xref: /openbmc/linux/drivers/hv/hv.c (revision bd329f028f1cd51c7623c326147af07c6d832193)
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/hyperv.h>
33 #include <asm/mshyperv.h>
34 #include "hyperv_vmbus.h"
35 
36 /* The one and only */
37 struct hv_context hv_context = {
38 	.synic_initialized	= false,
39 };
40 
41 /*
42  * If false, we're using the old mechanism for stimer0 interrupts
43  * where it sends a VMbus message when it expires. The old
44  * mechanism is used when running on older versions of Hyper-V
45  * that don't support Direct Mode. While Hyper-V provides
46  * four stimer's per CPU, Linux uses only stimer0.
47  */
48 static bool direct_mode_enabled;
49 static int stimer0_irq;
50 static int stimer0_vector;
51 
52 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
53 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
54 #define HV_MIN_DELTA_TICKS 1
55 
56 /*
57  * hv_init - Main initialization routine.
58  *
59  * This routine must be called before any other routines in here are called
60  */
61 int hv_init(void)
62 {
63 	hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
64 	if (!hv_context.cpu_context)
65 		return -ENOMEM;
66 
67 	direct_mode_enabled = ms_hyperv.misc_features &
68 			HV_X64_STIMER_DIRECT_MODE_AVAILABLE;
69 	return 0;
70 }
71 
72 /*
73  * hv_post_message - Post a message using the hypervisor message IPC.
74  *
75  * This involves a hypercall.
76  */
77 int hv_post_message(union hv_connection_id connection_id,
78 		  enum hv_message_type message_type,
79 		  void *payload, size_t payload_size)
80 {
81 	struct hv_input_post_message *aligned_msg;
82 	struct hv_per_cpu_context *hv_cpu;
83 	u64 status;
84 
85 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
86 		return -EMSGSIZE;
87 
88 	hv_cpu = get_cpu_ptr(hv_context.cpu_context);
89 	aligned_msg = hv_cpu->post_msg_page;
90 	aligned_msg->connectionid = connection_id;
91 	aligned_msg->reserved = 0;
92 	aligned_msg->message_type = message_type;
93 	aligned_msg->payload_size = payload_size;
94 	memcpy((void *)aligned_msg->payload, payload, payload_size);
95 
96 	status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
97 
98 	/* Preemption must remain disabled until after the hypercall
99 	 * so some other thread can't get scheduled onto this cpu and
100 	 * corrupt the per-cpu post_msg_page
101 	 */
102 	put_cpu_ptr(hv_cpu);
103 
104 	return status & 0xFFFF;
105 }
106 
107 /*
108  * ISR for when stimer0 is operating in Direct Mode.  Direct Mode
109  * does not use VMbus or any VMbus messages, so process here and not
110  * in the VMbus driver code.
111  */
112 
113 static void hv_stimer0_isr(void)
114 {
115 	struct hv_per_cpu_context *hv_cpu;
116 
117 	hv_cpu = this_cpu_ptr(hv_context.cpu_context);
118 	hv_cpu->clk_evt->event_handler(hv_cpu->clk_evt);
119 	add_interrupt_randomness(stimer0_vector, 0);
120 }
121 
122 static int hv_ce_set_next_event(unsigned long delta,
123 				struct clock_event_device *evt)
124 {
125 	u64 current_tick;
126 
127 	WARN_ON(!clockevent_state_oneshot(evt));
128 
129 	current_tick = hyperv_cs->read(NULL);
130 	current_tick += delta;
131 	hv_init_timer(HV_X64_MSR_STIMER0_COUNT, current_tick);
132 	return 0;
133 }
134 
135 static int hv_ce_shutdown(struct clock_event_device *evt)
136 {
137 	hv_init_timer(HV_X64_MSR_STIMER0_COUNT, 0);
138 	hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, 0);
139 	if (direct_mode_enabled)
140 		hv_disable_stimer0_percpu_irq(stimer0_irq);
141 
142 	return 0;
143 }
144 
145 static int hv_ce_set_oneshot(struct clock_event_device *evt)
146 {
147 	union hv_timer_config timer_cfg;
148 
149 	timer_cfg.as_uint64 = 0;
150 	timer_cfg.enable = 1;
151 	timer_cfg.auto_enable = 1;
152 	if (direct_mode_enabled) {
153 		/*
154 		 * When it expires, the timer will directly interrupt
155 		 * on the specified hardware vector/IRQ.
156 		 */
157 		timer_cfg.direct_mode = 1;
158 		timer_cfg.apic_vector = stimer0_vector;
159 		hv_enable_stimer0_percpu_irq(stimer0_irq);
160 	} else {
161 		/*
162 		 * When it expires, the timer will generate a VMbus message,
163 		 * to be handled by the normal VMbus interrupt handler.
164 		 */
165 		timer_cfg.direct_mode = 0;
166 		timer_cfg.sintx = VMBUS_MESSAGE_SINT;
167 	}
168 	hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
169 	return 0;
170 }
171 
172 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
173 {
174 	dev->name = "Hyper-V clockevent";
175 	dev->features = CLOCK_EVT_FEAT_ONESHOT;
176 	dev->cpumask = cpumask_of(cpu);
177 	dev->rating = 1000;
178 	/*
179 	 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
180 	 * result in clockevents_config_and_register() taking additional
181 	 * references to the hv_vmbus module making it impossible to unload.
182 	 */
183 
184 	dev->set_state_shutdown = hv_ce_shutdown;
185 	dev->set_state_oneshot = hv_ce_set_oneshot;
186 	dev->set_next_event = hv_ce_set_next_event;
187 }
188 
189 
190 int hv_synic_alloc(void)
191 {
192 	int cpu;
193 
194 	hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
195 					 GFP_KERNEL);
196 	if (hv_context.hv_numa_map == NULL) {
197 		pr_err("Unable to allocate NUMA map\n");
198 		goto err;
199 	}
200 
201 	for_each_present_cpu(cpu) {
202 		struct hv_per_cpu_context *hv_cpu
203 			= per_cpu_ptr(hv_context.cpu_context, cpu);
204 
205 		memset(hv_cpu, 0, sizeof(*hv_cpu));
206 		tasklet_init(&hv_cpu->msg_dpc,
207 			     vmbus_on_msg_dpc, (unsigned long) hv_cpu);
208 
209 		hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device),
210 					  GFP_KERNEL);
211 		if (hv_cpu->clk_evt == NULL) {
212 			pr_err("Unable to allocate clock event device\n");
213 			goto err;
214 		}
215 		hv_init_clockevent_device(hv_cpu->clk_evt, cpu);
216 
217 		hv_cpu->synic_message_page =
218 			(void *)get_zeroed_page(GFP_ATOMIC);
219 		if (hv_cpu->synic_message_page == NULL) {
220 			pr_err("Unable to allocate SYNIC message page\n");
221 			goto err;
222 		}
223 
224 		hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC);
225 		if (hv_cpu->synic_event_page == NULL) {
226 			pr_err("Unable to allocate SYNIC event page\n");
227 			goto err;
228 		}
229 
230 		hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
231 		if (hv_cpu->post_msg_page == NULL) {
232 			pr_err("Unable to allocate post msg page\n");
233 			goto err;
234 		}
235 
236 		INIT_LIST_HEAD(&hv_cpu->chan_list);
237 	}
238 
239 	if (direct_mode_enabled &&
240 	    hv_setup_stimer0_irq(&stimer0_irq, &stimer0_vector,
241 				hv_stimer0_isr))
242 		goto err;
243 
244 	return 0;
245 err:
246 	return -ENOMEM;
247 }
248 
249 
250 void hv_synic_free(void)
251 {
252 	int cpu;
253 
254 	for_each_present_cpu(cpu) {
255 		struct hv_per_cpu_context *hv_cpu
256 			= per_cpu_ptr(hv_context.cpu_context, cpu);
257 
258 		if (hv_cpu->synic_event_page)
259 			free_page((unsigned long)hv_cpu->synic_event_page);
260 		if (hv_cpu->synic_message_page)
261 			free_page((unsigned long)hv_cpu->synic_message_page);
262 		if (hv_cpu->post_msg_page)
263 			free_page((unsigned long)hv_cpu->post_msg_page);
264 	}
265 
266 	kfree(hv_context.hv_numa_map);
267 }
268 
269 /*
270  * hv_synic_init - Initialize the Synthetic Interrupt Controller.
271  *
272  * If it is already initialized by another entity (ie x2v shim), we need to
273  * retrieve the initialized message and event pages.  Otherwise, we create and
274  * initialize the message and event pages.
275  */
276 int hv_synic_init(unsigned int cpu)
277 {
278 	struct hv_per_cpu_context *hv_cpu
279 		= per_cpu_ptr(hv_context.cpu_context, cpu);
280 	union hv_synic_simp simp;
281 	union hv_synic_siefp siefp;
282 	union hv_synic_sint shared_sint;
283 	union hv_synic_scontrol sctrl;
284 
285 	/* Setup the Synic's message page */
286 	hv_get_simp(simp.as_uint64);
287 	simp.simp_enabled = 1;
288 	simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
289 		>> PAGE_SHIFT;
290 
291 	hv_set_simp(simp.as_uint64);
292 
293 	/* Setup the Synic's event page */
294 	hv_get_siefp(siefp.as_uint64);
295 	siefp.siefp_enabled = 1;
296 	siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
297 		>> PAGE_SHIFT;
298 
299 	hv_set_siefp(siefp.as_uint64);
300 
301 	/* Setup the shared SINT. */
302 	hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
303 			    shared_sint.as_uint64);
304 
305 	shared_sint.as_uint64 = 0;
306 	shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
307 	shared_sint.masked = false;
308 	if (ms_hyperv.hints & HV_X64_DEPRECATING_AEOI_RECOMMENDED)
309 		shared_sint.auto_eoi = false;
310 	else
311 		shared_sint.auto_eoi = true;
312 
313 	hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
314 			    shared_sint.as_uint64);
315 
316 	/* Enable the global synic bit */
317 	hv_get_synic_state(sctrl.as_uint64);
318 	sctrl.enable = 1;
319 
320 	hv_set_synic_state(sctrl.as_uint64);
321 
322 	hv_context.synic_initialized = true;
323 
324 	/*
325 	 * Register the per-cpu clockevent source.
326 	 */
327 	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
328 		clockevents_config_and_register(hv_cpu->clk_evt,
329 						HV_TIMER_FREQUENCY,
330 						HV_MIN_DELTA_TICKS,
331 						HV_MAX_MAX_DELTA_TICKS);
332 	return 0;
333 }
334 
335 /*
336  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
337  */
338 void hv_synic_clockevents_cleanup(void)
339 {
340 	int cpu;
341 
342 	if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
343 		return;
344 
345 	if (direct_mode_enabled)
346 		hv_remove_stimer0_irq(stimer0_irq);
347 
348 	for_each_present_cpu(cpu) {
349 		struct hv_per_cpu_context *hv_cpu
350 			= per_cpu_ptr(hv_context.cpu_context, cpu);
351 
352 		clockevents_unbind_device(hv_cpu->clk_evt, cpu);
353 	}
354 }
355 
356 /*
357  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
358  */
359 int hv_synic_cleanup(unsigned int cpu)
360 {
361 	union hv_synic_sint shared_sint;
362 	union hv_synic_simp simp;
363 	union hv_synic_siefp siefp;
364 	union hv_synic_scontrol sctrl;
365 	struct vmbus_channel *channel, *sc;
366 	bool channel_found = false;
367 	unsigned long flags;
368 
369 	if (!hv_context.synic_initialized)
370 		return -EFAULT;
371 
372 	/*
373 	 * Search for channels which are bound to the CPU we're about to
374 	 * cleanup. In case we find one and vmbus is still connected we need to
375 	 * fail, this will effectively prevent CPU offlining. There is no way
376 	 * we can re-bind channels to different CPUs for now.
377 	 */
378 	mutex_lock(&vmbus_connection.channel_mutex);
379 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
380 		if (channel->target_cpu == cpu) {
381 			channel_found = true;
382 			break;
383 		}
384 		spin_lock_irqsave(&channel->lock, flags);
385 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
386 			if (sc->target_cpu == cpu) {
387 				channel_found = true;
388 				break;
389 			}
390 		}
391 		spin_unlock_irqrestore(&channel->lock, flags);
392 		if (channel_found)
393 			break;
394 	}
395 	mutex_unlock(&vmbus_connection.channel_mutex);
396 
397 	if (channel_found && vmbus_connection.conn_state == CONNECTED)
398 		return -EBUSY;
399 
400 	/* Turn off clockevent device */
401 	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) {
402 		struct hv_per_cpu_context *hv_cpu
403 			= this_cpu_ptr(hv_context.cpu_context);
404 
405 		clockevents_unbind_device(hv_cpu->clk_evt, cpu);
406 		hv_ce_shutdown(hv_cpu->clk_evt);
407 		put_cpu_ptr(hv_cpu);
408 	}
409 
410 	hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
411 			    shared_sint.as_uint64);
412 
413 	shared_sint.masked = 1;
414 
415 	/* Need to correctly cleanup in the case of SMP!!! */
416 	/* Disable the interrupt */
417 	hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
418 			    shared_sint.as_uint64);
419 
420 	hv_get_simp(simp.as_uint64);
421 	simp.simp_enabled = 0;
422 	simp.base_simp_gpa = 0;
423 
424 	hv_set_simp(simp.as_uint64);
425 
426 	hv_get_siefp(siefp.as_uint64);
427 	siefp.siefp_enabled = 0;
428 	siefp.base_siefp_gpa = 0;
429 
430 	hv_set_siefp(siefp.as_uint64);
431 
432 	/* Disable the global synic bit */
433 	hv_get_synic_state(sctrl.as_uint64);
434 	sctrl.enable = 0;
435 	hv_set_synic_state(sctrl.as_uint64);
436 
437 	return 0;
438 }
439