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