ret) {
164 		if (ret == -ENXIO) {
165 			drm_dbg(&xe->drm, "Late binding not supported by firmware\n");
166 			ret = 0;
167 		}
168 		goto out;
169 	}
170 
171 	if (REG_FIELD_GET(V1_FAN_SUPPORTED, cap)) {
172 		ret = sysfs_create_file(&dev->kobj, &dev_attr_lb_fan_control_version.attr);
173 		if (ret)
174 			goto out;
175 	}
176 
177 	if (REG_FIELD_GET(VR_PARAMS_SUPPORTED, cap))
178 		ret = sysfs_create_file(&dev->kobj, &dev_attr_lb_voltage_regulator_version.attr);
179 out:
180 	xe_pm_runtime_put(xe);
181 
182 	return ret;
183 }
184 
late_bind_remove_files(struct device * dev)185 static void late_bind_remove_files(struct device *dev)
186 {
187 	struct xe_device *xe = pdev_to_xe_device(to_pci_dev(dev));
188 	struct xe_tile *root = xe_device_get_root_tile(xe);
189 	u32 cap;
190 	int ret;
191 
192 	xe_pm_runtime_get(xe);
193 
194 	ret = xe_pcode_read(root, PCODE_MBOX(PCODE_LATE_BINDING, GET_CAPABILITY_STATUS, 0),
195 			    &cap, NULL);
196 	if (ret)
197 		goto out;
198 
199 	if (REG_FIELD_GET(V1_FAN_SUPPORTED, cap))
200 		sysfs_remove_file(&dev->kobj, &dev_attr_lb_fan_control_version.attr);
201 
202 	if (REG_FIELD_GET(VR_PARAMS_SUPPORTED, cap))
203 		sysfs_remove_file(&dev->kobj, &dev_attr_lb_voltage_regulator_version.attr);
204 out:
205 	xe_pm_runtime_put(xe);
206 }
207 
208 /**
209  * DOC: PCIe Gen5 Limitations
210  *
211  * Default link speed of discrete GPUs is determined by configuration parameters
212  * stored in their flash memory, which are subject to override through user
213  * initiated firmware updates. It has been observed that devices configured with
214  * PCIe Gen5 as their default link speed can come across link quality issues due
215  * to host or motherboard limitations and may have to auto-downgrade their link
216  * to PCIe Gen4 speed when faced with unstable link at Gen5, which makes
217  * firmware updates rather risky on such setups. It is required to ensure that
218  * the device is capable of auto-downgrading its link to PCIe Gen4 speed before
219  * pushing the firmware image with PCIe Gen5 as default configuration. This can
220  * be done by reading ``auto_link_downgrade_capable`` sysfs entry, which will
221  * denote if the device is capable of auto-downgrading its link to PCIe Gen4
222  * speed with boolean output value of ``0`` or ``1``, meaning `incapable` or
223  * `capable` respectively.
224  *
225  * .. code-block:: shell
226  *
227  *    $ cat /sys/bus/pci/devices/<bdf>/auto_link_downgrade_capable
228  *
229  * Pushing the firmware image with PCIe Gen5 as default configuration on a auto
230  * link downgrade incapable device and facing link instability due to host or
231  * motherboard limitations can result in driver failing to bind to the device,
232  * making further firmware updates impossible with RMA being the only last
233  * resort.
234  *
235  * Link downgrade status of auto link downgrade capable devices is available
236  * through ``auto_link_downgrade_status`` sysfs entry with boolean output value
237  * of ``0`` or ``1``, where ``0`` means no auto-downgrading was required during
238  * link training (which is the optimal scenario) and ``1`` means the device has
239  * auto-downgraded its link to PCIe Gen4 speed due to unstable Gen5 link.
240  *
241  * .. code-block:: shell
242  *
243  *    $ cat /sys/bus/pci/devices/<bdf>/auto_link_downgrade_status
244  */
245 
246 static ssize_t
auto_link_downgrade_capable_show(struct device * dev,struct device_attribute * attr,char * buf)247 auto_link_downgrade_capable_show(struct device *dev, struct device_attribute *attr, char *buf)
248 {
249 	struct pci_dev *pdev = to_pci_dev(dev);
250 	struct xe_device *xe = pdev_to_xe_device(pdev);
251 	u32 cap, val;
252 
253 	xe_pm_runtime_get(xe);
254 	val = xe_mmio_read32(xe_root_tile_mmio(xe), BMG_PCIE_CAP);
255 	xe_pm_runtime_put(xe);
256 
257 	cap = REG_FIELD_GET(LINK_DOWNGRADE, val);
258 	return sysfs_emit(buf, "%u\n", cap == DOWNGRADE_CAPABLE);
259 }
260 static DEVICE_ATTR_ADMIN_RO(auto_link_downgrade_capable);
261 
262 static ssize_t
auto_link_downgrade_status_show(struct device * dev,struct device_attribute * attr,char * buf)263 auto_link_downgrade_status_show(struct device *dev, struct device_attribute *attr, char *buf)
264 {
265 	struct pci_dev *pdev = to_pci_dev(dev);
266 	struct xe_device *xe = pdev_to_xe_device(pdev);
267 	/* default the auto_link_downgrade status to 0 */
268 	u32 val = 0;
269 	int ret;
270 
271 	xe_pm_runtime_get(xe);
272 	ret = xe_pcode_read(xe_device_get_root_tile(xe),
273 			    PCODE_MBOX(DGFX_PCODE_STATUS, DGFX_GET_INIT_STATUS, 0),
274 			    &val, NULL);
275 	xe_pm_runtime_put(xe);
276 
277 	return ret ?: sysfs_emit(buf, "%u\n", REG_FIELD_GET(DGFX_LINK_DOWNGRADE_STATUS, val));
278 }
279 static DEVICE_ATTR_ADMIN_RO(auto_link_downgrade_status);
280 
281 static const struct attribute *auto_link_downgrade_attrs[] = {
282 	&dev_attr_auto_link_downgrade_capable.attr,
283 	&dev_attr_auto_link_downgrade_status.attr,
284 	NULL
285 };
286 
xe_device_sysfs_fini(void * arg)287 static void xe_device_sysfs_fini(void *arg)
288 {
289 	struct xe_device *xe = arg;
290 
291 	if (xe->d3cold.capable)
292 		sysfs_remove_file(&xe->drm.dev->kobj, &dev_attr_vram_d3cold_threshold.attr);
293 
294 	if (xe->info.platform == XE_BATTLEMAGE) {
295 		sysfs_remove_files(&xe->drm.dev->kobj, auto_link_downgrade_attrs);
296 		late_bind_remove_files(xe->drm.dev);
297 	}
298 }
299 
xe_device_sysfs_init(struct xe_device * xe)300 int xe_device_sysfs_init(struct xe_device *xe)
301 {
302 	struct device *dev = xe->drm.dev;
303 	int ret;
304 
305 	if (xe->d3cold.capable) {
306 		ret = sysfs_create_file(&dev->kobj, &dev_attr_vram_d3cold_threshold.attr);
307 		if (ret)
308 			return ret;
309 	}
310 
311 	if (xe->info.platform == XE_BATTLEMAGE) {
312 		ret = sysfs_create_files(&dev->kobj, auto_link_downgrade_attrs);
313 		if (ret)
314 			return ret;
315 
316 		ret = late_bind_create_files(dev);
317 		if (ret)
318 			return ret;
319 	}
320 
321 	return devm_add_action_or_reset(dev, xe_device_sysfs_fini, xe);
322 }
323