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("%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("%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("pfCurrentPage=%d\n",
1426 dev_priv->sarea_priv->pfCurrentPage);
1428 /* Do some trivial performance monitoring...
1430 if (dev_priv->do_boxes) {
1431 dev_priv->stats.boxes |= RADEON_BOX_FLIP;
1432 radeon_cp_performance_boxes(dev_priv);
1435 /* Update the frame offsets for both CRTCs
1439 RADEON_WAIT_UNTIL_3D_IDLE();
1440 OUT_RING_REG(RADEON_CRTC_OFFSET,
1441 ((sarea->frame.y * dev_priv->front_pitch +
1442 sarea->frame.x * (dev_priv->color_fmt - 2)) & ~7)
1444 OUT_RING_REG(RADEON_CRTC2_OFFSET, dev_priv->sarea_priv->crtc2_base
1449 /* Increment the frame counter. The client-side 3D driver must
1450 * throttle the framerate by waiting for this value before
1451 * performing the swapbuffer ioctl.
1453 dev_priv->sarea_priv->last_frame++;
1454 dev_priv->sarea_priv->pfCurrentPage =
1455 1 - dev_priv->sarea_priv->pfCurrentPage;
1459 RADEON_FRAME_AGE(dev_priv->sarea_priv->last_frame);
1464 static int bad_prim_vertex_nr(int primitive, int nr)
1466 switch (primitive & RADEON_PRIM_TYPE_MASK) {
1467 case RADEON_PRIM_TYPE_NONE:
1468 case RADEON_PRIM_TYPE_POINT:
1470 case RADEON_PRIM_TYPE_LINE:
1471 return (nr & 1) || nr == 0;
1472 case RADEON_PRIM_TYPE_LINE_STRIP:
1474 case RADEON_PRIM_TYPE_TRI_LIST:
1475 case RADEON_PRIM_TYPE_3VRT_POINT_LIST:
1476 case RADEON_PRIM_TYPE_3VRT_LINE_LIST:
1477 case RADEON_PRIM_TYPE_RECT_LIST:
1478 return nr % 3 || nr == 0;
1479 case RADEON_PRIM_TYPE_TRI_FAN:
1480 case RADEON_PRIM_TYPE_TRI_STRIP:
1489 unsigned int finish;
1491 unsigned int numverts;
1492 unsigned int offset;
1493 unsigned int vc_format;
1494 } drm_radeon_tcl_prim_t;
1496 static void radeon_cp_dispatch_vertex(struct drm_device * dev,
1497 struct drm_buf * buf,
1498 drm_radeon_tcl_prim_t * prim)
1500 drm_radeon_private_t *dev_priv = dev->dev_private;
1501 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1502 int offset = dev_priv->gart_buffers_offset + buf->offset + prim->start;
1503 int numverts = (int)prim->numverts;
1504 int nbox = sarea_priv->nbox;
1508 DRM_DEBUG("hwprim 0x%x vfmt 0x%x %d..%d %d verts\n",
1510 prim->vc_format, prim->start, prim->finish, prim->numverts);
1512 if (bad_prim_vertex_nr(prim->prim, prim->numverts)) {
1513 DRM_ERROR("bad prim %x numverts %d\n",
1514 prim->prim, prim->numverts);
1519 /* Emit the next cliprect */
1521 radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1524 /* Emit the vertex buffer rendering commands */
1527 OUT_RING(CP_PACKET3(RADEON_3D_RNDR_GEN_INDX_PRIM, 3));
1530 OUT_RING(prim->vc_format);
1531 OUT_RING(prim->prim | RADEON_PRIM_WALK_LIST |
1532 RADEON_COLOR_ORDER_RGBA |
1533 RADEON_VTX_FMT_RADEON_MODE |
1534 (numverts << RADEON_NUM_VERTICES_SHIFT));
1542 static void radeon_cp_discard_buffer(struct drm_device * dev, struct drm_buf * buf)
1544 drm_radeon_private_t *dev_priv = dev->dev_private;
1545 drm_radeon_buf_priv_t *buf_priv = buf->dev_private;
1548 buf_priv->age = ++dev_priv->sarea_priv->last_dispatch;
1550 /* Emit the vertex buffer age */
1552 RADEON_DISPATCH_AGE(buf_priv->age);
1559 static void radeon_cp_dispatch_indirect(struct drm_device * dev,
1560 struct drm_buf * buf, int start, int end)
1562 drm_radeon_private_t *dev_priv = dev->dev_private;
1564 DRM_DEBUG("buf=%d s=0x%x e=0x%x\n", buf->idx, start, end);
1567 int offset = (dev_priv->gart_buffers_offset
1568 + buf->offset + start);
1569 int dwords = (end - start + 3) / sizeof(u32);
1571 /* Indirect buffer data must be an even number of
1572 * dwords, so if we've been given an odd number we must
1573 * pad the data with a Type-2 CP packet.
1577 ((char *)dev->agp_buffer_map->handle
1578 + buf->offset + start);
1579 data[dwords++] = RADEON_CP_PACKET2;
1582 /* Fire off the indirect buffer */
1585 OUT_RING(CP_PACKET0(RADEON_CP_IB_BASE, 1));
1593 static void radeon_cp_dispatch_indices(struct drm_device * dev,
1594 struct drm_buf * elt_buf,
1595 drm_radeon_tcl_prim_t * prim)
1597 drm_radeon_private_t *dev_priv = dev->dev_private;
1598 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
1599 int offset = dev_priv->gart_buffers_offset + prim->offset;
1603 int start = prim->start + RADEON_INDEX_PRIM_OFFSET;
1604 int count = (prim->finish - start) / sizeof(u16);
1605 int nbox = sarea_priv->nbox;
1607 DRM_DEBUG("hwprim 0x%x vfmt 0x%x %d..%d offset: %x nr %d\n",
1610 prim->start, prim->finish, prim->offset, prim->numverts);
1612 if (bad_prim_vertex_nr(prim->prim, count)) {
1613 DRM_ERROR("bad prim %x count %d\n", prim->prim, count);
1617 if (start >= prim->finish || (prim->start & 0x7)) {
1618 DRM_ERROR("buffer prim %d\n", prim->prim);
1622 dwords = (prim->finish - prim->start + 3) / sizeof(u32);
1624 data = (u32 *) ((char *)dev->agp_buffer_map->handle +
1625 elt_buf->offset + prim->start);
1627 data[0] = CP_PACKET3(RADEON_3D_RNDR_GEN_INDX_PRIM, dwords - 2);
1629 data[2] = prim->numverts;
1630 data[3] = prim->vc_format;
1631 data[4] = (prim->prim |
1632 RADEON_PRIM_WALK_IND |
1633 RADEON_COLOR_ORDER_RGBA |
1634 RADEON_VTX_FMT_RADEON_MODE |
1635 (count << RADEON_NUM_VERTICES_SHIFT));
1639 radeon_emit_clip_rect(dev_priv, &sarea_priv->boxes[i]);
1641 radeon_cp_dispatch_indirect(dev, elt_buf,
1642 prim->start, prim->finish);
1649 #define RADEON_MAX_TEXTURE_SIZE RADEON_BUFFER_SIZE
1651 static int radeon_cp_dispatch_texture(struct drm_device * dev,
1652 struct drm_file *file_priv,
1653 drm_radeon_texture_t * tex,
1654 drm_radeon_tex_image_t * image)
1656 drm_radeon_private_t *dev_priv = dev->dev_private;
1657 struct drm_buf *buf;
1660 const u8 __user *data;
1661 int size, dwords, tex_width, blit_width, spitch;
1664 u32 texpitch, microtile;
1665 u32 offset, byte_offset;
1668 if (radeon_check_and_fixup_offset(dev_priv, file_priv, &tex->offset)) {
1669 DRM_ERROR("Invalid destination offset\n");
1673 dev_priv->stats.boxes |= RADEON_BOX_TEXTURE_LOAD;
1675 /* Flush the pixel cache. This ensures no pixel data gets mixed
1676 * up with the texture data from the host data blit, otherwise
1677 * part of the texture image may be corrupted.
1680 RADEON_FLUSH_CACHE();
1681 RADEON_WAIT_UNTIL_IDLE();
1684 /* The compiler won't optimize away a division by a variable,
1685 * even if the only legal values are powers of two. Thus, we'll
1686 * use a shift instead.
1688 switch (tex->format) {
1689 case RADEON_TXFORMAT_ARGB8888:
1690 case RADEON_TXFORMAT_RGBA8888:
1691 format = RADEON_COLOR_FORMAT_ARGB8888;
1692 tex_width = tex->width * 4;
1693 blit_width = image->width * 4;
1695 case RADEON_TXFORMAT_AI88:
1696 case RADEON_TXFORMAT_ARGB1555:
1697 case RADEON_TXFORMAT_RGB565:
1698 case RADEON_TXFORMAT_ARGB4444:
1699 case RADEON_TXFORMAT_VYUY422:
1700 case RADEON_TXFORMAT_YVYU422:
1701 format = RADEON_COLOR_FORMAT_RGB565;
1702 tex_width = tex->width * 2;
1703 blit_width = image->width * 2;
1705 case RADEON_TXFORMAT_I8:
1706 case RADEON_TXFORMAT_RGB332:
1707 format = RADEON_COLOR_FORMAT_CI8;
1708 tex_width = tex->width * 1;
1709 blit_width = image->width * 1;
1712 DRM_ERROR("invalid texture format %d\n", tex->format);
1715 spitch = blit_width >> 6;
1716 if (spitch == 0 && image->height > 1)
1719 texpitch = tex->pitch;
1720 if ((texpitch << 22) & RADEON_DST_TILE_MICRO) {
1722 if (tex_width < 64) {
1723 texpitch &= ~(RADEON_DST_TILE_MICRO >> 22);
1724 /* we got tiled coordinates, untile them */
1730 /* this might fail for zero-sized uploads - are those illegal? */
1731 if (!radeon_check_offset(dev_priv, tex->offset + image->height *
1733 DRM_ERROR("Invalid final destination offset\n");
1737 DRM_DEBUG("tex=%dx%d blit=%d\n", tex_width, tex->height, blit_width);
1740 DRM_DEBUG("tex: ofs=0x%x p=%d f=%d x=%hd y=%hd w=%hd h=%hd\n",
1741 tex->offset >> 10, tex->pitch, tex->format,
1742 image->x, image->y, image->width, image->height);
1744 /* Make a copy of some parameters in case we have to
1745 * update them for a multi-pass texture blit.
1747 height = image->height;
1748 data = (const u8 __user *)image->data;
1750 size = height * blit_width;
1752 if (size > RADEON_MAX_TEXTURE_SIZE) {
1753 height = RADEON_MAX_TEXTURE_SIZE / blit_width;
1754 size = height * blit_width;
1755 } else if (size < 4 && size > 0) {
1757 } else if (size == 0) {
1761 buf = radeon_freelist_get(dev);
1763 radeon_do_cp_idle(dev_priv);
1764 buf = radeon_freelist_get(dev);
1767 DRM_DEBUG("EAGAIN\n");
1768 if (DRM_COPY_TO_USER(tex->image, image, sizeof(*image)))
1773 /* Dispatch the indirect buffer.
1776 (u32 *) ((char *)dev->agp_buffer_map->handle + buf->offset);
1779 #define RADEON_COPY_MT(_buf, _data, _width) \
1781 if (DRM_COPY_FROM_USER(_buf, _data, (_width))) {\
1782 DRM_ERROR("EFAULT on pad, %d bytes\n", (_width)); \
1788 /* texture micro tiling in use, minimum texture width is thus 16 bytes.
1789 however, we cannot use blitter directly for texture width < 64 bytes,
1790 since minimum tex pitch is 64 bytes and we need this to match
1791 the texture width, otherwise the blitter will tile it wrong.
1792 Thus, tiling manually in this case. Additionally, need to special
1793 case tex height = 1, since our actual image will have height 2
1794 and we need to ensure we don't read beyond the texture size
1796 if (tex->height == 1) {
1797 if (tex_width >= 64 || tex_width <= 16) {
1798 RADEON_COPY_MT(buffer, data,
1799 (int)(tex_width * sizeof(u32)));
1800 } else if (tex_width == 32) {
1801 RADEON_COPY_MT(buffer, data, 16);
1802 RADEON_COPY_MT(buffer + 8,
1805 } else if (tex_width >= 64 || tex_width == 16) {
1806 RADEON_COPY_MT(buffer, data,
1807 (int)(dwords * sizeof(u32)));
1808 } else if (tex_width < 16) {
1809 for (i = 0; i < tex->height; i++) {
1810 RADEON_COPY_MT(buffer, data, tex_width);
1814 } else if (tex_width == 32) {
1815 /* TODO: make sure this works when not fitting in one buffer
1816 (i.e. 32bytes x 2048...) */
1817 for (i = 0; i < tex->height; i += 2) {
1818 RADEON_COPY_MT(buffer, data, 16);
1820 RADEON_COPY_MT(buffer + 8, data, 16);
1822 RADEON_COPY_MT(buffer + 4, data, 16);
1824 RADEON_COPY_MT(buffer + 12, data, 16);
1830 if (tex_width >= 32) {
1831 /* Texture image width is larger than the minimum, so we
1832 * can upload it directly.
1834 RADEON_COPY_MT(buffer, data,
1835 (int)(dwords * sizeof(u32)));
1837 /* Texture image width is less than the minimum, so we
1838 * need to pad out each image scanline to the minimum
1841 for (i = 0; i < tex->height; i++) {
1842 RADEON_COPY_MT(buffer, data, tex_width);
1849 #undef RADEON_COPY_MT
1850 byte_offset = (image->y & ~2047) * blit_width;
1851 buf->file_priv = file_priv;
1853 offset = dev_priv->gart_buffers_offset + buf->offset;
1855 OUT_RING(CP_PACKET3(RADEON_CNTL_BITBLT_MULTI, 5));
1856 OUT_RING(RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
1857 RADEON_GMC_DST_PITCH_OFFSET_CNTL |
1858 RADEON_GMC_BRUSH_NONE |
1860 RADEON_GMC_SRC_DATATYPE_COLOR |
1862 RADEON_DP_SRC_SOURCE_MEMORY |
1863 RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
1864 OUT_RING((spitch << 22) | (offset >> 10));
1865 OUT_RING((texpitch << 22) | ((tex->offset >> 10) + (byte_offset >> 10)));
1867 OUT_RING((image->x << 16) | (image->y % 2048));
1868 OUT_RING((image->width << 16) | height);
1869 RADEON_WAIT_UNTIL_2D_IDLE();
1873 radeon_cp_discard_buffer(dev, buf);
1875 /* Update the input parameters for next time */
1877 image->height -= height;
1878 image->data = (const u8 __user *)image->data + size;
1879 } while (image->height > 0);
1881 /* Flush the pixel cache after the blit completes. This ensures
1882 * the texture data is written out to memory before rendering
1886 RADEON_FLUSH_CACHE();
1887 RADEON_WAIT_UNTIL_2D_IDLE();
1894 static void radeon_cp_dispatch_stipple(struct drm_device * dev, u32 * stipple)
1896 drm_radeon_private_t *dev_priv = dev->dev_private;
1903 OUT_RING(CP_PACKET0(RADEON_RE_STIPPLE_ADDR, 0));
1904 OUT_RING(0x00000000);
1906 OUT_RING(CP_PACKET0_TABLE(RADEON_RE_STIPPLE_DATA, 31));
1907 for (i = 0; i < 32; i++) {
1908 OUT_RING(stipple[i]);
1914 static void radeon_apply_surface_regs(int surf_index,
1915 drm_radeon_private_t *dev_priv)
1917 if (!dev_priv->mmio)
1920 radeon_do_cp_idle(dev_priv);
1922 RADEON_WRITE(RADEON_SURFACE0_INFO + 16 * surf_index,
1923 dev_priv->surfaces[surf_index].flags);
1924 RADEON_WRITE(RADEON_SURFACE0_LOWER_BOUND + 16 * surf_index,
1925 dev_priv->surfaces[surf_index].lower);
1926 RADEON_WRITE(RADEON_SURFACE0_UPPER_BOUND + 16 * surf_index,
1927 dev_priv->surfaces[surf_index].upper);
1930 /* Allocates a virtual surface
1931 * doesn't always allocate a real surface, will stretch an existing
1932 * surface when possible.
1934 * Note that refcount can be at most 2, since during a free refcount=3
1935 * might mean we have to allocate a new surface which might not always
1937 * For example : we allocate three contigous surfaces ABC. If B is
1938 * freed, we suddenly need two surfaces to store A and C, which might
1939 * not always be available.
1941 static int alloc_surface(drm_radeon_surface_alloc_t *new,
1942 drm_radeon_private_t *dev_priv,
1943 struct drm_file *file_priv)
1945 struct radeon_virt_surface *s;
1947 int virt_surface_index;
1948 uint32_t new_upper, new_lower;
1950 new_lower = new->address;
1951 new_upper = new_lower + new->size - 1;
1954 if ((new_lower >= new_upper) || (new->flags == 0) || (new->size == 0) ||
1955 ((new_upper & RADEON_SURF_ADDRESS_FIXED_MASK) !=
1956 RADEON_SURF_ADDRESS_FIXED_MASK)
1957 || ((new_lower & RADEON_SURF_ADDRESS_FIXED_MASK) != 0))
1960 /* make sure there is no overlap with existing surfaces */
1961 for (i = 0; i < RADEON_MAX_SURFACES; i++) {
1962 if ((dev_priv->surfaces[i].refcount != 0) &&
1963 (((new_lower >= dev_priv->surfaces[i].lower) &&
1964 (new_lower < dev_priv->surfaces[i].upper)) ||
1965 ((new_lower < dev_priv->surfaces[i].lower) &&
1966 (new_upper > dev_priv->surfaces[i].lower)))) {
1971 /* find a virtual surface */
1972 for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++)
1973 if (dev_priv->virt_surfaces[i].file_priv == 0)
1975 if (i == 2 * RADEON_MAX_SURFACES) {
1978 virt_surface_index = i;
1980 /* try to reuse an existing surface */
1981 for (i = 0; i < RADEON_MAX_SURFACES; i++) {
1983 if ((dev_priv->surfaces[i].refcount == 1) &&
1984 (new->flags == dev_priv->surfaces[i].flags) &&
1985 (new_upper + 1 == dev_priv->surfaces[i].lower)) {
1986 s = &(dev_priv->virt_surfaces[virt_surface_index]);
1987 s->surface_index = i;
1988 s->lower = new_lower;
1989 s->upper = new_upper;
1990 s->flags = new->flags;
1991 s->file_priv = file_priv;
1992 dev_priv->surfaces[i].refcount++;
1993 dev_priv->surfaces[i].lower = s->lower;
1994 radeon_apply_surface_regs(s->surface_index, dev_priv);
1995 return virt_surface_index;
1999 if ((dev_priv->surfaces[i].refcount == 1) &&
2000 (new->flags == dev_priv->surfaces[i].flags) &&
2001 (new_lower == dev_priv->surfaces[i].upper + 1)) {
2002 s = &(dev_priv->virt_surfaces[virt_surface_index]);
2003 s->surface_index = i;
2004 s->lower = new_lower;
2005 s->upper = new_upper;
2006 s->flags = new->flags;
2007 s->file_priv = file_priv;
2008 dev_priv->surfaces[i].refcount++;
2009 dev_priv->surfaces[i].upper = s->upper;
2010 radeon_apply_surface_regs(s->surface_index, dev_priv);
2011 return virt_surface_index;
2015 /* okay, we need a new one */
2016 for (i = 0; i < RADEON_MAX_SURFACES; i++) {
2017 if (dev_priv->surfaces[i].refcount == 0) {
2018 s = &(dev_priv->virt_surfaces[virt_surface_index]);
2019 s->surface_index = i;
2020 s->lower = new_lower;
2021 s->upper = new_upper;
2022 s->flags = new->flags;
2023 s->file_priv = file_priv;
2024 dev_priv->surfaces[i].refcount = 1;
2025 dev_priv->surfaces[i].lower = s->lower;
2026 dev_priv->surfaces[i].upper = s->upper;
2027 dev_priv->surfaces[i].flags = s->flags;
2028 radeon_apply_surface_regs(s->surface_index, dev_priv);
2029 return virt_surface_index;
2033 /* we didn't find anything */
2037 static int free_surface(struct drm_file *file_priv,
2038 drm_radeon_private_t * dev_priv,
2041 struct radeon_virt_surface *s;
2043 /* find the virtual surface */
2044 for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++) {
2045 s = &(dev_priv->virt_surfaces[i]);
2047 if ((lower == s->lower) && (file_priv == s->file_priv))
2049 if (dev_priv->surfaces[s->surface_index].
2051 dev_priv->surfaces[s->surface_index].
2054 if (dev_priv->surfaces[s->surface_index].
2056 dev_priv->surfaces[s->surface_index].
2059 dev_priv->surfaces[s->surface_index].refcount--;
2060 if (dev_priv->surfaces[s->surface_index].
2062 dev_priv->surfaces[s->surface_index].
2064 s->file_priv = NULL;
2065 radeon_apply_surface_regs(s->surface_index,
2074 static void radeon_surfaces_release(struct drm_file *file_priv,
2075 drm_radeon_private_t * dev_priv)
2078 for (i = 0; i < 2 * RADEON_MAX_SURFACES; i++) {
2079 if (dev_priv->virt_surfaces[i].file_priv == file_priv)
2080 free_surface(file_priv, dev_priv,
2081 dev_priv->virt_surfaces[i].lower);
2085 /* ================================================================
2088 static int radeon_surface_alloc(struct drm_device *dev, void *data, struct drm_file *file_priv)
2090 drm_radeon_private_t *dev_priv = dev->dev_private;
2091 drm_radeon_surface_alloc_t *alloc = data;
2093 if (alloc_surface(alloc, dev_priv, file_priv) == -1)
2099 static int radeon_surface_free(struct drm_device *dev, void *data, struct drm_file *file_priv)
2101 drm_radeon_private_t *dev_priv = dev->dev_private;
2102 drm_radeon_surface_free_t *memfree = data;
2104 if (free_surface(file_priv, dev_priv, memfree->address))
2110 static int radeon_cp_clear(struct drm_device *dev, void *data, struct drm_file *file_priv)
2112 drm_radeon_private_t *dev_priv = dev->dev_private;
2113 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2114 drm_radeon_clear_t *clear = data;
2115 drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
2118 LOCK_TEST_WITH_RETURN(dev, file_priv);
2120 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2122 if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2123 sarea_priv->nbox = RADEON_NR_SAREA_CLIPRECTS;
2125 if (DRM_COPY_FROM_USER(&depth_boxes, clear->depth_boxes,
2126 sarea_priv->nbox * sizeof(depth_boxes[0])))
2129 radeon_cp_dispatch_clear(dev, clear, depth_boxes);
2135 /* Not sure why this isn't set all the time:
2137 static int radeon_do_init_pageflip(struct drm_device * dev)
2139 drm_radeon_private_t *dev_priv = dev->dev_private;
2145 RADEON_WAIT_UNTIL_3D_IDLE();
2146 OUT_RING(CP_PACKET0(RADEON_CRTC_OFFSET_CNTL, 0));
2147 OUT_RING(RADEON_READ(RADEON_CRTC_OFFSET_CNTL) |
2148 RADEON_CRTC_OFFSET_FLIP_CNTL);
2149 OUT_RING(CP_PACKET0(RADEON_CRTC2_OFFSET_CNTL, 0));
2150 OUT_RING(RADEON_READ(RADEON_CRTC2_OFFSET_CNTL) |
2151 RADEON_CRTC_OFFSET_FLIP_CNTL);
2154 dev_priv->page_flipping = 1;
2156 if (dev_priv->sarea_priv->pfCurrentPage != 1)
2157 dev_priv->sarea_priv->pfCurrentPage = 0;
2162 /* Swapping and flipping are different operations, need different ioctls.
2163 * They can & should be intermixed to support multiple 3d windows.
2165 static int radeon_cp_flip(struct drm_device *dev, void *data, struct drm_file *file_priv)
2167 drm_radeon_private_t *dev_priv = dev->dev_private;
2170 LOCK_TEST_WITH_RETURN(dev, file_priv);
2172 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2174 if (!dev_priv->page_flipping)
2175 radeon_do_init_pageflip(dev);
2177 radeon_cp_dispatch_flip(dev);
2183 static int radeon_cp_swap(struct drm_device *dev, void *data, struct drm_file *file_priv)
2185 drm_radeon_private_t *dev_priv = dev->dev_private;
2186 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2189 LOCK_TEST_WITH_RETURN(dev, file_priv);
2191 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2193 if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2194 sarea_priv->nbox = RADEON_NR_SAREA_CLIPRECTS;
2196 radeon_cp_dispatch_swap(dev);
2197 dev_priv->sarea_priv->ctx_owner = 0;
2203 static int radeon_cp_vertex(struct drm_device *dev, void *data, struct drm_file *file_priv)
2205 drm_radeon_private_t *dev_priv = dev->dev_private;
2206 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2207 struct drm_device_dma *dma = dev->dma;
2208 struct drm_buf *buf;
2209 drm_radeon_vertex_t *vertex = data;
2210 drm_radeon_tcl_prim_t prim;
2212 LOCK_TEST_WITH_RETURN(dev, file_priv);
2214 DRM_DEBUG("pid=%d index=%d count=%d discard=%d\n",
2215 DRM_CURRENTPID, vertex->idx, vertex->count, vertex->discard);
2217 if (vertex->idx < 0 || vertex->idx >= dma->buf_count) {
2218 DRM_ERROR("buffer index %d (of %d max)\n",
2219 vertex->idx, dma->buf_count - 1);
2222 if (vertex->prim < 0 || vertex->prim > RADEON_PRIM_TYPE_3VRT_LINE_LIST) {
2223 DRM_ERROR("buffer prim %d\n", vertex->prim);
2227 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2228 VB_AGE_TEST_WITH_RETURN(dev_priv);
2230 buf = dma->buflist[vertex->idx];
2232 if (buf->file_priv != file_priv) {
2233 DRM_ERROR("process %d using buffer owned by %p\n",
2234 DRM_CURRENTPID, buf->file_priv);
2238 DRM_ERROR("sending pending buffer %d\n", vertex->idx);
2242 /* Build up a prim_t record:
2244 if (vertex->count) {
2245 buf->used = vertex->count; /* not used? */
2247 if (sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS) {
2248 if (radeon_emit_state(dev_priv, file_priv,
2249 &sarea_priv->context_state,
2250 sarea_priv->tex_state,
2251 sarea_priv->dirty)) {
2252 DRM_ERROR("radeon_emit_state failed\n");
2256 sarea_priv->dirty &= ~(RADEON_UPLOAD_TEX0IMAGES |
2257 RADEON_UPLOAD_TEX1IMAGES |
2258 RADEON_UPLOAD_TEX2IMAGES |
2259 RADEON_REQUIRE_QUIESCENCE);
2263 prim.finish = vertex->count; /* unused */
2264 prim.prim = vertex->prim;
2265 prim.numverts = vertex->count;
2266 prim.vc_format = dev_priv->sarea_priv->vc_format;
2268 radeon_cp_dispatch_vertex(dev, buf, &prim);
2271 if (vertex->discard) {
2272 radeon_cp_discard_buffer(dev, buf);
2279 static int radeon_cp_indices(struct drm_device *dev, void *data, struct drm_file *file_priv)
2281 drm_radeon_private_t *dev_priv = dev->dev_private;
2282 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2283 struct drm_device_dma *dma = dev->dma;
2284 struct drm_buf *buf;
2285 drm_radeon_indices_t *elts = data;
2286 drm_radeon_tcl_prim_t prim;
2289 LOCK_TEST_WITH_RETURN(dev, file_priv);
2291 DRM_DEBUG("pid=%d index=%d start=%d end=%d discard=%d\n",
2292 DRM_CURRENTPID, elts->idx, elts->start, elts->end,
2295 if (elts->idx < 0 || elts->idx >= dma->buf_count) {
2296 DRM_ERROR("buffer index %d (of %d max)\n",
2297 elts->idx, dma->buf_count - 1);
2300 if (elts->prim < 0 || elts->prim > RADEON_PRIM_TYPE_3VRT_LINE_LIST) {
2301 DRM_ERROR("buffer prim %d\n", elts->prim);
2305 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2306 VB_AGE_TEST_WITH_RETURN(dev_priv);
2308 buf = dma->buflist[elts->idx];
2310 if (buf->file_priv != file_priv) {
2311 DRM_ERROR("process %d using buffer owned by %p\n",
2312 DRM_CURRENTPID, buf->file_priv);
2316 DRM_ERROR("sending pending buffer %d\n", elts->idx);
2320 count = (elts->end - elts->start) / sizeof(u16);
2321 elts->start -= RADEON_INDEX_PRIM_OFFSET;
2323 if (elts->start & 0x7) {
2324 DRM_ERROR("misaligned buffer 0x%x\n", elts->start);
2327 if (elts->start < buf->used) {
2328 DRM_ERROR("no header 0x%x - 0x%x\n", elts->start, buf->used);
2332 buf->used = elts->end;
2334 if (sarea_priv->dirty & ~RADEON_UPLOAD_CLIPRECTS) {
2335 if (radeon_emit_state(dev_priv, file_priv,
2336 &sarea_priv->context_state,
2337 sarea_priv->tex_state,
2338 sarea_priv->dirty)) {
2339 DRM_ERROR("radeon_emit_state failed\n");
2343 sarea_priv->dirty &= ~(RADEON_UPLOAD_TEX0IMAGES |
2344 RADEON_UPLOAD_TEX1IMAGES |
2345 RADEON_UPLOAD_TEX2IMAGES |
2346 RADEON_REQUIRE_QUIESCENCE);
2349 /* Build up a prim_t record:
2351 prim.start = elts->start;
2352 prim.finish = elts->end;
2353 prim.prim = elts->prim;
2354 prim.offset = 0; /* offset from start of dma buffers */
2355 prim.numverts = RADEON_MAX_VB_VERTS; /* duh */
2356 prim.vc_format = dev_priv->sarea_priv->vc_format;
2358 radeon_cp_dispatch_indices(dev, buf, &prim);
2359 if (elts->discard) {
2360 radeon_cp_discard_buffer(dev, buf);
2367 static int radeon_cp_texture(struct drm_device *dev, void *data, struct drm_file *file_priv)
2369 drm_radeon_private_t *dev_priv = dev->dev_private;
2370 drm_radeon_texture_t *tex = data;
2371 drm_radeon_tex_image_t image;
2374 LOCK_TEST_WITH_RETURN(dev, file_priv);
2376 if (tex->image == NULL) {
2377 DRM_ERROR("null texture image!\n");
2381 if (DRM_COPY_FROM_USER(&image,
2382 (drm_radeon_tex_image_t __user *) tex->image,
2386 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2387 VB_AGE_TEST_WITH_RETURN(dev_priv);
2389 ret = radeon_cp_dispatch_texture(dev, file_priv, tex, &image);
2394 static int radeon_cp_stipple(struct drm_device *dev, void *data, struct drm_file *file_priv)
2396 drm_radeon_private_t *dev_priv = dev->dev_private;
2397 drm_radeon_stipple_t *stipple = data;
2400 LOCK_TEST_WITH_RETURN(dev, file_priv);
2402 if (DRM_COPY_FROM_USER(&mask, stipple->mask, 32 * sizeof(u32)))
2405 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2407 radeon_cp_dispatch_stipple(dev, mask);
2413 static int radeon_cp_indirect(struct drm_device *dev, void *data, struct drm_file *file_priv)
2415 drm_radeon_private_t *dev_priv = dev->dev_private;
2416 struct drm_device_dma *dma = dev->dma;
2417 struct drm_buf *buf;
2418 drm_radeon_indirect_t *indirect = data;
2421 LOCK_TEST_WITH_RETURN(dev, file_priv);
2423 DRM_DEBUG("idx=%d s=%d e=%d d=%d\n",
2424 indirect->idx, indirect->start, indirect->end,
2427 if (indirect->idx < 0 || indirect->idx >= dma->buf_count) {
2428 DRM_ERROR("buffer index %d (of %d max)\n",
2429 indirect->idx, dma->buf_count - 1);
2433 buf = dma->buflist[indirect->idx];
2435 if (buf->file_priv != file_priv) {
2436 DRM_ERROR("process %d using buffer owned by %p\n",
2437 DRM_CURRENTPID, buf->file_priv);
2441 DRM_ERROR("sending pending buffer %d\n", indirect->idx);
2445 if (indirect->start < buf->used) {
2446 DRM_ERROR("reusing indirect: start=0x%x actual=0x%x\n",
2447 indirect->start, buf->used);
2451 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2452 VB_AGE_TEST_WITH_RETURN(dev_priv);
2454 buf->used = indirect->end;
2456 /* Wait for the 3D stream to idle before the indirect buffer
2457 * containing 2D acceleration commands is processed.
2461 RADEON_WAIT_UNTIL_3D_IDLE();
2465 /* Dispatch the indirect buffer full of commands from the
2466 * X server. This is insecure and is thus only available to
2467 * privileged clients.
2469 radeon_cp_dispatch_indirect(dev, buf, indirect->start, indirect->end);
2470 if (indirect->discard) {
2471 radeon_cp_discard_buffer(dev, buf);
2478 static int radeon_cp_vertex2(struct drm_device *dev, void *data, struct drm_file *file_priv)
2480 drm_radeon_private_t *dev_priv = dev->dev_private;
2481 drm_radeon_sarea_t *sarea_priv = dev_priv->sarea_priv;
2482 struct drm_device_dma *dma = dev->dma;
2483 struct drm_buf *buf;
2484 drm_radeon_vertex2_t *vertex = data;
2486 unsigned char laststate;
2488 LOCK_TEST_WITH_RETURN(dev, file_priv);
2490 DRM_DEBUG("pid=%d index=%d discard=%d\n",
2491 DRM_CURRENTPID, vertex->idx, vertex->discard);
2493 if (vertex->idx < 0 || vertex->idx >= dma->buf_count) {
2494 DRM_ERROR("buffer index %d (of %d max)\n",
2495 vertex->idx, dma->buf_count - 1);
2499 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2500 VB_AGE_TEST_WITH_RETURN(dev_priv);
2502 buf = dma->buflist[vertex->idx];
2504 if (buf->file_priv != file_priv) {
2505 DRM_ERROR("process %d using buffer owned by %p\n",
2506 DRM_CURRENTPID, buf->file_priv);
2511 DRM_ERROR("sending pending buffer %d\n", vertex->idx);
2515 if (sarea_priv->nbox > RADEON_NR_SAREA_CLIPRECTS)
2518 for (laststate = 0xff, i = 0; i < vertex->nr_prims; i++) {
2519 drm_radeon_prim_t prim;
2520 drm_radeon_tcl_prim_t tclprim;
2522 if (DRM_COPY_FROM_USER(&prim, &vertex->prim[i], sizeof(prim)))
2525 if (prim.stateidx != laststate) {
2526 drm_radeon_state_t state;
2528 if (DRM_COPY_FROM_USER(&state,
2529 &vertex->state[prim.stateidx],
2533 if (radeon_emit_state2(dev_priv, file_priv, &state)) {
2534 DRM_ERROR("radeon_emit_state2 failed\n");
2538 laststate = prim.stateidx;
2541 tclprim.start = prim.start;
2542 tclprim.finish = prim.finish;
2543 tclprim.prim = prim.prim;
2544 tclprim.vc_format = prim.vc_format;
2546 if (prim.prim & RADEON_PRIM_WALK_IND) {
2547 tclprim.offset = prim.numverts * 64;
2548 tclprim.numverts = RADEON_MAX_VB_VERTS; /* duh */
2550 radeon_cp_dispatch_indices(dev, buf, &tclprim);
2552 tclprim.numverts = prim.numverts;
2553 tclprim.offset = 0; /* not used */
2555 radeon_cp_dispatch_vertex(dev, buf, &tclprim);
2558 if (sarea_priv->nbox == 1)
2559 sarea_priv->nbox = 0;
2562 if (vertex->discard) {
2563 radeon_cp_discard_buffer(dev, buf);
2570 static int radeon_emit_packets(drm_radeon_private_t * dev_priv,
2571 struct drm_file *file_priv,
2572 drm_radeon_cmd_header_t header,
2573 drm_radeon_kcmd_buffer_t *cmdbuf)
2575 int id = (int)header.packet.packet_id;
2577 int *data = (int *)cmdbuf->buf;
2580 if (id >= RADEON_MAX_STATE_PACKETS)
2583 sz = packet[id].len;
2584 reg = packet[id].start;
2586 if (sz * sizeof(int) > cmdbuf->bufsz) {
2587 DRM_ERROR("Packet size provided larger than data provided\n");
2591 if (radeon_check_and_fixup_packets(dev_priv, file_priv, id, data)) {
2592 DRM_ERROR("Packet verification failed\n");
2597 OUT_RING(CP_PACKET0(reg, (sz - 1)));
2598 OUT_RING_TABLE(data, sz);
2601 cmdbuf->buf += sz * sizeof(int);
2602 cmdbuf->bufsz -= sz * sizeof(int);
2606 static __inline__ int radeon_emit_scalars(drm_radeon_private_t *dev_priv,
2607 drm_radeon_cmd_header_t header,
2608 drm_radeon_kcmd_buffer_t *cmdbuf)
2610 int sz = header.scalars.count;
2611 int start = header.scalars.offset;
2612 int stride = header.scalars.stride;
2616 OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
2617 OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
2618 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
2619 OUT_RING_TABLE(cmdbuf->buf, sz);
2621 cmdbuf->buf += sz * sizeof(int);
2622 cmdbuf->bufsz -= sz * sizeof(int);
2628 static __inline__ int radeon_emit_scalars2(drm_radeon_private_t *dev_priv,
2629 drm_radeon_cmd_header_t header,
2630 drm_radeon_kcmd_buffer_t *cmdbuf)
2632 int sz = header.scalars.count;
2633 int start = ((unsigned int)header.scalars.offset) + 0x100;
2634 int stride = header.scalars.stride;
2638 OUT_RING(CP_PACKET0(RADEON_SE_TCL_SCALAR_INDX_REG, 0));
2639 OUT_RING(start | (stride << RADEON_SCAL_INDX_DWORD_STRIDE_SHIFT));
2640 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_SCALAR_DATA_REG, sz - 1));
2641 OUT_RING_TABLE(cmdbuf->buf, sz);
2643 cmdbuf->buf += sz * sizeof(int);
2644 cmdbuf->bufsz -= sz * sizeof(int);
2648 static __inline__ int radeon_emit_vectors(drm_radeon_private_t *dev_priv,
2649 drm_radeon_cmd_header_t header,
2650 drm_radeon_kcmd_buffer_t *cmdbuf)
2652 int sz = header.vectors.count;
2653 int start = header.vectors.offset;
2654 int stride = header.vectors.stride;
2658 OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
2659 OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
2660 OUT_RING(start | (stride << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
2661 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
2662 OUT_RING_TABLE(cmdbuf->buf, sz);
2665 cmdbuf->buf += sz * sizeof(int);
2666 cmdbuf->bufsz -= sz * sizeof(int);
2670 static __inline__ int radeon_emit_veclinear(drm_radeon_private_t *dev_priv,
2671 drm_radeon_cmd_header_t header,
2672 drm_radeon_kcmd_buffer_t *cmdbuf)
2674 int sz = header.veclinear.count * 4;
2675 int start = header.veclinear.addr_lo | (header.veclinear.addr_hi << 8);
2680 if (sz * 4 > cmdbuf->bufsz)
2684 OUT_RING_REG(RADEON_SE_TCL_STATE_FLUSH, 0);
2685 OUT_RING(CP_PACKET0(RADEON_SE_TCL_VECTOR_INDX_REG, 0));
2686 OUT_RING(start | (1 << RADEON_VEC_INDX_OCTWORD_STRIDE_SHIFT));
2687 OUT_RING(CP_PACKET0_TABLE(RADEON_SE_TCL_VECTOR_DATA_REG, (sz - 1)));
2688 OUT_RING_TABLE(cmdbuf->buf, sz);
2691 cmdbuf->buf += sz * sizeof(int);
2692 cmdbuf->bufsz -= sz * sizeof(int);
2696 static int radeon_emit_packet3(struct drm_device * dev,
2697 struct drm_file *file_priv,
2698 drm_radeon_kcmd_buffer_t *cmdbuf)
2700 drm_radeon_private_t *dev_priv = dev->dev_private;
2707 if ((ret = radeon_check_and_fixup_packet3(dev_priv, file_priv,
2709 DRM_ERROR("Packet verification failed\n");
2714 OUT_RING_TABLE(cmdbuf->buf, cmdsz);
2717 cmdbuf->buf += cmdsz * 4;
2718 cmdbuf->bufsz -= cmdsz * 4;
2722 static int radeon_emit_packet3_cliprect(struct drm_device *dev,
2723 struct drm_file *file_priv,
2724 drm_radeon_kcmd_buffer_t *cmdbuf,
2727 drm_radeon_private_t *dev_priv = dev->dev_private;
2728 struct drm_clip_rect box;
2731 struct drm_clip_rect __user *boxes = cmdbuf->boxes;
2737 if ((ret = radeon_check_and_fixup_packet3(dev_priv, file_priv,
2739 DRM_ERROR("Packet verification failed\n");
2747 if (i < cmdbuf->nbox) {
2748 if (DRM_COPY_FROM_USER(&box, &boxes[i], sizeof(box)))
2750 /* FIXME The second and subsequent times round
2751 * this loop, send a WAIT_UNTIL_3D_IDLE before
2752 * calling emit_clip_rect(). This fixes a
2753 * lockup on fast machines when sending
2754 * several cliprects with a cmdbuf, as when
2755 * waving a 2D window over a 3D
2756 * window. Something in the commands from user
2757 * space seems to hang the card when they're
2758 * sent several times in a row. That would be
2759 * the correct place to fix it but this works
2760 * around it until I can figure that out - Tim
2764 RADEON_WAIT_UNTIL_3D_IDLE();
2767 radeon_emit_clip_rect(dev_priv, &box);
2771 OUT_RING_TABLE(cmdbuf->buf, cmdsz);
2774 } while (++i < cmdbuf->nbox);
2775 if (cmdbuf->nbox == 1)
2779 cmdbuf->buf += cmdsz * 4;
2780 cmdbuf->bufsz -= cmdsz * 4;
2784 static int radeon_emit_wait(struct drm_device * dev, int flags)
2786 drm_radeon_private_t *dev_priv = dev->dev_private;
2789 DRM_DEBUG("%x\n", flags);
2791 case RADEON_WAIT_2D:
2793 RADEON_WAIT_UNTIL_2D_IDLE();
2796 case RADEON_WAIT_3D:
2798 RADEON_WAIT_UNTIL_3D_IDLE();
2801 case RADEON_WAIT_2D | RADEON_WAIT_3D:
2803 RADEON_WAIT_UNTIL_IDLE();
2813 static int radeon_cp_cmdbuf(struct drm_device *dev, void *data, struct drm_file *file_priv)
2815 drm_radeon_private_t *dev_priv = dev->dev_private;
2816 struct drm_device_dma *dma = dev->dma;
2817 struct drm_buf *buf = NULL;
2819 drm_radeon_kcmd_buffer_t *cmdbuf = data;
2820 drm_radeon_cmd_header_t header;
2821 int orig_nbox, orig_bufsz;
2824 LOCK_TEST_WITH_RETURN(dev, file_priv);
2826 RING_SPACE_TEST_WITH_RETURN(dev_priv);
2827 VB_AGE_TEST_WITH_RETURN(dev_priv);
2829 if (cmdbuf->bufsz > 64 * 1024 || cmdbuf->bufsz < 0) {
2833 /* Allocate an in-kernel area and copy in the cmdbuf. Do this to avoid
2834 * races between checking values and using those values in other code,
2835 * and simply to avoid a lot of function calls to copy in data.
2837 orig_bufsz = cmdbuf->bufsz;
2838 if (orig_bufsz != 0) {
2839 kbuf = drm_alloc(cmdbuf->bufsz, DRM_MEM_DRIVER);
2842 if (DRM_COPY_FROM_USER(kbuf, (void __user *)cmdbuf->buf,
2844 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2850 orig_nbox = cmdbuf->nbox;
2852 if (dev_priv->microcode_version == UCODE_R300) {
2854 temp = r300_do_cp_cmdbuf(dev, file_priv, cmdbuf);
2856 if (orig_bufsz != 0)
2857 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2862 /* microcode_version != r300 */
2863 while (cmdbuf->bufsz >= sizeof(header)) {
2865 header.i = *(int *)cmdbuf->buf;
2866 cmdbuf->buf += sizeof(header);
2867 cmdbuf->bufsz -= sizeof(header);
2869 switch (header.header.cmd_type) {
2870 case RADEON_CMD_PACKET:
2871 DRM_DEBUG("RADEON_CMD_PACKET\n");
2872 if (radeon_emit_packets
2873 (dev_priv, file_priv, header, cmdbuf)) {
2874 DRM_ERROR("radeon_emit_packets failed\n");
2879 case RADEON_CMD_SCALARS:
2880 DRM_DEBUG("RADEON_CMD_SCALARS\n");
2881 if (radeon_emit_scalars(dev_priv, header, cmdbuf)) {
2882 DRM_ERROR("radeon_emit_scalars failed\n");
2887 case RADEON_CMD_VECTORS:
2888 DRM_DEBUG("RADEON_CMD_VECTORS\n");
2889 if (radeon_emit_vectors(dev_priv, header, cmdbuf)) {
2890 DRM_ERROR("radeon_emit_vectors failed\n");
2895 case RADEON_CMD_DMA_DISCARD:
2896 DRM_DEBUG("RADEON_CMD_DMA_DISCARD\n");
2897 idx = header.dma.buf_idx;
2898 if (idx < 0 || idx >= dma->buf_count) {
2899 DRM_ERROR("buffer index %d (of %d max)\n",
2900 idx, dma->buf_count - 1);
2904 buf = dma->buflist[idx];
2905 if (buf->file_priv != file_priv || buf->pending) {
2906 DRM_ERROR("bad buffer %p %p %d\n",
2907 buf->file_priv, file_priv,
2912 radeon_cp_discard_buffer(dev, buf);
2915 case RADEON_CMD_PACKET3:
2916 DRM_DEBUG("RADEON_CMD_PACKET3\n");
2917 if (radeon_emit_packet3(dev, file_priv, cmdbuf)) {
2918 DRM_ERROR("radeon_emit_packet3 failed\n");
2923 case RADEON_CMD_PACKET3_CLIP:
2924 DRM_DEBUG("RADEON_CMD_PACKET3_CLIP\n");
2925 if (radeon_emit_packet3_cliprect
2926 (dev, file_priv, cmdbuf, orig_nbox)) {
2927 DRM_ERROR("radeon_emit_packet3_clip failed\n");
2932 case RADEON_CMD_SCALARS2:
2933 DRM_DEBUG("RADEON_CMD_SCALARS2\n");
2934 if (radeon_emit_scalars2(dev_priv, header, cmdbuf)) {
2935 DRM_ERROR("radeon_emit_scalars2 failed\n");
2940 case RADEON_CMD_WAIT:
2941 DRM_DEBUG("RADEON_CMD_WAIT\n");
2942 if (radeon_emit_wait(dev, header.wait.flags)) {
2943 DRM_ERROR("radeon_emit_wait failed\n");
2947 case RADEON_CMD_VECLINEAR:
2948 DRM_DEBUG("RADEON_CMD_VECLINEAR\n");
2949 if (radeon_emit_veclinear(dev_priv, header, cmdbuf)) {
2950 DRM_ERROR("radeon_emit_veclinear failed\n");
2956 DRM_ERROR("bad cmd_type %d at %p\n",
2957 header.header.cmd_type,
2958 cmdbuf->buf - sizeof(header));
2963 if (orig_bufsz != 0)
2964 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2966 DRM_DEBUG("DONE\n");
2971 if (orig_bufsz != 0)
2972 drm_free(kbuf, orig_bufsz, DRM_MEM_DRIVER);
2976 static int radeon_cp_getparam(struct drm_device *dev, void *data, struct drm_file *file_priv)
2978 drm_radeon_private_t *dev_priv = dev->dev_private;
2979 drm_radeon_getparam_t *param = data;
2982 DRM_DEBUG("pid=%d\n", DRM_CURRENTPID);
2984 switch (param->param) {
2985 case RADEON_PARAM_GART_BUFFER_OFFSET:
2986 value = dev_priv->gart_buffers_offset;
2988 case RADEON_PARAM_LAST_FRAME:
2989 dev_priv->stats.last_frame_reads++;
2990 value = GET_SCRATCH(0);
2992 case RADEON_PARAM_LAST_DISPATCH:
2993 value = GET_SCRATCH(1);
2995 case RADEON_PARAM_LAST_CLEAR:
2996 dev_priv->stats.last_clear_reads++;
2997 value = GET_SCRATCH(2);
2999 case RADEON_PARAM_IRQ_NR:
3002 case RADEON_PARAM_GART_BASE:
3003 value = dev_priv->gart_vm_start;
3005 case RADEON_PARAM_REGISTER_HANDLE:
3006 value = dev_priv->mmio->offset;
3008 case RADEON_PARAM_STATUS_HANDLE:
3009 value = dev_priv->ring_rptr_offset;
3011 #if BITS_PER_LONG == 32
3013 * This ioctl() doesn't work on 64-bit platforms because hw_lock is a
3014 * pointer which can't fit into an int-sized variable. According to
3015 * Michel Dänzer, the ioctl() is only used on embedded platforms, so
3016 * not supporting it shouldn't be a problem. If the same functionality
3017 * is needed on 64-bit platforms, a new ioctl() would have to be added,
3018 * so backwards-compatibility for the embedded platforms can be
3019 * maintained. --davidm 4-Feb-2004.
3021 case RADEON_PARAM_SAREA_HANDLE:
3022 /* The lock is the first dword in the sarea. */
3023 value = (long)dev->lock.hw_lock;
3026 case RADEON_PARAM_GART_TEX_HANDLE:
3027 value = dev_priv->gart_textures_offset;
3029 case RADEON_PARAM_SCRATCH_OFFSET:
3030 if (!dev_priv->writeback_works)
3032 value = RADEON_SCRATCH_REG_OFFSET;
3034 case RADEON_PARAM_CARD_TYPE:
3035 if (dev_priv->flags & RADEON_IS_PCIE)
3036 value = RADEON_CARD_PCIE;
3037 else if (dev_priv->flags & RADEON_IS_AGP)
3038 value = RADEON_CARD_AGP;
3040 value = RADEON_CARD_PCI;
3042 case RADEON_PARAM_VBLANK_CRTC:
3043 value = radeon_vblank_crtc_get(dev);
3045 case RADEON_PARAM_FB_LOCATION:
3046 value = radeon_read_fb_location(dev_priv);
3048 case RADEON_PARAM_NUM_GB_PIPES:
3049 value = dev_priv->num_gb_pipes;
3052 DRM_DEBUG("Invalid parameter %d\n", param->param);
3056 if (DRM_COPY_TO_USER(param->value, &value, sizeof(int))) {
3057 DRM_ERROR("copy_to_user\n");
3064 static int radeon_cp_setparam(struct drm_device *dev, void *data, struct drm_file *file_priv)
3066 drm_radeon_private_t *dev_priv = dev->dev_private;
3067 drm_radeon_setparam_t *sp = data;
3068 struct drm_radeon_driver_file_fields *radeon_priv;
3070 switch (sp->param) {
3071 case RADEON_SETPARAM_FB_LOCATION:
3072 radeon_priv = file_priv->driver_priv;
3073 radeon_priv->radeon_fb_delta = dev_priv->fb_location -
3076 case RADEON_SETPARAM_SWITCH_TILING:
3077 if (sp->value == 0) {
3078 DRM_DEBUG("color tiling disabled\n");
3079 dev_priv->front_pitch_offset &= ~RADEON_DST_TILE_MACRO;
3080 dev_priv->back_pitch_offset &= ~RADEON_DST_TILE_MACRO;
3081 dev_priv->sarea_priv->tiling_enabled = 0;
3082 } else if (sp->value == 1) {
3083 DRM_DEBUG("color tiling enabled\n");
3084 dev_priv->front_pitch_offset |= RADEON_DST_TILE_MACRO;
3085 dev_priv->back_pitch_offset |= RADEON_DST_TILE_MACRO;
3086 dev_priv->sarea_priv->tiling_enabled = 1;
3089 case RADEON_SETPARAM_PCIGART_LOCATION:
3090 dev_priv->pcigart_offset = sp->value;
3091 dev_priv->pcigart_offset_set = 1;
3093 case RADEON_SETPARAM_NEW_MEMMAP:
3094 dev_priv->new_memmap = sp->value;
3096 case RADEON_SETPARAM_PCIGART_TABLE_SIZE:
3097 dev_priv->gart_info.table_size = sp->value;
3098 if (dev_priv->gart_info.table_size < RADEON_PCIGART_TABLE_SIZE)
3099 dev_priv->gart_info.table_size = RADEON_PCIGART_TABLE_SIZE;
3101 case RADEON_SETPARAM_VBLANK_CRTC:
3102 return radeon_vblank_crtc_set(dev, sp->value);
3105 DRM_DEBUG("Invalid parameter %d\n", sp->param);
3112 /* When a client dies:
3113 * - Check for and clean up flipped page state
3114 * - Free any alloced GART memory.
3115 * - Free any alloced radeon surfaces.
3117 * DRM infrastructure takes care of reclaiming dma buffers.
3119 void radeon_driver_preclose(struct drm_device *dev, struct drm_file *file_priv)
3121 if (dev->dev_private) {
3122 drm_radeon_private_t *dev_priv = dev->dev_private;
3123 dev_priv->page_flipping = 0;
3124 radeon_mem_release(file_priv, dev_priv->gart_heap);
3125 radeon_mem_release(file_priv, dev_priv->fb_heap);
3126 radeon_surfaces_release(file_priv, dev_priv);
3130 void radeon_driver_lastclose(struct drm_device *dev)
3132 if (dev->dev_private) {
3133 drm_radeon_private_t *dev_priv = dev->dev_private;
3135 if (dev_priv->sarea_priv &&
3136 dev_priv->sarea_priv->pfCurrentPage != 0)
3137 radeon_cp_dispatch_flip(dev);
3140 radeon_do_release(dev);
3143 int radeon_driver_open(struct drm_device *dev, struct drm_file *file_priv)
3145 drm_radeon_private_t *dev_priv = dev->dev_private;
3146 struct drm_radeon_driver_file_fields *radeon_priv;
3150 (struct drm_radeon_driver_file_fields *)
3151 drm_alloc(sizeof(*radeon_priv), DRM_MEM_FILES);
3156 file_priv->driver_priv = radeon_priv;
3159 radeon_priv->radeon_fb_delta = dev_priv->fb_location;
3161 radeon_priv->radeon_fb_delta = 0;
3165 void radeon_driver_postclose(struct drm_device *dev, struct drm_file *file_priv)
3167 struct drm_radeon_driver_file_fields *radeon_priv =
3168 file_priv->driver_priv;
3170 drm_free(radeon_priv, sizeof(*radeon_priv), DRM_MEM_FILES);
3173 struct drm_ioctl_desc radeon_ioctls[] = {
3174 DRM_IOCTL_DEF(DRM_RADEON_CP_INIT, radeon_cp_init, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3175 DRM_IOCTL_DEF(DRM_RADEON_CP_START, radeon_cp_start, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3176 DRM_IOCTL_DEF(DRM_RADEON_CP_STOP, radeon_cp_stop, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3177 DRM_IOCTL_DEF(DRM_RADEON_CP_RESET, radeon_cp_reset, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3178 DRM_IOCTL_DEF(DRM_RADEON_CP_IDLE, radeon_cp_idle, DRM_AUTH),
3179 DRM_IOCTL_DEF(DRM_RADEON_CP_RESUME, radeon_cp_resume, DRM_AUTH),
3180 DRM_IOCTL_DEF(DRM_RADEON_RESET, radeon_engine_reset, DRM_AUTH),
3181 DRM_IOCTL_DEF(DRM_RADEON_FULLSCREEN, radeon_fullscreen, DRM_AUTH),
3182 DRM_IOCTL_DEF(DRM_RADEON_SWAP, radeon_cp_swap, DRM_AUTH),
3183 DRM_IOCTL_DEF(DRM_RADEON_CLEAR, radeon_cp_clear, DRM_AUTH),
3184 DRM_IOCTL_DEF(DRM_RADEON_VERTEX, radeon_cp_vertex, DRM_AUTH),
3185 DRM_IOCTL_DEF(DRM_RADEON_INDICES, radeon_cp_indices, DRM_AUTH),
3186 DRM_IOCTL_DEF(DRM_RADEON_TEXTURE, radeon_cp_texture, DRM_AUTH),
3187 DRM_IOCTL_DEF(DRM_RADEON_STIPPLE, radeon_cp_stipple, DRM_AUTH),
3188 DRM_IOCTL_DEF(DRM_RADEON_INDIRECT, radeon_cp_indirect, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3189 DRM_IOCTL_DEF(DRM_RADEON_VERTEX2, radeon_cp_vertex2, DRM_AUTH),
3190 DRM_IOCTL_DEF(DRM_RADEON_CMDBUF, radeon_cp_cmdbuf, DRM_AUTH),
3191 DRM_IOCTL_DEF(DRM_RADEON_GETPARAM, radeon_cp_getparam, DRM_AUTH),
3192 DRM_IOCTL_DEF(DRM_RADEON_FLIP, radeon_cp_flip, DRM_AUTH),
3193 DRM_IOCTL_DEF(DRM_RADEON_ALLOC, radeon_mem_alloc, DRM_AUTH),
3194 DRM_IOCTL_DEF(DRM_RADEON_FREE, radeon_mem_free, DRM_AUTH),
3195 DRM_IOCTL_DEF(DRM_RADEON_INIT_HEAP, radeon_mem_init_heap, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
3196 DRM_IOCTL_DEF(DRM_RADEON_IRQ_EMIT, radeon_irq_emit, DRM_AUTH),
3197 DRM_IOCTL_DEF(DRM_RADEON_IRQ_WAIT, radeon_irq_wait, DRM_AUTH),
3198 DRM_IOCTL_DEF(DRM_RADEON_SETPARAM, radeon_cp_setparam, DRM_AUTH),
3199 DRM_IOCTL_DEF(DRM_RADEON_SURF_ALLOC, radeon_surface_alloc, DRM_AUTH),
3200 DRM_IOCTL_DEF(DRM_RADEON_SURF_FREE, radeon_surface_free, DRM_AUTH)
3203 int radeon_max_ioctl = DRM_ARRAY_SIZE(radeon_ioctls);