1 /*
2  * Copyright 2009 Red Hat Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  * Authors: Ben Skeggs
23  */
24 
25 /* NVIDIA context programs handle a number of other conditions which are
26  * not implemented in our versions.  It's not clear why NVIDIA context
27  * programs have this code, nor whether it's strictly necessary for
28  * correct operation.  We'll implement additional handling if/when we
29  * discover it's necessary.
30  *
31  * - On context save, NVIDIA set 0x400314 bit 0 to 1 if the "3D state"
32  *   flag is set, this gets saved into the context.
33  * - On context save, the context program for all cards load nsource
34  *   into a flag register and check for ILLEGAL_MTHD.  If it's set,
35  *   opcode 0x60000d is called before resuming normal operation.
36  * - Some context programs check more conditions than the above.  NV44
37  *   checks: ((nsource & 0x0857) || (0x400718 & 0x0100) || (intr & 0x0001))
38  *   and calls 0x60000d before resuming normal operation.
39  * - At the very beginning of NVIDIA's context programs, flag 9 is checked
40  *   and if true 0x800001 is called with count=0, pos=0, the flag is cleared
41  *   and then the ctxprog is aborted.  It looks like a complicated NOP,
42  *   its purpose is unknown.
43  * - In the section of code that loads the per-vs state, NVIDIA check
44  *   flag 10.  If it's set, they only transfer the small 0x300 byte block
45  *   of state + the state for a single vs as opposed to the state for
46  *   all vs units.  It doesn't seem likely that it'll occur in normal
47  *   operation, especially seeing as it appears NVIDIA may have screwed
48  *   up the ctxprogs for some cards and have an invalid instruction
49  *   rather than a cp_lsr(ctx, dwords_for_1_vs_unit) instruction.
50  * - There's a number of places where context offset 0 (where we place
51  *   the PRAMIN offset of the context) is loaded into either 0x408000,
52  *   0x408004 or 0x408008.  Not sure what's up there either.
53  * - The ctxprogs for some cards save 0x400a00 again during the cleanup
54  *   path for auto-loadctx.
55  */
56 
57 #define CP_FLAG_CLEAR                 0
58 #define CP_FLAG_SET                   1
59 #define CP_FLAG_SWAP_DIRECTION        ((0 * 32) + 0)
60 #define CP_FLAG_SWAP_DIRECTION_LOAD   0
61 #define CP_FLAG_SWAP_DIRECTION_SAVE   1
62 #define CP_FLAG_USER_SAVE             ((0 * 32) + 5)
63 #define CP_FLAG_USER_SAVE_NOT_PENDING 0
64 #define CP_FLAG_USER_SAVE_PENDING     1
65 #define CP_FLAG_USER_LOAD             ((0 * 32) + 6)
66 #define CP_FLAG_USER_LOAD_NOT_PENDING 0
67 #define CP_FLAG_USER_LOAD_PENDING     1
68 #define CP_FLAG_STATUS                ((3 * 32) + 0)
69 #define CP_FLAG_STATUS_IDLE           0
70 #define CP_FLAG_STATUS_BUSY           1
71 #define CP_FLAG_AUTO_SAVE             ((3 * 32) + 4)
72 #define CP_FLAG_AUTO_SAVE_NOT_PENDING 0
73 #define CP_FLAG_AUTO_SAVE_PENDING     1
74 #define CP_FLAG_AUTO_LOAD             ((3 * 32) + 5)
75 #define CP_FLAG_AUTO_LOAD_NOT_PENDING 0
76 #define CP_FLAG_AUTO_LOAD_PENDING     1
77 #define CP_FLAG_UNK54                 ((3 * 32) + 6)
78 #define CP_FLAG_UNK54_CLEAR           0
79 #define CP_FLAG_UNK54_SET             1
80 #define CP_FLAG_ALWAYS                ((3 * 32) + 8)
81 #define CP_FLAG_ALWAYS_FALSE          0
82 #define CP_FLAG_ALWAYS_TRUE           1
83 #define CP_FLAG_UNK57                 ((3 * 32) + 9)
84 #define CP_FLAG_UNK57_CLEAR           0
85 #define CP_FLAG_UNK57_SET             1
86 
87 #define CP_CTX                   0x00100000
88 #define CP_CTX_COUNT             0x000fc000
89 #define CP_CTX_COUNT_SHIFT               14
90 #define CP_CTX_REG               0x00003fff
91 #define CP_LOAD_SR               0x00200000
92 #define CP_LOAD_SR_VALUE         0x000fffff
93 #define CP_BRA                   0x00400000
94 #define CP_BRA_IP                0x0000ff00
95 #define CP_BRA_IP_SHIFT                   8
96 #define CP_BRA_IF_CLEAR          0x00000080
97 #define CP_BRA_FLAG              0x0000007f
98 #define CP_WAIT                  0x00500000
99 #define CP_WAIT_SET              0x00000080
100 #define CP_WAIT_FLAG             0x0000007f
101 #define CP_SET                   0x00700000
102 #define CP_SET_1                 0x00000080
103 #define CP_SET_FLAG              0x0000007f
104 #define CP_NEXT_TO_SWAP          0x00600007
105 #define CP_NEXT_TO_CURRENT       0x00600009
106 #define CP_SET_CONTEXT_POINTER   0x0060000a
107 #define CP_END                   0x0060000e
108 #define CP_LOAD_MAGIC_UNK01      0x00800001 /* unknown */
109 #define CP_LOAD_MAGIC_NV44TCL    0x00800029 /* per-vs state (0x4497) */
110 #define CP_LOAD_MAGIC_NV40TCL    0x00800041 /* per-vs state (0x4097) */
111 
112 #include "ctxnv40.h"
113 #include "nv40.h"
114 #include <core/device.h>
115 
116 /* TODO:
117  *  - get vs count from 0x1540
118  */
119 
120 static int
121 nv40_gr_vs_count(struct nvkm_device *device)
122 {
123 
124 	switch (device->chipset) {
125 	case 0x47:
126 	case 0x49:
127 	case 0x4b:
128 		return 8;
129 	case 0x40:
130 		return 6;
131 	case 0x41:
132 	case 0x42:
133 		return 5;
134 	case 0x43:
135 	case 0x44:
136 	case 0x46:
137 	case 0x4a:
138 		return 3;
139 	case 0x4c:
140 	case 0x4e:
141 	case 0x67:
142 	default:
143 		return 1;
144 	}
145 }
146 
147 
148 enum cp_label {
149 	cp_check_load = 1,
150 	cp_setup_auto_load,
151 	cp_setup_load,
152 	cp_setup_save,
153 	cp_swap_state,
154 	cp_swap_state3d_3_is_save,
155 	cp_prepare_exit,
156 	cp_exit,
157 };
158 
159 static void
160 nv40_gr_construct_general(struct nvkm_grctx *ctx)
161 {
162 	struct nvkm_device *device = ctx->device;
163 	int i;
164 
165 	cp_ctx(ctx, 0x4000a4, 1);
166 	gr_def(ctx, 0x4000a4, 0x00000008);
167 	cp_ctx(ctx, 0x400144, 58);
168 	gr_def(ctx, 0x400144, 0x00000001);
169 	cp_ctx(ctx, 0x400314, 1);
170 	gr_def(ctx, 0x400314, 0x00000000);
171 	cp_ctx(ctx, 0x400400, 10);
172 	cp_ctx(ctx, 0x400480, 10);
173 	cp_ctx(ctx, 0x400500, 19);
174 	gr_def(ctx, 0x400514, 0x00040000);
175 	gr_def(ctx, 0x400524, 0x55555555);
176 	gr_def(ctx, 0x400528, 0x55555555);
177 	gr_def(ctx, 0x40052c, 0x55555555);
178 	gr_def(ctx, 0x400530, 0x55555555);
179 	cp_ctx(ctx, 0x400560, 6);
180 	gr_def(ctx, 0x400568, 0x0000ffff);
181 	gr_def(ctx, 0x40056c, 0x0000ffff);
182 	cp_ctx(ctx, 0x40057c, 5);
183 	cp_ctx(ctx, 0x400710, 3);
184 	gr_def(ctx, 0x400710, 0x20010001);
185 	gr_def(ctx, 0x400714, 0x0f73ef00);
186 	cp_ctx(ctx, 0x400724, 1);
187 	gr_def(ctx, 0x400724, 0x02008821);
188 	cp_ctx(ctx, 0x400770, 3);
189 	if (device->chipset == 0x40) {
190 		cp_ctx(ctx, 0x400814, 4);
191 		cp_ctx(ctx, 0x400828, 5);
192 		cp_ctx(ctx, 0x400840, 5);
193 		gr_def(ctx, 0x400850, 0x00000040);
194 		cp_ctx(ctx, 0x400858, 4);
195 		gr_def(ctx, 0x400858, 0x00000040);
196 		gr_def(ctx, 0x40085c, 0x00000040);
197 		gr_def(ctx, 0x400864, 0x80000000);
198 		cp_ctx(ctx, 0x40086c, 9);
199 		gr_def(ctx, 0x40086c, 0x80000000);
200 		gr_def(ctx, 0x400870, 0x80000000);
201 		gr_def(ctx, 0x400874, 0x80000000);
202 		gr_def(ctx, 0x400878, 0x80000000);
203 		gr_def(ctx, 0x400888, 0x00000040);
204 		gr_def(ctx, 0x40088c, 0x80000000);
205 		cp_ctx(ctx, 0x4009c0, 8);
206 		gr_def(ctx, 0x4009cc, 0x80000000);
207 		gr_def(ctx, 0x4009dc, 0x80000000);
208 	} else {
209 		cp_ctx(ctx, 0x400840, 20);
210 		if (nv44_gr_class(ctx->device)) {
211 			for (i = 0; i < 8; i++)
212 				gr_def(ctx, 0x400860 + (i * 4), 0x00000001);
213 		}
214 		gr_def(ctx, 0x400880, 0x00000040);
215 		gr_def(ctx, 0x400884, 0x00000040);
216 		gr_def(ctx, 0x400888, 0x00000040);
217 		cp_ctx(ctx, 0x400894, 11);
218 		gr_def(ctx, 0x400894, 0x00000040);
219 		if (!nv44_gr_class(ctx->device)) {
220 			for (i = 0; i < 8; i++)
221 				gr_def(ctx, 0x4008a0 + (i * 4), 0x80000000);
222 		}
223 		cp_ctx(ctx, 0x4008e0, 2);
224 		cp_ctx(ctx, 0x4008f8, 2);
225 		if (device->chipset == 0x4c ||
226 		    (device->chipset & 0xf0) == 0x60)
227 			cp_ctx(ctx, 0x4009f8, 1);
228 	}
229 	cp_ctx(ctx, 0x400a00, 73);
230 	gr_def(ctx, 0x400b0c, 0x0b0b0b0c);
231 	cp_ctx(ctx, 0x401000, 4);
232 	cp_ctx(ctx, 0x405004, 1);
233 	switch (device->chipset) {
234 	case 0x47:
235 	case 0x49:
236 	case 0x4b:
237 		cp_ctx(ctx, 0x403448, 1);
238 		gr_def(ctx, 0x403448, 0x00001010);
239 		break;
240 	default:
241 		cp_ctx(ctx, 0x403440, 1);
242 		switch (device->chipset) {
243 		case 0x40:
244 			gr_def(ctx, 0x403440, 0x00000010);
245 			break;
246 		case 0x44:
247 		case 0x46:
248 		case 0x4a:
249 			gr_def(ctx, 0x403440, 0x00003010);
250 			break;
251 		case 0x41:
252 		case 0x42:
253 		case 0x43:
254 		case 0x4c:
255 		case 0x4e:
256 		case 0x67:
257 		default:
258 			gr_def(ctx, 0x403440, 0x00001010);
259 			break;
260 		}
261 		break;
262 	}
263 }
264 
265 static void
266 nv40_gr_construct_state3d(struct nvkm_grctx *ctx)
267 {
268 	struct nvkm_device *device = ctx->device;
269 	int i;
270 
271 	if (device->chipset == 0x40) {
272 		cp_ctx(ctx, 0x401880, 51);
273 		gr_def(ctx, 0x401940, 0x00000100);
274 	} else
275 	if (device->chipset == 0x46 || device->chipset == 0x47 ||
276 	    device->chipset == 0x49 || device->chipset == 0x4b) {
277 		cp_ctx(ctx, 0x401880, 32);
278 		for (i = 0; i < 16; i++)
279 			gr_def(ctx, 0x401880 + (i * 4), 0x00000111);
280 		if (device->chipset == 0x46)
281 			cp_ctx(ctx, 0x401900, 16);
282 		cp_ctx(ctx, 0x401940, 3);
283 	}
284 	cp_ctx(ctx, 0x40194c, 18);
285 	gr_def(ctx, 0x401954, 0x00000111);
286 	gr_def(ctx, 0x401958, 0x00080060);
287 	gr_def(ctx, 0x401974, 0x00000080);
288 	gr_def(ctx, 0x401978, 0xffff0000);
289 	gr_def(ctx, 0x40197c, 0x00000001);
290 	gr_def(ctx, 0x401990, 0x46400000);
291 	if (device->chipset == 0x40) {
292 		cp_ctx(ctx, 0x4019a0, 2);
293 		cp_ctx(ctx, 0x4019ac, 5);
294 	} else {
295 		cp_ctx(ctx, 0x4019a0, 1);
296 		cp_ctx(ctx, 0x4019b4, 3);
297 	}
298 	gr_def(ctx, 0x4019bc, 0xffff0000);
299 	switch (device->chipset) {
300 	case 0x46:
301 	case 0x47:
302 	case 0x49:
303 	case 0x4b:
304 		cp_ctx(ctx, 0x4019c0, 18);
305 		for (i = 0; i < 16; i++)
306 			gr_def(ctx, 0x4019c0 + (i * 4), 0x88888888);
307 		break;
308 	}
309 	cp_ctx(ctx, 0x401a08, 8);
310 	gr_def(ctx, 0x401a10, 0x0fff0000);
311 	gr_def(ctx, 0x401a14, 0x0fff0000);
312 	gr_def(ctx, 0x401a1c, 0x00011100);
313 	cp_ctx(ctx, 0x401a2c, 4);
314 	cp_ctx(ctx, 0x401a44, 26);
315 	for (i = 0; i < 16; i++)
316 		gr_def(ctx, 0x401a44 + (i * 4), 0x07ff0000);
317 	gr_def(ctx, 0x401a8c, 0x4b7fffff);
318 	if (device->chipset == 0x40) {
319 		cp_ctx(ctx, 0x401ab8, 3);
320 	} else {
321 		cp_ctx(ctx, 0x401ab8, 1);
322 		cp_ctx(ctx, 0x401ac0, 1);
323 	}
324 	cp_ctx(ctx, 0x401ad0, 8);
325 	gr_def(ctx, 0x401ad0, 0x30201000);
326 	gr_def(ctx, 0x401ad4, 0x70605040);
327 	gr_def(ctx, 0x401ad8, 0xb8a89888);
328 	gr_def(ctx, 0x401adc, 0xf8e8d8c8);
329 	cp_ctx(ctx, 0x401b10, device->chipset == 0x40 ? 2 : 1);
330 	gr_def(ctx, 0x401b10, 0x40100000);
331 	cp_ctx(ctx, 0x401b18, device->chipset == 0x40 ? 6 : 5);
332 	gr_def(ctx, 0x401b28, device->chipset == 0x40 ?
333 			      0x00000004 : 0x00000000);
334 	cp_ctx(ctx, 0x401b30, 25);
335 	gr_def(ctx, 0x401b34, 0x0000ffff);
336 	gr_def(ctx, 0x401b68, 0x435185d6);
337 	gr_def(ctx, 0x401b6c, 0x2155b699);
338 	gr_def(ctx, 0x401b70, 0xfedcba98);
339 	gr_def(ctx, 0x401b74, 0x00000098);
340 	gr_def(ctx, 0x401b84, 0xffffffff);
341 	gr_def(ctx, 0x401b88, 0x00ff7000);
342 	gr_def(ctx, 0x401b8c, 0x0000ffff);
343 	if (device->chipset != 0x44 && device->chipset != 0x4a &&
344 	    device->chipset != 0x4e)
345 		cp_ctx(ctx, 0x401b94, 1);
346 	cp_ctx(ctx, 0x401b98, 8);
347 	gr_def(ctx, 0x401b9c, 0x00ff0000);
348 	cp_ctx(ctx, 0x401bc0, 9);
349 	gr_def(ctx, 0x401be0, 0x00ffff00);
350 	cp_ctx(ctx, 0x401c00, 192);
351 	for (i = 0; i < 16; i++) { /* fragment texture units */
352 		gr_def(ctx, 0x401c40 + (i * 4), 0x00018488);
353 		gr_def(ctx, 0x401c80 + (i * 4), 0x00028202);
354 		gr_def(ctx, 0x401d00 + (i * 4), 0x0000aae4);
355 		gr_def(ctx, 0x401d40 + (i * 4), 0x01012000);
356 		gr_def(ctx, 0x401d80 + (i * 4), 0x00080008);
357 		gr_def(ctx, 0x401e00 + (i * 4), 0x00100008);
358 	}
359 	for (i = 0; i < 4; i++) { /* vertex texture units */
360 		gr_def(ctx, 0x401e90 + (i * 4), 0x0001bc80);
361 		gr_def(ctx, 0x401ea0 + (i * 4), 0x00000202);
362 		gr_def(ctx, 0x401ec0 + (i * 4), 0x00000008);
363 		gr_def(ctx, 0x401ee0 + (i * 4), 0x00080008);
364 	}
365 	cp_ctx(ctx, 0x400f5c, 3);
366 	gr_def(ctx, 0x400f5c, 0x00000002);
367 	cp_ctx(ctx, 0x400f84, 1);
368 }
369 
370 static void
371 nv40_gr_construct_state3d_2(struct nvkm_grctx *ctx)
372 {
373 	struct nvkm_device *device = ctx->device;
374 	int i;
375 
376 	cp_ctx(ctx, 0x402000, 1);
377 	cp_ctx(ctx, 0x402404, device->chipset == 0x40 ? 1 : 2);
378 	switch (device->chipset) {
379 	case 0x40:
380 		gr_def(ctx, 0x402404, 0x00000001);
381 		break;
382 	case 0x4c:
383 	case 0x4e:
384 	case 0x67:
385 		gr_def(ctx, 0x402404, 0x00000020);
386 		break;
387 	case 0x46:
388 	case 0x49:
389 	case 0x4b:
390 		gr_def(ctx, 0x402404, 0x00000421);
391 		break;
392 	default:
393 		gr_def(ctx, 0x402404, 0x00000021);
394 	}
395 	if (device->chipset != 0x40)
396 		gr_def(ctx, 0x402408, 0x030c30c3);
397 	switch (device->chipset) {
398 	case 0x44:
399 	case 0x46:
400 	case 0x4a:
401 	case 0x4c:
402 	case 0x4e:
403 	case 0x67:
404 		cp_ctx(ctx, 0x402440, 1);
405 		gr_def(ctx, 0x402440, 0x00011001);
406 		break;
407 	default:
408 		break;
409 	}
410 	cp_ctx(ctx, 0x402480, device->chipset == 0x40 ? 8 : 9);
411 	gr_def(ctx, 0x402488, 0x3e020200);
412 	gr_def(ctx, 0x40248c, 0x00ffffff);
413 	switch (device->chipset) {
414 	case 0x40:
415 		gr_def(ctx, 0x402490, 0x60103f00);
416 		break;
417 	case 0x47:
418 		gr_def(ctx, 0x402490, 0x40103f00);
419 		break;
420 	case 0x41:
421 	case 0x42:
422 	case 0x49:
423 	case 0x4b:
424 		gr_def(ctx, 0x402490, 0x20103f00);
425 		break;
426 	default:
427 		gr_def(ctx, 0x402490, 0x0c103f00);
428 		break;
429 	}
430 	gr_def(ctx, 0x40249c, device->chipset <= 0x43 ?
431 			      0x00020000 : 0x00040000);
432 	cp_ctx(ctx, 0x402500, 31);
433 	gr_def(ctx, 0x402530, 0x00008100);
434 	if (device->chipset == 0x40)
435 		cp_ctx(ctx, 0x40257c, 6);
436 	cp_ctx(ctx, 0x402594, 16);
437 	cp_ctx(ctx, 0x402800, 17);
438 	gr_def(ctx, 0x402800, 0x00000001);
439 	switch (device->chipset) {
440 	case 0x47:
441 	case 0x49:
442 	case 0x4b:
443 		cp_ctx(ctx, 0x402864, 1);
444 		gr_def(ctx, 0x402864, 0x00001001);
445 		cp_ctx(ctx, 0x402870, 3);
446 		gr_def(ctx, 0x402878, 0x00000003);
447 		if (device->chipset != 0x47) { /* belong at end!! */
448 			cp_ctx(ctx, 0x402900, 1);
449 			cp_ctx(ctx, 0x402940, 1);
450 			cp_ctx(ctx, 0x402980, 1);
451 			cp_ctx(ctx, 0x4029c0, 1);
452 			cp_ctx(ctx, 0x402a00, 1);
453 			cp_ctx(ctx, 0x402a40, 1);
454 			cp_ctx(ctx, 0x402a80, 1);
455 			cp_ctx(ctx, 0x402ac0, 1);
456 		}
457 		break;
458 	case 0x40:
459 		cp_ctx(ctx, 0x402844, 1);
460 		gr_def(ctx, 0x402844, 0x00000001);
461 		cp_ctx(ctx, 0x402850, 1);
462 		break;
463 	default:
464 		cp_ctx(ctx, 0x402844, 1);
465 		gr_def(ctx, 0x402844, 0x00001001);
466 		cp_ctx(ctx, 0x402850, 2);
467 		gr_def(ctx, 0x402854, 0x00000003);
468 		break;
469 	}
470 
471 	cp_ctx(ctx, 0x402c00, 4);
472 	gr_def(ctx, 0x402c00, device->chipset == 0x40 ?
473 			      0x80800001 : 0x00888001);
474 	switch (device->chipset) {
475 	case 0x47:
476 	case 0x49:
477 	case 0x4b:
478 		cp_ctx(ctx, 0x402c20, 40);
479 		for (i = 0; i < 32; i++)
480 			gr_def(ctx, 0x402c40 + (i * 4), 0xffffffff);
481 		cp_ctx(ctx, 0x4030b8, 13);
482 		gr_def(ctx, 0x4030dc, 0x00000005);
483 		gr_def(ctx, 0x4030e8, 0x0000ffff);
484 		break;
485 	default:
486 		cp_ctx(ctx, 0x402c10, 4);
487 		if (device->chipset == 0x40)
488 			cp_ctx(ctx, 0x402c20, 36);
489 		else
490 		if (device->chipset <= 0x42)
491 			cp_ctx(ctx, 0x402c20, 24);
492 		else
493 		if (device->chipset <= 0x4a)
494 			cp_ctx(ctx, 0x402c20, 16);
495 		else
496 			cp_ctx(ctx, 0x402c20, 8);
497 		cp_ctx(ctx, 0x402cb0, device->chipset == 0x40 ? 12 : 13);
498 		gr_def(ctx, 0x402cd4, 0x00000005);
499 		if (device->chipset != 0x40)
500 			gr_def(ctx, 0x402ce0, 0x0000ffff);
501 		break;
502 	}
503 
504 	cp_ctx(ctx, 0x403400, device->chipset == 0x40 ? 4 : 3);
505 	cp_ctx(ctx, 0x403410, device->chipset == 0x40 ? 4 : 3);
506 	cp_ctx(ctx, 0x403420, nv40_gr_vs_count(ctx->device));
507 	for (i = 0; i < nv40_gr_vs_count(ctx->device); i++)
508 		gr_def(ctx, 0x403420 + (i * 4), 0x00005555);
509 
510 	if (device->chipset != 0x40) {
511 		cp_ctx(ctx, 0x403600, 1);
512 		gr_def(ctx, 0x403600, 0x00000001);
513 	}
514 	cp_ctx(ctx, 0x403800, 1);
515 
516 	cp_ctx(ctx, 0x403c18, 1);
517 	gr_def(ctx, 0x403c18, 0x00000001);
518 	switch (device->chipset) {
519 	case 0x46:
520 	case 0x47:
521 	case 0x49:
522 	case 0x4b:
523 		cp_ctx(ctx, 0x405018, 1);
524 		gr_def(ctx, 0x405018, 0x08e00001);
525 		cp_ctx(ctx, 0x405c24, 1);
526 		gr_def(ctx, 0x405c24, 0x000e3000);
527 		break;
528 	}
529 	if (device->chipset != 0x4e)
530 		cp_ctx(ctx, 0x405800, 11);
531 	cp_ctx(ctx, 0x407000, 1);
532 }
533 
534 static void
535 nv40_gr_construct_state3d_3(struct nvkm_grctx *ctx)
536 {
537 	int len = nv44_gr_class(ctx->device) ? 0x0084 : 0x0684;
538 
539 	cp_out (ctx, 0x300000);
540 	cp_lsr (ctx, len - 4);
541 	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_swap_state3d_3_is_save);
542 	cp_lsr (ctx, len);
543 	cp_name(ctx, cp_swap_state3d_3_is_save);
544 	cp_out (ctx, 0x800001);
545 
546 	ctx->ctxvals_pos += len;
547 }
548 
549 static void
550 nv40_gr_construct_shader(struct nvkm_grctx *ctx)
551 {
552 	struct nvkm_device *device = ctx->device;
553 	struct nvkm_gpuobj *obj = ctx->data;
554 	int vs, vs_nr, vs_len, vs_nr_b0, vs_nr_b1, b0_offset, b1_offset;
555 	int offset, i;
556 
557 	vs_nr    = nv40_gr_vs_count(ctx->device);
558 	vs_nr_b0 = 363;
559 	vs_nr_b1 = device->chipset == 0x40 ? 128 : 64;
560 	if (device->chipset == 0x40) {
561 		b0_offset = 0x2200/4; /* 33a0 */
562 		b1_offset = 0x55a0/4; /* 1500 */
563 		vs_len = 0x6aa0/4;
564 	} else
565 	if (device->chipset == 0x41 || device->chipset == 0x42) {
566 		b0_offset = 0x2200/4; /* 2200 */
567 		b1_offset = 0x4400/4; /* 0b00 */
568 		vs_len = 0x4f00/4;
569 	} else {
570 		b0_offset = 0x1d40/4; /* 2200 */
571 		b1_offset = 0x3f40/4; /* 0b00 : 0a40 */
572 		vs_len = nv44_gr_class(device) ? 0x4980/4 : 0x4a40/4;
573 	}
574 
575 	cp_lsr(ctx, vs_len * vs_nr + 0x300/4);
576 	cp_out(ctx, nv44_gr_class(device) ? 0x800029 : 0x800041);
577 
578 	offset = ctx->ctxvals_pos;
579 	ctx->ctxvals_pos += (0x0300/4 + (vs_nr * vs_len));
580 
581 	if (ctx->mode != NVKM_GRCTX_VALS)
582 		return;
583 
584 	offset += 0x0280/4;
585 	for (i = 0; i < 16; i++, offset += 2)
586 		nv_wo32(obj, offset * 4, 0x3f800000);
587 
588 	for (vs = 0; vs < vs_nr; vs++, offset += vs_len) {
589 		for (i = 0; i < vs_nr_b0 * 6; i += 6)
590 			nv_wo32(obj, (offset + b0_offset + i) * 4, 0x00000001);
591 		for (i = 0; i < vs_nr_b1 * 4; i += 4)
592 			nv_wo32(obj, (offset + b1_offset + i) * 4, 0x3f800000);
593 	}
594 }
595 
596 static void
597 nv40_grctx_generate(struct nvkm_grctx *ctx)
598 {
599 	/* decide whether we're loading/unloading the context */
600 	cp_bra (ctx, AUTO_SAVE, PENDING, cp_setup_save);
601 	cp_bra (ctx, USER_SAVE, PENDING, cp_setup_save);
602 
603 	cp_name(ctx, cp_check_load);
604 	cp_bra (ctx, AUTO_LOAD, PENDING, cp_setup_auto_load);
605 	cp_bra (ctx, USER_LOAD, PENDING, cp_setup_load);
606 	cp_bra (ctx, ALWAYS, TRUE, cp_exit);
607 
608 	/* setup for context load */
609 	cp_name(ctx, cp_setup_auto_load);
610 	cp_wait(ctx, STATUS, IDLE);
611 	cp_out (ctx, CP_NEXT_TO_SWAP);
612 	cp_name(ctx, cp_setup_load);
613 	cp_wait(ctx, STATUS, IDLE);
614 	cp_set (ctx, SWAP_DIRECTION, LOAD);
615 	cp_out (ctx, 0x00910880); /* ?? */
616 	cp_out (ctx, 0x00901ffe); /* ?? */
617 	cp_out (ctx, 0x01940000); /* ?? */
618 	cp_lsr (ctx, 0x20);
619 	cp_out (ctx, 0x0060000b); /* ?? */
620 	cp_wait(ctx, UNK57, CLEAR);
621 	cp_out (ctx, 0x0060000c); /* ?? */
622 	cp_bra (ctx, ALWAYS, TRUE, cp_swap_state);
623 
624 	/* setup for context save */
625 	cp_name(ctx, cp_setup_save);
626 	cp_set (ctx, SWAP_DIRECTION, SAVE);
627 
628 	/* general PGRAPH state */
629 	cp_name(ctx, cp_swap_state);
630 	cp_pos (ctx, 0x00020/4);
631 	nv40_gr_construct_general(ctx);
632 	cp_wait(ctx, STATUS, IDLE);
633 
634 	/* 3D state, block 1 */
635 	cp_bra (ctx, UNK54, CLEAR, cp_prepare_exit);
636 	nv40_gr_construct_state3d(ctx);
637 	cp_wait(ctx, STATUS, IDLE);
638 
639 	/* 3D state, block 2 */
640 	nv40_gr_construct_state3d_2(ctx);
641 
642 	/* Some other block of "random" state */
643 	nv40_gr_construct_state3d_3(ctx);
644 
645 	/* Per-vertex shader state */
646 	cp_pos (ctx, ctx->ctxvals_pos);
647 	nv40_gr_construct_shader(ctx);
648 
649 	/* pre-exit state updates */
650 	cp_name(ctx, cp_prepare_exit);
651 	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_check_load);
652 	cp_bra (ctx, USER_SAVE, PENDING, cp_exit);
653 	cp_out (ctx, CP_NEXT_TO_CURRENT);
654 
655 	cp_name(ctx, cp_exit);
656 	cp_set (ctx, USER_SAVE, NOT_PENDING);
657 	cp_set (ctx, USER_LOAD, NOT_PENDING);
658 	cp_out (ctx, CP_END);
659 }
660 
661 void
662 nv40_grctx_fill(struct nvkm_device *device, struct nvkm_gpuobj *mem)
663 {
664 	nv40_grctx_generate(&(struct nvkm_grctx) {
665 			     .device = device,
666 			     .mode = NVKM_GRCTX_VALS,
667 			     .data = mem,
668 			   });
669 }
670 
671 int
672 nv40_grctx_init(struct nvkm_device *device, u32 *size)
673 {
674 	u32 *ctxprog = kmalloc(256 * 4, GFP_KERNEL), i;
675 	struct nvkm_grctx ctx = {
676 		.device = device,
677 		.mode = NVKM_GRCTX_PROG,
678 		.data = ctxprog,
679 		.ctxprog_max = 256,
680 	};
681 
682 	if (!ctxprog)
683 		return -ENOMEM;
684 
685 	nv40_grctx_generate(&ctx);
686 
687 	nv_wr32(device, 0x400324, 0);
688 	for (i = 0; i < ctx.ctxprog_len; i++)
689 		nv_wr32(device, 0x400328, ctxprog[i]);
690 	*size = ctx.ctxvals_pos * 4;
691 
692 	kfree(ctxprog);
693 	return 0;
694 }
695