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/random.h>
31 #include <linux/clockchips.h>
32 #include <asm/mshyperv.h>
33 #include "hyperv_vmbus.h"
34
35 /* The one and only */
36 struct hv_context hv_context = {
37 .synic_initialized = false,
38 };
39
40 /*
41 * If false, we're using the old mechanism for stimer0 interrupts
42 * where it sends a VMbus message when it expires. The old
43 * mechanism is used when running on older versions of Hyper-V
44 * that don't support Direct Mode. While Hyper-V provides
45 * four stimer's per CPU, Linux uses only stimer0.
46 */
47 static bool direct_mode_enabled;
48 static int stimer0_irq;
49 static int stimer0_vector;
50
51 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
52 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
53 #define HV_MIN_DELTA_TICKS 1
54
55 /*
56 * hv_init - Main initialization routine.
57 *
58 * This routine must be called before any other routines in here are called
59 */
hv_init(void)60 int hv_init(void)
61 {
62 hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
63 if (!hv_context.cpu_context)
64 return -ENOMEM;
65
66 direct_mode_enabled = ms_hyperv.misc_features &
67 HV_STIMER_DIRECT_MODE_AVAILABLE;
68 return 0;
69 }
70
71 /*
72 * hv_post_message - Post a message using the hypervisor message IPC.
73 *
74 * This involves a hypercall.
75 */
hv_post_message(union hv_connection_id connection_id,enum hv_message_type message_type,void * payload,size_t payload_size)76 int hv_post_message(union hv_connection_id connection_id,
77 enum hv_message_type message_type,
78 void *payload, size_t payload_size)
79 {
80 struct hv_input_post_message *aligned_msg;
81 struct hv_per_cpu_context *hv_cpu;
82 u64 status;
83
84 if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
85 return -EMSGSIZE;
86
87 hv_cpu = get_cpu_ptr(hv_context.cpu_context);
88 aligned_msg = hv_cpu->post_msg_page;
89 aligned_msg->connectionid = connection_id;
90 aligned_msg->reserved = 0;
91 aligned_msg->message_type = message_type;
92 aligned_msg->payload_size = payload_size;
93 memcpy((void *)aligned_msg->payload, payload, payload_size);
94
95 status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
96
97 /* Preemption must remain disabled until after the hypercall
98 * so some other thread can't get scheduled onto this cpu and
99 * corrupt the per-cpu post_msg_page
100 */
101 put_cpu_ptr(hv_cpu);
102
103 return status & 0xFFFF;
104 }
105
106 /*
107 * ISR for when stimer0 is operating in Direct Mode. Direct Mode
108 * does not use VMbus or any VMbus messages, so process here and not
109 * in the VMbus driver code.
110 */
111
hv_stimer0_isr(void)112 static void hv_stimer0_isr(void)
113 {
114 struct hv_per_cpu_context *hv_cpu;
115
116 hv_cpu = this_cpu_ptr(hv_context.cpu_context);
117 hv_cpu->clk_evt->event_handler(hv_cpu->clk_evt);
118 add_interrupt_randomness(stimer0_vector, 0);
119 }
120
hv_ce_set_next_event(unsigned long delta,struct clock_event_device * evt)121 static int hv_ce_set_next_event(unsigned long delta,
122 struct clock_event_device *evt)
123 {
124 u64 current_tick;
125
126 WARN_ON(!clockevent_state_oneshot(evt));
127
128 current_tick = hyperv_cs->read(NULL);
129 current_tick += delta;
130 hv_init_timer(0, current_tick);
131 return 0;
132 }
133
hv_ce_shutdown(struct clock_event_device * evt)134 static int hv_ce_shutdown(struct clock_event_device *evt)
135 {
136 hv_init_timer(0, 0);
137 hv_init_timer_config(0, 0);
138 if (direct_mode_enabled)
139 hv_disable_stimer0_percpu_irq(stimer0_irq);
140
141 return 0;
142 }
143
hv_ce_set_oneshot(struct clock_event_device * evt)144 static int hv_ce_set_oneshot(struct clock_event_device *evt)
145 {
146 union hv_timer_config timer_cfg;
147
148 timer_cfg.as_uint64 = 0;
149 timer_cfg.enable = 1;
150 timer_cfg.auto_enable = 1;
151 if (direct_mode_enabled) {
152 /*
153 * When it expires, the timer will directly interrupt
154 * on the specified hardware vector/IRQ.
155 */
156 timer_cfg.direct_mode = 1;
157 timer_cfg.apic_vector = stimer0_vector;
158 hv_enable_stimer0_percpu_irq(stimer0_irq);
159 } else {
160 /*
161 * When it expires, the timer will generate a VMbus message,
162 * to be handled by the normal VMbus interrupt handler.
163 */
164 timer_cfg.direct_mode = 0;
165 timer_cfg.sintx = VMBUS_MESSAGE_SINT;
166 }
167 hv_init_timer_config(0, timer_cfg.as_uint64);
168 return 0;
169 }
170
hv_init_clockevent_device(struct clock_event_device * dev,int cpu)171 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
172 {
173 dev->name = "Hyper-V clockevent";
174 dev->features = CLOCK_EVT_FEAT_ONESHOT;
175 dev->cpumask = cpumask_of(cpu);
176 dev->rating = 1000;
177 /*
178 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
179 * result in clockevents_config_and_register() taking additional
180 * references to the hv_vmbus module making it impossible to unload.
181 */
182
183 dev->set_state_shutdown = hv_ce_shutdown;
184 dev->set_state_oneshot = hv_ce_set_oneshot;
185 dev->set_next_event = hv_ce_set_next_event;
186 }
187
188
hv_synic_alloc(void)189 int hv_synic_alloc(void)
190 {
191 int cpu;
192 struct hv_per_cpu_context *hv_cpu;
193
194 /*
195 * First, zero all per-cpu memory areas so hv_synic_free() can
196 * detect what memory has been allocated and cleanup properly
197 * after any failures.
198 */
199 for_each_present_cpu(cpu) {
200 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
201 memset(hv_cpu, 0, sizeof(*hv_cpu));
202 }
203
204 hv_context.hv_numa_map = kcalloc(nr_node_ids, sizeof(struct cpumask),
205 GFP_KERNEL);
206 if (hv_context.hv_numa_map == NULL) {
207 pr_err("Unable to allocate NUMA map\n");
208 goto err;
209 }
210
211 for_each_present_cpu(cpu) {
212 hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
213
214 tasklet_init(&hv_cpu->msg_dpc,
215 vmbus_on_msg_dpc, (unsigned long) hv_cpu);
216
217 hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device),
218 GFP_KERNEL);
219 if (hv_cpu->clk_evt == NULL) {
220 pr_err("Unable to allocate clock event device\n");
221 goto err;
222 }
223 hv_init_clockevent_device(hv_cpu->clk_evt, cpu);
224
225 hv_cpu->synic_message_page =
226 (void *)get_zeroed_page(GFP_ATOMIC);
227 if (hv_cpu->synic_message_page == NULL) {
228 pr_err("Unable to allocate SYNIC message page\n");
229 goto err;
230 }
231
232 hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC);
233 if (hv_cpu->synic_event_page == NULL) {
234 pr_err("Unable to allocate SYNIC event page\n");
235 goto err;
236 }
237
238 hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
239 if (hv_cpu->post_msg_page == NULL) {
240 pr_err("Unable to allocate post msg page\n");
241 goto err;
242 }
243
244 INIT_LIST_HEAD(&hv_cpu->chan_list);
245 }
246
247 if (direct_mode_enabled &&
248 hv_setup_stimer0_irq(&stimer0_irq, &stimer0_vector,
249 hv_stimer0_isr))
250 goto err;
251
252 return 0;
253 err:
254 /*
255 * Any memory allocations that succeeded will be freed when
256 * the caller cleans up by calling hv_synic_free()
257 */
258 return -ENOMEM;
259 }
260
261
hv_synic_free(void)262 void hv_synic_free(void)
263 {
264 int cpu;
265
266 for_each_present_cpu(cpu) {
267 struct hv_per_cpu_context *hv_cpu
268 = per_cpu_ptr(hv_context.cpu_context, cpu);
269
270 kfree(hv_cpu->clk_evt);
271 free_page((unsigned long)hv_cpu->synic_event_page);
272 free_page((unsigned long)hv_cpu->synic_message_page);
273 free_page((unsigned long)hv_cpu->post_msg_page);
274 }
275
276 kfree(hv_context.hv_numa_map);
277 }
278
279 /*
280 * hv_synic_init - Initialize the Synthetic Interrupt Controller.
281 *
282 * If it is already initialized by another entity (ie x2v shim), we need to
283 * retrieve the initialized message and event pages. Otherwise, we create and
284 * initialize the message and event pages.
285 */
hv_synic_init(unsigned int cpu)286 int hv_synic_init(unsigned int cpu)
287 {
288 struct hv_per_cpu_context *hv_cpu
289 = per_cpu_ptr(hv_context.cpu_context, cpu);
290 union hv_synic_simp simp;
291 union hv_synic_siefp siefp;
292 union hv_synic_sint shared_sint;
293 union hv_synic_scontrol sctrl;
294
295 /* Setup the Synic's message page */
296 hv_get_simp(simp.as_uint64);
297 simp.simp_enabled = 1;
298 simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
299 >> PAGE_SHIFT;
300
301 hv_set_simp(simp.as_uint64);
302
303 /* Setup the Synic's event page */
304 hv_get_siefp(siefp.as_uint64);
305 siefp.siefp_enabled = 1;
306 siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
307 >> PAGE_SHIFT;
308
309 hv_set_siefp(siefp.as_uint64);
310
311 /* Setup the shared SINT. */
312 hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
313
314 shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
315 shared_sint.masked = false;
316 if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
317 shared_sint.auto_eoi = false;
318 else
319 shared_sint.auto_eoi = true;
320
321 hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
322
323 /* Enable the global synic bit */
324 hv_get_synic_state(sctrl.as_uint64);
325 sctrl.enable = 1;
326
327 hv_set_synic_state(sctrl.as_uint64);
328
329 hv_context.synic_initialized = true;
330
331 /*
332 * Register the per-cpu clockevent source.
333 */
334 if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE)
335 clockevents_config_and_register(hv_cpu->clk_evt,
336 HV_TIMER_FREQUENCY,
337 HV_MIN_DELTA_TICKS,
338 HV_MAX_MAX_DELTA_TICKS);
339 return 0;
340 }
341
342 /*
343 * hv_synic_clockevents_cleanup - Cleanup clockevent devices
344 */
hv_synic_clockevents_cleanup(void)345 void hv_synic_clockevents_cleanup(void)
346 {
347 int cpu;
348
349 if (!(ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE))
350 return;
351
352 if (direct_mode_enabled)
353 hv_remove_stimer0_irq(stimer0_irq);
354
355 for_each_present_cpu(cpu) {
356 struct hv_per_cpu_context *hv_cpu
357 = per_cpu_ptr(hv_context.cpu_context, cpu);
358
359 clockevents_unbind_device(hv_cpu->clk_evt, cpu);
360 }
361 }
362
363 /*
364 * hv_synic_cleanup - Cleanup routine for hv_synic_init().
365 */
hv_synic_cleanup(unsigned int cpu)366 int hv_synic_cleanup(unsigned int cpu)
367 {
368 union hv_synic_sint shared_sint;
369 union hv_synic_simp simp;
370 union hv_synic_siefp siefp;
371 union hv_synic_scontrol sctrl;
372 struct vmbus_channel *channel, *sc;
373 bool channel_found = false;
374 unsigned long flags;
375
376 if (!hv_context.synic_initialized)
377 return -EFAULT;
378
379 /*
380 * Search for channels which are bound to the CPU we're about to
381 * cleanup. In case we find one and vmbus is still connected we need to
382 * fail, this will effectively prevent CPU offlining. There is no way
383 * we can re-bind channels to different CPUs for now.
384 */
385 mutex_lock(&vmbus_connection.channel_mutex);
386 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
387 if (channel->target_cpu == cpu) {
388 channel_found = true;
389 break;
390 }
391 spin_lock_irqsave(&channel->lock, flags);
392 list_for_each_entry(sc, &channel->sc_list, sc_list) {
393 if (sc->target_cpu == cpu) {
394 channel_found = true;
395 break;
396 }
397 }
398 spin_unlock_irqrestore(&channel->lock, flags);
399 if (channel_found)
400 break;
401 }
402 mutex_unlock(&vmbus_connection.channel_mutex);
403
404 if (channel_found && vmbus_connection.conn_state == CONNECTED)
405 return -EBUSY;
406
407 /* Turn off clockevent device */
408 if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) {
409 struct hv_per_cpu_context *hv_cpu
410 = this_cpu_ptr(hv_context.cpu_context);
411
412 clockevents_unbind_device(hv_cpu->clk_evt, cpu);
413 hv_ce_shutdown(hv_cpu->clk_evt);
414 }
415
416 hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
417
418 shared_sint.masked = 1;
419
420 /* Need to correctly cleanup in the case of SMP!!! */
421 /* Disable the interrupt */
422 hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
423
424 hv_get_simp(simp.as_uint64);
425 simp.simp_enabled = 0;
426 simp.base_simp_gpa = 0;
427
428 hv_set_simp(simp.as_uint64);
429
430 hv_get_siefp(siefp.as_uint64);
431 siefp.siefp_enabled = 0;
432 siefp.base_siefp_gpa = 0;
433
434 hv_set_siefp(siefp.as_uint64);
435
436 /* Disable the global synic bit */
437 hv_get_synic_state(sctrl.as_uint64);
438 sctrl.enable = 0;
439 hv_set_synic_state(sctrl.as_uint64);
440
441 return 0;
442 }
443