1/*
2 * Tegra host1x Job
3 *
4 * Copyright (c) 2010-2013, NVIDIA Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17 */
18
19#include <linux/dma-mapping.h>
20#include <linux/err.h>
21#include <linux/host1x.h>
22#include <linux/kref.h>
23#include <linux/module.h>
24#include <linux/scatterlist.h>
25#include <linux/slab.h>
26#include <linux/vmalloc.h>
27#include <trace/events/host1x.h>
28
29#include "channel.h"
30#include "dev.h"
31#include "job.h"
32#include "syncpt.h"
33
34struct host1x_job *host1x_job_alloc(struct host1x_channel *ch,
35				    u32 num_cmdbufs, u32 num_relocs,
36				    u32 num_waitchks)
37{
38	struct host1x_job *job = NULL;
39	unsigned int num_unpins = num_cmdbufs + num_relocs;
40	u64 total;
41	void *mem;
42
43	/* Check that we're not going to overflow */
44	total = sizeof(struct host1x_job) +
45		(u64)num_relocs * sizeof(struct host1x_reloc) +
46		(u64)num_unpins * sizeof(struct host1x_job_unpin_data) +
47		(u64)num_waitchks * sizeof(struct host1x_waitchk) +
48		(u64)num_cmdbufs * sizeof(struct host1x_job_gather) +
49		(u64)num_unpins * sizeof(dma_addr_t) +
50		(u64)num_unpins * sizeof(u32 *);
51	if (total > ULONG_MAX)
52		return NULL;
53
54	mem = job = kzalloc(total, GFP_KERNEL);
55	if (!job)
56		return NULL;
57
58	kref_init(&job->ref);
59	job->channel = ch;
60
61	/* Redistribute memory to the structs  */
62	mem += sizeof(struct host1x_job);
63	job->relocarray = num_relocs ? mem : NULL;
64	mem += num_relocs * sizeof(struct host1x_reloc);
65	job->unpins = num_unpins ? mem : NULL;
66	mem += num_unpins * sizeof(struct host1x_job_unpin_data);
67	job->waitchk = num_waitchks ? mem : NULL;
68	mem += num_waitchks * sizeof(struct host1x_waitchk);
69	job->gathers = num_cmdbufs ? mem : NULL;
70	mem += num_cmdbufs * sizeof(struct host1x_job_gather);
71	job->addr_phys = num_unpins ? mem : NULL;
72
73	job->reloc_addr_phys = job->addr_phys;
74	job->gather_addr_phys = &job->addr_phys[num_relocs];
75
76	return job;
77}
78EXPORT_SYMBOL(host1x_job_alloc);
79
80struct host1x_job *host1x_job_get(struct host1x_job *job)
81{
82	kref_get(&job->ref);
83	return job;
84}
85EXPORT_SYMBOL(host1x_job_get);
86
87static void job_free(struct kref *ref)
88{
89	struct host1x_job *job = container_of(ref, struct host1x_job, ref);
90
91	kfree(job);
92}
93
94void host1x_job_put(struct host1x_job *job)
95{
96	kref_put(&job->ref, job_free);
97}
98EXPORT_SYMBOL(host1x_job_put);
99
100void host1x_job_add_gather(struct host1x_job *job, struct host1x_bo *bo,
101			   u32 words, u32 offset)
102{
103	struct host1x_job_gather *cur_gather = &job->gathers[job->num_gathers];
104
105	cur_gather->words = words;
106	cur_gather->bo = bo;
107	cur_gather->offset = offset;
108	job->num_gathers++;
109}
110EXPORT_SYMBOL(host1x_job_add_gather);
111
112/*
113 * NULL an already satisfied WAIT_SYNCPT host method, by patching its
114 * args in the command stream. The method data is changed to reference
115 * a reserved (never given out or incr) HOST1X_SYNCPT_RESERVED syncpt
116 * with a matching threshold value of 0, so is guaranteed to be popped
117 * by the host HW.
118 */
119static void host1x_syncpt_patch_offset(struct host1x_syncpt *sp,
120				       struct host1x_bo *h, u32 offset)
121{
122	void *patch_addr = NULL;
123
124	/* patch the wait */
125	patch_addr = host1x_bo_kmap(h, offset >> PAGE_SHIFT);
126	if (patch_addr) {
127		host1x_syncpt_patch_wait(sp,
128					 patch_addr + (offset & ~PAGE_MASK));
129		host1x_bo_kunmap(h, offset >> PAGE_SHIFT, patch_addr);
130	} else
131		pr_err("Could not map cmdbuf for wait check\n");
132}
133
134/*
135 * Check driver supplied waitchk structs for syncpt thresholds
136 * that have already been satisfied and NULL the comparison (to
137 * avoid a wrap condition in the HW).
138 */
139static int do_waitchks(struct host1x_job *job, struct host1x *host,
140		       struct host1x_bo *patch)
141{
142	int i;
143
144	/* compare syncpt vs wait threshold */
145	for (i = 0; i < job->num_waitchk; i++) {
146		struct host1x_waitchk *wait = &job->waitchk[i];
147		struct host1x_syncpt *sp =
148			host1x_syncpt_get(host, wait->syncpt_id);
149
150		/* validate syncpt id */
151		if (wait->syncpt_id > host1x_syncpt_nb_pts(host))
152			continue;
153
154		/* skip all other gathers */
155		if (patch != wait->bo)
156			continue;
157
158		trace_host1x_syncpt_wait_check(wait->bo, wait->offset,
159					       wait->syncpt_id, wait->thresh,
160					       host1x_syncpt_read_min(sp));
161
162		if (host1x_syncpt_is_expired(sp, wait->thresh)) {
163			dev_dbg(host->dev,
164				"drop WAIT id %d (%s) thresh 0x%x, min 0x%x\n",
165				wait->syncpt_id, sp->name, wait->thresh,
166				host1x_syncpt_read_min(sp));
167
168			host1x_syncpt_patch_offset(sp, patch, wait->offset);
169		}
170
171		wait->bo = NULL;
172	}
173
174	return 0;
175}
176
177static unsigned int pin_job(struct host1x_job *job)
178{
179	unsigned int i;
180
181	job->num_unpins = 0;
182
183	for (i = 0; i < job->num_relocs; i++) {
184		struct host1x_reloc *reloc = &job->relocarray[i];
185		struct sg_table *sgt;
186		dma_addr_t phys_addr;
187
188		reloc->target.bo = host1x_bo_get(reloc->target.bo);
189		if (!reloc->target.bo)
190			goto unpin;
191
192		phys_addr = host1x_bo_pin(reloc->target.bo, &sgt);
193		if (!phys_addr)
194			goto unpin;
195
196		job->addr_phys[job->num_unpins] = phys_addr;
197		job->unpins[job->num_unpins].bo = reloc->target.bo;
198		job->unpins[job->num_unpins].sgt = sgt;
199		job->num_unpins++;
200	}
201
202	for (i = 0; i < job->num_gathers; i++) {
203		struct host1x_job_gather *g = &job->gathers[i];
204		struct sg_table *sgt;
205		dma_addr_t phys_addr;
206
207		g->bo = host1x_bo_get(g->bo);
208		if (!g->bo)
209			goto unpin;
210
211		phys_addr = host1x_bo_pin(g->bo, &sgt);
212		if (!phys_addr)
213			goto unpin;
214
215		job->addr_phys[job->num_unpins] = phys_addr;
216		job->unpins[job->num_unpins].bo = g->bo;
217		job->unpins[job->num_unpins].sgt = sgt;
218		job->num_unpins++;
219	}
220
221	return job->num_unpins;
222
223unpin:
224	host1x_job_unpin(job);
225	return 0;
226}
227
228static unsigned int do_relocs(struct host1x_job *job, struct host1x_bo *cmdbuf)
229{
230	int i = 0;
231	u32 last_page = ~0;
232	void *cmdbuf_page_addr = NULL;
233
234	/* pin & patch the relocs for one gather */
235	for (i = 0; i < job->num_relocs; i++) {
236		struct host1x_reloc *reloc = &job->relocarray[i];
237		u32 reloc_addr = (job->reloc_addr_phys[i] +
238				  reloc->target.offset) >> reloc->shift;
239		u32 *target;
240
241		/* skip all other gathers */
242		if (cmdbuf != reloc->cmdbuf.bo)
243			continue;
244
245		if (last_page != reloc->cmdbuf.offset >> PAGE_SHIFT) {
246			if (cmdbuf_page_addr)
247				host1x_bo_kunmap(cmdbuf, last_page,
248						 cmdbuf_page_addr);
249
250			cmdbuf_page_addr = host1x_bo_kmap(cmdbuf,
251					reloc->cmdbuf.offset >> PAGE_SHIFT);
252			last_page = reloc->cmdbuf.offset >> PAGE_SHIFT;
253
254			if (unlikely(!cmdbuf_page_addr)) {
255				pr_err("Could not map cmdbuf for relocation\n");
256				return -ENOMEM;
257			}
258		}
259
260		target = cmdbuf_page_addr + (reloc->cmdbuf.offset & ~PAGE_MASK);
261		*target = reloc_addr;
262	}
263
264	if (cmdbuf_page_addr)
265		host1x_bo_kunmap(cmdbuf, last_page, cmdbuf_page_addr);
266
267	return 0;
268}
269
270static bool check_reloc(struct host1x_reloc *reloc, struct host1x_bo *cmdbuf,
271			unsigned int offset)
272{
273	offset *= sizeof(u32);
274
275	if (reloc->cmdbuf.bo != cmdbuf || reloc->cmdbuf.offset != offset)
276		return false;
277
278	return true;
279}
280
281struct host1x_firewall {
282	struct host1x_job *job;
283	struct device *dev;
284
285	unsigned int num_relocs;
286	struct host1x_reloc *reloc;
287
288	struct host1x_bo *cmdbuf;
289	unsigned int offset;
290
291	u32 words;
292	u32 class;
293	u32 reg;
294	u32 mask;
295	u32 count;
296};
297
298static int check_register(struct host1x_firewall *fw, unsigned long offset)
299{
300	if (fw->job->is_addr_reg(fw->dev, fw->class, offset)) {
301		if (!fw->num_relocs)
302			return -EINVAL;
303
304		if (!check_reloc(fw->reloc, fw->cmdbuf, fw->offset))
305			return -EINVAL;
306
307		fw->num_relocs--;
308		fw->reloc++;
309	}
310
311	return 0;
312}
313
314static int check_mask(struct host1x_firewall *fw)
315{
316	u32 mask = fw->mask;
317	u32 reg = fw->reg;
318	int ret;
319
320	while (mask) {
321		if (fw->words == 0)
322			return -EINVAL;
323
324		if (mask & 1) {
325			ret = check_register(fw, reg);
326			if (ret < 0)
327				return ret;
328
329			fw->words--;
330			fw->offset++;
331		}
332		mask >>= 1;
333		reg++;
334	}
335
336	return 0;
337}
338
339static int check_incr(struct host1x_firewall *fw)
340{
341	u32 count = fw->count;
342	u32 reg = fw->reg;
343	int ret;
344
345	while (count) {
346		if (fw->words == 0)
347			return -EINVAL;
348
349		ret = check_register(fw, reg);
350		if (ret < 0)
351			return ret;
352
353		reg++;
354		fw->words--;
355		fw->offset++;
356		count--;
357	}
358
359	return 0;
360}
361
362static int check_nonincr(struct host1x_firewall *fw)
363{
364	u32 count = fw->count;
365	int ret;
366
367	while (count) {
368		if (fw->words == 0)
369			return -EINVAL;
370
371		ret = check_register(fw, fw->reg);
372		if (ret < 0)
373			return ret;
374
375		fw->words--;
376		fw->offset++;
377		count--;
378	}
379
380	return 0;
381}
382
383static int validate(struct host1x_firewall *fw, struct host1x_job_gather *g)
384{
385	u32 *cmdbuf_base = (u32 *)fw->job->gather_copy_mapped +
386		(g->offset / sizeof(u32));
387	int err = 0;
388
389	if (!fw->job->is_addr_reg)
390		return 0;
391
392	fw->words = g->words;
393	fw->cmdbuf = g->bo;
394	fw->offset = 0;
395
396	while (fw->words && !err) {
397		u32 word = cmdbuf_base[fw->offset];
398		u32 opcode = (word & 0xf0000000) >> 28;
399
400		fw->mask = 0;
401		fw->reg = 0;
402		fw->count = 0;
403		fw->words--;
404		fw->offset++;
405
406		switch (opcode) {
407		case 0:
408			fw->class = word >> 6 & 0x3ff;
409			fw->mask = word & 0x3f;
410			fw->reg = word >> 16 & 0xfff;
411			err = check_mask(fw);
412			if (err)
413				goto out;
414			break;
415		case 1:
416			fw->reg = word >> 16 & 0xfff;
417			fw->count = word & 0xffff;
418			err = check_incr(fw);
419			if (err)
420				goto out;
421			break;
422
423		case 2:
424			fw->reg = word >> 16 & 0xfff;
425			fw->count = word & 0xffff;
426			err = check_nonincr(fw);
427			if (err)
428				goto out;
429			break;
430
431		case 3:
432			fw->mask = word & 0xffff;
433			fw->reg = word >> 16 & 0xfff;
434			err = check_mask(fw);
435			if (err)
436				goto out;
437			break;
438		case 4:
439		case 5:
440		case 14:
441			break;
442		default:
443			err = -EINVAL;
444			break;
445		}
446	}
447
448out:
449	return err;
450}
451
452static inline int copy_gathers(struct host1x_job *job, struct device *dev)
453{
454	struct host1x_firewall fw;
455	size_t size = 0;
456	size_t offset = 0;
457	int i;
458
459	fw.job = job;
460	fw.dev = dev;
461	fw.reloc = job->relocarray;
462	fw.num_relocs = job->num_relocs;
463	fw.class = 0;
464
465	for (i = 0; i < job->num_gathers; i++) {
466		struct host1x_job_gather *g = &job->gathers[i];
467		size += g->words * sizeof(u32);
468	}
469
470	job->gather_copy_mapped = dma_alloc_writecombine(dev, size,
471							 &job->gather_copy,
472							 GFP_KERNEL);
473	if (!job->gather_copy_mapped) {
474		job->gather_copy_mapped = NULL;
475		return -ENOMEM;
476	}
477
478	job->gather_copy_size = size;
479
480	for (i = 0; i < job->num_gathers; i++) {
481		struct host1x_job_gather *g = &job->gathers[i];
482		void *gather;
483
484		/* Copy the gather */
485		gather = host1x_bo_mmap(g->bo);
486		memcpy(job->gather_copy_mapped + offset, gather + g->offset,
487		       g->words * sizeof(u32));
488		host1x_bo_munmap(g->bo, gather);
489
490		/* Store the location in the buffer */
491		g->base = job->gather_copy;
492		g->offset = offset;
493
494		/* Validate the job */
495		if (validate(&fw, g))
496			return -EINVAL;
497
498		offset += g->words * sizeof(u32);
499	}
500
501	/* No relocs should remain at this point */
502	if (fw.num_relocs)
503		return -EINVAL;
504
505	return 0;
506}
507
508int host1x_job_pin(struct host1x_job *job, struct device *dev)
509{
510	int err;
511	unsigned int i, j;
512	struct host1x *host = dev_get_drvdata(dev->parent);
513	DECLARE_BITMAP(waitchk_mask, host1x_syncpt_nb_pts(host));
514
515	bitmap_zero(waitchk_mask, host1x_syncpt_nb_pts(host));
516	for (i = 0; i < job->num_waitchk; i++) {
517		u32 syncpt_id = job->waitchk[i].syncpt_id;
518		if (syncpt_id < host1x_syncpt_nb_pts(host))
519			set_bit(syncpt_id, waitchk_mask);
520	}
521
522	/* get current syncpt values for waitchk */
523	for_each_set_bit(i, waitchk_mask, host1x_syncpt_nb_pts(host))
524		host1x_syncpt_load(host->syncpt + i);
525
526	/* pin memory */
527	err = pin_job(job);
528	if (!err)
529		goto out;
530
531	/* patch gathers */
532	for (i = 0; i < job->num_gathers; i++) {
533		struct host1x_job_gather *g = &job->gathers[i];
534
535		/* process each gather mem only once */
536		if (g->handled)
537			continue;
538
539		g->base = job->gather_addr_phys[i];
540
541		for (j = i + 1; j < job->num_gathers; j++)
542			if (job->gathers[j].bo == g->bo)
543				job->gathers[j].handled = true;
544
545		err = do_relocs(job, g->bo);
546		if (err)
547			break;
548
549		err = do_waitchks(job, host, g->bo);
550		if (err)
551			break;
552	}
553
554	if (IS_ENABLED(CONFIG_TEGRA_HOST1X_FIREWALL) && !err) {
555		err = copy_gathers(job, dev);
556		if (err) {
557			host1x_job_unpin(job);
558			return err;
559		}
560	}
561
562out:
563	wmb();
564
565	return err;
566}
567EXPORT_SYMBOL(host1x_job_pin);
568
569void host1x_job_unpin(struct host1x_job *job)
570{
571	unsigned int i;
572
573	for (i = 0; i < job->num_unpins; i++) {
574		struct host1x_job_unpin_data *unpin = &job->unpins[i];
575		host1x_bo_unpin(unpin->bo, unpin->sgt);
576		host1x_bo_put(unpin->bo);
577	}
578	job->num_unpins = 0;
579
580	if (job->gather_copy_size)
581		dma_free_writecombine(job->channel->dev, job->gather_copy_size,
582				      job->gather_copy_mapped,
583				      job->gather_copy);
584}
585EXPORT_SYMBOL(host1x_job_unpin);
586
587/*
588 * Debug routine used to dump job entries
589 */
590void host1x_job_dump(struct device *dev, struct host1x_job *job)
591{
592	dev_dbg(dev, "    SYNCPT_ID   %d\n", job->syncpt_id);
593	dev_dbg(dev, "    SYNCPT_VAL  %d\n", job->syncpt_end);
594	dev_dbg(dev, "    FIRST_GET   0x%x\n", job->first_get);
595	dev_dbg(dev, "    TIMEOUT     %d\n", job->timeout);
596	dev_dbg(dev, "    NUM_SLOTS   %d\n", job->num_slots);
597	dev_dbg(dev, "    NUM_HANDLES %d\n", job->num_unpins);
598}
599