Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux
[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
30 #include "hyperv_vmbus.h"
31
32 void hv_begin_read(struct hv_ring_buffer_info *rbi)
33 {
34 rbi->ring_buffer->interrupt_mask = 1;
35 smp_mb();
36 }
37
38 u32 hv_end_read(struct hv_ring_buffer_info *rbi)
39 {
40 u32 read;
41 u32 write;
42
43 rbi->ring_buffer->interrupt_mask = 0;
44 smp_mb();
45
46 /*
47 * Now check to see if the ring buffer is still empty.
48 * If it is not, we raced and we need to process new
49 * incoming messages.
50 */
51 hv_get_ringbuffer_availbytes(rbi, &read, &write);
52
53 return read;
54 }
55
56 /*
57 * When we write to the ring buffer, check if the host needs to
58 * be signaled. Here is the details of this protocol:
59 *
60 * 1. The host guarantees that while it is draining the
61 * ring buffer, it will set the interrupt_mask to
62 * indicate it does not need to be interrupted when
63 * new data is placed.
64 *
65 * 2. The host guarantees that it will completely drain
66 * the ring buffer before exiting the read loop. Further,
67 * once the ring buffer is empty, it will clear the
68 * interrupt_mask and re-check to see if new data has
69 * arrived.
70 */
71
72 static bool hv_need_to_signal(u32 old_write, struct hv_ring_buffer_info *rbi)
73 {
74 if (rbi->ring_buffer->interrupt_mask)
75 return false;
76
77 /*
78 * This is the only case we need to signal when the
79 * ring transitions from being empty to non-empty.
80 */
81 if (old_write == rbi->ring_buffer->read_index)
82 return true;
83
84 return false;
85 }
86
87 /*
88 * To optimize the flow management on the send-side,
89 * when the sender is blocked because of lack of
90 * sufficient space in the ring buffer, potential the
91 * consumer of the ring buffer can signal the producer.
92 * This is controlled by the following parameters:
93 *
94 * 1. pending_send_sz: This is the size in bytes that the
95 * producer is trying to send.
96 * 2. The feature bit feat_pending_send_sz set to indicate if
97 * the consumer of the ring will signal when the ring
98 * state transitions from being full to a state where
99 * there is room for the producer to send the pending packet.
100 */
101
102 static bool hv_need_to_signal_on_read(u32 old_rd,
103 struct hv_ring_buffer_info *rbi)
104 {
105 u32 prev_write_sz;
106 u32 cur_write_sz;
107 u32 r_size;
108 u32 write_loc = rbi->ring_buffer->write_index;
109 u32 read_loc = rbi->ring_buffer->read_index;
110 u32 pending_sz = rbi->ring_buffer->pending_send_sz;
111
112 /*
113 * If the other end is not blocked on write don't bother.
114 */
115 if (pending_sz == 0)
116 return false;
117
118 r_size = rbi->ring_datasize;
119 cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) :
120 read_loc - write_loc;
121
122 prev_write_sz = write_loc >= old_rd ? r_size - (write_loc - old_rd) :
123 old_rd - write_loc;
124
125
126 if ((prev_write_sz < pending_sz) && (cur_write_sz >= pending_sz))
127 return true;
128
129 return false;
130 }
131
132 /*
133 * hv_get_next_write_location()
134 *
135 * Get the next write location for the specified ring buffer
136 *
137 */
138 static inline u32
139 hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
140 {
141 u32 next = ring_info->ring_buffer->write_index;
142
143 return next;
144 }
145
146 /*
147 * hv_set_next_write_location()
148 *
149 * Set the next write location for the specified ring buffer
150 *
151 */
152 static inline void
153 hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
154 u32 next_write_location)
155 {
156 ring_info->ring_buffer->write_index = next_write_location;
157 }
158
159 /*
160 * hv_get_next_read_location()
161 *
162 * Get the next read location for the specified ring buffer
163 */
164 static inline u32
165 hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
166 {
167 u32 next = ring_info->ring_buffer->read_index;
168
169 return next;
170 }
171
172 /*
173 * hv_get_next_readlocation_withoffset()
174 *
175 * Get the next read location + offset for the specified ring buffer.
176 * This allows the caller to skip
177 */
178 static inline u32
179 hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
180 u32 offset)
181 {
182 u32 next = ring_info->ring_buffer->read_index;
183
184 next += offset;
185 next %= ring_info->ring_datasize;
186
187 return next;
188 }
189
190 /*
191 *
192 * hv_set_next_read_location()
193 *
194 * Set the next read location for the specified ring buffer
195 *
196 */
197 static inline void
198 hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
199 u32 next_read_location)
200 {
201 ring_info->ring_buffer->read_index = next_read_location;
202 }
203
204
205 /*
206 *
207 * hv_get_ring_buffer()
208 *
209 * Get the start of the ring buffer
210 */
211 static inline void *
212 hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info)
213 {
214 return (void *)ring_info->ring_buffer->buffer;
215 }
216
217
218 /*
219 *
220 * hv_get_ring_buffersize()
221 *
222 * Get the size of the ring buffer
223 */
224 static inline u32
225 hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
226 {
227 return ring_info->ring_datasize;
228 }
229
230 /*
231 *
232 * hv_get_ring_bufferindices()
233 *
234 * Get the read and write indices as u64 of the specified ring buffer
235 *
236 */
237 static inline u64
238 hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
239 {
240 return (u64)ring_info->ring_buffer->write_index << 32;
241 }
242
243 /*
244 *
245 * hv_copyfrom_ringbuffer()
246 *
247 * Helper routine to copy to source from ring buffer.
248 * Assume there is enough room. Handles wrap-around in src case only!!
249 *
250 */
251 static u32 hv_copyfrom_ringbuffer(
252 struct hv_ring_buffer_info *ring_info,
253 void *dest,
254 u32 destlen,
255 u32 start_read_offset)
256 {
257 void *ring_buffer = hv_get_ring_buffer(ring_info);
258 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
259
260 u32 frag_len;
261
262 /* wrap-around detected at the src */
263 if (destlen > ring_buffer_size - start_read_offset) {
264 frag_len = ring_buffer_size - start_read_offset;
265
266 memcpy(dest, ring_buffer + start_read_offset, frag_len);
267 memcpy(dest + frag_len, ring_buffer, destlen - frag_len);
268 } else
269
270 memcpy(dest, ring_buffer + start_read_offset, destlen);
271
272
273 start_read_offset += destlen;
274 start_read_offset %= ring_buffer_size;
275
276 return start_read_offset;
277 }
278
279
280 /*
281 *
282 * hv_copyto_ringbuffer()
283 *
284 * Helper routine to copy from source to ring buffer.
285 * Assume there is enough room. Handles wrap-around in dest case only!!
286 *
287 */
288 static u32 hv_copyto_ringbuffer(
289 struct hv_ring_buffer_info *ring_info,
290 u32 start_write_offset,
291 void *src,
292 u32 srclen)
293 {
294 void *ring_buffer = hv_get_ring_buffer(ring_info);
295 u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
296 u32 frag_len;
297
298 /* wrap-around detected! */
299 if (srclen > ring_buffer_size - start_write_offset) {
300 frag_len = ring_buffer_size - start_write_offset;
301 memcpy(ring_buffer + start_write_offset, src, frag_len);
302 memcpy(ring_buffer, src + frag_len, srclen - frag_len);
303 } else
304 memcpy(ring_buffer + start_write_offset, src, srclen);
305
306 start_write_offset += srclen;
307 start_write_offset %= ring_buffer_size;
308
309 return start_write_offset;
310 }
311
312 /*
313 *
314 * hv_ringbuffer_get_debuginfo()
315 *
316 * Get various debug metrics for the specified ring buffer
317 *
318 */
319 void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
320 struct hv_ring_buffer_debug_info *debug_info)
321 {
322 u32 bytes_avail_towrite;
323 u32 bytes_avail_toread;
324
325 if (ring_info->ring_buffer) {
326 hv_get_ringbuffer_availbytes(ring_info,
327 &bytes_avail_toread,
328 &bytes_avail_towrite);
329
330 debug_info->bytes_avail_toread = bytes_avail_toread;
331 debug_info->bytes_avail_towrite = bytes_avail_towrite;
332 debug_info->current_read_index =
333 ring_info->ring_buffer->read_index;
334 debug_info->current_write_index =
335 ring_info->ring_buffer->write_index;
336 debug_info->current_interrupt_mask =
337 ring_info->ring_buffer->interrupt_mask;
338 }
339 }
340
341 /*
342 *
343 * hv_ringbuffer_init()
344 *
345 *Initialize the ring buffer
346 *
347 */
348 int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
349 void *buffer, u32 buflen)
350 {
351 if (sizeof(struct hv_ring_buffer) != PAGE_SIZE)
352 return -EINVAL;
353
354 memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
355
356 ring_info->ring_buffer = (struct hv_ring_buffer *)buffer;
357 ring_info->ring_buffer->read_index =
358 ring_info->ring_buffer->write_index = 0;
359
360 ring_info->ring_size = buflen;
361 ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer);
362
363 spin_lock_init(&ring_info->ring_lock);
364
365 return 0;
366 }
367
368 /*
369 *
370 * hv_ringbuffer_cleanup()
371 *
372 * Cleanup the ring buffer
373 *
374 */
375 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
376 {
377 }
378
379 /*
380 *
381 * hv_ringbuffer_write()
382 *
383 * Write to the ring buffer
384 *
385 */
386 int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info,
387 struct scatterlist *sglist, u32 sgcount, bool *signal)
388 {
389 int i = 0;
390 u32 bytes_avail_towrite;
391 u32 bytes_avail_toread;
392 u32 totalbytes_towrite = 0;
393
394 struct scatterlist *sg;
395 u32 next_write_location;
396 u32 old_write;
397 u64 prev_indices = 0;
398 unsigned long flags;
399
400 for_each_sg(sglist, sg, sgcount, i)
401 {
402 totalbytes_towrite += sg->length;
403 }
404
405 totalbytes_towrite += sizeof(u64);
406
407 spin_lock_irqsave(&outring_info->ring_lock, flags);
408
409 hv_get_ringbuffer_availbytes(outring_info,
410 &bytes_avail_toread,
411 &bytes_avail_towrite);
412
413
414 /* If there is only room for the packet, assume it is full. */
415 /* Otherwise, the next time around, we think the ring buffer */
416 /* is empty since the read index == write index */
417 if (bytes_avail_towrite <= totalbytes_towrite) {
418 spin_unlock_irqrestore(&outring_info->ring_lock, flags);
419 return -EAGAIN;
420 }
421
422 /* Write to the ring buffer */
423 next_write_location = hv_get_next_write_location(outring_info);
424
425 old_write = next_write_location;
426
427 for_each_sg(sglist, sg, sgcount, i)
428 {
429 next_write_location = hv_copyto_ringbuffer(outring_info,
430 next_write_location,
431 sg_virt(sg),
432 sg->length);
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 smp_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 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 smp_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 }
This page took 0.042792 seconds and 5 git commands to generate.