1 /* ldc.c: Logical Domain Channel link-layer protocol driver. 2 * 3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net> 4 */ 5 6 #include <linux/kernel.h> 7 #include <linux/export.h> 8 #include <linux/slab.h> 9 #include <linux/spinlock.h> 10 #include <linux/delay.h> 11 #include <linux/errno.h> 12 #include <linux/string.h> 13 #include <linux/scatterlist.h> 14 #include <linux/interrupt.h> 15 #include <linux/list.h> 16 #include <linux/init.h> 17 #include <linux/bitmap.h> 18 #include <linux/iommu-common.h> 19 20 #include <asm/hypervisor.h> 21 #include <asm/iommu.h> 22 #include <asm/page.h> 23 #include <asm/ldc.h> 24 #include <asm/mdesc.h> 25 26 #define DRV_MODULE_NAME "ldc" 27 #define PFX DRV_MODULE_NAME ": " 28 #define DRV_MODULE_VERSION "1.1" 29 #define DRV_MODULE_RELDATE "July 22, 2008" 30 31 #define COOKIE_PGSZ_CODE 0xf000000000000000ULL 32 #define COOKIE_PGSZ_CODE_SHIFT 60ULL 33 34 35 static char version[] = 36 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n"; 37 38 /* Packet header layout for unreliable and reliable mode frames. 39 * When in RAW mode, packets are simply straight 64-byte payloads 40 * with no headers. 41 */ 42 struct ldc_packet { 43 u8 type; 44 #define LDC_CTRL 0x01 45 #define LDC_DATA 0x02 46 #define LDC_ERR 0x10 47 48 u8 stype; 49 #define LDC_INFO 0x01 50 #define LDC_ACK 0x02 51 #define LDC_NACK 0x04 52 53 u8 ctrl; 54 #define LDC_VERS 0x01 /* Link Version */ 55 #define LDC_RTS 0x02 /* Request To Send */ 56 #define LDC_RTR 0x03 /* Ready To Receive */ 57 #define LDC_RDX 0x04 /* Ready for Data eXchange */ 58 #define LDC_CTRL_MSK 0x0f 59 60 u8 env; 61 #define LDC_LEN 0x3f 62 #define LDC_FRAG_MASK 0xc0 63 #define LDC_START 0x40 64 #define LDC_STOP 0x80 65 66 u32 seqid; 67 68 union { 69 u8 u_data[LDC_PACKET_SIZE - 8]; 70 struct { 71 u32 pad; 72 u32 ackid; 73 u8 r_data[LDC_PACKET_SIZE - 8 - 8]; 74 } r; 75 } u; 76 }; 77 78 struct ldc_version { 79 u16 major; 80 u16 minor; 81 }; 82 83 /* Ordered from largest major to lowest. */ 84 static struct ldc_version ver_arr[] = { 85 { .major = 1, .minor = 0 }, 86 }; 87 88 #define LDC_DEFAULT_MTU (4 * LDC_PACKET_SIZE) 89 #define LDC_DEFAULT_NUM_ENTRIES (PAGE_SIZE / LDC_PACKET_SIZE) 90 91 struct ldc_channel; 92 93 struct ldc_mode_ops { 94 int (*write)(struct ldc_channel *, const void *, unsigned int); 95 int (*read)(struct ldc_channel *, void *, unsigned int); 96 }; 97 98 static const struct ldc_mode_ops raw_ops; 99 static const struct ldc_mode_ops nonraw_ops; 100 static const struct ldc_mode_ops stream_ops; 101 102 int ldom_domaining_enabled; 103 104 struct ldc_iommu { 105 /* Protects ldc_unmap. */ 106 spinlock_t lock; 107 struct ldc_mtable_entry *page_table; 108 struct iommu_map_table iommu_map_table; 109 }; 110 111 struct ldc_channel { 112 /* Protects all operations that depend upon channel state. */ 113 spinlock_t lock; 114 115 unsigned long id; 116 117 u8 *mssbuf; 118 u32 mssbuf_len; 119 u32 mssbuf_off; 120 121 struct ldc_packet *tx_base; 122 unsigned long tx_head; 123 unsigned long tx_tail; 124 unsigned long tx_num_entries; 125 unsigned long tx_ra; 126 127 unsigned long tx_acked; 128 129 struct ldc_packet *rx_base; 130 unsigned long rx_head; 131 unsigned long rx_tail; 132 unsigned long rx_num_entries; 133 unsigned long rx_ra; 134 135 u32 rcv_nxt; 136 u32 snd_nxt; 137 138 unsigned long chan_state; 139 140 struct ldc_channel_config cfg; 141 void *event_arg; 142 143 const struct ldc_mode_ops *mops; 144 145 struct ldc_iommu iommu; 146 147 struct ldc_version ver; 148 149 u8 hs_state; 150 #define LDC_HS_CLOSED 0x00 151 #define LDC_HS_OPEN 0x01 152 #define LDC_HS_GOTVERS 0x02 153 #define LDC_HS_SENTRTR 0x03 154 #define LDC_HS_GOTRTR 0x04 155 #define LDC_HS_COMPLETE 0x10 156 157 u8 flags; 158 #define LDC_FLAG_ALLOCED_QUEUES 0x01 159 #define LDC_FLAG_REGISTERED_QUEUES 0x02 160 #define LDC_FLAG_REGISTERED_IRQS 0x04 161 #define LDC_FLAG_RESET 0x10 162 163 u8 mss; 164 u8 state; 165 166 #define LDC_IRQ_NAME_MAX 32 167 char rx_irq_name[LDC_IRQ_NAME_MAX]; 168 char tx_irq_name[LDC_IRQ_NAME_MAX]; 169 170 struct hlist_head mh_list; 171 172 struct hlist_node list; 173 }; 174 175 #define ldcdbg(TYPE, f, a...) \ 176 do { if (lp->cfg.debug & LDC_DEBUG_##TYPE) \ 177 printk(KERN_INFO PFX "ID[%lu] " f, lp->id, ## a); \ 178 } while (0) 179 180 #define LDC_ABORT(lp) ldc_abort((lp), __func__) 181 182 static const char *state_to_str(u8 state) 183 { 184 switch (state) { 185 case LDC_STATE_INVALID: 186 return "INVALID"; 187 case LDC_STATE_INIT: 188 return "INIT"; 189 case LDC_STATE_BOUND: 190 return "BOUND"; 191 case LDC_STATE_READY: 192 return "READY"; 193 case LDC_STATE_CONNECTED: 194 return "CONNECTED"; 195 default: 196 return "<UNKNOWN>"; 197 } 198 } 199 200 static unsigned long __advance(unsigned long off, unsigned long num_entries) 201 { 202 off += LDC_PACKET_SIZE; 203 if (off == (num_entries * LDC_PACKET_SIZE)) 204 off = 0; 205 206 return off; 207 } 208 209 static unsigned long rx_advance(struct ldc_channel *lp, unsigned long off) 210 { 211 return __advance(off, lp->rx_num_entries); 212 } 213 214 static unsigned long tx_advance(struct ldc_channel *lp, unsigned long off) 215 { 216 return __advance(off, lp->tx_num_entries); 217 } 218 219 static struct ldc_packet *handshake_get_tx_packet(struct ldc_channel *lp, 220 unsigned long *new_tail) 221 { 222 struct ldc_packet *p; 223 unsigned long t; 224 225 t = tx_advance(lp, lp->tx_tail); 226 if (t == lp->tx_head) 227 return NULL; 228 229 *new_tail = t; 230 231 p = lp->tx_base; 232 return p + (lp->tx_tail / LDC_PACKET_SIZE); 233 } 234 235 /* When we are in reliable or stream mode, have to track the next packet 236 * we haven't gotten an ACK for in the TX queue using tx_acked. We have 237 * to be careful not to stomp over the queue past that point. During 238 * the handshake, we don't have TX data packets pending in the queue 239 * and that's why handshake_get_tx_packet() need not be mindful of 240 * lp->tx_acked. 241 */ 242 static unsigned long head_for_data(struct ldc_channel *lp) 243 { 244 if (lp->cfg.mode == LDC_MODE_STREAM) 245 return lp->tx_acked; 246 return lp->tx_head; 247 } 248 249 static int tx_has_space_for(struct ldc_channel *lp, unsigned int size) 250 { 251 unsigned long limit, tail, new_tail, diff; 252 unsigned int mss; 253 254 limit = head_for_data(lp); 255 tail = lp->tx_tail; 256 new_tail = tx_advance(lp, tail); 257 if (new_tail == limit) 258 return 0; 259 260 if (limit > new_tail) 261 diff = limit - new_tail; 262 else 263 diff = (limit + 264 ((lp->tx_num_entries * LDC_PACKET_SIZE) - new_tail)); 265 diff /= LDC_PACKET_SIZE; 266 mss = lp->mss; 267 268 if (diff * mss < size) 269 return 0; 270 271 return 1; 272 } 273 274 static struct ldc_packet *data_get_tx_packet(struct ldc_channel *lp, 275 unsigned long *new_tail) 276 { 277 struct ldc_packet *p; 278 unsigned long h, t; 279 280 h = head_for_data(lp); 281 t = tx_advance(lp, lp->tx_tail); 282 if (t == h) 283 return NULL; 284 285 *new_tail = t; 286 287 p = lp->tx_base; 288 return p + (lp->tx_tail / LDC_PACKET_SIZE); 289 } 290 291 static int set_tx_tail(struct ldc_channel *lp, unsigned long tail) 292 { 293 unsigned long orig_tail = lp->tx_tail; 294 int limit = 1000; 295 296 lp->tx_tail = tail; 297 while (limit-- > 0) { 298 unsigned long err; 299 300 err = sun4v_ldc_tx_set_qtail(lp->id, tail); 301 if (!err) 302 return 0; 303 304 if (err != HV_EWOULDBLOCK) { 305 lp->tx_tail = orig_tail; 306 return -EINVAL; 307 } 308 udelay(1); 309 } 310 311 lp->tx_tail = orig_tail; 312 return -EBUSY; 313 } 314 315 /* This just updates the head value in the hypervisor using 316 * a polling loop with a timeout. The caller takes care of 317 * upating software state representing the head change, if any. 318 */ 319 static int __set_rx_head(struct ldc_channel *lp, unsigned long head) 320 { 321 int limit = 1000; 322 323 while (limit-- > 0) { 324 unsigned long err; 325 326 err = sun4v_ldc_rx_set_qhead(lp->id, head); 327 if (!err) 328 return 0; 329 330 if (err != HV_EWOULDBLOCK) 331 return -EINVAL; 332 333 udelay(1); 334 } 335 336 return -EBUSY; 337 } 338 339 static int send_tx_packet(struct ldc_channel *lp, 340 struct ldc_packet *p, 341 unsigned long new_tail) 342 { 343 BUG_ON(p != (lp->tx_base + (lp->tx_tail / LDC_PACKET_SIZE))); 344 345 return set_tx_tail(lp, new_tail); 346 } 347 348 static struct ldc_packet *handshake_compose_ctrl(struct ldc_channel *lp, 349 u8 stype, u8 ctrl, 350 void *data, int dlen, 351 unsigned long *new_tail) 352 { 353 struct ldc_packet *p = handshake_get_tx_packet(lp, new_tail); 354 355 if (p) { 356 memset(p, 0, sizeof(*p)); 357 p->type = LDC_CTRL; 358 p->stype = stype; 359 p->ctrl = ctrl; 360 if (data) 361 memcpy(p->u.u_data, data, dlen); 362 } 363 return p; 364 } 365 366 static int start_handshake(struct ldc_channel *lp) 367 { 368 struct ldc_packet *p; 369 struct ldc_version *ver; 370 unsigned long new_tail; 371 372 ver = &ver_arr[0]; 373 374 ldcdbg(HS, "SEND VER INFO maj[%u] min[%u]\n", 375 ver->major, ver->minor); 376 377 p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS, 378 ver, sizeof(*ver), &new_tail); 379 if (p) { 380 int err = send_tx_packet(lp, p, new_tail); 381 if (!err) 382 lp->flags &= ~LDC_FLAG_RESET; 383 return err; 384 } 385 return -EBUSY; 386 } 387 388 static int send_version_nack(struct ldc_channel *lp, 389 u16 major, u16 minor) 390 { 391 struct ldc_packet *p; 392 struct ldc_version ver; 393 unsigned long new_tail; 394 395 ver.major = major; 396 ver.minor = minor; 397 398 p = handshake_compose_ctrl(lp, LDC_NACK, LDC_VERS, 399 &ver, sizeof(ver), &new_tail); 400 if (p) { 401 ldcdbg(HS, "SEND VER NACK maj[%u] min[%u]\n", 402 ver.major, ver.minor); 403 404 return send_tx_packet(lp, p, new_tail); 405 } 406 return -EBUSY; 407 } 408 409 static int send_version_ack(struct ldc_channel *lp, 410 struct ldc_version *vp) 411 { 412 struct ldc_packet *p; 413 unsigned long new_tail; 414 415 p = handshake_compose_ctrl(lp, LDC_ACK, LDC_VERS, 416 vp, sizeof(*vp), &new_tail); 417 if (p) { 418 ldcdbg(HS, "SEND VER ACK maj[%u] min[%u]\n", 419 vp->major, vp->minor); 420 421 return send_tx_packet(lp, p, new_tail); 422 } 423 return -EBUSY; 424 } 425 426 static int send_rts(struct ldc_channel *lp) 427 { 428 struct ldc_packet *p; 429 unsigned long new_tail; 430 431 p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTS, NULL, 0, 432 &new_tail); 433 if (p) { 434 p->env = lp->cfg.mode; 435 p->seqid = 0; 436 lp->rcv_nxt = 0; 437 438 ldcdbg(HS, "SEND RTS env[0x%x] seqid[0x%x]\n", 439 p->env, p->seqid); 440 441 return send_tx_packet(lp, p, new_tail); 442 } 443 return -EBUSY; 444 } 445 446 static int send_rtr(struct ldc_channel *lp) 447 { 448 struct ldc_packet *p; 449 unsigned long new_tail; 450 451 p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTR, NULL, 0, 452 &new_tail); 453 if (p) { 454 p->env = lp->cfg.mode; 455 p->seqid = 0; 456 457 ldcdbg(HS, "SEND RTR env[0x%x] seqid[0x%x]\n", 458 p->env, p->seqid); 459 460 return send_tx_packet(lp, p, new_tail); 461 } 462 return -EBUSY; 463 } 464 465 static int send_rdx(struct ldc_channel *lp) 466 { 467 struct ldc_packet *p; 468 unsigned long new_tail; 469 470 p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RDX, NULL, 0, 471 &new_tail); 472 if (p) { 473 p->env = 0; 474 p->seqid = ++lp->snd_nxt; 475 p->u.r.ackid = lp->rcv_nxt; 476 477 ldcdbg(HS, "SEND RDX env[0x%x] seqid[0x%x] ackid[0x%x]\n", 478 p->env, p->seqid, p->u.r.ackid); 479 480 return send_tx_packet(lp, p, new_tail); 481 } 482 return -EBUSY; 483 } 484 485 static int send_data_nack(struct ldc_channel *lp, struct ldc_packet *data_pkt) 486 { 487 struct ldc_packet *p; 488 unsigned long new_tail; 489 int err; 490 491 p = data_get_tx_packet(lp, &new_tail); 492 if (!p) 493 return -EBUSY; 494 memset(p, 0, sizeof(*p)); 495 p->type = data_pkt->type; 496 p->stype = LDC_NACK; 497 p->ctrl = data_pkt->ctrl & LDC_CTRL_MSK; 498 p->seqid = lp->snd_nxt + 1; 499 p->u.r.ackid = lp->rcv_nxt; 500 501 ldcdbg(HS, "SEND DATA NACK type[0x%x] ctl[0x%x] seq[0x%x] ack[0x%x]\n", 502 p->type, p->ctrl, p->seqid, p->u.r.ackid); 503 504 err = send_tx_packet(lp, p, new_tail); 505 if (!err) 506 lp->snd_nxt++; 507 508 return err; 509 } 510 511 static int ldc_abort(struct ldc_channel *lp, const char *msg) 512 { 513 unsigned long hv_err; 514 515 ldcdbg(STATE, "ABORT[%s]\n", msg); 516 ldc_print(lp); 517 518 /* We report but do not act upon the hypervisor errors because 519 * there really isn't much we can do if they fail at this point. 520 */ 521 hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries); 522 if (hv_err) 523 printk(KERN_ERR PFX "ldc_abort: " 524 "sun4v_ldc_tx_qconf(%lx,%lx,%lx) failed, err=%lu\n", 525 lp->id, lp->tx_ra, lp->tx_num_entries, hv_err); 526 527 hv_err = sun4v_ldc_tx_get_state(lp->id, 528 &lp->tx_head, 529 &lp->tx_tail, 530 &lp->chan_state); 531 if (hv_err) 532 printk(KERN_ERR PFX "ldc_abort: " 533 "sun4v_ldc_tx_get_state(%lx,...) failed, err=%lu\n", 534 lp->id, hv_err); 535 536 hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries); 537 if (hv_err) 538 printk(KERN_ERR PFX "ldc_abort: " 539 "sun4v_ldc_rx_qconf(%lx,%lx,%lx) failed, err=%lu\n", 540 lp->id, lp->rx_ra, lp->rx_num_entries, hv_err); 541 542 /* Refetch the RX queue state as well, because we could be invoked 543 * here in the queue processing context. 544 */ 545 hv_err = sun4v_ldc_rx_get_state(lp->id, 546 &lp->rx_head, 547 &lp->rx_tail, 548 &lp->chan_state); 549 if (hv_err) 550 printk(KERN_ERR PFX "ldc_abort: " 551 "sun4v_ldc_rx_get_state(%lx,...) failed, err=%lu\n", 552 lp->id, hv_err); 553 554 return -ECONNRESET; 555 } 556 557 static struct ldc_version *find_by_major(u16 major) 558 { 559 struct ldc_version *ret = NULL; 560 int i; 561 562 for (i = 0; i < ARRAY_SIZE(ver_arr); i++) { 563 struct ldc_version *v = &ver_arr[i]; 564 if (v->major <= major) { 565 ret = v; 566 break; 567 } 568 } 569 return ret; 570 } 571 572 static int process_ver_info(struct ldc_channel *lp, struct ldc_version *vp) 573 { 574 struct ldc_version *vap; 575 int err; 576 577 ldcdbg(HS, "GOT VERSION INFO major[%x] minor[%x]\n", 578 vp->major, vp->minor); 579 580 if (lp->hs_state == LDC_HS_GOTVERS) { 581 lp->hs_state = LDC_HS_OPEN; 582 memset(&lp->ver, 0, sizeof(lp->ver)); 583 } 584 585 vap = find_by_major(vp->major); 586 if (!vap) { 587 err = send_version_nack(lp, 0, 0); 588 } else if (vap->major != vp->major) { 589 err = send_version_nack(lp, vap->major, vap->minor); 590 } else { 591 struct ldc_version ver = *vp; 592 if (ver.minor > vap->minor) 593 ver.minor = vap->minor; 594 err = send_version_ack(lp, &ver); 595 if (!err) { 596 lp->ver = ver; 597 lp->hs_state = LDC_HS_GOTVERS; 598 } 599 } 600 if (err) 601 return LDC_ABORT(lp); 602 603 return 0; 604 } 605 606 static int process_ver_ack(struct ldc_channel *lp, struct ldc_version *vp) 607 { 608 ldcdbg(HS, "GOT VERSION ACK major[%x] minor[%x]\n", 609 vp->major, vp->minor); 610 611 if (lp->hs_state == LDC_HS_GOTVERS) { 612 if (lp->ver.major != vp->major || 613 lp->ver.minor != vp->minor) 614 return LDC_ABORT(lp); 615 } else { 616 lp->ver = *vp; 617 lp->hs_state = LDC_HS_GOTVERS; 618 } 619 if (send_rts(lp)) 620 return LDC_ABORT(lp); 621 return 0; 622 } 623 624 static int process_ver_nack(struct ldc_channel *lp, struct ldc_version *vp) 625 { 626 struct ldc_version *vap; 627 struct ldc_packet *p; 628 unsigned long new_tail; 629 630 if (vp->major == 0 && vp->minor == 0) 631 return LDC_ABORT(lp); 632 633 vap = find_by_major(vp->major); 634 if (!vap) 635 return LDC_ABORT(lp); 636 637 p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS, 638 vap, sizeof(*vap), 639 &new_tail); 640 if (!p) 641 return LDC_ABORT(lp); 642 643 return send_tx_packet(lp, p, new_tail); 644 } 645 646 static int process_version(struct ldc_channel *lp, 647 struct ldc_packet *p) 648 { 649 struct ldc_version *vp; 650 651 vp = (struct ldc_version *) p->u.u_data; 652 653 switch (p->stype) { 654 case LDC_INFO: 655 return process_ver_info(lp, vp); 656 657 case LDC_ACK: 658 return process_ver_ack(lp, vp); 659 660 case LDC_NACK: 661 return process_ver_nack(lp, vp); 662 663 default: 664 return LDC_ABORT(lp); 665 } 666 } 667 668 static int process_rts(struct ldc_channel *lp, 669 struct ldc_packet *p) 670 { 671 ldcdbg(HS, "GOT RTS stype[%x] seqid[%x] env[%x]\n", 672 p->stype, p->seqid, p->env); 673 674 if (p->stype != LDC_INFO || 675 lp->hs_state != LDC_HS_GOTVERS || 676 p->env != lp->cfg.mode) 677 return LDC_ABORT(lp); 678 679 lp->snd_nxt = p->seqid; 680 lp->rcv_nxt = p->seqid; 681 lp->hs_state = LDC_HS_SENTRTR; 682 if (send_rtr(lp)) 683 return LDC_ABORT(lp); 684 685 return 0; 686 } 687 688 static int process_rtr(struct ldc_channel *lp, 689 struct ldc_packet *p) 690 { 691 ldcdbg(HS, "GOT RTR stype[%x] seqid[%x] env[%x]\n", 692 p->stype, p->seqid, p->env); 693 694 if (p->stype != LDC_INFO || 695 p->env != lp->cfg.mode) 696 return LDC_ABORT(lp); 697 698 lp->snd_nxt = p->seqid; 699 lp->hs_state = LDC_HS_COMPLETE; 700 ldc_set_state(lp, LDC_STATE_CONNECTED); 701 send_rdx(lp); 702 703 return LDC_EVENT_UP; 704 } 705 706 static int rx_seq_ok(struct ldc_channel *lp, u32 seqid) 707 { 708 return lp->rcv_nxt + 1 == seqid; 709 } 710 711 static int process_rdx(struct ldc_channel *lp, 712 struct ldc_packet *p) 713 { 714 ldcdbg(HS, "GOT RDX stype[%x] seqid[%x] env[%x] ackid[%x]\n", 715 p->stype, p->seqid, p->env, p->u.r.ackid); 716 717 if (p->stype != LDC_INFO || 718 !(rx_seq_ok(lp, p->seqid))) 719 return LDC_ABORT(lp); 720 721 lp->rcv_nxt = p->seqid; 722 723 lp->hs_state = LDC_HS_COMPLETE; 724 ldc_set_state(lp, LDC_STATE_CONNECTED); 725 726 return LDC_EVENT_UP; 727 } 728 729 static int process_control_frame(struct ldc_channel *lp, 730 struct ldc_packet *p) 731 { 732 switch (p->ctrl) { 733 case LDC_VERS: 734 return process_version(lp, p); 735 736 case LDC_RTS: 737 return process_rts(lp, p); 738 739 case LDC_RTR: 740 return process_rtr(lp, p); 741 742 case LDC_RDX: 743 return process_rdx(lp, p); 744 745 default: 746 return LDC_ABORT(lp); 747 } 748 } 749 750 static int process_error_frame(struct ldc_channel *lp, 751 struct ldc_packet *p) 752 { 753 return LDC_ABORT(lp); 754 } 755 756 static int process_data_ack(struct ldc_channel *lp, 757 struct ldc_packet *ack) 758 { 759 unsigned long head = lp->tx_acked; 760 u32 ackid = ack->u.r.ackid; 761 762 while (1) { 763 struct ldc_packet *p = lp->tx_base + (head / LDC_PACKET_SIZE); 764 765 head = tx_advance(lp, head); 766 767 if (p->seqid == ackid) { 768 lp->tx_acked = head; 769 return 0; 770 } 771 if (head == lp->tx_tail) 772 return LDC_ABORT(lp); 773 } 774 775 return 0; 776 } 777 778 static void send_events(struct ldc_channel *lp, unsigned int event_mask) 779 { 780 if (event_mask & LDC_EVENT_RESET) 781 lp->cfg.event(lp->event_arg, LDC_EVENT_RESET); 782 if (event_mask & LDC_EVENT_UP) 783 lp->cfg.event(lp->event_arg, LDC_EVENT_UP); 784 if (event_mask & LDC_EVENT_DATA_READY) 785 lp->cfg.event(lp->event_arg, LDC_EVENT_DATA_READY); 786 } 787 788 static irqreturn_t ldc_rx(int irq, void *dev_id) 789 { 790 struct ldc_channel *lp = dev_id; 791 unsigned long orig_state, flags; 792 unsigned int event_mask; 793 794 spin_lock_irqsave(&lp->lock, flags); 795 796 orig_state = lp->chan_state; 797 798 /* We should probably check for hypervisor errors here and 799 * reset the LDC channel if we get one. 800 */ 801 sun4v_ldc_rx_get_state(lp->id, 802 &lp->rx_head, 803 &lp->rx_tail, 804 &lp->chan_state); 805 806 ldcdbg(RX, "RX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n", 807 orig_state, lp->chan_state, lp->rx_head, lp->rx_tail); 808 809 event_mask = 0; 810 811 if (lp->cfg.mode == LDC_MODE_RAW && 812 lp->chan_state == LDC_CHANNEL_UP) { 813 lp->hs_state = LDC_HS_COMPLETE; 814 ldc_set_state(lp, LDC_STATE_CONNECTED); 815 816 /* 817 * Generate an LDC_EVENT_UP event if the channel 818 * was not already up. 819 */ 820 if (orig_state != LDC_CHANNEL_UP) { 821 event_mask |= LDC_EVENT_UP; 822 orig_state = lp->chan_state; 823 } 824 } 825 826 /* If we are in reset state, flush the RX queue and ignore 827 * everything. 828 */ 829 if (lp->flags & LDC_FLAG_RESET) { 830 (void) ldc_rx_reset(lp); 831 goto out; 832 } 833 834 /* Once we finish the handshake, we let the ldc_read() 835 * paths do all of the control frame and state management. 836 * Just trigger the callback. 837 */ 838 if (lp->hs_state == LDC_HS_COMPLETE) { 839 handshake_complete: 840 if (lp->chan_state != orig_state) { 841 unsigned int event = LDC_EVENT_RESET; 842 843 if (lp->chan_state == LDC_CHANNEL_UP) 844 event = LDC_EVENT_UP; 845 846 event_mask |= event; 847 } 848 if (lp->rx_head != lp->rx_tail) 849 event_mask |= LDC_EVENT_DATA_READY; 850 851 goto out; 852 } 853 854 if (lp->chan_state != orig_state) 855 goto out; 856 857 while (lp->rx_head != lp->rx_tail) { 858 struct ldc_packet *p; 859 unsigned long new; 860 int err; 861 862 p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE); 863 864 switch (p->type) { 865 case LDC_CTRL: 866 err = process_control_frame(lp, p); 867 if (err > 0) 868 event_mask |= err; 869 break; 870 871 case LDC_DATA: 872 event_mask |= LDC_EVENT_DATA_READY; 873 err = 0; 874 break; 875 876 case LDC_ERR: 877 err = process_error_frame(lp, p); 878 break; 879 880 default: 881 err = LDC_ABORT(lp); 882 break; 883 } 884 885 if (err < 0) 886 break; 887 888 new = lp->rx_head; 889 new += LDC_PACKET_SIZE; 890 if (new == (lp->rx_num_entries * LDC_PACKET_SIZE)) 891 new = 0; 892 lp->rx_head = new; 893 894 err = __set_rx_head(lp, new); 895 if (err < 0) { 896 (void) LDC_ABORT(lp); 897 break; 898 } 899 if (lp->hs_state == LDC_HS_COMPLETE) 900 goto handshake_complete; 901 } 902 903 out: 904 spin_unlock_irqrestore(&lp->lock, flags); 905 906 send_events(lp, event_mask); 907 908 return IRQ_HANDLED; 909 } 910 911 static irqreturn_t ldc_tx(int irq, void *dev_id) 912 { 913 struct ldc_channel *lp = dev_id; 914 unsigned long flags, orig_state; 915 unsigned int event_mask = 0; 916 917 spin_lock_irqsave(&lp->lock, flags); 918 919 orig_state = lp->chan_state; 920 921 /* We should probably check for hypervisor errors here and 922 * reset the LDC channel if we get one. 923 */ 924 sun4v_ldc_tx_get_state(lp->id, 925 &lp->tx_head, 926 &lp->tx_tail, 927 &lp->chan_state); 928 929 ldcdbg(TX, " TX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n", 930 orig_state, lp->chan_state, lp->tx_head, lp->tx_tail); 931 932 if (lp->cfg.mode == LDC_MODE_RAW && 933 lp->chan_state == LDC_CHANNEL_UP) { 934 lp->hs_state = LDC_HS_COMPLETE; 935 ldc_set_state(lp, LDC_STATE_CONNECTED); 936 937 /* 938 * Generate an LDC_EVENT_UP event if the channel 939 * was not already up. 940 */ 941 if (orig_state != LDC_CHANNEL_UP) { 942 event_mask |= LDC_EVENT_UP; 943 orig_state = lp->chan_state; 944 } 945 } 946 947 spin_unlock_irqrestore(&lp->lock, flags); 948 949 send_events(lp, event_mask); 950 951 return IRQ_HANDLED; 952 } 953 954 /* XXX ldc_alloc() and ldc_free() needs to run under a mutex so 955 * XXX that addition and removal from the ldc_channel_list has 956 * XXX atomicity, otherwise the __ldc_channel_exists() check is 957 * XXX totally pointless as another thread can slip into ldc_alloc() 958 * XXX and add a channel with the same ID. There also needs to be 959 * XXX a spinlock for ldc_channel_list. 960 */ 961 static HLIST_HEAD(ldc_channel_list); 962 963 static int __ldc_channel_exists(unsigned long id) 964 { 965 struct ldc_channel *lp; 966 967 hlist_for_each_entry(lp, &ldc_channel_list, list) { 968 if (lp->id == id) 969 return 1; 970 } 971 return 0; 972 } 973 974 static int alloc_queue(const char *name, unsigned long num_entries, 975 struct ldc_packet **base, unsigned long *ra) 976 { 977 unsigned long size, order; 978 void *q; 979 980 size = num_entries * LDC_PACKET_SIZE; 981 order = get_order(size); 982 983 q = (void *) __get_free_pages(GFP_KERNEL, order); 984 if (!q) { 985 printk(KERN_ERR PFX "Alloc of %s queue failed with " 986 "size=%lu order=%lu\n", name, size, order); 987 return -ENOMEM; 988 } 989 990 memset(q, 0, PAGE_SIZE << order); 991 992 *base = q; 993 *ra = __pa(q); 994 995 return 0; 996 } 997 998 static void free_queue(unsigned long num_entries, struct ldc_packet *q) 999 { 1000 unsigned long size, order; 1001 1002 if (!q) 1003 return; 1004 1005 size = num_entries * LDC_PACKET_SIZE; 1006 order = get_order(size); 1007 1008 free_pages((unsigned long)q, order); 1009 } 1010 1011 static unsigned long ldc_cookie_to_index(u64 cookie, void *arg) 1012 { 1013 u64 szcode = cookie >> COOKIE_PGSZ_CODE_SHIFT; 1014 /* struct ldc_iommu *ldc_iommu = (struct ldc_iommu *)arg; */ 1015 1016 cookie &= ~COOKIE_PGSZ_CODE; 1017 1018 return (cookie >> (13ULL + (szcode * 3ULL))); 1019 } 1020 1021 static void ldc_demap(struct ldc_iommu *iommu, unsigned long id, u64 cookie, 1022 unsigned long entry, unsigned long npages) 1023 { 1024 struct ldc_mtable_entry *base; 1025 unsigned long i, shift; 1026 1027 shift = (cookie >> COOKIE_PGSZ_CODE_SHIFT) * 3; 1028 base = iommu->page_table + entry; 1029 for (i = 0; i < npages; i++) { 1030 if (base->cookie) 1031 sun4v_ldc_revoke(id, cookie + (i << shift), 1032 base->cookie); 1033 base->mte = 0; 1034 } 1035 } 1036 1037 /* XXX Make this configurable... XXX */ 1038 #define LDC_IOTABLE_SIZE (8 * 1024) 1039 1040 static int ldc_iommu_init(const char *name, struct ldc_channel *lp) 1041 { 1042 unsigned long sz, num_tsb_entries, tsbsize, order; 1043 struct ldc_iommu *ldc_iommu = &lp->iommu; 1044 struct iommu_map_table *iommu = &ldc_iommu->iommu_map_table; 1045 struct ldc_mtable_entry *table; 1046 unsigned long hv_err; 1047 int err; 1048 1049 num_tsb_entries = LDC_IOTABLE_SIZE; 1050 tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry); 1051 spin_lock_init(&ldc_iommu->lock); 1052 1053 sz = num_tsb_entries / 8; 1054 sz = (sz + 7UL) & ~7UL; 1055 iommu->map = kzalloc(sz, GFP_KERNEL); 1056 if (!iommu->map) { 1057 printk(KERN_ERR PFX "Alloc of arena map failed, sz=%lu\n", sz); 1058 return -ENOMEM; 1059 } 1060 iommu_tbl_pool_init(iommu, num_tsb_entries, PAGE_SHIFT, 1061 NULL, false /* no large pool */, 1062 1 /* npools */, 1063 true /* skip span boundary check */); 1064 1065 order = get_order(tsbsize); 1066 1067 table = (struct ldc_mtable_entry *) 1068 __get_free_pages(GFP_KERNEL, order); 1069 err = -ENOMEM; 1070 if (!table) { 1071 printk(KERN_ERR PFX "Alloc of MTE table failed, " 1072 "size=%lu order=%lu\n", tsbsize, order); 1073 goto out_free_map; 1074 } 1075 1076 memset(table, 0, PAGE_SIZE << order); 1077 1078 ldc_iommu->page_table = table; 1079 1080 hv_err = sun4v_ldc_set_map_table(lp->id, __pa(table), 1081 num_tsb_entries); 1082 err = -EINVAL; 1083 if (hv_err) 1084 goto out_free_table; 1085 1086 return 0; 1087 1088 out_free_table: 1089 free_pages((unsigned long) table, order); 1090 ldc_iommu->page_table = NULL; 1091 1092 out_free_map: 1093 kfree(iommu->map); 1094 iommu->map = NULL; 1095 1096 return err; 1097 } 1098 1099 static void ldc_iommu_release(struct ldc_channel *lp) 1100 { 1101 struct ldc_iommu *ldc_iommu = &lp->iommu; 1102 struct iommu_map_table *iommu = &ldc_iommu->iommu_map_table; 1103 unsigned long num_tsb_entries, tsbsize, order; 1104 1105 (void) sun4v_ldc_set_map_table(lp->id, 0, 0); 1106 1107 num_tsb_entries = iommu->poolsize * iommu->nr_pools; 1108 tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry); 1109 order = get_order(tsbsize); 1110 1111 free_pages((unsigned long) ldc_iommu->page_table, order); 1112 ldc_iommu->page_table = NULL; 1113 1114 kfree(iommu->map); 1115 iommu->map = NULL; 1116 } 1117 1118 struct ldc_channel *ldc_alloc(unsigned long id, 1119 const struct ldc_channel_config *cfgp, 1120 void *event_arg, 1121 const char *name) 1122 { 1123 struct ldc_channel *lp; 1124 const struct ldc_mode_ops *mops; 1125 unsigned long dummy1, dummy2, hv_err; 1126 u8 mss, *mssbuf; 1127 int err; 1128 1129 err = -ENODEV; 1130 if (!ldom_domaining_enabled) 1131 goto out_err; 1132 1133 err = -EINVAL; 1134 if (!cfgp) 1135 goto out_err; 1136 if (!name) 1137 goto out_err; 1138 1139 switch (cfgp->mode) { 1140 case LDC_MODE_RAW: 1141 mops = &raw_ops; 1142 mss = LDC_PACKET_SIZE; 1143 break; 1144 1145 case LDC_MODE_UNRELIABLE: 1146 mops = &nonraw_ops; 1147 mss = LDC_PACKET_SIZE - 8; 1148 break; 1149 1150 case LDC_MODE_STREAM: 1151 mops = &stream_ops; 1152 mss = LDC_PACKET_SIZE - 8 - 8; 1153 break; 1154 1155 default: 1156 goto out_err; 1157 } 1158 1159 if (!cfgp->event || !event_arg || !cfgp->rx_irq || !cfgp->tx_irq) 1160 goto out_err; 1161 1162 hv_err = sun4v_ldc_tx_qinfo(id, &dummy1, &dummy2); 1163 err = -ENODEV; 1164 if (hv_err == HV_ECHANNEL) 1165 goto out_err; 1166 1167 err = -EEXIST; 1168 if (__ldc_channel_exists(id)) 1169 goto out_err; 1170 1171 mssbuf = NULL; 1172 1173 lp = kzalloc(sizeof(*lp), GFP_KERNEL); 1174 err = -ENOMEM; 1175 if (!lp) 1176 goto out_err; 1177 1178 spin_lock_init(&lp->lock); 1179 1180 lp->id = id; 1181 1182 err = ldc_iommu_init(name, lp); 1183 if (err) 1184 goto out_free_ldc; 1185 1186 lp->mops = mops; 1187 lp->mss = mss; 1188 1189 lp->cfg = *cfgp; 1190 if (!lp->cfg.mtu) 1191 lp->cfg.mtu = LDC_DEFAULT_MTU; 1192 1193 if (lp->cfg.mode == LDC_MODE_STREAM) { 1194 mssbuf = kzalloc(lp->cfg.mtu, GFP_KERNEL); 1195 if (!mssbuf) { 1196 err = -ENOMEM; 1197 goto out_free_iommu; 1198 } 1199 lp->mssbuf = mssbuf; 1200 } 1201 1202 lp->event_arg = event_arg; 1203 1204 /* XXX allow setting via ldc_channel_config to override defaults 1205 * XXX or use some formula based upon mtu 1206 */ 1207 lp->tx_num_entries = LDC_DEFAULT_NUM_ENTRIES; 1208 lp->rx_num_entries = LDC_DEFAULT_NUM_ENTRIES; 1209 1210 err = alloc_queue("TX", lp->tx_num_entries, 1211 &lp->tx_base, &lp->tx_ra); 1212 if (err) 1213 goto out_free_mssbuf; 1214 1215 err = alloc_queue("RX", lp->rx_num_entries, 1216 &lp->rx_base, &lp->rx_ra); 1217 if (err) 1218 goto out_free_txq; 1219 1220 lp->flags |= LDC_FLAG_ALLOCED_QUEUES; 1221 1222 lp->hs_state = LDC_HS_CLOSED; 1223 ldc_set_state(lp, LDC_STATE_INIT); 1224 1225 INIT_HLIST_NODE(&lp->list); 1226 hlist_add_head(&lp->list, &ldc_channel_list); 1227 1228 INIT_HLIST_HEAD(&lp->mh_list); 1229 1230 snprintf(lp->rx_irq_name, LDC_IRQ_NAME_MAX, "%s RX", name); 1231 snprintf(lp->tx_irq_name, LDC_IRQ_NAME_MAX, "%s TX", name); 1232 1233 err = request_irq(lp->cfg.rx_irq, ldc_rx, 0, 1234 lp->rx_irq_name, lp); 1235 if (err) 1236 goto out_free_txq; 1237 1238 err = request_irq(lp->cfg.tx_irq, ldc_tx, 0, 1239 lp->tx_irq_name, lp); 1240 if (err) { 1241 free_irq(lp->cfg.rx_irq, lp); 1242 goto out_free_txq; 1243 } 1244 1245 return lp; 1246 1247 out_free_txq: 1248 free_queue(lp->tx_num_entries, lp->tx_base); 1249 1250 out_free_mssbuf: 1251 kfree(mssbuf); 1252 1253 out_free_iommu: 1254 ldc_iommu_release(lp); 1255 1256 out_free_ldc: 1257 kfree(lp); 1258 1259 out_err: 1260 return ERR_PTR(err); 1261 } 1262 EXPORT_SYMBOL(ldc_alloc); 1263 1264 void ldc_unbind(struct ldc_channel *lp) 1265 { 1266 if (lp->flags & LDC_FLAG_REGISTERED_IRQS) { 1267 free_irq(lp->cfg.rx_irq, lp); 1268 free_irq(lp->cfg.tx_irq, lp); 1269 lp->flags &= ~LDC_FLAG_REGISTERED_IRQS; 1270 } 1271 1272 if (lp->flags & LDC_FLAG_REGISTERED_QUEUES) { 1273 sun4v_ldc_tx_qconf(lp->id, 0, 0); 1274 sun4v_ldc_rx_qconf(lp->id, 0, 0); 1275 lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES; 1276 } 1277 if (lp->flags & LDC_FLAG_ALLOCED_QUEUES) { 1278 free_queue(lp->tx_num_entries, lp->tx_base); 1279 free_queue(lp->rx_num_entries, lp->rx_base); 1280 lp->flags &= ~LDC_FLAG_ALLOCED_QUEUES; 1281 } 1282 1283 ldc_set_state(lp, LDC_STATE_INIT); 1284 } 1285 EXPORT_SYMBOL(ldc_unbind); 1286 1287 void ldc_free(struct ldc_channel *lp) 1288 { 1289 ldc_unbind(lp); 1290 hlist_del(&lp->list); 1291 kfree(lp->mssbuf); 1292 ldc_iommu_release(lp); 1293 1294 kfree(lp); 1295 } 1296 EXPORT_SYMBOL(ldc_free); 1297 1298 /* Bind the channel. This registers the LDC queues with 1299 * the hypervisor and puts the channel into a pseudo-listening 1300 * state. This does not initiate a handshake, ldc_connect() does 1301 * that. 1302 */ 1303 int ldc_bind(struct ldc_channel *lp) 1304 { 1305 unsigned long hv_err, flags; 1306 int err = -EINVAL; 1307 1308 if (lp->state != LDC_STATE_INIT) 1309 return -EINVAL; 1310 1311 spin_lock_irqsave(&lp->lock, flags); 1312 1313 enable_irq(lp->cfg.rx_irq); 1314 enable_irq(lp->cfg.tx_irq); 1315 1316 lp->flags |= LDC_FLAG_REGISTERED_IRQS; 1317 1318 err = -ENODEV; 1319 hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0); 1320 if (hv_err) 1321 goto out_free_irqs; 1322 1323 hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries); 1324 if (hv_err) 1325 goto out_free_irqs; 1326 1327 hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0); 1328 if (hv_err) 1329 goto out_unmap_tx; 1330 1331 hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries); 1332 if (hv_err) 1333 goto out_unmap_tx; 1334 1335 lp->flags |= LDC_FLAG_REGISTERED_QUEUES; 1336 1337 hv_err = sun4v_ldc_tx_get_state(lp->id, 1338 &lp->tx_head, 1339 &lp->tx_tail, 1340 &lp->chan_state); 1341 err = -EBUSY; 1342 if (hv_err) 1343 goto out_unmap_rx; 1344 1345 lp->tx_acked = lp->tx_head; 1346 1347 lp->hs_state = LDC_HS_OPEN; 1348 ldc_set_state(lp, LDC_STATE_BOUND); 1349 1350 if (lp->cfg.mode == LDC_MODE_RAW) { 1351 /* 1352 * There is no handshake in RAW mode, so handshake 1353 * is completed. 1354 */ 1355 lp->hs_state = LDC_HS_COMPLETE; 1356 } 1357 1358 spin_unlock_irqrestore(&lp->lock, flags); 1359 1360 return 0; 1361 1362 out_unmap_rx: 1363 lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES; 1364 sun4v_ldc_rx_qconf(lp->id, 0, 0); 1365 1366 out_unmap_tx: 1367 sun4v_ldc_tx_qconf(lp->id, 0, 0); 1368 1369 out_free_irqs: 1370 lp->flags &= ~LDC_FLAG_REGISTERED_IRQS; 1371 free_irq(lp->cfg.tx_irq, lp); 1372 free_irq(lp->cfg.rx_irq, lp); 1373 1374 spin_unlock_irqrestore(&lp->lock, flags); 1375 1376 return err; 1377 } 1378 EXPORT_SYMBOL(ldc_bind); 1379 1380 int ldc_connect(struct ldc_channel *lp) 1381 { 1382 unsigned long flags; 1383 int err; 1384 1385 if (lp->cfg.mode == LDC_MODE_RAW) 1386 return -EINVAL; 1387 1388 spin_lock_irqsave(&lp->lock, flags); 1389 1390 if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) || 1391 !(lp->flags & LDC_FLAG_REGISTERED_QUEUES) || 1392 lp->hs_state != LDC_HS_OPEN) 1393 err = ((lp->hs_state > LDC_HS_OPEN) ? 0 : -EINVAL); 1394 else 1395 err = start_handshake(lp); 1396 1397 spin_unlock_irqrestore(&lp->lock, flags); 1398 1399 return err; 1400 } 1401 EXPORT_SYMBOL(ldc_connect); 1402 1403 int ldc_disconnect(struct ldc_channel *lp) 1404 { 1405 unsigned long hv_err, flags; 1406 int err; 1407 1408 if (lp->cfg.mode == LDC_MODE_RAW) 1409 return -EINVAL; 1410 1411 if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) || 1412 !(lp->flags & LDC_FLAG_REGISTERED_QUEUES)) 1413 return -EINVAL; 1414 1415 spin_lock_irqsave(&lp->lock, flags); 1416 1417 err = -ENODEV; 1418 hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0); 1419 if (hv_err) 1420 goto out_err; 1421 1422 hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries); 1423 if (hv_err) 1424 goto out_err; 1425 1426 hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0); 1427 if (hv_err) 1428 goto out_err; 1429 1430 hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries); 1431 if (hv_err) 1432 goto out_err; 1433 1434 ldc_set_state(lp, LDC_STATE_BOUND); 1435 lp->hs_state = LDC_HS_OPEN; 1436 lp->flags |= LDC_FLAG_RESET; 1437 1438 spin_unlock_irqrestore(&lp->lock, flags); 1439 1440 return 0; 1441 1442 out_err: 1443 sun4v_ldc_tx_qconf(lp->id, 0, 0); 1444 sun4v_ldc_rx_qconf(lp->id, 0, 0); 1445 free_irq(lp->cfg.tx_irq, lp); 1446 free_irq(lp->cfg.rx_irq, lp); 1447 lp->flags &= ~(LDC_FLAG_REGISTERED_IRQS | 1448 LDC_FLAG_REGISTERED_QUEUES); 1449 ldc_set_state(lp, LDC_STATE_INIT); 1450 1451 spin_unlock_irqrestore(&lp->lock, flags); 1452 1453 return err; 1454 } 1455 EXPORT_SYMBOL(ldc_disconnect); 1456 1457 int ldc_state(struct ldc_channel *lp) 1458 { 1459 return lp->state; 1460 } 1461 EXPORT_SYMBOL(ldc_state); 1462 1463 void ldc_set_state(struct ldc_channel *lp, u8 state) 1464 { 1465 ldcdbg(STATE, "STATE (%s) --> (%s)\n", 1466 state_to_str(lp->state), 1467 state_to_str(state)); 1468 1469 lp->state = state; 1470 } 1471 EXPORT_SYMBOL(ldc_set_state); 1472 1473 int ldc_mode(struct ldc_channel *lp) 1474 { 1475 return lp->cfg.mode; 1476 } 1477 EXPORT_SYMBOL(ldc_mode); 1478 1479 int ldc_rx_reset(struct ldc_channel *lp) 1480 { 1481 return __set_rx_head(lp, lp->rx_tail); 1482 } 1483 EXPORT_SYMBOL(ldc_rx_reset); 1484 1485 void __ldc_print(struct ldc_channel *lp, const char *caller) 1486 { 1487 pr_info("%s: id=0x%lx flags=0x%x state=%s cstate=0x%lx hsstate=0x%x\n" 1488 "\trx_h=0x%lx rx_t=0x%lx rx_n=%ld\n" 1489 "\ttx_h=0x%lx tx_t=0x%lx tx_n=%ld\n" 1490 "\trcv_nxt=%u snd_nxt=%u\n", 1491 caller, lp->id, lp->flags, state_to_str(lp->state), 1492 lp->chan_state, lp->hs_state, 1493 lp->rx_head, lp->rx_tail, lp->rx_num_entries, 1494 lp->tx_head, lp->tx_tail, lp->tx_num_entries, 1495 lp->rcv_nxt, lp->snd_nxt); 1496 } 1497 EXPORT_SYMBOL(__ldc_print); 1498 1499 static int write_raw(struct ldc_channel *lp, const void *buf, unsigned int size) 1500 { 1501 struct ldc_packet *p; 1502 unsigned long new_tail, hv_err; 1503 int err; 1504 1505 hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail, 1506 &lp->chan_state); 1507 if (unlikely(hv_err)) 1508 return -EBUSY; 1509 1510 if (unlikely(lp->chan_state != LDC_CHANNEL_UP)) 1511 return LDC_ABORT(lp); 1512 1513 if (size > LDC_PACKET_SIZE) 1514 return -EMSGSIZE; 1515 1516 p = data_get_tx_packet(lp, &new_tail); 1517 if (!p) 1518 return -EAGAIN; 1519 1520 memcpy(p, buf, size); 1521 1522 err = send_tx_packet(lp, p, new_tail); 1523 if (!err) 1524 err = size; 1525 1526 return err; 1527 } 1528 1529 static int read_raw(struct ldc_channel *lp, void *buf, unsigned int size) 1530 { 1531 struct ldc_packet *p; 1532 unsigned long hv_err, new; 1533 int err; 1534 1535 if (size < LDC_PACKET_SIZE) 1536 return -EINVAL; 1537 1538 hv_err = sun4v_ldc_rx_get_state(lp->id, 1539 &lp->rx_head, 1540 &lp->rx_tail, 1541 &lp->chan_state); 1542 if (hv_err) 1543 return LDC_ABORT(lp); 1544 1545 if (lp->chan_state == LDC_CHANNEL_DOWN || 1546 lp->chan_state == LDC_CHANNEL_RESETTING) 1547 return -ECONNRESET; 1548 1549 if (lp->rx_head == lp->rx_tail) 1550 return 0; 1551 1552 p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE); 1553 memcpy(buf, p, LDC_PACKET_SIZE); 1554 1555 new = rx_advance(lp, lp->rx_head); 1556 lp->rx_head = new; 1557 1558 err = __set_rx_head(lp, new); 1559 if (err < 0) 1560 err = -ECONNRESET; 1561 else 1562 err = LDC_PACKET_SIZE; 1563 1564 return err; 1565 } 1566 1567 static const struct ldc_mode_ops raw_ops = { 1568 .write = write_raw, 1569 .read = read_raw, 1570 }; 1571 1572 static int write_nonraw(struct ldc_channel *lp, const void *buf, 1573 unsigned int size) 1574 { 1575 unsigned long hv_err, tail; 1576 unsigned int copied; 1577 u32 seq; 1578 int err; 1579 1580 hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail, 1581 &lp->chan_state); 1582 if (unlikely(hv_err)) 1583 return -EBUSY; 1584 1585 if (unlikely(lp->chan_state != LDC_CHANNEL_UP)) 1586 return LDC_ABORT(lp); 1587 1588 if (!tx_has_space_for(lp, size)) 1589 return -EAGAIN; 1590 1591 seq = lp->snd_nxt; 1592 copied = 0; 1593 tail = lp->tx_tail; 1594 while (copied < size) { 1595 struct ldc_packet *p = lp->tx_base + (tail / LDC_PACKET_SIZE); 1596 u8 *data = ((lp->cfg.mode == LDC_MODE_UNRELIABLE) ? 1597 p->u.u_data : 1598 p->u.r.r_data); 1599 int data_len; 1600 1601 p->type = LDC_DATA; 1602 p->stype = LDC_INFO; 1603 p->ctrl = 0; 1604 1605 data_len = size - copied; 1606 if (data_len > lp->mss) 1607 data_len = lp->mss; 1608 1609 BUG_ON(data_len > LDC_LEN); 1610 1611 p->env = (data_len | 1612 (copied == 0 ? LDC_START : 0) | 1613 (data_len == size - copied ? LDC_STOP : 0)); 1614 1615 p->seqid = ++seq; 1616 1617 ldcdbg(DATA, "SENT DATA [%02x:%02x:%02x:%02x:%08x]\n", 1618 p->type, 1619 p->stype, 1620 p->ctrl, 1621 p->env, 1622 p->seqid); 1623 1624 memcpy(data, buf, data_len); 1625 buf += data_len; 1626 copied += data_len; 1627 1628 tail = tx_advance(lp, tail); 1629 } 1630 1631 err = set_tx_tail(lp, tail); 1632 if (!err) { 1633 lp->snd_nxt = seq; 1634 err = size; 1635 } 1636 1637 return err; 1638 } 1639 1640 static int rx_bad_seq(struct ldc_channel *lp, struct ldc_packet *p, 1641 struct ldc_packet *first_frag) 1642 { 1643 int err; 1644 1645 if (first_frag) 1646 lp->rcv_nxt = first_frag->seqid - 1; 1647 1648 err = send_data_nack(lp, p); 1649 if (err) 1650 return err; 1651 1652 err = ldc_rx_reset(lp); 1653 if (err < 0) 1654 return LDC_ABORT(lp); 1655 1656 return 0; 1657 } 1658 1659 static int data_ack_nack(struct ldc_channel *lp, struct ldc_packet *p) 1660 { 1661 if (p->stype & LDC_ACK) { 1662 int err = process_data_ack(lp, p); 1663 if (err) 1664 return err; 1665 } 1666 if (p->stype & LDC_NACK) 1667 return LDC_ABORT(lp); 1668 1669 return 0; 1670 } 1671 1672 static int rx_data_wait(struct ldc_channel *lp, unsigned long cur_head) 1673 { 1674 unsigned long dummy; 1675 int limit = 1000; 1676 1677 ldcdbg(DATA, "DATA WAIT cur_head[%lx] rx_head[%lx] rx_tail[%lx]\n", 1678 cur_head, lp->rx_head, lp->rx_tail); 1679 while (limit-- > 0) { 1680 unsigned long hv_err; 1681 1682 hv_err = sun4v_ldc_rx_get_state(lp->id, 1683 &dummy, 1684 &lp->rx_tail, 1685 &lp->chan_state); 1686 if (hv_err) 1687 return LDC_ABORT(lp); 1688 1689 if (lp->chan_state == LDC_CHANNEL_DOWN || 1690 lp->chan_state == LDC_CHANNEL_RESETTING) 1691 return -ECONNRESET; 1692 1693 if (cur_head != lp->rx_tail) { 1694 ldcdbg(DATA, "DATA WAIT DONE " 1695 "head[%lx] tail[%lx] chan_state[%lx]\n", 1696 dummy, lp->rx_tail, lp->chan_state); 1697 return 0; 1698 } 1699 1700 udelay(1); 1701 } 1702 return -EAGAIN; 1703 } 1704 1705 static int rx_set_head(struct ldc_channel *lp, unsigned long head) 1706 { 1707 int err = __set_rx_head(lp, head); 1708 1709 if (err < 0) 1710 return LDC_ABORT(lp); 1711 1712 lp->rx_head = head; 1713 return 0; 1714 } 1715 1716 static void send_data_ack(struct ldc_channel *lp) 1717 { 1718 unsigned long new_tail; 1719 struct ldc_packet *p; 1720 1721 p = data_get_tx_packet(lp, &new_tail); 1722 if (likely(p)) { 1723 int err; 1724 1725 memset(p, 0, sizeof(*p)); 1726 p->type = LDC_DATA; 1727 p->stype = LDC_ACK; 1728 p->ctrl = 0; 1729 p->seqid = lp->snd_nxt + 1; 1730 p->u.r.ackid = lp->rcv_nxt; 1731 1732 err = send_tx_packet(lp, p, new_tail); 1733 if (!err) 1734 lp->snd_nxt++; 1735 } 1736 } 1737 1738 static int read_nonraw(struct ldc_channel *lp, void *buf, unsigned int size) 1739 { 1740 struct ldc_packet *first_frag; 1741 unsigned long hv_err, new; 1742 int err, copied; 1743 1744 hv_err = sun4v_ldc_rx_get_state(lp->id, 1745 &lp->rx_head, 1746 &lp->rx_tail, 1747 &lp->chan_state); 1748 if (hv_err) 1749 return LDC_ABORT(lp); 1750 1751 if (lp->chan_state == LDC_CHANNEL_DOWN || 1752 lp->chan_state == LDC_CHANNEL_RESETTING) 1753 return -ECONNRESET; 1754 1755 if (lp->rx_head == lp->rx_tail) 1756 return 0; 1757 1758 first_frag = NULL; 1759 copied = err = 0; 1760 new = lp->rx_head; 1761 while (1) { 1762 struct ldc_packet *p; 1763 int pkt_len; 1764 1765 BUG_ON(new == lp->rx_tail); 1766 p = lp->rx_base + (new / LDC_PACKET_SIZE); 1767 1768 ldcdbg(RX, "RX read pkt[%02x:%02x:%02x:%02x:%08x:%08x] " 1769 "rcv_nxt[%08x]\n", 1770 p->type, 1771 p->stype, 1772 p->ctrl, 1773 p->env, 1774 p->seqid, 1775 p->u.r.ackid, 1776 lp->rcv_nxt); 1777 1778 if (unlikely(!rx_seq_ok(lp, p->seqid))) { 1779 err = rx_bad_seq(lp, p, first_frag); 1780 copied = 0; 1781 break; 1782 } 1783 1784 if (p->type & LDC_CTRL) { 1785 err = process_control_frame(lp, p); 1786 if (err < 0) 1787 break; 1788 err = 0; 1789 } 1790 1791 lp->rcv_nxt = p->seqid; 1792 1793 /* 1794 * If this is a control-only packet, there is nothing 1795 * else to do but advance the rx queue since the packet 1796 * was already processed above. 1797 */ 1798 if (!(p->type & LDC_DATA)) { 1799 new = rx_advance(lp, new); 1800 break; 1801 } 1802 if (p->stype & (LDC_ACK | LDC_NACK)) { 1803 err = data_ack_nack(lp, p); 1804 if (err) 1805 break; 1806 } 1807 if (!(p->stype & LDC_INFO)) { 1808 new = rx_advance(lp, new); 1809 err = rx_set_head(lp, new); 1810 if (err) 1811 break; 1812 goto no_data; 1813 } 1814 1815 pkt_len = p->env & LDC_LEN; 1816 1817 /* Every initial packet starts with the START bit set. 1818 * 1819 * Singleton packets will have both START+STOP set. 1820 * 1821 * Fragments will have START set in the first frame, STOP 1822 * set in the last frame, and neither bit set in middle 1823 * frames of the packet. 1824 * 1825 * Therefore if we are at the beginning of a packet and 1826 * we don't see START, or we are in the middle of a fragmented 1827 * packet and do see START, we are unsynchronized and should 1828 * flush the RX queue. 1829 */ 1830 if ((first_frag == NULL && !(p->env & LDC_START)) || 1831 (first_frag != NULL && (p->env & LDC_START))) { 1832 if (!first_frag) 1833 new = rx_advance(lp, new); 1834 1835 err = rx_set_head(lp, new); 1836 if (err) 1837 break; 1838 1839 if (!first_frag) 1840 goto no_data; 1841 } 1842 if (!first_frag) 1843 first_frag = p; 1844 1845 if (pkt_len > size - copied) { 1846 /* User didn't give us a big enough buffer, 1847 * what to do? This is a pretty serious error. 1848 * 1849 * Since we haven't updated the RX ring head to 1850 * consume any of the packets, signal the error 1851 * to the user and just leave the RX ring alone. 1852 * 1853 * This seems the best behavior because this allows 1854 * a user of the LDC layer to start with a small 1855 * RX buffer for ldc_read() calls and use -EMSGSIZE 1856 * as a cue to enlarge it's read buffer. 1857 */ 1858 err = -EMSGSIZE; 1859 break; 1860 } 1861 1862 /* Ok, we are gonna eat this one. */ 1863 new = rx_advance(lp, new); 1864 1865 memcpy(buf, 1866 (lp->cfg.mode == LDC_MODE_UNRELIABLE ? 1867 p->u.u_data : p->u.r.r_data), pkt_len); 1868 buf += pkt_len; 1869 copied += pkt_len; 1870 1871 if (p->env & LDC_STOP) 1872 break; 1873 1874 no_data: 1875 if (new == lp->rx_tail) { 1876 err = rx_data_wait(lp, new); 1877 if (err) 1878 break; 1879 } 1880 } 1881 1882 if (!err) 1883 err = rx_set_head(lp, new); 1884 1885 if (err && first_frag) 1886 lp->rcv_nxt = first_frag->seqid - 1; 1887 1888 if (!err) { 1889 err = copied; 1890 if (err > 0 && lp->cfg.mode != LDC_MODE_UNRELIABLE) 1891 send_data_ack(lp); 1892 } 1893 1894 return err; 1895 } 1896 1897 static const struct ldc_mode_ops nonraw_ops = { 1898 .write = write_nonraw, 1899 .read = read_nonraw, 1900 }; 1901 1902 static int write_stream(struct ldc_channel *lp, const void *buf, 1903 unsigned int size) 1904 { 1905 if (size > lp->cfg.mtu) 1906 size = lp->cfg.mtu; 1907 return write_nonraw(lp, buf, size); 1908 } 1909 1910 static int read_stream(struct ldc_channel *lp, void *buf, unsigned int size) 1911 { 1912 if (!lp->mssbuf_len) { 1913 int err = read_nonraw(lp, lp->mssbuf, lp->cfg.mtu); 1914 if (err < 0) 1915 return err; 1916 1917 lp->mssbuf_len = err; 1918 lp->mssbuf_off = 0; 1919 } 1920 1921 if (size > lp->mssbuf_len) 1922 size = lp->mssbuf_len; 1923 memcpy(buf, lp->mssbuf + lp->mssbuf_off, size); 1924 1925 lp->mssbuf_off += size; 1926 lp->mssbuf_len -= size; 1927 1928 return size; 1929 } 1930 1931 static const struct ldc_mode_ops stream_ops = { 1932 .write = write_stream, 1933 .read = read_stream, 1934 }; 1935 1936 int ldc_write(struct ldc_channel *lp, const void *buf, unsigned int size) 1937 { 1938 unsigned long flags; 1939 int err; 1940 1941 if (!buf) 1942 return -EINVAL; 1943 1944 if (!size) 1945 return 0; 1946 1947 spin_lock_irqsave(&lp->lock, flags); 1948 1949 if (lp->hs_state != LDC_HS_COMPLETE) 1950 err = -ENOTCONN; 1951 else 1952 err = lp->mops->write(lp, buf, size); 1953 1954 spin_unlock_irqrestore(&lp->lock, flags); 1955 1956 return err; 1957 } 1958 EXPORT_SYMBOL(ldc_write); 1959 1960 int ldc_read(struct ldc_channel *lp, void *buf, unsigned int size) 1961 { 1962 unsigned long flags; 1963 int err; 1964 1965 ldcdbg(RX, "%s: entered size=%d\n", __func__, size); 1966 1967 if (!buf) 1968 return -EINVAL; 1969 1970 if (!size) 1971 return 0; 1972 1973 spin_lock_irqsave(&lp->lock, flags); 1974 1975 if (lp->hs_state != LDC_HS_COMPLETE) 1976 err = -ENOTCONN; 1977 else 1978 err = lp->mops->read(lp, buf, size); 1979 1980 spin_unlock_irqrestore(&lp->lock, flags); 1981 1982 ldcdbg(RX, "%s: mode=%d, head=%lu, tail=%lu rv=%d\n", __func__, 1983 lp->cfg.mode, lp->rx_head, lp->rx_tail, err); 1984 1985 return err; 1986 } 1987 EXPORT_SYMBOL(ldc_read); 1988 1989 static u64 pagesize_code(void) 1990 { 1991 switch (PAGE_SIZE) { 1992 default: 1993 case (8ULL * 1024ULL): 1994 return 0; 1995 case (64ULL * 1024ULL): 1996 return 1; 1997 case (512ULL * 1024ULL): 1998 return 2; 1999 case (4ULL * 1024ULL * 1024ULL): 2000 return 3; 2001 case (32ULL * 1024ULL * 1024ULL): 2002 return 4; 2003 case (256ULL * 1024ULL * 1024ULL): 2004 return 5; 2005 } 2006 } 2007 2008 static u64 make_cookie(u64 index, u64 pgsz_code, u64 page_offset) 2009 { 2010 return ((pgsz_code << COOKIE_PGSZ_CODE_SHIFT) | 2011 (index << PAGE_SHIFT) | 2012 page_offset); 2013 } 2014 2015 2016 static struct ldc_mtable_entry *alloc_npages(struct ldc_iommu *iommu, 2017 unsigned long npages) 2018 { 2019 long entry; 2020 2021 entry = iommu_tbl_range_alloc(NULL, &iommu->iommu_map_table, 2022 npages, NULL, (unsigned long)-1, 0); 2023 if (unlikely(entry == IOMMU_ERROR_CODE)) 2024 return NULL; 2025 2026 return iommu->page_table + entry; 2027 } 2028 2029 static u64 perm_to_mte(unsigned int map_perm) 2030 { 2031 u64 mte_base; 2032 2033 mte_base = pagesize_code(); 2034 2035 if (map_perm & LDC_MAP_SHADOW) { 2036 if (map_perm & LDC_MAP_R) 2037 mte_base |= LDC_MTE_COPY_R; 2038 if (map_perm & LDC_MAP_W) 2039 mte_base |= LDC_MTE_COPY_W; 2040 } 2041 if (map_perm & LDC_MAP_DIRECT) { 2042 if (map_perm & LDC_MAP_R) 2043 mte_base |= LDC_MTE_READ; 2044 if (map_perm & LDC_MAP_W) 2045 mte_base |= LDC_MTE_WRITE; 2046 if (map_perm & LDC_MAP_X) 2047 mte_base |= LDC_MTE_EXEC; 2048 } 2049 if (map_perm & LDC_MAP_IO) { 2050 if (map_perm & LDC_MAP_R) 2051 mte_base |= LDC_MTE_IOMMU_R; 2052 if (map_perm & LDC_MAP_W) 2053 mte_base |= LDC_MTE_IOMMU_W; 2054 } 2055 2056 return mte_base; 2057 } 2058 2059 static int pages_in_region(unsigned long base, long len) 2060 { 2061 int count = 0; 2062 2063 do { 2064 unsigned long new = (base + PAGE_SIZE) & PAGE_MASK; 2065 2066 len -= (new - base); 2067 base = new; 2068 count++; 2069 } while (len > 0); 2070 2071 return count; 2072 } 2073 2074 struct cookie_state { 2075 struct ldc_mtable_entry *page_table; 2076 struct ldc_trans_cookie *cookies; 2077 u64 mte_base; 2078 u64 prev_cookie; 2079 u32 pte_idx; 2080 u32 nc; 2081 }; 2082 2083 static void fill_cookies(struct cookie_state *sp, unsigned long pa, 2084 unsigned long off, unsigned long len) 2085 { 2086 do { 2087 unsigned long tlen, new = pa + PAGE_SIZE; 2088 u64 this_cookie; 2089 2090 sp->page_table[sp->pte_idx].mte = sp->mte_base | pa; 2091 2092 tlen = PAGE_SIZE; 2093 if (off) 2094 tlen = PAGE_SIZE - off; 2095 if (tlen > len) 2096 tlen = len; 2097 2098 this_cookie = make_cookie(sp->pte_idx, 2099 pagesize_code(), off); 2100 2101 off = 0; 2102 2103 if (this_cookie == sp->prev_cookie) { 2104 sp->cookies[sp->nc - 1].cookie_size += tlen; 2105 } else { 2106 sp->cookies[sp->nc].cookie_addr = this_cookie; 2107 sp->cookies[sp->nc].cookie_size = tlen; 2108 sp->nc++; 2109 } 2110 sp->prev_cookie = this_cookie + tlen; 2111 2112 sp->pte_idx++; 2113 2114 len -= tlen; 2115 pa = new; 2116 } while (len > 0); 2117 } 2118 2119 static int sg_count_one(struct scatterlist *sg) 2120 { 2121 unsigned long base = page_to_pfn(sg_page(sg)) << PAGE_SHIFT; 2122 long len = sg->length; 2123 2124 if ((sg->offset | len) & (8UL - 1)) 2125 return -EFAULT; 2126 2127 return pages_in_region(base + sg->offset, len); 2128 } 2129 2130 static int sg_count_pages(struct scatterlist *sg, int num_sg) 2131 { 2132 int count; 2133 int i; 2134 2135 count = 0; 2136 for (i = 0; i < num_sg; i++) { 2137 int err = sg_count_one(sg + i); 2138 if (err < 0) 2139 return err; 2140 count += err; 2141 } 2142 2143 return count; 2144 } 2145 2146 int ldc_map_sg(struct ldc_channel *lp, 2147 struct scatterlist *sg, int num_sg, 2148 struct ldc_trans_cookie *cookies, int ncookies, 2149 unsigned int map_perm) 2150 { 2151 unsigned long i, npages; 2152 struct ldc_mtable_entry *base; 2153 struct cookie_state state; 2154 struct ldc_iommu *iommu; 2155 int err; 2156 struct scatterlist *s; 2157 2158 if (map_perm & ~LDC_MAP_ALL) 2159 return -EINVAL; 2160 2161 err = sg_count_pages(sg, num_sg); 2162 if (err < 0) 2163 return err; 2164 2165 npages = err; 2166 if (err > ncookies) 2167 return -EMSGSIZE; 2168 2169 iommu = &lp->iommu; 2170 2171 base = alloc_npages(iommu, npages); 2172 2173 if (!base) 2174 return -ENOMEM; 2175 2176 state.page_table = iommu->page_table; 2177 state.cookies = cookies; 2178 state.mte_base = perm_to_mte(map_perm); 2179 state.prev_cookie = ~(u64)0; 2180 state.pte_idx = (base - iommu->page_table); 2181 state.nc = 0; 2182 2183 for_each_sg(sg, s, num_sg, i) { 2184 fill_cookies(&state, page_to_pfn(sg_page(s)) << PAGE_SHIFT, 2185 s->offset, s->length); 2186 } 2187 2188 return state.nc; 2189 } 2190 EXPORT_SYMBOL(ldc_map_sg); 2191 2192 int ldc_map_single(struct ldc_channel *lp, 2193 void *buf, unsigned int len, 2194 struct ldc_trans_cookie *cookies, int ncookies, 2195 unsigned int map_perm) 2196 { 2197 unsigned long npages, pa; 2198 struct ldc_mtable_entry *base; 2199 struct cookie_state state; 2200 struct ldc_iommu *iommu; 2201 2202 if ((map_perm & ~LDC_MAP_ALL) || (ncookies < 1)) 2203 return -EINVAL; 2204 2205 pa = __pa(buf); 2206 if ((pa | len) & (8UL - 1)) 2207 return -EFAULT; 2208 2209 npages = pages_in_region(pa, len); 2210 2211 iommu = &lp->iommu; 2212 2213 base = alloc_npages(iommu, npages); 2214 2215 if (!base) 2216 return -ENOMEM; 2217 2218 state.page_table = iommu->page_table; 2219 state.cookies = cookies; 2220 state.mte_base = perm_to_mte(map_perm); 2221 state.prev_cookie = ~(u64)0; 2222 state.pte_idx = (base - iommu->page_table); 2223 state.nc = 0; 2224 fill_cookies(&state, (pa & PAGE_MASK), (pa & ~PAGE_MASK), len); 2225 BUG_ON(state.nc > ncookies); 2226 2227 return state.nc; 2228 } 2229 EXPORT_SYMBOL(ldc_map_single); 2230 2231 2232 static void free_npages(unsigned long id, struct ldc_iommu *iommu, 2233 u64 cookie, u64 size) 2234 { 2235 unsigned long npages, entry; 2236 2237 npages = PAGE_ALIGN(((cookie & ~PAGE_MASK) + size)) >> PAGE_SHIFT; 2238 2239 entry = ldc_cookie_to_index(cookie, iommu); 2240 ldc_demap(iommu, id, cookie, entry, npages); 2241 iommu_tbl_range_free(&iommu->iommu_map_table, cookie, npages, entry); 2242 } 2243 2244 void ldc_unmap(struct ldc_channel *lp, struct ldc_trans_cookie *cookies, 2245 int ncookies) 2246 { 2247 struct ldc_iommu *iommu = &lp->iommu; 2248 int i; 2249 unsigned long flags; 2250 2251 spin_lock_irqsave(&iommu->lock, flags); 2252 for (i = 0; i < ncookies; i++) { 2253 u64 addr = cookies[i].cookie_addr; 2254 u64 size = cookies[i].cookie_size; 2255 2256 free_npages(lp->id, iommu, addr, size); 2257 } 2258 spin_unlock_irqrestore(&iommu->lock, flags); 2259 } 2260 EXPORT_SYMBOL(ldc_unmap); 2261 2262 int ldc_copy(struct ldc_channel *lp, int copy_dir, 2263 void *buf, unsigned int len, unsigned long offset, 2264 struct ldc_trans_cookie *cookies, int ncookies) 2265 { 2266 unsigned int orig_len; 2267 unsigned long ra; 2268 int i; 2269 2270 if (copy_dir != LDC_COPY_IN && copy_dir != LDC_COPY_OUT) { 2271 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Bad copy_dir[%d]\n", 2272 lp->id, copy_dir); 2273 return -EINVAL; 2274 } 2275 2276 ra = __pa(buf); 2277 if ((ra | len | offset) & (8UL - 1)) { 2278 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Unaligned buffer " 2279 "ra[%lx] len[%x] offset[%lx]\n", 2280 lp->id, ra, len, offset); 2281 return -EFAULT; 2282 } 2283 2284 if (lp->hs_state != LDC_HS_COMPLETE || 2285 (lp->flags & LDC_FLAG_RESET)) { 2286 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Link down hs_state[%x] " 2287 "flags[%x]\n", lp->id, lp->hs_state, lp->flags); 2288 return -ECONNRESET; 2289 } 2290 2291 orig_len = len; 2292 for (i = 0; i < ncookies; i++) { 2293 unsigned long cookie_raddr = cookies[i].cookie_addr; 2294 unsigned long this_len = cookies[i].cookie_size; 2295 unsigned long actual_len; 2296 2297 if (unlikely(offset)) { 2298 unsigned long this_off = offset; 2299 2300 if (this_off > this_len) 2301 this_off = this_len; 2302 2303 offset -= this_off; 2304 this_len -= this_off; 2305 if (!this_len) 2306 continue; 2307 cookie_raddr += this_off; 2308 } 2309 2310 if (this_len > len) 2311 this_len = len; 2312 2313 while (1) { 2314 unsigned long hv_err; 2315 2316 hv_err = sun4v_ldc_copy(lp->id, copy_dir, 2317 cookie_raddr, ra, 2318 this_len, &actual_len); 2319 if (unlikely(hv_err)) { 2320 printk(KERN_ERR PFX "ldc_copy: ID[%lu] " 2321 "HV error %lu\n", 2322 lp->id, hv_err); 2323 if (lp->hs_state != LDC_HS_COMPLETE || 2324 (lp->flags & LDC_FLAG_RESET)) 2325 return -ECONNRESET; 2326 else 2327 return -EFAULT; 2328 } 2329 2330 cookie_raddr += actual_len; 2331 ra += actual_len; 2332 len -= actual_len; 2333 if (actual_len == this_len) 2334 break; 2335 2336 this_len -= actual_len; 2337 } 2338 2339 if (!len) 2340 break; 2341 } 2342 2343 /* It is caller policy what to do about short copies. 2344 * For example, a networking driver can declare the 2345 * packet a runt and drop it. 2346 */ 2347 2348 return orig_len - len; 2349 } 2350 EXPORT_SYMBOL(ldc_copy); 2351 2352 void *ldc_alloc_exp_dring(struct ldc_channel *lp, unsigned int len, 2353 struct ldc_trans_cookie *cookies, int *ncookies, 2354 unsigned int map_perm) 2355 { 2356 void *buf; 2357 int err; 2358 2359 if (len & (8UL - 1)) 2360 return ERR_PTR(-EINVAL); 2361 2362 buf = kzalloc(len, GFP_ATOMIC); 2363 if (!buf) 2364 return ERR_PTR(-ENOMEM); 2365 2366 err = ldc_map_single(lp, buf, len, cookies, *ncookies, map_perm); 2367 if (err < 0) { 2368 kfree(buf); 2369 return ERR_PTR(err); 2370 } 2371 *ncookies = err; 2372 2373 return buf; 2374 } 2375 EXPORT_SYMBOL(ldc_alloc_exp_dring); 2376 2377 void ldc_free_exp_dring(struct ldc_channel *lp, void *buf, unsigned int len, 2378 struct ldc_trans_cookie *cookies, int ncookies) 2379 { 2380 ldc_unmap(lp, cookies, ncookies); 2381 kfree(buf); 2382 } 2383 EXPORT_SYMBOL(ldc_free_exp_dring); 2384 2385 static int __init ldc_init(void) 2386 { 2387 unsigned long major, minor; 2388 struct mdesc_handle *hp; 2389 const u64 *v; 2390 int err; 2391 u64 mp; 2392 2393 hp = mdesc_grab(); 2394 if (!hp) 2395 return -ENODEV; 2396 2397 mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform"); 2398 err = -ENODEV; 2399 if (mp == MDESC_NODE_NULL) 2400 goto out; 2401 2402 v = mdesc_get_property(hp, mp, "domaining-enabled", NULL); 2403 if (!v) 2404 goto out; 2405 2406 major = 1; 2407 minor = 0; 2408 if (sun4v_hvapi_register(HV_GRP_LDOM, major, &minor)) { 2409 printk(KERN_INFO PFX "Could not register LDOM hvapi.\n"); 2410 goto out; 2411 } 2412 2413 printk(KERN_INFO "%s", version); 2414 2415 if (!*v) { 2416 printk(KERN_INFO PFX "Domaining disabled.\n"); 2417 goto out; 2418 } 2419 ldom_domaining_enabled = 1; 2420 err = 0; 2421 2422 out: 2423 mdesc_release(hp); 2424 return err; 2425 } 2426 2427 core_initcall(ldc_init); 2428