Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
at v3.7-rc4 371 lines 8.4 kB view raw
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#include <linux/interrupt.h> 35#include <linux/spinlock.h> 36#include <linux/cpu.h> 37#include <linux/stat.h> 38#include <linux/capability.h> 39 40#include <xen/xen.h> 41#include <xen/xenbus.h> 42#include <xen/events.h> 43#include <xen/interface/platform.h> 44#include <asm/xen/hypervisor.h> 45#include <asm/xen/hypercall.h> 46 47#define XEN_PCPU "xen_cpu: " 48 49/* 50 * @cpu_id: Xen physical cpu logic number 51 * @flags: Xen physical cpu status flag 52 * - XEN_PCPU_FLAGS_ONLINE: cpu is online 53 * - XEN_PCPU_FLAGS_INVALID: cpu is not present 54 */ 55struct pcpu { 56 struct list_head list; 57 struct device dev; 58 uint32_t cpu_id; 59 uint32_t flags; 60}; 61 62static struct bus_type xen_pcpu_subsys = { 63 .name = "xen_cpu", 64 .dev_name = "xen_cpu", 65}; 66 67static DEFINE_MUTEX(xen_pcpu_lock); 68 69static LIST_HEAD(xen_pcpus); 70 71static int xen_pcpu_down(uint32_t cpu_id) 72{ 73 struct xen_platform_op op = { 74 .cmd = XENPF_cpu_offline, 75 .interface_version = XENPF_INTERFACE_VERSION, 76 .u.cpu_ol.cpuid = cpu_id, 77 }; 78 79 return HYPERVISOR_dom0_op(&op); 80} 81 82static int xen_pcpu_up(uint32_t cpu_id) 83{ 84 struct xen_platform_op op = { 85 .cmd = XENPF_cpu_online, 86 .interface_version = XENPF_INTERFACE_VERSION, 87 .u.cpu_ol.cpuid = cpu_id, 88 }; 89 90 return HYPERVISOR_dom0_op(&op); 91} 92 93static ssize_t show_online(struct device *dev, 94 struct device_attribute *attr, 95 char *buf) 96{ 97 struct pcpu *cpu = container_of(dev, struct pcpu, dev); 98 99 return sprintf(buf, "%u\n", !!(cpu->flags & XEN_PCPU_FLAGS_ONLINE)); 100} 101 102static ssize_t __ref store_online(struct device *dev, 103 struct device_attribute *attr, 104 const char *buf, size_t count) 105{ 106 struct pcpu *pcpu = container_of(dev, struct pcpu, dev); 107 unsigned long long val; 108 ssize_t ret; 109 110 if (!capable(CAP_SYS_ADMIN)) 111 return -EPERM; 112 113 if (kstrtoull(buf, 0, &val) < 0) 114 return -EINVAL; 115 116 switch (val) { 117 case 0: 118 ret = xen_pcpu_down(pcpu->cpu_id); 119 break; 120 case 1: 121 ret = xen_pcpu_up(pcpu->cpu_id); 122 break; 123 default: 124 ret = -EINVAL; 125 } 126 127 if (ret >= 0) 128 ret = count; 129 return ret; 130} 131static DEVICE_ATTR(online, S_IRUGO | S_IWUSR, show_online, store_online); 132 133static bool xen_pcpu_online(uint32_t flags) 134{ 135 return !!(flags & XEN_PCPU_FLAGS_ONLINE); 136} 137 138static void pcpu_online_status(struct xenpf_pcpuinfo *info, 139 struct pcpu *pcpu) 140{ 141 if (xen_pcpu_online(info->flags) && 142 !xen_pcpu_online(pcpu->flags)) { 143 /* the pcpu is onlined */ 144 pcpu->flags |= XEN_PCPU_FLAGS_ONLINE; 145 kobject_uevent(&pcpu->dev.kobj, KOBJ_ONLINE); 146 } else if (!xen_pcpu_online(info->flags) && 147 xen_pcpu_online(pcpu->flags)) { 148 /* The pcpu is offlined */ 149 pcpu->flags &= ~XEN_PCPU_FLAGS_ONLINE; 150 kobject_uevent(&pcpu->dev.kobj, KOBJ_OFFLINE); 151 } 152} 153 154static struct pcpu *get_pcpu(uint32_t cpu_id) 155{ 156 struct pcpu *pcpu; 157 158 list_for_each_entry(pcpu, &xen_pcpus, list) { 159 if (pcpu->cpu_id == cpu_id) 160 return pcpu; 161 } 162 163 return NULL; 164} 165 166static void pcpu_release(struct device *dev) 167{ 168 struct pcpu *pcpu = container_of(dev, struct pcpu, dev); 169 170 list_del(&pcpu->list); 171 kfree(pcpu); 172} 173 174static void unregister_and_remove_pcpu(struct pcpu *pcpu) 175{ 176 struct device *dev; 177 178 if (!pcpu) 179 return; 180 181 dev = &pcpu->dev; 182 if (dev->id) 183 device_remove_file(dev, &dev_attr_online); 184 185 /* pcpu remove would be implicitly done */ 186 device_unregister(dev); 187} 188 189static int register_pcpu(struct pcpu *pcpu) 190{ 191 struct device *dev; 192 int err = -EINVAL; 193 194 if (!pcpu) 195 return err; 196 197 dev = &pcpu->dev; 198 dev->bus = &xen_pcpu_subsys; 199 dev->id = pcpu->cpu_id; 200 dev->release = pcpu_release; 201 202 err = device_register(dev); 203 if (err) { 204 pcpu_release(dev); 205 return err; 206 } 207 208 /* 209 * Xen never offline cpu0 due to several restrictions 210 * and assumptions. This basically doesn't add a sys control 211 * to user, one cannot attempt to offline BSP. 212 */ 213 if (dev->id) { 214 err = device_create_file(dev, &dev_attr_online); 215 if (err) { 216 device_unregister(dev); 217 return err; 218 } 219 } 220 221 return 0; 222} 223 224static struct pcpu *create_and_register_pcpu(struct xenpf_pcpuinfo *info) 225{ 226 struct pcpu *pcpu; 227 int err; 228 229 if (info->flags & XEN_PCPU_FLAGS_INVALID) 230 return ERR_PTR(-ENODEV); 231 232 pcpu = kzalloc(sizeof(struct pcpu), GFP_KERNEL); 233 if (!pcpu) 234 return ERR_PTR(-ENOMEM); 235 236 INIT_LIST_HEAD(&pcpu->list); 237 pcpu->cpu_id = info->xen_cpuid; 238 pcpu->flags = info->flags; 239 240 /* Need hold on xen_pcpu_lock before pcpu list manipulations */ 241 list_add_tail(&pcpu->list, &xen_pcpus); 242 243 err = register_pcpu(pcpu); 244 if (err) { 245 pr_warning(XEN_PCPU "Failed to register pcpu%u\n", 246 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 */ 256static 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 if (pcpu) 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 */ 299static 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 324static void xen_pcpu_work_fn(struct work_struct *work) 325{ 326 xen_sync_pcpus(); 327} 328static DECLARE_WORK(xen_pcpu_work, xen_pcpu_work_fn); 329 330static irqreturn_t xen_pcpu_interrupt(int irq, void *dev_id) 331{ 332 schedule_work(&xen_pcpu_work); 333 return IRQ_HANDLED; 334} 335 336static int __init xen_pcpu_init(void) 337{ 338 int irq, ret; 339 340 if (!xen_initial_domain()) 341 return -ENODEV; 342 343 irq = bind_virq_to_irqhandler(VIRQ_PCPU_STATE, 0, 344 xen_pcpu_interrupt, 0, 345 "xen-pcpu", NULL); 346 if (irq < 0) { 347 pr_warning(XEN_PCPU "Failed to bind pcpu virq\n"); 348 return irq; 349 } 350 351 ret = subsys_system_register(&xen_pcpu_subsys, NULL); 352 if (ret) { 353 pr_warning(XEN_PCPU "Failed to register pcpu subsys\n"); 354 goto err1; 355 } 356 357 ret = xen_sync_pcpus(); 358 if (ret) { 359 pr_warning(XEN_PCPU "Failed to sync pcpu info\n"); 360 goto err2; 361 } 362 363 return 0; 364 365err2: 366 bus_unregister(&xen_pcpu_subsys); 367err1: 368 unbind_from_irqhandler(irq, NULL); 369 return ret; 370} 371arch_initcall(xen_pcpu_init);