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/xenbus.h> 44 #include <xen/events.h> 45 #include <xen/interface/platform.h> 46 #include <asm/xen/hypervisor.h> 47 #include <asm/xen/hypercall.h> 48 49 50 /* 51 * @cpu_id: Xen physical cpu logic number 52 * @flags: Xen physical cpu status flag 53 * - XEN_PCPU_FLAGS_ONLINE: cpu is online 54 * - XEN_PCPU_FLAGS_INVALID: cpu is not present 55 */ 56 struct pcpu { 57 struct list_head list; 58 struct device dev; 59 uint32_t cpu_id; 60 uint32_t flags; 61 }; 62 63 static struct bus_type xen_pcpu_subsys = { 64 .name = "xen_cpu", 65 .dev_name = "xen_cpu", 66 }; 67 68 static DEFINE_MUTEX(xen_pcpu_lock); 69 70 static LIST_HEAD(xen_pcpus); 71 72 static int xen_pcpu_down(uint32_t cpu_id) 73 { 74 struct xen_platform_op op = { 75 .cmd = XENPF_cpu_offline, 76 .interface_version = XENPF_INTERFACE_VERSION, 77 .u.cpu_ol.cpuid = cpu_id, 78 }; 79 80 return HYPERVISOR_dom0_op(&op); 81 } 82 83 static int xen_pcpu_up(uint32_t cpu_id) 84 { 85 struct xen_platform_op op = { 86 .cmd = XENPF_cpu_online, 87 .interface_version = XENPF_INTERFACE_VERSION, 88 .u.cpu_ol.cpuid = cpu_id, 89 }; 90 91 return HYPERVISOR_dom0_op(&op); 92 } 93 94 static ssize_t show_online(struct device *dev, 95 struct device_attribute *attr, 96 char *buf) 97 { 98 struct pcpu *cpu = container_of(dev, struct pcpu, dev); 99 100 return sprintf(buf, "%u\n", !!(cpu->flags & XEN_PCPU_FLAGS_ONLINE)); 101 } 102 103 static ssize_t __ref store_online(struct device *dev, 104 struct device_attribute *attr, 105 const char *buf, size_t count) 106 { 107 struct pcpu *pcpu = container_of(dev, struct pcpu, dev); 108 unsigned long long val; 109 ssize_t ret; 110 111 if (!capable(CAP_SYS_ADMIN)) 112 return -EPERM; 113 114 if (kstrtoull(buf, 0, &val) < 0) 115 return -EINVAL; 116 117 switch (val) { 118 case 0: 119 ret = xen_pcpu_down(pcpu->cpu_id); 120 break; 121 case 1: 122 ret = xen_pcpu_up(pcpu->cpu_id); 123 break; 124 default: 125 ret = -EINVAL; 126 } 127 128 if (ret >= 0) 129 ret = count; 130 return ret; 131 } 132 static DEVICE_ATTR(online, S_IRUGO | S_IWUSR, show_online, store_online); 133 134 static bool xen_pcpu_online(uint32_t flags) 135 { 136 return !!(flags & XEN_PCPU_FLAGS_ONLINE); 137 } 138 139 static void pcpu_online_status(struct xenpf_pcpuinfo *info, 140 struct pcpu *pcpu) 141 { 142 if (xen_pcpu_online(info->flags) && 143 !xen_pcpu_online(pcpu->flags)) { 144 /* the pcpu is onlined */ 145 pcpu->flags |= XEN_PCPU_FLAGS_ONLINE; 146 kobject_uevent(&pcpu->dev.kobj, KOBJ_ONLINE); 147 } else if (!xen_pcpu_online(info->flags) && 148 xen_pcpu_online(pcpu->flags)) { 149 /* The pcpu is offlined */ 150 pcpu->flags &= ~XEN_PCPU_FLAGS_ONLINE; 151 kobject_uevent(&pcpu->dev.kobj, KOBJ_OFFLINE); 152 } 153 } 154 155 static struct pcpu *get_pcpu(uint32_t cpu_id) 156 { 157 struct pcpu *pcpu; 158 159 list_for_each_entry(pcpu, &xen_pcpus, list) { 160 if (pcpu->cpu_id == cpu_id) 161 return pcpu; 162 } 163 164 return NULL; 165 } 166 167 static void pcpu_release(struct device *dev) 168 { 169 struct pcpu *pcpu = container_of(dev, struct pcpu, dev); 170 171 list_del(&pcpu->list); 172 kfree(pcpu); 173 } 174 175 static void unregister_and_remove_pcpu(struct pcpu *pcpu) 176 { 177 struct device *dev; 178 179 if (!pcpu) 180 return; 181 182 dev = &pcpu->dev; 183 if (dev->id) 184 device_remove_file(dev, &dev_attr_online); 185 186 /* pcpu remove would be implicitly done */ 187 device_unregister(dev); 188 } 189 190 static int register_pcpu(struct pcpu *pcpu) 191 { 192 struct device *dev; 193 int err = -EINVAL; 194 195 if (!pcpu) 196 return err; 197 198 dev = &pcpu->dev; 199 dev->bus = &xen_pcpu_subsys; 200 dev->id = pcpu->cpu_id; 201 dev->release = pcpu_release; 202 203 err = device_register(dev); 204 if (err) { 205 pcpu_release(dev); 206 return err; 207 } 208 209 /* 210 * Xen never offline cpu0 due to several restrictions 211 * and assumptions. This basically doesn't add a sys control 212 * to user, one cannot attempt to offline BSP. 213 */ 214 if (dev->id) { 215 err = device_create_file(dev, &dev_attr_online); 216 if (err) { 217 device_unregister(dev); 218 return err; 219 } 220 } 221 222 return 0; 223 } 224 225 static struct pcpu *create_and_register_pcpu(struct xenpf_pcpuinfo *info) 226 { 227 struct pcpu *pcpu; 228 int err; 229 230 if (info->flags & XEN_PCPU_FLAGS_INVALID) 231 return ERR_PTR(-ENODEV); 232 233 pcpu = kzalloc(sizeof(struct pcpu), GFP_KERNEL); 234 if (!pcpu) 235 return ERR_PTR(-ENOMEM); 236 237 INIT_LIST_HEAD(&pcpu->list); 238 pcpu->cpu_id = info->xen_cpuid; 239 pcpu->flags = info->flags; 240 241 /* Need hold on xen_pcpu_lock before pcpu list manipulations */ 242 list_add_tail(&pcpu->list, &xen_pcpus); 243 244 err = register_pcpu(pcpu); 245 if (err) { 246 pr_warn("Failed to register pcpu%u\n", info->xen_cpuid); 247 return ERR_PTR(-ENOENT); 248 } 249 250 return pcpu; 251 } 252 253 /* 254 * Caller should hold the xen_pcpu_lock 255 */ 256 static int sync_pcpu(uint32_t cpu, uint32_t *max_cpu) 257 { 258 int ret; 259 struct pcpu *pcpu = NULL; 260 struct xenpf_pcpuinfo *info; 261 struct xen_platform_op op = { 262 .cmd = XENPF_get_cpuinfo, 263 .interface_version = XENPF_INTERFACE_VERSION, 264 .u.pcpu_info.xen_cpuid = cpu, 265 }; 266 267 ret = HYPERVISOR_dom0_op(&op); 268 if (ret) 269 return ret; 270 271 info = &op.u.pcpu_info; 272 if (max_cpu) 273 *max_cpu = info->max_present; 274 275 pcpu = get_pcpu(cpu); 276 277 /* 278 * Only those at cpu present map has its sys interface. 279 */ 280 if (info->flags & XEN_PCPU_FLAGS_INVALID) { 281 unregister_and_remove_pcpu(pcpu); 282 return 0; 283 } 284 285 if (!pcpu) { 286 pcpu = create_and_register_pcpu(info); 287 if (IS_ERR_OR_NULL(pcpu)) 288 return -ENODEV; 289 } else 290 pcpu_online_status(info, pcpu); 291 292 return 0; 293 } 294 295 /* 296 * Sync dom0's pcpu information with xen hypervisor's 297 */ 298 static int xen_sync_pcpus(void) 299 { 300 /* 301 * Boot cpu always have cpu_id 0 in xen 302 */ 303 uint32_t cpu = 0, max_cpu = 0; 304 int err = 0; 305 struct pcpu *pcpu, *tmp; 306 307 mutex_lock(&xen_pcpu_lock); 308 309 while (!err && (cpu <= max_cpu)) { 310 err = sync_pcpu(cpu, &max_cpu); 311 cpu++; 312 } 313 314 if (err) 315 list_for_each_entry_safe(pcpu, tmp, &xen_pcpus, list) 316 unregister_and_remove_pcpu(pcpu); 317 318 mutex_unlock(&xen_pcpu_lock); 319 320 return err; 321 } 322 323 static void xen_pcpu_work_fn(struct work_struct *work) 324 { 325 xen_sync_pcpus(); 326 } 327 static DECLARE_WORK(xen_pcpu_work, xen_pcpu_work_fn); 328 329 static irqreturn_t xen_pcpu_interrupt(int irq, void *dev_id) 330 { 331 schedule_work(&xen_pcpu_work); 332 return IRQ_HANDLED; 333 } 334 335 /* Sync with Xen hypervisor after cpu hotadded */ 336 void xen_pcpu_hotplug_sync(void) 337 { 338 schedule_work(&xen_pcpu_work); 339 } 340 EXPORT_SYMBOL_GPL(xen_pcpu_hotplug_sync); 341 342 /* 343 * For hypervisor presented cpu, return logic cpu id; 344 * For hypervisor non-presented cpu, return -ENODEV. 345 */ 346 int xen_pcpu_id(uint32_t acpi_id) 347 { 348 int cpu_id = 0, max_id = 0; 349 struct xen_platform_op op; 350 351 op.cmd = XENPF_get_cpuinfo; 352 while (cpu_id <= max_id) { 353 op.u.pcpu_info.xen_cpuid = cpu_id; 354 if (HYPERVISOR_dom0_op(&op)) { 355 cpu_id++; 356 continue; 357 } 358 359 if (acpi_id == op.u.pcpu_info.acpi_id) 360 return cpu_id; 361 if (op.u.pcpu_info.max_present > max_id) 362 max_id = op.u.pcpu_info.max_present; 363 cpu_id++; 364 } 365 366 return -ENODEV; 367 } 368 EXPORT_SYMBOL_GPL(xen_pcpu_id); 369 370 static int __init xen_pcpu_init(void) 371 { 372 int irq, ret; 373 374 if (!xen_initial_domain()) 375 return -ENODEV; 376 377 irq = bind_virq_to_irqhandler(VIRQ_PCPU_STATE, 0, 378 xen_pcpu_interrupt, 0, 379 "xen-pcpu", NULL); 380 if (irq < 0) { 381 pr_warn("Failed to bind pcpu virq\n"); 382 return irq; 383 } 384 385 ret = subsys_system_register(&xen_pcpu_subsys, NULL); 386 if (ret) { 387 pr_warn("Failed to register pcpu subsys\n"); 388 goto err1; 389 } 390 391 ret = xen_sync_pcpus(); 392 if (ret) { 393 pr_warn("Failed to sync pcpu info\n"); 394 goto err2; 395 } 396 397 return 0; 398 399 err2: 400 bus_unregister(&xen_pcpu_subsys); 401 err1: 402 unbind_from_irqhandler(irq, NULL); 403 return ret; 404 } 405 arch_initcall(xen_pcpu_init); 406