xref: /linux/drivers/hv/hv.c (revision a67ff6a54095e27093ea501fb143fefe51a536c2)
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 <asm/hyperv.h>
30 #include "hyperv_vmbus.h"
31 
32 /* The one and only */
33 struct hv_context hv_context = {
34 	.synic_initialized	= false,
35 	.hypercall_page		= NULL,
36 	.signal_event_param	= NULL,
37 	.signal_event_buffer	= NULL,
38 };
39 
40 /*
41  * query_hypervisor_presence
42  * - Query the cpuid for presence of windows hypervisor
43  */
44 static int query_hypervisor_presence(void)
45 {
46 	unsigned int eax;
47 	unsigned int ebx;
48 	unsigned int ecx;
49 	unsigned int edx;
50 	unsigned int op;
51 
52 	eax = 0;
53 	ebx = 0;
54 	ecx = 0;
55 	edx = 0;
56 	op = HVCPUID_VERSION_FEATURES;
57 	cpuid(op, &eax, &ebx, &ecx, &edx);
58 
59 	return ecx & HV_PRESENT_BIT;
60 }
61 
62 /*
63  * query_hypervisor_info - Get version info of the windows hypervisor
64  */
65 static int query_hypervisor_info(void)
66 {
67 	unsigned int eax;
68 	unsigned int ebx;
69 	unsigned int ecx;
70 	unsigned int edx;
71 	unsigned int max_leaf;
72 	unsigned int op;
73 
74 	/*
75 	* Its assumed that this is called after confirming that Viridian
76 	* is present. Query id and revision.
77 	*/
78 	eax = 0;
79 	ebx = 0;
80 	ecx = 0;
81 	edx = 0;
82 	op = HVCPUID_VENDOR_MAXFUNCTION;
83 	cpuid(op, &eax, &ebx, &ecx, &edx);
84 
85 	max_leaf = eax;
86 
87 	if (max_leaf >= HVCPUID_VERSION) {
88 		eax = 0;
89 		ebx = 0;
90 		ecx = 0;
91 		edx = 0;
92 		op = HVCPUID_VERSION;
93 		cpuid(op, &eax, &ebx, &ecx, &edx);
94 		pr_info("Hyper-V Host OS Build:%d-%d.%d-%d-%d.%d\n",
95 			    eax,
96 			    ebx >> 16,
97 			    ebx & 0xFFFF,
98 			    ecx,
99 			    edx >> 24,
100 			    edx & 0xFFFFFF);
101 	}
102 	return max_leaf;
103 }
104 
105 /*
106  * do_hypercall- Invoke the specified hypercall
107  */
108 static u64 do_hypercall(u64 control, void *input, void *output)
109 {
110 #ifdef CONFIG_X86_64
111 	u64 hv_status = 0;
112 	u64 input_address = (input) ? virt_to_phys(input) : 0;
113 	u64 output_address = (output) ? virt_to_phys(output) : 0;
114 	void *hypercall_page = hv_context.hypercall_page;
115 
116 	__asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
117 	__asm__ __volatile__("call *%3" : "=a" (hv_status) :
118 			     "c" (control), "d" (input_address),
119 			     "m" (hypercall_page));
120 
121 	return hv_status;
122 
123 #else
124 
125 	u32 control_hi = control >> 32;
126 	u32 control_lo = control & 0xFFFFFFFF;
127 	u32 hv_status_hi = 1;
128 	u32 hv_status_lo = 1;
129 	u64 input_address = (input) ? virt_to_phys(input) : 0;
130 	u32 input_address_hi = input_address >> 32;
131 	u32 input_address_lo = input_address & 0xFFFFFFFF;
132 	u64 output_address = (output) ? virt_to_phys(output) : 0;
133 	u32 output_address_hi = output_address >> 32;
134 	u32 output_address_lo = output_address & 0xFFFFFFFF;
135 	void *hypercall_page = hv_context.hypercall_page;
136 
137 	__asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
138 			      "=a"(hv_status_lo) : "d" (control_hi),
139 			      "a" (control_lo), "b" (input_address_hi),
140 			      "c" (input_address_lo), "D"(output_address_hi),
141 			      "S"(output_address_lo), "m" (hypercall_page));
142 
143 	return hv_status_lo | ((u64)hv_status_hi << 32);
144 #endif /* !x86_64 */
145 }
146 
147 /*
148  * hv_init - Main initialization routine.
149  *
150  * This routine must be called before any other routines in here are called
151  */
152 int hv_init(void)
153 {
154 	int max_leaf;
155 	union hv_x64_msr_hypercall_contents hypercall_msr;
156 	void *virtaddr = NULL;
157 
158 	memset(hv_context.synic_event_page, 0, sizeof(void *) * MAX_NUM_CPUS);
159 	memset(hv_context.synic_message_page, 0,
160 	       sizeof(void *) * MAX_NUM_CPUS);
161 
162 	if (!query_hypervisor_presence())
163 		goto cleanup;
164 
165 	max_leaf = query_hypervisor_info();
166 
167 	rdmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
168 
169 	if (hv_context.guestid != 0)
170 		goto cleanup;
171 
172 	/* Write our OS info */
173 	wrmsrl(HV_X64_MSR_GUEST_OS_ID, HV_LINUX_GUEST_ID);
174 	hv_context.guestid = HV_LINUX_GUEST_ID;
175 
176 	/* See if the hypercall page is already set */
177 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
178 
179 	virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC);
180 
181 	if (!virtaddr)
182 		goto cleanup;
183 
184 	hypercall_msr.enable = 1;
185 
186 	hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
187 	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
188 
189 	/* Confirm that hypercall page did get setup. */
190 	hypercall_msr.as_uint64 = 0;
191 	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
192 
193 	if (!hypercall_msr.enable)
194 		goto cleanup;
195 
196 	hv_context.hypercall_page = virtaddr;
197 
198 	/* Setup the global signal event param for the signal event hypercall */
199 	hv_context.signal_event_buffer =
200 			kmalloc(sizeof(struct hv_input_signal_event_buffer),
201 				GFP_KERNEL);
202 	if (!hv_context.signal_event_buffer)
203 		goto cleanup;
204 
205 	hv_context.signal_event_param =
206 		(struct hv_input_signal_event *)
207 			(ALIGN((unsigned long)
208 				  hv_context.signal_event_buffer,
209 				  HV_HYPERCALL_PARAM_ALIGN));
210 	hv_context.signal_event_param->connectionid.asu32 = 0;
211 	hv_context.signal_event_param->connectionid.u.id =
212 						VMBUS_EVENT_CONNECTION_ID;
213 	hv_context.signal_event_param->flag_number = 0;
214 	hv_context.signal_event_param->rsvdz = 0;
215 
216 	return 0;
217 
218 cleanup:
219 	if (virtaddr) {
220 		if (hypercall_msr.enable) {
221 			hypercall_msr.as_uint64 = 0;
222 			wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
223 		}
224 
225 		vfree(virtaddr);
226 	}
227 
228 	return -ENOTSUPP;
229 }
230 
231 /*
232  * hv_cleanup - Cleanup routine.
233  *
234  * This routine is called normally during driver unloading or exiting.
235  */
236 void hv_cleanup(void)
237 {
238 	union hv_x64_msr_hypercall_contents hypercall_msr;
239 
240 	kfree(hv_context.signal_event_buffer);
241 	hv_context.signal_event_buffer = NULL;
242 	hv_context.signal_event_param = NULL;
243 
244 	if (hv_context.hypercall_page) {
245 		hypercall_msr.as_uint64 = 0;
246 		wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
247 		vfree(hv_context.hypercall_page);
248 		hv_context.hypercall_page = NULL;
249 	}
250 }
251 
252 /*
253  * hv_post_message - Post a message using the hypervisor message IPC.
254  *
255  * This involves a hypercall.
256  */
257 u16 hv_post_message(union hv_connection_id connection_id,
258 		  enum hv_message_type message_type,
259 		  void *payload, size_t payload_size)
260 {
261 	struct aligned_input {
262 		u64 alignment8;
263 		struct hv_input_post_message msg;
264 	};
265 
266 	struct hv_input_post_message *aligned_msg;
267 	u16 status;
268 	unsigned long addr;
269 
270 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
271 		return -EMSGSIZE;
272 
273 	addr = (unsigned long)kmalloc(sizeof(struct aligned_input), GFP_ATOMIC);
274 	if (!addr)
275 		return -ENOMEM;
276 
277 	aligned_msg = (struct hv_input_post_message *)
278 			(ALIGN(addr, HV_HYPERCALL_PARAM_ALIGN));
279 
280 	aligned_msg->connectionid = connection_id;
281 	aligned_msg->message_type = message_type;
282 	aligned_msg->payload_size = payload_size;
283 	memcpy((void *)aligned_msg->payload, payload, payload_size);
284 
285 	status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
286 		& 0xFFFF;
287 
288 	kfree((void *)addr);
289 
290 	return status;
291 }
292 
293 
294 /*
295  * hv_signal_event -
296  * Signal an event on the specified connection using the hypervisor event IPC.
297  *
298  * This involves a hypercall.
299  */
300 u16 hv_signal_event(void)
301 {
302 	u16 status;
303 
304 	status = do_hypercall(HVCALL_SIGNAL_EVENT,
305 			       hv_context.signal_event_param,
306 			       NULL) & 0xFFFF;
307 	return status;
308 }
309 
310 /*
311  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
312  *
313  * If it is already initialized by another entity (ie x2v shim), we need to
314  * retrieve the initialized message and event pages.  Otherwise, we create and
315  * initialize the message and event pages.
316  */
317 void hv_synic_init(void *irqarg)
318 {
319 	u64 version;
320 	union hv_synic_simp simp;
321 	union hv_synic_siefp siefp;
322 	union hv_synic_sint shared_sint;
323 	union hv_synic_scontrol sctrl;
324 
325 	u32 irq_vector = *((u32 *)(irqarg));
326 	int cpu = smp_processor_id();
327 
328 	if (!hv_context.hypercall_page)
329 		return;
330 
331 	/* Check the version */
332 	rdmsrl(HV_X64_MSR_SVERSION, version);
333 
334 	hv_context.synic_message_page[cpu] =
335 		(void *)get_zeroed_page(GFP_ATOMIC);
336 
337 	if (hv_context.synic_message_page[cpu] == NULL) {
338 		pr_err("Unable to allocate SYNIC message page\n");
339 		goto cleanup;
340 	}
341 
342 	hv_context.synic_event_page[cpu] =
343 		(void *)get_zeroed_page(GFP_ATOMIC);
344 
345 	if (hv_context.synic_event_page[cpu] == NULL) {
346 		pr_err("Unable to allocate SYNIC event page\n");
347 		goto cleanup;
348 	}
349 
350 	/* Setup the Synic's message page */
351 	rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
352 	simp.simp_enabled = 1;
353 	simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
354 		>> PAGE_SHIFT;
355 
356 	wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
357 
358 	/* Setup the Synic's event page */
359 	rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
360 	siefp.siefp_enabled = 1;
361 	siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
362 		>> PAGE_SHIFT;
363 
364 	wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
365 
366 	/* Setup the shared SINT. */
367 	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
368 
369 	shared_sint.as_uint64 = 0;
370 	shared_sint.vector = irq_vector; /* HV_SHARED_SINT_IDT_VECTOR + 0x20; */
371 	shared_sint.masked = false;
372 	shared_sint.auto_eoi = false;
373 
374 	wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
375 
376 	/* Enable the global synic bit */
377 	rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
378 	sctrl.enable = 1;
379 
380 	wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
381 
382 	hv_context.synic_initialized = true;
383 	return;
384 
385 cleanup:
386 	if (hv_context.synic_event_page[cpu])
387 		free_page((unsigned long)hv_context.synic_event_page[cpu]);
388 
389 	if (hv_context.synic_message_page[cpu])
390 		free_page((unsigned long)hv_context.synic_message_page[cpu]);
391 	return;
392 }
393 
394 /*
395  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
396  */
397 void hv_synic_cleanup(void *arg)
398 {
399 	union hv_synic_sint shared_sint;
400 	union hv_synic_simp simp;
401 	union hv_synic_siefp siefp;
402 	int cpu = smp_processor_id();
403 
404 	if (!hv_context.synic_initialized)
405 		return;
406 
407 	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
408 
409 	shared_sint.masked = 1;
410 
411 	/* Need to correctly cleanup in the case of SMP!!! */
412 	/* Disable the interrupt */
413 	wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
414 
415 	rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
416 	simp.simp_enabled = 0;
417 	simp.base_simp_gpa = 0;
418 
419 	wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
420 
421 	rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
422 	siefp.siefp_enabled = 0;
423 	siefp.base_siefp_gpa = 0;
424 
425 	wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
426 
427 	free_page((unsigned long)hv_context.synic_message_page[cpu]);
428 	free_page((unsigned long)hv_context.synic_event_page[cpu]);
429 }
430