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 | |
51 | /* |
52 | * @cpu_id: Xen physical cpu logic number |
53 | * @flags: Xen physical cpu status flag |
54 | * - XEN_PCPU_FLAGS_ONLINE: cpu is online |
55 | * - XEN_PCPU_FLAGS_INVALID: cpu is not present |
56 | */ |
57 | struct pcpu { |
58 | struct list_head list; |
59 | struct device dev; |
60 | uint32_t cpu_id; |
61 | uint32_t acpi_id; |
62 | uint32_t flags; |
63 | }; |
64 | |
65 | static struct bus_type xen_pcpu_subsys = { |
66 | .name = "xen_cpu" , |
67 | .dev_name = "xen_cpu" , |
68 | }; |
69 | |
70 | static DEFINE_MUTEX(xen_pcpu_lock); |
71 | |
72 | static LIST_HEAD(xen_pcpus); |
73 | |
74 | static int xen_pcpu_down(uint32_t cpu_id) |
75 | { |
76 | struct xen_platform_op op = { |
77 | .cmd = XENPF_cpu_offline, |
78 | .interface_version = XENPF_INTERFACE_VERSION, |
79 | .u.cpu_ol.cpuid = cpu_id, |
80 | }; |
81 | |
82 | return HYPERVISOR_platform_op(op: &op); |
83 | } |
84 | |
85 | static int xen_pcpu_up(uint32_t cpu_id) |
86 | { |
87 | struct xen_platform_op op = { |
88 | .cmd = XENPF_cpu_online, |
89 | .interface_version = XENPF_INTERFACE_VERSION, |
90 | .u.cpu_ol.cpuid = cpu_id, |
91 | }; |
92 | |
93 | return HYPERVISOR_platform_op(op: &op); |
94 | } |
95 | |
96 | static ssize_t online_show(struct device *dev, |
97 | struct device_attribute *attr, |
98 | char *buf) |
99 | { |
100 | struct pcpu *cpu = container_of(dev, struct pcpu, dev); |
101 | |
102 | return sprintf(buf, fmt: "%u\n" , !!(cpu->flags & XEN_PCPU_FLAGS_ONLINE)); |
103 | } |
104 | |
105 | static ssize_t __ref online_store(struct device *dev, |
106 | struct device_attribute *attr, |
107 | const char *buf, size_t count) |
108 | { |
109 | struct pcpu *pcpu = container_of(dev, struct pcpu, dev); |
110 | unsigned long long val; |
111 | ssize_t ret; |
112 | |
113 | if (!capable(CAP_SYS_ADMIN)) |
114 | return -EPERM; |
115 | |
116 | if (kstrtoull(s: buf, base: 0, res: &val) < 0) |
117 | return -EINVAL; |
118 | |
119 | switch (val) { |
120 | case 0: |
121 | ret = xen_pcpu_down(cpu_id: pcpu->cpu_id); |
122 | break; |
123 | case 1: |
124 | ret = xen_pcpu_up(cpu_id: pcpu->cpu_id); |
125 | break; |
126 | default: |
127 | ret = -EINVAL; |
128 | } |
129 | |
130 | if (ret >= 0) |
131 | ret = count; |
132 | return ret; |
133 | } |
134 | static DEVICE_ATTR_RW(online); |
135 | |
136 | static struct attribute *pcpu_dev_attrs[] = { |
137 | &dev_attr_online.attr, |
138 | NULL |
139 | }; |
140 | |
141 | static umode_t pcpu_dev_is_visible(struct kobject *kobj, |
142 | struct attribute *attr, int idx) |
143 | { |
144 | struct device *dev = kobj_to_dev(kobj); |
145 | /* |
146 | * Xen never offline cpu0 due to several restrictions |
147 | * and assumptions. This basically doesn't add a sys control |
148 | * to user, one cannot attempt to offline BSP. |
149 | */ |
150 | return dev->id ? attr->mode : 0; |
151 | } |
152 | |
153 | static const struct attribute_group pcpu_dev_group = { |
154 | .attrs = pcpu_dev_attrs, |
155 | .is_visible = pcpu_dev_is_visible, |
156 | }; |
157 | |
158 | static const struct attribute_group *pcpu_dev_groups[] = { |
159 | &pcpu_dev_group, |
160 | NULL |
161 | }; |
162 | |
163 | static bool xen_pcpu_online(uint32_t flags) |
164 | { |
165 | return !!(flags & XEN_PCPU_FLAGS_ONLINE); |
166 | } |
167 | |
168 | static void pcpu_online_status(struct xenpf_pcpuinfo *info, |
169 | struct pcpu *pcpu) |
170 | { |
171 | if (xen_pcpu_online(flags: info->flags) && |
172 | !xen_pcpu_online(flags: pcpu->flags)) { |
173 | /* the pcpu is onlined */ |
174 | pcpu->flags |= XEN_PCPU_FLAGS_ONLINE; |
175 | kobject_uevent(kobj: &pcpu->dev.kobj, action: KOBJ_ONLINE); |
176 | } else if (!xen_pcpu_online(flags: info->flags) && |
177 | xen_pcpu_online(flags: pcpu->flags)) { |
178 | /* The pcpu is offlined */ |
179 | pcpu->flags &= ~XEN_PCPU_FLAGS_ONLINE; |
180 | kobject_uevent(kobj: &pcpu->dev.kobj, action: KOBJ_OFFLINE); |
181 | } |
182 | } |
183 | |
184 | static struct pcpu *get_pcpu(uint32_t cpu_id) |
185 | { |
186 | struct pcpu *pcpu; |
187 | |
188 | list_for_each_entry(pcpu, &xen_pcpus, list) { |
189 | if (pcpu->cpu_id == cpu_id) |
190 | return pcpu; |
191 | } |
192 | |
193 | return NULL; |
194 | } |
195 | |
196 | static void pcpu_release(struct device *dev) |
197 | { |
198 | struct pcpu *pcpu = container_of(dev, struct pcpu, dev); |
199 | |
200 | list_del(entry: &pcpu->list); |
201 | kfree(objp: pcpu); |
202 | } |
203 | |
204 | static void unregister_and_remove_pcpu(struct pcpu *pcpu) |
205 | { |
206 | struct device *dev; |
207 | |
208 | if (!pcpu) |
209 | return; |
210 | |
211 | dev = &pcpu->dev; |
212 | /* pcpu remove would be implicitly done */ |
213 | device_unregister(dev); |
214 | } |
215 | |
216 | static int register_pcpu(struct pcpu *pcpu) |
217 | { |
218 | struct device *dev; |
219 | int err = -EINVAL; |
220 | |
221 | if (!pcpu) |
222 | return err; |
223 | |
224 | dev = &pcpu->dev; |
225 | dev->bus = &xen_pcpu_subsys; |
226 | dev->id = pcpu->cpu_id; |
227 | dev->release = pcpu_release; |
228 | dev->groups = pcpu_dev_groups; |
229 | |
230 | err = device_register(dev); |
231 | if (err) { |
232 | put_device(dev); |
233 | return err; |
234 | } |
235 | |
236 | return 0; |
237 | } |
238 | |
239 | static struct pcpu *create_and_register_pcpu(struct xenpf_pcpuinfo *info) |
240 | { |
241 | struct pcpu *pcpu; |
242 | int err; |
243 | |
244 | if (info->flags & XEN_PCPU_FLAGS_INVALID) |
245 | return ERR_PTR(error: -ENODEV); |
246 | |
247 | pcpu = kzalloc(size: sizeof(struct pcpu), GFP_KERNEL); |
248 | if (!pcpu) |
249 | return ERR_PTR(error: -ENOMEM); |
250 | |
251 | INIT_LIST_HEAD(list: &pcpu->list); |
252 | pcpu->cpu_id = info->xen_cpuid; |
253 | pcpu->acpi_id = info->acpi_id; |
254 | pcpu->flags = info->flags; |
255 | |
256 | /* Need hold on xen_pcpu_lock before pcpu list manipulations */ |
257 | list_add_tail(new: &pcpu->list, head: &xen_pcpus); |
258 | |
259 | err = register_pcpu(pcpu); |
260 | if (err) { |
261 | pr_warn("Failed to register pcpu%u\n" , info->xen_cpuid); |
262 | return ERR_PTR(error: -ENOENT); |
263 | } |
264 | |
265 | return pcpu; |
266 | } |
267 | |
268 | /* |
269 | * Caller should hold the xen_pcpu_lock |
270 | */ |
271 | static int sync_pcpu(uint32_t cpu, uint32_t *max_cpu) |
272 | { |
273 | int ret; |
274 | struct pcpu *pcpu = NULL; |
275 | struct xenpf_pcpuinfo *info; |
276 | struct xen_platform_op op = { |
277 | .cmd = XENPF_get_cpuinfo, |
278 | .interface_version = XENPF_INTERFACE_VERSION, |
279 | .u.pcpu_info.xen_cpuid = cpu, |
280 | }; |
281 | |
282 | ret = HYPERVISOR_platform_op(op: &op); |
283 | if (ret) |
284 | return ret; |
285 | |
286 | info = &op.u.pcpu_info; |
287 | if (max_cpu) |
288 | *max_cpu = info->max_present; |
289 | |
290 | pcpu = get_pcpu(cpu_id: cpu); |
291 | |
292 | /* |
293 | * Only those at cpu present map has its sys interface. |
294 | */ |
295 | if (info->flags & XEN_PCPU_FLAGS_INVALID) { |
296 | unregister_and_remove_pcpu(pcpu); |
297 | return 0; |
298 | } |
299 | |
300 | if (!pcpu) { |
301 | pcpu = create_and_register_pcpu(info); |
302 | if (IS_ERR_OR_NULL(ptr: pcpu)) |
303 | return -ENODEV; |
304 | } else |
305 | pcpu_online_status(info, pcpu); |
306 | |
307 | return 0; |
308 | } |
309 | |
310 | /* |
311 | * Sync dom0's pcpu information with xen hypervisor's |
312 | */ |
313 | static int xen_sync_pcpus(void) |
314 | { |
315 | /* |
316 | * Boot cpu always have cpu_id 0 in xen |
317 | */ |
318 | uint32_t cpu = 0, max_cpu = 0; |
319 | int err = 0; |
320 | struct pcpu *pcpu, *tmp; |
321 | |
322 | mutex_lock(&xen_pcpu_lock); |
323 | |
324 | while (!err && (cpu <= max_cpu)) { |
325 | err = sync_pcpu(cpu, max_cpu: &max_cpu); |
326 | cpu++; |
327 | } |
328 | |
329 | if (err) |
330 | list_for_each_entry_safe(pcpu, tmp, &xen_pcpus, list) |
331 | unregister_and_remove_pcpu(pcpu); |
332 | |
333 | mutex_unlock(lock: &xen_pcpu_lock); |
334 | |
335 | return err; |
336 | } |
337 | |
338 | static void xen_pcpu_work_fn(struct work_struct *work) |
339 | { |
340 | xen_sync_pcpus(); |
341 | } |
342 | static DECLARE_WORK(xen_pcpu_work, xen_pcpu_work_fn); |
343 | |
344 | static irqreturn_t xen_pcpu_interrupt(int irq, void *dev_id) |
345 | { |
346 | schedule_work(work: &xen_pcpu_work); |
347 | return IRQ_HANDLED; |
348 | } |
349 | |
350 | static int __init xen_pcpu_init(void) |
351 | { |
352 | int irq, ret; |
353 | |
354 | if (!xen_initial_domain()) |
355 | return -ENODEV; |
356 | |
357 | irq = bind_virq_to_irqhandler(VIRQ_PCPU_STATE, cpu: 0, |
358 | handler: xen_pcpu_interrupt, irqflags: 0, |
359 | devname: "xen-pcpu" , NULL); |
360 | if (irq < 0) { |
361 | pr_warn("Failed to bind pcpu virq\n" ); |
362 | return irq; |
363 | } |
364 | |
365 | ret = subsys_system_register(subsys: &xen_pcpu_subsys, NULL); |
366 | if (ret) { |
367 | pr_warn("Failed to register pcpu subsys\n" ); |
368 | goto err1; |
369 | } |
370 | |
371 | ret = xen_sync_pcpus(); |
372 | if (ret) { |
373 | pr_warn("Failed to sync pcpu info\n" ); |
374 | goto err2; |
375 | } |
376 | |
377 | return 0; |
378 | |
379 | err2: |
380 | bus_unregister(bus: &xen_pcpu_subsys); |
381 | err1: |
382 | unbind_from_irqhandler(irq, NULL); |
383 | return ret; |
384 | } |
385 | arch_initcall(xen_pcpu_init); |
386 | |
387 | #ifdef CONFIG_ACPI |
388 | bool __init xen_processor_present(uint32_t acpi_id) |
389 | { |
390 | const struct pcpu *pcpu; |
391 | bool online = false; |
392 | |
393 | mutex_lock(&xen_pcpu_lock); |
394 | list_for_each_entry(pcpu, &xen_pcpus, list) |
395 | if (pcpu->acpi_id == acpi_id) { |
396 | online = pcpu->flags & XEN_PCPU_FLAGS_ONLINE; |
397 | break; |
398 | } |
399 | mutex_unlock(lock: &xen_pcpu_lock); |
400 | |
401 | return online; |
402 | } |
403 | #endif |
404 | |