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 <asm/nospec-branch.h>
35 #include "hyperv_vmbus.h"
36
37 /* The one and only */
38 struct hv_context hv_context = {
39 .synic_initialized = false,
40 .hypercall_page = NULL,
41 };
42
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
46
47 /*
48 * query_hypervisor_info - Get version info of the windows hypervisor
49 */
50 unsigned int host_info_eax;
51 unsigned int host_info_ebx;
52 unsigned int host_info_ecx;
53 unsigned int host_info_edx;
54
query_hypervisor_info(void)55 static int query_hypervisor_info(void)
56 {
57 unsigned int eax;
58 unsigned int ebx;
59 unsigned int ecx;
60 unsigned int edx;
61 unsigned int max_leaf;
62 unsigned int op;
63
64 /*
65 * Its assumed that this is called after confirming that Viridian
66 * is present. Query id and revision.
67 */
68 eax = 0;
69 ebx = 0;
70 ecx = 0;
71 edx = 0;
72 op = HVCPUID_VENDOR_MAXFUNCTION;
73 cpuid(op, &eax, &ebx, &ecx, &edx);
74
75 max_leaf = eax;
76
77 if (max_leaf >= HVCPUID_VERSION) {
78 eax = 0;
79 ebx = 0;
80 ecx = 0;
81 edx = 0;
82 op = HVCPUID_VERSION;
83 cpuid(op, &eax, &ebx, &ecx, &edx);
84 host_info_eax = eax;
85 host_info_ebx = ebx;
86 host_info_ecx = ecx;
87 host_info_edx = edx;
88 }
89 return max_leaf;
90 }
91
92 /*
93 * do_hypercall- Invoke the specified hypercall
94 */
do_hypercall(u64 control,void * input,void * output)95 static u64 do_hypercall(u64 control, void *input, void *output)
96 {
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;
100 #ifdef CONFIG_X86_64
101 u64 hv_status = 0;
102
103 if (!hypercall_page)
104 return (u64)ULLONG_MAX;
105
106 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
107 __asm__ __volatile__(CALL_NOSPEC :
108 "=a" (hv_status) :
109 "c" (control), "d" (input_address),
110 THUNK_TARGET(hypercall_page));
111
112 return hv_status;
113
114 #else
115
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;
124
125 if (!hypercall_page)
126 return (u64)ULLONG_MAX;
127
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));
134
135 return hv_status_lo | ((u64)hv_status_hi << 32);
136 #endif /* !x86_64 */
137 }
138
139 #ifdef CONFIG_X86_64
read_hv_clock_tsc(struct clocksource * arg)140 static cycle_t read_hv_clock_tsc(struct clocksource *arg)
141 {
142 cycle_t current_tick;
143 struct ms_hyperv_tsc_page *tsc_pg = hv_context.tsc_page;
144
145 if (tsc_pg->tsc_sequence != -1) {
146 /*
147 * Use the tsc page to compute the value.
148 */
149
150 while (1) {
151 cycle_t tmp;
152 u32 sequence = tsc_pg->tsc_sequence;
153 u64 cur_tsc;
154 u64 scale = tsc_pg->tsc_scale;
155 s64 offset = tsc_pg->tsc_offset;
156
157 rdtscll(cur_tsc);
158 /* current_tick = ((cur_tsc *scale) >> 64) + offset */
159 asm("mulq %3"
160 : "=d" (current_tick), "=a" (tmp)
161 : "a" (cur_tsc), "r" (scale));
162
163 current_tick += offset;
164 if (tsc_pg->tsc_sequence == sequence)
165 return current_tick;
166
167 if (tsc_pg->tsc_sequence != -1)
168 continue;
169 /*
170 * Fallback using MSR method.
171 */
172 break;
173 }
174 }
175 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
176 return current_tick;
177 }
178
179 static struct clocksource hyperv_cs_tsc = {
180 .name = "hyperv_clocksource_tsc_page",
181 .rating = 425,
182 .read = read_hv_clock_tsc,
183 .mask = CLOCKSOURCE_MASK(64),
184 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
185 };
186 #endif
187
188
189 /*
190 * hv_init - Main initialization routine.
191 *
192 * This routine must be called before any other routines in here are called
193 */
hv_init(void)194 int hv_init(void)
195 {
196 int max_leaf;
197 union hv_x64_msr_hypercall_contents hypercall_msr;
198 void *virtaddr = NULL;
199
200 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
201 memset(hv_context.synic_message_page, 0,
202 sizeof(void *) * NR_CPUS);
203 memset(hv_context.post_msg_page, 0,
204 sizeof(void *) * NR_CPUS);
205 memset(hv_context.vp_index, 0,
206 sizeof(int) * NR_CPUS);
207 memset(hv_context.event_dpc, 0,
208 sizeof(void *) * NR_CPUS);
209 memset(hv_context.clk_evt, 0,
210 sizeof(void *) * NR_CPUS);
211
212 max_leaf = query_hypervisor_info();
213
214 /*
215 * Write our OS ID.
216 */
217 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
218 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
219
220 /* See if the hypercall page is already set */
221 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
222
223 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
224
225 if (!virtaddr)
226 goto cleanup;
227
228 hypercall_msr.enable = 1;
229
230 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
231 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
232
233 /* Confirm that hypercall page did get setup. */
234 hypercall_msr.as_uint64 = 0;
235 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
236
237 if (!hypercall_msr.enable)
238 goto cleanup;
239
240 hv_context.hypercall_page = virtaddr;
241
242 #ifdef CONFIG_X86_64
243 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
244 union hv_x64_msr_hypercall_contents tsc_msr;
245 void *va_tsc;
246
247 va_tsc = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
248 if (!va_tsc)
249 goto cleanup;
250 hv_context.tsc_page = va_tsc;
251
252 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
253
254 tsc_msr.enable = 1;
255 tsc_msr.guest_physical_address = vmalloc_to_pfn(va_tsc);
256
257 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
258 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
259 }
260 #endif
261 return 0;
262
263 cleanup:
264 if (virtaddr) {
265 if (hypercall_msr.enable) {
266 hypercall_msr.as_uint64 = 0;
267 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
268 }
269
270 vfree(virtaddr);
271 }
272
273 return -ENOTSUPP;
274 }
275
276 /*
277 * hv_cleanup - Cleanup routine.
278 *
279 * This routine is called normally during driver unloading or exiting.
280 */
hv_cleanup(bool crash)281 void hv_cleanup(bool crash)
282 {
283 union hv_x64_msr_hypercall_contents hypercall_msr;
284
285 /* Reset our OS id */
286 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
287
288 if (hv_context.hypercall_page) {
289 hypercall_msr.as_uint64 = 0;
290 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
291 if (!crash)
292 vfree(hv_context.hypercall_page);
293 hv_context.hypercall_page = NULL;
294 }
295
296 #ifdef CONFIG_X86_64
297 /*
298 * Cleanup the TSC page based CS.
299 */
300 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
301 /*
302 * Crash can happen in an interrupt context and unregistering
303 * a clocksource is impossible and redundant in this case.
304 */
305 if (!oops_in_progress) {
306 clocksource_change_rating(&hyperv_cs_tsc, 10);
307 clocksource_unregister(&hyperv_cs_tsc);
308 }
309
310 hypercall_msr.as_uint64 = 0;
311 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
312 if (!crash) {
313 vfree(hv_context.tsc_page);
314 hv_context.tsc_page = NULL;
315 }
316 }
317 #endif
318 }
319
320 /*
321 * hv_post_message - Post a message using the hypervisor message IPC.
322 *
323 * This involves a hypercall.
324 */
hv_post_message(union hv_connection_id connection_id,enum hv_message_type message_type,void * payload,size_t payload_size)325 int hv_post_message(union hv_connection_id connection_id,
326 enum hv_message_type message_type,
327 void *payload, size_t payload_size)
328 {
329
330 struct hv_input_post_message *aligned_msg;
331 u16 status;
332
333 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
334 return -EMSGSIZE;
335
336 aligned_msg = (struct hv_input_post_message *)
337 hv_context.post_msg_page[get_cpu()];
338
339 aligned_msg->connectionid = connection_id;
340 aligned_msg->reserved = 0;
341 aligned_msg->message_type = message_type;
342 aligned_msg->payload_size = payload_size;
343 memcpy((void *)aligned_msg->payload, payload, payload_size);
344
345 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
346 & 0xFFFF;
347
348 put_cpu();
349 return status;
350 }
351
352
353 /*
354 * hv_signal_event -
355 * Signal an event on the specified connection using the hypervisor event IPC.
356 *
357 * This involves a hypercall.
358 */
hv_signal_event(void * con_id)359 u16 hv_signal_event(void *con_id)
360 {
361 u16 status;
362
363 status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
364
365 return status;
366 }
367
hv_ce_set_next_event(unsigned long delta,struct clock_event_device * evt)368 static int hv_ce_set_next_event(unsigned long delta,
369 struct clock_event_device *evt)
370 {
371 cycle_t current_tick;
372
373 WARN_ON(!clockevent_state_oneshot(evt));
374
375 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
376 current_tick += delta;
377 wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
378 return 0;
379 }
380
hv_ce_shutdown(struct clock_event_device * evt)381 static int hv_ce_shutdown(struct clock_event_device *evt)
382 {
383 wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
384 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
385
386 return 0;
387 }
388
hv_ce_set_oneshot(struct clock_event_device * evt)389 static int hv_ce_set_oneshot(struct clock_event_device *evt)
390 {
391 union hv_timer_config timer_cfg;
392
393 timer_cfg.enable = 1;
394 timer_cfg.auto_enable = 1;
395 timer_cfg.sintx = VMBUS_MESSAGE_SINT;
396 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
397
398 return 0;
399 }
400
hv_init_clockevent_device(struct clock_event_device * dev,int cpu)401 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
402 {
403 dev->name = "Hyper-V clockevent";
404 dev->features = CLOCK_EVT_FEAT_ONESHOT;
405 dev->cpumask = cpumask_of(cpu);
406 dev->rating = 1000;
407 /*
408 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
409 * result in clockevents_config_and_register() taking additional
410 * references to the hv_vmbus module making it impossible to unload.
411 */
412
413 dev->set_state_shutdown = hv_ce_shutdown;
414 dev->set_state_oneshot = hv_ce_set_oneshot;
415 dev->set_next_event = hv_ce_set_next_event;
416 }
417
418
hv_synic_alloc(void)419 int hv_synic_alloc(void)
420 {
421 size_t size = sizeof(struct tasklet_struct);
422 size_t ced_size = sizeof(struct clock_event_device);
423 int cpu;
424
425 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
426 GFP_ATOMIC);
427 if (hv_context.hv_numa_map == NULL) {
428 pr_err("Unable to allocate NUMA map\n");
429 goto err;
430 }
431
432 for_each_present_cpu(cpu) {
433 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
434 if (hv_context.event_dpc[cpu] == NULL) {
435 pr_err("Unable to allocate event dpc\n");
436 goto err;
437 }
438 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
439
440 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
441 if (hv_context.clk_evt[cpu] == NULL) {
442 pr_err("Unable to allocate clock event device\n");
443 goto err;
444 }
445
446 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
447
448 hv_context.synic_message_page[cpu] =
449 (void *)get_zeroed_page(GFP_ATOMIC);
450
451 if (hv_context.synic_message_page[cpu] == NULL) {
452 pr_err("Unable to allocate SYNIC message page\n");
453 goto err;
454 }
455
456 hv_context.synic_event_page[cpu] =
457 (void *)get_zeroed_page(GFP_ATOMIC);
458
459 if (hv_context.synic_event_page[cpu] == NULL) {
460 pr_err("Unable to allocate SYNIC event page\n");
461 goto err;
462 }
463
464 hv_context.post_msg_page[cpu] =
465 (void *)get_zeroed_page(GFP_ATOMIC);
466
467 if (hv_context.post_msg_page[cpu] == NULL) {
468 pr_err("Unable to allocate post msg page\n");
469 goto err;
470 }
471
472 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
473 }
474
475 return 0;
476 err:
477 return -ENOMEM;
478 }
479
hv_synic_free_cpu(int cpu)480 static void hv_synic_free_cpu(int cpu)
481 {
482 kfree(hv_context.event_dpc[cpu]);
483 kfree(hv_context.clk_evt[cpu]);
484 if (hv_context.synic_event_page[cpu])
485 free_page((unsigned long)hv_context.synic_event_page[cpu]);
486 if (hv_context.synic_message_page[cpu])
487 free_page((unsigned long)hv_context.synic_message_page[cpu]);
488 if (hv_context.post_msg_page[cpu])
489 free_page((unsigned long)hv_context.post_msg_page[cpu]);
490 }
491
hv_synic_free(void)492 void hv_synic_free(void)
493 {
494 int cpu;
495
496 kfree(hv_context.hv_numa_map);
497 for_each_present_cpu(cpu)
498 hv_synic_free_cpu(cpu);
499 }
500
501 /*
502 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
503 *
504 * If it is already initialized by another entity (ie x2v shim), we need to
505 * retrieve the initialized message and event pages. Otherwise, we create and
506 * initialize the message and event pages.
507 */
hv_synic_init(void * arg)508 void hv_synic_init(void *arg)
509 {
510 u64 version;
511 union hv_synic_simp simp;
512 union hv_synic_siefp siefp;
513 union hv_synic_sint shared_sint;
514 union hv_synic_scontrol sctrl;
515 u64 vp_index;
516
517 int cpu = smp_processor_id();
518
519 if (!hv_context.hypercall_page)
520 return;
521
522 /* Check the version */
523 rdmsrl(HV_X64_MSR_SVERSION, version);
524
525 /* Setup the Synic's message page */
526 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
527 simp.simp_enabled = 1;
528 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
529 >> PAGE_SHIFT;
530
531 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
532
533 /* Setup the Synic's event page */
534 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
535 siefp.siefp_enabled = 1;
536 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
537 >> PAGE_SHIFT;
538
539 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
540
541 /* Setup the shared SINT. */
542 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
543
544 shared_sint.as_uint64 = 0;
545 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
546 shared_sint.masked = false;
547 shared_sint.auto_eoi = true;
548
549 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
550
551 /* Enable the global synic bit */
552 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
553 sctrl.enable = 1;
554
555 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
556
557 hv_context.synic_initialized = true;
558
559 /*
560 * Setup the mapping between Hyper-V's notion
561 * of cpuid and Linux' notion of cpuid.
562 * This array will be indexed using Linux cpuid.
563 */
564 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
565 hv_context.vp_index[cpu] = (u32)vp_index;
566
567 /*
568 * Register the per-cpu clockevent source.
569 */
570 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
571 clockevents_config_and_register(hv_context.clk_evt[cpu],
572 HV_TIMER_FREQUENCY,
573 HV_MIN_DELTA_TICKS,
574 HV_MAX_MAX_DELTA_TICKS);
575 return;
576 }
577
578 /*
579 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
580 */
hv_synic_clockevents_cleanup(void)581 void hv_synic_clockevents_cleanup(void)
582 {
583 int cpu;
584
585 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
586 return;
587
588 for_each_online_cpu(cpu)
589 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
590 }
591
592 /*
593 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
594 */
hv_synic_cleanup(void * arg)595 void hv_synic_cleanup(void *arg)
596 {
597 union hv_synic_sint shared_sint;
598 union hv_synic_simp simp;
599 union hv_synic_siefp siefp;
600 union hv_synic_scontrol sctrl;
601 int cpu = smp_processor_id();
602
603 if (!hv_context.synic_initialized)
604 return;
605
606 /* Turn off clockevent device */
607 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
608 hv_ce_shutdown(hv_context.clk_evt[cpu]);
609
610 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
611
612 shared_sint.masked = 1;
613
614 /* Need to correctly cleanup in the case of SMP!!! */
615 /* Disable the interrupt */
616 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
617
618 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
619 simp.simp_enabled = 0;
620 simp.base_simp_gpa = 0;
621
622 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
623
624 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
625 siefp.siefp_enabled = 0;
626 siefp.base_siefp_gpa = 0;
627
628 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
629
630 /* Disable the global synic bit */
631 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
632 sctrl.enable = 0;
633 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
634 }
635