1 /* 2 * Copyright (c) 2009, Microsoft Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 * 13 * You should have received a copy of the GNU General Public License along with 14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple 15 * Place - Suite 330, Boston, MA 02111-1307 USA. 16 * 17 * Authors: 18 * Haiyang Zhang <haiyangz@microsoft.com> 19 * Hank Janssen <hjanssen@microsoft.com> 20 * 21 */ 22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 23 24 #include <linux/kernel.h> 25 #include <linux/mm.h> 26 #include <linux/slab.h> 27 #include <linux/vmalloc.h> 28 #include <linux/hyperv.h> 29 #include <linux/version.h> 30 #include <linux/random.h> 31 #include <linux/clockchips.h> 32 #include <asm/mshyperv.h> 33 #include "hyperv_vmbus.h" 34 35 /* The one and only */ 36 struct hv_context hv_context; 37 38 /* 39 * If false, we're using the old mechanism for stimer0 interrupts 40 * where it sends a VMbus message when it expires. The old 41 * mechanism is used when running on older versions of Hyper-V 42 * that don't support Direct Mode. While Hyper-V provides 43 * four stimer's per CPU, Linux uses only stimer0. 44 */ 45 static bool direct_mode_enabled; 46 static int stimer0_irq; 47 static int stimer0_vector; 48 49 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */ 50 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff 51 #define HV_MIN_DELTA_TICKS 1 52 53 /* 54 * hv_init - Main initialization routine. 55 * 56 * This routine must be called before any other routines in here are called 57 */ 58 int hv_init(void) 59 { 60 hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context); 61 if (!hv_context.cpu_context) 62 return -ENOMEM; 63 64 direct_mode_enabled = ms_hyperv.misc_features & 65 HV_STIMER_DIRECT_MODE_AVAILABLE; 66 return 0; 67 } 68 69 /* 70 * hv_post_message - Post a message using the hypervisor message IPC. 71 * 72 * This involves a hypercall. 73 */ 74 int hv_post_message(union hv_connection_id connection_id, 75 enum hv_message_type message_type, 76 void *payload, size_t payload_size) 77 { 78 struct hv_input_post_message *aligned_msg; 79 struct hv_per_cpu_context *hv_cpu; 80 u64 status; 81 82 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 83 return -EMSGSIZE; 84 85 hv_cpu = get_cpu_ptr(hv_context.cpu_context); 86 aligned_msg = hv_cpu->post_msg_page; 87 aligned_msg->connectionid = connection_id; 88 aligned_msg->reserved = 0; 89 aligned_msg->message_type = message_type; 90 aligned_msg->payload_size = payload_size; 91 memcpy((void *)aligned_msg->payload, payload, payload_size); 92 93 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL); 94 95 /* Preemption must remain disabled until after the hypercall 96 * so some other thread can't get scheduled onto this cpu and 97 * corrupt the per-cpu post_msg_page 98 */ 99 put_cpu_ptr(hv_cpu); 100 101 return status & 0xFFFF; 102 } 103 104 /* 105 * ISR for when stimer0 is operating in Direct Mode. Direct Mode 106 * does not use VMbus or any VMbus messages, so process here and not 107 * in the VMbus driver code. 108 */ 109 110 static void hv_stimer0_isr(void) 111 { 112 struct hv_per_cpu_context *hv_cpu; 113 114 hv_cpu = this_cpu_ptr(hv_context.cpu_context); 115 hv_cpu->clk_evt->event_handler(hv_cpu->clk_evt); 116 add_interrupt_randomness(stimer0_vector, 0); 117 } 118 119 static int hv_ce_set_next_event(unsigned long delta, 120 struct clock_event_device *evt) 121 { 122 u64 current_tick; 123 124 WARN_ON(!clockevent_state_oneshot(evt)); 125 126 current_tick = hyperv_cs->read(NULL); 127 current_tick += delta; 128 hv_init_timer(0, current_tick); 129 return 0; 130 } 131 132 static int hv_ce_shutdown(struct clock_event_device *evt) 133 { 134 hv_init_timer(0, 0); 135 hv_init_timer_config(0, 0); 136 if (direct_mode_enabled) 137 hv_disable_stimer0_percpu_irq(stimer0_irq); 138 139 return 0; 140 } 141 142 static int hv_ce_set_oneshot(struct clock_event_device *evt) 143 { 144 union hv_stimer_config timer_cfg; 145 146 timer_cfg.as_uint64 = 0; 147 timer_cfg.enable = 1; 148 timer_cfg.auto_enable = 1; 149 if (direct_mode_enabled) { 150 /* 151 * When it expires, the timer will directly interrupt 152 * on the specified hardware vector/IRQ. 153 */ 154 timer_cfg.direct_mode = 1; 155 timer_cfg.apic_vector = stimer0_vector; 156 hv_enable_stimer0_percpu_irq(stimer0_irq); 157 } else { 158 /* 159 * When it expires, the timer will generate a VMbus message, 160 * to be handled by the normal VMbus interrupt handler. 161 */ 162 timer_cfg.direct_mode = 0; 163 timer_cfg.sintx = VMBUS_MESSAGE_SINT; 164 } 165 hv_init_timer_config(0, timer_cfg.as_uint64); 166 return 0; 167 } 168 169 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu) 170 { 171 dev->name = "Hyper-V clockevent"; 172 dev->features = CLOCK_EVT_FEAT_ONESHOT; 173 dev->cpumask = cpumask_of(cpu); 174 dev->rating = 1000; 175 /* 176 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will 177 * result in clockevents_config_and_register() taking additional 178 * references to the hv_vmbus module making it impossible to unload. 179 */ 180 181 dev->set_state_shutdown = hv_ce_shutdown; 182 dev->set_state_oneshot = hv_ce_set_oneshot; 183 dev->set_next_event = hv_ce_set_next_event; 184 } 185 186 187 int hv_synic_alloc(void) 188 { 189 int cpu; 190 struct hv_per_cpu_context *hv_cpu; 191 192 /* 193 * First, zero all per-cpu memory areas so hv_synic_free() can 194 * detect what memory has been allocated and cleanup properly 195 * after any failures. 196 */ 197 for_each_present_cpu(cpu) { 198 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu); 199 memset(hv_cpu, 0, sizeof(*hv_cpu)); 200 } 201 202 hv_context.hv_numa_map = kcalloc(nr_node_ids, sizeof(struct cpumask), 203 GFP_KERNEL); 204 if (hv_context.hv_numa_map == NULL) { 205 pr_err("Unable to allocate NUMA map\n"); 206 goto err; 207 } 208 209 for_each_present_cpu(cpu) { 210 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu); 211 212 tasklet_init(&hv_cpu->msg_dpc, 213 vmbus_on_msg_dpc, (unsigned long) hv_cpu); 214 215 hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device), 216 GFP_KERNEL); 217 if (hv_cpu->clk_evt == NULL) { 218 pr_err("Unable to allocate clock event device\n"); 219 goto err; 220 } 221 hv_init_clockevent_device(hv_cpu->clk_evt, cpu); 222 223 hv_cpu->synic_message_page = 224 (void *)get_zeroed_page(GFP_ATOMIC); 225 if (hv_cpu->synic_message_page == NULL) { 226 pr_err("Unable to allocate SYNIC message page\n"); 227 goto err; 228 } 229 230 hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC); 231 if (hv_cpu->synic_event_page == NULL) { 232 pr_err("Unable to allocate SYNIC event page\n"); 233 goto err; 234 } 235 236 hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC); 237 if (hv_cpu->post_msg_page == NULL) { 238 pr_err("Unable to allocate post msg page\n"); 239 goto err; 240 } 241 242 INIT_LIST_HEAD(&hv_cpu->chan_list); 243 } 244 245 if (direct_mode_enabled && 246 hv_setup_stimer0_irq(&stimer0_irq, &stimer0_vector, 247 hv_stimer0_isr)) 248 goto err; 249 250 return 0; 251 err: 252 /* 253 * Any memory allocations that succeeded will be freed when 254 * the caller cleans up by calling hv_synic_free() 255 */ 256 return -ENOMEM; 257 } 258 259 260 void hv_synic_free(void) 261 { 262 int cpu; 263 264 for_each_present_cpu(cpu) { 265 struct hv_per_cpu_context *hv_cpu 266 = per_cpu_ptr(hv_context.cpu_context, cpu); 267 268 kfree(hv_cpu->clk_evt); 269 free_page((unsigned long)hv_cpu->synic_event_page); 270 free_page((unsigned long)hv_cpu->synic_message_page); 271 free_page((unsigned long)hv_cpu->post_msg_page); 272 } 273 274 kfree(hv_context.hv_numa_map); 275 } 276 277 /* 278 * hv_synic_init - Initialize the Synthetic Interrupt Controller. 279 * 280 * If it is already initialized by another entity (ie x2v shim), we need to 281 * retrieve the initialized message and event pages. Otherwise, we create and 282 * initialize the message and event pages. 283 */ 284 int hv_synic_init(unsigned int cpu) 285 { 286 struct hv_per_cpu_context *hv_cpu 287 = per_cpu_ptr(hv_context.cpu_context, cpu); 288 union hv_synic_simp simp; 289 union hv_synic_siefp siefp; 290 union hv_synic_sint shared_sint; 291 union hv_synic_scontrol sctrl; 292 293 /* Setup the Synic's message page */ 294 hv_get_simp(simp.as_uint64); 295 simp.simp_enabled = 1; 296 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page) 297 >> PAGE_SHIFT; 298 299 hv_set_simp(simp.as_uint64); 300 301 /* Setup the Synic's event page */ 302 hv_get_siefp(siefp.as_uint64); 303 siefp.siefp_enabled = 1; 304 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page) 305 >> PAGE_SHIFT; 306 307 hv_set_siefp(siefp.as_uint64); 308 309 /* Setup the shared SINT. */ 310 hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 311 312 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR; 313 shared_sint.masked = false; 314 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED) 315 shared_sint.auto_eoi = false; 316 else 317 shared_sint.auto_eoi = true; 318 319 hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 320 321 /* Enable the global synic bit */ 322 hv_get_synic_state(sctrl.as_uint64); 323 sctrl.enable = 1; 324 325 hv_set_synic_state(sctrl.as_uint64); 326 327 /* 328 * Register the per-cpu clockevent source. 329 */ 330 if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) 331 clockevents_config_and_register(hv_cpu->clk_evt, 332 HV_TIMER_FREQUENCY, 333 HV_MIN_DELTA_TICKS, 334 HV_MAX_MAX_DELTA_TICKS); 335 return 0; 336 } 337 338 /* 339 * hv_synic_clockevents_cleanup - Cleanup clockevent devices 340 */ 341 void hv_synic_clockevents_cleanup(void) 342 { 343 int cpu; 344 345 if (!(ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE)) 346 return; 347 348 if (direct_mode_enabled) 349 hv_remove_stimer0_irq(stimer0_irq); 350 351 for_each_present_cpu(cpu) { 352 struct hv_per_cpu_context *hv_cpu 353 = per_cpu_ptr(hv_context.cpu_context, cpu); 354 355 clockevents_unbind_device(hv_cpu->clk_evt, cpu); 356 } 357 } 358 359 /* 360 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 361 */ 362 int hv_synic_cleanup(unsigned int cpu) 363 { 364 union hv_synic_sint shared_sint; 365 union hv_synic_simp simp; 366 union hv_synic_siefp siefp; 367 union hv_synic_scontrol sctrl; 368 struct vmbus_channel *channel, *sc; 369 bool channel_found = false; 370 unsigned long flags; 371 372 hv_get_synic_state(sctrl.as_uint64); 373 if (sctrl.enable != 1) 374 return -EFAULT; 375 376 /* 377 * Search for channels which are bound to the CPU we're about to 378 * cleanup. In case we find one and vmbus is still connected we need to 379 * fail, this will effectively prevent CPU offlining. There is no way 380 * we can re-bind channels to different CPUs for now. 381 */ 382 mutex_lock(&vmbus_connection.channel_mutex); 383 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 384 if (channel->target_cpu == cpu) { 385 channel_found = true; 386 break; 387 } 388 spin_lock_irqsave(&channel->lock, flags); 389 list_for_each_entry(sc, &channel->sc_list, sc_list) { 390 if (sc->target_cpu == cpu) { 391 channel_found = true; 392 break; 393 } 394 } 395 spin_unlock_irqrestore(&channel->lock, flags); 396 if (channel_found) 397 break; 398 } 399 mutex_unlock(&vmbus_connection.channel_mutex); 400 401 if (channel_found && vmbus_connection.conn_state == CONNECTED) 402 return -EBUSY; 403 404 /* Turn off clockevent device */ 405 if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) { 406 struct hv_per_cpu_context *hv_cpu 407 = this_cpu_ptr(hv_context.cpu_context); 408 409 clockevents_unbind_device(hv_cpu->clk_evt, cpu); 410 hv_ce_shutdown(hv_cpu->clk_evt); 411 } 412 413 hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 414 415 shared_sint.masked = 1; 416 417 /* Need to correctly cleanup in the case of SMP!!! */ 418 /* Disable the interrupt */ 419 hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64); 420 421 hv_get_simp(simp.as_uint64); 422 simp.simp_enabled = 0; 423 simp.base_simp_gpa = 0; 424 425 hv_set_simp(simp.as_uint64); 426 427 hv_get_siefp(siefp.as_uint64); 428 siefp.siefp_enabled = 0; 429 siefp.base_siefp_gpa = 0; 430 431 hv_set_siefp(siefp.as_uint64); 432 433 /* Disable the global synic bit */ 434 sctrl.enable = 0; 435 hv_set_synic_state(sctrl.as_uint64); 436 437 return 0; 438 } 439