Loading...
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 <asm/hyperv.h>
32#include "hyperv_vmbus.h"
33
34/* The one and only */
35struct hv_context hv_context = {
36 .synic_initialized = false,
37 .hypercall_page = NULL,
38};
39
40/*
41 * query_hypervisor_info - Get version info of the windows hypervisor
42 */
43unsigned int host_info_eax;
44unsigned int host_info_ebx;
45unsigned int host_info_ecx;
46unsigned int host_info_edx;
47
48static int query_hypervisor_info(void)
49{
50 unsigned int eax;
51 unsigned int ebx;
52 unsigned int ecx;
53 unsigned int edx;
54 unsigned int max_leaf;
55 unsigned int op;
56
57 /*
58 * Its assumed that this is called after confirming that Viridian
59 * is present. Query id and revision.
60 */
61 eax = 0;
62 ebx = 0;
63 ecx = 0;
64 edx = 0;
65 op = HVCPUID_VENDOR_MAXFUNCTION;
66 cpuid(op, &eax, &ebx, &ecx, &edx);
67
68 max_leaf = eax;
69
70 if (max_leaf >= HVCPUID_VERSION) {
71 eax = 0;
72 ebx = 0;
73 ecx = 0;
74 edx = 0;
75 op = HVCPUID_VERSION;
76 cpuid(op, &eax, &ebx, &ecx, &edx);
77 host_info_eax = eax;
78 host_info_ebx = ebx;
79 host_info_ecx = ecx;
80 host_info_edx = edx;
81 }
82 return max_leaf;
83}
84
85/*
86 * do_hypercall- Invoke the specified hypercall
87 */
88static u64 do_hypercall(u64 control, void *input, void *output)
89{
90#ifdef CONFIG_X86_64
91 u64 hv_status = 0;
92 u64 input_address = (input) ? virt_to_phys(input) : 0;
93 u64 output_address = (output) ? virt_to_phys(output) : 0;
94 void *hypercall_page = hv_context.hypercall_page;
95
96 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
97 __asm__ __volatile__("call *%3" : "=a" (hv_status) :
98 "c" (control), "d" (input_address),
99 "m" (hypercall_page));
100
101 return hv_status;
102
103#else
104
105 u32 control_hi = control >> 32;
106 u32 control_lo = control & 0xFFFFFFFF;
107 u32 hv_status_hi = 1;
108 u32 hv_status_lo = 1;
109 u64 input_address = (input) ? virt_to_phys(input) : 0;
110 u32 input_address_hi = input_address >> 32;
111 u32 input_address_lo = input_address & 0xFFFFFFFF;
112 u64 output_address = (output) ? virt_to_phys(output) : 0;
113 u32 output_address_hi = output_address >> 32;
114 u32 output_address_lo = output_address & 0xFFFFFFFF;
115 void *hypercall_page = hv_context.hypercall_page;
116
117 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
118 "=a"(hv_status_lo) : "d" (control_hi),
119 "a" (control_lo), "b" (input_address_hi),
120 "c" (input_address_lo), "D"(output_address_hi),
121 "S"(output_address_lo), "m" (hypercall_page));
122
123 return hv_status_lo | ((u64)hv_status_hi << 32);
124#endif /* !x86_64 */
125}
126
127/*
128 * hv_init - Main initialization routine.
129 *
130 * This routine must be called before any other routines in here are called
131 */
132int hv_init(void)
133{
134 int max_leaf;
135 union hv_x64_msr_hypercall_contents hypercall_msr;
136 void *virtaddr = NULL;
137
138 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
139 memset(hv_context.synic_message_page, 0,
140 sizeof(void *) * NR_CPUS);
141 memset(hv_context.vp_index, 0,
142 sizeof(int) * NR_CPUS);
143 memset(hv_context.event_dpc, 0,
144 sizeof(void *) * NR_CPUS);
145
146 max_leaf = query_hypervisor_info();
147
148 /*
149 * Write our OS ID.
150 */
151 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
152 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
153
154 /* See if the hypercall page is already set */
155 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
156
157 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_EXEC);
158
159 if (!virtaddr)
160 goto cleanup;
161
162 hypercall_msr.enable = 1;
163
164 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
165 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
166
167 /* Confirm that hypercall page did get setup. */
168 hypercall_msr.as_uint64 = 0;
169 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
170
171 if (!hypercall_msr.enable)
172 goto cleanup;
173
174 hv_context.hypercall_page = virtaddr;
175
176 return 0;
177
178cleanup:
179 if (virtaddr) {
180 if (hypercall_msr.enable) {
181 hypercall_msr.as_uint64 = 0;
182 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
183 }
184
185 vfree(virtaddr);
186 }
187
188 return -ENOTSUPP;
189}
190
191/*
192 * hv_cleanup - Cleanup routine.
193 *
194 * This routine is called normally during driver unloading or exiting.
195 */
196void hv_cleanup(void)
197{
198 union hv_x64_msr_hypercall_contents hypercall_msr;
199
200 /* Reset our OS id */
201 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
202
203 if (hv_context.hypercall_page) {
204 hypercall_msr.as_uint64 = 0;
205 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
206 vfree(hv_context.hypercall_page);
207 hv_context.hypercall_page = NULL;
208 }
209}
210
211/*
212 * hv_post_message - Post a message using the hypervisor message IPC.
213 *
214 * This involves a hypercall.
215 */
216int hv_post_message(union hv_connection_id connection_id,
217 enum hv_message_type message_type,
218 void *payload, size_t payload_size)
219{
220 struct aligned_input {
221 u64 alignment8;
222 struct hv_input_post_message msg;
223 };
224
225 struct hv_input_post_message *aligned_msg;
226 u16 status;
227 unsigned long addr;
228
229 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
230 return -EMSGSIZE;
231
232 addr = (unsigned long)kmalloc(sizeof(struct aligned_input), GFP_ATOMIC);
233 if (!addr)
234 return -ENOMEM;
235
236 aligned_msg = (struct hv_input_post_message *)
237 (ALIGN(addr, HV_HYPERCALL_PARAM_ALIGN));
238
239 aligned_msg->connectionid = connection_id;
240 aligned_msg->message_type = message_type;
241 aligned_msg->payload_size = payload_size;
242 memcpy((void *)aligned_msg->payload, payload, payload_size);
243
244 status = do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL)
245 & 0xFFFF;
246
247 kfree((void *)addr);
248
249 return status;
250}
251
252
253/*
254 * hv_signal_event -
255 * Signal an event on the specified connection using the hypervisor event IPC.
256 *
257 * This involves a hypercall.
258 */
259u16 hv_signal_event(void *con_id)
260{
261 u16 status;
262
263 status = (do_hypercall(HVCALL_SIGNAL_EVENT, con_id, NULL) & 0xFFFF);
264
265 return status;
266}
267
268
269int hv_synic_alloc(void)
270{
271 size_t size = sizeof(struct tasklet_struct);
272 int cpu;
273
274 for_each_online_cpu(cpu) {
275 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
276 if (hv_context.event_dpc[cpu] == NULL) {
277 pr_err("Unable to allocate event dpc\n");
278 goto err;
279 }
280 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
281
282 hv_context.synic_message_page[cpu] =
283 (void *)get_zeroed_page(GFP_ATOMIC);
284
285 if (hv_context.synic_message_page[cpu] == NULL) {
286 pr_err("Unable to allocate SYNIC message page\n");
287 goto err;
288 }
289
290 hv_context.synic_event_page[cpu] =
291 (void *)get_zeroed_page(GFP_ATOMIC);
292
293 if (hv_context.synic_event_page[cpu] == NULL) {
294 pr_err("Unable to allocate SYNIC event page\n");
295 goto err;
296 }
297 }
298
299 return 0;
300err:
301 return -ENOMEM;
302}
303
304static void hv_synic_free_cpu(int cpu)
305{
306 kfree(hv_context.event_dpc[cpu]);
307 if (hv_context.synic_event_page[cpu])
308 free_page((unsigned long)hv_context.synic_event_page[cpu]);
309 if (hv_context.synic_message_page[cpu])
310 free_page((unsigned long)hv_context.synic_message_page[cpu]);
311}
312
313void hv_synic_free(void)
314{
315 int cpu;
316
317 for_each_online_cpu(cpu)
318 hv_synic_free_cpu(cpu);
319}
320
321/*
322 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
323 *
324 * If it is already initialized by another entity (ie x2v shim), we need to
325 * retrieve the initialized message and event pages. Otherwise, we create and
326 * initialize the message and event pages.
327 */
328void hv_synic_init(void *arg)
329{
330 u64 version;
331 union hv_synic_simp simp;
332 union hv_synic_siefp siefp;
333 union hv_synic_sint shared_sint;
334 union hv_synic_scontrol sctrl;
335 u64 vp_index;
336
337 int cpu = smp_processor_id();
338
339 if (!hv_context.hypercall_page)
340 return;
341
342 /* Check the version */
343 rdmsrl(HV_X64_MSR_SVERSION, version);
344
345 /* Setup the Synic's message page */
346 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
347 simp.simp_enabled = 1;
348 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
349 >> PAGE_SHIFT;
350
351 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
352
353 /* Setup the Synic's event page */
354 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
355 siefp.siefp_enabled = 1;
356 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
357 >> PAGE_SHIFT;
358
359 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
360
361 /* Setup the shared SINT. */
362 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
363
364 shared_sint.as_uint64 = 0;
365 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
366 shared_sint.masked = false;
367 shared_sint.auto_eoi = true;
368
369 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
370
371 /* Enable the global synic bit */
372 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
373 sctrl.enable = 1;
374
375 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
376
377 hv_context.synic_initialized = true;
378
379 /*
380 * Setup the mapping between Hyper-V's notion
381 * of cpuid and Linux' notion of cpuid.
382 * This array will be indexed using Linux cpuid.
383 */
384 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
385 hv_context.vp_index[cpu] = (u32)vp_index;
386 return;
387}
388
389/*
390 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
391 */
392void hv_synic_cleanup(void *arg)
393{
394 union hv_synic_sint shared_sint;
395 union hv_synic_simp simp;
396 union hv_synic_siefp siefp;
397 int cpu = smp_processor_id();
398
399 if (!hv_context.synic_initialized)
400 return;
401
402 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
403
404 shared_sint.masked = 1;
405
406 /* Need to correctly cleanup in the case of SMP!!! */
407 /* Disable the interrupt */
408 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
409
410 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
411 simp.simp_enabled = 0;
412 simp.base_simp_gpa = 0;
413
414 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
415
416 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
417 siefp.siefp_enabled = 0;
418 siefp.base_siefp_gpa = 0;
419
420 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
421
422 free_page((unsigned long)hv_context.synic_message_page[cpu]);
423 free_page((unsigned long)hv_context.synic_event_page[cpu]);
424}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/io.h>
12#include <linux/kernel.h>
13#include <linux/mm.h>
14#include <linux/slab.h>
15#include <linux/vmalloc.h>
16#include <linux/hyperv.h>
17#include <linux/random.h>
18#include <linux/clockchips.h>
19#include <linux/delay.h>
20#include <linux/interrupt.h>
21#include <clocksource/hyperv_timer.h>
22#include <asm/mshyperv.h>
23#include "hyperv_vmbus.h"
24
25/* The one and only */
26struct hv_context hv_context;
27
28/*
29 * hv_init - Main initialization routine.
30 *
31 * This routine must be called before any other routines in here are called
32 */
33int hv_init(void)
34{
35 hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
36 if (!hv_context.cpu_context)
37 return -ENOMEM;
38 return 0;
39}
40
41/*
42 * Functions for allocating and freeing memory with size and
43 * alignment HV_HYP_PAGE_SIZE. These functions are needed because
44 * the guest page size may not be the same as the Hyper-V page
45 * size. We depend upon kmalloc() aligning power-of-two size
46 * allocations to the allocation size boundary, so that the
47 * allocated memory appears to Hyper-V as a page of the size
48 * it expects.
49 */
50
51void *hv_alloc_hyperv_page(void)
52{
53 BUILD_BUG_ON(PAGE_SIZE < HV_HYP_PAGE_SIZE);
54
55 if (PAGE_SIZE == HV_HYP_PAGE_SIZE)
56 return (void *)__get_free_page(GFP_KERNEL);
57 else
58 return kmalloc(HV_HYP_PAGE_SIZE, GFP_KERNEL);
59}
60
61void *hv_alloc_hyperv_zeroed_page(void)
62{
63 if (PAGE_SIZE == HV_HYP_PAGE_SIZE)
64 return (void *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
65 else
66 return kzalloc(HV_HYP_PAGE_SIZE, GFP_KERNEL);
67}
68
69void hv_free_hyperv_page(unsigned long addr)
70{
71 if (PAGE_SIZE == HV_HYP_PAGE_SIZE)
72 free_page(addr);
73 else
74 kfree((void *)addr);
75}
76
77/*
78 * hv_post_message - Post a message using the hypervisor message IPC.
79 *
80 * This involves a hypercall.
81 */
82int hv_post_message(union hv_connection_id connection_id,
83 enum hv_message_type message_type,
84 void *payload, size_t payload_size)
85{
86 struct hv_input_post_message *aligned_msg;
87 struct hv_per_cpu_context *hv_cpu;
88 u64 status;
89
90 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
91 return -EMSGSIZE;
92
93 hv_cpu = get_cpu_ptr(hv_context.cpu_context);
94 aligned_msg = hv_cpu->post_msg_page;
95 aligned_msg->connectionid = connection_id;
96 aligned_msg->reserved = 0;
97 aligned_msg->message_type = message_type;
98 aligned_msg->payload_size = payload_size;
99 memcpy((void *)aligned_msg->payload, payload, payload_size);
100
101 if (hv_isolation_type_snp())
102 status = hv_ghcb_hypercall(HVCALL_POST_MESSAGE,
103 (void *)aligned_msg, NULL,
104 sizeof(*aligned_msg));
105 else
106 status = hv_do_hypercall(HVCALL_POST_MESSAGE,
107 aligned_msg, NULL);
108
109 /* Preemption must remain disabled until after the hypercall
110 * so some other thread can't get scheduled onto this cpu and
111 * corrupt the per-cpu post_msg_page
112 */
113 put_cpu_ptr(hv_cpu);
114
115 return hv_result(status);
116}
117
118int hv_synic_alloc(void)
119{
120 int cpu;
121 struct hv_per_cpu_context *hv_cpu;
122
123 /*
124 * First, zero all per-cpu memory areas so hv_synic_free() can
125 * detect what memory has been allocated and cleanup properly
126 * after any failures.
127 */
128 for_each_present_cpu(cpu) {
129 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
130 memset(hv_cpu, 0, sizeof(*hv_cpu));
131 }
132
133 hv_context.hv_numa_map = kcalloc(nr_node_ids, sizeof(struct cpumask),
134 GFP_KERNEL);
135 if (hv_context.hv_numa_map == NULL) {
136 pr_err("Unable to allocate NUMA map\n");
137 goto err;
138 }
139
140 for_each_present_cpu(cpu) {
141 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
142
143 tasklet_init(&hv_cpu->msg_dpc,
144 vmbus_on_msg_dpc, (unsigned long) hv_cpu);
145
146 /*
147 * Synic message and event pages are allocated by paravisor.
148 * Skip these pages allocation here.
149 */
150 if (!hv_isolation_type_snp()) {
151 hv_cpu->synic_message_page =
152 (void *)get_zeroed_page(GFP_ATOMIC);
153 if (hv_cpu->synic_message_page == NULL) {
154 pr_err("Unable to allocate SYNIC message page\n");
155 goto err;
156 }
157
158 hv_cpu->synic_event_page =
159 (void *)get_zeroed_page(GFP_ATOMIC);
160 if (hv_cpu->synic_event_page == NULL) {
161 pr_err("Unable to allocate SYNIC event page\n");
162 goto err;
163 }
164 }
165
166 hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
167 if (hv_cpu->post_msg_page == NULL) {
168 pr_err("Unable to allocate post msg page\n");
169 goto err;
170 }
171 }
172
173 return 0;
174err:
175 /*
176 * Any memory allocations that succeeded will be freed when
177 * the caller cleans up by calling hv_synic_free()
178 */
179 return -ENOMEM;
180}
181
182
183void hv_synic_free(void)
184{
185 int cpu;
186
187 for_each_present_cpu(cpu) {
188 struct hv_per_cpu_context *hv_cpu
189 = per_cpu_ptr(hv_context.cpu_context, cpu);
190
191 free_page((unsigned long)hv_cpu->synic_event_page);
192 free_page((unsigned long)hv_cpu->synic_message_page);
193 free_page((unsigned long)hv_cpu->post_msg_page);
194 }
195
196 kfree(hv_context.hv_numa_map);
197}
198
199/*
200 * hv_synic_init - Initialize the Synthetic Interrupt Controller.
201 *
202 * If it is already initialized by another entity (ie x2v shim), we need to
203 * retrieve the initialized message and event pages. Otherwise, we create and
204 * initialize the message and event pages.
205 */
206void hv_synic_enable_regs(unsigned int cpu)
207{
208 struct hv_per_cpu_context *hv_cpu
209 = per_cpu_ptr(hv_context.cpu_context, cpu);
210 union hv_synic_simp simp;
211 union hv_synic_siefp siefp;
212 union hv_synic_sint shared_sint;
213 union hv_synic_scontrol sctrl;
214
215 /* Setup the Synic's message page */
216 simp.as_uint64 = hv_get_register(HV_REGISTER_SIMP);
217 simp.simp_enabled = 1;
218
219 if (hv_isolation_type_snp()) {
220 hv_cpu->synic_message_page
221 = memremap(simp.base_simp_gpa << HV_HYP_PAGE_SHIFT,
222 HV_HYP_PAGE_SIZE, MEMREMAP_WB);
223 if (!hv_cpu->synic_message_page)
224 pr_err("Fail to map syinc message page.\n");
225 } else {
226 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
227 >> HV_HYP_PAGE_SHIFT;
228 }
229
230 hv_set_register(HV_REGISTER_SIMP, simp.as_uint64);
231
232 /* Setup the Synic's event page */
233 siefp.as_uint64 = hv_get_register(HV_REGISTER_SIEFP);
234 siefp.siefp_enabled = 1;
235
236 if (hv_isolation_type_snp()) {
237 hv_cpu->synic_event_page =
238 memremap(siefp.base_siefp_gpa << HV_HYP_PAGE_SHIFT,
239 HV_HYP_PAGE_SIZE, MEMREMAP_WB);
240
241 if (!hv_cpu->synic_event_page)
242 pr_err("Fail to map syinc event page.\n");
243 } else {
244 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
245 >> HV_HYP_PAGE_SHIFT;
246 }
247
248 hv_set_register(HV_REGISTER_SIEFP, siefp.as_uint64);
249
250 /* Setup the shared SINT. */
251 if (vmbus_irq != -1)
252 enable_percpu_irq(vmbus_irq, 0);
253 shared_sint.as_uint64 = hv_get_register(HV_REGISTER_SINT0 +
254 VMBUS_MESSAGE_SINT);
255
256 shared_sint.vector = vmbus_interrupt;
257 shared_sint.masked = false;
258
259 /*
260 * On architectures where Hyper-V doesn't support AEOI (e.g., ARM64),
261 * it doesn't provide a recommendation flag and AEOI must be disabled.
262 */
263#ifdef HV_DEPRECATING_AEOI_RECOMMENDED
264 shared_sint.auto_eoi =
265 !(ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED);
266#else
267 shared_sint.auto_eoi = 0;
268#endif
269 hv_set_register(HV_REGISTER_SINT0 + VMBUS_MESSAGE_SINT,
270 shared_sint.as_uint64);
271
272 /* Enable the global synic bit */
273 sctrl.as_uint64 = hv_get_register(HV_REGISTER_SCONTROL);
274 sctrl.enable = 1;
275
276 hv_set_register(HV_REGISTER_SCONTROL, sctrl.as_uint64);
277}
278
279int hv_synic_init(unsigned int cpu)
280{
281 hv_synic_enable_regs(cpu);
282
283 hv_stimer_legacy_init(cpu, VMBUS_MESSAGE_SINT);
284
285 return 0;
286}
287
288/*
289 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
290 */
291void hv_synic_disable_regs(unsigned int cpu)
292{
293 struct hv_per_cpu_context *hv_cpu
294 = per_cpu_ptr(hv_context.cpu_context, cpu);
295 union hv_synic_sint shared_sint;
296 union hv_synic_simp simp;
297 union hv_synic_siefp siefp;
298 union hv_synic_scontrol sctrl;
299
300 shared_sint.as_uint64 = hv_get_register(HV_REGISTER_SINT0 +
301 VMBUS_MESSAGE_SINT);
302
303 shared_sint.masked = 1;
304
305 /* Need to correctly cleanup in the case of SMP!!! */
306 /* Disable the interrupt */
307 hv_set_register(HV_REGISTER_SINT0 + VMBUS_MESSAGE_SINT,
308 shared_sint.as_uint64);
309
310 simp.as_uint64 = hv_get_register(HV_REGISTER_SIMP);
311 /*
312 * In Isolation VM, sim and sief pages are allocated by
313 * paravisor. These pages also will be used by kdump
314 * kernel. So just reset enable bit here and keep page
315 * addresses.
316 */
317 simp.simp_enabled = 0;
318 if (hv_isolation_type_snp())
319 memunmap(hv_cpu->synic_message_page);
320 else
321 simp.base_simp_gpa = 0;
322
323 hv_set_register(HV_REGISTER_SIMP, simp.as_uint64);
324
325 siefp.as_uint64 = hv_get_register(HV_REGISTER_SIEFP);
326 siefp.siefp_enabled = 0;
327
328 if (hv_isolation_type_snp())
329 memunmap(hv_cpu->synic_event_page);
330 else
331 siefp.base_siefp_gpa = 0;
332
333 hv_set_register(HV_REGISTER_SIEFP, siefp.as_uint64);
334
335 /* Disable the global synic bit */
336 sctrl.as_uint64 = hv_get_register(HV_REGISTER_SCONTROL);
337 sctrl.enable = 0;
338 hv_set_register(HV_REGISTER_SCONTROL, sctrl.as_uint64);
339
340 if (vmbus_irq != -1)
341 disable_percpu_irq(vmbus_irq);
342}
343
344#define HV_MAX_TRIES 3
345/*
346 * Scan the event flags page of 'this' CPU looking for any bit that is set. If we find one
347 * bit set, then wait for a few milliseconds. Repeat these steps for a maximum of 3 times.
348 * Return 'true', if there is still any set bit after this operation; 'false', otherwise.
349 *
350 * If a bit is set, that means there is a pending channel interrupt. The expectation is
351 * that the normal interrupt handling mechanism will find and process the channel interrupt
352 * "very soon", and in the process clear the bit.
353 */
354static bool hv_synic_event_pending(void)
355{
356 struct hv_per_cpu_context *hv_cpu = this_cpu_ptr(hv_context.cpu_context);
357 union hv_synic_event_flags *event =
358 (union hv_synic_event_flags *)hv_cpu->synic_event_page + VMBUS_MESSAGE_SINT;
359 unsigned long *recv_int_page = event->flags; /* assumes VMBus version >= VERSION_WIN8 */
360 bool pending;
361 u32 relid;
362 int tries = 0;
363
364retry:
365 pending = false;
366 for_each_set_bit(relid, recv_int_page, HV_EVENT_FLAGS_COUNT) {
367 /* Special case - VMBus channel protocol messages */
368 if (relid == 0)
369 continue;
370 pending = true;
371 break;
372 }
373 if (pending && tries++ < HV_MAX_TRIES) {
374 usleep_range(10000, 20000);
375 goto retry;
376 }
377 return pending;
378}
379
380int hv_synic_cleanup(unsigned int cpu)
381{
382 struct vmbus_channel *channel, *sc;
383 bool channel_found = false;
384
385 if (vmbus_connection.conn_state != CONNECTED)
386 goto always_cleanup;
387
388 /*
389 * Hyper-V does not provide a way to change the connect CPU once
390 * it is set; we must prevent the connect CPU from going offline
391 * while the VM is running normally. But in the panic or kexec()
392 * path where the vmbus is already disconnected, the CPU must be
393 * allowed to shut down.
394 */
395 if (cpu == VMBUS_CONNECT_CPU)
396 return -EBUSY;
397
398 /*
399 * Search for channels which are bound to the CPU we're about to
400 * cleanup. In case we find one and vmbus is still connected, we
401 * fail; this will effectively prevent CPU offlining.
402 *
403 * TODO: Re-bind the channels to different CPUs.
404 */
405 mutex_lock(&vmbus_connection.channel_mutex);
406 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
407 if (channel->target_cpu == cpu) {
408 channel_found = true;
409 break;
410 }
411 list_for_each_entry(sc, &channel->sc_list, sc_list) {
412 if (sc->target_cpu == cpu) {
413 channel_found = true;
414 break;
415 }
416 }
417 if (channel_found)
418 break;
419 }
420 mutex_unlock(&vmbus_connection.channel_mutex);
421
422 if (channel_found)
423 return -EBUSY;
424
425 /*
426 * channel_found == false means that any channels that were previously
427 * assigned to the CPU have been reassigned elsewhere with a call of
428 * vmbus_send_modifychannel(). Scan the event flags page looking for
429 * bits that are set and waiting with a timeout for vmbus_chan_sched()
430 * to process such bits. If bits are still set after this operation
431 * and VMBus is connected, fail the CPU offlining operation.
432 */
433 if (vmbus_proto_version >= VERSION_WIN10_V4_1 && hv_synic_event_pending())
434 return -EBUSY;
435
436always_cleanup:
437 hv_stimer_legacy_cleanup(cpu);
438
439 hv_synic_disable_regs(cpu);
440
441 return 0;
442}