xref: /linux/drivers/hv/hv.c (revision 18f90d372cf35b387663f1567de701e5393f6eb5)
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 
38 /*
39  * If false, we're using the old mechanism for stimer0 interrupts
40  * where it sends a VMbus message when it expires. The old
41  * mechanism is used when running on older versions of Hyper-V
42  * that don't support Direct Mode. While Hyper-V provides
43  * four stimer's per CPU, Linux uses only stimer0.
44  */
45 static bool direct_mode_enabled;
46 static int stimer0_irq;
47 static int stimer0_vector;
48 
49 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
50 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
51 #define HV_MIN_DELTA_TICKS 1
52 
53 /*
54  * hv_init - Main initialization routine.
55  *
56  * This routine must be called before any other routines in here are called
57  */
58 int hv_init(void)
59 {
60 	hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
61 	if (!hv_context.cpu_context)
62 		return -ENOMEM;
63 
64 	direct_mode_enabled = ms_hyperv.misc_features &
65 			HV_STIMER_DIRECT_MODE_AVAILABLE;
66 	return 0;
67 }
68 
69 /*
70  * hv_post_message - Post a message using the hypervisor message IPC.
71  *
72  * This involves a hypercall.
73  */
74 int hv_post_message(union hv_connection_id connection_id,
75 		  enum hv_message_type message_type,
76 		  void *payload, size_t payload_size)
77 {
78 	struct hv_input_post_message *aligned_msg;
79 	struct hv_per_cpu_context *hv_cpu;
80 	u64 status;
81 
82 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
83 		return -EMSGSIZE;
84 
85 	hv_cpu = get_cpu_ptr(hv_context.cpu_context);
86 	aligned_msg = hv_cpu->post_msg_page;
87 	aligned_msg->connectionid = connection_id;
88 	aligned_msg->reserved = 0;
89 	aligned_msg->message_type = message_type;
90 	aligned_msg->payload_size = payload_size;
91 	memcpy((void *)aligned_msg->payload, payload, payload_size);
92 
93 	status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
94 
95 	/* Preemption must remain disabled until after the hypercall
96 	 * so some other thread can't get scheduled onto this cpu and
97 	 * corrupt the per-cpu post_msg_page
98 	 */
99 	put_cpu_ptr(hv_cpu);
100 
101 	return status & 0xFFFF;
102 }
103 
104 /*
105  * ISR for when stimer0 is operating in Direct Mode.  Direct Mode
106  * does not use VMbus or any VMbus messages, so process here and not
107  * in the VMbus driver code.
108  */
109 
110 static void hv_stimer0_isr(void)
111 {
112 	struct hv_per_cpu_context *hv_cpu;
113 
114 	hv_cpu = this_cpu_ptr(hv_context.cpu_context);
115 	hv_cpu->clk_evt->event_handler(hv_cpu->clk_evt);
116 	add_interrupt_randomness(stimer0_vector, 0);
117 }
118 
119 static int hv_ce_set_next_event(unsigned long delta,
120 				struct clock_event_device *evt)
121 {
122 	u64 current_tick;
123 
124 	WARN_ON(!clockevent_state_oneshot(evt));
125 
126 	current_tick = hyperv_cs->read(NULL);
127 	current_tick += delta;
128 	hv_init_timer(0, current_tick);
129 	return 0;
130 }
131 
132 static int hv_ce_shutdown(struct clock_event_device *evt)
133 {
134 	hv_init_timer(0, 0);
135 	hv_init_timer_config(0, 0);
136 	if (direct_mode_enabled)
137 		hv_disable_stimer0_percpu_irq(stimer0_irq);
138 
139 	return 0;
140 }
141 
142 static int hv_ce_set_oneshot(struct clock_event_device *evt)
143 {
144 	union hv_stimer_config timer_cfg;
145 
146 	timer_cfg.as_uint64 = 0;
147 	timer_cfg.enable = 1;
148 	timer_cfg.auto_enable = 1;
149 	if (direct_mode_enabled) {
150 		/*
151 		 * When it expires, the timer will directly interrupt
152 		 * on the specified hardware vector/IRQ.
153 		 */
154 		timer_cfg.direct_mode = 1;
155 		timer_cfg.apic_vector = stimer0_vector;
156 		hv_enable_stimer0_percpu_irq(stimer0_irq);
157 	} else {
158 		/*
159 		 * When it expires, the timer will generate a VMbus message,
160 		 * to be handled by the normal VMbus interrupt handler.
161 		 */
162 		timer_cfg.direct_mode = 0;
163 		timer_cfg.sintx = VMBUS_MESSAGE_SINT;
164 	}
165 	hv_init_timer_config(0, timer_cfg.as_uint64);
166 	return 0;
167 }
168 
169 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
170 {
171 	dev->name = "Hyper-V clockevent";
172 	dev->features = CLOCK_EVT_FEAT_ONESHOT;
173 	dev->cpumask = cpumask_of(cpu);
174 	dev->rating = 1000;
175 	/*
176 	 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
177 	 * result in clockevents_config_and_register() taking additional
178 	 * references to the hv_vmbus module making it impossible to unload.
179 	 */
180 
181 	dev->set_state_shutdown = hv_ce_shutdown;
182 	dev->set_state_oneshot = hv_ce_set_oneshot;
183 	dev->set_next_event = hv_ce_set_next_event;
184 }
185 
186 
187 int hv_synic_alloc(void)
188 {
189 	int cpu;
190 	struct hv_per_cpu_context *hv_cpu;
191 
192 	/*
193 	 * First, zero all per-cpu memory areas so hv_synic_free() can
194 	 * detect what memory has been allocated and cleanup properly
195 	 * after any failures.
196 	 */
197 	for_each_present_cpu(cpu) {
198 		hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
199 		memset(hv_cpu, 0, sizeof(*hv_cpu));
200 	}
201 
202 	hv_context.hv_numa_map = kcalloc(nr_node_ids, sizeof(struct cpumask),
203 					 GFP_KERNEL);
204 	if (hv_context.hv_numa_map == NULL) {
205 		pr_err("Unable to allocate NUMA map\n");
206 		goto err;
207 	}
208 
209 	for_each_present_cpu(cpu) {
210 		hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
211 
212 		tasklet_init(&hv_cpu->msg_dpc,
213 			     vmbus_on_msg_dpc, (unsigned long) hv_cpu);
214 
215 		hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device),
216 					  GFP_KERNEL);
217 		if (hv_cpu->clk_evt == NULL) {
218 			pr_err("Unable to allocate clock event device\n");
219 			goto err;
220 		}
221 		hv_init_clockevent_device(hv_cpu->clk_evt, cpu);
222 
223 		hv_cpu->synic_message_page =
224 			(void *)get_zeroed_page(GFP_ATOMIC);
225 		if (hv_cpu->synic_message_page == NULL) {
226 			pr_err("Unable to allocate SYNIC message page\n");
227 			goto err;
228 		}
229 
230 		hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC);
231 		if (hv_cpu->synic_event_page == NULL) {
232 			pr_err("Unable to allocate SYNIC event page\n");
233 			goto err;
234 		}
235 
236 		hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
237 		if (hv_cpu->post_msg_page == NULL) {
238 			pr_err("Unable to allocate post msg page\n");
239 			goto err;
240 		}
241 
242 		INIT_LIST_HEAD(&hv_cpu->chan_list);
243 	}
244 
245 	if (direct_mode_enabled &&
246 	    hv_setup_stimer0_irq(&stimer0_irq, &stimer0_vector,
247 				hv_stimer0_isr))
248 		goto err;
249 
250 	return 0;
251 err:
252 	/*
253 	 * Any memory allocations that succeeded will be freed when
254 	 * the caller cleans up by calling hv_synic_free()
255 	 */
256 	return -ENOMEM;
257 }
258 
259 
260 void hv_synic_free(void)
261 {
262 	int cpu;
263 
264 	for_each_present_cpu(cpu) {
265 		struct hv_per_cpu_context *hv_cpu
266 			= per_cpu_ptr(hv_context.cpu_context, cpu);
267 
268 		kfree(hv_cpu->clk_evt);
269 		free_page((unsigned long)hv_cpu->synic_event_page);
270 		free_page((unsigned long)hv_cpu->synic_message_page);
271 		free_page((unsigned long)hv_cpu->post_msg_page);
272 	}
273 
274 	kfree(hv_context.hv_numa_map);
275 }
276 
277 /*
278  * hv_synic_init - Initialize the Synthetic Interrupt Controller.
279  *
280  * If it is already initialized by another entity (ie x2v shim), we need to
281  * retrieve the initialized message and event pages.  Otherwise, we create and
282  * initialize the message and event pages.
283  */
284 int hv_synic_init(unsigned int cpu)
285 {
286 	struct hv_per_cpu_context *hv_cpu
287 		= per_cpu_ptr(hv_context.cpu_context, cpu);
288 	union hv_synic_simp simp;
289 	union hv_synic_siefp siefp;
290 	union hv_synic_sint shared_sint;
291 	union hv_synic_scontrol sctrl;
292 
293 	/* Setup the Synic's message page */
294 	hv_get_simp(simp.as_uint64);
295 	simp.simp_enabled = 1;
296 	simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
297 		>> PAGE_SHIFT;
298 
299 	hv_set_simp(simp.as_uint64);
300 
301 	/* Setup the Synic's event page */
302 	hv_get_siefp(siefp.as_uint64);
303 	siefp.siefp_enabled = 1;
304 	siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
305 		>> PAGE_SHIFT;
306 
307 	hv_set_siefp(siefp.as_uint64);
308 
309 	/* Setup the shared SINT. */
310 	hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
311 
312 	shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
313 	shared_sint.masked = false;
314 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
315 		shared_sint.auto_eoi = false;
316 	else
317 		shared_sint.auto_eoi = true;
318 
319 	hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
320 
321 	/* Enable the global synic bit */
322 	hv_get_synic_state(sctrl.as_uint64);
323 	sctrl.enable = 1;
324 
325 	hv_set_synic_state(sctrl.as_uint64);
326 
327 	/*
328 	 * Register the per-cpu clockevent source.
329 	 */
330 	if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE)
331 		clockevents_config_and_register(hv_cpu->clk_evt,
332 						HV_TIMER_FREQUENCY,
333 						HV_MIN_DELTA_TICKS,
334 						HV_MAX_MAX_DELTA_TICKS);
335 	return 0;
336 }
337 
338 /*
339  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
340  */
341 void hv_synic_clockevents_cleanup(void)
342 {
343 	int cpu;
344 
345 	if (!(ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE))
346 		return;
347 
348 	if (direct_mode_enabled)
349 		hv_remove_stimer0_irq(stimer0_irq);
350 
351 	for_each_present_cpu(cpu) {
352 		struct hv_per_cpu_context *hv_cpu
353 			= per_cpu_ptr(hv_context.cpu_context, cpu);
354 
355 		clockevents_unbind_device(hv_cpu->clk_evt, cpu);
356 	}
357 }
358 
359 /*
360  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
361  */
362 int hv_synic_cleanup(unsigned int cpu)
363 {
364 	union hv_synic_sint shared_sint;
365 	union hv_synic_simp simp;
366 	union hv_synic_siefp siefp;
367 	union hv_synic_scontrol sctrl;
368 	struct vmbus_channel *channel, *sc;
369 	bool channel_found = false;
370 	unsigned long flags;
371 
372 	hv_get_synic_state(sctrl.as_uint64);
373 	if (sctrl.enable != 1)
374 		return -EFAULT;
375 
376 	/*
377 	 * Search for channels which are bound to the CPU we're about to
378 	 * cleanup. In case we find one and vmbus is still connected we need to
379 	 * fail, this will effectively prevent CPU offlining. There is no way
380 	 * we can re-bind channels to different CPUs for now.
381 	 */
382 	mutex_lock(&vmbus_connection.channel_mutex);
383 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
384 		if (channel->target_cpu == cpu) {
385 			channel_found = true;
386 			break;
387 		}
388 		spin_lock_irqsave(&channel->lock, flags);
389 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
390 			if (sc->target_cpu == cpu) {
391 				channel_found = true;
392 				break;
393 			}
394 		}
395 		spin_unlock_irqrestore(&channel->lock, flags);
396 		if (channel_found)
397 			break;
398 	}
399 	mutex_unlock(&vmbus_connection.channel_mutex);
400 
401 	if (channel_found && vmbus_connection.conn_state == CONNECTED)
402 		return -EBUSY;
403 
404 	/* Turn off clockevent device */
405 	if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) {
406 		struct hv_per_cpu_context *hv_cpu
407 			= this_cpu_ptr(hv_context.cpu_context);
408 
409 		clockevents_unbind_device(hv_cpu->clk_evt, cpu);
410 		hv_ce_shutdown(hv_cpu->clk_evt);
411 		put_cpu_ptr(hv_cpu);
412 	}
413 
414 	hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
415 
416 	shared_sint.masked = 1;
417 
418 	/* Need to correctly cleanup in the case of SMP!!! */
419 	/* Disable the interrupt */
420 	hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
421 
422 	hv_get_simp(simp.as_uint64);
423 	simp.simp_enabled = 0;
424 	simp.base_simp_gpa = 0;
425 
426 	hv_set_simp(simp.as_uint64);
427 
428 	hv_get_siefp(siefp.as_uint64);
429 	siefp.siefp_enabled = 0;
430 	siefp.base_siefp_gpa = 0;
431 
432 	hv_set_siefp(siefp.as_uint64);
433 
434 	/* Disable the global synic bit */
435 	sctrl.enable = 0;
436 	hv_set_synic_state(sctrl.as_uint64);
437 
438 	return 0;
439 }
440