Commit | Line | Data |
---|---|---|
3e7ee490 HJ |
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> | |
b2a5a585 | 21 | * K. Y. Srinivasan <kys@microsoft.com> |
3e7ee490 HJ |
22 | * |
23 | */ | |
0a46618d | 24 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
3e7ee490 | 25 | |
a0086dc5 GKH |
26 | #include <linux/kernel.h> |
27 | #include <linux/mm.h> | |
46a97191 | 28 | #include <linux/hyperv.h> |
011a7c3c | 29 | #include <linux/uio.h> |
3f335ea2 | 30 | |
0f2a6619 | 31 | #include "hyperv_vmbus.h" |
3e7ee490 | 32 | |
6fdf3b21 S |
33 | void hv_begin_read(struct hv_ring_buffer_info *rbi) |
34 | { | |
35 | rbi->ring_buffer->interrupt_mask = 1; | |
35848f68 | 36 | mb(); |
6fdf3b21 S |
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; | |
35848f68 | 45 | mb(); |
6fdf3b21 S |
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 | ||
98fa8cf4 S |
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 | { | |
35848f68 | 75 | mb(); |
98fa8cf4 S |
76 | if (rbi->ring_buffer->interrupt_mask) |
77 | return false; | |
78 | ||
e91e84fa JW |
79 | /* check interrupt_mask before read_index */ |
80 | rmb(); | |
98fa8cf4 S |
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 | ||
c2b8e520 S |
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 | ||
1db488d1 | 106 | static bool hv_need_to_signal_on_read(struct hv_ring_buffer_info *rbi) |
c2b8e520 | 107 | { |
c2b8e520 S |
108 | u32 cur_write_sz; |
109 | u32 r_size; | |
1db488d1 | 110 | u32 write_loc; |
c2b8e520 | 111 | u32 read_loc = rbi->ring_buffer->read_index; |
1db488d1 | 112 | u32 pending_sz; |
c2b8e520 | 113 | |
1db488d1 S |
114 | /* |
115 | * Issue a full memory barrier before making the signaling decision. | |
116 | * Here is the reason for having this barrier: | |
117 | * If the reading of the pend_sz (in this function) | |
118 | * were to be reordered and read before we commit the new read | |
119 | * index (in the calling function) we could | |
120 | * have a problem. If the host were to set the pending_sz after we | |
121 | * have sampled pending_sz and go to sleep before we commit the | |
122 | * read index, we could miss sending the interrupt. Issue a full | |
123 | * memory barrier to address this. | |
124 | */ | |
125 | mb(); | |
126 | ||
127 | pending_sz = rbi->ring_buffer->pending_send_sz; | |
128 | write_loc = rbi->ring_buffer->write_index; | |
822f18d4 | 129 | /* If the other end is not blocked on write don't bother. */ |
c2b8e520 S |
130 | if (pending_sz == 0) |
131 | return false; | |
132 | ||
133 | r_size = rbi->ring_datasize; | |
134 | cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) : | |
135 | read_loc - write_loc; | |
136 | ||
1db488d1 | 137 | if (cur_write_sz >= pending_sz) |
c2b8e520 S |
138 | return true; |
139 | ||
140 | return false; | |
141 | } | |
3e7ee490 | 142 | |
822f18d4 | 143 | /* Get the next write location for the specified ring buffer. */ |
4d643114 | 144 | static inline u32 |
2b8a912e | 145 | hv_get_next_write_location(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 146 | { |
fc8c72eb | 147 | u32 next = ring_info->ring_buffer->write_index; |
3e7ee490 | 148 | |
3e7ee490 HJ |
149 | return next; |
150 | } | |
151 | ||
822f18d4 | 152 | /* Set the next write location for the specified ring buffer. */ |
3e7ee490 | 153 | static inline void |
2b8a912e | 154 | hv_set_next_write_location(struct hv_ring_buffer_info *ring_info, |
fc8c72eb | 155 | u32 next_write_location) |
3e7ee490 | 156 | { |
fc8c72eb | 157 | ring_info->ring_buffer->write_index = next_write_location; |
3e7ee490 HJ |
158 | } |
159 | ||
822f18d4 | 160 | /* Get the next read location for the specified ring buffer. */ |
4d643114 | 161 | static inline u32 |
2b8a912e | 162 | hv_get_next_read_location(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 163 | { |
fc8c72eb | 164 | u32 next = ring_info->ring_buffer->read_index; |
3e7ee490 | 165 | |
3e7ee490 HJ |
166 | return next; |
167 | } | |
168 | ||
b2a5a585 | 169 | /* |
b2a5a585 | 170 | * Get the next read location + offset for the specified ring buffer. |
822f18d4 | 171 | * This allows the caller to skip. |
b2a5a585 | 172 | */ |
4d643114 | 173 | static inline u32 |
2b8a912e | 174 | hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info, |
1ac58644 | 175 | u32 offset) |
3e7ee490 | 176 | { |
fc8c72eb | 177 | u32 next = ring_info->ring_buffer->read_index; |
3e7ee490 | 178 | |
fc8c72eb HZ |
179 | next += offset; |
180 | next %= ring_info->ring_datasize; | |
3e7ee490 HJ |
181 | |
182 | return next; | |
183 | } | |
184 | ||
822f18d4 | 185 | /* Set the next read location for the specified ring buffer. */ |
3e7ee490 | 186 | static inline void |
2b8a912e | 187 | hv_set_next_read_location(struct hv_ring_buffer_info *ring_info, |
fc8c72eb | 188 | u32 next_read_location) |
3e7ee490 | 189 | { |
fc8c72eb | 190 | ring_info->ring_buffer->read_index = next_read_location; |
3e7ee490 HJ |
191 | } |
192 | ||
193 | ||
822f18d4 | 194 | /* Get the start of the ring buffer. */ |
8282c400 | 195 | static inline void * |
2b8a912e | 196 | hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 197 | { |
fc8c72eb | 198 | return (void *)ring_info->ring_buffer->buffer; |
3e7ee490 HJ |
199 | } |
200 | ||
201 | ||
822f18d4 | 202 | /* Get the size of the ring buffer. */ |
4d643114 | 203 | static inline u32 |
2b8a912e | 204 | hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 205 | { |
fc8c72eb | 206 | return ring_info->ring_datasize; |
3e7ee490 HJ |
207 | } |
208 | ||
822f18d4 | 209 | /* Get the read and write indices as u64 of the specified ring buffer. */ |
59471438 | 210 | static inline u64 |
2b8a912e | 211 | hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 212 | { |
fc8c72eb | 213 | return (u64)ring_info->ring_buffer->write_index << 32; |
3e7ee490 HJ |
214 | } |
215 | ||
8f1136ae | 216 | /* |
8f1136ae S |
217 | * Helper routine to copy to source from ring buffer. |
218 | * Assume there is enough room. Handles wrap-around in src case only!! | |
8f1136ae S |
219 | */ |
220 | static u32 hv_copyfrom_ringbuffer( | |
221 | struct hv_ring_buffer_info *ring_info, | |
222 | void *dest, | |
223 | u32 destlen, | |
224 | u32 start_read_offset) | |
225 | { | |
226 | void *ring_buffer = hv_get_ring_buffer(ring_info); | |
227 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); | |
228 | ||
229 | u32 frag_len; | |
230 | ||
231 | /* wrap-around detected at the src */ | |
232 | if (destlen > ring_buffer_size - start_read_offset) { | |
233 | frag_len = ring_buffer_size - start_read_offset; | |
234 | ||
235 | memcpy(dest, ring_buffer + start_read_offset, frag_len); | |
236 | memcpy(dest + frag_len, ring_buffer, destlen - frag_len); | |
237 | } else | |
238 | ||
239 | memcpy(dest, ring_buffer + start_read_offset, destlen); | |
240 | ||
241 | ||
242 | start_read_offset += destlen; | |
243 | start_read_offset %= ring_buffer_size; | |
244 | ||
245 | return start_read_offset; | |
246 | } | |
247 | ||
248 | ||
7581578d | 249 | /* |
7581578d S |
250 | * Helper routine to copy from source to ring buffer. |
251 | * Assume there is enough room. Handles wrap-around in dest case only!! | |
7581578d S |
252 | */ |
253 | static u32 hv_copyto_ringbuffer( | |
fc8c72eb HZ |
254 | struct hv_ring_buffer_info *ring_info, |
255 | u32 start_write_offset, | |
256 | void *src, | |
7581578d S |
257 | u32 srclen) |
258 | { | |
259 | void *ring_buffer = hv_get_ring_buffer(ring_info); | |
260 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); | |
261 | u32 frag_len; | |
262 | ||
263 | /* wrap-around detected! */ | |
264 | if (srclen > ring_buffer_size - start_write_offset) { | |
265 | frag_len = ring_buffer_size - start_write_offset; | |
266 | memcpy(ring_buffer + start_write_offset, src, frag_len); | |
267 | memcpy(ring_buffer, src + frag_len, srclen - frag_len); | |
268 | } else | |
269 | memcpy(ring_buffer + start_write_offset, src, srclen); | |
3e7ee490 | 270 | |
7581578d S |
271 | start_write_offset += srclen; |
272 | start_write_offset %= ring_buffer_size; | |
273 | ||
274 | return start_write_offset; | |
275 | } | |
3e7ee490 | 276 | |
822f18d4 | 277 | /* Get various debug metrics for the specified ring buffer. */ |
a75b61d5 | 278 | void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info, |
80682b7a | 279 | struct hv_ring_buffer_debug_info *debug_info) |
3e7ee490 | 280 | { |
fc8c72eb HZ |
281 | u32 bytes_avail_towrite; |
282 | u32 bytes_avail_toread; | |
3e7ee490 | 283 | |
fc8c72eb | 284 | if (ring_info->ring_buffer) { |
2b8a912e | 285 | hv_get_ringbuffer_availbytes(ring_info, |
fc8c72eb HZ |
286 | &bytes_avail_toread, |
287 | &bytes_avail_towrite); | |
3e7ee490 | 288 | |
fc8c72eb HZ |
289 | debug_info->bytes_avail_toread = bytes_avail_toread; |
290 | debug_info->bytes_avail_towrite = bytes_avail_towrite; | |
82f8bd40 | 291 | debug_info->current_read_index = |
fc8c72eb | 292 | ring_info->ring_buffer->read_index; |
82f8bd40 | 293 | debug_info->current_write_index = |
fc8c72eb | 294 | ring_info->ring_buffer->write_index; |
82f8bd40 | 295 | debug_info->current_interrupt_mask = |
fc8c72eb | 296 | ring_info->ring_buffer->interrupt_mask; |
3e7ee490 HJ |
297 | } |
298 | } | |
299 | ||
822f18d4 | 300 | /* Initialize the ring buffer. */ |
72a95cbc | 301 | int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info, |
fc8c72eb | 302 | void *buffer, u32 buflen) |
3e7ee490 | 303 | { |
4a1b3acc | 304 | if (sizeof(struct hv_ring_buffer) != PAGE_SIZE) |
3324fb40 | 305 | return -EINVAL; |
3e7ee490 | 306 | |
fc8c72eb | 307 | memset(ring_info, 0, sizeof(struct hv_ring_buffer_info)); |
3e7ee490 | 308 | |
fc8c72eb HZ |
309 | ring_info->ring_buffer = (struct hv_ring_buffer *)buffer; |
310 | ring_info->ring_buffer->read_index = | |
311 | ring_info->ring_buffer->write_index = 0; | |
3e7ee490 | 312 | |
822f18d4 | 313 | /* Set the feature bit for enabling flow control. */ |
046c7911 S |
314 | ring_info->ring_buffer->feature_bits.value = 1; |
315 | ||
fc8c72eb HZ |
316 | ring_info->ring_size = buflen; |
317 | ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer); | |
3e7ee490 | 318 | |
fc8c72eb | 319 | spin_lock_init(&ring_info->ring_lock); |
3e7ee490 HJ |
320 | |
321 | return 0; | |
322 | } | |
323 | ||
822f18d4 | 324 | /* Cleanup the ring buffer. */ |
2dba688b | 325 | void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 326 | { |
3e7ee490 HJ |
327 | } |
328 | ||
822f18d4 | 329 | /* Write to the ring buffer. */ |
633c4dce | 330 | int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info, |
fe760e4d | 331 | struct kvec *kv_list, u32 kv_count, bool *signal, bool lock) |
3e7ee490 | 332 | { |
4408f531 | 333 | int i = 0; |
fc8c72eb HZ |
334 | u32 bytes_avail_towrite; |
335 | u32 bytes_avail_toread; | |
336 | u32 totalbytes_towrite = 0; | |
3e7ee490 | 337 | |
66a60543 | 338 | u32 next_write_location; |
98fa8cf4 | 339 | u32 old_write; |
fc8c72eb | 340 | u64 prev_indices = 0; |
fe760e4d | 341 | unsigned long flags = 0; |
3e7ee490 | 342 | |
011a7c3c S |
343 | for (i = 0; i < kv_count; i++) |
344 | totalbytes_towrite += kv_list[i].iov_len; | |
3e7ee490 | 345 | |
fc8c72eb | 346 | totalbytes_towrite += sizeof(u64); |
3e7ee490 | 347 | |
fe760e4d S |
348 | if (lock) |
349 | spin_lock_irqsave(&outring_info->ring_lock, flags); | |
3e7ee490 | 350 | |
2b8a912e | 351 | hv_get_ringbuffer_availbytes(outring_info, |
fc8c72eb HZ |
352 | &bytes_avail_toread, |
353 | &bytes_avail_towrite); | |
3e7ee490 | 354 | |
822f18d4 VK |
355 | /* |
356 | * If there is only room for the packet, assume it is full. | |
357 | * Otherwise, the next time around, we think the ring buffer | |
358 | * is empty since the read index == write index. | |
359 | */ | |
fc8c72eb | 360 | if (bytes_avail_towrite <= totalbytes_towrite) { |
fe760e4d S |
361 | if (lock) |
362 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); | |
d2598f01 | 363 | return -EAGAIN; |
3e7ee490 HJ |
364 | } |
365 | ||
454f18a9 | 366 | /* Write to the ring buffer */ |
2b8a912e | 367 | next_write_location = hv_get_next_write_location(outring_info); |
3e7ee490 | 368 | |
98fa8cf4 S |
369 | old_write = next_write_location; |
370 | ||
011a7c3c | 371 | for (i = 0; i < kv_count; i++) { |
2b8a912e | 372 | next_write_location = hv_copyto_ringbuffer(outring_info, |
fc8c72eb | 373 | next_write_location, |
011a7c3c S |
374 | kv_list[i].iov_base, |
375 | kv_list[i].iov_len); | |
3e7ee490 HJ |
376 | } |
377 | ||
454f18a9 | 378 | /* Set previous packet start */ |
2b8a912e | 379 | prev_indices = hv_get_ring_bufferindices(outring_info); |
3e7ee490 | 380 | |
2b8a912e | 381 | next_write_location = hv_copyto_ringbuffer(outring_info, |
fc8c72eb HZ |
382 | next_write_location, |
383 | &prev_indices, | |
b219b3f7 | 384 | sizeof(u64)); |
3e7ee490 | 385 | |
98fa8cf4 | 386 | /* Issue a full memory barrier before updating the write index */ |
35848f68 | 387 | mb(); |
3e7ee490 | 388 | |
454f18a9 | 389 | /* Now, update the write location */ |
2b8a912e | 390 | hv_set_next_write_location(outring_info, next_write_location); |
3e7ee490 | 391 | |
3e7ee490 | 392 | |
fe760e4d S |
393 | if (lock) |
394 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); | |
98fa8cf4 S |
395 | |
396 | *signal = hv_need_to_signal(old_write, outring_info); | |
3e7ee490 HJ |
397 | return 0; |
398 | } | |
399 | ||
940b68e2 VK |
400 | int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info, |
401 | void *buffer, u32 buflen, u32 *buffer_actual_len, | |
402 | u64 *requestid, bool *signal, bool raw) | |
3e7ee490 | 403 | { |
fc8c72eb HZ |
404 | u32 bytes_avail_towrite; |
405 | u32 bytes_avail_toread; | |
406 | u32 next_read_location = 0; | |
407 | u64 prev_indices = 0; | |
940b68e2 VK |
408 | struct vmpacket_descriptor desc; |
409 | u32 offset; | |
410 | u32 packetlen; | |
411 | int ret = 0; | |
3e7ee490 | 412 | |
fc8c72eb | 413 | if (buflen <= 0) |
a16e1485 | 414 | return -EINVAL; |
3e7ee490 | 415 | |
3e7ee490 | 416 | |
940b68e2 VK |
417 | *buffer_actual_len = 0; |
418 | *requestid = 0; | |
419 | ||
2b8a912e | 420 | hv_get_ringbuffer_availbytes(inring_info, |
fc8c72eb HZ |
421 | &bytes_avail_toread, |
422 | &bytes_avail_towrite); | |
3e7ee490 | 423 | |
454f18a9 | 424 | /* Make sure there is something to read */ |
940b68e2 VK |
425 | if (bytes_avail_toread < sizeof(desc)) { |
426 | /* | |
427 | * No error is set when there is even no header, drivers are | |
428 | * supposed to analyze buffer_actual_len. | |
429 | */ | |
3eba9a77 | 430 | return ret; |
940b68e2 | 431 | } |
3e7ee490 | 432 | |
940b68e2 VK |
433 | next_read_location = hv_get_next_read_location(inring_info); |
434 | next_read_location = hv_copyfrom_ringbuffer(inring_info, &desc, | |
435 | sizeof(desc), | |
436 | next_read_location); | |
437 | ||
438 | offset = raw ? 0 : (desc.offset8 << 3); | |
439 | packetlen = (desc.len8 << 3) - offset; | |
440 | *buffer_actual_len = packetlen; | |
441 | *requestid = desc.trans_id; | |
442 | ||
3eba9a77 S |
443 | if (bytes_avail_toread < packetlen + offset) |
444 | return -EAGAIN; | |
940b68e2 | 445 | |
3eba9a77 S |
446 | if (packetlen > buflen) |
447 | return -ENOBUFS; | |
3e7ee490 | 448 | |
1ac58644 | 449 | next_read_location = |
2b8a912e | 450 | hv_get_next_readlocation_withoffset(inring_info, offset); |
3e7ee490 | 451 | |
2b8a912e | 452 | next_read_location = hv_copyfrom_ringbuffer(inring_info, |
fc8c72eb | 453 | buffer, |
940b68e2 | 454 | packetlen, |
fc8c72eb | 455 | next_read_location); |
3e7ee490 | 456 | |
2b8a912e | 457 | next_read_location = hv_copyfrom_ringbuffer(inring_info, |
fc8c72eb | 458 | &prev_indices, |
4408f531 | 459 | sizeof(u64), |
fc8c72eb | 460 | next_read_location); |
3e7ee490 | 461 | |
822f18d4 VK |
462 | /* |
463 | * Make sure all reads are done before we update the read index since | |
464 | * the writer may start writing to the read area once the read index | |
465 | * is updated. | |
466 | */ | |
35848f68 | 467 | mb(); |
3e7ee490 | 468 | |
454f18a9 | 469 | /* Update the read index */ |
2b8a912e | 470 | hv_set_next_read_location(inring_info, next_read_location); |
3e7ee490 | 471 | |
1db488d1 | 472 | *signal = hv_need_to_signal_on_read(inring_info); |
c2b8e520 | 473 | |
940b68e2 | 474 | return ret; |
b5f53dde | 475 | } |