Loading...
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 * K. Y. Srinivasan <kys@microsoft.com>
22 *
23 */
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/kernel.h>
27#include <linux/mm.h>
28#include <linux/hyperv.h>
29#include <linux/uio.h>
30
31#include "hyperv_vmbus.h"
32
33void hv_begin_read(struct hv_ring_buffer_info *rbi)
34{
35 rbi->ring_buffer->interrupt_mask = 1;
36 mb();
37}
38
39u32 hv_end_read(struct hv_ring_buffer_info *rbi)
40{
41 u32 read;
42 u32 write;
43
44 rbi->ring_buffer->interrupt_mask = 0;
45 mb();
46
47 /*
48 * Now check to see if the ring buffer is still empty.
49 * If it is not, we raced and we need to process new
50 * incoming messages.
51 */
52 hv_get_ringbuffer_availbytes(rbi, &read, &write);
53
54 return read;
55}
56
57/*
58 * When we write to the ring buffer, check if the host needs to
59 * be signaled. Here is the details of this protocol:
60 *
61 * 1. The host guarantees that while it is draining the
62 * ring buffer, it will set the interrupt_mask to
63 * indicate it does not need to be interrupted when
64 * new data is placed.
65 *
66 * 2. The host guarantees that it will completely drain
67 * the ring buffer before exiting the read loop. Further,
68 * once the ring buffer is empty, it will clear the
69 * interrupt_mask and re-check to see if new data has
70 * arrived.
71 */
72
73static bool hv_need_to_signal(u32 old_write, struct hv_ring_buffer_info *rbi)
74{
75 mb();
76 if (rbi->ring_buffer->interrupt_mask)
77 return false;
78
79 /* check interrupt_mask before read_index */
80 rmb();
81 /*
82 * This is the only case we need to signal when the
83 * ring transitions from being empty to non-empty.
84 */
85 if (old_write == rbi->ring_buffer->read_index)
86 return true;
87
88 return false;
89}
90
91/*
92 * To optimize the flow management on the send-side,
93 * when the sender is blocked because of lack of
94 * sufficient space in the ring buffer, potential the
95 * consumer of the ring buffer can signal the producer.
96 * This is controlled by the following parameters:
97 *
98 * 1. pending_send_sz: This is the size in bytes that the
99 * producer is trying to send.
100 * 2. The feature bit feat_pending_send_sz set to indicate if
101 * the consumer of the ring will signal when the ring
102 * state transitions from being full to a state where
103 * there is room for the producer to send the pending packet.
104 */
105
106static bool hv_need_to_signal_on_read(u32 old_rd,
107 struct hv_ring_buffer_info *rbi)
108{
109 u32 prev_write_sz;
110 u32 cur_write_sz;
111 u32 r_size;
112 u32 write_loc = rbi->ring_buffer->write_index;
113 u32 read_loc = rbi->ring_buffer->read_index;
114 u32 pending_sz = rbi->ring_buffer->pending_send_sz;
115
116 /*
117 * If the other end is not blocked on write don't bother.
118 */
119 if (pending_sz == 0)
120 return false;
121
122 r_size = rbi->ring_datasize;
123 cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) :
124 read_loc - write_loc;
125
126 prev_write_sz = write_loc >= old_rd ? r_size - (write_loc - old_rd) :
127 old_rd - write_loc;
128
129
130 if ((prev_write_sz < pending_sz) && (cur_write_sz >= pending_sz))
131 return true;
132
133 return false;
134}
135
136/*
137 * hv_get_next_write_location()
138 *
139 * Get the next write location for the specified ring buffer
140 *
141 */
142static inline u32
143hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
144{
145 u32 next = ring_info->ring_buffer->write_index;
146
147 return next;
148}
149
150/*
151 * hv_set_next_write_location()
152 *
153 * Set the next write location for the specified ring buffer
154 *
155 */
156static inline void
157hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
158 u32 next_write_location)
159{
160 ring_info->ring_buffer->write_index = next_write_location;
161}
162
163/*
164 * hv_get_next_read_location()
165 *
166 * Get the next read location for the specified ring buffer
167 */
168static inline u32
169hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
170{
171 u32 next = ring_info->ring_buffer->read_index;
172
173 return next;
174}
175
176/*
177 * hv_get_next_readlocation_withoffset()
178 *
179 * Get the next read location + offset for the specified ring buffer.
180 * This allows the caller to skip
181 */
182static inline u32
183hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
184 u32 offset)
185{
186 u32 next = ring_info->ring_buffer->read_index;
187
188 next += offset;
189 next %= ring_info->ring_datasize;
190
191 return next;
192}
193
194/*
195 *
196 * hv_set_next_read_location()
197 *
198 * Set the next read location for the specified ring buffer
199 *
200 */
201static inline void
202hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
203 u32 next_read_location)
204{
205 ring_info->ring_buffer->read_index = next_read_location;
206}
207
208
209/*
210 *
211 * hv_get_ring_buffer()
212 *
213 * Get the start of the ring buffer
214 */
215static inline void *
216hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info)
217{
218 return (void *)ring_info->ring_buffer->buffer;
219}
220
221
222/*
223 *
224 * hv_get_ring_buffersize()
225 *
226 * Get the size of the ring buffer
227 */
228static inline u32
229hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
230{
231 return ring_info->ring_datasize;
232}
233
234/*
235 *
236 * hv_get_ring_bufferindices()
237 *
238 * Get the read and write indices as u64 of the specified ring buffer
239 *
240 */
241static inline u64
242hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
243{
244 return (u64)ring_info->ring_buffer->write_index << 32;
245}
246
247/*
248 *
249 * hv_copyfrom_ringbuffer()
250 *
251 * Helper routine to copy to source from ring buffer.
252 * Assume there is enough room. Handles wrap-around in src case only!!
253 *
254 */
255static u32 hv_copyfrom_ringbuffer(
256 struct hv_ring_buffer_info *ring_info,
257 void *dest,
258 u32 destlen,
259 u32 start_read_offset)
260{
261 void *ring_buffer = hv_get_ring_buffer(ring_info);
262 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
263
264 u32 frag_len;
265
266 /* wrap-around detected at the src */
267 if (destlen > ring_buffer_size - start_read_offset) {
268 frag_len = ring_buffer_size - start_read_offset;
269
270 memcpy(dest, ring_buffer + start_read_offset, frag_len);
271 memcpy(dest + frag_len, ring_buffer, destlen - frag_len);
272 } else
273
274 memcpy(dest, ring_buffer + start_read_offset, destlen);
275
276
277 start_read_offset += destlen;
278 start_read_offset %= ring_buffer_size;
279
280 return start_read_offset;
281}
282
283
284/*
285 *
286 * hv_copyto_ringbuffer()
287 *
288 * Helper routine to copy from source to ring buffer.
289 * Assume there is enough room. Handles wrap-around in dest case only!!
290 *
291 */
292static u32 hv_copyto_ringbuffer(
293 struct hv_ring_buffer_info *ring_info,
294 u32 start_write_offset,
295 void *src,
296 u32 srclen)
297{
298 void *ring_buffer = hv_get_ring_buffer(ring_info);
299 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
300 u32 frag_len;
301
302 /* wrap-around detected! */
303 if (srclen > ring_buffer_size - start_write_offset) {
304 frag_len = ring_buffer_size - start_write_offset;
305 memcpy(ring_buffer + start_write_offset, src, frag_len);
306 memcpy(ring_buffer, src + frag_len, srclen - frag_len);
307 } else
308 memcpy(ring_buffer + start_write_offset, src, srclen);
309
310 start_write_offset += srclen;
311 start_write_offset %= ring_buffer_size;
312
313 return start_write_offset;
314}
315
316/*
317 *
318 * hv_ringbuffer_get_debuginfo()
319 *
320 * Get various debug metrics for the specified ring buffer
321 *
322 */
323void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
324 struct hv_ring_buffer_debug_info *debug_info)
325{
326 u32 bytes_avail_towrite;
327 u32 bytes_avail_toread;
328
329 if (ring_info->ring_buffer) {
330 hv_get_ringbuffer_availbytes(ring_info,
331 &bytes_avail_toread,
332 &bytes_avail_towrite);
333
334 debug_info->bytes_avail_toread = bytes_avail_toread;
335 debug_info->bytes_avail_towrite = bytes_avail_towrite;
336 debug_info->current_read_index =
337 ring_info->ring_buffer->read_index;
338 debug_info->current_write_index =
339 ring_info->ring_buffer->write_index;
340 debug_info->current_interrupt_mask =
341 ring_info->ring_buffer->interrupt_mask;
342 }
343}
344
345/*
346 *
347 * hv_ringbuffer_init()
348 *
349 *Initialize the ring buffer
350 *
351 */
352int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
353 void *buffer, u32 buflen)
354{
355 if (sizeof(struct hv_ring_buffer) != PAGE_SIZE)
356 return -EINVAL;
357
358 memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
359
360 ring_info->ring_buffer = (struct hv_ring_buffer *)buffer;
361 ring_info->ring_buffer->read_index =
362 ring_info->ring_buffer->write_index = 0;
363
364 ring_info->ring_size = buflen;
365 ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer);
366
367 spin_lock_init(&ring_info->ring_lock);
368
369 return 0;
370}
371
372/*
373 *
374 * hv_ringbuffer_cleanup()
375 *
376 * Cleanup the ring buffer
377 *
378 */
379void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
380{
381}
382
383/*
384 *
385 * hv_ringbuffer_write()
386 *
387 * Write to the ring buffer
388 *
389 */
390int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info,
391 struct kvec *kv_list, u32 kv_count, bool *signal)
392{
393 int i = 0;
394 u32 bytes_avail_towrite;
395 u32 bytes_avail_toread;
396 u32 totalbytes_towrite = 0;
397
398 u32 next_write_location;
399 u32 old_write;
400 u64 prev_indices = 0;
401 unsigned long flags;
402
403 for (i = 0; i < kv_count; i++)
404 totalbytes_towrite += kv_list[i].iov_len;
405
406 totalbytes_towrite += sizeof(u64);
407
408 spin_lock_irqsave(&outring_info->ring_lock, flags);
409
410 hv_get_ringbuffer_availbytes(outring_info,
411 &bytes_avail_toread,
412 &bytes_avail_towrite);
413
414
415 /* If there is only room for the packet, assume it is full. */
416 /* Otherwise, the next time around, we think the ring buffer */
417 /* is empty since the read index == write index */
418 if (bytes_avail_towrite <= totalbytes_towrite) {
419 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
420 return -EAGAIN;
421 }
422
423 /* Write to the ring buffer */
424 next_write_location = hv_get_next_write_location(outring_info);
425
426 old_write = next_write_location;
427
428 for (i = 0; i < kv_count; i++) {
429 next_write_location = hv_copyto_ringbuffer(outring_info,
430 next_write_location,
431 kv_list[i].iov_base,
432 kv_list[i].iov_len);
433 }
434
435 /* Set previous packet start */
436 prev_indices = hv_get_ring_bufferindices(outring_info);
437
438 next_write_location = hv_copyto_ringbuffer(outring_info,
439 next_write_location,
440 &prev_indices,
441 sizeof(u64));
442
443 /* Issue a full memory barrier before updating the write index */
444 mb();
445
446 /* Now, update the write location */
447 hv_set_next_write_location(outring_info, next_write_location);
448
449
450 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
451
452 *signal = hv_need_to_signal(old_write, outring_info);
453 return 0;
454}
455
456
457/*
458 *
459 * hv_ringbuffer_peek()
460 *
461 * Read without advancing the read index
462 *
463 */
464int hv_ringbuffer_peek(struct hv_ring_buffer_info *Inring_info,
465 void *Buffer, u32 buflen)
466{
467 u32 bytes_avail_towrite;
468 u32 bytes_avail_toread;
469 u32 next_read_location = 0;
470 unsigned long flags;
471
472 spin_lock_irqsave(&Inring_info->ring_lock, flags);
473
474 hv_get_ringbuffer_availbytes(Inring_info,
475 &bytes_avail_toread,
476 &bytes_avail_towrite);
477
478 /* Make sure there is something to read */
479 if (bytes_avail_toread < buflen) {
480
481 spin_unlock_irqrestore(&Inring_info->ring_lock, flags);
482
483 return -EAGAIN;
484 }
485
486 /* Convert to byte offset */
487 next_read_location = hv_get_next_read_location(Inring_info);
488
489 next_read_location = hv_copyfrom_ringbuffer(Inring_info,
490 Buffer,
491 buflen,
492 next_read_location);
493
494 spin_unlock_irqrestore(&Inring_info->ring_lock, flags);
495
496 return 0;
497}
498
499
500/*
501 *
502 * hv_ringbuffer_read()
503 *
504 * Read and advance the read index
505 *
506 */
507int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info, void *buffer,
508 u32 buflen, u32 offset, bool *signal)
509{
510 u32 bytes_avail_towrite;
511 u32 bytes_avail_toread;
512 u32 next_read_location = 0;
513 u64 prev_indices = 0;
514 unsigned long flags;
515 u32 old_read;
516
517 if (buflen <= 0)
518 return -EINVAL;
519
520 spin_lock_irqsave(&inring_info->ring_lock, flags);
521
522 hv_get_ringbuffer_availbytes(inring_info,
523 &bytes_avail_toread,
524 &bytes_avail_towrite);
525
526 old_read = bytes_avail_toread;
527
528 /* Make sure there is something to read */
529 if (bytes_avail_toread < buflen) {
530 spin_unlock_irqrestore(&inring_info->ring_lock, flags);
531
532 return -EAGAIN;
533 }
534
535 next_read_location =
536 hv_get_next_readlocation_withoffset(inring_info, offset);
537
538 next_read_location = hv_copyfrom_ringbuffer(inring_info,
539 buffer,
540 buflen,
541 next_read_location);
542
543 next_read_location = hv_copyfrom_ringbuffer(inring_info,
544 &prev_indices,
545 sizeof(u64),
546 next_read_location);
547
548 /* Make sure all reads are done before we update the read index since */
549 /* the writer may start writing to the read area once the read index */
550 /*is updated */
551 mb();
552
553 /* Update the read index */
554 hv_set_next_read_location(inring_info, next_read_location);
555
556 spin_unlock_irqrestore(&inring_info->ring_lock, flags);
557
558 *signal = hv_need_to_signal_on_read(old_read, inring_info);
559
560 return 0;
561}
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 * K. Y. Srinivasan <kys@microsoft.com>
22 *
23 */
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/kernel.h>
27#include <linux/mm.h>
28#include <linux/hyperv.h>
29#include <linux/uio.h>
30#include <linux/vmalloc.h>
31#include <linux/slab.h>
32#include <linux/prefetch.h>
33
34#include "hyperv_vmbus.h"
35
36#define VMBUS_PKT_TRAILER 8
37
38/*
39 * When we write to the ring buffer, check if the host needs to
40 * be signaled. Here is the details of this protocol:
41 *
42 * 1. The host guarantees that while it is draining the
43 * ring buffer, it will set the interrupt_mask to
44 * indicate it does not need to be interrupted when
45 * new data is placed.
46 *
47 * 2. The host guarantees that it will completely drain
48 * the ring buffer before exiting the read loop. Further,
49 * once the ring buffer is empty, it will clear the
50 * interrupt_mask and re-check to see if new data has
51 * arrived.
52 *
53 * KYS: Oct. 30, 2016:
54 * It looks like Windows hosts have logic to deal with DOS attacks that
55 * can be triggered if it receives interrupts when it is not expecting
56 * the interrupt. The host expects interrupts only when the ring
57 * transitions from empty to non-empty (or full to non full on the guest
58 * to host ring).
59 * So, base the signaling decision solely on the ring state until the
60 * host logic is fixed.
61 */
62
63static void hv_signal_on_write(u32 old_write, struct vmbus_channel *channel)
64{
65 struct hv_ring_buffer_info *rbi = &channel->outbound;
66
67 virt_mb();
68 if (READ_ONCE(rbi->ring_buffer->interrupt_mask))
69 return;
70
71 /* check interrupt_mask before read_index */
72 virt_rmb();
73 /*
74 * This is the only case we need to signal when the
75 * ring transitions from being empty to non-empty.
76 */
77 if (old_write == READ_ONCE(rbi->ring_buffer->read_index))
78 vmbus_setevent(channel);
79}
80
81/* Get the next write location for the specified ring buffer. */
82static inline u32
83hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
84{
85 u32 next = ring_info->ring_buffer->write_index;
86
87 return next;
88}
89
90/* Set the next write location for the specified ring buffer. */
91static inline void
92hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
93 u32 next_write_location)
94{
95 ring_info->ring_buffer->write_index = next_write_location;
96}
97
98/* Set the next read location for the specified ring buffer. */
99static inline void
100hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
101 u32 next_read_location)
102{
103 ring_info->ring_buffer->read_index = next_read_location;
104 ring_info->priv_read_index = next_read_location;
105}
106
107/* Get the size of the ring buffer. */
108static inline u32
109hv_get_ring_buffersize(const struct hv_ring_buffer_info *ring_info)
110{
111 return ring_info->ring_datasize;
112}
113
114/* Get the read and write indices as u64 of the specified ring buffer. */
115static inline u64
116hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
117{
118 return (u64)ring_info->ring_buffer->write_index << 32;
119}
120
121/*
122 * Helper routine to copy from source to ring buffer.
123 * Assume there is enough room. Handles wrap-around in dest case only!!
124 */
125static u32 hv_copyto_ringbuffer(
126 struct hv_ring_buffer_info *ring_info,
127 u32 start_write_offset,
128 const void *src,
129 u32 srclen)
130{
131 void *ring_buffer = hv_get_ring_buffer(ring_info);
132 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
133
134 memcpy(ring_buffer + start_write_offset, src, srclen);
135
136 start_write_offset += srclen;
137 if (start_write_offset >= ring_buffer_size)
138 start_write_offset -= ring_buffer_size;
139
140 return start_write_offset;
141}
142
143/*
144 *
145 * hv_get_ringbuffer_availbytes()
146 *
147 * Get number of bytes available to read and to write to
148 * for the specified ring buffer
149 */
150static void
151hv_get_ringbuffer_availbytes(const struct hv_ring_buffer_info *rbi,
152 u32 *read, u32 *write)
153{
154 u32 read_loc, write_loc, dsize;
155
156 /* Capture the read/write indices before they changed */
157 read_loc = READ_ONCE(rbi->ring_buffer->read_index);
158 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
159 dsize = rbi->ring_datasize;
160
161 *write = write_loc >= read_loc ? dsize - (write_loc - read_loc) :
162 read_loc - write_loc;
163 *read = dsize - *write;
164}
165
166/* Get various debug metrics for the specified ring buffer. */
167void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info,
168 struct hv_ring_buffer_debug_info *debug_info)
169{
170 u32 bytes_avail_towrite;
171 u32 bytes_avail_toread;
172
173 if (ring_info->ring_buffer) {
174 hv_get_ringbuffer_availbytes(ring_info,
175 &bytes_avail_toread,
176 &bytes_avail_towrite);
177
178 debug_info->bytes_avail_toread = bytes_avail_toread;
179 debug_info->bytes_avail_towrite = bytes_avail_towrite;
180 debug_info->current_read_index =
181 ring_info->ring_buffer->read_index;
182 debug_info->current_write_index =
183 ring_info->ring_buffer->write_index;
184 debug_info->current_interrupt_mask =
185 ring_info->ring_buffer->interrupt_mask;
186 }
187}
188EXPORT_SYMBOL_GPL(hv_ringbuffer_get_debuginfo);
189
190/* Initialize the ring buffer. */
191int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
192 struct page *pages, u32 page_cnt)
193{
194 int i;
195 struct page **pages_wraparound;
196
197 BUILD_BUG_ON((sizeof(struct hv_ring_buffer) != PAGE_SIZE));
198
199 memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
200
201 /*
202 * First page holds struct hv_ring_buffer, do wraparound mapping for
203 * the rest.
204 */
205 pages_wraparound = kzalloc(sizeof(struct page *) * (page_cnt * 2 - 1),
206 GFP_KERNEL);
207 if (!pages_wraparound)
208 return -ENOMEM;
209
210 pages_wraparound[0] = pages;
211 for (i = 0; i < 2 * (page_cnt - 1); i++)
212 pages_wraparound[i + 1] = &pages[i % (page_cnt - 1) + 1];
213
214 ring_info->ring_buffer = (struct hv_ring_buffer *)
215 vmap(pages_wraparound, page_cnt * 2 - 1, VM_MAP, PAGE_KERNEL);
216
217 kfree(pages_wraparound);
218
219
220 if (!ring_info->ring_buffer)
221 return -ENOMEM;
222
223 ring_info->ring_buffer->read_index =
224 ring_info->ring_buffer->write_index = 0;
225
226 /* Set the feature bit for enabling flow control. */
227 ring_info->ring_buffer->feature_bits.value = 1;
228
229 ring_info->ring_size = page_cnt << PAGE_SHIFT;
230 ring_info->ring_datasize = ring_info->ring_size -
231 sizeof(struct hv_ring_buffer);
232
233 spin_lock_init(&ring_info->ring_lock);
234
235 return 0;
236}
237
238/* Cleanup the ring buffer. */
239void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
240{
241 vunmap(ring_info->ring_buffer);
242}
243
244/* Write to the ring buffer. */
245int hv_ringbuffer_write(struct vmbus_channel *channel,
246 const struct kvec *kv_list, u32 kv_count)
247{
248 int i;
249 u32 bytes_avail_towrite;
250 u32 totalbytes_towrite = sizeof(u64);
251 u32 next_write_location;
252 u32 old_write;
253 u64 prev_indices;
254 unsigned long flags;
255 struct hv_ring_buffer_info *outring_info = &channel->outbound;
256
257 if (channel->rescind)
258 return -ENODEV;
259
260 for (i = 0; i < kv_count; i++)
261 totalbytes_towrite += kv_list[i].iov_len;
262
263 spin_lock_irqsave(&outring_info->ring_lock, flags);
264
265 bytes_avail_towrite = hv_get_bytes_to_write(outring_info);
266
267 /*
268 * If there is only room for the packet, assume it is full.
269 * Otherwise, the next time around, we think the ring buffer
270 * is empty since the read index == write index.
271 */
272 if (bytes_avail_towrite <= totalbytes_towrite) {
273 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
274 return -EAGAIN;
275 }
276
277 /* Write to the ring buffer */
278 next_write_location = hv_get_next_write_location(outring_info);
279
280 old_write = next_write_location;
281
282 for (i = 0; i < kv_count; i++) {
283 next_write_location = hv_copyto_ringbuffer(outring_info,
284 next_write_location,
285 kv_list[i].iov_base,
286 kv_list[i].iov_len);
287 }
288
289 /* Set previous packet start */
290 prev_indices = hv_get_ring_bufferindices(outring_info);
291
292 next_write_location = hv_copyto_ringbuffer(outring_info,
293 next_write_location,
294 &prev_indices,
295 sizeof(u64));
296
297 /* Issue a full memory barrier before updating the write index */
298 virt_mb();
299
300 /* Now, update the write location */
301 hv_set_next_write_location(outring_info, next_write_location);
302
303
304 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
305
306 hv_signal_on_write(old_write, channel);
307
308 if (channel->rescind)
309 return -ENODEV;
310
311 return 0;
312}
313
314int hv_ringbuffer_read(struct vmbus_channel *channel,
315 void *buffer, u32 buflen, u32 *buffer_actual_len,
316 u64 *requestid, bool raw)
317{
318 struct vmpacket_descriptor *desc;
319 u32 packetlen, offset;
320
321 if (unlikely(buflen == 0))
322 return -EINVAL;
323
324 *buffer_actual_len = 0;
325 *requestid = 0;
326
327 /* Make sure there is something to read */
328 desc = hv_pkt_iter_first(channel);
329 if (desc == NULL) {
330 /*
331 * No error is set when there is even no header, drivers are
332 * supposed to analyze buffer_actual_len.
333 */
334 return 0;
335 }
336
337 offset = raw ? 0 : (desc->offset8 << 3);
338 packetlen = (desc->len8 << 3) - offset;
339 *buffer_actual_len = packetlen;
340 *requestid = desc->trans_id;
341
342 if (unlikely(packetlen > buflen))
343 return -ENOBUFS;
344
345 /* since ring is double mapped, only one copy is necessary */
346 memcpy(buffer, (const char *)desc + offset, packetlen);
347
348 /* Advance ring index to next packet descriptor */
349 __hv_pkt_iter_next(channel, desc);
350
351 /* Notify host of update */
352 hv_pkt_iter_close(channel);
353
354 return 0;
355}
356
357/*
358 * Determine number of bytes available in ring buffer after
359 * the current iterator (priv_read_index) location.
360 *
361 * This is similar to hv_get_bytes_to_read but with private
362 * read index instead.
363 */
364static u32 hv_pkt_iter_avail(const struct hv_ring_buffer_info *rbi)
365{
366 u32 priv_read_loc = rbi->priv_read_index;
367 u32 write_loc = READ_ONCE(rbi->ring_buffer->write_index);
368
369 if (write_loc >= priv_read_loc)
370 return write_loc - priv_read_loc;
371 else
372 return (rbi->ring_datasize - priv_read_loc) + write_loc;
373}
374
375/*
376 * Get first vmbus packet from ring buffer after read_index
377 *
378 * If ring buffer is empty, returns NULL and no other action needed.
379 */
380struct vmpacket_descriptor *hv_pkt_iter_first(struct vmbus_channel *channel)
381{
382 struct hv_ring_buffer_info *rbi = &channel->inbound;
383 struct vmpacket_descriptor *desc;
384
385 if (hv_pkt_iter_avail(rbi) < sizeof(struct vmpacket_descriptor))
386 return NULL;
387
388 desc = hv_get_ring_buffer(rbi) + rbi->priv_read_index;
389 if (desc)
390 prefetch((char *)desc + (desc->len8 << 3));
391
392 return desc;
393}
394EXPORT_SYMBOL_GPL(hv_pkt_iter_first);
395
396/*
397 * Get next vmbus packet from ring buffer.
398 *
399 * Advances the current location (priv_read_index) and checks for more
400 * data. If the end of the ring buffer is reached, then return NULL.
401 */
402struct vmpacket_descriptor *
403__hv_pkt_iter_next(struct vmbus_channel *channel,
404 const struct vmpacket_descriptor *desc)
405{
406 struct hv_ring_buffer_info *rbi = &channel->inbound;
407 u32 packetlen = desc->len8 << 3;
408 u32 dsize = rbi->ring_datasize;
409
410 /* bump offset to next potential packet */
411 rbi->priv_read_index += packetlen + VMBUS_PKT_TRAILER;
412 if (rbi->priv_read_index >= dsize)
413 rbi->priv_read_index -= dsize;
414
415 /* more data? */
416 return hv_pkt_iter_first(channel);
417}
418EXPORT_SYMBOL_GPL(__hv_pkt_iter_next);
419
420/* How many bytes were read in this iterator cycle */
421static u32 hv_pkt_iter_bytes_read(const struct hv_ring_buffer_info *rbi,
422 u32 start_read_index)
423{
424 if (rbi->priv_read_index >= start_read_index)
425 return rbi->priv_read_index - start_read_index;
426 else
427 return rbi->ring_datasize - start_read_index +
428 rbi->priv_read_index;
429}
430
431/*
432 * Update host ring buffer after iterating over packets.
433 */
434void hv_pkt_iter_close(struct vmbus_channel *channel)
435{
436 struct hv_ring_buffer_info *rbi = &channel->inbound;
437 u32 curr_write_sz, pending_sz, bytes_read, start_read_index;
438
439 /*
440 * Make sure all reads are done before we update the read index since
441 * the writer may start writing to the read area once the read index
442 * is updated.
443 */
444 virt_rmb();
445 start_read_index = rbi->ring_buffer->read_index;
446 rbi->ring_buffer->read_index = rbi->priv_read_index;
447
448 if (!rbi->ring_buffer->feature_bits.feat_pending_send_sz)
449 return;
450
451 /*
452 * Issue a full memory barrier before making the signaling decision.
453 * Here is the reason for having this barrier:
454 * If the reading of the pend_sz (in this function)
455 * were to be reordered and read before we commit the new read
456 * index (in the calling function) we could
457 * have a problem. If the host were to set the pending_sz after we
458 * have sampled pending_sz and go to sleep before we commit the
459 * read index, we could miss sending the interrupt. Issue a full
460 * memory barrier to address this.
461 */
462 virt_mb();
463
464 pending_sz = READ_ONCE(rbi->ring_buffer->pending_send_sz);
465 if (!pending_sz)
466 return;
467
468 /*
469 * Ensure the read of write_index in hv_get_bytes_to_write()
470 * happens after the read of pending_send_sz.
471 */
472 virt_rmb();
473 curr_write_sz = hv_get_bytes_to_write(rbi);
474 bytes_read = hv_pkt_iter_bytes_read(rbi, start_read_index);
475
476 /*
477 * If there was space before we began iteration,
478 * then host was not blocked.
479 */
480
481 if (curr_write_sz - bytes_read > pending_sz)
482 return;
483
484 /* If pending write will not fit, don't give false hope. */
485 if (curr_write_sz <= pending_sz)
486 return;
487
488 vmbus_setevent(channel);
489}
490EXPORT_SYMBOL_GPL(hv_pkt_iter_close);