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