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/interrupt.h> 31 #include <linux/clockchips.h> 32 #include <asm/hyperv.h> 33 #include <asm/mshyperv.h> 34 #include "hyperv_vmbus.h" 35 36 /* The one and only */ 37 struct hv_context hv_context = { 38 .synic_initialized = false, 39 }; 40 41 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */ 42 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff 43 #define HV_MIN_DELTA_TICKS 1 44 45 /* 46 * hv_init - Main initialization routine. 47 * 48 * This routine must be called before any other routines in here are called 49 */ 50 int hv_init(void) 51 { 52 if (!hv_is_hypercall_page_setup()) 53 return -ENOTSUPP; 54 55 hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context); 56 if (!hv_context.cpu_context) 57 return -ENOMEM; 58 59 return 0; 60 } 61 62 /* 63 * hv_post_message - Post a message using the hypervisor message IPC. 64 * 65 * This involves a hypercall. 66 */ 67 int hv_post_message(union hv_connection_id connection_id, 68 enum hv_message_type message_type, 69 void *payload, size_t payload_size) 70 { 71 struct hv_input_post_message *aligned_msg; 72 struct hv_per_cpu_context *hv_cpu; 73 u64 status; 74 75 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) 76 return -EMSGSIZE; 77 78 hv_cpu = get_cpu_ptr(hv_context.cpu_context); 79 aligned_msg = hv_cpu->post_msg_page; 80 aligned_msg->connectionid = connection_id; 81 aligned_msg->reserved = 0; 82 aligned_msg->message_type = message_type; 83 aligned_msg->payload_size = payload_size; 84 memcpy((void *)aligned_msg->payload, payload, payload_size); 85 86 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL); 87 88 /* Preemption must remain disabled until after the hypercall 89 * so some other thread can't get scheduled onto this cpu and 90 * corrupt the per-cpu post_msg_page 91 */ 92 put_cpu_ptr(hv_cpu); 93 94 return status & 0xFFFF; 95 } 96 97 static int hv_ce_set_next_event(unsigned long delta, 98 struct clock_event_device *evt) 99 { 100 u64 current_tick; 101 102 WARN_ON(!clockevent_state_oneshot(evt)); 103 104 current_tick = hyperv_cs->read(NULL); 105 current_tick += delta; 106 hv_init_timer(HV_X64_MSR_STIMER0_COUNT, current_tick); 107 return 0; 108 } 109 110 static int hv_ce_shutdown(struct clock_event_device *evt) 111 { 112 hv_init_timer(HV_X64_MSR_STIMER0_COUNT, 0); 113 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, 0); 114 115 return 0; 116 } 117 118 static int hv_ce_set_oneshot(struct clock_event_device *evt) 119 { 120 union hv_timer_config timer_cfg; 121 122 timer_cfg.enable = 1; 123 timer_cfg.auto_enable = 1; 124 timer_cfg.sintx = VMBUS_MESSAGE_SINT; 125 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64); 126 127 return 0; 128 } 129 130 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu) 131 { 132 dev->name = "Hyper-V clockevent"; 133 dev->features = CLOCK_EVT_FEAT_ONESHOT; 134 dev->cpumask = cpumask_of(cpu); 135 dev->rating = 1000; 136 /* 137 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will 138 * result in clockevents_config_and_register() taking additional 139 * references to the hv_vmbus module making it impossible to unload. 140 */ 141 142 dev->set_state_shutdown = hv_ce_shutdown; 143 dev->set_state_oneshot = hv_ce_set_oneshot; 144 dev->set_next_event = hv_ce_set_next_event; 145 } 146 147 148 int hv_synic_alloc(void) 149 { 150 int cpu; 151 152 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids, 153 GFP_ATOMIC); 154 if (hv_context.hv_numa_map == NULL) { 155 pr_err("Unable to allocate NUMA map\n"); 156 goto err; 157 } 158 159 for_each_present_cpu(cpu) { 160 struct hv_per_cpu_context *hv_cpu 161 = per_cpu_ptr(hv_context.cpu_context, cpu); 162 163 memset(hv_cpu, 0, sizeof(*hv_cpu)); 164 tasklet_init(&hv_cpu->msg_dpc, 165 vmbus_on_msg_dpc, (unsigned long) hv_cpu); 166 167 hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device), 168 GFP_KERNEL); 169 if (hv_cpu->clk_evt == NULL) { 170 pr_err("Unable to allocate clock event device\n"); 171 goto err; 172 } 173 hv_init_clockevent_device(hv_cpu->clk_evt, cpu); 174 175 hv_cpu->synic_message_page = 176 (void *)get_zeroed_page(GFP_ATOMIC); 177 if (hv_cpu->synic_message_page == NULL) { 178 pr_err("Unable to allocate SYNIC message page\n"); 179 goto err; 180 } 181 182 hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC); 183 if (hv_cpu->synic_event_page == NULL) { 184 pr_err("Unable to allocate SYNIC event page\n"); 185 goto err; 186 } 187 188 hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC); 189 if (hv_cpu->post_msg_page == NULL) { 190 pr_err("Unable to allocate post msg page\n"); 191 goto err; 192 } 193 194 INIT_LIST_HEAD(&hv_cpu->chan_list); 195 } 196 197 return 0; 198 err: 199 return -ENOMEM; 200 } 201 202 203 void hv_synic_free(void) 204 { 205 int cpu; 206 207 for_each_present_cpu(cpu) { 208 struct hv_per_cpu_context *hv_cpu 209 = per_cpu_ptr(hv_context.cpu_context, cpu); 210 211 if (hv_cpu->synic_event_page) 212 free_page((unsigned long)hv_cpu->synic_event_page); 213 if (hv_cpu->synic_message_page) 214 free_page((unsigned long)hv_cpu->synic_message_page); 215 if (hv_cpu->post_msg_page) 216 free_page((unsigned long)hv_cpu->post_msg_page); 217 } 218 219 kfree(hv_context.hv_numa_map); 220 } 221 222 /* 223 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 224 * 225 * If it is already initialized by another entity (ie x2v shim), we need to 226 * retrieve the initialized message and event pages. Otherwise, we create and 227 * initialize the message and event pages. 228 */ 229 int hv_synic_init(unsigned int cpu) 230 { 231 struct hv_per_cpu_context *hv_cpu 232 = per_cpu_ptr(hv_context.cpu_context, cpu); 233 union hv_synic_simp simp; 234 union hv_synic_siefp siefp; 235 union hv_synic_sint shared_sint; 236 union hv_synic_scontrol sctrl; 237 238 /* Setup the Synic's message page */ 239 hv_get_simp(simp.as_uint64); 240 simp.simp_enabled = 1; 241 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page) 242 >> PAGE_SHIFT; 243 244 hv_set_simp(simp.as_uint64); 245 246 /* Setup the Synic's event page */ 247 hv_get_siefp(siefp.as_uint64); 248 siefp.siefp_enabled = 1; 249 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page) 250 >> PAGE_SHIFT; 251 252 hv_set_siefp(siefp.as_uint64); 253 254 /* Setup the shared SINT. */ 255 hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 256 shared_sint.as_uint64); 257 258 shared_sint.as_uint64 = 0; 259 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR; 260 shared_sint.masked = false; 261 if (ms_hyperv.hints & HV_X64_DEPRECATING_AEOI_RECOMMENDED) 262 shared_sint.auto_eoi = false; 263 else 264 shared_sint.auto_eoi = true; 265 266 hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 267 shared_sint.as_uint64); 268 269 /* Enable the global synic bit */ 270 hv_get_synic_state(sctrl.as_uint64); 271 sctrl.enable = 1; 272 273 hv_set_synic_state(sctrl.as_uint64); 274 275 hv_context.synic_initialized = true; 276 277 /* 278 * Register the per-cpu clockevent source. 279 */ 280 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) 281 clockevents_config_and_register(hv_cpu->clk_evt, 282 HV_TIMER_FREQUENCY, 283 HV_MIN_DELTA_TICKS, 284 HV_MAX_MAX_DELTA_TICKS); 285 return 0; 286 } 287 288 /* 289 * hv_synic_clockevents_cleanup - Cleanup clockevent devices 290 */ 291 void hv_synic_clockevents_cleanup(void) 292 { 293 int cpu; 294 295 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)) 296 return; 297 298 for_each_present_cpu(cpu) { 299 struct hv_per_cpu_context *hv_cpu 300 = per_cpu_ptr(hv_context.cpu_context, cpu); 301 302 clockevents_unbind_device(hv_cpu->clk_evt, cpu); 303 } 304 } 305 306 /* 307 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 308 */ 309 int hv_synic_cleanup(unsigned int cpu) 310 { 311 union hv_synic_sint shared_sint; 312 union hv_synic_simp simp; 313 union hv_synic_siefp siefp; 314 union hv_synic_scontrol sctrl; 315 struct vmbus_channel *channel, *sc; 316 bool channel_found = false; 317 unsigned long flags; 318 319 if (!hv_context.synic_initialized) 320 return -EFAULT; 321 322 /* 323 * Search for channels which are bound to the CPU we're about to 324 * cleanup. In case we find one and vmbus is still connected we need to 325 * fail, this will effectively prevent CPU offlining. There is no way 326 * we can re-bind channels to different CPUs for now. 327 */ 328 mutex_lock(&vmbus_connection.channel_mutex); 329 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 330 if (channel->target_cpu == cpu) { 331 channel_found = true; 332 break; 333 } 334 spin_lock_irqsave(&channel->lock, flags); 335 list_for_each_entry(sc, &channel->sc_list, sc_list) { 336 if (sc->target_cpu == cpu) { 337 channel_found = true; 338 break; 339 } 340 } 341 spin_unlock_irqrestore(&channel->lock, flags); 342 if (channel_found) 343 break; 344 } 345 mutex_unlock(&vmbus_connection.channel_mutex); 346 347 if (channel_found && vmbus_connection.conn_state == CONNECTED) 348 return -EBUSY; 349 350 /* Turn off clockevent device */ 351 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) { 352 struct hv_per_cpu_context *hv_cpu 353 = this_cpu_ptr(hv_context.cpu_context); 354 355 clockevents_unbind_device(hv_cpu->clk_evt, cpu); 356 hv_ce_shutdown(hv_cpu->clk_evt); 357 put_cpu_ptr(hv_cpu); 358 } 359 360 hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 361 shared_sint.as_uint64); 362 363 shared_sint.masked = 1; 364 365 /* Need to correctly cleanup in the case of SMP!!! */ 366 /* Disable the interrupt */ 367 hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 368 shared_sint.as_uint64); 369 370 hv_get_simp(simp.as_uint64); 371 simp.simp_enabled = 0; 372 simp.base_simp_gpa = 0; 373 374 hv_set_simp(simp.as_uint64); 375 376 hv_get_siefp(siefp.as_uint64); 377 siefp.siefp_enabled = 0; 378 siefp.base_siefp_gpa = 0; 379 380 hv_set_siefp(siefp.as_uint64); 381 382 /* Disable the global synic bit */ 383 hv_get_synic_state(sctrl.as_uint64); 384 sctrl.enable = 0; 385 hv_set_synic_state(sctrl.as_uint64); 386 387 return 0; 388 } 389