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