xref: /linux/drivers/hv/hv.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
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 #ifdef CONFIG_X86_64
97 	u64 hv_status = 0;
98 	u64 input_address = (input) ? virt_to_phys(input) : 0;
99 	u64 output_address = (output) ? virt_to_phys(output) : 0;
100 	void *hypercall_page = hv_context.hypercall_page;
101 
102 	__asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
103 	__asm__ __volatile__("call *%3" : "=a" (hv_status) :
104 			     "c" (control), "d" (input_address),
105 			     "m" (hypercall_page));
106 
107 	return hv_status;
108 
109 #else
110 
111 	u32 control_hi = control >> 32;
112 	u32 control_lo = control & 0xFFFFFFFF;
113 	u32 hv_status_hi = 1;
114 	u32 hv_status_lo = 1;
115 	u64 input_address = (input) ? virt_to_phys(input) : 0;
116 	u32 input_address_hi = input_address >> 32;
117 	u32 input_address_lo = input_address & 0xFFFFFFFF;
118 	u64 output_address = (output) ? virt_to_phys(output) : 0;
119 	u32 output_address_hi = output_address >> 32;
120 	u32 output_address_lo = output_address & 0xFFFFFFFF;
121 	void *hypercall_page = hv_context.hypercall_page;
122 
123 	__asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
124 			      "=a"(hv_status_lo) : "d" (control_hi),
125 			      "a" (control_lo), "b" (input_address_hi),
126 			      "c" (input_address_lo), "D"(output_address_hi),
127 			      "S"(output_address_lo), "m" (hypercall_page));
128 
129 	return hv_status_lo | ((u64)hv_status_hi << 32);
130 #endif /* !x86_64 */
131 }
132 
133 /*
134  * hv_init - Main initialization routine.
135  *
136  * This routine must be called before any other routines in here are called
137  */
138 int hv_init(void)
139 {
140 	int max_leaf;
141 	union hv_x64_msr_hypercall_contents hypercall_msr;
142 	void *virtaddr = NULL;
143 
144 	memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
145 	memset(hv_context.synic_message_page, 0,
146 	       sizeof(void *) * NR_CPUS);
147 	memset(hv_context.post_msg_page, 0,
148 	       sizeof(void *) * NR_CPUS);
149 	memset(hv_context.vp_index, 0,
150 	       sizeof(int) * NR_CPUS);
151 	memset(hv_context.event_dpc, 0,
152 	       sizeof(void *) * NR_CPUS);
153 	memset(hv_context.clk_evt, 0,
154 	       sizeof(void *) * NR_CPUS);
155 
156 	max_leaf = query_hypervisor_info();
157 
158 	/*
159 	 * Write our OS ID.
160 	 */
161 	hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
162 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
163 
164 	/* See if the hypercall page is already set */
165 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
166 
167 	virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC);
168 
169 	if (!virtaddr)
170 		goto cleanup;
171 
172 	hypercall_msr.enable = 1;
173 
174 	hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
175 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
176 
177 	/* Confirm that hypercall page did get setup. */
178 	hypercall_msr.as_uint64 = 0;
179 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
180 
181 	if (!hypercall_msr.enable)
182 		goto cleanup;
183 
184 	hv_context.hypercall_page = virtaddr;
185 
186 	return 0;
187 
188 cleanup:
189 	if (virtaddr) {
190 		if (hypercall_msr.enable) {
191 			hypercall_msr.as_uint64 = 0;
192 			wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
193 		}
194 
195 		vfree(virtaddr);
196 	}
197 
198 	return -ENOTSUPP;
199 }
200 
201 /*
202  * hv_cleanup - Cleanup routine.
203  *
204  * This routine is called normally during driver unloading or exiting.
205  */
206 void hv_cleanup(void)
207 {
208 	union hv_x64_msr_hypercall_contents hypercall_msr;
209 
210 	/* Reset our OS id */
211 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
212 
213 	if (hv_context.hypercall_page) {
214 		hypercall_msr.as_uint64 = 0;
215 		wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
216 		vfree(hv_context.hypercall_page);
217 		hv_context.hypercall_page = NULL;
218 	}
219 }
220 
221 /*
222  * hv_post_message - Post a message using the hypervisor message IPC.
223  *
224  * This involves a hypercall.
225  */
226 int hv_post_message(union hv_connection_id connection_id,
227 		  enum hv_message_type message_type,
228 		  void *payload, size_t payload_size)
229 {
230 
231 	struct hv_input_post_message *aligned_msg;
232 	u16 status;
233 
234 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
235 		return -EMSGSIZE;
236 
237 	aligned_msg = (struct hv_input_post_message *)
238 			hv_context.post_msg_page[get_cpu()];
239 
240 	aligned_msg->connectionid = connection_id;
241 	aligned_msg->reserved = 0;
242 	aligned_msg->message_type = message_type;
243 	aligned_msg->payload_size = payload_size;
244 	memcpy((void *)aligned_msg->payload, payload, payload_size);
245 
246 	status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
247 		& 0xFFFF;
248 
249 	put_cpu();
250 	return status;
251 }
252 
253 
254 /*
255  * hv_signal_event -
256  * Signal an event on the specified connection using the hypervisor event IPC.
257  *
258  * This involves a hypercall.
259  */
260 u16 hv_signal_event(void *con_id)
261 {
262 	u16 status;
263 
264 	status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
265 
266 	return status;
267 }
268 
269 static int hv_ce_set_next_event(unsigned long delta,
270 				struct clock_event_device *evt)
271 {
272 	cycle_t current_tick;
273 
274 	WARN_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT);
275 
276 	rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
277 	current_tick += delta;
278 	wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
279 	return 0;
280 }
281 
282 static void hv_ce_setmode(enum clock_event_mode mode,
283 			  struct clock_event_device *evt)
284 {
285 	union hv_timer_config timer_cfg;
286 
287 	switch (mode) {
288 	case CLOCK_EVT_MODE_PERIODIC:
289 		/* unsupported */
290 		break;
291 
292 	case CLOCK_EVT_MODE_ONESHOT:
293 		timer_cfg.enable = 1;
294 		timer_cfg.auto_enable = 1;
295 		timer_cfg.sintx = VMBUS_MESSAGE_SINT;
296 		wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
297 		break;
298 
299 	case CLOCK_EVT_MODE_UNUSED:
300 	case CLOCK_EVT_MODE_SHUTDOWN:
301 		wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
302 		wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
303 		break;
304 	case CLOCK_EVT_MODE_RESUME:
305 		break;
306 	}
307 }
308 
309 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
310 {
311 	dev->name = "Hyper-V clockevent";
312 	dev->features = CLOCK_EVT_FEAT_ONESHOT;
313 	dev->cpumask = cpumask_of(cpu);
314 	dev->rating = 1000;
315 	/*
316 	 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
317 	 * result in clockevents_config_and_register() taking additional
318 	 * references to the hv_vmbus module making it impossible to unload.
319 	 */
320 
321 	dev->set_mode = hv_ce_setmode;
322 	dev->set_next_event = hv_ce_set_next_event;
323 }
324 
325 
326 int hv_synic_alloc(void)
327 {
328 	size_t size = sizeof(struct tasklet_struct);
329 	size_t ced_size = sizeof(struct clock_event_device);
330 	int cpu;
331 
332 	for_each_online_cpu(cpu) {
333 		hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
334 		if (hv_context.event_dpc[cpu] == NULL) {
335 			pr_err("Unable to allocate event dpc\n");
336 			goto err;
337 		}
338 		tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
339 
340 		hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
341 		if (hv_context.clk_evt[cpu] == NULL) {
342 			pr_err("Unable to allocate clock event device\n");
343 			goto err;
344 		}
345 		hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
346 
347 		hv_context.synic_message_page[cpu] =
348 			(void *)get_zeroed_page(GFP_ATOMIC);
349 
350 		if (hv_context.synic_message_page[cpu] == NULL) {
351 			pr_err("Unable to allocate SYNIC message page\n");
352 			goto err;
353 		}
354 
355 		hv_context.synic_event_page[cpu] =
356 			(void *)get_zeroed_page(GFP_ATOMIC);
357 
358 		if (hv_context.synic_event_page[cpu] == NULL) {
359 			pr_err("Unable to allocate SYNIC event page\n");
360 			goto err;
361 		}
362 
363 		hv_context.post_msg_page[cpu] =
364 			(void *)get_zeroed_page(GFP_ATOMIC);
365 
366 		if (hv_context.post_msg_page[cpu] == NULL) {
367 			pr_err("Unable to allocate post msg page\n");
368 			goto err;
369 		}
370 	}
371 
372 	return 0;
373 err:
374 	return -ENOMEM;
375 }
376 
377 static void hv_synic_free_cpu(int cpu)
378 {
379 	kfree(hv_context.event_dpc[cpu]);
380 	kfree(hv_context.clk_evt[cpu]);
381 	if (hv_context.synic_event_page[cpu])
382 		free_page((unsigned long)hv_context.synic_event_page[cpu]);
383 	if (hv_context.synic_message_page[cpu])
384 		free_page((unsigned long)hv_context.synic_message_page[cpu]);
385 	if (hv_context.post_msg_page[cpu])
386 		free_page((unsigned long)hv_context.post_msg_page[cpu]);
387 }
388 
389 void hv_synic_free(void)
390 {
391 	int cpu;
392 
393 	for_each_online_cpu(cpu)
394 		hv_synic_free_cpu(cpu);
395 }
396 
397 /*
398  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
399  *
400  * If it is already initialized by another entity (ie x2v shim), we need to
401  * retrieve the initialized message and event pages.  Otherwise, we create and
402  * initialize the message and event pages.
403  */
404 void hv_synic_init(void *arg)
405 {
406 	u64 version;
407 	union hv_synic_simp simp;
408 	union hv_synic_siefp siefp;
409 	union hv_synic_sint shared_sint;
410 	union hv_synic_scontrol sctrl;
411 	u64 vp_index;
412 
413 	int cpu = smp_processor_id();
414 
415 	if (!hv_context.hypercall_page)
416 		return;
417 
418 	/* Check the version */
419 	rdmsrl(HV_X64_MSR_SVERSION, version);
420 
421 	/* Setup the Synic's message page */
422 	rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
423 	simp.simp_enabled = 1;
424 	simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
425 		>> PAGE_SHIFT;
426 
427 	wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
428 
429 	/* Setup the Synic's event page */
430 	rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
431 	siefp.siefp_enabled = 1;
432 	siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
433 		>> PAGE_SHIFT;
434 
435 	wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
436 
437 	/* Setup the shared SINT. */
438 	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
439 
440 	shared_sint.as_uint64 = 0;
441 	shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
442 	shared_sint.masked = false;
443 	shared_sint.auto_eoi = true;
444 
445 	wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
446 
447 	/* Enable the global synic bit */
448 	rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
449 	sctrl.enable = 1;
450 
451 	wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
452 
453 	hv_context.synic_initialized = true;
454 
455 	/*
456 	 * Setup the mapping between Hyper-V's notion
457 	 * of cpuid and Linux' notion of cpuid.
458 	 * This array will be indexed using Linux cpuid.
459 	 */
460 	rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
461 	hv_context.vp_index[cpu] = (u32)vp_index;
462 
463 	INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
464 
465 	/*
466 	 * Register the per-cpu clockevent source.
467 	 */
468 	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
469 		clockevents_config_and_register(hv_context.clk_evt[cpu],
470 						HV_TIMER_FREQUENCY,
471 						HV_MIN_DELTA_TICKS,
472 						HV_MAX_MAX_DELTA_TICKS);
473 	return;
474 }
475 
476 /*
477  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
478  */
479 void hv_synic_clockevents_cleanup(void)
480 {
481 	int cpu;
482 
483 	if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
484 		return;
485 
486 	for_each_online_cpu(cpu)
487 		clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
488 }
489 
490 /*
491  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
492  */
493 void hv_synic_cleanup(void *arg)
494 {
495 	union hv_synic_sint shared_sint;
496 	union hv_synic_simp simp;
497 	union hv_synic_siefp siefp;
498 	union hv_synic_scontrol sctrl;
499 	int cpu = smp_processor_id();
500 
501 	if (!hv_context.synic_initialized)
502 		return;
503 
504 	/* Turn off clockevent device */
505 	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
506 		hv_ce_setmode(CLOCK_EVT_MODE_SHUTDOWN,
507 			      hv_context.clk_evt[cpu]);
508 
509 	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
510 
511 	shared_sint.masked = 1;
512 
513 	/* Need to correctly cleanup in the case of SMP!!! */
514 	/* Disable the interrupt */
515 	wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
516 
517 	rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
518 	simp.simp_enabled = 0;
519 	simp.base_simp_gpa = 0;
520 
521 	wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
522 
523 	rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
524 	siefp.siefp_enabled = 0;
525 	siefp.base_siefp_gpa = 0;
526 
527 	wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
528 
529 	/* Disable the global synic bit */
530 	rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
531 	sctrl.enable = 0;
532 	wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
533 
534 	hv_synic_free_cpu(cpu);
535 }
536