Merge branch 'linux-linaro-lsk-v4.4' into linux-linaro-lsk-v4.4-android
[firefly-linux-kernel-4.4.55.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         }
313 #endif
314 }
315
316 /*
317  * hv_post_message - Post a message using the hypervisor message IPC.
318  *
319  * This involves a hypercall.
320  */
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)
324 {
325
326         struct hv_input_post_message *aligned_msg;
327         u16 status;
328
329         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
330                 return -EMSGSIZE;
331
332         aligned_msg = (struct hv_input_post_message *)
333                         hv_context.post_msg_page[get_cpu()];
334
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);
340
341         status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
342                 & 0xFFFF;
343
344         put_cpu();
345         return status;
346 }
347
348
349 /*
350  * hv_signal_event -
351  * Signal an event on the specified connection using the hypervisor event IPC.
352  *
353  * This involves a hypercall.
354  */
355 u16 hv_signal_event(void *con_id)
356 {
357         u16 status;
358
359         status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
360
361         return status;
362 }
363
364 static int hv_ce_set_next_event(unsigned long delta,
365                                 struct clock_event_device *evt)
366 {
367         cycle_t current_tick;
368
369         WARN_ON(!clockevent_state_oneshot(evt));
370
371         rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
372         current_tick += delta;
373         wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
374         return 0;
375 }
376
377 static int hv_ce_shutdown(struct clock_event_device *evt)
378 {
379         wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
380         wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
381
382         return 0;
383 }
384
385 static int hv_ce_set_oneshot(struct clock_event_device *evt)
386 {
387         union hv_timer_config timer_cfg;
388
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);
393
394         return 0;
395 }
396
397 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
398 {
399         dev->name = "Hyper-V clockevent";
400         dev->features = CLOCK_EVT_FEAT_ONESHOT;
401         dev->cpumask = cpumask_of(cpu);
402         dev->rating = 1000;
403         /*
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.
407          */
408
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;
412 }
413
414
415 int hv_synic_alloc(void)
416 {
417         size_t size = sizeof(struct tasklet_struct);
418         size_t ced_size = sizeof(struct clock_event_device);
419         int cpu;
420
421         hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
422                                          GFP_ATOMIC);
423         if (hv_context.hv_numa_map == NULL) {
424                 pr_err("Unable to allocate NUMA map\n");
425                 goto err;
426         }
427
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");
432                         goto err;
433                 }
434                 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
435
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");
439                         goto err;
440                 }
441
442                 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
443
444                 hv_context.synic_message_page[cpu] =
445                         (void *)get_zeroed_page(GFP_ATOMIC);
446
447                 if (hv_context.synic_message_page[cpu] == NULL) {
448                         pr_err("Unable to allocate SYNIC message page\n");
449                         goto err;
450                 }
451
452                 hv_context.synic_event_page[cpu] =
453                         (void *)get_zeroed_page(GFP_ATOMIC);
454
455                 if (hv_context.synic_event_page[cpu] == NULL) {
456                         pr_err("Unable to allocate SYNIC event page\n");
457                         goto err;
458                 }
459
460                 hv_context.post_msg_page[cpu] =
461                         (void *)get_zeroed_page(GFP_ATOMIC);
462
463                 if (hv_context.post_msg_page[cpu] == NULL) {
464                         pr_err("Unable to allocate post msg page\n");
465                         goto err;
466                 }
467
468                 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
469         }
470
471         return 0;
472 err:
473         return -ENOMEM;
474 }
475
476 static void hv_synic_free_cpu(int cpu)
477 {
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]);
486 }
487
488 void hv_synic_free(void)
489 {
490         int cpu;
491
492         kfree(hv_context.hv_numa_map);
493         for_each_present_cpu(cpu)
494                 hv_synic_free_cpu(cpu);
495 }
496
497 /*
498  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
499  *
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.
503  */
504 void hv_synic_init(void *arg)
505 {
506         u64 version;
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;
511         u64 vp_index;
512
513         int cpu = smp_processor_id();
514
515         if (!hv_context.hypercall_page)
516                 return;
517
518         /* Check the version */
519         rdmsrl(HV_X64_MSR_SVERSION, version);
520
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])
525                 >> PAGE_SHIFT;
526
527         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
528
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])
533                 >> PAGE_SHIFT;
534
535         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
536
537         /* Setup the shared SINT. */
538         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
539
540         shared_sint.as_uint64 = 0;
541         shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
542         shared_sint.masked = false;
543         shared_sint.auto_eoi = true;
544
545         wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
546
547         /* Enable the global synic bit */
548         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
549         sctrl.enable = 1;
550
551         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
552
553         hv_context.synic_initialized = true;
554
555         /*
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.
559          */
560         rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
561         hv_context.vp_index[cpu] = (u32)vp_index;
562
563         /*
564          * Register the per-cpu clockevent source.
565          */
566         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
567                 clockevents_config_and_register(hv_context.clk_evt[cpu],
568                                                 HV_TIMER_FREQUENCY,
569                                                 HV_MIN_DELTA_TICKS,
570                                                 HV_MAX_MAX_DELTA_TICKS);
571         return;
572 }
573
574 /*
575  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
576  */
577 void hv_synic_clockevents_cleanup(void)
578 {
579         int cpu;
580
581         if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
582                 return;
583
584         for_each_online_cpu(cpu)
585                 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
586 }
587
588 /*
589  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
590  */
591 void hv_synic_cleanup(void *arg)
592 {
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();
598
599         if (!hv_context.synic_initialized)
600                 return;
601
602         /* Turn off clockevent device */
603         if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
604                 hv_ce_shutdown(hv_context.clk_evt[cpu]);
605
606         rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
607
608         shared_sint.masked = 1;
609
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);
613
614         rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
615         simp.simp_enabled = 0;
616         simp.base_simp_gpa = 0;
617
618         wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
619
620         rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
621         siefp.siefp_enabled = 0;
622         siefp.base_siefp_gpa = 0;
623
624         wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
625
626         /* Disable the global synic bit */
627         rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
628         sctrl.enable = 0;
629         wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
630 }