2 * Copyright (c) 2009, Microsoft Corporation.
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.
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
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.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
22 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 #include <linux/kernel.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 <asm/nospec-branch.h>
35 #include "hyperv_vmbus.h"
37 /* The one and only */
38 struct hv_context hv_context = {
39 .synic_initialized = false,
40 .hypercall_page = NULL,
43 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
44 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
45 #define HV_MIN_DELTA_TICKS 1
48 * query_hypervisor_info - Get version info of the windows hypervisor
50 unsigned int host_info_eax;
51 unsigned int host_info_ebx;
52 unsigned int host_info_ecx;
53 unsigned int host_info_edx;
55 static int query_hypervisor_info(void)
61 unsigned int max_leaf;
65 * Its assumed that this is called after confirming that Viridian
66 * is present. Query id and revision.
72 op = HVCPUID_VENDOR_MAXFUNCTION;
73 cpuid(op, &eax, &ebx, &ecx, &edx);
77 if (max_leaf >= HVCPUID_VERSION) {
83 cpuid(op, &eax, &ebx, &ecx, &edx);
93 * hv_do_hypercall- Invoke the specified hypercall
95 u64 hv_do_hypercall(u64 control, void *input, void *output)
97 u64 input_address = (input) ? virt_to_phys(input) : 0;
98 u64 output_address = (output) ? virt_to_phys(output) : 0;
99 void *hypercall_page = hv_context.hypercall_page;
104 return (u64)ULLONG_MAX;
106 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
107 __asm__ __volatile__(CALL_NOSPEC :
109 "c" (control), "d" (input_address),
110 THUNK_TARGET(hypercall_page));
116 u32 control_hi = control >> 32;
117 u32 control_lo = control & 0xFFFFFFFF;
118 u32 hv_status_hi = 1;
119 u32 hv_status_lo = 1;
120 u32 input_address_hi = input_address >> 32;
121 u32 input_address_lo = input_address & 0xFFFFFFFF;
122 u32 output_address_hi = output_address >> 32;
123 u32 output_address_lo = output_address & 0xFFFFFFFF;
126 return (u64)ULLONG_MAX;
128 __asm__ __volatile__ (CALL_NOSPEC : "=d"(hv_status_hi),
129 "=a"(hv_status_lo) : "d" (control_hi),
130 "a" (control_lo), "b" (input_address_hi),
131 "c" (input_address_lo), "D"(output_address_hi),
132 "S"(output_address_lo),
133 THUNK_TARGET(hypercall_page));
135 return hv_status_lo | ((u64)hv_status_hi << 32);
138 EXPORT_SYMBOL_GPL(hv_do_hypercall);
141 static cycle_t read_hv_clock_tsc(struct clocksource *arg)
143 cycle_t current_tick;
144 struct ms_hyperv_tsc_page *tsc_pg = hv_context.tsc_page;
146 if (tsc_pg->tsc_sequence != 0) {
148 * Use the tsc page to compute the value.
153 u32 sequence = tsc_pg->tsc_sequence;
155 u64 scale = tsc_pg->tsc_scale;
156 s64 offset = tsc_pg->tsc_offset;
159 /* current_tick = ((cur_tsc *scale) >> 64) + offset */
161 : "=d" (current_tick), "=a" (tmp)
162 : "a" (cur_tsc), "r" (scale));
164 current_tick += offset;
165 if (tsc_pg->tsc_sequence == sequence)
168 if (tsc_pg->tsc_sequence != 0)
171 * Fallback using MSR method.
176 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
180 static struct clocksource hyperv_cs_tsc = {
181 .name = "hyperv_clocksource_tsc_page",
183 .read = read_hv_clock_tsc,
184 .mask = CLOCKSOURCE_MASK(64),
185 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
191 * hv_init - Main initialization routine.
193 * This routine must be called before any other routines in here are called
198 union hv_x64_msr_hypercall_contents hypercall_msr;
199 void *virtaddr = NULL;
201 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
202 memset(hv_context.synic_message_page, 0,
203 sizeof(void *) * NR_CPUS);
204 memset(hv_context.post_msg_page, 0,
205 sizeof(void *) * NR_CPUS);
206 memset(hv_context.vp_index, 0,
207 sizeof(int) * NR_CPUS);
208 memset(hv_context.event_dpc, 0,
209 sizeof(void *) * NR_CPUS);
210 memset(hv_context.msg_dpc, 0,
211 sizeof(void *) * NR_CPUS);
212 memset(hv_context.clk_evt, 0,
213 sizeof(void *) * NR_CPUS);
215 max_leaf = query_hypervisor_info();
220 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
221 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
223 /* See if the hypercall page is already set */
224 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
226 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
231 hypercall_msr.enable = 1;
233 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
234 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
236 /* Confirm that hypercall page did get setup. */
237 hypercall_msr.as_uint64 = 0;
238 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
240 if (!hypercall_msr.enable)
243 hv_context.hypercall_page = virtaddr;
246 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
247 union hv_x64_msr_hypercall_contents tsc_msr;
250 va_tsc = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
253 hv_context.tsc_page = va_tsc;
255 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
258 tsc_msr.guest_physical_address = vmalloc_to_pfn(va_tsc);
260 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
261 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
268 if (hypercall_msr.enable) {
269 hypercall_msr.as_uint64 = 0;
270 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
280 * hv_cleanup - Cleanup routine.
282 * This routine is called normally during driver unloading or exiting.
284 void hv_cleanup(bool crash)
286 union hv_x64_msr_hypercall_contents hypercall_msr;
288 /* Reset our OS id */
289 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
291 if (hv_context.hypercall_page) {
292 hypercall_msr.as_uint64 = 0;
293 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
295 vfree(hv_context.hypercall_page);
296 hv_context.hypercall_page = NULL;
301 * Cleanup the TSC page based CS.
303 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
305 * Crash can happen in an interrupt context and unregistering
306 * a clocksource is impossible and redundant in this case.
308 if (!oops_in_progress) {
309 clocksource_change_rating(&hyperv_cs_tsc, 10);
310 clocksource_unregister(&hyperv_cs_tsc);
313 hypercall_msr.as_uint64 = 0;
314 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
316 vfree(hv_context.tsc_page);
317 hv_context.tsc_page = NULL;
324 * hv_post_message - Post a message using the hypervisor message IPC.
326 * This involves a hypercall.
328 int hv_post_message(union hv_connection_id connection_id,
329 enum hv_message_type message_type,
330 void *payload, size_t payload_size)
333 struct hv_input_post_message *aligned_msg;
336 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
339 aligned_msg = (struct hv_input_post_message *)
340 hv_context.post_msg_page[get_cpu()];
342 aligned_msg->connectionid = connection_id;
343 aligned_msg->reserved = 0;
344 aligned_msg->message_type = message_type;
345 aligned_msg->payload_size = payload_size;
346 memcpy((void *)aligned_msg->payload, payload, payload_size);
348 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
351 return status & 0xFFFF;
354 static int hv_ce_set_next_event(unsigned long delta,
355 struct clock_event_device *evt)
357 cycle_t current_tick;
359 WARN_ON(!clockevent_state_oneshot(evt));
361 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
362 current_tick += delta;
363 wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
367 static int hv_ce_shutdown(struct clock_event_device *evt)
369 wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
370 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
375 static int hv_ce_set_oneshot(struct clock_event_device *evt)
377 union hv_timer_config timer_cfg;
379 timer_cfg.enable = 1;
380 timer_cfg.auto_enable = 1;
381 timer_cfg.sintx = VMBUS_MESSAGE_SINT;
382 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
387 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
389 dev->name = "Hyper-V clockevent";
390 dev->features = CLOCK_EVT_FEAT_ONESHOT;
391 dev->cpumask = cpumask_of(cpu);
394 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
395 * result in clockevents_config_and_register() taking additional
396 * references to the hv_vmbus module making it impossible to unload.
399 dev->set_state_shutdown = hv_ce_shutdown;
400 dev->set_state_oneshot = hv_ce_set_oneshot;
401 dev->set_next_event = hv_ce_set_next_event;
405 int hv_synic_alloc(void)
407 size_t size = sizeof(struct tasklet_struct);
408 size_t ced_size = sizeof(struct clock_event_device);
411 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
413 if (hv_context.hv_numa_map == NULL) {
414 pr_err("Unable to allocate NUMA map\n");
418 for_each_present_cpu(cpu) {
419 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
420 if (hv_context.event_dpc[cpu] == NULL) {
421 pr_err("Unable to allocate event dpc\n");
424 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
426 hv_context.msg_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
427 if (hv_context.msg_dpc[cpu] == NULL) {
428 pr_err("Unable to allocate event dpc\n");
431 tasklet_init(hv_context.msg_dpc[cpu], vmbus_on_msg_dpc, cpu);
433 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
434 if (hv_context.clk_evt[cpu] == NULL) {
435 pr_err("Unable to allocate clock event device\n");
439 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
441 hv_context.synic_message_page[cpu] =
442 (void *)get_zeroed_page(GFP_ATOMIC);
444 if (hv_context.synic_message_page[cpu] == NULL) {
445 pr_err("Unable to allocate SYNIC message page\n");
449 hv_context.synic_event_page[cpu] =
450 (void *)get_zeroed_page(GFP_ATOMIC);
452 if (hv_context.synic_event_page[cpu] == NULL) {
453 pr_err("Unable to allocate SYNIC event page\n");
457 hv_context.post_msg_page[cpu] =
458 (void *)get_zeroed_page(GFP_ATOMIC);
460 if (hv_context.post_msg_page[cpu] == NULL) {
461 pr_err("Unable to allocate post msg page\n");
465 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
473 static void hv_synic_free_cpu(int cpu)
475 kfree(hv_context.event_dpc[cpu]);
476 kfree(hv_context.msg_dpc[cpu]);
477 kfree(hv_context.clk_evt[cpu]);
478 if (hv_context.synic_event_page[cpu])
479 free_page((unsigned long)hv_context.synic_event_page[cpu]);
480 if (hv_context.synic_message_page[cpu])
481 free_page((unsigned long)hv_context.synic_message_page[cpu]);
482 if (hv_context.post_msg_page[cpu])
483 free_page((unsigned long)hv_context.post_msg_page[cpu]);
486 void hv_synic_free(void)
490 kfree(hv_context.hv_numa_map);
491 for_each_present_cpu(cpu)
492 hv_synic_free_cpu(cpu);
496 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
498 * If it is already initialized by another entity (ie x2v shim), we need to
499 * retrieve the initialized message and event pages. Otherwise, we create and
500 * initialize the message and event pages.
502 void hv_synic_init(void *arg)
505 union hv_synic_simp simp;
506 union hv_synic_siefp siefp;
507 union hv_synic_sint shared_sint;
508 union hv_synic_scontrol sctrl;
511 int cpu = smp_processor_id();
513 if (!hv_context.hypercall_page)
516 /* Check the version */
517 rdmsrl(HV_X64_MSR_SVERSION, version);
519 /* Setup the Synic's message page */
520 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
521 simp.simp_enabled = 1;
522 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
525 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
527 /* Setup the Synic's event page */
528 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
529 siefp.siefp_enabled = 1;
530 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
533 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
535 /* Setup the shared SINT. */
536 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
538 shared_sint.as_uint64 = 0;
539 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
540 shared_sint.masked = false;
541 shared_sint.auto_eoi = true;
543 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
545 /* Enable the global synic bit */
546 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
549 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
551 hv_context.synic_initialized = true;
554 * Setup the mapping between Hyper-V's notion
555 * of cpuid and Linux' notion of cpuid.
556 * This array will be indexed using Linux cpuid.
558 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
559 hv_context.vp_index[cpu] = (u32)vp_index;
562 * Register the per-cpu clockevent source.
564 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
565 clockevents_config_and_register(hv_context.clk_evt[cpu],
568 HV_MAX_MAX_DELTA_TICKS);
573 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
575 void hv_synic_clockevents_cleanup(void)
579 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
582 for_each_online_cpu(cpu)
583 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
587 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
589 void hv_synic_cleanup(void *arg)
591 union hv_synic_sint shared_sint;
592 union hv_synic_simp simp;
593 union hv_synic_siefp siefp;
594 union hv_synic_scontrol sctrl;
595 int cpu = smp_processor_id();
597 if (!hv_context.synic_initialized)
600 /* Turn off clockevent device */
601 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
602 hv_ce_shutdown(hv_context.clk_evt[cpu]);
604 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
606 shared_sint.masked = 1;
608 /* Need to correctly cleanup in the case of SMP!!! */
609 /* Disable the interrupt */
610 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
612 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
613 simp.simp_enabled = 0;
614 simp.base_simp_gpa = 0;
616 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
618 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
619 siefp.siefp_enabled = 0;
620 siefp.base_siefp_gpa = 0;
622 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
624 /* Disable the global synic bit */
625 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
627 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);