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 */
26 struct 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 */
hv_init(void)33 int 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
hv_alloc_hyperv_page(void)51 void *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
hv_alloc_hyperv_zeroed_page(void)61 void *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
hv_free_hyperv_page(unsigned long addr)69 void 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 */
hv_post_message(union hv_connection_id connection_id,enum hv_message_type message_type,void * payload,size_t payload_size)82 int 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
hv_synic_alloc(void)118 int 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() && !hv_root_partition) {
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;
174 err:
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
hv_synic_free(void)183 void 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 */
hv_synic_enable_regs(unsigned int cpu)206 void 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() || hv_root_partition) {
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() || hv_root_partition) {
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
hv_synic_init(unsigned int cpu)279 int 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 */
hv_synic_disable_regs(unsigned int cpu)291 void 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() || hv_root_partition) {
319 memunmap(hv_cpu->synic_message_page);
320 hv_cpu->synic_message_page = NULL;
321 } else {
322 simp.base_simp_gpa = 0;
323 }
324
325 hv_set_register(HV_REGISTER_SIMP, simp.as_uint64);
326
327 siefp.as_uint64 = hv_get_register(HV_REGISTER_SIEFP);
328 siefp.siefp_enabled = 0;
329
330 if (hv_isolation_type_snp() || hv_root_partition) {
331 memunmap(hv_cpu->synic_event_page);
332 hv_cpu->synic_event_page = NULL;
333 } else {
334 siefp.base_siefp_gpa = 0;
335 }
336
337 hv_set_register(HV_REGISTER_SIEFP, siefp.as_uint64);
338
339 /* Disable the global synic bit */
340 sctrl.as_uint64 = hv_get_register(HV_REGISTER_SCONTROL);
341 sctrl.enable = 0;
342 hv_set_register(HV_REGISTER_SCONTROL, sctrl.as_uint64);
343
344 if (vmbus_irq != -1)
345 disable_percpu_irq(vmbus_irq);
346 }
347
348 #define HV_MAX_TRIES 3
349 /*
350 * Scan the event flags page of 'this' CPU looking for any bit that is set. If we find one
351 * bit set, then wait for a few milliseconds. Repeat these steps for a maximum of 3 times.
352 * Return 'true', if there is still any set bit after this operation; 'false', otherwise.
353 *
354 * If a bit is set, that means there is a pending channel interrupt. The expectation is
355 * that the normal interrupt handling mechanism will find and process the channel interrupt
356 * "very soon", and in the process clear the bit.
357 */
hv_synic_event_pending(void)358 static bool hv_synic_event_pending(void)
359 {
360 struct hv_per_cpu_context *hv_cpu = this_cpu_ptr(hv_context.cpu_context);
361 union hv_synic_event_flags *event =
362 (union hv_synic_event_flags *)hv_cpu->synic_event_page + VMBUS_MESSAGE_SINT;
363 unsigned long *recv_int_page = event->flags; /* assumes VMBus version >= VERSION_WIN8 */
364 bool pending;
365 u32 relid;
366 int tries = 0;
367
368 retry:
369 pending = false;
370 for_each_set_bit(relid, recv_int_page, HV_EVENT_FLAGS_COUNT) {
371 /* Special case - VMBus channel protocol messages */
372 if (relid == 0)
373 continue;
374 pending = true;
375 break;
376 }
377 if (pending && tries++ < HV_MAX_TRIES) {
378 usleep_range(10000, 20000);
379 goto retry;
380 }
381 return pending;
382 }
383
hv_synic_cleanup(unsigned int cpu)384 int hv_synic_cleanup(unsigned int cpu)
385 {
386 struct vmbus_channel *channel, *sc;
387 bool channel_found = false;
388
389 if (vmbus_connection.conn_state != CONNECTED)
390 goto always_cleanup;
391
392 /*
393 * Hyper-V does not provide a way to change the connect CPU once
394 * it is set; we must prevent the connect CPU from going offline
395 * while the VM is running normally. But in the panic or kexec()
396 * path where the vmbus is already disconnected, the CPU must be
397 * allowed to shut down.
398 */
399 if (cpu == VMBUS_CONNECT_CPU)
400 return -EBUSY;
401
402 /*
403 * Search for channels which are bound to the CPU we're about to
404 * cleanup. In case we find one and vmbus is still connected, we
405 * fail; this will effectively prevent CPU offlining.
406 *
407 * TODO: Re-bind the channels to different CPUs.
408 */
409 mutex_lock(&vmbus_connection.channel_mutex);
410 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
411 if (channel->target_cpu == cpu) {
412 channel_found = true;
413 break;
414 }
415 list_for_each_entry(sc, &channel->sc_list, sc_list) {
416 if (sc->target_cpu == cpu) {
417 channel_found = true;
418 break;
419 }
420 }
421 if (channel_found)
422 break;
423 }
424 mutex_unlock(&vmbus_connection.channel_mutex);
425
426 if (channel_found)
427 return -EBUSY;
428
429 /*
430 * channel_found == false means that any channels that were previously
431 * assigned to the CPU have been reassigned elsewhere with a call of
432 * vmbus_send_modifychannel(). Scan the event flags page looking for
433 * bits that are set and waiting with a timeout for vmbus_chan_sched()
434 * to process such bits. If bits are still set after this operation
435 * and VMBus is connected, fail the CPU offlining operation.
436 */
437 if (vmbus_proto_version >= VERSION_WIN10_V4_1 && hv_synic_event_pending())
438 return -EBUSY;
439
440 always_cleanup:
441 hv_stimer_legacy_cleanup(cpu);
442
443 hv_synic_disable_regs(cpu);
444
445 return 0;
446 }
447