hv.c revision 1f42248d724a413baaafd5f83a8f4746bc6f51a5
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 <asm/hyperv.h>
31#include "hyperv_vmbus.h"
32
33/* The one and only */
34struct hv_context hv_context = {
35	.synic_initialized	= false,
36	.hypercall_page		= NULL,
37	.signal_event_param	= NULL,
38	.signal_event_buffer	= NULL,
39};
40
41/*
42 * query_hypervisor_info - Get version info of the windows hypervisor
43 */
44static int query_hypervisor_info(void)
45{
46	unsigned int eax;
47	unsigned int ebx;
48	unsigned int ecx;
49	unsigned int edx;
50	unsigned int max_leaf;
51	unsigned int op;
52
53	/*
54	* Its assumed that this is called after confirming that Viridian
55	* is present. Query id and revision.
56	*/
57	eax = 0;
58	ebx = 0;
59	ecx = 0;
60	edx = 0;
61	op = HVCPUID_VENDOR_MAXFUNCTION;
62	cpuid(op, &eax, &ebx, &ecx, &edx);
63
64	max_leaf = eax;
65
66	if (max_leaf >= HVCPUID_VERSION) {
67		eax = 0;
68		ebx = 0;
69		ecx = 0;
70		edx = 0;
71		op = HVCPUID_VERSION;
72		cpuid(op, &eax, &ebx, &ecx, &edx);
73		pr_info("Hyper-V Host OS Build:%d-%d.%d-%d-%d.%d\n",
74			    eax,
75			    ebx >> 16,
76			    ebx & 0xFFFF,
77			    ecx,
78			    edx >> 24,
79			    edx & 0xFFFFFF);
80	}
81	return max_leaf;
82}
83
84/*
85 * do_hypercall- Invoke the specified hypercall
86 */
87static u64 do_hypercall(u64 control, void *input, void *output)
88{
89#ifdef CONFIG_X86_64
90	u64 hv_status = 0;
91	u64 input_address = (input) ? virt_to_phys(input) : 0;
92	u64 output_address = (output) ? virt_to_phys(output) : 0;
93	void *hypercall_page = hv_context.hypercall_page;
94
95	__asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
96	__asm__ __volatile__("call *%3" : "=a" (hv_status) :
97			     "c" (control), "d" (input_address),
98			     "m" (hypercall_page));
99
100	return hv_status;
101
102#else
103
104	u32 control_hi = control >> 32;
105	u32 control_lo = control & 0xFFFFFFFF;
106	u32 hv_status_hi = 1;
107	u32 hv_status_lo = 1;
108	u64 input_address = (input) ? virt_to_phys(input) : 0;
109	u32 input_address_hi = input_address >> 32;
110	u32 input_address_lo = input_address & 0xFFFFFFFF;
111	u64 output_address = (output) ? virt_to_phys(output) : 0;
112	u32 output_address_hi = output_address >> 32;
113	u32 output_address_lo = output_address & 0xFFFFFFFF;
114	void *hypercall_page = hv_context.hypercall_page;
115
116	__asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
117			      "=a"(hv_status_lo) : "d" (control_hi),
118			      "a" (control_lo), "b" (input_address_hi),
119			      "c" (input_address_lo), "D"(output_address_hi),
120			      "S"(output_address_lo), "m" (hypercall_page));
121
122	return hv_status_lo | ((u64)hv_status_hi << 32);
123#endif /* !x86_64 */
124}
125
126/*
127 * hv_init - Main initialization routine.
128 *
129 * This routine must be called before any other routines in here are called
130 */
131int hv_init(void)
132{
133	int max_leaf;
134	union hv_x64_msr_hypercall_contents hypercall_msr;
135	void *virtaddr = NULL;
136
137	memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
138	memset(hv_context.synic_message_page, 0,
139	       sizeof(void *) * NR_CPUS);
140
141	max_leaf = query_hypervisor_info();
142
143	/*
144	 * Write our OS ID.
145	 */
146	hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
147	wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
148
149	/* See if the hypercall page is already set */
150	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
151
152	virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC);
153
154	if (!virtaddr)
155		goto cleanup;
156
157	hypercall_msr.enable = 1;
158
159	hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
160	wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
161
162	/* Confirm that hypercall page did get setup. */
163	hypercall_msr.as_uint64 = 0;
164	rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
165
166	if (!hypercall_msr.enable)
167		goto cleanup;
168
169	hv_context.hypercall_page = virtaddr;
170
171	/* Setup the global signal event param for the signal event hypercall */
172	hv_context.signal_event_buffer =
173			kmalloc(sizeof(struct hv_input_signal_event_buffer),
174				GFP_KERNEL);
175	if (!hv_context.signal_event_buffer)
176		goto cleanup;
177
178	hv_context.signal_event_param =
179		(struct hv_input_signal_event *)
180			(ALIGN((unsigned long)
181				  hv_context.signal_event_buffer,
182				  HV_HYPERCALL_PARAM_ALIGN));
183	hv_context.signal_event_param->connectionid.asu32 = 0;
184	hv_context.signal_event_param->connectionid.u.id =
185						VMBUS_EVENT_CONNECTION_ID;
186	hv_context.signal_event_param->flag_number = 0;
187	hv_context.signal_event_param->rsvdz = 0;
188
189	return 0;
190
191cleanup:
192	if (virtaddr) {
193		if (hypercall_msr.enable) {
194			hypercall_msr.as_uint64 = 0;
195			wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
196		}
197
198		vfree(virtaddr);
199	}
200
201	return -ENOTSUPP;
202}
203
204/*
205 * hv_cleanup - Cleanup routine.
206 *
207 * This routine is called normally during driver unloading or exiting.
208 */
209void hv_cleanup(void)
210{
211	union hv_x64_msr_hypercall_contents hypercall_msr;
212
213	/* Reset our OS id */
214	wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
215
216	kfree(hv_context.signal_event_buffer);
217	hv_context.signal_event_buffer = NULL;
218	hv_context.signal_event_param = NULL;
219
220	if (hv_context.hypercall_page) {
221		hypercall_msr.as_uint64 = 0;
222		wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
223		vfree(hv_context.hypercall_page);
224		hv_context.hypercall_page = NULL;
225	}
226}
227
228/*
229 * hv_post_message - Post a message using the hypervisor message IPC.
230 *
231 * This involves a hypercall.
232 */
233int hv_post_message(union hv_connection_id connection_id,
234		  enum hv_message_type message_type,
235		  void *payload, size_t payload_size)
236{
237	struct aligned_input {
238		u64 alignment8;
239		struct hv_input_post_message msg;
240	};
241
242	struct hv_input_post_message *aligned_msg;
243	u16 status;
244	unsigned long addr;
245
246	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
247		return -EMSGSIZE;
248
249	addr = (unsigned long)kmalloc(sizeof(struct aligned_input), GFP_ATOMIC);
250	if (!addr)
251		return -ENOMEM;
252
253	aligned_msg = (struct hv_input_post_message *)
254			(ALIGN(addr, HV_HYPERCALL_PARAM_ALIGN));
255
256	aligned_msg->connectionid = connection_id;
257	aligned_msg->message_type = message_type;
258	aligned_msg->payload_size = payload_size;
259	memcpy((void *)aligned_msg->payload, payload, payload_size);
260
261	status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
262		& 0xFFFF;
263
264	kfree((void *)addr);
265
266	return status;
267}
268
269
270/*
271 * hv_signal_event -
272 * Signal an event on the specified connection using the hypervisor event IPC.
273 *
274 * This involves a hypercall.
275 */
276u16 hv_signal_event(void *con_id)
277{
278	u16 status;
279
280	status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
281
282	return status;
283}
284
285/*
286 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
287 *
288 * If it is already initialized by another entity (ie x2v shim), we need to
289 * retrieve the initialized message and event pages.  Otherwise, we create and
290 * initialize the message and event pages.
291 */
292void hv_synic_init(void *irqarg)
293{
294	u64 version;
295	union hv_synic_simp simp;
296	union hv_synic_siefp siefp;
297	union hv_synic_sint shared_sint;
298	union hv_synic_scontrol sctrl;
299
300	u32 irq_vector = *((u32 *)(irqarg));
301	int cpu = smp_processor_id();
302
303	if (!hv_context.hypercall_page)
304		return;
305
306	/* Check the version */
307	rdmsrl(HV_X64_MSR_SVERSION, version);
308
309	hv_context.synic_message_page[cpu] =
310		(void *)get_zeroed_page(GFP_ATOMIC);
311
312	if (hv_context.synic_message_page[cpu] == NULL) {
313		pr_err("Unable to allocate SYNIC message page\n");
314		goto cleanup;
315	}
316
317	hv_context.synic_event_page[cpu] =
318		(void *)get_zeroed_page(GFP_ATOMIC);
319
320	if (hv_context.synic_event_page[cpu] == NULL) {
321		pr_err("Unable to allocate SYNIC event page\n");
322		goto cleanup;
323	}
324
325	/* Setup the Synic's message page */
326	rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
327	simp.simp_enabled = 1;
328	simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
329		>> PAGE_SHIFT;
330
331	wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
332
333	/* Setup the Synic's event page */
334	rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
335	siefp.siefp_enabled = 1;
336	siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
337		>> PAGE_SHIFT;
338
339	wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
340
341	/* Setup the shared SINT. */
342	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
343
344	shared_sint.as_uint64 = 0;
345	shared_sint.vector = irq_vector; /* HV_SHARED_SINT_IDT_VECTOR + 0x20; */
346	shared_sint.masked = false;
347	shared_sint.auto_eoi = false;
348
349	wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
350
351	/* Enable the global synic bit */
352	rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
353	sctrl.enable = 1;
354
355	wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
356
357	hv_context.synic_initialized = true;
358	return;
359
360cleanup:
361	if (hv_context.synic_event_page[cpu])
362		free_page((unsigned long)hv_context.synic_event_page[cpu]);
363
364	if (hv_context.synic_message_page[cpu])
365		free_page((unsigned long)hv_context.synic_message_page[cpu]);
366	return;
367}
368
369/*
370 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
371 */
372void hv_synic_cleanup(void *arg)
373{
374	union hv_synic_sint shared_sint;
375	union hv_synic_simp simp;
376	union hv_synic_siefp siefp;
377	int cpu = smp_processor_id();
378
379	if (!hv_context.synic_initialized)
380		return;
381
382	rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
383
384	shared_sint.masked = 1;
385
386	/* Need to correctly cleanup in the case of SMP!!! */
387	/* Disable the interrupt */
388	wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
389
390	rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
391	simp.simp_enabled = 0;
392	simp.base_simp_gpa = 0;
393
394	wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
395
396	rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
397	siefp.siefp_enabled = 0;
398	siefp.base_siefp_gpa = 0;
399
400	wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
401
402	free_page((unsigned long)hv_context.synic_message_page[cpu]);
403	free_page((unsigned long)hv_context.synic_event_page[cpu]);
404}
405