xref: /openbmc/linux/drivers/hv/hv.c (revision e6dec923)
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 */
37 struct hv_context hv_context = {
38 	.synic_initialized	= false,
39 };
40 
41 #define HV_TIMER_FREQUENCY (10 * 1000 * 1000) /* 100ns period */
42 #define HV_MAX_MAX_DELTA_TICKS 0xffffffff
43 #define HV_MIN_DELTA_TICKS 1
44 
45 /*
46  * hv_init - Main initialization routine.
47  *
48  * This routine must be called before any other routines in here are called
49  */
50 int hv_init(void)
51 {
52 	if (!hv_is_hypercall_page_setup())
53 		return -ENOTSUPP;
54 
55 	hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
56 	if (!hv_context.cpu_context)
57 		return -ENOMEM;
58 
59 	return 0;
60 }
61 
62 /*
63  * hv_post_message - Post a message using the hypervisor message IPC.
64  *
65  * This involves a hypercall.
66  */
67 int hv_post_message(union hv_connection_id connection_id,
68 		  enum hv_message_type message_type,
69 		  void *payload, size_t payload_size)
70 {
71 	struct hv_input_post_message *aligned_msg;
72 	struct hv_per_cpu_context *hv_cpu;
73 	u64 status;
74 
75 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
76 		return -EMSGSIZE;
77 
78 	hv_cpu = get_cpu_ptr(hv_context.cpu_context);
79 	aligned_msg = hv_cpu->post_msg_page;
80 	aligned_msg->connectionid = connection_id;
81 	aligned_msg->reserved = 0;
82 	aligned_msg->message_type = message_type;
83 	aligned_msg->payload_size = payload_size;
84 	memcpy((void *)aligned_msg->payload, payload, payload_size);
85 
86 	status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
87 
88 	/* Preemption must remain disabled until after the hypercall
89 	 * so some other thread can't get scheduled onto this cpu and
90 	 * corrupt the per-cpu post_msg_page
91 	 */
92 	put_cpu_ptr(hv_cpu);
93 
94 	return status & 0xFFFF;
95 }
96 
97 static int hv_ce_set_next_event(unsigned long delta,
98 				struct clock_event_device *evt)
99 {
100 	u64 current_tick;
101 
102 	WARN_ON(!clockevent_state_oneshot(evt));
103 
104 	current_tick = hyperv_cs->read(NULL);
105 	current_tick += delta;
106 	hv_init_timer(HV_X64_MSR_STIMER0_COUNT, current_tick);
107 	return 0;
108 }
109 
110 static int hv_ce_shutdown(struct clock_event_device *evt)
111 {
112 	hv_init_timer(HV_X64_MSR_STIMER0_COUNT, 0);
113 	hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, 0);
114 
115 	return 0;
116 }
117 
118 static int hv_ce_set_oneshot(struct clock_event_device *evt)
119 {
120 	union hv_timer_config timer_cfg;
121 
122 	timer_cfg.enable = 1;
123 	timer_cfg.auto_enable = 1;
124 	timer_cfg.sintx = VMBUS_MESSAGE_SINT;
125 	hv_init_timer_config(HV_X64_MSR_STIMER0_CONFIG, timer_cfg.as_uint64);
126 
127 	return 0;
128 }
129 
130 static void hv_init_clockevent_device(struct clock_event_device *dev, int cpu)
131 {
132 	dev->name = "Hyper-V clockevent";
133 	dev->features = CLOCK_EVT_FEAT_ONESHOT;
134 	dev->cpumask = cpumask_of(cpu);
135 	dev->rating = 1000;
136 	/*
137 	 * Avoid settint dev->owner = THIS_MODULE deliberately as doing so will
138 	 * result in clockevents_config_and_register() taking additional
139 	 * references to the hv_vmbus module making it impossible to unload.
140 	 */
141 
142 	dev->set_state_shutdown = hv_ce_shutdown;
143 	dev->set_state_oneshot = hv_ce_set_oneshot;
144 	dev->set_next_event = hv_ce_set_next_event;
145 }
146 
147 
148 int hv_synic_alloc(void)
149 {
150 	int cpu;
151 
152 	hv_context.hv_numa_map = kzalloc(sizeof(struct cpumask) * nr_node_ids,
153 					 GFP_ATOMIC);
154 	if (hv_context.hv_numa_map == NULL) {
155 		pr_err("Unable to allocate NUMA map\n");
156 		goto err;
157 	}
158 
159 	for_each_present_cpu(cpu) {
160 		struct hv_per_cpu_context *hv_cpu
161 			= per_cpu_ptr(hv_context.cpu_context, cpu);
162 
163 		memset(hv_cpu, 0, sizeof(*hv_cpu));
164 		tasklet_init(&hv_cpu->msg_dpc,
165 			     vmbus_on_msg_dpc, (unsigned long) hv_cpu);
166 
167 		hv_cpu->clk_evt = kzalloc(sizeof(struct clock_event_device),
168 					  GFP_KERNEL);
169 		if (hv_cpu->clk_evt == NULL) {
170 			pr_err("Unable to allocate clock event device\n");
171 			goto err;
172 		}
173 		hv_init_clockevent_device(hv_cpu->clk_evt, cpu);
174 
175 		hv_cpu->synic_message_page =
176 			(void *)get_zeroed_page(GFP_ATOMIC);
177 		if (hv_cpu->synic_message_page == NULL) {
178 			pr_err("Unable to allocate SYNIC message page\n");
179 			goto err;
180 		}
181 
182 		hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC);
183 		if (hv_cpu->synic_event_page == NULL) {
184 			pr_err("Unable to allocate SYNIC event page\n");
185 			goto err;
186 		}
187 
188 		hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
189 		if (hv_cpu->post_msg_page == NULL) {
190 			pr_err("Unable to allocate post msg page\n");
191 			goto err;
192 		}
193 
194 		INIT_LIST_HEAD(&hv_cpu->chan_list);
195 	}
196 
197 	return 0;
198 err:
199 	return -ENOMEM;
200 }
201 
202 
203 void hv_synic_free(void)
204 {
205 	int cpu;
206 
207 	for_each_present_cpu(cpu) {
208 		struct hv_per_cpu_context *hv_cpu
209 			= per_cpu_ptr(hv_context.cpu_context, cpu);
210 
211 		if (hv_cpu->synic_event_page)
212 			free_page((unsigned long)hv_cpu->synic_event_page);
213 		if (hv_cpu->synic_message_page)
214 			free_page((unsigned long)hv_cpu->synic_message_page);
215 		if (hv_cpu->post_msg_page)
216 			free_page((unsigned long)hv_cpu->post_msg_page);
217 	}
218 
219 	kfree(hv_context.hv_numa_map);
220 }
221 
222 /*
223  * hv_synic_init - Initialize the Synthethic Interrupt Controller.
224  *
225  * If it is already initialized by another entity (ie x2v shim), we need to
226  * retrieve the initialized message and event pages.  Otherwise, we create and
227  * initialize the message and event pages.
228  */
229 int hv_synic_init(unsigned int cpu)
230 {
231 	struct hv_per_cpu_context *hv_cpu
232 		= per_cpu_ptr(hv_context.cpu_context, cpu);
233 	union hv_synic_simp simp;
234 	union hv_synic_siefp siefp;
235 	union hv_synic_sint shared_sint;
236 	union hv_synic_scontrol sctrl;
237 	u64 vp_index;
238 
239 	/* Setup the Synic's message page */
240 	hv_get_simp(simp.as_uint64);
241 	simp.simp_enabled = 1;
242 	simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
243 		>> PAGE_SHIFT;
244 
245 	hv_set_simp(simp.as_uint64);
246 
247 	/* Setup the Synic's event page */
248 	hv_get_siefp(siefp.as_uint64);
249 	siefp.siefp_enabled = 1;
250 	siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
251 		>> PAGE_SHIFT;
252 
253 	hv_set_siefp(siefp.as_uint64);
254 
255 	/* Setup the shared SINT. */
256 	hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
257 			    shared_sint.as_uint64);
258 
259 	shared_sint.as_uint64 = 0;
260 	shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
261 	shared_sint.masked = false;
262 	if (ms_hyperv.hints & HV_X64_DEPRECATING_AEOI_RECOMMENDED)
263 		shared_sint.auto_eoi = false;
264 	else
265 		shared_sint.auto_eoi = true;
266 
267 	hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
268 			    shared_sint.as_uint64);
269 
270 	/* Enable the global synic bit */
271 	hv_get_synic_state(sctrl.as_uint64);
272 	sctrl.enable = 1;
273 
274 	hv_set_synic_state(sctrl.as_uint64);
275 
276 	hv_context.synic_initialized = true;
277 
278 	/*
279 	 * Setup the mapping between Hyper-V's notion
280 	 * of cpuid and Linux' notion of cpuid.
281 	 * This array will be indexed using Linux cpuid.
282 	 */
283 	hv_get_vp_index(vp_index);
284 	hv_context.vp_index[cpu] = (u32)vp_index;
285 
286 	/*
287 	 * Register the per-cpu clockevent source.
288 	 */
289 	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE)
290 		clockevents_config_and_register(hv_cpu->clk_evt,
291 						HV_TIMER_FREQUENCY,
292 						HV_MIN_DELTA_TICKS,
293 						HV_MAX_MAX_DELTA_TICKS);
294 	return 0;
295 }
296 
297 /*
298  * hv_synic_clockevents_cleanup - Cleanup clockevent devices
299  */
300 void hv_synic_clockevents_cleanup(void)
301 {
302 	int cpu;
303 
304 	if (!(ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE))
305 		return;
306 
307 	for_each_present_cpu(cpu) {
308 		struct hv_per_cpu_context *hv_cpu
309 			= per_cpu_ptr(hv_context.cpu_context, cpu);
310 
311 		clockevents_unbind_device(hv_cpu->clk_evt, cpu);
312 	}
313 }
314 
315 /*
316  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
317  */
318 int hv_synic_cleanup(unsigned int cpu)
319 {
320 	union hv_synic_sint shared_sint;
321 	union hv_synic_simp simp;
322 	union hv_synic_siefp siefp;
323 	union hv_synic_scontrol sctrl;
324 	struct vmbus_channel *channel, *sc;
325 	bool channel_found = false;
326 	unsigned long flags;
327 
328 	if (!hv_context.synic_initialized)
329 		return -EFAULT;
330 
331 	/*
332 	 * Search for channels which are bound to the CPU we're about to
333 	 * cleanup. In case we find one and vmbus is still connected we need to
334 	 * fail, this will effectively prevent CPU offlining. There is no way
335 	 * we can re-bind channels to different CPUs for now.
336 	 */
337 	mutex_lock(&vmbus_connection.channel_mutex);
338 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
339 		if (channel->target_cpu == cpu) {
340 			channel_found = true;
341 			break;
342 		}
343 		spin_lock_irqsave(&channel->lock, flags);
344 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
345 			if (sc->target_cpu == cpu) {
346 				channel_found = true;
347 				break;
348 			}
349 		}
350 		spin_unlock_irqrestore(&channel->lock, flags);
351 		if (channel_found)
352 			break;
353 	}
354 	mutex_unlock(&vmbus_connection.channel_mutex);
355 
356 	if (channel_found && vmbus_connection.conn_state == CONNECTED)
357 		return -EBUSY;
358 
359 	/* Turn off clockevent device */
360 	if (ms_hyperv.features & HV_X64_MSR_SYNTIMER_AVAILABLE) {
361 		struct hv_per_cpu_context *hv_cpu
362 			= this_cpu_ptr(hv_context.cpu_context);
363 
364 		clockevents_unbind_device(hv_cpu->clk_evt, cpu);
365 		hv_ce_shutdown(hv_cpu->clk_evt);
366 		put_cpu_ptr(hv_cpu);
367 	}
368 
369 	hv_get_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
370 			    shared_sint.as_uint64);
371 
372 	shared_sint.masked = 1;
373 
374 	/* Need to correctly cleanup in the case of SMP!!! */
375 	/* Disable the interrupt */
376 	hv_set_synint_state(HV_X64_MSR_SINT0 + VMBUS_MESSAGE_SINT,
377 			    shared_sint.as_uint64);
378 
379 	hv_get_simp(simp.as_uint64);
380 	simp.simp_enabled = 0;
381 	simp.base_simp_gpa = 0;
382 
383 	hv_set_simp(simp.as_uint64);
384 
385 	hv_get_siefp(siefp.as_uint64);
386 	siefp.siefp_enabled = 0;
387 	siefp.base_siefp_gpa = 0;
388 
389 	hv_set_siefp(siefp.as_uint64);
390 
391 	/* Disable the global synic bit */
392 	hv_get_synic_state(sctrl.as_uint64);
393 	sctrl.enable = 0;
394 	hv_set_synic_state(sctrl.as_uint64);
395 
396 	return 0;
397 }
398