xref: /titanic_44/usr/src/uts/intel/io/drm/radeon_state.c (revision 4a6ec905b96eb96a398c346f59e034a90ce8ad37)
1 
2 /*
3  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
4  * Use is subject to license terms.
5  */
6 /* radeon_state.c -- State support for Radeon -*- linux-c -*- */
7 /*
8  * Copyright 2000 VA Linux Systems, Inc., Fremont, California.
9  * All Rights Reserved.
10  *
11  * Permission is hereby granted, free of charge, to any person obtaining a
12  * copy of this software and associated documentation files (the "Software"),
13  * to deal in the Software without restriction, including without limitation
14  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
15  * and/or sell copies of the Software, and to permit persons to whom the
16  * Software is furnished to do so, subject to the following conditions:
17  *
18  * The above copyright notice and this permission notice (including the next
19  * paragraph) shall be included in all copies or substantial portions of the
20  * Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
25  * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
26  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
27  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
28  * DEALINGS IN THE SOFTWARE.
29  *
30  * Authors:
31  *    Gareth Hughes <gareth@valinux.com>
32  *    Kevin E. Martin <martin@valinux.com>
33  */
34 
35 #pragma ident	"%Z%%M%	%I%	%E% SMI"
36 
37 #include "drmP.h"
38 #include "drm.h"
39 #include "drm_sarea.h"
40 #include "radeon_drm.h"
41 #include "radeon_drv.h"
42 #include "radeon_io32.h"
43 
44 /*
45  * Helper functions for client state checking and fixup
46  */
47 
48 static inline int
49 radeon_check_and_fixup_offset(drm_radeon_private_t *dev_priv,
50     drm_file_t *filp_priv, u32 *offset)
51 {
52 	u64 off = *offset;
53 	u32 fb_end = dev_priv->fb_location + dev_priv->fb_size - 1;
54 	struct drm_radeon_driver_file_fields *radeon_priv;
55 
56 	/*
57 	 * Hrm ... the story of the offset ... So this function converts
58 	 * the various ideas of what userland clients might have for an
59 	 * offset in the card address space into an offset into the card
60 	 * address space :) So with a sane client, it should just keep
61 	 * the value intact and just do some boundary checking. However,
62 	 * not all clients are sane. Some older clients pass us 0 based
63 	 * offsets relative to the start of the framebuffer and some may
64 	 * assume the AGP aperture it appended to the framebuffer, so we
65 	 * try to detect those cases and fix them up.
66 	 *
67 	 * Note: It might be a good idea here to make sure the offset lands
68 	 * in some "allowed" area to protect things like the PCIE GART...
69 	 */
70 
71 	/*
72 	 * First, the best case, the offset already lands in either the
73 	 * framebuffer or the GART mapped space
74 	 */
75 
76 	if (radeon_check_offset(dev_priv, off))
77 		return (0);
78 
79 	/*
80 	 * Ok, that didn't happen... now check if we have a zero based
81 	 * offset that fits in the framebuffer + gart space, apply the
82 	 * magic offset we get from SETPARAM or calculated from fb_location
83 	 */
84 	if (off < (dev_priv->fb_size + dev_priv->gart_size)) {
85 		radeon_priv = filp_priv->driver_priv;
86 		off += radeon_priv->radeon_fb_delta;
87 	}
88 
89 	/* Finally, assume we aimed at a GART offset if beyond the fb */
90 	if (off > fb_end)
91 		off = off - fb_end - 1 + dev_priv->gart_vm_start;
92 
93 	/* Now recheck and fail if out of bounds */
94 	if (radeon_check_offset(dev_priv, off)) {
95 		DRM_DEBUG("offset fixed up to 0x%x\n", off);
96 		*offset = (uint32_t)off;
97 		return (0);
98 	}
99 	return (EINVAL);
100 }
101 
102 static inline int
103 radeon_check_and_fixup_packets(drm_radeon_private_t *dev_priv,
104     drm_file_t *filp_priv, int id, u32 *data)
105 {
106 	switch (id) {
107 
108 	case RADEON_EMIT_PP_MISC:
109 		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
110 		    &data[(RADEON_RB3D_DEPTHOFFSET - RADEON_PP_MISC) / 4])) {
111 			DRM_ERROR("Invalid depth buffer offset\n");
112 			return (EINVAL);
113 		}
114 		break;
115 
116 	case RADEON_EMIT_PP_CNTL:
117 		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
118 		    &data[(RADEON_RB3D_COLOROFFSET - RADEON_PP_CNTL) / 4])) {
119 			DRM_ERROR("Invalid colour buffer offset\n");
120 			return (EINVAL);
121 		}
122 		break;
123 
124 	case R200_EMIT_PP_TXOFFSET_0:
125 	case R200_EMIT_PP_TXOFFSET_1:
126 	case R200_EMIT_PP_TXOFFSET_2:
127 	case R200_EMIT_PP_TXOFFSET_3:
128 	case R200_EMIT_PP_TXOFFSET_4:
129 	case R200_EMIT_PP_TXOFFSET_5:
130 		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
131 		    &data[0])) {
132 			DRM_ERROR("Invalid R200 texture offset\n");
133 			return (EINVAL);
134 		}
135 		break;
136 
137 	case RADEON_EMIT_PP_TXFILTER_0:
138 	case RADEON_EMIT_PP_TXFILTER_1:
139 	case RADEON_EMIT_PP_TXFILTER_2:
140 		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
141 		    &data[(RADEON_PP_TXOFFSET_0 - RADEON_PP_TXFILTER_0) / 4])) {
142 			DRM_ERROR("Invalid R100 texture offset\n");
143 			return (EINVAL);
144 		}
145 		break;
146 
147 	case R200_EMIT_PP_CUBIC_OFFSETS_0:
148 	case R200_EMIT_PP_CUBIC_OFFSETS_1:
149 	case R200_EMIT_PP_CUBIC_OFFSETS_2:
150 	case R200_EMIT_PP_CUBIC_OFFSETS_3:
151 	case R200_EMIT_PP_CUBIC_OFFSETS_4:
152 	case R200_EMIT_PP_CUBIC_OFFSETS_5: {
153 			int i;
154 			for (i = 0; i < 5; i++) {
155 				if (radeon_check_and_fixup_offset(dev_priv,
156 				    filp_priv, &data[i])) {
157 					DRM_ERROR("Invalid R200 cubic"
158 					    " texture offset\n");
159 					return (EINVAL);
160 				}
161 			}
162 			break;
163 		}
164 
165 	case RADEON_EMIT_PP_CUBIC_OFFSETS_T0:
166 	case RADEON_EMIT_PP_CUBIC_OFFSETS_T1:
167 	case RADEON_EMIT_PP_CUBIC_OFFSETS_T2:
168 	{
169 			int i;
170 			for (i = 0; i < 5; i++) {
171 				if (radeon_check_and_fixup_offset(dev_priv,
172 				    filp_priv, &data[i])) {
173 					DRM_ERROR("Invalid R100 cubic"
174 					    " texture offset\n");
175 					return (EINVAL);
176 				}
177 			}
178 	}
179 		break;
180 
181 	case R200_EMIT_VAP_CTL:
182 	{
183 		RING_LOCALS;
184 		BEGIN_RING(2);
185 		OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
186 		ADVANCE_RING();
187 	}
188 		break;
189 
190 	case RADEON_EMIT_RB3D_COLORPITCH:
191 	case RADEON_EMIT_RE_LINE_PATTERN:
192 	case RADEON_EMIT_SE_LINE_WIDTH:
193 	case RADEON_EMIT_PP_LUM_MATRIX:
194 	case RADEON_EMIT_PP_ROT_MATRIX_0:
195 	case RADEON_EMIT_RB3D_STENCILREFMASK:
196 	case RADEON_EMIT_SE_VPORT_XSCALE:
197 	case RADEON_EMIT_SE_CNTL:
198 	case RADEON_EMIT_SE_CNTL_STATUS:
199 	case RADEON_EMIT_RE_MISC:
200 	case RADEON_EMIT_PP_BORDER_COLOR_0:
201 	case RADEON_EMIT_PP_BORDER_COLOR_1:
202 	case RADEON_EMIT_PP_BORDER_COLOR_2:
203 	case RADEON_EMIT_SE_ZBIAS_FACTOR:
204 	case RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT:
205 	case RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED:
206 	case R200_EMIT_PP_TXCBLEND_0:
207 	case R200_EMIT_PP_TXCBLEND_1:
208 	case R200_EMIT_PP_TXCBLEND_2:
209 	case R200_EMIT_PP_TXCBLEND_3:
210 	case R200_EMIT_PP_TXCBLEND_4:
211 	case R200_EMIT_PP_TXCBLEND_5:
212 	case R200_EMIT_PP_TXCBLEND_6:
213 	case R200_EMIT_PP_TXCBLEND_7:
214 	case R200_EMIT_TCL_LIGHT_MODEL_CTL_0:
215 	case R200_EMIT_TFACTOR_0:
216 	case R200_EMIT_VTX_FMT_0:
217 	case R200_EMIT_MATRIX_SELECT_0:
218 	case R200_EMIT_TEX_PROC_CTL_2:
219 	case R200_EMIT_TCL_UCP_VERT_BLEND_CTL:
220 	case R200_EMIT_PP_TXFILTER_0:
221 	case R200_EMIT_PP_TXFILTER_1:
222 	case R200_EMIT_PP_TXFILTER_2:
223 	case R200_EMIT_PP_TXFILTER_3:
224 	case R200_EMIT_PP_TXFILTER_4:
225 	case R200_EMIT_PP_TXFILTER_5:
226 	case R200_EMIT_VTE_CNTL:
227 	case R200_EMIT_OUTPUT_VTX_COMP_SEL:
228 	case R200_EMIT_PP_TAM_DEBUG3:
229 	case R200_EMIT_PP_CNTL_X:
230 	case R200_EMIT_RB3D_DEPTHXY_OFFSET:
231 	case R200_EMIT_RE_AUX_SCISSOR_CNTL:
232 	case R200_EMIT_RE_SCISSOR_TL_0:
233 	case R200_EMIT_RE_SCISSOR_TL_1:
234 	case R200_EMIT_RE_SCISSOR_TL_2:
235 	case R200_EMIT_SE_VAP_CNTL_STATUS:
236 	case R200_EMIT_SE_VTX_STATE_CNTL:
237 	case R200_EMIT_RE_POINTSIZE:
238 	case R200_EMIT_TCL_INPUT_VTX_VECTOR_ADDR_0:
239 	case R200_EMIT_PP_CUBIC_FACES_0:
240 	case R200_EMIT_PP_CUBIC_FACES_1:
241 	case R200_EMIT_PP_CUBIC_FACES_2:
242 	case R200_EMIT_PP_CUBIC_FACES_3:
243 	case R200_EMIT_PP_CUBIC_FACES_4:
244 	case R200_EMIT_PP_CUBIC_FACES_5:
245 	case RADEON_EMIT_PP_TEX_SIZE_0:
246 	case RADEON_EMIT_PP_TEX_SIZE_1:
247 	case RADEON_EMIT_PP_TEX_SIZE_2:
248 	case R200_EMIT_RB3D_BLENDCOLOR:
249 	case R200_EMIT_TCL_POINT_SPRITE_CNTL:
250 	case RADEON_EMIT_PP_CUBIC_FACES_0:
251 	case RADEON_EMIT_PP_CUBIC_FACES_1:
252 	case RADEON_EMIT_PP_CUBIC_FACES_2:
253 	case R200_EMIT_PP_TRI_PERF_CNTL:
254 	case R200_EMIT_PP_AFS_0:
255 	case R200_EMIT_PP_AFS_1:
256 	case R200_EMIT_ATF_TFACTOR:
257 	case R200_EMIT_PP_TXCTLALL_0:
258 	case R200_EMIT_PP_TXCTLALL_1:
259 	case R200_EMIT_PP_TXCTLALL_2:
260 	case R200_EMIT_PP_TXCTLALL_3:
261 	case R200_EMIT_PP_TXCTLALL_4:
262 	case R200_EMIT_PP_TXCTLALL_5:
263 	case R200_EMIT_VAP_PVS_CNTL:
264 		/* These packets don't contain memory offsets */
265 		break;
266 
267 	default:
268 		DRM_ERROR("Unknown state packet ID %d\n", id);
269 		return (EINVAL);
270 	}
271 
272 	return (0);
273 }
274 
275 static inline int
276 radeon_check_and_fixup_packet3(drm_radeon_private_t *dev_priv,
277     drm_file_t *filp_priv, drm_radeon_kcmd_buffer_t *cmdbuf,
278     unsigned int *cmdsz)
279 {
280 	u32 *cmd = (u32 *)(uintptr_t)cmdbuf->buf;
281 	u32 offset, narrays;
282 	int count, i, k;
283 
284 	*cmdsz = 2 + ((cmd[0] & RADEON_CP_PACKET_COUNT_MASK) >> 16);
285 
286 	if ((cmd[0] & 0xc0000000) != RADEON_CP_PACKET3) {
287 		DRM_ERROR("Not a type 3 packet\n");
288 		return (EINVAL);
289 	}
290 
291 	if (4 * *cmdsz > cmdbuf->bufsz) {
292 		DRM_ERROR("Packet size larger than size of data provided\n");
293 		return (EINVAL);
294 	}
295 
296 
297 	switch (cmd[0] & 0xff00) {
298 	/* XXX Are there old drivers needing other packets? */
299 
300 	case RADEON_3D_DRAW_IMMD:
301 	case RADEON_3D_DRAW_VBUF:
302 	case RADEON_3D_DRAW_INDX:
303 	case RADEON_WAIT_FOR_IDLE:
304 	case RADEON_CP_NOP:
305 	case RADEON_3D_CLEAR_ZMASK:
306 #if 0
307 	case RADEON_CP_NEXT_CHAR:
308 	case RADEON_CP_PLY_NEXTSCAN:
309 	case RADEON_CP_SET_SCISSORS:
310 	/* probably safe but will never need them? */
311 #endif
312 /* these packets are safe */
313 		break;
314 
315 	case RADEON_CP_3D_DRAW_IMMD_2:
316 	case RADEON_CP_3D_DRAW_VBUF_2:
317 	case RADEON_CP_3D_DRAW_INDX_2:
318 	case RADEON_3D_CLEAR_HIZ:
319 		/* safe but r200 only */
320 		if (dev_priv->microcode_version != UCODE_R200) {
321 			DRM_ERROR("Invalid 3d packet for r100-class chip\n");
322 			return (EINVAL);
323 		}
324 		break;
325 
326 	case RADEON_3D_LOAD_VBPNTR:
327 		count = (cmd[0] >> 16) & 0x3fff;
328 
329 		if (count > 18) { /* 12 arrays max */
330 			DRM_ERROR(
331 			    "Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
332 			    count);
333 			return (EINVAL);
334 		}
335 
336 		/* carefully check packet contents */
337 		narrays = cmd[1] & ~0xc000;
338 		k = 0;
339 		i = 2;
340 		while ((k < narrays) && (i < (count + 2))) {
341 			i++;		/* skip attribute field */
342 			if (radeon_check_and_fixup_offset(dev_priv,
343 			    filp_priv, &cmd[i])) {
344 				DRM_ERROR(
345 				    "Invalid offset (k=%d i=%d) ini"
346 				    " 3D_LOAD_VBPNTR packet.\n", k, i);
347 				return (EINVAL);
348 			}
349 			k++;
350 			i++;
351 			if (k == narrays)
352 				break;
353 			/* have one more to process, they come in pairs */
354 			if (radeon_check_and_fixup_offset(dev_priv,
355 			    filp_priv, &cmd[i])) {
356 				DRM_ERROR(
357 				    "Invalid offset (k=%d i=%d) in"
358 				    " 3D_LOAD_VBPNTR packet.\n", k, i);
359 				return (EINVAL);
360 			}
361 			k++;
362 			i++;
363 		}
364 		/* do the counts match what we expect ? */
365 		if ((k != narrays) || (i != (count + 2))) {
366 			DRM_ERROR(
367 			    "Malformed 3D_LOAD_VBPNTR packet"
368 			    "(k=%d i=%d narrays=%d count+1=%d).\n",
369 			    k, i, narrays, count + 1);
370 			return (EINVAL);
371 		}
372 		break;
373 
374 	case RADEON_3D_RNDR_GEN_INDX_PRIM:
375 		if (dev_priv->microcode_version != UCODE_R100) {
376 			DRM_ERROR("Invalid 3d packet for r200-class chip\n");
377 			return (EINVAL);
378 		}
379 		if (radeon_check_and_fixup_offset(dev_priv,
380 		    filp_priv, &cmd[1])) {
381 				DRM_ERROR("Invalid rndr_gen_indx offset\n");
382 				return (EINVAL);
383 		}
384 		break;
385 
386 	case RADEON_CP_INDX_BUFFER:
387 		if (dev_priv->microcode_version != UCODE_R200) {
388 			DRM_ERROR("Invalid 3d packet for r100-class chip\n");
389 			return (EINVAL);
390 		}
391 		if ((cmd[1] & 0x8000ffff) != 0x80000810) {
392 			DRM_ERROR(
393 			    "Invalid indx_buffer reg address %08X\n", cmd[1]);
394 			return (EINVAL);
395 		}
396 		if (radeon_check_and_fixup_offset(dev_priv,
397 		    filp_priv, &cmd[2])) {
398 			DRM_ERROR(
399 			    "Invalid indx_buffer offset is %08X\n", cmd[2]);
400 			return (EINVAL);
401 		}
402 		break;
403 
404 	case RADEON_CNTL_HOSTDATA_BLT:
405 	case RADEON_CNTL_PAINT_MULTI:
406 	case RADEON_CNTL_BITBLT_MULTI:
407 		/* MSB of opcode: next DWORD GUI_CNTL */
408 		if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
409 		    RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
410 			offset = cmd[2] << 10;
411 			if (radeon_check_and_fixup_offset
412 			    (dev_priv, filp_priv, &offset)) {
413 				DRM_ERROR("Invalid first packet offset\n");
414 				return (EINVAL);
415 			}
416 			cmd[2] = (cmd[2] & 0xffc00000) | offset >> 10;
417 		}
418 
419 		if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
420 		    (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
421 			offset = cmd[3] << 10;
422 			if (radeon_check_and_fixup_offset
423 			    (dev_priv, filp_priv, &offset)) {
424 				DRM_ERROR("Invalid second packet offset\n");
425 				return (EINVAL);
426 			}
427 			cmd[3] = (cmd[3] & 0xffc00000) | offset >> 10;
428 		}
429 		break;
430 
431 	default:
432 		DRM_ERROR("Invalid packet type %x\n", cmd[0] & 0xff00);
433 		return (EINVAL);
434 	}
435 
436 	return (0);
437 }
438 
439 /*
440  * CP hardware state programming functions
441  */
442 
443 static inline void radeon_emit_clip_rect(drm_radeon_private_t *dev_priv,
444     drm_clip_rect_t *box)
445 {
446 	RING_LOCALS;
447 
448 	DRM_DEBUG("   box:  x1=%d y1=%d  x2=%d y2=%d\n",
449 	    box->x1, box->y1, box->x2, box->y2);
450 
451 	BEGIN_RING(4);
452 	OUT_RING(CP_PACKET0(RADEON_RE_TOP_LEFT, 0));
453 	OUT_RING((box->y1 << 16) | box->x1);
454 	OUT_RING(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0));
455 	OUT_RING(((box->y2 - 1) << 16) | (box->x2 - 1));
456 	ADVANCE_RING();
457 }
458 
459 /* Emit 1.1 state */
460 static int radeon_emit_state(drm_radeon_private_t *dev_priv,
461     drm_file_t *filp_priv, drm_radeon_context_regs_t *ctx,
462     drm_radeon_texture_regs_t *tex, unsigned int dirty)
463 {
464 	RING_LOCALS;
465 	DRM_DEBUG("dirty=0x%08x\n", dirty);
466 
467 	if (dirty & RADEON_UPLOAD_CONTEXT) {
468 		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
469 		    &ctx->rb3d_depthoffset)) {
470 			DRM_ERROR("Invalid depth buffer offset\n");
471 			return (EINVAL);
472 		}
473 
474 		if (radeon_check_and_fixup_offset(dev_priv, filp_priv,
475 		    &ctx->rb3d_coloroffset)) {
476 			DRM_ERROR("Invalid depth buffer offset\n");
477 			return (EINVAL);
478 		}
479 
480 		BEGIN_RING(14);
481 		OUT_RING(CP_PACKET0(RADEON_PP_MISC, 6));
482 		OUT_RING(ctx->pp_misc);
483 		OUT_RING(ctx->pp_fog_color);
484 		OUT_RING(ctx->re_solid_color);
485 		OUT_RING(ctx->rb3d_blendcntl);
486 		OUT_RING(ctx->rb3d_depthoffset);
487 		OUT_RING(ctx->rb3d_depthpitch);
488 		OUT_RING(ctx->rb3d_zstencilcntl);
489 		OUT_RING(CP_PACKET0(RADEON_PP_CNTL, 2));
490 		OUT_RING(ctx->pp_cntl);
491 		OUT_RING(ctx->rb3d_cntl);
492 		OUT_RING(ctx->rb3d_coloroffset);
493 		OUT_RING(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0));
494 		OUT_RING(ctx->rb3d_colorpitch);
495 		ADVANCE_RING();
496 	}
497 
498 	if (dirty & RADEON_UPLOAD_VERTFMT) {
499 		BEGIN_RING(2);
500 		OUT_RING(CP_PACKET0(RADEON_SE_COORD_FMT, 0));
501 		OUT_RING(ctx->se_coord_fmt);
502 		ADVANCE_RING();
503 	}
504 
505 	if (dirty & RADEON_UPLOAD_LINE) {
506 		BEGIN_RING(5);
507 		OUT_RING(CP_PACKET0(RADEON_RE_LINE_PATTERN, 1));
508 		OUT_RING(ctx->re_line_pattern);
509 		OUT_RING(ctx->re_line_state);
510 		OUT_RING(CP_PACKET0(RADEON_SE_LINE_WIDTH, 0));
511 		OUT_RING(ctx->se_line_width);
512 		ADVANCE_RING();
513 	}
514 
515 	if (dirty & RADEON_UPLOAD_BUMPMAP) {
516 		BEGIN_RING(5);
517 		OUT_RING(CP_PACKET0(RADEON_PP_LUM_MATRIX, 0));
518 		OUT_RING(ctx->pp_lum_matrix);
519 		OUT_RING(CP_PACKET0(RADEON_PP_ROT_MATRIX_0, 1));
520 		OUT_RING(ctx->pp_rot_matrix_0);
521 		OUT_RING(ctx->pp_rot_matrix_1);
522 		ADVANCE_RING();
523 	}
524 
525 	if (dirty & RADEON_UPLOAD_MASKS) {
526 		BEGIN_RING(4);
527 		OUT_RING(CP_PACKET0(RADEON_RB3D_STENCILREFMASK, 2));
528 		OUT_RING(ctx->rb3d_stencilrefmask);
529 		OUT_RING(ctx->rb3d_ropcntl);
530 		OUT_RING(ctx->rb3d_planemask);
531 		ADVANCE_RING();
532 	}
533 
534 	if (dirty & RADEON_UPLOAD_VIEWPORT) {
535 		BEGIN_RING(7);
536 		OUT_RING(CP_PACKET0(RADEON_SE_VPORT_XSCALE, 5));
537 		OUT_RING(ctx->se_vport_xscale);
538 		OUT_RING(ctx->se_vport_xoffset);
539 		OUT_RING(ctx->se_vport_yscale);
540 		OUT_RING(ctx->se_vport_yoffset);
541 		OUT_RING(ctx->se_vport_zscale);
542 		OUT_RING(ctx->se_vport_zoffset);
543 		ADVANCE_RING();
544 	}
545 
546 	if (dirty & RADEON_UPLOAD_SETUP) {
547 		BEGIN_RING(4);
548 		OUT_RING(CP_PACKET0(RADEON_SE_CNTL, 0));
549 		OUT_RING(ctx->se_cntl);
550 		OUT_RING(CP_PACKET0(RADEON_SE_CNTL_STATUS, 0));
551 		OUT_RING(ctx->se_cntl_status);
552 		ADVANCE_RING();
553 	}
554 
555 	if (dirty & RADEON_UPLOAD_MISC) {
556 		BEGIN_RING(2);
557 		OUT_RING(CP_PACKET0(RADEON_RE_MISC, 0));
558 		OUT_RING(ctx->re_misc);
559 		ADVANCE_RING();
560 	}
561 
562 	if (dirty & RADEON_UPLOAD_TEX0) {
563 		if (radeon_check_and_fixup_offset(dev_priv,
564 		    filp_priv, &tex[0].pp_txoffset)) {
565 			DRM_ERROR("Invalid texture offset for unit 0\n");
566 			return (EINVAL);
567 		}
568 
569 		BEGIN_RING(9);
570 		OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_0, 5));
571 		OUT_RING(tex[0].pp_txfilter);
572 		OUT_RING(tex[0].pp_txformat);
573 		OUT_RING(tex[0].pp_txoffset);
574 		OUT_RING(tex[0].pp_txcblend);
575 		OUT_RING(tex[0].pp_txablend);
576 		OUT_RING(tex[0].pp_tfactor);
577 		OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_0, 0));
578 		OUT_RING(tex[0].pp_border_color);
579 		ADVANCE_RING();
580 	}
581 
582 	if (dirty & RADEON_UPLOAD_TEX1) {
583 		if (radeon_check_and_fixup_offset(dev_priv,
584 		    filp_priv, &tex[1].pp_txoffset)) {
585 			DRM_ERROR("Invalid texture offset for unit 1\n");
586 			return (EINVAL);
587 		}
588 
589 		BEGIN_RING(9);
590 		OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_1, 5));
591 		OUT_RING(tex[1].pp_txfilter);
592 		OUT_RING(tex[1].pp_txformat);
593 		OUT_RING(tex[1].pp_txoffset);
594 		OUT_RING(tex[1].pp_txcblend);
595 		OUT_RING(tex[1].pp_txablend);
596 		OUT_RING(tex[1].pp_tfactor);
597 		OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_1, 0));
598 		OUT_RING(tex[1].pp_border_color);
599 		ADVANCE_RING();
600 	}
601 
602 	if (dirty & RADEON_UPLOAD_TEX2) {
603 		if (radeon_check_and_fixup_offset(dev_priv,
604 		    filp_priv, &tex[2].pp_txoffset)) {
605 			DRM_ERROR("Invalid texture offset for unit 2\n");
606 			return (EINVAL);
607 		}
608 
609 		BEGIN_RING(9);
610 		OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_2, 5));
611 		OUT_RING(tex[2].pp_txfilter);
612 		OUT_RING(tex[2].pp_txformat);
613 		OUT_RING(tex[2].pp_txoffset);
614 		OUT_RING(tex[2].pp_txcblend);
615 		OUT_RING(tex[2].pp_txablend);
616 		OUT_RING(tex[2].pp_tfactor);
617 		OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_2, 0));
618 		OUT_RING(tex[2].pp_border_color);
619 		ADVANCE_RING();
620 	}
621 
622 	return (0);
623 }
624 
625 /* Emit 1.2 state */
626 static int radeon_emit_state2(drm_radeon_private_t *dev_priv,
627     drm_file_t *filp_priv, drm_radeon_state_t *state)
628 {
629 	RING_LOCALS;
630 
631 	if (state->dirty & RADEON_UPLOAD_ZBIAS) {
632 		BEGIN_RING(3);
633 		OUT_RING(CP_PACKET0(RADEON_SE_ZBIAS_FACTOR, 1));
634 		OUT_RING(state->context2.se_zbias_factor);
635 		OUT_RING(state->context2.se_zbias_constant);
636 		ADVANCE_RING();
637 	}
638 
639 	return (radeon_emit_state(dev_priv, filp_priv,
640 	    &state->context, state->tex, state->dirty));
641 }
642 
643 /*
644  * New (1.3) state mechanism.  3 commands (packet, scalar, vector) in
645  * 1.3 cmdbuffers allow all previous state to be updated as well as
646  * the tcl scalar and vector areas.
647  */
648 static struct {
649 	int start;
650 	int len;
651 	const char *name;
652 } packet[RADEON_MAX_STATE_PACKETS] = {
653 	{RADEON_PP_MISC, 7, "RADEON_PP_MISC"},
654 	{RADEON_PP_CNTL, 3, "RADEON_PP_CNTL"},
655 	{RADEON_RB3D_COLORPITCH, 1, "RADEON_RB3D_COLORPITCH"},
656 	{RADEON_RE_LINE_PATTERN, 2, "RADEON_RE_LINE_PATTERN"},
657 	{RADEON_SE_LINE_WIDTH, 1, "RADEON_SE_LINE_WIDTH"},
658 	{RADEON_PP_LUM_MATRIX, 1, "RADEON_PP_LUM_MATRIX"},
659 	{RADEON_PP_ROT_MATRIX_0, 2, "RADEON_PP_ROT_MATRIX_0"},
660 	{RADEON_RB3D_STENCILREFMASK, 3, "RADEON_RB3D_STENCILREFMASK"},
661 	{RADEON_SE_VPORT_XSCALE, 6, "RADEON_SE_VPORT_XSCALE"},
662 	{RADEON_SE_CNTL, 2, "RADEON_SE_CNTL"},
663 	{RADEON_SE_CNTL_STATUS, 1, "RADEON_SE_CNTL_STATUS"},
664 	{RADEON_RE_MISC, 1, "RADEON_RE_MISC"},
665 	{RADEON_PP_TXFILTER_0, 6, "RADEON_PP_TXFILTER_0"},
666 	{RADEON_PP_BORDER_COLOR_0, 1, "RADEON_PP_BORDER_COLOR_0"},
667 	{RADEON_PP_TXFILTER_1, 6, "RADEON_PP_TXFILTER_1"},
668 	{RADEON_PP_BORDER_COLOR_1, 1, "RADEON_PP_BORDER_COLOR_1"},
669 	{RADEON_PP_TXFILTER_2, 6, "RADEON_PP_TXFILTER_2"},
670 	{RADEON_PP_BORDER_COLOR_2, 1, "RADEON_PP_BORDER_COLOR_2"},
671 	{RADEON_SE_ZBIAS_FACTOR, 2, "RADEON_SE_ZBIAS_FACTOR"},
672 	{RADEON_SE_TCL_OUTPUT_VTX_FMT, 11, "RADEON_SE_TCL_OUTPUT_VTX_FMT"},
673 	{RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, 17,
674 		    "RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED"},
675 	{R200_PP_TXCBLEND_0, 4, "R200_PP_TXCBLEND_0"},
676 	{R200_PP_TXCBLEND_1, 4, "R200_PP_TXCBLEND_1"},
677 	{R200_PP_TXCBLEND_2, 4, "R200_PP_TXCBLEND_2"},
678 	{R200_PP_TXCBLEND_3, 4, "R200_PP_TXCBLEND_3"},
679 	{R200_PP_TXCBLEND_4, 4, "R200_PP_TXCBLEND_4"},
680 	{R200_PP_TXCBLEND_5, 4, "R200_PP_TXCBLEND_5"},
681 	{R200_PP_TXCBLEND_6, 4, "R200_PP_TXCBLEND_6"},
682 	{R200_PP_TXCBLEND_7, 4, "R200_PP_TXCBLEND_7"},
683 	{R200_SE_TCL_LIGHT_MODEL_CTL_0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0"},
684 	{R200_PP_TFACTOR_0, 6, "R200_PP_TFACTOR_0"},
685 	{R200_SE_VTX_FMT_0, 4, "R200_SE_VTX_FMT_0"},
686 	{R200_SE_VAP_CNTL, 1, "R200_SE_VAP_CNTL"},
687 	{R200_SE_TCL_MATRIX_SEL_0, 5, "R200_SE_TCL_MATRIX_SEL_0"},
688 	{R200_SE_TCL_TEX_PROC_CTL_2, 5, "R200_SE_TCL_TEX_PROC_CTL_2"},
689 	{R200_SE_TCL_UCP_VERT_BLEND_CTL, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL"},
690 	{R200_PP_TXFILTER_0, 6, "R200_PP_TXFILTER_0"},
691 	{R200_PP_TXFILTER_1, 6, "R200_PP_TXFILTER_1"},
692 	{R200_PP_TXFILTER_2, 6, "R200_PP_TXFILTER_2"},
693 	{R200_PP_TXFILTER_3, 6, "R200_PP_TXFILTER_3"},
694 	{R200_PP_TXFILTER_4, 6, "R200_PP_TXFILTER_4"},
695 	{R200_PP_TXFILTER_5, 6, "R200_PP_TXFILTER_5"},
696 	{R200_PP_TXOFFSET_0, 1, "R200_PP_TXOFFSET_0"},
697 	{R200_PP_TXOFFSET_1, 1, "R200_PP_TXOFFSET_1"},
698 	{R200_PP_TXOFFSET_2, 1, "R200_PP_TXOFFSET_2"},
699 	{R200_PP_TXOFFSET_3, 1, "R200_PP_TXOFFSET_3"},
700 	{R200_PP_TXOFFSET_4, 1, "R200_PP_TXOFFSET_4"},
701 	{R200_PP_TXOFFSET_5, 1, "R200_PP_TXOFFSET_5"},
702 	{R200_SE_VTE_CNTL, 1, "R200_SE_VTE_CNTL"},
703 	{R200_SE_TCL_OUTPUT_VTX_COMP_SEL, 1,
704 	"R200_SE_TCL_OUTPUT_VTX_COMP_SEL"},
705 	{R200_PP_TAM_DEBUG3, 1, "R200_PP_TAM_DEBUG3"},
706 	{R200_PP_CNTL_X, 1, "R200_PP_CNTL_X"},
707 	{R200_RB3D_DEPTHXY_OFFSET, 1, "R200_RB3D_DEPTHXY_OFFSET"},
708 	{R200_RE_AUX_SCISSOR_CNTL, 1, "R200_RE_AUX_SCISSOR_CNTL"},
709 	{R200_RE_SCISSOR_TL_0, 2, "R200_RE_SCISSOR_TL_0"},
710 	{R200_RE_SCISSOR_TL_1, 2, "R200_RE_SCISSOR_TL_1"},
711 	{R200_RE_SCISSOR_TL_2, 2, "R200_RE_SCISSOR_TL_2"},
712 	{R200_SE_VAP_CNTL_STATUS, 1, "R200_SE_VAP_CNTL_STATUS"},
713 	{R200_SE_VTX_STATE_CNTL, 1, "R200_SE_VTX_STATE_CNTL"},
714 	{R200_RE_POINTSIZE, 1, "R200_RE_POINTSIZE"},
715 	{R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0, 4,
716 		    "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0"},
717 	{R200_PP_CUBIC_FACES_0, 1, "R200_PP_CUBIC_FACES_0"},	/* 61 */
718 	{R200_PP_CUBIC_OFFSET_F1_0, 5, "R200_PP_CUBIC_OFFSET_F1_0"}, /* 62 */
719 	{R200_PP_CUBIC_FACES_1, 1, "R200_PP_CUBIC_FACES_1"},
720 	{R200_PP_CUBIC_OFFSET_F1_1, 5, "R200_PP_CUBIC_OFFSET_F1_1"},
721 	{R200_PP_CUBIC_FACES_2, 1, "R200_PP_CUBIC_FACES_2"},
722 	{R200_PP_CUBIC_OFFSET_F1_2, 5, "R200_PP_CUBIC_OFFSET_F1_2"},
723 	{R200_PP_CUBIC_FACES_3, 1, "R200_PP_CUBIC_FACES_3"},
724 	{R200_PP_CUBIC_OFFSET_F1_3, 5, "R200_PP_CUBIC_OFFSET_F1_3"},
725 	{R200_PP_CUBIC_FACES_4, 1, "R200_PP_CUBIC_FACES_4"},
726 	{R200_PP_CUBIC_OFFSET_F1_4, 5, "R200_PP_CUBIC_OFFSET_F1_4"},
727 	{R200_PP_CUBIC_FACES_5, 1, "R200_PP_CUBIC_FACES_5"},
728 	{R200_PP_CUBIC_OFFSET_F1_5, 5, "R200_PP_CUBIC_OFFSET_F1_5"},
729 	{RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0"},
730 	{RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1"},
731 	{RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_2"},
732 	{R200_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"},
733 	{R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"},
734 	{RADEON_PP_CUBIC_FACES_0, 1, "RADEON_PP_CUBIC_FACES_0"},
735 	{RADEON_PP_CUBIC_OFFSET_T0_0, 5, "RADEON_PP_CUBIC_OFFSET_T0_0"},
736 	{RADEON_PP_CUBIC_FACES_1, 1, "RADEON_PP_CUBIC_FACES_1"},
737 	{RADEON_PP_CUBIC_OFFSET_T1_0, 5, "RADEON_PP_CUBIC_OFFSET_T1_0"},
738 	{RADEON_PP_CUBIC_FACES_2, 1, "RADEON_PP_CUBIC_FACES_2"},
739 	{RADEON_PP_CUBIC_OFFSET_T2_0, 5, "RADEON_PP_CUBIC_OFFSET_T2_0"},
740 	{R200_PP_TRI_PERF, 2, "R200_PP_TRI_PERF"},
741 	{R200_PP_AFS_0, 32, "R200_PP_AFS_0"},	/* 85 */
742 	{R200_PP_AFS_1, 32, "R200_PP_AFS_1"},
743 	{R200_PP_TFACTOR_0, 8, "R200_ATF_TFACTOR"},
744 	{R200_PP_TXFILTER_0, 8, "R200_PP_TXCTLALL_0"},
745 	{R200_PP_TXFILTER_1, 8, "R200_PP_TXCTLALL_1"},
746 	{R200_PP_TXFILTER_2, 8, "R200_PP_TXCTLALL_2"},
747 	{R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"},
748 	{R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"},
749 	{R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"},
750 	{R200_VAP_PVS_CNTL_1, 2, "R200_VAP_PVS_CNTL"},
751 };
752 
753 /*
754  * Performance monitoring functions
755  */
756 
757 static void radeon_clear_box(drm_radeon_private_t *dev_priv,
758     int x, int y, int w, int h, int r, int g, int b)
759 {
760 	u32 color;
761 	RING_LOCALS;
762 
763 	x += dev_priv->sarea_priv->boxes[0].x1;
764 	y += dev_priv->sarea_priv->boxes[0].y1;
765 
766 	switch (dev_priv->color_fmt) {
767 	case RADEON_COLOR_FORMAT_RGB565:
768 		color = (((r & 0xf8) << 8) |
769 		    ((g & 0xfc) << 3) | ((b & 0xf8) >> 3));
770 		break;
771 	case RADEON_COLOR_FORMAT_ARGB8888:
772 	default:
773 		color = (((0xfful) << 24) | (r << 16) | (g << 8) | b);
774 		break;
775 	}
776 
777 	BEGIN_RING(4);
778 	RADEON_WAIT_UNTIL_3D_IDLE();
779 	OUT_RING(CP_PACKET0(RADEON_DP_WRITE_MASK, 0));
780 	OUT_RING(0xffffffff);
781 	ADVANCE_RING();
782 
783 	BEGIN_RING(6);
784 
785 	OUT_RING(CP_PACKET3(RADEON_CNTL_PAINT_MULTI, 4));
786 	OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
787 	    RADEON_GMC_BRUSH_SOLID_COLOR |
788 	    (dev_priv->color_fmt << 8) |
789 	    RADEON_GMC_SRC_DATATYPE_COLOR |
790 	    RADEON_ROP3_P | RADEON_GMC_CLR_CMP_CNTL_DIS);
791 
792 	if (dev_priv->page_flipping && dev_priv->current_page == 1) {
793 		OUT_RING(dev_priv->front_pitch_offset);
794 	} else {
795 		OUT_RING(dev_priv->back_pitch_offset);
796 	}
797 
798 	OUT_RING(color);
799 
800 	OUT_RING((x << 16) | y);
801 	OUT_RING((w << 16) | h);
802 
803 	ADVANCE_RING();
804 }
805 
806 static void radeon_cp_performance_boxes(drm_radeon_private_t *dev_priv)
807 {
808 	/*
809 	 * Collapse various things into a wait flag -- trying to
810 	 * guess if userspase slept -- better just to have them tell us.
811 	 */
812 	if (dev_priv->stats.last_frame_reads > 1 ||
813 	    dev_priv->stats.last_clear_reads > dev_priv->stats.clears) {
814 		dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
815 	}
816 
817 	if (dev_priv->stats.freelist_loops) {
818 		dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
819 	}
820 
821 	/* Purple box for page flipping */
822 	if (dev_priv->stats.boxes & RADEON_BOX_FLIP)
823 		radeon_clear_box(dev_priv, 4, 4, 8, 8, 255, 0, 255);
824 
825 	/* Red box if we have to wait for idle at any point */
826 	if (dev_priv->stats.boxes & RADEON_BOX_WAIT_IDLE)
827 		radeon_clear_box(dev_priv, 16, 4, 8, 8, 255, 0, 0);
828 
829 	/* Blue box: lost context? */
830 
831 	/* Yellow box for texture swaps */
832 	if (dev_priv->stats.boxes & RADEON_BOX_TEXTURE_LOAD)
833 		radeon_clear_box(dev_priv, 40, 4, 8, 8, 255, 255, 0);
834 
835 	/* Green box if hardware never idles (as far as we can tell) */
836 	if (!(dev_priv->stats.boxes & RADEON_BOX_DMA_IDLE))
837 		radeon_clear_box(dev_priv, 64, 4, 8, 8, 0, 255, 0);
838 
839 	/*
840 	 * Draw bars indicating number of buffers allocated
841 	 * (not a great measure, easily confused)
842 	 */
843 	if (dev_priv->stats.requested_bufs) {
844 		if (dev_priv->stats.requested_bufs > 100)
845 			dev_priv->stats.requested_bufs = 100;
846 
847 		radeon_clear_box(dev_priv, 4, 16,
848 		    dev_priv->stats.requested_bufs, 4, 196, 128, 128);
849 	}
850 
851 	(void) memset(&dev_priv->stats, 0, sizeof (dev_priv->stats));
852 
853 }
854 
855 /*
856  * CP command dispatch functions
857  */
858 
859 static void radeon_cp_dispatch_clear(drm_device_t *dev,
860     drm_radeon_clear_t *clear, drm_radeon_clear_rect_t *depth_boxes)
861 {
862 	drm_radeon_private_t *dev_priv = dev->dev_private;
863 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
864 	drm_radeon_depth_clear_t *depth_clear = &dev_priv->depth_clear;
865 	int nbox = sarea_priv->nbox;
866 	drm_clip_rect_t *pbox = sarea_priv->boxes;
867 	unsigned int flags = clear->flags;
868 	u32 rb3d_cntl = 0, rb3d_stencilrefmask = 0;
869 	int i;
870 	RING_LOCALS;
871 	DRM_DEBUG("flags = 0x%x\n", flags);
872 
873 	dev_priv->stats.clears++;
874 
875 	if (dev_priv->page_flipping && dev_priv->current_page == 1) {
876 		unsigned int tmp = flags;
877 
878 		flags &= ~(RADEON_FRONT | RADEON_BACK);
879 		if (tmp & RADEON_FRONT)
880 			flags |= RADEON_BACK;
881 		if (tmp & RADEON_BACK)
882 			flags |= RADEON_FRONT;
883 	}
884 
885 	if (flags & (RADEON_FRONT | RADEON_BACK)) {
886 
887 		BEGIN_RING(4);
888 
889 		/*
890 		 * Ensure the 3D stream is idle before doing a
891 		 * 2D fill to clear the front or back buffer.
892 		 */
893 		RADEON_WAIT_UNTIL_3D_IDLE();
894 
895 		OUT_RING(CP_PACKET0(RADEON_DP_WRITE_MASK, 0));
896 		OUT_RING(clear->color_mask);
897 
898 		ADVANCE_RING();
899 
900 		/* Make sure we restore the 3D state next time.  */
901 		dev_priv->sarea_priv->ctx_owner = 0;
902 
903 		for (i = 0; i < nbox; i++) {
904 			int x = pbox[i].x1;
905 			int y = pbox[i].y1;
906 			int w = pbox[i].x2 - x;
907 			int h = pbox[i].y2 - y;
908 
909 			DRM_DEBUG("dispatch clear %d,%d-%d,%d flags 0x%x\n",
910 			    x, y, w, h, flags);
911 
912 			if (flags & RADEON_FRONT) {
913 				BEGIN_RING(6);
914 
915 				OUT_RING(CP_PACKET3
916 				    (RADEON_CNTL_PAINT_MULTI, 4));
917 				OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
918 				    RADEON_GMC_BRUSH_SOLID_COLOR |
919 				    (dev_priv-> color_fmt << 8) |
920 				    RADEON_GMC_SRC_DATATYPE_COLOR |
921 				    RADEON_ROP3_P |
922 				    RADEON_GMC_CLR_CMP_CNTL_DIS);
923 
924 				OUT_RING(dev_priv->front_pitch_offset);
925 				OUT_RING(clear->clear_color);
926 
927 				OUT_RING((x << 16) | y);
928 				OUT_RING((w << 16) | h);
929 
930 				ADVANCE_RING();
931 			}
932 
933 			if (flags & RADEON_BACK) {
934 				BEGIN_RING(6);
935 
936 				OUT_RING(CP_PACKET3
937 				    (RADEON_CNTL_PAINT_MULTI, 4));
938 				OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
939 				    RADEON_GMC_BRUSH_SOLID_COLOR |
940 				    (dev_priv-> color_fmt << 8) |
941 				    RADEON_GMC_SRC_DATATYPE_COLOR |
942 				    RADEON_ROP3_P |
943 				    RADEON_GMC_CLR_CMP_CNTL_DIS);
944 
945 				OUT_RING(dev_priv->back_pitch_offset);
946 				OUT_RING(clear->clear_color);
947 
948 				OUT_RING((x << 16) | y);
949 				OUT_RING((w << 16) | h);
950 
951 				ADVANCE_RING();
952 			}
953 		}
954 	}
955 
956 	/* hyper z clear */
957 	/*
958 	 * no docs available, based on reverse engeneering
959 	 * by Stephane Marchesin
960 	 */
961 	if ((flags & (RADEON_DEPTH | RADEON_STENCIL)) &&
962 	    (flags & RADEON_CLEAR_FASTZ)) {
963 
964 		int i;
965 		int depthpixperline =
966 		    dev_priv->depth_fmt ==
967 		    RADEON_DEPTH_FORMAT_16BIT_INT_Z ?
968 		    (dev_priv->depth_pitch / 2) :
969 		    (dev_priv-> depth_pitch / 4);
970 
971 		u32 clearmask;
972 
973 		u32 tempRB3D_DEPTHCLEARVALUE = clear->clear_depth |
974 		    ((clear->depth_mask & 0xff) << 24);
975 
976 		/*
977 		 * Make sure we restore the 3D state next time.
978 		 * we haven't touched any "normal" state - still
979 		 * need this?
980 		 */
981 		dev_priv->sarea_priv->ctx_owner = 0;
982 
983 		if ((dev_priv->flags & RADEON_HAS_HIERZ) &&
984 		    (flags & RADEON_USE_HIERZ)) {
985 			/* FIXME : reverse engineer that for Rx00 cards */
986 			/*
987 			 * FIXME : the mask supposedly contains low-res
988 			 * z values. So can't set just to the max (0xff?
989 			 * or actually 0x3fff?), need to take z clear
990 			 * value into account?
991 			 */
992 			/*
993 			 * pattern seems to work for r100, though get
994 			 * slight rendering errors with glxgears. If
995 			 * hierz is not enabled for r100, only 4 bits
996 			 * which indicate clear (15,16,31,32, all zero)
997 			 * matter, the other ones are ignored, and the
998 			 * same clear mask can be used. That's very
999 			 * different behaviour than R200 which needs
1000 			 * different clear mask and different number
1001 			 * of tiles to clear if hierz is enabled or not !?!
1002 			 */
1003 			clearmask = (0xff << 22) | (0xff << 6) | 0x003f003f;
1004 		} else {
1005 		/*
1006 		 * clear mask : chooses the clearing pattern.
1007 		 * rv250: could be used to clear only parts of macrotiles
1008 		 * (but that would get really complicated...)?
1009 		 * bit 0 and 1 (either or both of them ?!?!) are used to
1010 		 * not clear tile (or maybe one of the bits indicates if
1011 		 * the tile is compressed or not), bit 2 and 3 to not
1012 		 * clear tile 1,...,.
1013 		 * Pattern is as follows:
1014 		 * | 0,1 | 4,5 | 8,9 |12,13|16,17|20,21|24,25|28,29|
1015 		 * bits -------------------------------------------------
1016 		 * | 2,3 | 6,7 |10,11|14,15|18,19|22,23|26,27|30,31|
1017 		 * rv100: clearmask covers 2x8 4x1 tiles, but one clear
1018 		 * still covers 256 pixels ?!?
1019 		 */
1020 			clearmask = 0x0;
1021 		}
1022 
1023 		BEGIN_RING(8);
1024 		RADEON_WAIT_UNTIL_2D_IDLE();
1025 		OUT_RING_REG(RADEON_RB3D_DEPTHCLEARVALUE,
1026 		    tempRB3D_DEPTHCLEARVALUE);
1027 		/* what offset is this exactly ? */
1028 		OUT_RING_REG(RADEON_RB3D_ZMASKOFFSET, 0);
1029 		/* need ctlstat, otherwise get some strange black flickering */
1030 		OUT_RING_REG(RADEON_RB3D_ZCACHE_CTLSTAT,
1031 		    RADEON_RB3D_ZC_FLUSH_ALL);
1032 		ADVANCE_RING();
1033 
1034 		for (i = 0; i < nbox; i++) {
1035 			int tileoffset, nrtilesx, nrtilesy, j;
1036 			/*
1037 			 * it looks like r200 needs rv-style clears, at
1038 			 * least if hierz is not enabled?
1039 			 */
1040 			if ((dev_priv->flags & RADEON_HAS_HIERZ) &&
1041 			    !(dev_priv->microcode_version == UCODE_R200)) {
1042 				/*
1043 				 * FIXME : figure this out for r200 (when hierz
1044 				 * is enabled). Or maybe r200 actually doesn't
1045 				 * need to put the low-res z value into the tile
1046 				 * cache like r100, but just needs to clear the
1047 				 * hi-level z-buffer? Works for R100, both with
1048 				 * hierz and without.R100 seems to operate on
1049 				 * 2x1 8x8 tiles, but... odd: offset/nrtiles
1050 				 * need to be 64 pix (4 blocka) aligned?
1051 				 * Potentially problematic with resolutions
1052 				 * which are not 64 pix aligned?
1053 				 */
1054 				tileoffset =
1055 				    ((pbox[i].y1 >> 3) * depthpixperline +
1056 				    pbox[i].x1) >> 6;
1057 				nrtilesx =
1058 				    ((pbox[i].x2 & ~63) -
1059 				    (pbox[i].x1 & ~63)) >> 4;
1060 				nrtilesy =
1061 				    (pbox[i].y2 >> 3) - (pbox[i].y1 >> 3);
1062 				for (j = 0; j <= nrtilesy; j++) {
1063 					BEGIN_RING(4);
1064 					OUT_RING(CP_PACKET3
1065 					    (RADEON_3D_CLEAR_ZMASK, 2));
1066 					/* first tile */
1067 					OUT_RING(tileoffset * 8);
1068 					/* the number of tiles to clear */
1069 					OUT_RING(nrtilesx + 4);
1070 					/*
1071 					 * clear mask :
1072 					 * chooses the clearing pattern.
1073 					 */
1074 					OUT_RING(clearmask);
1075 					ADVANCE_RING();
1076 					tileoffset += depthpixperline >> 6;
1077 				}
1078 			} else if (dev_priv->microcode_version == UCODE_R200) {
1079 				/* works for rv250. */
1080 				/*
1081 				 * find first macro tile
1082 				 * (8x2 4x4 z-pixels on rv250)
1083 				 */
1084 				tileoffset =
1085 				    ((pbox[i].y1 >> 3) * depthpixperline +
1086 				    pbox[i].x1) >> 5;
1087 				nrtilesx =
1088 				    (pbox[i].x2 >> 5) - (pbox[i].x1 >> 5);
1089 				nrtilesy =
1090 				    (pbox[i].y2 >> 3) - (pbox[i].y1 >> 3);
1091 				for (j = 0; j <= nrtilesy; j++) {
1092 					BEGIN_RING(4);
1093 					OUT_RING(CP_PACKET3
1094 					    (RADEON_3D_CLEAR_ZMASK, 2));
1095 					/* first tile */
1096 					/*
1097 					 * judging by the first tile
1098 					 * offset needed, could possibly
1099 					 * directly address/clear 4x4
1100 					 * tiles instead of 8x2 * 4x4
1101 					 * macro tiles, though would
1102 					 * still need clear mask for
1103 					 * right/bottom if truely 4x4
1104 					 * granularity is desired ?
1105 					 */
1106 					OUT_RING(tileoffset * 16);
1107 					/* the number of tiles to clear */
1108 					OUT_RING(nrtilesx + 1);
1109 					/*
1110 					 * clear mask :
1111 					 * chooses the clearing pattern.
1112 					 */
1113 					OUT_RING(clearmask);
1114 					ADVANCE_RING();
1115 					tileoffset += depthpixperline >> 5;
1116 				}
1117 			} else {	/* rv 100 */
1118 				/* rv100 might not need 64 pix alignment */
1119 				/* offsets are, hmm, weird */
1120 				tileoffset =
1121 				    ((pbox[i].y1 >> 4) * depthpixperline +
1122 				    pbox[i].x1) >> 6;
1123 				nrtilesx =
1124 				    ((pbox[i].x2 & ~63) -
1125 				    (pbox[i].x1 & ~63)) >> 4;
1126 				nrtilesy =
1127 				    (pbox[i].y2 >> 4) - (pbox[i].y1 >> 4);
1128 				for (j = 0; j <= nrtilesy; j++) {
1129 					BEGIN_RING(4);
1130 					OUT_RING(CP_PACKET3
1131 					    (RADEON_3D_CLEAR_ZMASK, 2));
1132 					OUT_RING(tileoffset * 128);
1133 					/* the number of tiles to clear */
1134 					OUT_RING(nrtilesx + 4);
1135 					/*
1136 					 * clear mask :
1137 					 * chooses the clearing pattern.
1138 					 */
1139 					OUT_RING(clearmask);
1140 					ADVANCE_RING();
1141 					tileoffset += depthpixperline >> 6;
1142 				}
1143 			}
1144 		}
1145 
1146 		/* TODO don't always clear all hi-level z tiles */
1147 		if ((dev_priv->flags & RADEON_HAS_HIERZ) &&
1148 		    (dev_priv->microcode_version == UCODE_R200) &&
1149 		    (flags & RADEON_USE_HIERZ))
1150 			/*
1151 			 * r100 and cards without hierarchical z-buffer
1152 			 * have no high-level z-buffer
1153 			 */
1154 			/*
1155 			 * FIXME : the mask supposedly contains low-res
1156 			 * z values. So can't set just to the max (0xff?
1157 			 * or actually 0x3fff?), need to take z clear value
1158 			 * into account?
1159 			 */
1160 		{
1161 			BEGIN_RING(4);
1162 			OUT_RING(CP_PACKET3(RADEON_3D_CLEAR_HIZ, 2));
1163 			OUT_RING(0x0);	/* First tile */
1164 			OUT_RING(0x3cc0);
1165 			OUT_RING((0xff << 22) | (0xff << 6) | 0x003f003f);
1166 			ADVANCE_RING();
1167 		}
1168 	}
1169 
1170 	/*
1171 	 * We have to clear the depth and/or stencil buffers by
1172 	 * rendering a quad into just those buffers.  Thus, we have to
1173 	 * make sure the 3D engine is configured correctly.
1174 	 */
1175 	else if ((dev_priv->microcode_version == UCODE_R200) &&
1176 	    (flags & (RADEON_DEPTH | RADEON_STENCIL))) {
1177 
1178 		int tempPP_CNTL;
1179 		int tempRE_CNTL;
1180 		int tempRB3D_CNTL;
1181 		int tempRB3D_ZSTENCILCNTL;
1182 		int tempRB3D_STENCILREFMASK;
1183 		int tempRB3D_PLANEMASK;
1184 		int tempSE_CNTL;
1185 		int tempSE_VTE_CNTL;
1186 		int tempSE_VTX_FMT_0;
1187 		int tempSE_VTX_FMT_1;
1188 		int tempSE_VAP_CNTL;
1189 		int tempRE_AUX_SCISSOR_CNTL;
1190 
1191 		tempPP_CNTL = 0;
1192 		tempRE_CNTL = 0;
1193 
1194 		tempRB3D_CNTL = depth_clear->rb3d_cntl;
1195 
1196 		tempRB3D_ZSTENCILCNTL = depth_clear->rb3d_zstencilcntl;
1197 		tempRB3D_STENCILREFMASK = 0x0;
1198 
1199 		tempSE_CNTL = depth_clear->se_cntl;
1200 
1201 		/* Disable TCL */
1202 
1203 		tempSE_VAP_CNTL =
1204 		    (/* SE_VAP_CNTL__FORCE_W_TO_ONE_MASK |  */
1205 		    (0x9 << SE_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT));
1206 
1207 		tempRB3D_PLANEMASK = 0x0;
1208 
1209 		tempRE_AUX_SCISSOR_CNTL = 0x0;
1210 
1211 		tempSE_VTE_CNTL =
1212 		    SE_VTE_CNTL__VTX_XY_FMT_MASK | SE_VTE_CNTL__VTX_Z_FMT_MASK;
1213 
1214 		/* Vertex format (X, Y, Z, W) */
1215 		tempSE_VTX_FMT_0 =
1216 		    SE_VTX_FMT_0__VTX_Z0_PRESENT_MASK |
1217 		    SE_VTX_FMT_0__VTX_W0_PRESENT_MASK;
1218 		tempSE_VTX_FMT_1 = 0x0;
1219 
1220 		/*
1221 		 * Depth buffer specific enables
1222 		 */
1223 		if (flags & RADEON_DEPTH) {
1224 			/* Enable depth buffer */
1225 			tempRB3D_CNTL |= RADEON_Z_ENABLE;
1226 		} else {
1227 			/* Disable depth buffer */
1228 			tempRB3D_CNTL &= ~RADEON_Z_ENABLE;
1229 		}
1230 
1231 		/*
1232 		 * Stencil buffer specific enables
1233 		 */
1234 		if (flags & RADEON_STENCIL) {
1235 			tempRB3D_CNTL |= RADEON_STENCIL_ENABLE;
1236 			tempRB3D_STENCILREFMASK = clear->depth_mask;
1237 		} else {
1238 			tempRB3D_CNTL &= ~RADEON_STENCIL_ENABLE;
1239 			tempRB3D_STENCILREFMASK = 0x00000000;
1240 		}
1241 
1242 		if (flags & RADEON_USE_COMP_ZBUF) {
1243 			tempRB3D_ZSTENCILCNTL |= RADEON_Z_COMPRESSION_ENABLE |
1244 			    RADEON_Z_DECOMPRESSION_ENABLE;
1245 		}
1246 		if (flags & RADEON_USE_HIERZ) {
1247 			tempRB3D_ZSTENCILCNTL |= RADEON_Z_HIERARCHY_ENABLE;
1248 		}
1249 
1250 		BEGIN_RING(26);
1251 		RADEON_WAIT_UNTIL_2D_IDLE();
1252 
1253 		OUT_RING_REG(RADEON_PP_CNTL, tempPP_CNTL);
1254 		OUT_RING_REG(R200_RE_CNTL, tempRE_CNTL);
1255 		OUT_RING_REG(RADEON_RB3D_CNTL, tempRB3D_CNTL);
1256 		OUT_RING_REG(RADEON_RB3D_ZSTENCILCNTL, tempRB3D_ZSTENCILCNTL);
1257 		OUT_RING_REG(RADEON_RB3D_STENCILREFMASK,
1258 		    tempRB3D_STENCILREFMASK);
1259 		OUT_RING_REG(RADEON_RB3D_PLANEMASK, tempRB3D_PLANEMASK);
1260 		OUT_RING_REG(RADEON_SE_CNTL, tempSE_CNTL);
1261 		OUT_RING_REG(R200_SE_VTE_CNTL, tempSE_VTE_CNTL);
1262 		OUT_RING_REG(R200_SE_VTX_FMT_0, tempSE_VTX_FMT_0);
1263 		OUT_RING_REG(R200_SE_VTX_FMT_1, tempSE_VTX_FMT_1);
1264 		OUT_RING_REG(R200_SE_VAP_CNTL, tempSE_VAP_CNTL);
1265 		OUT_RING_REG(R200_RE_AUX_SCISSOR_CNTL, tempRE_AUX_SCISSOR_CNTL);
1266 		ADVANCE_RING();
1267 
1268 		/* Make sure we restore the 3D state next time. */
1269 		dev_priv->sarea_priv->ctx_owner = 0;
1270 
1271 		for (i = 0; i < nbox; i++) {
1272 
1273 			/*
1274 			 * Funny that this should be required --
1275 			 *  sets top-left?
1276 			 */
1277 			radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1278 
1279 			BEGIN_RING(14);
1280 			OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 12));
1281 			OUT_RING((RADEON_PRIM_TYPE_RECT_LIST |
1282 			    RADEON_PRIM_WALK_RING |
1283 			    (3 << RADEON_NUM_VERTICES_SHIFT)));
1284 			OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1285 			OUT_RING(depth_boxes[i].ui[CLEAR_Y1]);
1286 			OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1287 			OUT_RING(0x3f800000);
1288 			OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1289 			OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1290 			OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1291 			OUT_RING(0x3f800000);
1292 			OUT_RING(depth_boxes[i].ui[CLEAR_X2]);
1293 			OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1294 			OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1295 			OUT_RING(0x3f800000);
1296 			ADVANCE_RING();
1297 		}
1298 	} else if ((flags & (RADEON_DEPTH | RADEON_STENCIL))) {
1299 
1300 		int tempRB3D_ZSTENCILCNTL = depth_clear->rb3d_zstencilcntl;
1301 
1302 		rb3d_cntl = depth_clear->rb3d_cntl;
1303 
1304 		if (flags & RADEON_DEPTH) {
1305 			rb3d_cntl |= RADEON_Z_ENABLE;
1306 		} else {
1307 			rb3d_cntl &= ~RADEON_Z_ENABLE;
1308 		}
1309 
1310 		if (flags & RADEON_STENCIL) {
1311 			rb3d_cntl |= RADEON_STENCIL_ENABLE;
1312 
1313 			/* misnamed field */
1314 			rb3d_stencilrefmask = clear->depth_mask;
1315 
1316 		} else {
1317 			rb3d_cntl &= ~RADEON_STENCIL_ENABLE;
1318 			rb3d_stencilrefmask = 0x00000000;
1319 		}
1320 
1321 		if (flags & RADEON_USE_COMP_ZBUF) {
1322 			tempRB3D_ZSTENCILCNTL |= RADEON_Z_COMPRESSION_ENABLE |
1323 			    RADEON_Z_DECOMPRESSION_ENABLE;
1324 		}
1325 		if (flags & RADEON_USE_HIERZ) {
1326 			tempRB3D_ZSTENCILCNTL |= RADEON_Z_HIERARCHY_ENABLE;
1327 		}
1328 
1329 		BEGIN_RING(13);
1330 		RADEON_WAIT_UNTIL_2D_IDLE();
1331 
1332 		OUT_RING(CP_PACKET0(RADEON_PP_CNTL, 1));
1333 		OUT_RING(0x00000000);
1334 		OUT_RING(rb3d_cntl);
1335 
1336 		OUT_RING_REG(RADEON_RB3D_ZSTENCILCNTL, tempRB3D_ZSTENCILCNTL);
1337 		OUT_RING_REG(RADEON_RB3D_STENCILREFMASK, rb3d_stencilrefmask);
1338 		OUT_RING_REG(RADEON_RB3D_PLANEMASK, 0x00000000);
1339 		OUT_RING_REG(RADEON_SE_CNTL, depth_clear->se_cntl);
1340 		ADVANCE_RING();
1341 
1342 		/* Make sure we restore the 3D state next time.  */
1343 		dev_priv->sarea_priv->ctx_owner = 0;
1344 
1345 		for (i = 0; i < nbox; i++) {
1346 
1347 			/*
1348 			 * Funny that this should be required --
1349 			 *  sets top-left?
1350 			 */
1351 			radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1352 
1353 			BEGIN_RING(15);
1354 
1355 			OUT_RING(CP_PACKET3(RADEON_3D_DRAW_IMMD, 13));
1356 			OUT_RING(RADEON_VTX_Z_PRESENT |
1357 			    RADEON_VTX_PKCOLOR_PRESENT);
1358 			OUT_RING((RADEON_PRIM_TYPE_RECT_LIST |
1359 			    RADEON_PRIM_WALK_RING |
1360 			    RADEON_MAOS_ENABLE |
1361 			    RADEON_VTX_FMT_RADEON_MODE |
1362 			    (3 << RADEON_NUM_VERTICES_SHIFT)));
1363 
1364 			OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1365 			OUT_RING(depth_boxes[i].ui[CLEAR_Y1]);
1366 			OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1367 			OUT_RING(0x0);
1368 
1369 			OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1370 			OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1371 			OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1372 			OUT_RING(0x0);
1373 
1374 			OUT_RING(depth_boxes[i].ui[CLEAR_X2]);
1375 			OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1376 			OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1377 			OUT_RING(0x0);
1378 
1379 			ADVANCE_RING();
1380 		}
1381 	}
1382 
1383 	/*
1384 	 * Increment the clear counter.  The client-side 3D driver must
1385 	 * wait on this value before performing the clear ioctl.  We
1386 	 * need this because the card's so damned fast...
1387 	 */
1388 	dev_priv->sarea_priv->last_clear++;
1389 
1390 	BEGIN_RING(4);
1391 
1392 	RADEON_CLEAR_AGE(dev_priv->sarea_priv->last_clear);
1393 	RADEON_WAIT_UNTIL_IDLE();
1394 
1395 	ADVANCE_RING();
1396 }
1397 
1398 static void radeon_cp_dispatch_swap(drm_device_t *dev)
1399 {
1400 	drm_radeon_private_t *dev_priv = dev->dev_private;
1401 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1402 	int nbox = sarea_priv->nbox;
1403 	drm_clip_rect_t *pbox = sarea_priv->boxes;
1404 	int i;
1405 	RING_LOCALS;
1406 
1407 	/* Do some trivial performance monitoring... */
1408 	if (dev_priv->do_boxes)
1409 		radeon_cp_performance_boxes(dev_priv);
1410 
1411 	/*
1412 	 * Wait for the 3D stream to idle before dispatching the bitblt.
1413 	 * This will prevent data corruption between the two streams.
1414 	 */
1415 	BEGIN_RING(2);
1416 
1417 	RADEON_WAIT_UNTIL_3D_IDLE();
1418 
1419 	ADVANCE_RING();
1420 
1421 	for (i = 0; i < nbox; i++) {
1422 		int x = pbox[i].x1;
1423 		int y = pbox[i].y1;
1424 		int w = pbox[i].x2 - x;
1425 		int h = pbox[i].y2 - y;
1426 
1427 		DRM_DEBUG("dispatch swap %d,%d-%d,%d\n", x, y, w, h);
1428 
1429 		BEGIN_RING(9);
1430 
1431 		OUT_RING(CP_PACKET0(RADEON_DP_GUI_MASTER_CNTL, 0));
1432 		OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
1433 		    RADEON_GMC_DST_PITCH_OFFSET_CNTL |
1434 		    RADEON_GMC_BRUSH_NONE |
1435 		    (dev_priv->color_fmt << 8) |
1436 		    RADEON_GMC_SRC_DATATYPE_COLOR |
1437 		    RADEON_ROP3_S |
1438 		    RADEON_DP_SRC_SOURCE_MEMORY |
1439 		    RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
1440 
1441 		/* Make this work even if front & back are flipped: */
1442 		OUT_RING(CP_PACKET0(RADEON_SRC_PITCH_OFFSET, 1));
1443 		if (dev_priv->current_page == 0) {
1444 			OUT_RING(dev_priv->back_pitch_offset);
1445 			OUT_RING(dev_priv->front_pitch_offset);
1446 		} else {
1447 			OUT_RING(dev_priv->front_pitch_offset);
1448 			OUT_RING(dev_priv->back_pitch_offset);
1449 		}
1450 
1451 		OUT_RING(CP_PACKET0(RADEON_SRC_X_Y, 2));
1452 		OUT_RING((x << 16) | y);
1453 		OUT_RING((x << 16) | y);
1454 		OUT_RING((w << 16) | h);
1455 
1456 		ADVANCE_RING();
1457 	}
1458 
1459 	/*
1460 	 * Increment the frame counter.  The client-side 3D driver must
1461 	 * throttle the framerate by waiting for this value before
1462 	 * performing the swapbuffer ioctl.
1463 	 */
1464 	dev_priv->sarea_priv->last_frame ++;
1465 
1466 	BEGIN_RING(4);
1467 
1468 	RADEON_FRAME_AGE(dev_priv->sarea_priv->last_frame);
1469 	RADEON_WAIT_UNTIL_2D_IDLE();
1470 
1471 	ADVANCE_RING();
1472 }
1473 
1474 static void radeon_cp_dispatch_flip(drm_device_t *dev)
1475 {
1476 	drm_radeon_private_t *dev_priv = dev->dev_private;
1477 	drm_sarea_t *sarea = (drm_sarea_t *)dev_priv->sarea->handle;
1478 	int offset = (dev_priv->current_page == 1)
1479 	    ? dev_priv->front_offset : dev_priv->back_offset;
1480 	RING_LOCALS;
1481 	DRM_DEBUG("%s: page=%d pfCurrentPage=%d\n",
1482 	    __FUNCTION__,
1483 	    dev_priv->current_page, dev_priv->sarea_priv->pfCurrentPage);
1484 
1485 	/* Do some trivial performance monitoring... */
1486 	if (dev_priv->do_boxes) {
1487 		dev_priv->stats.boxes |= RADEON_BOX_FLIP;
1488 		radeon_cp_performance_boxes(dev_priv);
1489 	}
1490 
1491 	/* Update the frame offsets for both CRTCs */
1492 	BEGIN_RING(6);
1493 
1494 	RADEON_WAIT_UNTIL_3D_IDLE();
1495 	OUT_RING_REG(RADEON_CRTC_OFFSET,
1496 	    ((sarea->frame.y * dev_priv->front_pitch +
1497 	    sarea->frame.x * (dev_priv->color_fmt - 2)) & ~7) + offset);
1498 	OUT_RING_REG(RADEON_CRTC2_OFFSET,
1499 	    dev_priv->sarea_priv->crtc2_base + offset);
1500 
1501 	ADVANCE_RING();
1502 
1503 	/*
1504 	 * Increment the frame counter.  The client-side 3D driver must
1505 	 * throttle the framerate by waiting for this value before
1506 	 * performing the swapbuffer ioctl.
1507 	 */
1508 	dev_priv->sarea_priv->last_frame ++;
1509 	dev_priv->sarea_priv->pfCurrentPage = dev_priv->current_page =
1510 	    1 - dev_priv->current_page;
1511 
1512 	BEGIN_RING(2);
1513 
1514 	RADEON_FRAME_AGE(dev_priv->sarea_priv->last_frame);
1515 
1516 	ADVANCE_RING();
1517 }
1518 
1519 static int bad_prim_vertex_nr(int primitive, int nr)
1520 {
1521 	switch (primitive & RADEON_PRIM_TYPE_MASK) {
1522 	case RADEON_PRIM_TYPE_NONE:
1523 	case RADEON_PRIM_TYPE_POINT:
1524 		return (nr < 1);
1525 	case RADEON_PRIM_TYPE_LINE:
1526 		return ((nr & 1) || nr == 0);
1527 	case RADEON_PRIM_TYPE_LINE_STRIP:
1528 		return (nr < 2);
1529 	case RADEON_PRIM_TYPE_TRI_LIST:
1530 	case RADEON_PRIM_TYPE_3VRT_POINT_LIST:
1531 	case RADEON_PRIM_TYPE_3VRT_LINE_LIST:
1532 	case RADEON_PRIM_TYPE_RECT_LIST:
1533 		return (nr % 3 || nr == 0);
1534 	case RADEON_PRIM_TYPE_TRI_FAN:
1535 	case RADEON_PRIM_TYPE_TRI_STRIP:
1536 		return (nr < 3);
1537 	default:
1538 		return (1);
1539 	}
1540 }
1541 
1542 typedef struct {
1543 	unsigned int start;
1544 	unsigned int finish;
1545 	unsigned int prim;
1546 	unsigned int numverts;
1547 	unsigned int offset;
1548 	unsigned int vc_format;
1549 } drm_radeon_tcl_prim_t;
1550 
1551 static void radeon_cp_dispatch_vertex(drm_device_t *dev,
1552     drm_buf_t *buf, drm_radeon_tcl_prim_t *prim)
1553 {
1554 	drm_radeon_private_t *dev_priv = dev->dev_private;
1555 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1556 	int offset = dev_priv->gart_buffers_offset + buf->offset + prim->start;
1557 	int numverts = (int)prim->numverts;
1558 	int nbox = sarea_priv->nbox;
1559 	int i = 0;
1560 	RING_LOCALS;
1561 
1562 	DRM_DEBUG("hwprim 0x%x vfmt 0x%x %d..%d %d verts\n",
1563 	    prim->prim, prim->vc_format, prim->start,
1564 	    prim->finish, prim->numverts);
1565 
1566 	if (bad_prim_vertex_nr(prim->prim, prim->numverts)) {
1567 		DRM_ERROR("bad prim %x numverts %d\n",
1568 		    prim->prim, prim->numverts);
1569 		return;
1570 	}
1571 
1572 	do {
1573 		/* Emit the next cliprect */
1574 		if (i < nbox) {
1575 			radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1576 		}
1577 
1578 		/* Emit the vertex buffer rendering commands */
1579 		BEGIN_RING(5);
1580 
1581 		OUT_RING(CP_PACKET3(RADEON_3D_RNDR_GEN_INDX_PRIM, 3));
1582 		OUT_RING(offset);
1583 		OUT_RING(numverts);
1584 		OUT_RING(prim->vc_format);
1585 		OUT_RING(prim->prim | RADEON_PRIM_WALK_LIST |
1586 		    RADEON_COLOR_ORDER_RGBA |
1587 		    RADEON_VTX_FMT_RADEON_MODE |
1588 		    (numverts << RADEON_NUM_VERTICES_SHIFT));
1589 
1590 		ADVANCE_RING();
1591 
1592 		i++;
1593 	} while (i < nbox);
1594 }
1595 
1596 static void radeon_cp_discard_buffer(drm_device_t *dev, drm_buf_t *buf)
1597 {
1598 	drm_radeon_private_t *dev_priv = dev->dev_private;
1599 	drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
1600 	RING_LOCALS;
1601 
1602 	buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
1603 
1604 	/* Emit the vertex buffer age */
1605 	BEGIN_RING(2);
1606 	RADEON_DISPATCH_AGE(buf_priv->age);
1607 	ADVANCE_RING();
1608 
1609 	buf->pending = 1;
1610 	buf->used = 0;
1611 }
1612 
1613 static void radeon_cp_dispatch_indirect(drm_device_t *dev,
1614     drm_buf_t *buf, int start, int end)
1615 {
1616 	drm_radeon_private_t *dev_priv = dev->dev_private;
1617 	RING_LOCALS;
1618 	DRM_DEBUG("indirect: buf=%d s=0x%x e=0x%x\n", buf->idx, start, end);
1619 
1620 	if (start != end) {
1621 		int offset = (dev_priv->gart_buffers_offset +
1622 		    buf->offset + start);
1623 		int dwords = (end - start + 3) / sizeof (u32);
1624 
1625 		/*
1626 		 * Indirect buffer data must be an even number of
1627 		 * dwords, so if we've been given an odd number we must
1628 		 * pad the data with a Type-2 CP packet.
1629 		 */
1630 		if (dwords & 1) {
1631 			u32 *data = (u32 *)(uintptr_t)
1632 			    ((char *)dev->agp_buffer_map->handle
1633 			    + buf->offset + start);
1634 			data[dwords++] = RADEON_CP_PACKET2;
1635 		}
1636 
1637 		/* Fire off the indirect buffer */
1638 		BEGIN_RING(3);
1639 
1640 		OUT_RING(CP_PACKET0(RADEON_CP_IB_BASE, 1));
1641 		OUT_RING(offset);
1642 		OUT_RING(dwords);
1643 
1644 		ADVANCE_RING();
1645 	}
1646 }
1647 
1648 static void radeon_cp_dispatch_indices(drm_device_t *dev,
1649     drm_buf_t *elt_buf, drm_radeon_tcl_prim_t *prim)
1650 {
1651 	drm_radeon_private_t *dev_priv = dev->dev_private;
1652 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1653 	int offset = dev_priv->gart_buffers_offset + prim->offset;
1654 	u32 *data;
1655 	int dwords;
1656 	int i = 0;
1657 	int start = prim->start + RADEON_INDEX_PRIM_OFFSET;
1658 	int count = (prim->finish - start) / sizeof (u16);
1659 	int nbox = sarea_priv->nbox;
1660 
1661 	DRM_DEBUG("hwprim 0x%x vfmt 0x%x %d..%d offset: %x nr %d\n",
1662 	    prim->prim, prim->vc_format, prim->start,
1663 	    prim->finish, prim->offset, prim->numverts);
1664 
1665 	if (bad_prim_vertex_nr(prim->prim, count)) {
1666 		DRM_ERROR("bad prim %x count %d\n", prim->prim, count);
1667 		return;
1668 	}
1669 
1670 	if (start >= prim->finish || (prim->start & 0x7)) {
1671 		DRM_ERROR("buffer prim %d\n", prim->prim);
1672 		return;
1673 	}
1674 
1675 	dwords = (prim->finish - prim->start + 3) / sizeof (u32);
1676 
1677 	data = (u32 *)(uintptr_t)((char *)dev->agp_buffer_map->handle +
1678 	    elt_buf->offset + prim->start);
1679 
1680 	data[0] = CP_PACKET3(RADEON_3D_RNDR_GEN_INDX_PRIM, dwords - 2);
1681 	data[1] = offset;
1682 	data[2] = prim->numverts;
1683 	data[3] = prim->vc_format;
1684 	data[4] = (prim->prim |
1685 	    RADEON_PRIM_WALK_IND |
1686 	    RADEON_COLOR_ORDER_RGBA |
1687 	    RADEON_VTX_FMT_RADEON_MODE |
1688 	    (count << RADEON_NUM_VERTICES_SHIFT));
1689 
1690 	do {
1691 		if (i < nbox)
1692 			radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1693 
1694 		radeon_cp_dispatch_indirect(dev, elt_buf,
1695 		    prim->start, prim->finish);
1696 
1697 		i++;
1698 	} while (i < nbox);
1699 
1700 }
1701 
1702 #define	RADEON_MAX_TEXTURE_SIZE RADEON_BUFFER_SIZE
1703 
1704 /*ARGSUSED*/
1705 static int radeon_cp_dispatch_texture(drm_file_t *fpriv,
1706     drm_device_t *dev, drm_radeon_texture_t *tex,
1707     drm_radeon_tex_image_t *image, int mode)
1708 {
1709 	drm_radeon_private_t *dev_priv = dev->dev_private;
1710 	drm_buf_t *buf;
1711 	u32 format;
1712 	u32 *buffer;
1713 	const u8 __user *data;
1714 	int size, dwords, tex_width, blit_width, spitch;
1715 	u32 height;
1716 	int i;
1717 	u32 texpitch, microtile;
1718 	u32 offset;
1719 	RING_LOCALS;
1720 
1721 
1722 	if (radeon_check_and_fixup_offset(dev_priv, fpriv, &tex->offset)) {
1723 		DRM_ERROR("Invalid destination offset\n");
1724 		return (EINVAL);
1725 	}
1726 
1727 	dev_priv->stats.boxes |= RADEON_BOX_TEXTURE_LOAD;
1728 
1729 	/*
1730 	 * Flush the pixel cache.  This ensures no pixel data gets mixed
1731 	 * up with the texture data from the host data blit, otherwise
1732 	 * part of the texture image may be corrupted.
1733 	 */
1734 	BEGIN_RING(4);
1735 	RADEON_FLUSH_CACHE();
1736 	RADEON_WAIT_UNTIL_IDLE();
1737 	ADVANCE_RING();
1738 
1739 	/*
1740 	 * The compiler won't optimize away a division by a variable,
1741 	 * even if the only legal values are powers of two.  Thus, we'll
1742 	 * use a shift instead.
1743 	 */
1744 	switch (tex->format) {
1745 	case RADEON_TXFORMAT_ARGB8888:
1746 	case RADEON_TXFORMAT_RGBA8888:
1747 		format = RADEON_COLOR_FORMAT_ARGB8888;
1748 		tex_width = tex->width * 4;
1749 		blit_width = image->width * 4;
1750 		break;
1751 	case RADEON_TXFORMAT_AI88:
1752 	case RADEON_TXFORMAT_ARGB1555:
1753 	case RADEON_TXFORMAT_RGB565:
1754 	case RADEON_TXFORMAT_ARGB4444:
1755 	case RADEON_TXFORMAT_VYUY422:
1756 	case RADEON_TXFORMAT_YVYU422:
1757 		format = RADEON_COLOR_FORMAT_RGB565;
1758 		tex_width = tex->width * 2;
1759 		blit_width = image->width * 2;
1760 		break;
1761 	case RADEON_TXFORMAT_I8:
1762 	case RADEON_TXFORMAT_RGB332:
1763 		format = RADEON_COLOR_FORMAT_CI8;
1764 		tex_width = tex->width * 1;
1765 		blit_width = image->width * 1;
1766 		break;
1767 	default:
1768 		DRM_ERROR("invalid texture format %d\n", tex->format);
1769 		return (EINVAL);
1770 	}
1771 	spitch = blit_width >> 6;
1772 	if (spitch == 0 && image->height > 1)
1773 		return (EINVAL);
1774 
1775 	texpitch = tex->pitch;
1776 	if ((texpitch << 22) & RADEON_DST_TILE_MICRO) {
1777 		microtile = 1;
1778 		if (tex_width < 64) {
1779 			texpitch &= ~(RADEON_DST_TILE_MICRO >> 22);
1780 			/* we got tiled coordinates, untile them */
1781 			image->x *= 2;
1782 		}
1783 	} else
1784 		microtile = 0;
1785 
1786 	DRM_DEBUG("tex=%dx%d blit=%d\n", tex_width, tex->height, blit_width);
1787 
1788 	do {
1789 		DRM_DEBUG("tex: ofs=0x%x p=%d f=%d x=%hd y=%hd w=%hd h=%hd\n",
1790 		    tex->offset >> 10, tex->pitch, tex->format,
1791 		    image->x, image->y, image->width, image->height);
1792 
1793 		/*
1794 		 * Make a copy of some parameters in case we have to
1795 		 * update them for a multi-pass texture blit.
1796 		 */
1797 		height = image->height;
1798 		data = (const u8 __user *)image->data;
1799 
1800 		size = height * blit_width;
1801 
1802 		if (size > RADEON_MAX_TEXTURE_SIZE) {
1803 			height = RADEON_MAX_TEXTURE_SIZE / blit_width;
1804 			size = height * blit_width;
1805 		} else if (size < 4 && size > 0) {
1806 			size = 4;
1807 		} else if (size == 0) {
1808 			return (0);
1809 		}
1810 
1811 		buf = radeon_freelist_get(dev);
1812 #if 0
1813 		if (0 && !buf) {
1814 			radeon_do_cp_idle(dev_priv);
1815 			buf = radeon_freelist_get(dev);
1816 		}
1817 #endif
1818 		if (!buf) {
1819 			DRM_DEBUG("radeon_cp_dispatch_texture: EAGAIN\n");
1820 
1821 #ifdef _MULTI_DATAMODEL
1822 			if (ddi_model_convert_from(mode & FMODELS) ==
1823 			    DDI_MODEL_ILP32) {
1824 				drm_radeon_tex_image_32_t image32;
1825 				image32.x = image->x;
1826 				image32.y = image->y;
1827 				image32.width = image->width;
1828 				image32.height = image->height;
1829 				image32.data = (uint32_t)(uintptr_t)image->data;
1830 				DRM_COPYTO_WITH_RETURN(tex->image, &image32,
1831 				    sizeof (image32));
1832 			} else {
1833 #endif
1834 				DRM_COPYTO_WITH_RETURN(tex->image, image,
1835 				    sizeof (*image));
1836 #ifdef _MULTI_DATAMODEL
1837 			}
1838 #endif
1839 			return (EAGAIN);
1840 		}
1841 
1842 		/*
1843 		 * Dispatch the indirect buffer.
1844 		 */
1845 		buffer = (u32 *)(uintptr_t)
1846 		    ((char *)dev->agp_buffer_map->handle + buf->offset);
1847 
1848 		dwords = size / 4;
1849 
1850 #define	RADEON_COPY_MT(_buf, _data, _width) \
1851 	do { \
1852 		if (DRM_COPY_FROM_USER(_buf, _data, (_width))) {\
1853 			DRM_ERROR("%d: EFAULT on pad, %d bytes\n", \
1854 			    __LINE__, (_width)); \
1855 			return (EFAULT); \
1856 		} \
1857 	} while (*"\0")
1858 
1859 		if (microtile) {
1860 			/*
1861 			 * texture micro tiling in use, minimum texture
1862 			 * width is thus 16 bytes. however, we cannot use
1863 			 * blitter directly for texture width < 64 bytes,
1864 			 * since minimum tex pitch is 64 bytes and we need
1865 			 * this to match the texture width, otherwise the
1866 			 * blitter will tile it wrong. Thus, tiling manually
1867 			 * in this case. Additionally, need to special case
1868 			 * tex height = 1, since our actual image will have
1869 			 * height 2 and we need to ensure we don't read
1870 			 * beyond the texture size from user space.
1871 			 */
1872 			if (tex->height == 1) {
1873 				if (tex_width >= 64 || tex_width <= 16) {
1874 					RADEON_COPY_MT(buffer, data,
1875 					    (int)(tex_width * sizeof (u32)));
1876 				} else if (tex_width == 32) {
1877 					RADEON_COPY_MT(buffer, data, 16);
1878 					RADEON_COPY_MT(buffer + 8,
1879 					    data + 16, 16);
1880 				}
1881 			} else if (tex_width >= 64 || tex_width == 16) {
1882 				RADEON_COPY_MT(buffer, data,
1883 				    (int)(dwords * sizeof (u32)));
1884 			} else if (tex_width < 16) {
1885 				for (i = 0; i < tex->height; i++) {
1886 					RADEON_COPY_MT(buffer, data, tex_width);
1887 					buffer += 4;
1888 					data += tex_width;
1889 				}
1890 			} else if (tex_width == 32) {
1891 				/*
1892 				 * TODO: make sure this works when not
1893 				 * fitting in one buffer
1894 				 *  (i.e. 32bytes x 2048...)
1895 				 */
1896 				for (i = 0; i < tex->height; i += 2) {
1897 					RADEON_COPY_MT(buffer, data, 16);
1898 					data += 16;
1899 					RADEON_COPY_MT(buffer + 8, data, 16);
1900 					data += 16;
1901 					RADEON_COPY_MT(buffer + 4, data, 16);
1902 					data += 16;
1903 					RADEON_COPY_MT(buffer + 12, data, 16);
1904 					data += 16;
1905 					buffer += 16;
1906 				}
1907 			}
1908 		} else {
1909 			if (tex_width >= 32) {
1910 				/*
1911 				 * Texture image width is larger than the
1912 				 * minimum, so we can upload it directly.
1913 				 */
1914 				RADEON_COPY_MT(buffer, data,
1915 				    (int)(dwords * sizeof (u32)));
1916 			} else {
1917 				/*
1918 				 * Texture image width is less than the minimum,
1919 				 * so we need to pad out each image scanline to
1920 				 * the minimum width.
1921 				 */
1922 				for (i = 0; i < tex->height; i++) {
1923 					RADEON_COPY_MT(buffer, data, tex_width);
1924 					buffer += 8;
1925 					data += tex_width;
1926 				}
1927 			}
1928 		}
1929 
1930 #undef RADEON_COPY_MT
1931 		buf->filp = fpriv;
1932 		buf->used = size;
1933 		offset = dev_priv->gart_buffers_offset + buf->offset;
1934 
1935 		BEGIN_RING(9);
1936 		OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 5));
1937 		OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
1938 		    RADEON_GMC_DST_PITCH_OFFSET_CNTL |
1939 		    RADEON_GMC_BRUSH_NONE |
1940 		    (format << 8) |
1941 		    RADEON_GMC_SRC_DATATYPE_COLOR |
1942 		    RADEON_ROP3_S |
1943 		    RADEON_DP_SRC_SOURCE_MEMORY |
1944 		    RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
1945 		OUT_RING((spitch << 22) | (offset >> 10));
1946 		OUT_RING((texpitch << 22) | (tex->offset >> 10));
1947 		OUT_RING(0);
1948 		OUT_RING((image->x << 16) | image->y);
1949 		OUT_RING((image->width << 16) | height);
1950 		RADEON_WAIT_UNTIL_2D_IDLE();
1951 		ADVANCE_RING();
1952 		COMMIT_RING();
1953 
1954 
1955 		radeon_cp_discard_buffer(dev, buf);
1956 
1957 		/* Update the input parameters for next time */
1958 		image->y += height;
1959 		image->height -= height;
1960 		image->data = (const u8 __user *)image->data + size;
1961 	} while (image->height > 0);
1962 
1963 	/*
1964 	 * Flush the pixel cache after the blit completes.  This ensures
1965 	 * the texture data is written out to memory before rendering
1966 	 * continues.
1967 	 */
1968 	BEGIN_RING(4);
1969 	RADEON_FLUSH_CACHE();
1970 	RADEON_WAIT_UNTIL_2D_IDLE();
1971 	ADVANCE_RING();
1972 	COMMIT_RING();
1973 	return (0);
1974 }
1975 
1976 static void radeon_cp_dispatch_stipple(drm_device_t *dev, u32 *stipple)
1977 {
1978 	drm_radeon_private_t *dev_priv = dev->dev_private;
1979 	int i;
1980 	RING_LOCALS;
1981 	DRM_DEBUG("\n");
1982 
1983 	BEGIN_RING(35);
1984 
1985 	OUT_RING(CP_PACKET0(RADEON_RE_STIPPLE_ADDR, 0));
1986 	OUT_RING(0x00000000);
1987 
1988 	OUT_RING(CP_PACKET0_TABLE(RADEON_RE_STIPPLE_DATA, 31));
1989 	for (i = 0; i < 32; i++) {
1990 		OUT_RING(stipple[i]);
1991 	}
1992 
1993 	ADVANCE_RING();
1994 }
1995 
1996 static void radeon_apply_surface_regs(int surf_index,
1997     drm_radeon_private_t *dev_priv)
1998 {
1999 	if (!dev_priv->mmio)
2000 		return;
2001 
2002 	(void) radeon_do_cp_idle(dev_priv);
2003 
2004 	RADEON_WRITE(RADEON_SURFACE0_INFO + 16 * surf_index,
2005 	    dev_priv->surfaces[surf_index].flags);
2006 	RADEON_WRITE(RADEON_SURFACE0_LOWER_BOUND + 16 * surf_index,
2007 	    dev_priv->surfaces[surf_index].lower);
2008 	RADEON_WRITE(RADEON_SURFACE0_UPPER_BOUND + 16 * surf_index,
2009 	    dev_priv->surfaces[surf_index].upper);
2010 }
2011 
2012 /*
2013  * Allocates a virtual surface
2014  * doesn't always allocate a real surface, will stretch an existing
2015  * surface when possible.
2016  *
2017  * Note that refcount can be at most 2, since during a free refcount=3
2018  * might mean we have to allocate a new surface which might not always
2019  * be available.
2020  * For example : we allocate three contigous surfaces ABC. If B is
2021  * freed, we suddenly need two surfaces to store A and C, which might
2022  * not always be available.
2023  */
2024 static int alloc_surface(drm_radeon_surface_alloc_t *new,
2025     drm_radeon_private_t *dev_priv, drm_file_t *filp)
2026 {
2027 	struct radeon_virt_surface *s;
2028 	int i;
2029 	int virt_surface_index;
2030 	uint32_t new_upper, new_lower;
2031 
2032 	new_lower = new->address;
2033 	new_upper = new_lower + new->size - 1;
2034 
2035 	/* sanity check */
2036 	if ((new_lower >= new_upper) || (new->flags == 0) || (new->size == 0) ||
2037 	    ((new_upper & RADEON_SURF_ADDRESS_FIXED_MASK) !=
2038 	    RADEON_SURF_ADDRESS_FIXED_MASK) ||
2039 	    ((new_lower & RADEON_SURF_ADDRESS_FIXED_MASK) != 0))
2040 		return (-1);
2041 
2042 	/* make sure there is no overlap with existing surfaces */
2043 	for (i = 0; i < RADEON_MAX_SURFACES; i++) {
2044 		if ((dev_priv->surfaces[i].refcount != 0) &&
2045 		    (((new_lower >= dev_priv->surfaces[i].lower) &&
2046 		    (new_lower < dev_priv->surfaces[i].upper)) ||
2047 		    ((new_lower < dev_priv->surfaces[i].lower) &&
2048 		    (new_upper > dev_priv->surfaces[i].lower)))) {
2049 			return (-1);
2050 		}
2051 	}
2052 
2053 	/* find a virtual surface */
2054 	for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++)
2055 		if (dev_priv->virt_surfaces[i].filp == 0)
2056 			break;
2057 	if (i == 2 * RADEON_MAX_SURFACES) {
2058 		return (-1);
2059 	}
2060 	virt_surface_index = i;
2061 
2062 	/* try to reuse an existing surface */
2063 	for (i = 0; i < RADEON_MAX_SURFACES; i++) {
2064 		/* extend before */
2065 		if ((dev_priv->surfaces[i].refcount == 1) &&
2066 		    (new->flags == dev_priv->surfaces[i].flags) &&
2067 		    (new_upper + 1 == dev_priv->surfaces[i].lower)) {
2068 			s = &(dev_priv->virt_surfaces[virt_surface_index]);
2069 			s->surface_index = i;
2070 			s->lower = new_lower;
2071 			s->upper = new_upper;
2072 			s->flags = new->flags;
2073 			s->filp = filp;
2074 			dev_priv->surfaces[i].refcount++;
2075 			dev_priv->surfaces[i].lower = s->lower;
2076 			radeon_apply_surface_regs(s->surface_index, dev_priv);
2077 			return (virt_surface_index);
2078 		}
2079 
2080 		/* extend after */
2081 		if ((dev_priv->surfaces[i].refcount == 1) &&
2082 		    (new->flags == dev_priv->surfaces[i].flags) &&
2083 		    (new_lower == dev_priv->surfaces[i].upper + 1)) {
2084 			s = &(dev_priv->virt_surfaces[virt_surface_index]);
2085 			s->surface_index = i;
2086 			s->lower = new_lower;
2087 			s->upper = new_upper;
2088 			s->flags = new->flags;
2089 			s->filp = filp;
2090 			dev_priv->surfaces[i].refcount++;
2091 			dev_priv->surfaces[i].upper = s->upper;
2092 			radeon_apply_surface_regs(s->surface_index, dev_priv);
2093 			return (virt_surface_index);
2094 		}
2095 	}
2096 
2097 	/* okay, we need a new one */
2098 	for (i = 0; i < RADEON_MAX_SURFACES; i++) {
2099 		if (dev_priv->surfaces[i].refcount == 0) {
2100 			s = &(dev_priv->virt_surfaces[virt_surface_index]);
2101 			s->surface_index = i;
2102 			s->lower = new_lower;
2103 			s->upper = new_upper;
2104 			s->flags = new->flags;
2105 			s->filp = filp;
2106 			dev_priv->surfaces[i].refcount = 1;
2107 			dev_priv->surfaces[i].lower = s->lower;
2108 			dev_priv->surfaces[i].upper = s->upper;
2109 			dev_priv->surfaces[i].flags = s->flags;
2110 			radeon_apply_surface_regs(s->surface_index, dev_priv);
2111 			return (virt_surface_index);
2112 		}
2113 	}
2114 
2115 	/* we didn't find anything */
2116 	return (-1);
2117 }
2118 
2119 static int
2120 free_surface(drm_file_t *filp, drm_radeon_private_t *dev_priv, int lower)
2121 {
2122 	struct radeon_virt_surface *s;
2123 	int i;
2124 
2125 	/* find the virtual surface */
2126 	for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++) {
2127 		s = &(dev_priv->virt_surfaces[i]);
2128 		if (s->filp) {
2129 			if ((lower == s->lower) && (filp == s->filp)) {
2130 				if (dev_priv->surfaces[s->surface_index].
2131 				    lower == s->lower)
2132 					dev_priv->surfaces[s->surface_index].
2133 					    lower = s->upper;
2134 
2135 				if (dev_priv->surfaces[s->surface_index].
2136 				    upper == s->upper)
2137 					dev_priv->surfaces[s->surface_index].
2138 					    upper = s->lower;
2139 
2140 				dev_priv->surfaces[s->surface_index].refcount--;
2141 				if (dev_priv->surfaces[s->surface_index].
2142 				    refcount == 0)
2143 					dev_priv->surfaces[s->surface_index].
2144 					    flags = 0;
2145 				s->filp = NULL;
2146 				radeon_apply_surface_regs(s->surface_index,
2147 				    dev_priv);
2148 				return (0);
2149 			}
2150 		}
2151 	}
2152 
2153 	return (1);
2154 }
2155 
2156 static void radeon_surfaces_release(drm_file_t *filp,
2157     drm_radeon_private_t *dev_priv)
2158 {
2159 	int i;
2160 
2161 	for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++) {
2162 		if (dev_priv->virt_surfaces[i].filp == filp)
2163 			(void) free_surface(filp, dev_priv,
2164 			    dev_priv->virt_surfaces[i].lower);
2165 	}
2166 }
2167 
2168 /*
2169  * IOCTL functions
2170  */
2171 /*ARGSUSED*/
2172 static int radeon_surface_alloc(DRM_IOCTL_ARGS)
2173 {
2174 	DRM_DEVICE;
2175 	drm_radeon_private_t *dev_priv = dev->dev_private;
2176 	drm_radeon_surface_alloc_t alloc;
2177 
2178 	if (!dev_priv) {
2179 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
2180 		return (EINVAL);
2181 	}
2182 
2183 	DRM_COPYFROM_WITH_RETURN(&alloc, (void *)data, sizeof (alloc));
2184 
2185 	if (alloc_surface(&alloc, dev_priv, fpriv) == -1)
2186 		return (EINVAL);
2187 	else
2188 		return (0);
2189 }
2190 
2191 /*ARGSUSED*/
2192 static int radeon_surface_free(DRM_IOCTL_ARGS)
2193 {
2194 	DRM_DEVICE;
2195 	drm_radeon_private_t *dev_priv = dev->dev_private;
2196 	drm_radeon_surface_free_t memfree;
2197 
2198 	if (!dev_priv) {
2199 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
2200 		return (EINVAL);
2201 	}
2202 
2203 	DRM_COPYFROM_WITH_RETURN(&memfree, (void *)data, sizeof (memfree));
2204 	if (free_surface(fpriv, dev_priv, memfree.address)) {
2205 		return (EINVAL);
2206 	}
2207 	else
2208 		return (0);
2209 }
2210 
2211 /*ARGSUSED*/
2212 static int radeon_cp_clear(DRM_IOCTL_ARGS)
2213 {
2214 	DRM_DEVICE;
2215 	drm_radeon_private_t *dev_priv = dev->dev_private;
2216 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2217 	drm_radeon_clear_t clear;
2218 	drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
2219 
2220 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2221 
2222 #ifdef _MULTI_DATAMODEL
2223 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
2224 		drm_radeon_clear_32_t	clear32;
2225 		DRM_COPYFROM_WITH_RETURN(&clear32, (void *)data,
2226 		    sizeof (clear32));
2227 		clear.flags = clear32.flags;
2228 		clear.clear_color = clear32.clear_color;
2229 		clear.clear_depth = clear32.clear_depth;
2230 		clear.color_mask = clear32.color_mask;
2231 		clear.depth_mask = clear32.depth_mask;
2232 		clear.depth_boxes = (void*)(uintptr_t)clear32.depth_boxes;
2233 	} else {
2234 #endif
2235 		DRM_COPYFROM_WITH_RETURN(&clear, (void *)data, sizeof (clear));
2236 #ifdef _MULTI_DATAMODEL
2237 	}
2238 #endif
2239 
2240 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2241 
2242 	if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2243 		sarea_priv->nbox = RADEON_NR_SAREA_CLIPRECTS;
2244 
2245 	if (DRM_COPY_FROM_USER(&depth_boxes, clear.depth_boxes,
2246 	    sarea_priv->nbox * sizeof (depth_boxes[0])))
2247 	return (EFAULT);
2248 
2249 	radeon_cp_dispatch_clear(dev, &clear, depth_boxes);
2250 
2251 	COMMIT_RING();
2252 	return (0);
2253 }
2254 
2255 /*
2256  * Not sure why this isn't set all the time:
2257  */
2258 static int radeon_do_init_pageflip(drm_device_t *dev)
2259 {
2260 	drm_radeon_private_t *dev_priv = dev->dev_private;
2261 	RING_LOCALS;
2262 
2263 	BEGIN_RING(6);
2264 	RADEON_WAIT_UNTIL_3D_IDLE();
2265 	OUT_RING(CP_PACKET0(RADEON_CRTC_OFFSET_CNTL, 0));
2266 	OUT_RING(RADEON_READ(RADEON_CRTC_OFFSET_CNTL) |
2267 	    RADEON_CRTC_OFFSET_FLIP_CNTL);
2268 	OUT_RING(CP_PACKET0(RADEON_CRTC2_OFFSET_CNTL, 0));
2269 	OUT_RING(RADEON_READ(RADEON_CRTC2_OFFSET_CNTL) |
2270 	    RADEON_CRTC_OFFSET_FLIP_CNTL);
2271 	ADVANCE_RING();
2272 
2273 	dev_priv->page_flipping = 1;
2274 	dev_priv->current_page = 0;
2275 	dev_priv->sarea_priv->pfCurrentPage = dev_priv->current_page;
2276 
2277 	return (0);
2278 }
2279 
2280 /*
2281  * Called whenever a client dies, from drm_release.
2282  * NOTE:  Lock isn't necessarily held when this is called!
2283  */
2284 static int radeon_do_cleanup_pageflip(drm_device_t *dev)
2285 {
2286 	drm_radeon_private_t *dev_priv = dev->dev_private;
2287 
2288 	if (dev_priv->current_page != 0)
2289 		radeon_cp_dispatch_flip(dev);
2290 
2291 	dev_priv->page_flipping = 0;
2292 	return (0);
2293 }
2294 
2295 /*
2296  * Swapping and flipping are different operations, need different ioctls.
2297  * They can & should be intermixed to support multiple 3d windows.
2298  */
2299 /*ARGSUSED*/
2300 static int radeon_cp_flip(DRM_IOCTL_ARGS)
2301 {
2302 	DRM_DEVICE;
2303 	drm_radeon_private_t *dev_priv = dev->dev_private;
2304 
2305 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2306 
2307 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2308 
2309 	if (!dev_priv->page_flipping)
2310 		(void) radeon_do_init_pageflip(dev);
2311 
2312 	radeon_cp_dispatch_flip(dev);
2313 
2314 	COMMIT_RING();
2315 	return (0);
2316 }
2317 
2318 /*ARGSUSED*/
2319 static int radeon_cp_swap(DRM_IOCTL_ARGS)
2320 {
2321 	DRM_DEVICE;
2322 	drm_radeon_private_t *dev_priv = dev->dev_private;
2323 	drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2324 
2325 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2326 
2327 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2328 
2329 	if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2330 		sarea_priv->nbox = RADEON_NR_SAREA_CLIPRECTS;
2331 
2332 	radeon_cp_dispatch_swap(dev);
2333 	dev_priv->sarea_priv->ctx_owner = 0;
2334 
2335 	COMMIT_RING();
2336 	return (0);
2337 }
2338 
2339 /*ARGSUSED*/
2340 static int radeon_cp_vertex(DRM_IOCTL_ARGS)
2341 {
2342 	DRM_DEVICE;
2343 	drm_radeon_private_t *dev_priv = dev->dev_private;
2344 	drm_radeon_sarea_t *sarea_priv;
2345 	drm_device_dma_t *dma = dev->dma;
2346 	drm_buf_t *buf;
2347 	drm_radeon_vertex_t vertex;
2348 	drm_radeon_tcl_prim_t prim;
2349 
2350 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2351 
2352 	if (!dev_priv) {
2353 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
2354 		return (EINVAL);
2355 	}
2356 
2357 	sarea_priv = dev_priv->sarea_priv;
2358 
2359 	DRM_COPYFROM_WITH_RETURN(&vertex, (void *)data, sizeof (vertex));
2360 
2361 	DRM_DEBUG("pid=%d index=%d count=%d discard=%d\n",
2362 	    DRM_CURRENTPID, vertex.idx, vertex.count, vertex.discard);
2363 
2364 	if (vertex.idx < 0 || vertex.idx >= dma->buf_count) {
2365 		DRM_ERROR("buffer index %d (of %d max)\n",
2366 		    vertex.idx, dma->buf_count - 1);
2367 		return (EINVAL);
2368 	}
2369 	if (vertex.prim < 0 || vertex.prim > RADEON_PRIM_TYPE_3VRT_LINE_LIST) {
2370 		DRM_ERROR("buffer prim %d\n", vertex.prim);
2371 		return (EINVAL);
2372 	}
2373 
2374 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2375 	VB_AGE_TEST_WITH_RETURN(dev_priv);
2376 
2377 	buf = dma->buflist[vertex.idx];
2378 
2379 	if (buf->filp != fpriv) {
2380 		DRM_ERROR("process %d using buffer owned by %p\n",
2381 		    DRM_CURRENTPID, buf->filp);
2382 		return (EINVAL);
2383 	}
2384 	if (buf->pending) {
2385 		DRM_ERROR("sending pending buffer %d\n", vertex.idx);
2386 		return (EINVAL);
2387 	}
2388 
2389 	/*
2390 	 * Build up a prim_t record:
2391 	 */
2392 	if (vertex.count) {
2393 		buf->used = vertex.count;	/* not used? */
2394 
2395 		if (sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS) {
2396 			if (radeon_emit_state(dev_priv, fpriv,
2397 			    &sarea_priv->context_state,
2398 			    sarea_priv->tex_state,
2399 			    sarea_priv->dirty)) {
2400 				DRM_ERROR("radeon_emit_state failed\n");
2401 				return (EINVAL);
2402 			}
2403 
2404 			sarea_priv->dirty &= ~(RADEON_UPLOAD_TEX0IMAGES |
2405 			    RADEON_UPLOAD_TEX1IMAGES |
2406 			    RADEON_UPLOAD_TEX2IMAGES |
2407 			    RADEON_REQUIRE_QUIESCENCE);
2408 		}
2409 
2410 		prim.start = 0;
2411 		prim.finish = vertex.count;	/* unused */
2412 		prim.prim = vertex.prim;
2413 		prim.numverts = vertex.count;
2414 		prim.vc_format = dev_priv->sarea_priv->vc_format;
2415 
2416 		radeon_cp_dispatch_vertex(dev, buf, &prim);
2417 	}
2418 
2419 	if (vertex.discard) {
2420 		radeon_cp_discard_buffer(dev, buf);
2421 	}
2422 
2423 	COMMIT_RING();
2424 	return (0);
2425 }
2426 
2427 /*ARGSUSED*/
2428 static int radeon_cp_indices(DRM_IOCTL_ARGS)
2429 {
2430 	DRM_DEVICE;
2431 	drm_radeon_private_t *dev_priv = dev->dev_private;
2432 	drm_radeon_sarea_t *sarea_priv;
2433 	drm_device_dma_t *dma = dev->dma;
2434 	drm_buf_t *buf;
2435 	drm_radeon_indices_t elts;
2436 	drm_radeon_tcl_prim_t prim;
2437 /*	int count; */
2438 
2439 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2440 
2441 	if (!dev_priv) {
2442 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
2443 		return (EINVAL);
2444 	}
2445 	sarea_priv = dev_priv->sarea_priv;
2446 
2447 	DRM_COPYFROM_WITH_RETURN(&elts, (void *)data, sizeof (elts));
2448 
2449 	DRM_DEBUG("pid=%d index=%d start=%d end=%d discard=%d\n",
2450 	    DRM_CURRENTPID, elts.idx, elts.start, elts.end, elts.discard);
2451 
2452 	if (elts.idx < 0 || elts.idx >= dma->buf_count) {
2453 		DRM_ERROR("buffer index %d (of %d max)\n",
2454 		    elts.idx, dma->buf_count - 1);
2455 		return (EINVAL);
2456 	}
2457 	if (elts.prim < 0 || elts.prim > RADEON_PRIM_TYPE_3VRT_LINE_LIST) {
2458 		DRM_ERROR("buffer prim %d\n", elts.prim);
2459 		return (EINVAL);
2460 	}
2461 
2462 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2463 	VB_AGE_TEST_WITH_RETURN(dev_priv);
2464 
2465 	buf = dma->buflist[elts.idx];
2466 
2467 	if (buf->filp != fpriv) {
2468 		DRM_ERROR("process %d using buffer owned by %p\n",
2469 		    DRM_CURRENTPID, buf->filp);
2470 		return (EINVAL);
2471 	}
2472 	if (buf->pending) {
2473 		DRM_ERROR("sending pending buffer %d\n", elts.idx);
2474 		return (EINVAL);
2475 	}
2476 
2477 /*	count = (elts.end - elts.start) / sizeof(u16); */
2478 	elts.start -= RADEON_INDEX_PRIM_OFFSET;
2479 
2480 	if (elts.start & 0x7) {
2481 		DRM_ERROR("misaligned buffer 0x%x\n", elts.start);
2482 		return (EINVAL);
2483 	}
2484 	if (elts.start < buf->used) {
2485 		DRM_ERROR("no header 0x%x - 0x%x\n", elts.start, buf->used);
2486 		return (EINVAL);
2487 	}
2488 
2489 	buf->used = elts.end;
2490 
2491 	if (sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS) {
2492 		if (radeon_emit_state(dev_priv, fpriv,
2493 		    &sarea_priv->context_state,
2494 		    sarea_priv->tex_state,
2495 		    sarea_priv->dirty)) {
2496 			DRM_ERROR("radeon_emit_state failed\n");
2497 			return (EINVAL);
2498 		}
2499 
2500 		sarea_priv->dirty &= ~(RADEON_UPLOAD_TEX0IMAGES |
2501 		    RADEON_UPLOAD_TEX1IMAGES |
2502 		    RADEON_UPLOAD_TEX2IMAGES |
2503 		    RADEON_REQUIRE_QUIESCENCE);
2504 	}
2505 
2506 	/*
2507 	 * Build up a prim_t record:
2508 	 */
2509 	prim.start = elts.start;
2510 	prim.finish = elts.end;
2511 	prim.prim = elts.prim;
2512 	prim.offset = 0;	/* offset from start of dma buffers */
2513 	prim.numverts = RADEON_MAX_VB_VERTS;	/* duh */
2514 	prim.vc_format = dev_priv->sarea_priv->vc_format;
2515 
2516 	radeon_cp_dispatch_indices(dev, buf, &prim);
2517 	if (elts.discard) {
2518 		radeon_cp_discard_buffer(dev, buf);
2519 	}
2520 
2521 	COMMIT_RING();
2522 	return (0);
2523 }
2524 
2525 /*ARGSUSED*/
2526 static int radeon_cp_texture(DRM_IOCTL_ARGS)
2527 {
2528 	DRM_DEVICE;
2529 	drm_radeon_private_t *dev_priv = dev->dev_private;
2530 	drm_radeon_texture_t tex;
2531 	drm_radeon_tex_image_t image;
2532 	int ret;
2533 
2534 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2535 
2536 #ifdef _MULTI_DATAMODEL
2537 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
2538 		drm_radeon_texture_32_t tex32;
2539 		drm_radeon_tex_image_32_t image32;
2540 
2541 		DRM_COPYFROM_WITH_RETURN(&tex32, (void *)data, sizeof (tex32));
2542 		if (tex32.image == 0) {
2543 			DRM_ERROR("null texture image!\n");
2544 			return (EINVAL);
2545 		}
2546 		if (DRM_COPY_FROM_USER(&image32,
2547 		    (void *)(uintptr_t)tex32.image, sizeof (image32))) {
2548 			cmn_err(CE_WARN, "copyin32 failed");
2549 			return (EFAULT);
2550 		}
2551 
2552 		tex.offset = tex32.offset;
2553 		tex.pitch = tex32.pitch;
2554 		tex.format = tex32.format;
2555 		tex.width = tex32.width;
2556 		tex.height = tex32.height;
2557 		tex.image = (void*)(uintptr_t)tex32.image;
2558 
2559 		image.x = image32.x;
2560 		image.y = image32.y;
2561 		image.width = image32.width;
2562 		image.height = image32.height;
2563 		image.data = (void*)(uintptr_t)image32.data;
2564 
2565 	} else {
2566 #endif
2567 		DRM_COPYFROM_WITH_RETURN(&tex, (void *)data, sizeof (tex));
2568 		if (tex.image == NULL) {
2569 			return (EINVAL);
2570 		}
2571 		if (DRM_COPY_FROM_USER(&image,
2572 		    (drm_radeon_tex_image_t *)tex.image, sizeof (image))) {
2573 			return (EFAULT);
2574 		}
2575 #ifdef _MULTI_DATAMODEL
2576 	}
2577 #endif
2578 
2579 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2580 	VB_AGE_TEST_WITH_RETURN(dev_priv);
2581 
2582 	ret = radeon_cp_dispatch_texture(fpriv, dev, &tex, &image, mode);
2583 
2584 	COMMIT_RING();
2585 	return (ret);
2586 }
2587 
2588 /*ARGSUSED*/
2589 static int radeon_cp_stipple(DRM_IOCTL_ARGS)
2590 {
2591 	DRM_DEVICE;
2592 	drm_radeon_private_t *dev_priv = dev->dev_private;
2593 	drm_radeon_stipple_t stipple;
2594 	u32 mask[32];
2595 
2596 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2597 
2598 #ifdef _MULTI_DATAMODEL
2599 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
2600 		drm_radeon_stipple_32_t stipple32;
2601 		DRM_COPYFROM_WITH_RETURN(&stipple32, (void *)data,
2602 		    sizeof (stipple32));
2603 		stipple.mask = (void *)(uintptr_t)stipple32.mask;
2604 	} else {
2605 #endif
2606 		DRM_COPYFROM_WITH_RETURN(&stipple, (void *)data,
2607 		    sizeof (stipple));
2608 #ifdef _MULTI_DATAMODEL
2609 	}
2610 #endif
2611 	if (DRM_COPY_FROM_USER(&mask, stipple.mask, 32 * sizeof (u32)))
2612 		return (EFAULT);
2613 
2614 
2615 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2616 
2617 	radeon_cp_dispatch_stipple(dev, mask);
2618 
2619 	COMMIT_RING();
2620 	return (0);
2621 }
2622 
2623 /*ARGSUSED*/
2624 static int radeon_cp_indirect(DRM_IOCTL_ARGS)
2625 {
2626 	DRM_DEVICE;
2627 	drm_radeon_private_t *dev_priv = dev->dev_private;
2628 	drm_device_dma_t *dma = dev->dma;
2629 	drm_buf_t *buf;
2630 	drm_radeon_indirect_t indirect;
2631 	RING_LOCALS;
2632 
2633 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2634 
2635 	if (!dev_priv) {
2636 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
2637 		return (EINVAL);
2638 	}
2639 
2640 	DRM_COPYFROM_WITH_RETURN(&indirect, (void *) data, sizeof (indirect));
2641 
2642 	DRM_DEBUG("indirect: idx=%d s=%d e=%d d=%d\n",
2643 	    indirect.idx, indirect.start, indirect.end, indirect.discard);
2644 
2645 	if (indirect.idx < 0 || indirect.idx >= dma->buf_count) {
2646 		DRM_ERROR("buffer index %d (of %d max)\n",
2647 		    indirect.idx, dma->buf_count - 1);
2648 		return (EINVAL);
2649 	}
2650 
2651 	buf = dma->buflist[indirect.idx];
2652 
2653 	if (buf->filp != fpriv) {
2654 		DRM_ERROR("process %d using buffer owned by %p\n",
2655 		    DRM_CURRENTPID, buf->filp);
2656 		return (EINVAL);
2657 	}
2658 	if (buf->pending) {
2659 		DRM_ERROR("sending pending buffer %d\n", indirect.idx);
2660 		return (EINVAL);
2661 	}
2662 
2663 	if (indirect.start < buf->used) {
2664 		DRM_ERROR("reusing indirect: start=0x%x actual=0x%x\n",
2665 		    indirect.start, buf->used);
2666 		return (EINVAL);
2667 	}
2668 
2669 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2670 	VB_AGE_TEST_WITH_RETURN(dev_priv);
2671 
2672 	buf->used = indirect.end;
2673 
2674 	/*
2675 	 * Wait for the 3D stream to idle before the indirect buffer
2676 	 * containing 2D acceleration commands is processed.
2677 	 */
2678 	BEGIN_RING(2);
2679 
2680 	RADEON_WAIT_UNTIL_3D_IDLE();
2681 
2682 	ADVANCE_RING();
2683 
2684 	/*
2685 	 * Dispatch the indirect buffer full of commands from the
2686 	 * X server.  This is insecure and is thus only available to
2687 	 * privileged clients.
2688 	 */
2689 	radeon_cp_dispatch_indirect(dev, buf, indirect.start, indirect.end);
2690 	if (indirect.discard) {
2691 		radeon_cp_discard_buffer(dev, buf);
2692 	}
2693 
2694 	COMMIT_RING();
2695 	return (0);
2696 }
2697 
2698 /*ARGSUSED*/
2699 static int radeon_cp_vertex2(DRM_IOCTL_ARGS)
2700 {
2701 	DRM_DEVICE;
2702 	drm_radeon_private_t *dev_priv = dev->dev_private;
2703 	drm_radeon_sarea_t *sarea_priv;
2704 	drm_device_dma_t *dma = dev->dma;
2705 	drm_buf_t *buf;
2706 	drm_radeon_vertex2_t vertex;
2707 	int i;
2708 	unsigned char laststate;
2709 
2710 	LOCK_TEST_WITH_RETURN(dev, fpriv);
2711 
2712 	if (!dev_priv) {
2713 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
2714 		return (EINVAL);
2715 	}
2716 
2717 	sarea_priv = dev_priv->sarea_priv;
2718 
2719 
2720 #ifdef _MULTI_DATAMODEL
2721 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
2722 		drm_radeon_vertex2_32_t	vertex32;
2723 
2724 		DRM_COPYFROM_WITH_RETURN(&vertex32, (void *) data,
2725 		    sizeof (vertex32));
2726 		vertex.idx = vertex32.idx;
2727 		vertex.discard = vertex32.discard;
2728 		vertex.nr_states = vertex32.nr_states;
2729 		vertex.state = (void *) (uintptr_t)vertex32.state;
2730 		vertex.nr_prims = vertex32.nr_prims;
2731 		vertex.prim = (void *)(uintptr_t)vertex32.prim;
2732 	} else {
2733 #endif
2734 		DRM_COPYFROM_WITH_RETURN(&vertex, (void *) data,
2735 		    sizeof (vertex));
2736 #ifdef _MULTI_DATAMODEL
2737 	}
2738 #endif
2739 
2740 	DRM_DEBUG("pid=%d index=%d discard=%d\n",
2741 	    DRM_CURRENTPID, vertex.idx, vertex.discard);
2742 
2743 	if (vertex.idx < 0 || vertex.idx >= dma->buf_count) {
2744 		DRM_ERROR("buffer index %d (of %d max)\n",
2745 		    vertex.idx, dma->buf_count - 1);
2746 		return (EINVAL);
2747 	}
2748 
2749 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
2750 	VB_AGE_TEST_WITH_RETURN(dev_priv);
2751 
2752 	buf = dma->buflist[vertex.idx];
2753 
2754 	if (buf->filp != fpriv) {
2755 		DRM_ERROR("process %d using buffer owned by %p\n",
2756 		    DRM_CURRENTPID, buf->filp);
2757 		return (EINVAL);
2758 	}
2759 
2760 	if (buf->pending) {
2761 		DRM_ERROR("sending pending buffer %d\n", vertex.idx);
2762 		return (EINVAL);
2763 	}
2764 
2765 	if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2766 		return (EINVAL);
2767 
2768 	for (laststate = 0xff, i = 0; i < vertex.nr_prims; i++) {
2769 		drm_radeon_prim_t prim;
2770 		drm_radeon_tcl_prim_t tclprim;
2771 
2772 		if (DRM_COPY_FROM_USER(&prim, &vertex.prim[i], sizeof (prim)))
2773 			return (EFAULT);
2774 
2775 		if (prim.stateidx != laststate) {
2776 			drm_radeon_state_t state;
2777 
2778 			if (DRM_COPY_FROM_USER(&state,
2779 			    &vertex.state[prim.stateidx], sizeof (state)))
2780 				return (EFAULT);
2781 
2782 			if (radeon_emit_state2(dev_priv, fpriv, &state)) {
2783 				DRM_ERROR("radeon_emit_state2 failed\n");
2784 				return (EINVAL);
2785 			}
2786 
2787 			laststate = prim.stateidx;
2788 		}
2789 
2790 		tclprim.start = prim.start;
2791 		tclprim.finish = prim.finish;
2792 		tclprim.prim = prim.prim;
2793 		tclprim.vc_format = prim.vc_format;
2794 
2795 		if (prim.prim & RADEON_PRIM_WALK_IND) {
2796 			tclprim.offset = prim.numverts * 64;
2797 			tclprim.numverts = RADEON_MAX_VB_VERTS;	/* duh */
2798 
2799 			radeon_cp_dispatch_indices(dev, buf, &tclprim);
2800 		} else {
2801 			tclprim.numverts = prim.numverts;
2802 			tclprim.offset = 0;	/* not used */
2803 
2804 			radeon_cp_dispatch_vertex(dev, buf, &tclprim);
2805 		}
2806 
2807 		if (sarea_priv->nbox == 1)
2808 			sarea_priv->nbox = 0;
2809 	}
2810 
2811 	if (vertex.discard) {
2812 		radeon_cp_discard_buffer(dev, buf);
2813 	}
2814 
2815 	COMMIT_RING();
2816 	return (0);
2817 }
2818 
2819 static int radeon_emit_packets(drm_radeon_private_t *dev_priv,
2820     drm_file_t *filp_priv, drm_radeon_cmd_header_t header,
2821     drm_radeon_kcmd_buffer_t *cmdbuf)
2822 {
2823 	int id = (int)header.packet.packet_id;
2824 	int sz, reg;
2825 	u32 *data = (u32 *)(uintptr_t)cmdbuf->buf;
2826 	RING_LOCALS;
2827 
2828 	if (id >= RADEON_MAX_STATE_PACKETS)
2829 		return (EINVAL);
2830 
2831 	sz = packet[id].len;
2832 	reg = packet[id].start;
2833 
2834 	if (sz * sizeof (int) > cmdbuf->bufsz) {
2835 		DRM_ERROR("Packet size provided larger than data provided\n");
2836 		return (EINVAL);
2837 	}
2838 
2839 	if (radeon_check_and_fixup_packets(dev_priv, filp_priv, id, data)) {
2840 		DRM_ERROR("Packet verification failed\n");
2841 		return (EINVAL);
2842 	}
2843 
2844 	BEGIN_RING(sz + 1);
2845 	OUT_RING(CP_PACKET0(reg, (sz - 1)));
2846 	OUT_RING_TABLE(data, sz);
2847 	ADVANCE_RING();
2848 
2849 	cmdbuf->buf += sz * sizeof (int);
2850 	cmdbuf->bufsz -= sz * sizeof (int);
2851 	return (0);
2852 }
2853 
2854 static inline int
2855 radeon_emit_scalars(drm_radeon_private_t *dev_priv,
2856     drm_radeon_cmd_header_t header, drm_radeon_kcmd_buffer_t *cmdbuf)
2857 {
2858 	int sz = header.scalars.count;
2859 	int start = header.scalars.offset;
2860 	int stride = header.scalars.stride;
2861 	RING_LOCALS;
2862 
2863 	BEGIN_RING(3 + sz);
2864 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
2865 	OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
2866 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
2867 	OUT_RING_TABLE(cmdbuf->buf, sz);
2868 	ADVANCE_RING();
2869 	cmdbuf->buf += sz * sizeof (int);
2870 	cmdbuf->bufsz -= sz * sizeof (int);
2871 	return (0);
2872 }
2873 
2874 /*
2875  * God this is ugly
2876  */
2877 static inline int
2878 radeon_emit_scalars2(drm_radeon_private_t *dev_priv,
2879     drm_radeon_cmd_header_t header, drm_radeon_kcmd_buffer_t *cmdbuf)
2880 {
2881 	int sz = header.scalars.count;
2882 	int start = ((unsigned int)header.scalars.offset) + 0x100;
2883 	int stride = header.scalars.stride;
2884 	RING_LOCALS;
2885 
2886 	BEGIN_RING(3 + sz);
2887 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
2888 	OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
2889 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
2890 	OUT_RING_TABLE(cmdbuf->buf, sz);
2891 	ADVANCE_RING();
2892 	cmdbuf->buf += sz * sizeof (int);
2893 	cmdbuf->bufsz -= sz * sizeof (int);
2894 	return (0);
2895 }
2896 
2897 static inline int
2898 radeon_emit_vectors(drm_radeon_private_t *dev_priv,
2899     drm_radeon_cmd_header_t header, drm_radeon_kcmd_buffer_t *cmdbuf)
2900 {
2901 	int sz = header.vectors.count;
2902 	int start = header.vectors.offset;
2903 	int stride = header.vectors.stride;
2904 	RING_LOCALS;
2905 
2906 	BEGIN_RING(5 + sz);
2907 	OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
2908 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
2909 	OUT_RING(start | (stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
2910 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
2911 	OUT_RING_TABLE(cmdbuf->buf, sz);
2912 	ADVANCE_RING();
2913 
2914 	cmdbuf->buf += sz * sizeof (int);
2915 	cmdbuf->bufsz -= sz * sizeof (int);
2916 	return (0);
2917 }
2918 
2919 static inline int
2920 radeon_emit_veclinear(drm_radeon_private_t *dev_priv,
2921     drm_radeon_cmd_header_t header, drm_radeon_kcmd_buffer_t *cmdbuf)
2922 {
2923 	int sz = header.veclinear.count * 4;
2924 	int start = header.veclinear.addr_lo | (header.veclinear.addr_hi << 8);
2925 	RING_LOCALS;
2926 
2927 		if (!sz)
2928 			return (0);
2929 		if (sz * 4 > cmdbuf->bufsz)
2930 			return (EINVAL);
2931 
2932 	BEGIN_RING(5 + sz);
2933 	OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
2934 	OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
2935 	OUT_RING(start | (1 << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
2936 	OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
2937 	OUT_RING_TABLE(cmdbuf->buf, sz);
2938 	ADVANCE_RING();
2939 
2940 	cmdbuf->buf += sz * sizeof (int);
2941 	cmdbuf->bufsz -= sz * sizeof (int);
2942 	return (0);
2943 }
2944 
2945 static int
2946 radeon_emit_packet3(drm_device_t *dev, drm_file_t *filp_priv,
2947     drm_radeon_kcmd_buffer_t *cmdbuf)
2948 {
2949 	drm_radeon_private_t *dev_priv = dev->dev_private;
2950 	unsigned int cmdsz;
2951 	int ret;
2952 	RING_LOCALS;
2953 
2954 
2955 	if ((ret = radeon_check_and_fixup_packet3(dev_priv,
2956 	    filp_priv, cmdbuf, &cmdsz))) {
2957 		DRM_ERROR("Packet verification failed\n");
2958 		return (ret);
2959 	}
2960 
2961 	BEGIN_RING(cmdsz);
2962 	OUT_RING_TABLE(cmdbuf->buf, cmdsz);
2963 	ADVANCE_RING();
2964 
2965 	cmdbuf->buf += cmdsz * 4;
2966 	cmdbuf->bufsz -= cmdsz * 4;
2967 	return (0);
2968 }
2969 
2970 static int radeon_emit_packet3_cliprect(drm_device_t *dev,
2971 					drm_file_t *filp_priv,
2972 					drm_radeon_kcmd_buffer_t *cmdbuf,
2973 					int orig_nbox)
2974 {
2975 	drm_radeon_private_t *dev_priv = dev->dev_private;
2976 	drm_clip_rect_t box;
2977 	unsigned int cmdsz;
2978 	int ret;
2979 	drm_clip_rect_t __user *boxes = cmdbuf->boxes;
2980 	int i = 0;
2981 	RING_LOCALS;
2982 
2983 	if ((ret = radeon_check_and_fixup_packet3(dev_priv,
2984 	    filp_priv, cmdbuf, &cmdsz))) {
2985 		DRM_ERROR("Packet verification failed\n");
2986 		return (ret);
2987 	}
2988 
2989 	if (!orig_nbox)
2990 		goto out;
2991 
2992 	do {
2993 		if (i < cmdbuf->nbox) {
2994 			if (DRM_COPY_FROM_USER(&box, &boxes[i], sizeof (box)))
2995 				return (EFAULT);
2996 			/*
2997 			 * FIXME The second and subsequent times round
2998 			 * this loop, send a WAIT_UNTIL_3D_IDLE before
2999 			 * calling emit_clip_rect(). This fixes a
3000 			 * lockup on fast machines when sending
3001 			 * several cliprects with a cmdbuf, as when
3002 			 * waving a 2D window over a 3D
3003 			 * window. Something in the commands from user
3004 			 * space seems to hang the card when they're
3005 			 * sent several times in a row. That would be
3006 			 * the correct place to fix it but this works
3007 			 * around it until I can figure that out - Tim
3008 			 * Smith
3009 			 */
3010 			if (i) {
3011 				BEGIN_RING(2);
3012 				RADEON_WAIT_UNTIL_3D_IDLE();
3013 				ADVANCE_RING();
3014 			}
3015 			radeon_emit_clip_rect(dev_priv, &box);
3016 		}
3017 
3018 		BEGIN_RING(cmdsz);
3019 		OUT_RING_TABLE(cmdbuf->buf, cmdsz);
3020 		ADVANCE_RING();
3021 
3022 	} while (++i < cmdbuf->nbox);
3023 	if (cmdbuf->nbox == 1)
3024 		cmdbuf->nbox = 0;
3025 
3026 out:
3027 	cmdbuf->buf += cmdsz * 4;
3028 	cmdbuf->bufsz -= cmdsz * 4;
3029 	return (0);
3030 }
3031 
3032 static int
3033 radeon_emit_wait(drm_device_t *dev, int flags)
3034 {
3035 	drm_radeon_private_t *dev_priv = dev->dev_private;
3036 	RING_LOCALS;
3037 
3038 	DRM_DEBUG("%s: %x\n", __FUNCTION__, flags);
3039 	switch (flags) {
3040 	case RADEON_WAIT_2D:
3041 		BEGIN_RING(2);
3042 		RADEON_WAIT_UNTIL_2D_IDLE();
3043 		ADVANCE_RING();
3044 		break;
3045 	case RADEON_WAIT_3D:
3046 		BEGIN_RING(2);
3047 		RADEON_WAIT_UNTIL_3D_IDLE();
3048 		ADVANCE_RING();
3049 		break;
3050 	case RADEON_WAIT_2D | RADEON_WAIT_3D:
3051 		BEGIN_RING(2);
3052 		RADEON_WAIT_UNTIL_IDLE();
3053 		ADVANCE_RING();
3054 		break;
3055 	default:
3056 		return (EINVAL);
3057 	}
3058 
3059 	return (0);
3060 }
3061 
3062 /*ARGSUSED*/
3063 static int radeon_cp_cmdbuf(DRM_IOCTL_ARGS)
3064 {
3065 	DRM_DEVICE;
3066 	drm_radeon_private_t *dev_priv = dev->dev_private;
3067 	drm_device_dma_t *dma = dev->dma;
3068 	drm_buf_t *buf = NULL;
3069 	int idx;
3070 	drm_radeon_kcmd_buffer_t cmdbuf;
3071 	drm_radeon_cmd_header_t header;
3072 	int orig_nbox, orig_bufsz;
3073 	char *kbuf = NULL;
3074 
3075 	LOCK_TEST_WITH_RETURN(dev, fpriv);
3076 
3077 	if (!dev_priv) {
3078 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
3079 		return (EINVAL);
3080 	}
3081 
3082 #ifdef _MULTI_DATAMODEL
3083 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
3084 		drm_radeon_kcmd_buffer_32_t cmdbuf32;
3085 
3086 		DRM_COPYFROM_WITH_RETURN(&cmdbuf32, (void *)data,
3087 		    sizeof (cmdbuf32));
3088 		cmdbuf.bufsz = cmdbuf32.bufsz;
3089 		cmdbuf.buf = (void *)(uintptr_t)cmdbuf32.buf;
3090 		cmdbuf.nbox = cmdbuf32.nbox;
3091 		cmdbuf.boxes = (void *)(uintptr_t)cmdbuf32.boxes;
3092 	} else {
3093 #endif
3094 		DRM_COPYFROM_WITH_RETURN(&cmdbuf, (void *) data,
3095 		    sizeof (cmdbuf));
3096 #ifdef _MULTI_DATAMODEL
3097 	}
3098 #endif
3099 	RING_SPACE_TEST_WITH_RETURN(dev_priv);
3100 	VB_AGE_TEST_WITH_RETURN(dev_priv);
3101 
3102 	if (cmdbuf.bufsz > 64 * 1024 || cmdbuf.bufsz < 0) {
3103 		return (EINVAL);
3104 	}
3105 
3106 	/*
3107 	 * Allocate an in-kernel area and copy in the cmdbuf. Do this
3108 	 * to avoid races between checking values and using those values
3109 	 * in other code, and simply to avoid a lot of function calls
3110 	 * to copy in data.
3111 	 */
3112 	orig_bufsz = cmdbuf.bufsz;
3113 	if (orig_bufsz != 0) {
3114 		kbuf = drm_alloc(cmdbuf.bufsz, DRM_MEM_DRIVER);
3115 		if (kbuf == NULL)
3116 			return (ENOMEM);
3117 		if (DRM_COPY_FROM_USER(kbuf, (void *)cmdbuf.buf,
3118 		    cmdbuf.bufsz)) {
3119 			drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
3120 			return (EFAULT);
3121 		}
3122 		cmdbuf.buf = kbuf;
3123 	}
3124 
3125 	orig_nbox = cmdbuf.nbox;
3126 
3127 	if (dev_priv->microcode_version == UCODE_R300) {
3128 		int temp;
3129 		temp = r300_do_cp_cmdbuf(dev, fpriv, &cmdbuf);
3130 
3131 		if (orig_bufsz != 0)
3132 			drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
3133 
3134 		return (temp);
3135 	}
3136 
3137 	/* microcode_version != r300 */
3138 	while (cmdbuf.bufsz >= sizeof (header)) {
3139 
3140 		header.i = *(int *)(uintptr_t)cmdbuf.buf;
3141 		cmdbuf.buf += sizeof (header);
3142 		cmdbuf.bufsz -= sizeof (header);
3143 
3144 		switch (header.header.cmd_type) {
3145 		case RADEON_CMD_PACKET:
3146 			DRM_DEBUG("RADEON_CMD_PACKET\n");
3147 			if (radeon_emit_packets
3148 			    (dev_priv, fpriv, header, &cmdbuf)) {
3149 				DRM_ERROR("radeon_emit_packets failed\n");
3150 				goto err;
3151 			}
3152 			break;
3153 
3154 		case RADEON_CMD_SCALARS:
3155 			DRM_DEBUG("RADEON_CMD_SCALARS\n");
3156 			if (radeon_emit_scalars(dev_priv, header, &cmdbuf)) {
3157 				DRM_ERROR("radeon_emit_scalars failed\n");
3158 				goto err;
3159 			}
3160 			break;
3161 
3162 		case RADEON_CMD_VECTORS:
3163 			DRM_DEBUG("RADEON_CMD_VECTORS\n");
3164 			if (radeon_emit_vectors(dev_priv, header, &cmdbuf)) {
3165 				DRM_ERROR("radeon_emit_vectors failed\n");
3166 				goto err;
3167 			}
3168 			break;
3169 
3170 		case RADEON_CMD_DMA_DISCARD:
3171 			DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
3172 			idx = header.dma.buf_idx;
3173 			if (idx < 0 || idx >= dma->buf_count) {
3174 				DRM_ERROR("buffer index %d (of %d max)\n",
3175 				    idx, dma->buf_count - 1);
3176 				goto err;
3177 			}
3178 
3179 			buf = dma->buflist[idx];
3180 			if (buf->filp != fpriv || buf->pending) {
3181 				DRM_ERROR("bad buffer %p %p %d\n",
3182 				    buf->filp, fpriv, buf->pending);
3183 				goto err;
3184 			}
3185 
3186 			radeon_cp_discard_buffer(dev, buf);
3187 			break;
3188 
3189 		case RADEON_CMD_PACKET3:
3190 			DRM_DEBUG("RADEON_CMD_PACKET3\n");
3191 			if (radeon_emit_packet3(dev, fpriv, &cmdbuf)) {
3192 				DRM_ERROR("radeon_emit_packet3 failed\n");
3193 				goto err;
3194 			}
3195 			break;
3196 
3197 		case RADEON_CMD_PACKET3_CLIP:
3198 			DRM_DEBUG("RADEON_CMD_PACKET3_CLIP\n");
3199 			if (radeon_emit_packet3_cliprect
3200 			    (dev, fpriv, &cmdbuf, orig_nbox)) {
3201 				DRM_ERROR("radeon_emit_packet3_clip failed\n");
3202 				goto err;
3203 			}
3204 			break;
3205 
3206 		case RADEON_CMD_SCALARS2:
3207 			DRM_DEBUG("RADEON_CMD_SCALARS2\n");
3208 			if (radeon_emit_scalars2(dev_priv, header, &cmdbuf)) {
3209 				DRM_ERROR("radeon_emit_scalars2 failed\n");
3210 				goto err;
3211 			}
3212 			break;
3213 
3214 		case RADEON_CMD_WAIT:
3215 			DRM_DEBUG("RADEON_CMD_WAIT\n");
3216 			if (radeon_emit_wait(dev, header.wait.flags)) {
3217 				DRM_ERROR("radeon_emit_wait failed\n");
3218 				goto err;
3219 			}
3220 			break;
3221 		case RADEON_CMD_VECLINEAR:
3222 			DRM_DEBUG("RADEON_CMD_VECLINEAR\n");
3223 			if (radeon_emit_veclinear(dev_priv, header, &cmdbuf)) {
3224 				DRM_ERROR("radeon_emit_veclinear failed\n");
3225 				goto err;
3226 			}
3227 			break;
3228 
3229 		default:
3230 			DRM_ERROR("bad cmd_type %d at %p\n",
3231 			    header.header.cmd_type,
3232 			    cmdbuf.buf - sizeof (header));
3233 			goto err;
3234 		}
3235 	}
3236 
3237 	if (orig_bufsz != 0)
3238 		drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
3239 
3240 	COMMIT_RING();
3241 	return (0);
3242 
3243 err:
3244 	if (orig_bufsz != 0)
3245 		drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
3246 	return (EINVAL);
3247 }
3248 
3249 /*ARGSUSED*/
3250 static int radeon_cp_getparam(DRM_IOCTL_ARGS)
3251 {
3252 	DRM_DEVICE;
3253 	drm_radeon_private_t *dev_priv = dev->dev_private;
3254 	drm_radeon_getparam_t param;
3255 	int value;
3256 
3257 	if (!dev_priv) {
3258 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
3259 		return (EINVAL);
3260 	}
3261 
3262 #ifdef _MULTI_DATAMODEL
3263 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
3264 		drm_radeon_getparam_32_t param32;
3265 
3266 		DRM_COPYFROM_WITH_RETURN(&param32,
3267 		    (drm_radeon_getparam_32_t *)data, sizeof (param32));
3268 		param.param = param32.param;
3269 		param.value = (void *)(uintptr_t)param32.value;
3270 	} else {
3271 #endif
3272 		DRM_COPYFROM_WITH_RETURN(&param,
3273 		    (drm_radeon_getparam_t *)data, sizeof (param));
3274 #ifdef _MULTI_DATAMODEL
3275 	}
3276 #endif
3277 	DRM_DEBUG("pid=%d\n", DRM_CURRENTPID);
3278 
3279 	switch (param.param) {
3280 	case RADEON_PARAM_GART_BUFFER_OFFSET:
3281 		value = dev_priv->gart_buffers_offset;
3282 		break;
3283 	case RADEON_PARAM_LAST_FRAME:
3284 		dev_priv->stats.last_frame_reads++;
3285 		value = GET_SCRATCH(0);
3286 		break;
3287 	case RADEON_PARAM_LAST_DISPATCH:
3288 		value = GET_SCRATCH(1);
3289 		break;
3290 	case RADEON_PARAM_LAST_CLEAR:
3291 		dev_priv->stats.last_clear_reads++;
3292 		value = GET_SCRATCH(2);
3293 		break;
3294 	case RADEON_PARAM_IRQ_NR:
3295 		value = dev->irq;
3296 		break;
3297 	case RADEON_PARAM_GART_BASE:
3298 		value = dev_priv->gart_vm_start;
3299 		break;
3300 	case RADEON_PARAM_REGISTER_HANDLE:
3301 		value = dev_priv->mmio->offset;
3302 		break;
3303 	case RADEON_PARAM_STATUS_HANDLE:
3304 		value = dev_priv->ring_rptr_offset;
3305 		break;
3306 #ifndef __LP64__
3307 		/*
3308 		 * This ioctl() doesn't work on 64-bit platforms because
3309 		 * hw_lock is a pointer which can't fit into an int-sized
3310 		 * variable.  According to Michel Dänzer, the ioctl) is
3311 		 * only used on embedded platforms, so not supporting it
3312 		 * shouldn't be a problem.  If the same functionality is
3313 		 * needed on 64-bit platforms, a new ioctl() would have
3314 		 * to be added, so backwards-compatibility for the embedded
3315 		 * platforms can be maintained.  --davidm 4-Feb-2004.
3316 		 */
3317 	case RADEON_PARAM_SAREA_HANDLE:
3318 		/* The lock is the first dword in the sarea. */
3319 		value = (long)dev->lock.hw_lock;
3320 		break;
3321 #endif
3322 	case RADEON_PARAM_GART_TEX_HANDLE:
3323 		value = dev_priv->gart_textures_offset;
3324 		break;
3325 	case RADEON_PARAM_SCRATCH_OFFSET:
3326 		if (!dev_priv->writeback_works)
3327 			return (EINVAL);
3328 		value = RADEON_SCRATCH_REG_OFFSET;
3329 		break;
3330 
3331 	case RADEON_PARAM_CARD_TYPE:
3332 		if (dev_priv->flags & RADEON_IS_PCIE)
3333 			value = RADEON_CARD_PCIE;
3334 		else if (dev_priv->flags & RADEON_IS_AGP)
3335 			value = RADEON_CARD_AGP;
3336 		else
3337 			value = RADEON_CARD_PCI;
3338 		break;
3339 	case RADEON_PARAM_VBLANK_CRTC:
3340 		value = radeon_vblank_crtc_get(dev);
3341 		break;
3342 	default:
3343 		return (EINVAL);
3344 	}
3345 
3346 	if (DRM_COPY_TO_USER(param.value, &value, sizeof (int))) {
3347 		DRM_ERROR("copy_to_user\n");
3348 		return (EFAULT);
3349 	}
3350 	return (0);
3351 }
3352 
3353 /*ARGSUSED*/
3354 static int radeon_cp_setparam(DRM_IOCTL_ARGS)
3355 {
3356 	DRM_DEVICE;
3357 	drm_radeon_private_t *dev_priv = dev->dev_private;
3358 	drm_radeon_setparam_t sp;
3359 	struct drm_radeon_driver_file_fields *radeon_priv;
3360 
3361 	if (!dev_priv) {
3362 		DRM_ERROR("%s called with no initialization\n", __FUNCTION__);
3363 		return (EINVAL);
3364 	}
3365 
3366 #ifdef _MULTI_DATAMODEL
3367 	if (ddi_model_convert_from(mode & FMODELS) == DDI_MODEL_ILP32) {
3368 		drm_radeon_setparam_32_t sp32;
3369 
3370 		DRM_COPYFROM_WITH_RETURN(&sp32, (void *) data, sizeof (sp32));
3371 		sp.param = sp32.param;
3372 		sp.value = sp32.value;
3373 	} else {
3374 #endif
3375 	DRM_COPYFROM_WITH_RETURN(&sp, (void *) data, sizeof (sp));
3376 #ifdef _MULTI_DATAMODEL
3377 	}
3378 #endif
3379 	switch (sp.param) {
3380 	case RADEON_SETPARAM_FB_LOCATION:
3381 		radeon_priv = fpriv->driver_priv;
3382 		radeon_priv->radeon_fb_delta = dev_priv->fb_location - sp.value;
3383 		break;
3384 	case RADEON_SETPARAM_SWITCH_TILING:
3385 		if (sp.value == 0) {
3386 			DRM_DEBUG("color tiling disabled\n");
3387 			dev_priv->front_pitch_offset &= ~RADEON_DST_TILE_MACRO;
3388 			dev_priv->back_pitch_offset &= ~RADEON_DST_TILE_MACRO;
3389 			dev_priv->sarea_priv->tiling_enabled = 0;
3390 		} else if (sp.value == 1) {
3391 			DRM_DEBUG("color tiling enabled\n");
3392 			dev_priv->front_pitch_offset |= RADEON_DST_TILE_MACRO;
3393 			dev_priv->back_pitch_offset |= RADEON_DST_TILE_MACRO;
3394 			dev_priv->sarea_priv->tiling_enabled = 1;
3395 		}
3396 		break;
3397 	case RADEON_SETPARAM_PCIGART_LOCATION:
3398 		dev_priv->pcigart_offset = (unsigned long)sp.value;
3399 		break;
3400 	case RADEON_SETPARAM_NEW_MEMMAP:
3401 		dev_priv->new_memmap = (int)sp.value;
3402 		break;
3403 	case RADEON_SETPARAM_VBLANK_CRTC:
3404 		return (radeon_vblank_crtc_set(dev, sp.value));
3405 	default:
3406 		DRM_DEBUG("Invalid parameter %d\n", sp.param);
3407 		return (EINVAL);
3408 	}
3409 
3410 	return (0);
3411 }
3412 
3413 /*
3414  * When a client dies:
3415  *    - Check for and clean up flipped page state
3416  *    - Free any alloced GART memory.
3417  *    - Free any alloced radeon surfaces.
3418  *
3419  * DRM infrastructure takes care of reclaiming dma buffers.
3420  */
3421 void
3422 radeon_driver_preclose(drm_device_t *dev, drm_file_t *filp)
3423 {
3424 	if (dev->dev_private) {
3425 		drm_radeon_private_t *dev_priv = dev->dev_private;
3426 		if (dev_priv->page_flipping) {
3427 			(void) radeon_do_cleanup_pageflip(dev);
3428 		}
3429 		radeon_mem_release(filp, dev_priv->gart_heap);
3430 		radeon_mem_release(filp, dev_priv->fb_heap);
3431 		radeon_surfaces_release(filp, dev_priv);
3432 	}
3433 }
3434 
3435 void
3436 radeon_driver_lastclose(drm_device_t *dev)
3437 {
3438 	radeon_do_release(dev);
3439 }
3440 
3441 int
3442 radeon_driver_open(drm_device_t *dev, drm_file_t *filp_priv)
3443 {
3444 	drm_radeon_private_t *dev_priv = dev->dev_private;
3445 	struct drm_radeon_driver_file_fields *radeon_priv;
3446 
3447 	radeon_priv =
3448 	    (struct drm_radeon_driver_file_fields *)
3449 	    drm_alloc(sizeof (*radeon_priv), DRM_MEM_FILES);
3450 
3451 	if (!radeon_priv)
3452 		return (-ENOMEM);
3453 
3454 	filp_priv->driver_priv = radeon_priv;
3455 
3456 	if (dev_priv)
3457 		radeon_priv->radeon_fb_delta = dev_priv->fb_location;
3458 	else
3459 		radeon_priv->radeon_fb_delta = 0;
3460 	return (0);
3461 }
3462 
3463 /*ARGSUSED*/
3464 void
3465 radeon_driver_postclose(drm_device_t *dev, drm_file_t *filp_priv)
3466 {
3467 	struct drm_radeon_driver_file_fields *radeon_priv =
3468 	    filp_priv->driver_priv;
3469 
3470 	drm_free(radeon_priv, sizeof (* radeon_priv), DRM_MEM_FILES);
3471 }
3472 
3473 drm_ioctl_desc_t radeon_ioctls[] = {
3474 	[DRM_IOCTL_NR(DRM_RADEON_CP_INIT)] =
3475 	    {radeon_cp_init, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
3476 	[DRM_IOCTL_NR(DRM_RADEON_CP_START)] =
3477 	    {radeon_cp_start, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
3478 	[DRM_IOCTL_NR(DRM_RADEON_CP_STOP)] =
3479 	    {radeon_cp_stop, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
3480 	[DRM_IOCTL_NR(DRM_RADEON_CP_RESET)] =
3481 	    {radeon_cp_reset, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
3482 	[DRM_IOCTL_NR(DRM_RADEON_CP_IDLE)] =
3483 	    {radeon_cp_idle, DRM_AUTH},
3484 	[DRM_IOCTL_NR(DRM_RADEON_CP_RESUME)] =
3485 	    {radeon_cp_resume, DRM_AUTH},
3486 	[DRM_IOCTL_NR(DRM_RADEON_RESET)] =
3487 	    {radeon_engine_reset, DRM_AUTH},
3488 	[DRM_IOCTL_NR(DRM_RADEON_FULLSCREEN)] =
3489 	    {radeon_fullscreen, DRM_AUTH},
3490 	[DRM_IOCTL_NR(DRM_RADEON_SWAP)] =
3491 	    {radeon_cp_swap, DRM_AUTH},
3492 	[DRM_IOCTL_NR(DRM_RADEON_CLEAR)] =
3493 	    {radeon_cp_clear, DRM_AUTH},
3494 	[DRM_IOCTL_NR(DRM_RADEON_VERTEX)] =
3495 	    {radeon_cp_vertex, DRM_AUTH},
3496 	[DRM_IOCTL_NR(DRM_RADEON_INDICES)] =
3497 	    {radeon_cp_indices, DRM_AUTH},
3498 	[DRM_IOCTL_NR(DRM_RADEON_TEXTURE)] =
3499 	    {radeon_cp_texture, DRM_AUTH},
3500 	[DRM_IOCTL_NR(DRM_RADEON_STIPPLE)] =
3501 	    {radeon_cp_stipple, DRM_AUTH},
3502 	[DRM_IOCTL_NR(DRM_RADEON_INDIRECT)] =
3503 	    {radeon_cp_indirect, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
3504 	[DRM_IOCTL_NR(DRM_RADEON_VERTEX2)] =
3505 	    {radeon_cp_vertex2, DRM_AUTH},
3506 	[DRM_IOCTL_NR(DRM_RADEON_CMDBUF)] =
3507 	    {radeon_cp_cmdbuf, DRM_AUTH},
3508 	[DRM_IOCTL_NR(DRM_RADEON_GETPARAM)] =
3509 	    {radeon_cp_getparam, DRM_AUTH},
3510 	[DRM_IOCTL_NR(DRM_RADEON_FLIP)] =
3511 	    {radeon_cp_flip, DRM_AUTH},
3512 	[DRM_IOCTL_NR(DRM_RADEON_ALLOC)] =
3513 	    {radeon_mem_alloc, DRM_AUTH},
3514 	[DRM_IOCTL_NR(DRM_RADEON_FREE)] =
3515 	    {radeon_mem_free, DRM_AUTH},
3516 	[DRM_IOCTL_NR(DRM_RADEON_INIT_HEAP)] =
3517 	    {radeon_mem_init_heap, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY},
3518 	[DRM_IOCTL_NR(DRM_RADEON_IRQ_EMIT)] =
3519 	    {radeon_irq_emit, DRM_AUTH},
3520 	[DRM_IOCTL_NR(DRM_RADEON_IRQ_WAIT)] =
3521 	    {radeon_irq_wait, DRM_AUTH},
3522 	[DRM_IOCTL_NR(DRM_RADEON_SETPARAM)] =
3523 	    {radeon_cp_setparam, DRM_AUTH},
3524 	[DRM_IOCTL_NR(DRM_RADEON_SURF_ALLOC)] =
3525 	    {radeon_surface_alloc, DRM_AUTH},
3526 	[DRM_IOCTL_NR(DRM_RADEON_SURF_FREE)] =
3527 	    {radeon_surface_free, DRM_AUTH}
3528 };
3529 
3530 int radeon_max_ioctl = DRM_ARRAY_SIZE(radeon_ioctls);
3531