1 /* via_dma.c -- DMA support for the VIA Unichrome/Pro
2  *
3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
4  * All Rights Reserved.
5  *
6  * Copyright 2004 Digeo, Inc., Palo Alto, CA, U.S.A.
7  * All Rights Reserved.
8  *
9  * Copyright 2004 The Unichrome project.
10  * All Rights Reserved.
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a
13  * copy of this software and associated documentation files (the "Software"),
14  * to deal in the Software without restriction, including without limitation
15  * the rights to use, copy, modify, merge, publish, distribute, sub license,
16  * and/or sell copies of the Software, and to permit persons to whom the
17  * Software is furnished to do so, subject to the following conditions:
18  *
19  * The above copyright notice and this permission notice (including the
20  * next paragraph) shall be included in all copies or substantial portions
21  * of the Software.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
26  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
27  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
28  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
29  * USE OR OTHER DEALINGS IN THE SOFTWARE.
30  *
31  * Authors:
32  *    Tungsten Graphics,
33  *    Erdi Chen,
34  *    Thomas Hellstrom.
35  */
36 
37 #include <linux/delay.h>
38 #include <linux/uaccess.h>
39 
40 #include <drm/drm.h>
41 #include <drm/drm_device.h>
42 #include <drm/drm_file.h>
43 #include <drm/via_drm.h>
44 
45 #include "via_drv.h"
46 #include "via_3d_reg.h"
47 
48 #define CMDBUF_ALIGNMENT_SIZE   (0x100)
49 #define CMDBUF_ALIGNMENT_MASK   (0x0ff)
50 
51 /* defines for VIA 3D registers */
52 #define VIA_REG_STATUS          0x400
53 #define VIA_REG_TRANSET         0x43C
54 #define VIA_REG_TRANSPACE       0x440
55 
56 /* VIA_REG_STATUS(0x400): Engine Status */
57 #define VIA_CMD_RGTR_BUSY       0x00000080	/* Command Regulator is busy */
58 #define VIA_2D_ENG_BUSY         0x00000001	/* 2D Engine is busy */
59 #define VIA_3D_ENG_BUSY         0x00000002	/* 3D Engine is busy */
60 #define VIA_VR_QUEUE_BUSY       0x00020000	/* Virtual Queue is busy */
61 
62 #define SetReg2DAGP(nReg, nData) {				\
63 	*((uint32_t *)(vb)) = ((nReg) >> 2) | HALCYON_HEADER1;	\
64 	*((uint32_t *)(vb) + 1) = (nData);			\
65 	vb = ((uint32_t *)vb) + 2;				\
66 	dev_priv->dma_low += 8;					\
67 }
68 
69 #define via_flush_write_combine() mb()
70 
71 #define VIA_OUT_RING_QW(w1, w2)	do {		\
72 	*vb++ = (w1);				\
73 	*vb++ = (w2);				\
74 	dev_priv->dma_low += 8;			\
75 } while (0)
76 
77 static void via_cmdbuf_start(drm_via_private_t *dev_priv);
78 static void via_cmdbuf_pause(drm_via_private_t *dev_priv);
79 static void via_cmdbuf_reset(drm_via_private_t *dev_priv);
80 static void via_cmdbuf_rewind(drm_via_private_t *dev_priv);
81 static int via_wait_idle(drm_via_private_t *dev_priv);
82 static void via_pad_cache(drm_via_private_t *dev_priv, int qwords);
83 
84 /*
85  * Free space in command buffer.
86  */
87 
via_cmdbuf_space(drm_via_private_t * dev_priv)88 static uint32_t via_cmdbuf_space(drm_via_private_t *dev_priv)
89 {
90 	uint32_t agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr;
91 	uint32_t hw_addr = *(dev_priv->hw_addr_ptr) - agp_base;
92 
93 	return ((hw_addr <= dev_priv->dma_low) ?
94 		(dev_priv->dma_high + hw_addr - dev_priv->dma_low) :
95 		(hw_addr - dev_priv->dma_low));
96 }
97 
98 /*
99  * How much does the command regulator lag behind?
100  */
101 
via_cmdbuf_lag(drm_via_private_t * dev_priv)102 static uint32_t via_cmdbuf_lag(drm_via_private_t *dev_priv)
103 {
104 	uint32_t agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr;
105 	uint32_t hw_addr = *(dev_priv->hw_addr_ptr) - agp_base;
106 
107 	return ((hw_addr <= dev_priv->dma_low) ?
108 		(dev_priv->dma_low - hw_addr) :
109 		(dev_priv->dma_wrap + dev_priv->dma_low - hw_addr));
110 }
111 
112 /*
113  * Check that the given size fits in the buffer, otherwise wait.
114  */
115 
116 static inline int
via_cmdbuf_wait(drm_via_private_t * dev_priv,unsigned int size)117 via_cmdbuf_wait(drm_via_private_t *dev_priv, unsigned int size)
118 {
119 	uint32_t agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr;
120 	uint32_t cur_addr, hw_addr, next_addr;
121 	volatile uint32_t *hw_addr_ptr;
122 	uint32_t count;
123 	hw_addr_ptr = dev_priv->hw_addr_ptr;
124 	cur_addr = dev_priv->dma_low;
125 	next_addr = cur_addr + size + 512 * 1024;
126 	count = 1000000;
127 	do {
128 		hw_addr = *hw_addr_ptr - agp_base;
129 		if (count-- == 0) {
130 			DRM_ERROR
131 			    ("via_cmdbuf_wait timed out hw %x cur_addr %x next_addr %x\n",
132 			     hw_addr, cur_addr, next_addr);
133 			return -1;
134 		}
135 		if  ((cur_addr < hw_addr) && (next_addr >= hw_addr))
136 			msleep(1);
137 	} while ((cur_addr < hw_addr) && (next_addr >= hw_addr));
138 	return 0;
139 }
140 
141 /*
142  * Checks whether buffer head has reach the end. Rewind the ring buffer
143  * when necessary.
144  *
145  * Returns virtual pointer to ring buffer.
146  */
147 
via_check_dma(drm_via_private_t * dev_priv,unsigned int size)148 static inline uint32_t *via_check_dma(drm_via_private_t * dev_priv,
149 				      unsigned int size)
150 {
151 	if ((dev_priv->dma_low + size + 4 * CMDBUF_ALIGNMENT_SIZE) >
152 	    dev_priv->dma_high) {
153 		via_cmdbuf_rewind(dev_priv);
154 	}
155 	if (via_cmdbuf_wait(dev_priv, size) != 0)
156 		return NULL;
157 
158 	return (uint32_t *) (dev_priv->dma_ptr + dev_priv->dma_low);
159 }
160 
via_dma_cleanup(struct drm_device * dev)161 int via_dma_cleanup(struct drm_device *dev)
162 {
163 	if (dev->dev_private) {
164 		drm_via_private_t *dev_priv =
165 		    (drm_via_private_t *) dev->dev_private;
166 
167 		if (dev_priv->ring.virtual_start) {
168 			via_cmdbuf_reset(dev_priv);
169 
170 			drm_legacy_ioremapfree(&dev_priv->ring.map, dev);
171 			dev_priv->ring.virtual_start = NULL;
172 		}
173 
174 	}
175 
176 	return 0;
177 }
178 
via_initialize(struct drm_device * dev,drm_via_private_t * dev_priv,drm_via_dma_init_t * init)179 static int via_initialize(struct drm_device *dev,
180 			  drm_via_private_t *dev_priv,
181 			  drm_via_dma_init_t *init)
182 {
183 	if (!dev_priv || !dev_priv->mmio) {
184 		DRM_ERROR("via_dma_init called before via_map_init\n");
185 		return -EFAULT;
186 	}
187 
188 	if (dev_priv->ring.virtual_start != NULL) {
189 		DRM_ERROR("called again without calling cleanup\n");
190 		return -EFAULT;
191 	}
192 
193 	if (!dev->agp || !dev->agp->base) {
194 		DRM_ERROR("called with no agp memory available\n");
195 		return -EFAULT;
196 	}
197 
198 	if (dev_priv->chipset == VIA_DX9_0) {
199 		DRM_ERROR("AGP DMA is not supported on this chip\n");
200 		return -EINVAL;
201 	}
202 
203 	dev_priv->ring.map.offset = dev->agp->base + init->offset;
204 	dev_priv->ring.map.size = init->size;
205 	dev_priv->ring.map.type = 0;
206 	dev_priv->ring.map.flags = 0;
207 	dev_priv->ring.map.mtrr = 0;
208 
209 	drm_legacy_ioremap(&dev_priv->ring.map, dev);
210 
211 	if (dev_priv->ring.map.handle == NULL) {
212 		via_dma_cleanup(dev);
213 		DRM_ERROR("can not ioremap virtual address for"
214 			  " ring buffer\n");
215 		return -ENOMEM;
216 	}
217 
218 	dev_priv->ring.virtual_start = dev_priv->ring.map.handle;
219 
220 	dev_priv->dma_ptr = dev_priv->ring.virtual_start;
221 	dev_priv->dma_low = 0;
222 	dev_priv->dma_high = init->size;
223 	dev_priv->dma_wrap = init->size;
224 	dev_priv->dma_offset = init->offset;
225 	dev_priv->last_pause_ptr = NULL;
226 	dev_priv->hw_addr_ptr =
227 		(volatile uint32_t *)((char *)dev_priv->mmio->handle +
228 		init->reg_pause_addr);
229 
230 	via_cmdbuf_start(dev_priv);
231 
232 	return 0;
233 }
234 
via_dma_init(struct drm_device * dev,void * data,struct drm_file * file_priv)235 static int via_dma_init(struct drm_device *dev, void *data, struct drm_file *file_priv)
236 {
237 	drm_via_private_t *dev_priv = (drm_via_private_t *) dev->dev_private;
238 	drm_via_dma_init_t *init = data;
239 	int retcode = 0;
240 
241 	switch (init->func) {
242 	case VIA_INIT_DMA:
243 		if (!capable(CAP_SYS_ADMIN))
244 			retcode = -EPERM;
245 		else
246 			retcode = via_initialize(dev, dev_priv, init);
247 		break;
248 	case VIA_CLEANUP_DMA:
249 		if (!capable(CAP_SYS_ADMIN))
250 			retcode = -EPERM;
251 		else
252 			retcode = via_dma_cleanup(dev);
253 		break;
254 	case VIA_DMA_INITIALIZED:
255 		retcode = (dev_priv->ring.virtual_start != NULL) ?
256 			0 : -EFAULT;
257 		break;
258 	default:
259 		retcode = -EINVAL;
260 		break;
261 	}
262 
263 	return retcode;
264 }
265 
via_dispatch_cmdbuffer(struct drm_device * dev,drm_via_cmdbuffer_t * cmd)266 static int via_dispatch_cmdbuffer(struct drm_device *dev, drm_via_cmdbuffer_t *cmd)
267 {
268 	drm_via_private_t *dev_priv;
269 	uint32_t *vb;
270 	int ret;
271 
272 	dev_priv = (drm_via_private_t *) dev->dev_private;
273 
274 	if (dev_priv->ring.virtual_start == NULL) {
275 		DRM_ERROR("called without initializing AGP ring buffer.\n");
276 		return -EFAULT;
277 	}
278 
279 	if (cmd->size > VIA_PCI_BUF_SIZE)
280 		return -ENOMEM;
281 
282 	if (copy_from_user(dev_priv->pci_buf, cmd->buf, cmd->size))
283 		return -EFAULT;
284 
285 	/*
286 	 * Running this function on AGP memory is dead slow. Therefore
287 	 * we run it on a temporary cacheable system memory buffer and
288 	 * copy it to AGP memory when ready.
289 	 */
290 
291 	if ((ret =
292 	     via_verify_command_stream((uint32_t *) dev_priv->pci_buf,
293 				       cmd->size, dev, 1))) {
294 		return ret;
295 	}
296 
297 	vb = via_check_dma(dev_priv, (cmd->size < 0x100) ? 0x102 : cmd->size);
298 	if (vb == NULL)
299 		return -EAGAIN;
300 
301 	memcpy(vb, dev_priv->pci_buf, cmd->size);
302 
303 	dev_priv->dma_low += cmd->size;
304 
305 	/*
306 	 * Small submissions somehow stalls the CPU. (AGP cache effects?)
307 	 * pad to greater size.
308 	 */
309 
310 	if (cmd->size < 0x100)
311 		via_pad_cache(dev_priv, (0x100 - cmd->size) >> 3);
312 	via_cmdbuf_pause(dev_priv);
313 
314 	return 0;
315 }
316 
via_driver_dma_quiescent(struct drm_device * dev)317 int via_driver_dma_quiescent(struct drm_device *dev)
318 {
319 	drm_via_private_t *dev_priv = dev->dev_private;
320 
321 	if (!via_wait_idle(dev_priv))
322 		return -EBUSY;
323 	return 0;
324 }
325 
via_flush_ioctl(struct drm_device * dev,void * data,struct drm_file * file_priv)326 static int via_flush_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
327 {
328 
329 	LOCK_TEST_WITH_RETURN(dev, file_priv);
330 
331 	return via_driver_dma_quiescent(dev);
332 }
333 
via_cmdbuffer(struct drm_device * dev,void * data,struct drm_file * file_priv)334 static int via_cmdbuffer(struct drm_device *dev, void *data, struct drm_file *file_priv)
335 {
336 	drm_via_cmdbuffer_t *cmdbuf = data;
337 	int ret;
338 
339 	LOCK_TEST_WITH_RETURN(dev, file_priv);
340 
341 	DRM_DEBUG("buf %p size %lu\n", cmdbuf->buf, cmdbuf->size);
342 
343 	ret = via_dispatch_cmdbuffer(dev, cmdbuf);
344 	return ret;
345 }
346 
via_dispatch_pci_cmdbuffer(struct drm_device * dev,drm_via_cmdbuffer_t * cmd)347 static int via_dispatch_pci_cmdbuffer(struct drm_device *dev,
348 				      drm_via_cmdbuffer_t *cmd)
349 {
350 	drm_via_private_t *dev_priv = dev->dev_private;
351 	int ret;
352 
353 	if (cmd->size > VIA_PCI_BUF_SIZE)
354 		return -ENOMEM;
355 	if (copy_from_user(dev_priv->pci_buf, cmd->buf, cmd->size))
356 		return -EFAULT;
357 
358 	if ((ret =
359 	     via_verify_command_stream((uint32_t *) dev_priv->pci_buf,
360 				       cmd->size, dev, 0))) {
361 		return ret;
362 	}
363 
364 	ret =
365 	    via_parse_command_stream(dev, (const uint32_t *)dev_priv->pci_buf,
366 				     cmd->size);
367 	return ret;
368 }
369 
via_pci_cmdbuffer(struct drm_device * dev,void * data,struct drm_file * file_priv)370 static int via_pci_cmdbuffer(struct drm_device *dev, void *data, struct drm_file *file_priv)
371 {
372 	drm_via_cmdbuffer_t *cmdbuf = data;
373 	int ret;
374 
375 	LOCK_TEST_WITH_RETURN(dev, file_priv);
376 
377 	DRM_DEBUG("buf %p size %lu\n", cmdbuf->buf, cmdbuf->size);
378 
379 	ret = via_dispatch_pci_cmdbuffer(dev, cmdbuf);
380 	return ret;
381 }
382 
via_align_buffer(drm_via_private_t * dev_priv,uint32_t * vb,int qw_count)383 static inline uint32_t *via_align_buffer(drm_via_private_t *dev_priv,
384 					 uint32_t * vb, int qw_count)
385 {
386 	for (; qw_count > 0; --qw_count)
387 		VIA_OUT_RING_QW(HC_DUMMY, HC_DUMMY);
388 	return vb;
389 }
390 
391 /*
392  * This function is used internally by ring buffer management code.
393  *
394  * Returns virtual pointer to ring buffer.
395  */
via_get_dma(drm_via_private_t * dev_priv)396 static inline uint32_t *via_get_dma(drm_via_private_t *dev_priv)
397 {
398 	return (uint32_t *) (dev_priv->dma_ptr + dev_priv->dma_low);
399 }
400 
401 /*
402  * Hooks a segment of data into the tail of the ring-buffer by
403  * modifying the pause address stored in the buffer itself. If
404  * the regulator has already paused, restart it.
405  */
via_hook_segment(drm_via_private_t * dev_priv,uint32_t pause_addr_hi,uint32_t pause_addr_lo,int no_pci_fire)406 static int via_hook_segment(drm_via_private_t *dev_priv,
407 			    uint32_t pause_addr_hi, uint32_t pause_addr_lo,
408 			    int no_pci_fire)
409 {
410 	int paused, count;
411 	volatile uint32_t *paused_at = dev_priv->last_pause_ptr;
412 	uint32_t reader, ptr;
413 	uint32_t diff;
414 
415 	paused = 0;
416 	via_flush_write_combine();
417 	(void) *(volatile uint32_t *)(via_get_dma(dev_priv) - 1);
418 
419 	*paused_at = pause_addr_lo;
420 	via_flush_write_combine();
421 	(void) *paused_at;
422 
423 	reader = *(dev_priv->hw_addr_ptr);
424 	ptr = ((volatile char *)paused_at - dev_priv->dma_ptr) +
425 		dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr + 4;
426 
427 	dev_priv->last_pause_ptr = via_get_dma(dev_priv) - 1;
428 
429 	/*
430 	 * If there is a possibility that the command reader will
431 	 * miss the new pause address and pause on the old one,
432 	 * In that case we need to program the new start address
433 	 * using PCI.
434 	 */
435 
436 	diff = (uint32_t) (ptr - reader) - dev_priv->dma_diff;
437 	count = 10000000;
438 	while (diff == 0 && count--) {
439 		paused = (via_read(dev_priv, 0x41c) & 0x80000000);
440 		if (paused)
441 			break;
442 		reader = *(dev_priv->hw_addr_ptr);
443 		diff = (uint32_t) (ptr - reader) - dev_priv->dma_diff;
444 	}
445 
446 	paused = via_read(dev_priv, 0x41c) & 0x80000000;
447 
448 	if (paused && !no_pci_fire) {
449 		reader = *(dev_priv->hw_addr_ptr);
450 		diff = (uint32_t) (ptr - reader) - dev_priv->dma_diff;
451 		diff &= (dev_priv->dma_high - 1);
452 		if (diff != 0 && diff < (dev_priv->dma_high >> 1)) {
453 			DRM_ERROR("Paused at incorrect address. "
454 				  "0x%08x, 0x%08x 0x%08x\n",
455 				  ptr, reader, dev_priv->dma_diff);
456 		} else if (diff == 0) {
457 			/*
458 			 * There is a concern that these writes may stall the PCI bus
459 			 * if the GPU is not idle. However, idling the GPU first
460 			 * doesn't make a difference.
461 			 */
462 
463 			via_write(dev_priv, VIA_REG_TRANSET, (HC_ParaType_PreCR << 16));
464 			via_write(dev_priv, VIA_REG_TRANSPACE, pause_addr_hi);
465 			via_write(dev_priv, VIA_REG_TRANSPACE, pause_addr_lo);
466 			via_read(dev_priv, VIA_REG_TRANSPACE);
467 		}
468 	}
469 	return paused;
470 }
471 
via_wait_idle(drm_via_private_t * dev_priv)472 static int via_wait_idle(drm_via_private_t *dev_priv)
473 {
474 	int count = 10000000;
475 
476 	while (!(via_read(dev_priv, VIA_REG_STATUS) & VIA_VR_QUEUE_BUSY) && --count)
477 		;
478 
479 	while (count && (via_read(dev_priv, VIA_REG_STATUS) &
480 			   (VIA_CMD_RGTR_BUSY | VIA_2D_ENG_BUSY |
481 			    VIA_3D_ENG_BUSY)))
482 		--count;
483 	return count;
484 }
485 
via_align_cmd(drm_via_private_t * dev_priv,uint32_t cmd_type,uint32_t addr,uint32_t * cmd_addr_hi,uint32_t * cmd_addr_lo,int skip_wait)486 static uint32_t *via_align_cmd(drm_via_private_t *dev_priv, uint32_t cmd_type,
487 			       uint32_t addr, uint32_t *cmd_addr_hi,
488 			       uint32_t *cmd_addr_lo, int skip_wait)
489 {
490 	uint32_t agp_base;
491 	uint32_t cmd_addr, addr_lo, addr_hi;
492 	uint32_t *vb;
493 	uint32_t qw_pad_count;
494 
495 	if (!skip_wait)
496 		via_cmdbuf_wait(dev_priv, 2 * CMDBUF_ALIGNMENT_SIZE);
497 
498 	vb = via_get_dma(dev_priv);
499 	VIA_OUT_RING_QW(HC_HEADER2 | ((VIA_REG_TRANSET >> 2) << 12) |
500 			(VIA_REG_TRANSPACE >> 2), HC_ParaType_PreCR << 16);
501 	agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr;
502 	qw_pad_count = (CMDBUF_ALIGNMENT_SIZE >> 3) -
503 	    ((dev_priv->dma_low & CMDBUF_ALIGNMENT_MASK) >> 3);
504 
505 	cmd_addr = (addr) ? addr :
506 	    agp_base + dev_priv->dma_low - 8 + (qw_pad_count << 3);
507 	addr_lo = ((HC_SubA_HAGPBpL << 24) | (cmd_type & HC_HAGPBpID_MASK) |
508 		   (cmd_addr & HC_HAGPBpL_MASK));
509 	addr_hi = ((HC_SubA_HAGPBpH << 24) | (cmd_addr >> 24));
510 
511 	vb = via_align_buffer(dev_priv, vb, qw_pad_count - 1);
512 	VIA_OUT_RING_QW(*cmd_addr_hi = addr_hi, *cmd_addr_lo = addr_lo);
513 	return vb;
514 }
515 
via_cmdbuf_start(drm_via_private_t * dev_priv)516 static void via_cmdbuf_start(drm_via_private_t *dev_priv)
517 {
518 	uint32_t pause_addr_lo, pause_addr_hi;
519 	uint32_t start_addr, start_addr_lo;
520 	uint32_t end_addr, end_addr_lo;
521 	uint32_t command;
522 	uint32_t agp_base;
523 	uint32_t ptr;
524 	uint32_t reader;
525 	int count;
526 
527 	dev_priv->dma_low = 0;
528 
529 	agp_base = dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr;
530 	start_addr = agp_base;
531 	end_addr = agp_base + dev_priv->dma_high;
532 
533 	start_addr_lo = ((HC_SubA_HAGPBstL << 24) | (start_addr & 0xFFFFFF));
534 	end_addr_lo = ((HC_SubA_HAGPBendL << 24) | (end_addr & 0xFFFFFF));
535 	command = ((HC_SubA_HAGPCMNT << 24) | (start_addr >> 24) |
536 		   ((end_addr & 0xff000000) >> 16));
537 
538 	dev_priv->last_pause_ptr =
539 	    via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0,
540 			  &pause_addr_hi, &pause_addr_lo, 1) - 1;
541 
542 	via_flush_write_combine();
543 	(void) *(volatile uint32_t *)dev_priv->last_pause_ptr;
544 
545 	via_write(dev_priv, VIA_REG_TRANSET, (HC_ParaType_PreCR << 16));
546 	via_write(dev_priv, VIA_REG_TRANSPACE, command);
547 	via_write(dev_priv, VIA_REG_TRANSPACE, start_addr_lo);
548 	via_write(dev_priv, VIA_REG_TRANSPACE, end_addr_lo);
549 
550 	via_write(dev_priv, VIA_REG_TRANSPACE, pause_addr_hi);
551 	via_write(dev_priv, VIA_REG_TRANSPACE, pause_addr_lo);
552 	wmb();
553 	via_write(dev_priv, VIA_REG_TRANSPACE, command | HC_HAGPCMNT_MASK);
554 	via_read(dev_priv, VIA_REG_TRANSPACE);
555 
556 	dev_priv->dma_diff = 0;
557 
558 	count = 10000000;
559 	while (!(via_read(dev_priv, 0x41c) & 0x80000000) && count--);
560 
561 	reader = *(dev_priv->hw_addr_ptr);
562 	ptr = ((volatile char *)dev_priv->last_pause_ptr - dev_priv->dma_ptr) +
563 	    dev_priv->dma_offset + (uint32_t) dev_priv->agpAddr + 4;
564 
565 	/*
566 	 * This is the difference between where we tell the
567 	 * command reader to pause and where it actually pauses.
568 	 * This differs between hw implementation so we need to
569 	 * detect it.
570 	 */
571 
572 	dev_priv->dma_diff = ptr - reader;
573 }
574 
via_pad_cache(drm_via_private_t * dev_priv,int qwords)575 static void via_pad_cache(drm_via_private_t *dev_priv, int qwords)
576 {
577 	uint32_t *vb;
578 
579 	via_cmdbuf_wait(dev_priv, qwords + 2);
580 	vb = via_get_dma(dev_priv);
581 	VIA_OUT_RING_QW(HC_HEADER2, HC_ParaType_NotTex << 16);
582 	via_align_buffer(dev_priv, vb, qwords);
583 }
584 
via_dummy_bitblt(drm_via_private_t * dev_priv)585 static inline void via_dummy_bitblt(drm_via_private_t *dev_priv)
586 {
587 	uint32_t *vb = via_get_dma(dev_priv);
588 	SetReg2DAGP(0x0C, (0 | (0 << 16)));
589 	SetReg2DAGP(0x10, 0 | (0 << 16));
590 	SetReg2DAGP(0x0, 0x1 | 0x2000 | 0xAA000000);
591 }
592 
via_cmdbuf_jump(drm_via_private_t * dev_priv)593 static void via_cmdbuf_jump(drm_via_private_t *dev_priv)
594 {
595 	uint32_t pause_addr_lo, pause_addr_hi;
596 	uint32_t jump_addr_lo, jump_addr_hi;
597 	volatile uint32_t *last_pause_ptr;
598 	uint32_t dma_low_save1, dma_low_save2;
599 
600 	via_align_cmd(dev_priv, HC_HAGPBpID_JUMP, 0, &jump_addr_hi,
601 		      &jump_addr_lo, 0);
602 
603 	dev_priv->dma_wrap = dev_priv->dma_low;
604 
605 	/*
606 	 * Wrap command buffer to the beginning.
607 	 */
608 
609 	dev_priv->dma_low = 0;
610 	if (via_cmdbuf_wait(dev_priv, CMDBUF_ALIGNMENT_SIZE) != 0)
611 		DRM_ERROR("via_cmdbuf_jump failed\n");
612 
613 	via_dummy_bitblt(dev_priv);
614 	via_dummy_bitblt(dev_priv);
615 
616 	last_pause_ptr =
617 	    via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi,
618 			  &pause_addr_lo, 0) - 1;
619 	via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi,
620 		      &pause_addr_lo, 0);
621 
622 	*last_pause_ptr = pause_addr_lo;
623 	dma_low_save1 = dev_priv->dma_low;
624 
625 	/*
626 	 * Now, set a trap that will pause the regulator if it tries to rerun the old
627 	 * command buffer. (Which may happen if via_hook_segment detecs a command regulator pause
628 	 * and reissues the jump command over PCI, while the regulator has already taken the jump
629 	 * and actually paused at the current buffer end).
630 	 * There appears to be no other way to detect this condition, since the hw_addr_pointer
631 	 * does not seem to get updated immediately when a jump occurs.
632 	 */
633 
634 	last_pause_ptr =
635 		via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi,
636 			      &pause_addr_lo, 0) - 1;
637 	via_align_cmd(dev_priv, HC_HAGPBpID_PAUSE, 0, &pause_addr_hi,
638 		      &pause_addr_lo, 0);
639 	*last_pause_ptr = pause_addr_lo;
640 
641 	dma_low_save2 = dev_priv->dma_low;
642 	dev_priv->dma_low = dma_low_save1;
643 	via_hook_segment(dev_priv, jump_addr_hi, jump_addr_lo, 0);
644 	dev_priv->dma_low = dma_low_save2;
645 	via_hook_segment(dev_priv, pause_addr_hi, pause_addr_lo, 0);
646 }
647 
648 
via_cmdbuf_rewind(drm_via_private_t * dev_priv)649 static void via_cmdbuf_rewind(drm_via_private_t *dev_priv)
650 {
651 	via_cmdbuf_jump(dev_priv);
652 }
653 
via_cmdbuf_flush(drm_via_private_t * dev_priv,uint32_t cmd_type)654 static void via_cmdbuf_flush(drm_via_private_t *dev_priv, uint32_t cmd_type)
655 {
656 	uint32_t pause_addr_lo, pause_addr_hi;
657 
658 	via_align_cmd(dev_priv, cmd_type, 0, &pause_addr_hi, &pause_addr_lo, 0);
659 	via_hook_segment(dev_priv, pause_addr_hi, pause_addr_lo, 0);
660 }
661 
via_cmdbuf_pause(drm_via_private_t * dev_priv)662 static void via_cmdbuf_pause(drm_via_private_t *dev_priv)
663 {
664 	via_cmdbuf_flush(dev_priv, HC_HAGPBpID_PAUSE);
665 }
666 
via_cmdbuf_reset(drm_via_private_t * dev_priv)667 static void via_cmdbuf_reset(drm_via_private_t *dev_priv)
668 {
669 	via_cmdbuf_flush(dev_priv, HC_HAGPBpID_STOP);
670 	via_wait_idle(dev_priv);
671 }
672 
673 /*
674  * User interface to the space and lag functions.
675  */
676 
via_cmdbuf_size(struct drm_device * dev,void * data,struct drm_file * file_priv)677 static int via_cmdbuf_size(struct drm_device *dev, void *data, struct drm_file *file_priv)
678 {
679 	drm_via_cmdbuf_size_t *d_siz = data;
680 	int ret = 0;
681 	uint32_t tmp_size, count;
682 	drm_via_private_t *dev_priv;
683 
684 	DRM_DEBUG("\n");
685 	LOCK_TEST_WITH_RETURN(dev, file_priv);
686 
687 	dev_priv = (drm_via_private_t *) dev->dev_private;
688 
689 	if (dev_priv->ring.virtual_start == NULL) {
690 		DRM_ERROR("called without initializing AGP ring buffer.\n");
691 		return -EFAULT;
692 	}
693 
694 	count = 1000000;
695 	tmp_size = d_siz->size;
696 	switch (d_siz->func) {
697 	case VIA_CMDBUF_SPACE:
698 		while (((tmp_size = via_cmdbuf_space(dev_priv)) < d_siz->size)
699 		       && --count) {
700 			if (!d_siz->wait)
701 				break;
702 		}
703 		if (!count) {
704 			DRM_ERROR("VIA_CMDBUF_SPACE timed out.\n");
705 			ret = -EAGAIN;
706 		}
707 		break;
708 	case VIA_CMDBUF_LAG:
709 		while (((tmp_size = via_cmdbuf_lag(dev_priv)) > d_siz->size)
710 		       && --count) {
711 			if (!d_siz->wait)
712 				break;
713 		}
714 		if (!count) {
715 			DRM_ERROR("VIA_CMDBUF_LAG timed out.\n");
716 			ret = -EAGAIN;
717 		}
718 		break;
719 	default:
720 		ret = -EFAULT;
721 	}
722 	d_siz->size = tmp_size;
723 
724 	return ret;
725 }
726 
727 const struct drm_ioctl_desc via_ioctls[] = {
728 	DRM_IOCTL_DEF_DRV(VIA_ALLOCMEM, via_mem_alloc, DRM_AUTH),
729 	DRM_IOCTL_DEF_DRV(VIA_FREEMEM, via_mem_free, DRM_AUTH),
730 	DRM_IOCTL_DEF_DRV(VIA_AGP_INIT, via_agp_init, DRM_AUTH|DRM_MASTER),
731 	DRM_IOCTL_DEF_DRV(VIA_FB_INIT, via_fb_init, DRM_AUTH|DRM_MASTER),
732 	DRM_IOCTL_DEF_DRV(VIA_MAP_INIT, via_map_init, DRM_AUTH|DRM_MASTER),
733 	DRM_IOCTL_DEF_DRV(VIA_DEC_FUTEX, via_decoder_futex, DRM_AUTH),
734 	DRM_IOCTL_DEF_DRV(VIA_DMA_INIT, via_dma_init, DRM_AUTH),
735 	DRM_IOCTL_DEF_DRV(VIA_CMDBUFFER, via_cmdbuffer, DRM_AUTH),
736 	DRM_IOCTL_DEF_DRV(VIA_FLUSH, via_flush_ioctl, DRM_AUTH),
737 	DRM_IOCTL_DEF_DRV(VIA_PCICMD, via_pci_cmdbuffer, DRM_AUTH),
738 	DRM_IOCTL_DEF_DRV(VIA_CMDBUF_SIZE, via_cmdbuf_size, DRM_AUTH),
739 	DRM_IOCTL_DEF_DRV(VIA_WAIT_IRQ, via_wait_irq, DRM_AUTH),
740 	DRM_IOCTL_DEF_DRV(VIA_DMA_BLIT, via_dma_blit, DRM_AUTH),
741 	DRM_IOCTL_DEF_DRV(VIA_BLIT_SYNC, via_dma_blit_sync, DRM_AUTH)
742 };
743 
744 int via_max_ioctl = ARRAY_SIZE(via_ioctls);
745