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 put_cpu_ptr(hv_cpu); 86 87 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL); 88 89 return status & 0xFFFF; 90 } 91 92 static int hv_ce_set_next_event(unsigned long delta, 93 struct clock_event_device *evt) 94 { 95 u64 current_tick; 96 97 WARN_ON(!clockevent_state_oneshot(evt)); 98 99 hv_get_current_tick(current_tick); 100 current_tick += delta; 101 hv_init_timer(HV_X64_MSR_STIMER0_COUNT, current_tick); 102 return 0; 103 } 104 105 static int hv_ce_shutdown(struct clock_event_device *evt) 106 { 107 hv_init_timer(HV_X64_MSR_STIMER0_COUNT, 0); 108 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, 0); 109 110 return 0; 111 } 112 113 static int hv_ce_set_oneshot(struct clock_event_device *evt) 114 { 115 union hv_timer_config timer_cfg; 116 117 timer_cfg.enable = 1; 118 timer_cfg.auto_enable = 1; 119 timer_cfg.sintx = VMBUS_MESSAGE_SINT; 120 hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64); 121 122 return 0; 123 } 124 125 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu) 126 { 127 dev->name = "Hyper-V clockevent"; 128 dev->features = CLOCK_EVT_FEAT_ONESHOT; 129 dev->cpumask = cpumask_of(cpu); 130 dev->rating = 1000; 131 /* 132 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will 133 * result in clockevents_config_and_register() taking additional 134 * references to the hv_vmbus module making it impossible to unload. 135 */ 136 137 dev->set_state_shutdown = hv_ce_shutdown; 138 dev->set_state_oneshot = hv_ce_set_oneshot; 139 dev->set_next_event = hv_ce_set_next_event; 140 } 141 142 143 int hv_synic_alloc(void) 144 { 145 int cpu; 146 147 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids, 148 GFP_ATOMIC); 149 if (hv_context.hv_numa_map == NULL) { 150 pr_err("Unable to allocate NUMA map\n"); 151 goto err; 152 } 153 154 for_each_present_cpu(cpu) { 155 struct hv_per_cpu_context *hv_cpu 156 = per_cpu_ptr(hv_context.cpu_context, cpu); 157 158 memset(hv_cpu, 0, sizeof(*hv_cpu)); 159 tasklet_init(&hv_cpu->msg_dpc, 160 vmbus_on_msg_dpc, (unsigned long) hv_cpu); 161 162 hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device), 163 GFP_KERNEL); 164 if (hv_cpu->clk_evt == NULL) { 165 pr_err("Unable to allocate clock event device\n"); 166 goto err; 167 } 168 hv_init_clockevent_device(hv_cpu->clk_evt, cpu); 169 170 hv_cpu->synic_message_page = 171 (void *)get_zeroed_page(GFP_ATOMIC); 172 if (hv_cpu->synic_message_page == NULL) { 173 pr_err("Unable to allocate SYNIC message page\n"); 174 goto err; 175 } 176 177 hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC); 178 if (hv_cpu->synic_event_page == NULL) { 179 pr_err("Unable to allocate SYNIC event page\n"); 180 goto err; 181 } 182 183 hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC); 184 if (hv_cpu->post_msg_page == NULL) { 185 pr_err("Unable to allocate post msg page\n"); 186 goto err; 187 } 188 189 INIT_LIST_HEAD(&hv_cpu->chan_list); 190 } 191 192 return 0; 193 err: 194 return -ENOMEM; 195 } 196 197 198 void hv_synic_free(void) 199 { 200 int cpu; 201 202 for_each_present_cpu(cpu) { 203 struct hv_per_cpu_context *hv_cpu 204 = per_cpu_ptr(hv_context.cpu_context, cpu); 205 206 if (hv_cpu->synic_event_page) 207 free_page((unsigned long)hv_cpu->synic_event_page); 208 if (hv_cpu->synic_message_page) 209 free_page((unsigned long)hv_cpu->synic_message_page); 210 if (hv_cpu->post_msg_page) 211 free_page((unsigned long)hv_cpu->post_msg_page); 212 } 213 214 kfree(hv_context.hv_numa_map); 215 } 216 217 /* 218 * hv_synic_init - Initialize the Synthethic Interrupt Controller. 219 * 220 * If it is already initialized by another entity (ie x2v shim), we need to 221 * retrieve the initialized message and event pages. Otherwise, we create and 222 * initialize the message and event pages. 223 */ 224 int hv_synic_init(unsigned int cpu) 225 { 226 struct hv_per_cpu_context *hv_cpu 227 = per_cpu_ptr(hv_context.cpu_context, cpu); 228 union hv_synic_simp simp; 229 union hv_synic_siefp siefp; 230 union hv_synic_sint shared_sint; 231 union hv_synic_scontrol sctrl; 232 u64 vp_index; 233 234 /* Setup the Synic's message page */ 235 hv_get_simp(simp.as_uint64); 236 simp.simp_enabled = 1; 237 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page) 238 >> PAGE_SHIFT; 239 240 hv_set_simp(simp.as_uint64); 241 242 /* Setup the Synic's event page */ 243 hv_get_siefp(siefp.as_uint64); 244 siefp.siefp_enabled = 1; 245 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page) 246 >> PAGE_SHIFT; 247 248 hv_set_siefp(siefp.as_uint64); 249 250 /* Setup the shared SINT. */ 251 hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 252 shared_sint.as_uint64); 253 254 shared_sint.as_uint64 = 0; 255 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR; 256 shared_sint.masked = false; 257 if (ms_hyperv.hints & HV_X64_DEPRECATING_AEOI_RECOMMENDED) 258 shared_sint.auto_eoi = false; 259 else 260 shared_sint.auto_eoi = true; 261 262 hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 263 shared_sint.as_uint64); 264 265 /* Enable the global synic bit */ 266 hv_get_synic_state(sctrl.as_uint64); 267 sctrl.enable = 1; 268 269 hv_set_synic_state(sctrl.as_uint64); 270 271 hv_context.synic_initialized = true; 272 273 /* 274 * Setup the mapping between Hyper-V's notion 275 * of cpuid and Linux' notion of cpuid. 276 * This array will be indexed using Linux cpuid. 277 */ 278 hv_get_vp_index(vp_index); 279 hv_context.vp_index[cpu] = (u32)vp_index; 280 281 /* 282 * Register the per-cpu clockevent source. 283 */ 284 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) 285 clockevents_config_and_register(hv_cpu->clk_evt, 286 HV_TIMER_FREQUENCY, 287 HV_MIN_DELTA_TICKS, 288 HV_MAX_MAX_DELTA_TICKS); 289 return 0; 290 } 291 292 /* 293 * hv_synic_clockevents_cleanup - Cleanup clockevent devices 294 */ 295 void hv_synic_clockevents_cleanup(void) 296 { 297 int cpu; 298 299 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)) 300 return; 301 302 for_each_present_cpu(cpu) { 303 struct hv_per_cpu_context *hv_cpu 304 = per_cpu_ptr(hv_context.cpu_context, cpu); 305 306 clockevents_unbind_device(hv_cpu->clk_evt, cpu); 307 } 308 } 309 310 /* 311 * hv_synic_cleanup - Cleanup routine for hv_synic_init(). 312 */ 313 int hv_synic_cleanup(unsigned int cpu) 314 { 315 union hv_synic_sint shared_sint; 316 union hv_synic_simp simp; 317 union hv_synic_siefp siefp; 318 union hv_synic_scontrol sctrl; 319 struct vmbus_channel *channel, *sc; 320 bool channel_found = false; 321 unsigned long flags; 322 323 if (!hv_context.synic_initialized) 324 return -EFAULT; 325 326 /* 327 * Search for channels which are bound to the CPU we're about to 328 * cleanup. In case we find one and vmbus is still connected we need to 329 * fail, this will effectively prevent CPU offlining. There is no way 330 * we can re-bind channels to different CPUs for now. 331 */ 332 mutex_lock(&vmbus_connection.channel_mutex); 333 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 334 if (channel->target_cpu == cpu) { 335 channel_found = true; 336 break; 337 } 338 spin_lock_irqsave(&channel->lock, flags); 339 list_for_each_entry(sc, &channel->sc_list, sc_list) { 340 if (sc->target_cpu == cpu) { 341 channel_found = true; 342 break; 343 } 344 } 345 spin_unlock_irqrestore(&channel->lock, flags); 346 if (channel_found) 347 break; 348 } 349 mutex_unlock(&vmbus_connection.channel_mutex); 350 351 if (channel_found && vmbus_connection.conn_state == CONNECTED) 352 return -EBUSY; 353 354 /* Turn off clockevent device */ 355 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) { 356 struct hv_per_cpu_context *hv_cpu 357 = this_cpu_ptr(hv_context.cpu_context); 358 359 clockevents_unbind_device(hv_cpu->clk_evt, cpu); 360 hv_ce_shutdown(hv_cpu->clk_evt); 361 put_cpu_ptr(hv_cpu); 362 } 363 364 hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 365 shared_sint.as_uint64); 366 367 shared_sint.masked = 1; 368 369 /* Need to correctly cleanup in the case of SMP!!! */ 370 /* Disable the interrupt */ 371 hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, 372 shared_sint.as_uint64); 373 374 hv_get_simp(simp.as_uint64); 375 simp.simp_enabled = 0; 376 simp.base_simp_gpa = 0; 377 378 hv_set_simp(simp.as_uint64); 379 380 hv_get_siefp(siefp.as_uint64); 381 siefp.siefp_enabled = 0; 382 siefp.base_siefp_gpa = 0; 383 384 hv_set_siefp(siefp.as_uint64); 385 386 /* Disable the global synic bit */ 387 hv_get_synic_state(sctrl.as_uint64); 388 sctrl.enable = 0; 389 hv_set_synic_state(sctrl.as_uint64); 390 391 return 0; 392 } 393