err.no Git - linux-2.6/blob - arch/sparc64/kernel/ldc.c

   1 /* ldc.c: Logical Domain Channel link-layer protocol driver.
   2  *
   3  * Copyright (C) 2007 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
  18 #include <asm/hypervisor.h>
  19 #include <asm/iommu.h>
  20 #include <asm/page.h>
  21 #include <asm/ldc.h>
  22 #include <asm/mdesc.h>
  23
  24 #define DRV_MODULE_NAME         "ldc"
  25 #define PFX DRV_MODULE_NAME     ": "
  26 #define DRV_MODULE_VERSION      "1.0"
  27 #define DRV_MODULE_RELDATE      "June 25, 2007"
  28
  29 static char version[] __devinitdata =
  30         DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
  31 #define LDC_PACKET_SIZE         64
  32
  33 /* Packet header layout for unreliable and reliable mode frames.
  34  * When in RAW mode, packets are simply straight 64-byte payloads
  35  * with no headers.
  36  */
  37 struct ldc_packet {
  38         u8                      type;
  39 #define LDC_CTRL                0x01
  40 #define LDC_DATA                0x02
  41 #define LDC_ERR                 0x10
  42
  43         u8                      stype;
  44 #define LDC_INFO                0x01
  45 #define LDC_ACK                 0x02
  46 #define LDC_NACK                0x04
  47
  48         u8                      ctrl;
  49 #define LDC_VERS                0x01 /* Link Version            */
  50 #define LDC_RTS                 0x02 /* Request To Send         */
  51 #define LDC_RTR                 0x03 /* Ready To Receive        */
  52 #define LDC_RDX                 0x04 /* Ready for Data eXchange */
  53 #define LDC_CTRL_MSK            0x0f
  54
  55         u8                      env;
  56 #define LDC_LEN                 0x3f
  57 #define LDC_FRAG_MASK           0xc0
  58 #define LDC_START               0x40
  59 #define LDC_STOP                0x80
  60
  61         u32                     seqid;
  62
  63         union {
  64                 u8              u_data[LDC_PACKET_SIZE - 8];
  65                 struct {
  66                         u32     pad;
  67                         u32     ackid;
  68                         u8      r_data[LDC_PACKET_SIZE - 8 - 8];
  69                 } r;
  70         } u;
  71 };
  72
  73 struct ldc_version {
  74         u16 major;
  75         u16 minor;
  76 };
  77
  78 /* Ordered from largest major to lowest.  */
  79 static struct ldc_version ver_arr[] = {
  80         { .major = 1, .minor = 0 },
  81 };
  82
  83 #define LDC_DEFAULT_MTU                 (4 * LDC_PACKET_SIZE)
  84 #define LDC_DEFAULT_NUM_ENTRIES         (PAGE_SIZE / LDC_PACKET_SIZE)
  85
  86 struct ldc_channel;
  87
  88 struct ldc_mode_ops {
  89         int (*write)(struct ldc_channel *, const void *, unsigned int);
  90         int (*read)(struct ldc_channel *, void *, unsigned int);
  91 };
  92
  93 static const struct ldc_mode_ops raw_ops;
  94 static const struct ldc_mode_ops nonraw_ops;
  95 static const struct ldc_mode_ops stream_ops;
  96
  97 int ldom_domaining_enabled;
  98
  99 struct ldc_iommu {
 100         /* Protects arena alloc/free.  */
 101         spinlock_t                      lock;
 102         struct iommu_arena              arena;
 103         struct ldc_mtable_entry         *page_table;
 104 };
 105
 106 struct ldc_channel {
 107         /* Protects all operations that depend upon channel state.  */
 108         spinlock_t                      lock;
 109
 110         unsigned long                   id;
 111
 112         u8                              *mssbuf;
 113         u32                             mssbuf_len;
 114         u32                             mssbuf_off;
 115
 116         struct ldc_packet               *tx_base;
 117         unsigned long                   tx_head;
 118         unsigned long                   tx_tail;
 119         unsigned long                   tx_num_entries;
 120         unsigned long                   tx_ra;
 121
 122         unsigned long                   tx_acked;
 123
 124         struct ldc_packet               *rx_base;
 125         unsigned long                   rx_head;
 126         unsigned long                   rx_tail;
 127         unsigned long                   rx_num_entries;
 128         unsigned long                   rx_ra;
 129
 130         u32                             rcv_nxt;
 131         u32                             snd_nxt;
 132
 133         unsigned long                   chan_state;
 134
 135         struct ldc_channel_config       cfg;
 136         void                            *event_arg;
 137
 138         const struct ldc_mode_ops       *mops;
 139
 140         struct ldc_iommu                iommu;
 141
 142         struct ldc_version              ver;
 143
 144         u8                              hs_state;
 145 #define LDC_HS_CLOSED                   0x00
 146 #define LDC_HS_OPEN                     0x01
 147 #define LDC_HS_GOTVERS                  0x02
 148 #define LDC_HS_SENTRTR                  0x03
 149 #define LDC_HS_GOTRTR                   0x04
 150 #define LDC_HS_COMPLETE                 0x10
 151
 152         u8                              flags;
 153 #define LDC_FLAG_ALLOCED_QUEUES         0x01
 154 #define LDC_FLAG_REGISTERED_QUEUES      0x02
 155 #define LDC_FLAG_REGISTERED_IRQS        0x04
 156 #define LDC_FLAG_RESET                  0x10
 157
 158         u8                              mss;
 159         u8                              state;
 160
 161         struct hlist_head               mh_list;
 162
 163         struct hlist_node               list;
 164 };
 165
 166 #define ldcdbg(TYPE, f, a...) \
 167 do {    if (lp->cfg.debug & LDC_DEBUG_##TYPE) \
 168                 printk(KERN_INFO PFX "ID[%lu] " f, lp->id, ## a); \
 169 } while (0)
 170
 171 static const char *state_to_str(u8 state)
 172 {
 173         switch (state) {
 174         case LDC_STATE_INVALID:
 175                 return "INVALID";
 176         case LDC_STATE_INIT:
 177                 return "INIT";
 178         case LDC_STATE_BOUND:
 179                 return "BOUND";
 180         case LDC_STATE_READY:
 181                 return "READY";
 182         case LDC_STATE_CONNECTED:
 183                 return "CONNECTED";
 184         default:
 185                 return "<UNKNOWN>";
 186         }
 187 }
 188
 189 static void ldc_set_state(struct ldc_channel *lp, u8 state)
 190 {
 191         ldcdbg(STATE, "STATE (%s) --> (%s)\n",
 192                state_to_str(lp->state),
 193                state_to_str(state));
 194
 195         lp->state = state;
 196 }
 197
 198 static unsigned long __advance(unsigned long off, unsigned long num_entries)
 199 {
 200         off += LDC_PACKET_SIZE;
 201         if (off == (num_entries * LDC_PACKET_SIZE))
 202                 off = 0;
 203
 204         return off;
 205 }
 206
 207 static unsigned long rx_advance(struct ldc_channel *lp, unsigned long off)
 208 {
 209         return __advance(off, lp->rx_num_entries);
 210 }
 211
 212 static unsigned long tx_advance(struct ldc_channel *lp, unsigned long off)
 213 {
 214         return __advance(off, lp->tx_num_entries);
 215 }
 216
 217 static struct ldc_packet *handshake_get_tx_packet(struct ldc_channel *lp,
 218                                                   unsigned long *new_tail)
 219 {
 220         struct ldc_packet *p;
 221         unsigned long t;
 222
 223         t = tx_advance(lp, lp->tx_tail);
 224         if (t == lp->tx_head)
 225                 return NULL;
 226
 227         *new_tail = t;
 228
 229         p = lp->tx_base;
 230         return p + (lp->tx_tail / LDC_PACKET_SIZE);
 231 }
 232
 233 /* When we are in reliable or stream mode, have to track the next packet
 234  * we haven't gotten an ACK for in the TX queue using tx_acked.  We have
 235  * to be careful not to stomp over the queue past that point.  During
 236  * the handshake, we don't have TX data packets pending in the queue
 237  * and that's why handshake_get_tx_packet() need not be mindful of
 238  * lp->tx_acked.
 239  */
 240 static unsigned long head_for_data(struct ldc_channel *lp)
 241 {
 242         if (lp->cfg.mode == LDC_MODE_RELIABLE ||
 243             lp->cfg.mode == LDC_MODE_STREAM)
 244                 return lp->tx_acked;
 245         return lp->tx_head;
 246 }
 247
 248 static int tx_has_space_for(struct ldc_channel *lp, unsigned int size)
 249 {
 250         unsigned long limit, tail, new_tail, diff;
 251         unsigned int mss;
 252
 253         limit = head_for_data(lp);
 254         tail = lp->tx_tail;
 255         new_tail = tx_advance(lp, tail);
 256         if (new_tail == limit)
 257                 return 0;
 258
 259         if (limit > new_tail)
 260                 diff = limit - new_tail;
 261         else
 262                 diff = (limit +
 263                         ((lp->tx_num_entries * LDC_PACKET_SIZE) - new_tail));
 264         diff /= LDC_PACKET_SIZE;
 265         mss = lp->mss;
 266
 267         if (diff * mss < size)
 268                 return 0;
 269
 270         return 1;
 271 }
 272
 273 static struct ldc_packet *data_get_tx_packet(struct ldc_channel *lp,
 274                                              unsigned long *new_tail)
 275 {
 276         struct ldc_packet *p;
 277         unsigned long h, t;
 278
 279         h = head_for_data(lp);
 280         t = tx_advance(lp, lp->tx_tail);
 281         if (t == h)
 282                 return NULL;
 283
 284         *new_tail = t;
 285
 286         p = lp->tx_base;
 287         return p + (lp->tx_tail / LDC_PACKET_SIZE);
 288 }
 289
 290 static int set_tx_tail(struct ldc_channel *lp, unsigned long tail)
 291 {
 292         unsigned long orig_tail = lp->tx_tail;
 293         int limit = 1000;
 294
 295         lp->tx_tail = tail;
 296         while (limit-- > 0) {
 297                 unsigned long err;
 298
 299                 err = sun4v_ldc_tx_set_qtail(lp->id, tail);
 300                 if (!err)
 301                         return 0;
 302
 303                 if (err != HV_EWOULDBLOCK) {
 304                         lp->tx_tail = orig_tail;
 305                         return -EINVAL;
 306                 }
 307                 udelay(1);
 308         }
 309
 310         lp->tx_tail = orig_tail;
 311         return -EBUSY;
 312 }
 313
 314 /* This just updates the head value in the hypervisor using
 315  * a polling loop with a timeout.  The caller takes care of
 316  * upating software state representing the head change, if any.
 317  */
 318 static int __set_rx_head(struct ldc_channel *lp, unsigned long head)
 319 {
 320         int limit = 1000;
 321
 322         while (limit-- > 0) {
 323                 unsigned long err;
 324
 325                 err = sun4v_ldc_rx_set_qhead(lp->id, head);
 326                 if (!err)
 327                         return 0;
 328
 329                 if (err != HV_EWOULDBLOCK)
 330                         return -EINVAL;
 331
 332                 udelay(1);
 333         }
 334
 335         return -EBUSY;
 336 }
 337
 338 static int send_tx_packet(struct ldc_channel *lp,
 339                           struct ldc_packet *p,
 340                           unsigned long new_tail)
 341 {
 342         BUG_ON(p != (lp->tx_base + (lp->tx_tail / LDC_PACKET_SIZE)));
 343
 344         return set_tx_tail(lp, new_tail);
 345 }
 346
 347 static struct ldc_packet *handshake_compose_ctrl(struct ldc_channel *lp,
 348                                                  u8 stype, u8 ctrl,
 349                                                  void *data, int dlen,
 350                                                  unsigned long *new_tail)
 351 {
 352         struct ldc_packet *p = handshake_get_tx_packet(lp, new_tail);
 353
 354         if (p) {
 355                 memset(p, 0, sizeof(*p));
 356                 p->type = LDC_CTRL;
 357                 p->stype = stype;
 358                 p->ctrl = ctrl;
 359                 if (data)
 360                         memcpy(p->u.u_data, data, dlen);
 361         }
 362         return p;
 363 }
 364
 365 static int start_handshake(struct ldc_channel *lp)
 366 {
 367         struct ldc_packet *p;
 368         struct ldc_version *ver;
 369         unsigned long new_tail;
 370
 371         ver = &ver_arr[0];
 372
 373         ldcdbg(HS, "SEND VER INFO maj[%u] min[%u]\n",
 374                ver->major, ver->minor);
 375
 376         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
 377                                    ver, sizeof(*ver), &new_tail);
 378         if (p) {
 379                 int err = send_tx_packet(lp, p, new_tail);
 380                 if (!err)
 381                         lp->flags &= ~LDC_FLAG_RESET;
 382                 return err;
 383         }
 384         return -EBUSY;
 385 }
 386
 387 static int send_version_nack(struct ldc_channel *lp,
 388                              u16 major, u16 minor)
 389 {
 390         struct ldc_packet *p;
 391         struct ldc_version ver;
 392         unsigned long new_tail;
 393
 394         ver.major = major;
 395         ver.minor = minor;
 396
 397         p = handshake_compose_ctrl(lp, LDC_NACK, LDC_VERS,
 398                                    &ver, sizeof(ver), &new_tail);
 399         if (p) {
 400                 ldcdbg(HS, "SEND VER NACK maj[%u] min[%u]\n",
 401                        ver.major, ver.minor);
 402
 403                 return send_tx_packet(lp, p, new_tail);
 404         }
 405         return -EBUSY;
 406 }
 407
 408 static int send_version_ack(struct ldc_channel *lp,
 409                             struct ldc_version *vp)
 410 {
 411         struct ldc_packet *p;
 412         unsigned long new_tail;
 413
 414         p = handshake_compose_ctrl(lp, LDC_ACK, LDC_VERS,
 415                                    vp, sizeof(*vp), &new_tail);
 416         if (p) {
 417                 ldcdbg(HS, "SEND VER ACK maj[%u] min[%u]\n",
 418                        vp->major, vp->minor);
 419
 420                 return send_tx_packet(lp, p, new_tail);
 421         }
 422         return -EBUSY;
 423 }
 424
 425 static int send_rts(struct ldc_channel *lp)
 426 {
 427         struct ldc_packet *p;
 428         unsigned long new_tail;
 429
 430         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTS, NULL, 0,
 431                                    &new_tail);
 432         if (p) {
 433                 p->env = lp->cfg.mode;
 434                 p->seqid = 0;
 435                 lp->rcv_nxt = 0;
 436
 437                 ldcdbg(HS, "SEND RTS env[0x%x] seqid[0x%x]\n",
 438                        p->env, p->seqid);
 439
 440                 return send_tx_packet(lp, p, new_tail);
 441         }
 442         return -EBUSY;
 443 }
 444
 445 static int send_rtr(struct ldc_channel *lp)
 446 {
 447         struct ldc_packet *p;
 448         unsigned long new_tail;
 449
 450         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RTR, NULL, 0,
 451                                    &new_tail);
 452         if (p) {
 453                 p->env = lp->cfg.mode;
 454                 p->seqid = 0;
 455
 456                 ldcdbg(HS, "SEND RTR env[0x%x] seqid[0x%x]\n",
 457                        p->env, p->seqid);
 458
 459                 return send_tx_packet(lp, p, new_tail);
 460         }
 461         return -EBUSY;
 462 }
 463
 464 static int send_rdx(struct ldc_channel *lp)
 465 {
 466         struct ldc_packet *p;
 467         unsigned long new_tail;
 468
 469         p = handshake_compose_ctrl(lp, LDC_INFO, LDC_RDX, NULL, 0,
 470                                    &new_tail);
 471         if (p) {
 472                 p->env = 0;
 473                 p->seqid = ++lp->snd_nxt;
 474                 p->u.r.ackid = lp->rcv_nxt;
 475
 476                 ldcdbg(HS, "SEND RDX env[0x%x] seqid[0x%x] ackid[0x%x]\n",
 477                        p->env, p->seqid, p->u.r.ackid);
 478
 479                 return send_tx_packet(lp, p, new_tail);
 480         }
 481         return -EBUSY;
 482 }
 483
 484 static int send_data_nack(struct ldc_channel *lp, struct ldc_packet *data_pkt)
 485 {
 486         struct ldc_packet *p;
 487         unsigned long new_tail;
 488         int err;
 489
 490         p = data_get_tx_packet(lp, &new_tail);
 491         if (!p)
 492                 return -EBUSY;
 493         memset(p, 0, sizeof(*p));
 494         p->type = data_pkt->type;
 495         p->stype = LDC_NACK;
 496         p->ctrl = data_pkt->ctrl & LDC_CTRL_MSK;
 497         p->seqid = lp->snd_nxt;
 498         p->u.r.ackid = lp->rcv_nxt;
 499
 500         ldcdbg(HS, "SEND DATA NACK type[0x%x] ctl[0x%x] seq[0x%x] ack[0x%x]\n",
 501                p->type, p->ctrl, p->seqid, p->u.r.ackid);
 502
 503         err = send_tx_packet(lp, p, new_tail);
 504         if (!err)
 505                 lp->snd_nxt++;
 506
 507         return err;
 508 }
 509
 510 static int ldc_abort(struct ldc_channel *lp)
 511 {
 512         unsigned long hv_err;
 513
 514         ldcdbg(STATE, "ABORT\n");
 515
 516         /* We report but do not act upon the hypervisor errors because
 517          * there really isn't much we can do if they fail at this point.
 518          */
 519         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
 520         if (hv_err)
 521                 printk(KERN_ERR PFX "ldc_abort: "
 522                        "sun4v_ldc_tx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
 523                        lp->id, lp->tx_ra, lp->tx_num_entries, hv_err);
 524
 525         hv_err = sun4v_ldc_tx_get_state(lp->id,
 526                                         &lp->tx_head,
 527                                         &lp->tx_tail,
 528                                         &lp->chan_state);
 529         if (hv_err)
 530                 printk(KERN_ERR PFX "ldc_abort: "
 531                        "sun4v_ldc_tx_get_state(%lx,...) failed, err=%lu\n",
 532                        lp->id, hv_err);
 533
 534         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
 535         if (hv_err)
 536                 printk(KERN_ERR PFX "ldc_abort: "
 537                        "sun4v_ldc_rx_qconf(%lx,%lx,%lx) failed, err=%lu\n",
 538                        lp->id, lp->rx_ra, lp->rx_num_entries, hv_err);
 539
 540         /* Refetch the RX queue state as well, because we could be invoked
 541          * here in the queue processing context.
 542          */
 543         hv_err = sun4v_ldc_rx_get_state(lp->id,
 544                                         &lp->rx_head,
 545                                         &lp->rx_tail,
 546                                         &lp->chan_state);
 547         if (hv_err)
 548                 printk(KERN_ERR PFX "ldc_abort: "
 549                        "sun4v_ldc_rx_get_state(%lx,...) failed, err=%lu\n",
 550                        lp->id, hv_err);
 551
 552         return -ECONNRESET;
 553 }
 554
 555 static struct ldc_version *find_by_major(u16 major)
 556 {
 557         struct ldc_version *ret = NULL;
 558         int i;
 559
 560         for (i = 0; i < ARRAY_SIZE(ver_arr); i++) {
 561                 struct ldc_version *v = &ver_arr[i];
 562                 if (v->major <= major) {
 563                         ret = v;
 564                         break;
 565                 }
 566         }
 567         return ret;
 568 }
 569
 570 static int process_ver_info(struct ldc_channel *lp, struct ldc_version *vp)
 571 {
 572         struct ldc_version *vap;
 573         int err;
 574
 575         ldcdbg(HS, "GOT VERSION INFO major[%x] minor[%x]\n",
 576                vp->major, vp->minor);
 577
 578         if (lp->hs_state == LDC_HS_GOTVERS) {
 579                 lp->hs_state = LDC_HS_OPEN;
 580                 memset(&lp->ver, 0, sizeof(lp->ver));
 581         }
 582
 583         vap = find_by_major(vp->major);
 584         if (!vap) {
 585                 err = send_version_nack(lp, 0, 0);
 586         } else if (vap->major != vp->major) {
 587                 err = send_version_nack(lp, vap->major, vap->minor);
 588         } else {
 589                 struct ldc_version ver = *vp;
 590                 if (ver.minor > vap->minor)
 591                         ver.minor = vap->minor;
 592                 err = send_version_ack(lp, &ver);
 593                 if (!err) {
 594                         lp->ver = ver;
 595                         lp->hs_state = LDC_HS_GOTVERS;
 596                 }
 597         }
 598         if (err)
 599                 return ldc_abort(lp);
 600
 601         return 0;
 602 }
 603
 604 static int process_ver_ack(struct ldc_channel *lp, struct ldc_version *vp)
 605 {
 606         ldcdbg(HS, "GOT VERSION ACK major[%x] minor[%x]\n",
 607                vp->major, vp->minor);
 608
 609         if (lp->hs_state == LDC_HS_GOTVERS) {
 610                 if (lp->ver.major != vp->major ||
 611                     lp->ver.minor != vp->minor)
 612                         return ldc_abort(lp);
 613         } else {
 614                 lp->ver = *vp;
 615                 lp->hs_state = LDC_HS_GOTVERS;
 616         }
 617         if (send_rts(lp))
 618                 return ldc_abort(lp);
 619         return 0;
 620 }
 621
 622 static int process_ver_nack(struct ldc_channel *lp, struct ldc_version *vp)
 623 {
 624         struct ldc_version *vap;
 625
 626         if ((vp->major == 0 && vp->minor == 0) ||
 627             !(vap = find_by_major(vp->major))) {
 628                 return ldc_abort(lp);
 629         } else {
 630                 struct ldc_packet *p;
 631                 unsigned long new_tail;
 632
 633                 p = handshake_compose_ctrl(lp, LDC_INFO, LDC_VERS,
 634                                            vap, sizeof(*vap),
 635                                            &new_tail);
 636                 if (p)
 637                         return send_tx_packet(lp, p, new_tail);
 638                 else
 639                         return ldc_abort(lp);
 640         }
 641 }
 642
 643 static int process_version(struct ldc_channel *lp,
 644                            struct ldc_packet *p)
 645 {
 646         struct ldc_version *vp;
 647
 648         vp = (struct ldc_version *) p->u.u_data;
 649
 650         switch (p->stype) {
 651         case LDC_INFO:
 652                 return process_ver_info(lp, vp);
 653
 654         case LDC_ACK:
 655                 return process_ver_ack(lp, vp);
 656
 657         case LDC_NACK:
 658                 return process_ver_nack(lp, vp);
 659
 660         default:
 661                 return ldc_abort(lp);
 662         }
 663 }
 664
 665 static int process_rts(struct ldc_channel *lp,
 666                        struct ldc_packet *p)
 667 {
 668         ldcdbg(HS, "GOT RTS stype[%x] seqid[%x] env[%x]\n",
 669                p->stype, p->seqid, p->env);
 670
 671         if (p->stype     != LDC_INFO       ||
 672             lp->hs_state != LDC_HS_GOTVERS ||
 673             p->env       != lp->cfg.mode)
 674                 return ldc_abort(lp);
 675
 676         lp->snd_nxt = p->seqid;
 677         lp->rcv_nxt = p->seqid;
 678         lp->hs_state = LDC_HS_SENTRTR;
 679         if (send_rtr(lp))
 680                 return ldc_abort(lp);
 681
 682         return 0;
 683 }
 684
 685 static int process_rtr(struct ldc_channel *lp,
 686                        struct ldc_packet *p)
 687 {
 688         ldcdbg(HS, "GOT RTR stype[%x] seqid[%x] env[%x]\n",
 689                p->stype, p->seqid, p->env);
 690
 691         if (p->stype     != LDC_INFO ||
 692             p->env       != lp->cfg.mode)
 693                 return ldc_abort(lp);
 694
 695         lp->snd_nxt = p->seqid;
 696         lp->hs_state = LDC_HS_COMPLETE;
 697         ldc_set_state(lp, LDC_STATE_CONNECTED);
 698         send_rdx(lp);
 699
 700         return LDC_EVENT_UP;
 701 }
 702
 703 static int rx_seq_ok(struct ldc_channel *lp, u32 seqid)
 704 {
 705         return lp->rcv_nxt + 1 == seqid;
 706 }
 707
 708 static int process_rdx(struct ldc_channel *lp,
 709                        struct ldc_packet *p)
 710 {
 711         ldcdbg(HS, "GOT RDX stype[%x] seqid[%x] env[%x] ackid[%x]\n",
 712                p->stype, p->seqid, p->env, p->u.r.ackid);
 713
 714         if (p->stype != LDC_INFO ||
 715             !(rx_seq_ok(lp, p->seqid)))
 716                 return ldc_abort(lp);
 717
 718         lp->rcv_nxt = p->seqid;
 719
 720         lp->hs_state = LDC_HS_COMPLETE;
 721         ldc_set_state(lp, LDC_STATE_CONNECTED);
 722
 723         return LDC_EVENT_UP;
 724 }
 725
 726 static int process_control_frame(struct ldc_channel *lp,
 727                                  struct ldc_packet *p)
 728 {
 729         switch (p->ctrl) {
 730         case LDC_VERS:
 731                 return process_version(lp, p);
 732
 733         case LDC_RTS:
 734                 return process_rts(lp, p);
 735
 736         case LDC_RTR:
 737                 return process_rtr(lp, p);
 738
 739         case LDC_RDX:
 740                 return process_rdx(lp, p);
 741
 742         default:
 743                 return ldc_abort(lp);
 744         }
 745 }
 746
 747 static int process_error_frame(struct ldc_channel *lp,
 748                                struct ldc_packet *p)
 749 {
 750         return ldc_abort(lp);
 751 }
 752
 753 static int process_data_ack(struct ldc_channel *lp,
 754                             struct ldc_packet *ack)
 755 {
 756         unsigned long head = lp->tx_acked;
 757         u32 ackid = ack->u.r.ackid;
 758
 759         while (1) {
 760                 struct ldc_packet *p = lp->tx_base + (head / LDC_PACKET_SIZE);
 761
 762                 head = tx_advance(lp, head);
 763
 764                 if (p->seqid == ackid) {
 765                         lp->tx_acked = head;
 766                         return 0;
 767                 }
 768                 if (head == lp->tx_head)
 769                         return ldc_abort(lp);
 770         }
 771
 772         return 0;
 773 }
 774
 775 static void send_events(struct ldc_channel *lp, unsigned int event_mask)
 776 {
 777         if (event_mask & LDC_EVENT_RESET)
 778                 lp->cfg.event(lp->event_arg, LDC_EVENT_RESET);
 779         if (event_mask & LDC_EVENT_UP)
 780                 lp->cfg.event(lp->event_arg, LDC_EVENT_UP);
 781         if (event_mask & LDC_EVENT_DATA_READY)
 782                 lp->cfg.event(lp->event_arg, LDC_EVENT_DATA_READY);
 783 }
 784
 785 static irqreturn_t ldc_rx(int irq, void *dev_id)
 786 {
 787         struct ldc_channel *lp = dev_id;
 788         unsigned long orig_state, hv_err, flags;
 789         unsigned int event_mask;
 790
 791         spin_lock_irqsave(&lp->lock, flags);
 792
 793         orig_state = lp->chan_state;
 794         hv_err = sun4v_ldc_rx_get_state(lp->id,
 795                                         &lp->rx_head,
 796                                         &lp->rx_tail,
 797                                         &lp->chan_state);
 798
 799         ldcdbg(RX, "RX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
 800                orig_state, lp->chan_state, lp->rx_head, lp->rx_tail);
 801
 802         event_mask = 0;
 803
 804         if (lp->cfg.mode == LDC_MODE_RAW &&
 805             lp->chan_state == LDC_CHANNEL_UP) {
 806                 lp->hs_state = LDC_HS_COMPLETE;
 807                 ldc_set_state(lp, LDC_STATE_CONNECTED);
 808
 809                 event_mask |= LDC_EVENT_UP;
 810
 811                 orig_state = lp->chan_state;
 812         }
 813
 814         /* If we are in reset state, flush the RX queue and ignore
 815          * everything.
 816          */
 817         if (lp->flags & LDC_FLAG_RESET) {
 818                 (void) __set_rx_head(lp, lp->rx_tail);
 819                 goto out;
 820         }
 821
 822         /* Once we finish the handshake, we let the ldc_read()
 823          * paths do all of the control frame and state management.
 824          * Just trigger the callback.
 825          */
 826         if (lp->hs_state == LDC_HS_COMPLETE) {
 827 handshake_complete:
 828                 if (lp->chan_state != orig_state) {
 829                         unsigned int event = LDC_EVENT_RESET;
 830
 831                         if (lp->chan_state == LDC_CHANNEL_UP)
 832                                 event = LDC_EVENT_UP;
 833
 834                         event_mask |= event;
 835                 }
 836                 if (lp->rx_head != lp->rx_tail)
 837                         event_mask |= LDC_EVENT_DATA_READY;
 838
 839                 goto out;
 840         }
 841
 842         if (lp->chan_state != orig_state)
 843                 goto out;
 844
 845         while (lp->rx_head != lp->rx_tail) {
 846                 struct ldc_packet *p;
 847                 unsigned long new;
 848                 int err;
 849
 850                 p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
 851
 852                 switch (p->type) {
 853                 case LDC_CTRL:
 854                         err = process_control_frame(lp, p);
 855                         if (err > 0)
 856                                 event_mask |= err;
 857                         break;
 858
 859                 case LDC_DATA:
 860                         event_mask |= LDC_EVENT_DATA_READY;
 861                         err = 0;
 862                         break;
 863
 864                 case LDC_ERR:
 865                         err = process_error_frame(lp, p);
 866                         break;
 867
 868                 default:
 869                         err = ldc_abort(lp);
 870                         break;
 871                 }
 872
 873                 if (err < 0)
 874                         break;
 875
 876                 new = lp->rx_head;
 877                 new += LDC_PACKET_SIZE;
 878                 if (new == (lp->rx_num_entries * LDC_PACKET_SIZE))
 879                         new = 0;
 880                 lp->rx_head = new;
 881
 882                 err = __set_rx_head(lp, new);
 883                 if (err < 0) {
 884                         (void) ldc_abort(lp);
 885                         break;
 886                 }
 887                 if (lp->hs_state == LDC_HS_COMPLETE)
 888                         goto handshake_complete;
 889         }
 890
 891 out:
 892         spin_unlock_irqrestore(&lp->lock, flags);
 893
 894         send_events(lp, event_mask);
 895
 896         return IRQ_HANDLED;
 897 }
 898
 899 static irqreturn_t ldc_tx(int irq, void *dev_id)
 900 {
 901         struct ldc_channel *lp = dev_id;
 902         unsigned long flags, hv_err, orig_state;
 903         unsigned int event_mask = 0;
 904
 905         spin_lock_irqsave(&lp->lock, flags);
 906
 907         orig_state = lp->chan_state;
 908         hv_err = sun4v_ldc_tx_get_state(lp->id,
 909                                         &lp->tx_head,
 910                                         &lp->tx_tail,
 911                                         &lp->chan_state);
 912
 913         ldcdbg(TX, " TX state[0x%02lx:0x%02lx] head[0x%04lx] tail[0x%04lx]\n",
 914                orig_state, lp->chan_state, lp->tx_head, lp->tx_tail);
 915
 916         if (lp->cfg.mode == LDC_MODE_RAW &&
 917             lp->chan_state == LDC_CHANNEL_UP) {
 918                 lp->hs_state = LDC_HS_COMPLETE;
 919                 ldc_set_state(lp, LDC_STATE_CONNECTED);
 920
 921                 event_mask |= LDC_EVENT_UP;
 922         }
 923
 924         spin_unlock_irqrestore(&lp->lock, flags);
 925
 926         send_events(lp, event_mask);
 927
 928         return IRQ_HANDLED;
 929 }
 930
 931 /* XXX ldc_alloc() and ldc_free() needs to run under a mutex so
 932  * XXX that addition and removal from the ldc_channel_list has
 933  * XXX atomicity, otherwise the __ldc_channel_exists() check is
 934  * XXX totally pointless as another thread can slip into ldc_alloc()
 935  * XXX and add a channel with the same ID.  There also needs to be
 936  * XXX a spinlock for ldc_channel_list.
 937  */
 938 static HLIST_HEAD(ldc_channel_list);
 939
 940 static int __ldc_channel_exists(unsigned long id)
 941 {
 942         struct ldc_channel *lp;
 943         struct hlist_node *n;
 944
 945         hlist_for_each_entry(lp, n, &ldc_channel_list, list) {
 946                 if (lp->id == id)
 947                         return 1;
 948         }
 949         return 0;
 950 }
 951
 952 static int alloc_queue(const char *name, unsigned long num_entries,
 953                        struct ldc_packet **base, unsigned long *ra)
 954 {
 955         unsigned long size, order;
 956         void *q;
 957
 958         size = num_entries * LDC_PACKET_SIZE;
 959         order = get_order(size);
 960
 961         q = (void *) __get_free_pages(GFP_KERNEL, order);
 962         if (!q) {
 963                 printk(KERN_ERR PFX "Alloc of %s queue failed with "
 964                        "size=%lu order=%lu\n", name, size, order);
 965                 return -ENOMEM;
 966         }
 967
 968         memset(q, 0, PAGE_SIZE << order);
 969
 970         *base = q;
 971         *ra = __pa(q);
 972
 973         return 0;
 974 }
 975
 976 static void free_queue(unsigned long num_entries, struct ldc_packet *q)
 977 {
 978         unsigned long size, order;
 979
 980         if (!q)
 981                 return;
 982
 983         size = num_entries * LDC_PACKET_SIZE;
 984         order = get_order(size);
 985
 986         free_pages((unsigned long)q, order);
 987 }
 988
 989 /* XXX Make this configurable... XXX */
 990 #define LDC_IOTABLE_SIZE        (8 * 1024)
 991
 992 static int ldc_iommu_init(struct ldc_channel *lp)
 993 {
 994         unsigned long sz, num_tsb_entries, tsbsize, order;
 995         struct ldc_iommu *iommu = &lp->iommu;
 996         struct ldc_mtable_entry *table;
 997         unsigned long hv_err;
 998         int err;
 999
1000         num_tsb_entries = LDC_IOTABLE_SIZE;
1001         tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1002
1003         spin_lock_init(&iommu->lock);
1004
1005         sz = num_tsb_entries / 8;
1006         sz = (sz + 7UL) & ~7UL;
1007         iommu->arena.map = kzalloc(sz, GFP_KERNEL);
1008         if (!iommu->arena.map) {
1009                 printk(KERN_ERR PFX "Alloc of arena map failed, sz=%lu\n", sz);
1010                 return -ENOMEM;
1011         }
1012
1013         iommu->arena.limit = num_tsb_entries;
1014
1015         order = get_order(tsbsize);
1016
1017         table = (struct ldc_mtable_entry *)
1018                 __get_free_pages(GFP_KERNEL, order);
1019         err = -ENOMEM;
1020         if (!table) {
1021                 printk(KERN_ERR PFX "Alloc of MTE table failed, "
1022                        "size=%lu order=%lu\n", tsbsize, order);
1023                 goto out_free_map;
1024         }
1025
1026         memset(table, 0, PAGE_SIZE << order);
1027
1028         iommu->page_table = table;
1029
1030         hv_err = sun4v_ldc_set_map_table(lp->id, __pa(table),
1031                                          num_tsb_entries);
1032         err = -EINVAL;
1033         if (hv_err)
1034                 goto out_free_table;
1035
1036         return 0;
1037
1038 out_free_table:
1039         free_pages((unsigned long) table, order);
1040         iommu->page_table = NULL;
1041
1042 out_free_map:
1043         kfree(iommu->arena.map);
1044         iommu->arena.map = NULL;
1045
1046         return err;
1047 }
1048
1049 static void ldc_iommu_release(struct ldc_channel *lp)
1050 {
1051         struct ldc_iommu *iommu = &lp->iommu;
1052         unsigned long num_tsb_entries, tsbsize, order;
1053
1054         (void) sun4v_ldc_set_map_table(lp->id, 0, 0);
1055
1056         num_tsb_entries = iommu->arena.limit;
1057         tsbsize = num_tsb_entries * sizeof(struct ldc_mtable_entry);
1058         order = get_order(tsbsize);
1059
1060         free_pages((unsigned long) iommu->page_table, order);
1061         iommu->page_table = NULL;
1062
1063         kfree(iommu->arena.map);
1064         iommu->arena.map = NULL;
1065 }
1066
1067 struct ldc_channel *ldc_alloc(unsigned long id,
1068                               const struct ldc_channel_config *cfgp,
1069                               void *event_arg)
1070 {
1071         struct ldc_channel *lp;
1072         const struct ldc_mode_ops *mops;
1073         unsigned long dummy1, dummy2, hv_err;
1074         u8 mss, *mssbuf;
1075         int err;
1076
1077         err = -ENODEV;
1078         if (!ldom_domaining_enabled)
1079                 goto out_err;
1080
1081         err = -EINVAL;
1082         if (!cfgp)
1083                 goto out_err;
1084
1085         switch (cfgp->mode) {
1086         case LDC_MODE_RAW:
1087                 mops = &raw_ops;
1088                 mss = LDC_PACKET_SIZE;
1089                 break;
1090
1091         case LDC_MODE_UNRELIABLE:
1092                 mops = &nonraw_ops;
1093                 mss = LDC_PACKET_SIZE - 8;
1094                 break;
1095
1096         case LDC_MODE_RELIABLE:
1097                 mops = &nonraw_ops;
1098                 mss = LDC_PACKET_SIZE - 8 - 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         if (mssbuf)
1188                 kfree(mssbuf);
1189
1190 out_free_iommu:
1191         ldc_iommu_release(lp);
1192
1193 out_free_ldc:
1194         kfree(lp);
1195
1196 out_err:
1197         return ERR_PTR(err);
1198 }
1199 EXPORT_SYMBOL(ldc_alloc);
1200
1201 void ldc_free(struct ldc_channel *lp)
1202 {
1203         if (lp->flags & LDC_FLAG_REGISTERED_IRQS) {
1204                 free_irq(lp->cfg.rx_irq, lp);
1205                 free_irq(lp->cfg.tx_irq, lp);
1206         }
1207
1208         if (lp->flags & LDC_FLAG_REGISTERED_QUEUES) {
1209                 sun4v_ldc_tx_qconf(lp->id, 0, 0);
1210                 sun4v_ldc_rx_qconf(lp->id, 0, 0);
1211                 lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1212         }
1213         if (lp->flags & LDC_FLAG_ALLOCED_QUEUES) {
1214                 free_queue(lp->tx_num_entries, lp->tx_base);
1215                 free_queue(lp->rx_num_entries, lp->rx_base);
1216                 lp->flags &= ~LDC_FLAG_ALLOCED_QUEUES;
1217         }
1218
1219         hlist_del(&lp->list);
1220
1221         if (lp->mssbuf)
1222                 kfree(lp->mssbuf);
1223
1224         ldc_iommu_release(lp);
1225
1226         kfree(lp);
1227 }
1228 EXPORT_SYMBOL(ldc_free);
1229
1230 /* Bind the channel.  This registers the LDC queues with
1231  * the hypervisor and puts the channel into a pseudo-listening
1232  * state.  This does not initiate a handshake, ldc_connect() does
1233  * that.
1234  */
1235 int ldc_bind(struct ldc_channel *lp)
1236 {
1237         unsigned long hv_err, flags;
1238         int err = -EINVAL;
1239
1240         spin_lock_irqsave(&lp->lock, flags);
1241
1242         if (lp->state != LDC_STATE_INIT)
1243                 goto out_err;
1244
1245         err = request_irq(lp->cfg.rx_irq, ldc_rx,
1246                           IRQF_SAMPLE_RANDOM | IRQF_SHARED,
1247                           "LDC RX", lp);
1248         if (err)
1249                 goto out_err;
1250
1251         err = request_irq(lp->cfg.tx_irq, ldc_tx,
1252                           IRQF_SAMPLE_RANDOM | IRQF_SHARED,
1253                           "LDC TX", lp);
1254         if (err)
1255                 goto out_free_rx_irq;
1256
1257
1258         lp->flags |= LDC_FLAG_REGISTERED_IRQS;
1259
1260         err = -ENODEV;
1261         hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1262         if (hv_err)
1263                 goto out_free_tx_irq;
1264
1265         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1266         if (hv_err)
1267                 goto out_free_tx_irq;
1268
1269         hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1270         if (hv_err)
1271                 goto out_unmap_tx;
1272
1273         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1274         if (hv_err)
1275                 goto out_unmap_tx;
1276
1277         lp->flags |= LDC_FLAG_REGISTERED_QUEUES;
1278
1279         hv_err = sun4v_ldc_tx_get_state(lp->id,
1280                                         &lp->tx_head,
1281                                         &lp->tx_tail,
1282                                         &lp->chan_state);
1283         err = -EBUSY;
1284         if (hv_err)
1285                 goto out_unmap_rx;
1286
1287         lp->tx_acked = lp->tx_head;
1288
1289         lp->hs_state = LDC_HS_OPEN;
1290         ldc_set_state(lp, LDC_STATE_BOUND);
1291
1292         spin_unlock_irqrestore(&lp->lock, flags);
1293
1294         return 0;
1295
1296 out_unmap_rx:
1297         lp->flags &= ~LDC_FLAG_REGISTERED_QUEUES;
1298         sun4v_ldc_rx_qconf(lp->id, 0, 0);
1299
1300 out_unmap_tx:
1301         sun4v_ldc_tx_qconf(lp->id, 0, 0);
1302
1303 out_free_tx_irq:
1304         lp->flags &= ~LDC_FLAG_REGISTERED_IRQS;
1305         free_irq(lp->cfg.tx_irq, lp);
1306
1307 out_free_rx_irq:
1308         free_irq(lp->cfg.rx_irq, lp);
1309
1310 out_err:
1311         spin_unlock_irqrestore(&lp->lock, flags);
1312
1313         return err;
1314 }
1315 EXPORT_SYMBOL(ldc_bind);
1316
1317 int ldc_connect(struct ldc_channel *lp)
1318 {
1319         unsigned long flags;
1320         int err;
1321
1322         if (lp->cfg.mode == LDC_MODE_RAW)
1323                 return -EINVAL;
1324
1325         spin_lock_irqsave(&lp->lock, flags);
1326
1327         if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1328             !(lp->flags & LDC_FLAG_REGISTERED_QUEUES) ||
1329             lp->hs_state != LDC_HS_OPEN)
1330                 err = -EINVAL;
1331         else
1332                 err = start_handshake(lp);
1333
1334         spin_unlock_irqrestore(&lp->lock, flags);
1335
1336         return err;
1337 }
1338 EXPORT_SYMBOL(ldc_connect);
1339
1340 int ldc_disconnect(struct ldc_channel *lp)
1341 {
1342         unsigned long hv_err, flags;
1343         int err;
1344
1345         if (lp->cfg.mode == LDC_MODE_RAW)
1346                 return -EINVAL;
1347
1348         if (!(lp->flags & LDC_FLAG_ALLOCED_QUEUES) ||
1349             !(lp->flags & LDC_FLAG_REGISTERED_QUEUES))
1350                 return -EINVAL;
1351
1352         spin_lock_irqsave(&lp->lock, flags);
1353
1354         err = -ENODEV;
1355         hv_err = sun4v_ldc_tx_qconf(lp->id, 0, 0);
1356         if (hv_err)
1357                 goto out_err;
1358
1359         hv_err = sun4v_ldc_tx_qconf(lp->id, lp->tx_ra, lp->tx_num_entries);
1360         if (hv_err)
1361                 goto out_err;
1362
1363         hv_err = sun4v_ldc_rx_qconf(lp->id, 0, 0);
1364         if (hv_err)
1365                 goto out_err;
1366
1367         hv_err = sun4v_ldc_rx_qconf(lp->id, lp->rx_ra, lp->rx_num_entries);
1368         if (hv_err)
1369                 goto out_err;
1370
1371         ldc_set_state(lp, LDC_STATE_BOUND);
1372         lp->hs_state = LDC_HS_OPEN;
1373         lp->flags |= LDC_FLAG_RESET;
1374
1375         spin_unlock_irqrestore(&lp->lock, flags);
1376
1377         return 0;
1378
1379 out_err:
1380         sun4v_ldc_tx_qconf(lp->id, 0, 0);
1381         sun4v_ldc_rx_qconf(lp->id, 0, 0);
1382         free_irq(lp->cfg.tx_irq, lp);
1383         free_irq(lp->cfg.rx_irq, lp);
1384         lp->flags &= ~(LDC_FLAG_REGISTERED_IRQS |
1385                        LDC_FLAG_REGISTERED_QUEUES);
1386         ldc_set_state(lp, LDC_STATE_INIT);
1387
1388         spin_unlock_irqrestore(&lp->lock, flags);
1389
1390         return err;
1391 }
1392 EXPORT_SYMBOL(ldc_disconnect);
1393
1394 int ldc_state(struct ldc_channel *lp)
1395 {
1396         return lp->state;
1397 }
1398 EXPORT_SYMBOL(ldc_state);
1399
1400 static int write_raw(struct ldc_channel *lp, const void *buf, unsigned int size)
1401 {
1402         struct ldc_packet *p;
1403         unsigned long new_tail;
1404         int err;
1405
1406         if (size > LDC_PACKET_SIZE)
1407                 return -EMSGSIZE;
1408
1409         p = data_get_tx_packet(lp, &new_tail);
1410         if (!p)
1411                 return -EAGAIN;
1412
1413         memcpy(p, buf, size);
1414
1415         err = send_tx_packet(lp, p, new_tail);
1416         if (!err)
1417                 err = size;
1418
1419         return err;
1420 }
1421
1422 static int read_raw(struct ldc_channel *lp, void *buf, unsigned int size)
1423 {
1424         struct ldc_packet *p;
1425         unsigned long hv_err, new;
1426         int err;
1427
1428         if (size < LDC_PACKET_SIZE)
1429                 return -EINVAL;
1430
1431         hv_err = sun4v_ldc_rx_get_state(lp->id,
1432                                         &lp->rx_head,
1433                                         &lp->rx_tail,
1434                                         &lp->chan_state);
1435         if (hv_err)
1436                 return ldc_abort(lp);
1437
1438         if (lp->chan_state == LDC_CHANNEL_DOWN ||
1439             lp->chan_state == LDC_CHANNEL_RESETTING)
1440                 return -ECONNRESET;
1441
1442         if (lp->rx_head == lp->rx_tail)
1443                 return 0;
1444
1445         p = lp->rx_base + (lp->rx_head / LDC_PACKET_SIZE);
1446         memcpy(buf, p, LDC_PACKET_SIZE);
1447
1448         new = rx_advance(lp, lp->rx_head);
1449         lp->rx_head = new;
1450
1451         err = __set_rx_head(lp, new);
1452         if (err < 0)
1453                 err = -ECONNRESET;
1454         else
1455                 err = LDC_PACKET_SIZE;
1456
1457         return err;
1458 }
1459
1460 static const struct ldc_mode_ops raw_ops = {
1461         .write          =       write_raw,
1462         .read           =       read_raw,
1463 };
1464
1465 static int write_nonraw(struct ldc_channel *lp, const void *buf,
1466                         unsigned int size)
1467 {
1468         unsigned long hv_err, tail;
1469         unsigned int copied;
1470         u32 seq;
1471         int err;
1472
1473         hv_err = sun4v_ldc_tx_get_state(lp->id, &lp->tx_head, &lp->tx_tail,
1474                                         &lp->chan_state);
1475         if (unlikely(hv_err))
1476                 return -EBUSY;
1477
1478         if (unlikely(lp->chan_state != LDC_CHANNEL_UP))
1479                 return ldc_abort(lp);
1480
1481         if (!tx_has_space_for(lp, size))
1482                 return -EAGAIN;
1483
1484         seq = lp->snd_nxt;
1485         copied = 0;
1486         tail = lp->tx_tail;
1487         while (copied < size) {
1488                 struct ldc_packet *p = lp->tx_base + (tail / LDC_PACKET_SIZE);
1489                 u8 *data = ((lp->cfg.mode == LDC_MODE_UNRELIABLE) ?
1490                             p->u.u_data :
1491                             p->u.r.r_data);
1492                 int data_len;
1493
1494                 p->type = LDC_DATA;
1495                 p->stype = LDC_INFO;
1496                 p->ctrl = 0;
1497
1498                 data_len = size - copied;
1499                 if (data_len > lp->mss)
1500                         data_len = lp->mss;
1501
1502                 BUG_ON(data_len > LDC_LEN);
1503
1504                 p->env = (data_len |
1505                           (copied == 0 ? LDC_START : 0) |
1506                           (data_len == size - copied ? LDC_STOP : 0));
1507
1508                 p->seqid = ++seq;
1509
1510                 ldcdbg(DATA, "SENT DATA [%02x:%02x:%02x:%02x:%08x]\n",
1511                        p->type,
1512                        p->stype,
1513                        p->ctrl,
1514                        p->env,
1515                        p->seqid);
1516
1517                 memcpy(data, buf, data_len);
1518                 buf += data_len;
1519                 copied += data_len;
1520
1521                 tail = tx_advance(lp, tail);
1522         }
1523
1524         err = set_tx_tail(lp, tail);
1525         if (!err) {
1526                 lp->snd_nxt = seq;
1527                 err = size;
1528         }
1529
1530         return err;
1531 }
1532
1533 static int rx_bad_seq(struct ldc_channel *lp, struct ldc_packet *p,
1534                       struct ldc_packet *first_frag)
1535 {
1536         int err;
1537
1538         if (first_frag)
1539                 lp->rcv_nxt = first_frag->seqid - 1;
1540
1541         err = send_data_nack(lp, p);
1542         if (err)
1543                 return err;
1544
1545         err = __set_rx_head(lp, lp->rx_tail);
1546         if (err < 0)
1547                 return ldc_abort(lp);
1548
1549         return 0;
1550 }
1551
1552 static int data_ack_nack(struct ldc_channel *lp, struct ldc_packet *p)
1553 {
1554         if (p->stype & LDC_ACK) {
1555                 int err = process_data_ack(lp, p);
1556                 if (err)
1557                         return err;
1558         }
1559         if (p->stype & LDC_NACK)
1560                 return ldc_abort(lp);
1561
1562         return 0;
1563 }
1564
1565 static int rx_data_wait(struct ldc_channel *lp, unsigned long cur_head)
1566 {
1567         unsigned long dummy;
1568         int limit = 1000;
1569
1570         ldcdbg(DATA, "DATA WAIT cur_head[%lx] rx_head[%lx] rx_tail[%lx]\n",
1571                cur_head, lp->rx_head, lp->rx_tail);
1572         while (limit-- > 0) {
1573                 unsigned long hv_err;
1574
1575                 hv_err = sun4v_ldc_rx_get_state(lp->id,
1576                                                 &dummy,
1577                                                 &lp->rx_tail,
1578                                                 &lp->chan_state);
1579                 if (hv_err)
1580                         return ldc_abort(lp);
1581
1582                 ldcdbg(DATA, "REREAD head[%lx] tail[%lx] chan_state[%lx]\n",
1583                        dummy, lp->rx_tail, lp->chan_state);
1584
1585                 if (lp->chan_state == LDC_CHANNEL_DOWN ||
1586                     lp->chan_state == LDC_CHANNEL_RESETTING)
1587                         return -ECONNRESET;
1588
1589                 if (cur_head != lp->rx_tail) {
1590                         ldcdbg(DATA, "DATA WAIT DONE\n");
1591                         return 0;
1592                 }
1593
1594                 udelay(1);
1595         }
1596         return -EAGAIN;
1597 }
1598
1599 static int rx_set_head(struct ldc_channel *lp, unsigned long head)
1600 {
1601         int err = __set_rx_head(lp, head);
1602
1603         if (err < 0)
1604                 return ldc_abort(lp);
1605
1606         lp->rx_head = head;
1607         return 0;
1608 }
1609
1610 static int read_nonraw(struct ldc_channel *lp, void *buf, unsigned int size)
1611 {
1612         struct ldc_packet *first_frag;
1613         unsigned long hv_err, new;
1614         int err, copied;
1615
1616         hv_err = sun4v_ldc_rx_get_state(lp->id,
1617                                         &lp->rx_head,
1618                                         &lp->rx_tail,
1619                                         &lp->chan_state);
1620         if (hv_err)
1621                 return ldc_abort(lp);
1622
1623         if (lp->chan_state == LDC_CHANNEL_DOWN ||
1624             lp->chan_state == LDC_CHANNEL_RESETTING)
1625                 return -ECONNRESET;
1626
1627         if (lp->rx_head == lp->rx_tail)
1628                 return 0;
1629
1630         first_frag = NULL;
1631         copied = err = 0;
1632         new = lp->rx_head;
1633         while (1) {
1634                 struct ldc_packet *p;
1635                 int pkt_len;
1636
1637                 BUG_ON(new == lp->rx_tail);
1638                 p = lp->rx_base + (new / LDC_PACKET_SIZE);
1639
1640                 ldcdbg(RX, "RX read pkt[%02x:%02x:%02x:%02x:%08x] "
1641                        "rcv_nxt[%08x]\n",
1642                        p->type,
1643                        p->stype,
1644                        p->ctrl,
1645                        p->env,
1646                        p->seqid,
1647                        lp->rcv_nxt);
1648
1649                 if (unlikely(!rx_seq_ok(lp, p->seqid))) {
1650                         err = rx_bad_seq(lp, p, first_frag);
1651                         copied = 0;
1652                         break;
1653                 }
1654
1655                 if (p->type & LDC_CTRL) {
1656                         err = process_control_frame(lp, p);
1657                         if (err < 0)
1658                                 break;
1659                         err = 0;
1660                 }
1661
1662                 lp->rcv_nxt = p->seqid;
1663
1664                 if (!(p->type & LDC_DATA)) {
1665                         new = rx_advance(lp, new);
1666                         goto no_data;
1667                 }
1668                 if (p->stype & (LDC_ACK | LDC_NACK)) {
1669                         err = data_ack_nack(lp, p);
1670                         if (err)
1671                                 break;
1672                 }
1673                 if (!(p->stype & LDC_INFO)) {
1674                         new = rx_advance(lp, new);
1675                         goto no_data;
1676                 }
1677
1678                 pkt_len = p->env & LDC_LEN;
1679
1680                 /* Every initial packet starts with the START bit set.
1681                  *
1682                  * Singleton packets will have both START+STOP set.
1683                  *
1684                  * Fragments will have START set in the first frame, STOP
1685                  * set in the last frame, and neither bit set in middle
1686                  * frames of the packet.
1687                  *
1688                  * Therefore if we are at the beginning of a packet and
1689                  * we don't see START, or we are in the middle of a fragmented
1690                  * packet and do see START, we are unsynchronized and should
1691                  * flush the RX queue.
1692                  */
1693                 if ((first_frag == NULL && !(p->env & LDC_START)) ||
1694                     (first_frag != NULL &&  (p->env & LDC_START))) {
1695                         if (!first_frag)
1696                                 new = rx_advance(lp, new);
1697
1698                         err = rx_set_head(lp, new);
1699                         if (err)
1700                                 break;
1701
1702                         if (!first_frag)
1703                                 goto no_data;
1704                 }
1705                 if (!first_frag)
1706                         first_frag = p;
1707
1708                 if (pkt_len > size - copied) {
1709                         /* User didn't give us a big enough buffer,
1710                          * what to do?  This is a pretty serious error.
1711                          *
1712                          * Since we haven't updated the RX ring head to
1713                          * consume any of the packets, signal the error
1714                          * to the user and just leave the RX ring alone.
1715                          *
1716                          * This seems the best behavior because this allows
1717                          * a user of the LDC layer to start with a small
1718                          * RX buffer for ldc_read() calls and use -EMSGSIZE
1719                          * as a cue to enlarge it's read buffer.
1720                          */
1721                         err = -EMSGSIZE;
1722                         break;
1723                 }
1724
1725                 /* Ok, we are gonna eat this one.  */
1726                 new = rx_advance(lp, new);
1727
1728                 memcpy(buf,
1729                        (lp->cfg.mode == LDC_MODE_UNRELIABLE ?
1730                         p->u.u_data : p->u.r.r_data), pkt_len);
1731                 buf += pkt_len;
1732                 copied += pkt_len;
1733
1734                 if (p->env & LDC_STOP)
1735                         break;
1736
1737 no_data:
1738                 if (new == lp->rx_tail) {
1739                         err = rx_data_wait(lp, new);
1740                         if (err)
1741                                 break;
1742                 }
1743         }
1744
1745         if (!err)
1746                 err = rx_set_head(lp, new);
1747
1748         if (err && first_frag)
1749                 lp->rcv_nxt = first_frag->seqid - 1;
1750
1751         if (!err)
1752                 err = copied;
1753
1754         return err;
1755 }
1756
1757 static const struct ldc_mode_ops nonraw_ops = {
1758         .write          =       write_nonraw,
1759         .read           =       read_nonraw,
1760 };
1761
1762 static int write_stream(struct ldc_channel *lp, const void *buf,
1763                         unsigned int size)
1764 {
1765         if (size > lp->cfg.mtu)
1766                 size = lp->cfg.mtu;
1767         return write_nonraw(lp, buf, size);
1768 }
1769
1770 static int read_stream(struct ldc_channel *lp, void *buf, unsigned int size)
1771 {
1772         if (!lp->mssbuf_len) {
1773                 int err = read_nonraw(lp, lp->mssbuf,
1774                                       (size > lp->cfg.mtu ?
1775                                        lp->cfg.mtu : size));
1776                 if (err < 0)
1777                         return err;
1778
1779                 lp->mssbuf_len = err;
1780                 lp->mssbuf_off = 0;
1781         }
1782
1783         if (size > lp->mssbuf_len)
1784                 size = lp->mssbuf_len;
1785         memcpy(buf, lp->mssbuf + lp->mssbuf_off, size);
1786
1787         lp->mssbuf_off += size;
1788         lp->mssbuf_len -= size;
1789
1790         return size;
1791 }
1792
1793 static const struct ldc_mode_ops stream_ops = {
1794         .write          =       write_stream,
1795         .read           =       read_stream,
1796 };
1797
1798 int ldc_write(struct ldc_channel *lp, const void *buf, unsigned int size)
1799 {
1800         unsigned long flags;
1801         int err;
1802
1803         if (!buf)
1804                 return -EINVAL;
1805
1806         if (!size)
1807                 return 0;
1808
1809         spin_lock_irqsave(&lp->lock, flags);
1810
1811         if (lp->hs_state != LDC_HS_COMPLETE)
1812                 err = -ENOTCONN;
1813         else
1814                 err = lp->mops->write(lp, buf, size);
1815
1816         spin_unlock_irqrestore(&lp->lock, flags);
1817
1818         return err;
1819 }
1820 EXPORT_SYMBOL(ldc_write);
1821
1822 int ldc_read(struct ldc_channel *lp, void *buf, unsigned int size)
1823 {
1824         unsigned long flags;
1825         int err;
1826
1827         if (!buf)
1828                 return -EINVAL;
1829
1830         if (!size)
1831                 return 0;
1832
1833         spin_lock_irqsave(&lp->lock, flags);
1834
1835         if (lp->hs_state != LDC_HS_COMPLETE)
1836                 err = -ENOTCONN;
1837         else
1838                 err = lp->mops->read(lp, buf, size);
1839
1840         spin_unlock_irqrestore(&lp->lock, flags);
1841
1842         return err;
1843 }
1844 EXPORT_SYMBOL(ldc_read);
1845
1846 static long arena_alloc(struct ldc_iommu *iommu, unsigned long npages)
1847 {
1848         struct iommu_arena *arena = &iommu->arena;
1849         unsigned long n, i, start, end, limit;
1850         int pass;
1851
1852         limit = arena->limit;
1853         start = arena->hint;
1854         pass = 0;
1855
1856 again:
1857         n = find_next_zero_bit(arena->map, limit, start);
1858         end = n + npages;
1859         if (unlikely(end >= limit)) {
1860                 if (likely(pass < 1)) {
1861                         limit = start;
1862                         start = 0;
1863                         pass++;
1864                         goto again;
1865                 } else {
1866                         /* Scanned the whole thing, give up. */
1867                         return -1;
1868                 }
1869         }
1870
1871         for (i = n; i < end; i++) {
1872                 if (test_bit(i, arena->map)) {
1873                         start = i + 1;
1874                         goto again;
1875                 }
1876         }
1877
1878         for (i = n; i < end; i++)
1879                 __set_bit(i, arena->map);
1880
1881         arena->hint = end;
1882
1883         return n;
1884 }
1885
1886 #define COOKIE_PGSZ_CODE        0xf000000000000000ULL
1887 #define COOKIE_PGSZ_CODE_SHIFT  60ULL
1888
1889 static u64 pagesize_code(void)
1890 {
1891         switch (PAGE_SIZE) {
1892         default:
1893         case (8ULL * 1024ULL):
1894                 return 0;
1895         case (64ULL * 1024ULL):
1896                 return 1;
1897         case (512ULL * 1024ULL):
1898                 return 2;
1899         case (4ULL * 1024ULL * 1024ULL):
1900                 return 3;
1901         case (32ULL * 1024ULL * 1024ULL):
1902                 return 4;
1903         case (256ULL * 1024ULL * 1024ULL):
1904                 return 5;
1905         }
1906 }
1907
1908 static u64 make_cookie(u64 index, u64 pgsz_code, u64 page_offset)
1909 {
1910         return ((pgsz_code << COOKIE_PGSZ_CODE_SHIFT) |
1911                 (index << PAGE_SHIFT) |
1912                 page_offset);
1913 }
1914
1915 static u64 cookie_to_index(u64 cookie, unsigned long *shift)
1916 {
1917         u64 szcode = cookie >> COOKIE_PGSZ_CODE_SHIFT;
1918
1919         cookie &= ~COOKIE_PGSZ_CODE;
1920
1921         *shift = szcode * 3;
1922
1923         return (cookie >> (13ULL + (szcode * 3ULL)));
1924 }
1925
1926 static struct ldc_mtable_entry *alloc_npages(struct ldc_iommu *iommu,
1927                                              unsigned long npages)
1928 {
1929         long entry;
1930
1931         entry = arena_alloc(iommu, npages);
1932         if (unlikely(entry < 0))
1933                 return NULL;
1934
1935         return iommu->page_table + entry;
1936 }
1937
1938 static u64 perm_to_mte(unsigned int map_perm)
1939 {
1940         u64 mte_base;
1941
1942         mte_base = pagesize_code();
1943
1944         if (map_perm & LDC_MAP_SHADOW) {
1945                 if (map_perm & LDC_MAP_R)
1946                         mte_base |= LDC_MTE_COPY_R;
1947                 if (map_perm & LDC_MAP_W)
1948                         mte_base |= LDC_MTE_COPY_W;
1949         }
1950         if (map_perm & LDC_MAP_DIRECT) {
1951                 if (map_perm & LDC_MAP_R)
1952                         mte_base |= LDC_MTE_READ;
1953                 if (map_perm & LDC_MAP_W)
1954                         mte_base |= LDC_MTE_WRITE;
1955                 if (map_perm & LDC_MAP_X)
1956                         mte_base |= LDC_MTE_EXEC;
1957         }
1958         if (map_perm & LDC_MAP_IO) {
1959                 if (map_perm & LDC_MAP_R)
1960                         mte_base |= LDC_MTE_IOMMU_R;
1961                 if (map_perm & LDC_MAP_W)
1962                         mte_base |= LDC_MTE_IOMMU_W;
1963         }
1964
1965         return mte_base;
1966 }
1967
1968 static int pages_in_region(unsigned long base, long len)
1969 {
1970         int count = 0;
1971
1972         do {
1973                 unsigned long new = (base + PAGE_SIZE) & PAGE_MASK;
1974
1975                 len -= (new - base);
1976                 base = new;
1977                 count++;
1978         } while (len > 0);
1979
1980         return count;
1981 }
1982
1983 struct cookie_state {
1984         struct ldc_mtable_entry         *page_table;
1985         struct ldc_trans_cookie         *cookies;
1986         u64                             mte_base;
1987         u64                             prev_cookie;
1988         u32                             pte_idx;
1989         u32                             nc;
1990 };
1991
1992 static void fill_cookies(struct cookie_state *sp, unsigned long pa,
1993                          unsigned long off, unsigned long len)
1994 {
1995         do {
1996                 unsigned long tlen, new = pa + PAGE_SIZE;
1997                 u64 this_cookie;
1998
1999                 sp->page_table[sp->pte_idx].mte = sp->mte_base | pa;
2000
2001                 tlen = PAGE_SIZE;
2002                 if (off)
2003                         tlen = PAGE_SIZE - off;
2004                 if (tlen > len)
2005                         tlen = len;
2006
2007                 this_cookie = make_cookie(sp->pte_idx,
2008                                           pagesize_code(), off);
2009
2010                 off = 0;
2011
2012                 if (this_cookie == sp->prev_cookie) {
2013                         sp->cookies[sp->nc - 1].cookie_size += tlen;
2014                 } else {
2015                         sp->cookies[sp->nc].cookie_addr = this_cookie;
2016                         sp->cookies[sp->nc].cookie_size = tlen;
2017                         sp->nc++;
2018                 }
2019                 sp->prev_cookie = this_cookie + tlen;
2020
2021                 sp->pte_idx++;
2022
2023                 len -= tlen;
2024                 pa = new;
2025         } while (len > 0);
2026 }
2027
2028 static int sg_count_one(struct scatterlist *sg)
2029 {
2030         unsigned long base = page_to_pfn(sg->page) << PAGE_SHIFT;
2031         long len = sg->length;
2032
2033         if ((sg->offset | len) & (8UL - 1))
2034                 return -EFAULT;
2035
2036         return pages_in_region(base + sg->offset, len);
2037 }
2038
2039 static int sg_count_pages(struct scatterlist *sg, int num_sg)
2040 {
2041         int count;
2042         int i;
2043
2044         count = 0;
2045         for (i = 0; i < num_sg; i++) {
2046                 int err = sg_count_one(sg + i);
2047                 if (err < 0)
2048                         return err;
2049                 count += err;
2050         }
2051
2052         return count;
2053 }
2054
2055 int ldc_map_sg(struct ldc_channel *lp,
2056                struct scatterlist *sg, int num_sg,
2057                struct ldc_trans_cookie *cookies, int ncookies,
2058                unsigned int map_perm)
2059 {
2060         unsigned long i, npages, flags;
2061         struct ldc_mtable_entry *base;
2062         struct cookie_state state;
2063         struct ldc_iommu *iommu;
2064         int err;
2065
2066         if (map_perm & ~LDC_MAP_ALL)
2067                 return -EINVAL;
2068
2069         err = sg_count_pages(sg, num_sg);
2070         if (err < 0)
2071                 return err;
2072
2073         npages = err;
2074         if (err > ncookies)
2075                 return -EMSGSIZE;
2076
2077         iommu = &lp->iommu;
2078
2079         spin_lock_irqsave(&iommu->lock, flags);
2080         base = alloc_npages(iommu, npages);
2081         spin_unlock_irqrestore(&iommu->lock, flags);
2082
2083         if (!base)
2084                 return -ENOMEM;
2085
2086         state.page_table = iommu->page_table;
2087         state.cookies = cookies;
2088         state.mte_base = perm_to_mte(map_perm);
2089         state.prev_cookie = ~(u64)0;
2090         state.pte_idx = (base - iommu->page_table);
2091         state.nc = 0;
2092
2093         for (i = 0; i < num_sg; i++)
2094                 fill_cookies(&state, page_to_pfn(sg[i].page) << PAGE_SHIFT,
2095                              sg[i].offset, sg[i].length);
2096
2097         return state.nc;
2098 }
2099 EXPORT_SYMBOL(ldc_map_sg);
2100
2101 int ldc_map_single(struct ldc_channel *lp,
2102                    void *buf, unsigned int len,
2103                    struct ldc_trans_cookie *cookies, int ncookies,
2104                    unsigned int map_perm)
2105 {
2106         unsigned long npages, pa, flags;
2107         struct ldc_mtable_entry *base;
2108         struct cookie_state state;
2109         struct ldc_iommu *iommu;
2110
2111         if ((map_perm & ~LDC_MAP_ALL) || (ncookies < 1))
2112                 return -EINVAL;
2113
2114         pa = __pa(buf);
2115         if ((pa | len) & (8UL - 1))
2116                 return -EFAULT;
2117
2118         npages = pages_in_region(pa, len);
2119
2120         iommu = &lp->iommu;
2121
2122         spin_lock_irqsave(&iommu->lock, flags);
2123         base = alloc_npages(iommu, npages);
2124         spin_unlock_irqrestore(&iommu->lock, flags);
2125
2126         if (!base)
2127                 return -ENOMEM;
2128
2129         state.page_table = iommu->page_table;
2130         state.cookies = cookies;
2131         state.mte_base = perm_to_mte(map_perm);
2132         state.prev_cookie = ~(u64)0;
2133         state.pte_idx = (base - iommu->page_table);
2134         state.nc = 0;
2135         fill_cookies(&state, (pa & PAGE_MASK), (pa & ~PAGE_MASK), len);
2136         BUG_ON(state.nc != 1);
2137
2138         return state.nc;
2139 }
2140 EXPORT_SYMBOL(ldc_map_single);
2141
2142 static void free_npages(unsigned long id, struct ldc_iommu *iommu,
2143                         u64 cookie, u64 size)
2144 {
2145         struct iommu_arena *arena = &iommu->arena;
2146         unsigned long i, shift, index, npages;
2147         struct ldc_mtable_entry *base;
2148
2149         npages = PAGE_ALIGN(((cookie & ~PAGE_MASK) + size)) >> PAGE_SHIFT;
2150         index = cookie_to_index(cookie, &shift);
2151         base = iommu->page_table + index;
2152
2153         BUG_ON(index > arena->limit ||
2154                (index + npages) > arena->limit);
2155
2156         for (i = 0; i < npages; i++) {
2157                 if (base->cookie)
2158                         sun4v_ldc_revoke(id, cookie + (i << shift),
2159                                          base->cookie);
2160                 base->mte = 0;
2161                 __clear_bit(index + i, arena->map);
2162         }
2163 }
2164
2165 void ldc_unmap(struct ldc_channel *lp, struct ldc_trans_cookie *cookies,
2166                int ncookies)
2167 {
2168         struct ldc_iommu *iommu = &lp->iommu;
2169         unsigned long flags;
2170         int i;
2171
2172         spin_lock_irqsave(&iommu->lock, flags);
2173         for (i = 0; i < ncookies; i++) {
2174                 u64 addr = cookies[i].cookie_addr;
2175                 u64 size = cookies[i].cookie_size;
2176
2177                 free_npages(lp->id, iommu, addr, size);
2178         }
2179         spin_unlock_irqrestore(&iommu->lock, flags);
2180 }
2181 EXPORT_SYMBOL(ldc_unmap);
2182
2183 int ldc_copy(struct ldc_channel *lp, int copy_dir,
2184              void *buf, unsigned int len, unsigned long offset,
2185              struct ldc_trans_cookie *cookies, int ncookies)
2186 {
2187         unsigned int orig_len;
2188         unsigned long ra;
2189         int i;
2190
2191         if (copy_dir != LDC_COPY_IN && copy_dir != LDC_COPY_OUT) {
2192                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Bad copy_dir[%d]\n",
2193                        lp->id, copy_dir);
2194                 return -EINVAL;
2195         }
2196
2197         ra = __pa(buf);
2198         if ((ra | len | offset) & (8UL - 1)) {
2199                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Unaligned buffer "
2200                        "ra[%lx] len[%x] offset[%lx]\n",
2201                        lp->id, ra, len, offset);
2202                 return -EFAULT;
2203         }
2204
2205         if (lp->hs_state != LDC_HS_COMPLETE ||
2206             (lp->flags & LDC_FLAG_RESET)) {
2207                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] Link down hs_state[%x] "
2208                        "flags[%x]\n", lp->id, lp->hs_state, lp->flags);
2209                 return -ECONNRESET;
2210         }
2211
2212         orig_len = len;
2213         for (i = 0; i < ncookies; i++) {
2214                 unsigned long cookie_raddr = cookies[i].cookie_addr;
2215                 unsigned long this_len = cookies[i].cookie_size;
2216                 unsigned long actual_len;
2217
2218                 if (unlikely(offset)) {
2219                         unsigned long this_off = offset;
2220
2221                         if (this_off > this_len)
2222                                 this_off = this_len;
2223
2224                         offset -= this_off;
2225                         this_len -= this_off;
2226                         if (!this_len)
2227                                 continue;
2228                         cookie_raddr += this_off;
2229                 }
2230
2231                 if (this_len > len)
2232                         this_len = len;
2233
2234                 while (1) {
2235                         unsigned long hv_err;
2236
2237                         hv_err = sun4v_ldc_copy(lp->id, copy_dir,
2238                                                 cookie_raddr, ra,
2239                                                 this_len, &actual_len);
2240                         if (unlikely(hv_err)) {
2241                                 printk(KERN_ERR PFX "ldc_copy: ID[%lu] "
2242                                        "HV error %lu\n",
2243                                        lp->id, hv_err);
2244                                 if (lp->hs_state != LDC_HS_COMPLETE ||
2245                                     (lp->flags & LDC_FLAG_RESET))
2246                                         return -ECONNRESET;
2247                                 else
2248                                         return -EFAULT;
2249                         }
2250
2251                         cookie_raddr += actual_len;
2252                         ra += actual_len;
2253                         len -= actual_len;
2254                         if (actual_len == this_len)
2255                                 break;
2256
2257                         this_len -= actual_len;
2258                 }
2259
2260                 if (!len)
2261                         break;
2262         }
2263
2264         /* It is caller policy what to do about short copies.
2265          * For example, a networking driver can declare the
2266          * packet a runt and drop it.
2267          */
2268
2269         return orig_len - len;
2270 }
2271 EXPORT_SYMBOL(ldc_copy);
2272
2273 void *ldc_alloc_exp_dring(struct ldc_channel *lp, unsigned int len,
2274                           struct ldc_trans_cookie *cookies, int *ncookies,
2275                           unsigned int map_perm)
2276 {
2277         void *buf;
2278         int err;
2279
2280         if (len & (8UL - 1))
2281                 return ERR_PTR(-EINVAL);
2282
2283         buf = kzalloc(len, GFP_KERNEL);
2284         if (!buf)
2285                 return ERR_PTR(-ENOMEM);
2286
2287         err = ldc_map_single(lp, buf, len, cookies, *ncookies, map_perm);
2288         if (err < 0) {
2289                 kfree(buf);
2290                 return ERR_PTR(err);
2291         }
2292         *ncookies = err;
2293
2294         return buf;
2295 }
2296 EXPORT_SYMBOL(ldc_alloc_exp_dring);
2297
2298 void ldc_free_exp_dring(struct ldc_channel *lp, void *buf, unsigned int len,
2299                         struct ldc_trans_cookie *cookies, int ncookies)
2300 {
2301         ldc_unmap(lp, cookies, ncookies);
2302         kfree(buf);
2303 }
2304 EXPORT_SYMBOL(ldc_free_exp_dring);
2305
2306 static int __init ldc_init(void)
2307 {
2308         struct mdesc_node *mp;
2309         unsigned long major, minor;
2310         const u64 *v;
2311
2312         mp = md_find_node_by_name(NULL, "platform");
2313         if (!mp)
2314                 return -ENODEV;
2315
2316         v = md_get_property(mp, "domaining-enabled", NULL);
2317         if (!v)
2318                 return -ENODEV;
2319
2320         major = 1;
2321         minor = 0;
2322         if (sun4v_hvapi_register(HV_GRP_LDOM, major, &minor)) {
2323                 printk(KERN_INFO PFX "Could not register LDOM hvapi.\n");
2324                 return -ENODEV;
2325         }
2326
2327         printk(KERN_INFO "%s", version);
2328
2329         if (!*v) {
2330                 printk(KERN_INFO PFX "Domaining disabled.\n");
2331                 return -ENODEV;
2332         }
2333         ldom_domaining_enabled = 1;
2334
2335         return 0;
2336 }
2337
2338 core_initcall(ldc_init);