drivers/dma/iop-adma.c

   1 /*
   2  * offload engine driver for the Intel Xscale series of i/o processors
   3  * Copyright © 2006, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along with
  15  * this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  17  *
  18  */
  19
  20 /*
  21  * This driver supports the asynchrounous DMA copy and RAID engines available
  22  * on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
  23  */
  24
  25 #include <linux/init.h>
  26 #include <linux/module.h>
  27 #include <linux/async_tx.h>
  28 #include <linux/delay.h>
  29 #include <linux/dma-mapping.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/platform_device.h>
  33 #include <linux/memory.h>
  34 #include <linux/ioport.h>
  35
  36 #include <asm/arch/adma.h>
  37
  38 #define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
  39 #define to_iop_adma_device(dev) \
  40         container_of(dev, struct iop_adma_device, common)
  41 #define tx_to_iop_adma_slot(tx) \
  42         container_of(tx, struct iop_adma_desc_slot, async_tx)
  43
  44 /**
  45  * iop_adma_free_slots - flags descriptor slots for reuse
  46  * @slot: Slot to free
  47  * Caller must hold &iop_chan->lock while calling this function
  48  */
  49 static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
  50 {
  51         int stride = slot->slots_per_op;
  52
  53         while (stride--) {
  54                 slot->slots_per_op = 0;
  55                 slot = list_entry(slot->slot_node.next,
  56                                 struct iop_adma_desc_slot,
  57                                 slot_node);
  58         }
  59 }
  60
  61 static dma_cookie_t
  62 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
  63         struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
  64 {
  65         BUG_ON(desc->async_tx.cookie < 0);
  66         if (desc->async_tx.cookie > 0) {
  67                 cookie = desc->async_tx.cookie;
  68                 desc->async_tx.cookie = 0;
  69
  70                 /* call the callback (must not sleep or submit new
  71                  * operations to this channel)
  72                  */
  73                 if (desc->async_tx.callback)
  74                         desc->async_tx.callback(
  75                                 desc->async_tx.callback_param);
  76
  77                 /* unmap dma addresses
  78                  * (unmap_single vs unmap_page?)
  79                  */
  80                 if (desc->group_head && desc->unmap_len) {
  81                         struct iop_adma_desc_slot *unmap = desc->group_head;
  82                         struct device *dev =
  83                                 &iop_chan->device->pdev->dev;
  84                         u32 len = unmap->unmap_len;
  85                         u32 src_cnt = unmap->unmap_src_cnt;
  86                         dma_addr_t addr = iop_desc_get_dest_addr(unmap,
  87                                 iop_chan);
  88
  89                         dma_unmap_page(dev, addr, len, DMA_FROM_DEVICE);
  90                         while (src_cnt--) {
  91                                 addr = iop_desc_get_src_addr(unmap,
  92                                                         iop_chan,
  93                                                         src_cnt);
  94                                 dma_unmap_page(dev, addr, len,
  95                                         DMA_TO_DEVICE);
  96                         }
  97                         desc->group_head = NULL;
  98                 }
  99         }
 100
 101         /* run dependent operations */
 102         async_tx_run_dependencies(&desc->async_tx);
 103
 104         return cookie;
 105 }
 106
 107 static int
 108 iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
 109         struct iop_adma_chan *iop_chan)
 110 {
 111         /* the client is allowed to attach dependent operations
 112          * until 'ack' is set
 113          */
 114         if (!async_tx_test_ack(&desc->async_tx))
 115                 return 0;
 116
 117         /* leave the last descriptor in the chain
 118          * so we can append to it
 119          */
 120         if (desc->chain_node.next == &iop_chan->chain)
 121                 return 1;
 122
 123         dev_dbg(iop_chan->device->common.dev,
 124                 "\tfree slot: %d slots_per_op: %d\n",
 125                 desc->idx, desc->slots_per_op);
 126
 127         list_del(&desc->chain_node);
 128         iop_adma_free_slots(desc);
 129
 130         return 0;
 131 }
 132
 133 static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
 134 {
 135         struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
 136         dma_cookie_t cookie = 0;
 137         u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
 138         int busy = iop_chan_is_busy(iop_chan);
 139         int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
 140
 141         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
 142         /* free completed slots from the chain starting with
 143          * the oldest descriptor
 144          */
 145         list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
 146                                         chain_node) {
 147                 pr_debug("\tcookie: %d slot: %d busy: %d "
 148                         "this_desc: %#x next_desc: %#x ack: %d\n",
 149                         iter->async_tx.cookie, iter->idx, busy,
 150                         iter->async_tx.phys, iop_desc_get_next_desc(iter),
 151                         async_tx_test_ack(&iter->async_tx));
 152                 prefetch(_iter);
 153                 prefetch(&_iter->async_tx);
 154
 155                 /* do not advance past the current descriptor loaded into the
 156                  * hardware channel, subsequent descriptors are either in
 157                  * process or have not been submitted
 158                  */
 159                 if (seen_current)
 160                         break;
 161
 162                 /* stop the search if we reach the current descriptor and the
 163                  * channel is busy, or if it appears that the current descriptor
 164                  * needs to be re-read (i.e. has been appended to)
 165                  */
 166                 if (iter->async_tx.phys == current_desc) {
 167                         BUG_ON(seen_current++);
 168                         if (busy || iop_desc_get_next_desc(iter))
 169                                 break;
 170                 }
 171
 172                 /* detect the start of a group transaction */
 173                 if (!slot_cnt && !slots_per_op) {
 174                         slot_cnt = iter->slot_cnt;
 175                         slots_per_op = iter->slots_per_op;
 176                         if (slot_cnt <= slots_per_op) {
 177                                 slot_cnt = 0;
 178                                 slots_per_op = 0;
 179                         }
 180                 }
 181
 182                 if (slot_cnt) {
 183                         pr_debug("\tgroup++\n");
 184                         if (!grp_start)
 185                                 grp_start = iter;
 186                         slot_cnt -= slots_per_op;
 187                 }
 188
 189                 /* all the members of a group are complete */
 190                 if (slots_per_op != 0 && slot_cnt == 0) {
 191                         struct iop_adma_desc_slot *grp_iter, *_grp_iter;
 192                         int end_of_chain = 0;
 193                         pr_debug("\tgroup end\n");
 194
 195                         /* collect the total results */
 196                         if (grp_start->xor_check_result) {
 197                                 u32 zero_sum_result = 0;
 198                                 slot_cnt = grp_start->slot_cnt;
 199                                 grp_iter = grp_start;
 200
 201                                 list_for_each_entry_from(grp_iter,
 202                                         &iop_chan->chain, chain_node) {
 203                                         zero_sum_result |=
 204                                             iop_desc_get_zero_result(grp_iter);
 205                                             pr_debug("\titer%d result: %d\n",
 206                                             grp_iter->idx, zero_sum_result);
 207                                         slot_cnt -= slots_per_op;
 208                                         if (slot_cnt == 0)
 209                                                 break;
 210                                 }
 211                                 pr_debug("\tgrp_start->xor_check_result: %p\n",
 212                                         grp_start->xor_check_result);
 213                                 *grp_start->xor_check_result = zero_sum_result;
 214                         }
 215
 216                         /* clean up the group */
 217                         slot_cnt = grp_start->slot_cnt;
 218                         grp_iter = grp_start;
 219                         list_for_each_entry_safe_from(grp_iter, _grp_iter,
 220                                 &iop_chan->chain, chain_node) {
 221                                 cookie = iop_adma_run_tx_complete_actions(
 222                                         grp_iter, iop_chan, cookie);
 223
 224                                 slot_cnt -= slots_per_op;
 225                                 end_of_chain = iop_adma_clean_slot(grp_iter,
 226                                         iop_chan);
 227
 228                                 if (slot_cnt == 0 || end_of_chain)
 229                                         break;
 230                         }
 231
 232                         /* the group should be complete at this point */
 233                         BUG_ON(slot_cnt);
 234
 235                         slots_per_op = 0;
 236                         grp_start = NULL;
 237                         if (end_of_chain)
 238                                 break;
 239                         else
 240                                 continue;
 241                 } else if (slots_per_op) /* wait for group completion */
 242                         continue;
 243
 244                 /* write back zero sum results (single descriptor case) */
 245                 if (iter->xor_check_result && iter->async_tx.cookie)
 246                         *iter->xor_check_result =
 247                                 iop_desc_get_zero_result(iter);
 248
 249                 cookie = iop_adma_run_tx_complete_actions(
 250                                         iter, iop_chan, cookie);
 251
 252                 if (iop_adma_clean_slot(iter, iop_chan))
 253                         break;
 254         }
 255
 256         BUG_ON(!seen_current);
 257
 258         if (cookie > 0) {
 259                 iop_chan->completed_cookie = cookie;
 260                 pr_debug("\tcompleted cookie %d\n", cookie);
 261         }
 262 }
 263
 264 static void
 265 iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
 266 {
 267         spin_lock_bh(&iop_chan->lock);
 268         __iop_adma_slot_cleanup(iop_chan);
 269         spin_unlock_bh(&iop_chan->lock);
 270 }
 271
 272 static void iop_adma_tasklet(unsigned long data)
 273 {
 274         struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
 275
 276         spin_lock(&iop_chan->lock);
 277         __iop_adma_slot_cleanup(iop_chan);
 278         spin_unlock(&iop_chan->lock);
 279 }
 280
 281 static struct iop_adma_desc_slot *
 282 iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
 283                         int slots_per_op)
 284 {
 285         struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
 286         LIST_HEAD(chain);
 287         int slots_found, retry = 0;
 288
 289         /* start search from the last allocated descrtiptor
 290          * if a contiguous allocation can not be found start searching
 291          * from the beginning of the list
 292          */
 293 retry:
 294         slots_found = 0;
 295         if (retry == 0)
 296                 iter = iop_chan->last_used;
 297         else
 298                 iter = list_entry(&iop_chan->all_slots,
 299                         struct iop_adma_desc_slot,
 300                         slot_node);
 301
 302         list_for_each_entry_safe_continue(
 303                 iter, _iter, &iop_chan->all_slots, slot_node) {
 304                 prefetch(_iter);
 305                 prefetch(&_iter->async_tx);
 306                 if (iter->slots_per_op) {
 307                         /* give up after finding the first busy slot
 308                          * on the second pass through the list
 309                          */
 310                         if (retry)
 311                                 break;
 312
 313                         slots_found = 0;
 314                         continue;
 315                 }
 316
 317                 /* start the allocation if the slot is correctly aligned */
 318                 if (!slots_found++) {
 319                         if (iop_desc_is_aligned(iter, slots_per_op))
 320                                 alloc_start = iter;
 321                         else {
 322                                 slots_found = 0;
 323                                 continue;
 324                         }
 325                 }
 326
 327                 if (slots_found == num_slots) {
 328                         struct iop_adma_desc_slot *alloc_tail = NULL;
 329                         struct iop_adma_desc_slot *last_used = NULL;
 330                         iter = alloc_start;
 331                         while (num_slots) {
 332                                 int i;
 333                                 dev_dbg(iop_chan->device->common.dev,
 334                                         "allocated slot: %d "
 335                                         "(desc %p phys: %#x) slots_per_op %d\n",
 336                                         iter->idx, iter->hw_desc,
 337                                         iter->async_tx.phys, slots_per_op);
 338
 339                                 /* pre-ack all but the last descriptor */
 340                                 if (num_slots != slots_per_op)
 341                                         async_tx_ack(&iter->async_tx);
 342
 343                                 list_add_tail(&iter->chain_node, &chain);
 344                                 alloc_tail = iter;
 345                                 iter->async_tx.cookie = 0;
 346                                 iter->slot_cnt = num_slots;
 347                                 iter->xor_check_result = NULL;
 348                                 for (i = 0; i < slots_per_op; i++) {
 349                                         iter->slots_per_op = slots_per_op - i;
 350                                         last_used = iter;
 351                                         iter = list_entry(iter->slot_node.next,
 352                                                 struct iop_adma_desc_slot,
 353                                                 slot_node);
 354                                 }
 355                                 num_slots -= slots_per_op;
 356                         }
 357                         alloc_tail->group_head = alloc_start;
 358                         alloc_tail->async_tx.cookie = -EBUSY;
 359                         list_splice(&chain, &alloc_tail->async_tx.tx_list);
 360                         iop_chan->last_used = last_used;
 361                         iop_desc_clear_next_desc(alloc_start);
 362                         iop_desc_clear_next_desc(alloc_tail);
 363                         return alloc_tail;
 364                 }
 365         }
 366         if (!retry++)
 367                 goto retry;
 368
 369         /* try to free some slots if the allocation fails */
 370         tasklet_schedule(&iop_chan->irq_tasklet);
 371
 372         return NULL;
 373 }
 374
 375 static dma_cookie_t
 376 iop_desc_assign_cookie(struct iop_adma_chan *iop_chan,
 377         struct iop_adma_desc_slot *desc)
 378 {
 379         dma_cookie_t cookie = iop_chan->common.cookie;
 380         cookie++;
 381         if (cookie < 0)
 382                 cookie = 1;
 383         iop_chan->common.cookie = desc->async_tx.cookie = cookie;
 384         return cookie;
 385 }
 386
 387 static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
 388 {
 389         dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
 390                 iop_chan->pending);
 391
 392         if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
 393                 iop_chan->pending = 0;
 394                 iop_chan_append(iop_chan);
 395         }
 396 }
 397
 398 static dma_cookie_t
 399 iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
 400 {
 401         struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
 402         struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
 403         struct iop_adma_desc_slot *grp_start, *old_chain_tail;
 404         int slot_cnt;
 405         int slots_per_op;
 406         dma_cookie_t cookie;
 407
 408         grp_start = sw_desc->group_head;
 409         slot_cnt = grp_start->slot_cnt;
 410         slots_per_op = grp_start->slots_per_op;
 411
 412         spin_lock_bh(&iop_chan->lock);
 413         cookie = iop_desc_assign_cookie(iop_chan, sw_desc);
 414
 415         old_chain_tail = list_entry(iop_chan->chain.prev,
 416                 struct iop_adma_desc_slot, chain_node);
 417         list_splice_init(&sw_desc->async_tx.tx_list,
 418                          &old_chain_tail->chain_node);
 419
 420         /* fix up the hardware chain */
 421         iop_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
 422
 423         /* 1/ don't add pre-chained descriptors
 424          * 2/ dummy read to flush next_desc write
 425          */
 426         BUG_ON(iop_desc_get_next_desc(sw_desc));
 427
 428         /* increment the pending count by the number of slots
 429          * memcpy operations have a 1:1 (slot:operation) relation
 430          * other operations are heavier and will pop the threshold
 431          * more often.
 432          */
 433         iop_chan->pending += slot_cnt;
 434         iop_adma_check_threshold(iop_chan);
 435         spin_unlock_bh(&iop_chan->lock);
 436
 437         dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
 438                 __func__, sw_desc->async_tx.cookie, sw_desc->idx);
 439
 440         return cookie;
 441 }
 442
 443 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
 444 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
 445
 446 /* returns the number of allocated descriptors */
 447 static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
 448 {
 449         char *hw_desc;
 450         int idx;
 451         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 452         struct iop_adma_desc_slot *slot = NULL;
 453         int init = iop_chan->slots_allocated ? 0 : 1;
 454         struct iop_adma_platform_data *plat_data =
 455                 iop_chan->device->pdev->dev.platform_data;
 456         int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
 457
 458         /* Allocate descriptor slots */
 459         do {
 460                 idx = iop_chan->slots_allocated;
 461                 if (idx == num_descs_in_pool)
 462                         break;
 463
 464                 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
 465                 if (!slot) {
 466                         printk(KERN_INFO "IOP ADMA Channel only initialized"
 467                                 " %d descriptor slots", idx);
 468                         break;
 469                 }
 470                 hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
 471                 slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
 472
 473                 dma_async_tx_descriptor_init(&slot->async_tx, chan);
 474                 slot->async_tx.tx_submit = iop_adma_tx_submit;
 475                 INIT_LIST_HEAD(&slot->chain_node);
 476                 INIT_LIST_HEAD(&slot->slot_node);
 477                 INIT_LIST_HEAD(&slot->async_tx.tx_list);
 478                 hw_desc = (char *) iop_chan->device->dma_desc_pool;
 479                 slot->async_tx.phys =
 480                         (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
 481                 slot->idx = idx;
 482
 483                 spin_lock_bh(&iop_chan->lock);
 484                 iop_chan->slots_allocated++;
 485                 list_add_tail(&slot->slot_node, &iop_chan->all_slots);
 486                 spin_unlock_bh(&iop_chan->lock);
 487         } while (iop_chan->slots_allocated < num_descs_in_pool);
 488
 489         if (idx && !iop_chan->last_used)
 490                 iop_chan->last_used = list_entry(iop_chan->all_slots.next,
 491                                         struct iop_adma_desc_slot,
 492                                         slot_node);
 493
 494         dev_dbg(iop_chan->device->common.dev,
 495                 "allocated %d descriptor slots last_used: %p\n",
 496                 iop_chan->slots_allocated, iop_chan->last_used);
 497
 498         /* initialize the channel and the chain with a null operation */
 499         if (init) {
 500                 if (dma_has_cap(DMA_MEMCPY,
 501                         iop_chan->device->common.cap_mask))
 502                         iop_chan_start_null_memcpy(iop_chan);
 503                 else if (dma_has_cap(DMA_XOR,
 504                         iop_chan->device->common.cap_mask))
 505                         iop_chan_start_null_xor(iop_chan);
 506                 else
 507                         BUG();
 508         }
 509
 510         return (idx > 0) ? idx : -ENOMEM;
 511 }
 512
 513 static struct dma_async_tx_descriptor *
 514 iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags)
 515 {
 516         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 517         struct iop_adma_desc_slot *sw_desc, *grp_start;
 518         int slot_cnt, slots_per_op;
 519
 520         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
 521
 522         spin_lock_bh(&iop_chan->lock);
 523         slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
 524         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
 525         if (sw_desc) {
 526                 grp_start = sw_desc->group_head;
 527                 iop_desc_init_interrupt(grp_start, iop_chan);
 528                 grp_start->unmap_len = 0;
 529                 sw_desc->async_tx.flags = flags;
 530         }
 531         spin_unlock_bh(&iop_chan->lock);
 532
 533         return sw_desc ? &sw_desc->async_tx : NULL;
 534 }
 535
 536 static struct dma_async_tx_descriptor *
 537 iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
 538                          dma_addr_t dma_src, size_t len, unsigned long flags)
 539 {
 540         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 541         struct iop_adma_desc_slot *sw_desc, *grp_start;
 542         int slot_cnt, slots_per_op;
 543
 544         if (unlikely(!len))
 545                 return NULL;
 546         BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
 547
 548         dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
 549                 __func__, len);
 550
 551         spin_lock_bh(&iop_chan->lock);
 552         slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
 553         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
 554         if (sw_desc) {
 555                 grp_start = sw_desc->group_head;
 556                 iop_desc_init_memcpy(grp_start, flags);
 557                 iop_desc_set_byte_count(grp_start, iop_chan, len);
 558                 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
 559                 iop_desc_set_memcpy_src_addr(grp_start, dma_src);
 560                 sw_desc->unmap_src_cnt = 1;
 561                 sw_desc->unmap_len = len;
 562                 sw_desc->async_tx.flags = flags;
 563         }
 564         spin_unlock_bh(&iop_chan->lock);
 565
 566         return sw_desc ? &sw_desc->async_tx : NULL;
 567 }
 568
 569 static struct dma_async_tx_descriptor *
 570 iop_adma_prep_dma_memset(struct dma_chan *chan, dma_addr_t dma_dest,
 571                          int value, size_t len, unsigned long flags)
 572 {
 573         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 574         struct iop_adma_desc_slot *sw_desc, *grp_start;
 575         int slot_cnt, slots_per_op;
 576
 577         if (unlikely(!len))
 578                 return NULL;
 579         BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
 580
 581         dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
 582                 __func__, len);
 583
 584         spin_lock_bh(&iop_chan->lock);
 585         slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
 586         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
 587         if (sw_desc) {
 588                 grp_start = sw_desc->group_head;
 589                 iop_desc_init_memset(grp_start, flags);
 590                 iop_desc_set_byte_count(grp_start, iop_chan, len);
 591                 iop_desc_set_block_fill_val(grp_start, value);
 592                 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
 593                 sw_desc->unmap_src_cnt = 1;
 594                 sw_desc->unmap_len = len;
 595                 sw_desc->async_tx.flags = flags;
 596         }
 597         spin_unlock_bh(&iop_chan->lock);
 598
 599         return sw_desc ? &sw_desc->async_tx : NULL;
 600 }
 601
 602 static struct dma_async_tx_descriptor *
 603 iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest,
 604                       dma_addr_t *dma_src, unsigned int src_cnt, size_t len,
 605                       unsigned long flags)
 606 {
 607         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 608         struct iop_adma_desc_slot *sw_desc, *grp_start;
 609         int slot_cnt, slots_per_op;
 610
 611         if (unlikely(!len))
 612                 return NULL;
 613         BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT));
 614
 615         dev_dbg(iop_chan->device->common.dev,
 616                 "%s src_cnt: %d len: %u flags: %lx\n",
 617                 __func__, src_cnt, len, flags);
 618
 619         spin_lock_bh(&iop_chan->lock);
 620         slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
 621         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
 622         if (sw_desc) {
 623                 grp_start = sw_desc->group_head;
 624                 iop_desc_init_xor(grp_start, src_cnt, flags);
 625                 iop_desc_set_byte_count(grp_start, iop_chan, len);
 626                 iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest);
 627                 sw_desc->unmap_src_cnt = src_cnt;
 628                 sw_desc->unmap_len = len;
 629                 sw_desc->async_tx.flags = flags;
 630                 while (src_cnt--)
 631                         iop_desc_set_xor_src_addr(grp_start, src_cnt,
 632                                                   dma_src[src_cnt]);
 633         }
 634         spin_unlock_bh(&iop_chan->lock);
 635
 636         return sw_desc ? &sw_desc->async_tx : NULL;
 637 }
 638
 639 static struct dma_async_tx_descriptor *
 640 iop_adma_prep_dma_zero_sum(struct dma_chan *chan, dma_addr_t *dma_src,
 641                            unsigned int src_cnt, size_t len, u32 *result,
 642                            unsigned long flags)
 643 {
 644         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 645         struct iop_adma_desc_slot *sw_desc, *grp_start;
 646         int slot_cnt, slots_per_op;
 647
 648         if (unlikely(!len))
 649                 return NULL;
 650
 651         dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
 652                 __func__, src_cnt, len);
 653
 654         spin_lock_bh(&iop_chan->lock);
 655         slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
 656         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
 657         if (sw_desc) {
 658                 grp_start = sw_desc->group_head;
 659                 iop_desc_init_zero_sum(grp_start, src_cnt, flags);
 660                 iop_desc_set_zero_sum_byte_count(grp_start, len);
 661                 grp_start->xor_check_result = result;
 662                 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
 663                         __func__, grp_start->xor_check_result);
 664                 sw_desc->unmap_src_cnt = src_cnt;
 665                 sw_desc->unmap_len = len;
 666                 sw_desc->async_tx.flags = flags;
 667                 while (src_cnt--)
 668                         iop_desc_set_zero_sum_src_addr(grp_start, src_cnt,
 669                                                        dma_src[src_cnt]);
 670         }
 671         spin_unlock_bh(&iop_chan->lock);
 672
 673         return sw_desc ? &sw_desc->async_tx : NULL;
 674 }
 675
 676 static void iop_adma_free_chan_resources(struct dma_chan *chan)
 677 {
 678         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 679         struct iop_adma_desc_slot *iter, *_iter;
 680         int in_use_descs = 0;
 681
 682         iop_adma_slot_cleanup(iop_chan);
 683
 684         spin_lock_bh(&iop_chan->lock);
 685         list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
 686                                         chain_node) {
 687                 in_use_descs++;
 688                 list_del(&iter->chain_node);
 689         }
 690         list_for_each_entry_safe_reverse(
 691                 iter, _iter, &iop_chan->all_slots, slot_node) {
 692                 list_del(&iter->slot_node);
 693                 kfree(iter);
 694                 iop_chan->slots_allocated--;
 695         }
 696         iop_chan->last_used = NULL;
 697
 698         dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
 699                 __func__, iop_chan->slots_allocated);
 700         spin_unlock_bh(&iop_chan->lock);
 701
 702         /* one is ok since we left it on there on purpose */
 703         if (in_use_descs > 1)
 704                 printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
 705                         in_use_descs - 1);
 706 }
 707
 708 /**
 709  * iop_adma_is_complete - poll the status of an ADMA transaction
 710  * @chan: ADMA channel handle
 711  * @cookie: ADMA transaction identifier
 712  */
 713 static enum dma_status iop_adma_is_complete(struct dma_chan *chan,
 714                                         dma_cookie_t cookie,
 715                                         dma_cookie_t *done,
 716                                         dma_cookie_t *used)
 717 {
 718         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 719         dma_cookie_t last_used;
 720         dma_cookie_t last_complete;
 721         enum dma_status ret;
 722
 723         last_used = chan->cookie;
 724         last_complete = iop_chan->completed_cookie;
 725
 726         if (done)
 727                 *done = last_complete;
 728         if (used)
 729                 *used = last_used;
 730
 731         ret = dma_async_is_complete(cookie, last_complete, last_used);
 732         if (ret == DMA_SUCCESS)
 733                 return ret;
 734
 735         iop_adma_slot_cleanup(iop_chan);
 736
 737         last_used = chan->cookie;
 738         last_complete = iop_chan->completed_cookie;
 739
 740         if (done)
 741                 *done = last_complete;
 742         if (used)
 743                 *used = last_used;
 744
 745         return dma_async_is_complete(cookie, last_complete, last_used);
 746 }
 747
 748 static irqreturn_t iop_adma_eot_handler(int irq, void *data)
 749 {
 750         struct iop_adma_chan *chan = data;
 751
 752         dev_dbg(chan->device->common.dev, "%s\n", __func__);
 753
 754         tasklet_schedule(&chan->irq_tasklet);
 755
 756         iop_adma_device_clear_eot_status(chan);
 757
 758         return IRQ_HANDLED;
 759 }
 760
 761 static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
 762 {
 763         struct iop_adma_chan *chan = data;
 764
 765         dev_dbg(chan->device->common.dev, "%s\n", __func__);
 766
 767         tasklet_schedule(&chan->irq_tasklet);
 768
 769         iop_adma_device_clear_eoc_status(chan);
 770
 771         return IRQ_HANDLED;
 772 }
 773
 774 static irqreturn_t iop_adma_err_handler(int irq, void *data)
 775 {
 776         struct iop_adma_chan *chan = data;
 777         unsigned long status = iop_chan_get_status(chan);
 778
 779         dev_printk(KERN_ERR, chan->device->common.dev,
 780                 "error ( %s%s%s%s%s%s%s)\n",
 781                 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
 782                 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
 783                 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
 784                 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
 785                 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
 786                 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
 787                 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
 788
 789         iop_adma_device_clear_err_status(chan);
 790
 791         BUG();
 792
 793         return IRQ_HANDLED;
 794 }
 795
 796 static void iop_adma_issue_pending(struct dma_chan *chan)
 797 {
 798         struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
 799
 800         if (iop_chan->pending) {
 801                 iop_chan->pending = 0;
 802                 iop_chan_append(iop_chan);
 803         }
 804 }
 805
 806 /*
 807  * Perform a transaction to verify the HW works.
 808  */
 809 #define IOP_ADMA_TEST_SIZE 2000
 810
 811 static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
 812 {
 813         int i;
 814         void *src, *dest;
 815         dma_addr_t src_dma, dest_dma;
 816         struct dma_chan *dma_chan;
 817         dma_cookie_t cookie;
 818         struct dma_async_tx_descriptor *tx;
 819         int err = 0;
 820         struct iop_adma_chan *iop_chan;
 821
 822         dev_dbg(device->common.dev, "%s\n", __func__);
 823
 824         src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
 825         if (!src)
 826                 return -ENOMEM;
 827         dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL);
 828         if (!dest) {
 829                 kfree(src);
 830                 return -ENOMEM;
 831         }
 832
 833         /* Fill in src buffer */
 834         for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
 835                 ((u8 *) src)[i] = (u8)i;
 836
 837         /* Start copy, using first DMA channel */
 838         dma_chan = container_of(device->common.channels.next,
 839                                 struct dma_chan,
 840                                 device_node);
 841         if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
 842                 err = -ENODEV;
 843                 goto out;
 844         }
 845
 846         dest_dma = dma_map_single(dma_chan->device->dev, dest,
 847                                 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
 848         src_dma = dma_map_single(dma_chan->device->dev, src,
 849                                 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
 850         tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma,
 851                                       IOP_ADMA_TEST_SIZE,
 852                                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
 853
 854         cookie = iop_adma_tx_submit(tx);
 855         iop_adma_issue_pending(dma_chan);
 856         msleep(1);
 857
 858         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
 859                         DMA_SUCCESS) {
 860                 dev_printk(KERN_ERR, dma_chan->device->dev,
 861                         "Self-test copy timed out, disabling\n");
 862                 err = -ENODEV;
 863                 goto free_resources;
 864         }
 865
 866         iop_chan = to_iop_adma_chan(dma_chan);
 867         dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
 868                 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
 869         if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
 870                 dev_printk(KERN_ERR, dma_chan->device->dev,
 871                         "Self-test copy failed compare, disabling\n");
 872                 err = -ENODEV;
 873                 goto free_resources;
 874         }
 875
 876 free_resources:
 877         iop_adma_free_chan_resources(dma_chan);
 878 out:
 879         kfree(src);
 880         kfree(dest);
 881         return err;
 882 }
 883
 884 #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
 885 static int __devinit
 886 iop_adma_xor_zero_sum_self_test(struct iop_adma_device *device)
 887 {
 888         int i, src_idx;
 889         struct page *dest;
 890         struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
 891         struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
 892         dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
 893         dma_addr_t dma_addr, dest_dma;
 894         struct dma_async_tx_descriptor *tx;
 895         struct dma_chan *dma_chan;
 896         dma_cookie_t cookie;
 897         u8 cmp_byte = 0;
 898         u32 cmp_word;
 899         u32 zero_sum_result;
 900         int err = 0;
 901         struct iop_adma_chan *iop_chan;
 902
 903         dev_dbg(device->common.dev, "%s\n", __func__);
 904
 905         for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
 906                 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
 907                 if (!xor_srcs[src_idx])
 908                         while (src_idx--) {
 909                                 __free_page(xor_srcs[src_idx]);
 910                                 return -ENOMEM;
 911                         }
 912         }
 913
 914         dest = alloc_page(GFP_KERNEL);
 915         if (!dest)
 916                 while (src_idx--) {
 917                         __free_page(xor_srcs[src_idx]);
 918                         return -ENOMEM;
 919                 }
 920
 921         /* Fill in src buffers */
 922         for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
 923                 u8 *ptr = page_address(xor_srcs[src_idx]);
 924                 for (i = 0; i < PAGE_SIZE; i++)
 925                         ptr[i] = (1 << src_idx);
 926         }
 927
 928         for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
 929                 cmp_byte ^= (u8) (1 << src_idx);
 930
 931         cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
 932                         (cmp_byte << 8) | cmp_byte;
 933
 934         memset(page_address(dest), 0, PAGE_SIZE);
 935
 936         dma_chan = container_of(device->common.channels.next,
 937                                 struct dma_chan,
 938                                 device_node);
 939         if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
 940                 err = -ENODEV;
 941                 goto out;
 942         }
 943
 944         /* test xor */
 945         dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
 946                                 PAGE_SIZE, DMA_FROM_DEVICE);
 947         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
 948                 dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i],
 949                                            0, PAGE_SIZE, DMA_TO_DEVICE);
 950         tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs,
 951                                    IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE,
 952                                    DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
 953
 954         cookie = iop_adma_tx_submit(tx);
 955         iop_adma_issue_pending(dma_chan);
 956         msleep(8);
 957
 958         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
 959                 DMA_SUCCESS) {
 960                 dev_printk(KERN_ERR, dma_chan->device->dev,
 961                         "Self-test xor timed out, disabling\n");
 962                 err = -ENODEV;
 963                 goto free_resources;
 964         }
 965
 966         iop_chan = to_iop_adma_chan(dma_chan);
 967         dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
 968                 PAGE_SIZE, DMA_FROM_DEVICE);
 969         for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
 970                 u32 *ptr = page_address(dest);
 971                 if (ptr[i] != cmp_word) {
 972                         dev_printk(KERN_ERR, dma_chan->device->dev,
 973                                 "Self-test xor failed compare, disabling\n");
 974                         err = -ENODEV;
 975                         goto free_resources;
 976                 }
 977         }
 978         dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
 979                 PAGE_SIZE, DMA_TO_DEVICE);
 980
 981         /* skip zero sum if the capability is not present */
 982         if (!dma_has_cap(DMA_ZERO_SUM, dma_chan->device->cap_mask))
 983                 goto free_resources;
 984
 985         /* zero sum the sources with the destintation page */
 986         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
 987                 zero_sum_srcs[i] = xor_srcs[i];
 988         zero_sum_srcs[i] = dest;
 989
 990         zero_sum_result = 1;
 991
 992         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
 993                 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
 994                                            zero_sum_srcs[i], 0, PAGE_SIZE,
 995                                            DMA_TO_DEVICE);
 996         tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs,
 997                                         IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
 998                                         &zero_sum_result,
 999                                         DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1000
1001         cookie = iop_adma_tx_submit(tx);
1002         iop_adma_issue_pending(dma_chan);
1003         msleep(8);
1004
1005         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1006                 dev_printk(KERN_ERR, dma_chan->device->dev,
1007                         "Self-test zero sum timed out, disabling\n");
1008                 err = -ENODEV;
1009                 goto free_resources;
1010         }
1011
1012         if (zero_sum_result != 0) {
1013                 dev_printk(KERN_ERR, dma_chan->device->dev,
1014                         "Self-test zero sum failed compare, disabling\n");
1015                 err = -ENODEV;
1016                 goto free_resources;
1017         }
1018
1019         /* test memset */
1020         dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1021                         PAGE_SIZE, DMA_FROM_DEVICE);
1022         tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE,
1023                                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1024
1025         cookie = iop_adma_tx_submit(tx);
1026         iop_adma_issue_pending(dma_chan);
1027         msleep(8);
1028
1029         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1030                 dev_printk(KERN_ERR, dma_chan->device->dev,
1031                         "Self-test memset timed out, disabling\n");
1032                 err = -ENODEV;
1033                 goto free_resources;
1034         }
1035
1036         for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1037                 u32 *ptr = page_address(dest);
1038                 if (ptr[i]) {
1039                         dev_printk(KERN_ERR, dma_chan->device->dev,
1040                                 "Self-test memset failed compare, disabling\n");
1041                         err = -ENODEV;
1042                         goto free_resources;
1043                 }
1044         }
1045
1046         /* test for non-zero parity sum */
1047         zero_sum_result = 0;
1048         for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++)
1049                 dma_srcs[i] = dma_map_page(dma_chan->device->dev,
1050                                            zero_sum_srcs[i], 0, PAGE_SIZE,
1051                                            DMA_TO_DEVICE);
1052         tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs,
1053                                         IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE,
1054                                         &zero_sum_result,
1055                                         DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
1056
1057         cookie = iop_adma_tx_submit(tx);
1058         iop_adma_issue_pending(dma_chan);
1059         msleep(8);
1060
1061         if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1062                 dev_printk(KERN_ERR, dma_chan->device->dev,
1063                         "Self-test non-zero sum timed out, disabling\n");
1064                 err = -ENODEV;
1065                 goto free_resources;
1066         }
1067
1068         if (zero_sum_result != 1) {
1069                 dev_printk(KERN_ERR, dma_chan->device->dev,
1070                         "Self-test non-zero sum failed compare, disabling\n");
1071                 err = -ENODEV;
1072                 goto free_resources;
1073         }
1074
1075 free_resources:
1076         iop_adma_free_chan_resources(dma_chan);
1077 out:
1078         src_idx = IOP_ADMA_NUM_SRC_TEST;
1079         while (src_idx--)
1080                 __free_page(xor_srcs[src_idx]);
1081         __free_page(dest);
1082         return err;
1083 }
1084
1085 static int __devexit iop_adma_remove(struct platform_device *dev)
1086 {
1087         struct iop_adma_device *device = platform_get_drvdata(dev);
1088         struct dma_chan *chan, *_chan;
1089         struct iop_adma_chan *iop_chan;
1090         int i;
1091         struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1092
1093         dma_async_device_unregister(&device->common);
1094
1095         for (i = 0; i < 3; i++) {
1096                 unsigned int irq;
1097                 irq = platform_get_irq(dev, i);
1098                 free_irq(irq, device);
1099         }
1100
1101         dma_free_coherent(&dev->dev, plat_data->pool_size,
1102                         device->dma_desc_pool_virt, device->dma_desc_pool);
1103
1104         do {
1105                 struct resource *res;
1106                 res = platform_get_resource(dev, IORESOURCE_MEM, 0);
1107                 release_mem_region(res->start, res->end - res->start);
1108         } while (0);
1109
1110         list_for_each_entry_safe(chan, _chan, &device->common.channels,
1111                                 device_node) {
1112                 iop_chan = to_iop_adma_chan(chan);
1113                 list_del(&chan->device_node);
1114                 kfree(iop_chan);
1115         }
1116         kfree(device);
1117
1118         return 0;
1119 }
1120
1121 static int __devinit iop_adma_probe(struct platform_device *pdev)
1122 {
1123         struct resource *res;
1124         int ret = 0, i;
1125         struct iop_adma_device *adev;
1126         struct iop_adma_chan *iop_chan;
1127         struct dma_device *dma_dev;
1128         struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1129
1130         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1131         if (!res)
1132                 return -ENODEV;
1133
1134         if (!devm_request_mem_region(&pdev->dev, res->start,
1135                                 res->end - res->start, pdev->name))
1136                 return -EBUSY;
1137
1138         adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1139         if (!adev)
1140                 return -ENOMEM;
1141         dma_dev = &adev->common;
1142
1143         /* allocate coherent memory for hardware descriptors
1144          * note: writecombine gives slightly better performance, but
1145          * requires that we explicitly flush the writes
1146          */
1147         if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1148                                         plat_data->pool_size,
1149                                         &adev->dma_desc_pool,
1150                                         GFP_KERNEL)) == NULL) {
1151                 ret = -ENOMEM;
1152                 goto err_free_adev;
1153         }
1154
1155         dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
1156                 __func__, adev->dma_desc_pool_virt,
1157                 (void *) adev->dma_desc_pool);
1158
1159         adev->id = plat_data->hw_id;
1160
1161         /* discover transaction capabilites from the platform data */
1162         dma_dev->cap_mask = plat_data->cap_mask;
1163
1164         adev->pdev = pdev;
1165         platform_set_drvdata(pdev, adev);
1166
1167         INIT_LIST_HEAD(&dma_dev->channels);
1168
1169         /* set base routines */
1170         dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1171         dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1172         dma_dev->device_is_tx_complete = iop_adma_is_complete;
1173         dma_dev->device_issue_pending = iop_adma_issue_pending;
1174         dma_dev->dev = &pdev->dev;
1175
1176         /* set prep routines based on capability */
1177         if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1178                 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1179         if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1180                 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1181         if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1182                 dma_dev->max_xor = iop_adma_get_max_xor();
1183                 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1184         }
1185         if (dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask))
1186                 dma_dev->device_prep_dma_zero_sum =
1187                         iop_adma_prep_dma_zero_sum;
1188         if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1189                 dma_dev->device_prep_dma_interrupt =
1190                         iop_adma_prep_dma_interrupt;
1191
1192         iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1193         if (!iop_chan) {
1194                 ret = -ENOMEM;
1195                 goto err_free_dma;
1196         }
1197         iop_chan->device = adev;
1198
1199         iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1200                                         res->end - res->start);
1201         if (!iop_chan->mmr_base) {
1202                 ret = -ENOMEM;
1203                 goto err_free_iop_chan;
1204         }
1205         tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1206                 iop_chan);
1207
1208         /* clear errors before enabling interrupts */
1209         iop_adma_device_clear_err_status(iop_chan);
1210
1211         for (i = 0; i < 3; i++) {
1212                 irq_handler_t handler[] = { iop_adma_eot_handler,
1213                                         iop_adma_eoc_handler,
1214                                         iop_adma_err_handler };
1215                 int irq = platform_get_irq(pdev, i);
1216                 if (irq < 0) {
1217                         ret = -ENXIO;
1218                         goto err_free_iop_chan;
1219                 } else {
1220                         ret = devm_request_irq(&pdev->dev, irq,
1221                                         handler[i], 0, pdev->name, iop_chan);
1222                         if (ret)
1223                                 goto err_free_iop_chan;
1224                 }
1225         }
1226
1227         spin_lock_init(&iop_chan->lock);
1228         INIT_LIST_HEAD(&iop_chan->chain);
1229         INIT_LIST_HEAD(&iop_chan->all_slots);
1230         INIT_RCU_HEAD(&iop_chan->common.rcu);
1231         iop_chan->common.device = dma_dev;
1232         list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1233
1234         if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1235                 ret = iop_adma_memcpy_self_test(adev);
1236                 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1237                 if (ret)
1238                         goto err_free_iop_chan;
1239         }
1240
1241         if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1242                 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1243                 ret = iop_adma_xor_zero_sum_self_test(adev);
1244                 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1245                 if (ret)
1246                         goto err_free_iop_chan;
1247         }
1248
1249         dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
1250           "( %s%s%s%s%s%s%s%s%s%s)\n",
1251           dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "",
1252           dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "",
1253           dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "",
1254           dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1255           dma_has_cap(DMA_DUAL_XOR, dma_dev->cap_mask) ? "dual_xor " : "",
1256           dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask) ? "xor_zero_sum " : "",
1257           dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)  ? "fill " : "",
1258           dma_has_cap(DMA_MEMCPY_CRC32C, dma_dev->cap_mask) ? "cpy+crc " : "",
1259           dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1260           dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1261
1262         dma_async_device_register(dma_dev);
1263         goto out;
1264
1265  err_free_iop_chan:
1266         kfree(iop_chan);
1267  err_free_dma:
1268         dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1269                         adev->dma_desc_pool_virt, adev->dma_desc_pool);
1270  err_free_adev:
1271         kfree(adev);
1272  out:
1273         return ret;
1274 }
1275
1276 static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1277 {
1278         struct iop_adma_desc_slot *sw_desc, *grp_start;
1279         dma_cookie_t cookie;
1280         int slot_cnt, slots_per_op;
1281
1282         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1283
1284         spin_lock_bh(&iop_chan->lock);
1285         slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1286         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1287         if (sw_desc) {
1288                 grp_start = sw_desc->group_head;
1289
1290                 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1291                 async_tx_ack(&sw_desc->async_tx);
1292                 iop_desc_init_memcpy(grp_start, 0);
1293                 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1294                 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1295                 iop_desc_set_memcpy_src_addr(grp_start, 0);
1296
1297                 cookie = iop_chan->common.cookie;
1298                 cookie++;
1299                 if (cookie <= 1)
1300                         cookie = 2;
1301
1302                 /* initialize the completed cookie to be less than
1303                  * the most recently used cookie
1304                  */
1305                 iop_chan->completed_cookie = cookie - 1;
1306                 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1307
1308                 /* channel should not be busy */
1309                 BUG_ON(iop_chan_is_busy(iop_chan));
1310
1311                 /* clear any prior error-status bits */
1312                 iop_adma_device_clear_err_status(iop_chan);
1313
1314                 /* disable operation */
1315                 iop_chan_disable(iop_chan);
1316
1317                 /* set the descriptor address */
1318                 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1319
1320                 /* 1/ don't add pre-chained descriptors
1321                  * 2/ dummy read to flush next_desc write
1322                  */
1323                 BUG_ON(iop_desc_get_next_desc(sw_desc));
1324
1325                 /* run the descriptor */
1326                 iop_chan_enable(iop_chan);
1327         } else
1328                 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1329                          "failed to allocate null descriptor\n");
1330         spin_unlock_bh(&iop_chan->lock);
1331 }
1332
1333 static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1334 {
1335         struct iop_adma_desc_slot *sw_desc, *grp_start;
1336         dma_cookie_t cookie;
1337         int slot_cnt, slots_per_op;
1338
1339         dev_dbg(iop_chan->device->common.dev, "%s\n", __func__);
1340
1341         spin_lock_bh(&iop_chan->lock);
1342         slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1343         sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1344         if (sw_desc) {
1345                 grp_start = sw_desc->group_head;
1346                 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1347                 async_tx_ack(&sw_desc->async_tx);
1348                 iop_desc_init_null_xor(grp_start, 2, 0);
1349                 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1350                 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1351                 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1352                 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1353
1354                 cookie = iop_chan->common.cookie;
1355                 cookie++;
1356                 if (cookie <= 1)
1357                         cookie = 2;
1358
1359                 /* initialize the completed cookie to be less than
1360                  * the most recently used cookie
1361                  */
1362                 iop_chan->completed_cookie = cookie - 1;
1363                 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1364
1365                 /* channel should not be busy */
1366                 BUG_ON(iop_chan_is_busy(iop_chan));
1367
1368                 /* clear any prior error-status bits */
1369                 iop_adma_device_clear_err_status(iop_chan);
1370
1371                 /* disable operation */
1372                 iop_chan_disable(iop_chan);
1373
1374                 /* set the descriptor address */
1375                 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1376
1377                 /* 1/ don't add pre-chained descriptors
1378                  * 2/ dummy read to flush next_desc write
1379                  */
1380                 BUG_ON(iop_desc_get_next_desc(sw_desc));
1381
1382                 /* run the descriptor */
1383                 iop_chan_enable(iop_chan);
1384         } else
1385                 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1386                         "failed to allocate null descriptor\n");
1387         spin_unlock_bh(&iop_chan->lock);
1388 }
1389
1390 static struct platform_driver iop_adma_driver = {
1391         .probe          = iop_adma_probe,
1392         .remove         = iop_adma_remove,
1393         .driver         = {
1394                 .owner  = THIS_MODULE,
1395                 .name   = "iop-adma",
1396         },
1397 };
1398
1399 static int __init iop_adma_init (void)
1400 {
1401         return platform_driver_register(&iop_adma_driver);
1402 }
1403
1404 /* it's currently unsafe to unload this module */
1405 #if 0
1406 static void __exit iop_adma_exit (void)
1407 {
1408         platform_driver_unregister(&iop_adma_driver);
1409         return;
1410 }
1411 module_exit(iop_adma_exit);
1412 #endif
1413
1414 module_init(iop_adma_init);
1415
1416 MODULE_AUTHOR("Intel Corporation");
1417 MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1418 MODULE_LICENSE("GPL");