xref: /linux/drivers/hv/hv.c (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
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/kernel.h>
12 #include <linux/mm.h>
13 #include <linux/slab.h>
14 #include <linux/vmalloc.h>
15 #include <linux/hyperv.h>
16 #include <linux/version.h>
17 #include <linux/random.h>
18 #include <linux/clockchips.h>
19 #include <clocksource/hyperv_timer.h>
20 #include <asm/mshyperv.h>
21 #include "hyperv_vmbus.h"
22 
23 /* The one and only */
24 struct hv_context hv_context;
25 
26 /*
27  * hv_init - Main initialization routine.
28  *
29  * This routine must be called before any other routines in here are called
30  */
31 int hv_init(void)
32 {
33 	hv_context.cpu_context = alloc_percpu(struct hv_per_cpu_context);
34 	if (!hv_context.cpu_context)
35 		return -ENOMEM;
36 	return 0;
37 }
38 
39 /*
40  * hv_post_message - Post a message using the hypervisor message IPC.
41  *
42  * This involves a hypercall.
43  */
44 int hv_post_message(union hv_connection_id connection_id,
45 		  enum hv_message_type message_type,
46 		  void *payload, size_t payload_size)
47 {
48 	struct hv_input_post_message *aligned_msg;
49 	struct hv_per_cpu_context *hv_cpu;
50 	u64 status;
51 
52 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
53 		return -EMSGSIZE;
54 
55 	hv_cpu = get_cpu_ptr(hv_context.cpu_context);
56 	aligned_msg = hv_cpu->post_msg_page;
57 	aligned_msg->connectionid = connection_id;
58 	aligned_msg->reserved = 0;
59 	aligned_msg->message_type = message_type;
60 	aligned_msg->payload_size = payload_size;
61 	memcpy((void *)aligned_msg->payload, payload, payload_size);
62 
63 	status = hv_do_hypercall(HVCALL_POST_MESSAGE, aligned_msg, NULL);
64 
65 	/* Preemption must remain disabled until after the hypercall
66 	 * so some other thread can't get scheduled onto this cpu and
67 	 * corrupt the per-cpu post_msg_page
68 	 */
69 	put_cpu_ptr(hv_cpu);
70 
71 	return status & 0xFFFF;
72 }
73 
74 int hv_synic_alloc(void)
75 {
76 	int cpu;
77 	struct hv_per_cpu_context *hv_cpu;
78 
79 	/*
80 	 * First, zero all per-cpu memory areas so hv_synic_free() can
81 	 * detect what memory has been allocated and cleanup properly
82 	 * after any failures.
83 	 */
84 	for_each_present_cpu(cpu) {
85 		hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
86 		memset(hv_cpu, 0, sizeof(*hv_cpu));
87 	}
88 
89 	hv_context.hv_numa_map = kcalloc(nr_node_ids, sizeof(struct cpumask),
90 					 GFP_KERNEL);
91 	if (hv_context.hv_numa_map == NULL) {
92 		pr_err("Unable to allocate NUMA map\n");
93 		goto err;
94 	}
95 
96 	for_each_present_cpu(cpu) {
97 		hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
98 
99 		tasklet_init(&hv_cpu->msg_dpc,
100 			     vmbus_on_msg_dpc, (unsigned long) hv_cpu);
101 
102 		hv_cpu->synic_message_page =
103 			(void *)get_zeroed_page(GFP_ATOMIC);
104 		if (hv_cpu->synic_message_page == NULL) {
105 			pr_err("Unable to allocate SYNIC message page\n");
106 			goto err;
107 		}
108 
109 		hv_cpu->synic_event_page = (void *)get_zeroed_page(GFP_ATOMIC);
110 		if (hv_cpu->synic_event_page == NULL) {
111 			pr_err("Unable to allocate SYNIC event page\n");
112 			goto err;
113 		}
114 
115 		hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
116 		if (hv_cpu->post_msg_page == NULL) {
117 			pr_err("Unable to allocate post msg page\n");
118 			goto err;
119 		}
120 
121 		INIT_LIST_HEAD(&hv_cpu->chan_list);
122 	}
123 
124 	return 0;
125 err:
126 	/*
127 	 * Any memory allocations that succeeded will be freed when
128 	 * the caller cleans up by calling hv_synic_free()
129 	 */
130 	return -ENOMEM;
131 }
132 
133 
134 void hv_synic_free(void)
135 {
136 	int cpu;
137 
138 	for_each_present_cpu(cpu) {
139 		struct hv_per_cpu_context *hv_cpu
140 			= per_cpu_ptr(hv_context.cpu_context, cpu);
141 
142 		free_page((unsigned long)hv_cpu->synic_event_page);
143 		free_page((unsigned long)hv_cpu->synic_message_page);
144 		free_page((unsigned long)hv_cpu->post_msg_page);
145 	}
146 
147 	kfree(hv_context.hv_numa_map);
148 }
149 
150 /*
151  * hv_synic_init - Initialize the Synthetic Interrupt Controller.
152  *
153  * If it is already initialized by another entity (ie x2v shim), we need to
154  * retrieve the initialized message and event pages.  Otherwise, we create and
155  * initialize the message and event pages.
156  */
157 void hv_synic_enable_regs(unsigned int cpu)
158 {
159 	struct hv_per_cpu_context *hv_cpu
160 		= per_cpu_ptr(hv_context.cpu_context, cpu);
161 	union hv_synic_simp simp;
162 	union hv_synic_siefp siefp;
163 	union hv_synic_sint shared_sint;
164 	union hv_synic_scontrol sctrl;
165 
166 	/* Setup the Synic's message page */
167 	hv_get_simp(simp.as_uint64);
168 	simp.simp_enabled = 1;
169 	simp.base_simp_gpa = virt_to_phys(hv_cpu->synic_message_page)
170 		>> PAGE_SHIFT;
171 
172 	hv_set_simp(simp.as_uint64);
173 
174 	/* Setup the Synic's event page */
175 	hv_get_siefp(siefp.as_uint64);
176 	siefp.siefp_enabled = 1;
177 	siefp.base_siefp_gpa = virt_to_phys(hv_cpu->synic_event_page)
178 		>> PAGE_SHIFT;
179 
180 	hv_set_siefp(siefp.as_uint64);
181 
182 	/* Setup the shared SINT. */
183 	hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
184 
185 	shared_sint.vector = HYPERVISOR_CALLBACK_VECTOR;
186 	shared_sint.masked = false;
187 	if (ms_hyperv.hints & HV_DEPRECATING_AEOI_RECOMMENDED)
188 		shared_sint.auto_eoi = false;
189 	else
190 		shared_sint.auto_eoi = true;
191 
192 	hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
193 
194 	/* Enable the global synic bit */
195 	hv_get_synic_state(sctrl.as_uint64);
196 	sctrl.enable = 1;
197 
198 	hv_set_synic_state(sctrl.as_uint64);
199 }
200 
201 int hv_synic_init(unsigned int cpu)
202 {
203 	hv_synic_enable_regs(cpu);
204 
205 	hv_stimer_init(cpu);
206 
207 	return 0;
208 }
209 
210 /*
211  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
212  */
213 void hv_synic_disable_regs(unsigned int cpu)
214 {
215 	union hv_synic_sint shared_sint;
216 	union hv_synic_simp simp;
217 	union hv_synic_siefp siefp;
218 	union hv_synic_scontrol sctrl;
219 
220 	hv_get_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
221 
222 	shared_sint.masked = 1;
223 
224 	/* Need to correctly cleanup in the case of SMP!!! */
225 	/* Disable the interrupt */
226 	hv_set_synint_state(VMBUS_MESSAGE_SINT, shared_sint.as_uint64);
227 
228 	hv_get_simp(simp.as_uint64);
229 	simp.simp_enabled = 0;
230 	simp.base_simp_gpa = 0;
231 
232 	hv_set_simp(simp.as_uint64);
233 
234 	hv_get_siefp(siefp.as_uint64);
235 	siefp.siefp_enabled = 0;
236 	siefp.base_siefp_gpa = 0;
237 
238 	hv_set_siefp(siefp.as_uint64);
239 
240 	/* Disable the global synic bit */
241 	hv_get_synic_state(sctrl.as_uint64);
242 	sctrl.enable = 0;
243 	hv_set_synic_state(sctrl.as_uint64);
244 }
245 
246 int hv_synic_cleanup(unsigned int cpu)
247 {
248 	struct vmbus_channel *channel, *sc;
249 	bool channel_found = false;
250 	unsigned long flags;
251 
252 	/*
253 	 * Search for channels which are bound to the CPU we're about to
254 	 * cleanup. In case we find one and vmbus is still connected we need to
255 	 * fail, this will effectively prevent CPU offlining. There is no way
256 	 * we can re-bind channels to different CPUs for now.
257 	 */
258 	mutex_lock(&vmbus_connection.channel_mutex);
259 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
260 		if (channel->target_cpu == cpu) {
261 			channel_found = true;
262 			break;
263 		}
264 		spin_lock_irqsave(&channel->lock, flags);
265 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
266 			if (sc->target_cpu == cpu) {
267 				channel_found = true;
268 				break;
269 			}
270 		}
271 		spin_unlock_irqrestore(&channel->lock, flags);
272 		if (channel_found)
273 			break;
274 	}
275 	mutex_unlock(&vmbus_connection.channel_mutex);
276 
277 	if (channel_found && vmbus_connection.conn_state == CONNECTED)
278 		return -EBUSY;
279 
280 	hv_stimer_cleanup(cpu);
281 
282 	hv_synic_disable_regs(cpu);
283 
284 	return 0;
285 }
286