xref: /netbsd-src/sys/external/bsd/drm/dist/shared-core/savage_state.c (revision 76b938c44aa92027e7964b0265cb60ce2c71788c)
1 /* savage_state.c -- State and drawing 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 #include "drmP.h"
26 #include "savage_drm.h"
27 #include "savage_drv.h"
28 
savage_emit_clip_rect_s3d(drm_savage_private_t * dev_priv,const struct drm_clip_rect * pbox)29 void savage_emit_clip_rect_s3d(drm_savage_private_t *dev_priv,
30 			       const struct drm_clip_rect *pbox)
31 {
32 	uint32_t scstart = dev_priv->state.s3d.new_scstart;
33 	uint32_t scend = dev_priv->state.s3d.new_scend;
34 	scstart = (scstart & ~SAVAGE_SCISSOR_MASK_S3D) |
35 		((uint32_t)pbox->x1 & 0x000007ff) |
36 		(((uint32_t)pbox->y1 << 16) & 0x07ff0000);
37 	scend   = (scend & ~SAVAGE_SCISSOR_MASK_S3D) |
38 		(((uint32_t)pbox->x2 - 1) & 0x000007ff) |
39 		((((uint32_t)pbox->y2 - 1) << 16) & 0x07ff0000);
40 	if (scstart != dev_priv->state.s3d.scstart ||
41 	    scend   != dev_priv->state.s3d.scend) {
42 		DMA_LOCALS;
43 		BEGIN_DMA(4);
44 		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
45 		DMA_SET_REGISTERS(SAVAGE_SCSTART_S3D, 2);
46 		DMA_WRITE(scstart);
47 		DMA_WRITE(scend);
48 		dev_priv->state.s3d.scstart = scstart;
49 		dev_priv->state.s3d.scend = scend;
50 		dev_priv->waiting = 1;
51 		DMA_COMMIT();
52 	}
53 }
54 
savage_emit_clip_rect_s4(drm_savage_private_t * dev_priv,const struct drm_clip_rect * pbox)55 void savage_emit_clip_rect_s4(drm_savage_private_t *dev_priv,
56 			      const struct drm_clip_rect *pbox)
57 {
58 	uint32_t drawctrl0 = dev_priv->state.s4.new_drawctrl0;
59 	uint32_t drawctrl1 = dev_priv->state.s4.new_drawctrl1;
60 	drawctrl0 = (drawctrl0 & ~SAVAGE_SCISSOR_MASK_S4) |
61 		((uint32_t)pbox->x1 & 0x000007ff) |
62 		(((uint32_t)pbox->y1 << 12) & 0x00fff000);
63 	drawctrl1 = (drawctrl1 & ~SAVAGE_SCISSOR_MASK_S4) |
64 		(((uint32_t)pbox->x2 - 1) & 0x000007ff) |
65 		((((uint32_t)pbox->y2 - 1) << 12) & 0x00fff000);
66 	if (drawctrl0 != dev_priv->state.s4.drawctrl0 ||
67 	    drawctrl1 != dev_priv->state.s4.drawctrl1) {
68 		DMA_LOCALS;
69 		BEGIN_DMA(4);
70 		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
71 		DMA_SET_REGISTERS(SAVAGE_DRAWCTRL0_S4, 2);
72 		DMA_WRITE(drawctrl0);
73 		DMA_WRITE(drawctrl1);
74 		dev_priv->state.s4.drawctrl0 = drawctrl0;
75 		dev_priv->state.s4.drawctrl1 = drawctrl1;
76 		dev_priv->waiting = 1;
77 		DMA_COMMIT();
78 	}
79 }
80 
savage_verify_texaddr(drm_savage_private_t * dev_priv,int unit,uint32_t addr)81 static int savage_verify_texaddr(drm_savage_private_t *dev_priv, int unit,
82 				 uint32_t addr)
83 {
84 	if ((addr & 6) != 2) { /* reserved bits */
85 		DRM_ERROR("bad texAddr%d %08x (reserved bits)\n", unit, addr);
86 		return -EINVAL;
87 	}
88 	if (!(addr & 1)) { /* local */
89 		addr &= ~7;
90 		if (addr < dev_priv->texture_offset ||
91 		    addr >= dev_priv->texture_offset + dev_priv->texture_size) {
92 			DRM_ERROR
93 			    ("bad texAddr%d %08x (local addr out of range)\n",
94 			     unit, addr);
95 			return -EINVAL;
96 		}
97 	} else { /* AGP */
98 		if (!dev_priv->agp_textures) {
99 			DRM_ERROR("bad texAddr%d %08x (AGP not available)\n",
100 				  unit, addr);
101 			return -EINVAL;
102 		}
103 		addr &= ~7;
104 		if (addr < dev_priv->agp_textures->offset ||
105 		    addr >= (dev_priv->agp_textures->offset +
106 			     dev_priv->agp_textures->size)) {
107 			DRM_ERROR
108 			    ("bad texAddr%d %08x (AGP addr out of range)\n",
109 			     unit, addr);
110 			return -EINVAL;
111 		}
112 	}
113 	return 0;
114 }
115 
116 #define SAVE_STATE(reg,where)			\
117 	if(start <= reg && start + count > reg)	\
118 		dev_priv->state.where = regs[reg - start]
119 #define SAVE_STATE_MASK(reg,where,mask) do {			\
120 	if(start <= reg && start + count > reg) {			\
121 		uint32_t tmp;					\
122 		tmp = regs[reg - start];			\
123 		dev_priv->state.where = (tmp & (mask)) |	\
124 			(dev_priv->state.where & ~(mask));	\
125 	}							\
126 } while (0)
savage_verify_state_s3d(drm_savage_private_t * dev_priv,unsigned int start,unsigned int count,const uint32_t * regs)127 static int savage_verify_state_s3d(drm_savage_private_t *dev_priv,
128 				   unsigned int start, unsigned int count,
129 				   const uint32_t *regs)
130 {
131 	if (start < SAVAGE_TEXPALADDR_S3D ||
132 	    start + count - 1 > SAVAGE_DESTTEXRWWATERMARK_S3D) {
133 		DRM_ERROR("invalid register range (0x%04x-0x%04x)\n",
134 			  start, start + count - 1);
135 		return -EINVAL;
136 	}
137 
138 	SAVE_STATE_MASK(SAVAGE_SCSTART_S3D, s3d.new_scstart,
139 			~SAVAGE_SCISSOR_MASK_S3D);
140 	SAVE_STATE_MASK(SAVAGE_SCEND_S3D, s3d.new_scend,
141 			~SAVAGE_SCISSOR_MASK_S3D);
142 
143 	/* if any texture regs were changed ... */
144 	if (start <= SAVAGE_TEXCTRL_S3D &&
145 	    start + count > SAVAGE_TEXPALADDR_S3D) {
146 		/* ... check texture state */
147 		SAVE_STATE(SAVAGE_TEXCTRL_S3D, s3d.texctrl);
148 		SAVE_STATE(SAVAGE_TEXADDR_S3D, s3d.texaddr);
149 		if (dev_priv->state.s3d.texctrl & SAVAGE_TEXCTRL_TEXEN_MASK)
150 			return savage_verify_texaddr(dev_priv, 0,
151 						dev_priv->state.s3d.texaddr);
152 	}
153 
154 	return 0;
155 }
156 
savage_verify_state_s4(drm_savage_private_t * dev_priv,unsigned int start,unsigned int count,const uint32_t * regs)157 static int savage_verify_state_s4(drm_savage_private_t *dev_priv,
158 				  unsigned int start, unsigned int count,
159 				  const uint32_t *regs)
160 {
161 	int ret = 0;
162 
163 	if (start < SAVAGE_DRAWLOCALCTRL_S4 ||
164 	    start + count - 1 > SAVAGE_TEXBLENDCOLOR_S4) {
165 		DRM_ERROR("invalid register range (0x%04x-0x%04x)\n",
166 			  start, start + count - 1);
167 		return -EINVAL;
168 	}
169 
170 	SAVE_STATE_MASK(SAVAGE_DRAWCTRL0_S4, s4.new_drawctrl0,
171 			~SAVAGE_SCISSOR_MASK_S4);
172 	SAVE_STATE_MASK(SAVAGE_DRAWCTRL1_S4, s4.new_drawctrl1,
173 			~SAVAGE_SCISSOR_MASK_S4);
174 
175 	/* if any texture regs were changed ... */
176 	if (start <= SAVAGE_TEXDESCR_S4 &&
177 	    start + count > SAVAGE_TEXPALADDR_S4) {
178 		/* ... check texture state */
179 		SAVE_STATE(SAVAGE_TEXDESCR_S4, s4.texdescr);
180 		SAVE_STATE(SAVAGE_TEXADDR0_S4, s4.texaddr0);
181 		SAVE_STATE(SAVAGE_TEXADDR1_S4, s4.texaddr1);
182 		if (dev_priv->state.s4.texdescr & SAVAGE_TEXDESCR_TEX0EN_MASK)
183 			ret |= savage_verify_texaddr(dev_priv, 0,
184 						dev_priv->state.s4.texaddr0);
185 		if (dev_priv->state.s4.texdescr & SAVAGE_TEXDESCR_TEX1EN_MASK)
186 			ret |= savage_verify_texaddr(dev_priv, 1,
187 						dev_priv->state.s4.texaddr1);
188 	}
189 
190 	return ret;
191 }
192 #undef SAVE_STATE
193 #undef SAVE_STATE_MASK
194 
savage_dispatch_state(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * cmd_header,const uint32_t * regs)195 static int savage_dispatch_state(drm_savage_private_t *dev_priv,
196 				 const drm_savage_cmd_header_t *cmd_header,
197 				 const uint32_t *regs)
198 {
199 	unsigned int count = cmd_header->state.count;
200 	unsigned int start = cmd_header->state.start;
201 	unsigned int count2 = 0;
202 	unsigned int bci_size;
203 	int ret;
204 	DMA_LOCALS;
205 
206 	if (!count)
207 		return 0;
208 
209 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
210 		ret = savage_verify_state_s3d(dev_priv, start, count, regs);
211 		if (ret != 0)
212 			return ret;
213 		/* scissor regs are emitted in savage_dispatch_draw */
214 		if (start < SAVAGE_SCSTART_S3D) {
215 			if (start + count > SAVAGE_SCEND_S3D + 1)
216 				count2 = count - (SAVAGE_SCEND_S3D + 1 - start);
217 			if (start + count > SAVAGE_SCSTART_S3D)
218 				count = SAVAGE_SCSTART_S3D - start;
219 		} else if (start <= SAVAGE_SCEND_S3D) {
220 			if (start + count > SAVAGE_SCEND_S3D + 1) {
221 				count -= SAVAGE_SCEND_S3D + 1 - start;
222 				start = SAVAGE_SCEND_S3D + 1;
223 			} else
224 				return 0;
225 		}
226 	} else {
227 		ret = savage_verify_state_s4(dev_priv, start, count, regs);
228 		if (ret != 0)
229 			return ret;
230 		/* scissor regs are emitted in savage_dispatch_draw */
231 		if (start < SAVAGE_DRAWCTRL0_S4) {
232 			if (start + count > SAVAGE_DRAWCTRL1_S4 + 1)
233 				count2 = count -
234 					 (SAVAGE_DRAWCTRL1_S4 + 1 - start);
235 			if (start + count > SAVAGE_DRAWCTRL0_S4)
236 				count = SAVAGE_DRAWCTRL0_S4 - start;
237 		} else if (start <= SAVAGE_DRAWCTRL1_S4) {
238 			if (start + count > SAVAGE_DRAWCTRL1_S4 + 1) {
239 				count -= SAVAGE_DRAWCTRL1_S4 + 1 - start;
240 				start = SAVAGE_DRAWCTRL1_S4 + 1;
241 			} else
242 				return 0;
243 		}
244 	}
245 
246 	bci_size = count + (count + 254) / 255 + count2 + (count2 + 254) / 255;
247 
248 	if (cmd_header->state.global) {
249 		BEGIN_DMA(bci_size + 1);
250 		DMA_WRITE(BCI_CMD_WAIT | BCI_CMD_WAIT_3D);
251 		dev_priv->waiting = 1;
252 	} else {
253 		BEGIN_DMA(bci_size);
254 	}
255 
256 	do {
257 		while (count > 0) {
258 			unsigned int n = count < 255 ? count : 255;
259 			DMA_SET_REGISTERS(start, n);
260 			DMA_COPY(regs, n);
261 			count -= n;
262 			start += n;
263 			regs += n;
264 		}
265 		start += 2;
266 		regs += 2;
267 		count = count2;
268 		count2 = 0;
269 	} while (count);
270 
271 	DMA_COMMIT();
272 
273 	return 0;
274 }
275 
savage_dispatch_dma_prim(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * cmd_header,const struct drm_buf * dmabuf)276 static int savage_dispatch_dma_prim(drm_savage_private_t *dev_priv,
277 				    const drm_savage_cmd_header_t *cmd_header,
278 				    const struct drm_buf *dmabuf)
279 {
280 	unsigned char reorder = 0;
281 	unsigned int prim = cmd_header->prim.prim;
282 	unsigned int skip = cmd_header->prim.skip;
283 	unsigned int n = cmd_header->prim.count;
284 	unsigned int start = cmd_header->prim.start;
285 	unsigned int i;
286 	BCI_LOCALS;
287 
288 	if (!dmabuf) {
289 		DRM_ERROR("called without dma buffers!\n");
290 		return -EINVAL;
291 	}
292 
293 	if (!n)
294 		return 0;
295 
296 	switch (prim) {
297 	case SAVAGE_PRIM_TRILIST_201:
298 		reorder = 1;
299 		prim = SAVAGE_PRIM_TRILIST;
300 		/* FALLTHROUGH */
301 	case SAVAGE_PRIM_TRILIST:
302 		if (n % 3 != 0) {
303 			DRM_ERROR("wrong number of vertices %u in TRILIST\n",
304 				  n);
305 			return -EINVAL;
306 		}
307 		break;
308 	case SAVAGE_PRIM_TRISTRIP:
309 	case SAVAGE_PRIM_TRIFAN:
310 		if (n < 3) {
311 			DRM_ERROR
312 			   ("wrong number of vertices %u in TRIFAN/STRIP\n",
313 			    n);
314 			return -EINVAL;
315 		}
316 		break;
317 	default:
318 		DRM_ERROR("invalid primitive type %u\n", prim);
319 		return -EINVAL;
320 	}
321 
322 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
323 		if (skip != 0) {
324 			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
325 			return -EINVAL;
326 		}
327 	} else {
328 		unsigned int size = 10 - (skip & 1) - (skip >> 1 & 1) -
329 			(skip >> 2 & 1) - (skip >> 3 & 1) - (skip >> 4 & 1) -
330 			(skip >> 5 & 1) - (skip >> 6 & 1) - (skip >> 7 & 1);
331 		if (skip > SAVAGE_SKIP_ALL_S4 || size != 8) {
332 			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
333 			return -EINVAL;
334 		}
335 		if (reorder) {
336 			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
337 			return -EINVAL;
338 		}
339 	}
340 
341 	if (start + n > dmabuf->total / 32) {
342 		DRM_ERROR("vertex indices (%u-%u) out of range (0-%u)\n",
343 			  start, start + n - 1, dmabuf->total / 32);
344 		return -EINVAL;
345 	}
346 
347 	/* Vertex DMA doesn't work with command DMA at the same time,
348 	 * so we use BCI_... to submit commands here. Flush buffered
349 	 * faked DMA first. */
350 	DMA_FLUSH();
351 
352 	if (dmabuf->bus_address != dev_priv->state.common.vbaddr) {
353 		BEGIN_BCI(2);
354 		BCI_SET_REGISTERS(SAVAGE_VERTBUFADDR, 1);
355 		BCI_WRITE(dmabuf->bus_address | dev_priv->dma_type);
356 		dev_priv->state.common.vbaddr = dmabuf->bus_address;
357 	}
358 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset) && dev_priv->waiting) {
359 		/* Workaround for what looks like a hardware bug. If a
360 		 * WAIT_3D_IDLE was emitted some time before the
361 		 * indexed drawing command then the engine will lock
362 		 * up. There are two known workarounds:
363 		 * WAIT_IDLE_EMPTY or emit at least 63 NOPs. */
364 		BEGIN_BCI(63);
365 		for (i = 0; i < 63; ++i)
366 			BCI_WRITE(BCI_CMD_WAIT);
367 		dev_priv->waiting = 0;
368 	}
369 
370 	prim <<= 25;
371 	while (n != 0) {
372 		/* Can emit up to 255 indices (85 triangles) at once. */
373 		unsigned int count = n > 255 ? 255 : n;
374 		if (reorder) {
375 			/* Need to reorder indices for correct flat
376 			 * shading while preserving the clock sense
377 			 * for correct culling. Only on Savage3D. */
378 			int reorderarr[3] = { -1, -1, -1 };
379 			reorderarr[start % 3] = 2;
380 
381 			BEGIN_BCI((count + 1 + 1) / 2);
382 			BCI_DRAW_INDICES_S3D(count, prim, start + 2);
383 
384 			for (i = start + 1; i + 1 < start + count; i += 2)
385 				BCI_WRITE((i + reorderarr[i % 3]) |
386 					  ((i + 1 +
387 					    reorderarr[(i + 1) % 3]) << 16));
388 			if (i < start + count)
389 				BCI_WRITE(i + reorderarr[i % 3]);
390 		} else if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
391 			BEGIN_BCI((count + 1 + 1) / 2);
392 			BCI_DRAW_INDICES_S3D(count, prim, start);
393 
394 			for (i = start + 1; i + 1 < start + count; i += 2)
395 				BCI_WRITE(i | ((i + 1) << 16));
396 			if (i < start + count)
397 				BCI_WRITE(i);
398 		} else {
399 			BEGIN_BCI((count + 2 + 1) / 2);
400 			BCI_DRAW_INDICES_S4(count, prim, skip);
401 
402 			for (i = start; i + 1 < start + count; i += 2)
403 				BCI_WRITE(i | ((i + 1) << 16));
404 			if (i < start + count)
405 				BCI_WRITE(i);
406 		}
407 
408 		start += count;
409 		n -= count;
410 
411 		prim |= BCI_CMD_DRAW_CONT;
412 	}
413 
414 	return 0;
415 }
416 
savage_dispatch_vb_prim(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * cmd_header,const uint32_t * vtxbuf,unsigned int vb_size,unsigned int vb_stride)417 static int savage_dispatch_vb_prim(drm_savage_private_t *dev_priv,
418 				   const drm_savage_cmd_header_t *cmd_header,
419 				   const uint32_t *vtxbuf, unsigned int vb_size,
420 				   unsigned int vb_stride)
421 {
422 	unsigned char reorder = 0;
423 	unsigned int prim = cmd_header->prim.prim;
424 	unsigned int skip = cmd_header->prim.skip;
425 	unsigned int n = cmd_header->prim.count;
426 	unsigned int start = cmd_header->prim.start;
427 	unsigned int vtx_size;
428 	unsigned int i;
429 	DMA_LOCALS;
430 
431 	if (!n)
432 		return 0;
433 
434 	switch (prim) {
435 	case SAVAGE_PRIM_TRILIST_201:
436 		reorder = 1;
437 		prim = SAVAGE_PRIM_TRILIST;
438 		/* FALLTHROUGH */
439 	case SAVAGE_PRIM_TRILIST:
440 		if (n % 3 != 0) {
441 			DRM_ERROR("wrong number of vertices %u in TRILIST\n",
442 				  n);
443 			return -EINVAL;
444 		}
445 		break;
446 	case SAVAGE_PRIM_TRISTRIP:
447 	case SAVAGE_PRIM_TRIFAN:
448 		if (n < 3) {
449 			DRM_ERROR
450 			    ("wrong number of vertices %u in TRIFAN/STRIP\n",
451 			     n);
452 			return -EINVAL;
453 		}
454 		break;
455 	default:
456 		DRM_ERROR("invalid primitive type %u\n", prim);
457 		return -EINVAL;
458 	}
459 
460 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
461 		if (skip > SAVAGE_SKIP_ALL_S3D) {
462 			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
463 			return -EINVAL;
464 		}
465 		vtx_size = 8; /* full vertex */
466 	} else {
467 		if (skip > SAVAGE_SKIP_ALL_S4) {
468 			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
469 			return -EINVAL;
470 		}
471 		vtx_size = 10; /* full vertex */
472 	}
473 
474 	vtx_size -= (skip & 1) + (skip >> 1 & 1) +
475 		(skip >> 2 & 1) + (skip >> 3 & 1) + (skip >> 4 & 1) +
476 		(skip >> 5 & 1) + (skip >> 6 & 1) + (skip >> 7 & 1);
477 
478 	if (vtx_size > vb_stride) {
479 		DRM_ERROR("vertex size greater than vb stride (%u > %u)\n",
480 			  vtx_size, vb_stride);
481 		return -EINVAL;
482 	}
483 
484 	if (start + n > vb_size / (vb_stride * 4)) {
485 		DRM_ERROR("vertex indices (%u-%u) out of range (0-%u)\n",
486 			  start, start + n - 1, vb_size / (vb_stride * 4));
487 		return -EINVAL;
488 	}
489 
490 	prim <<= 25;
491 	while (n != 0) {
492 		/* Can emit up to 255 vertices (85 triangles) at once. */
493 		unsigned int count = n > 255 ? 255 : n;
494 		if (reorder) {
495 			/* Need to reorder vertices for correct flat
496 			 * shading while preserving the clock sense
497 			 * for correct culling. Only on Savage3D. */
498 			int reorderarr[3] = { -1, -1, -1 };
499 			reorderarr[start % 3] = 2;
500 
501 			BEGIN_DMA(count * vtx_size + 1);
502 			DMA_DRAW_PRIMITIVE(count, prim, skip);
503 
504 			for (i = start; i < start + count; ++i) {
505 				unsigned int j = i + reorderarr[i % 3];
506 				DMA_COPY(&vtxbuf[vb_stride * j], vtx_size);
507 			}
508 
509 			DMA_COMMIT();
510 		} else {
511 			BEGIN_DMA(count * vtx_size + 1);
512 			DMA_DRAW_PRIMITIVE(count, prim, skip);
513 
514 			if (vb_stride == vtx_size) {
515 				DMA_COPY(&vtxbuf[vb_stride * start],
516 					 vtx_size * count);
517 			} else {
518 				for (i = start; i < start + count; ++i) {
519 					DMA_COPY(&vtxbuf[vb_stride * i],
520 						 vtx_size);
521 				}
522 			}
523 
524 			DMA_COMMIT();
525 		}
526 
527 		start += count;
528 		n -= count;
529 
530 		prim |= BCI_CMD_DRAW_CONT;
531 	}
532 
533 	return 0;
534 }
535 
savage_dispatch_dma_idx(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * cmd_header,const uint16_t * idx,const struct drm_buf * dmabuf)536 static int savage_dispatch_dma_idx(drm_savage_private_t *dev_priv,
537 				   const drm_savage_cmd_header_t *cmd_header,
538 				   const uint16_t *idx,
539 				   const struct drm_buf *dmabuf)
540 {
541 	unsigned char reorder = 0;
542 	unsigned int prim = cmd_header->idx.prim;
543 	unsigned int skip = cmd_header->idx.skip;
544 	unsigned int n = cmd_header->idx.count;
545 	unsigned int i;
546 	BCI_LOCALS;
547 
548 	if (!dmabuf) {
549 		DRM_ERROR("called without dma buffers!\n");
550 		return -EINVAL;
551 	}
552 
553 	if (!n)
554 		return 0;
555 
556 	switch (prim) {
557 	case SAVAGE_PRIM_TRILIST_201:
558 		reorder = 1;
559 		prim = SAVAGE_PRIM_TRILIST;
560 		/* FALLTHROUGH */
561 	case SAVAGE_PRIM_TRILIST:
562 		if (n % 3 != 0) {
563 			DRM_ERROR("wrong number of indices %u in TRILIST\n", n);
564 			return -EINVAL;
565 		}
566 		break;
567 	case SAVAGE_PRIM_TRISTRIP:
568 	case SAVAGE_PRIM_TRIFAN:
569 		if (n < 3) {
570 			DRM_ERROR
571 			    ("wrong number of indices %u in TRIFAN/STRIP\n", n);
572 			return -EINVAL;
573 		}
574 		break;
575 	default:
576 		DRM_ERROR("invalid primitive type %u\n", prim);
577 		return -EINVAL;
578 	}
579 
580 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
581 		if (skip != 0) {
582 			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
583 			return -EINVAL;
584 		}
585 	} else {
586 		unsigned int size = 10 - (skip & 1) - (skip >> 1 & 1) -
587 			(skip >> 2 & 1) - (skip >> 3 & 1) - (skip >> 4 & 1) -
588 			(skip >> 5 & 1) - (skip >> 6 & 1) - (skip >> 7 & 1);
589 		if (skip > SAVAGE_SKIP_ALL_S4 || size != 8) {
590 			DRM_ERROR("invalid skip flags 0x%04x for DMA\n", skip);
591 			return -EINVAL;
592 		}
593 		if (reorder) {
594 			DRM_ERROR("TRILIST_201 used on Savage4 hardware\n");
595 			return -EINVAL;
596 		}
597 	}
598 
599 	/* Vertex DMA doesn't work with command DMA at the same time,
600 	 * so we use BCI_... to submit commands here. Flush buffered
601 	 * faked DMA first. */
602 	DMA_FLUSH();
603 
604 	if (dmabuf->bus_address != dev_priv->state.common.vbaddr) {
605 		BEGIN_BCI(2);
606 		BCI_SET_REGISTERS(SAVAGE_VERTBUFADDR, 1);
607 		BCI_WRITE(dmabuf->bus_address | dev_priv->dma_type);
608 		dev_priv->state.common.vbaddr = dmabuf->bus_address;
609 	}
610 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset) && dev_priv->waiting) {
611 		/* Workaround for what looks like a hardware bug. If a
612 		 * WAIT_3D_IDLE was emitted some time before the
613 		 * indexed drawing command then the engine will lock
614 		 * up. There are two known workarounds:
615 		 * WAIT_IDLE_EMPTY or emit at least 63 NOPs. */
616 		BEGIN_BCI(63);
617 		for (i = 0; i < 63; ++i)
618 			BCI_WRITE(BCI_CMD_WAIT);
619 		dev_priv->waiting = 0;
620 	}
621 
622 	prim <<= 25;
623 	while (n != 0) {
624 		/* Can emit up to 255 indices (85 triangles) at once. */
625 		unsigned int count = n > 255 ? 255 : n;
626 
627 		/* check indices */
628 		for (i = 0; i < count; ++i) {
629 			if (idx[i] > dmabuf->total / 32) {
630 				DRM_ERROR("idx[%u]=%u out of range (0-%u)\n",
631 					  i, idx[i], dmabuf->total / 32);
632 				return -EINVAL;
633 			}
634 		}
635 
636 		if (reorder) {
637 			/* Need to reorder indices for correct flat
638 			 * shading while preserving the clock sense
639 			 * for correct culling. Only on Savage3D. */
640 			int reorderarr[3] = { 2, -1, -1 };
641 
642 			BEGIN_BCI((count + 1 + 1) / 2);
643 			BCI_DRAW_INDICES_S3D(count, prim, idx[2]);
644 
645 			for (i = 1; i + 1 < count; i += 2)
646 				BCI_WRITE(idx[i + reorderarr[i % 3]] |
647 					  (idx[i + 1 +
648 					   reorderarr[(i + 1) % 3]] << 16));
649 			if (i < count)
650 				BCI_WRITE(idx[i + reorderarr[i % 3]]);
651 		} else if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
652 			BEGIN_BCI((count + 1 + 1) / 2);
653 			BCI_DRAW_INDICES_S3D(count, prim, idx[0]);
654 
655 			for (i = 1; i + 1 < count; i += 2)
656 				BCI_WRITE(idx[i] | (idx[i + 1] << 16));
657 			if (i < count)
658 				BCI_WRITE(idx[i]);
659 		} else {
660 			BEGIN_BCI((count + 2 + 1) / 2);
661 			BCI_DRAW_INDICES_S4(count, prim, skip);
662 
663 			for (i = 0; i + 1 < count; i += 2)
664 				BCI_WRITE(idx[i] | (idx[i + 1] << 16));
665 			if (i < count)
666 				BCI_WRITE(idx[i]);
667 		}
668 
669 		idx += count;
670 		n -= count;
671 
672 		prim |= BCI_CMD_DRAW_CONT;
673 	}
674 
675 	return 0;
676 }
677 
savage_dispatch_vb_idx(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * cmd_header,const uint16_t * idx,const uint32_t * vtxbuf,unsigned int vb_size,unsigned int vb_stride)678 static int savage_dispatch_vb_idx(drm_savage_private_t *dev_priv,
679 				  const drm_savage_cmd_header_t *cmd_header,
680 				  const uint16_t *idx,
681 				  const uint32_t *vtxbuf,
682 				  unsigned int vb_size, unsigned int vb_stride)
683 {
684 	unsigned char reorder = 0;
685 	unsigned int prim = cmd_header->idx.prim;
686 	unsigned int skip = cmd_header->idx.skip;
687 	unsigned int n = cmd_header->idx.count;
688 	unsigned int vtx_size;
689 	unsigned int i;
690 	DMA_LOCALS;
691 
692 	if (!n)
693 		return 0;
694 
695 	switch (prim) {
696 	case SAVAGE_PRIM_TRILIST_201:
697 		reorder = 1;
698 		prim = SAVAGE_PRIM_TRILIST;
699 		/* FALLTHROUGH */
700 	case SAVAGE_PRIM_TRILIST:
701 		if (n % 3 != 0) {
702 			DRM_ERROR("wrong number of indices %u in TRILIST\n", n);
703 			return -EINVAL;
704 		}
705 		break;
706 	case SAVAGE_PRIM_TRISTRIP:
707 	case SAVAGE_PRIM_TRIFAN:
708 		if (n < 3) {
709 			DRM_ERROR
710 			    ("wrong number of indices %u in TRIFAN/STRIP\n", n);
711 			return -EINVAL;
712 		}
713 		break;
714 	default:
715 		DRM_ERROR("invalid primitive type %u\n", prim);
716 		return -EINVAL;
717 	}
718 
719 	if (S3_SAVAGE3D_SERIES(dev_priv->chipset)) {
720 		if (skip > SAVAGE_SKIP_ALL_S3D) {
721 			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
722 			return -EINVAL;
723 		}
724 		vtx_size = 8; /* full vertex */
725 	} else {
726 		if (skip > SAVAGE_SKIP_ALL_S4) {
727 			DRM_ERROR("invalid skip flags 0x%04x\n", skip);
728 			return -EINVAL;
729 		}
730 		vtx_size = 10; /* full vertex */
731 	}
732 
733 	vtx_size -= (skip & 1) + (skip >> 1 & 1) +
734 		(skip >> 2 & 1) + (skip >> 3 & 1) + (skip >> 4 & 1) +
735 		(skip >> 5 & 1) + (skip >> 6 & 1) + (skip >> 7 & 1);
736 
737 	if (vtx_size > vb_stride) {
738 		DRM_ERROR("vertex size greater than vb stride (%u > %u)\n",
739 			  vtx_size, vb_stride);
740 		return -EINVAL;
741 	}
742 
743 	prim <<= 25;
744 	while (n != 0) {
745 		/* Can emit up to 255 vertices (85 triangles) at once. */
746 		unsigned int count = n > 255 ? 255 : n;
747 
748 		/* Check indices */
749 		for (i = 0; i < count; ++i) {
750 			if (idx[i] > vb_size / (vb_stride * 4)) {
751 				DRM_ERROR("idx[%u]=%u out of range (0-%u)\n",
752 					  i, idx[i],  vb_size / (vb_stride * 4));
753 				return -EINVAL;
754 			}
755 		}
756 
757 		if (reorder) {
758 			/* Need to reorder vertices for correct flat
759 			 * shading while preserving the clock sense
760 			 * for correct culling. Only on Savage3D. */
761 			int reorderarr[3] = { 2, -1, -1 };
762 
763 			BEGIN_DMA(count * vtx_size + 1);
764 			DMA_DRAW_PRIMITIVE(count, prim, skip);
765 
766 			for (i = 0; i < count; ++i) {
767 				unsigned int j = idx[i + reorderarr[i % 3]];
768 				DMA_COPY(&vtxbuf[vb_stride * j], vtx_size);
769 			}
770 
771 			DMA_COMMIT();
772 		} else {
773 			BEGIN_DMA(count * vtx_size + 1);
774 			DMA_DRAW_PRIMITIVE(count, prim, skip);
775 
776 			for (i = 0; i < count; ++i) {
777 				unsigned int j = idx[i];
778 				DMA_COPY(&vtxbuf[vb_stride * j], vtx_size);
779 			}
780 
781 			DMA_COMMIT();
782 		}
783 
784 		idx += count;
785 		n -= count;
786 
787 		prim |= BCI_CMD_DRAW_CONT;
788 	}
789 
790 	return 0;
791 }
792 
savage_dispatch_clear(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * cmd_header,const drm_savage_cmd_header_t * data,unsigned int nbox,const struct drm_clip_rect * boxes)793 static int savage_dispatch_clear(drm_savage_private_t *dev_priv,
794 				 const drm_savage_cmd_header_t *cmd_header,
795 				 const drm_savage_cmd_header_t *data,
796 				 unsigned int nbox,
797 				 const struct drm_clip_rect *boxes)
798 {
799 	unsigned int flags = cmd_header->clear0.flags;
800 	unsigned int clear_cmd;
801 	unsigned int i, nbufs;
802 	DMA_LOCALS;
803 
804 	if (nbox == 0)
805 		return 0;
806 
807 	clear_cmd = BCI_CMD_RECT | BCI_CMD_RECT_XP | BCI_CMD_RECT_YP |
808 		BCI_CMD_SEND_COLOR | BCI_CMD_DEST_PBD_NEW;
809 	BCI_CMD_SET_ROP(clear_cmd,0xCC);
810 
811 	nbufs = ((flags & SAVAGE_FRONT) ? 1 : 0) +
812 	    ((flags & SAVAGE_BACK) ? 1 : 0) + ((flags & SAVAGE_DEPTH) ? 1 : 0);
813 	if (nbufs == 0)
814 		return 0;
815 
816 	if (data->clear1.mask != 0xffffffff) {
817 		/* set mask */
818 		BEGIN_DMA(2);
819 		DMA_SET_REGISTERS(SAVAGE_BITPLANEWTMASK, 1);
820 		DMA_WRITE(data->clear1.mask);
821 		DMA_COMMIT();
822 	}
823 	for (i = 0; i < nbox; ++i) {
824 		unsigned int x, y, w, h;
825 		unsigned int buf;
826 
827 		x = boxes[i].x1, y = boxes[i].y1;
828 		w = boxes[i].x2 - boxes[i].x1;
829 		h = boxes[i].y2 - boxes[i].y1;
830 		BEGIN_DMA(nbufs * 6);
831 		for (buf = SAVAGE_FRONT; buf <= SAVAGE_DEPTH; buf <<= 1) {
832 			if (!(flags & buf))
833 				continue;
834 			DMA_WRITE(clear_cmd);
835 			switch (buf) {
836 			case SAVAGE_FRONT:
837 				DMA_WRITE(dev_priv->front_offset);
838 				DMA_WRITE(dev_priv->front_bd);
839 				break;
840 			case SAVAGE_BACK:
841 				DMA_WRITE(dev_priv->back_offset);
842 				DMA_WRITE(dev_priv->back_bd);
843 				break;
844 			case SAVAGE_DEPTH:
845 				DMA_WRITE(dev_priv->depth_offset);
846 				DMA_WRITE(dev_priv->depth_bd);
847 				break;
848 			}
849 			DMA_WRITE(data->clear1.value);
850 			DMA_WRITE(BCI_X_Y(x, y));
851 			DMA_WRITE(BCI_W_H(w, h));
852 		}
853 		DMA_COMMIT();
854 	}
855 	if (data->clear1.mask != 0xffffffff) {
856 		/* reset mask */
857 		BEGIN_DMA(2);
858 		DMA_SET_REGISTERS(SAVAGE_BITPLANEWTMASK, 1);
859 		DMA_WRITE(0xffffffff);
860 		DMA_COMMIT();
861 	}
862 
863 	return 0;
864 }
865 
savage_dispatch_swap(drm_savage_private_t * dev_priv,unsigned int nbox,const struct drm_clip_rect * boxes)866 static int savage_dispatch_swap(drm_savage_private_t *dev_priv,
867 				unsigned int nbox, const struct drm_clip_rect *boxes)
868 {
869 	unsigned int swap_cmd;
870 	unsigned int i;
871 	DMA_LOCALS;
872 
873 	if (nbox == 0)
874 		return 0;
875 
876 	swap_cmd = BCI_CMD_RECT | BCI_CMD_RECT_XP | BCI_CMD_RECT_YP |
877 		BCI_CMD_SRC_PBD_COLOR_NEW | BCI_CMD_DEST_GBD;
878 	BCI_CMD_SET_ROP(swap_cmd,0xCC);
879 
880 	for (i = 0; i < nbox; ++i) {
881 		BEGIN_DMA(6);
882 		DMA_WRITE(swap_cmd);
883 		DMA_WRITE(dev_priv->back_offset);
884 		DMA_WRITE(dev_priv->back_bd);
885 		DMA_WRITE(BCI_X_Y(boxes[i].x1, boxes[i].y1));
886 		DMA_WRITE(BCI_X_Y(boxes[i].x1, boxes[i].y1));
887 		DMA_WRITE(BCI_W_H(boxes[i].x2 - boxes[i].x1,
888 				  boxes[i].y2 - boxes[i].y1));
889 		DMA_COMMIT();
890 	}
891 
892 	return 0;
893 }
894 
savage_dispatch_draw(drm_savage_private_t * dev_priv,const drm_savage_cmd_header_t * start,const drm_savage_cmd_header_t * end,const struct drm_buf * dmabuf,const unsigned int * vtxbuf,unsigned int vb_size,unsigned int vb_stride,unsigned int nbox,const struct drm_clip_rect * boxes)895 static int savage_dispatch_draw(drm_savage_private_t *dev_priv,
896 				const drm_savage_cmd_header_t *start,
897 				const drm_savage_cmd_header_t *end,
898 				const struct drm_buf *dmabuf,
899 				const unsigned int *vtxbuf,
900 				unsigned int vb_size, unsigned int vb_stride,
901 				unsigned int nbox,
902 				const struct drm_clip_rect *boxes)
903 {
904 	unsigned int i, j;
905 	int ret;
906 
907 	for (i = 0; i < nbox; ++i) {
908 		const drm_savage_cmd_header_t *cmdbuf;
909 		dev_priv->emit_clip_rect(dev_priv, &boxes[i]);
910 
911 		cmdbuf = start;
912 		while (cmdbuf < end) {
913 			drm_savage_cmd_header_t cmd_header;
914 			cmd_header = *cmdbuf;
915 			cmdbuf++;
916 			switch (cmd_header.cmd.cmd) {
917 			case SAVAGE_CMD_DMA_PRIM:
918 				ret = savage_dispatch_dma_prim(
919 					dev_priv, &cmd_header, dmabuf);
920 				break;
921 			case SAVAGE_CMD_VB_PRIM:
922 				ret = savage_dispatch_vb_prim(
923 					dev_priv, &cmd_header,
924 					vtxbuf, vb_size, vb_stride);
925 				break;
926 			case SAVAGE_CMD_DMA_IDX:
927 				j = (cmd_header.idx.count + 3) / 4;
928 				/* j was check in savage_bci_cmdbuf */
929 				ret = savage_dispatch_dma_idx(dev_priv,
930 					&cmd_header, (const uint16_t *)cmdbuf,
931 					dmabuf);
932 				cmdbuf += j;
933 				break;
934 			case SAVAGE_CMD_VB_IDX:
935 				j = (cmd_header.idx.count + 3) / 4;
936 				/* j was check in savage_bci_cmdbuf */
937 				ret = savage_dispatch_vb_idx(dev_priv,
938 					&cmd_header, (const uint16_t *)cmdbuf,
939 					(const uint32_t *)vtxbuf, vb_size,
940 					vb_stride);
941 				cmdbuf += j;
942 				break;
943 			default:
944 				/* What's the best return code? EFAULT? */
945 				DRM_ERROR("IMPLEMENTATION ERROR: "
946 					  "non-drawing-command %d\n",
947 					  cmd_header.cmd.cmd);
948 				return -EINVAL;
949 			}
950 
951 			if (ret != 0)
952 				return ret;
953 		}
954 	}
955 
956 	return 0;
957 }
958 
savage_bci_cmdbuf(struct drm_device * dev,void * data,struct drm_file * file_priv)959 int savage_bci_cmdbuf(struct drm_device *dev, void *data, struct drm_file *file_priv)
960 {
961 	drm_savage_private_t *dev_priv = dev->dev_private;
962 	struct drm_device_dma *dma = dev->dma;
963 	struct drm_buf *dmabuf;
964 	drm_savage_cmdbuf_t *cmdbuf = data;
965 	drm_savage_cmd_header_t *kcmd_addr = NULL;
966 	drm_savage_cmd_header_t *first_draw_cmd;
967 	unsigned int *kvb_addr = NULL;
968 	struct drm_clip_rect *kbox_addr = NULL;
969 	unsigned int i, j;
970 	int ret = 0;
971 
972 	DRM_DEBUG("\n");
973 
974 	LOCK_TEST_WITH_RETURN(dev, file_priv);
975 
976 	if (dma && dma->buflist) {
977 		if (cmdbuf->dma_idx > dma->buf_count) {
978 			DRM_ERROR
979 			    ("vertex buffer index %u out of range (0-%u)\n",
980 			     cmdbuf->dma_idx, dma->buf_count - 1);
981 			return -EINVAL;
982 		}
983 		dmabuf = dma->buflist[cmdbuf->dma_idx];
984 	} else {
985 		dmabuf = NULL;
986 	}
987 
988 	/* Copy the user buffers into kernel temporary areas.  This hasn't been
989 	 * a performance loss compared to VERIFYAREA_READ/
990 	 * COPY_FROM_USER_UNCHECKED when done in other drivers, and is correct
991 	 * for locking on FreeBSD.
992 	 */
993 	if (cmdbuf->size) {
994 		kcmd_addr = drm_alloc(cmdbuf->size * 8, DRM_MEM_DRIVER);
995 		if (kcmd_addr == NULL)
996 			return -ENOMEM;
997 
998 		if (DRM_COPY_FROM_USER(kcmd_addr, cmdbuf->cmd_addr,
999 				       cmdbuf->size * 8))
1000 		{
1001 			drm_free(kcmd_addr, cmdbuf->size * 8, DRM_MEM_DRIVER);
1002 			return -EFAULT;
1003 		}
1004 		cmdbuf->cmd_addr = kcmd_addr;
1005 	}
1006 	if (cmdbuf->vb_size) {
1007 		kvb_addr = drm_alloc(cmdbuf->vb_size, DRM_MEM_DRIVER);
1008 		if (kvb_addr == NULL) {
1009 			ret = -ENOMEM;
1010 			goto done;
1011 		}
1012 
1013 		if (DRM_COPY_FROM_USER(kvb_addr, cmdbuf->vb_addr,
1014 				       cmdbuf->vb_size)) {
1015 			ret = -EFAULT;
1016 			goto done;
1017 		}
1018 		cmdbuf->vb_addr = kvb_addr;
1019 	}
1020 	if (cmdbuf->nbox) {
1021 		kbox_addr = drm_alloc(cmdbuf->nbox *
1022 				      sizeof(struct drm_clip_rect),
1023 				      DRM_MEM_DRIVER);
1024 		if (kbox_addr == NULL) {
1025 			ret = -ENOMEM;
1026 			goto done;
1027 		}
1028 
1029 		if (DRM_COPY_FROM_USER(kbox_addr, cmdbuf->box_addr,
1030 				       cmdbuf->nbox *
1031 				       sizeof(struct drm_clip_rect))) {
1032 			ret = -EFAULT;
1033 			goto done;
1034 		}
1035 		cmdbuf->box_addr = kbox_addr;
1036 	}
1037 
1038 	/* Make sure writes to DMA buffers are finished before sending
1039 	 * DMA commands to the graphics hardware. */
1040 	DRM_MEMORYBARRIER();
1041 
1042 	/* Coming from user space. Don't know if the Xserver has
1043 	 * emitted wait commands. Assuming the worst. */
1044 	dev_priv->waiting = 1;
1045 
1046 	i = 0;
1047 	first_draw_cmd = NULL;
1048 	while (i < cmdbuf->size) {
1049 		drm_savage_cmd_header_t cmd_header;
1050 		cmd_header = *(drm_savage_cmd_header_t *)cmdbuf->cmd_addr;
1051 		cmdbuf->cmd_addr++;
1052 		i++;
1053 
1054 		/* Group drawing commands with same state to minimize
1055 		 * iterations over clip rects. */
1056 		j = 0;
1057 		switch (cmd_header.cmd.cmd) {
1058 		case SAVAGE_CMD_DMA_IDX:
1059 		case SAVAGE_CMD_VB_IDX:
1060 			j = (cmd_header.idx.count + 3) / 4;
1061 			if (i + j > cmdbuf->size) {
1062 				DRM_ERROR("indexed drawing command extends "
1063 					  "beyond end of command buffer\n");
1064 				DMA_FLUSH();
1065 				return -EINVAL;
1066 			}
1067 			/* FALLTHROUGH */
1068 		case SAVAGE_CMD_DMA_PRIM:
1069 		case SAVAGE_CMD_VB_PRIM:
1070 			if (!first_draw_cmd)
1071 				first_draw_cmd = cmdbuf->cmd_addr - 1;
1072 			cmdbuf->cmd_addr += j;
1073 			i += j;
1074 			break;
1075 		default:
1076 			if (first_draw_cmd) {
1077 				ret = savage_dispatch_draw(
1078 					dev_priv, first_draw_cmd,
1079 					cmdbuf->cmd_addr - 1,
1080 					dmabuf, cmdbuf->vb_addr,
1081 					cmdbuf->vb_size,
1082 					cmdbuf->vb_stride,
1083 					cmdbuf->nbox, cmdbuf->box_addr);
1084 				if (ret != 0)
1085 					return ret;
1086 				first_draw_cmd = NULL;
1087 			}
1088 		}
1089 		if (first_draw_cmd)
1090 			continue;
1091 
1092 		switch (cmd_header.cmd.cmd) {
1093 		case SAVAGE_CMD_STATE:
1094 			j = (cmd_header.state.count + 1) / 2;
1095 			if (i + j > cmdbuf->size) {
1096 				DRM_ERROR("command SAVAGE_CMD_STATE extends "
1097 					  "beyond end of command buffer\n");
1098 				DMA_FLUSH();
1099 				ret = -EINVAL;
1100 				goto done;
1101 			}
1102 			ret = savage_dispatch_state(dev_priv, &cmd_header,
1103 				(const uint32_t *)cmdbuf->cmd_addr);
1104 			cmdbuf->cmd_addr += j;
1105 			i += j;
1106 			break;
1107 		case SAVAGE_CMD_CLEAR:
1108 			if (i + 1 > cmdbuf->size) {
1109 				DRM_ERROR("command SAVAGE_CMD_CLEAR extends "
1110 					  "beyond end of command buffer\n");
1111 				DMA_FLUSH();
1112 				ret = -EINVAL;
1113 				goto done;
1114 			}
1115 			ret = savage_dispatch_clear(dev_priv, &cmd_header,
1116 						    cmdbuf->cmd_addr,
1117 						    cmdbuf->nbox,
1118 						    cmdbuf->box_addr);
1119 			cmdbuf->cmd_addr++;
1120 			i++;
1121 			break;
1122 		case SAVAGE_CMD_SWAP:
1123 			ret = savage_dispatch_swap(dev_priv, cmdbuf->nbox,
1124 						   cmdbuf->box_addr);
1125 			break;
1126 		default:
1127 			DRM_ERROR("invalid command 0x%x\n",
1128 				  cmd_header.cmd.cmd);
1129 			DMA_FLUSH();
1130 			ret = -EINVAL;
1131 			goto done;
1132 		}
1133 
1134 		if (ret != 0) {
1135 			DMA_FLUSH();
1136 			goto done;
1137 		}
1138 	}
1139 
1140 	if (first_draw_cmd) {
1141 		ret = savage_dispatch_draw(
1142 			dev_priv, first_draw_cmd, cmdbuf->cmd_addr, dmabuf,
1143 			cmdbuf->vb_addr, cmdbuf->vb_size, cmdbuf->vb_stride,
1144 			cmdbuf->nbox, cmdbuf->box_addr);
1145 		if (ret != 0) {
1146 			DMA_FLUSH();
1147 			goto done;
1148 		}
1149 	}
1150 
1151 	DMA_FLUSH();
1152 
1153 	if (dmabuf && cmdbuf->discard) {
1154 		drm_savage_buf_priv_t *buf_priv = dmabuf->dev_private;
1155 		uint16_t event;
1156 		event = savage_bci_emit_event(dev_priv, SAVAGE_WAIT_3D);
1157 		SET_AGE(&buf_priv->age, event, dev_priv->event_wrap);
1158 		savage_freelist_put(dev, dmabuf);
1159 	}
1160 
1161 done:
1162 	/* If we didn't need to allocate them, these'll be NULL */
1163 	drm_free(kcmd_addr, cmdbuf->size * 8, DRM_MEM_DRIVER);
1164 	drm_free(kvb_addr, cmdbuf->vb_size, DRM_MEM_DRIVER);
1165 	drm_free(kbox_addr, cmdbuf->nbox * sizeof(struct drm_clip_rect),
1166 		 DRM_MEM_DRIVER);
1167 
1168 	return ret;
1169 }
1170