drivers/gpu/drm/omapdrm/omap_dmm_tiler.c

   1 /*
   2  * DMM IOMMU driver support functions for TI OMAP processors.
   3  *
   4  * Copyright (C) 2011 Texas Instruments Incorporated - http://www.ti.com/
   5  * Author: Rob Clark <rob@ti.com>
   6  *         Andy Gross <andy.gross@ti.com>
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License as
  10  * published by the Free Software Foundation version 2.
  11  *
  12  * This program is distributed "as is" WITHOUT ANY WARRANTY of any
  13  * kind, whether express or implied; without even the implied warranty
  14  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  */
  17
  18 #include <linux/completion.h>
  19 #include <linux/delay.h>
  20 #include <linux/dma-mapping.h>
  21 #include <linux/errno.h>
  22 #include <linux/init.h>
  23 #include <linux/interrupt.h>
  24 #include <linux/list.h>
  25 #include <linux/mm.h>
  26 #include <linux/module.h>
  27 #include <linux/platform_device.h> /* platform_device() */
  28 #include <linux/sched.h>
  29 #include <linux/seq_file.h>
  30 #include <linux/slab.h>
  31 #include <linux/time.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/wait.h>
  34
  35 #include "omap_dmm_tiler.h"
  36 #include "omap_dmm_priv.h"
  37
  38 #define DMM_DRIVER_NAME "dmm"
  39
  40 /* mappings for associating views to luts */
  41 static struct tcm *containers[TILFMT_NFORMATS];
  42 static struct dmm *omap_dmm;
  43
  44 #if defined(CONFIG_OF)
  45 static const struct of_device_id dmm_of_match[];
  46 #endif
  47
  48 /* global spinlock for protecting lists */
  49 static DEFINE_SPINLOCK(list_lock);
  50
  51 /* Geometry table */
  52 #define GEOM(xshift, yshift, bytes_per_pixel) { \
  53                 .x_shft = (xshift), \
  54                 .y_shft = (yshift), \
  55                 .cpp    = (bytes_per_pixel), \
  56                 .slot_w = 1 << (SLOT_WIDTH_BITS - (xshift)), \
  57                 .slot_h = 1 << (SLOT_HEIGHT_BITS - (yshift)), \
  58         }
  59
  60 static const struct {
  61         u32 x_shft;     /* unused X-bits (as part of bpp) */
  62         u32 y_shft;     /* unused Y-bits (as part of bpp) */
  63         u32 cpp;                /* bytes/chars per pixel */
  64         u32 slot_w;     /* width of each slot (in pixels) */
  65         u32 slot_h;     /* height of each slot (in pixels) */
  66 } geom[TILFMT_NFORMATS] = {
  67         [TILFMT_8BIT]  = GEOM(0, 0, 1),
  68         [TILFMT_16BIT] = GEOM(0, 1, 2),
  69         [TILFMT_32BIT] = GEOM(1, 1, 4),
  70         [TILFMT_PAGE]  = GEOM(SLOT_WIDTH_BITS, SLOT_HEIGHT_BITS, 1),
  71 };
  72
  73
  74 /* lookup table for registers w/ per-engine instances */
  75 static const u32 reg[][4] = {
  76         [PAT_STATUS] = {DMM_PAT_STATUS__0, DMM_PAT_STATUS__1,
  77                         DMM_PAT_STATUS__2, DMM_PAT_STATUS__3},
  78         [PAT_DESCR]  = {DMM_PAT_DESCR__0, DMM_PAT_DESCR__1,
  79                         DMM_PAT_DESCR__2, DMM_PAT_DESCR__3},
  80 };
  81
  82 static u32 dmm_read(struct dmm *dmm, u32 reg)
  83 {
  84         return readl(dmm->base + reg);
  85 }
  86
  87 static void dmm_write(struct dmm *dmm, u32 val, u32 reg)
  88 {
  89         writel(val, dmm->base + reg);
  90 }
  91
  92 /* simple allocator to grab next 16 byte aligned memory from txn */
  93 static void *alloc_dma(struct dmm_txn *txn, size_t sz, dma_addr_t *pa)
  94 {
  95         void *ptr;
  96         struct refill_engine *engine = txn->engine_handle;
  97
  98         /* dmm programming requires 16 byte aligned addresses */
  99         txn->current_pa = round_up(txn->current_pa, 16);
 100         txn->current_va = (void *)round_up((long)txn->current_va, 16);
 101
 102         ptr = txn->current_va;
 103         *pa = txn->current_pa;
 104
 105         txn->current_pa += sz;
 106         txn->current_va += sz;
 107
 108         BUG_ON((txn->current_va - engine->refill_va) > REFILL_BUFFER_SIZE);
 109
 110         return ptr;
 111 }
 112
 113 /* check status and spin until wait_mask comes true */
 114 static int wait_status(struct refill_engine *engine, u32 wait_mask)
 115 {
 116         struct dmm *dmm = engine->dmm;
 117         u32 r = 0, err, i;
 118
 119         i = DMM_FIXED_RETRY_COUNT;
 120         while (true) {
 121                 r = dmm_read(dmm, reg[PAT_STATUS][engine->id]);
 122                 err = r & DMM_PATSTATUS_ERR;
 123                 if (err) {
 124                         dev_err(dmm->dev,
 125                                 "%s: error (engine%d). PAT_STATUS: 0x%08x\n",
 126                                 __func__, engine->id, r);
 127                         return -EFAULT;
 128                 }
 129
 130                 if ((r & wait_mask) == wait_mask)
 131                         break;
 132
 133                 if (--i == 0) {
 134                         dev_err(dmm->dev,
 135                                 "%s: timeout (engine%d). PAT_STATUS: 0x%08x\n",
 136                                 __func__, engine->id, r);
 137                         return -ETIMEDOUT;
 138                 }
 139
 140                 udelay(1);
 141         }
 142
 143         return 0;
 144 }
 145
 146 static void release_engine(struct refill_engine *engine)
 147 {
 148         unsigned long flags;
 149
 150         spin_lock_irqsave(&list_lock, flags);
 151         list_add(&engine->idle_node, &omap_dmm->idle_head);
 152         spin_unlock_irqrestore(&list_lock, flags);
 153
 154         atomic_inc(&omap_dmm->engine_counter);
 155         wake_up_interruptible(&omap_dmm->engine_queue);
 156 }
 157
 158 static irqreturn_t omap_dmm_irq_handler(int irq, void *arg)
 159 {
 160         struct dmm *dmm = arg;
 161         u32 status = dmm_read(dmm, DMM_PAT_IRQSTATUS);
 162         int i;
 163
 164         /* ack IRQ */
 165         dmm_write(dmm, status, DMM_PAT_IRQSTATUS);
 166
 167         for (i = 0; i < dmm->num_engines; i++) {
 168                 if (status & DMM_IRQSTAT_ERR_MASK)
 169                         dev_err(dmm->dev,
 170                                 "irq error(engine%d): IRQSTAT 0x%02x\n",
 171                                 i, status & 0xff);
 172
 173                 if (status & DMM_IRQSTAT_LST) {
 174                         if (dmm->engines[i].async)
 175                                 release_engine(&dmm->engines[i]);
 176
 177                         complete(&dmm->engines[i].compl);
 178                 }
 179
 180                 status >>= 8;
 181         }
 182
 183         return IRQ_HANDLED;
 184 }
 185
 186 /**
 187  * Get a handle for a DMM transaction
 188  */
 189 static struct dmm_txn *dmm_txn_init(struct dmm *dmm, struct tcm *tcm)
 190 {
 191         struct dmm_txn *txn = NULL;
 192         struct refill_engine *engine = NULL;
 193         int ret;
 194         unsigned long flags;
 195
 196
 197         /* wait until an engine is available */
 198         ret = wait_event_interruptible(omap_dmm->engine_queue,
 199                 atomic_add_unless(&omap_dmm->engine_counter, -1, 0));
 200         if (ret)
 201                 return ERR_PTR(ret);
 202
 203         /* grab an idle engine */
 204         spin_lock_irqsave(&list_lock, flags);
 205         if (!list_empty(&dmm->idle_head)) {
 206                 engine = list_entry(dmm->idle_head.next, struct refill_engine,
 207                                         idle_node);
 208                 list_del(&engine->idle_node);
 209         }
 210         spin_unlock_irqrestore(&list_lock, flags);
 211
 212         BUG_ON(!engine);
 213
 214         txn = &engine->txn;
 215         engine->tcm = tcm;
 216         txn->engine_handle = engine;
 217         txn->last_pat = NULL;
 218         txn->current_va = engine->refill_va;
 219         txn->current_pa = engine->refill_pa;
 220
 221         return txn;
 222 }
 223
 224 /**
 225  * Add region to DMM transaction.  If pages or pages[i] is NULL, then the
 226  * corresponding slot is cleared (ie. dummy_pa is programmed)
 227  */
 228 static void dmm_txn_append(struct dmm_txn *txn, struct pat_area *area,
 229                 struct page **pages, u32 npages, u32 roll)
 230 {
 231         dma_addr_t pat_pa = 0, data_pa = 0;
 232         u32 *data;
 233         struct pat *pat;
 234         struct refill_engine *engine = txn->engine_handle;
 235         int columns = (1 + area->x1 - area->x0);
 236         int rows = (1 + area->y1 - area->y0);
 237         int i = columns*rows;
 238
 239         pat = alloc_dma(txn, sizeof(*pat), &pat_pa);
 240
 241         if (txn->last_pat)
 242                 txn->last_pat->next_pa = (u32)pat_pa;
 243
 244         pat->area = *area;
 245
 246         /* adjust Y coordinates based off of container parameters */
 247         pat->area.y0 += engine->tcm->y_offset;
 248         pat->area.y1 += engine->tcm->y_offset;
 249
 250         pat->ctrl = (struct pat_ctrl){
 251                         .start = 1,
 252                         .lut_id = engine->tcm->lut_id,
 253                 };
 254
 255         data = alloc_dma(txn, 4*i, &data_pa);
 256         /* FIXME: what if data_pa is more than 32-bit ? */
 257         pat->data_pa = data_pa;
 258
 259         while (i--) {
 260                 int n = i + roll;
 261                 if (n >= npages)
 262                         n -= npages;
 263                 data[i] = (pages && pages[n]) ?
 264                         page_to_phys(pages[n]) : engine->dmm->dummy_pa;
 265         }
 266
 267         txn->last_pat = pat;
 268
 269         return;
 270 }
 271
 272 /**
 273  * Commit the DMM transaction.
 274  */
 275 static int dmm_txn_commit(struct dmm_txn *txn, bool wait)
 276 {
 277         int ret = 0;
 278         struct refill_engine *engine = txn->engine_handle;
 279         struct dmm *dmm = engine->dmm;
 280
 281         if (!txn->last_pat) {
 282                 dev_err(engine->dmm->dev, "need at least one txn\n");
 283                 ret = -EINVAL;
 284                 goto cleanup;
 285         }
 286
 287         txn->last_pat->next_pa = 0;
 288
 289         /* write to PAT_DESCR to clear out any pending transaction */
 290         dmm_write(dmm, 0x0, reg[PAT_DESCR][engine->id]);
 291
 292         /* wait for engine ready: */
 293         ret = wait_status(engine, DMM_PATSTATUS_READY);
 294         if (ret) {
 295                 ret = -EFAULT;
 296                 goto cleanup;
 297         }
 298
 299         /* mark whether it is async to denote list management in IRQ handler */
 300         engine->async = wait ? false : true;
 301         reinit_completion(&engine->compl);
 302         /* verify that the irq handler sees the 'async' and completion value */
 303         smp_mb();
 304
 305         /* kick reload */
 306         dmm_write(dmm, engine->refill_pa, reg[PAT_DESCR][engine->id]);
 307
 308         if (wait) {
 309                 if (!wait_for_completion_timeout(&engine->compl,
 310                                 msecs_to_jiffies(100))) {
 311                         dev_err(dmm->dev, "timed out waiting for done\n");
 312                         ret = -ETIMEDOUT;
 313                         goto cleanup;
 314                 }
 315
 316                 /* Check the engine status before continue */
 317                 ret = wait_status(engine, DMM_PATSTATUS_READY |
 318                                   DMM_PATSTATUS_VALID | DMM_PATSTATUS_DONE);
 319         }
 320
 321 cleanup:
 322         /* only place engine back on list if we are done with it */
 323         if (ret || wait)
 324                 release_engine(engine);
 325
 326         return ret;
 327 }
 328
 329 /*
 330  * DMM programming
 331  */
 332 static int fill(struct tcm_area *area, struct page **pages,
 333                 u32 npages, u32 roll, bool wait)
 334 {
 335         int ret = 0;
 336         struct tcm_area slice, area_s;
 337         struct dmm_txn *txn;
 338
 339         /*
 340          * FIXME
 341          *
 342          * Asynchronous fill does not work reliably, as the driver does not
 343          * handle errors in the async code paths. The fill operation may
 344          * silently fail, leading to leaking DMM engines, which may eventually
 345          * lead to deadlock if we run out of DMM engines.
 346          *
 347          * For now, always set 'wait' so that we only use sync fills. Async
 348          * fills should be fixed, or alternatively we could decide to only
 349          * support sync fills and so the whole async code path could be removed.
 350          */
 351
 352         wait = true;
 353
 354         txn = dmm_txn_init(omap_dmm, area->tcm);
 355         if (IS_ERR_OR_NULL(txn))
 356                 return -ENOMEM;
 357
 358         tcm_for_each_slice(slice, *area, area_s) {
 359                 struct pat_area p_area = {
 360                                 .x0 = slice.p0.x,  .y0 = slice.p0.y,
 361                                 .x1 = slice.p1.x,  .y1 = slice.p1.y,
 362                 };
 363
 364                 dmm_txn_append(txn, &p_area, pages, npages, roll);
 365
 366                 roll += tcm_sizeof(slice);
 367         }
 368
 369         ret = dmm_txn_commit(txn, wait);
 370
 371         return ret;
 372 }
 373
 374 /*
 375  * Pin/unpin
 376  */
 377
 378 /* note: slots for which pages[i] == NULL are filled w/ dummy page
 379  */
 380 int tiler_pin(struct tiler_block *block, struct page **pages,
 381                 u32 npages, u32 roll, bool wait)
 382 {
 383         int ret;
 384
 385         ret = fill(&block->area, pages, npages, roll, wait);
 386
 387         if (ret)
 388                 tiler_unpin(block);
 389
 390         return ret;
 391 }
 392
 393 int tiler_unpin(struct tiler_block *block)
 394 {
 395         return fill(&block->area, NULL, 0, 0, false);
 396 }
 397
 398 /*
 399  * Reserve/release
 400  */
 401 struct tiler_block *tiler_reserve_2d(enum tiler_fmt fmt, u16 w,
 402                 u16 h, u16 align)
 403 {
 404         struct tiler_block *block;
 405         u32 min_align = 128;
 406         int ret;
 407         unsigned long flags;
 408         u32 slot_bytes;
 409
 410         block = kzalloc(sizeof(*block), GFP_KERNEL);
 411         if (!block)
 412                 return ERR_PTR(-ENOMEM);
 413
 414         BUG_ON(!validfmt(fmt));
 415
 416         /* convert width/height to slots */
 417         w = DIV_ROUND_UP(w, geom[fmt].slot_w);
 418         h = DIV_ROUND_UP(h, geom[fmt].slot_h);
 419
 420         /* convert alignment to slots */
 421         slot_bytes = geom[fmt].slot_w * geom[fmt].cpp;
 422         min_align = max(min_align, slot_bytes);
 423         align = (align > min_align) ? ALIGN(align, min_align) : min_align;
 424         align /= slot_bytes;
 425
 426         block->fmt = fmt;
 427
 428         ret = tcm_reserve_2d(containers[fmt], w, h, align, -1, slot_bytes,
 429                         &block->area);
 430         if (ret) {
 431                 kfree(block);
 432                 return ERR_PTR(-ENOMEM);
 433         }
 434
 435         /* add to allocation list */
 436         spin_lock_irqsave(&list_lock, flags);
 437         list_add(&block->alloc_node, &omap_dmm->alloc_head);
 438         spin_unlock_irqrestore(&list_lock, flags);
 439
 440         return block;
 441 }
 442
 443 struct tiler_block *tiler_reserve_1d(size_t size)
 444 {
 445         struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
 446         int num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 447         unsigned long flags;
 448
 449         if (!block)
 450                 return ERR_PTR(-ENOMEM);
 451
 452         block->fmt = TILFMT_PAGE;
 453
 454         if (tcm_reserve_1d(containers[TILFMT_PAGE], num_pages,
 455                                 &block->area)) {
 456                 kfree(block);
 457                 return ERR_PTR(-ENOMEM);
 458         }
 459
 460         spin_lock_irqsave(&list_lock, flags);
 461         list_add(&block->alloc_node, &omap_dmm->alloc_head);
 462         spin_unlock_irqrestore(&list_lock, flags);
 463
 464         return block;
 465 }
 466
 467 /* note: if you have pin'd pages, you should have already unpin'd first! */
 468 int tiler_release(struct tiler_block *block)
 469 {
 470         int ret = tcm_free(&block->area);
 471         unsigned long flags;
 472
 473         if (block->area.tcm)
 474                 dev_err(omap_dmm->dev, "failed to release block\n");
 475
 476         spin_lock_irqsave(&list_lock, flags);
 477         list_del(&block->alloc_node);
 478         spin_unlock_irqrestore(&list_lock, flags);
 479
 480         kfree(block);
 481         return ret;
 482 }
 483
 484 /*
 485  * Utils
 486  */
 487
 488 /* calculate the tiler space address of a pixel in a view orientation...
 489  * below description copied from the display subsystem section of TRM:
 490  *
 491  * When the TILER is addressed, the bits:
 492  *   [28:27] = 0x0 for 8-bit tiled
 493  *             0x1 for 16-bit tiled
 494  *             0x2 for 32-bit tiled
 495  *             0x3 for page mode
 496  *   [31:29] = 0x0 for 0-degree view
 497  *             0x1 for 180-degree view + mirroring
 498  *             0x2 for 0-degree view + mirroring
 499  *             0x3 for 180-degree view
 500  *             0x4 for 270-degree view + mirroring
 501  *             0x5 for 270-degree view
 502  *             0x6 for 90-degree view
 503  *             0x7 for 90-degree view + mirroring
 504  * Otherwise the bits indicated the corresponding bit address to access
 505  * the SDRAM.
 506  */
 507 static u32 tiler_get_address(enum tiler_fmt fmt, u32 orient, u32 x, u32 y)
 508 {
 509         u32 x_bits, y_bits, tmp, x_mask, y_mask, alignment;
 510
 511         x_bits = CONT_WIDTH_BITS - geom[fmt].x_shft;
 512         y_bits = CONT_HEIGHT_BITS - geom[fmt].y_shft;
 513         alignment = geom[fmt].x_shft + geom[fmt].y_shft;
 514
 515         /* validate coordinate */
 516         x_mask = MASK(x_bits);
 517         y_mask = MASK(y_bits);
 518
 519         if (x < 0 || x > x_mask || y < 0 || y > y_mask) {
 520                 DBG("invalid coords: %u < 0 || %u > %u || %u < 0 || %u > %u",
 521                                 x, x, x_mask, y, y, y_mask);
 522                 return 0;
 523         }
 524
 525         /* account for mirroring */
 526         if (orient & MASK_X_INVERT)
 527                 x ^= x_mask;
 528         if (orient & MASK_Y_INVERT)
 529                 y ^= y_mask;
 530
 531         /* get coordinate address */
 532         if (orient & MASK_XY_FLIP)
 533                 tmp = ((x << y_bits) + y);
 534         else
 535                 tmp = ((y << x_bits) + x);
 536
 537         return TIL_ADDR((tmp << alignment), orient, fmt);
 538 }
 539
 540 dma_addr_t tiler_ssptr(struct tiler_block *block)
 541 {
 542         BUG_ON(!validfmt(block->fmt));
 543
 544         return TILVIEW_8BIT + tiler_get_address(block->fmt, 0,
 545                         block->area.p0.x * geom[block->fmt].slot_w,
 546                         block->area.p0.y * geom[block->fmt].slot_h);
 547 }
 548
 549 dma_addr_t tiler_tsptr(struct tiler_block *block, u32 orient,
 550                 u32 x, u32 y)
 551 {
 552         struct tcm_pt *p = &block->area.p0;
 553         BUG_ON(!validfmt(block->fmt));
 554
 555         return tiler_get_address(block->fmt, orient,
 556                         (p->x * geom[block->fmt].slot_w) + x,
 557                         (p->y * geom[block->fmt].slot_h) + y);
 558 }
 559
 560 void tiler_align(enum tiler_fmt fmt, u16 *w, u16 *h)
 561 {
 562         BUG_ON(!validfmt(fmt));
 563         *w = round_up(*w, geom[fmt].slot_w);
 564         *h = round_up(*h, geom[fmt].slot_h);
 565 }
 566
 567 u32 tiler_stride(enum tiler_fmt fmt, u32 orient)
 568 {
 569         BUG_ON(!validfmt(fmt));
 570
 571         if (orient & MASK_XY_FLIP)
 572                 return 1 << (CONT_HEIGHT_BITS + geom[fmt].x_shft);
 573         else
 574                 return 1 << (CONT_WIDTH_BITS + geom[fmt].y_shft);
 575 }
 576
 577 size_t tiler_size(enum tiler_fmt fmt, u16 w, u16 h)
 578 {
 579         tiler_align(fmt, &w, &h);
 580         return geom[fmt].cpp * w * h;
 581 }
 582
 583 size_t tiler_vsize(enum tiler_fmt fmt, u16 w, u16 h)
 584 {
 585         BUG_ON(!validfmt(fmt));
 586         return round_up(geom[fmt].cpp * w, PAGE_SIZE) * h;
 587 }
 588
 589 u32 tiler_get_cpu_cache_flags(void)
 590 {
 591         return omap_dmm->plat_data->cpu_cache_flags;
 592 }
 593
 594 bool dmm_is_available(void)
 595 {
 596         return omap_dmm ? true : false;
 597 }
 598
 599 static int omap_dmm_remove(struct platform_device *dev)
 600 {
 601         struct tiler_block *block, *_block;
 602         int i;
 603         unsigned long flags;
 604
 605         if (omap_dmm) {
 606                 /* free all area regions */
 607                 spin_lock_irqsave(&list_lock, flags);
 608                 list_for_each_entry_safe(block, _block, &omap_dmm->alloc_head,
 609                                         alloc_node) {
 610                         list_del(&block->alloc_node);
 611                         kfree(block);
 612                 }
 613                 spin_unlock_irqrestore(&list_lock, flags);
 614
 615                 for (i = 0; i < omap_dmm->num_lut; i++)
 616                         if (omap_dmm->tcm && omap_dmm->tcm[i])
 617                                 omap_dmm->tcm[i]->deinit(omap_dmm->tcm[i]);
 618                 kfree(omap_dmm->tcm);
 619
 620                 kfree(omap_dmm->engines);
 621                 if (omap_dmm->refill_va)
 622                         dma_free_wc(omap_dmm->dev,
 623                                     REFILL_BUFFER_SIZE * omap_dmm->num_engines,
 624                                     omap_dmm->refill_va, omap_dmm->refill_pa);
 625                 if (omap_dmm->dummy_page)
 626                         __free_page(omap_dmm->dummy_page);
 627
 628                 if (omap_dmm->irq > 0)
 629                         free_irq(omap_dmm->irq, omap_dmm);
 630
 631                 iounmap(omap_dmm->base);
 632                 kfree(omap_dmm);
 633                 omap_dmm = NULL;
 634         }
 635
 636         return 0;
 637 }
 638
 639 static int omap_dmm_probe(struct platform_device *dev)
 640 {
 641         int ret = -EFAULT, i;
 642         struct tcm_area area = {0};
 643         u32 hwinfo, pat_geom;
 644         struct resource *mem;
 645
 646         omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL);
 647         if (!omap_dmm)
 648                 goto fail;
 649
 650         /* initialize lists */
 651         INIT_LIST_HEAD(&omap_dmm->alloc_head);
 652         INIT_LIST_HEAD(&omap_dmm->idle_head);
 653
 654         init_waitqueue_head(&omap_dmm->engine_queue);
 655
 656         if (dev->dev.of_node) {
 657                 const struct of_device_id *match;
 658
 659                 match = of_match_node(dmm_of_match, dev->dev.of_node);
 660                 if (!match) {
 661                         dev_err(&dev->dev, "failed to find matching device node\n");
 662                         ret = -ENODEV;
 663                         goto fail;
 664                 }
 665
 666                 omap_dmm->plat_data = match->data;
 667         }
 668
 669         /* lookup hwmod data - base address and irq */
 670         mem = platform_get_resource(dev, IORESOURCE_MEM, 0);
 671         if (!mem) {
 672                 dev_err(&dev->dev, "failed to get base address resource\n");
 673                 goto fail;
 674         }
 675
 676         omap_dmm->base = ioremap(mem->start, SZ_2K);
 677
 678         if (!omap_dmm->base) {
 679                 dev_err(&dev->dev, "failed to get dmm base address\n");
 680                 goto fail;
 681         }
 682
 683         omap_dmm->irq = platform_get_irq(dev, 0);
 684         if (omap_dmm->irq < 0) {
 685                 dev_err(&dev->dev, "failed to get IRQ resource\n");
 686                 goto fail;
 687         }
 688
 689         omap_dmm->dev = &dev->dev;
 690
 691         hwinfo = dmm_read(omap_dmm, DMM_PAT_HWINFO);
 692         omap_dmm->num_engines = (hwinfo >> 24) & 0x1F;
 693         omap_dmm->num_lut = (hwinfo >> 16) & 0x1F;
 694         omap_dmm->container_width = 256;
 695         omap_dmm->container_height = 128;
 696
 697         atomic_set(&omap_dmm->engine_counter, omap_dmm->num_engines);
 698
 699         /* read out actual LUT width and height */
 700         pat_geom = dmm_read(omap_dmm, DMM_PAT_GEOMETRY);
 701         omap_dmm->lut_width = ((pat_geom >> 16) & 0xF) << 5;
 702         omap_dmm->lut_height = ((pat_geom >> 24) & 0xF) << 5;
 703
 704         /* increment LUT by one if on OMAP5 */
 705         /* LUT has twice the height, and is split into a separate container */
 706         if (omap_dmm->lut_height != omap_dmm->container_height)
 707                 omap_dmm->num_lut++;
 708
 709         /* initialize DMM registers */
 710         dmm_write(omap_dmm, 0x88888888, DMM_PAT_VIEW__0);
 711         dmm_write(omap_dmm, 0x88888888, DMM_PAT_VIEW__1);
 712         dmm_write(omap_dmm, 0x80808080, DMM_PAT_VIEW_MAP__0);
 713         dmm_write(omap_dmm, 0x80000000, DMM_PAT_VIEW_MAP_BASE);
 714         dmm_write(omap_dmm, 0x88888888, DMM_TILER_OR__0);
 715         dmm_write(omap_dmm, 0x88888888, DMM_TILER_OR__1);
 716
 717         ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
 718                                 "omap_dmm_irq_handler", omap_dmm);
 719
 720         if (ret) {
 721                 dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
 722                         omap_dmm->irq, ret);
 723                 omap_dmm->irq = -1;
 724                 goto fail;
 725         }
 726
 727         /* Enable all interrupts for each refill engine except
 728          * ERR_LUT_MISS<n> (which is just advisory, and we don't care
 729          * about because we want to be able to refill live scanout
 730          * buffers for accelerated pan/scroll) and FILL_DSC<n> which
 731          * we just generally don't care about.
 732          */
 733         dmm_write(omap_dmm, 0x7e7e7e7e, DMM_PAT_IRQENABLE_SET);
 734
 735         omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32);
 736         if (!omap_dmm->dummy_page) {
 737                 dev_err(&dev->dev, "could not allocate dummy page\n");
 738                 ret = -ENOMEM;
 739                 goto fail;
 740         }
 741
 742         /* set dma mask for device */
 743         ret = dma_set_coherent_mask(&dev->dev, DMA_BIT_MASK(32));
 744         if (ret)
 745                 goto fail;
 746
 747         omap_dmm->dummy_pa = page_to_phys(omap_dmm->dummy_page);
 748
 749         /* alloc refill memory */
 750         omap_dmm->refill_va = dma_alloc_wc(&dev->dev,
 751                                            REFILL_BUFFER_SIZE * omap_dmm->num_engines,
 752                                            &omap_dmm->refill_pa, GFP_KERNEL);
 753         if (!omap_dmm->refill_va) {
 754                 dev_err(&dev->dev, "could not allocate refill memory\n");
 755                 goto fail;
 756         }
 757
 758         /* alloc engines */
 759         omap_dmm->engines = kcalloc(omap_dmm->num_engines,
 760                                     sizeof(*omap_dmm->engines), GFP_KERNEL);
 761         if (!omap_dmm->engines) {
 762                 ret = -ENOMEM;
 763                 goto fail;
 764         }
 765
 766         for (i = 0; i < omap_dmm->num_engines; i++) {
 767                 omap_dmm->engines[i].id = i;
 768                 omap_dmm->engines[i].dmm = omap_dmm;
 769                 omap_dmm->engines[i].refill_va = omap_dmm->refill_va +
 770                                                 (REFILL_BUFFER_SIZE * i);
 771                 omap_dmm->engines[i].refill_pa = omap_dmm->refill_pa +
 772                                                 (REFILL_BUFFER_SIZE * i);
 773                 init_completion(&omap_dmm->engines[i].compl);
 774
 775                 list_add(&omap_dmm->engines[i].idle_node, &omap_dmm->idle_head);
 776         }
 777
 778         omap_dmm->tcm = kcalloc(omap_dmm->num_lut, sizeof(*omap_dmm->tcm),
 779                                 GFP_KERNEL);
 780         if (!omap_dmm->tcm) {
 781                 ret = -ENOMEM;
 782                 goto fail;
 783         }
 784
 785         /* init containers */
 786         /* Each LUT is associated with a TCM (container manager).  We use the
 787            lut_id to denote the lut_id used to identify the correct LUT for
 788            programming during reill operations */
 789         for (i = 0; i < omap_dmm->num_lut; i++) {
 790                 omap_dmm->tcm[i] = sita_init(omap_dmm->container_width,
 791                                                 omap_dmm->container_height);
 792
 793                 if (!omap_dmm->tcm[i]) {
 794                         dev_err(&dev->dev, "failed to allocate container\n");
 795                         ret = -ENOMEM;
 796                         goto fail;
 797                 }
 798
 799                 omap_dmm->tcm[i]->lut_id = i;
 800         }
 801
 802         /* assign access mode containers to applicable tcm container */
 803         /* OMAP 4 has 1 container for all 4 views */
 804         /* OMAP 5 has 2 containers, 1 for 2D and 1 for 1D */
 805         containers[TILFMT_8BIT] = omap_dmm->tcm[0];
 806         containers[TILFMT_16BIT] = omap_dmm->tcm[0];
 807         containers[TILFMT_32BIT] = omap_dmm->tcm[0];
 808
 809         if (omap_dmm->container_height != omap_dmm->lut_height) {
 810                 /* second LUT is used for PAGE mode.  Programming must use
 811                    y offset that is added to all y coordinates.  LUT id is still
 812                    0, because it is the same LUT, just the upper 128 lines */
 813                 containers[TILFMT_PAGE] = omap_dmm->tcm[1];
 814                 omap_dmm->tcm[1]->y_offset = OMAP5_LUT_OFFSET;
 815                 omap_dmm->tcm[1]->lut_id = 0;
 816         } else {
 817                 containers[TILFMT_PAGE] = omap_dmm->tcm[0];
 818         }
 819
 820         area = (struct tcm_area) {
 821                 .tcm = NULL,
 822                 .p1.x = omap_dmm->container_width - 1,
 823                 .p1.y = omap_dmm->container_height - 1,
 824         };
 825
 826         /* initialize all LUTs to dummy page entries */
 827         for (i = 0; i < omap_dmm->num_lut; i++) {
 828                 area.tcm = omap_dmm->tcm[i];
 829                 if (fill(&area, NULL, 0, 0, true))
 830                         dev_err(omap_dmm->dev, "refill failed");
 831         }
 832
 833         dev_info(omap_dmm->dev, "initialized all PAT entries\n");
 834
 835         return 0;
 836
 837 fail:
 838         if (omap_dmm_remove(dev))
 839                 dev_err(&dev->dev, "cleanup failed\n");
 840         return ret;
 841 }
 842
 843 /*
 844  * debugfs support
 845  */
 846
 847 #ifdef CONFIG_DEBUG_FS
 848
 849 static const char *alphabet = "abcdefghijklmnopqrstuvwxyz"
 850                                 "ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
 851 static const char *special = ".,:;'\"`~!^-+";
 852
 853 static void fill_map(char **map, int xdiv, int ydiv, struct tcm_area *a,
 854                                                         char c, bool ovw)
 855 {
 856         int x, y;
 857         for (y = a->p0.y / ydiv; y <= a->p1.y / ydiv; y++)
 858                 for (x = a->p0.x / xdiv; x <= a->p1.x / xdiv; x++)
 859                         if (map[y][x] == ' ' || ovw)
 860                                 map[y][x] = c;
 861 }
 862
 863 static void fill_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p,
 864                                                                         char c)
 865 {
 866         map[p->y / ydiv][p->x / xdiv] = c;
 867 }
 868
 869 static char read_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p)
 870 {
 871         return map[p->y / ydiv][p->x / xdiv];
 872 }
 873
 874 static int map_width(int xdiv, int x0, int x1)
 875 {
 876         return (x1 / xdiv) - (x0 / xdiv) + 1;
 877 }
 878
 879 static void text_map(char **map, int xdiv, char *nice, int yd, int x0, int x1)
 880 {
 881         char *p = map[yd] + (x0 / xdiv);
 882         int w = (map_width(xdiv, x0, x1) - strlen(nice)) / 2;
 883         if (w >= 0) {
 884                 p += w;
 885                 while (*nice)
 886                         *p++ = *nice++;
 887         }
 888 }
 889
 890 static void map_1d_info(char **map, int xdiv, int ydiv, char *nice,
 891                                                         struct tcm_area *a)
 892 {
 893         sprintf(nice, "%dK", tcm_sizeof(*a) * 4);
 894         if (a->p0.y + 1 < a->p1.y) {
 895                 text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv, 0,
 896                                                         256 - 1);
 897         } else if (a->p0.y < a->p1.y) {
 898                 if (strlen(nice) < map_width(xdiv, a->p0.x, 256 - 1))
 899                         text_map(map, xdiv, nice, a->p0.y / ydiv,
 900                                         a->p0.x + xdiv, 256 - 1);
 901                 else if (strlen(nice) < map_width(xdiv, 0, a->p1.x))
 902                         text_map(map, xdiv, nice, a->p1.y / ydiv,
 903                                         0, a->p1.y - xdiv);
 904         } else if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x)) {
 905                 text_map(map, xdiv, nice, a->p0.y / ydiv, a->p0.x, a->p1.x);
 906         }
 907 }
 908
 909 static void map_2d_info(char **map, int xdiv, int ydiv, char *nice,
 910                                                         struct tcm_area *a)
 911 {
 912         sprintf(nice, "(%d*%d)", tcm_awidth(*a), tcm_aheight(*a));
 913         if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x))
 914                 text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv,
 915                                                         a->p0.x, a->p1.x);
 916 }
 917
 918 int tiler_map_show(struct seq_file *s, void *arg)
 919 {
 920         int xdiv = 2, ydiv = 1;
 921         char **map = NULL, *global_map;
 922         struct tiler_block *block;
 923         struct tcm_area a, p;
 924         int i;
 925         const char *m2d = alphabet;
 926         const char *a2d = special;
 927         const char *m2dp = m2d, *a2dp = a2d;
 928         char nice[128];
 929         int h_adj;
 930         int w_adj;
 931         unsigned long flags;
 932         int lut_idx;
 933
 934
 935         if (!omap_dmm) {
 936                 /* early return if dmm/tiler device is not initialized */
 937                 return 0;
 938         }
 939
 940         h_adj = omap_dmm->container_height / ydiv;
 941         w_adj = omap_dmm->container_width / xdiv;
 942
 943         map = kmalloc_array(h_adj, sizeof(*map), GFP_KERNEL);
 944         global_map = kmalloc_array(w_adj + 1, h_adj, GFP_KERNEL);
 945
 946         if (!map || !global_map)
 947                 goto error;
 948
 949         for (lut_idx = 0; lut_idx < omap_dmm->num_lut; lut_idx++) {
 950                 memset(map, 0, h_adj * sizeof(*map));
 951                 memset(global_map, ' ', (w_adj + 1) * h_adj);
 952
 953                 for (i = 0; i < omap_dmm->container_height; i++) {
 954                         map[i] = global_map + i * (w_adj + 1);
 955                         map[i][w_adj] = 0;
 956                 }
 957
 958                 spin_lock_irqsave(&list_lock, flags);
 959
 960                 list_for_each_entry(block, &omap_dmm->alloc_head, alloc_node) {
 961                         if (block->area.tcm == omap_dmm->tcm[lut_idx]) {
 962                                 if (block->fmt != TILFMT_PAGE) {
 963                                         fill_map(map, xdiv, ydiv, &block->area,
 964                                                 *m2dp, true);
 965                                         if (!*++a2dp)
 966                                                 a2dp = a2d;
 967                                         if (!*++m2dp)
 968                                                 m2dp = m2d;
 969                                         map_2d_info(map, xdiv, ydiv, nice,
 970                                                         &block->area);
 971                                 } else {
 972                                         bool start = read_map_pt(map, xdiv,
 973                                                 ydiv, &block->area.p0) == ' ';
 974                                         bool end = read_map_pt(map, xdiv, ydiv,
 975                                                         &block->area.p1) == ' ';
 976
 977                                         tcm_for_each_slice(a, block->area, p)
 978                                                 fill_map(map, xdiv, ydiv, &a,
 979                                                         '=', true);
 980                                         fill_map_pt(map, xdiv, ydiv,
 981                                                         &block->area.p0,
 982                                                         start ? '<' : 'X');
 983                                         fill_map_pt(map, xdiv, ydiv,
 984                                                         &block->area.p1,
 985                                                         end ? '>' : 'X');
 986                                         map_1d_info(map, xdiv, ydiv, nice,
 987                                                         &block->area);
 988                                 }
 989                         }
 990                 }
 991
 992                 spin_unlock_irqrestore(&list_lock, flags);
 993
 994                 if (s) {
 995                         seq_printf(s, "CONTAINER %d DUMP BEGIN\n", lut_idx);
 996                         for (i = 0; i < 128; i++)
 997                                 seq_printf(s, "%03d:%s\n", i, map[i]);
 998                         seq_printf(s, "CONTAINER %d DUMP END\n", lut_idx);
 999                 } else {
1000                         dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP BEGIN\n",
1001                                 lut_idx);
1002                         for (i = 0; i < 128; i++)
1003                                 dev_dbg(omap_dmm->dev, "%03d:%s\n", i, map[i]);
1004                         dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP END\n",
1005                                 lut_idx);
1006                 }
1007         }
1008
1009 error:
1010         kfree(map);
1011         kfree(global_map);
1012
1013         return 0;
1014 }
1015 #endif
1016
1017 #ifdef CONFIG_PM_SLEEP
1018 static int omap_dmm_resume(struct device *dev)
1019 {
1020         struct tcm_area area;
1021         int i;
1022
1023         if (!omap_dmm)
1024                 return -ENODEV;
1025
1026         area = (struct tcm_area) {
1027                 .tcm = NULL,
1028                 .p1.x = omap_dmm->container_width - 1,
1029                 .p1.y = omap_dmm->container_height - 1,
1030         };
1031
1032         /* initialize all LUTs to dummy page entries */
1033         for (i = 0; i < omap_dmm->num_lut; i++) {
1034                 area.tcm = omap_dmm->tcm[i];
1035                 if (fill(&area, NULL, 0, 0, true))
1036                         dev_err(dev, "refill failed");
1037         }
1038
1039         return 0;
1040 }
1041 #endif
1042
1043 static SIMPLE_DEV_PM_OPS(omap_dmm_pm_ops, NULL, omap_dmm_resume);
1044
1045 #if defined(CONFIG_OF)
1046 static const struct dmm_platform_data dmm_omap4_platform_data = {
1047         .cpu_cache_flags = OMAP_BO_WC,
1048 };
1049
1050 static const struct dmm_platform_data dmm_omap5_platform_data = {
1051         .cpu_cache_flags = OMAP_BO_UNCACHED,
1052 };
1053
1054 static const struct of_device_id dmm_of_match[] = {
1055         {
1056                 .compatible = "ti,omap4-dmm",
1057                 .data = &dmm_omap4_platform_data,
1058         },
1059         {
1060                 .compatible = "ti,omap5-dmm",
1061                 .data = &dmm_omap5_platform_data,
1062         },
1063         {},
1064 };
1065 #endif
1066
1067 struct platform_driver omap_dmm_driver = {
1068         .probe = omap_dmm_probe,
1069         .remove = omap_dmm_remove,
1070         .driver = {
1071                 .owner = THIS_MODULE,
1072                 .name = DMM_DRIVER_NAME,
1073                 .of_match_table = of_match_ptr(dmm_of_match),
1074                 .pm = &omap_dmm_pm_ops,
1075         },
1076 };
1077
1078 MODULE_LICENSE("GPL v2");
1079 MODULE_AUTHOR("Andy Gross <andy.gross@ti.com>");
1080 MODULE_DESCRIPTION("OMAP DMM/Tiler Driver");