drivers/hv/hv.c

   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         union hv_x64_msr_hypercall_contents hypercall_msr;
  53
  54         memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
  55         memset(hv_context.synic_message_page, 0,
  56                sizeof(void *) * NR_CPUS);
  57         memset(hv_context.post_msg_page, 0,
  58                sizeof(void *) * NR_CPUS);
  59         memset(hv_context.vp_index, 0,
  60                sizeof(int) * NR_CPUS);
  61         memset(hv_context.event_dpc, 0,
  62                sizeof(void *) * NR_CPUS);
  63         memset(hv_context.msg_dpc, 0,
  64                sizeof(void *) * NR_CPUS);
  65         memset(hv_context.clk_evt, 0,
  66                sizeof(void *) * NR_CPUS);
  67
  68         /* See if the hypercall page is already set */
  69         hypercall_msr.as_uint64 = 0;
  70         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
  71
  72         if (!hypercall_msr.enable)
  73                 return -ENOTSUPP;
  74
  75         return 0;
  76 }
  77
  78 /*
  79  * hv_cleanup - Cleanup routine.
  80  *
  81  * This routine is called normally during driver unloading or exiting.
  82  */
  83 void hv_cleanup(bool crash)
  84 {
  85
  86 }
  87
  88 /*
  89  * hv_post_message - Post a message using the hypervisor message IPC.
  90  *
  91  * This involves a hypercall.
  92  */
  93 int hv_post_message(union hv_connection_id connection_id,
  94                   enum hv_message_type message_type,
  95                   void *payload, size_t payload_size)
  96 {
  97
  98         struct hv_input_post_message *aligned_msg;
  99         u64 status;
 100
 101         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
 102                 return -EMSGSIZE;
 103
 104         aligned_msg = (struct hv_input_post_message *)
 105                         hv_context.post_msg_page[get_cpu()];
 106
 107         aligned_msg->connectionid = connection_id;
 108         aligned_msg->reserved = 0;
 109         aligned_msg->message_type = message_type;
 110         aligned_msg->payload_size = payload_size;
 111         memcpy((void *)aligned_msg->payload, payload, payload_size);
 112
 113         status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
 114
 115         put_cpu();
 116         return status & 0xFFFF;
 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         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
 127         current_tick += delta;
 128         wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
 129         return 0;
 130 }
 131
 132 static int hv_ce_shutdown(struct clock_event_device *evt)
 133 {
 134         wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
 135         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
 136
 137         return 0;
 138 }
 139
 140 static int hv_ce_set_oneshot(struct clock_event_device *evt)
 141 {
 142         union hv_timer_config timer_cfg;
 143
 144         timer_cfg.enable = 1;
 145         timer_cfg.auto_enable = 1;
 146         timer_cfg.sintx = VMBUS_MESSAGE_SINT;
 147         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
 148
 149         return 0;
 150 }
 151
 152 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
 153 {
 154         dev->name = "Hyper-V clockevent";
 155         dev->features = CLOCK_EVT_FEAT_ONESHOT;
 156         dev->cpumask = cpumask_of(cpu);
 157         dev->rating = 1000;
 158         /*
 159          * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
 160          * result in clockevents_config_and_register() taking additional
 161          * references to the hv_vmbus module making it impossible to unload.
 162          */
 163
 164         dev->set_state_shutdown = hv_ce_shutdown;
 165         dev->set_state_oneshot = hv_ce_set_oneshot;
 166         dev->set_next_event = hv_ce_set_next_event;
 167 }
 168
 169
 170 int hv_synic_alloc(void)
 171 {
 172         size_t size = sizeof(struct tasklet_struct);
 173         size_t ced_size = sizeof(struct clock_event_device);
 174         int cpu;
 175
 176         hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
 177                                          GFP_ATOMIC);
 178         if (hv_context.hv_numa_map == NULL) {
 179                 pr_err("Unable to allocate NUMA map\n");
 180                 goto err;
 181         }
 182
 183         for_each_present_cpu(cpu) {
 184                 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
 185                 if (hv_context.event_dpc[cpu] == NULL) {
 186                         pr_err("Unable to allocate event dpc\n");
 187                         goto err;
 188                 }
 189                 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
 190
 191                 hv_context.msg_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
 192                 if (hv_context.msg_dpc[cpu] == NULL) {
 193                         pr_err("Unable to allocate event dpc\n");
 194                         goto err;
 195                 }
 196                 tasklet_init(hv_context.msg_dpc[cpu], vmbus_on_msg_dpc, cpu);
 197
 198                 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
 199                 if (hv_context.clk_evt[cpu] == NULL) {
 200                         pr_err("Unable to allocate clock event device\n");
 201                         goto err;
 202                 }
 203
 204                 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
 205
 206                 hv_context.synic_message_page[cpu] =
 207                         (void *)get_zeroed_page(GFP_ATOMIC);
 208
 209                 if (hv_context.synic_message_page[cpu] == NULL) {
 210                         pr_err("Unable to allocate SYNIC message page\n");
 211                         goto err;
 212                 }
 213
 214                 hv_context.synic_event_page[cpu] =
 215                         (void *)get_zeroed_page(GFP_ATOMIC);
 216
 217                 if (hv_context.synic_event_page[cpu] == NULL) {
 218                         pr_err("Unable to allocate SYNIC event page\n");
 219                         goto err;
 220                 }
 221
 222                 hv_context.post_msg_page[cpu] =
 223                         (void *)get_zeroed_page(GFP_ATOMIC);
 224
 225                 if (hv_context.post_msg_page[cpu] == NULL) {
 226                         pr_err("Unable to allocate post msg page\n");
 227                         goto err;
 228                 }
 229
 230                 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
 231         }
 232
 233         return 0;
 234 err:
 235         return -ENOMEM;
 236 }
 237
 238 static void hv_synic_free_cpu(int cpu)
 239 {
 240         kfree(hv_context.event_dpc[cpu]);
 241         kfree(hv_context.msg_dpc[cpu]);
 242         kfree(hv_context.clk_evt[cpu]);
 243         if (hv_context.synic_event_page[cpu])
 244                 free_page((unsigned long)hv_context.synic_event_page[cpu]);
 245         if (hv_context.synic_message_page[cpu])
 246                 free_page((unsigned long)hv_context.synic_message_page[cpu]);
 247         if (hv_context.post_msg_page[cpu])
 248                 free_page((unsigned long)hv_context.post_msg_page[cpu]);
 249 }
 250
 251 void hv_synic_free(void)
 252 {
 253         int cpu;
 254
 255         kfree(hv_context.hv_numa_map);
 256         for_each_present_cpu(cpu)
 257                 hv_synic_free_cpu(cpu);
 258 }
 259
 260 /*
 261  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
 262  *
 263  * If it is already initialized by another entity (ie x2v shim), we need to
 264  * retrieve the initialized message and event pages.  Otherwise, we create and
 265  * initialize the message and event pages.
 266  */
 267 int hv_synic_init(unsigned int cpu)
 268 {
 269         u64 version;
 270         union hv_synic_simp simp;
 271         union hv_synic_siefp siefp;
 272         union hv_synic_sint shared_sint;
 273         union hv_synic_scontrol sctrl;
 274         u64 vp_index;
 275
 276         /* Check the version */
 277         rdmsrl(HV_X64_MSR_SVERSION, version);
 278
 279         /* Setup the Synic's message page */
 280         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 281         simp.simp_enabled = 1;
 282         simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
 283                 >> PAGE_SHIFT;
 284
 285         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 286
 287         /* Setup the Synic's event page */
 288         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 289         siefp.siefp_enabled = 1;
 290         siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
 291                 >> PAGE_SHIFT;
 292
 293         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 294
 295         /* Setup the shared SINT. */
 296         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 297
 298         shared_sint.as_uint64 = 0;
 299         shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
 300         shared_sint.masked = false;
 301         shared_sint.auto_eoi = true;
 302
 303         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 304
 305         /* Enable the global synic bit */
 306         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 307         sctrl.enable = 1;
 308
 309         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 310
 311         hv_context.synic_initialized = true;
 312
 313         /*
 314          * Setup the mapping between Hyper-V's notion
 315          * of cpuid and Linux' notion of cpuid.
 316          * This array will be indexed using Linux cpuid.
 317          */
 318         rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
 319         hv_context.vp_index[cpu] = (u32)vp_index;
 320
 321         /*
 322          * Register the per-cpu clockevent source.
 323          */
 324         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
 325                 clockevents_config_and_register(hv_context.clk_evt[cpu],
 326                                                 HV_TIMER_FREQUENCY,
 327                                                 HV_MIN_DELTA_TICKS,
 328                                                 HV_MAX_MAX_DELTA_TICKS);
 329         return 0;
 330 }
 331
 332 /*
 333  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
 334  */
 335 void hv_synic_clockevents_cleanup(void)
 336 {
 337         int cpu;
 338
 339         if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
 340                 return;
 341
 342         for_each_present_cpu(cpu)
 343                 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
 344 }
 345
 346 /*
 347  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
 348  */
 349 int hv_synic_cleanup(unsigned int cpu)
 350 {
 351         union hv_synic_sint shared_sint;
 352         union hv_synic_simp simp;
 353         union hv_synic_siefp siefp;
 354         union hv_synic_scontrol sctrl;
 355         struct vmbus_channel *channel, *sc;
 356         bool channel_found = false;
 357         unsigned long flags;
 358
 359         if (!hv_context.synic_initialized)
 360                 return -EFAULT;
 361
 362         /*
 363          * Search for channels which are bound to the CPU we're about to
 364          * cleanup. In case we find one and vmbus is still connected we need to
 365          * fail, this will effectively prevent CPU offlining. There is no way
 366          * we can re-bind channels to different CPUs for now.
 367          */
 368         mutex_lock(&vmbus_connection.channel_mutex);
 369         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 370                 if (channel->target_cpu == cpu) {
 371                         channel_found = true;
 372                         break;
 373                 }
 374                 spin_lock_irqsave(&channel->lock, flags);
 375                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
 376                         if (sc->target_cpu == cpu) {
 377                                 channel_found = true;
 378                                 break;
 379                         }
 380                 }
 381                 spin_unlock_irqrestore(&channel->lock, flags);
 382                 if (channel_found)
 383                         break;
 384         }
 385         mutex_unlock(&vmbus_connection.channel_mutex);
 386
 387         if (channel_found && vmbus_connection.conn_state == CONNECTED)
 388                 return -EBUSY;
 389
 390         /* Turn off clockevent device */
 391         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) {
 392                 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
 393                 hv_ce_shutdown(hv_context.clk_evt[cpu]);
 394         }
 395
 396         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 397
 398         shared_sint.masked = 1;
 399
 400         /* Need to correctly cleanup in the case of SMP!!! */
 401         /* Disable the interrupt */
 402         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 403
 404         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 405         simp.simp_enabled = 0;
 406         simp.base_simp_gpa = 0;
 407
 408         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 409
 410         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 411         siefp.siefp_enabled = 0;
 412         siefp.base_siefp_gpa = 0;
 413
 414         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 415
 416         /* Disable the global synic bit */
 417         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 418         sctrl.enable = 0;
 419         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 420
 421         return 0;
 422 }