| 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 | /* check interrupt_mask before read_index */ |
| 79 | rmb(); |
| 80 | /* |
| 81 | * This is the only case we need to signal when the |
| 82 | * ring transitions from being empty to non-empty. |
| 83 | */ |
| 84 | if (old_write == rbi->ring_buffer->read_index) |
| 85 | return true; |
| 86 | |
| 87 | return false; |
| 88 | } |
| 89 | |
| 90 | /* |
| 91 | * To optimize the flow management on the send-side, |
| 92 | * when the sender is blocked because of lack of |
| 93 | * sufficient space in the ring buffer, potential the |
| 94 | * consumer of the ring buffer can signal the producer. |
| 95 | * This is controlled by the following parameters: |
| 96 | * |
| 97 | * 1. pending_send_sz: This is the size in bytes that the |
| 98 | * producer is trying to send. |
| 99 | * 2. The feature bit feat_pending_send_sz set to indicate if |
| 100 | * the consumer of the ring will signal when the ring |
| 101 | * state transitions from being full to a state where |
| 102 | * there is room for the producer to send the pending packet. |
| 103 | */ |
| 104 | |
| 105 | static bool hv_need_to_signal_on_read(u32 old_rd, |
| 106 | struct hv_ring_buffer_info *rbi) |
| 107 | { |
| 108 | u32 prev_write_sz; |
| 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 | prev_write_sz = write_loc >= old_rd ? r_size - (write_loc - old_rd) : |
| 126 | old_rd - write_loc; |
| 127 | |
| 128 | |
| 129 | if ((prev_write_sz < pending_sz) && (cur_write_sz >= pending_sz)) |
| 130 | return true; |
| 131 | |
| 132 | return false; |
| 133 | } |
| 134 | |
| 135 | /* |
| 136 | * hv_get_next_write_location() |
| 137 | * |
| 138 | * Get the next write location for the specified ring buffer |
| 139 | * |
| 140 | */ |
| 141 | static inline u32 |
| 142 | hv_get_next_write_location(struct hv_ring_buffer_info *ring_info) |
| 143 | { |
| 144 | u32 next = ring_info->ring_buffer->write_index; |
| 145 | |
| 146 | return next; |
| 147 | } |
| 148 | |
| 149 | /* |
| 150 | * hv_set_next_write_location() |
| 151 | * |
| 152 | * Set the next write location for the specified ring buffer |
| 153 | * |
| 154 | */ |
| 155 | static inline void |
| 156 | hv_set_next_write_location(struct hv_ring_buffer_info *ring_info, |
| 157 | u32 next_write_location) |
| 158 | { |
| 159 | ring_info->ring_buffer->write_index = next_write_location; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * hv_get_next_read_location() |
| 164 | * |
| 165 | * Get the next read location for the specified ring buffer |
| 166 | */ |
| 167 | static inline u32 |
| 168 | hv_get_next_read_location(struct hv_ring_buffer_info *ring_info) |
| 169 | { |
| 170 | u32 next = ring_info->ring_buffer->read_index; |
| 171 | |
| 172 | return next; |
| 173 | } |
| 174 | |
| 175 | /* |
| 176 | * hv_get_next_readlocation_withoffset() |
| 177 | * |
| 178 | * Get the next read location + offset for the specified ring buffer. |
| 179 | * This allows the caller to skip |
| 180 | */ |
| 181 | static inline u32 |
| 182 | hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info, |
| 183 | u32 offset) |
| 184 | { |
| 185 | u32 next = ring_info->ring_buffer->read_index; |
| 186 | |
| 187 | next += offset; |
| 188 | next %= ring_info->ring_datasize; |
| 189 | |
| 190 | return next; |
| 191 | } |
| 192 | |
| 193 | /* |
| 194 | * |
| 195 | * hv_set_next_read_location() |
| 196 | * |
| 197 | * Set the next read location for the specified ring buffer |
| 198 | * |
| 199 | */ |
| 200 | static inline void |
| 201 | hv_set_next_read_location(struct hv_ring_buffer_info *ring_info, |
| 202 | u32 next_read_location) |
| 203 | { |
| 204 | ring_info->ring_buffer->read_index = next_read_location; |
| 205 | } |
| 206 | |
| 207 | |
| 208 | /* |
| 209 | * |
| 210 | * hv_get_ring_buffer() |
| 211 | * |
| 212 | * Get the start of the ring buffer |
| 213 | */ |
| 214 | static inline void * |
| 215 | hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info) |
| 216 | { |
| 217 | return (void *)ring_info->ring_buffer->buffer; |
| 218 | } |
| 219 | |
| 220 | |
| 221 | /* |
| 222 | * |
| 223 | * hv_get_ring_buffersize() |
| 224 | * |
| 225 | * Get the size of the ring buffer |
| 226 | */ |
| 227 | static inline u32 |
| 228 | hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info) |
| 229 | { |
| 230 | return ring_info->ring_datasize; |
| 231 | } |
| 232 | |
| 233 | /* |
| 234 | * |
| 235 | * hv_get_ring_bufferindices() |
| 236 | * |
| 237 | * Get the read and write indices as u64 of the specified ring buffer |
| 238 | * |
| 239 | */ |
| 240 | static inline u64 |
| 241 | hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info) |
| 242 | { |
| 243 | return (u64)ring_info->ring_buffer->write_index << 32; |
| 244 | } |
| 245 | |
| 246 | /* |
| 247 | * |
| 248 | * hv_copyfrom_ringbuffer() |
| 249 | * |
| 250 | * Helper routine to copy to source from ring buffer. |
| 251 | * Assume there is enough room. Handles wrap-around in src case only!! |
| 252 | * |
| 253 | */ |
| 254 | static u32 hv_copyfrom_ringbuffer( |
| 255 | struct hv_ring_buffer_info *ring_info, |
| 256 | void *dest, |
| 257 | u32 destlen, |
| 258 | u32 start_read_offset) |
| 259 | { |
| 260 | void *ring_buffer = hv_get_ring_buffer(ring_info); |
| 261 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); |
| 262 | |
| 263 | u32 frag_len; |
| 264 | |
| 265 | /* wrap-around detected at the src */ |
| 266 | if (destlen > ring_buffer_size - start_read_offset) { |
| 267 | frag_len = ring_buffer_size - start_read_offset; |
| 268 | |
| 269 | memcpy(dest, ring_buffer + start_read_offset, frag_len); |
| 270 | memcpy(dest + frag_len, ring_buffer, destlen - frag_len); |
| 271 | } else |
| 272 | |
| 273 | memcpy(dest, ring_buffer + start_read_offset, destlen); |
| 274 | |
| 275 | |
| 276 | start_read_offset += destlen; |
| 277 | start_read_offset %= ring_buffer_size; |
| 278 | |
| 279 | return start_read_offset; |
| 280 | } |
| 281 | |
| 282 | |
| 283 | /* |
| 284 | * |
| 285 | * hv_copyto_ringbuffer() |
| 286 | * |
| 287 | * Helper routine to copy from source to ring buffer. |
| 288 | * Assume there is enough room. Handles wrap-around in dest case only!! |
| 289 | * |
| 290 | */ |
| 291 | static u32 hv_copyto_ringbuffer( |
| 292 | struct hv_ring_buffer_info *ring_info, |
| 293 | u32 start_write_offset, |
| 294 | void *src, |
| 295 | u32 srclen) |
| 296 | { |
| 297 | void *ring_buffer = hv_get_ring_buffer(ring_info); |
| 298 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); |
| 299 | u32 frag_len; |
| 300 | |
| 301 | /* wrap-around detected! */ |
| 302 | if (srclen > ring_buffer_size - start_write_offset) { |
| 303 | frag_len = ring_buffer_size - start_write_offset; |
| 304 | memcpy(ring_buffer + start_write_offset, src, frag_len); |
| 305 | memcpy(ring_buffer, src + frag_len, srclen - frag_len); |
| 306 | } else |
| 307 | memcpy(ring_buffer + start_write_offset, src, srclen); |
| 308 | |
| 309 | start_write_offset += srclen; |
| 310 | start_write_offset %= ring_buffer_size; |
| 311 | |
| 312 | return start_write_offset; |
| 313 | } |
| 314 | |
| 315 | /* |
| 316 | * |
| 317 | * hv_ringbuffer_get_debuginfo() |
| 318 | * |
| 319 | * Get various debug metrics for the specified ring buffer |
| 320 | * |
| 321 | */ |
| 322 | void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info, |
| 323 | struct hv_ring_buffer_debug_info *debug_info) |
| 324 | { |
| 325 | u32 bytes_avail_towrite; |
| 326 | u32 bytes_avail_toread; |
| 327 | |
| 328 | if (ring_info->ring_buffer) { |
| 329 | hv_get_ringbuffer_availbytes(ring_info, |
| 330 | &bytes_avail_toread, |
| 331 | &bytes_avail_towrite); |
| 332 | |
| 333 | debug_info->bytes_avail_toread = bytes_avail_toread; |
| 334 | debug_info->bytes_avail_towrite = bytes_avail_towrite; |
| 335 | debug_info->current_read_index = |
| 336 | ring_info->ring_buffer->read_index; |
| 337 | debug_info->current_write_index = |
| 338 | ring_info->ring_buffer->write_index; |
| 339 | debug_info->current_interrupt_mask = |
| 340 | ring_info->ring_buffer->interrupt_mask; |
| 341 | } |
| 342 | } |
| 343 | |
| 344 | /* |
| 345 | * |
| 346 | * hv_ringbuffer_init() |
| 347 | * |
| 348 | *Initialize the ring buffer |
| 349 | * |
| 350 | */ |
| 351 | int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info, |
| 352 | void *buffer, u32 buflen) |
| 353 | { |
| 354 | if (sizeof(struct hv_ring_buffer) != PAGE_SIZE) |
| 355 | return -EINVAL; |
| 356 | |
| 357 | memset(ring_info, 0, sizeof(struct hv_ring_buffer_info)); |
| 358 | |
| 359 | ring_info->ring_buffer = (struct hv_ring_buffer *)buffer; |
| 360 | ring_info->ring_buffer->read_index = |
| 361 | ring_info->ring_buffer->write_index = 0; |
| 362 | |
| 363 | ring_info->ring_size = buflen; |
| 364 | ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer); |
| 365 | |
| 366 | spin_lock_init(&ring_info->ring_lock); |
| 367 | |
| 368 | return 0; |
| 369 | } |
| 370 | |
| 371 | /* |
| 372 | * |
| 373 | * hv_ringbuffer_cleanup() |
| 374 | * |
| 375 | * Cleanup the ring buffer |
| 376 | * |
| 377 | */ |
| 378 | void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info) |
| 379 | { |
| 380 | } |
| 381 | |
| 382 | /* |
| 383 | * |
| 384 | * hv_ringbuffer_write() |
| 385 | * |
| 386 | * Write to the ring buffer |
| 387 | * |
| 388 | */ |
| 389 | int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info, |
| 390 | struct scatterlist *sglist, u32 sgcount, bool *signal) |
| 391 | { |
| 392 | int i = 0; |
| 393 | u32 bytes_avail_towrite; |
| 394 | u32 bytes_avail_toread; |
| 395 | u32 totalbytes_towrite = 0; |
| 396 | |
| 397 | struct scatterlist *sg; |
| 398 | u32 next_write_location; |
| 399 | u32 old_write; |
| 400 | u64 prev_indices = 0; |
| 401 | unsigned long flags; |
| 402 | |
| 403 | for_each_sg(sglist, sg, sgcount, i) |
| 404 | { |
| 405 | totalbytes_towrite += sg->length; |
| 406 | } |
| 407 | |
| 408 | totalbytes_towrite += sizeof(u64); |
| 409 | |
| 410 | spin_lock_irqsave(&outring_info->ring_lock, flags); |
| 411 | |
| 412 | hv_get_ringbuffer_availbytes(outring_info, |
| 413 | &bytes_avail_toread, |
| 414 | &bytes_avail_towrite); |
| 415 | |
| 416 | |
| 417 | /* If there is only room for the packet, assume it is full. */ |
| 418 | /* Otherwise, the next time around, we think the ring buffer */ |
| 419 | /* is empty since the read index == write index */ |
| 420 | if (bytes_avail_towrite <= totalbytes_towrite) { |
| 421 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); |
| 422 | return -EAGAIN; |
| 423 | } |
| 424 | |
| 425 | /* Write to the ring buffer */ |
| 426 | next_write_location = hv_get_next_write_location(outring_info); |
| 427 | |
| 428 | old_write = next_write_location; |
| 429 | |
| 430 | for_each_sg(sglist, sg, sgcount, i) |
| 431 | { |
| 432 | next_write_location = hv_copyto_ringbuffer(outring_info, |
| 433 | next_write_location, |
| 434 | sg_virt(sg), |
| 435 | sg->length); |
| 436 | } |
| 437 | |
| 438 | /* Set previous packet start */ |
| 439 | prev_indices = hv_get_ring_bufferindices(outring_info); |
| 440 | |
| 441 | next_write_location = hv_copyto_ringbuffer(outring_info, |
| 442 | next_write_location, |
| 443 | &prev_indices, |
| 444 | sizeof(u64)); |
| 445 | |
| 446 | /* Issue a full memory barrier before updating the write index */ |
| 447 | mb(); |
| 448 | |
| 449 | /* Now, update the write location */ |
| 450 | hv_set_next_write_location(outring_info, next_write_location); |
| 451 | |
| 452 | |
| 453 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); |
| 454 | |
| 455 | *signal = hv_need_to_signal(old_write, outring_info); |
| 456 | return 0; |
| 457 | } |
| 458 | |
| 459 | |
| 460 | /* |
| 461 | * |
| 462 | * hv_ringbuffer_peek() |
| 463 | * |
| 464 | * Read without advancing the read index |
| 465 | * |
| 466 | */ |
| 467 | int hv_ringbuffer_peek(struct hv_ring_buffer_info *Inring_info, |
| 468 | void *Buffer, u32 buflen) |
| 469 | { |
| 470 | u32 bytes_avail_towrite; |
| 471 | u32 bytes_avail_toread; |
| 472 | u32 next_read_location = 0; |
| 473 | unsigned long flags; |
| 474 | |
| 475 | spin_lock_irqsave(&Inring_info->ring_lock, flags); |
| 476 | |
| 477 | hv_get_ringbuffer_availbytes(Inring_info, |
| 478 | &bytes_avail_toread, |
| 479 | &bytes_avail_towrite); |
| 480 | |
| 481 | /* Make sure there is something to read */ |
| 482 | if (bytes_avail_toread < buflen) { |
| 483 | |
| 484 | spin_unlock_irqrestore(&Inring_info->ring_lock, flags); |
| 485 | |
| 486 | return -EAGAIN; |
| 487 | } |
| 488 | |
| 489 | /* Convert to byte offset */ |
| 490 | next_read_location = hv_get_next_read_location(Inring_info); |
| 491 | |
| 492 | next_read_location = hv_copyfrom_ringbuffer(Inring_info, |
| 493 | Buffer, |
| 494 | buflen, |
| 495 | next_read_location); |
| 496 | |
| 497 | spin_unlock_irqrestore(&Inring_info->ring_lock, flags); |
| 498 | |
| 499 | return 0; |
| 500 | } |
| 501 | |
| 502 | |
| 503 | /* |
| 504 | * |
| 505 | * hv_ringbuffer_read() |
| 506 | * |
| 507 | * Read and advance the read index |
| 508 | * |
| 509 | */ |
| 510 | int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info, void *buffer, |
| 511 | u32 buflen, u32 offset, bool *signal) |
| 512 | { |
| 513 | u32 bytes_avail_towrite; |
| 514 | u32 bytes_avail_toread; |
| 515 | u32 next_read_location = 0; |
| 516 | u64 prev_indices = 0; |
| 517 | unsigned long flags; |
| 518 | u32 old_read; |
| 519 | |
| 520 | if (buflen <= 0) |
| 521 | return -EINVAL; |
| 522 | |
| 523 | spin_lock_irqsave(&inring_info->ring_lock, flags); |
| 524 | |
| 525 | hv_get_ringbuffer_availbytes(inring_info, |
| 526 | &bytes_avail_toread, |
| 527 | &bytes_avail_towrite); |
| 528 | |
| 529 | old_read = bytes_avail_toread; |
| 530 | |
| 531 | /* Make sure there is something to read */ |
| 532 | if (bytes_avail_toread < buflen) { |
| 533 | spin_unlock_irqrestore(&inring_info->ring_lock, flags); |
| 534 | |
| 535 | return -EAGAIN; |
| 536 | } |
| 537 | |
| 538 | next_read_location = |
| 539 | hv_get_next_readlocation_withoffset(inring_info, offset); |
| 540 | |
| 541 | next_read_location = hv_copyfrom_ringbuffer(inring_info, |
| 542 | buffer, |
| 543 | buflen, |
| 544 | next_read_location); |
| 545 | |
| 546 | next_read_location = hv_copyfrom_ringbuffer(inring_info, |
| 547 | &prev_indices, |
| 548 | sizeof(u64), |
| 549 | next_read_location); |
| 550 | |
| 551 | /* Make sure all reads are done before we update the read index since */ |
| 552 | /* the writer may start writing to the read area once the read index */ |
| 553 | /*is updated */ |
| 554 | mb(); |
| 555 | |
| 556 | /* Update the read index */ |
| 557 | hv_set_next_read_location(inring_info, next_read_location); |
| 558 | |
| 559 | spin_unlock_irqrestore(&inring_info->ring_lock, flags); |
| 560 | |
| 561 | *signal = hv_need_to_signal_on_read(old_read, inring_info); |
| 562 | |
| 563 | return 0; |
| 564 | } |