1 /* radeon_state.c -- State support for Radeon -*- linux-c -*- */
3 * Copyright 2000 VA Linux Systems, Inc., Fremont, California.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
21 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
22 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
23 * DEALINGS IN THE SOFTWARE.
26 * Gareth Hughes <gareth@valinux.com>
27 * Kevin E. Martin <martin@valinux.com>
32 #include "drm_sarea.h"
33 #include "radeon_drm.h"
34 #include "radeon_drv.h"
36 /* ================================================================
37 * Helper functions for client state checking and fixup
40 static __inline__ int radeon_check_and_fixup_offset(drm_radeon_private_t *
42 struct drm_file * file_priv,
46 u32 fb_end = dev_priv->fb_location + dev_priv->fb_size - 1;
47 struct drm_radeon_driver_file_fields *radeon_priv;
49 /* Hrm ... the story of the offset ... So this function converts
50 * the various ideas of what userland clients might have for an
51 * offset in the card address space into an offset into the card
52 * address space :) So with a sane client, it should just keep
53 * the value intact and just do some boundary checking. However,
54 * not all clients are sane. Some older clients pass us 0 based
55 * offsets relative to the start of the framebuffer and some may
56 * assume the AGP aperture it appended to the framebuffer, so we
57 * try to detect those cases and fix them up.
59 * Note: It might be a good idea here to make sure the offset lands
60 * in some "allowed" area to protect things like the PCIE GART...
63 /* First, the best case, the offset already lands in either the
64 * framebuffer or the GART mapped space
66 if (radeon_check_offset(dev_priv, off))
69 /* Ok, that didn't happen... now check if we have a zero based
70 * offset that fits in the framebuffer + gart space, apply the
71 * magic offset we get from SETPARAM or calculated from fb_location
73 if (off < (dev_priv->fb_size + dev_priv->gart_size)) {
74 radeon_priv = file_priv->driver_priv;
75 off += radeon_priv->radeon_fb_delta;
78 /* Finally, assume we aimed at a GART offset if beyond the fb */
80 off = off - fb_end - 1 + dev_priv->gart_vm_start;
82 /* Now recheck and fail if out of bounds */
83 if (radeon_check_offset(dev_priv, off)) {
84 DRM_DEBUG("offset fixed up to 0x%x\n", (unsigned int)off);
91 static __inline__ int radeon_check_and_fixup_packets(drm_radeon_private_t *
93 struct drm_file *file_priv,
98 case RADEON_EMIT_PP_MISC:
99 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
100 &data[(RADEON_RB3D_DEPTHOFFSET - RADEON_PP_MISC) / 4])) {
101 DRM_ERROR("Invalid depth buffer offset\n");
106 case RADEON_EMIT_PP_CNTL:
107 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
108 &data[(RADEON_RB3D_COLOROFFSET - RADEON_PP_CNTL) / 4])) {
109 DRM_ERROR("Invalid colour buffer offset\n");
114 case R200_EMIT_PP_TXOFFSET_0:
115 case R200_EMIT_PP_TXOFFSET_1:
116 case R200_EMIT_PP_TXOFFSET_2:
117 case R200_EMIT_PP_TXOFFSET_3:
118 case R200_EMIT_PP_TXOFFSET_4:
119 case R200_EMIT_PP_TXOFFSET_5:
120 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
122 DRM_ERROR("Invalid R200 texture offset\n");
127 case RADEON_EMIT_PP_TXFILTER_0:
128 case RADEON_EMIT_PP_TXFILTER_1:
129 case RADEON_EMIT_PP_TXFILTER_2:
130 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
131 &data[(RADEON_PP_TXOFFSET_0 - RADEON_PP_TXFILTER_0) / 4])) {
132 DRM_ERROR("Invalid R100 texture offset\n");
137 case R200_EMIT_PP_CUBIC_OFFSETS_0:
138 case R200_EMIT_PP_CUBIC_OFFSETS_1:
139 case R200_EMIT_PP_CUBIC_OFFSETS_2:
140 case R200_EMIT_PP_CUBIC_OFFSETS_3:
141 case R200_EMIT_PP_CUBIC_OFFSETS_4:
142 case R200_EMIT_PP_CUBIC_OFFSETS_5:{
144 for (i = 0; i < 5; i++) {
145 if (radeon_check_and_fixup_offset(dev_priv,
149 ("Invalid R200 cubic texture offset\n");
156 case RADEON_EMIT_PP_CUBIC_OFFSETS_T0:
157 case RADEON_EMIT_PP_CUBIC_OFFSETS_T1:
158 case RADEON_EMIT_PP_CUBIC_OFFSETS_T2:{
160 for (i = 0; i < 5; i++) {
161 if (radeon_check_and_fixup_offset(dev_priv,
165 ("Invalid R100 cubic texture offset\n");
172 case R200_EMIT_VAP_CTL:{
175 OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
180 case RADEON_EMIT_RB3D_COLORPITCH:
181 case RADEON_EMIT_RE_LINE_PATTERN:
182 case RADEON_EMIT_SE_LINE_WIDTH:
183 case RADEON_EMIT_PP_LUM_MATRIX:
184 case RADEON_EMIT_PP_ROT_MATRIX_0:
185 case RADEON_EMIT_RB3D_STENCILREFMASK:
186 case RADEON_EMIT_SE_VPORT_XSCALE:
187 case RADEON_EMIT_SE_CNTL:
188 case RADEON_EMIT_SE_CNTL_STATUS:
189 case RADEON_EMIT_RE_MISC:
190 case RADEON_EMIT_PP_BORDER_COLOR_0:
191 case RADEON_EMIT_PP_BORDER_COLOR_1:
192 case RADEON_EMIT_PP_BORDER_COLOR_2:
193 case RADEON_EMIT_SE_ZBIAS_FACTOR:
194 case RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT:
195 case RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED:
196 case R200_EMIT_PP_TXCBLEND_0:
197 case R200_EMIT_PP_TXCBLEND_1:
198 case R200_EMIT_PP_TXCBLEND_2:
199 case R200_EMIT_PP_TXCBLEND_3:
200 case R200_EMIT_PP_TXCBLEND_4:
201 case R200_EMIT_PP_TXCBLEND_5:
202 case R200_EMIT_PP_TXCBLEND_6:
203 case R200_EMIT_PP_TXCBLEND_7:
204 case R200_EMIT_TCL_LIGHT_MODEL_CTL_0:
205 case R200_EMIT_TFACTOR_0:
206 case R200_EMIT_VTX_FMT_0:
207 case R200_EMIT_MATRIX_SELECT_0:
208 case R200_EMIT_TEX_PROC_CTL_2:
209 case R200_EMIT_TCL_UCP_VERT_BLEND_CTL:
210 case R200_EMIT_PP_TXFILTER_0:
211 case R200_EMIT_PP_TXFILTER_1:
212 case R200_EMIT_PP_TXFILTER_2:
213 case R200_EMIT_PP_TXFILTER_3:
214 case R200_EMIT_PP_TXFILTER_4:
215 case R200_EMIT_PP_TXFILTER_5:
216 case R200_EMIT_VTE_CNTL:
217 case R200_EMIT_OUTPUT_VTX_COMP_SEL:
218 case R200_EMIT_PP_TAM_DEBUG3:
219 case R200_EMIT_PP_CNTL_X:
220 case R200_EMIT_RB3D_DEPTHXY_OFFSET:
221 case R200_EMIT_RE_AUX_SCISSOR_CNTL:
222 case R200_EMIT_RE_SCISSOR_TL_0:
223 case R200_EMIT_RE_SCISSOR_TL_1:
224 case R200_EMIT_RE_SCISSOR_TL_2:
225 case R200_EMIT_SE_VAP_CNTL_STATUS:
226 case R200_EMIT_SE_VTX_STATE_CNTL:
227 case R200_EMIT_RE_POINTSIZE:
228 case R200_EMIT_TCL_INPUT_VTX_VECTOR_ADDR_0:
229 case R200_EMIT_PP_CUBIC_FACES_0:
230 case R200_EMIT_PP_CUBIC_FACES_1:
231 case R200_EMIT_PP_CUBIC_FACES_2:
232 case R200_EMIT_PP_CUBIC_FACES_3:
233 case R200_EMIT_PP_CUBIC_FACES_4:
234 case R200_EMIT_PP_CUBIC_FACES_5:
235 case RADEON_EMIT_PP_TEX_SIZE_0:
236 case RADEON_EMIT_PP_TEX_SIZE_1:
237 case RADEON_EMIT_PP_TEX_SIZE_2:
238 case R200_EMIT_RB3D_BLENDCOLOR:
239 case R200_EMIT_TCL_POINT_SPRITE_CNTL:
240 case RADEON_EMIT_PP_CUBIC_FACES_0:
241 case RADEON_EMIT_PP_CUBIC_FACES_1:
242 case RADEON_EMIT_PP_CUBIC_FACES_2:
243 case R200_EMIT_PP_TRI_PERF_CNTL:
244 case R200_EMIT_PP_AFS_0:
245 case R200_EMIT_PP_AFS_1:
246 case R200_EMIT_ATF_TFACTOR:
247 case R200_EMIT_PP_TXCTLALL_0:
248 case R200_EMIT_PP_TXCTLALL_1:
249 case R200_EMIT_PP_TXCTLALL_2:
250 case R200_EMIT_PP_TXCTLALL_3:
251 case R200_EMIT_PP_TXCTLALL_4:
252 case R200_EMIT_PP_TXCTLALL_5:
253 case R200_EMIT_VAP_PVS_CNTL:
254 /* These packets don't contain memory offsets */
258 DRM_ERROR("Unknown state packet ID %d\n", id);
265 static __inline__ int radeon_check_and_fixup_packet3(drm_radeon_private_t *
267 struct drm_file *file_priv,
268 drm_radeon_kcmd_buffer_t *
272 u32 *cmd = (u32 *) cmdbuf->buf;
276 *cmdsz = 2 + ((cmd[0] & RADEON_CP_PACKET_COUNT_MASK) >> 16);
278 if ((cmd[0] & 0xc0000000) != RADEON_CP_PACKET3) {
279 DRM_ERROR("Not a type 3 packet\n");
283 if (4 * *cmdsz > cmdbuf->bufsz) {
284 DRM_ERROR("Packet size larger than size of data provided\n");
288 switch(cmd[0] & 0xff00) {
289 /* XXX Are there old drivers needing other packets? */
291 case RADEON_3D_DRAW_IMMD:
292 case RADEON_3D_DRAW_VBUF:
293 case RADEON_3D_DRAW_INDX:
294 case RADEON_WAIT_FOR_IDLE:
296 case RADEON_3D_CLEAR_ZMASK:
297 /* case RADEON_CP_NEXT_CHAR:
298 case RADEON_CP_PLY_NEXTSCAN:
299 case RADEON_CP_SET_SCISSORS: */ /* probably safe but will never need them? */
300 /* these packets are safe */
303 case RADEON_CP_3D_DRAW_IMMD_2:
304 case RADEON_CP_3D_DRAW_VBUF_2:
305 case RADEON_CP_3D_DRAW_INDX_2:
306 case RADEON_3D_CLEAR_HIZ:
307 /* safe but r200 only */
308 if (dev_priv->microcode_version != UCODE_R200) {
309 DRM_ERROR("Invalid 3d packet for r100-class chip\n");
314 case RADEON_3D_LOAD_VBPNTR:
315 count = (cmd[0] >> 16) & 0x3fff;
317 if (count > 18) { /* 12 arrays max */
318 DRM_ERROR("Too large payload in 3D_LOAD_VBPNTR (count=%d)\n",
323 /* carefully check packet contents */
324 narrays = cmd[1] & ~0xc000;
327 while ((k < narrays) && (i < (count + 2))) {
328 i++; /* skip attribute field */
329 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
332 ("Invalid offset (k=%d i=%d) in 3D_LOAD_VBPNTR packet.\n",
340 /* have one more to process, they come in pairs */
341 if (radeon_check_and_fixup_offset(dev_priv,
345 ("Invalid offset (k=%d i=%d) in 3D_LOAD_VBPNTR packet.\n",
352 /* do the counts match what we expect ? */
353 if ((k != narrays) || (i != (count + 2))) {
355 ("Malformed 3D_LOAD_VBPNTR packet (k=%d i=%d narrays=%d count+1=%d).\n",
356 k, i, narrays, count + 1);
361 case RADEON_3D_RNDR_GEN_INDX_PRIM:
362 if (dev_priv->microcode_version != UCODE_R100) {
363 DRM_ERROR("Invalid 3d packet for r200-class chip\n");
366 if (radeon_check_and_fixup_offset(dev_priv, file_priv, &cmd[1])) {
367 DRM_ERROR("Invalid rndr_gen_indx offset\n");
372 case RADEON_CP_INDX_BUFFER:
373 if (dev_priv->microcode_version != UCODE_R200) {
374 DRM_ERROR("Invalid 3d packet for r100-class chip\n");
377 if ((cmd[1] & 0x8000ffff) != 0x80000810) {
378 DRM_ERROR("Invalid indx_buffer reg address %08X\n", cmd[1]);
381 if (radeon_check_and_fixup_offset(dev_priv, file_priv, &cmd[2])) {
382 DRM_ERROR("Invalid indx_buffer offset is %08X\n", cmd[2]);
387 case RADEON_CNTL_HOSTDATA_BLT:
388 case RADEON_CNTL_PAINT_MULTI:
389 case RADEON_CNTL_BITBLT_MULTI:
390 /* MSB of opcode: next DWORD GUI_CNTL */
391 if (cmd[1] & (RADEON_GMC_SRC_PITCH_OFFSET_CNTL
392 | RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
393 offset = cmd[2] << 10;
394 if (radeon_check_and_fixup_offset
395 (dev_priv, file_priv, &offset)) {
396 DRM_ERROR("Invalid first packet offset\n");
399 cmd[2] = (cmd[2] & 0xffc00000) | offset >> 10;
402 if ((cmd[1] & RADEON_GMC_SRC_PITCH_OFFSET_CNTL) &&
403 (cmd[1] & RADEON_GMC_DST_PITCH_OFFSET_CNTL)) {
404 offset = cmd[3] << 10;
405 if (radeon_check_and_fixup_offset
406 (dev_priv, file_priv, &offset)) {
407 DRM_ERROR("Invalid second packet offset\n");
410 cmd[3] = (cmd[3] & 0xffc00000) | offset >> 10;
415 DRM_ERROR("Invalid packet type %x\n", cmd[0] & 0xff00);
422 /* ================================================================
423 * CP hardware state programming functions
426 static __inline__ void radeon_emit_clip_rect(drm_radeon_private_t * dev_priv,
427 struct drm_clip_rect * box)
431 DRM_DEBUG(" box: x1=%d y1=%d x2=%d y2=%d\n",
432 box->x1, box->y1, box->x2, box->y2);
435 OUT_RING(CP_PACKET0(RADEON_RE_TOP_LEFT, 0));
436 OUT_RING((box->y1 << 16) | box->x1);
437 OUT_RING(CP_PACKET0(RADEON_RE_WIDTH_HEIGHT, 0));
438 OUT_RING(((box->y2 - 1) << 16) | (box->x2 - 1));
444 static int radeon_emit_state(drm_radeon_private_t * dev_priv,
445 struct drm_file *file_priv,
446 drm_radeon_context_regs_t * ctx,
447 drm_radeon_texture_regs_t * tex,
451 DRM_DEBUG("dirty=0x%08x\n", dirty);
453 if (dirty & RADEON_UPLOAD_CONTEXT) {
454 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
455 &ctx->rb3d_depthoffset)) {
456 DRM_ERROR("Invalid depth buffer offset\n");
460 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
461 &ctx->rb3d_coloroffset)) {
462 DRM_ERROR("Invalid depth buffer offset\n");
467 OUT_RING(CP_PACKET0(RADEON_PP_MISC, 6));
468 OUT_RING(ctx->pp_misc);
469 OUT_RING(ctx->pp_fog_color);
470 OUT_RING(ctx->re_solid_color);
471 OUT_RING(ctx->rb3d_blendcntl);
472 OUT_RING(ctx->rb3d_depthoffset);
473 OUT_RING(ctx->rb3d_depthpitch);
474 OUT_RING(ctx->rb3d_zstencilcntl);
475 OUT_RING(CP_PACKET0(RADEON_PP_CNTL, 2));
476 OUT_RING(ctx->pp_cntl);
477 OUT_RING(ctx->rb3d_cntl);
478 OUT_RING(ctx->rb3d_coloroffset);
479 OUT_RING(CP_PACKET0(RADEON_RB3D_COLORPITCH, 0));
480 OUT_RING(ctx->rb3d_colorpitch);
484 if (dirty & RADEON_UPLOAD_VERTFMT) {
486 OUT_RING(CP_PACKET0(RADEON_SE_COORD_FMT, 0));
487 OUT_RING(ctx->se_coord_fmt);
491 if (dirty & RADEON_UPLOAD_LINE) {
493 OUT_RING(CP_PACKET0(RADEON_RE_LINE_PATTERN, 1));
494 OUT_RING(ctx->re_line_pattern);
495 OUT_RING(ctx->re_line_state);
496 OUT_RING(CP_PACKET0(RADEON_SE_LINE_WIDTH, 0));
497 OUT_RING(ctx->se_line_width);
501 if (dirty & RADEON_UPLOAD_BUMPMAP) {
503 OUT_RING(CP_PACKET0(RADEON_PP_LUM_MATRIX, 0));
504 OUT_RING(ctx->pp_lum_matrix);
505 OUT_RING(CP_PACKET0(RADEON_PP_ROT_MATRIX_0, 1));
506 OUT_RING(ctx->pp_rot_matrix_0);
507 OUT_RING(ctx->pp_rot_matrix_1);
511 if (dirty & RADEON_UPLOAD_MASKS) {
513 OUT_RING(CP_PACKET0(RADEON_RB3D_STENCILREFMASK, 2));
514 OUT_RING(ctx->rb3d_stencilrefmask);
515 OUT_RING(ctx->rb3d_ropcntl);
516 OUT_RING(ctx->rb3d_planemask);
520 if (dirty & RADEON_UPLOAD_VIEWPORT) {
522 OUT_RING(CP_PACKET0(RADEON_SE_VPORT_XSCALE, 5));
523 OUT_RING(ctx->se_vport_xscale);
524 OUT_RING(ctx->se_vport_xoffset);
525 OUT_RING(ctx->se_vport_yscale);
526 OUT_RING(ctx->se_vport_yoffset);
527 OUT_RING(ctx->se_vport_zscale);
528 OUT_RING(ctx->se_vport_zoffset);
532 if (dirty & RADEON_UPLOAD_SETUP) {
534 OUT_RING(CP_PACKET0(RADEON_SE_CNTL, 0));
535 OUT_RING(ctx->se_cntl);
536 OUT_RING(CP_PACKET0(RADEON_SE_CNTL_STATUS, 0));
537 OUT_RING(ctx->se_cntl_status);
541 if (dirty & RADEON_UPLOAD_MISC) {
543 OUT_RING(CP_PACKET0(RADEON_RE_MISC, 0));
544 OUT_RING(ctx->re_misc);
548 if (dirty & RADEON_UPLOAD_TEX0) {
549 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
550 &tex[0].pp_txoffset)) {
551 DRM_ERROR("Invalid texture offset for unit 0\n");
556 OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_0, 5));
557 OUT_RING(tex[0].pp_txfilter);
558 OUT_RING(tex[0].pp_txformat);
559 OUT_RING(tex[0].pp_txoffset);
560 OUT_RING(tex[0].pp_txcblend);
561 OUT_RING(tex[0].pp_txablend);
562 OUT_RING(tex[0].pp_tfactor);
563 OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_0, 0));
564 OUT_RING(tex[0].pp_border_color);
568 if (dirty & RADEON_UPLOAD_TEX1) {
569 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
570 &tex[1].pp_txoffset)) {
571 DRM_ERROR("Invalid texture offset for unit 1\n");
576 OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_1, 5));
577 OUT_RING(tex[1].pp_txfilter);
578 OUT_RING(tex[1].pp_txformat);
579 OUT_RING(tex[1].pp_txoffset);
580 OUT_RING(tex[1].pp_txcblend);
581 OUT_RING(tex[1].pp_txablend);
582 OUT_RING(tex[1].pp_tfactor);
583 OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_1, 0));
584 OUT_RING(tex[1].pp_border_color);
588 if (dirty & RADEON_UPLOAD_TEX2) {
589 if (radeon_check_and_fixup_offset(dev_priv, file_priv,
590 &tex[2].pp_txoffset)) {
591 DRM_ERROR("Invalid texture offset for unit 2\n");
596 OUT_RING(CP_PACKET0(RADEON_PP_TXFILTER_2, 5));
597 OUT_RING(tex[2].pp_txfilter);
598 OUT_RING(tex[2].pp_txformat);
599 OUT_RING(tex[2].pp_txoffset);
600 OUT_RING(tex[2].pp_txcblend);
601 OUT_RING(tex[2].pp_txablend);
602 OUT_RING(tex[2].pp_tfactor);
603 OUT_RING(CP_PACKET0(RADEON_PP_BORDER_COLOR_2, 0));
604 OUT_RING(tex[2].pp_border_color);
613 static int radeon_emit_state2(drm_radeon_private_t * dev_priv,
614 struct drm_file *file_priv,
615 drm_radeon_state_t * state)
619 if (state->dirty & RADEON_UPLOAD_ZBIAS) {
621 OUT_RING(CP_PACKET0(RADEON_SE_ZBIAS_FACTOR, 1));
622 OUT_RING(state->context2.se_zbias_factor);
623 OUT_RING(state->context2.se_zbias_constant);
627 return radeon_emit_state(dev_priv, file_priv, &state->context,
628 state->tex, state->dirty);
631 /* New (1.3) state mechanism. 3 commands (packet, scalar, vector) in
632 * 1.3 cmdbuffers allow all previous state to be updated as well as
633 * the tcl scalar and vector areas.
639 } packet[RADEON_MAX_STATE_PACKETS] = {
640 {RADEON_PP_MISC, 7, "RADEON_PP_MISC"},
641 {RADEON_PP_CNTL, 3, "RADEON_PP_CNTL"},
642 {RADEON_RB3D_COLORPITCH, 1, "RADEON_RB3D_COLORPITCH"},
643 {RADEON_RE_LINE_PATTERN, 2, "RADEON_RE_LINE_PATTERN"},
644 {RADEON_SE_LINE_WIDTH, 1, "RADEON_SE_LINE_WIDTH"},
645 {RADEON_PP_LUM_MATRIX, 1, "RADEON_PP_LUM_MATRIX"},
646 {RADEON_PP_ROT_MATRIX_0, 2, "RADEON_PP_ROT_MATRIX_0"},
647 {RADEON_RB3D_STENCILREFMASK, 3, "RADEON_RB3D_STENCILREFMASK"},
648 {RADEON_SE_VPORT_XSCALE, 6, "RADEON_SE_VPORT_XSCALE"},
649 {RADEON_SE_CNTL, 2, "RADEON_SE_CNTL"},
650 {RADEON_SE_CNTL_STATUS, 1, "RADEON_SE_CNTL_STATUS"},
651 {RADEON_RE_MISC, 1, "RADEON_RE_MISC"},
652 {RADEON_PP_TXFILTER_0, 6, "RADEON_PP_TXFILTER_0"},
653 {RADEON_PP_BORDER_COLOR_0, 1, "RADEON_PP_BORDER_COLOR_0"},
654 {RADEON_PP_TXFILTER_1, 6, "RADEON_PP_TXFILTER_1"},
655 {RADEON_PP_BORDER_COLOR_1, 1, "RADEON_PP_BORDER_COLOR_1"},
656 {RADEON_PP_TXFILTER_2, 6, "RADEON_PP_TXFILTER_2"},
657 {RADEON_PP_BORDER_COLOR_2, 1, "RADEON_PP_BORDER_COLOR_2"},
658 {RADEON_SE_ZBIAS_FACTOR, 2, "RADEON_SE_ZBIAS_FACTOR"},
659 {RADEON_SE_TCL_OUTPUT_VTX_FMT, 11, "RADEON_SE_TCL_OUTPUT_VTX_FMT"},
660 {RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, 17,
661 "RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED"},
662 {R200_PP_TXCBLEND_0, 4, "R200_PP_TXCBLEND_0"},
663 {R200_PP_TXCBLEND_1, 4, "R200_PP_TXCBLEND_1"},
664 {R200_PP_TXCBLEND_2, 4, "R200_PP_TXCBLEND_2"},
665 {R200_PP_TXCBLEND_3, 4, "R200_PP_TXCBLEND_3"},
666 {R200_PP_TXCBLEND_4, 4, "R200_PP_TXCBLEND_4"},
667 {R200_PP_TXCBLEND_5, 4, "R200_PP_TXCBLEND_5"},
668 {R200_PP_TXCBLEND_6, 4, "R200_PP_TXCBLEND_6"},
669 {R200_PP_TXCBLEND_7, 4, "R200_PP_TXCBLEND_7"},
670 {R200_SE_TCL_LIGHT_MODEL_CTL_0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0"},
671 {R200_PP_TFACTOR_0, 6, "R200_PP_TFACTOR_0"},
672 {R200_SE_VTX_FMT_0, 4, "R200_SE_VTX_FMT_0"},
673 {R200_SE_VAP_CNTL, 1, "R200_SE_VAP_CNTL"},
674 {R200_SE_TCL_MATRIX_SEL_0, 5, "R200_SE_TCL_MATRIX_SEL_0"},
675 {R200_SE_TCL_TEX_PROC_CTL_2, 5, "R200_SE_TCL_TEX_PROC_CTL_2"},
676 {R200_SE_TCL_UCP_VERT_BLEND_CTL, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL"},
677 {R200_PP_TXFILTER_0, 6, "R200_PP_TXFILTER_0"},
678 {R200_PP_TXFILTER_1, 6, "R200_PP_TXFILTER_1"},
679 {R200_PP_TXFILTER_2, 6, "R200_PP_TXFILTER_2"},
680 {R200_PP_TXFILTER_3, 6, "R200_PP_TXFILTER_3"},
681 {R200_PP_TXFILTER_4, 6, "R200_PP_TXFILTER_4"},
682 {R200_PP_TXFILTER_5, 6, "R200_PP_TXFILTER_5"},
683 {R200_PP_TXOFFSET_0, 1, "R200_PP_TXOFFSET_0"},
684 {R200_PP_TXOFFSET_1, 1, "R200_PP_TXOFFSET_1"},
685 {R200_PP_TXOFFSET_2, 1, "R200_PP_TXOFFSET_2"},
686 {R200_PP_TXOFFSET_3, 1, "R200_PP_TXOFFSET_3"},
687 {R200_PP_TXOFFSET_4, 1, "R200_PP_TXOFFSET_4"},
688 {R200_PP_TXOFFSET_5, 1, "R200_PP_TXOFFSET_5"},
689 {R200_SE_VTE_CNTL, 1, "R200_SE_VTE_CNTL"},
690 {R200_SE_TCL_OUTPUT_VTX_COMP_SEL, 1,
691 "R200_SE_TCL_OUTPUT_VTX_COMP_SEL"},
692 {R200_PP_TAM_DEBUG3, 1, "R200_PP_TAM_DEBUG3"},
693 {R200_PP_CNTL_X, 1, "R200_PP_CNTL_X"},
694 {R200_RB3D_DEPTHXY_OFFSET, 1, "R200_RB3D_DEPTHXY_OFFSET"},
695 {R200_RE_AUX_SCISSOR_CNTL, 1, "R200_RE_AUX_SCISSOR_CNTL"},
696 {R200_RE_SCISSOR_TL_0, 2, "R200_RE_SCISSOR_TL_0"},
697 {R200_RE_SCISSOR_TL_1, 2, "R200_RE_SCISSOR_TL_1"},
698 {R200_RE_SCISSOR_TL_2, 2, "R200_RE_SCISSOR_TL_2"},
699 {R200_SE_VAP_CNTL_STATUS, 1, "R200_SE_VAP_CNTL_STATUS"},
700 {R200_SE_VTX_STATE_CNTL, 1, "R200_SE_VTX_STATE_CNTL"},
701 {R200_RE_POINTSIZE, 1, "R200_RE_POINTSIZE"},
702 {R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0, 4,
703 "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0"},
704 {R200_PP_CUBIC_FACES_0, 1, "R200_PP_CUBIC_FACES_0"}, /* 61 */
705 {R200_PP_CUBIC_OFFSET_F1_0, 5, "R200_PP_CUBIC_OFFSET_F1_0"}, /* 62 */
706 {R200_PP_CUBIC_FACES_1, 1, "R200_PP_CUBIC_FACES_1"},
707 {R200_PP_CUBIC_OFFSET_F1_1, 5, "R200_PP_CUBIC_OFFSET_F1_1"},
708 {R200_PP_CUBIC_FACES_2, 1, "R200_PP_CUBIC_FACES_2"},
709 {R200_PP_CUBIC_OFFSET_F1_2, 5, "R200_PP_CUBIC_OFFSET_F1_2"},
710 {R200_PP_CUBIC_FACES_3, 1, "R200_PP_CUBIC_FACES_3"},
711 {R200_PP_CUBIC_OFFSET_F1_3, 5, "R200_PP_CUBIC_OFFSET_F1_3"},
712 {R200_PP_CUBIC_FACES_4, 1, "R200_PP_CUBIC_FACES_4"},
713 {R200_PP_CUBIC_OFFSET_F1_4, 5, "R200_PP_CUBIC_OFFSET_F1_4"},
714 {R200_PP_CUBIC_FACES_5, 1, "R200_PP_CUBIC_FACES_5"},
715 {R200_PP_CUBIC_OFFSET_F1_5, 5, "R200_PP_CUBIC_OFFSET_F1_5"},
716 {RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0"},
717 {RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1"},
718 {RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_2"},
719 {R200_RB3D_BLENDCOLOR, 3, "R200_RB3D_BLENDCOLOR"},
720 {R200_SE_TCL_POINT_SPRITE_CNTL, 1, "R200_SE_TCL_POINT_SPRITE_CNTL"},
721 {RADEON_PP_CUBIC_FACES_0, 1, "RADEON_PP_CUBIC_FACES_0"},
722 {RADEON_PP_CUBIC_OFFSET_T0_0, 5, "RADEON_PP_CUBIC_OFFSET_T0_0"},
723 {RADEON_PP_CUBIC_FACES_1, 1, "RADEON_PP_CUBIC_FACES_1"},
724 {RADEON_PP_CUBIC_OFFSET_T1_0, 5, "RADEON_PP_CUBIC_OFFSET_T1_0"},
725 {RADEON_PP_CUBIC_FACES_2, 1, "RADEON_PP_CUBIC_FACES_2"},
726 {RADEON_PP_CUBIC_OFFSET_T2_0, 5, "RADEON_PP_CUBIC_OFFSET_T2_0"},
727 {R200_PP_TRI_PERF, 2, "R200_PP_TRI_PERF"},
728 {R200_PP_AFS_0, 32, "R200_PP_AFS_0"}, /* 85 */
729 {R200_PP_AFS_1, 32, "R200_PP_AFS_1"},
730 {R200_PP_TFACTOR_0, 8, "R200_ATF_TFACTOR"},
731 {R200_PP_TXFILTER_0, 8, "R200_PP_TXCTLALL_0"},
732 {R200_PP_TXFILTER_1, 8, "R200_PP_TXCTLALL_1"},
733 {R200_PP_TXFILTER_2, 8, "R200_PP_TXCTLALL_2"},
734 {R200_PP_TXFILTER_3, 8, "R200_PP_TXCTLALL_3"},
735 {R200_PP_TXFILTER_4, 8, "R200_PP_TXCTLALL_4"},
736 {R200_PP_TXFILTER_5, 8, "R200_PP_TXCTLALL_5"},
737 {R200_VAP_PVS_CNTL_1, 2, "R200_VAP_PVS_CNTL"},
740 /* ================================================================
741 * Performance monitoring functions
744 static void radeon_clear_box(drm_radeon_private_t * dev_priv,
745 int x, int y, int w, int h, int r, int g, int b)
750 x += dev_priv->sarea_priv->boxes[0].x1;
751 y += dev_priv->sarea_priv->boxes[0].y1;
753 switch (dev_priv->color_fmt) {
754 case RADEON_COLOR_FORMAT_RGB565:
755 color = (((r & 0xf8) << 8) |
756 ((g & 0xfc) << 3) | ((b & 0xf8) >> 3));
758 case RADEON_COLOR_FORMAT_ARGB8888:
760 color = (((0xff) << 24) | (r << 16) | (g << 8) | b);
765 RADEON_WAIT_UNTIL_3D_IDLE();
766 OUT_RING(CP_PACKET0(RADEON_DP_WRITE_MASK, 0));
767 OUT_RING(0xffffffff);
772 OUT_RING(CP_PACKET3(RADEON_CNTL_PAINT_MULTI, 4));
773 OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
774 RADEON_GMC_BRUSH_SOLID_COLOR |
775 (dev_priv->color_fmt << 8) |
776 RADEON_GMC_SRC_DATATYPE_COLOR |
777 RADEON_ROP3_P | RADEON_GMC_CLR_CMP_CNTL_DIS);
779 if (dev_priv->sarea_priv->pfCurrentPage == 1) {
780 OUT_RING(dev_priv->front_pitch_offset);
782 OUT_RING(dev_priv->back_pitch_offset);
787 OUT_RING((x << 16) | y);
788 OUT_RING((w << 16) | h);
793 static void radeon_cp_performance_boxes(drm_radeon_private_t * dev_priv)
795 /* Collapse various things into a wait flag -- trying to
796 * guess if userspase slept -- better just to have them tell us.
798 if (dev_priv->stats.last_frame_reads > 1 ||
799 dev_priv->stats.last_clear_reads > dev_priv->stats.clears) {
800 dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
803 if (dev_priv->stats.freelist_loops) {
804 dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
807 /* Purple box for page flipping
809 if (dev_priv->stats.boxes & RADEON_BOX_FLIP)
810 radeon_clear_box(dev_priv, 4, 4, 8, 8, 255, 0, 255);
812 /* Red box if we have to wait for idle at any point
814 if (dev_priv->stats.boxes & RADEON_BOX_WAIT_IDLE)
815 radeon_clear_box(dev_priv, 16, 4, 8, 8, 255, 0, 0);
817 /* Blue box: lost context?
820 /* Yellow box for texture swaps
822 if (dev_priv->stats.boxes & RADEON_BOX_TEXTURE_LOAD)
823 radeon_clear_box(dev_priv, 40, 4, 8, 8, 255, 255, 0);
825 /* Green box if hardware never idles (as far as we can tell)
827 if (!(dev_priv->stats.boxes & RADEON_BOX_DMA_IDLE))
828 radeon_clear_box(dev_priv, 64, 4, 8, 8, 0, 255, 0);
830 /* Draw bars indicating number of buffers allocated
831 * (not a great measure, easily confused)
833 if (dev_priv->stats.requested_bufs) {
834 if (dev_priv->stats.requested_bufs > 100)
835 dev_priv->stats.requested_bufs = 100;
837 radeon_clear_box(dev_priv, 4, 16,
838 dev_priv->stats.requested_bufs, 4,
842 memset(&dev_priv->stats, 0, sizeof(dev_priv->stats));
846 /* ================================================================
847 * CP command dispatch functions
850 static void radeon_cp_dispatch_clear(struct drm_device * dev,
851 drm_radeon_clear_t * clear,
852 drm_radeon_clear_rect_t * depth_boxes)
854 drm_radeon_private_t *dev_priv = dev->dev_private;
855 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
856 drm_radeon_depth_clear_t *depth_clear = &dev_priv->depth_clear;
857 int nbox = sarea_priv->nbox;
858 struct drm_clip_rect *pbox = sarea_priv->boxes;
859 unsigned int flags = clear->flags;
860 u32 rb3d_cntl = 0, rb3d_stencilrefmask = 0;
863 DRM_DEBUG("flags = 0x%x\n", flags);
865 dev_priv->stats.clears++;
867 if (dev_priv->sarea_priv->pfCurrentPage == 1) {
868 unsigned int tmp = flags;
870 flags &= ~(RADEON_FRONT | RADEON_BACK);
871 if (tmp & RADEON_FRONT)
872 flags |= RADEON_BACK;
873 if (tmp & RADEON_BACK)
874 flags |= RADEON_FRONT;
877 if (flags & (RADEON_FRONT | RADEON_BACK)) {
881 /* Ensure the 3D stream is idle before doing a
882 * 2D fill to clear the front or back buffer.
884 RADEON_WAIT_UNTIL_3D_IDLE();
886 OUT_RING(CP_PACKET0(RADEON_DP_WRITE_MASK, 0));
887 OUT_RING(clear->color_mask);
891 /* Make sure we restore the 3D state next time.
893 dev_priv->sarea_priv->ctx_owner = 0;
895 for (i = 0; i < nbox; i++) {
898 int w = pbox[i].x2 - x;
899 int h = pbox[i].y2 - y;
901 DRM_DEBUG("dispatch clear %d,%d-%d,%d flags 0x%x\n",
904 if (flags & RADEON_FRONT) {
908 (RADEON_CNTL_PAINT_MULTI, 4));
909 OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
910 RADEON_GMC_BRUSH_SOLID_COLOR |
913 RADEON_GMC_SRC_DATATYPE_COLOR |
915 RADEON_GMC_CLR_CMP_CNTL_DIS);
917 OUT_RING(dev_priv->front_pitch_offset);
918 OUT_RING(clear->clear_color);
920 OUT_RING((x << 16) | y);
921 OUT_RING((w << 16) | h);
926 if (flags & RADEON_BACK) {
930 (RADEON_CNTL_PAINT_MULTI, 4));
931 OUT_RING(RADEON_GMC_DST_PITCH_OFFSET_CNTL |
932 RADEON_GMC_BRUSH_SOLID_COLOR |
935 RADEON_GMC_SRC_DATATYPE_COLOR |
937 RADEON_GMC_CLR_CMP_CNTL_DIS);
939 OUT_RING(dev_priv->back_pitch_offset);
940 OUT_RING(clear->clear_color);
942 OUT_RING((x << 16) | y);
943 OUT_RING((w << 16) | h);
951 /* no docs available, based on reverse engeneering by Stephane Marchesin */
952 if ((flags & (RADEON_DEPTH | RADEON_STENCIL))
953 && (flags & RADEON_CLEAR_FASTZ)) {
956 int depthpixperline =
957 dev_priv->depth_fmt ==
958 RADEON_DEPTH_FORMAT_16BIT_INT_Z ? (dev_priv->depth_pitch /
964 u32 tempRB3D_DEPTHCLEARVALUE = clear->clear_depth |
965 ((clear->depth_mask & 0xff) << 24);
967 /* Make sure we restore the 3D state next time.
968 * we haven't touched any "normal" state - still need this?
970 dev_priv->sarea_priv->ctx_owner = 0;
972 if ((dev_priv->flags & RADEON_HAS_HIERZ)
973 && (flags & RADEON_USE_HIERZ)) {
974 /* FIXME : reverse engineer that for Rx00 cards */
975 /* FIXME : the mask supposedly contains low-res z values. So can't set
976 just to the max (0xff? or actually 0x3fff?), need to take z clear
977 value into account? */
978 /* pattern seems to work for r100, though get slight
979 rendering errors with glxgears. If hierz is not enabled for r100,
980 only 4 bits which indicate clear (15,16,31,32, all zero) matter, the
981 other ones are ignored, and the same clear mask can be used. That's
982 very different behaviour than R200 which needs different clear mask
983 and different number of tiles to clear if hierz is enabled or not !?!
985 clearmask = (0xff << 22) | (0xff << 6) | 0x003f003f;
987 /* clear mask : chooses the clearing pattern.
988 rv250: could be used to clear only parts of macrotiles
989 (but that would get really complicated...)?
990 bit 0 and 1 (either or both of them ?!?!) are used to
991 not clear tile (or maybe one of the bits indicates if the tile is
992 compressed or not), bit 2 and 3 to not clear tile 1,...,.
993 Pattern is as follows:
994 | 0,1 | 4,5 | 8,9 |12,13|16,17|20,21|24,25|28,29|
995 bits -------------------------------------------------
996 | 2,3 | 6,7 |10,11|14,15|18,19|22,23|26,27|30,31|
997 rv100: clearmask covers 2x8 4x1 tiles, but one clear still
998 covers 256 pixels ?!?
1004 RADEON_WAIT_UNTIL_2D_IDLE();
1005 OUT_RING_REG(RADEON_RB3D_DEPTHCLEARVALUE,
1006 tempRB3D_DEPTHCLEARVALUE);
1007 /* what offset is this exactly ? */
1008 OUT_RING_REG(RADEON_RB3D_ZMASKOFFSET, 0);
1009 /* need ctlstat, otherwise get some strange black flickering */
1010 OUT_RING_REG(RADEON_RB3D_ZCACHE_CTLSTAT,
1011 RADEON_RB3D_ZC_FLUSH_ALL);
1014 for (i = 0; i < nbox; i++) {
1015 int tileoffset, nrtilesx, nrtilesy, j;
1016 /* it looks like r200 needs rv-style clears, at least if hierz is not enabled? */
1017 if ((dev_priv->flags & RADEON_HAS_HIERZ)
1018 && !(dev_priv->microcode_version == UCODE_R200)) {
1019 /* FIXME : figure this out for r200 (when hierz is enabled). Or
1020 maybe r200 actually doesn't need to put the low-res z value into
1021 the tile cache like r100, but just needs to clear the hi-level z-buffer?
1022 Works for R100, both with hierz and without.
1023 R100 seems to operate on 2x1 8x8 tiles, but...
1024 odd: offset/nrtiles need to be 64 pix (4 block) aligned? Potentially
1025 problematic with resolutions which are not 64 pix aligned? */
1027 ((pbox[i].y1 >> 3) * depthpixperline +
1030 ((pbox[i].x2 & ~63) -
1031 (pbox[i].x1 & ~63)) >> 4;
1033 (pbox[i].y2 >> 3) - (pbox[i].y1 >> 3);
1034 for (j = 0; j <= nrtilesy; j++) {
1037 (RADEON_3D_CLEAR_ZMASK, 2));
1039 OUT_RING(tileoffset * 8);
1040 /* the number of tiles to clear */
1041 OUT_RING(nrtilesx + 4);
1042 /* clear mask : chooses the clearing pattern. */
1043 OUT_RING(clearmask);
1045 tileoffset += depthpixperline >> 6;
1047 } else if (dev_priv->microcode_version == UCODE_R200) {
1048 /* works for rv250. */
1049 /* find first macro tile (8x2 4x4 z-pixels on rv250) */
1051 ((pbox[i].y1 >> 3) * depthpixperline +
1054 (pbox[i].x2 >> 5) - (pbox[i].x1 >> 5);
1056 (pbox[i].y2 >> 3) - (pbox[i].y1 >> 3);
1057 for (j = 0; j <= nrtilesy; j++) {
1060 (RADEON_3D_CLEAR_ZMASK, 2));
1062 /* judging by the first tile offset needed, could possibly
1063 directly address/clear 4x4 tiles instead of 8x2 * 4x4
1064 macro tiles, though would still need clear mask for
1065 right/bottom if truely 4x4 granularity is desired ? */
1066 OUT_RING(tileoffset * 16);
1067 /* the number of tiles to clear */
1068 OUT_RING(nrtilesx + 1);
1069 /* clear mask : chooses the clearing pattern. */
1070 OUT_RING(clearmask);
1072 tileoffset += depthpixperline >> 5;
1074 } else { /* rv 100 */
1075 /* rv100 might not need 64 pix alignment, who knows */
1076 /* offsets are, hmm, weird */
1078 ((pbox[i].y1 >> 4) * depthpixperline +
1081 ((pbox[i].x2 & ~63) -
1082 (pbox[i].x1 & ~63)) >> 4;
1084 (pbox[i].y2 >> 4) - (pbox[i].y1 >> 4);
1085 for (j = 0; j <= nrtilesy; j++) {
1088 (RADEON_3D_CLEAR_ZMASK, 2));
1089 OUT_RING(tileoffset * 128);
1090 /* the number of tiles to clear */
1091 OUT_RING(nrtilesx + 4);
1092 /* clear mask : chooses the clearing pattern. */
1093 OUT_RING(clearmask);
1095 tileoffset += depthpixperline >> 6;
1100 /* TODO don't always clear all hi-level z tiles */
1101 if ((dev_priv->flags & RADEON_HAS_HIERZ)
1102 && (dev_priv->microcode_version == UCODE_R200)
1103 && (flags & RADEON_USE_HIERZ))
1104 /* r100 and cards without hierarchical z-buffer have no high-level z-buffer */
1105 /* FIXME : the mask supposedly contains low-res z values. So can't set
1106 just to the max (0xff? or actually 0x3fff?), need to take z clear
1107 value into account? */
1110 OUT_RING(CP_PACKET3(RADEON_3D_CLEAR_HIZ, 2));
1111 OUT_RING(0x0); /* First tile */
1113 OUT_RING((0xff << 22) | (0xff << 6) | 0x003f003f);
1118 /* We have to clear the depth and/or stencil buffers by
1119 * rendering a quad into just those buffers. Thus, we have to
1120 * make sure the 3D engine is configured correctly.
1122 else if ((dev_priv->microcode_version == UCODE_R200) &&
1123 (flags & (RADEON_DEPTH | RADEON_STENCIL))) {
1128 int tempRB3D_ZSTENCILCNTL;
1129 int tempRB3D_STENCILREFMASK;
1130 int tempRB3D_PLANEMASK;
1132 int tempSE_VTE_CNTL;
1133 int tempSE_VTX_FMT_0;
1134 int tempSE_VTX_FMT_1;
1135 int tempSE_VAP_CNTL;
1136 int tempRE_AUX_SCISSOR_CNTL;
1141 tempRB3D_CNTL = depth_clear->rb3d_cntl;
1143 tempRB3D_ZSTENCILCNTL = depth_clear->rb3d_zstencilcntl;
1144 tempRB3D_STENCILREFMASK = 0x0;
1146 tempSE_CNTL = depth_clear->se_cntl;
1150 tempSE_VAP_CNTL = ( /* SE_VAP_CNTL__FORCE_W_TO_ONE_MASK | */
1152 SE_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT));
1154 tempRB3D_PLANEMASK = 0x0;
1156 tempRE_AUX_SCISSOR_CNTL = 0x0;
1159 SE_VTE_CNTL__VTX_XY_FMT_MASK | SE_VTE_CNTL__VTX_Z_FMT_MASK;
1161 /* Vertex format (X, Y, Z, W) */
1163 SE_VTX_FMT_0__VTX_Z0_PRESENT_MASK |
1164 SE_VTX_FMT_0__VTX_W0_PRESENT_MASK;
1165 tempSE_VTX_FMT_1 = 0x0;
1168 * Depth buffer specific enables
1170 if (flags & RADEON_DEPTH) {
1171 /* Enable depth buffer */
1172 tempRB3D_CNTL |= RADEON_Z_ENABLE;
1174 /* Disable depth buffer */
1175 tempRB3D_CNTL &= ~RADEON_Z_ENABLE;
1179 * Stencil buffer specific enables
1181 if (flags & RADEON_STENCIL) {
1182 tempRB3D_CNTL |= RADEON_STENCIL_ENABLE;
1183 tempRB3D_STENCILREFMASK = clear->depth_mask;
1185 tempRB3D_CNTL &= ~RADEON_STENCIL_ENABLE;
1186 tempRB3D_STENCILREFMASK = 0x00000000;
1189 if (flags & RADEON_USE_COMP_ZBUF) {
1190 tempRB3D_ZSTENCILCNTL |= RADEON_Z_COMPRESSION_ENABLE |
1191 RADEON_Z_DECOMPRESSION_ENABLE;
1193 if (flags & RADEON_USE_HIERZ) {
1194 tempRB3D_ZSTENCILCNTL |= RADEON_Z_HIERARCHY_ENABLE;
1198 RADEON_WAIT_UNTIL_2D_IDLE();
1200 OUT_RING_REG(RADEON_PP_CNTL, tempPP_CNTL);
1201 OUT_RING_REG(R200_RE_CNTL, tempRE_CNTL);
1202 OUT_RING_REG(RADEON_RB3D_CNTL, tempRB3D_CNTL);
1203 OUT_RING_REG(RADEON_RB3D_ZSTENCILCNTL, tempRB3D_ZSTENCILCNTL);
1204 OUT_RING_REG(RADEON_RB3D_STENCILREFMASK,
1205 tempRB3D_STENCILREFMASK);
1206 OUT_RING_REG(RADEON_RB3D_PLANEMASK, tempRB3D_PLANEMASK);
1207 OUT_RING_REG(RADEON_SE_CNTL, tempSE_CNTL);
1208 OUT_RING_REG(R200_SE_VTE_CNTL, tempSE_VTE_CNTL);
1209 OUT_RING_REG(R200_SE_VTX_FMT_0, tempSE_VTX_FMT_0);
1210 OUT_RING_REG(R200_SE_VTX_FMT_1, tempSE_VTX_FMT_1);
1211 OUT_RING_REG(R200_SE_VAP_CNTL, tempSE_VAP_CNTL);
1212 OUT_RING_REG(R200_RE_AUX_SCISSOR_CNTL, tempRE_AUX_SCISSOR_CNTL);
1215 /* Make sure we restore the 3D state next time.
1217 dev_priv->sarea_priv->ctx_owner = 0;
1219 for (i = 0; i < nbox; i++) {
1221 /* Funny that this should be required --
1224 radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1227 OUT_RING(CP_PACKET3(R200_3D_DRAW_IMMD_2, 12));
1228 OUT_RING((RADEON_PRIM_TYPE_RECT_LIST |
1229 RADEON_PRIM_WALK_RING |
1230 (3 << RADEON_NUM_VERTICES_SHIFT)));
1231 OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1232 OUT_RING(depth_boxes[i].ui[CLEAR_Y1]);
1233 OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1234 OUT_RING(0x3f800000);
1235 OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1236 OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1237 OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1238 OUT_RING(0x3f800000);
1239 OUT_RING(depth_boxes[i].ui[CLEAR_X2]);
1240 OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1241 OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1242 OUT_RING(0x3f800000);
1245 } else if ((flags & (RADEON_DEPTH | RADEON_STENCIL))) {
1247 int tempRB3D_ZSTENCILCNTL = depth_clear->rb3d_zstencilcntl;
1249 rb3d_cntl = depth_clear->rb3d_cntl;
1251 if (flags & RADEON_DEPTH) {
1252 rb3d_cntl |= RADEON_Z_ENABLE;
1254 rb3d_cntl &= ~RADEON_Z_ENABLE;
1257 if (flags & RADEON_STENCIL) {
1258 rb3d_cntl |= RADEON_STENCIL_ENABLE;
1259 rb3d_stencilrefmask = clear->depth_mask; /* misnamed field */
1261 rb3d_cntl &= ~RADEON_STENCIL_ENABLE;
1262 rb3d_stencilrefmask = 0x00000000;
1265 if (flags & RADEON_USE_COMP_ZBUF) {
1266 tempRB3D_ZSTENCILCNTL |= RADEON_Z_COMPRESSION_ENABLE |
1267 RADEON_Z_DECOMPRESSION_ENABLE;
1269 if (flags & RADEON_USE_HIERZ) {
1270 tempRB3D_ZSTENCILCNTL |= RADEON_Z_HIERARCHY_ENABLE;
1274 RADEON_WAIT_UNTIL_2D_IDLE();
1276 OUT_RING(CP_PACKET0(RADEON_PP_CNTL, 1));
1277 OUT_RING(0x00000000);
1278 OUT_RING(rb3d_cntl);
1280 OUT_RING_REG(RADEON_RB3D_ZSTENCILCNTL, tempRB3D_ZSTENCILCNTL);
1281 OUT_RING_REG(RADEON_RB3D_STENCILREFMASK, rb3d_stencilrefmask);
1282 OUT_RING_REG(RADEON_RB3D_PLANEMASK, 0x00000000);
1283 OUT_RING_REG(RADEON_SE_CNTL, depth_clear->se_cntl);
1286 /* Make sure we restore the 3D state next time.
1288 dev_priv->sarea_priv->ctx_owner = 0;
1290 for (i = 0; i < nbox; i++) {
1292 /* Funny that this should be required --
1295 radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1299 OUT_RING(CP_PACKET3(RADEON_3D_DRAW_IMMD, 13));
1300 OUT_RING(RADEON_VTX_Z_PRESENT |
1301 RADEON_VTX_PKCOLOR_PRESENT);
1302 OUT_RING((RADEON_PRIM_TYPE_RECT_LIST |
1303 RADEON_PRIM_WALK_RING |
1304 RADEON_MAOS_ENABLE |
1305 RADEON_VTX_FMT_RADEON_MODE |
1306 (3 << RADEON_NUM_VERTICES_SHIFT)));
1308 OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1309 OUT_RING(depth_boxes[i].ui[CLEAR_Y1]);
1310 OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1313 OUT_RING(depth_boxes[i].ui[CLEAR_X1]);
1314 OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1315 OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1318 OUT_RING(depth_boxes[i].ui[CLEAR_X2]);
1319 OUT_RING(depth_boxes[i].ui[CLEAR_Y2]);
1320 OUT_RING(depth_boxes[i].ui[CLEAR_DEPTH]);
1327 /* Increment the clear counter. The client-side 3D driver must
1328 * wait on this value before performing the clear ioctl. We
1329 * need this because the card's so damned fast...
1331 dev_priv->sarea_priv->last_clear++;
1335 RADEON_CLEAR_AGE(dev_priv->sarea_priv->last_clear);
1336 RADEON_WAIT_UNTIL_IDLE();
1341 static void radeon_cp_dispatch_swap(struct drm_device * dev)
1343 drm_radeon_private_t *dev_priv = dev->dev_private;
1344 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1345 int nbox = sarea_priv->nbox;
1346 struct drm_clip_rect *pbox = sarea_priv->boxes;
1351 /* Do some trivial performance monitoring...
1353 if (dev_priv->do_boxes)
1354 radeon_cp_performance_boxes(dev_priv);
1356 /* Wait for the 3D stream to idle before dispatching the bitblt.
1357 * This will prevent data corruption between the two streams.
1361 RADEON_WAIT_UNTIL_3D_IDLE();
1365 for (i = 0; i < nbox; i++) {
1368 int w = pbox[i].x2 - x;
1369 int h = pbox[i].y2 - y;
1371 DRM_DEBUG("dispatch swap %d,%d-%d,%d\n", x, y, w, h);
1375 OUT_RING(CP_PACKET0(RADEON_DP_GUI_MASTER_CNTL, 0));
1376 OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
1377 RADEON_GMC_DST_PITCH_OFFSET_CNTL |
1378 RADEON_GMC_BRUSH_NONE |
1379 (dev_priv->color_fmt << 8) |
1380 RADEON_GMC_SRC_DATATYPE_COLOR |
1382 RADEON_DP_SRC_SOURCE_MEMORY |
1383 RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
1385 /* Make this work even if front & back are flipped:
1387 OUT_RING(CP_PACKET0(RADEON_SRC_PITCH_OFFSET, 1));
1388 if (dev_priv->sarea_priv->pfCurrentPage == 0) {
1389 OUT_RING(dev_priv->back_pitch_offset);
1390 OUT_RING(dev_priv->front_pitch_offset);
1392 OUT_RING(dev_priv->front_pitch_offset);
1393 OUT_RING(dev_priv->back_pitch_offset);
1396 OUT_RING(CP_PACKET0(RADEON_SRC_X_Y, 2));
1397 OUT_RING((x << 16) | y);
1398 OUT_RING((x << 16) | y);
1399 OUT_RING((w << 16) | h);
1404 /* Increment the frame counter. The client-side 3D driver must
1405 * throttle the framerate by waiting for this value before
1406 * performing the swapbuffer ioctl.
1408 dev_priv->sarea_priv->last_frame++;
1412 RADEON_FRAME_AGE(dev_priv->sarea_priv->last_frame);
1413 RADEON_WAIT_UNTIL_2D_IDLE();
1418 static void radeon_cp_dispatch_flip(struct drm_device * dev)
1420 drm_radeon_private_t *dev_priv = dev->dev_private;
1421 struct drm_sarea *sarea = (struct drm_sarea *) dev_priv->sarea->handle;
1422 int offset = (dev_priv->sarea_priv->pfCurrentPage == 1)
1423 ? dev_priv->front_offset : dev_priv->back_offset;
1425 DRM_DEBUG("%s: pfCurrentPage=%d\n",
1427 dev_priv->sarea_priv->pfCurrentPage);
1429 /* Do some trivial performance monitoring...
1431 if (dev_priv->do_boxes) {
1432 dev_priv->stats.boxes |= RADEON_BOX_FLIP;
1433 radeon_cp_performance_boxes(dev_priv);
1436 /* Update the frame offsets for both CRTCs
1440 RADEON_WAIT_UNTIL_3D_IDLE();
1441 OUT_RING_REG(RADEON_CRTC_OFFSET,
1442 ((sarea->frame.y * dev_priv->front_pitch +
1443 sarea->frame.x * (dev_priv->color_fmt - 2)) & ~7)
1445 OUT_RING_REG(RADEON_CRTC2_OFFSET, dev_priv->sarea_priv->crtc2_base
1450 /* Increment the frame counter. The client-side 3D driver must
1451 * throttle the framerate by waiting for this value before
1452 * performing the swapbuffer ioctl.
1454 dev_priv->sarea_priv->last_frame++;
1455 dev_priv->sarea_priv->pfCurrentPage =
1456 1 - dev_priv->sarea_priv->pfCurrentPage;
1460 RADEON_FRAME_AGE(dev_priv->sarea_priv->last_frame);
1465 static int bad_prim_vertex_nr(int primitive, int nr)
1467 switch (primitive & RADEON_PRIM_TYPE_MASK) {
1468 case RADEON_PRIM_TYPE_NONE:
1469 case RADEON_PRIM_TYPE_POINT:
1471 case RADEON_PRIM_TYPE_LINE:
1472 return (nr & 1) || nr == 0;
1473 case RADEON_PRIM_TYPE_LINE_STRIP:
1475 case RADEON_PRIM_TYPE_TRI_LIST:
1476 case RADEON_PRIM_TYPE_3VRT_POINT_LIST:
1477 case RADEON_PRIM_TYPE_3VRT_LINE_LIST:
1478 case RADEON_PRIM_TYPE_RECT_LIST:
1479 return nr % 3 || nr == 0;
1480 case RADEON_PRIM_TYPE_TRI_FAN:
1481 case RADEON_PRIM_TYPE_TRI_STRIP:
1490 unsigned int finish;
1492 unsigned int numverts;
1493 unsigned int offset;
1494 unsigned int vc_format;
1495 } drm_radeon_tcl_prim_t;
1497 static void radeon_cp_dispatch_vertex(struct drm_device * dev,
1498 struct drm_buf * buf,
1499 drm_radeon_tcl_prim_t * prim)
1501 drm_radeon_private_t *dev_priv = dev->dev_private;
1502 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1503 int offset = dev_priv->gart_buffers_offset + buf->offset + prim->start;
1504 int numverts = (int)prim->numverts;
1505 int nbox = sarea_priv->nbox;
1509 DRM_DEBUG("hwprim 0x%x vfmt 0x%x %d..%d %d verts\n",
1511 prim->vc_format, prim->start, prim->finish, prim->numverts);
1513 if (bad_prim_vertex_nr(prim->prim, prim->numverts)) {
1514 DRM_ERROR("bad prim %x numverts %d\n",
1515 prim->prim, prim->numverts);
1520 /* Emit the next cliprect */
1522 radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1525 /* Emit the vertex buffer rendering commands */
1528 OUT_RING(CP_PACKET3(RADEON_3D_RNDR_GEN_INDX_PRIM, 3));
1531 OUT_RING(prim->vc_format);
1532 OUT_RING(prim->prim | RADEON_PRIM_WALK_LIST |
1533 RADEON_COLOR_ORDER_RGBA |
1534 RADEON_VTX_FMT_RADEON_MODE |
1535 (numverts << RADEON_NUM_VERTICES_SHIFT));
1543 static void radeon_cp_discard_buffer(struct drm_device * dev, struct drm_buf * buf)
1545 drm_radeon_private_t *dev_priv = dev->dev_private;
1546 drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
1549 buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
1551 /* Emit the vertex buffer age */
1553 RADEON_DISPATCH_AGE(buf_priv->age);
1560 static void radeon_cp_dispatch_indirect(struct drm_device * dev,
1561 struct drm_buf * buf, int start, int end)
1563 drm_radeon_private_t *dev_priv = dev->dev_private;
1565 DRM_DEBUG("indirect: buf=%d s=0x%x e=0x%x\n", buf->idx, start, end);
1568 int offset = (dev_priv->gart_buffers_offset
1569 + buf->offset + start);
1570 int dwords = (end - start + 3) / sizeof(u32);
1572 /* Indirect buffer data must be an even number of
1573 * dwords, so if we've been given an odd number we must
1574 * pad the data with a Type-2 CP packet.
1578 ((char *)dev->agp_buffer_map->handle
1579 + buf->offset + start);
1580 data[dwords++] = RADEON_CP_PACKET2;
1583 /* Fire off the indirect buffer */
1586 OUT_RING(CP_PACKET0(RADEON_CP_IB_BASE, 1));
1594 static void radeon_cp_dispatch_indices(struct drm_device * dev,
1595 struct drm_buf * elt_buf,
1596 drm_radeon_tcl_prim_t * prim)
1598 drm_radeon_private_t *dev_priv = dev->dev_private;
1599 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1600 int offset = dev_priv->gart_buffers_offset + prim->offset;
1604 int start = prim->start + RADEON_INDEX_PRIM_OFFSET;
1605 int count = (prim->finish - start) / sizeof(u16);
1606 int nbox = sarea_priv->nbox;
1608 DRM_DEBUG("hwprim 0x%x vfmt 0x%x %d..%d offset: %x nr %d\n",
1611 prim->start, prim->finish, prim->offset, prim->numverts);
1613 if (bad_prim_vertex_nr(prim->prim, count)) {
1614 DRM_ERROR("bad prim %x count %d\n", prim->prim, count);
1618 if (start >= prim->finish || (prim->start & 0x7)) {
1619 DRM_ERROR("buffer prim %d\n", prim->prim);
1623 dwords = (prim->finish - prim->start + 3) / sizeof(u32);
1625 data = (u32 *) ((char *)dev->agp_buffer_map->handle +
1626 elt_buf->offset + prim->start);
1628 data[0] = CP_PACKET3(RADEON_3D_RNDR_GEN_INDX_PRIM, dwords - 2);
1630 data[2] = prim->numverts;
1631 data[3] = prim->vc_format;
1632 data[4] = (prim->prim |
1633 RADEON_PRIM_WALK_IND |
1634 RADEON_COLOR_ORDER_RGBA |
1635 RADEON_VTX_FMT_RADEON_MODE |
1636 (count << RADEON_NUM_VERTICES_SHIFT));
1640 radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1642 radeon_cp_dispatch_indirect(dev, elt_buf,
1643 prim->start, prim->finish);
1650 #define RADEON_MAX_TEXTURE_SIZE RADEON_BUFFER_SIZE
1652 static int radeon_cp_dispatch_texture(struct drm_device * dev,
1653 struct drm_file *file_priv,
1654 drm_radeon_texture_t * tex,
1655 drm_radeon_tex_image_t * image)
1657 drm_radeon_private_t *dev_priv = dev->dev_private;
1658 struct drm_buf *buf;
1661 const u8 __user *data;
1662 int size, dwords, tex_width, blit_width, spitch;
1665 u32 texpitch, microtile;
1669 if (radeon_check_and_fixup_offset(dev_priv, file_priv, &tex->offset)) {
1670 DRM_ERROR("Invalid destination offset\n");
1674 dev_priv->stats.boxes |= RADEON_BOX_TEXTURE_LOAD;
1676 /* Flush the pixel cache. This ensures no pixel data gets mixed
1677 * up with the texture data from the host data blit, otherwise
1678 * part of the texture image may be corrupted.
1681 RADEON_FLUSH_CACHE();
1682 RADEON_WAIT_UNTIL_IDLE();
1685 /* The compiler won't optimize away a division by a variable,
1686 * even if the only legal values are powers of two. Thus, we'll
1687 * use a shift instead.
1689 switch (tex->format) {
1690 case RADEON_TXFORMAT_ARGB8888:
1691 case RADEON_TXFORMAT_RGBA8888:
1692 format = RADEON_COLOR_FORMAT_ARGB8888;
1693 tex_width = tex->width * 4;
1694 blit_width = image->width * 4;
1696 case RADEON_TXFORMAT_AI88:
1697 case RADEON_TXFORMAT_ARGB1555:
1698 case RADEON_TXFORMAT_RGB565:
1699 case RADEON_TXFORMAT_ARGB4444:
1700 case RADEON_TXFORMAT_VYUY422:
1701 case RADEON_TXFORMAT_YVYU422:
1702 format = RADEON_COLOR_FORMAT_RGB565;
1703 tex_width = tex->width * 2;
1704 blit_width = image->width * 2;
1706 case RADEON_TXFORMAT_I8:
1707 case RADEON_TXFORMAT_RGB332:
1708 format = RADEON_COLOR_FORMAT_CI8;
1709 tex_width = tex->width * 1;
1710 blit_width = image->width * 1;
1713 DRM_ERROR("invalid texture format %d\n", tex->format);
1716 spitch = blit_width >> 6;
1717 if (spitch == 0 && image->height > 1)
1720 texpitch = tex->pitch;
1721 if ((texpitch << 22) & RADEON_DST_TILE_MICRO) {
1723 if (tex_width < 64) {
1724 texpitch &= ~(RADEON_DST_TILE_MICRO >> 22);
1725 /* we got tiled coordinates, untile them */
1731 DRM_DEBUG("tex=%dx%d blit=%d\n", tex_width, tex->height, blit_width);
1734 DRM_DEBUG("tex: ofs=0x%x p=%d f=%d x=%hd y=%hd w=%hd h=%hd\n",
1735 tex->offset >> 10, tex->pitch, tex->format,
1736 image->x, image->y, image->width, image->height);
1738 /* Make a copy of some parameters in case we have to
1739 * update them for a multi-pass texture blit.
1741 height = image->height;
1742 data = (const u8 __user *)image->data;
1744 size = height * blit_width;
1746 if (size > RADEON_MAX_TEXTURE_SIZE) {
1747 height = RADEON_MAX_TEXTURE_SIZE / blit_width;
1748 size = height * blit_width;
1749 } else if (size < 4 && size > 0) {
1751 } else if (size == 0) {
1755 buf = radeon_freelist_get(dev);
1757 radeon_do_cp_idle(dev_priv);
1758 buf = radeon_freelist_get(dev);
1761 DRM_DEBUG("radeon_cp_dispatch_texture: EAGAIN\n");
1762 if (DRM_COPY_TO_USER(tex->image, image, sizeof(*image)))
1767 /* Dispatch the indirect buffer.
1770 (u32 *) ((char *)dev->agp_buffer_map->handle + buf->offset);
1773 #define RADEON_COPY_MT(_buf, _data, _width) \
1775 if (DRM_COPY_FROM_USER(_buf, _data, (_width))) {\
1776 DRM_ERROR("EFAULT on pad, %d bytes\n", (_width)); \
1782 /* texture micro tiling in use, minimum texture width is thus 16 bytes.
1783 however, we cannot use blitter directly for texture width < 64 bytes,
1784 since minimum tex pitch is 64 bytes and we need this to match
1785 the texture width, otherwise the blitter will tile it wrong.
1786 Thus, tiling manually in this case. Additionally, need to special
1787 case tex height = 1, since our actual image will have height 2
1788 and we need to ensure we don't read beyond the texture size
1790 if (tex->height == 1) {
1791 if (tex_width >= 64 || tex_width <= 16) {
1792 RADEON_COPY_MT(buffer, data,
1793 (int)(tex_width * sizeof(u32)));
1794 } else if (tex_width == 32) {
1795 RADEON_COPY_MT(buffer, data, 16);
1796 RADEON_COPY_MT(buffer + 8,
1799 } else if (tex_width >= 64 || tex_width == 16) {
1800 RADEON_COPY_MT(buffer, data,
1801 (int)(dwords * sizeof(u32)));
1802 } else if (tex_width < 16) {
1803 for (i = 0; i < tex->height; i++) {
1804 RADEON_COPY_MT(buffer, data, tex_width);
1808 } else if (tex_width == 32) {
1809 /* TODO: make sure this works when not fitting in one buffer
1810 (i.e. 32bytes x 2048...) */
1811 for (i = 0; i < tex->height; i += 2) {
1812 RADEON_COPY_MT(buffer, data, 16);
1814 RADEON_COPY_MT(buffer + 8, data, 16);
1816 RADEON_COPY_MT(buffer + 4, data, 16);
1818 RADEON_COPY_MT(buffer + 12, data, 16);
1824 if (tex_width >= 32) {
1825 /* Texture image width is larger than the minimum, so we
1826 * can upload it directly.
1828 RADEON_COPY_MT(buffer, data,
1829 (int)(dwords * sizeof(u32)));
1831 /* Texture image width is less than the minimum, so we
1832 * need to pad out each image scanline to the minimum
1835 for (i = 0; i < tex->height; i++) {
1836 RADEON_COPY_MT(buffer, data, tex_width);
1843 #undef RADEON_COPY_MT
1844 buf->file_priv = file_priv;
1846 offset = dev_priv->gart_buffers_offset + buf->offset;
1848 OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 5));
1849 OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
1850 RADEON_GMC_DST_PITCH_OFFSET_CNTL |
1851 RADEON_GMC_BRUSH_NONE |
1853 RADEON_GMC_SRC_DATATYPE_COLOR |
1855 RADEON_DP_SRC_SOURCE_MEMORY |
1856 RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
1857 OUT_RING((spitch << 22) | (offset >> 10));
1858 OUT_RING((texpitch << 22) | (tex->offset >> 10));
1860 OUT_RING((image->x << 16) | image->y);
1861 OUT_RING((image->width << 16) | height);
1862 RADEON_WAIT_UNTIL_2D_IDLE();
1866 radeon_cp_discard_buffer(dev, buf);
1868 /* Update the input parameters for next time */
1870 image->height -= height;
1871 image->data = (const u8 __user *)image->data + size;
1872 } while (image->height > 0);
1874 /* Flush the pixel cache after the blit completes. This ensures
1875 * the texture data is written out to memory before rendering
1879 RADEON_FLUSH_CACHE();
1880 RADEON_WAIT_UNTIL_2D_IDLE();
1887 static void radeon_cp_dispatch_stipple(struct drm_device * dev, u32 * stipple)
1889 drm_radeon_private_t *dev_priv = dev->dev_private;
1896 OUT_RING(CP_PACKET0(RADEON_RE_STIPPLE_ADDR, 0));
1897 OUT_RING(0x00000000);
1899 OUT_RING(CP_PACKET0_TABLE(RADEON_RE_STIPPLE_DATA, 31));
1900 for (i = 0; i < 32; i++) {
1901 OUT_RING(stipple[i]);
1907 static void radeon_apply_surface_regs(int surf_index,
1908 drm_radeon_private_t *dev_priv)
1910 if (!dev_priv->mmio)
1913 radeon_do_cp_idle(dev_priv);
1915 RADEON_WRITE(RADEON_SURFACE0_INFO + 16 * surf_index,
1916 dev_priv->surfaces[surf_index].flags);
1917 RADEON_WRITE(RADEON_SURFACE0_LOWER_BOUND + 16 * surf_index,
1918 dev_priv->surfaces[surf_index].lower);
1919 RADEON_WRITE(RADEON_SURFACE0_UPPER_BOUND + 16 * surf_index,
1920 dev_priv->surfaces[surf_index].upper);
1923 /* Allocates a virtual surface
1924 * doesn't always allocate a real surface, will stretch an existing
1925 * surface when possible.
1927 * Note that refcount can be at most 2, since during a free refcount=3
1928 * might mean we have to allocate a new surface which might not always
1930 * For example : we allocate three contigous surfaces ABC. If B is
1931 * freed, we suddenly need two surfaces to store A and C, which might
1932 * not always be available.
1934 static int alloc_surface(drm_radeon_surface_alloc_t *new,
1935 drm_radeon_private_t *dev_priv,
1936 struct drm_file *file_priv)
1938 struct radeon_virt_surface *s;
1940 int virt_surface_index;
1941 uint32_t new_upper, new_lower;
1943 new_lower = new->address;
1944 new_upper = new_lower + new->size - 1;
1947 if ((new_lower >= new_upper) || (new->flags == 0) || (new->size == 0) ||
1948 ((new_upper & RADEON_SURF_ADDRESS_FIXED_MASK) !=
1949 RADEON_SURF_ADDRESS_FIXED_MASK)
1950 || ((new_lower & RADEON_SURF_ADDRESS_FIXED_MASK) != 0))
1953 /* make sure there is no overlap with existing surfaces */
1954 for (i = 0; i < RADEON_MAX_SURFACES; i++) {
1955 if ((dev_priv->surfaces[i].refcount != 0) &&
1956 (((new_lower >= dev_priv->surfaces[i].lower) &&
1957 (new_lower < dev_priv->surfaces[i].upper)) ||
1958 ((new_lower < dev_priv->surfaces[i].lower) &&
1959 (new_upper > dev_priv->surfaces[i].lower)))) {
1964 /* find a virtual surface */
1965 for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++)
1966 if (dev_priv->virt_surfaces[i].file_priv == 0)
1968 if (i == 2 * RADEON_MAX_SURFACES) {
1971 virt_surface_index = i;
1973 /* try to reuse an existing surface */
1974 for (i = 0; i < RADEON_MAX_SURFACES; i++) {
1976 if ((dev_priv->surfaces[i].refcount == 1) &&
1977 (new->flags == dev_priv->surfaces[i].flags) &&
1978 (new_upper + 1 == dev_priv->surfaces[i].lower)) {
1979 s = &(dev_priv->virt_surfaces[virt_surface_index]);
1980 s->surface_index = i;
1981 s->lower = new_lower;
1982 s->upper = new_upper;
1983 s->flags = new->flags;
1984 s->file_priv = file_priv;
1985 dev_priv->surfaces[i].refcount++;
1986 dev_priv->surfaces[i].lower = s->lower;
1987 radeon_apply_surface_regs(s->surface_index, dev_priv);
1988 return virt_surface_index;
1992 if ((dev_priv->surfaces[i].refcount == 1) &&
1993 (new->flags == dev_priv->surfaces[i].flags) &&
1994 (new_lower == dev_priv->surfaces[i].upper + 1)) {
1995 s = &(dev_priv->virt_surfaces[virt_surface_index]);
1996 s->surface_index = i;
1997 s->lower = new_lower;
1998 s->upper = new_upper;
1999 s->flags = new->flags;
2000 s->file_priv = file_priv;
2001 dev_priv->surfaces[i].refcount++;
2002 dev_priv->surfaces[i].upper = s->upper;
2003 radeon_apply_surface_regs(s->surface_index, dev_priv);
2004 return virt_surface_index;
2008 /* okay, we need a new one */
2009 for (i = 0; i < RADEON_MAX_SURFACES; i++) {
2010 if (dev_priv->surfaces[i].refcount == 0) {
2011 s = &(dev_priv->virt_surfaces[virt_surface_index]);
2012 s->surface_index = i;
2013 s->lower = new_lower;
2014 s->upper = new_upper;
2015 s->flags = new->flags;
2016 s->file_priv = file_priv;
2017 dev_priv->surfaces[i].refcount = 1;
2018 dev_priv->surfaces[i].lower = s->lower;
2019 dev_priv->surfaces[i].upper = s->upper;
2020 dev_priv->surfaces[i].flags = s->flags;
2021 radeon_apply_surface_regs(s->surface_index, dev_priv);
2022 return virt_surface_index;
2026 /* we didn't find anything */
2030 static int free_surface(struct drm_file *file_priv,
2031 drm_radeon_private_t * dev_priv,
2034 struct radeon_virt_surface *s;
2036 /* find the virtual surface */
2037 for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++) {
2038 s = &(dev_priv->virt_surfaces[i]);
2040 if ((lower == s->lower) && (file_priv == s->file_priv))
2042 if (dev_priv->surfaces[s->surface_index].
2044 dev_priv->surfaces[s->surface_index].
2047 if (dev_priv->surfaces[s->surface_index].
2049 dev_priv->surfaces[s->surface_index].
2052 dev_priv->surfaces[s->surface_index].refcount--;
2053 if (dev_priv->surfaces[s->surface_index].
2055 dev_priv->surfaces[s->surface_index].
2057 s->file_priv = NULL;
2058 radeon_apply_surface_regs(s->surface_index,
2067 static void radeon_surfaces_release(struct drm_file *file_priv,
2068 drm_radeon_private_t * dev_priv)
2071 for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++) {
2072 if (dev_priv->virt_surfaces[i].file_priv == file_priv)
2073 free_surface(file_priv, dev_priv,
2074 dev_priv->virt_surfaces[i].lower);
2078 /* ================================================================
2081 static int radeon_surface_alloc(struct drm_device *dev, void *data, struct drm_file *file_priv)
2083 drm_radeon_private_t *dev_priv = dev->dev_private;
2084 drm_radeon_surface_alloc_t *alloc = data;
2086 if (alloc_surface(alloc, dev_priv, file_priv) == -1)
2092 static int radeon_surface_free(struct drm_device *dev, void *data, struct drm_file *file_priv)
2094 drm_radeon_private_t *dev_priv = dev->dev_private;
2095 drm_radeon_surface_free_t *memfree = data;
2097 if (free_surface(file_priv, dev_priv, memfree->address))
2103 static int radeon_cp_clear(struct drm_device *dev, void *data, struct drm_file *file_priv)
2105 drm_radeon_private_t *dev_priv = dev->dev_private;
2106 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2107 drm_radeon_clear_t *clear = data;
2108 drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
2111 LOCK_TEST_WITH_RETURN(dev, file_priv);
2113 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2115 if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2116 sarea_priv->nbox = RADEON_NR_SAREA_CLIPRECTS;
2118 if (DRM_COPY_FROM_USER(&depth_boxes, clear->depth_boxes,
2119 sarea_priv->nbox * sizeof(depth_boxes[0])))
2122 radeon_cp_dispatch_clear(dev, clear, depth_boxes);
2128 /* Not sure why this isn't set all the time:
2130 static int radeon_do_init_pageflip(struct drm_device * dev)
2132 drm_radeon_private_t *dev_priv = dev->dev_private;
2138 RADEON_WAIT_UNTIL_3D_IDLE();
2139 OUT_RING(CP_PACKET0(RADEON_CRTC_OFFSET_CNTL, 0));
2140 OUT_RING(RADEON_READ(RADEON_CRTC_OFFSET_CNTL) |
2141 RADEON_CRTC_OFFSET_FLIP_CNTL);
2142 OUT_RING(CP_PACKET0(RADEON_CRTC2_OFFSET_CNTL, 0));
2143 OUT_RING(RADEON_READ(RADEON_CRTC2_OFFSET_CNTL) |
2144 RADEON_CRTC_OFFSET_FLIP_CNTL);
2147 dev_priv->page_flipping = 1;
2149 if (dev_priv->sarea_priv->pfCurrentPage != 1)
2150 dev_priv->sarea_priv->pfCurrentPage = 0;
2155 /* Swapping and flipping are different operations, need different ioctls.
2156 * They can & should be intermixed to support multiple 3d windows.
2158 static int radeon_cp_flip(struct drm_device *dev, void *data, struct drm_file *file_priv)
2160 drm_radeon_private_t *dev_priv = dev->dev_private;
2163 LOCK_TEST_WITH_RETURN(dev, file_priv);
2165 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2167 if (!dev_priv->page_flipping)
2168 radeon_do_init_pageflip(dev);
2170 radeon_cp_dispatch_flip(dev);
2176 static int radeon_cp_swap(struct drm_device *dev, void *data, struct drm_file *file_priv)
2178 drm_radeon_private_t *dev_priv = dev->dev_private;
2179 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2182 LOCK_TEST_WITH_RETURN(dev, file_priv);
2184 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2186 if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2187 sarea_priv->nbox = RADEON_NR_SAREA_CLIPRECTS;
2189 radeon_cp_dispatch_swap(dev);
2190 dev_priv->sarea_priv->ctx_owner = 0;
2196 static int radeon_cp_vertex(struct drm_device *dev, void *data, struct drm_file *file_priv)
2198 drm_radeon_private_t *dev_priv = dev->dev_private;
2199 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2200 struct drm_device_dma *dma = dev->dma;
2201 struct drm_buf *buf;
2202 drm_radeon_vertex_t *vertex = data;
2203 drm_radeon_tcl_prim_t prim;
2205 LOCK_TEST_WITH_RETURN(dev, file_priv);
2207 DRM_DEBUG("pid=%d index=%d count=%d discard=%d\n",
2208 DRM_CURRENTPID, vertex->idx, vertex->count, vertex->discard);
2210 if (vertex->idx < 0 || vertex->idx >= dma->buf_count) {
2211 DRM_ERROR("buffer index %d (of %d max)\n",
2212 vertex->idx, dma->buf_count - 1);
2215 if (vertex->prim < 0 || vertex->prim > RADEON_PRIM_TYPE_3VRT_LINE_LIST) {
2216 DRM_ERROR("buffer prim %d\n", vertex->prim);
2220 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2221 VB_AGE_TEST_WITH_RETURN(dev_priv);
2223 buf = dma->buflist[vertex->idx];
2225 if (buf->file_priv != file_priv) {
2226 DRM_ERROR("process %d using buffer owned by %p\n",
2227 DRM_CURRENTPID, buf->file_priv);
2231 DRM_ERROR("sending pending buffer %d\n", vertex->idx);
2235 /* Build up a prim_t record:
2237 if (vertex->count) {
2238 buf->used = vertex->count; /* not used? */
2240 if (sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS) {
2241 if (radeon_emit_state(dev_priv, file_priv,
2242 &sarea_priv->context_state,
2243 sarea_priv->tex_state,
2244 sarea_priv->dirty)) {
2245 DRM_ERROR("radeon_emit_state failed\n");
2249 sarea_priv->dirty &= ~(RADEON_UPLOAD_TEX0IMAGES |
2250 RADEON_UPLOAD_TEX1IMAGES |
2251 RADEON_UPLOAD_TEX2IMAGES |
2252 RADEON_REQUIRE_QUIESCENCE);
2256 prim.finish = vertex->count; /* unused */
2257 prim.prim = vertex->prim;
2258 prim.numverts = vertex->count;
2259 prim.vc_format = dev_priv->sarea_priv->vc_format;
2261 radeon_cp_dispatch_vertex(dev, buf, &prim);
2264 if (vertex->discard) {
2265 radeon_cp_discard_buffer(dev, buf);
2272 static int radeon_cp_indices(struct drm_device *dev, void *data, struct drm_file *file_priv)
2274 drm_radeon_private_t *dev_priv = dev->dev_private;
2275 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2276 struct drm_device_dma *dma = dev->dma;
2277 struct drm_buf *buf;
2278 drm_radeon_indices_t *elts = data;
2279 drm_radeon_tcl_prim_t prim;
2282 LOCK_TEST_WITH_RETURN(dev, file_priv);
2284 DRM_DEBUG("pid=%d index=%d start=%d end=%d discard=%d\n",
2285 DRM_CURRENTPID, elts->idx, elts->start, elts->end,
2288 if (elts->idx < 0 || elts->idx >= dma->buf_count) {
2289 DRM_ERROR("buffer index %d (of %d max)\n",
2290 elts->idx, dma->buf_count - 1);
2293 if (elts->prim < 0 || elts->prim > RADEON_PRIM_TYPE_3VRT_LINE_LIST) {
2294 DRM_ERROR("buffer prim %d\n", elts->prim);
2298 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2299 VB_AGE_TEST_WITH_RETURN(dev_priv);
2301 buf = dma->buflist[elts->idx];
2303 if (buf->file_priv != file_priv) {
2304 DRM_ERROR("process %d using buffer owned by %p\n",
2305 DRM_CURRENTPID, buf->file_priv);
2309 DRM_ERROR("sending pending buffer %d\n", elts->idx);
2313 count = (elts->end - elts->start) / sizeof(u16);
2314 elts->start -= RADEON_INDEX_PRIM_OFFSET;
2316 if (elts->start & 0x7) {
2317 DRM_ERROR("misaligned buffer 0x%x\n", elts->start);
2320 if (elts->start < buf->used) {
2321 DRM_ERROR("no header 0x%x - 0x%x\n", elts->start, buf->used);
2325 buf->used = elts->end;
2327 if (sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS) {
2328 if (radeon_emit_state(dev_priv, file_priv,
2329 &sarea_priv->context_state,
2330 sarea_priv->tex_state,
2331 sarea_priv->dirty)) {
2332 DRM_ERROR("radeon_emit_state failed\n");
2336 sarea_priv->dirty &= ~(RADEON_UPLOAD_TEX0IMAGES |
2337 RADEON_UPLOAD_TEX1IMAGES |
2338 RADEON_UPLOAD_TEX2IMAGES |
2339 RADEON_REQUIRE_QUIESCENCE);
2342 /* Build up a prim_t record:
2344 prim.start = elts->start;
2345 prim.finish = elts->end;
2346 prim.prim = elts->prim;
2347 prim.offset = 0; /* offset from start of dma buffers */
2348 prim.numverts = RADEON_MAX_VB_VERTS; /* duh */
2349 prim.vc_format = dev_priv->sarea_priv->vc_format;
2351 radeon_cp_dispatch_indices(dev, buf, &prim);
2352 if (elts->discard) {
2353 radeon_cp_discard_buffer(dev, buf);
2360 static int radeon_cp_texture(struct drm_device *dev, void *data, struct drm_file *file_priv)
2362 drm_radeon_private_t *dev_priv = dev->dev_private;
2363 drm_radeon_texture_t *tex = data;
2364 drm_radeon_tex_image_t image;
2367 LOCK_TEST_WITH_RETURN(dev, file_priv);
2369 if (tex->image == NULL) {
2370 DRM_ERROR("null texture image!\n");
2374 if (DRM_COPY_FROM_USER(&image,
2375 (drm_radeon_tex_image_t __user *) tex->image,
2379 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2380 VB_AGE_TEST_WITH_RETURN(dev_priv);
2382 ret = radeon_cp_dispatch_texture(dev, file_priv, tex, &image);
2387 static int radeon_cp_stipple(struct drm_device *dev, void *data, struct drm_file *file_priv)
2389 drm_radeon_private_t *dev_priv = dev->dev_private;
2390 drm_radeon_stipple_t *stipple = data;
2393 LOCK_TEST_WITH_RETURN(dev, file_priv);
2395 if (DRM_COPY_FROM_USER(&mask, stipple->mask, 32 * sizeof(u32)))
2398 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2400 radeon_cp_dispatch_stipple(dev, mask);
2406 static int radeon_cp_indirect(struct drm_device *dev, void *data, struct drm_file *file_priv)
2408 drm_radeon_private_t *dev_priv = dev->dev_private;
2409 struct drm_device_dma *dma = dev->dma;
2410 struct drm_buf *buf;
2411 drm_radeon_indirect_t *indirect = data;
2414 LOCK_TEST_WITH_RETURN(dev, file_priv);
2416 DRM_DEBUG("indirect: idx=%d s=%d e=%d d=%d\n",
2417 indirect->idx, indirect->start, indirect->end,
2420 if (indirect->idx < 0 || indirect->idx >= dma->buf_count) {
2421 DRM_ERROR("buffer index %d (of %d max)\n",
2422 indirect->idx, dma->buf_count - 1);
2426 buf = dma->buflist[indirect->idx];
2428 if (buf->file_priv != file_priv) {
2429 DRM_ERROR("process %d using buffer owned by %p\n",
2430 DRM_CURRENTPID, buf->file_priv);
2434 DRM_ERROR("sending pending buffer %d\n", indirect->idx);
2438 if (indirect->start < buf->used) {
2439 DRM_ERROR("reusing indirect: start=0x%x actual=0x%x\n",
2440 indirect->start, buf->used);
2444 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2445 VB_AGE_TEST_WITH_RETURN(dev_priv);
2447 buf->used = indirect->end;
2449 /* Wait for the 3D stream to idle before the indirect buffer
2450 * containing 2D acceleration commands is processed.
2454 RADEON_WAIT_UNTIL_3D_IDLE();
2458 /* Dispatch the indirect buffer full of commands from the
2459 * X server. This is insecure and is thus only available to
2460 * privileged clients.
2462 radeon_cp_dispatch_indirect(dev, buf, indirect->start, indirect->end);
2463 if (indirect->discard) {
2464 radeon_cp_discard_buffer(dev, buf);
2471 static int radeon_cp_vertex2(struct drm_device *dev, void *data, struct drm_file *file_priv)
2473 drm_radeon_private_t *dev_priv = dev->dev_private;
2474 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2475 struct drm_device_dma *dma = dev->dma;
2476 struct drm_buf *buf;
2477 drm_radeon_vertex2_t *vertex = data;
2479 unsigned char laststate;
2481 LOCK_TEST_WITH_RETURN(dev, file_priv);
2483 DRM_DEBUG("pid=%d index=%d discard=%d\n",
2484 DRM_CURRENTPID, vertex->idx, vertex->discard);
2486 if (vertex->idx < 0 || vertex->idx >= dma->buf_count) {
2487 DRM_ERROR("buffer index %d (of %d max)\n",
2488 vertex->idx, dma->buf_count - 1);
2492 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2493 VB_AGE_TEST_WITH_RETURN(dev_priv);
2495 buf = dma->buflist[vertex->idx];
2497 if (buf->file_priv != file_priv) {
2498 DRM_ERROR("process %d using buffer owned by %p\n",
2499 DRM_CURRENTPID, buf->file_priv);
2504 DRM_ERROR("sending pending buffer %d\n", vertex->idx);
2508 if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2511 for (laststate = 0xff, i = 0; i < vertex->nr_prims; i++) {
2512 drm_radeon_prim_t prim;
2513 drm_radeon_tcl_prim_t tclprim;
2515 if (DRM_COPY_FROM_USER(&prim, &vertex->prim[i], sizeof(prim)))
2518 if (prim.stateidx != laststate) {
2519 drm_radeon_state_t state;
2521 if (DRM_COPY_FROM_USER(&state,
2522 &vertex->state[prim.stateidx],
2526 if (radeon_emit_state2(dev_priv, file_priv, &state)) {
2527 DRM_ERROR("radeon_emit_state2 failed\n");
2531 laststate = prim.stateidx;
2534 tclprim.start = prim.start;
2535 tclprim.finish = prim.finish;
2536 tclprim.prim = prim.prim;
2537 tclprim.vc_format = prim.vc_format;
2539 if (prim.prim & RADEON_PRIM_WALK_IND) {
2540 tclprim.offset = prim.numverts * 64;
2541 tclprim.numverts = RADEON_MAX_VB_VERTS; /* duh */
2543 radeon_cp_dispatch_indices(dev, buf, &tclprim);
2545 tclprim.numverts = prim.numverts;
2546 tclprim.offset = 0; /* not used */
2548 radeon_cp_dispatch_vertex(dev, buf, &tclprim);
2551 if (sarea_priv->nbox == 1)
2552 sarea_priv->nbox = 0;
2555 if (vertex->discard) {
2556 radeon_cp_discard_buffer(dev, buf);
2563 static int radeon_emit_packets(drm_radeon_private_t * dev_priv,
2564 struct drm_file *file_priv,
2565 drm_radeon_cmd_header_t header,
2566 drm_radeon_kcmd_buffer_t *cmdbuf)
2568 int id = (int)header.packet.packet_id;
2570 int *data = (int *)cmdbuf->buf;
2573 if (id >= RADEON_MAX_STATE_PACKETS)
2576 sz = packet[id].len;
2577 reg = packet[id].start;
2579 if (sz * sizeof(int) > cmdbuf->bufsz) {
2580 DRM_ERROR("Packet size provided larger than data provided\n");
2584 if (radeon_check_and_fixup_packets(dev_priv, file_priv, id, data)) {
2585 DRM_ERROR("Packet verification failed\n");
2590 OUT_RING(CP_PACKET0(reg, (sz - 1)));
2591 OUT_RING_TABLE(data, sz);
2594 cmdbuf->buf += sz * sizeof(int);
2595 cmdbuf->bufsz -= sz * sizeof(int);
2599 static __inline__ int radeon_emit_scalars(drm_radeon_private_t *dev_priv,
2600 drm_radeon_cmd_header_t header,
2601 drm_radeon_kcmd_buffer_t *cmdbuf)
2603 int sz = header.scalars.count;
2604 int start = header.scalars.offset;
2605 int stride = header.scalars.stride;
2609 OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
2610 OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
2611 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
2612 OUT_RING_TABLE(cmdbuf->buf, sz);
2614 cmdbuf->buf += sz * sizeof(int);
2615 cmdbuf->bufsz -= sz * sizeof(int);
2621 static __inline__ int radeon_emit_scalars2(drm_radeon_private_t *dev_priv,
2622 drm_radeon_cmd_header_t header,
2623 drm_radeon_kcmd_buffer_t *cmdbuf)
2625 int sz = header.scalars.count;
2626 int start = ((unsigned int)header.scalars.offset) + 0x100;
2627 int stride = header.scalars.stride;
2631 OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
2632 OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
2633 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
2634 OUT_RING_TABLE(cmdbuf->buf, sz);
2636 cmdbuf->buf += sz * sizeof(int);
2637 cmdbuf->bufsz -= sz * sizeof(int);
2641 static __inline__ int radeon_emit_vectors(drm_radeon_private_t *dev_priv,
2642 drm_radeon_cmd_header_t header,
2643 drm_radeon_kcmd_buffer_t *cmdbuf)
2645 int sz = header.vectors.count;
2646 int start = header.vectors.offset;
2647 int stride = header.vectors.stride;
2651 OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
2652 OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
2653 OUT_RING(start | (stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
2654 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
2655 OUT_RING_TABLE(cmdbuf->buf, sz);
2658 cmdbuf->buf += sz * sizeof(int);
2659 cmdbuf->bufsz -= sz * sizeof(int);
2663 static __inline__ int radeon_emit_veclinear(drm_radeon_private_t *dev_priv,
2664 drm_radeon_cmd_header_t header,
2665 drm_radeon_kcmd_buffer_t *cmdbuf)
2667 int sz = header.veclinear.count * 4;
2668 int start = header.veclinear.addr_lo | (header.veclinear.addr_hi << 8);
2673 if (sz * 4 > cmdbuf->bufsz)
2677 OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
2678 OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
2679 OUT_RING(start | (1 << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
2680 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
2681 OUT_RING_TABLE(cmdbuf->buf, sz);
2684 cmdbuf->buf += sz * sizeof(int);
2685 cmdbuf->bufsz -= sz * sizeof(int);
2689 static int radeon_emit_packet3(struct drm_device * dev,
2690 struct drm_file *file_priv,
2691 drm_radeon_kcmd_buffer_t *cmdbuf)
2693 drm_radeon_private_t *dev_priv = dev->dev_private;
2700 if ((ret = radeon_check_and_fixup_packet3(dev_priv, file_priv,
2702 DRM_ERROR("Packet verification failed\n");
2707 OUT_RING_TABLE(cmdbuf->buf, cmdsz);
2710 cmdbuf->buf += cmdsz * 4;
2711 cmdbuf->bufsz -= cmdsz * 4;
2715 static int radeon_emit_packet3_cliprect(struct drm_device *dev,
2716 struct drm_file *file_priv,
2717 drm_radeon_kcmd_buffer_t *cmdbuf,
2720 drm_radeon_private_t *dev_priv = dev->dev_private;
2721 struct drm_clip_rect box;
2724 struct drm_clip_rect __user *boxes = cmdbuf->boxes;
2730 if ((ret = radeon_check_and_fixup_packet3(dev_priv, file_priv,
2732 DRM_ERROR("Packet verification failed\n");
2740 if (i < cmdbuf->nbox) {
2741 if (DRM_COPY_FROM_USER(&box, &boxes[i], sizeof(box)))
2743 /* FIXME The second and subsequent times round
2744 * this loop, send a WAIT_UNTIL_3D_IDLE before
2745 * calling emit_clip_rect(). This fixes a
2746 * lockup on fast machines when sending
2747 * several cliprects with a cmdbuf, as when
2748 * waving a 2D window over a 3D
2749 * window. Something in the commands from user
2750 * space seems to hang the card when they're
2751 * sent several times in a row. That would be
2752 * the correct place to fix it but this works
2753 * around it until I can figure that out - Tim
2757 RADEON_WAIT_UNTIL_3D_IDLE();
2760 radeon_emit_clip_rect(dev_priv, &box);
2764 OUT_RING_TABLE(cmdbuf->buf, cmdsz);
2767 } while (++i < cmdbuf->nbox);
2768 if (cmdbuf->nbox == 1)
2772 cmdbuf->buf += cmdsz * 4;
2773 cmdbuf->bufsz -= cmdsz * 4;
2777 static int radeon_emit_wait(struct drm_device * dev, int flags)
2779 drm_radeon_private_t *dev_priv = dev->dev_private;
2782 DRM_DEBUG("%s: %x\n", __FUNCTION__, flags);
2784 case RADEON_WAIT_2D:
2786 RADEON_WAIT_UNTIL_2D_IDLE();
2789 case RADEON_WAIT_3D:
2791 RADEON_WAIT_UNTIL_3D_IDLE();
2794 case RADEON_WAIT_2D | RADEON_WAIT_3D:
2796 RADEON_WAIT_UNTIL_IDLE();
2806 static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file *file_priv)
2808 drm_radeon_private_t *dev_priv = dev->dev_private;
2809 struct drm_device_dma *dma = dev->dma;
2810 struct drm_buf *buf = NULL;
2812 drm_radeon_kcmd_buffer_t *cmdbuf = data;
2813 drm_radeon_cmd_header_t header;
2814 int orig_nbox, orig_bufsz;
2817 LOCK_TEST_WITH_RETURN(dev, file_priv);
2819 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2820 VB_AGE_TEST_WITH_RETURN(dev_priv);
2822 if (cmdbuf->bufsz > 64 * 1024 || cmdbuf->bufsz < 0) {
2826 /* Allocate an in-kernel area and copy in the cmdbuf. Do this to avoid
2827 * races between checking values and using those values in other code,
2828 * and simply to avoid a lot of function calls to copy in data.
2830 orig_bufsz = cmdbuf->bufsz;
2831 if (orig_bufsz != 0) {
2832 kbuf = drm_alloc(cmdbuf->bufsz, DRM_MEM_DRIVER);
2835 if (DRM_COPY_FROM_USER(kbuf, (void __user *)cmdbuf->buf,
2837 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2843 orig_nbox = cmdbuf->nbox;
2845 if (dev_priv->microcode_version == UCODE_R300) {
2847 temp = r300_do_cp_cmdbuf(dev, file_priv, cmdbuf);
2849 if (orig_bufsz != 0)
2850 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2855 /* microcode_version != r300 */
2856 while (cmdbuf->bufsz >= sizeof(header)) {
2858 header.i = *(int *)cmdbuf->buf;
2859 cmdbuf->buf += sizeof(header);
2860 cmdbuf->bufsz -= sizeof(header);
2862 switch (header.header.cmd_type) {
2863 case RADEON_CMD_PACKET:
2864 DRM_DEBUG("RADEON_CMD_PACKET\n");
2865 if (radeon_emit_packets
2866 (dev_priv, file_priv, header, cmdbuf)) {
2867 DRM_ERROR("radeon_emit_packets failed\n");
2872 case RADEON_CMD_SCALARS:
2873 DRM_DEBUG("RADEON_CMD_SCALARS\n");
2874 if (radeon_emit_scalars(dev_priv, header, cmdbuf)) {
2875 DRM_ERROR("radeon_emit_scalars failed\n");
2880 case RADEON_CMD_VECTORS:
2881 DRM_DEBUG("RADEON_CMD_VECTORS\n");
2882 if (radeon_emit_vectors(dev_priv, header, cmdbuf)) {
2883 DRM_ERROR("radeon_emit_vectors failed\n");
2888 case RADEON_CMD_DMA_DISCARD:
2889 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
2890 idx = header.dma.buf_idx;
2891 if (idx < 0 || idx >= dma->buf_count) {
2892 DRM_ERROR("buffer index %d (of %d max)\n",
2893 idx, dma->buf_count - 1);
2897 buf = dma->buflist[idx];
2898 if (buf->file_priv != file_priv || buf->pending) {
2899 DRM_ERROR("bad buffer %p %p %d\n",
2900 buf->file_priv, file_priv,
2905 radeon_cp_discard_buffer(dev, buf);
2908 case RADEON_CMD_PACKET3:
2909 DRM_DEBUG("RADEON_CMD_PACKET3\n");
2910 if (radeon_emit_packet3(dev, file_priv, cmdbuf)) {
2911 DRM_ERROR("radeon_emit_packet3 failed\n");
2916 case RADEON_CMD_PACKET3_CLIP:
2917 DRM_DEBUG("RADEON_CMD_PACKET3_CLIP\n");
2918 if (radeon_emit_packet3_cliprect
2919 (dev, file_priv, cmdbuf, orig_nbox)) {
2920 DRM_ERROR("radeon_emit_packet3_clip failed\n");
2925 case RADEON_CMD_SCALARS2:
2926 DRM_DEBUG("RADEON_CMD_SCALARS2\n");
2927 if (radeon_emit_scalars2(dev_priv, header, cmdbuf)) {
2928 DRM_ERROR("radeon_emit_scalars2 failed\n");
2933 case RADEON_CMD_WAIT:
2934 DRM_DEBUG("RADEON_CMD_WAIT\n");
2935 if (radeon_emit_wait(dev, header.wait.flags)) {
2936 DRM_ERROR("radeon_emit_wait failed\n");
2940 case RADEON_CMD_VECLINEAR:
2941 DRM_DEBUG("RADEON_CMD_VECLINEAR\n");
2942 if (radeon_emit_veclinear(dev_priv, header, cmdbuf)) {
2943 DRM_ERROR("radeon_emit_veclinear failed\n");
2949 DRM_ERROR("bad cmd_type %d at %p\n",
2950 header.header.cmd_type,
2951 cmdbuf->buf - sizeof(header));
2956 if (orig_bufsz != 0)
2957 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2959 DRM_DEBUG("DONE\n");
2964 if (orig_bufsz != 0)
2965 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2969 static int radeon_cp_getparam(struct drm_device *dev, void *data, struct drm_file *file_priv)
2971 drm_radeon_private_t *dev_priv = dev->dev_private;
2972 drm_radeon_getparam_t *param = data;
2975 DRM_DEBUG("pid=%d\n", DRM_CURRENTPID);
2977 switch (param->param) {
2978 case RADEON_PARAM_GART_BUFFER_OFFSET:
2979 value = dev_priv->gart_buffers_offset;
2981 case RADEON_PARAM_LAST_FRAME:
2982 dev_priv->stats.last_frame_reads++;
2983 value = GET_SCRATCH(0);
2985 case RADEON_PARAM_LAST_DISPATCH:
2986 value = GET_SCRATCH(1);
2988 case RADEON_PARAM_LAST_CLEAR:
2989 dev_priv->stats.last_clear_reads++;
2990 value = GET_SCRATCH(2);
2992 case RADEON_PARAM_IRQ_NR:
2995 case RADEON_PARAM_GART_BASE:
2996 value = dev_priv->gart_vm_start;
2998 case RADEON_PARAM_REGISTER_HANDLE:
2999 value = dev_priv->mmio->offset;
3001 case RADEON_PARAM_STATUS_HANDLE:
3002 value = dev_priv->ring_rptr_offset;
3004 #if BITS_PER_LONG == 32
3006 * This ioctl() doesn't work on 64-bit platforms because hw_lock is a
3007 * pointer which can't fit into an int-sized variable. According to
3008 * Michel Dänzer, the ioctl() is only used on embedded platforms, so
3009 * not supporting it shouldn't be a problem. If the same functionality
3010 * is needed on 64-bit platforms, a new ioctl() would have to be added,
3011 * so backwards-compatibility for the embedded platforms can be
3012 * maintained. --davidm 4-Feb-2004.
3014 case RADEON_PARAM_SAREA_HANDLE:
3015 /* The lock is the first dword in the sarea. */
3016 value = (long)dev->lock.hw_lock;
3019 case RADEON_PARAM_GART_TEX_HANDLE:
3020 value = dev_priv->gart_textures_offset;
3022 case RADEON_PARAM_SCRATCH_OFFSET:
3023 if (!dev_priv->writeback_works)
3025 value = RADEON_SCRATCH_REG_OFFSET;
3027 case RADEON_PARAM_CARD_TYPE:
3028 if (dev_priv->flags & RADEON_IS_PCIE)
3029 value = RADEON_CARD_PCIE;
3030 else if (dev_priv->flags & RADEON_IS_AGP)
3031 value = RADEON_CARD_AGP;
3033 value = RADEON_CARD_PCI;
3035 case RADEON_PARAM_VBLANK_CRTC:
3036 value = radeon_vblank_crtc_get(dev);
3039 DRM_DEBUG("Invalid parameter %d\n", param->param);
3043 if (DRM_COPY_TO_USER(param->value, &value, sizeof(int))) {
3044 DRM_ERROR("copy_to_user\n");
3051 static int radeon_cp_setparam(struct drm_device *dev, void *data, struct drm_file *file_priv)
3053 drm_radeon_private_t *dev_priv = dev->dev_private;
3054 drm_radeon_setparam_t *sp = data;
3055 struct drm_radeon_driver_file_fields *radeon_priv;
3057 switch (sp->param) {
3058 case RADEON_SETPARAM_FB_LOCATION:
3059 radeon_priv = file_priv->driver_priv;
3060 radeon_priv->radeon_fb_delta = dev_priv->fb_location -
3063 case RADEON_SETPARAM_SWITCH_TILING:
3064 if (sp->value == 0) {
3065 DRM_DEBUG("color tiling disabled\n");
3066 dev_priv->front_pitch_offset &= ~RADEON_DST_TILE_MACRO;
3067 dev_priv->back_pitch_offset &= ~RADEON_DST_TILE_MACRO;
3068 dev_priv->sarea_priv->tiling_enabled = 0;
3069 } else if (sp->value == 1) {
3070 DRM_DEBUG("color tiling enabled\n");
3071 dev_priv->front_pitch_offset |= RADEON_DST_TILE_MACRO;
3072 dev_priv->back_pitch_offset |= RADEON_DST_TILE_MACRO;
3073 dev_priv->sarea_priv->tiling_enabled = 1;
3076 case RADEON_SETPARAM_PCIGART_LOCATION:
3077 dev_priv->pcigart_offset = sp->value;
3078 dev_priv->pcigart_offset_set = 1;
3080 case RADEON_SETPARAM_NEW_MEMMAP:
3081 dev_priv->new_memmap = sp->value;
3083 case RADEON_SETPARAM_PCIGART_TABLE_SIZE:
3084 dev_priv->gart_info.table_size = sp->value;
3085 if (dev_priv->gart_info.table_size < RADEON_PCIGART_TABLE_SIZE)
3086 dev_priv->gart_info.table_size = RADEON_PCIGART_TABLE_SIZE;
3088 case RADEON_SETPARAM_VBLANK_CRTC:
3089 return radeon_vblank_crtc_set(dev, sp->value);
3092 DRM_DEBUG("Invalid parameter %d\n", sp->param);
3099 /* When a client dies:
3100 * - Check for and clean up flipped page state
3101 * - Free any alloced GART memory.
3102 * - Free any alloced radeon surfaces.
3104 * DRM infrastructure takes care of reclaiming dma buffers.
3106 void radeon_driver_preclose(struct drm_device *dev, struct drm_file *file_priv)
3108 if (dev->dev_private) {
3109 drm_radeon_private_t *dev_priv = dev->dev_private;
3110 dev_priv->page_flipping = 0;
3111 radeon_mem_release(file_priv, dev_priv->gart_heap);
3112 radeon_mem_release(file_priv, dev_priv->fb_heap);
3113 radeon_surfaces_release(file_priv, dev_priv);
3117 void radeon_driver_lastclose(struct drm_device *dev)
3119 if (dev->dev_private) {
3120 drm_radeon_private_t *dev_priv = dev->dev_private;
3122 if (dev_priv->sarea_priv &&
3123 dev_priv->sarea_priv->pfCurrentPage != 0)
3124 radeon_cp_dispatch_flip(dev);
3127 radeon_do_release(dev);
3130 int radeon_driver_open(struct drm_device *dev, struct drm_file *file_priv)
3132 drm_radeon_private_t *dev_priv = dev->dev_private;
3133 struct drm_radeon_driver_file_fields *radeon_priv;
3137 (struct drm_radeon_driver_file_fields *)
3138 drm_alloc(sizeof(*radeon_priv), DRM_MEM_FILES);
3143 file_priv->driver_priv = radeon_priv;
3146 radeon_priv->radeon_fb_delta = dev_priv->fb_location;
3148 radeon_priv->radeon_fb_delta = 0;
3152 void radeon_driver_postclose(struct drm_device *dev, struct drm_file *file_priv)
3154 struct drm_radeon_driver_file_fields *radeon_priv =
3155 file_priv->driver_priv;
3157 drm_free(radeon_priv, sizeof(*radeon_priv), DRM_MEM_FILES);
3160 struct drm_ioctl_desc radeon_ioctls[] = {
3161 DRM_IOCTL_DEF(DRM_RADEON_CP_INIT, radeon_cp_init, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3162 DRM_IOCTL_DEF(DRM_RADEON_CP_START, radeon_cp_start, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3163 DRM_IOCTL_DEF(DRM_RADEON_CP_STOP, radeon_cp_stop, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3164 DRM_IOCTL_DEF(DRM_RADEON_CP_RESET, radeon_cp_reset, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3165 DRM_IOCTL_DEF(DRM_RADEON_CP_IDLE, radeon_cp_idle, DRM_AUTH),
3166 DRM_IOCTL_DEF(DRM_RADEON_CP_RESUME, radeon_cp_resume, DRM_AUTH),
3167 DRM_IOCTL_DEF(DRM_RADEON_RESET, radeon_engine_reset, DRM_AUTH),
3168 DRM_IOCTL_DEF(DRM_RADEON_FULLSCREEN, radeon_fullscreen, DRM_AUTH),
3169 DRM_IOCTL_DEF(DRM_RADEON_SWAP, radeon_cp_swap, DRM_AUTH),
3170 DRM_IOCTL_DEF(DRM_RADEON_CLEAR, radeon_cp_clear, DRM_AUTH),
3171 DRM_IOCTL_DEF(DRM_RADEON_VERTEX, radeon_cp_vertex, DRM_AUTH),
3172 DRM_IOCTL_DEF(DRM_RADEON_INDICES, radeon_cp_indices, DRM_AUTH),
3173 DRM_IOCTL_DEF(DRM_RADEON_TEXTURE, radeon_cp_texture, DRM_AUTH),
3174 DRM_IOCTL_DEF(DRM_RADEON_STIPPLE, radeon_cp_stipple, DRM_AUTH),
3175 DRM_IOCTL_DEF(DRM_RADEON_INDIRECT, radeon_cp_indirect, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3176 DRM_IOCTL_DEF(DRM_RADEON_VERTEX2, radeon_cp_vertex2, DRM_AUTH),
3177 DRM_IOCTL_DEF(DRM_RADEON_CMDBUF, radeon_cp_cmdbuf, DRM_AUTH),
3178 DRM_IOCTL_DEF(DRM_RADEON_GETPARAM, radeon_cp_getparam, DRM_AUTH),
3179 DRM_IOCTL_DEF(DRM_RADEON_FLIP, radeon_cp_flip, DRM_AUTH),
3180 DRM_IOCTL_DEF(DRM_RADEON_ALLOC, radeon_mem_alloc, DRM_AUTH),
3181 DRM_IOCTL_DEF(DRM_RADEON_FREE, radeon_mem_free, DRM_AUTH),
3182 DRM_IOCTL_DEF(DRM_RADEON_INIT_HEAP, radeon_mem_init_heap, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3183 DRM_IOCTL_DEF(DRM_RADEON_IRQ_EMIT, radeon_irq_emit, DRM_AUTH),
3184 DRM_IOCTL_DEF(DRM_RADEON_IRQ_WAIT, radeon_irq_wait, DRM_AUTH),
3185 DRM_IOCTL_DEF(DRM_RADEON_SETPARAM, radeon_cp_setparam, DRM_AUTH),
3186 DRM_IOCTL_DEF(DRM_RADEON_SURF_ALLOC, radeon_surface_alloc, DRM_AUTH),
3187 DRM_IOCTL_DEF(DRM_RADEON_SURF_FREE, radeon_surface_free, DRM_AUTH)
3190 int radeon_max_ioctl = DRM_ARRAY_SIZE(radeon_ioctls);