1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 *  cx18 buffer queues
4 *
5 *  Derived from ivtv-queue.c
6 *
7 *  Copyright (C) 2007  Hans Verkuil <hverkuil@xs4all.nl>
8 *  Copyright (C) 2008  Andy Walls <awalls@md.metrocast.net>
9 */
10
11#include "cx18-driver.h"
12#include "cx18-queue.h"
13#include "cx18-streams.h"
14#include "cx18-scb.h"
15#include "cx18-io.h"
16
17void cx18_buf_swap(struct cx18_buffer *buf)
18{
19	int i;
20
21	for (i = 0; i < buf->bytesused; i += 4)
22		swab32s((u32 *)(buf->buf + i));
23}
24
25void _cx18_mdl_swap(struct cx18_mdl *mdl)
26{
27	struct cx18_buffer *buf;
28
29	list_for_each_entry(buf, &mdl->buf_list, list) {
30		if (buf->bytesused == 0)
31			break;
32		cx18_buf_swap(buf);
33	}
34}
35
36void cx18_queue_init(struct cx18_queue *q)
37{
38	INIT_LIST_HEAD(&q->list);
39	atomic_set(&q->depth, 0);
40	q->bytesused = 0;
41}
42
43struct cx18_queue *_cx18_enqueue(struct cx18_stream *s, struct cx18_mdl *mdl,
44				 struct cx18_queue *q, int to_front)
45{
46	/* clear the mdl if it is not to be enqueued to the full queue */
47	if (q != &s->q_full) {
48		mdl->bytesused = 0;
49		mdl->readpos = 0;
50		mdl->m_flags = 0;
51		mdl->skipped = 0;
52		mdl->curr_buf = NULL;
53	}
54
55	/* q_busy is restricted to a max buffer count imposed by firmware */
56	if (q == &s->q_busy &&
57	    atomic_read(&q->depth) >= CX18_MAX_FW_MDLS_PER_STREAM)
58		q = &s->q_free;
59
60	spin_lock(&q->lock);
61
62	if (to_front)
63		list_add(&mdl->list, &q->list); /* LIFO */
64	else
65		list_add_tail(&mdl->list, &q->list); /* FIFO */
66	q->bytesused += mdl->bytesused - mdl->readpos;
67	atomic_inc(&q->depth);
68
69	spin_unlock(&q->lock);
70	return q;
71}
72
73struct cx18_mdl *cx18_dequeue(struct cx18_stream *s, struct cx18_queue *q)
74{
75	struct cx18_mdl *mdl = NULL;
76
77	spin_lock(&q->lock);
78	if (!list_empty(&q->list)) {
79		mdl = list_first_entry(&q->list, struct cx18_mdl, list);
80		list_del_init(&mdl->list);
81		q->bytesused -= mdl->bytesused - mdl->readpos;
82		mdl->skipped = 0;
83		atomic_dec(&q->depth);
84	}
85	spin_unlock(&q->lock);
86	return mdl;
87}
88
89static void _cx18_mdl_update_bufs_for_cpu(struct cx18_stream *s,
90					  struct cx18_mdl *mdl)
91{
92	struct cx18_buffer *buf;
93	u32 buf_size = s->buf_size;
94	u32 bytesused = mdl->bytesused;
95
96	list_for_each_entry(buf, &mdl->buf_list, list) {
97		buf->readpos = 0;
98		if (bytesused >= buf_size) {
99			buf->bytesused = buf_size;
100			bytesused -= buf_size;
101		} else {
102			buf->bytesused = bytesused;
103			bytesused = 0;
104		}
105		cx18_buf_sync_for_cpu(s, buf);
106	}
107}
108
109static inline void cx18_mdl_update_bufs_for_cpu(struct cx18_stream *s,
110						struct cx18_mdl *mdl)
111{
112	struct cx18_buffer *buf;
113
114	if (list_is_singular(&mdl->buf_list)) {
115		buf = list_first_entry(&mdl->buf_list, struct cx18_buffer,
116				       list);
117		buf->bytesused = mdl->bytesused;
118		buf->readpos = 0;
119		cx18_buf_sync_for_cpu(s, buf);
120	} else {
121		_cx18_mdl_update_bufs_for_cpu(s, mdl);
122	}
123}
124
125struct cx18_mdl *cx18_queue_get_mdl(struct cx18_stream *s, u32 id,
126	u32 bytesused)
127{
128	struct cx18 *cx = s->cx;
129	struct cx18_mdl *mdl;
130	struct cx18_mdl *tmp;
131	struct cx18_mdl *ret = NULL;
132	LIST_HEAD(sweep_up);
133
134	/*
135	 * We don't have to acquire multiple q locks here, because we are
136	 * serialized by the single threaded work handler.
137	 * MDLs from the firmware will thus remain in order as
138	 * they are moved from q_busy to q_full or to the dvb ring buffer.
139	 */
140	spin_lock(&s->q_busy.lock);
141	list_for_each_entry_safe(mdl, tmp, &s->q_busy.list, list) {
142		/*
143		 * We should find what the firmware told us is done,
144		 * right at the front of the queue.  If we don't, we likely have
145		 * missed an mdl done message from the firmware.
146		 * Once we skip an mdl repeatedly, relative to the size of
147		 * q_busy, we have high confidence we've missed it.
148		 */
149		if (mdl->id != id) {
150			mdl->skipped++;
151			if (mdl->skipped >= atomic_read(&s->q_busy.depth)-1) {
152				/* mdl must have fallen out of rotation */
153				CX18_WARN("Skipped %s, MDL %d, %d times - it must have dropped out of rotation\n",
154					  s->name, mdl->id,
155					  mdl->skipped);
156				/* Sweep it up to put it back into rotation */
157				list_move_tail(&mdl->list, &sweep_up);
158				atomic_dec(&s->q_busy.depth);
159			}
160			continue;
161		}
162		/*
163		 * We pull the desired mdl off of the queue here.  Something
164		 * will have to put it back on a queue later.
165		 */
166		list_del_init(&mdl->list);
167		atomic_dec(&s->q_busy.depth);
168		ret = mdl;
169		break;
170	}
171	spin_unlock(&s->q_busy.lock);
172
173	/*
174	 * We found the mdl for which we were looking.  Get it ready for
175	 * the caller to put on q_full or in the dvb ring buffer.
176	 */
177	if (ret != NULL) {
178		ret->bytesused = bytesused;
179		ret->skipped = 0;
180		/* 0'ed readpos, m_flags & curr_buf when mdl went on q_busy */
181		cx18_mdl_update_bufs_for_cpu(s, ret);
182		if (s->type != CX18_ENC_STREAM_TYPE_TS)
183			set_bit(CX18_F_M_NEED_SWAP, &ret->m_flags);
184	}
185
186	/* Put any mdls the firmware is ignoring back into normal rotation */
187	list_for_each_entry_safe(mdl, tmp, &sweep_up, list) {
188		list_del_init(&mdl->list);
189		cx18_enqueue(s, mdl, &s->q_free);
190	}
191	return ret;
192}
193
194/* Move all mdls of a queue, while flushing the mdl */
195static void cx18_queue_flush(struct cx18_stream *s,
196			     struct cx18_queue *q_src, struct cx18_queue *q_dst)
197{
198	struct cx18_mdl *mdl;
199
200	/* It only makes sense to flush to q_free or q_idle */
201	if (q_src == q_dst || q_dst == &s->q_full || q_dst == &s->q_busy)
202		return;
203
204	spin_lock(&q_src->lock);
205	spin_lock(&q_dst->lock);
206	while (!list_empty(&q_src->list)) {
207		mdl = list_first_entry(&q_src->list, struct cx18_mdl, list);
208		list_move_tail(&mdl->list, &q_dst->list);
209		mdl->bytesused = 0;
210		mdl->readpos = 0;
211		mdl->m_flags = 0;
212		mdl->skipped = 0;
213		mdl->curr_buf = NULL;
214		atomic_inc(&q_dst->depth);
215	}
216	cx18_queue_init(q_src);
217	spin_unlock(&q_src->lock);
218	spin_unlock(&q_dst->lock);
219}
220
221void cx18_flush_queues(struct cx18_stream *s)
222{
223	cx18_queue_flush(s, &s->q_busy, &s->q_free);
224	cx18_queue_flush(s, &s->q_full, &s->q_free);
225}
226
227/*
228 * Note, s->buf_pool is not protected by a lock,
229 * the stream better not have *anything* going on when calling this
230 */
231void cx18_unload_queues(struct cx18_stream *s)
232{
233	struct cx18_queue *q_idle = &s->q_idle;
234	struct cx18_mdl *mdl;
235	struct cx18_buffer *buf;
236
237	/* Move all MDLS to q_idle */
238	cx18_queue_flush(s, &s->q_busy, q_idle);
239	cx18_queue_flush(s, &s->q_full, q_idle);
240	cx18_queue_flush(s, &s->q_free, q_idle);
241
242	/* Reset MDL id's and move all buffers back to the stream's buf_pool */
243	spin_lock(&q_idle->lock);
244	list_for_each_entry(mdl, &q_idle->list, list) {
245		while (!list_empty(&mdl->buf_list)) {
246			buf = list_first_entry(&mdl->buf_list,
247					       struct cx18_buffer, list);
248			list_move_tail(&buf->list, &s->buf_pool);
249			buf->bytesused = 0;
250			buf->readpos = 0;
251		}
252		mdl->id = s->mdl_base_idx; /* reset id to a "safe" value */
253		/* all other mdl fields were cleared by cx18_queue_flush() */
254	}
255	spin_unlock(&q_idle->lock);
256}
257
258/*
259 * Note, s->buf_pool is not protected by a lock,
260 * the stream better not have *anything* going on when calling this
261 */
262void cx18_load_queues(struct cx18_stream *s)
263{
264	struct cx18 *cx = s->cx;
265	struct cx18_mdl *mdl;
266	struct cx18_buffer *buf;
267	int mdl_id;
268	int i;
269	u32 partial_buf_size;
270
271	/*
272	 * Attach buffers to MDLs, give the MDLs ids, and add MDLs to q_free
273	 * Excess MDLs are left on q_idle
274	 * Excess buffers are left in buf_pool and/or on an MDL in q_idle
275	 */
276	mdl_id = s->mdl_base_idx;
277	for (mdl = cx18_dequeue(s, &s->q_idle), i = s->bufs_per_mdl;
278	     mdl != NULL && i == s->bufs_per_mdl;
279	     mdl = cx18_dequeue(s, &s->q_idle)) {
280
281		mdl->id = mdl_id;
282
283		for (i = 0; i < s->bufs_per_mdl; i++) {
284			if (list_empty(&s->buf_pool))
285				break;
286
287			buf = list_first_entry(&s->buf_pool, struct cx18_buffer,
288					       list);
289			list_move_tail(&buf->list, &mdl->buf_list);
290
291			/* update the firmware's MDL array with this buffer */
292			cx18_writel(cx, buf->dma_handle,
293				    &cx->scb->cpu_mdl[mdl_id + i].paddr);
294			cx18_writel(cx, s->buf_size,
295				    &cx->scb->cpu_mdl[mdl_id + i].length);
296		}
297
298		if (i == s->bufs_per_mdl) {
299			/*
300			 * The encoder doesn't honor s->mdl_size.  So in the
301			 * case of a non-integral number of buffers to meet
302			 * mdl_size, we lie about the size of the last buffer
303			 * in the MDL to get the encoder to really only send
304			 * us mdl_size bytes per MDL transfer.
305			 */
306			partial_buf_size = s->mdl_size % s->buf_size;
307			if (partial_buf_size) {
308				cx18_writel(cx, partial_buf_size,
309				      &cx->scb->cpu_mdl[mdl_id + i - 1].length);
310			}
311			cx18_enqueue(s, mdl, &s->q_free);
312		} else {
313			/* Not enough buffers for this MDL; we won't use it */
314			cx18_push(s, mdl, &s->q_idle);
315		}
316		mdl_id += i;
317	}
318}
319
320void _cx18_mdl_sync_for_device(struct cx18_stream *s, struct cx18_mdl *mdl)
321{
322	int dma = s->dma;
323	u32 buf_size = s->buf_size;
324	struct pci_dev *pci_dev = s->cx->pci_dev;
325	struct cx18_buffer *buf;
326
327	list_for_each_entry(buf, &mdl->buf_list, list)
328		dma_sync_single_for_device(&pci_dev->dev, buf->dma_handle,
329					   buf_size, dma);
330}
331
332int cx18_stream_alloc(struct cx18_stream *s)
333{
334	struct cx18 *cx = s->cx;
335	int i;
336
337	if (s->buffers == 0)
338		return 0;
339
340	CX18_DEBUG_INFO("Allocate %s stream: %d x %d buffers (%d.%02d kB total)\n",
341		s->name, s->buffers, s->buf_size,
342		s->buffers * s->buf_size / 1024,
343		(s->buffers * s->buf_size * 100 / 1024) % 100);
344
345	if (((char __iomem *)&cx->scb->cpu_mdl[cx->free_mdl_idx + s->buffers] -
346				(char __iomem *)cx->scb) > SCB_RESERVED_SIZE) {
347		unsigned bufsz = (((char __iomem *)cx->scb) + SCB_RESERVED_SIZE -
348					((char __iomem *)cx->scb->cpu_mdl));
349
350		CX18_ERR("Too many buffers, cannot fit in SCB area\n");
351		CX18_ERR("Max buffers = %zu\n",
352			bufsz / sizeof(struct cx18_mdl_ent));
353		return -ENOMEM;
354	}
355
356	s->mdl_base_idx = cx->free_mdl_idx;
357
358	/* allocate stream buffers and MDLs */
359	for (i = 0; i < s->buffers; i++) {
360		struct cx18_mdl *mdl;
361		struct cx18_buffer *buf;
362
363		/* 1 MDL per buffer to handle the worst & also default case */
364		mdl = kzalloc(sizeof(struct cx18_mdl), GFP_KERNEL|__GFP_NOWARN);
365		if (mdl == NULL)
366			break;
367
368		buf = kzalloc(sizeof(struct cx18_buffer),
369				GFP_KERNEL|__GFP_NOWARN);
370		if (buf == NULL) {
371			kfree(mdl);
372			break;
373		}
374
375		buf->buf = kmalloc(s->buf_size, GFP_KERNEL|__GFP_NOWARN);
376		if (buf->buf == NULL) {
377			kfree(mdl);
378			kfree(buf);
379			break;
380		}
381
382		INIT_LIST_HEAD(&mdl->list);
383		INIT_LIST_HEAD(&mdl->buf_list);
384		mdl->id = s->mdl_base_idx; /* a somewhat safe value */
385		cx18_enqueue(s, mdl, &s->q_idle);
386
387		INIT_LIST_HEAD(&buf->list);
388		buf->dma_handle = dma_map_single(&s->cx->pci_dev->dev,
389						 buf->buf, s->buf_size,
390						 s->dma);
391		cx18_buf_sync_for_cpu(s, buf);
392		list_add_tail(&buf->list, &s->buf_pool);
393	}
394	if (i == s->buffers) {
395		cx->free_mdl_idx += s->buffers;
396		return 0;
397	}
398	CX18_ERR("Couldn't allocate buffers for %s stream\n", s->name);
399	cx18_stream_free(s);
400	return -ENOMEM;
401}
402
403void cx18_stream_free(struct cx18_stream *s)
404{
405	struct cx18_mdl *mdl;
406	struct cx18_buffer *buf;
407	struct cx18 *cx = s->cx;
408
409	CX18_DEBUG_INFO("Deallocating buffers for %s stream\n", s->name);
410
411	/* move all buffers to buf_pool and all MDLs to q_idle */
412	cx18_unload_queues(s);
413
414	/* empty q_idle */
415	while ((mdl = cx18_dequeue(s, &s->q_idle)))
416		kfree(mdl);
417
418	/* empty buf_pool */
419	while (!list_empty(&s->buf_pool)) {
420		buf = list_first_entry(&s->buf_pool, struct cx18_buffer, list);
421		list_del_init(&buf->list);
422
423		dma_unmap_single(&s->cx->pci_dev->dev, buf->dma_handle,
424				 s->buf_size, s->dma);
425		kfree(buf->buf);
426		kfree(buf);
427	}
428}
429