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 "hyperv_vmbus.h"
36 /* The one and only */
37 struct hv_context hv_context = {
38 .synic_initialized = false,
39 .hypercall_page = NULL,
42 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
43 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
44 #define HV_MIN_DELTA_TICKS 1
47 * query_hypervisor_info - Get version info of the windows hypervisor
49 unsigned int host_info_eax;
50 unsigned int host_info_ebx;
51 unsigned int host_info_ecx;
52 unsigned int host_info_edx;
54 static int query_hypervisor_info(void)
60 unsigned int max_leaf;
64 * Its assumed that this is called after confirming that Viridian
65 * is present. Query id and revision.
71 op = HVCPUID_VENDOR_MAXFUNCTION;
72 cpuid(op, &eax, &ebx, &ecx, &edx);
76 if (max_leaf >= HVCPUID_VERSION) {
82 cpuid(op, &eax, &ebx, &ecx, &edx);
92 * do_hypercall- Invoke the specified hypercall
94 static u64 do_hypercall(u64 control, void *input, void *output)
96 u64 input_address = (input) ? virt_to_phys(input) : 0;
97 u64 output_address = (output) ? virt_to_phys(output) : 0;
98 void *hypercall_page = hv_context.hypercall_page;
103 return (u64)ULLONG_MAX;
105 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
106 __asm__ __volatile__("call *%3" : "=a" (hv_status) :
107 "c" (control), "d" (input_address),
108 "m" (hypercall_page));
114 u32 control_hi = control >> 32;
115 u32 control_lo = control & 0xFFFFFFFF;
116 u32 hv_status_hi = 1;
117 u32 hv_status_lo = 1;
118 u32 input_address_hi = input_address >> 32;
119 u32 input_address_lo = input_address & 0xFFFFFFFF;
120 u32 output_address_hi = output_address >> 32;
121 u32 output_address_lo = output_address & 0xFFFFFFFF;
124 return (u64)ULLONG_MAX;
126 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
127 "=a"(hv_status_lo) : "d" (control_hi),
128 "a" (control_lo), "b" (input_address_hi),
129 "c" (input_address_lo), "D"(output_address_hi),
130 "S"(output_address_lo), "m" (hypercall_page));
132 return hv_status_lo | ((u64)hv_status_hi << 32);
137 static cycle_t read_hv_clock_tsc(struct clocksource *arg)
139 cycle_t current_tick;
140 struct ms_hyperv_tsc_page *tsc_pg = hv_context.tsc_page;
142 if (tsc_pg->tsc_sequence != -1) {
144 * Use the tsc page to compute the value.
149 u32 sequence = tsc_pg->tsc_sequence;
151 u64 scale = tsc_pg->tsc_scale;
152 s64 offset = tsc_pg->tsc_offset;
155 /* current_tick = ((cur_tsc *scale) >> 64) + offset */
157 : "=d" (current_tick), "=a" (tmp)
158 : "a" (cur_tsc), "r" (scale));
160 current_tick += offset;
161 if (tsc_pg->tsc_sequence == sequence)
164 if (tsc_pg->tsc_sequence != -1)
167 * Fallback using MSR method.
172 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
176 static struct clocksource hyperv_cs_tsc = {
177 .name = "hyperv_clocksource_tsc_page",
179 .read = read_hv_clock_tsc,
180 .mask = CLOCKSOURCE_MASK(64),
181 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
187 * hv_init - Main initialization routine.
189 * This routine must be called before any other routines in here are called
194 union hv_x64_msr_hypercall_contents hypercall_msr;
195 union hv_x64_msr_hypercall_contents tsc_msr;
196 void *virtaddr = NULL;
199 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
200 memset(hv_context.synic_message_page, 0,
201 sizeof(void *) * NR_CPUS);
202 memset(hv_context.post_msg_page, 0,
203 sizeof(void *) * NR_CPUS);
204 memset(hv_context.vp_index, 0,
205 sizeof(int) * NR_CPUS);
206 memset(hv_context.event_dpc, 0,
207 sizeof(void *) * NR_CPUS);
208 memset(hv_context.clk_evt, 0,
209 sizeof(void *) * NR_CPUS);
211 max_leaf = query_hypervisor_info();
216 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
217 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
219 /* See if the hypercall page is already set */
220 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
222 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
227 hypercall_msr.enable = 1;
229 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
230 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
232 /* Confirm that hypercall page did get setup. */
233 hypercall_msr.as_uint64 = 0;
234 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
236 if (!hypercall_msr.enable)
239 hv_context.hypercall_page = virtaddr;
242 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
243 va_tsc = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
246 hv_context.tsc_page = va_tsc;
248 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
251 tsc_msr.guest_physical_address = vmalloc_to_pfn(va_tsc);
253 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
254 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
261 if (hypercall_msr.enable) {
262 hypercall_msr.as_uint64 = 0;
263 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
273 * hv_cleanup - Cleanup routine.
275 * This routine is called normally during driver unloading or exiting.
277 void hv_cleanup(bool crash)
279 union hv_x64_msr_hypercall_contents hypercall_msr;
281 /* Reset our OS id */
282 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
284 if (hv_context.hypercall_page) {
285 hypercall_msr.as_uint64 = 0;
286 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
288 vfree(hv_context.hypercall_page);
289 hv_context.hypercall_page = NULL;
294 * Cleanup the TSC page based CS.
296 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
298 * Crash can happen in an interrupt context and unregistering
299 * a clocksource is impossible and redundant in this case.
301 if (!oops_in_progress) {
302 clocksource_change_rating(&hyperv_cs_tsc, 10);
303 clocksource_unregister(&hyperv_cs_tsc);
306 hypercall_msr.as_uint64 = 0;
307 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
309 vfree(hv_context.tsc_page);
310 hv_context.tsc_page = NULL;
317 * hv_post_message - Post a message using the hypervisor message IPC.
319 * This involves a hypercall.
321 int hv_post_message(union hv_connection_id connection_id,
322 enum hv_message_type message_type,
323 void *payload, size_t payload_size)
326 struct hv_input_post_message *aligned_msg;
329 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
332 aligned_msg = (struct hv_input_post_message *)
333 hv_context.post_msg_page[get_cpu()];
335 aligned_msg->connectionid = connection_id;
336 aligned_msg->reserved = 0;
337 aligned_msg->message_type = message_type;
338 aligned_msg->payload_size = payload_size;
339 memcpy((void *)aligned_msg->payload, payload, payload_size);
341 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
351 * Signal an event on the specified connection using the hypervisor event IPC.
353 * This involves a hypercall.
355 u16 hv_signal_event(void *con_id)
359 status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
364 static int hv_ce_set_next_event(unsigned long delta,
365 struct clock_event_device *evt)
367 cycle_t current_tick;
369 WARN_ON(!clockevent_state_oneshot(evt));
371 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
372 current_tick += delta;
373 wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
377 static int hv_ce_shutdown(struct clock_event_device *evt)
379 wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
380 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
385 static int hv_ce_set_oneshot(struct clock_event_device *evt)
387 union hv_timer_config timer_cfg;
389 timer_cfg.enable = 1;
390 timer_cfg.auto_enable = 1;
391 timer_cfg.sintx = VMBUS_MESSAGE_SINT;
392 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
397 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
399 dev->name = "Hyper-V clockevent";
400 dev->features = CLOCK_EVT_FEAT_ONESHOT;
401 dev->cpumask = cpumask_of(cpu);
404 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
405 * result in clockevents_config_and_register() taking additional
406 * references to the hv_vmbus module making it impossible to unload.
409 dev->set_state_shutdown = hv_ce_shutdown;
410 dev->set_state_oneshot = hv_ce_set_oneshot;
411 dev->set_next_event = hv_ce_set_next_event;
415 int hv_synic_alloc(void)
417 size_t size = sizeof(struct tasklet_struct);
418 size_t ced_size = sizeof(struct clock_event_device);
421 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
423 if (hv_context.hv_numa_map == NULL) {
424 pr_err("Unable to allocate NUMA map\n");
428 for_each_present_cpu(cpu) {
429 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
430 if (hv_context.event_dpc[cpu] == NULL) {
431 pr_err("Unable to allocate event dpc\n");
434 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
436 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
437 if (hv_context.clk_evt[cpu] == NULL) {
438 pr_err("Unable to allocate clock event device\n");
442 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
444 hv_context.synic_message_page[cpu] =
445 (void *)get_zeroed_page(GFP_ATOMIC);
447 if (hv_context.synic_message_page[cpu] == NULL) {
448 pr_err("Unable to allocate SYNIC message page\n");
452 hv_context.synic_event_page[cpu] =
453 (void *)get_zeroed_page(GFP_ATOMIC);
455 if (hv_context.synic_event_page[cpu] == NULL) {
456 pr_err("Unable to allocate SYNIC event page\n");
460 hv_context.post_msg_page[cpu] =
461 (void *)get_zeroed_page(GFP_ATOMIC);
463 if (hv_context.post_msg_page[cpu] == NULL) {
464 pr_err("Unable to allocate post msg page\n");
468 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
476 static void hv_synic_free_cpu(int cpu)
478 kfree(hv_context.event_dpc[cpu]);
479 kfree(hv_context.clk_evt[cpu]);
480 if (hv_context.synic_event_page[cpu])
481 free_page((unsigned long)hv_context.synic_event_page[cpu]);
482 if (hv_context.synic_message_page[cpu])
483 free_page((unsigned long)hv_context.synic_message_page[cpu]);
484 if (hv_context.post_msg_page[cpu])
485 free_page((unsigned long)hv_context.post_msg_page[cpu]);
488 void hv_synic_free(void)
492 kfree(hv_context.hv_numa_map);
493 for_each_present_cpu(cpu)
494 hv_synic_free_cpu(cpu);
498 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
500 * If it is already initialized by another entity (ie x2v shim), we need to
501 * retrieve the initialized message and event pages. Otherwise, we create and
502 * initialize the message and event pages.
504 void hv_synic_init(void *arg)
507 union hv_synic_simp simp;
508 union hv_synic_siefp siefp;
509 union hv_synic_sint shared_sint;
510 union hv_synic_scontrol sctrl;
513 int cpu = smp_processor_id();
515 if (!hv_context.hypercall_page)
518 /* Check the version */
519 rdmsrl(HV_X64_MSR_SVERSION, version);
521 /* Setup the Synic's message page */
522 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
523 simp.simp_enabled = 1;
524 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
527 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
529 /* Setup the Synic's event page */
530 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
531 siefp.siefp_enabled = 1;
532 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
535 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
537 /* Setup the shared SINT. */
538 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
540 shared_sint.as_uint64 = 0;
541 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
542 shared_sint.masked = false;
543 shared_sint.auto_eoi = true;
545 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
547 /* Enable the global synic bit */
548 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
551 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
553 hv_context.synic_initialized = true;
556 * Setup the mapping between Hyper-V's notion
557 * of cpuid and Linux' notion of cpuid.
558 * This array will be indexed using Linux cpuid.
560 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
561 hv_context.vp_index[cpu] = (u32)vp_index;
564 * Register the per-cpu clockevent source.
566 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
567 clockevents_config_and_register(hv_context.clk_evt[cpu],
570 HV_MAX_MAX_DELTA_TICKS);
575 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
577 void hv_synic_clockevents_cleanup(void)
581 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
584 for_each_online_cpu(cpu)
585 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
589 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
591 void hv_synic_cleanup(void *arg)
593 union hv_synic_sint shared_sint;
594 union hv_synic_simp simp;
595 union hv_synic_siefp siefp;
596 union hv_synic_scontrol sctrl;
597 int cpu = smp_processor_id();
599 if (!hv_context.synic_initialized)
602 /* Turn off clockevent device */
603 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
604 hv_ce_shutdown(hv_context.clk_evt[cpu]);
606 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
608 shared_sint.masked = 1;
610 /* Need to correctly cleanup in the case of SMP!!! */
611 /* Disable the interrupt */
612 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
614 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
615 simp.simp_enabled = 0;
616 simp.base_simp_gpa = 0;
618 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
620 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
621 siefp.siefp_enabled = 0;
622 siefp.base_siefp_gpa = 0;
624 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
626 /* Disable the global synic bit */
627 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
629 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);