Linux-libre 4.4.62-gnu
[librecmc/linux-libre.git] / 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         .hypercall_page         = NULL,
40 };
41
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
45
46 /*
47  * query_hypervisor_info - Get version info of the windows hypervisor
48  */
49 unsigned int host_info_eax;
50 unsigned int host_info_ebx;
51 unsigned int host_info_ecx;
52 unsigned int host_info_edx;
53
54 static int query_hypervisor_info(void)
55 {
56         unsigned int eax;
57         unsigned int ebx;
58         unsigned int ecx;
59         unsigned int edx;
60         unsigned int max_leaf;
61         unsigned int op;
62
63         /*
64         * Its assumed that this is called after confirming that Viridian
65         * is present. Query id and revision.
66         */
67         eax = 0;
68         ebx = 0;
69         ecx = 0;
70         edx = 0;
71         op = HVCPUID_VENDOR_MAXFUNCTION;
72         cpuid(op, &eax, &ebx, &ecx, &edx);
73
74         max_leaf = eax;
75
76         if (max_leaf >= HVCPUID_VERSION) {
77                 eax = 0;
78                 ebx = 0;
79                 ecx = 0;
80                 edx = 0;
81                 op = HVCPUID_VERSION;
82                 cpuid(op, &eax, &ebx, &ecx, &edx);
83                 host_info_eax = eax;
84                 host_info_ebx = ebx;
85                 host_info_ecx = ecx;
86                 host_info_edx = edx;
87         }
88         return max_leaf;
89 }
90
91 /*
92  * do_hypercall- Invoke the specified hypercall
93  */
94 static u64 do_hypercall(u64 control, void *input, void *output)
95 {
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;
99 #ifdef CONFIG_X86_64
100         u64 hv_status = 0;
101
102         if (!hypercall_page)
103                 return (u64)ULLONG_MAX;
104
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));
109
110         return hv_status;
111
112 #else
113
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;
122
123         if (!hypercall_page)
124                 return (u64)ULLONG_MAX;
125
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));
131
132         return hv_status_lo | ((u64)hv_status_hi << 32);
133 #endif /* !x86_64 */
134 }
135
136 #ifdef CONFIG_X86_64
137 static cycle_t read_hv_clock_tsc(struct clocksource *arg)
138 {
139         cycle_t current_tick;
140         struct ms_hyperv_tsc_page *tsc_pg = hv_context.tsc_page;
141
142         if (tsc_pg->tsc_sequence != -1) {
143                 /*
144                  * Use the tsc page to compute the value.
145                  */
146
147                 while (1) {
148                         cycle_t tmp;
149                         u32 sequence = tsc_pg->tsc_sequence;
150                         u64 cur_tsc;
151                         u64 scale = tsc_pg->tsc_scale;
152                         s64 offset = tsc_pg->tsc_offset;
153
154                         rdtscll(cur_tsc);
155                         /* current_tick = ((cur_tsc *scale) >> 64) + offset */
156                         asm("mulq %3"
157                                 : "=d" (current_tick), "=a" (tmp)
158                                 : "a" (cur_tsc), "r" (scale));
159
160                         current_tick += offset;
161                         if (tsc_pg->tsc_sequence == sequence)
162                                 return current_tick;
163
164                         if (tsc_pg->tsc_sequence != -1)
165                                 continue;
166                         /*
167                          * Fallback using MSR method.
168                          */
169                         break;
170                 }
171         }
172         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
173         return current_tick;
174 }
175
176 static struct clocksource hyperv_cs_tsc = {
177                 .name           = "hyperv_clocksource_tsc_page",
178                 .rating         = 425,
179                 .read           = read_hv_clock_tsc,
180                 .mask           = CLOCKSOURCE_MASK(64),
181                 .flags          = CLOCK_SOURCE_IS_CONTINUOUS,
182 };
183 #endif
184
185
186 /*
187  * hv_init - Main initialization routine.
188  *
189  * This routine must be called before any other routines in here are called
190  */
191 int hv_init(void)
192 {
193         int max_leaf;
194         union hv_x64_msr_hypercall_contents hypercall_msr;
195         union hv_x64_msr_hypercall_contents tsc_msr;
196         void *virtaddr = NULL;
197         void *va_tsc = NULL;
198
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);
210
211         max_leaf = query_hypervisor_info();
212
213         /*
214          * Write our OS ID.
215          */
216         hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
217         wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
218
219         /* See if the hypercall page is already set */
220         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
221
222         virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
223
224         if (!virtaddr)
225                 goto cleanup;
226
227         hypercall_msr.enable = 1;
228
229         hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
230         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
231
232         /* Confirm that hypercall page did get setup. */
233         hypercall_msr.as_uint64 = 0;
234         rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
235
236         if (!hypercall_msr.enable)
237                 goto cleanup;
238
239         hv_context.hypercall_page = virtaddr;
240
241 #ifdef CONFIG_X86_64
242         if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
243                 va_tsc = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
244                 if (!va_tsc)
245                         goto cleanup;
246                 hv_context.tsc_page = va_tsc;
247
248                 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
249
250                 tsc_msr.enable = 1;
251                 tsc_msr.guest_physical_address = vmalloc_to_pfn(va_tsc);
252
253                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
254                 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
255         }
256 #endif
257         return 0;
258
259 cleanup:
260         if (virtaddr) {
261                 if (hypercall_msr.enable) {
262                         hypercall_msr.as_uint64 = 0;
263                         wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
264                 }
265
266                 vfree(virtaddr);
267         }
268
269         return -ENOTSUPP;
270 }
271
272 /*
273  * hv_cleanup - Cleanup routine.
274  *
275  * This routine is called normally during driver unloading or exiting.
276  */
277 void hv_cleanup(bool crash)
278 {
279         union hv_x64_msr_hypercall_contents hypercall_msr;
280
281         /* Reset our OS id */
282         wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
283
284         if (hv_context.hypercall_page) {
285                 hypercall_msr.as_uint64 = 0;
286                 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
287                 if (!crash)
288                         vfree(hv_context.hypercall_page);
289                 hv_context.hypercall_page = NULL;
290         }
291
292 #ifdef CONFIG_X86_64
293         /*
294          * Cleanup the TSC page based CS.
295          */
296         if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
297                 /*
298                  * Crash can happen in an interrupt context and unregistering
299                  * a clocksource is impossible and redundant in this case.
300                  */
301                 if (!oops_in_progress) {
302                         clocksource_change_rating(&hyperv_cs_tsc, 10);
303                         clocksource_unregister(&hyperv_cs_tsc);
304                 }
305
306                 hypercall_msr.as_uint64 = 0;
307                 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
308                 if (!crash)
309                         vfree(hv_context.tsc_page);
310                 hv_context.tsc_page = NULL;
311         }
312 #endif
313 }
314
315 /*
316  * hv_post_message - Post a message using the hypervisor message IPC.
317  *
318  * This involves a hypercall.
319  */
320 int hv_post_message(union hv_connection_id connection_id,
321                   enum hv_message_type message_type,
322                   void *payload, size_t payload_size)
323 {
324
325         struct hv_input_post_message *aligned_msg;
326         u16 status;
327
328         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
329                 return -EMSGSIZE;
330
331         aligned_msg = (struct hv_input_post_message *)
332                         hv_context.post_msg_page[get_cpu()];
333
334         aligned_msg->connectionid = connection_id;
335         aligned_msg->reserved = 0;
336         aligned_msg->message_type = message_type;
337         aligned_msg->payload_size = payload_size;
338         memcpy((void *)aligned_msg->payload, payload, payload_size);
339
340         status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
341                 & 0xFFFF;
342
343         put_cpu();
344         return status;
345 }
346
347
348 /*
349  * hv_signal_event -
350  * Signal an event on the specified connection using the hypervisor event IPC.
351  *
352  * This involves a hypercall.
353  */
354 u16 hv_signal_event(void *con_id)
355 {
356         u16 status;
357
358         status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
359
360         return status;
361 }
362
363 static int hv_ce_set_next_event(unsigned long delta,
364                                 struct clock_event_device *evt)
365 {
366         cycle_t current_tick;
367
368         WARN_ON(!clockevent_state_oneshot(evt));
369
370         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
371         current_tick += delta;
372         wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
373         return 0;
374 }
375
376 static int hv_ce_shutdown(struct clock_event_device *evt)
377 {
378         wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
379         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
380
381         return 0;
382 }
383
384 static int hv_ce_set_oneshot(struct clock_event_device *evt)
385 {
386         union hv_timer_config timer_cfg;
387
388         timer_cfg.enable = 1;
389         timer_cfg.auto_enable = 1;
390         timer_cfg.sintx = VMBUS_MESSAGE_SINT;
391         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
392
393         return 0;
394 }
395
396 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
397 {
398         dev->name = "Hyper-V clockevent";
399         dev->features = CLOCK_EVT_FEAT_ONESHOT;
400         dev->cpumask = cpumask_of(cpu);
401         dev->rating = 1000;
402         /*
403          * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
404          * result in clockevents_config_and_register() taking additional
405          * references to the hv_vmbus module making it impossible to unload.
406          */
407
408         dev->set_state_shutdown = hv_ce_shutdown;
409         dev->set_state_oneshot = hv_ce_set_oneshot;
410         dev->set_next_event = hv_ce_set_next_event;
411 }
412
413
414 int hv_synic_alloc(void)
415 {
416         size_t size = sizeof(struct tasklet_struct);
417         size_t ced_size = sizeof(struct clock_event_device);
418         int cpu;
419
420         hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
421                                          GFP_ATOMIC);
422         if (hv_context.hv_numa_map == NULL) {
423                 pr_err("Unable to allocate NUMA map\n");
424                 goto err;
425         }
426
427         for_each_present_cpu(cpu) {
428                 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
429                 if (hv_context.event_dpc[cpu] == NULL) {
430                         pr_err("Unable to allocate event dpc\n");
431                         goto err;
432                 }
433                 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
434
435                 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
436                 if (hv_context.clk_evt[cpu] == NULL) {
437                         pr_err("Unable to allocate clock event device\n");
438                         goto err;
439                 }
440
441                 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
442
443                 hv_context.synic_message_page[cpu] =
444                         (void *)get_zeroed_page(GFP_ATOMIC);
445
446                 if (hv_context.synic_message_page[cpu] == NULL) {
447                         pr_err("Unable to allocate SYNIC message page\n");
448                         goto err;
449                 }
450
451                 hv_context.synic_event_page[cpu] =
452                         (void *)get_zeroed_page(GFP_ATOMIC);
453
454                 if (hv_context.synic_event_page[cpu] == NULL) {
455                         pr_err("Unable to allocate SYNIC event page\n");
456                         goto err;
457                 }
458
459                 hv_context.post_msg_page[cpu] =
460                         (void *)get_zeroed_page(GFP_ATOMIC);
461
462                 if (hv_context.post_msg_page[cpu] == NULL) {
463                         pr_err("Unable to allocate post msg page\n");
464                         goto err;
465                 }
466
467                 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
468         }
469
470         return 0;
471 err:
472         return -ENOMEM;
473 }
474
475 static void hv_synic_free_cpu(int cpu)
476 {
477         kfree(hv_context.event_dpc[cpu]);
478         kfree(hv_context.clk_evt[cpu]);
479         if (hv_context.synic_event_page[cpu])
480                 free_page((unsigned long)hv_context.synic_event_page[cpu]);
481         if (hv_context.synic_message_page[cpu])
482                 free_page((unsigned long)hv_context.synic_message_page[cpu]);
483         if (hv_context.post_msg_page[cpu])
484                 free_page((unsigned long)hv_context.post_msg_page[cpu]);
485 }
486
487 void hv_synic_free(void)
488 {
489         int cpu;
490
491         kfree(hv_context.hv_numa_map);
492         for_each_present_cpu(cpu)
493                 hv_synic_free_cpu(cpu);
494 }
495
496 /*
497  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
498  *
499  * If it is already initialized by another entity (ie x2v shim), we need to
500  * retrieve the initialized message and event pages.  Otherwise, we create and
501  * initialize the message and event pages.
502  */
503 void hv_synic_init(void *arg)
504 {
505         u64 version;
506         union hv_synic_simp simp;
507         union hv_synic_siefp siefp;
508         union hv_synic_sint shared_sint;
509         union hv_synic_scontrol sctrl;
510         u64 vp_index;
511
512         int cpu = smp_processor_id();
513
514         if (!hv_context.hypercall_page)
515                 return;
516
517         /* Check the version */
518         rdmsrl(HV_X64_MSR_SVERSION, version);
519
520         /* Setup the Synic's message page */
521         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
522         simp.simp_enabled = 1;
523         simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
524                 >> PAGE_SHIFT;
525
526         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
527
528         /* Setup the Synic's event page */
529         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
530         siefp.siefp_enabled = 1;
531         siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
532                 >> PAGE_SHIFT;
533
534         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
535
536         /* Setup the shared SINT. */
537         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
538
539         shared_sint.as_uint64 = 0;
540         shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
541         shared_sint.masked = false;
542         shared_sint.auto_eoi = true;
543
544         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
545
546         /* Enable the global synic bit */
547         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
548         sctrl.enable = 1;
549
550         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
551
552         hv_context.synic_initialized = true;
553
554         /*
555          * Setup the mapping between Hyper-V's notion
556          * of cpuid and Linux' notion of cpuid.
557          * This array will be indexed using Linux cpuid.
558          */
559         rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
560         hv_context.vp_index[cpu] = (u32)vp_index;
561
562         /*
563          * Register the per-cpu clockevent source.
564          */
565         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
566                 clockevents_config_and_register(hv_context.clk_evt[cpu],
567                                                 HV_TIMER_FREQUENCY,
568                                                 HV_MIN_DELTA_TICKS,
569                                                 HV_MAX_MAX_DELTA_TICKS);
570         return;
571 }
572
573 /*
574  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
575  */
576 void hv_synic_clockevents_cleanup(void)
577 {
578         int cpu;
579
580         if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
581                 return;
582
583         for_each_online_cpu(cpu)
584                 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
585 }
586
587 /*
588  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
589  */
590 void hv_synic_cleanup(void *arg)
591 {
592         union hv_synic_sint shared_sint;
593         union hv_synic_simp simp;
594         union hv_synic_siefp siefp;
595         union hv_synic_scontrol sctrl;
596         int cpu = smp_processor_id();
597
598         if (!hv_context.synic_initialized)
599                 return;
600
601         /* Turn off clockevent device */
602         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
603                 hv_ce_shutdown(hv_context.clk_evt[cpu]);
604
605         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
606
607         shared_sint.masked = 1;
608
609         /* Need to correctly cleanup in the case of SMP!!! */
610         /* Disable the interrupt */
611         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
612
613         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
614         simp.simp_enabled = 0;
615         simp.base_simp_gpa = 0;
616
617         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
618
619         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
620         siefp.siefp_enabled = 0;
621         siefp.base_siefp_gpa = 0;
622
623         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
624
625         /* Disable the global synic bit */
626         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
627         sctrl.enable = 0;
628         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
629 }