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 <asm/hyperv.h>
  30 #include "hyperv_vmbus.h"
  31
  32 /* The one and only */
  33 struct hv_context hv_context = {
  34         .synic_initialized      = false,
  35         .hypercall_page         = NULL,
  36         .signal_event_param     = NULL,
  37         .signal_event_buffer    = NULL,
  38 };
  39
  40 /*
  41  * query_hypervisor_presence
  42  * - Query the cpuid for presence of windows hypervisor
  43  */
  44 static int query_hypervisor_presence(void)
  45 {
  46         unsigned int eax;
  47         unsigned int ebx;
  48         unsigned int ecx;
  49         unsigned int edx;
  50         unsigned int op;
  51
  52         eax = 0;
  53         ebx = 0;
  54         ecx = 0;
  55         edx = 0;
  56         op = HVCPUID_VERSION_FEATURES;
  57         cpuid(op, &eax, &ebx, &ecx, &edx);
  58
  59         return ecx & HV_PRESENT_BIT;
  60 }
  61
  62 /*
  63  * query_hypervisor_info - Get version info of the windows hypervisor
  64  */
  65 static int query_hypervisor_info(void)
  66 {
  67         unsigned int eax;
  68         unsigned int ebx;
  69         unsigned int ecx;
  70         unsigned int edx;
  71         unsigned int max_leaf;
  72         unsigned int op;
  73
  74         /*
  75         * Its assumed that this is called after confirming that Viridian
  76         * is present. Query id and revision.
  77         */
  78         eax = 0;
  79         ebx = 0;
  80         ecx = 0;
  81         edx = 0;
  82         op = HVCPUID_VENDOR_MAXFUNCTION;
  83         cpuid(op, &eax, &ebx, &ecx, &edx);
  84
  85         max_leaf = eax;
  86
  87         if (max_leaf >= HVCPUID_VERSION) {
  88                 eax = 0;
  89                 ebx = 0;
  90                 ecx = 0;
  91                 edx = 0;
  92                 op = HVCPUID_VERSION;
  93                 cpuid(op, &eax, &ebx, &ecx, &edx);
  94                 pr_info("Hyper-V Host OS Build:%d-%d.%d-%d-%d.%d\n",
  95                             eax,
  96                             ebx >> 16,
  97                             ebx & 0xFFFF,
  98                             ecx,
  99                             edx >> 24,
 100                             edx & 0xFFFFFF);
 101         }
 102         return max_leaf;
 103 }
 104
 105 /*
 106  * do_hypercall- Invoke the specified hypercall
 107  */
 108 static u64 do_hypercall(u64 control, void *input, void *output)
 109 {
 110 #ifdef CONFIG_X86_64
 111         u64 hv_status = 0;
 112         u64 input_address = (input) ? virt_to_phys(input) : 0;
 113         u64 output_address = (output) ? virt_to_phys(output) : 0;
 114         void *hypercall_page = hv_context.hypercall_page;
 115
 116         __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
 117         __asm__ __volatile__("call *%3" : "=a" (hv_status) :
 118                              "c" (control), "d" (input_address),
 119                              "m" (hypercall_page));
 120
 121         return hv_status;
 122
 123 #else
 124
 125         u32 control_hi = control >> 32;
 126         u32 control_lo = control & 0xFFFFFFFF;
 127         u32 hv_status_hi = 1;
 128         u32 hv_status_lo = 1;
 129         u64 input_address = (input) ? virt_to_phys(input) : 0;
 130         u32 input_address_hi = input_address >> 32;
 131         u32 input_address_lo = input_address & 0xFFFFFFFF;
 132         u64 output_address = (output) ? virt_to_phys(output) : 0;
 133         u32 output_address_hi = output_address >> 32;
 134         u32 output_address_lo = output_address & 0xFFFFFFFF;
 135         void *hypercall_page = hv_context.hypercall_page;
 136
 137         __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
 138                               "=a"(hv_status_lo) : "d" (control_hi),
 139                               "a" (control_lo), "b" (input_address_hi),
 140                               "c" (input_address_lo), "D"(output_address_hi),
 141                               "S"(output_address_lo), "m" (hypercall_page));
 142
 143         return hv_status_lo | ((u64)hv_status_hi << 32);
 144 #endif /* !x86_64 */
 145 }
 146
 147 /*
 148  * hv_init - Main initialization routine.
 149  *
 150  * This routine must be called before any other routines in here are called
 151  */
 152 int hv_init(void)
 153 {
 154         int max_leaf;
 155         union hv_x64_msr_hypercall_contents hypercall_msr;
 156         void *virtaddr = NULL;
 157
 158         memset(hv_context.synic_event_page, 0, sizeof(void *) * MAX_NUM_CPUS);
 159         memset(hv_context.synic_message_page, 0,
 160                sizeof(void *) * MAX_NUM_CPUS);
 161         memset(hv_context.post_msg_page, 0,
 162                sizeof(void *) * MAX_NUM_CPUS);
 163
 164         if (!query_hypervisor_presence())
 165                 goto cleanup;
 166
 167         max_leaf = query_hypervisor_info();
 168
 169         rdmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
 170
 171         if (hv_context.guestid != 0)
 172                 goto cleanup;
 173
 174         /* Write our OS info */
 175         wrmsrl(HV_X64_MSR_GUEST_OS_ID, HV_LINUX_GUEST_ID);
 176         hv_context.guestid = HV_LINUX_GUEST_ID;
 177
 178         /* See if the hypercall page is already set */
 179         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 180
 181         virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC);
 182
 183         if (!virtaddr)
 184                 goto cleanup;
 185
 186         hypercall_msr.enable = 1;
 187
 188         hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
 189         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 190
 191         /* Confirm that hypercall page did get setup. */
 192         hypercall_msr.as_uint64 = 0;
 193         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 194
 195         if (!hypercall_msr.enable)
 196                 goto cleanup;
 197
 198         hv_context.hypercall_page = virtaddr;
 199
 200         /* Setup the global signal event param for the signal event hypercall */
 201         hv_context.signal_event_buffer =
 202                         kmalloc(sizeof(struct hv_input_signal_event_buffer),
 203                                 GFP_KERNEL);
 204         if (!hv_context.signal_event_buffer)
 205                 goto cleanup;
 206
 207         hv_context.signal_event_param =
 208                 (struct hv_input_signal_event *)
 209                         (ALIGN((unsigned long)
 210                                   hv_context.signal_event_buffer,
 211                                   HV_HYPERCALL_PARAM_ALIGN));
 212         hv_context.signal_event_param->connectionid.asu32 = 0;
 213         hv_context.signal_event_param->connectionid.u.id =
 214                                                 VMBUS_EVENT_CONNECTION_ID;
 215         hv_context.signal_event_param->flag_number = 0;
 216         hv_context.signal_event_param->rsvdz = 0;
 217
 218         return 0;
 219
 220 cleanup:
 221         if (virtaddr) {
 222                 if (hypercall_msr.enable) {
 223                         hypercall_msr.as_uint64 = 0;
 224                         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 225                 }
 226
 227                 vfree(virtaddr);
 228         }
 229
 230         return -ENOTSUPP;
 231 }
 232
 233 /*
 234  * hv_cleanup - Cleanup routine.
 235  *
 236  * This routine is called normally during driver unloading or exiting.
 237  */
 238 void hv_cleanup(void)
 239 {
 240         union hv_x64_msr_hypercall_contents hypercall_msr;
 241
 242         kfree(hv_context.signal_event_buffer);
 243         hv_context.signal_event_buffer = NULL;
 244         hv_context.signal_event_param = NULL;
 245
 246         if (hv_context.hypercall_page) {
 247                 hypercall_msr.as_uint64 = 0;
 248                 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
 249                 vfree(hv_context.hypercall_page);
 250                 hv_context.hypercall_page = NULL;
 251         }
 252 }
 253
 254 /*
 255  * hv_post_message - Post a message using the hypervisor message IPC.
 256  *
 257  * This involves a hypercall.
 258  */
 259 u16 hv_post_message(union hv_connection_id connection_id,
 260                   enum hv_message_type message_type,
 261                   void *payload, size_t payload_size)
 262 {
 263
 264         struct hv_input_post_message *aligned_msg;
 265         u16 status;
 266
 267         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
 268                 return -EMSGSIZE;
 269
 270         aligned_msg = (struct hv_input_post_message *)
 271                         hv_context.post_msg_page[get_cpu()];
 272
 273         aligned_msg->connectionid = connection_id;
 274         aligned_msg->reserved = 0;
 275         aligned_msg->message_type = message_type;
 276         aligned_msg->payload_size = payload_size;
 277         memcpy((void *)aligned_msg->payload, payload, payload_size);
 278
 279         status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
 280                 & 0xFFFF;
 281
 282         put_cpu();
 283         return status;
 284 }
 285
 286
 287 /*
 288  * hv_signal_event -
 289  * Signal an event on the specified connection using the hypervisor event IPC.
 290  *
 291  * This involves a hypercall.
 292  */
 293 u16 hv_signal_event(void)
 294 {
 295         u16 status;
 296
 297         status = do_hypercall(HVCALL_SIGNAL_EVENT,
 298                                hv_context.signal_event_param,
 299                                NULL) & 0xFFFF;
 300         return status;
 301 }
 302
 303 /*
 304  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
 305  *
 306  * If it is already initialized by another entity (ie x2v shim), we need to
 307  * retrieve the initialized message and event pages.  Otherwise, we create and
 308  * initialize the message and event pages.
 309  */
 310 void hv_synic_init(void *irqarg)
 311 {
 312         u64 version;
 313         union hv_synic_simp simp;
 314         union hv_synic_siefp siefp;
 315         union hv_synic_sint shared_sint;
 316         union hv_synic_scontrol sctrl;
 317
 318         u32 irq_vector = *((u32 *)(irqarg));
 319         int cpu = smp_processor_id();
 320
 321         if (!hv_context.hypercall_page)
 322                 return;
 323
 324         /* Check the version */
 325         rdmsrl(HV_X64_MSR_SVERSION, version);
 326
 327         hv_context.synic_message_page[cpu] =
 328                 (void *)get_zeroed_page(GFP_ATOMIC);
 329
 330         if (hv_context.synic_message_page[cpu] == NULL) {
 331                 pr_err("Unable to allocate SYNIC message page\n");
 332                 goto cleanup;
 333         }
 334
 335         hv_context.synic_event_page[cpu] =
 336                 (void *)get_zeroed_page(GFP_ATOMIC);
 337
 338         if (hv_context.synic_event_page[cpu] == NULL) {
 339                 pr_err("Unable to allocate SYNIC event page\n");
 340                 goto cleanup;
 341         }
 342
 343         hv_context.post_msg_page[cpu] =
 344                 (void *)get_zeroed_page(GFP_ATOMIC);
 345
 346         if (hv_context.post_msg_page[cpu] == NULL) {
 347                 pr_err("Unable to allocate post msg page\n");
 348                 goto cleanup;
 349         }
 350
 351         /* Setup the Synic's message page */
 352         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 353         simp.simp_enabled = 1;
 354         simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
 355                 >> PAGE_SHIFT;
 356
 357         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 358
 359         /* Setup the Synic's event page */
 360         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 361         siefp.siefp_enabled = 1;
 362         siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
 363                 >> PAGE_SHIFT;
 364
 365         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 366
 367         /* Setup the shared SINT. */
 368         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 369
 370         shared_sint.as_uint64 = 0;
 371         shared_sint.vector = irq_vector; /* HV_SHARED_SINT_IDT_VECTOR + 0x20; */
 372         shared_sint.masked = false;
 373         shared_sint.auto_eoi = false;
 374
 375         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 376
 377         /* Enable the global synic bit */
 378         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 379         sctrl.enable = 1;
 380
 381         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
 382
 383         hv_context.synic_initialized = true;
 384         return;
 385
 386 cleanup:
 387         if (hv_context.synic_event_page[cpu])
 388                 free_page((unsigned long)hv_context.synic_event_page[cpu]);
 389
 390         if (hv_context.synic_message_page[cpu])
 391                 free_page((unsigned long)hv_context.synic_message_page[cpu]);
 392         if (hv_context.post_msg_page[cpu])
 393                 free_page((unsigned long)hv_context.post_msg_page[cpu]);
 394         return;
 395 }
 396
 397 /*
 398  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
 399  */
 400 void hv_synic_cleanup(void *arg)
 401 {
 402         union hv_synic_sint shared_sint;
 403         union hv_synic_simp simp;
 404         union hv_synic_siefp siefp;
 405         int cpu = smp_processor_id();
 406
 407         if (!hv_context.synic_initialized)
 408                 return;
 409
 410         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 411
 412         shared_sint.masked = 1;
 413
 414         /* Need to correctly cleanup in the case of SMP!!! */
 415         /* Disable the interrupt */
 416         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
 417
 418         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 419         simp.simp_enabled = 0;
 420         simp.base_simp_gpa = 0;
 421
 422         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
 423
 424         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 425         siefp.siefp_enabled = 0;
 426         siefp.base_siefp_gpa = 0;
 427
 428         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
 429
 430         free_page((unsigned long)hv_context.synic_message_page[cpu]);
 431         free_page((unsigned long)hv_context.synic_event_page[cpu]);
 432         free_page((unsigned long)hv_context.post_msg_page[cpu]);
 433 }