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 <linux/set_memory.h>
24 #include "hyperv_vmbus.h"
25
26 /* The one and only */
27 struct hv_context hv_context;
28
29 /*
30 * hv_init - Main initialization routine.
31 *
32 * This routine must be called before any other routines in here are called
33 */
hv_init(void)34 int hv_init(void)
35 {
36 hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
37 if (!hv_context.cpu_context)
38 return -ENOMEM;
39 return 0;
40 }
41
42 /*
43 * hv_post_message - Post a message using the hypervisor message IPC.
44 *
45 * This involves a hypercall.
46 */
hv_post_message(union hv_connection_id connection_id,enum hv_message_type message_type,void * payload,size_t payload_size)47 int hv_post_message(union hv_connection_id connection_id,
48 enum hv_message_type message_type,
49 void *payload, size_t payload_size)
50 {
51 struct hv_input_post_message *aligned_msg;
52 unsigned long flags;
53 u64 status;
54
55 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
56 return -EMSGSIZE;
57
58 local_irq_save(flags);
59
60 /*
61 * A TDX VM with the paravisor must use the decrypted post_msg_page: see
62 * the comment in struct hv_per_cpu_context. A SNP VM with the paravisor
63 * can use the encrypted hyperv_pcpu_input_arg because it copies the
64 * input into the GHCB page, which has been decrypted by the paravisor.
65 */
66 if (hv_isolation_type_tdx() && ms_hyperv.paravisor_present)
67 aligned_msg = this_cpu_ptr(hv_context.cpu_context)->post_msg_page;
68 else
69 aligned_msg = *this_cpu_ptr(hyperv_pcpu_input_arg);
70
71 aligned_msg->connectionid = connection_id;
72 aligned_msg->reserved = 0;
73 aligned_msg->message_type = message_type;
74 aligned_msg->payload_size = payload_size;
75 memcpy((void *)aligned_msg->payload, payload, payload_size);
76
77 if (ms_hyperv.paravisor_present) {
78 if (hv_isolation_type_tdx())
79 status = hv_tdx_hypercall(HVCALL_POST_MESSAGE,
80 virt_to_phys(aligned_msg), 0);
81 else if (hv_isolation_type_snp())
82 status = hv_ghcb_hypercall(HVCALL_POST_MESSAGE,
83 aligned_msg, NULL,
84 sizeof(*aligned_msg));
85 else
86 status = HV_STATUS_INVALID_PARAMETER;
87 } else {
88 status = hv_do_hypercall(HVCALL_POST_MESSAGE,
89 aligned_msg, NULL);
90 }
91
92 local_irq_restore(flags);
93
94 return hv_result(status);
95 }
96
hv_synic_alloc(void)97 int hv_synic_alloc(void)
98 {
99 int cpu, ret = -ENOMEM;
100 struct hv_per_cpu_context *hv_cpu;
101
102 /*
103 * First, zero all per-cpu memory areas so hv_synic_free() can
104 * detect what memory has been allocated and cleanup properly
105 * after any failures.
106 */
107 for_each_present_cpu(cpu) {
108 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
109 memset(hv_cpu, 0, sizeof(*hv_cpu));
110 }
111
112 hv_context.hv_numa_map = kcalloc(nr_node_ids, sizeof(struct cpumask),
113 GFP_KERNEL);
114 if (!hv_context.hv_numa_map) {
115 pr_err("Unable to allocate NUMA map\n");
116 goto err;
117 }
118
119 for_each_present_cpu(cpu) {
120 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
121
122 tasklet_init(&hv_cpu->msg_dpc,
123 vmbus_on_msg_dpc, (unsigned long)hv_cpu);
124
125 if (ms_hyperv.paravisor_present && hv_isolation_type_tdx()) {
126 hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
127 if (!hv_cpu->post_msg_page) {
128 pr_err("Unable to allocate post msg page\n");
129 goto err;
130 }
131
132 ret = set_memory_decrypted((unsigned long)hv_cpu->post_msg_page, 1);
133 if (ret) {
134 pr_err("Failed to decrypt post msg page: %d\n", ret);
135 /* Just leak the page, as it's unsafe to free the page. */
136 hv_cpu->post_msg_page = NULL;
137 goto err;
138 }
139
140 memset(hv_cpu->post_msg_page, 0, PAGE_SIZE);
141 }
142
143 /*
144 * Synic message and event pages are allocated by paravisor.
145 * Skip these pages allocation here.
146 */
147 if (!ms_hyperv.paravisor_present && !hv_root_partition) {
148 hv_cpu->synic_message_page =
149 (void *)get_zeroed_page(GFP_ATOMIC);
150 if (!hv_cpu->synic_message_page) {
151 pr_err("Unable to allocate SYNIC message page\n");
152 goto err;
153 }
154
155 hv_cpu->synic_event_page =
156 (void *)get_zeroed_page(GFP_ATOMIC);
157 if (!hv_cpu->synic_event_page) {
158 pr_err("Unable to allocate SYNIC event page\n");
159
160 free_page((unsigned long)hv_cpu->synic_message_page);
161 hv_cpu->synic_message_page = NULL;
162 goto err;
163 }
164 }
165
166 if (!ms_hyperv.paravisor_present &&
167 (hv_isolation_type_snp() || hv_isolation_type_tdx())) {
168 ret = set_memory_decrypted((unsigned long)
169 hv_cpu->synic_message_page, 1);
170 if (ret) {
171 pr_err("Failed to decrypt SYNIC msg page: %d\n", ret);
172 hv_cpu->synic_message_page = NULL;
173
174 /*
175 * Free the event page here so that hv_synic_free()
176 * won't later try to re-encrypt it.
177 */
178 free_page((unsigned long)hv_cpu->synic_event_page);
179 hv_cpu->synic_event_page = NULL;
180 goto err;
181 }
182
183 ret = set_memory_decrypted((unsigned long)
184 hv_cpu->synic_event_page, 1);
185 if (ret) {
186 pr_err("Failed to decrypt SYNIC event page: %d\n", ret);
187 hv_cpu->synic_event_page = NULL;
188 goto err;
189 }
190
191 memset(hv_cpu->synic_message_page, 0, PAGE_SIZE);
192 memset(hv_cpu->synic_event_page, 0, PAGE_SIZE);
193 }
194 }
195
196 return 0;
197
198 err:
199 /*
200 * Any memory allocations that succeeded will be freed when
201 * the caller cleans up by calling hv_synic_free()
202 */
203 return ret;
204 }
205
hv_synic_free(void)206 void hv_synic_free(void)
207 {
208 int cpu, ret;
209
210 for_each_present_cpu(cpu) {
211 struct hv_per_cpu_context *hv_cpu =
212 per_cpu_ptr(hv_context.cpu_context, cpu);
213
214 /* It's better to leak the page if the encryption fails. */
215 if (ms_hyperv.paravisor_present && hv_isolation_type_tdx()) {
216 if (hv_cpu->post_msg_page) {
217 ret = set_memory_encrypted((unsigned long)
218 hv_cpu->post_msg_page, 1);
219 if (ret) {
220 pr_err("Failed to encrypt post msg page: %d\n", ret);
221 hv_cpu->post_msg_page = NULL;
222 }
223 }
224 }
225
226 if (!ms_hyperv.paravisor_present &&
227 (hv_isolation_type_snp() || hv_isolation_type_tdx())) {
228 if (hv_cpu->synic_message_page) {
229 ret = set_memory_encrypted((unsigned long)
230 hv_cpu->synic_message_page, 1);
231 if (ret) {
232 pr_err("Failed to encrypt SYNIC msg page: %d\n", ret);
233 hv_cpu->synic_message_page = NULL;
234 }
235 }
236
237 if (hv_cpu->synic_event_page) {
238 ret = set_memory_encrypted((unsigned long)
239 hv_cpu->synic_event_page, 1);
240 if (ret) {
241 pr_err("Failed to encrypt SYNIC event page: %d\n", ret);
242 hv_cpu->synic_event_page = NULL;
243 }
244 }
245 }
246
247 free_page((unsigned long)hv_cpu->post_msg_page);
248 free_page((unsigned long)hv_cpu->synic_event_page);
249 free_page((unsigned long)hv_cpu->synic_message_page);
250 }
251
252 kfree(hv_context.hv_numa_map);
253 }
254
255 /*
256 * hv_synic_init - Initialize the Synthetic Interrupt Controller.
257 *
258 * If it is already initialized by another entity (ie x2v shim), we need to
259 * retrieve the initialized message and event pages. Otherwise, we create and
260 * initialize the message and event pages.
261 */
hv_synic_enable_regs(unsigned int cpu)262 void hv_synic_enable_regs(unsigned int cpu)
263 {
264 struct hv_per_cpu_context *hv_cpu =
265 per_cpu_ptr(hv_context.cpu_context, cpu);
266 union hv_synic_simp simp;
267 union hv_synic_siefp siefp;
268 union hv_synic_sint shared_sint;
269 union hv_synic_scontrol sctrl;
270
271 /* Setup the Synic's message page */
272 simp.as_uint64 = hv_get_msr(HV_MSR_SIMP);
273 simp.simp_enabled = 1;
274
275 if (ms_hyperv.paravisor_present || hv_root_partition) {
276 /* Mask out vTOM bit. ioremap_cache() maps decrypted */
277 u64 base = (simp.base_simp_gpa << HV_HYP_PAGE_SHIFT) &
278 ~ms_hyperv.shared_gpa_boundary;
279 hv_cpu->synic_message_page =
280 (void *)ioremap_cache(base, HV_HYP_PAGE_SIZE);
281 if (!hv_cpu->synic_message_page)
282 pr_err("Fail to map synic message page.\n");
283 } else {
284 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
285 >> HV_HYP_PAGE_SHIFT;
286 }
287
288 hv_set_msr(HV_MSR_SIMP, simp.as_uint64);
289
290 /* Setup the Synic's event page */
291 siefp.as_uint64 = hv_get_msr(HV_MSR_SIEFP);
292 siefp.siefp_enabled = 1;
293
294 if (ms_hyperv.paravisor_present || hv_root_partition) {
295 /* Mask out vTOM bit. ioremap_cache() maps decrypted */
296 u64 base = (siefp.base_siefp_gpa << HV_HYP_PAGE_SHIFT) &
297 ~ms_hyperv.shared_gpa_boundary;
298 hv_cpu->synic_event_page =
299 (void *)ioremap_cache(base, HV_HYP_PAGE_SIZE);
300 if (!hv_cpu->synic_event_page)
301 pr_err("Fail to map synic event page.\n");
302 } else {
303 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
304 >> HV_HYP_PAGE_SHIFT;
305 }
306
307 hv_set_msr(HV_MSR_SIEFP, siefp.as_uint64);
308
309 /* Setup the shared SINT. */
310 if (vmbus_irq != -1)
311 enable_percpu_irq(vmbus_irq, 0);
312 shared_sint.as_uint64 = hv_get_msr(HV_MSR_SINT0 + VMBUS_MESSAGE_SINT);
313
314 shared_sint.vector = vmbus_interrupt;
315 shared_sint.masked = false;
316
317 /*
318 * On architectures where Hyper-V doesn't support AEOI (e.g., ARM64),
319 * it doesn't provide a recommendation flag and AEOI must be disabled.
320 */
321 #ifdef HV_DEPRECATING_AEOI_RECOMMENDED
322 shared_sint.auto_eoi =
323 !(ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED);
324 #else
325 shared_sint.auto_eoi = 0;
326 #endif
327 hv_set_msr(HV_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
328
329 /* Enable the global synic bit */
330 sctrl.as_uint64 = hv_get_msr(HV_MSR_SCONTROL);
331 sctrl.enable = 1;
332
333 hv_set_msr(HV_MSR_SCONTROL, sctrl.as_uint64);
334 }
335
hv_synic_init(unsigned int cpu)336 int hv_synic_init(unsigned int cpu)
337 {
338 hv_synic_enable_regs(cpu);
339
340 hv_stimer_legacy_init(cpu, VMBUS_MESSAGE_SINT);
341
342 return 0;
343 }
344
hv_synic_disable_regs(unsigned int cpu)345 void hv_synic_disable_regs(unsigned int cpu)
346 {
347 struct hv_per_cpu_context *hv_cpu =
348 per_cpu_ptr(hv_context.cpu_context, cpu);
349 union hv_synic_sint shared_sint;
350 union hv_synic_simp simp;
351 union hv_synic_siefp siefp;
352 union hv_synic_scontrol sctrl;
353
354 shared_sint.as_uint64 = hv_get_msr(HV_MSR_SINT0 + VMBUS_MESSAGE_SINT);
355
356 shared_sint.masked = 1;
357
358 /* Need to correctly cleanup in the case of SMP!!! */
359 /* Disable the interrupt */
360 hv_set_msr(HV_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
361
362 simp.as_uint64 = hv_get_msr(HV_MSR_SIMP);
363 /*
364 * In Isolation VM, sim and sief pages are allocated by
365 * paravisor. These pages also will be used by kdump
366 * kernel. So just reset enable bit here and keep page
367 * addresses.
368 */
369 simp.simp_enabled = 0;
370 if (ms_hyperv.paravisor_present || hv_root_partition) {
371 iounmap(hv_cpu->synic_message_page);
372 hv_cpu->synic_message_page = NULL;
373 } else {
374 simp.base_simp_gpa = 0;
375 }
376
377 hv_set_msr(HV_MSR_SIMP, simp.as_uint64);
378
379 siefp.as_uint64 = hv_get_msr(HV_MSR_SIEFP);
380 siefp.siefp_enabled = 0;
381
382 if (ms_hyperv.paravisor_present || hv_root_partition) {
383 iounmap(hv_cpu->synic_event_page);
384 hv_cpu->synic_event_page = NULL;
385 } else {
386 siefp.base_siefp_gpa = 0;
387 }
388
389 hv_set_msr(HV_MSR_SIEFP, siefp.as_uint64);
390
391 /* Disable the global synic bit */
392 sctrl.as_uint64 = hv_get_msr(HV_MSR_SCONTROL);
393 sctrl.enable = 0;
394 hv_set_msr(HV_MSR_SCONTROL, sctrl.as_uint64);
395
396 if (vmbus_irq != -1)
397 disable_percpu_irq(vmbus_irq);
398 }
399
400 #define HV_MAX_TRIES 3
401 /*
402 * Scan the event flags page of 'this' CPU looking for any bit that is set. If we find one
403 * bit set, then wait for a few milliseconds. Repeat these steps for a maximum of 3 times.
404 * Return 'true', if there is still any set bit after this operation; 'false', otherwise.
405 *
406 * If a bit is set, that means there is a pending channel interrupt. The expectation is
407 * that the normal interrupt handling mechanism will find and process the channel interrupt
408 * "very soon", and in the process clear the bit.
409 */
hv_synic_event_pending(void)410 static bool hv_synic_event_pending(void)
411 {
412 struct hv_per_cpu_context *hv_cpu = this_cpu_ptr(hv_context.cpu_context);
413 union hv_synic_event_flags *event =
414 (union hv_synic_event_flags *)hv_cpu->synic_event_page + VMBUS_MESSAGE_SINT;
415 unsigned long *recv_int_page = event->flags; /* assumes VMBus version >= VERSION_WIN8 */
416 bool pending;
417 u32 relid;
418 int tries = 0;
419
420 retry:
421 pending = false;
422 for_each_set_bit(relid, recv_int_page, HV_EVENT_FLAGS_COUNT) {
423 /* Special case - VMBus channel protocol messages */
424 if (relid == 0)
425 continue;
426 pending = true;
427 break;
428 }
429 if (pending && tries++ < HV_MAX_TRIES) {
430 usleep_range(10000, 20000);
431 goto retry;
432 }
433 return pending;
434 }
435
436 /*
437 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
438 */
hv_synic_cleanup(unsigned int cpu)439 int hv_synic_cleanup(unsigned int cpu)
440 {
441 struct vmbus_channel *channel, *sc;
442 bool channel_found = false;
443
444 if (vmbus_connection.conn_state != CONNECTED)
445 goto always_cleanup;
446
447 /*
448 * Hyper-V does not provide a way to change the connect CPU once
449 * it is set; we must prevent the connect CPU from going offline
450 * while the VM is running normally. But in the panic or kexec()
451 * path where the vmbus is already disconnected, the CPU must be
452 * allowed to shut down.
453 */
454 if (cpu == VMBUS_CONNECT_CPU)
455 return -EBUSY;
456
457 /*
458 * Search for channels which are bound to the CPU we're about to
459 * cleanup. In case we find one and vmbus is still connected, we
460 * fail; this will effectively prevent CPU offlining.
461 *
462 * TODO: Re-bind the channels to different CPUs.
463 */
464 mutex_lock(&vmbus_connection.channel_mutex);
465 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
466 if (channel->target_cpu == cpu) {
467 channel_found = true;
468 break;
469 }
470 list_for_each_entry(sc, &channel->sc_list, sc_list) {
471 if (sc->target_cpu == cpu) {
472 channel_found = true;
473 break;
474 }
475 }
476 if (channel_found)
477 break;
478 }
479 mutex_unlock(&vmbus_connection.channel_mutex);
480
481 if (channel_found)
482 return -EBUSY;
483
484 /*
485 * channel_found == false means that any channels that were previously
486 * assigned to the CPU have been reassigned elsewhere with a call of
487 * vmbus_send_modifychannel(). Scan the event flags page looking for
488 * bits that are set and waiting with a timeout for vmbus_chan_sched()
489 * to process such bits. If bits are still set after this operation
490 * and VMBus is connected, fail the CPU offlining operation.
491 */
492 if (vmbus_proto_version >= VERSION_WIN10_V4_1 && hv_synic_event_pending())
493 return -EBUSY;
494
495 always_cleanup:
496 hv_stimer_legacy_cleanup(cpu);
497
498 hv_synic_disable_regs(cpu);
499
500 return 0;
501 }
502