cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

ctxnv40.c (20120B)


      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
    115/* TODO:
    116 *  - get vs count from 0x1540
    117 */
    118
    119static int
    120nv40_gr_vs_count(struct nvkm_device *device)
    121{
    122
    123	switch (device->chipset) {
    124	case 0x47:
    125	case 0x49:
    126	case 0x4b:
    127		return 8;
    128	case 0x40:
    129		return 6;
    130	case 0x41:
    131	case 0x42:
    132		return 5;
    133	case 0x43:
    134	case 0x44:
    135	case 0x46:
    136	case 0x4a:
    137		return 3;
    138	case 0x4c:
    139	case 0x4e:
    140	case 0x67:
    141	default:
    142		return 1;
    143	}
    144}
    145
    146
    147enum cp_label {
    148	cp_check_load = 1,
    149	cp_setup_auto_load,
    150	cp_setup_load,
    151	cp_setup_save,
    152	cp_swap_state,
    153	cp_swap_state3d_3_is_save,
    154	cp_prepare_exit,
    155	cp_exit,
    156};
    157
    158static void
    159nv40_gr_construct_general(struct nvkm_grctx *ctx)
    160{
    161	struct nvkm_device *device = ctx->device;
    162	int i;
    163
    164	cp_ctx(ctx, 0x4000a4, 1);
    165	gr_def(ctx, 0x4000a4, 0x00000008);
    166	cp_ctx(ctx, 0x400144, 58);
    167	gr_def(ctx, 0x400144, 0x00000001);
    168	cp_ctx(ctx, 0x400314, 1);
    169	gr_def(ctx, 0x400314, 0x00000000);
    170	cp_ctx(ctx, 0x400400, 10);
    171	cp_ctx(ctx, 0x400480, 10);
    172	cp_ctx(ctx, 0x400500, 19);
    173	gr_def(ctx, 0x400514, 0x00040000);
    174	gr_def(ctx, 0x400524, 0x55555555);
    175	gr_def(ctx, 0x400528, 0x55555555);
    176	gr_def(ctx, 0x40052c, 0x55555555);
    177	gr_def(ctx, 0x400530, 0x55555555);
    178	cp_ctx(ctx, 0x400560, 6);
    179	gr_def(ctx, 0x400568, 0x0000ffff);
    180	gr_def(ctx, 0x40056c, 0x0000ffff);
    181	cp_ctx(ctx, 0x40057c, 5);
    182	cp_ctx(ctx, 0x400710, 3);
    183	gr_def(ctx, 0x400710, 0x20010001);
    184	gr_def(ctx, 0x400714, 0x0f73ef00);
    185	cp_ctx(ctx, 0x400724, 1);
    186	gr_def(ctx, 0x400724, 0x02008821);
    187	cp_ctx(ctx, 0x400770, 3);
    188	if (device->chipset == 0x40) {
    189		cp_ctx(ctx, 0x400814, 4);
    190		cp_ctx(ctx, 0x400828, 5);
    191		cp_ctx(ctx, 0x400840, 5);
    192		gr_def(ctx, 0x400850, 0x00000040);
    193		cp_ctx(ctx, 0x400858, 4);
    194		gr_def(ctx, 0x400858, 0x00000040);
    195		gr_def(ctx, 0x40085c, 0x00000040);
    196		gr_def(ctx, 0x400864, 0x80000000);
    197		cp_ctx(ctx, 0x40086c, 9);
    198		gr_def(ctx, 0x40086c, 0x80000000);
    199		gr_def(ctx, 0x400870, 0x80000000);
    200		gr_def(ctx, 0x400874, 0x80000000);
    201		gr_def(ctx, 0x400878, 0x80000000);
    202		gr_def(ctx, 0x400888, 0x00000040);
    203		gr_def(ctx, 0x40088c, 0x80000000);
    204		cp_ctx(ctx, 0x4009c0, 8);
    205		gr_def(ctx, 0x4009cc, 0x80000000);
    206		gr_def(ctx, 0x4009dc, 0x80000000);
    207	} else {
    208		cp_ctx(ctx, 0x400840, 20);
    209		if (nv44_gr_class(ctx->device)) {
    210			for (i = 0; i < 8; i++)
    211				gr_def(ctx, 0x400860 + (i * 4), 0x00000001);
    212		}
    213		gr_def(ctx, 0x400880, 0x00000040);
    214		gr_def(ctx, 0x400884, 0x00000040);
    215		gr_def(ctx, 0x400888, 0x00000040);
    216		cp_ctx(ctx, 0x400894, 11);
    217		gr_def(ctx, 0x400894, 0x00000040);
    218		if (!nv44_gr_class(ctx->device)) {
    219			for (i = 0; i < 8; i++)
    220				gr_def(ctx, 0x4008a0 + (i * 4), 0x80000000);
    221		}
    222		cp_ctx(ctx, 0x4008e0, 2);
    223		cp_ctx(ctx, 0x4008f8, 2);
    224		if (device->chipset == 0x4c ||
    225		    (device->chipset & 0xf0) == 0x60)
    226			cp_ctx(ctx, 0x4009f8, 1);
    227	}
    228	cp_ctx(ctx, 0x400a00, 73);
    229	gr_def(ctx, 0x400b0c, 0x0b0b0b0c);
    230	cp_ctx(ctx, 0x401000, 4);
    231	cp_ctx(ctx, 0x405004, 1);
    232	switch (device->chipset) {
    233	case 0x47:
    234	case 0x49:
    235	case 0x4b:
    236		cp_ctx(ctx, 0x403448, 1);
    237		gr_def(ctx, 0x403448, 0x00001010);
    238		break;
    239	default:
    240		cp_ctx(ctx, 0x403440, 1);
    241		switch (device->chipset) {
    242		case 0x40:
    243			gr_def(ctx, 0x403440, 0x00000010);
    244			break;
    245		case 0x44:
    246		case 0x46:
    247		case 0x4a:
    248			gr_def(ctx, 0x403440, 0x00003010);
    249			break;
    250		case 0x41:
    251		case 0x42:
    252		case 0x43:
    253		case 0x4c:
    254		case 0x4e:
    255		case 0x67:
    256		default:
    257			gr_def(ctx, 0x403440, 0x00001010);
    258			break;
    259		}
    260		break;
    261	}
    262}
    263
    264static void
    265nv40_gr_construct_state3d(struct nvkm_grctx *ctx)
    266{
    267	struct nvkm_device *device = ctx->device;
    268	int i;
    269
    270	if (device->chipset == 0x40) {
    271		cp_ctx(ctx, 0x401880, 51);
    272		gr_def(ctx, 0x401940, 0x00000100);
    273	} else
    274	if (device->chipset == 0x46 || device->chipset == 0x47 ||
    275	    device->chipset == 0x49 || device->chipset == 0x4b) {
    276		cp_ctx(ctx, 0x401880, 32);
    277		for (i = 0; i < 16; i++)
    278			gr_def(ctx, 0x401880 + (i * 4), 0x00000111);
    279		if (device->chipset == 0x46)
    280			cp_ctx(ctx, 0x401900, 16);
    281		cp_ctx(ctx, 0x401940, 3);
    282	}
    283	cp_ctx(ctx, 0x40194c, 18);
    284	gr_def(ctx, 0x401954, 0x00000111);
    285	gr_def(ctx, 0x401958, 0x00080060);
    286	gr_def(ctx, 0x401974, 0x00000080);
    287	gr_def(ctx, 0x401978, 0xffff0000);
    288	gr_def(ctx, 0x40197c, 0x00000001);
    289	gr_def(ctx, 0x401990, 0x46400000);
    290	if (device->chipset == 0x40) {
    291		cp_ctx(ctx, 0x4019a0, 2);
    292		cp_ctx(ctx, 0x4019ac, 5);
    293	} else {
    294		cp_ctx(ctx, 0x4019a0, 1);
    295		cp_ctx(ctx, 0x4019b4, 3);
    296	}
    297	gr_def(ctx, 0x4019bc, 0xffff0000);
    298	switch (device->chipset) {
    299	case 0x46:
    300	case 0x47:
    301	case 0x49:
    302	case 0x4b:
    303		cp_ctx(ctx, 0x4019c0, 18);
    304		for (i = 0; i < 16; i++)
    305			gr_def(ctx, 0x4019c0 + (i * 4), 0x88888888);
    306		break;
    307	}
    308	cp_ctx(ctx, 0x401a08, 8);
    309	gr_def(ctx, 0x401a10, 0x0fff0000);
    310	gr_def(ctx, 0x401a14, 0x0fff0000);
    311	gr_def(ctx, 0x401a1c, 0x00011100);
    312	cp_ctx(ctx, 0x401a2c, 4);
    313	cp_ctx(ctx, 0x401a44, 26);
    314	for (i = 0; i < 16; i++)
    315		gr_def(ctx, 0x401a44 + (i * 4), 0x07ff0000);
    316	gr_def(ctx, 0x401a8c, 0x4b7fffff);
    317	if (device->chipset == 0x40) {
    318		cp_ctx(ctx, 0x401ab8, 3);
    319	} else {
    320		cp_ctx(ctx, 0x401ab8, 1);
    321		cp_ctx(ctx, 0x401ac0, 1);
    322	}
    323	cp_ctx(ctx, 0x401ad0, 8);
    324	gr_def(ctx, 0x401ad0, 0x30201000);
    325	gr_def(ctx, 0x401ad4, 0x70605040);
    326	gr_def(ctx, 0x401ad8, 0xb8a89888);
    327	gr_def(ctx, 0x401adc, 0xf8e8d8c8);
    328	cp_ctx(ctx, 0x401b10, device->chipset == 0x40 ? 2 : 1);
    329	gr_def(ctx, 0x401b10, 0x40100000);
    330	cp_ctx(ctx, 0x401b18, device->chipset == 0x40 ? 6 : 5);
    331	gr_def(ctx, 0x401b28, device->chipset == 0x40 ?
    332			      0x00000004 : 0x00000000);
    333	cp_ctx(ctx, 0x401b30, 25);
    334	gr_def(ctx, 0x401b34, 0x0000ffff);
    335	gr_def(ctx, 0x401b68, 0x435185d6);
    336	gr_def(ctx, 0x401b6c, 0x2155b699);
    337	gr_def(ctx, 0x401b70, 0xfedcba98);
    338	gr_def(ctx, 0x401b74, 0x00000098);
    339	gr_def(ctx, 0x401b84, 0xffffffff);
    340	gr_def(ctx, 0x401b88, 0x00ff7000);
    341	gr_def(ctx, 0x401b8c, 0x0000ffff);
    342	if (device->chipset != 0x44 && device->chipset != 0x4a &&
    343	    device->chipset != 0x4e)
    344		cp_ctx(ctx, 0x401b94, 1);
    345	cp_ctx(ctx, 0x401b98, 8);
    346	gr_def(ctx, 0x401b9c, 0x00ff0000);
    347	cp_ctx(ctx, 0x401bc0, 9);
    348	gr_def(ctx, 0x401be0, 0x00ffff00);
    349	cp_ctx(ctx, 0x401c00, 192);
    350	for (i = 0; i < 16; i++) { /* fragment texture units */
    351		gr_def(ctx, 0x401c40 + (i * 4), 0x00018488);
    352		gr_def(ctx, 0x401c80 + (i * 4), 0x00028202);
    353		gr_def(ctx, 0x401d00 + (i * 4), 0x0000aae4);
    354		gr_def(ctx, 0x401d40 + (i * 4), 0x01012000);
    355		gr_def(ctx, 0x401d80 + (i * 4), 0x00080008);
    356		gr_def(ctx, 0x401e00 + (i * 4), 0x00100008);
    357	}
    358	for (i = 0; i < 4; i++) { /* vertex texture units */
    359		gr_def(ctx, 0x401e90 + (i * 4), 0x0001bc80);
    360		gr_def(ctx, 0x401ea0 + (i * 4), 0x00000202);
    361		gr_def(ctx, 0x401ec0 + (i * 4), 0x00000008);
    362		gr_def(ctx, 0x401ee0 + (i * 4), 0x00080008);
    363	}
    364	cp_ctx(ctx, 0x400f5c, 3);
    365	gr_def(ctx, 0x400f5c, 0x00000002);
    366	cp_ctx(ctx, 0x400f84, 1);
    367}
    368
    369static void
    370nv40_gr_construct_state3d_2(struct nvkm_grctx *ctx)
    371{
    372	struct nvkm_device *device = ctx->device;
    373	int i;
    374
    375	cp_ctx(ctx, 0x402000, 1);
    376	cp_ctx(ctx, 0x402404, device->chipset == 0x40 ? 1 : 2);
    377	switch (device->chipset) {
    378	case 0x40:
    379		gr_def(ctx, 0x402404, 0x00000001);
    380		break;
    381	case 0x4c:
    382	case 0x4e:
    383	case 0x67:
    384		gr_def(ctx, 0x402404, 0x00000020);
    385		break;
    386	case 0x46:
    387	case 0x49:
    388	case 0x4b:
    389		gr_def(ctx, 0x402404, 0x00000421);
    390		break;
    391	default:
    392		gr_def(ctx, 0x402404, 0x00000021);
    393	}
    394	if (device->chipset != 0x40)
    395		gr_def(ctx, 0x402408, 0x030c30c3);
    396	switch (device->chipset) {
    397	case 0x44:
    398	case 0x46:
    399	case 0x4a:
    400	case 0x4c:
    401	case 0x4e:
    402	case 0x67:
    403		cp_ctx(ctx, 0x402440, 1);
    404		gr_def(ctx, 0x402440, 0x00011001);
    405		break;
    406	default:
    407		break;
    408	}
    409	cp_ctx(ctx, 0x402480, device->chipset == 0x40 ? 8 : 9);
    410	gr_def(ctx, 0x402488, 0x3e020200);
    411	gr_def(ctx, 0x40248c, 0x00ffffff);
    412	switch (device->chipset) {
    413	case 0x40:
    414		gr_def(ctx, 0x402490, 0x60103f00);
    415		break;
    416	case 0x47:
    417		gr_def(ctx, 0x402490, 0x40103f00);
    418		break;
    419	case 0x41:
    420	case 0x42:
    421	case 0x49:
    422	case 0x4b:
    423		gr_def(ctx, 0x402490, 0x20103f00);
    424		break;
    425	default:
    426		gr_def(ctx, 0x402490, 0x0c103f00);
    427		break;
    428	}
    429	gr_def(ctx, 0x40249c, device->chipset <= 0x43 ?
    430			      0x00020000 : 0x00040000);
    431	cp_ctx(ctx, 0x402500, 31);
    432	gr_def(ctx, 0x402530, 0x00008100);
    433	if (device->chipset == 0x40)
    434		cp_ctx(ctx, 0x40257c, 6);
    435	cp_ctx(ctx, 0x402594, 16);
    436	cp_ctx(ctx, 0x402800, 17);
    437	gr_def(ctx, 0x402800, 0x00000001);
    438	switch (device->chipset) {
    439	case 0x47:
    440	case 0x49:
    441	case 0x4b:
    442		cp_ctx(ctx, 0x402864, 1);
    443		gr_def(ctx, 0x402864, 0x00001001);
    444		cp_ctx(ctx, 0x402870, 3);
    445		gr_def(ctx, 0x402878, 0x00000003);
    446		if (device->chipset != 0x47) { /* belong at end!! */
    447			cp_ctx(ctx, 0x402900, 1);
    448			cp_ctx(ctx, 0x402940, 1);
    449			cp_ctx(ctx, 0x402980, 1);
    450			cp_ctx(ctx, 0x4029c0, 1);
    451			cp_ctx(ctx, 0x402a00, 1);
    452			cp_ctx(ctx, 0x402a40, 1);
    453			cp_ctx(ctx, 0x402a80, 1);
    454			cp_ctx(ctx, 0x402ac0, 1);
    455		}
    456		break;
    457	case 0x40:
    458		cp_ctx(ctx, 0x402844, 1);
    459		gr_def(ctx, 0x402844, 0x00000001);
    460		cp_ctx(ctx, 0x402850, 1);
    461		break;
    462	default:
    463		cp_ctx(ctx, 0x402844, 1);
    464		gr_def(ctx, 0x402844, 0x00001001);
    465		cp_ctx(ctx, 0x402850, 2);
    466		gr_def(ctx, 0x402854, 0x00000003);
    467		break;
    468	}
    469
    470	cp_ctx(ctx, 0x402c00, 4);
    471	gr_def(ctx, 0x402c00, device->chipset == 0x40 ?
    472			      0x80800001 : 0x00888001);
    473	switch (device->chipset) {
    474	case 0x47:
    475	case 0x49:
    476	case 0x4b:
    477		cp_ctx(ctx, 0x402c20, 40);
    478		for (i = 0; i < 32; i++)
    479			gr_def(ctx, 0x402c40 + (i * 4), 0xffffffff);
    480		cp_ctx(ctx, 0x4030b8, 13);
    481		gr_def(ctx, 0x4030dc, 0x00000005);
    482		gr_def(ctx, 0x4030e8, 0x0000ffff);
    483		break;
    484	default:
    485		cp_ctx(ctx, 0x402c10, 4);
    486		if (device->chipset == 0x40)
    487			cp_ctx(ctx, 0x402c20, 36);
    488		else
    489		if (device->chipset <= 0x42)
    490			cp_ctx(ctx, 0x402c20, 24);
    491		else
    492		if (device->chipset <= 0x4a)
    493			cp_ctx(ctx, 0x402c20, 16);
    494		else
    495			cp_ctx(ctx, 0x402c20, 8);
    496		cp_ctx(ctx, 0x402cb0, device->chipset == 0x40 ? 12 : 13);
    497		gr_def(ctx, 0x402cd4, 0x00000005);
    498		if (device->chipset != 0x40)
    499			gr_def(ctx, 0x402ce0, 0x0000ffff);
    500		break;
    501	}
    502
    503	cp_ctx(ctx, 0x403400, device->chipset == 0x40 ? 4 : 3);
    504	cp_ctx(ctx, 0x403410, device->chipset == 0x40 ? 4 : 3);
    505	cp_ctx(ctx, 0x403420, nv40_gr_vs_count(ctx->device));
    506	for (i = 0; i < nv40_gr_vs_count(ctx->device); i++)
    507		gr_def(ctx, 0x403420 + (i * 4), 0x00005555);
    508
    509	if (device->chipset != 0x40) {
    510		cp_ctx(ctx, 0x403600, 1);
    511		gr_def(ctx, 0x403600, 0x00000001);
    512	}
    513	cp_ctx(ctx, 0x403800, 1);
    514
    515	cp_ctx(ctx, 0x403c18, 1);
    516	gr_def(ctx, 0x403c18, 0x00000001);
    517	switch (device->chipset) {
    518	case 0x46:
    519	case 0x47:
    520	case 0x49:
    521	case 0x4b:
    522		cp_ctx(ctx, 0x405018, 1);
    523		gr_def(ctx, 0x405018, 0x08e00001);
    524		cp_ctx(ctx, 0x405c24, 1);
    525		gr_def(ctx, 0x405c24, 0x000e3000);
    526		break;
    527	}
    528	if (device->chipset != 0x4e)
    529		cp_ctx(ctx, 0x405800, 11);
    530	cp_ctx(ctx, 0x407000, 1);
    531}
    532
    533static void
    534nv40_gr_construct_state3d_3(struct nvkm_grctx *ctx)
    535{
    536	int len = nv44_gr_class(ctx->device) ? 0x0084 : 0x0684;
    537
    538	cp_out (ctx, 0x300000);
    539	cp_lsr (ctx, len - 4);
    540	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_swap_state3d_3_is_save);
    541	cp_lsr (ctx, len);
    542	cp_name(ctx, cp_swap_state3d_3_is_save);
    543	cp_out (ctx, 0x800001);
    544
    545	ctx->ctxvals_pos += len;
    546}
    547
    548static void
    549nv40_gr_construct_shader(struct nvkm_grctx *ctx)
    550{
    551	struct nvkm_device *device = ctx->device;
    552	struct nvkm_gpuobj *obj = ctx->data;
    553	int vs, vs_nr, vs_len, vs_nr_b0, vs_nr_b1, b0_offset, b1_offset;
    554	int offset, i;
    555
    556	vs_nr    = nv40_gr_vs_count(ctx->device);
    557	vs_nr_b0 = 363;
    558	vs_nr_b1 = device->chipset == 0x40 ? 128 : 64;
    559	if (device->chipset == 0x40) {
    560		b0_offset = 0x2200/4; /* 33a0 */
    561		b1_offset = 0x55a0/4; /* 1500 */
    562		vs_len = 0x6aa0/4;
    563	} else
    564	if (device->chipset == 0x41 || device->chipset == 0x42) {
    565		b0_offset = 0x2200/4; /* 2200 */
    566		b1_offset = 0x4400/4; /* 0b00 */
    567		vs_len = 0x4f00/4;
    568	} else {
    569		b0_offset = 0x1d40/4; /* 2200 */
    570		b1_offset = 0x3f40/4; /* 0b00 : 0a40 */
    571		vs_len = nv44_gr_class(device) ? 0x4980/4 : 0x4a40/4;
    572	}
    573
    574	cp_lsr(ctx, vs_len * vs_nr + 0x300/4);
    575	cp_out(ctx, nv44_gr_class(device) ? 0x800029 : 0x800041);
    576
    577	offset = ctx->ctxvals_pos;
    578	ctx->ctxvals_pos += (0x0300/4 + (vs_nr * vs_len));
    579
    580	if (ctx->mode != NVKM_GRCTX_VALS)
    581		return;
    582
    583	offset += 0x0280/4;
    584	for (i = 0; i < 16; i++, offset += 2)
    585		nvkm_wo32(obj, offset * 4, 0x3f800000);
    586
    587	for (vs = 0; vs < vs_nr; vs++, offset += vs_len) {
    588		for (i = 0; i < vs_nr_b0 * 6; i += 6)
    589			nvkm_wo32(obj, (offset + b0_offset + i) * 4, 0x00000001);
    590		for (i = 0; i < vs_nr_b1 * 4; i += 4)
    591			nvkm_wo32(obj, (offset + b1_offset + i) * 4, 0x3f800000);
    592	}
    593}
    594
    595static void
    596nv40_grctx_generate(struct nvkm_grctx *ctx)
    597{
    598	/* decide whether we're loading/unloading the context */
    599	cp_bra (ctx, AUTO_SAVE, PENDING, cp_setup_save);
    600	cp_bra (ctx, USER_SAVE, PENDING, cp_setup_save);
    601
    602	cp_name(ctx, cp_check_load);
    603	cp_bra (ctx, AUTO_LOAD, PENDING, cp_setup_auto_load);
    604	cp_bra (ctx, USER_LOAD, PENDING, cp_setup_load);
    605	cp_bra (ctx, ALWAYS, TRUE, cp_exit);
    606
    607	/* setup for context load */
    608	cp_name(ctx, cp_setup_auto_load);
    609	cp_wait(ctx, STATUS, IDLE);
    610	cp_out (ctx, CP_NEXT_TO_SWAP);
    611	cp_name(ctx, cp_setup_load);
    612	cp_wait(ctx, STATUS, IDLE);
    613	cp_set (ctx, SWAP_DIRECTION, LOAD);
    614	cp_out (ctx, 0x00910880); /* ?? */
    615	cp_out (ctx, 0x00901ffe); /* ?? */
    616	cp_out (ctx, 0x01940000); /* ?? */
    617	cp_lsr (ctx, 0x20);
    618	cp_out (ctx, 0x0060000b); /* ?? */
    619	cp_wait(ctx, UNK57, CLEAR);
    620	cp_out (ctx, 0x0060000c); /* ?? */
    621	cp_bra (ctx, ALWAYS, TRUE, cp_swap_state);
    622
    623	/* setup for context save */
    624	cp_name(ctx, cp_setup_save);
    625	cp_set (ctx, SWAP_DIRECTION, SAVE);
    626
    627	/* general PGRAPH state */
    628	cp_name(ctx, cp_swap_state);
    629	cp_pos (ctx, 0x00020/4);
    630	nv40_gr_construct_general(ctx);
    631	cp_wait(ctx, STATUS, IDLE);
    632
    633	/* 3D state, block 1 */
    634	cp_bra (ctx, UNK54, CLEAR, cp_prepare_exit);
    635	nv40_gr_construct_state3d(ctx);
    636	cp_wait(ctx, STATUS, IDLE);
    637
    638	/* 3D state, block 2 */
    639	nv40_gr_construct_state3d_2(ctx);
    640
    641	/* Some other block of "random" state */
    642	nv40_gr_construct_state3d_3(ctx);
    643
    644	/* Per-vertex shader state */
    645	cp_pos (ctx, ctx->ctxvals_pos);
    646	nv40_gr_construct_shader(ctx);
    647
    648	/* pre-exit state updates */
    649	cp_name(ctx, cp_prepare_exit);
    650	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_check_load);
    651	cp_bra (ctx, USER_SAVE, PENDING, cp_exit);
    652	cp_out (ctx, CP_NEXT_TO_CURRENT);
    653
    654	cp_name(ctx, cp_exit);
    655	cp_set (ctx, USER_SAVE, NOT_PENDING);
    656	cp_set (ctx, USER_LOAD, NOT_PENDING);
    657	cp_out (ctx, CP_END);
    658}
    659
    660void
    661nv40_grctx_fill(struct nvkm_device *device, struct nvkm_gpuobj *mem)
    662{
    663	nv40_grctx_generate(&(struct nvkm_grctx) {
    664			     .device = device,
    665			     .mode = NVKM_GRCTX_VALS,
    666			     .data = mem,
    667			   });
    668}
    669
    670int
    671nv40_grctx_init(struct nvkm_device *device, u32 *size)
    672{
    673	u32 *ctxprog = kmalloc(256 * 4, GFP_KERNEL), i;
    674	struct nvkm_grctx ctx = {
    675		.device = device,
    676		.mode = NVKM_GRCTX_PROG,
    677		.ucode = ctxprog,
    678		.ctxprog_max = 256,
    679	};
    680
    681	if (!ctxprog)
    682		return -ENOMEM;
    683
    684	nv40_grctx_generate(&ctx);
    685
    686	nvkm_wr32(device, 0x400324, 0);
    687	for (i = 0; i < ctx.ctxprog_len; i++)
    688		nvkm_wr32(device, 0x400328, ctxprog[i]);
    689	*size = ctx.ctxvals_pos * 4;
    690
    691	kfree(ctxprog);
    692	return 0;
    693}