1/* savage_bci.c -- BCI support for Savage
2 *
3 * Copyright 2004  Felix Kuehling
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sub license,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial portions
15 * of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
18 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
20 * NON-INFRINGEMENT. IN NO EVENT SHALL FELIX KUEHLING BE LIABLE FOR
21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
22 * CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
23 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
24 */
25
26#include <linux/delay.h>
27#include <linux/pci.h>
28#include <linux/slab.h>
29#include <linux/uaccess.h>
30
31#include <drm/drm_device.h>
32#include <drm/drm_file.h>
33#include <drm/drm_print.h>
34#include <drm/savage_drm.h>
35
36#include "savage_drv.h"
37
38/* Need a long timeout for shadow status updates can take a while
39 * and so can waiting for events when the queue is full. */
40#define SAVAGE_DEFAULT_USEC_TIMEOUT	1000000	/* 1s */
41#define SAVAGE_EVENT_USEC_TIMEOUT	5000000	/* 5s */
42#define SAVAGE_FREELIST_DEBUG		0
43
44static int savage_do_cleanup_bci(struct drm_device *dev);
45
46static int
47savage_bci_wait_fifo_shadow(drm_savage_private_t * dev_priv, unsigned int n)
48{
49	uint32_t mask = dev_priv->status_used_mask;
50	uint32_t threshold = dev_priv->bci_threshold_hi;
51	uint32_t status;
52	int i;
53
54#if SAVAGE_BCI_DEBUG
55	if (n > dev_priv->cob_size + SAVAGE_BCI_FIFO_SIZE - threshold)
56		DRM_ERROR("Trying to emit %d words "
57			  "(more than guaranteed space in COB)\n", n);
58#endif
59
60	for (i = 0; i < SAVAGE_DEFAULT_USEC_TIMEOUT; i++) {
61		mb();
62		status = dev_priv->status_ptr[0];
63		if ((status & mask) < threshold)
64			return 0;
65		udelay(1);
66	}
67
68#if SAVAGE_BCI_DEBUG
69	DRM_ERROR("failed!\n");
70	DRM_INFO("   status=0x%08x, threshold=0x%08x\n", status, threshold);
71#endif
72	return -EBUSY;
73}
74
75static int
76savage_bci_wait_fifo_s3d(drm_savage_private_t * dev_priv, unsigned int n)
77{
78	uint32_t maxUsed = dev_priv->cob_size + SAVAGE_BCI_FIFO_SIZE - n;
79	uint32_t status;
80	int i;
81
82	for (i = 0; i < SAVAGE_DEFAULT_USEC_TIMEOUT; i++) {
83		status = SAVAGE_READ(SAVAGE_STATUS_WORD0);
84		if ((status & SAVAGE_FIFO_USED_MASK_S3D) <= maxUsed)
85			return 0;
86		udelay(1);
87	}
88
89#if SAVAGE_BCI_DEBUG
90	DRM_ERROR("failed!\n");
91	DRM_INFO("   status=0x%08x\n", status);
92#endif
93	return -EBUSY;
94}
95
96static int
97savage_bci_wait_fifo_s4(drm_savage_private_t * dev_priv, unsigned int n)
98{
99	uint32_t maxUsed = dev_priv->cob_size + SAVAGE_BCI_FIFO_SIZE - n;
100	uint32_t status;
101	int i;
102
103	for (i = 0; i < SAVAGE_DEFAULT_USEC_TIMEOUT; i++) {
104		status = SAVAGE_READ(SAVAGE_ALT_STATUS_WORD0);
105		if ((status & SAVAGE_FIFO_USED_MASK_S4) <= maxUsed)
106			return 0;
107		udelay(1);
108	}
109
110#if SAVAGE_BCI_DEBUG
111	DRM_ERROR("failed!\n");
112	DRM_INFO("   status=0x%08x\n", status);
113#endif
114	return -EBUSY;
115}
116
117/*
118 * Waiting for events.
119 *
120 * The BIOSresets the event tag to 0 on mode changes. Therefore we
121 * never emit 0 to the event tag. If we find a 0 event tag we know the
122 * BIOS stomped on it and return success assuming that the BIOS waited
123 * for engine idle.
124 *
125 * Note: if the Xserver uses the event tag it has to follow the same
126 * rule. Otherwise there may be glitches every 2^16 events.
127 */
128static int
129savage_bci_wait_event_shadow(drm_savage_private_t * dev_priv, uint16_t e)
130{
131	uint32_t status;
132	int i;
133
134	for (i = 0; i < SAVAGE_EVENT_USEC_TIMEOUT; i++) {
135		mb();
136		status = dev_priv->status_ptr[1];
137		if ((((status & 0xffff) - e) & 0xffff) <= 0x7fff ||
138		    (status & 0xffff) == 0)
139			return 0;
140		udelay(1);
141	}
142
143#if SAVAGE_BCI_DEBUG
144	DRM_ERROR("failed!\n");
145	DRM_INFO("   status=0x%08x, e=0x%04x\n", status, e);
146#endif
147
148	return -EBUSY;
149}
150
151static int
152savage_bci_wait_event_reg(drm_savage_private_t * dev_priv, uint16_t e)
153{
154	uint32_t status;
155	int i;
156
157	for (i = 0; i < SAVAGE_EVENT_USEC_TIMEOUT; i++) {
158		status = SAVAGE_READ(SAVAGE_STATUS_WORD1);
159		if ((((status & 0xffff) - e) & 0xffff) <= 0x7fff ||
160		    (status & 0xffff) == 0)
161			return 0;
162		udelay(1);
163	}
164
165#if SAVAGE_BCI_DEBUG
166	DRM_ERROR("failed!\n");
167	DRM_INFO("   status=0x%08x, e=0x%04x\n", status, e);
168#endif
169
170	return -EBUSY;
171}
172
173uint16_t savage_bci_emit_event(drm_savage_private_t * dev_priv,
174			       unsigned int flags)
175{
176	uint16_t count;
177	BCI_LOCALS;
178
179	if (dev_priv->status_ptr) {
180		/* coordinate with Xserver */
181		count = dev_priv->status_ptr[1023];
182		if (count < dev_priv->event_counter)
183			dev_priv->event_wrap++;
184	} else {
185		count = dev_priv->event_counter;
186	}
187	count = (count + 1) & 0xffff;
188	if (count == 0) {
189		count++;	/* See the comment above savage_wait_event_*. */
190		dev_priv->event_wrap++;
191	}
192	dev_priv->event_counter = count;
193	if (dev_priv->status_ptr)
194		dev_priv->status_ptr[1023] = (uint32_t) count;
195
196	if ((flags & (SAVAGE_WAIT_2D | SAVAGE_WAIT_3D))) {
197		unsigned int wait_cmd = BCI_CMD_WAIT;
198		if ((flags & SAVAGE_WAIT_2D))
199			wait_cmd |= BCI_CMD_WAIT_2D;
200		if ((flags & SAVAGE_WAIT_3D))
201			wait_cmd |= BCI_CMD_WAIT_3D;
202		BEGIN_BCI(2);
203		BCI_WRITE(wait_cmd);
204	} else {
205		BEGIN_BCI(1);
206	}
207	BCI_WRITE(BCI_CMD_UPDATE_EVENT_TAG | (uint32_t) count);
208
209	return count;
210}
211
212/*
213 * Freelist management
214 */
215static int savage_freelist_init(struct drm_device * dev)
216{
217	drm_savage_private_t *dev_priv = dev->dev_private;
218	struct drm_device_dma *dma = dev->dma;
219	struct drm_buf *buf;
220	drm_savage_buf_priv_t *entry;
221	int i;
222	DRM_DEBUG("count=%d\n", dma->buf_count);
223
224	dev_priv->head.next = &dev_priv->tail;
225	dev_priv->head.prev = NULL;
226	dev_priv->head.buf = NULL;
227
228	dev_priv->tail.next = NULL;
229	dev_priv->tail.prev = &dev_priv->head;
230	dev_priv->tail.buf = NULL;
231
232	for (i = 0; i < dma->buf_count; i++) {
233		buf = dma->buflist[i];
234		entry = buf->dev_private;
235
236		SET_AGE(&entry->age, 0, 0);
237		entry->buf = buf;
238
239		entry->next = dev_priv->head.next;
240		entry->prev = &dev_priv->head;
241		dev_priv->head.next->prev = entry;
242		dev_priv->head.next = entry;
243	}
244
245	return 0;
246}
247
248static struct drm_buf *savage_freelist_get(struct drm_device * dev)
249{
250	drm_savage_private_t *dev_priv = dev->dev_private;
251	drm_savage_buf_priv_t *tail = dev_priv->tail.prev;
252	uint16_t event;
253	unsigned int wrap;
254	DRM_DEBUG("\n");
255
256	UPDATE_EVENT_COUNTER();
257	if (dev_priv->status_ptr)
258		event = dev_priv->status_ptr[1] & 0xffff;
259	else
260		event = SAVAGE_READ(SAVAGE_STATUS_WORD1) & 0xffff;
261	wrap = dev_priv->event_wrap;
262	if (event > dev_priv->event_counter)
263		wrap--;		/* hardware hasn't passed the last wrap yet */
264
265	DRM_DEBUG("   tail=0x%04x %d\n", tail->age.event, tail->age.wrap);
266	DRM_DEBUG("   head=0x%04x %d\n", event, wrap);
267
268	if (tail->buf && (TEST_AGE(&tail->age, event, wrap) || event == 0)) {
269		drm_savage_buf_priv_t *next = tail->next;
270		drm_savage_buf_priv_t *prev = tail->prev;
271		prev->next = next;
272		next->prev = prev;
273		tail->next = tail->prev = NULL;
274		return tail->buf;
275	}
276
277	DRM_DEBUG("returning NULL, tail->buf=%p!\n", tail->buf);
278	return NULL;
279}
280
281void savage_freelist_put(struct drm_device * dev, struct drm_buf * buf)
282{
283	drm_savage_private_t *dev_priv = dev->dev_private;
284	drm_savage_buf_priv_t *entry = buf->dev_private, *prev, *next;
285
286	DRM_DEBUG("age=0x%04x wrap=%d\n", entry->age.event, entry->age.wrap);
287
288	if (entry->next != NULL || entry->prev != NULL) {
289		DRM_ERROR("entry already on freelist.\n");
290		return;
291	}
292
293	prev = &dev_priv->head;
294	next = prev->next;
295	prev->next = entry;
296	next->prev = entry;
297	entry->prev = prev;
298	entry->next = next;
299}
300
301/*
302 * Command DMA
303 */
304static int savage_dma_init(drm_savage_private_t * dev_priv)
305{
306	unsigned int i;
307
308	dev_priv->nr_dma_pages = dev_priv->cmd_dma->size /
309	    (SAVAGE_DMA_PAGE_SIZE * 4);
310	dev_priv->dma_pages = kmalloc_array(dev_priv->nr_dma_pages,
311					    sizeof(drm_savage_dma_page_t),
312					    GFP_KERNEL);
313	if (dev_priv->dma_pages == NULL)
314		return -ENOMEM;
315
316	for (i = 0; i < dev_priv->nr_dma_pages; ++i) {
317		SET_AGE(&dev_priv->dma_pages[i].age, 0, 0);
318		dev_priv->dma_pages[i].used = 0;
319		dev_priv->dma_pages[i].flushed = 0;
320	}
321	SET_AGE(&dev_priv->last_dma_age, 0, 0);
322
323	dev_priv->first_dma_page = 0;
324	dev_priv->current_dma_page = 0;
325
326	return 0;
327}
328
329void savage_dma_reset(drm_savage_private_t * dev_priv)
330{
331	uint16_t event;
332	unsigned int wrap, i;
333	event = savage_bci_emit_event(dev_priv, 0);
334	wrap = dev_priv->event_wrap;
335	for (i = 0; i < dev_priv->nr_dma_pages; ++i) {
336		SET_AGE(&dev_priv->dma_pages[i].age, event, wrap);
337		dev_priv->dma_pages[i].used = 0;
338		dev_priv->dma_pages[i].flushed = 0;
339	}
340	SET_AGE(&dev_priv->last_dma_age, event, wrap);
341	dev_priv->first_dma_page = dev_priv->current_dma_page = 0;
342}
343
344void savage_dma_wait(drm_savage_private_t * dev_priv, unsigned int page)
345{
346	uint16_t event;
347	unsigned int wrap;
348
349	/* Faked DMA buffer pages don't age. */
350	if (dev_priv->cmd_dma == &dev_priv->fake_dma)
351		return;
352
353	UPDATE_EVENT_COUNTER();
354	if (dev_priv->status_ptr)
355		event = dev_priv->status_ptr[1] & 0xffff;
356	else
357		event = SAVAGE_READ(SAVAGE_STATUS_WORD1) & 0xffff;
358	wrap = dev_priv->event_wrap;
359	if (event > dev_priv->event_counter)
360		wrap--;		/* hardware hasn't passed the last wrap yet */
361
362	if (dev_priv->dma_pages[page].age.wrap > wrap ||
363	    (dev_priv->dma_pages[page].age.wrap == wrap &&
364	     dev_priv->dma_pages[page].age.event > event)) {
365		if (dev_priv->wait_evnt(dev_priv,
366					dev_priv->dma_pages[page].age.event)
367		    < 0)
368			DRM_ERROR("wait_evnt failed!\n");
369	}
370}
371
372uint32_t *savage_dma_alloc(drm_savage_private_t * dev_priv, unsigned int n)
373{
374	unsigned int cur = dev_priv->current_dma_page;
375	unsigned int rest = SAVAGE_DMA_PAGE_SIZE -
376	    dev_priv->dma_pages[cur].used;
377	unsigned int nr_pages = (n - rest + SAVAGE_DMA_PAGE_SIZE - 1) /
378	    SAVAGE_DMA_PAGE_SIZE;
379	uint32_t *dma_ptr;
380	unsigned int i;
381
382	DRM_DEBUG("cur=%u, cur->used=%u, n=%u, rest=%u, nr_pages=%u\n",
383		  cur, dev_priv->dma_pages[cur].used, n, rest, nr_pages);
384
385	if (cur + nr_pages < dev_priv->nr_dma_pages) {
386		dma_ptr = (uint32_t *) dev_priv->cmd_dma->handle +
387		    cur * SAVAGE_DMA_PAGE_SIZE + dev_priv->dma_pages[cur].used;
388		if (n < rest)
389			rest = n;
390		dev_priv->dma_pages[cur].used += rest;
391		n -= rest;
392		cur++;
393	} else {
394		dev_priv->dma_flush(dev_priv);
395		nr_pages =
396		    (n + SAVAGE_DMA_PAGE_SIZE - 1) / SAVAGE_DMA_PAGE_SIZE;
397		for (i = cur; i < dev_priv->nr_dma_pages; ++i) {
398			dev_priv->dma_pages[i].age = dev_priv->last_dma_age;
399			dev_priv->dma_pages[i].used = 0;
400			dev_priv->dma_pages[i].flushed = 0;
401		}
402		dma_ptr = (uint32_t *) dev_priv->cmd_dma->handle;
403		dev_priv->first_dma_page = cur = 0;
404	}
405	for (i = cur; nr_pages > 0; ++i, --nr_pages) {
406#if SAVAGE_DMA_DEBUG
407		if (dev_priv->dma_pages[i].used) {
408			DRM_ERROR("unflushed page %u: used=%u\n",
409				  i, dev_priv->dma_pages[i].used);
410		}
411#endif
412		if (n > SAVAGE_DMA_PAGE_SIZE)
413			dev_priv->dma_pages[i].used = SAVAGE_DMA_PAGE_SIZE;
414		else
415			dev_priv->dma_pages[i].used = n;
416		n -= SAVAGE_DMA_PAGE_SIZE;
417	}
418	dev_priv->current_dma_page = --i;
419
420	DRM_DEBUG("cur=%u, cur->used=%u, n=%u\n",
421		  i, dev_priv->dma_pages[i].used, n);
422
423	savage_dma_wait(dev_priv, dev_priv->current_dma_page);
424
425	return dma_ptr;
426}
427
428static void savage_dma_flush(drm_savage_private_t * dev_priv)
429{
430	unsigned int first = dev_priv->first_dma_page;
431	unsigned int cur = dev_priv->current_dma_page;
432	uint16_t event;
433	unsigned int wrap, pad, align, len, i;
434	unsigned long phys_addr;
435	BCI_LOCALS;
436
437	if (first == cur &&
438	    dev_priv->dma_pages[cur].used == dev_priv->dma_pages[cur].flushed)
439		return;
440
441	/* pad length to multiples of 2 entries
442	 * align start of next DMA block to multiles of 8 entries */
443	pad = -dev_priv->dma_pages[cur].used & 1;
444	align = -(dev_priv->dma_pages[cur].used + pad) & 7;
445
446	DRM_DEBUG("first=%u, cur=%u, first->flushed=%u, cur->used=%u, "
447		  "pad=%u, align=%u\n",
448		  first, cur, dev_priv->dma_pages[first].flushed,
449		  dev_priv->dma_pages[cur].used, pad, align);
450
451	/* pad with noops */
452	if (pad) {
453		uint32_t *dma_ptr = (uint32_t *) dev_priv->cmd_dma->handle +
454		    cur * SAVAGE_DMA_PAGE_SIZE + dev_priv->dma_pages[cur].used;
455		dev_priv->dma_pages[cur].used += pad;
456		while (pad != 0) {
457			*dma_ptr++ = BCI_CMD_WAIT;
458			pad--;
459		}
460	}
461
462	mb();
463
464	/* do flush ... */
465	phys_addr = dev_priv->cmd_dma->offset +
466	    (first * SAVAGE_DMA_PAGE_SIZE +
467	     dev_priv->dma_pages[first].flushed) * 4;
468	len = (cur - first) * SAVAGE_DMA_PAGE_SIZE +
469	    dev_priv->dma_pages[cur].used - dev_priv->dma_pages[first].flushed;
470
471	DRM_DEBUG("phys_addr=%lx, len=%u\n",
472		  phys_addr | dev_priv->dma_type, len);
473
474	BEGIN_BCI(3);
475	BCI_SET_REGISTERS(SAVAGE_DMABUFADDR, 1);
476	BCI_WRITE(phys_addr | dev_priv->dma_type);
477	BCI_DMA(len);
478
479	/* fix alignment of the start of the next block */
480	dev_priv->dma_pages[cur].used += align;
481
482	/* age DMA pages */
483	event = savage_bci_emit_event(dev_priv, 0);
484	wrap = dev_priv->event_wrap;
485	for (i = first; i < cur; ++i) {
486		SET_AGE(&dev_priv->dma_pages[i].age, event, wrap);
487		dev_priv->dma_pages[i].used = 0;
488		dev_priv->dma_pages[i].flushed = 0;
489	}
490	/* age the current page only when it's full */
491	if (dev_priv->dma_pages[cur].used == SAVAGE_DMA_PAGE_SIZE) {
492		SET_AGE(&dev_priv->dma_pages[cur].age, event, wrap);
493		dev_priv->dma_pages[cur].used = 0;
494		dev_priv->dma_pages[cur].flushed = 0;
495		/* advance to next page */
496		cur++;
497		if (cur == dev_priv->nr_dma_pages)
498			cur = 0;
499		dev_priv->first_dma_page = dev_priv->current_dma_page = cur;
500	} else {
501		dev_priv->first_dma_page = cur;
502		dev_priv->dma_pages[cur].flushed = dev_priv->dma_pages[i].used;
503	}
504	SET_AGE(&dev_priv->last_dma_age, event, wrap);
505
506	DRM_DEBUG("first=cur=%u, cur->used=%u, cur->flushed=%u\n", cur,
507		  dev_priv->dma_pages[cur].used,
508		  dev_priv->dma_pages[cur].flushed);
509}
510
511static void savage_fake_dma_flush(drm_savage_private_t * dev_priv)
512{
513	unsigned int i, j;
514	BCI_LOCALS;
515
516	if (dev_priv->first_dma_page == dev_priv->current_dma_page &&
517	    dev_priv->dma_pages[dev_priv->current_dma_page].used == 0)
518		return;
519
520	DRM_DEBUG("first=%u, cur=%u, cur->used=%u\n",
521		  dev_priv->first_dma_page, dev_priv->current_dma_page,
522		  dev_priv->dma_pages[dev_priv->current_dma_page].used);
523
524	for (i = dev_priv->first_dma_page;
525	     i <= dev_priv->current_dma_page && dev_priv->dma_pages[i].used;
526	     ++i) {
527		uint32_t *dma_ptr = (uint32_t *) dev_priv->cmd_dma->handle +
528		    i * SAVAGE_DMA_PAGE_SIZE;
529#if SAVAGE_DMA_DEBUG
530		/* Sanity check: all pages except the last one must be full. */
531		if (i < dev_priv->current_dma_page &&
532		    dev_priv->dma_pages[i].used != SAVAGE_DMA_PAGE_SIZE) {
533			DRM_ERROR("partial DMA page %u: used=%u",
534				  i, dev_priv->dma_pages[i].used);
535		}
536#endif
537		BEGIN_BCI(dev_priv->dma_pages[i].used);
538		for (j = 0; j < dev_priv->dma_pages[i].used; ++j) {
539			BCI_WRITE(dma_ptr[j]);
540		}
541		dev_priv->dma_pages[i].used = 0;
542	}
543
544	/* reset to first page */
545	dev_priv->first_dma_page = dev_priv->current_dma_page = 0;
546}
547
548int savage_driver_load(struct drm_device *dev, unsigned long chipset)
549{
550	drm_savage_private_t *dev_priv;
551
552	dev_priv = kzalloc(sizeof(drm_savage_private_t), GFP_KERNEL);
553	if (dev_priv == NULL)
554		return -ENOMEM;
555
556	dev->dev_private = (void *)dev_priv;
557
558	dev_priv->chipset = (enum savage_family)chipset;
559
560	pci_set_master(dev->pdev);
561
562	return 0;
563}
564
565
566/*
567 * Initialize mappings. On Savage4 and SavageIX the alignment
568 * and size of the aperture is not suitable for automatic MTRR setup
569 * in drm_legacy_addmap. Therefore we add them manually before the maps are
570 * initialized, and tear them down on last close.
571 */
572int savage_driver_firstopen(struct drm_device *dev)
573{
574	drm_savage_private_t *dev_priv = dev->dev_private;
575	unsigned long mmio_base, fb_base, fb_size, aperture_base;
576	/* fb_rsrc and aper_rsrc aren't really used currently, but still exist
577	 * in case we decide we need information on the BAR for BSD in the
578	 * future.
579	 */
580	unsigned int fb_rsrc, aper_rsrc;
581	int ret = 0;
582
583	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
584		fb_rsrc = 0;
585		fb_base = pci_resource_start(dev->pdev, 0);
586		fb_size = SAVAGE_FB_SIZE_S3;
587		mmio_base = fb_base + SAVAGE_FB_SIZE_S3;
588		aper_rsrc = 0;
589		aperture_base = fb_base + SAVAGE_APERTURE_OFFSET;
590		/* this should always be true */
591		if (pci_resource_len(dev->pdev, 0) == 0x08000000) {
592			/* Don't make MMIO write-cobining! We need 3
593			 * MTRRs. */
594			dev_priv->mtrr_handles[0] =
595				arch_phys_wc_add(fb_base, 0x01000000);
596			dev_priv->mtrr_handles[1] =
597				arch_phys_wc_add(fb_base + 0x02000000,
598						 0x02000000);
599			dev_priv->mtrr_handles[2] =
600				arch_phys_wc_add(fb_base + 0x04000000,
601						0x04000000);
602		} else {
603			DRM_ERROR("strange pci_resource_len %08llx\n",
604				  (unsigned long long)
605				  pci_resource_len(dev->pdev, 0));
606		}
607	} else if (dev_priv->chipset != S3_SUPERSAVAGE &&
608		   dev_priv->chipset != S3_SAVAGE2000) {
609		mmio_base = pci_resource_start(dev->pdev, 0);
610		fb_rsrc = 1;
611		fb_base = pci_resource_start(dev->pdev, 1);
612		fb_size = SAVAGE_FB_SIZE_S4;
613		aper_rsrc = 1;
614		aperture_base = fb_base + SAVAGE_APERTURE_OFFSET;
615		/* this should always be true */
616		if (pci_resource_len(dev->pdev, 1) == 0x08000000) {
617			/* Can use one MTRR to cover both fb and
618			 * aperture. */
619			dev_priv->mtrr_handles[0] =
620				arch_phys_wc_add(fb_base,
621						 0x08000000);
622		} else {
623			DRM_ERROR("strange pci_resource_len %08llx\n",
624				  (unsigned long long)
625				  pci_resource_len(dev->pdev, 1));
626		}
627	} else {
628		mmio_base = pci_resource_start(dev->pdev, 0);
629		fb_rsrc = 1;
630		fb_base = pci_resource_start(dev->pdev, 1);
631		fb_size = pci_resource_len(dev->pdev, 1);
632		aper_rsrc = 2;
633		aperture_base = pci_resource_start(dev->pdev, 2);
634		/* Automatic MTRR setup will do the right thing. */
635	}
636
637	ret = drm_legacy_addmap(dev, mmio_base, SAVAGE_MMIO_SIZE,
638				_DRM_REGISTERS, _DRM_READ_ONLY,
639				&dev_priv->mmio);
640	if (ret)
641		return ret;
642
643	ret = drm_legacy_addmap(dev, fb_base, fb_size, _DRM_FRAME_BUFFER,
644				_DRM_WRITE_COMBINING, &dev_priv->fb);
645	if (ret)
646		return ret;
647
648	ret = drm_legacy_addmap(dev, aperture_base, SAVAGE_APERTURE_SIZE,
649				_DRM_FRAME_BUFFER, _DRM_WRITE_COMBINING,
650				&dev_priv->aperture);
651	return ret;
652}
653
654/*
655 * Delete MTRRs and free device-private data.
656 */
657void savage_driver_lastclose(struct drm_device *dev)
658{
659	drm_savage_private_t *dev_priv = dev->dev_private;
660	int i;
661
662	for (i = 0; i < 3; ++i) {
663		arch_phys_wc_del(dev_priv->mtrr_handles[i]);
664		dev_priv->mtrr_handles[i] = 0;
665	}
666}
667
668void savage_driver_unload(struct drm_device *dev)
669{
670	drm_savage_private_t *dev_priv = dev->dev_private;
671
672	kfree(dev_priv);
673}
674
675static int savage_do_init_bci(struct drm_device * dev, drm_savage_init_t * init)
676{
677	drm_savage_private_t *dev_priv = dev->dev_private;
678
679	if (init->fb_bpp != 16 && init->fb_bpp != 32) {
680		DRM_ERROR("invalid frame buffer bpp %d!\n", init->fb_bpp);
681		return -EINVAL;
682	}
683	if (init->depth_bpp != 16 && init->depth_bpp != 32) {
684		DRM_ERROR("invalid depth buffer bpp %d!\n", init->fb_bpp);
685		return -EINVAL;
686	}
687	if (init->dma_type != SAVAGE_DMA_AGP &&
688	    init->dma_type != SAVAGE_DMA_PCI) {
689		DRM_ERROR("invalid dma memory type %d!\n", init->dma_type);
690		return -EINVAL;
691	}
692
693	dev_priv->cob_size = init->cob_size;
694	dev_priv->bci_threshold_lo = init->bci_threshold_lo;
695	dev_priv->bci_threshold_hi = init->bci_threshold_hi;
696	dev_priv->dma_type = init->dma_type;
697
698	dev_priv->fb_bpp = init->fb_bpp;
699	dev_priv->front_offset = init->front_offset;
700	dev_priv->front_pitch = init->front_pitch;
701	dev_priv->back_offset = init->back_offset;
702	dev_priv->back_pitch = init->back_pitch;
703	dev_priv->depth_bpp = init->depth_bpp;
704	dev_priv->depth_offset = init->depth_offset;
705	dev_priv->depth_pitch = init->depth_pitch;
706
707	dev_priv->texture_offset = init->texture_offset;
708	dev_priv->texture_size = init->texture_size;
709
710	dev_priv->sarea = drm_legacy_getsarea(dev);
711	if (!dev_priv->sarea) {
712		DRM_ERROR("could not find sarea!\n");
713		savage_do_cleanup_bci(dev);
714		return -EINVAL;
715	}
716	if (init->status_offset != 0) {
717		dev_priv->status = drm_legacy_findmap(dev, init->status_offset);
718		if (!dev_priv->status) {
719			DRM_ERROR("could not find shadow status region!\n");
720			savage_do_cleanup_bci(dev);
721			return -EINVAL;
722		}
723	} else {
724		dev_priv->status = NULL;
725	}
726	if (dev_priv->dma_type == SAVAGE_DMA_AGP && init->buffers_offset) {
727		dev->agp_buffer_token = init->buffers_offset;
728		dev->agp_buffer_map = drm_legacy_findmap(dev,
729						       init->buffers_offset);
730		if (!dev->agp_buffer_map) {
731			DRM_ERROR("could not find DMA buffer region!\n");
732			savage_do_cleanup_bci(dev);
733			return -EINVAL;
734		}
735		drm_legacy_ioremap(dev->agp_buffer_map, dev);
736		if (!dev->agp_buffer_map->handle) {
737			DRM_ERROR("failed to ioremap DMA buffer region!\n");
738			savage_do_cleanup_bci(dev);
739			return -ENOMEM;
740		}
741	}
742	if (init->agp_textures_offset) {
743		dev_priv->agp_textures =
744		    drm_legacy_findmap(dev, init->agp_textures_offset);
745		if (!dev_priv->agp_textures) {
746			DRM_ERROR("could not find agp texture region!\n");
747			savage_do_cleanup_bci(dev);
748			return -EINVAL;
749		}
750	} else {
751		dev_priv->agp_textures = NULL;
752	}
753
754	if (init->cmd_dma_offset) {
755		if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
756			DRM_ERROR("command DMA not supported on "
757				  "Savage3D/MX/IX.\n");
758			savage_do_cleanup_bci(dev);
759			return -EINVAL;
760		}
761		if (dev->dma && dev->dma->buflist) {
762			DRM_ERROR("command and vertex DMA not supported "
763				  "at the same time.\n");
764			savage_do_cleanup_bci(dev);
765			return -EINVAL;
766		}
767		dev_priv->cmd_dma = drm_legacy_findmap(dev, init->cmd_dma_offset);
768		if (!dev_priv->cmd_dma) {
769			DRM_ERROR("could not find command DMA region!\n");
770			savage_do_cleanup_bci(dev);
771			return -EINVAL;
772		}
773		if (dev_priv->dma_type == SAVAGE_DMA_AGP) {
774			if (dev_priv->cmd_dma->type != _DRM_AGP) {
775				DRM_ERROR("AGP command DMA region is not a "
776					  "_DRM_AGP map!\n");
777				savage_do_cleanup_bci(dev);
778				return -EINVAL;
779			}
780			drm_legacy_ioremap(dev_priv->cmd_dma, dev);
781			if (!dev_priv->cmd_dma->handle) {
782				DRM_ERROR("failed to ioremap command "
783					  "DMA region!\n");
784				savage_do_cleanup_bci(dev);
785				return -ENOMEM;
786			}
787		} else if (dev_priv->cmd_dma->type != _DRM_CONSISTENT) {
788			DRM_ERROR("PCI command DMA region is not a "
789				  "_DRM_CONSISTENT map!\n");
790			savage_do_cleanup_bci(dev);
791			return -EINVAL;
792		}
793	} else {
794		dev_priv->cmd_dma = NULL;
795	}
796
797	dev_priv->dma_flush = savage_dma_flush;
798	if (!dev_priv->cmd_dma) {
799		DRM_DEBUG("falling back to faked command DMA.\n");
800		dev_priv->fake_dma.offset = 0;
801		dev_priv->fake_dma.size = SAVAGE_FAKE_DMA_SIZE;
802		dev_priv->fake_dma.type = _DRM_SHM;
803		dev_priv->fake_dma.handle = kmalloc(SAVAGE_FAKE_DMA_SIZE,
804						    GFP_KERNEL);
805		if (!dev_priv->fake_dma.handle) {
806			DRM_ERROR("could not allocate faked DMA buffer!\n");
807			savage_do_cleanup_bci(dev);
808			return -ENOMEM;
809		}
810		dev_priv->cmd_dma = &dev_priv->fake_dma;
811		dev_priv->dma_flush = savage_fake_dma_flush;
812	}
813
814	dev_priv->sarea_priv =
815	    (drm_savage_sarea_t *) ((uint8_t *) dev_priv->sarea->handle +
816				    init->sarea_priv_offset);
817
818	/* setup bitmap descriptors */
819	{
820		unsigned int color_tile_format;
821		unsigned int depth_tile_format;
822		unsigned int front_stride, back_stride, depth_stride;
823		if (dev_priv->chipset <= S3_SAVAGE4) {
824			color_tile_format = dev_priv->fb_bpp == 16 ?
825			    SAVAGE_BD_TILE_16BPP : SAVAGE_BD_TILE_32BPP;
826			depth_tile_format = dev_priv->depth_bpp == 16 ?
827			    SAVAGE_BD_TILE_16BPP : SAVAGE_BD_TILE_32BPP;
828		} else {
829			color_tile_format = SAVAGE_BD_TILE_DEST;
830			depth_tile_format = SAVAGE_BD_TILE_DEST;
831		}
832		front_stride = dev_priv->front_pitch / (dev_priv->fb_bpp / 8);
833		back_stride = dev_priv->back_pitch / (dev_priv->fb_bpp / 8);
834		depth_stride =
835		    dev_priv->depth_pitch / (dev_priv->depth_bpp / 8);
836
837		dev_priv->front_bd = front_stride | SAVAGE_BD_BW_DISABLE |
838		    (dev_priv->fb_bpp << SAVAGE_BD_BPP_SHIFT) |
839		    (color_tile_format << SAVAGE_BD_TILE_SHIFT);
840
841		dev_priv->back_bd = back_stride | SAVAGE_BD_BW_DISABLE |
842		    (dev_priv->fb_bpp << SAVAGE_BD_BPP_SHIFT) |
843		    (color_tile_format << SAVAGE_BD_TILE_SHIFT);
844
845		dev_priv->depth_bd = depth_stride | SAVAGE_BD_BW_DISABLE |
846		    (dev_priv->depth_bpp << SAVAGE_BD_BPP_SHIFT) |
847		    (depth_tile_format << SAVAGE_BD_TILE_SHIFT);
848	}
849
850	/* setup status and bci ptr */
851	dev_priv->event_counter = 0;
852	dev_priv->event_wrap = 0;
853	dev_priv->bci_ptr = (volatile uint32_t *)
854	    ((uint8_t *) dev_priv->mmio->handle + SAVAGE_BCI_OFFSET);
855	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
856		dev_priv->status_used_mask = SAVAGE_FIFO_USED_MASK_S3D;
857	} else {
858		dev_priv->status_used_mask = SAVAGE_FIFO_USED_MASK_S4;
859	}
860	if (dev_priv->status != NULL) {
861		dev_priv->status_ptr =
862		    (volatile uint32_t *)dev_priv->status->handle;
863		dev_priv->wait_fifo = savage_bci_wait_fifo_shadow;
864		dev_priv->wait_evnt = savage_bci_wait_event_shadow;
865		dev_priv->status_ptr[1023] = dev_priv->event_counter;
866	} else {
867		dev_priv->status_ptr = NULL;
868		if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
869			dev_priv->wait_fifo = savage_bci_wait_fifo_s3d;
870		} else {
871			dev_priv->wait_fifo = savage_bci_wait_fifo_s4;
872		}
873		dev_priv->wait_evnt = savage_bci_wait_event_reg;
874	}
875
876	/* cliprect functions */
877	if (S3_SAVAGE3D_SERIES(dev_priv->chipset))
878		dev_priv->emit_clip_rect = savage_emit_clip_rect_s3d;
879	else
880		dev_priv->emit_clip_rect = savage_emit_clip_rect_s4;
881
882	if (savage_freelist_init(dev) < 0) {
883		DRM_ERROR("could not initialize freelist\n");
884		savage_do_cleanup_bci(dev);
885		return -ENOMEM;
886	}
887
888	if (savage_dma_init(dev_priv) < 0) {
889		DRM_ERROR("could not initialize command DMA\n");
890		savage_do_cleanup_bci(dev);
891		return -ENOMEM;
892	}
893
894	return 0;
895}
896
897static int savage_do_cleanup_bci(struct drm_device * dev)
898{
899	drm_savage_private_t *dev_priv = dev->dev_private;
900
901	if (dev_priv->cmd_dma == &dev_priv->fake_dma) {
902		kfree(dev_priv->fake_dma.handle);
903	} else if (dev_priv->cmd_dma && dev_priv->cmd_dma->handle &&
904		   dev_priv->cmd_dma->type == _DRM_AGP &&
905		   dev_priv->dma_type == SAVAGE_DMA_AGP)
906		drm_legacy_ioremapfree(dev_priv->cmd_dma, dev);
907
908	if (dev_priv->dma_type == SAVAGE_DMA_AGP &&
909	    dev->agp_buffer_map && dev->agp_buffer_map->handle) {
910		drm_legacy_ioremapfree(dev->agp_buffer_map, dev);
911		/* make sure the next instance (which may be running
912		 * in PCI mode) doesn't try to use an old
913		 * agp_buffer_map. */
914		dev->agp_buffer_map = NULL;
915	}
916
917	kfree(dev_priv->dma_pages);
918
919	return 0;
920}
921
922static int savage_bci_init(struct drm_device *dev, void *data, struct drm_file *file_priv)
923{
924	drm_savage_init_t *init = data;
925
926	LOCK_TEST_WITH_RETURN(dev, file_priv);
927
928	switch (init->func) {
929	case SAVAGE_INIT_BCI:
930		return savage_do_init_bci(dev, init);
931	case SAVAGE_CLEANUP_BCI:
932		return savage_do_cleanup_bci(dev);
933	}
934
935	return -EINVAL;
936}
937
938static int savage_bci_event_emit(struct drm_device *dev, void *data, struct drm_file *file_priv)
939{
940	drm_savage_private_t *dev_priv = dev->dev_private;
941	drm_savage_event_emit_t *event = data;
942
943	DRM_DEBUG("\n");
944
945	LOCK_TEST_WITH_RETURN(dev, file_priv);
946
947	event->count = savage_bci_emit_event(dev_priv, event->flags);
948	event->count |= dev_priv->event_wrap << 16;
949
950	return 0;
951}
952
953static int savage_bci_event_wait(struct drm_device *dev, void *data, struct drm_file *file_priv)
954{
955	drm_savage_private_t *dev_priv = dev->dev_private;
956	drm_savage_event_wait_t *event = data;
957	unsigned int event_e, hw_e;
958	unsigned int event_w, hw_w;
959
960	DRM_DEBUG("\n");
961
962	UPDATE_EVENT_COUNTER();
963	if (dev_priv->status_ptr)
964		hw_e = dev_priv->status_ptr[1] & 0xffff;
965	else
966		hw_e = SAVAGE_READ(SAVAGE_STATUS_WORD1) & 0xffff;
967	hw_w = dev_priv->event_wrap;
968	if (hw_e > dev_priv->event_counter)
969		hw_w--;		/* hardware hasn't passed the last wrap yet */
970
971	event_e = event->count & 0xffff;
972	event_w = event->count >> 16;
973
974	/* Don't need to wait if
975	 * - event counter wrapped since the event was emitted or
976	 * - the hardware has advanced up to or over the event to wait for.
977	 */
978	if (event_w < hw_w || (event_w == hw_w && event_e <= hw_e))
979		return 0;
980	else
981		return dev_priv->wait_evnt(dev_priv, event_e);
982}
983
984/*
985 * DMA buffer management
986 */
987
988static int savage_bci_get_buffers(struct drm_device *dev,
989				  struct drm_file *file_priv,
990				  struct drm_dma *d)
991{
992	struct drm_buf *buf;
993	int i;
994
995	for (i = d->granted_count; i < d->request_count; i++) {
996		buf = savage_freelist_get(dev);
997		if (!buf)
998			return -EAGAIN;
999
1000		buf->file_priv = file_priv;
1001
1002		if (copy_to_user(&d->request_indices[i],
1003				     &buf->idx, sizeof(buf->idx)))
1004			return -EFAULT;
1005		if (copy_to_user(&d->request_sizes[i],
1006				     &buf->total, sizeof(buf->total)))
1007			return -EFAULT;
1008
1009		d->granted_count++;
1010	}
1011	return 0;
1012}
1013
1014int savage_bci_buffers(struct drm_device *dev, void *data, struct drm_file *file_priv)
1015{
1016	struct drm_device_dma *dma = dev->dma;
1017	struct drm_dma *d = data;
1018	int ret = 0;
1019
1020	LOCK_TEST_WITH_RETURN(dev, file_priv);
1021
1022	/* Please don't send us buffers.
1023	 */
1024	if (d->send_count != 0) {
1025		DRM_ERROR("Process %d trying to send %d buffers via drmDMA\n",
1026			  task_pid_nr(current), d->send_count);
1027		return -EINVAL;
1028	}
1029
1030	/* We'll send you buffers.
1031	 */
1032	if (d->request_count < 0 || d->request_count > dma->buf_count) {
1033		DRM_ERROR("Process %d trying to get %d buffers (of %d max)\n",
1034			  task_pid_nr(current), d->request_count, dma->buf_count);
1035		return -EINVAL;
1036	}
1037
1038	d->granted_count = 0;
1039
1040	if (d->request_count) {
1041		ret = savage_bci_get_buffers(dev, file_priv, d);
1042	}
1043
1044	return ret;
1045}
1046
1047void savage_reclaim_buffers(struct drm_device *dev, struct drm_file *file_priv)
1048{
1049	struct drm_device_dma *dma = dev->dma;
1050	drm_savage_private_t *dev_priv = dev->dev_private;
1051	int release_idlelock = 0;
1052	int i;
1053
1054	if (!dma)
1055		return;
1056	if (!dev_priv)
1057		return;
1058	if (!dma->buflist)
1059		return;
1060
1061	if (file_priv->master && file_priv->master->lock.hw_lock) {
1062		drm_legacy_idlelock_take(&file_priv->master->lock);
1063		release_idlelock = 1;
1064	}
1065
1066	for (i = 0; i < dma->buf_count; i++) {
1067		struct drm_buf *buf = dma->buflist[i];
1068		drm_savage_buf_priv_t *buf_priv = buf->dev_private;
1069
1070		if (buf->file_priv == file_priv && buf_priv &&
1071		    buf_priv->next == NULL && buf_priv->prev == NULL) {
1072			uint16_t event;
1073			DRM_DEBUG("reclaimed from client\n");
1074			event = savage_bci_emit_event(dev_priv, SAVAGE_WAIT_3D);
1075			SET_AGE(&buf_priv->age, event, dev_priv->event_wrap);
1076			savage_freelist_put(dev, buf);
1077		}
1078	}
1079
1080	if (release_idlelock)
1081		drm_legacy_idlelock_release(&file_priv->master->lock);
1082}
1083
1084const struct drm_ioctl_desc savage_ioctls[] = {
1085	DRM_IOCTL_DEF_DRV(SAVAGE_BCI_INIT, savage_bci_init, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
1086	DRM_IOCTL_DEF_DRV(SAVAGE_BCI_CMDBUF, savage_bci_cmdbuf, DRM_AUTH),
1087	DRM_IOCTL_DEF_DRV(SAVAGE_BCI_EVENT_EMIT, savage_bci_event_emit, DRM_AUTH),
1088	DRM_IOCTL_DEF_DRV(SAVAGE_BCI_EVENT_WAIT, savage_bci_event_wait, DRM_AUTH),
1089};
1090
1091int savage_max_ioctl = ARRAY_SIZE(savage_ioctls);
1092