7dffc57fa2291dab04c9a4cb56ecb6779b9ed63c
[linux-2.6.git] / drivers / gpu / drm / radeon / ni.c
1 /*
2  * Copyright 2010 Advanced Micro Devices, 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: Alex Deucher
23  */
24 #include <linux/firmware.h>
25 #include <linux/platform_device.h>
26 #include <linux/slab.h>
27 #include <linux/module.h>
28 #include "drmP.h"
29 #include "radeon.h"
30 #include "radeon_asic.h"
31 #include "radeon_drm.h"
32 #include "nid.h"
33 #include "atom.h"
34 #include "ni_reg.h"
35 #include "cayman_blit_shaders.h"
36
37 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
38 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
39 extern int evergreen_mc_wait_for_idle(struct radeon_device *rdev);
40 extern void evergreen_mc_program(struct radeon_device *rdev);
41 extern void evergreen_irq_suspend(struct radeon_device *rdev);
42 extern int evergreen_mc_init(struct radeon_device *rdev);
43 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
44 extern void evergreen_pcie_gen2_enable(struct radeon_device *rdev);
45 extern void si_rlc_fini(struct radeon_device *rdev);
46 extern int si_rlc_init(struct radeon_device *rdev);
47
48 #define EVERGREEN_PFP_UCODE_SIZE 1120
49 #define EVERGREEN_PM4_UCODE_SIZE 1376
50 #define EVERGREEN_RLC_UCODE_SIZE 768
51 #define BTC_MC_UCODE_SIZE 6024
52
53 #define CAYMAN_PFP_UCODE_SIZE 2176
54 #define CAYMAN_PM4_UCODE_SIZE 2176
55 #define CAYMAN_RLC_UCODE_SIZE 1024
56 #define CAYMAN_MC_UCODE_SIZE 6037
57
58 #define ARUBA_RLC_UCODE_SIZE 1536
59
60 /* Firmware Names */
61 MODULE_FIRMWARE("radeon/BARTS_pfp.bin");
62 MODULE_FIRMWARE("radeon/BARTS_me.bin");
63 MODULE_FIRMWARE("radeon/BARTS_mc.bin");
64 MODULE_FIRMWARE("radeon/BTC_rlc.bin");
65 MODULE_FIRMWARE("radeon/TURKS_pfp.bin");
66 MODULE_FIRMWARE("radeon/TURKS_me.bin");
67 MODULE_FIRMWARE("radeon/TURKS_mc.bin");
68 MODULE_FIRMWARE("radeon/CAICOS_pfp.bin");
69 MODULE_FIRMWARE("radeon/CAICOS_me.bin");
70 MODULE_FIRMWARE("radeon/CAICOS_mc.bin");
71 MODULE_FIRMWARE("radeon/CAYMAN_pfp.bin");
72 MODULE_FIRMWARE("radeon/CAYMAN_me.bin");
73 MODULE_FIRMWARE("radeon/CAYMAN_mc.bin");
74 MODULE_FIRMWARE("radeon/CAYMAN_rlc.bin");
75 MODULE_FIRMWARE("radeon/ARUBA_pfp.bin");
76 MODULE_FIRMWARE("radeon/ARUBA_me.bin");
77 MODULE_FIRMWARE("radeon/ARUBA_rlc.bin");
78
79 #define BTC_IO_MC_REGS_SIZE 29
80
81 static const u32 barts_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
82         {0x00000077, 0xff010100},
83         {0x00000078, 0x00000000},
84         {0x00000079, 0x00001434},
85         {0x0000007a, 0xcc08ec08},
86         {0x0000007b, 0x00040000},
87         {0x0000007c, 0x000080c0},
88         {0x0000007d, 0x09000000},
89         {0x0000007e, 0x00210404},
90         {0x00000081, 0x08a8e800},
91         {0x00000082, 0x00030444},
92         {0x00000083, 0x00000000},
93         {0x00000085, 0x00000001},
94         {0x00000086, 0x00000002},
95         {0x00000087, 0x48490000},
96         {0x00000088, 0x20244647},
97         {0x00000089, 0x00000005},
98         {0x0000008b, 0x66030000},
99         {0x0000008c, 0x00006603},
100         {0x0000008d, 0x00000100},
101         {0x0000008f, 0x00001c0a},
102         {0x00000090, 0xff000001},
103         {0x00000094, 0x00101101},
104         {0x00000095, 0x00000fff},
105         {0x00000096, 0x00116fff},
106         {0x00000097, 0x60010000},
107         {0x00000098, 0x10010000},
108         {0x00000099, 0x00006000},
109         {0x0000009a, 0x00001000},
110         {0x0000009f, 0x00946a00}
111 };
112
113 static const u32 turks_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
114         {0x00000077, 0xff010100},
115         {0x00000078, 0x00000000},
116         {0x00000079, 0x00001434},
117         {0x0000007a, 0xcc08ec08},
118         {0x0000007b, 0x00040000},
119         {0x0000007c, 0x000080c0},
120         {0x0000007d, 0x09000000},
121         {0x0000007e, 0x00210404},
122         {0x00000081, 0x08a8e800},
123         {0x00000082, 0x00030444},
124         {0x00000083, 0x00000000},
125         {0x00000085, 0x00000001},
126         {0x00000086, 0x00000002},
127         {0x00000087, 0x48490000},
128         {0x00000088, 0x20244647},
129         {0x00000089, 0x00000005},
130         {0x0000008b, 0x66030000},
131         {0x0000008c, 0x00006603},
132         {0x0000008d, 0x00000100},
133         {0x0000008f, 0x00001c0a},
134         {0x00000090, 0xff000001},
135         {0x00000094, 0x00101101},
136         {0x00000095, 0x00000fff},
137         {0x00000096, 0x00116fff},
138         {0x00000097, 0x60010000},
139         {0x00000098, 0x10010000},
140         {0x00000099, 0x00006000},
141         {0x0000009a, 0x00001000},
142         {0x0000009f, 0x00936a00}
143 };
144
145 static const u32 caicos_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
146         {0x00000077, 0xff010100},
147         {0x00000078, 0x00000000},
148         {0x00000079, 0x00001434},
149         {0x0000007a, 0xcc08ec08},
150         {0x0000007b, 0x00040000},
151         {0x0000007c, 0x000080c0},
152         {0x0000007d, 0x09000000},
153         {0x0000007e, 0x00210404},
154         {0x00000081, 0x08a8e800},
155         {0x00000082, 0x00030444},
156         {0x00000083, 0x00000000},
157         {0x00000085, 0x00000001},
158         {0x00000086, 0x00000002},
159         {0x00000087, 0x48490000},
160         {0x00000088, 0x20244647},
161         {0x00000089, 0x00000005},
162         {0x0000008b, 0x66030000},
163         {0x0000008c, 0x00006603},
164         {0x0000008d, 0x00000100},
165         {0x0000008f, 0x00001c0a},
166         {0x00000090, 0xff000001},
167         {0x00000094, 0x00101101},
168         {0x00000095, 0x00000fff},
169         {0x00000096, 0x00116fff},
170         {0x00000097, 0x60010000},
171         {0x00000098, 0x10010000},
172         {0x00000099, 0x00006000},
173         {0x0000009a, 0x00001000},
174         {0x0000009f, 0x00916a00}
175 };
176
177 static const u32 cayman_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = {
178         {0x00000077, 0xff010100},
179         {0x00000078, 0x00000000},
180         {0x00000079, 0x00001434},
181         {0x0000007a, 0xcc08ec08},
182         {0x0000007b, 0x00040000},
183         {0x0000007c, 0x000080c0},
184         {0x0000007d, 0x09000000},
185         {0x0000007e, 0x00210404},
186         {0x00000081, 0x08a8e800},
187         {0x00000082, 0x00030444},
188         {0x00000083, 0x00000000},
189         {0x00000085, 0x00000001},
190         {0x00000086, 0x00000002},
191         {0x00000087, 0x48490000},
192         {0x00000088, 0x20244647},
193         {0x00000089, 0x00000005},
194         {0x0000008b, 0x66030000},
195         {0x0000008c, 0x00006603},
196         {0x0000008d, 0x00000100},
197         {0x0000008f, 0x00001c0a},
198         {0x00000090, 0xff000001},
199         {0x00000094, 0x00101101},
200         {0x00000095, 0x00000fff},
201         {0x00000096, 0x00116fff},
202         {0x00000097, 0x60010000},
203         {0x00000098, 0x10010000},
204         {0x00000099, 0x00006000},
205         {0x0000009a, 0x00001000},
206         {0x0000009f, 0x00976b00}
207 };
208
209 int ni_mc_load_microcode(struct radeon_device *rdev)
210 {
211         const __be32 *fw_data;
212         u32 mem_type, running, blackout = 0;
213         u32 *io_mc_regs;
214         int i, ucode_size, regs_size;
215
216         if (!rdev->mc_fw)
217                 return -EINVAL;
218
219         switch (rdev->family) {
220         case CHIP_BARTS:
221                 io_mc_regs = (u32 *)&barts_io_mc_regs;
222                 ucode_size = BTC_MC_UCODE_SIZE;
223                 regs_size = BTC_IO_MC_REGS_SIZE;
224                 break;
225         case CHIP_TURKS:
226                 io_mc_regs = (u32 *)&turks_io_mc_regs;
227                 ucode_size = BTC_MC_UCODE_SIZE;
228                 regs_size = BTC_IO_MC_REGS_SIZE;
229                 break;
230         case CHIP_CAICOS:
231         default:
232                 io_mc_regs = (u32 *)&caicos_io_mc_regs;
233                 ucode_size = BTC_MC_UCODE_SIZE;
234                 regs_size = BTC_IO_MC_REGS_SIZE;
235                 break;
236         case CHIP_CAYMAN:
237                 io_mc_regs = (u32 *)&cayman_io_mc_regs;
238                 ucode_size = CAYMAN_MC_UCODE_SIZE;
239                 regs_size = BTC_IO_MC_REGS_SIZE;
240                 break;
241         }
242
243         mem_type = (RREG32(MC_SEQ_MISC0) & MC_SEQ_MISC0_GDDR5_MASK) >> MC_SEQ_MISC0_GDDR5_SHIFT;
244         running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
245
246         if ((mem_type == MC_SEQ_MISC0_GDDR5_VALUE) && (running == 0)) {
247                 if (running) {
248                         blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
249                         WREG32(MC_SHARED_BLACKOUT_CNTL, 1);
250                 }
251
252                 /* reset the engine and set to writable */
253                 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
254                 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
255
256                 /* load mc io regs */
257                 for (i = 0; i < regs_size; i++) {
258                         WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
259                         WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
260                 }
261                 /* load the MC ucode */
262                 fw_data = (const __be32 *)rdev->mc_fw->data;
263                 for (i = 0; i < ucode_size; i++)
264                         WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
265
266                 /* put the engine back into the active state */
267                 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
268                 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
269                 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
270
271                 /* wait for training to complete */
272                 for (i = 0; i < rdev->usec_timeout; i++) {
273                         if (RREG32(MC_IO_PAD_CNTL_D0) & MEM_FALL_OUT_CMD)
274                                 break;
275                         udelay(1);
276                 }
277
278                 if (running)
279                         WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
280         }
281
282         return 0;
283 }
284
285 int ni_init_microcode(struct radeon_device *rdev)
286 {
287         struct platform_device *pdev;
288         const char *chip_name;
289         const char *rlc_chip_name;
290         size_t pfp_req_size, me_req_size, rlc_req_size, mc_req_size;
291         char fw_name[30];
292         int err;
293
294         DRM_DEBUG("\n");
295
296         pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
297         err = IS_ERR(pdev);
298         if (err) {
299                 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
300                 return -EINVAL;
301         }
302
303         switch (rdev->family) {
304         case CHIP_BARTS:
305                 chip_name = "BARTS";
306                 rlc_chip_name = "BTC";
307                 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
308                 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
309                 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
310                 mc_req_size = BTC_MC_UCODE_SIZE * 4;
311                 break;
312         case CHIP_TURKS:
313                 chip_name = "TURKS";
314                 rlc_chip_name = "BTC";
315                 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
316                 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
317                 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
318                 mc_req_size = BTC_MC_UCODE_SIZE * 4;
319                 break;
320         case CHIP_CAICOS:
321                 chip_name = "CAICOS";
322                 rlc_chip_name = "BTC";
323                 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4;
324                 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4;
325                 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4;
326                 mc_req_size = BTC_MC_UCODE_SIZE * 4;
327                 break;
328         case CHIP_CAYMAN:
329                 chip_name = "CAYMAN";
330                 rlc_chip_name = "CAYMAN";
331                 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4;
332                 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4;
333                 rlc_req_size = CAYMAN_RLC_UCODE_SIZE * 4;
334                 mc_req_size = CAYMAN_MC_UCODE_SIZE * 4;
335                 break;
336         case CHIP_ARUBA:
337                 chip_name = "ARUBA";
338                 rlc_chip_name = "ARUBA";
339                 /* pfp/me same size as CAYMAN */
340                 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4;
341                 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4;
342                 rlc_req_size = ARUBA_RLC_UCODE_SIZE * 4;
343                 mc_req_size = 0;
344                 break;
345         default: BUG();
346         }
347
348         DRM_INFO("Loading %s Microcode\n", chip_name);
349
350         snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
351         err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
352         if (err)
353                 goto out;
354         if (rdev->pfp_fw->size != pfp_req_size) {
355                 printk(KERN_ERR
356                        "ni_cp: Bogus length %zu in firmware \"%s\"\n",
357                        rdev->pfp_fw->size, fw_name);
358                 err = -EINVAL;
359                 goto out;
360         }
361
362         snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
363         err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
364         if (err)
365                 goto out;
366         if (rdev->me_fw->size != me_req_size) {
367                 printk(KERN_ERR
368                        "ni_cp: Bogus length %zu in firmware \"%s\"\n",
369                        rdev->me_fw->size, fw_name);
370                 err = -EINVAL;
371         }
372
373         snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
374         err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
375         if (err)
376                 goto out;
377         if (rdev->rlc_fw->size != rlc_req_size) {
378                 printk(KERN_ERR
379                        "ni_rlc: Bogus length %zu in firmware \"%s\"\n",
380                        rdev->rlc_fw->size, fw_name);
381                 err = -EINVAL;
382         }
383
384         /* no MC ucode on TN */
385         if (!(rdev->flags & RADEON_IS_IGP)) {
386                 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
387                 err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev);
388                 if (err)
389                         goto out;
390                 if (rdev->mc_fw->size != mc_req_size) {
391                         printk(KERN_ERR
392                                "ni_mc: Bogus length %zu in firmware \"%s\"\n",
393                                rdev->mc_fw->size, fw_name);
394                         err = -EINVAL;
395                 }
396         }
397 out:
398         platform_device_unregister(pdev);
399
400         if (err) {
401                 if (err != -EINVAL)
402                         printk(KERN_ERR
403                                "ni_cp: Failed to load firmware \"%s\"\n",
404                                fw_name);
405                 release_firmware(rdev->pfp_fw);
406                 rdev->pfp_fw = NULL;
407                 release_firmware(rdev->me_fw);
408                 rdev->me_fw = NULL;
409                 release_firmware(rdev->rlc_fw);
410                 rdev->rlc_fw = NULL;
411                 release_firmware(rdev->mc_fw);
412                 rdev->mc_fw = NULL;
413         }
414         return err;
415 }
416
417 /*
418  * Core functions
419  */
420 static u32 cayman_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
421                                                u32 num_tile_pipes,
422                                                u32 num_backends_per_asic,
423                                                u32 *backend_disable_mask_per_asic,
424                                                u32 num_shader_engines)
425 {
426         u32 backend_map = 0;
427         u32 enabled_backends_mask = 0;
428         u32 enabled_backends_count = 0;
429         u32 num_backends_per_se;
430         u32 cur_pipe;
431         u32 swizzle_pipe[CAYMAN_MAX_PIPES];
432         u32 cur_backend = 0;
433         u32 i;
434         bool force_no_swizzle;
435
436         /* force legal values */
437         if (num_tile_pipes < 1)
438                 num_tile_pipes = 1;
439         if (num_tile_pipes > rdev->config.cayman.max_tile_pipes)
440                 num_tile_pipes = rdev->config.cayman.max_tile_pipes;
441         if (num_shader_engines < 1)
442                 num_shader_engines = 1;
443         if (num_shader_engines > rdev->config.cayman.max_shader_engines)
444                 num_shader_engines = rdev->config.cayman.max_shader_engines;
445         if (num_backends_per_asic < num_shader_engines)
446                 num_backends_per_asic = num_shader_engines;
447         if (num_backends_per_asic > (rdev->config.cayman.max_backends_per_se * num_shader_engines))
448                 num_backends_per_asic = rdev->config.cayman.max_backends_per_se * num_shader_engines;
449
450         /* make sure we have the same number of backends per se */
451         num_backends_per_asic = ALIGN(num_backends_per_asic, num_shader_engines);
452         /* set up the number of backends per se */
453         num_backends_per_se = num_backends_per_asic / num_shader_engines;
454         if (num_backends_per_se > rdev->config.cayman.max_backends_per_se) {
455                 num_backends_per_se = rdev->config.cayman.max_backends_per_se;
456                 num_backends_per_asic = num_backends_per_se * num_shader_engines;
457         }
458
459         /* create enable mask and count for enabled backends */
460         for (i = 0; i < CAYMAN_MAX_BACKENDS; ++i) {
461                 if (((*backend_disable_mask_per_asic >> i) & 1) == 0) {
462                         enabled_backends_mask |= (1 << i);
463                         ++enabled_backends_count;
464                 }
465                 if (enabled_backends_count == num_backends_per_asic)
466                         break;
467         }
468
469         /* force the backends mask to match the current number of backends */
470         if (enabled_backends_count != num_backends_per_asic) {
471                 u32 this_backend_enabled;
472                 u32 shader_engine;
473                 u32 backend_per_se;
474
475                 enabled_backends_mask = 0;
476                 enabled_backends_count = 0;
477                 *backend_disable_mask_per_asic = CAYMAN_MAX_BACKENDS_MASK;
478                 for (i = 0; i < CAYMAN_MAX_BACKENDS; ++i) {
479                         /* calc the current se */
480                         shader_engine = i / rdev->config.cayman.max_backends_per_se;
481                         /* calc the backend per se */
482                         backend_per_se = i % rdev->config.cayman.max_backends_per_se;
483                         /* default to not enabled */
484                         this_backend_enabled = 0;
485                         if ((shader_engine < num_shader_engines) &&
486                             (backend_per_se < num_backends_per_se))
487                                 this_backend_enabled = 1;
488                         if (this_backend_enabled) {
489                                 enabled_backends_mask |= (1 << i);
490                                 *backend_disable_mask_per_asic &= ~(1 << i);
491                                 ++enabled_backends_count;
492                         }
493                 }
494         }
495
496
497         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * CAYMAN_MAX_PIPES);
498         switch (rdev->family) {
499         case CHIP_CAYMAN:
500         case CHIP_ARUBA:
501                 force_no_swizzle = true;
502                 break;
503         default:
504                 force_no_swizzle = false;
505                 break;
506         }
507         if (force_no_swizzle) {
508                 bool last_backend_enabled = false;
509
510                 force_no_swizzle = false;
511                 for (i = 0; i < CAYMAN_MAX_BACKENDS; ++i) {
512                         if (((enabled_backends_mask >> i) & 1) == 1) {
513                                 if (last_backend_enabled)
514                                         force_no_swizzle = true;
515                                 last_backend_enabled = true;
516                         } else
517                                 last_backend_enabled = false;
518                 }
519         }
520
521         switch (num_tile_pipes) {
522         case 1:
523         case 3:
524         case 5:
525         case 7:
526                 DRM_ERROR("odd number of pipes!\n");
527                 break;
528         case 2:
529                 swizzle_pipe[0] = 0;
530                 swizzle_pipe[1] = 1;
531                 break;
532         case 4:
533                 if (force_no_swizzle) {
534                         swizzle_pipe[0] = 0;
535                         swizzle_pipe[1] = 1;
536                         swizzle_pipe[2] = 2;
537                         swizzle_pipe[3] = 3;
538                 } else {
539                         swizzle_pipe[0] = 0;
540                         swizzle_pipe[1] = 2;
541                         swizzle_pipe[2] = 1;
542                         swizzle_pipe[3] = 3;
543                 }
544                 break;
545         case 6:
546                 if (force_no_swizzle) {
547                         swizzle_pipe[0] = 0;
548                         swizzle_pipe[1] = 1;
549                         swizzle_pipe[2] = 2;
550                         swizzle_pipe[3] = 3;
551                         swizzle_pipe[4] = 4;
552                         swizzle_pipe[5] = 5;
553                 } else {
554                         swizzle_pipe[0] = 0;
555                         swizzle_pipe[1] = 2;
556                         swizzle_pipe[2] = 4;
557                         swizzle_pipe[3] = 1;
558                         swizzle_pipe[4] = 3;
559                         swizzle_pipe[5] = 5;
560                 }
561                 break;
562         case 8:
563                 if (force_no_swizzle) {
564                         swizzle_pipe[0] = 0;
565                         swizzle_pipe[1] = 1;
566                         swizzle_pipe[2] = 2;
567                         swizzle_pipe[3] = 3;
568                         swizzle_pipe[4] = 4;
569                         swizzle_pipe[5] = 5;
570                         swizzle_pipe[6] = 6;
571                         swizzle_pipe[7] = 7;
572                 } else {
573                         swizzle_pipe[0] = 0;
574                         swizzle_pipe[1] = 2;
575                         swizzle_pipe[2] = 4;
576                         swizzle_pipe[3] = 6;
577                         swizzle_pipe[4] = 1;
578                         swizzle_pipe[5] = 3;
579                         swizzle_pipe[6] = 5;
580                         swizzle_pipe[7] = 7;
581                 }
582                 break;
583         }
584
585         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
586                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
587                         cur_backend = (cur_backend + 1) % CAYMAN_MAX_BACKENDS;
588
589                 backend_map |= (((cur_backend & 0xf) << (swizzle_pipe[cur_pipe] * 4)));
590
591                 cur_backend = (cur_backend + 1) % CAYMAN_MAX_BACKENDS;
592         }
593
594         return backend_map;
595 }
596
597 static u32 cayman_get_disable_mask_per_asic(struct radeon_device *rdev,
598                                             u32 disable_mask_per_se,
599                                             u32 max_disable_mask_per_se,
600                                             u32 num_shader_engines)
601 {
602         u32 disable_field_width_per_se = r600_count_pipe_bits(disable_mask_per_se);
603         u32 disable_mask_per_asic = disable_mask_per_se & max_disable_mask_per_se;
604
605         if (num_shader_engines == 1)
606                 return disable_mask_per_asic;
607         else if (num_shader_engines == 2)
608                 return disable_mask_per_asic | (disable_mask_per_asic << disable_field_width_per_se);
609         else
610                 return 0xffffffff;
611 }
612
613 static void cayman_gpu_init(struct radeon_device *rdev)
614 {
615         u32 cc_rb_backend_disable = 0;
616         u32 cc_gc_shader_pipe_config;
617         u32 gb_addr_config = 0;
618         u32 mc_shared_chmap, mc_arb_ramcfg;
619         u32 gb_backend_map;
620         u32 cgts_tcc_disable;
621         u32 sx_debug_1;
622         u32 smx_dc_ctl0;
623         u32 gc_user_shader_pipe_config;
624         u32 gc_user_rb_backend_disable;
625         u32 cgts_user_tcc_disable;
626         u32 cgts_sm_ctrl_reg;
627         u32 hdp_host_path_cntl;
628         u32 tmp;
629         int i, j;
630
631         switch (rdev->family) {
632         case CHIP_CAYMAN:
633                 rdev->config.cayman.max_shader_engines = 2;
634                 rdev->config.cayman.max_pipes_per_simd = 4;
635                 rdev->config.cayman.max_tile_pipes = 8;
636                 rdev->config.cayman.max_simds_per_se = 12;
637                 rdev->config.cayman.max_backends_per_se = 4;
638                 rdev->config.cayman.max_texture_channel_caches = 8;
639                 rdev->config.cayman.max_gprs = 256;
640                 rdev->config.cayman.max_threads = 256;
641                 rdev->config.cayman.max_gs_threads = 32;
642                 rdev->config.cayman.max_stack_entries = 512;
643                 rdev->config.cayman.sx_num_of_sets = 8;
644                 rdev->config.cayman.sx_max_export_size = 256;
645                 rdev->config.cayman.sx_max_export_pos_size = 64;
646                 rdev->config.cayman.sx_max_export_smx_size = 192;
647                 rdev->config.cayman.max_hw_contexts = 8;
648                 rdev->config.cayman.sq_num_cf_insts = 2;
649
650                 rdev->config.cayman.sc_prim_fifo_size = 0x100;
651                 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30;
652                 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130;
653                 break;
654         case CHIP_ARUBA:
655         default:
656                 rdev->config.cayman.max_shader_engines = 1;
657                 rdev->config.cayman.max_pipes_per_simd = 4;
658                 rdev->config.cayman.max_tile_pipes = 2;
659                 if ((rdev->pdev->device == 0x9900) ||
660                     (rdev->pdev->device == 0x9901) ||
661                     (rdev->pdev->device == 0x9905) ||
662                     (rdev->pdev->device == 0x9906) ||
663                     (rdev->pdev->device == 0x9907) ||
664                     (rdev->pdev->device == 0x9908) ||
665                     (rdev->pdev->device == 0x9909) ||
666                     (rdev->pdev->device == 0x990B) ||
667                     (rdev->pdev->device == 0x990C) ||
668                     (rdev->pdev->device == 0x990F) ||
669                     (rdev->pdev->device == 0x9910) ||
670                     (rdev->pdev->device == 0x9917) ||
671                     (rdev->pdev->device == 0x9999)) {
672                         rdev->config.cayman.max_simds_per_se = 6;
673                         rdev->config.cayman.max_backends_per_se = 2;
674                 } else if ((rdev->pdev->device == 0x9903) ||
675                            (rdev->pdev->device == 0x9904) ||
676                            (rdev->pdev->device == 0x990A) ||
677                            (rdev->pdev->device == 0x990D) ||
678                            (rdev->pdev->device == 0x990E) ||
679                            (rdev->pdev->device == 0x9913) ||
680                            (rdev->pdev->device == 0x9918)) {
681                         rdev->config.cayman.max_simds_per_se = 4;
682                         rdev->config.cayman.max_backends_per_se = 2;
683                 } else if ((rdev->pdev->device == 0x9919) ||
684                            (rdev->pdev->device == 0x9990) ||
685                            (rdev->pdev->device == 0x9991) ||
686                            (rdev->pdev->device == 0x9994) ||
687                            (rdev->pdev->device == 0x9995) ||
688                            (rdev->pdev->device == 0x9996) ||
689                            (rdev->pdev->device == 0x999A) ||
690                            (rdev->pdev->device == 0x99A0)) {
691                         rdev->config.cayman.max_simds_per_se = 3;
692                         rdev->config.cayman.max_backends_per_se = 1;
693                 } else {
694                         rdev->config.cayman.max_simds_per_se = 2;
695                         rdev->config.cayman.max_backends_per_se = 1;
696                 }
697                 rdev->config.cayman.max_texture_channel_caches = 2;
698                 rdev->config.cayman.max_gprs = 256;
699                 rdev->config.cayman.max_threads = 256;
700                 rdev->config.cayman.max_gs_threads = 32;
701                 rdev->config.cayman.max_stack_entries = 512;
702                 rdev->config.cayman.sx_num_of_sets = 8;
703                 rdev->config.cayman.sx_max_export_size = 256;
704                 rdev->config.cayman.sx_max_export_pos_size = 64;
705                 rdev->config.cayman.sx_max_export_smx_size = 192;
706                 rdev->config.cayman.max_hw_contexts = 8;
707                 rdev->config.cayman.sq_num_cf_insts = 2;
708
709                 rdev->config.cayman.sc_prim_fifo_size = 0x40;
710                 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30;
711                 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130;
712                 break;
713         }
714
715         /* Initialize HDP */
716         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
717                 WREG32((0x2c14 + j), 0x00000000);
718                 WREG32((0x2c18 + j), 0x00000000);
719                 WREG32((0x2c1c + j), 0x00000000);
720                 WREG32((0x2c20 + j), 0x00000000);
721                 WREG32((0x2c24 + j), 0x00000000);
722         }
723
724         WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
725
726         evergreen_fix_pci_max_read_req_size(rdev);
727
728         mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
729         mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
730
731         cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE);
732         cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG);
733         cgts_tcc_disable = 0xffff0000;
734         for (i = 0; i < rdev->config.cayman.max_texture_channel_caches; i++)
735                 cgts_tcc_disable &= ~(1 << (16 + i));
736         gc_user_rb_backend_disable = RREG32(GC_USER_RB_BACKEND_DISABLE);
737         gc_user_shader_pipe_config = RREG32(GC_USER_SHADER_PIPE_CONFIG);
738         cgts_user_tcc_disable = RREG32(CGTS_USER_TCC_DISABLE);
739
740         rdev->config.cayman.num_shader_engines = rdev->config.cayman.max_shader_engines;
741         tmp = ((~gc_user_shader_pipe_config) & INACTIVE_QD_PIPES_MASK) >> INACTIVE_QD_PIPES_SHIFT;
742         rdev->config.cayman.num_shader_pipes_per_simd = r600_count_pipe_bits(tmp);
743         rdev->config.cayman.num_tile_pipes = rdev->config.cayman.max_tile_pipes;
744         tmp = ((~gc_user_shader_pipe_config) & INACTIVE_SIMDS_MASK) >> INACTIVE_SIMDS_SHIFT;
745         rdev->config.cayman.num_simds_per_se = r600_count_pipe_bits(tmp);
746         tmp = ((~gc_user_rb_backend_disable) & BACKEND_DISABLE_MASK) >> BACKEND_DISABLE_SHIFT;
747         rdev->config.cayman.num_backends_per_se = r600_count_pipe_bits(tmp);
748         tmp = (gc_user_rb_backend_disable & BACKEND_DISABLE_MASK) >> BACKEND_DISABLE_SHIFT;
749         rdev->config.cayman.backend_disable_mask_per_asic =
750                 cayman_get_disable_mask_per_asic(rdev, tmp, CAYMAN_MAX_BACKENDS_PER_SE_MASK,
751                                                  rdev->config.cayman.num_shader_engines);
752         rdev->config.cayman.backend_map =
753                 cayman_get_tile_pipe_to_backend_map(rdev, rdev->config.cayman.num_tile_pipes,
754                                                     rdev->config.cayman.num_backends_per_se *
755                                                     rdev->config.cayman.num_shader_engines,
756                                                     &rdev->config.cayman.backend_disable_mask_per_asic,
757                                                     rdev->config.cayman.num_shader_engines);
758         tmp = ((~cgts_user_tcc_disable) & TCC_DISABLE_MASK) >> TCC_DISABLE_SHIFT;
759         rdev->config.cayman.num_texture_channel_caches = r600_count_pipe_bits(tmp);
760         tmp = (mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT;
761         rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256;
762         if (rdev->config.cayman.mem_max_burst_length_bytes > 512)
763                 rdev->config.cayman.mem_max_burst_length_bytes = 512;
764         tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
765         rdev->config.cayman.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
766         if (rdev->config.cayman.mem_row_size_in_kb > 4)
767                 rdev->config.cayman.mem_row_size_in_kb = 4;
768         /* XXX use MC settings? */
769         rdev->config.cayman.shader_engine_tile_size = 32;
770         rdev->config.cayman.num_gpus = 1;
771         rdev->config.cayman.multi_gpu_tile_size = 64;
772
773         //gb_addr_config = 0x02011003
774 #if 0
775         gb_addr_config = RREG32(GB_ADDR_CONFIG);
776 #else
777         gb_addr_config = 0;
778         switch (rdev->config.cayman.num_tile_pipes) {
779         case 1:
780         default:
781                 gb_addr_config |= NUM_PIPES(0);
782                 break;
783         case 2:
784                 gb_addr_config |= NUM_PIPES(1);
785                 break;
786         case 4:
787                 gb_addr_config |= NUM_PIPES(2);
788                 break;
789         case 8:
790                 gb_addr_config |= NUM_PIPES(3);
791                 break;
792         }
793
794         tmp = (rdev->config.cayman.mem_max_burst_length_bytes / 256) - 1;
795         gb_addr_config |= PIPE_INTERLEAVE_SIZE(tmp);
796         gb_addr_config |= NUM_SHADER_ENGINES(rdev->config.cayman.num_shader_engines - 1);
797         tmp = (rdev->config.cayman.shader_engine_tile_size / 16) - 1;
798         gb_addr_config |= SHADER_ENGINE_TILE_SIZE(tmp);
799         switch (rdev->config.cayman.num_gpus) {
800         case 1:
801         default:
802                 gb_addr_config |= NUM_GPUS(0);
803                 break;
804         case 2:
805                 gb_addr_config |= NUM_GPUS(1);
806                 break;
807         case 4:
808                 gb_addr_config |= NUM_GPUS(2);
809                 break;
810         }
811         switch (rdev->config.cayman.multi_gpu_tile_size) {
812         case 16:
813                 gb_addr_config |= MULTI_GPU_TILE_SIZE(0);
814                 break;
815         case 32:
816         default:
817                 gb_addr_config |= MULTI_GPU_TILE_SIZE(1);
818                 break;
819         case 64:
820                 gb_addr_config |= MULTI_GPU_TILE_SIZE(2);
821                 break;
822         case 128:
823                 gb_addr_config |= MULTI_GPU_TILE_SIZE(3);
824                 break;
825         }
826         switch (rdev->config.cayman.mem_row_size_in_kb) {
827         case 1:
828         default:
829                 gb_addr_config |= ROW_SIZE(0);
830                 break;
831         case 2:
832                 gb_addr_config |= ROW_SIZE(1);
833                 break;
834         case 4:
835                 gb_addr_config |= ROW_SIZE(2);
836                 break;
837         }
838 #endif
839
840         tmp = (gb_addr_config & NUM_PIPES_MASK) >> NUM_PIPES_SHIFT;
841         rdev->config.cayman.num_tile_pipes = (1 << tmp);
842         tmp = (gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT;
843         rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256;
844         tmp = (gb_addr_config & NUM_SHADER_ENGINES_MASK) >> NUM_SHADER_ENGINES_SHIFT;
845         rdev->config.cayman.num_shader_engines = tmp + 1;
846         tmp = (gb_addr_config & NUM_GPUS_MASK) >> NUM_GPUS_SHIFT;
847         rdev->config.cayman.num_gpus = tmp + 1;
848         tmp = (gb_addr_config & MULTI_GPU_TILE_SIZE_MASK) >> MULTI_GPU_TILE_SIZE_SHIFT;
849         rdev->config.cayman.multi_gpu_tile_size = 1 << tmp;
850         tmp = (gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT;
851         rdev->config.cayman.mem_row_size_in_kb = 1 << tmp;
852
853         //gb_backend_map = 0x76541032;
854 #if 0
855         gb_backend_map = RREG32(GB_BACKEND_MAP);
856 #else
857         gb_backend_map =
858                 cayman_get_tile_pipe_to_backend_map(rdev, rdev->config.cayman.num_tile_pipes,
859                                                     rdev->config.cayman.num_backends_per_se *
860                                                     rdev->config.cayman.num_shader_engines,
861                                                     &rdev->config.cayman.backend_disable_mask_per_asic,
862                                                     rdev->config.cayman.num_shader_engines);
863 #endif
864         /* setup tiling info dword.  gb_addr_config is not adequate since it does
865          * not have bank info, so create a custom tiling dword.
866          * bits 3:0   num_pipes
867          * bits 7:4   num_banks
868          * bits 11:8  group_size
869          * bits 15:12 row_size
870          */
871         rdev->config.cayman.tile_config = 0;
872         switch (rdev->config.cayman.num_tile_pipes) {
873         case 1:
874         default:
875                 rdev->config.cayman.tile_config |= (0 << 0);
876                 break;
877         case 2:
878                 rdev->config.cayman.tile_config |= (1 << 0);
879                 break;
880         case 4:
881                 rdev->config.cayman.tile_config |= (2 << 0);
882                 break;
883         case 8:
884                 rdev->config.cayman.tile_config |= (3 << 0);
885                 break;
886         }
887
888         /* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */
889         if (rdev->flags & RADEON_IS_IGP)
890                 rdev->config.cayman.tile_config |= 1 << 4;
891         else {
892                 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
893                 case 0: /* four banks */
894                         rdev->config.cayman.tile_config |= 0 << 4;
895                         break;
896                 case 1: /* eight banks */
897                         rdev->config.cayman.tile_config |= 1 << 4;
898                         break;
899                 case 2: /* sixteen banks */
900                 default:
901                         rdev->config.cayman.tile_config |= 2 << 4;
902                         break;
903                 }
904         }
905         rdev->config.cayman.tile_config |=
906                 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
907         rdev->config.cayman.tile_config |=
908                 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
909
910         rdev->config.cayman.backend_map = gb_backend_map;
911         WREG32(GB_BACKEND_MAP, gb_backend_map);
912         WREG32(GB_ADDR_CONFIG, gb_addr_config);
913         WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
914         WREG32(HDP_ADDR_CONFIG, gb_addr_config);
915
916         /* primary versions */
917         WREG32(CC_RB_BACKEND_DISABLE, cc_rb_backend_disable);
918         WREG32(CC_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable);
919         WREG32(CC_GC_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
920
921         WREG32(CGTS_TCC_DISABLE, cgts_tcc_disable);
922         WREG32(CGTS_SYS_TCC_DISABLE, cgts_tcc_disable);
923
924         /* user versions */
925         WREG32(GC_USER_RB_BACKEND_DISABLE, cc_rb_backend_disable);
926         WREG32(GC_USER_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable);
927         WREG32(GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
928
929         WREG32(CGTS_USER_SYS_TCC_DISABLE, cgts_tcc_disable);
930         WREG32(CGTS_USER_TCC_DISABLE, cgts_tcc_disable);
931
932         /* reprogram the shader complex */
933         cgts_sm_ctrl_reg = RREG32(CGTS_SM_CTRL_REG);
934         for (i = 0; i < 16; i++)
935                 WREG32(CGTS_SM_CTRL_REG, OVERRIDE);
936         WREG32(CGTS_SM_CTRL_REG, cgts_sm_ctrl_reg);
937
938         /* set HW defaults for 3D engine */
939         WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
940
941         sx_debug_1 = RREG32(SX_DEBUG_1);
942         sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS;
943         WREG32(SX_DEBUG_1, sx_debug_1);
944
945         smx_dc_ctl0 = RREG32(SMX_DC_CTL0);
946         smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff);
947         smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.cayman.sx_num_of_sets);
948         WREG32(SMX_DC_CTL0, smx_dc_ctl0);
949
950         WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4) | CRC_SIMD_ID_WADDR_DISABLE);
951
952         /* need to be explicitly zero-ed */
953         WREG32(VGT_OFFCHIP_LDS_BASE, 0);
954         WREG32(SQ_LSTMP_RING_BASE, 0);
955         WREG32(SQ_HSTMP_RING_BASE, 0);
956         WREG32(SQ_ESTMP_RING_BASE, 0);
957         WREG32(SQ_GSTMP_RING_BASE, 0);
958         WREG32(SQ_VSTMP_RING_BASE, 0);
959         WREG32(SQ_PSTMP_RING_BASE, 0);
960
961         WREG32(TA_CNTL_AUX, DISABLE_CUBE_ANISO);
962
963         WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.cayman.sx_max_export_size / 4) - 1) |
964                                         POSITION_BUFFER_SIZE((rdev->config.cayman.sx_max_export_pos_size / 4) - 1) |
965                                         SMX_BUFFER_SIZE((rdev->config.cayman.sx_max_export_smx_size / 4) - 1)));
966
967         WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.cayman.sc_prim_fifo_size) |
968                                  SC_HIZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_hiz_tile_fifo_size) |
969                                  SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_earlyz_tile_fifo_size)));
970
971
972         WREG32(VGT_NUM_INSTANCES, 1);
973
974         WREG32(CP_PERFMON_CNTL, 0);
975
976         WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.cayman.sq_num_cf_insts) |
977                                   FETCH_FIFO_HIWATER(0x4) |
978                                   DONE_FIFO_HIWATER(0xe0) |
979                                   ALU_UPDATE_FIFO_HIWATER(0x8)));
980
981         WREG32(SQ_GPR_RESOURCE_MGMT_1, NUM_CLAUSE_TEMP_GPRS(4));
982         WREG32(SQ_CONFIG, (VC_ENABLE |
983                            EXPORT_SRC_C |
984                            GFX_PRIO(0) |
985                            CS1_PRIO(0) |
986                            CS2_PRIO(1)));
987         WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, DYN_GPR_ENABLE);
988
989         WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
990                                           FORCE_EOV_MAX_REZ_CNT(255)));
991
992         WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
993                AUTO_INVLD_EN(ES_AND_GS_AUTO));
994
995         WREG32(VGT_GS_VERTEX_REUSE, 16);
996         WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
997
998         WREG32(CB_PERF_CTR0_SEL_0, 0);
999         WREG32(CB_PERF_CTR0_SEL_1, 0);
1000         WREG32(CB_PERF_CTR1_SEL_0, 0);
1001         WREG32(CB_PERF_CTR1_SEL_1, 0);
1002         WREG32(CB_PERF_CTR2_SEL_0, 0);
1003         WREG32(CB_PERF_CTR2_SEL_1, 0);
1004         WREG32(CB_PERF_CTR3_SEL_0, 0);
1005         WREG32(CB_PERF_CTR3_SEL_1, 0);
1006
1007         tmp = RREG32(HDP_MISC_CNTL);
1008         tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1009         WREG32(HDP_MISC_CNTL, tmp);
1010
1011         hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1012         WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1013
1014         WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1015
1016         udelay(50);
1017 }
1018
1019 /*
1020  * GART
1021  */
1022 void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev)
1023 {
1024         /* flush hdp cache */
1025         WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1026
1027         /* bits 0-7 are the VM contexts0-7 */
1028         WREG32(VM_INVALIDATE_REQUEST, 1);
1029 }
1030
1031 int cayman_pcie_gart_enable(struct radeon_device *rdev)
1032 {
1033         int i, r;
1034
1035         if (rdev->gart.robj == NULL) {
1036                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
1037                 return -EINVAL;
1038         }
1039         r = radeon_gart_table_vram_pin(rdev);
1040         if (r)
1041                 return r;
1042         radeon_gart_restore(rdev);
1043         /* Setup TLB control */
1044         WREG32(MC_VM_MX_L1_TLB_CNTL,
1045                (0xA << 7) |
1046                ENABLE_L1_TLB |
1047                ENABLE_L1_FRAGMENT_PROCESSING |
1048                SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1049                ENABLE_ADVANCED_DRIVER_MODEL |
1050                SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
1051         /* Setup L2 cache */
1052         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
1053                ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1054                ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
1055                EFFECTIVE_L2_QUEUE_SIZE(7) |
1056                CONTEXT1_IDENTITY_ACCESS_MODE(1));
1057         WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
1058         WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
1059                L2_CACHE_BIGK_FRAGMENT_SIZE(6));
1060         /* setup context0 */
1061         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
1062         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
1063         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
1064         WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
1065                         (u32)(rdev->dummy_page.addr >> 12));
1066         WREG32(VM_CONTEXT0_CNTL2, 0);
1067         WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
1068                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
1069
1070         WREG32(0x15D4, 0);
1071         WREG32(0x15D8, 0);
1072         WREG32(0x15DC, 0);
1073
1074         /* empty context1-7 */
1075         for (i = 1; i < 8; i++) {
1076                 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR + (i << 2), 0);
1077                 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR + (i << 2), 0);
1078                 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
1079                         rdev->gart.table_addr >> 12);
1080         }
1081
1082         /* enable context1-7 */
1083         WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
1084                (u32)(rdev->dummy_page.addr >> 12));
1085         WREG32(VM_CONTEXT1_CNTL2, 0);
1086         WREG32(VM_CONTEXT1_CNTL, 0);
1087         WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
1088                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
1089
1090         cayman_pcie_gart_tlb_flush(rdev);
1091         DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1092                  (unsigned)(rdev->mc.gtt_size >> 20),
1093                  (unsigned long long)rdev->gart.table_addr);
1094         rdev->gart.ready = true;
1095         return 0;
1096 }
1097
1098 void cayman_pcie_gart_disable(struct radeon_device *rdev)
1099 {
1100         /* Disable all tables */
1101         WREG32(VM_CONTEXT0_CNTL, 0);
1102         WREG32(VM_CONTEXT1_CNTL, 0);
1103         /* Setup TLB control */
1104         WREG32(MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING |
1105                SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1106                SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
1107         /* Setup L2 cache */
1108         WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1109                ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
1110                EFFECTIVE_L2_QUEUE_SIZE(7) |
1111                CONTEXT1_IDENTITY_ACCESS_MODE(1));
1112         WREG32(VM_L2_CNTL2, 0);
1113         WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
1114                L2_CACHE_BIGK_FRAGMENT_SIZE(6));
1115         radeon_gart_table_vram_unpin(rdev);
1116 }
1117
1118 void cayman_pcie_gart_fini(struct radeon_device *rdev)
1119 {
1120         cayman_pcie_gart_disable(rdev);
1121         radeon_gart_table_vram_free(rdev);
1122         radeon_gart_fini(rdev);
1123 }
1124
1125 void cayman_cp_int_cntl_setup(struct radeon_device *rdev,
1126                               int ring, u32 cp_int_cntl)
1127 {
1128         u32 srbm_gfx_cntl = RREG32(SRBM_GFX_CNTL) & ~3;
1129
1130         WREG32(SRBM_GFX_CNTL, srbm_gfx_cntl | (ring & 3));
1131         WREG32(CP_INT_CNTL, cp_int_cntl);
1132 }
1133
1134 /*
1135  * CP.
1136  */
1137 void cayman_fence_ring_emit(struct radeon_device *rdev,
1138                             struct radeon_fence *fence)
1139 {
1140         struct radeon_ring *ring = &rdev->ring[fence->ring];
1141         u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1142
1143         /* flush read cache over gart for this vmid */
1144         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1145         radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1146         radeon_ring_write(ring, 0);
1147         radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1148         radeon_ring_write(ring, PACKET3_TC_ACTION_ENA | PACKET3_SH_ACTION_ENA);
1149         radeon_ring_write(ring, 0xFFFFFFFF);
1150         radeon_ring_write(ring, 0);
1151         radeon_ring_write(ring, 10); /* poll interval */
1152         /* EVENT_WRITE_EOP - flush caches, send int */
1153         radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1154         radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT_TS) | EVENT_INDEX(5));
1155         radeon_ring_write(ring, addr & 0xffffffff);
1156         radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1157         radeon_ring_write(ring, fence->seq);
1158         radeon_ring_write(ring, 0);
1159 }
1160
1161 void cayman_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1162 {
1163         struct radeon_ring *ring = &rdev->ring[ib->fence->ring];
1164
1165         /* set to DX10/11 mode */
1166         radeon_ring_write(ring, PACKET3(PACKET3_MODE_CONTROL, 0));
1167         radeon_ring_write(ring, 1);
1168         radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
1169         radeon_ring_write(ring,
1170 #ifdef __BIG_ENDIAN
1171                           (2 << 0) |
1172 #endif
1173                           (ib->gpu_addr & 0xFFFFFFFC));
1174         radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF);
1175         radeon_ring_write(ring, ib->length_dw | (ib->vm_id << 24));
1176
1177         /* flush read cache over gart for this vmid */
1178         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1179         radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1180         radeon_ring_write(ring, ib->vm_id);
1181         radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1182         radeon_ring_write(ring, PACKET3_TC_ACTION_ENA | PACKET3_SH_ACTION_ENA);
1183         radeon_ring_write(ring, 0xFFFFFFFF);
1184         radeon_ring_write(ring, 0);
1185         radeon_ring_write(ring, 10); /* poll interval */
1186 }
1187
1188 static void cayman_cp_enable(struct radeon_device *rdev, bool enable)
1189 {
1190         if (enable)
1191                 WREG32(CP_ME_CNTL, 0);
1192         else {
1193                 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1194                 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT));
1195                 WREG32(SCRATCH_UMSK, 0);
1196         }
1197 }
1198
1199 static int cayman_cp_load_microcode(struct radeon_device *rdev)
1200 {
1201         const __be32 *fw_data;
1202         int i;
1203
1204         if (!rdev->me_fw || !rdev->pfp_fw)
1205                 return -EINVAL;
1206
1207         cayman_cp_enable(rdev, false);
1208
1209         fw_data = (const __be32 *)rdev->pfp_fw->data;
1210         WREG32(CP_PFP_UCODE_ADDR, 0);
1211         for (i = 0; i < CAYMAN_PFP_UCODE_SIZE; i++)
1212                 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1213         WREG32(CP_PFP_UCODE_ADDR, 0);
1214
1215         fw_data = (const __be32 *)rdev->me_fw->data;
1216         WREG32(CP_ME_RAM_WADDR, 0);
1217         for (i = 0; i < CAYMAN_PM4_UCODE_SIZE; i++)
1218                 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1219
1220         WREG32(CP_PFP_UCODE_ADDR, 0);
1221         WREG32(CP_ME_RAM_WADDR, 0);
1222         WREG32(CP_ME_RAM_RADDR, 0);
1223         return 0;
1224 }
1225
1226 static int cayman_cp_start(struct radeon_device *rdev)
1227 {
1228         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1229         int r, i;
1230
1231         r = radeon_ring_lock(rdev, ring, 7);
1232         if (r) {
1233                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1234                 return r;
1235         }
1236         radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
1237         radeon_ring_write(ring, 0x1);
1238         radeon_ring_write(ring, 0x0);
1239         radeon_ring_write(ring, rdev->config.cayman.max_hw_contexts - 1);
1240         radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1241         radeon_ring_write(ring, 0);
1242         radeon_ring_write(ring, 0);
1243         radeon_ring_unlock_commit(rdev, ring);
1244
1245         cayman_cp_enable(rdev, true);
1246
1247         r = radeon_ring_lock(rdev, ring, cayman_default_size + 19);
1248         if (r) {
1249                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1250                 return r;
1251         }
1252
1253         /* setup clear context state */
1254         radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1255         radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1256
1257         for (i = 0; i < cayman_default_size; i++)
1258                 radeon_ring_write(ring, cayman_default_state[i]);
1259
1260         radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1261         radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
1262
1263         /* set clear context state */
1264         radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
1265         radeon_ring_write(ring, 0);
1266
1267         /* SQ_VTX_BASE_VTX_LOC */
1268         radeon_ring_write(ring, 0xc0026f00);
1269         radeon_ring_write(ring, 0x00000000);
1270         radeon_ring_write(ring, 0x00000000);
1271         radeon_ring_write(ring, 0x00000000);
1272
1273         /* Clear consts */
1274         radeon_ring_write(ring, 0xc0036f00);
1275         radeon_ring_write(ring, 0x00000bc4);
1276         radeon_ring_write(ring, 0xffffffff);
1277         radeon_ring_write(ring, 0xffffffff);
1278         radeon_ring_write(ring, 0xffffffff);
1279
1280         radeon_ring_write(ring, 0xc0026900);
1281         radeon_ring_write(ring, 0x00000316);
1282         radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
1283         radeon_ring_write(ring, 0x00000010); /*  */
1284
1285         radeon_ring_unlock_commit(rdev, ring);
1286
1287         /* XXX init other rings */
1288
1289         return 0;
1290 }
1291
1292 static void cayman_cp_fini(struct radeon_device *rdev)
1293 {
1294         cayman_cp_enable(rdev, false);
1295         radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1296 }
1297
1298 int cayman_cp_resume(struct radeon_device *rdev)
1299 {
1300         struct radeon_ring *ring;
1301         u32 tmp;
1302         u32 rb_bufsz;
1303         int r;
1304
1305         /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
1306         WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
1307                                  SOFT_RESET_PA |
1308                                  SOFT_RESET_SH |
1309                                  SOFT_RESET_VGT |
1310                                  SOFT_RESET_SPI |
1311                                  SOFT_RESET_SX));
1312         RREG32(GRBM_SOFT_RESET);
1313         mdelay(15);
1314         WREG32(GRBM_SOFT_RESET, 0);
1315         RREG32(GRBM_SOFT_RESET);
1316
1317         WREG32(CP_SEM_WAIT_TIMER, 0x0);
1318         WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
1319
1320         /* Set the write pointer delay */
1321         WREG32(CP_RB_WPTR_DELAY, 0);
1322
1323         WREG32(CP_DEBUG, (1 << 27));
1324
1325         /* ring 0 - compute and gfx */
1326         /* Set ring buffer size */
1327         ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1328         rb_bufsz = drm_order(ring->ring_size / 8);
1329         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1330 #ifdef __BIG_ENDIAN
1331         tmp |= BUF_SWAP_32BIT;
1332 #endif
1333         WREG32(CP_RB0_CNTL, tmp);
1334
1335         /* Initialize the ring buffer's read and write pointers */
1336         WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
1337         ring->wptr = 0;
1338         WREG32(CP_RB0_WPTR, ring->wptr);
1339
1340         /* set the wb address wether it's enabled or not */
1341         WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
1342         WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
1343         WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
1344
1345         if (rdev->wb.enabled)
1346                 WREG32(SCRATCH_UMSK, 0xff);
1347         else {
1348                 tmp |= RB_NO_UPDATE;
1349                 WREG32(SCRATCH_UMSK, 0);
1350         }
1351
1352         mdelay(1);
1353         WREG32(CP_RB0_CNTL, tmp);
1354
1355         WREG32(CP_RB0_BASE, ring->gpu_addr >> 8);
1356
1357         ring->rptr = RREG32(CP_RB0_RPTR);
1358
1359         /* ring1  - compute only */
1360         /* Set ring buffer size */
1361         ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
1362         rb_bufsz = drm_order(ring->ring_size / 8);
1363         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1364 #ifdef __BIG_ENDIAN
1365         tmp |= BUF_SWAP_32BIT;
1366 #endif
1367         WREG32(CP_RB1_CNTL, tmp);
1368
1369         /* Initialize the ring buffer's read and write pointers */
1370         WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA);
1371         ring->wptr = 0;
1372         WREG32(CP_RB1_WPTR, ring->wptr);
1373
1374         /* set the wb address wether it's enabled or not */
1375         WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC);
1376         WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF);
1377
1378         mdelay(1);
1379         WREG32(CP_RB1_CNTL, tmp);
1380
1381         WREG32(CP_RB1_BASE, ring->gpu_addr >> 8);
1382
1383         ring->rptr = RREG32(CP_RB1_RPTR);
1384
1385         /* ring2 - compute only */
1386         /* Set ring buffer size */
1387         ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
1388         rb_bufsz = drm_order(ring->ring_size / 8);
1389         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1390 #ifdef __BIG_ENDIAN
1391         tmp |= BUF_SWAP_32BIT;
1392 #endif
1393         WREG32(CP_RB2_CNTL, tmp);
1394
1395         /* Initialize the ring buffer's read and write pointers */
1396         WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA);
1397         ring->wptr = 0;
1398         WREG32(CP_RB2_WPTR, ring->wptr);
1399
1400         /* set the wb address wether it's enabled or not */
1401         WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC);
1402         WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF);
1403
1404         mdelay(1);
1405         WREG32(CP_RB2_CNTL, tmp);
1406
1407         WREG32(CP_RB2_BASE, ring->gpu_addr >> 8);
1408
1409         ring->rptr = RREG32(CP_RB2_RPTR);
1410
1411         /* start the rings */
1412         cayman_cp_start(rdev);
1413         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
1414         rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1415         rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1416         /* this only test cp0 */
1417         r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1418         if (r) {
1419                 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1420                 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1421                 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1422                 return r;
1423         }
1424
1425         return 0;
1426 }
1427
1428 bool cayman_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
1429 {
1430         u32 srbm_status;
1431         u32 grbm_status;
1432         u32 grbm_status_se0, grbm_status_se1;
1433         struct r100_gpu_lockup *lockup = &rdev->config.cayman.lockup;
1434         int r;
1435
1436         srbm_status = RREG32(SRBM_STATUS);
1437         grbm_status = RREG32(GRBM_STATUS);
1438         grbm_status_se0 = RREG32(GRBM_STATUS_SE0);
1439         grbm_status_se1 = RREG32(GRBM_STATUS_SE1);
1440         if (!(grbm_status & GUI_ACTIVE)) {
1441                 r100_gpu_lockup_update(lockup, ring);
1442                 return false;
1443         }
1444         /* force CP activities */
1445         r = radeon_ring_lock(rdev, ring, 2);
1446         if (!r) {
1447                 /* PACKET2 NOP */
1448                 radeon_ring_write(ring, 0x80000000);
1449                 radeon_ring_write(ring, 0x80000000);
1450                 radeon_ring_unlock_commit(rdev, ring);
1451         }
1452         /* XXX deal with CP0,1,2 */
1453         ring->rptr = RREG32(ring->rptr_reg);
1454         return r100_gpu_cp_is_lockup(rdev, lockup, ring);
1455 }
1456
1457 static int cayman_gpu_soft_reset(struct radeon_device *rdev)
1458 {
1459         struct evergreen_mc_save save;
1460         u32 grbm_reset = 0;
1461
1462         if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
1463                 return 0;
1464
1465         dev_info(rdev->dev, "GPU softreset \n");
1466         dev_info(rdev->dev, "  GRBM_STATUS=0x%08X\n",
1467                 RREG32(GRBM_STATUS));
1468         dev_info(rdev->dev, "  GRBM_STATUS_SE0=0x%08X\n",
1469                 RREG32(GRBM_STATUS_SE0));
1470         dev_info(rdev->dev, "  GRBM_STATUS_SE1=0x%08X\n",
1471                 RREG32(GRBM_STATUS_SE1));
1472         dev_info(rdev->dev, "  SRBM_STATUS=0x%08X\n",
1473                 RREG32(SRBM_STATUS));
1474         dev_info(rdev->dev, "  VM_CONTEXT0_PROTECTION_FAULT_ADDR   0x%08X\n",
1475                  RREG32(0x14F8));
1476         dev_info(rdev->dev, "  VM_CONTEXT0_PROTECTION_FAULT_STATUS 0x%08X\n",
1477                  RREG32(0x14D8));
1478         dev_info(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1479                  RREG32(0x14FC));
1480         dev_info(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1481                  RREG32(0x14DC));
1482
1483         evergreen_mc_stop(rdev, &save);
1484         if (evergreen_mc_wait_for_idle(rdev)) {
1485                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1486         }
1487         /* Disable CP parsing/prefetching */
1488         WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
1489
1490         /* reset all the gfx blocks */
1491         grbm_reset = (SOFT_RESET_CP |
1492                       SOFT_RESET_CB |
1493                       SOFT_RESET_DB |
1494                       SOFT_RESET_GDS |
1495                       SOFT_RESET_PA |
1496                       SOFT_RESET_SC |
1497                       SOFT_RESET_SPI |
1498                       SOFT_RESET_SH |
1499                       SOFT_RESET_SX |
1500                       SOFT_RESET_TC |
1501                       SOFT_RESET_TA |
1502                       SOFT_RESET_VGT |
1503                       SOFT_RESET_IA);
1504
1505         dev_info(rdev->dev, "  GRBM_SOFT_RESET=0x%08X\n", grbm_reset);
1506         WREG32(GRBM_SOFT_RESET, grbm_reset);
1507         (void)RREG32(GRBM_SOFT_RESET);
1508         udelay(50);
1509         WREG32(GRBM_SOFT_RESET, 0);
1510         (void)RREG32(GRBM_SOFT_RESET);
1511         /* Wait a little for things to settle down */
1512         udelay(50);
1513
1514         dev_info(rdev->dev, "  GRBM_STATUS=0x%08X\n",
1515                 RREG32(GRBM_STATUS));
1516         dev_info(rdev->dev, "  GRBM_STATUS_SE0=0x%08X\n",
1517                 RREG32(GRBM_STATUS_SE0));
1518         dev_info(rdev->dev, "  GRBM_STATUS_SE1=0x%08X\n",
1519                 RREG32(GRBM_STATUS_SE1));
1520         dev_info(rdev->dev, "  SRBM_STATUS=0x%08X\n",
1521                 RREG32(SRBM_STATUS));
1522         evergreen_mc_resume(rdev, &save);
1523         return 0;
1524 }
1525
1526 int cayman_asic_reset(struct radeon_device *rdev)
1527 {
1528         return cayman_gpu_soft_reset(rdev);
1529 }
1530
1531 static int cayman_startup(struct radeon_device *rdev)
1532 {
1533         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1534         int r;
1535
1536         /* enable pcie gen2 link */
1537         evergreen_pcie_gen2_enable(rdev);
1538
1539         if (rdev->flags & RADEON_IS_IGP) {
1540                 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
1541                         r = ni_init_microcode(rdev);
1542                         if (r) {
1543                                 DRM_ERROR("Failed to load firmware!\n");
1544                                 return r;
1545                         }
1546                 }
1547         } else {
1548                 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) {
1549                         r = ni_init_microcode(rdev);
1550                         if (r) {
1551                                 DRM_ERROR("Failed to load firmware!\n");
1552                                 return r;
1553                         }
1554                 }
1555
1556                 r = ni_mc_load_microcode(rdev);
1557                 if (r) {
1558                         DRM_ERROR("Failed to load MC firmware!\n");
1559                         return r;
1560                 }
1561         }
1562
1563         r = r600_vram_scratch_init(rdev);
1564         if (r)
1565                 return r;
1566
1567         evergreen_mc_program(rdev);
1568         r = cayman_pcie_gart_enable(rdev);
1569         if (r)
1570                 return r;
1571         cayman_gpu_init(rdev);
1572
1573         r = evergreen_blit_init(rdev);
1574         if (r) {
1575                 r600_blit_fini(rdev);
1576                 rdev->asic->copy.copy = NULL;
1577                 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
1578         }
1579
1580         /* allocate rlc buffers */
1581         if (rdev->flags & RADEON_IS_IGP) {
1582                 r = si_rlc_init(rdev);
1583                 if (r) {
1584                         DRM_ERROR("Failed to init rlc BOs!\n");
1585                         return r;
1586                 }
1587         }
1588
1589         /* allocate wb buffer */
1590         r = radeon_wb_init(rdev);
1591         if (r)
1592                 return r;
1593
1594         r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
1595         if (r) {
1596                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
1597                 return r;
1598         }
1599
1600         r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
1601         if (r) {
1602                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
1603                 return r;
1604         }
1605
1606         r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
1607         if (r) {
1608                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
1609                 return r;
1610         }
1611
1612         /* Enable IRQ */
1613         r = r600_irq_init(rdev);
1614         if (r) {
1615                 DRM_ERROR("radeon: IH init failed (%d).\n", r);
1616                 radeon_irq_kms_fini(rdev);
1617                 return r;
1618         }
1619         evergreen_irq_set(rdev);
1620
1621         r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
1622                              CP_RB0_RPTR, CP_RB0_WPTR,
1623                              0, 0xfffff, RADEON_CP_PACKET2);
1624         if (r)
1625                 return r;
1626         r = cayman_cp_load_microcode(rdev);
1627         if (r)
1628                 return r;
1629         r = cayman_cp_resume(rdev);
1630         if (r)
1631                 return r;
1632
1633         r = radeon_ib_pool_start(rdev);
1634         if (r)
1635                 return r;
1636
1637         r = radeon_ib_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
1638         if (r) {
1639                 DRM_ERROR("radeon: failed testing IB (%d).\n", r);
1640                 rdev->accel_working = false;
1641                 return r;
1642         }
1643
1644         r = radeon_vm_manager_start(rdev);
1645         if (r)
1646                 return r;
1647
1648         return 0;
1649 }
1650
1651 int cayman_resume(struct radeon_device *rdev)
1652 {
1653         int r;
1654
1655         /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
1656          * posting will perform necessary task to bring back GPU into good
1657          * shape.
1658          */
1659         /* post card */
1660         atom_asic_init(rdev->mode_info.atom_context);
1661
1662         rdev->accel_working = true;
1663         r = cayman_startup(rdev);
1664         if (r) {
1665                 DRM_ERROR("cayman startup failed on resume\n");
1666                 rdev->accel_working = false;
1667                 return r;
1668         }
1669         return r;
1670 }
1671
1672 int cayman_suspend(struct radeon_device *rdev)
1673 {
1674         /* FIXME: we should wait for ring to be empty */
1675         radeon_ib_pool_suspend(rdev);
1676         radeon_vm_manager_suspend(rdev);
1677         r600_blit_suspend(rdev);
1678         cayman_cp_enable(rdev, false);
1679         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1680         evergreen_irq_suspend(rdev);
1681         radeon_wb_disable(rdev);
1682         cayman_pcie_gart_disable(rdev);
1683         return 0;
1684 }
1685
1686 /* Plan is to move initialization in that function and use
1687  * helper function so that radeon_device_init pretty much
1688  * do nothing more than calling asic specific function. This
1689  * should also allow to remove a bunch of callback function
1690  * like vram_info.
1691  */
1692 int cayman_init(struct radeon_device *rdev)
1693 {
1694         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1695         int r;
1696
1697         /* This don't do much */
1698         r = radeon_gem_init(rdev);
1699         if (r)
1700                 return r;
1701         /* Read BIOS */
1702         if (!radeon_get_bios(rdev)) {
1703                 if (ASIC_IS_AVIVO(rdev))
1704                         return -EINVAL;
1705         }
1706         /* Must be an ATOMBIOS */
1707         if (!rdev->is_atom_bios) {
1708                 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
1709                 return -EINVAL;
1710         }
1711         r = radeon_atombios_init(rdev);
1712         if (r)
1713                 return r;
1714
1715         /* Post card if necessary */
1716         if (!radeon_card_posted(rdev)) {
1717                 if (!rdev->bios) {
1718                         dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
1719                         return -EINVAL;
1720                 }
1721                 DRM_INFO("GPU not posted. posting now...\n");
1722                 atom_asic_init(rdev->mode_info.atom_context);
1723         }
1724         /* Initialize scratch registers */
1725         r600_scratch_init(rdev);
1726         /* Initialize surface registers */
1727         radeon_surface_init(rdev);
1728         /* Initialize clocks */
1729         radeon_get_clock_info(rdev->ddev);
1730         /* Fence driver */
1731         r = radeon_fence_driver_init(rdev);
1732         if (r)
1733                 return r;
1734         /* initialize memory controller */
1735         r = evergreen_mc_init(rdev);
1736         if (r)
1737                 return r;
1738         /* Memory manager */
1739         r = radeon_bo_init(rdev);
1740         if (r)
1741                 return r;
1742
1743         r = radeon_irq_kms_init(rdev);
1744         if (r)
1745                 return r;
1746
1747         ring->ring_obj = NULL;
1748         r600_ring_init(rdev, ring, 1024 * 1024);
1749
1750         rdev->ih.ring_obj = NULL;
1751         r600_ih_ring_init(rdev, 64 * 1024);
1752
1753         r = r600_pcie_gart_init(rdev);
1754         if (r)
1755                 return r;
1756
1757         r = radeon_ib_pool_init(rdev);
1758         rdev->accel_working = true;
1759         if (r) {
1760                 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
1761                 rdev->accel_working = false;
1762         }
1763         r = radeon_vm_manager_init(rdev);
1764         if (r) {
1765                 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
1766         }
1767
1768         r = cayman_startup(rdev);
1769         if (r) {
1770                 dev_err(rdev->dev, "disabling GPU acceleration\n");
1771                 cayman_cp_fini(rdev);
1772                 r600_irq_fini(rdev);
1773                 if (rdev->flags & RADEON_IS_IGP)
1774                         si_rlc_fini(rdev);
1775                 radeon_wb_fini(rdev);
1776                 r100_ib_fini(rdev);
1777                 radeon_vm_manager_fini(rdev);
1778                 radeon_irq_kms_fini(rdev);
1779                 cayman_pcie_gart_fini(rdev);
1780                 rdev->accel_working = false;
1781         }
1782
1783         /* Don't start up if the MC ucode is missing.
1784          * The default clocks and voltages before the MC ucode
1785          * is loaded are not suffient for advanced operations.
1786          *
1787          * We can skip this check for TN, because there is no MC
1788          * ucode.
1789          */
1790         if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) {
1791                 DRM_ERROR("radeon: MC ucode required for NI+.\n");
1792                 return -EINVAL;
1793         }
1794
1795         return 0;
1796 }
1797
1798 void cayman_fini(struct radeon_device *rdev)
1799 {
1800         r600_blit_fini(rdev);
1801         cayman_cp_fini(rdev);
1802         r600_irq_fini(rdev);
1803         if (rdev->flags & RADEON_IS_IGP)
1804                 si_rlc_fini(rdev);
1805         radeon_wb_fini(rdev);
1806         radeon_vm_manager_fini(rdev);
1807         r100_ib_fini(rdev);
1808         radeon_irq_kms_fini(rdev);
1809         cayman_pcie_gart_fini(rdev);
1810         r600_vram_scratch_fini(rdev);
1811         radeon_gem_fini(rdev);
1812         radeon_semaphore_driver_fini(rdev);
1813         radeon_fence_driver_fini(rdev);
1814         radeon_bo_fini(rdev);
1815         radeon_atombios_fini(rdev);
1816         kfree(rdev->bios);
1817         rdev->bios = NULL;
1818 }
1819
1820 /*
1821  * vm
1822  */
1823 int cayman_vm_init(struct radeon_device *rdev)
1824 {
1825         /* number of VMs */
1826         rdev->vm_manager.nvm = 8;
1827         /* base offset of vram pages */
1828         if (rdev->flags & RADEON_IS_IGP) {
1829                 u64 tmp = RREG32(FUS_MC_VM_FB_OFFSET);
1830                 tmp <<= 22;
1831                 rdev->vm_manager.vram_base_offset = tmp;
1832         } else
1833                 rdev->vm_manager.vram_base_offset = 0;
1834         return 0;
1835 }
1836
1837 void cayman_vm_fini(struct radeon_device *rdev)
1838 {
1839 }
1840
1841 int cayman_vm_bind(struct radeon_device *rdev, struct radeon_vm *vm, int id)
1842 {
1843         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR + (id << 2), 0);
1844         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR + (id << 2), vm->last_pfn);
1845         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (id << 2), vm->pt_gpu_addr >> 12);
1846         /* flush hdp cache */
1847         WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1848         /* bits 0-7 are the VM contexts0-7 */
1849         WREG32(VM_INVALIDATE_REQUEST, 1 << id);
1850         return 0;
1851 }
1852
1853 void cayman_vm_unbind(struct radeon_device *rdev, struct radeon_vm *vm)
1854 {
1855         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR + (vm->id << 2), 0);
1856         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR + (vm->id << 2), 0);
1857         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2), 0);
1858         /* flush hdp cache */
1859         WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1860         /* bits 0-7 are the VM contexts0-7 */
1861         WREG32(VM_INVALIDATE_REQUEST, 1 << vm->id);
1862 }
1863
1864 void cayman_vm_tlb_flush(struct radeon_device *rdev, struct radeon_vm *vm)
1865 {
1866         if (vm->id == -1)
1867                 return;
1868
1869         /* flush hdp cache */
1870         WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1871         /* bits 0-7 are the VM contexts0-7 */
1872         WREG32(VM_INVALIDATE_REQUEST, 1 << vm->id);
1873 }
1874
1875 #define R600_PTE_VALID     (1 << 0)
1876 #define R600_PTE_SYSTEM    (1 << 1)
1877 #define R600_PTE_SNOOPED   (1 << 2)
1878 #define R600_PTE_READABLE  (1 << 5)
1879 #define R600_PTE_WRITEABLE (1 << 6)
1880
1881 uint32_t cayman_vm_page_flags(struct radeon_device *rdev,
1882                               struct radeon_vm *vm,
1883                               uint32_t flags)
1884 {
1885         uint32_t r600_flags = 0;
1886
1887         r600_flags |= (flags & RADEON_VM_PAGE_VALID) ? R600_PTE_VALID : 0;
1888         r600_flags |= (flags & RADEON_VM_PAGE_READABLE) ? R600_PTE_READABLE : 0;
1889         r600_flags |= (flags & RADEON_VM_PAGE_WRITEABLE) ? R600_PTE_WRITEABLE : 0;
1890         if (flags & RADEON_VM_PAGE_SYSTEM) {
1891                 r600_flags |= R600_PTE_SYSTEM;
1892                 r600_flags |= (flags & RADEON_VM_PAGE_SNOOPED) ? R600_PTE_SNOOPED : 0;
1893         }
1894         return r600_flags;
1895 }
1896
1897 void cayman_vm_set_page(struct radeon_device *rdev, struct radeon_vm *vm,
1898                         unsigned pfn, uint64_t addr, uint32_t flags)
1899 {
1900         void __iomem *ptr = (void *)vm->pt;
1901
1902         addr = addr & 0xFFFFFFFFFFFFF000ULL;
1903         addr |= flags;
1904         writeq(addr, ptr + (pfn * 8));
1905 }