Merge tag 'iommu-updates-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/joro...
[deliverable/linux.git] / drivers / hv / ring_buffer.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 * 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
33 void hv_begin_read(struct hv_ring_buffer_info *rbi)
34 {
35 rbi->ring_buffer->interrupt_mask = 1;
36 mb();
37 }
38
39 u32 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
73 static 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
106 static bool hv_need_to_signal_on_read(u32 prev_write_sz,
107 struct hv_ring_buffer_info *rbi)
108 {
109 u32 cur_write_sz;
110 u32 r_size;
111 u32 write_loc = rbi->ring_buffer->write_index;
112 u32 read_loc = rbi->ring_buffer->read_index;
113 u32 pending_sz = rbi->ring_buffer->pending_send_sz;
114
115 /*
116 * If the other end is not blocked on write don't bother.
117 */
118 if (pending_sz == 0)
119 return false;
120
121 r_size = rbi->ring_datasize;
122 cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) :
123 read_loc - write_loc;
124
125 if ((prev_write_sz < pending_sz) && (cur_write_sz >= pending_sz))
126 return true;
127
128 return false;
129 }
130
131 /*
132 * hv_get_next_write_location()
133 *
134 * Get the next write location for the specified ring buffer
135 *
136 */
137 static inline u32
138 hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
139 {
140 u32 next = ring_info->ring_buffer->write_index;
141
142 return next;
143 }
144
145 /*
146 * hv_set_next_write_location()
147 *
148 * Set the next write location for the specified ring buffer
149 *
150 */
151 static inline void
152 hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
153 u32 next_write_location)
154 {
155 ring_info->ring_buffer->write_index = next_write_location;
156 }
157
158 /*
159 * hv_get_next_read_location()
160 *
161 * Get the next read location for the specified ring buffer
162 */
163 static inline u32
164 hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
165 {
166 u32 next = ring_info->ring_buffer->read_index;
167
168 return next;
169 }
170
171 /*
172 * hv_get_next_readlocation_withoffset()
173 *
174 * Get the next read location + offset for the specified ring buffer.
175 * This allows the caller to skip
176 */
177 static inline u32
178 hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
179 u32 offset)
180 {
181 u32 next = ring_info->ring_buffer->read_index;
182
183 next += offset;
184 next %= ring_info->ring_datasize;
185
186 return next;
187 }
188
189 /*
190 *
191 * hv_set_next_read_location()
192 *
193 * Set the next read location for the specified ring buffer
194 *
195 */
196 static inline void
197 hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
198 u32 next_read_location)
199 {
200 ring_info->ring_buffer->read_index = next_read_location;
201 }
202
203
204 /*
205 *
206 * hv_get_ring_buffer()
207 *
208 * Get the start of the ring buffer
209 */
210 static inline void *
211 hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info)
212 {
213 return (void *)ring_info->ring_buffer->buffer;
214 }
215
216
217 /*
218 *
219 * hv_get_ring_buffersize()
220 *
221 * Get the size of the ring buffer
222 */
223 static inline u32
224 hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
225 {
226 return ring_info->ring_datasize;
227 }
228
229 /*
230 *
231 * hv_get_ring_bufferindices()
232 *
233 * Get the read and write indices as u64 of the specified ring buffer
234 *
235 */
236 static inline u64
237 hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
238 {
239 return (u64)ring_info->ring_buffer->write_index << 32;
240 }
241
242 /*
243 *
244 * hv_copyfrom_ringbuffer()
245 *
246 * Helper routine to copy to source from ring buffer.
247 * Assume there is enough room. Handles wrap-around in src case only!!
248 *
249 */
250 static u32 hv_copyfrom_ringbuffer(
251 struct hv_ring_buffer_info *ring_info,
252 void *dest,
253 u32 destlen,
254 u32 start_read_offset)
255 {
256 void *ring_buffer = hv_get_ring_buffer(ring_info);
257 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
258
259 u32 frag_len;
260
261 /* wrap-around detected at the src */
262 if (destlen > ring_buffer_size - start_read_offset) {
263 frag_len = ring_buffer_size - start_read_offset;
264
265 memcpy(dest, ring_buffer + start_read_offset, frag_len);
266 memcpy(dest + frag_len, ring_buffer, destlen - frag_len);
267 } else
268
269 memcpy(dest, ring_buffer + start_read_offset, destlen);
270
271
272 start_read_offset += destlen;
273 start_read_offset %= ring_buffer_size;
274
275 return start_read_offset;
276 }
277
278
279 /*
280 *
281 * hv_copyto_ringbuffer()
282 *
283 * Helper routine to copy from source to ring buffer.
284 * Assume there is enough room. Handles wrap-around in dest case only!!
285 *
286 */
287 static u32 hv_copyto_ringbuffer(
288 struct hv_ring_buffer_info *ring_info,
289 u32 start_write_offset,
290 void *src,
291 u32 srclen)
292 {
293 void *ring_buffer = hv_get_ring_buffer(ring_info);
294 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
295 u32 frag_len;
296
297 /* wrap-around detected! */
298 if (srclen > ring_buffer_size - start_write_offset) {
299 frag_len = ring_buffer_size - start_write_offset;
300 memcpy(ring_buffer + start_write_offset, src, frag_len);
301 memcpy(ring_buffer, src + frag_len, srclen - frag_len);
302 } else
303 memcpy(ring_buffer + start_write_offset, src, srclen);
304
305 start_write_offset += srclen;
306 start_write_offset %= ring_buffer_size;
307
308 return start_write_offset;
309 }
310
311 /*
312 *
313 * hv_ringbuffer_get_debuginfo()
314 *
315 * Get various debug metrics for the specified ring buffer
316 *
317 */
318 void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
319 struct hv_ring_buffer_debug_info *debug_info)
320 {
321 u32 bytes_avail_towrite;
322 u32 bytes_avail_toread;
323
324 if (ring_info->ring_buffer) {
325 hv_get_ringbuffer_availbytes(ring_info,
326 &bytes_avail_toread,
327 &bytes_avail_towrite);
328
329 debug_info->bytes_avail_toread = bytes_avail_toread;
330 debug_info->bytes_avail_towrite = bytes_avail_towrite;
331 debug_info->current_read_index =
332 ring_info->ring_buffer->read_index;
333 debug_info->current_write_index =
334 ring_info->ring_buffer->write_index;
335 debug_info->current_interrupt_mask =
336 ring_info->ring_buffer->interrupt_mask;
337 }
338 }
339
340 /*
341 *
342 * hv_ringbuffer_init()
343 *
344 *Initialize the ring buffer
345 *
346 */
347 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
348 void *buffer, u32 buflen)
349 {
350 if (sizeof(struct hv_ring_buffer) != PAGE_SIZE)
351 return -EINVAL;
352
353 memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
354
355 ring_info->ring_buffer = (struct hv_ring_buffer *)buffer;
356 ring_info->ring_buffer->read_index =
357 ring_info->ring_buffer->write_index = 0;
358
359 /*
360 * Set the feature bit for enabling flow control.
361 */
362 ring_info->ring_buffer->feature_bits.value = 1;
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 */
379 void 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 */
390 int 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 */
464 int 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 */
507 int 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
516 if (buflen <= 0)
517 return -EINVAL;
518
519 spin_lock_irqsave(&inring_info->ring_lock, flags);
520
521 hv_get_ringbuffer_availbytes(inring_info,
522 &bytes_avail_toread,
523 &bytes_avail_towrite);
524
525 /* Make sure there is something to read */
526 if (bytes_avail_toread < buflen) {
527 spin_unlock_irqrestore(&inring_info->ring_lock, flags);
528
529 return -EAGAIN;
530 }
531
532 next_read_location =
533 hv_get_next_readlocation_withoffset(inring_info, offset);
534
535 next_read_location = hv_copyfrom_ringbuffer(inring_info,
536 buffer,
537 buflen,
538 next_read_location);
539
540 next_read_location = hv_copyfrom_ringbuffer(inring_info,
541 &prev_indices,
542 sizeof(u64),
543 next_read_location);
544
545 /* Make sure all reads are done before we update the read index since */
546 /* the writer may start writing to the read area once the read index */
547 /*is updated */
548 mb();
549
550 /* Update the read index */
551 hv_set_next_read_location(inring_info, next_read_location);
552
553 spin_unlock_irqrestore(&inring_info->ring_lock, flags);
554
555 *signal = hv_need_to_signal_on_read(bytes_avail_towrite, inring_info);
556
557 return 0;
558 }
This page took 0.041761 seconds and 5 git commands to generate.