Merge tag 'omap-for-v4.7-dts-fixes1' of git://git.kernel.org/pub/scm/linux/kernel...
[deliverable/linux.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
43 /*
44 * Negotiated protocol version with the host.
45 */
46 __u32 vmbus_proto_version;
47 EXPORT_SYMBOL_GPL(vmbus_proto_version);
48
49 static __u32 vmbus_get_next_version(__u32 current_version)
50 {
51 switch (current_version) {
52 case (VERSION_WIN7):
53 return VERSION_WS2008;
54
55 case (VERSION_WIN8):
56 return VERSION_WIN7;
57
58 case (VERSION_WIN8_1):
59 return VERSION_WIN8;
60
61 case (VERSION_WIN10):
62 return VERSION_WIN8_1;
63
64 case (VERSION_WS2008):
65 default:
66 return VERSION_INVAL;
67 }
68 }
69
70 static int vmbus_negotiate_version(struct vmbus_channel_msginfo *msginfo,
71 __u32 version)
72 {
73 int ret = 0;
74 struct vmbus_channel_initiate_contact *msg;
75 unsigned long flags;
76
77 init_completion(&msginfo->waitevent);
78
79 msg = (struct vmbus_channel_initiate_contact *)msginfo->msg;
80
81 msg->header.msgtype = CHANNELMSG_INITIATE_CONTACT;
82 msg->vmbus_version_requested = version;
83 msg->interrupt_page = virt_to_phys(vmbus_connection.int_page);
84 msg->monitor_page1 = virt_to_phys(vmbus_connection.monitor_pages[0]);
85 msg->monitor_page2 = virt_to_phys(vmbus_connection.monitor_pages[1]);
86 /*
87 * We want all channel messages to be delivered on CPU 0.
88 * This has been the behavior pre-win8. This is not
89 * perf issue and having all channel messages delivered on CPU 0
90 * would be ok.
91 * For post win8 hosts, we support receiving channel messagges on
92 * all the CPUs. This is needed for kexec to work correctly where
93 * the CPU attempting to connect may not be CPU 0.
94 */
95 if (version >= VERSION_WIN8_1) {
96 msg->target_vcpu = hv_context.vp_index[get_cpu()];
97 put_cpu();
98 } else {
99 msg->target_vcpu = 0;
100 }
101
102 /*
103 * Add to list before we send the request since we may
104 * receive the response before returning from this routine
105 */
106 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
107 list_add_tail(&msginfo->msglistentry,
108 &vmbus_connection.chn_msg_list);
109
110 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
111
112 ret = vmbus_post_msg(msg,
113 sizeof(struct vmbus_channel_initiate_contact));
114 if (ret != 0) {
115 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
116 list_del(&msginfo->msglistentry);
117 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock,
118 flags);
119 return ret;
120 }
121
122 /* Wait for the connection response */
123 wait_for_completion(&msginfo->waitevent);
124
125 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
126 list_del(&msginfo->msglistentry);
127 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
128
129 /* Check if successful */
130 if (msginfo->response.version_response.version_supported) {
131 vmbus_connection.conn_state = CONNECTED;
132 } else {
133 return -ECONNREFUSED;
134 }
135
136 return ret;
137 }
138
139 /*
140 * vmbus_connect - Sends a connect request on the partition service connection
141 */
142 int vmbus_connect(void)
143 {
144 int ret = 0;
145 struct vmbus_channel_msginfo *msginfo = NULL;
146 __u32 version;
147
148 /* Initialize the vmbus connection */
149 vmbus_connection.conn_state = CONNECTING;
150 vmbus_connection.work_queue = create_workqueue("hv_vmbus_con");
151 if (!vmbus_connection.work_queue) {
152 ret = -ENOMEM;
153 goto cleanup;
154 }
155
156 INIT_LIST_HEAD(&vmbus_connection.chn_msg_list);
157 spin_lock_init(&vmbus_connection.channelmsg_lock);
158
159 INIT_LIST_HEAD(&vmbus_connection.chn_list);
160 mutex_init(&vmbus_connection.channel_mutex);
161
162 /*
163 * Setup the vmbus event connection for channel interrupt
164 * abstraction stuff
165 */
166 vmbus_connection.int_page =
167 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, 0);
168 if (vmbus_connection.int_page == NULL) {
169 ret = -ENOMEM;
170 goto cleanup;
171 }
172
173 vmbus_connection.recv_int_page = vmbus_connection.int_page;
174 vmbus_connection.send_int_page =
175 (void *)((unsigned long)vmbus_connection.int_page +
176 (PAGE_SIZE >> 1));
177
178 /*
179 * Setup the monitor notification facility. The 1st page for
180 * parent->child and the 2nd page for child->parent
181 */
182 vmbus_connection.monitor_pages[0] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0);
183 vmbus_connection.monitor_pages[1] = (void *)__get_free_pages((GFP_KERNEL|__GFP_ZERO), 0);
184 if ((vmbus_connection.monitor_pages[0] == NULL) ||
185 (vmbus_connection.monitor_pages[1] == NULL)) {
186 ret = -ENOMEM;
187 goto cleanup;
188 }
189
190 msginfo = kzalloc(sizeof(*msginfo) +
191 sizeof(struct vmbus_channel_initiate_contact),
192 GFP_KERNEL);
193 if (msginfo == NULL) {
194 ret = -ENOMEM;
195 goto cleanup;
196 }
197
198 /*
199 * Negotiate a compatible VMBUS version number with the
200 * host. We start with the highest number we can support
201 * and work our way down until we negotiate a compatible
202 * version.
203 */
204
205 version = VERSION_CURRENT;
206
207 do {
208 ret = vmbus_negotiate_version(msginfo, version);
209 if (ret == -ETIMEDOUT)
210 goto cleanup;
211
212 if (vmbus_connection.conn_state == CONNECTED)
213 break;
214
215 version = vmbus_get_next_version(version);
216 } while (version != VERSION_INVAL);
217
218 if (version == VERSION_INVAL)
219 goto cleanup;
220
221 vmbus_proto_version = version;
222 pr_info("Hyper-V Host Build:%d-%d.%d-%d-%d.%d; Vmbus version:%d.%d\n",
223 host_info_eax, host_info_ebx >> 16,
224 host_info_ebx & 0xFFFF, host_info_ecx,
225 host_info_edx >> 24, host_info_edx & 0xFFFFFF,
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 * Map the given relid to the corresponding channel based on the
267 * per-cpu list of channels that have been affinitized to this CPU.
268 * This will be used in the channel callback path as we can do this
269 * mapping in a lock-free fashion.
270 */
271 static struct vmbus_channel *pcpu_relid2channel(u32 relid)
272 {
273 struct vmbus_channel *channel;
274 struct vmbus_channel *found_channel = NULL;
275 int cpu = smp_processor_id();
276 struct list_head *pcpu_head = &hv_context.percpu_list[cpu];
277
278 list_for_each_entry(channel, pcpu_head, percpu_list) {
279 if (channel->offermsg.child_relid == relid) {
280 found_channel = channel;
281 break;
282 }
283 }
284
285 return found_channel;
286 }
287
288 /*
289 * relid2channel - Get the channel object given its
290 * child relative id (ie channel id)
291 */
292 struct vmbus_channel *relid2channel(u32 relid)
293 {
294 struct vmbus_channel *channel;
295 struct vmbus_channel *found_channel = NULL;
296 struct list_head *cur, *tmp;
297 struct vmbus_channel *cur_sc;
298
299 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
300
301 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
302 if (channel->offermsg.child_relid == relid) {
303 found_channel = channel;
304 break;
305 } else if (!list_empty(&channel->sc_list)) {
306 /*
307 * Deal with sub-channels.
308 */
309 list_for_each_safe(cur, tmp, &channel->sc_list) {
310 cur_sc = list_entry(cur, struct vmbus_channel,
311 sc_list);
312 if (cur_sc->offermsg.child_relid == relid) {
313 found_channel = cur_sc;
314 break;
315 }
316 }
317 }
318 }
319
320 return found_channel;
321 }
322
323 /*
324 * process_chn_event - Process a channel event notification
325 */
326 static void process_chn_event(u32 relid)
327 {
328 struct vmbus_channel *channel;
329 void *arg;
330 bool read_state;
331 u32 bytes_to_read;
332
333 /*
334 * Find the channel based on this relid and invokes the
335 * channel callback to process the event
336 */
337 channel = pcpu_relid2channel(relid);
338
339 if (!channel)
340 return;
341
342 /*
343 * A channel once created is persistent even when there
344 * is no driver handling the device. An unloading driver
345 * sets the onchannel_callback to NULL on the same CPU
346 * as where this interrupt is handled (in an interrupt context).
347 * Thus, checking and invoking the driver specific callback takes
348 * care of orderly unloading of the driver.
349 */
350
351 if (channel->onchannel_callback != NULL) {
352 arg = channel->channel_callback_context;
353 read_state = channel->batched_reading;
354 /*
355 * This callback reads the messages sent by the host.
356 * We can optimize host to guest signaling by ensuring:
357 * 1. While reading the channel, we disable interrupts from
358 * host.
359 * 2. Ensure that we process all posted messages from the host
360 * before returning from this callback.
361 * 3. Once we return, enable signaling from the host. Once this
362 * state is set we check to see if additional packets are
363 * available to read. In this case we repeat the process.
364 */
365
366 do {
367 if (read_state)
368 hv_begin_read(&channel->inbound);
369 channel->onchannel_callback(arg);
370 if (read_state)
371 bytes_to_read = hv_end_read(&channel->inbound);
372 else
373 bytes_to_read = 0;
374 } while (read_state && (bytes_to_read != 0));
375 }
376 }
377
378 /*
379 * vmbus_on_event - Handler for events
380 */
381 void vmbus_on_event(unsigned long data)
382 {
383 u32 dword;
384 u32 maxdword;
385 int bit;
386 u32 relid;
387 u32 *recv_int_page = NULL;
388 void *page_addr;
389 int cpu = smp_processor_id();
390 union hv_synic_event_flags *event;
391
392 if (vmbus_proto_version < VERSION_WIN8) {
393 maxdword = MAX_NUM_CHANNELS_SUPPORTED >> 5;
394 recv_int_page = vmbus_connection.recv_int_page;
395 } else {
396 /*
397 * When the host is win8 and beyond, the event page
398 * can be directly checked to get the id of the channel
399 * that has the interrupt pending.
400 */
401 maxdword = HV_EVENT_FLAGS_DWORD_COUNT;
402 page_addr = hv_context.synic_event_page[cpu];
403 event = (union hv_synic_event_flags *)page_addr +
404 VMBUS_MESSAGE_SINT;
405 recv_int_page = event->flags32;
406 }
407
408
409
410 /* Check events */
411 if (!recv_int_page)
412 return;
413 for (dword = 0; dword < maxdword; dword++) {
414 if (!recv_int_page[dword])
415 continue;
416 for (bit = 0; bit < 32; bit++) {
417 if (sync_test_and_clear_bit(bit,
418 (unsigned long *)&recv_int_page[dword])) {
419 relid = (dword << 5) + bit;
420
421 if (relid == 0)
422 /*
423 * Special case - vmbus
424 * channel protocol msg
425 */
426 continue;
427
428 process_chn_event(relid);
429 }
430 }
431 }
432 }
433
434 /*
435 * vmbus_post_msg - Send a msg on the vmbus's message connection
436 */
437 int vmbus_post_msg(void *buffer, size_t buflen)
438 {
439 union hv_connection_id conn_id;
440 int ret = 0;
441 int retries = 0;
442 u32 msec = 1;
443
444 conn_id.asu32 = 0;
445 conn_id.u.id = VMBUS_MESSAGE_CONNECTION_ID;
446
447 /*
448 * hv_post_message() can have transient failures because of
449 * insufficient resources. Retry the operation a couple of
450 * times before giving up.
451 */
452 while (retries < 20) {
453 ret = hv_post_message(conn_id, 1, buffer, buflen);
454
455 switch (ret) {
456 case HV_STATUS_INVALID_CONNECTION_ID:
457 /*
458 * We could get this if we send messages too
459 * frequently.
460 */
461 ret = -EAGAIN;
462 break;
463 case HV_STATUS_INSUFFICIENT_MEMORY:
464 case HV_STATUS_INSUFFICIENT_BUFFERS:
465 ret = -ENOMEM;
466 break;
467 case HV_STATUS_SUCCESS:
468 return ret;
469 default:
470 pr_err("hv_post_msg() failed; error code:%d\n", ret);
471 return -EINVAL;
472 }
473
474 retries++;
475 msleep(msec);
476 if (msec < 2048)
477 msec *= 2;
478 }
479 return ret;
480 }
481
482 /*
483 * vmbus_set_event - Send an event notification to the parent
484 */
485 void vmbus_set_event(struct vmbus_channel *channel)
486 {
487 u32 child_relid = channel->offermsg.child_relid;
488
489 if (!channel->is_dedicated_interrupt) {
490 /* Each u32 represents 32 channels */
491 sync_set_bit(child_relid & 31,
492 (unsigned long *)vmbus_connection.send_int_page +
493 (child_relid >> 5));
494 }
495
496 hv_do_hypercall(HVCALL_SIGNAL_EVENT, channel->sig_event, NULL);
497 }
498 EXPORT_SYMBOL_GPL(vmbus_set_event);
This page took 0.071422 seconds and 5 git commands to generate.