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 flags; 62 }; 63 64 static struct bus_type xen_pcpu_subsys = { 65 .name = "xen_cpu", 66 .dev_name = "xen_cpu", 67 }; 68 69 static DEFINE_MUTEX(xen_pcpu_lock); 70 71 static LIST_HEAD(xen_pcpus); 72 73 static int xen_pcpu_down(uint32_t cpu_id) 74 { 75 struct xen_platform_op op = { 76 .cmd = XENPF_cpu_offline, 77 .interface_version = XENPF_INTERFACE_VERSION, 78 .u.cpu_ol.cpuid = cpu_id, 79 }; 80 81 return HYPERVISOR_dom0_op(&op); 82 } 83 84 static int xen_pcpu_up(uint32_t cpu_id) 85 { 86 struct xen_platform_op op = { 87 .cmd = XENPF_cpu_online, 88 .interface_version = XENPF_INTERFACE_VERSION, 89 .u.cpu_ol.cpuid = cpu_id, 90 }; 91 92 return HYPERVISOR_dom0_op(&op); 93 } 94 95 static ssize_t show_online(struct device *dev, 96 struct device_attribute *attr, 97 char *buf) 98 { 99 struct pcpu *cpu = container_of(dev, struct pcpu, dev); 100 101 return sprintf(buf, "%u\n", !!(cpu->flags & XEN_PCPU_FLAGS_ONLINE)); 102 } 103 104 static ssize_t __ref store_online(struct device *dev, 105 struct device_attribute *attr, 106 const char *buf, size_t count) 107 { 108 struct pcpu *pcpu = container_of(dev, struct pcpu, dev); 109 unsigned long long val; 110 ssize_t ret; 111 112 if (!capable(CAP_SYS_ADMIN)) 113 return -EPERM; 114 115 if (kstrtoull(buf, 0, &val) < 0) 116 return -EINVAL; 117 118 switch (val) { 119 case 0: 120 ret = xen_pcpu_down(pcpu->cpu_id); 121 break; 122 case 1: 123 ret = xen_pcpu_up(pcpu->cpu_id); 124 break; 125 default: 126 ret = -EINVAL; 127 } 128 129 if (ret >= 0) 130 ret = count; 131 return ret; 132 } 133 static DEVICE_ATTR(online, S_IRUGO | S_IWUSR, show_online, store_online); 134 135 static bool xen_pcpu_online(uint32_t flags) 136 { 137 return !!(flags & XEN_PCPU_FLAGS_ONLINE); 138 } 139 140 static void pcpu_online_status(struct xenpf_pcpuinfo *info, 141 struct pcpu *pcpu) 142 { 143 if (xen_pcpu_online(info->flags) && 144 !xen_pcpu_online(pcpu->flags)) { 145 /* the pcpu is onlined */ 146 pcpu->flags |= XEN_PCPU_FLAGS_ONLINE; 147 kobject_uevent(&pcpu->dev.kobj, KOBJ_ONLINE); 148 } else if (!xen_pcpu_online(info->flags) && 149 xen_pcpu_online(pcpu->flags)) { 150 /* The pcpu is offlined */ 151 pcpu->flags &= ~XEN_PCPU_FLAGS_ONLINE; 152 kobject_uevent(&pcpu->dev.kobj, KOBJ_OFFLINE); 153 } 154 } 155 156 static struct pcpu *get_pcpu(uint32_t cpu_id) 157 { 158 struct pcpu *pcpu; 159 160 list_for_each_entry(pcpu, &xen_pcpus, list) { 161 if (pcpu->cpu_id == cpu_id) 162 return pcpu; 163 } 164 165 return NULL; 166 } 167 168 static void pcpu_release(struct device *dev) 169 { 170 struct pcpu *pcpu = container_of(dev, struct pcpu, dev); 171 172 list_del(&pcpu->list); 173 kfree(pcpu); 174 } 175 176 static void unregister_and_remove_pcpu(struct pcpu *pcpu) 177 { 178 struct device *dev; 179 180 if (!pcpu) 181 return; 182 183 dev = &pcpu->dev; 184 if (dev->id) 185 device_remove_file(dev, &dev_attr_online); 186 187 /* pcpu remove would be implicitly done */ 188 device_unregister(dev); 189 } 190 191 static int register_pcpu(struct pcpu *pcpu) 192 { 193 struct device *dev; 194 int err = -EINVAL; 195 196 if (!pcpu) 197 return err; 198 199 dev = &pcpu->dev; 200 dev->bus = &xen_pcpu_subsys; 201 dev->id = pcpu->cpu_id; 202 dev->release = pcpu_release; 203 204 err = device_register(dev); 205 if (err) { 206 pcpu_release(dev); 207 return err; 208 } 209 210 /* 211 * Xen never offline cpu0 due to several restrictions 212 * and assumptions. This basically doesn't add a sys control 213 * to user, one cannot attempt to offline BSP. 214 */ 215 if (dev->id) { 216 err = device_create_file(dev, &dev_attr_online); 217 if (err) { 218 device_unregister(dev); 219 return err; 220 } 221 } 222 223 return 0; 224 } 225 226 static struct pcpu *create_and_register_pcpu(struct xenpf_pcpuinfo *info) 227 { 228 struct pcpu *pcpu; 229 int err; 230 231 if (info->flags & XEN_PCPU_FLAGS_INVALID) 232 return ERR_PTR(-ENODEV); 233 234 pcpu = kzalloc(sizeof(struct pcpu), GFP_KERNEL); 235 if (!pcpu) 236 return ERR_PTR(-ENOMEM); 237 238 INIT_LIST_HEAD(&pcpu->list); 239 pcpu->cpu_id = info->xen_cpuid; 240 pcpu->flags = info->flags; 241 242 /* Need hold on xen_pcpu_lock before pcpu list manipulations */ 243 list_add_tail(&pcpu->list, &xen_pcpus); 244 245 err = register_pcpu(pcpu); 246 if (err) { 247 pr_warn("Failed to register pcpu%u\n", info->xen_cpuid); 248 return ERR_PTR(-ENOENT); 249 } 250 251 return pcpu; 252 } 253 254 /* 255 * Caller should hold the xen_pcpu_lock 256 */ 257 static int sync_pcpu(uint32_t cpu, uint32_t *max_cpu) 258 { 259 int ret; 260 struct pcpu *pcpu = NULL; 261 struct xenpf_pcpuinfo *info; 262 struct xen_platform_op op = { 263 .cmd = XENPF_get_cpuinfo, 264 .interface_version = XENPF_INTERFACE_VERSION, 265 .u.pcpu_info.xen_cpuid = cpu, 266 }; 267 268 ret = HYPERVISOR_dom0_op(&op); 269 if (ret) 270 return ret; 271 272 info = &op.u.pcpu_info; 273 if (max_cpu) 274 *max_cpu = info->max_present; 275 276 pcpu = get_pcpu(cpu); 277 278 /* 279 * Only those at cpu present map has its sys interface. 280 */ 281 if (info->flags & XEN_PCPU_FLAGS_INVALID) { 282 unregister_and_remove_pcpu(pcpu); 283 return 0; 284 } 285 286 if (!pcpu) { 287 pcpu = create_and_register_pcpu(info); 288 if (IS_ERR_OR_NULL(pcpu)) 289 return -ENODEV; 290 } else 291 pcpu_online_status(info, pcpu); 292 293 return 0; 294 } 295 296 /* 297 * Sync dom0's pcpu information with xen hypervisor's 298 */ 299 static int xen_sync_pcpus(void) 300 { 301 /* 302 * Boot cpu always have cpu_id 0 in xen 303 */ 304 uint32_t cpu = 0, max_cpu = 0; 305 int err = 0; 306 struct pcpu *pcpu, *tmp; 307 308 mutex_lock(&xen_pcpu_lock); 309 310 while (!err && (cpu <= max_cpu)) { 311 err = sync_pcpu(cpu, &max_cpu); 312 cpu++; 313 } 314 315 if (err) 316 list_for_each_entry_safe(pcpu, tmp, &xen_pcpus, list) 317 unregister_and_remove_pcpu(pcpu); 318 319 mutex_unlock(&xen_pcpu_lock); 320 321 return err; 322 } 323 324 static void xen_pcpu_work_fn(struct work_struct *work) 325 { 326 xen_sync_pcpus(); 327 } 328 static DECLARE_WORK(xen_pcpu_work, xen_pcpu_work_fn); 329 330 static irqreturn_t xen_pcpu_interrupt(int irq, void *dev_id) 331 { 332 schedule_work(&xen_pcpu_work); 333 return IRQ_HANDLED; 334 } 335 336 /* Sync with Xen hypervisor after cpu hotadded */ 337 void xen_pcpu_hotplug_sync(void) 338 { 339 schedule_work(&xen_pcpu_work); 340 } 341 EXPORT_SYMBOL_GPL(xen_pcpu_hotplug_sync); 342 343 /* 344 * For hypervisor presented cpu, return logic cpu id; 345 * For hypervisor non-presented cpu, return -ENODEV. 346 */ 347 int xen_pcpu_id(uint32_t acpi_id) 348 { 349 int cpu_id = 0, max_id = 0; 350 struct xen_platform_op op; 351 352 op.cmd = XENPF_get_cpuinfo; 353 while (cpu_id <= max_id) { 354 op.u.pcpu_info.xen_cpuid = cpu_id; 355 if (HYPERVISOR_dom0_op(&op)) { 356 cpu_id++; 357 continue; 358 } 359 360 if (acpi_id == op.u.pcpu_info.acpi_id) 361 return cpu_id; 362 if (op.u.pcpu_info.max_present > max_id) 363 max_id = op.u.pcpu_info.max_present; 364 cpu_id++; 365 } 366 367 return -ENODEV; 368 } 369 EXPORT_SYMBOL_GPL(xen_pcpu_id); 370 371 static int __init xen_pcpu_init(void) 372 { 373 int irq, ret; 374 375 if (!xen_initial_domain()) 376 return -ENODEV; 377 378 irq = bind_virq_to_irqhandler(VIRQ_PCPU_STATE, 0, 379 xen_pcpu_interrupt, 0, 380 "xen-pcpu", NULL); 381 if (irq < 0) { 382 pr_warn("Failed to bind pcpu virq\n"); 383 return irq; 384 } 385 386 ret = subsys_system_register(&xen_pcpu_subsys, NULL); 387 if (ret) { 388 pr_warn("Failed to register pcpu subsys\n"); 389 goto err1; 390 } 391 392 ret = xen_sync_pcpus(); 393 if (ret) { 394 pr_warn("Failed to sync pcpu info\n"); 395 goto err2; 396 } 397 398 return 0; 399 400 err2: 401 bus_unregister(&xen_pcpu_subsys); 402 err1: 403 unbind_from_irqhandler(irq, NULL); 404 return ret; 405 } 406 arch_initcall(xen_pcpu_init); 407