1 /*
2  * DMM IOMMU driver support functions for TI OMAP processors.
3  *
4  * Author: Rob Clark <rob@ti.com>
5  *         Andy Gross <andy.gross@ti.com>
6  *
7  * Copyright (C) 2011 Texas Instruments Incorporated - http://www.ti.com/
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License as
11  * published by the Free Software Foundation version 2.
12  *
13  * This program is distributed "as is" WITHOUT ANY WARRANTY of any
14  * kind, whether express or implied; without even the implied warranty
15  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  */
18 
19 #include <linux/completion.h>
20 #include <linux/delay.h>
21 #include <linux/dma-mapping.h>
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/interrupt.h>
25 #include <linux/list.h>
26 #include <linux/mm.h>
27 #include <linux/module.h>
28 #include <linux/platform_device.h> /* platform_device() */
29 #include <linux/sched.h>
30 #include <linux/seq_file.h>
31 #include <linux/slab.h>
32 #include <linux/time.h>
33 #include <linux/vmalloc.h>
34 #include <linux/wait.h>
35 
36 #include "omap_dmm_tiler.h"
37 #include "omap_dmm_priv.h"
38 
39 #define DMM_DRIVER_NAME "dmm"
40 
41 /* mappings for associating views to luts */
42 static struct tcm *containers[TILFMT_NFORMATS];
43 static struct dmm *omap_dmm;
44 
45 #if defined(CONFIG_OF)
46 static const struct of_device_id dmm_of_match[];
47 #endif
48 
49 /* global spinlock for protecting lists */
50 static DEFINE_SPINLOCK(list_lock);
51 
52 /* Geometry table */
53 #define GEOM(xshift, yshift, bytes_per_pixel) { \
54 		.x_shft = (xshift), \
55 		.y_shft = (yshift), \
56 		.cpp    = (bytes_per_pixel), \
57 		.slot_w = 1 << (SLOT_WIDTH_BITS - (xshift)), \
58 		.slot_h = 1 << (SLOT_HEIGHT_BITS - (yshift)), \
59 	}
60 
61 static const struct {
62 	uint32_t x_shft;	/* unused X-bits (as part of bpp) */
63 	uint32_t y_shft;	/* unused Y-bits (as part of bpp) */
64 	uint32_t cpp;		/* bytes/chars per pixel */
65 	uint32_t slot_w;	/* width of each slot (in pixels) */
66 	uint32_t slot_h;	/* height of each slot (in pixels) */
67 } geom[TILFMT_NFORMATS] = {
68 	[TILFMT_8BIT]  = GEOM(0, 0, 1),
69 	[TILFMT_16BIT] = GEOM(0, 1, 2),
70 	[TILFMT_32BIT] = GEOM(1, 1, 4),
71 	[TILFMT_PAGE]  = GEOM(SLOT_WIDTH_BITS, SLOT_HEIGHT_BITS, 1),
72 };
73 
74 
75 /* lookup table for registers w/ per-engine instances */
76 static const uint32_t reg[][4] = {
77 	[PAT_STATUS] = {DMM_PAT_STATUS__0, DMM_PAT_STATUS__1,
78 			DMM_PAT_STATUS__2, DMM_PAT_STATUS__3},
79 	[PAT_DESCR]  = {DMM_PAT_DESCR__0, DMM_PAT_DESCR__1,
80 			DMM_PAT_DESCR__2, DMM_PAT_DESCR__3},
81 };
82 
83 static u32 dmm_read(struct dmm *dmm, u32 reg)
84 {
85 	return readl(dmm->base + reg);
86 }
87 
88 static void dmm_write(struct dmm *dmm, u32 val, u32 reg)
89 {
90 	writel(val, dmm->base + reg);
91 }
92 
93 /* simple allocator to grab next 16 byte aligned memory from txn */
94 static void *alloc_dma(struct dmm_txn *txn, size_t sz, dma_addr_t *pa)
95 {
96 	void *ptr;
97 	struct refill_engine *engine = txn->engine_handle;
98 
99 	/* dmm programming requires 16 byte aligned addresses */
100 	txn->current_pa = round_up(txn->current_pa, 16);
101 	txn->current_va = (void *)round_up((long)txn->current_va, 16);
102 
103 	ptr = txn->current_va;
104 	*pa = txn->current_pa;
105 
106 	txn->current_pa += sz;
107 	txn->current_va += sz;
108 
109 	BUG_ON((txn->current_va - engine->refill_va) > REFILL_BUFFER_SIZE);
110 
111 	return ptr;
112 }
113 
114 /* check status and spin until wait_mask comes true */
115 static int wait_status(struct refill_engine *engine, uint32_t wait_mask)
116 {
117 	struct dmm *dmm = engine->dmm;
118 	uint32_t r = 0, err, i;
119 
120 	i = DMM_FIXED_RETRY_COUNT;
121 	while (true) {
122 		r = dmm_read(dmm, reg[PAT_STATUS][engine->id]);
123 		err = r & DMM_PATSTATUS_ERR;
124 		if (err)
125 			return -EFAULT;
126 
127 		if ((r & wait_mask) == wait_mask)
128 			break;
129 
130 		if (--i == 0)
131 			return -ETIMEDOUT;
132 
133 		udelay(1);
134 	}
135 
136 	return 0;
137 }
138 
139 static void release_engine(struct refill_engine *engine)
140 {
141 	unsigned long flags;
142 
143 	spin_lock_irqsave(&list_lock, flags);
144 	list_add(&engine->idle_node, &omap_dmm->idle_head);
145 	spin_unlock_irqrestore(&list_lock, flags);
146 
147 	atomic_inc(&omap_dmm->engine_counter);
148 	wake_up_interruptible(&omap_dmm->engine_queue);
149 }
150 
151 static irqreturn_t omap_dmm_irq_handler(int irq, void *arg)
152 {
153 	struct dmm *dmm = arg;
154 	uint32_t status = dmm_read(dmm, DMM_PAT_IRQSTATUS);
155 	int i;
156 
157 	/* ack IRQ */
158 	dmm_write(dmm, status, DMM_PAT_IRQSTATUS);
159 
160 	for (i = 0; i < dmm->num_engines; i++) {
161 		if (status & DMM_IRQSTAT_LST) {
162 			if (dmm->engines[i].async)
163 				release_engine(&dmm->engines[i]);
164 
165 			complete(&dmm->engines[i].compl);
166 		}
167 
168 		status >>= 8;
169 	}
170 
171 	return IRQ_HANDLED;
172 }
173 
174 /**
175  * Get a handle for a DMM transaction
176  */
177 static struct dmm_txn *dmm_txn_init(struct dmm *dmm, struct tcm *tcm)
178 {
179 	struct dmm_txn *txn = NULL;
180 	struct refill_engine *engine = NULL;
181 	int ret;
182 	unsigned long flags;
183 
184 
185 	/* wait until an engine is available */
186 	ret = wait_event_interruptible(omap_dmm->engine_queue,
187 		atomic_add_unless(&omap_dmm->engine_counter, -1, 0));
188 	if (ret)
189 		return ERR_PTR(ret);
190 
191 	/* grab an idle engine */
192 	spin_lock_irqsave(&list_lock, flags);
193 	if (!list_empty(&dmm->idle_head)) {
194 		engine = list_entry(dmm->idle_head.next, struct refill_engine,
195 					idle_node);
196 		list_del(&engine->idle_node);
197 	}
198 	spin_unlock_irqrestore(&list_lock, flags);
199 
200 	BUG_ON(!engine);
201 
202 	txn = &engine->txn;
203 	engine->tcm = tcm;
204 	txn->engine_handle = engine;
205 	txn->last_pat = NULL;
206 	txn->current_va = engine->refill_va;
207 	txn->current_pa = engine->refill_pa;
208 
209 	return txn;
210 }
211 
212 /**
213  * Add region to DMM transaction.  If pages or pages[i] is NULL, then the
214  * corresponding slot is cleared (ie. dummy_pa is programmed)
215  */
216 static void dmm_txn_append(struct dmm_txn *txn, struct pat_area *area,
217 		struct page **pages, uint32_t npages, uint32_t roll)
218 {
219 	dma_addr_t pat_pa = 0, data_pa = 0;
220 	uint32_t *data;
221 	struct pat *pat;
222 	struct refill_engine *engine = txn->engine_handle;
223 	int columns = (1 + area->x1 - area->x0);
224 	int rows = (1 + area->y1 - area->y0);
225 	int i = columns*rows;
226 
227 	pat = alloc_dma(txn, sizeof(struct pat), &pat_pa);
228 
229 	if (txn->last_pat)
230 		txn->last_pat->next_pa = (uint32_t)pat_pa;
231 
232 	pat->area = *area;
233 
234 	/* adjust Y coordinates based off of container parameters */
235 	pat->area.y0 += engine->tcm->y_offset;
236 	pat->area.y1 += engine->tcm->y_offset;
237 
238 	pat->ctrl = (struct pat_ctrl){
239 			.start = 1,
240 			.lut_id = engine->tcm->lut_id,
241 		};
242 
243 	data = alloc_dma(txn, 4*i, &data_pa);
244 	/* FIXME: what if data_pa is more than 32-bit ? */
245 	pat->data_pa = data_pa;
246 
247 	while (i--) {
248 		int n = i + roll;
249 		if (n >= npages)
250 			n -= npages;
251 		data[i] = (pages && pages[n]) ?
252 			page_to_phys(pages[n]) : engine->dmm->dummy_pa;
253 	}
254 
255 	txn->last_pat = pat;
256 
257 	return;
258 }
259 
260 /**
261  * Commit the DMM transaction.
262  */
263 static int dmm_txn_commit(struct dmm_txn *txn, bool wait)
264 {
265 	int ret = 0;
266 	struct refill_engine *engine = txn->engine_handle;
267 	struct dmm *dmm = engine->dmm;
268 
269 	if (!txn->last_pat) {
270 		dev_err(engine->dmm->dev, "need at least one txn\n");
271 		ret = -EINVAL;
272 		goto cleanup;
273 	}
274 
275 	txn->last_pat->next_pa = 0;
276 
277 	/* write to PAT_DESCR to clear out any pending transaction */
278 	dmm_write(dmm, 0x0, reg[PAT_DESCR][engine->id]);
279 
280 	/* wait for engine ready: */
281 	ret = wait_status(engine, DMM_PATSTATUS_READY);
282 	if (ret) {
283 		ret = -EFAULT;
284 		goto cleanup;
285 	}
286 
287 	/* mark whether it is async to denote list management in IRQ handler */
288 	engine->async = wait ? false : true;
289 	reinit_completion(&engine->compl);
290 	/* verify that the irq handler sees the 'async' and completion value */
291 	smp_mb();
292 
293 	/* kick reload */
294 	dmm_write(dmm, engine->refill_pa, reg[PAT_DESCR][engine->id]);
295 
296 	if (wait) {
297 		if (!wait_for_completion_timeout(&engine->compl,
298 				msecs_to_jiffies(100))) {
299 			dev_err(dmm->dev, "timed out waiting for done\n");
300 			ret = -ETIMEDOUT;
301 		}
302 	}
303 
304 cleanup:
305 	/* only place engine back on list if we are done with it */
306 	if (ret || wait)
307 		release_engine(engine);
308 
309 	return ret;
310 }
311 
312 /*
313  * DMM programming
314  */
315 static int fill(struct tcm_area *area, struct page **pages,
316 		uint32_t npages, uint32_t roll, bool wait)
317 {
318 	int ret = 0;
319 	struct tcm_area slice, area_s;
320 	struct dmm_txn *txn;
321 
322 	/*
323 	 * FIXME
324 	 *
325 	 * Asynchronous fill does not work reliably, as the driver does not
326 	 * handle errors in the async code paths. The fill operation may
327 	 * silently fail, leading to leaking DMM engines, which may eventually
328 	 * lead to deadlock if we run out of DMM engines.
329 	 *
330 	 * For now, always set 'wait' so that we only use sync fills. Async
331 	 * fills should be fixed, or alternatively we could decide to only
332 	 * support sync fills and so the whole async code path could be removed.
333 	 */
334 
335 	wait = true;
336 
337 	txn = dmm_txn_init(omap_dmm, area->tcm);
338 	if (IS_ERR_OR_NULL(txn))
339 		return -ENOMEM;
340 
341 	tcm_for_each_slice(slice, *area, area_s) {
342 		struct pat_area p_area = {
343 				.x0 = slice.p0.x,  .y0 = slice.p0.y,
344 				.x1 = slice.p1.x,  .y1 = slice.p1.y,
345 		};
346 
347 		dmm_txn_append(txn, &p_area, pages, npages, roll);
348 
349 		roll += tcm_sizeof(slice);
350 	}
351 
352 	ret = dmm_txn_commit(txn, wait);
353 
354 	return ret;
355 }
356 
357 /*
358  * Pin/unpin
359  */
360 
361 /* note: slots for which pages[i] == NULL are filled w/ dummy page
362  */
363 int tiler_pin(struct tiler_block *block, struct page **pages,
364 		uint32_t npages, uint32_t roll, bool wait)
365 {
366 	int ret;
367 
368 	ret = fill(&block->area, pages, npages, roll, wait);
369 
370 	if (ret)
371 		tiler_unpin(block);
372 
373 	return ret;
374 }
375 
376 int tiler_unpin(struct tiler_block *block)
377 {
378 	return fill(&block->area, NULL, 0, 0, false);
379 }
380 
381 /*
382  * Reserve/release
383  */
384 struct tiler_block *tiler_reserve_2d(enum tiler_fmt fmt, uint16_t w,
385 		uint16_t h, uint16_t align)
386 {
387 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
388 	u32 min_align = 128;
389 	int ret;
390 	unsigned long flags;
391 	size_t slot_bytes;
392 
393 	BUG_ON(!validfmt(fmt));
394 
395 	/* convert width/height to slots */
396 	w = DIV_ROUND_UP(w, geom[fmt].slot_w);
397 	h = DIV_ROUND_UP(h, geom[fmt].slot_h);
398 
399 	/* convert alignment to slots */
400 	slot_bytes = geom[fmt].slot_w * geom[fmt].cpp;
401 	min_align = max(min_align, slot_bytes);
402 	align = (align > min_align) ? ALIGN(align, min_align) : min_align;
403 	align /= slot_bytes;
404 
405 	block->fmt = fmt;
406 
407 	ret = tcm_reserve_2d(containers[fmt], w, h, align, -1, slot_bytes,
408 			&block->area);
409 	if (ret) {
410 		kfree(block);
411 		return ERR_PTR(-ENOMEM);
412 	}
413 
414 	/* add to allocation list */
415 	spin_lock_irqsave(&list_lock, flags);
416 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
417 	spin_unlock_irqrestore(&list_lock, flags);
418 
419 	return block;
420 }
421 
422 struct tiler_block *tiler_reserve_1d(size_t size)
423 {
424 	struct tiler_block *block = kzalloc(sizeof(*block), GFP_KERNEL);
425 	int num_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
426 	unsigned long flags;
427 
428 	if (!block)
429 		return ERR_PTR(-ENOMEM);
430 
431 	block->fmt = TILFMT_PAGE;
432 
433 	if (tcm_reserve_1d(containers[TILFMT_PAGE], num_pages,
434 				&block->area)) {
435 		kfree(block);
436 		return ERR_PTR(-ENOMEM);
437 	}
438 
439 	spin_lock_irqsave(&list_lock, flags);
440 	list_add(&block->alloc_node, &omap_dmm->alloc_head);
441 	spin_unlock_irqrestore(&list_lock, flags);
442 
443 	return block;
444 }
445 
446 /* note: if you have pin'd pages, you should have already unpin'd first! */
447 int tiler_release(struct tiler_block *block)
448 {
449 	int ret = tcm_free(&block->area);
450 	unsigned long flags;
451 
452 	if (block->area.tcm)
453 		dev_err(omap_dmm->dev, "failed to release block\n");
454 
455 	spin_lock_irqsave(&list_lock, flags);
456 	list_del(&block->alloc_node);
457 	spin_unlock_irqrestore(&list_lock, flags);
458 
459 	kfree(block);
460 	return ret;
461 }
462 
463 /*
464  * Utils
465  */
466 
467 /* calculate the tiler space address of a pixel in a view orientation...
468  * below description copied from the display subsystem section of TRM:
469  *
470  * When the TILER is addressed, the bits:
471  *   [28:27] = 0x0 for 8-bit tiled
472  *             0x1 for 16-bit tiled
473  *             0x2 for 32-bit tiled
474  *             0x3 for page mode
475  *   [31:29] = 0x0 for 0-degree view
476  *             0x1 for 180-degree view + mirroring
477  *             0x2 for 0-degree view + mirroring
478  *             0x3 for 180-degree view
479  *             0x4 for 270-degree view + mirroring
480  *             0x5 for 270-degree view
481  *             0x6 for 90-degree view
482  *             0x7 for 90-degree view + mirroring
483  * Otherwise the bits indicated the corresponding bit address to access
484  * the SDRAM.
485  */
486 static u32 tiler_get_address(enum tiler_fmt fmt, u32 orient, u32 x, u32 y)
487 {
488 	u32 x_bits, y_bits, tmp, x_mask, y_mask, alignment;
489 
490 	x_bits = CONT_WIDTH_BITS - geom[fmt].x_shft;
491 	y_bits = CONT_HEIGHT_BITS - geom[fmt].y_shft;
492 	alignment = geom[fmt].x_shft + geom[fmt].y_shft;
493 
494 	/* validate coordinate */
495 	x_mask = MASK(x_bits);
496 	y_mask = MASK(y_bits);
497 
498 	if (x < 0 || x > x_mask || y < 0 || y > y_mask) {
499 		DBG("invalid coords: %u < 0 || %u > %u || %u < 0 || %u > %u",
500 				x, x, x_mask, y, y, y_mask);
501 		return 0;
502 	}
503 
504 	/* account for mirroring */
505 	if (orient & MASK_X_INVERT)
506 		x ^= x_mask;
507 	if (orient & MASK_Y_INVERT)
508 		y ^= y_mask;
509 
510 	/* get coordinate address */
511 	if (orient & MASK_XY_FLIP)
512 		tmp = ((x << y_bits) + y);
513 	else
514 		tmp = ((y << x_bits) + x);
515 
516 	return TIL_ADDR((tmp << alignment), orient, fmt);
517 }
518 
519 dma_addr_t tiler_ssptr(struct tiler_block *block)
520 {
521 	BUG_ON(!validfmt(block->fmt));
522 
523 	return TILVIEW_8BIT + tiler_get_address(block->fmt, 0,
524 			block->area.p0.x * geom[block->fmt].slot_w,
525 			block->area.p0.y * geom[block->fmt].slot_h);
526 }
527 
528 dma_addr_t tiler_tsptr(struct tiler_block *block, uint32_t orient,
529 		uint32_t x, uint32_t y)
530 {
531 	struct tcm_pt *p = &block->area.p0;
532 	BUG_ON(!validfmt(block->fmt));
533 
534 	return tiler_get_address(block->fmt, orient,
535 			(p->x * geom[block->fmt].slot_w) + x,
536 			(p->y * geom[block->fmt].slot_h) + y);
537 }
538 
539 void tiler_align(enum tiler_fmt fmt, uint16_t *w, uint16_t *h)
540 {
541 	BUG_ON(!validfmt(fmt));
542 	*w = round_up(*w, geom[fmt].slot_w);
543 	*h = round_up(*h, geom[fmt].slot_h);
544 }
545 
546 uint32_t tiler_stride(enum tiler_fmt fmt, uint32_t orient)
547 {
548 	BUG_ON(!validfmt(fmt));
549 
550 	if (orient & MASK_XY_FLIP)
551 		return 1 << (CONT_HEIGHT_BITS + geom[fmt].x_shft);
552 	else
553 		return 1 << (CONT_WIDTH_BITS + geom[fmt].y_shft);
554 }
555 
556 size_t tiler_size(enum tiler_fmt fmt, uint16_t w, uint16_t h)
557 {
558 	tiler_align(fmt, &w, &h);
559 	return geom[fmt].cpp * w * h;
560 }
561 
562 size_t tiler_vsize(enum tiler_fmt fmt, uint16_t w, uint16_t h)
563 {
564 	BUG_ON(!validfmt(fmt));
565 	return round_up(geom[fmt].cpp * w, PAGE_SIZE) * h;
566 }
567 
568 uint32_t tiler_get_cpu_cache_flags(void)
569 {
570 	return omap_dmm->plat_data->cpu_cache_flags;
571 }
572 
573 bool dmm_is_available(void)
574 {
575 	return omap_dmm ? true : false;
576 }
577 
578 static int omap_dmm_remove(struct platform_device *dev)
579 {
580 	struct tiler_block *block, *_block;
581 	int i;
582 	unsigned long flags;
583 
584 	if (omap_dmm) {
585 		/* free all area regions */
586 		spin_lock_irqsave(&list_lock, flags);
587 		list_for_each_entry_safe(block, _block, &omap_dmm->alloc_head,
588 					alloc_node) {
589 			list_del(&block->alloc_node);
590 			kfree(block);
591 		}
592 		spin_unlock_irqrestore(&list_lock, flags);
593 
594 		for (i = 0; i < omap_dmm->num_lut; i++)
595 			if (omap_dmm->tcm && omap_dmm->tcm[i])
596 				omap_dmm->tcm[i]->deinit(omap_dmm->tcm[i]);
597 		kfree(omap_dmm->tcm);
598 
599 		kfree(omap_dmm->engines);
600 		if (omap_dmm->refill_va)
601 			dma_free_wc(omap_dmm->dev,
602 				    REFILL_BUFFER_SIZE * omap_dmm->num_engines,
603 				    omap_dmm->refill_va, omap_dmm->refill_pa);
604 		if (omap_dmm->dummy_page)
605 			__free_page(omap_dmm->dummy_page);
606 
607 		if (omap_dmm->irq > 0)
608 			free_irq(omap_dmm->irq, omap_dmm);
609 
610 		iounmap(omap_dmm->base);
611 		kfree(omap_dmm);
612 		omap_dmm = NULL;
613 	}
614 
615 	return 0;
616 }
617 
618 static int omap_dmm_probe(struct platform_device *dev)
619 {
620 	int ret = -EFAULT, i;
621 	struct tcm_area area = {0};
622 	u32 hwinfo, pat_geom;
623 	struct resource *mem;
624 
625 	omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL);
626 	if (!omap_dmm)
627 		goto fail;
628 
629 	/* initialize lists */
630 	INIT_LIST_HEAD(&omap_dmm->alloc_head);
631 	INIT_LIST_HEAD(&omap_dmm->idle_head);
632 
633 	init_waitqueue_head(&omap_dmm->engine_queue);
634 
635 	if (dev->dev.of_node) {
636 		const struct of_device_id *match;
637 
638 		match = of_match_node(dmm_of_match, dev->dev.of_node);
639 		if (!match) {
640 			dev_err(&dev->dev, "failed to find matching device node\n");
641 			return -ENODEV;
642 		}
643 
644 		omap_dmm->plat_data = match->data;
645 	}
646 
647 	/* lookup hwmod data - base address and irq */
648 	mem = platform_get_resource(dev, IORESOURCE_MEM, 0);
649 	if (!mem) {
650 		dev_err(&dev->dev, "failed to get base address resource\n");
651 		goto fail;
652 	}
653 
654 	omap_dmm->base = ioremap(mem->start, SZ_2K);
655 
656 	if (!omap_dmm->base) {
657 		dev_err(&dev->dev, "failed to get dmm base address\n");
658 		goto fail;
659 	}
660 
661 	omap_dmm->irq = platform_get_irq(dev, 0);
662 	if (omap_dmm->irq < 0) {
663 		dev_err(&dev->dev, "failed to get IRQ resource\n");
664 		goto fail;
665 	}
666 
667 	omap_dmm->dev = &dev->dev;
668 
669 	hwinfo = dmm_read(omap_dmm, DMM_PAT_HWINFO);
670 	omap_dmm->num_engines = (hwinfo >> 24) & 0x1F;
671 	omap_dmm->num_lut = (hwinfo >> 16) & 0x1F;
672 	omap_dmm->container_width = 256;
673 	omap_dmm->container_height = 128;
674 
675 	atomic_set(&omap_dmm->engine_counter, omap_dmm->num_engines);
676 
677 	/* read out actual LUT width and height */
678 	pat_geom = dmm_read(omap_dmm, DMM_PAT_GEOMETRY);
679 	omap_dmm->lut_width = ((pat_geom >> 16) & 0xF) << 5;
680 	omap_dmm->lut_height = ((pat_geom >> 24) & 0xF) << 5;
681 
682 	/* increment LUT by one if on OMAP5 */
683 	/* LUT has twice the height, and is split into a separate container */
684 	if (omap_dmm->lut_height != omap_dmm->container_height)
685 		omap_dmm->num_lut++;
686 
687 	/* initialize DMM registers */
688 	dmm_write(omap_dmm, 0x88888888, DMM_PAT_VIEW__0);
689 	dmm_write(omap_dmm, 0x88888888, DMM_PAT_VIEW__1);
690 	dmm_write(omap_dmm, 0x80808080, DMM_PAT_VIEW_MAP__0);
691 	dmm_write(omap_dmm, 0x80000000, DMM_PAT_VIEW_MAP_BASE);
692 	dmm_write(omap_dmm, 0x88888888, DMM_TILER_OR__0);
693 	dmm_write(omap_dmm, 0x88888888, DMM_TILER_OR__1);
694 
695 	ret = request_irq(omap_dmm->irq, omap_dmm_irq_handler, IRQF_SHARED,
696 				"omap_dmm_irq_handler", omap_dmm);
697 
698 	if (ret) {
699 		dev_err(&dev->dev, "couldn't register IRQ %d, error %d\n",
700 			omap_dmm->irq, ret);
701 		omap_dmm->irq = -1;
702 		goto fail;
703 	}
704 
705 	/* Enable all interrupts for each refill engine except
706 	 * ERR_LUT_MISS<n> (which is just advisory, and we don't care
707 	 * about because we want to be able to refill live scanout
708 	 * buffers for accelerated pan/scroll) and FILL_DSC<n> which
709 	 * we just generally don't care about.
710 	 */
711 	dmm_write(omap_dmm, 0x7e7e7e7e, DMM_PAT_IRQENABLE_SET);
712 
713 	omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32);
714 	if (!omap_dmm->dummy_page) {
715 		dev_err(&dev->dev, "could not allocate dummy page\n");
716 		ret = -ENOMEM;
717 		goto fail;
718 	}
719 
720 	/* set dma mask for device */
721 	ret = dma_set_coherent_mask(&dev->dev, DMA_BIT_MASK(32));
722 	if (ret)
723 		goto fail;
724 
725 	omap_dmm->dummy_pa = page_to_phys(omap_dmm->dummy_page);
726 
727 	/* alloc refill memory */
728 	omap_dmm->refill_va = dma_alloc_wc(&dev->dev,
729 					   REFILL_BUFFER_SIZE * omap_dmm->num_engines,
730 					   &omap_dmm->refill_pa, GFP_KERNEL);
731 	if (!omap_dmm->refill_va) {
732 		dev_err(&dev->dev, "could not allocate refill memory\n");
733 		goto fail;
734 	}
735 
736 	/* alloc engines */
737 	omap_dmm->engines = kcalloc(omap_dmm->num_engines,
738 				    sizeof(struct refill_engine), GFP_KERNEL);
739 	if (!omap_dmm->engines) {
740 		ret = -ENOMEM;
741 		goto fail;
742 	}
743 
744 	for (i = 0; i < omap_dmm->num_engines; i++) {
745 		omap_dmm->engines[i].id = i;
746 		omap_dmm->engines[i].dmm = omap_dmm;
747 		omap_dmm->engines[i].refill_va = omap_dmm->refill_va +
748 						(REFILL_BUFFER_SIZE * i);
749 		omap_dmm->engines[i].refill_pa = omap_dmm->refill_pa +
750 						(REFILL_BUFFER_SIZE * i);
751 		init_completion(&omap_dmm->engines[i].compl);
752 
753 		list_add(&omap_dmm->engines[i].idle_node, &omap_dmm->idle_head);
754 	}
755 
756 	omap_dmm->tcm = kcalloc(omap_dmm->num_lut, sizeof(*omap_dmm->tcm),
757 				GFP_KERNEL);
758 	if (!omap_dmm->tcm) {
759 		ret = -ENOMEM;
760 		goto fail;
761 	}
762 
763 	/* init containers */
764 	/* Each LUT is associated with a TCM (container manager).  We use the
765 	   lut_id to denote the lut_id used to identify the correct LUT for
766 	   programming during reill operations */
767 	for (i = 0; i < omap_dmm->num_lut; i++) {
768 		omap_dmm->tcm[i] = sita_init(omap_dmm->container_width,
769 						omap_dmm->container_height);
770 
771 		if (!omap_dmm->tcm[i]) {
772 			dev_err(&dev->dev, "failed to allocate container\n");
773 			ret = -ENOMEM;
774 			goto fail;
775 		}
776 
777 		omap_dmm->tcm[i]->lut_id = i;
778 	}
779 
780 	/* assign access mode containers to applicable tcm container */
781 	/* OMAP 4 has 1 container for all 4 views */
782 	/* OMAP 5 has 2 containers, 1 for 2D and 1 for 1D */
783 	containers[TILFMT_8BIT] = omap_dmm->tcm[0];
784 	containers[TILFMT_16BIT] = omap_dmm->tcm[0];
785 	containers[TILFMT_32BIT] = omap_dmm->tcm[0];
786 
787 	if (omap_dmm->container_height != omap_dmm->lut_height) {
788 		/* second LUT is used for PAGE mode.  Programming must use
789 		   y offset that is added to all y coordinates.  LUT id is still
790 		   0, because it is the same LUT, just the upper 128 lines */
791 		containers[TILFMT_PAGE] = omap_dmm->tcm[1];
792 		omap_dmm->tcm[1]->y_offset = OMAP5_LUT_OFFSET;
793 		omap_dmm->tcm[1]->lut_id = 0;
794 	} else {
795 		containers[TILFMT_PAGE] = omap_dmm->tcm[0];
796 	}
797 
798 	area = (struct tcm_area) {
799 		.tcm = NULL,
800 		.p1.x = omap_dmm->container_width - 1,
801 		.p1.y = omap_dmm->container_height - 1,
802 	};
803 
804 	/* initialize all LUTs to dummy page entries */
805 	for (i = 0; i < omap_dmm->num_lut; i++) {
806 		area.tcm = omap_dmm->tcm[i];
807 		if (fill(&area, NULL, 0, 0, true))
808 			dev_err(omap_dmm->dev, "refill failed");
809 	}
810 
811 	dev_info(omap_dmm->dev, "initialized all PAT entries\n");
812 
813 	return 0;
814 
815 fail:
816 	if (omap_dmm_remove(dev))
817 		dev_err(&dev->dev, "cleanup failed\n");
818 	return ret;
819 }
820 
821 /*
822  * debugfs support
823  */
824 
825 #ifdef CONFIG_DEBUG_FS
826 
827 static const char *alphabet = "abcdefghijklmnopqrstuvwxyz"
828 				"ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
829 static const char *special = ".,:;'\"`~!^-+";
830 
831 static void fill_map(char **map, int xdiv, int ydiv, struct tcm_area *a,
832 							char c, bool ovw)
833 {
834 	int x, y;
835 	for (y = a->p0.y / ydiv; y <= a->p1.y / ydiv; y++)
836 		for (x = a->p0.x / xdiv; x <= a->p1.x / xdiv; x++)
837 			if (map[y][x] == ' ' || ovw)
838 				map[y][x] = c;
839 }
840 
841 static void fill_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p,
842 									char c)
843 {
844 	map[p->y / ydiv][p->x / xdiv] = c;
845 }
846 
847 static char read_map_pt(char **map, int xdiv, int ydiv, struct tcm_pt *p)
848 {
849 	return map[p->y / ydiv][p->x / xdiv];
850 }
851 
852 static int map_width(int xdiv, int x0, int x1)
853 {
854 	return (x1 / xdiv) - (x0 / xdiv) + 1;
855 }
856 
857 static void text_map(char **map, int xdiv, char *nice, int yd, int x0, int x1)
858 {
859 	char *p = map[yd] + (x0 / xdiv);
860 	int w = (map_width(xdiv, x0, x1) - strlen(nice)) / 2;
861 	if (w >= 0) {
862 		p += w;
863 		while (*nice)
864 			*p++ = *nice++;
865 	}
866 }
867 
868 static void map_1d_info(char **map, int xdiv, int ydiv, char *nice,
869 							struct tcm_area *a)
870 {
871 	sprintf(nice, "%dK", tcm_sizeof(*a) * 4);
872 	if (a->p0.y + 1 < a->p1.y) {
873 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv, 0,
874 							256 - 1);
875 	} else if (a->p0.y < a->p1.y) {
876 		if (strlen(nice) < map_width(xdiv, a->p0.x, 256 - 1))
877 			text_map(map, xdiv, nice, a->p0.y / ydiv,
878 					a->p0.x + xdiv,	256 - 1);
879 		else if (strlen(nice) < map_width(xdiv, 0, a->p1.x))
880 			text_map(map, xdiv, nice, a->p1.y / ydiv,
881 					0, a->p1.y - xdiv);
882 	} else if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x)) {
883 		text_map(map, xdiv, nice, a->p0.y / ydiv, a->p0.x, a->p1.x);
884 	}
885 }
886 
887 static void map_2d_info(char **map, int xdiv, int ydiv, char *nice,
888 							struct tcm_area *a)
889 {
890 	sprintf(nice, "(%d*%d)", tcm_awidth(*a), tcm_aheight(*a));
891 	if (strlen(nice) + 1 < map_width(xdiv, a->p0.x, a->p1.x))
892 		text_map(map, xdiv, nice, (a->p0.y + a->p1.y) / 2 / ydiv,
893 							a->p0.x, a->p1.x);
894 }
895 
896 int tiler_map_show(struct seq_file *s, void *arg)
897 {
898 	int xdiv = 2, ydiv = 1;
899 	char **map = NULL, *global_map;
900 	struct tiler_block *block;
901 	struct tcm_area a, p;
902 	int i;
903 	const char *m2d = alphabet;
904 	const char *a2d = special;
905 	const char *m2dp = m2d, *a2dp = a2d;
906 	char nice[128];
907 	int h_adj;
908 	int w_adj;
909 	unsigned long flags;
910 	int lut_idx;
911 
912 
913 	if (!omap_dmm) {
914 		/* early return if dmm/tiler device is not initialized */
915 		return 0;
916 	}
917 
918 	h_adj = omap_dmm->container_height / ydiv;
919 	w_adj = omap_dmm->container_width / xdiv;
920 
921 	map = kmalloc(h_adj * sizeof(*map), GFP_KERNEL);
922 	global_map = kmalloc((w_adj + 1) * h_adj, GFP_KERNEL);
923 
924 	if (!map || !global_map)
925 		goto error;
926 
927 	for (lut_idx = 0; lut_idx < omap_dmm->num_lut; lut_idx++) {
928 		memset(map, 0, h_adj * sizeof(*map));
929 		memset(global_map, ' ', (w_adj + 1) * h_adj);
930 
931 		for (i = 0; i < omap_dmm->container_height; i++) {
932 			map[i] = global_map + i * (w_adj + 1);
933 			map[i][w_adj] = 0;
934 		}
935 
936 		spin_lock_irqsave(&list_lock, flags);
937 
938 		list_for_each_entry(block, &omap_dmm->alloc_head, alloc_node) {
939 			if (block->area.tcm == omap_dmm->tcm[lut_idx]) {
940 				if (block->fmt != TILFMT_PAGE) {
941 					fill_map(map, xdiv, ydiv, &block->area,
942 						*m2dp, true);
943 					if (!*++a2dp)
944 						a2dp = a2d;
945 					if (!*++m2dp)
946 						m2dp = m2d;
947 					map_2d_info(map, xdiv, ydiv, nice,
948 							&block->area);
949 				} else {
950 					bool start = read_map_pt(map, xdiv,
951 						ydiv, &block->area.p0) == ' ';
952 					bool end = read_map_pt(map, xdiv, ydiv,
953 							&block->area.p1) == ' ';
954 
955 					tcm_for_each_slice(a, block->area, p)
956 						fill_map(map, xdiv, ydiv, &a,
957 							'=', true);
958 					fill_map_pt(map, xdiv, ydiv,
959 							&block->area.p0,
960 							start ? '<' : 'X');
961 					fill_map_pt(map, xdiv, ydiv,
962 							&block->area.p1,
963 							end ? '>' : 'X');
964 					map_1d_info(map, xdiv, ydiv, nice,
965 							&block->area);
966 				}
967 			}
968 		}
969 
970 		spin_unlock_irqrestore(&list_lock, flags);
971 
972 		if (s) {
973 			seq_printf(s, "CONTAINER %d DUMP BEGIN\n", lut_idx);
974 			for (i = 0; i < 128; i++)
975 				seq_printf(s, "%03d:%s\n", i, map[i]);
976 			seq_printf(s, "CONTAINER %d DUMP END\n", lut_idx);
977 		} else {
978 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP BEGIN\n",
979 				lut_idx);
980 			for (i = 0; i < 128; i++)
981 				dev_dbg(omap_dmm->dev, "%03d:%s\n", i, map[i]);
982 			dev_dbg(omap_dmm->dev, "CONTAINER %d DUMP END\n",
983 				lut_idx);
984 		}
985 	}
986 
987 error:
988 	kfree(map);
989 	kfree(global_map);
990 
991 	return 0;
992 }
993 #endif
994 
995 #ifdef CONFIG_PM_SLEEP
996 static int omap_dmm_resume(struct device *dev)
997 {
998 	struct tcm_area area;
999 	int i;
1000 
1001 	if (!omap_dmm)
1002 		return -ENODEV;
1003 
1004 	area = (struct tcm_area) {
1005 		.tcm = NULL,
1006 		.p1.x = omap_dmm->container_width - 1,
1007 		.p1.y = omap_dmm->container_height - 1,
1008 	};
1009 
1010 	/* initialize all LUTs to dummy page entries */
1011 	for (i = 0; i < omap_dmm->num_lut; i++) {
1012 		area.tcm = omap_dmm->tcm[i];
1013 		if (fill(&area, NULL, 0, 0, true))
1014 			dev_err(dev, "refill failed");
1015 	}
1016 
1017 	return 0;
1018 }
1019 #endif
1020 
1021 static SIMPLE_DEV_PM_OPS(omap_dmm_pm_ops, NULL, omap_dmm_resume);
1022 
1023 #if defined(CONFIG_OF)
1024 static const struct dmm_platform_data dmm_omap4_platform_data = {
1025 	.cpu_cache_flags = OMAP_BO_WC,
1026 };
1027 
1028 static const struct dmm_platform_data dmm_omap5_platform_data = {
1029 	.cpu_cache_flags = OMAP_BO_UNCACHED,
1030 };
1031 
1032 static const struct of_device_id dmm_of_match[] = {
1033 	{
1034 		.compatible = "ti,omap4-dmm",
1035 		.data = &dmm_omap4_platform_data,
1036 	},
1037 	{
1038 		.compatible = "ti,omap5-dmm",
1039 		.data = &dmm_omap5_platform_data,
1040 	},
1041 	{},
1042 };
1043 #endif
1044 
1045 struct platform_driver omap_dmm_driver = {
1046 	.probe = omap_dmm_probe,
1047 	.remove = omap_dmm_remove,
1048 	.driver = {
1049 		.owner = THIS_MODULE,
1050 		.name = DMM_DRIVER_NAME,
1051 		.of_match_table = of_match_ptr(dmm_of_match),
1052 		.pm = &omap_dmm_pm_ops,
1053 	},
1054 };
1055 
1056 MODULE_LICENSE("GPL v2");
1057 MODULE_AUTHOR("Andy Gross <andy.gross@ti.com>");
1058 MODULE_DESCRIPTION("OMAP DMM/Tiler Driver");
1059