Loading...
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 */
27struct 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 */
34int 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 */
47int 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
97int 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 == NULL) {
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 == NULL) {
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 == NULL) {
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 == NULL) {
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
198err:
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
206
207void hv_synic_free(void)
208{
209 int cpu, ret;
210
211 for_each_present_cpu(cpu) {
212 struct hv_per_cpu_context *hv_cpu
213 = per_cpu_ptr(hv_context.cpu_context, cpu);
214
215 /* It's better to leak the page if the encryption fails. */
216 if (ms_hyperv.paravisor_present && hv_isolation_type_tdx()) {
217 if (hv_cpu->post_msg_page) {
218 ret = set_memory_encrypted((unsigned long)
219 hv_cpu->post_msg_page, 1);
220 if (ret) {
221 pr_err("Failed to encrypt post msg page: %d\n", ret);
222 hv_cpu->post_msg_page = NULL;
223 }
224 }
225 }
226
227 if (!ms_hyperv.paravisor_present &&
228 (hv_isolation_type_snp() || hv_isolation_type_tdx())) {
229 if (hv_cpu->synic_message_page) {
230 ret = set_memory_encrypted((unsigned long)
231 hv_cpu->synic_message_page, 1);
232 if (ret) {
233 pr_err("Failed to encrypt SYNIC msg page: %d\n", ret);
234 hv_cpu->synic_message_page = NULL;
235 }
236 }
237
238 if (hv_cpu->synic_event_page) {
239 ret = set_memory_encrypted((unsigned long)
240 hv_cpu->synic_event_page, 1);
241 if (ret) {
242 pr_err("Failed to encrypt SYNIC event page: %d\n", ret);
243 hv_cpu->synic_event_page = NULL;
244 }
245 }
246 }
247
248 free_page((unsigned long)hv_cpu->post_msg_page);
249 free_page((unsigned long)hv_cpu->synic_event_page);
250 free_page((unsigned long)hv_cpu->synic_message_page);
251 }
252
253 kfree(hv_context.hv_numa_map);
254}
255
256/*
257 * hv_synic_init - Initialize the Synthetic Interrupt Controller.
258 *
259 * If it is already initialized by another entity (ie x2v shim), we need to
260 * retrieve the initialized message and event pages. Otherwise, we create and
261 * initialize the message and event pages.
262 */
263void hv_synic_enable_regs(unsigned int cpu)
264{
265 struct hv_per_cpu_context *hv_cpu
266 = per_cpu_ptr(hv_context.cpu_context, cpu);
267 union hv_synic_simp simp;
268 union hv_synic_siefp siefp;
269 union hv_synic_sint shared_sint;
270 union hv_synic_scontrol sctrl;
271
272 /* Setup the Synic's message page */
273 simp.as_uint64 = hv_get_register(HV_REGISTER_SIMP);
274 simp.simp_enabled = 1;
275
276 if (ms_hyperv.paravisor_present || hv_root_partition) {
277 /* Mask out vTOM bit. ioremap_cache() maps decrypted */
278 u64 base = (simp.base_simp_gpa << HV_HYP_PAGE_SHIFT) &
279 ~ms_hyperv.shared_gpa_boundary;
280 hv_cpu->synic_message_page
281 = (void *)ioremap_cache(base, HV_HYP_PAGE_SIZE);
282 if (!hv_cpu->synic_message_page)
283 pr_err("Fail to map synic message page.\n");
284 } else {
285 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
286 >> HV_HYP_PAGE_SHIFT;
287 }
288
289 hv_set_register(HV_REGISTER_SIMP, simp.as_uint64);
290
291 /* Setup the Synic's event page */
292 siefp.as_uint64 = hv_get_register(HV_REGISTER_SIEFP);
293 siefp.siefp_enabled = 1;
294
295 if (ms_hyperv.paravisor_present || hv_root_partition) {
296 /* Mask out vTOM bit. ioremap_cache() maps decrypted */
297 u64 base = (siefp.base_siefp_gpa << HV_HYP_PAGE_SHIFT) &
298 ~ms_hyperv.shared_gpa_boundary;
299 hv_cpu->synic_event_page
300 = (void *)ioremap_cache(base, HV_HYP_PAGE_SIZE);
301 if (!hv_cpu->synic_event_page)
302 pr_err("Fail to map synic event page.\n");
303 } else {
304 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
305 >> HV_HYP_PAGE_SHIFT;
306 }
307
308 hv_set_register(HV_REGISTER_SIEFP, siefp.as_uint64);
309
310 /* Setup the shared SINT. */
311 if (vmbus_irq != -1)
312 enable_percpu_irq(vmbus_irq, 0);
313 shared_sint.as_uint64 = hv_get_register(HV_REGISTER_SINT0 +
314 VMBUS_MESSAGE_SINT);
315
316 shared_sint.vector = vmbus_interrupt;
317 shared_sint.masked = false;
318
319 /*
320 * On architectures where Hyper-V doesn't support AEOI (e.g., ARM64),
321 * it doesn't provide a recommendation flag and AEOI must be disabled.
322 */
323#ifdef HV_DEPRECATING_AEOI_RECOMMENDED
324 shared_sint.auto_eoi =
325 !(ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED);
326#else
327 shared_sint.auto_eoi = 0;
328#endif
329 hv_set_register(HV_REGISTER_SINT0 + VMBUS_MESSAGE_SINT,
330 shared_sint.as_uint64);
331
332 /* Enable the global synic bit */
333 sctrl.as_uint64 = hv_get_register(HV_REGISTER_SCONTROL);
334 sctrl.enable = 1;
335
336 hv_set_register(HV_REGISTER_SCONTROL, sctrl.as_uint64);
337}
338
339int hv_synic_init(unsigned int cpu)
340{
341 hv_synic_enable_regs(cpu);
342
343 hv_stimer_legacy_init(cpu, VMBUS_MESSAGE_SINT);
344
345 return 0;
346}
347
348/*
349 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
350 */
351void hv_synic_disable_regs(unsigned int cpu)
352{
353 struct hv_per_cpu_context *hv_cpu
354 = per_cpu_ptr(hv_context.cpu_context, cpu);
355 union hv_synic_sint shared_sint;
356 union hv_synic_simp simp;
357 union hv_synic_siefp siefp;
358 union hv_synic_scontrol sctrl;
359
360 shared_sint.as_uint64 = hv_get_register(HV_REGISTER_SINT0 +
361 VMBUS_MESSAGE_SINT);
362
363 shared_sint.masked = 1;
364
365 /* Need to correctly cleanup in the case of SMP!!! */
366 /* Disable the interrupt */
367 hv_set_register(HV_REGISTER_SINT0 + VMBUS_MESSAGE_SINT,
368 shared_sint.as_uint64);
369
370 simp.as_uint64 = hv_get_register(HV_REGISTER_SIMP);
371 /*
372 * In Isolation VM, sim and sief pages are allocated by
373 * paravisor. These pages also will be used by kdump
374 * kernel. So just reset enable bit here and keep page
375 * addresses.
376 */
377 simp.simp_enabled = 0;
378 if (ms_hyperv.paravisor_present || hv_root_partition) {
379 iounmap(hv_cpu->synic_message_page);
380 hv_cpu->synic_message_page = NULL;
381 } else {
382 simp.base_simp_gpa = 0;
383 }
384
385 hv_set_register(HV_REGISTER_SIMP, simp.as_uint64);
386
387 siefp.as_uint64 = hv_get_register(HV_REGISTER_SIEFP);
388 siefp.siefp_enabled = 0;
389
390 if (ms_hyperv.paravisor_present || hv_root_partition) {
391 iounmap(hv_cpu->synic_event_page);
392 hv_cpu->synic_event_page = NULL;
393 } else {
394 siefp.base_siefp_gpa = 0;
395 }
396
397 hv_set_register(HV_REGISTER_SIEFP, siefp.as_uint64);
398
399 /* Disable the global synic bit */
400 sctrl.as_uint64 = hv_get_register(HV_REGISTER_SCONTROL);
401 sctrl.enable = 0;
402 hv_set_register(HV_REGISTER_SCONTROL, sctrl.as_uint64);
403
404 if (vmbus_irq != -1)
405 disable_percpu_irq(vmbus_irq);
406}
407
408#define HV_MAX_TRIES 3
409/*
410 * Scan the event flags page of 'this' CPU looking for any bit that is set. If we find one
411 * bit set, then wait for a few milliseconds. Repeat these steps for a maximum of 3 times.
412 * Return 'true', if there is still any set bit after this operation; 'false', otherwise.
413 *
414 * If a bit is set, that means there is a pending channel interrupt. The expectation is
415 * that the normal interrupt handling mechanism will find and process the channel interrupt
416 * "very soon", and in the process clear the bit.
417 */
418static bool hv_synic_event_pending(void)
419{
420 struct hv_per_cpu_context *hv_cpu = this_cpu_ptr(hv_context.cpu_context);
421 union hv_synic_event_flags *event =
422 (union hv_synic_event_flags *)hv_cpu->synic_event_page + VMBUS_MESSAGE_SINT;
423 unsigned long *recv_int_page = event->flags; /* assumes VMBus version >= VERSION_WIN8 */
424 bool pending;
425 u32 relid;
426 int tries = 0;
427
428retry:
429 pending = false;
430 for_each_set_bit(relid, recv_int_page, HV_EVENT_FLAGS_COUNT) {
431 /* Special case - VMBus channel protocol messages */
432 if (relid == 0)
433 continue;
434 pending = true;
435 break;
436 }
437 if (pending && tries++ < HV_MAX_TRIES) {
438 usleep_range(10000, 20000);
439 goto retry;
440 }
441 return pending;
442}
443
444int hv_synic_cleanup(unsigned int cpu)
445{
446 struct vmbus_channel *channel, *sc;
447 bool channel_found = false;
448
449 if (vmbus_connection.conn_state != CONNECTED)
450 goto always_cleanup;
451
452 /*
453 * Hyper-V does not provide a way to change the connect CPU once
454 * it is set; we must prevent the connect CPU from going offline
455 * while the VM is running normally. But in the panic or kexec()
456 * path where the vmbus is already disconnected, the CPU must be
457 * allowed to shut down.
458 */
459 if (cpu == VMBUS_CONNECT_CPU)
460 return -EBUSY;
461
462 /*
463 * Search for channels which are bound to the CPU we're about to
464 * cleanup. In case we find one and vmbus is still connected, we
465 * fail; this will effectively prevent CPU offlining.
466 *
467 * TODO: Re-bind the channels to different CPUs.
468 */
469 mutex_lock(&vmbus_connection.channel_mutex);
470 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
471 if (channel->target_cpu == cpu) {
472 channel_found = true;
473 break;
474 }
475 list_for_each_entry(sc, &channel->sc_list, sc_list) {
476 if (sc->target_cpu == cpu) {
477 channel_found = true;
478 break;
479 }
480 }
481 if (channel_found)
482 break;
483 }
484 mutex_unlock(&vmbus_connection.channel_mutex);
485
486 if (channel_found)
487 return -EBUSY;
488
489 /*
490 * channel_found == false means that any channels that were previously
491 * assigned to the CPU have been reassigned elsewhere with a call of
492 * vmbus_send_modifychannel(). Scan the event flags page looking for
493 * bits that are set and waiting with a timeout for vmbus_chan_sched()
494 * to process such bits. If bits are still set after this operation
495 * and VMBus is connected, fail the CPU offlining operation.
496 */
497 if (vmbus_proto_version >= VERSION_WIN10_V4_1 && hv_synic_event_pending())
498 return -EBUSY;
499
500always_cleanup:
501 hv_stimer_legacy_cleanup(cpu);
502
503 hv_synic_disable_regs(cpu);
504
505 return 0;
506}
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 <linux/clockchips.h>
32#include <asm/hyperv.h>
33#include <asm/mshyperv.h>
34#include "hyperv_vmbus.h"
35
36/* The one and only */
37struct hv_context hv_context = {
38 .synic_initialized = false,
39 .hypercall_page = NULL,
40};
41
42#define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
43#define HV_MAX_MAX_DELTA_TICKS 0xffffffff
44#define HV_MIN_DELTA_TICKS 1
45
46/*
47 * query_hypervisor_info - Get version info of the windows hypervisor
48 */
49unsigned int host_info_eax;
50unsigned int host_info_ebx;
51unsigned int host_info_ecx;
52unsigned int host_info_edx;
53
54static int query_hypervisor_info(void)
55{
56 unsigned int eax;
57 unsigned int ebx;
58 unsigned int ecx;
59 unsigned int edx;
60 unsigned int max_leaf;
61 unsigned int op;
62
63 /*
64 * Its assumed that this is called after confirming that Viridian
65 * is present. Query id and revision.
66 */
67 eax = 0;
68 ebx = 0;
69 ecx = 0;
70 edx = 0;
71 op = HVCPUID_VENDOR_MAXFUNCTION;
72 cpuid(op, &eax, &ebx, &ecx, &edx);
73
74 max_leaf = eax;
75
76 if (max_leaf >= HVCPUID_VERSION) {
77 eax = 0;
78 ebx = 0;
79 ecx = 0;
80 edx = 0;
81 op = HVCPUID_VERSION;
82 cpuid(op, &eax, &ebx, &ecx, &edx);
83 host_info_eax = eax;
84 host_info_ebx = ebx;
85 host_info_ecx = ecx;
86 host_info_edx = edx;
87 }
88 return max_leaf;
89}
90
91/*
92 * hv_do_hypercall- Invoke the specified hypercall
93 */
94u64 hv_do_hypercall(u64 control, void *input, void *output)
95{
96 u64 input_address = (input) ? virt_to_phys(input) : 0;
97 u64 output_address = (output) ? virt_to_phys(output) : 0;
98 void *hypercall_page = hv_context.hypercall_page;
99#ifdef CONFIG_X86_64
100 u64 hv_status = 0;
101
102 if (!hypercall_page)
103 return (u64)ULLONG_MAX;
104
105 __asm__ __volatile__("mov %0, %%r8" : : "r" (output_address) : "r8");
106 __asm__ __volatile__("call *%3" : "=a" (hv_status) :
107 "c" (control), "d" (input_address),
108 "m" (hypercall_page));
109
110 return hv_status;
111
112#else
113
114 u32 control_hi = control >> 32;
115 u32 control_lo = control & 0xFFFFFFFF;
116 u32 hv_status_hi = 1;
117 u32 hv_status_lo = 1;
118 u32 input_address_hi = input_address >> 32;
119 u32 input_address_lo = input_address & 0xFFFFFFFF;
120 u32 output_address_hi = output_address >> 32;
121 u32 output_address_lo = output_address & 0xFFFFFFFF;
122
123 if (!hypercall_page)
124 return (u64)ULLONG_MAX;
125
126 __asm__ __volatile__ ("call *%8" : "=d"(hv_status_hi),
127 "=a"(hv_status_lo) : "d" (control_hi),
128 "a" (control_lo), "b" (input_address_hi),
129 "c" (input_address_lo), "D"(output_address_hi),
130 "S"(output_address_lo), "m" (hypercall_page));
131
132 return hv_status_lo | ((u64)hv_status_hi << 32);
133#endif /* !x86_64 */
134}
135EXPORT_SYMBOL_GPL(hv_do_hypercall);
136
137#ifdef CONFIG_X86_64
138static u64 read_hv_clock_tsc(struct clocksource *arg)
139{
140 u64 current_tick;
141 struct ms_hyperv_tsc_page *tsc_pg = hv_context.tsc_page;
142
143 if (tsc_pg->tsc_sequence != 0) {
144 /*
145 * Use the tsc page to compute the value.
146 */
147
148 while (1) {
149 u64 tmp;
150 u32 sequence = tsc_pg->tsc_sequence;
151 u64 cur_tsc;
152 u64 scale = tsc_pg->tsc_scale;
153 s64 offset = tsc_pg->tsc_offset;
154
155 rdtscll(cur_tsc);
156 /* current_tick = ((cur_tsc *scale) >> 64) + offset */
157 asm("mulq %3"
158 : "=d" (current_tick), "=a" (tmp)
159 : "a" (cur_tsc), "r" (scale));
160
161 current_tick += offset;
162 if (tsc_pg->tsc_sequence == sequence)
163 return current_tick;
164
165 if (tsc_pg->tsc_sequence != 0)
166 continue;
167 /*
168 * Fallback using MSR method.
169 */
170 break;
171 }
172 }
173 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
174 return current_tick;
175}
176
177static struct clocksource hyperv_cs_tsc = {
178 .name = "hyperv_clocksource_tsc_page",
179 .rating = 425,
180 .read = read_hv_clock_tsc,
181 .mask = CLOCKSOURCE_MASK(64),
182 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
183};
184#endif
185
186
187/*
188 * hv_init - Main initialization routine.
189 *
190 * This routine must be called before any other routines in here are called
191 */
192int hv_init(void)
193{
194 int max_leaf;
195 union hv_x64_msr_hypercall_contents hypercall_msr;
196 void *virtaddr = NULL;
197
198 memset(hv_context.synic_event_page, 0, sizeof(void *) * NR_CPUS);
199 memset(hv_context.synic_message_page, 0,
200 sizeof(void *) * NR_CPUS);
201 memset(hv_context.post_msg_page, 0,
202 sizeof(void *) * NR_CPUS);
203 memset(hv_context.vp_index, 0,
204 sizeof(int) * NR_CPUS);
205 memset(hv_context.event_dpc, 0,
206 sizeof(void *) * NR_CPUS);
207 memset(hv_context.msg_dpc, 0,
208 sizeof(void *) * NR_CPUS);
209 memset(hv_context.clk_evt, 0,
210 sizeof(void *) * NR_CPUS);
211
212 max_leaf = query_hypervisor_info();
213
214 /*
215 * Write our OS ID.
216 */
217 hv_context.guestid = generate_guest_id(0, LINUX_VERSION_CODE, 0);
218 wrmsrl(HV_X64_MSR_GUEST_OS_ID, hv_context.guestid);
219
220 /* See if the hypercall page is already set */
221 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
222
223 virtaddr = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL_RX);
224
225 if (!virtaddr)
226 goto cleanup;
227
228 hypercall_msr.enable = 1;
229
230 hypercall_msr.guest_physical_address = vmalloc_to_pfn(virtaddr);
231 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
232
233 /* Confirm that hypercall page did get setup. */
234 hypercall_msr.as_uint64 = 0;
235 rdmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
236
237 if (!hypercall_msr.enable)
238 goto cleanup;
239
240 hv_context.hypercall_page = virtaddr;
241
242#ifdef CONFIG_X86_64
243 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
244 union hv_x64_msr_hypercall_contents tsc_msr;
245 void *va_tsc;
246
247 va_tsc = __vmalloc(PAGE_SIZE, GFP_KERNEL, PAGE_KERNEL);
248 if (!va_tsc)
249 goto cleanup;
250 hv_context.tsc_page = va_tsc;
251
252 rdmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
253
254 tsc_msr.enable = 1;
255 tsc_msr.guest_physical_address = vmalloc_to_pfn(va_tsc);
256
257 wrmsrl(HV_X64_MSR_REFERENCE_TSC, tsc_msr.as_uint64);
258 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
259 }
260#endif
261 return 0;
262
263cleanup:
264 if (virtaddr) {
265 if (hypercall_msr.enable) {
266 hypercall_msr.as_uint64 = 0;
267 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
268 }
269
270 vfree(virtaddr);
271 }
272
273 return -ENOTSUPP;
274}
275
276/*
277 * hv_cleanup - Cleanup routine.
278 *
279 * This routine is called normally during driver unloading or exiting.
280 */
281void hv_cleanup(bool crash)
282{
283 union hv_x64_msr_hypercall_contents hypercall_msr;
284
285 /* Reset our OS id */
286 wrmsrl(HV_X64_MSR_GUEST_OS_ID, 0);
287
288 if (hv_context.hypercall_page) {
289 hypercall_msr.as_uint64 = 0;
290 wrmsrl(HV_X64_MSR_HYPERCALL, hypercall_msr.as_uint64);
291 if (!crash)
292 vfree(hv_context.hypercall_page);
293 hv_context.hypercall_page = NULL;
294 }
295
296#ifdef CONFIG_X86_64
297 /*
298 * Cleanup the TSC page based CS.
299 */
300 if (ms_hyperv.features & HV_X64_MSR_REFERENCE_TSC_AVAILABLE) {
301 /*
302 * Crash can happen in an interrupt context and unregistering
303 * a clocksource is impossible and redundant in this case.
304 */
305 if (!oops_in_progress) {
306 clocksource_change_rating(&hyperv_cs_tsc, 10);
307 clocksource_unregister(&hyperv_cs_tsc);
308 }
309
310 hypercall_msr.as_uint64 = 0;
311 wrmsrl(HV_X64_MSR_REFERENCE_TSC, hypercall_msr.as_uint64);
312 if (!crash) {
313 vfree(hv_context.tsc_page);
314 hv_context.tsc_page = NULL;
315 }
316 }
317#endif
318}
319
320/*
321 * hv_post_message - Post a message using the hypervisor message IPC.
322 *
323 * This involves a hypercall.
324 */
325int hv_post_message(union hv_connection_id connection_id,
326 enum hv_message_type message_type,
327 void *payload, size_t payload_size)
328{
329
330 struct hv_input_post_message *aligned_msg;
331 u64 status;
332
333 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
334 return -EMSGSIZE;
335
336 aligned_msg = (struct hv_input_post_message *)
337 hv_context.post_msg_page[get_cpu()];
338
339 aligned_msg->connectionid = connection_id;
340 aligned_msg->reserved = 0;
341 aligned_msg->message_type = message_type;
342 aligned_msg->payload_size = payload_size;
343 memcpy((void *)aligned_msg->payload, payload, payload_size);
344
345 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
346
347 put_cpu();
348 return status & 0xFFFF;
349}
350
351static int hv_ce_set_next_event(unsigned long delta,
352 struct clock_event_device *evt)
353{
354 u64 current_tick;
355
356 WARN_ON(!clockevent_state_oneshot(evt));
357
358 rdmsrl(HV_X64_MSR_TIME_REF_COUNT, current_tick);
359 current_tick += delta;
360 wrmsrl(HV_X64_MSR_STIMER0_COUNT, current_tick);
361 return 0;
362}
363
364static int hv_ce_shutdown(struct clock_event_device *evt)
365{
366 wrmsrl(HV_X64_MSR_STIMER0_COUNT, 0);
367 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, 0);
368
369 return 0;
370}
371
372static int hv_ce_set_oneshot(struct clock_event_device *evt)
373{
374 union hv_timer_config timer_cfg;
375
376 timer_cfg.enable = 1;
377 timer_cfg.auto_enable = 1;
378 timer_cfg.sintx = VMBUS_MESSAGE_SINT;
379 wrmsrl(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
380
381 return 0;
382}
383
384static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
385{
386 dev->name = "Hyper-V clockevent";
387 dev->features = CLOCK_EVT_FEAT_ONESHOT;
388 dev->cpumask = cpumask_of(cpu);
389 dev->rating = 1000;
390 /*
391 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
392 * result in clockevents_config_and_register() taking additional
393 * references to the hv_vmbus module making it impossible to unload.
394 */
395
396 dev->set_state_shutdown = hv_ce_shutdown;
397 dev->set_state_oneshot = hv_ce_set_oneshot;
398 dev->set_next_event = hv_ce_set_next_event;
399}
400
401
402int hv_synic_alloc(void)
403{
404 size_t size = sizeof(struct tasklet_struct);
405 size_t ced_size = sizeof(struct clock_event_device);
406 int cpu;
407
408 hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
409 GFP_ATOMIC);
410 if (hv_context.hv_numa_map == NULL) {
411 pr_err("Unable to allocate NUMA map\n");
412 goto err;
413 }
414
415 for_each_present_cpu(cpu) {
416 hv_context.event_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
417 if (hv_context.event_dpc[cpu] == NULL) {
418 pr_err("Unable to allocate event dpc\n");
419 goto err;
420 }
421 tasklet_init(hv_context.event_dpc[cpu], vmbus_on_event, cpu);
422
423 hv_context.msg_dpc[cpu] = kmalloc(size, GFP_ATOMIC);
424 if (hv_context.msg_dpc[cpu] == NULL) {
425 pr_err("Unable to allocate event dpc\n");
426 goto err;
427 }
428 tasklet_init(hv_context.msg_dpc[cpu], vmbus_on_msg_dpc, cpu);
429
430 hv_context.clk_evt[cpu] = kzalloc(ced_size, GFP_ATOMIC);
431 if (hv_context.clk_evt[cpu] == NULL) {
432 pr_err("Unable to allocate clock event device\n");
433 goto err;
434 }
435
436 hv_init_clockevent_device(hv_context.clk_evt[cpu], cpu);
437
438 hv_context.synic_message_page[cpu] =
439 (void *)get_zeroed_page(GFP_ATOMIC);
440
441 if (hv_context.synic_message_page[cpu] == NULL) {
442 pr_err("Unable to allocate SYNIC message page\n");
443 goto err;
444 }
445
446 hv_context.synic_event_page[cpu] =
447 (void *)get_zeroed_page(GFP_ATOMIC);
448
449 if (hv_context.synic_event_page[cpu] == NULL) {
450 pr_err("Unable to allocate SYNIC event page\n");
451 goto err;
452 }
453
454 hv_context.post_msg_page[cpu] =
455 (void *)get_zeroed_page(GFP_ATOMIC);
456
457 if (hv_context.post_msg_page[cpu] == NULL) {
458 pr_err("Unable to allocate post msg page\n");
459 goto err;
460 }
461
462 INIT_LIST_HEAD(&hv_context.percpu_list[cpu]);
463 }
464
465 return 0;
466err:
467 return -ENOMEM;
468}
469
470static void hv_synic_free_cpu(int cpu)
471{
472 kfree(hv_context.event_dpc[cpu]);
473 kfree(hv_context.msg_dpc[cpu]);
474 kfree(hv_context.clk_evt[cpu]);
475 if (hv_context.synic_event_page[cpu])
476 free_page((unsigned long)hv_context.synic_event_page[cpu]);
477 if (hv_context.synic_message_page[cpu])
478 free_page((unsigned long)hv_context.synic_message_page[cpu]);
479 if (hv_context.post_msg_page[cpu])
480 free_page((unsigned long)hv_context.post_msg_page[cpu]);
481}
482
483void hv_synic_free(void)
484{
485 int cpu;
486
487 kfree(hv_context.hv_numa_map);
488 for_each_present_cpu(cpu)
489 hv_synic_free_cpu(cpu);
490}
491
492/*
493 * hv_synic_init - Initialize the Synthethic Interrupt Controller.
494 *
495 * If it is already initialized by another entity (ie x2v shim), we need to
496 * retrieve the initialized message and event pages. Otherwise, we create and
497 * initialize the message and event pages.
498 */
499void hv_synic_init(void *arg)
500{
501 u64 version;
502 union hv_synic_simp simp;
503 union hv_synic_siefp siefp;
504 union hv_synic_sint shared_sint;
505 union hv_synic_scontrol sctrl;
506 u64 vp_index;
507
508 int cpu = smp_processor_id();
509
510 if (!hv_context.hypercall_page)
511 return;
512
513 /* Check the version */
514 rdmsrl(HV_X64_MSR_SVERSION, version);
515
516 /* Setup the Synic's message page */
517 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
518 simp.simp_enabled = 1;
519 simp.base_simp_gpa = virt_to_phys(hv_context.synic_message_page[cpu])
520 >> PAGE_SHIFT;
521
522 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
523
524 /* Setup the Synic's event page */
525 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
526 siefp.siefp_enabled = 1;
527 siefp.base_siefp_gpa = virt_to_phys(hv_context.synic_event_page[cpu])
528 >> PAGE_SHIFT;
529
530 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
531
532 /* Setup the shared SINT. */
533 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
534
535 shared_sint.as_uint64 = 0;
536 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
537 shared_sint.masked = false;
538 shared_sint.auto_eoi = true;
539
540 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
541
542 /* Enable the global synic bit */
543 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
544 sctrl.enable = 1;
545
546 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
547
548 hv_context.synic_initialized = true;
549
550 /*
551 * Setup the mapping between Hyper-V's notion
552 * of cpuid and Linux' notion of cpuid.
553 * This array will be indexed using Linux cpuid.
554 */
555 rdmsrl(HV_X64_MSR_VP_INDEX, vp_index);
556 hv_context.vp_index[cpu] = (u32)vp_index;
557
558 /*
559 * Register the per-cpu clockevent source.
560 */
561 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
562 clockevents_config_and_register(hv_context.clk_evt[cpu],
563 HV_TIMER_FREQUENCY,
564 HV_MIN_DELTA_TICKS,
565 HV_MAX_MAX_DELTA_TICKS);
566 return;
567}
568
569/*
570 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
571 */
572void hv_synic_clockevents_cleanup(void)
573{
574 int cpu;
575
576 if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
577 return;
578
579 for_each_present_cpu(cpu)
580 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
581}
582
583/*
584 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
585 */
586void hv_synic_cleanup(void *arg)
587{
588 union hv_synic_sint shared_sint;
589 union hv_synic_simp simp;
590 union hv_synic_siefp siefp;
591 union hv_synic_scontrol sctrl;
592 int cpu = smp_processor_id();
593
594 if (!hv_context.synic_initialized)
595 return;
596
597 /* Turn off clockevent device */
598 if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) {
599 clockevents_unbind_device(hv_context.clk_evt[cpu], cpu);
600 hv_ce_shutdown(hv_context.clk_evt[cpu]);
601 }
602
603 rdmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
604
605 shared_sint.masked = 1;
606
607 /* Need to correctly cleanup in the case of SMP!!! */
608 /* Disable the interrupt */
609 wrmsrl(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
610
611 rdmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
612 simp.simp_enabled = 0;
613 simp.base_simp_gpa = 0;
614
615 wrmsrl(HV_X64_MSR_SIMP, simp.as_uint64);
616
617 rdmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
618 siefp.siefp_enabled = 0;
619 siefp.base_siefp_gpa = 0;
620
621 wrmsrl(HV_X64_MSR_SIEFP, siefp.as_uint64);
622
623 /* Disable the global synic bit */
624 rdmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
625 sctrl.enable = 0;
626 wrmsrl(HV_X64_MSR_SCONTROL, sctrl.as_uint64);
627}