]>
Commit | Line | Data |
---|---|---|
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 | 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 | 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 | mb(); | |
75 | if (rbi->ring_buffer->interrupt_mask) | |
76 | return false; | |
77 | ||
78 | /* | |
79 | * This is the only case we need to signal when the | |
80 | * ring transitions from being empty to non-empty. | |
81 | */ | |
82 | if (old_write == rbi->ring_buffer->read_index) | |
83 | return true; | |
84 | ||
85 | return false; | |
86 | } | |
87 | ||
88 | /* | |
89 | * To optimize the flow management on the send-side, | |
90 | * when the sender is blocked because of lack of | |
91 | * sufficient space in the ring buffer, potential the | |
92 | * consumer of the ring buffer can signal the producer. | |
93 | * This is controlled by the following parameters: | |
94 | * | |
95 | * 1. pending_send_sz: This is the size in bytes that the | |
96 | * producer is trying to send. | |
97 | * 2. The feature bit feat_pending_send_sz set to indicate if | |
98 | * the consumer of the ring will signal when the ring | |
99 | * state transitions from being full to a state where | |
100 | * there is room for the producer to send the pending packet. | |
101 | */ | |
102 | ||
103 | static bool hv_need_to_signal_on_read(u32 old_rd, | |
104 | struct hv_ring_buffer_info *rbi) | |
105 | { | |
106 | u32 prev_write_sz; | |
107 | u32 cur_write_sz; | |
108 | u32 r_size; | |
109 | u32 write_loc = rbi->ring_buffer->write_index; | |
110 | u32 read_loc = rbi->ring_buffer->read_index; | |
111 | u32 pending_sz = rbi->ring_buffer->pending_send_sz; | |
112 | ||
113 | /* | |
114 | * If the other end is not blocked on write don't bother. | |
115 | */ | |
116 | if (pending_sz == 0) | |
117 | return false; | |
118 | ||
119 | r_size = rbi->ring_datasize; | |
120 | cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) : | |
121 | read_loc - write_loc; | |
122 | ||
123 | prev_write_sz = write_loc >= old_rd ? r_size - (write_loc - old_rd) : | |
124 | old_rd - write_loc; | |
125 | ||
126 | ||
127 | if ((prev_write_sz < pending_sz) && (cur_write_sz >= pending_sz)) | |
128 | return true; | |
129 | ||
130 | return false; | |
131 | } | |
132 | ||
133 | /* | |
134 | * hv_get_next_write_location() | |
135 | * | |
136 | * Get the next write location for the specified ring buffer | |
137 | * | |
138 | */ | |
139 | static inline u32 | |
140 | hv_get_next_write_location(struct hv_ring_buffer_info *ring_info) | |
141 | { | |
142 | u32 next = ring_info->ring_buffer->write_index; | |
143 | ||
144 | return next; | |
145 | } | |
146 | ||
147 | /* | |
148 | * hv_set_next_write_location() | |
149 | * | |
150 | * Set the next write location for the specified ring buffer | |
151 | * | |
152 | */ | |
153 | static inline void | |
154 | hv_set_next_write_location(struct hv_ring_buffer_info *ring_info, | |
155 | u32 next_write_location) | |
156 | { | |
157 | ring_info->ring_buffer->write_index = next_write_location; | |
158 | } | |
159 | ||
160 | /* | |
161 | * hv_get_next_read_location() | |
162 | * | |
163 | * Get the next read location for the specified ring buffer | |
164 | */ | |
165 | static inline u32 | |
166 | hv_get_next_read_location(struct hv_ring_buffer_info *ring_info) | |
167 | { | |
168 | u32 next = ring_info->ring_buffer->read_index; | |
169 | ||
170 | return next; | |
171 | } | |
172 | ||
173 | /* | |
174 | * hv_get_next_readlocation_withoffset() | |
175 | * | |
176 | * Get the next read location + offset for the specified ring buffer. | |
177 | * This allows the caller to skip | |
178 | */ | |
179 | static inline u32 | |
180 | hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info, | |
181 | u32 offset) | |
182 | { | |
183 | u32 next = ring_info->ring_buffer->read_index; | |
184 | ||
185 | next += offset; | |
186 | next %= ring_info->ring_datasize; | |
187 | ||
188 | return next; | |
189 | } | |
190 | ||
191 | /* | |
192 | * | |
193 | * hv_set_next_read_location() | |
194 | * | |
195 | * Set the next read location for the specified ring buffer | |
196 | * | |
197 | */ | |
198 | static inline void | |
199 | hv_set_next_read_location(struct hv_ring_buffer_info *ring_info, | |
200 | u32 next_read_location) | |
201 | { | |
202 | ring_info->ring_buffer->read_index = next_read_location; | |
203 | } | |
204 | ||
205 | ||
206 | /* | |
207 | * | |
208 | * hv_get_ring_buffer() | |
209 | * | |
210 | * Get the start of the ring buffer | |
211 | */ | |
212 | static inline void * | |
213 | hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info) | |
214 | { | |
215 | return (void *)ring_info->ring_buffer->buffer; | |
216 | } | |
217 | ||
218 | ||
219 | /* | |
220 | * | |
221 | * hv_get_ring_buffersize() | |
222 | * | |
223 | * Get the size of the ring buffer | |
224 | */ | |
225 | static inline u32 | |
226 | hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info) | |
227 | { | |
228 | return ring_info->ring_datasize; | |
229 | } | |
230 | ||
231 | /* | |
232 | * | |
233 | * hv_get_ring_bufferindices() | |
234 | * | |
235 | * Get the read and write indices as u64 of the specified ring buffer | |
236 | * | |
237 | */ | |
238 | static inline u64 | |
239 | hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info) | |
240 | { | |
241 | return (u64)ring_info->ring_buffer->write_index << 32; | |
242 | } | |
243 | ||
244 | /* | |
245 | * | |
246 | * hv_copyfrom_ringbuffer() | |
247 | * | |
248 | * Helper routine to copy to source from ring buffer. | |
249 | * Assume there is enough room. Handles wrap-around in src case only!! | |
250 | * | |
251 | */ | |
252 | static u32 hv_copyfrom_ringbuffer( | |
253 | struct hv_ring_buffer_info *ring_info, | |
254 | void *dest, | |
255 | u32 destlen, | |
256 | u32 start_read_offset) | |
257 | { | |
258 | void *ring_buffer = hv_get_ring_buffer(ring_info); | |
259 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); | |
260 | ||
261 | u32 frag_len; | |
262 | ||
263 | /* wrap-around detected at the src */ | |
264 | if (destlen > ring_buffer_size - start_read_offset) { | |
265 | frag_len = ring_buffer_size - start_read_offset; | |
266 | ||
267 | memcpy(dest, ring_buffer + start_read_offset, frag_len); | |
268 | memcpy(dest + frag_len, ring_buffer, destlen - frag_len); | |
269 | } else | |
270 | ||
271 | memcpy(dest, ring_buffer + start_read_offset, destlen); | |
272 | ||
273 | ||
274 | start_read_offset += destlen; | |
275 | start_read_offset %= ring_buffer_size; | |
276 | ||
277 | return start_read_offset; | |
278 | } | |
279 | ||
280 | ||
281 | /* | |
282 | * | |
283 | * hv_copyto_ringbuffer() | |
284 | * | |
285 | * Helper routine to copy from source to ring buffer. | |
286 | * Assume there is enough room. Handles wrap-around in dest case only!! | |
287 | * | |
288 | */ | |
289 | static u32 hv_copyto_ringbuffer( | |
290 | struct hv_ring_buffer_info *ring_info, | |
291 | u32 start_write_offset, | |
292 | void *src, | |
293 | u32 srclen) | |
294 | { | |
295 | void *ring_buffer = hv_get_ring_buffer(ring_info); | |
296 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); | |
297 | u32 frag_len; | |
298 | ||
299 | /* wrap-around detected! */ | |
300 | if (srclen > ring_buffer_size - start_write_offset) { | |
301 | frag_len = ring_buffer_size - start_write_offset; | |
302 | memcpy(ring_buffer + start_write_offset, src, frag_len); | |
303 | memcpy(ring_buffer, src + frag_len, srclen - frag_len); | |
304 | } else | |
305 | memcpy(ring_buffer + start_write_offset, src, srclen); | |
306 | ||
307 | start_write_offset += srclen; | |
308 | start_write_offset %= ring_buffer_size; | |
309 | ||
310 | return start_write_offset; | |
311 | } | |
312 | ||
313 | /* | |
314 | * | |
315 | * hv_ringbuffer_get_debuginfo() | |
316 | * | |
317 | * Get various debug metrics for the specified ring buffer | |
318 | * | |
319 | */ | |
320 | void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info, | |
321 | struct hv_ring_buffer_debug_info *debug_info) | |
322 | { | |
323 | u32 bytes_avail_towrite; | |
324 | u32 bytes_avail_toread; | |
325 | ||
326 | if (ring_info->ring_buffer) { | |
327 | hv_get_ringbuffer_availbytes(ring_info, | |
328 | &bytes_avail_toread, | |
329 | &bytes_avail_towrite); | |
330 | ||
331 | debug_info->bytes_avail_toread = bytes_avail_toread; | |
332 | debug_info->bytes_avail_towrite = bytes_avail_towrite; | |
333 | debug_info->current_read_index = | |
334 | ring_info->ring_buffer->read_index; | |
335 | debug_info->current_write_index = | |
336 | ring_info->ring_buffer->write_index; | |
337 | debug_info->current_interrupt_mask = | |
338 | ring_info->ring_buffer->interrupt_mask; | |
339 | } | |
340 | } | |
341 | ||
342 | /* | |
343 | * | |
344 | * hv_ringbuffer_init() | |
345 | * | |
346 | *Initialize the ring buffer | |
347 | * | |
348 | */ | |
349 | int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info, | |
350 | void *buffer, u32 buflen) | |
351 | { | |
352 | if (sizeof(struct hv_ring_buffer) != PAGE_SIZE) | |
353 | return -EINVAL; | |
354 | ||
355 | memset(ring_info, 0, sizeof(struct hv_ring_buffer_info)); | |
356 | ||
357 | ring_info->ring_buffer = (struct hv_ring_buffer *)buffer; | |
358 | ring_info->ring_buffer->read_index = | |
359 | ring_info->ring_buffer->write_index = 0; | |
360 | ||
361 | ring_info->ring_size = buflen; | |
362 | ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer); | |
363 | ||
364 | spin_lock_init(&ring_info->ring_lock); | |
365 | ||
366 | return 0; | |
367 | } | |
368 | ||
369 | /* | |
370 | * | |
371 | * hv_ringbuffer_cleanup() | |
372 | * | |
373 | * Cleanup the ring buffer | |
374 | * | |
375 | */ | |
376 | void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info) | |
377 | { | |
378 | } | |
379 | ||
380 | /* | |
381 | * | |
382 | * hv_ringbuffer_write() | |
383 | * | |
384 | * Write to the ring buffer | |
385 | * | |
386 | */ | |
387 | int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info, | |
388 | struct scatterlist *sglist, u32 sgcount, bool *signal) | |
389 | { | |
390 | int i = 0; | |
391 | u32 bytes_avail_towrite; | |
392 | u32 bytes_avail_toread; | |
393 | u32 totalbytes_towrite = 0; | |
394 | ||
395 | struct scatterlist *sg; | |
396 | u32 next_write_location; | |
397 | u32 old_write; | |
398 | u64 prev_indices = 0; | |
399 | unsigned long flags; | |
400 | ||
401 | for_each_sg(sglist, sg, sgcount, i) | |
402 | { | |
403 | totalbytes_towrite += sg->length; | |
404 | } | |
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_each_sg(sglist, sg, sgcount, i) | |
429 | { | |
430 | next_write_location = hv_copyto_ringbuffer(outring_info, | |
431 | next_write_location, | |
432 | sg_virt(sg), | |
433 | sg->length); | |
434 | } | |
435 | ||
436 | /* Set previous packet start */ | |
437 | prev_indices = hv_get_ring_bufferindices(outring_info); | |
438 | ||
439 | next_write_location = hv_copyto_ringbuffer(outring_info, | |
440 | next_write_location, | |
441 | &prev_indices, | |
442 | sizeof(u64)); | |
443 | ||
444 | /* Issue a full memory barrier before updating the write index */ | |
445 | mb(); | |
446 | ||
447 | /* Now, update the write location */ | |
448 | hv_set_next_write_location(outring_info, next_write_location); | |
449 | ||
450 | ||
451 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); | |
452 | ||
453 | *signal = hv_need_to_signal(old_write, outring_info); | |
454 | return 0; | |
455 | } | |
456 | ||
457 | ||
458 | /* | |
459 | * | |
460 | * hv_ringbuffer_peek() | |
461 | * | |
462 | * Read without advancing the read index | |
463 | * | |
464 | */ | |
465 | int hv_ringbuffer_peek(struct hv_ring_buffer_info *Inring_info, | |
466 | void *Buffer, u32 buflen) | |
467 | { | |
468 | u32 bytes_avail_towrite; | |
469 | u32 bytes_avail_toread; | |
470 | u32 next_read_location = 0; | |
471 | unsigned long flags; | |
472 | ||
473 | spin_lock_irqsave(&Inring_info->ring_lock, flags); | |
474 | ||
475 | hv_get_ringbuffer_availbytes(Inring_info, | |
476 | &bytes_avail_toread, | |
477 | &bytes_avail_towrite); | |
478 | ||
479 | /* Make sure there is something to read */ | |
480 | if (bytes_avail_toread < buflen) { | |
481 | ||
482 | spin_unlock_irqrestore(&Inring_info->ring_lock, flags); | |
483 | ||
484 | return -EAGAIN; | |
485 | } | |
486 | ||
487 | /* Convert to byte offset */ | |
488 | next_read_location = hv_get_next_read_location(Inring_info); | |
489 | ||
490 | next_read_location = hv_copyfrom_ringbuffer(Inring_info, | |
491 | Buffer, | |
492 | buflen, | |
493 | next_read_location); | |
494 | ||
495 | spin_unlock_irqrestore(&Inring_info->ring_lock, flags); | |
496 | ||
497 | return 0; | |
498 | } | |
499 | ||
500 | ||
501 | /* | |
502 | * | |
503 | * hv_ringbuffer_read() | |
504 | * | |
505 | * Read and advance the read index | |
506 | * | |
507 | */ | |
508 | int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info, void *buffer, | |
509 | u32 buflen, u32 offset, bool *signal) | |
510 | { | |
511 | u32 bytes_avail_towrite; | |
512 | u32 bytes_avail_toread; | |
513 | u32 next_read_location = 0; | |
514 | u64 prev_indices = 0; | |
515 | unsigned long flags; | |
516 | u32 old_read; | |
517 | ||
518 | if (buflen <= 0) | |
519 | return -EINVAL; | |
520 | ||
521 | spin_lock_irqsave(&inring_info->ring_lock, flags); | |
522 | ||
523 | hv_get_ringbuffer_availbytes(inring_info, | |
524 | &bytes_avail_toread, | |
525 | &bytes_avail_towrite); | |
526 | ||
527 | old_read = bytes_avail_toread; | |
528 | ||
529 | /* Make sure there is something to read */ | |
530 | if (bytes_avail_toread < buflen) { | |
531 | spin_unlock_irqrestore(&inring_info->ring_lock, flags); | |
532 | ||
533 | return -EAGAIN; | |
534 | } | |
535 | ||
536 | next_read_location = | |
537 | hv_get_next_readlocation_withoffset(inring_info, offset); | |
538 | ||
539 | next_read_location = hv_copyfrom_ringbuffer(inring_info, | |
540 | buffer, | |
541 | buflen, | |
542 | next_read_location); | |
543 | ||
544 | next_read_location = hv_copyfrom_ringbuffer(inring_info, | |
545 | &prev_indices, | |
546 | sizeof(u64), | |
547 | next_read_location); | |
548 | ||
549 | /* Make sure all reads are done before we update the read index since */ | |
550 | /* the writer may start writing to the read area once the read index */ | |
551 | /*is updated */ | |
552 | mb(); | |
553 | ||
554 | /* Update the read index */ | |
555 | hv_set_next_read_location(inring_info, next_read_location); | |
556 | ||
557 | spin_unlock_irqrestore(&inring_info->ring_lock, flags); | |
558 | ||
559 | *signal = hv_need_to_signal_on_read(old_read, inring_info); | |
560 | ||
561 | return 0; | |
562 | } |