]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/hv/connection.c
Merge tag 'vfio-ccw-20170724' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms39...
[mirror_ubuntu-artful-kernel.git] / drivers / hv / connection.c
1 /*
2 *
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
16 * Place - Suite 330, Boston, MA 02111-1307 USA.
17 *
18 * Authors:
19 * Haiyang Zhang <haiyangz@microsoft.com>
20 * Hank Janssen <hjanssen@microsoft.com>
21 *
22 */
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
25 #include <linux/kernel.h>
26 #include <linux/sched.h>
27 #include <linux/wait.h>
28 #include <linux/delay.h>
29 #include <linux/mm.h>
30 #include <linux/slab.h>
31 #include <linux/vmalloc.h>
32 #include <linux/hyperv.h>
33 #include <linux/export.h>
34 #include <asm/hyperv.h>
35 #include "hyperv_vmbus.h"
36
37
38 struct vmbus_connection vmbus_connection = {
39 .conn_state = DISCONNECTED,
40 .next_gpadl_handle = ATOMIC_INIT(0xE1E10),
41 };
42 EXPORT_SYMBOL_GPL(vmbus_connection);
43
44 /*
45 * Negotiated protocol version with the host.
46 */
47 __u32 vmbus_proto_version;
48 EXPORT_SYMBOL_GPL(vmbus_proto_version);
49
50 static __u32 vmbus_get_next_version(__u32 current_version)
51 {
52 switch (current_version) {
53 case (VERSION_WIN7):
54 return VERSION_WS2008;
55
56 case (VERSION_WIN8):
57 return VERSION_WIN7;
58
59 case (VERSION_WIN8_1):
60 return VERSION_WIN8;
61
62 case (VERSION_WIN10):
63 return VERSION_WIN8_1;
64
65 case (VERSION_WS2008):
66 default:
67 return VERSION_INVAL;
68 }
69 }
70
71 static int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo,
72 __u32 version)
73 {
74 int ret = 0;
75 struct vmbus_channel_initiate_contact *msg;
76 unsigned long flags;
77
78 init_completion(&msginfo->waitevent);
79
80 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
81
82 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
83 msg->vmbus_version_requested = version;
84 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
85 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]);
86 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]);
87 /*
88 * We want all channel messages to be delivered on CPU 0.
89 * This has been the behavior pre-win8. This is not
90 * perf issue and having all channel messages delivered on CPU 0
91 * would be ok.
92 * For post win8 hosts, we support receiving channel messagges on
93 * all the CPUs. This is needed for kexec to work correctly where
94 * the CPU attempting to connect may not be CPU 0.
95 */
96 if (version >= VERSION_WIN8_1) {
97 msg->target_vcpu = hv_context.vp_index[smp_processor_id()];
98 vmbus_connection.connect_cpu = smp_processor_id();
99 } else {
100 msg->target_vcpu = 0;
101 vmbus_connection.connect_cpu = 0;
102 }
103
104 /*
105 * Add to list before we send the request since we may
106 * receive the response before returning from this routine
107 */
108 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
109 list_add_tail(&msginfo->msglistentry,
110 &vmbus_connection.chn_msg_list);
111
112 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
113
114 ret = vmbus_post_msg(msg,
115 sizeof(struct vmbus_channel_initiate_contact),
116 true);
117 if (ret != 0) {
118 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
119 list_del(&msginfo->msglistentry);
120 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
121 flags);
122 return ret;
123 }
124
125 /* Wait for the connection response */
126 wait_for_completion(&msginfo->waitevent);
127
128 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
129 list_del(&msginfo->msglistentry);
130 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
131
132 /* Check if successful */
133 if (msginfo->response.version_response.version_supported) {
134 vmbus_connection.conn_state = CONNECTED;
135 } else {
136 return -ECONNREFUSED;
137 }
138
139 return ret;
140 }
141
142 /*
143 * vmbus_connect - Sends a connect request on the partition service connection
144 */
145 int vmbus_connect(void)
146 {
147 int ret = 0;
148 struct vmbus_channel_msginfo *msginfo = NULL;
149 __u32 version;
150
151 /* Initialize the vmbus connection */
152 vmbus_connection.conn_state = CONNECTING;
153 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
154 if (!vmbus_connection.work_queue) {
155 ret = -ENOMEM;
156 goto cleanup;
157 }
158
159 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
160 spin_lock_init(&vmbus_connection.channelmsg_lock);
161
162 INIT_LIST_HEAD(&vmbus_connection.chn_list);
163 mutex_init(&vmbus_connection.channel_mutex);
164
165 /*
166 * Setup the vmbus event connection for channel interrupt
167 * abstraction stuff
168 */
169 vmbus_connection.int_page =
170 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0);
171 if (vmbus_connection.int_page == NULL) {
172 ret = -ENOMEM;
173 goto cleanup;
174 }
175
176 vmbus_connection.recv_int_page = vmbus_connection.int_page;
177 vmbus_connection.send_int_page =
178 (void *)((unsigned long)vmbus_connection.int_page +
179 (PAGE_SIZE >> 1));
180
181 /*
182 * Setup the monitor notification facility. The 1st page for
183 * parent->child and the 2nd page for child->parent
184 */
185 vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0);
186 vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0);
187 if ((vmbus_connection.monitor_pages[0] == NULL) ||
188 (vmbus_connection.monitor_pages[1] == NULL)) {
189 ret = -ENOMEM;
190 goto cleanup;
191 }
192
193 msginfo = kzalloc(sizeof(*msginfo) +
194 sizeof(struct vmbus_channel_initiate_contact),
195 GFP_KERNEL);
196 if (msginfo == NULL) {
197 ret = -ENOMEM;
198 goto cleanup;
199 }
200
201 /*
202 * Negotiate a compatible VMBUS version number with the
203 * host. We start with the highest number we can support
204 * and work our way down until we negotiate a compatible
205 * version.
206 */
207
208 version = VERSION_CURRENT;
209
210 do {
211 ret = vmbus_negotiate_version(msginfo, version);
212 if (ret == -ETIMEDOUT)
213 goto cleanup;
214
215 if (vmbus_connection.conn_state == CONNECTED)
216 break;
217
218 version = vmbus_get_next_version(version);
219 } while (version != VERSION_INVAL);
220
221 if (version == VERSION_INVAL)
222 goto cleanup;
223
224 vmbus_proto_version = version;
225 pr_info("Vmbus version:%d.%d\n",
226 version >> 16, version & 0xFFFF);
227
228 kfree(msginfo);
229 return 0;
230
231 cleanup:
232 pr_err("Unable to connect to host\n");
233
234 vmbus_connection.conn_state = DISCONNECTED;
235 vmbus_disconnect();
236
237 kfree(msginfo);
238
239 return ret;
240 }
241
242 void vmbus_disconnect(void)
243 {
244 /*
245 * First send the unload request to the host.
246 */
247 vmbus_initiate_unload(false);
248
249 if (vmbus_connection.work_queue) {
250 drain_workqueue(vmbus_connection.work_queue);
251 destroy_workqueue(vmbus_connection.work_queue);
252 }
253
254 if (vmbus_connection.int_page) {
255 free_pages((unsigned long)vmbus_connection.int_page, 0);
256 vmbus_connection.int_page = NULL;
257 }
258
259 free_pages((unsigned long)vmbus_connection.monitor_pages[0], 0);
260 free_pages((unsigned long)vmbus_connection.monitor_pages[1], 0);
261 vmbus_connection.monitor_pages[0] = NULL;
262 vmbus_connection.monitor_pages[1] = NULL;
263 }
264
265 /*
266 * relid2channel - Get the channel object given its
267 * child relative id (ie channel id)
268 */
269 struct vmbus_channel *relid2channel(u32 relid)
270 {
271 struct vmbus_channel *channel;
272 struct vmbus_channel *found_channel = NULL;
273 struct list_head *cur, *tmp;
274 struct vmbus_channel *cur_sc;
275
276 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
277
278 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
279 if (channel->offermsg.child_relid == relid) {
280 found_channel = channel;
281 break;
282 } else if (!list_empty(&channel->sc_list)) {
283 /*
284 * Deal with sub-channels.
285 */
286 list_for_each_safe(cur, tmp, &channel->sc_list) {
287 cur_sc = list_entry(cur, struct vmbus_channel,
288 sc_list);
289 if (cur_sc->offermsg.child_relid == relid) {
290 found_channel = cur_sc;
291 break;
292 }
293 }
294 }
295 }
296
297 return found_channel;
298 }
299
300 /*
301 * vmbus_on_event - Process a channel event notification
302 *
303 * For batched channels (default) optimize host to guest signaling
304 * by ensuring:
305 * 1. While reading the channel, we disable interrupts from host.
306 * 2. Ensure that we process all posted messages from the host
307 * before returning from this callback.
308 * 3. Once we return, enable signaling from the host. Once this
309 * state is set we check to see if additional packets are
310 * available to read. In this case we repeat the process.
311 * If this tasklet has been running for a long time
312 * then reschedule ourselves.
313 */
314 void vmbus_on_event(unsigned long data)
315 {
316 struct vmbus_channel *channel = (void *) data;
317 unsigned long time_limit = jiffies + 2;
318
319 do {
320 void (*callback_fn)(void *);
321
322 /* A channel once created is persistent even when
323 * there is no driver handling the device. An
324 * unloading driver sets the onchannel_callback to NULL.
325 */
326 callback_fn = READ_ONCE(channel->onchannel_callback);
327 if (unlikely(callback_fn == NULL))
328 return;
329
330 (*callback_fn)(channel->channel_callback_context);
331
332 if (channel->callback_mode != HV_CALL_BATCHED)
333 return;
334
335 if (likely(hv_end_read(&channel->inbound) == 0))
336 return;
337
338 hv_begin_read(&channel->inbound);
339 } while (likely(time_before(jiffies, time_limit)));
340
341 /* The time limit (2 jiffies) has been reached */
342 tasklet_schedule(&channel->callback_event);
343 }
344
345 /*
346 * vmbus_post_msg - Send a msg on the vmbus's message connection
347 */
348 int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep)
349 {
350 union hv_connection_id conn_id;
351 int ret = 0;
352 int retries = 0;
353 u32 usec = 1;
354
355 conn_id.asu32 = 0;
356 conn_id.u.id = VMBUS_MESSAGE_CONNECTION_ID;
357
358 /*
359 * hv_post_message() can have transient failures because of
360 * insufficient resources. Retry the operation a couple of
361 * times before giving up.
362 */
363 while (retries < 100) {
364 ret = hv_post_message(conn_id, 1, buffer, buflen);
365
366 switch (ret) {
367 case HV_STATUS_INVALID_CONNECTION_ID:
368 /*
369 * We could get this if we send messages too
370 * frequently.
371 */
372 ret = -EAGAIN;
373 break;
374 case HV_STATUS_INSUFFICIENT_MEMORY:
375 case HV_STATUS_INSUFFICIENT_BUFFERS:
376 ret = -ENOBUFS;
377 break;
378 case HV_STATUS_SUCCESS:
379 return ret;
380 default:
381 pr_err("hv_post_msg() failed; error code:%d\n", ret);
382 return -EINVAL;
383 }
384
385 retries++;
386 if (can_sleep && usec > 1000)
387 msleep(usec / 1000);
388 else if (usec < MAX_UDELAY_MS * 1000)
389 udelay(usec);
390 else
391 mdelay(usec / 1000);
392
393 if (retries < 22)
394 usec *= 2;
395 }
396 return ret;
397 }
398
399 /*
400 * vmbus_set_event - Send an event notification to the parent
401 */
402 void vmbus_set_event(struct vmbus_channel *channel)
403 {
404 u32 child_relid = channel->offermsg.child_relid;
405
406 if (!channel->is_dedicated_interrupt)
407 vmbus_send_interrupt(child_relid);
408
409 hv_do_hypercall(HVCALL_SIGNAL_EVENT, channel->sig_event, NULL);
410 }
411 EXPORT_SYMBOL_GPL(vmbus_set_event);