1 | /****************************************************************************** |
2 | * pcpu.c |
3 | * Management physical cpu in dom0, get pcpu info and provide sys interface |
4 | * |
5 | * Copyright (c) 2012 Intel Corporation |
6 | * Author: Liu, Jinsong <jinsong.liu@intel.com> |
7 | * Author: Jiang, Yunhong <yunhong.jiang@intel.com> |
8 | * |
9 | * This program is free software; you can redistribute it and/or |
10 | * modify it under the terms of the GNU General Public License version 2 |
11 | * as published by the Free Software Foundation; or, when distributed |
12 | * separately from the Linux kernel or incorporated into other |
13 | * software packages, subject to the following license: |
14 | * |
15 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
16 | * of this source file (the "Software"), to deal in the Software without |
17 | * restriction, including without limitation the rights to use, copy, modify, |
18 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, |
19 | * and to permit persons to whom the Software is furnished to do so, subject to |
20 | * the following conditions: |
21 | * |
22 | * The above copyright notice and this permission notice shall be included in |
23 | * all copies or substantial portions of the Software. |
24 | * |
25 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
26 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
27 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
28 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
29 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
30 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
31 | * IN THE SOFTWARE. |
32 | */ |
33 | |
34 | #define pr_fmt(fmt) "xen_cpu: " fmt |
35 | |
36 | #include <linux/interrupt.h> |
37 | #include <linux/spinlock.h> |
38 | #include <linux/cpu.h> |
39 | #include <linux/stat.h> |
40 | #include <linux/capability.h> |
41 | |
42 | #include <xen/xen.h> |
43 | #include <xen/acpi.h> |
44 | #include <xen/xenbus.h> |
45 | #include <xen/events.h> |
46 | #include <xen/interface/platform.h> |
47 | #include <asm/xen/hypervisor.h> |
48 | #include <asm/xen/hypercall.h> |
49 | |
50 | #ifdef CONFIG_ACPI |
51 | #include <acpi/processor.h> |
52 | #endif |
53 | |
54 | /* |
55 | * @cpu_id: Xen physical cpu logic number |
56 | * @flags: Xen physical cpu status flag |
57 | * - XEN_PCPU_FLAGS_ONLINE: cpu is online |
58 | * - XEN_PCPU_FLAGS_INVALID: cpu is not present |
59 | */ |
60 | struct pcpu { |
61 | struct list_head list; |
62 | struct device dev; |
63 | uint32_t cpu_id; |
64 | uint32_t acpi_id; |
65 | uint32_t flags; |
66 | }; |
67 | |
68 | static const struct bus_type xen_pcpu_subsys = { |
69 | .name = "xen_cpu" , |
70 | .dev_name = "xen_cpu" , |
71 | }; |
72 | |
73 | static DEFINE_MUTEX(xen_pcpu_lock); |
74 | |
75 | static LIST_HEAD(xen_pcpus); |
76 | |
77 | static int xen_pcpu_down(uint32_t cpu_id) |
78 | { |
79 | struct xen_platform_op op = { |
80 | .cmd = XENPF_cpu_offline, |
81 | .interface_version = XENPF_INTERFACE_VERSION, |
82 | .u.cpu_ol.cpuid = cpu_id, |
83 | }; |
84 | |
85 | return HYPERVISOR_platform_op(op: &op); |
86 | } |
87 | |
88 | static int xen_pcpu_up(uint32_t cpu_id) |
89 | { |
90 | struct xen_platform_op op = { |
91 | .cmd = XENPF_cpu_online, |
92 | .interface_version = XENPF_INTERFACE_VERSION, |
93 | .u.cpu_ol.cpuid = cpu_id, |
94 | }; |
95 | |
96 | return HYPERVISOR_platform_op(op: &op); |
97 | } |
98 | |
99 | static ssize_t online_show(struct device *dev, |
100 | struct device_attribute *attr, |
101 | char *buf) |
102 | { |
103 | struct pcpu *cpu = container_of(dev, struct pcpu, dev); |
104 | |
105 | return sprintf(buf, fmt: "%u\n" , !!(cpu->flags & XEN_PCPU_FLAGS_ONLINE)); |
106 | } |
107 | |
108 | static ssize_t __ref online_store(struct device *dev, |
109 | struct device_attribute *attr, |
110 | const char *buf, size_t count) |
111 | { |
112 | struct pcpu *pcpu = container_of(dev, struct pcpu, dev); |
113 | unsigned long long val; |
114 | ssize_t ret; |
115 | |
116 | if (!capable(CAP_SYS_ADMIN)) |
117 | return -EPERM; |
118 | |
119 | if (kstrtoull(s: buf, base: 0, res: &val) < 0) |
120 | return -EINVAL; |
121 | |
122 | switch (val) { |
123 | case 0: |
124 | ret = xen_pcpu_down(cpu_id: pcpu->cpu_id); |
125 | break; |
126 | case 1: |
127 | ret = xen_pcpu_up(cpu_id: pcpu->cpu_id); |
128 | break; |
129 | default: |
130 | ret = -EINVAL; |
131 | } |
132 | |
133 | if (ret >= 0) |
134 | ret = count; |
135 | return ret; |
136 | } |
137 | static DEVICE_ATTR_RW(online); |
138 | |
139 | static struct attribute *pcpu_dev_attrs[] = { |
140 | &dev_attr_online.attr, |
141 | NULL |
142 | }; |
143 | |
144 | static umode_t pcpu_dev_is_visible(struct kobject *kobj, |
145 | struct attribute *attr, int idx) |
146 | { |
147 | struct device *dev = kobj_to_dev(kobj); |
148 | /* |
149 | * Xen never offline cpu0 due to several restrictions |
150 | * and assumptions. This basically doesn't add a sys control |
151 | * to user, one cannot attempt to offline BSP. |
152 | */ |
153 | return dev->id ? attr->mode : 0; |
154 | } |
155 | |
156 | static const struct attribute_group pcpu_dev_group = { |
157 | .attrs = pcpu_dev_attrs, |
158 | .is_visible = pcpu_dev_is_visible, |
159 | }; |
160 | |
161 | static const struct attribute_group *pcpu_dev_groups[] = { |
162 | &pcpu_dev_group, |
163 | NULL |
164 | }; |
165 | |
166 | static bool xen_pcpu_online(uint32_t flags) |
167 | { |
168 | return !!(flags & XEN_PCPU_FLAGS_ONLINE); |
169 | } |
170 | |
171 | static void pcpu_online_status(struct xenpf_pcpuinfo *info, |
172 | struct pcpu *pcpu) |
173 | { |
174 | if (xen_pcpu_online(flags: info->flags) && |
175 | !xen_pcpu_online(flags: pcpu->flags)) { |
176 | /* the pcpu is onlined */ |
177 | pcpu->flags |= XEN_PCPU_FLAGS_ONLINE; |
178 | kobject_uevent(kobj: &pcpu->dev.kobj, action: KOBJ_ONLINE); |
179 | } else if (!xen_pcpu_online(flags: info->flags) && |
180 | xen_pcpu_online(flags: pcpu->flags)) { |
181 | /* The pcpu is offlined */ |
182 | pcpu->flags &= ~XEN_PCPU_FLAGS_ONLINE; |
183 | kobject_uevent(kobj: &pcpu->dev.kobj, action: KOBJ_OFFLINE); |
184 | } |
185 | } |
186 | |
187 | static struct pcpu *get_pcpu(uint32_t cpu_id) |
188 | { |
189 | struct pcpu *pcpu; |
190 | |
191 | list_for_each_entry(pcpu, &xen_pcpus, list) { |
192 | if (pcpu->cpu_id == cpu_id) |
193 | return pcpu; |
194 | } |
195 | |
196 | return NULL; |
197 | } |
198 | |
199 | static void pcpu_release(struct device *dev) |
200 | { |
201 | struct pcpu *pcpu = container_of(dev, struct pcpu, dev); |
202 | |
203 | list_del(entry: &pcpu->list); |
204 | kfree(objp: pcpu); |
205 | } |
206 | |
207 | static void unregister_and_remove_pcpu(struct pcpu *pcpu) |
208 | { |
209 | struct device *dev; |
210 | |
211 | if (!pcpu) |
212 | return; |
213 | |
214 | dev = &pcpu->dev; |
215 | /* pcpu remove would be implicitly done */ |
216 | device_unregister(dev); |
217 | } |
218 | |
219 | static int register_pcpu(struct pcpu *pcpu) |
220 | { |
221 | struct device *dev; |
222 | int err = -EINVAL; |
223 | |
224 | if (!pcpu) |
225 | return err; |
226 | |
227 | dev = &pcpu->dev; |
228 | dev->bus = &xen_pcpu_subsys; |
229 | dev->id = pcpu->cpu_id; |
230 | dev->release = pcpu_release; |
231 | dev->groups = pcpu_dev_groups; |
232 | |
233 | err = device_register(dev); |
234 | if (err) { |
235 | put_device(dev); |
236 | return err; |
237 | } |
238 | |
239 | return 0; |
240 | } |
241 | |
242 | static struct pcpu *create_and_register_pcpu(struct xenpf_pcpuinfo *info) |
243 | { |
244 | struct pcpu *pcpu; |
245 | int err; |
246 | |
247 | if (info->flags & XEN_PCPU_FLAGS_INVALID) |
248 | return ERR_PTR(error: -ENODEV); |
249 | |
250 | pcpu = kzalloc(size: sizeof(struct pcpu), GFP_KERNEL); |
251 | if (!pcpu) |
252 | return ERR_PTR(error: -ENOMEM); |
253 | |
254 | INIT_LIST_HEAD(list: &pcpu->list); |
255 | pcpu->cpu_id = info->xen_cpuid; |
256 | pcpu->acpi_id = info->acpi_id; |
257 | pcpu->flags = info->flags; |
258 | |
259 | /* Need hold on xen_pcpu_lock before pcpu list manipulations */ |
260 | list_add_tail(new: &pcpu->list, head: &xen_pcpus); |
261 | |
262 | err = register_pcpu(pcpu); |
263 | if (err) { |
264 | pr_warn("Failed to register pcpu%u\n" , info->xen_cpuid); |
265 | return ERR_PTR(error: -ENOENT); |
266 | } |
267 | |
268 | return pcpu; |
269 | } |
270 | |
271 | /* |
272 | * Caller should hold the xen_pcpu_lock |
273 | */ |
274 | static int sync_pcpu(uint32_t cpu, uint32_t *max_cpu) |
275 | { |
276 | int ret; |
277 | struct pcpu *pcpu = NULL; |
278 | struct xenpf_pcpuinfo *info; |
279 | struct xen_platform_op op = { |
280 | .cmd = XENPF_get_cpuinfo, |
281 | .interface_version = XENPF_INTERFACE_VERSION, |
282 | .u.pcpu_info.xen_cpuid = cpu, |
283 | }; |
284 | |
285 | ret = HYPERVISOR_platform_op(op: &op); |
286 | if (ret) |
287 | return ret; |
288 | |
289 | info = &op.u.pcpu_info; |
290 | if (max_cpu) |
291 | *max_cpu = info->max_present; |
292 | |
293 | pcpu = get_pcpu(cpu_id: cpu); |
294 | |
295 | /* |
296 | * Only those at cpu present map has its sys interface. |
297 | */ |
298 | if (info->flags & XEN_PCPU_FLAGS_INVALID) { |
299 | unregister_and_remove_pcpu(pcpu); |
300 | return 0; |
301 | } |
302 | |
303 | if (!pcpu) { |
304 | pcpu = create_and_register_pcpu(info); |
305 | if (IS_ERR_OR_NULL(ptr: pcpu)) |
306 | return -ENODEV; |
307 | } else |
308 | pcpu_online_status(info, pcpu); |
309 | |
310 | return 0; |
311 | } |
312 | |
313 | /* |
314 | * Sync dom0's pcpu information with xen hypervisor's |
315 | */ |
316 | static int xen_sync_pcpus(void) |
317 | { |
318 | /* |
319 | * Boot cpu always have cpu_id 0 in xen |
320 | */ |
321 | uint32_t cpu = 0, max_cpu = 0; |
322 | int err = 0; |
323 | struct pcpu *pcpu, *tmp; |
324 | |
325 | mutex_lock(&xen_pcpu_lock); |
326 | |
327 | while (!err && (cpu <= max_cpu)) { |
328 | err = sync_pcpu(cpu, max_cpu: &max_cpu); |
329 | cpu++; |
330 | } |
331 | |
332 | if (err) |
333 | list_for_each_entry_safe(pcpu, tmp, &xen_pcpus, list) |
334 | unregister_and_remove_pcpu(pcpu); |
335 | |
336 | mutex_unlock(lock: &xen_pcpu_lock); |
337 | |
338 | return err; |
339 | } |
340 | |
341 | static void xen_pcpu_work_fn(struct work_struct *work) |
342 | { |
343 | xen_sync_pcpus(); |
344 | } |
345 | static DECLARE_WORK(xen_pcpu_work, xen_pcpu_work_fn); |
346 | |
347 | static irqreturn_t xen_pcpu_interrupt(int irq, void *dev_id) |
348 | { |
349 | schedule_work(work: &xen_pcpu_work); |
350 | return IRQ_HANDLED; |
351 | } |
352 | |
353 | static int __init xen_pcpu_init(void) |
354 | { |
355 | int irq, ret; |
356 | |
357 | if (!xen_initial_domain()) |
358 | return -ENODEV; |
359 | |
360 | irq = bind_virq_to_irqhandler(VIRQ_PCPU_STATE, cpu: 0, |
361 | handler: xen_pcpu_interrupt, irqflags: 0, |
362 | devname: "xen-pcpu" , NULL); |
363 | if (irq < 0) { |
364 | pr_warn("Failed to bind pcpu virq\n" ); |
365 | return irq; |
366 | } |
367 | |
368 | ret = subsys_system_register(subsys: &xen_pcpu_subsys, NULL); |
369 | if (ret) { |
370 | pr_warn("Failed to register pcpu subsys\n" ); |
371 | goto err1; |
372 | } |
373 | |
374 | ret = xen_sync_pcpus(); |
375 | if (ret) { |
376 | pr_warn("Failed to sync pcpu info\n" ); |
377 | goto err2; |
378 | } |
379 | |
380 | return 0; |
381 | |
382 | err2: |
383 | bus_unregister(bus: &xen_pcpu_subsys); |
384 | err1: |
385 | unbind_from_irqhandler(irq, NULL); |
386 | return ret; |
387 | } |
388 | arch_initcall(xen_pcpu_init); |
389 | |
390 | #ifdef CONFIG_ACPI |
391 | bool __init xen_processor_present(uint32_t acpi_id) |
392 | { |
393 | const struct pcpu *pcpu; |
394 | bool online = false; |
395 | |
396 | mutex_lock(&xen_pcpu_lock); |
397 | list_for_each_entry(pcpu, &xen_pcpus, list) |
398 | if (pcpu->acpi_id == acpi_id) { |
399 | online = pcpu->flags & XEN_PCPU_FLAGS_ONLINE; |
400 | break; |
401 | } |
402 | mutex_unlock(lock: &xen_pcpu_lock); |
403 | |
404 | return online; |
405 | } |
406 | |
407 | void xen_sanitize_proc_cap_bits(uint32_t *cap) |
408 | { |
409 | struct xen_platform_op op = { |
410 | .cmd = XENPF_set_processor_pminfo, |
411 | .u.set_pminfo.id = -1, |
412 | .u.set_pminfo.type = XEN_PM_PDC, |
413 | }; |
414 | u32 buf[3] = { ACPI_PDC_REVISION_ID, 1, *cap }; |
415 | int ret; |
416 | |
417 | set_xen_guest_handle(op.u.set_pminfo.pdc, buf); |
418 | ret = HYPERVISOR_platform_op(op: &op); |
419 | if (ret) |
420 | pr_err("sanitize of _PDC buffer bits from Xen failed: %d\n" , |
421 | ret); |
422 | else |
423 | *cap = buf[2]; |
424 | } |
425 | #endif |
426 | |