arch/sparc/kernel/ldc.c

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