2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
24 * Authors: Dave Airlie
28 #include <linux/firmware.h>
29 #include <linux/platform_device.h>
32 #include "radeon_drm.h"
37 #define R700_PFP_UCODE_SIZE 848
38 #define R700_PM4_UCODE_SIZE 1360
40 static void rv770_gpu_init(struct radeon_device
*rdev
);
41 void rv770_fini(struct radeon_device
*rdev
);
47 int rv770_pcie_gart_enable(struct radeon_device
*rdev
)
52 if (rdev
->gart
.table
.vram
.robj
== NULL
) {
53 dev_err(rdev
->dev
, "No VRAM object for PCIE GART.\n");
56 r
= radeon_gart_table_vram_pin(rdev
);
60 WREG32(VM_L2_CNTL
, ENABLE_L2_CACHE
| ENABLE_L2_FRAGMENT_PROCESSING
|
61 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE
|
62 EFFECTIVE_L2_QUEUE_SIZE(7));
63 WREG32(VM_L2_CNTL2
, 0);
64 WREG32(VM_L2_CNTL3
, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
65 /* Setup TLB control */
66 tmp
= ENABLE_L1_TLB
| ENABLE_L1_FRAGMENT_PROCESSING
|
67 SYSTEM_ACCESS_MODE_NOT_IN_SYS
|
68 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU
|
69 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
70 WREG32(MC_VM_MD_L1_TLB0_CNTL
, tmp
);
71 WREG32(MC_VM_MD_L1_TLB1_CNTL
, tmp
);
72 WREG32(MC_VM_MD_L1_TLB2_CNTL
, tmp
);
73 WREG32(MC_VM_MB_L1_TLB0_CNTL
, tmp
);
74 WREG32(MC_VM_MB_L1_TLB1_CNTL
, tmp
);
75 WREG32(MC_VM_MB_L1_TLB2_CNTL
, tmp
);
76 WREG32(MC_VM_MB_L1_TLB3_CNTL
, tmp
);
77 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR
, rdev
->mc
.gtt_start
>> 12);
78 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR
, rdev
->mc
.gtt_end
>> 12);
79 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR
, rdev
->gart
.table_addr
>> 12);
80 WREG32(VM_CONTEXT0_CNTL
, ENABLE_CONTEXT
| PAGE_TABLE_DEPTH(0) |
81 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT
);
82 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR
,
83 (u32
)(rdev
->dummy_page
.addr
>> 12));
84 for (i
= 1; i
< 7; i
++)
85 WREG32(VM_CONTEXT0_CNTL
+ (i
* 4), 0);
87 r600_pcie_gart_tlb_flush(rdev
);
88 rdev
->gart
.ready
= true;
92 void rv770_pcie_gart_disable(struct radeon_device
*rdev
)
97 /* Disable all tables */
98 for (i
= 0; i
< 7; i
++)
99 WREG32(VM_CONTEXT0_CNTL
+ (i
* 4), 0);
102 WREG32(VM_L2_CNTL
, ENABLE_L2_FRAGMENT_PROCESSING
|
103 EFFECTIVE_L2_QUEUE_SIZE(7));
104 WREG32(VM_L2_CNTL2
, 0);
105 WREG32(VM_L2_CNTL3
, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
106 /* Setup TLB control */
107 tmp
= EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
108 WREG32(MC_VM_MD_L1_TLB0_CNTL
, tmp
);
109 WREG32(MC_VM_MD_L1_TLB1_CNTL
, tmp
);
110 WREG32(MC_VM_MD_L1_TLB2_CNTL
, tmp
);
111 WREG32(MC_VM_MB_L1_TLB0_CNTL
, tmp
);
112 WREG32(MC_VM_MB_L1_TLB1_CNTL
, tmp
);
113 WREG32(MC_VM_MB_L1_TLB2_CNTL
, tmp
);
114 WREG32(MC_VM_MB_L1_TLB3_CNTL
, tmp
);
115 if (rdev
->gart
.table
.vram
.robj
) {
116 radeon_object_kunmap(rdev
->gart
.table
.vram
.robj
);
117 radeon_object_unpin(rdev
->gart
.table
.vram
.robj
);
121 void rv770_pcie_gart_fini(struct radeon_device
*rdev
)
123 rv770_pcie_gart_disable(rdev
);
124 radeon_gart_table_vram_free(rdev
);
125 radeon_gart_fini(rdev
);
129 void rv770_agp_enable(struct radeon_device
*rdev
)
135 WREG32(VM_L2_CNTL
, ENABLE_L2_CACHE
| ENABLE_L2_FRAGMENT_PROCESSING
|
136 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE
|
137 EFFECTIVE_L2_QUEUE_SIZE(7));
138 WREG32(VM_L2_CNTL2
, 0);
139 WREG32(VM_L2_CNTL3
, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
140 /* Setup TLB control */
141 tmp
= ENABLE_L1_TLB
| ENABLE_L1_FRAGMENT_PROCESSING
|
142 SYSTEM_ACCESS_MODE_NOT_IN_SYS
|
143 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU
|
144 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
145 WREG32(MC_VM_MD_L1_TLB0_CNTL
, tmp
);
146 WREG32(MC_VM_MD_L1_TLB1_CNTL
, tmp
);
147 WREG32(MC_VM_MD_L1_TLB2_CNTL
, tmp
);
148 WREG32(MC_VM_MB_L1_TLB0_CNTL
, tmp
);
149 WREG32(MC_VM_MB_L1_TLB1_CNTL
, tmp
);
150 WREG32(MC_VM_MB_L1_TLB2_CNTL
, tmp
);
151 WREG32(MC_VM_MB_L1_TLB3_CNTL
, tmp
);
152 for (i
= 0; i
< 7; i
++)
153 WREG32(VM_CONTEXT0_CNTL
+ (i
* 4), 0);
156 static void rv770_mc_program(struct radeon_device
*rdev
)
158 struct rv515_mc_save save
;
163 for (i
= 0, j
= 0; i
< 32; i
++, j
+= 0x18) {
164 WREG32((0x2c14 + j
), 0x00000000);
165 WREG32((0x2c18 + j
), 0x00000000);
166 WREG32((0x2c1c + j
), 0x00000000);
167 WREG32((0x2c20 + j
), 0x00000000);
168 WREG32((0x2c24 + j
), 0x00000000);
170 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL
, 0);
172 rv515_mc_stop(rdev
, &save
);
173 if (r600_mc_wait_for_idle(rdev
)) {
174 dev_warn(rdev
->dev
, "Wait for MC idle timedout !\n");
176 /* Lockout access through VGA aperture*/
177 WREG32(VGA_HDP_CONTROL
, VGA_MEMORY_DISABLE
);
178 /* Update configuration */
179 if (rdev
->flags
& RADEON_IS_AGP
) {
180 if (rdev
->mc
.vram_start
< rdev
->mc
.gtt_start
) {
181 /* VRAM before AGP */
182 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
,
183 rdev
->mc
.vram_start
>> 12);
184 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
185 rdev
->mc
.gtt_end
>> 12);
188 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
,
189 rdev
->mc
.gtt_start
>> 12);
190 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
191 rdev
->mc
.vram_end
>> 12);
194 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
,
195 rdev
->mc
.vram_start
>> 12);
196 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
197 rdev
->mc
.vram_end
>> 12);
199 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR
, 0);
200 tmp
= ((rdev
->mc
.vram_end
>> 24) & 0xFFFF) << 16;
201 tmp
|= ((rdev
->mc
.vram_start
>> 24) & 0xFFFF);
202 WREG32(MC_VM_FB_LOCATION
, tmp
);
203 WREG32(HDP_NONSURFACE_BASE
, (rdev
->mc
.vram_start
>> 8));
204 WREG32(HDP_NONSURFACE_INFO
, (2 << 7));
205 WREG32(HDP_NONSURFACE_SIZE
, (rdev
->mc
.mc_vram_size
- 1) | 0x3FF);
206 if (rdev
->flags
& RADEON_IS_AGP
) {
207 WREG32(MC_VM_AGP_TOP
, rdev
->mc
.gtt_end
>> 16);
208 WREG32(MC_VM_AGP_BOT
, rdev
->mc
.gtt_start
>> 16);
209 WREG32(MC_VM_AGP_BASE
, rdev
->mc
.agp_base
>> 22);
211 WREG32(MC_VM_AGP_BASE
, 0);
212 WREG32(MC_VM_AGP_TOP
, 0x0FFFFFFF);
213 WREG32(MC_VM_AGP_BOT
, 0x0FFFFFFF);
215 if (r600_mc_wait_for_idle(rdev
)) {
216 dev_warn(rdev
->dev
, "Wait for MC idle timedout !\n");
218 rv515_mc_resume(rdev
, &save
);
219 /* we need to own VRAM, so turn off the VGA renderer here
220 * to stop it overwriting our objects */
221 rv515_vga_render_disable(rdev
);
228 void r700_cp_stop(struct radeon_device
*rdev
)
230 WREG32(CP_ME_CNTL
, (CP_ME_HALT
| CP_PFP_HALT
));
234 static int rv770_cp_load_microcode(struct radeon_device
*rdev
)
236 const __be32
*fw_data
;
239 if (!rdev
->me_fw
|| !rdev
->pfp_fw
)
243 WREG32(CP_RB_CNTL
, RB_NO_UPDATE
| (15 << 8) | (3 << 0));
246 WREG32(GRBM_SOFT_RESET
, SOFT_RESET_CP
);
247 RREG32(GRBM_SOFT_RESET
);
249 WREG32(GRBM_SOFT_RESET
, 0);
251 fw_data
= (const __be32
*)rdev
->pfp_fw
->data
;
252 WREG32(CP_PFP_UCODE_ADDR
, 0);
253 for (i
= 0; i
< R700_PFP_UCODE_SIZE
; i
++)
254 WREG32(CP_PFP_UCODE_DATA
, be32_to_cpup(fw_data
++));
255 WREG32(CP_PFP_UCODE_ADDR
, 0);
257 fw_data
= (const __be32
*)rdev
->me_fw
->data
;
258 WREG32(CP_ME_RAM_WADDR
, 0);
259 for (i
= 0; i
< R700_PM4_UCODE_SIZE
; i
++)
260 WREG32(CP_ME_RAM_DATA
, be32_to_cpup(fw_data
++));
262 WREG32(CP_PFP_UCODE_ADDR
, 0);
263 WREG32(CP_ME_RAM_WADDR
, 0);
264 WREG32(CP_ME_RAM_RADDR
, 0);
272 static u32
r700_get_tile_pipe_to_backend_map(u32 num_tile_pipes
,
274 u32 backend_disable_mask
)
277 u32 enabled_backends_mask
;
278 u32 enabled_backends_count
;
280 u32 swizzle_pipe
[R7XX_MAX_PIPES
];
284 if (num_tile_pipes
> R7XX_MAX_PIPES
)
285 num_tile_pipes
= R7XX_MAX_PIPES
;
286 if (num_tile_pipes
< 1)
288 if (num_backends
> R7XX_MAX_BACKENDS
)
289 num_backends
= R7XX_MAX_BACKENDS
;
290 if (num_backends
< 1)
293 enabled_backends_mask
= 0;
294 enabled_backends_count
= 0;
295 for (i
= 0; i
< R7XX_MAX_BACKENDS
; ++i
) {
296 if (((backend_disable_mask
>> i
) & 1) == 0) {
297 enabled_backends_mask
|= (1 << i
);
298 ++enabled_backends_count
;
300 if (enabled_backends_count
== num_backends
)
304 if (enabled_backends_count
== 0) {
305 enabled_backends_mask
= 1;
306 enabled_backends_count
= 1;
309 if (enabled_backends_count
!= num_backends
)
310 num_backends
= enabled_backends_count
;
312 memset((uint8_t *)&swizzle_pipe
[0], 0, sizeof(u32
) * R7XX_MAX_PIPES
);
313 switch (num_tile_pipes
) {
369 for (cur_pipe
= 0; cur_pipe
< num_tile_pipes
; ++cur_pipe
) {
370 while (((1 << cur_backend
) & enabled_backends_mask
) == 0)
371 cur_backend
= (cur_backend
+ 1) % R7XX_MAX_BACKENDS
;
373 backend_map
|= (u32
)(((cur_backend
& 3) << (swizzle_pipe
[cur_pipe
] * 2)));
375 cur_backend
= (cur_backend
+ 1) % R7XX_MAX_BACKENDS
;
381 static void rv770_gpu_init(struct radeon_device
*rdev
)
383 int i
, j
, num_qd_pipes
;
386 u32 num_gs_verts_per_thread
;
388 u32 gs_prim_buffer_depth
= 0;
389 u32 sq_ms_fifo_sizes
;
391 u32 sq_thread_resource_mgmt
;
392 u32 hdp_host_path_cntl
;
393 u32 sq_dyn_gpr_size_simd_ab_0
;
395 u32 gb_tiling_config
= 0;
396 u32 cc_rb_backend_disable
= 0;
397 u32 cc_gc_shader_pipe_config
= 0;
401 /* setup chip specs */
402 switch (rdev
->family
) {
404 rdev
->config
.rv770
.max_pipes
= 4;
405 rdev
->config
.rv770
.max_tile_pipes
= 8;
406 rdev
->config
.rv770
.max_simds
= 10;
407 rdev
->config
.rv770
.max_backends
= 4;
408 rdev
->config
.rv770
.max_gprs
= 256;
409 rdev
->config
.rv770
.max_threads
= 248;
410 rdev
->config
.rv770
.max_stack_entries
= 512;
411 rdev
->config
.rv770
.max_hw_contexts
= 8;
412 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
413 rdev
->config
.rv770
.sx_max_export_size
= 128;
414 rdev
->config
.rv770
.sx_max_export_pos_size
= 16;
415 rdev
->config
.rv770
.sx_max_export_smx_size
= 112;
416 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
418 rdev
->config
.rv770
.sx_num_of_sets
= 7;
419 rdev
->config
.rv770
.sc_prim_fifo_size
= 0xF9;
420 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
421 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
424 rdev
->config
.rv770
.max_pipes
= 2;
425 rdev
->config
.rv770
.max_tile_pipes
= 4;
426 rdev
->config
.rv770
.max_simds
= 8;
427 rdev
->config
.rv770
.max_backends
= 2;
428 rdev
->config
.rv770
.max_gprs
= 128;
429 rdev
->config
.rv770
.max_threads
= 248;
430 rdev
->config
.rv770
.max_stack_entries
= 256;
431 rdev
->config
.rv770
.max_hw_contexts
= 8;
432 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
433 rdev
->config
.rv770
.sx_max_export_size
= 256;
434 rdev
->config
.rv770
.sx_max_export_pos_size
= 32;
435 rdev
->config
.rv770
.sx_max_export_smx_size
= 224;
436 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
438 rdev
->config
.rv770
.sx_num_of_sets
= 7;
439 rdev
->config
.rv770
.sc_prim_fifo_size
= 0xf9;
440 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
441 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
442 if (rdev
->config
.rv770
.sx_max_export_pos_size
> 16) {
443 rdev
->config
.rv770
.sx_max_export_pos_size
-= 16;
444 rdev
->config
.rv770
.sx_max_export_smx_size
+= 16;
448 rdev
->config
.rv770
.max_pipes
= 2;
449 rdev
->config
.rv770
.max_tile_pipes
= 2;
450 rdev
->config
.rv770
.max_simds
= 2;
451 rdev
->config
.rv770
.max_backends
= 1;
452 rdev
->config
.rv770
.max_gprs
= 256;
453 rdev
->config
.rv770
.max_threads
= 192;
454 rdev
->config
.rv770
.max_stack_entries
= 256;
455 rdev
->config
.rv770
.max_hw_contexts
= 4;
456 rdev
->config
.rv770
.max_gs_threads
= 8 * 2;
457 rdev
->config
.rv770
.sx_max_export_size
= 128;
458 rdev
->config
.rv770
.sx_max_export_pos_size
= 16;
459 rdev
->config
.rv770
.sx_max_export_smx_size
= 112;
460 rdev
->config
.rv770
.sq_num_cf_insts
= 1;
462 rdev
->config
.rv770
.sx_num_of_sets
= 7;
463 rdev
->config
.rv770
.sc_prim_fifo_size
= 0x40;
464 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
465 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
468 rdev
->config
.rv770
.max_pipes
= 4;
469 rdev
->config
.rv770
.max_tile_pipes
= 4;
470 rdev
->config
.rv770
.max_simds
= 8;
471 rdev
->config
.rv770
.max_backends
= 4;
472 rdev
->config
.rv770
.max_gprs
= 256;
473 rdev
->config
.rv770
.max_threads
= 248;
474 rdev
->config
.rv770
.max_stack_entries
= 512;
475 rdev
->config
.rv770
.max_hw_contexts
= 8;
476 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
477 rdev
->config
.rv770
.sx_max_export_size
= 256;
478 rdev
->config
.rv770
.sx_max_export_pos_size
= 32;
479 rdev
->config
.rv770
.sx_max_export_smx_size
= 224;
480 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
482 rdev
->config
.rv770
.sx_num_of_sets
= 7;
483 rdev
->config
.rv770
.sc_prim_fifo_size
= 0x100;
484 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
485 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
487 if (rdev
->config
.rv770
.sx_max_export_pos_size
> 16) {
488 rdev
->config
.rv770
.sx_max_export_pos_size
-= 16;
489 rdev
->config
.rv770
.sx_max_export_smx_size
+= 16;
498 for (i
= 0; i
< 32; i
++) {
499 WREG32((0x2c14 + j
), 0x00000000);
500 WREG32((0x2c18 + j
), 0x00000000);
501 WREG32((0x2c1c + j
), 0x00000000);
502 WREG32((0x2c20 + j
), 0x00000000);
503 WREG32((0x2c24 + j
), 0x00000000);
507 WREG32(GRBM_CNTL
, GRBM_READ_TIMEOUT(0xff));
509 /* setup tiling, simd, pipe config */
510 mc_arb_ramcfg
= RREG32(MC_ARB_RAMCFG
);
512 switch (rdev
->config
.rv770
.max_tile_pipes
) {
514 gb_tiling_config
|= PIPE_TILING(0);
517 gb_tiling_config
|= PIPE_TILING(1);
520 gb_tiling_config
|= PIPE_TILING(2);
523 gb_tiling_config
|= PIPE_TILING(3);
529 if (rdev
->family
== CHIP_RV770
)
530 gb_tiling_config
|= BANK_TILING(1);
532 gb_tiling_config
|= BANK_TILING((mc_arb_ramcfg
& NOOFBANK_SHIFT
) >> NOOFBANK_MASK
);
534 gb_tiling_config
|= GROUP_SIZE(0);
536 if (((mc_arb_ramcfg
& NOOFROWS_MASK
) & NOOFROWS_SHIFT
) > 3) {
537 gb_tiling_config
|= ROW_TILING(3);
538 gb_tiling_config
|= SAMPLE_SPLIT(3);
541 ROW_TILING(((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
));
543 SAMPLE_SPLIT(((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
));
546 gb_tiling_config
|= BANK_SWAPS(1);
548 backend_map
= r700_get_tile_pipe_to_backend_map(rdev
->config
.rv770
.max_tile_pipes
,
549 rdev
->config
.rv770
.max_backends
,
550 (0xff << rdev
->config
.rv770
.max_backends
) & 0xff);
551 gb_tiling_config
|= BACKEND_MAP(backend_map
);
553 cc_gc_shader_pipe_config
=
554 INACTIVE_QD_PIPES((R7XX_MAX_PIPES_MASK
<< rdev
->config
.rv770
.max_pipes
) & R7XX_MAX_PIPES_MASK
);
555 cc_gc_shader_pipe_config
|=
556 INACTIVE_SIMDS((R7XX_MAX_SIMDS_MASK
<< rdev
->config
.rv770
.max_simds
) & R7XX_MAX_SIMDS_MASK
);
558 cc_rb_backend_disable
=
559 BACKEND_DISABLE((R7XX_MAX_BACKENDS_MASK
<< rdev
->config
.rv770
.max_backends
) & R7XX_MAX_BACKENDS_MASK
);
561 WREG32(GB_TILING_CONFIG
, gb_tiling_config
);
562 WREG32(DCP_TILING_CONFIG
, (gb_tiling_config
& 0xffff));
563 WREG32(HDP_TILING_CONFIG
, (gb_tiling_config
& 0xffff));
565 WREG32(CC_RB_BACKEND_DISABLE
, cc_rb_backend_disable
);
566 WREG32(CC_GC_SHADER_PIPE_CONFIG
, cc_gc_shader_pipe_config
);
567 WREG32(GC_USER_SHADER_PIPE_CONFIG
, cc_gc_shader_pipe_config
);
569 WREG32(CC_SYS_RB_BACKEND_DISABLE
, cc_rb_backend_disable
);
570 WREG32(CGTS_SYS_TCC_DISABLE
, 0);
571 WREG32(CGTS_TCC_DISABLE
, 0);
572 WREG32(CGTS_USER_SYS_TCC_DISABLE
, 0);
573 WREG32(CGTS_USER_TCC_DISABLE
, 0);
576 R7XX_MAX_BACKENDS
- r600_count_pipe_bits(cc_gc_shader_pipe_config
& INACTIVE_QD_PIPES_MASK
);
577 WREG32(VGT_OUT_DEALLOC_CNTL
, (num_qd_pipes
* 4) & DEALLOC_DIST_MASK
);
578 WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL
, ((num_qd_pipes
* 4) - 2) & VTX_REUSE_DEPTH_MASK
);
580 /* set HW defaults for 3D engine */
581 WREG32(CP_QUEUE_THRESHOLDS
, (ROQ_IB1_START(0x16) |
582 ROQ_IB2_START(0x2b)));
584 WREG32(CP_MEQ_THRESHOLDS
, STQ_SPLIT(0x30));
586 WREG32(TA_CNTL_AUX
, (DISABLE_CUBE_ANISO
|
591 sx_debug_1
= RREG32(SX_DEBUG_1
);
592 sx_debug_1
|= ENABLE_NEW_SMX_ADDRESS
;
593 WREG32(SX_DEBUG_1
, sx_debug_1
);
595 smx_dc_ctl0
= RREG32(SMX_DC_CTL0
);
596 smx_dc_ctl0
&= ~CACHE_DEPTH(0x1ff);
597 smx_dc_ctl0
|= CACHE_DEPTH((rdev
->config
.rv770
.sx_num_of_sets
* 64) - 1);
598 WREG32(SMX_DC_CTL0
, smx_dc_ctl0
);
600 WREG32(SMX_EVENT_CTL
, (ES_FLUSH_CTL(4) |
605 if (rdev
->family
== CHIP_RV770
)
606 WREG32(DB_DEBUG3
, DB_CLK_OFF_DELAY(0x1f));
608 db_debug4
= RREG32(DB_DEBUG4
);
609 db_debug4
|= DISABLE_TILE_COVERED_FOR_PS_ITER
;
610 WREG32(DB_DEBUG4
, db_debug4
);
613 WREG32(SX_EXPORT_BUFFER_SIZES
, (COLOR_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_size
/ 4) - 1) |
614 POSITION_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_pos_size
/ 4) - 1) |
615 SMX_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_smx_size
/ 4) - 1)));
617 WREG32(PA_SC_FIFO_SIZE
, (SC_PRIM_FIFO_SIZE(rdev
->config
.rv770
.sc_prim_fifo_size
) |
618 SC_HIZ_TILE_FIFO_SIZE(rdev
->config
.rv770
.sc_hiz_tile_fifo_size
) |
619 SC_EARLYZ_TILE_FIFO_SIZE(rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
)));
621 WREG32(PA_SC_MULTI_CHIP_CNTL
, 0);
623 WREG32(VGT_NUM_INSTANCES
, 1);
625 WREG32(SPI_CONFIG_CNTL
, GPR_WRITE_PRIORITY(0));
627 WREG32(SPI_CONFIG_CNTL_1
, VTX_DONE_DELAY(4));
629 WREG32(CP_PERFMON_CNTL
, 0);
631 sq_ms_fifo_sizes
= (CACHE_FIFO_SIZE(16 * rdev
->config
.rv770
.sq_num_cf_insts
) |
632 DONE_FIFO_HIWATER(0xe0) |
633 ALU_UPDATE_FIFO_HIWATER(0x8));
634 switch (rdev
->family
) {
636 sq_ms_fifo_sizes
|= FETCH_FIFO_HIWATER(0x1);
642 sq_ms_fifo_sizes
|= FETCH_FIFO_HIWATER(0x4);
645 WREG32(SQ_MS_FIFO_SIZES
, sq_ms_fifo_sizes
);
647 /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
648 * should be adjusted as needed by the 2D/3D drivers. This just sets default values
650 sq_config
= RREG32(SQ_CONFIG
);
651 sq_config
&= ~(PS_PRIO(3) |
655 sq_config
|= (DX9_CONSTS
|
662 if (rdev
->family
== CHIP_RV710
)
663 /* no vertex cache */
664 sq_config
&= ~VC_ENABLE
;
666 WREG32(SQ_CONFIG
, sq_config
);
668 WREG32(SQ_GPR_RESOURCE_MGMT_1
, (NUM_PS_GPRS((rdev
->config
.rv770
.max_gprs
* 24)/64) |
669 NUM_VS_GPRS((rdev
->config
.rv770
.max_gprs
* 24)/64) |
670 NUM_CLAUSE_TEMP_GPRS(((rdev
->config
.rv770
.max_gprs
* 24)/64)/2)));
672 WREG32(SQ_GPR_RESOURCE_MGMT_2
, (NUM_GS_GPRS((rdev
->config
.rv770
.max_gprs
* 7)/64) |
673 NUM_ES_GPRS((rdev
->config
.rv770
.max_gprs
* 7)/64)));
675 sq_thread_resource_mgmt
= (NUM_PS_THREADS((rdev
->config
.rv770
.max_threads
* 4)/8) |
676 NUM_VS_THREADS((rdev
->config
.rv770
.max_threads
* 2)/8) |
677 NUM_ES_THREADS((rdev
->config
.rv770
.max_threads
* 1)/8));
678 if (((rdev
->config
.rv770
.max_threads
* 1) / 8) > rdev
->config
.rv770
.max_gs_threads
)
679 sq_thread_resource_mgmt
|= NUM_GS_THREADS(rdev
->config
.rv770
.max_gs_threads
);
681 sq_thread_resource_mgmt
|= NUM_GS_THREADS((rdev
->config
.rv770
.max_gs_threads
* 1)/8);
682 WREG32(SQ_THREAD_RESOURCE_MGMT
, sq_thread_resource_mgmt
);
684 WREG32(SQ_STACK_RESOURCE_MGMT_1
, (NUM_PS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4) |
685 NUM_VS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4)));
687 WREG32(SQ_STACK_RESOURCE_MGMT_2
, (NUM_GS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4) |
688 NUM_ES_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4)));
690 sq_dyn_gpr_size_simd_ab_0
= (SIMDA_RING0((rdev
->config
.rv770
.max_gprs
* 38)/64) |
691 SIMDA_RING1((rdev
->config
.rv770
.max_gprs
* 38)/64) |
692 SIMDB_RING0((rdev
->config
.rv770
.max_gprs
* 38)/64) |
693 SIMDB_RING1((rdev
->config
.rv770
.max_gprs
* 38)/64));
695 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_0
, sq_dyn_gpr_size_simd_ab_0
);
696 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_1
, sq_dyn_gpr_size_simd_ab_0
);
697 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_2
, sq_dyn_gpr_size_simd_ab_0
);
698 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_3
, sq_dyn_gpr_size_simd_ab_0
);
699 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_4
, sq_dyn_gpr_size_simd_ab_0
);
700 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_5
, sq_dyn_gpr_size_simd_ab_0
);
701 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_6
, sq_dyn_gpr_size_simd_ab_0
);
702 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_7
, sq_dyn_gpr_size_simd_ab_0
);
704 WREG32(PA_SC_FORCE_EOV_MAX_CNTS
, (FORCE_EOV_MAX_CLK_CNT(4095) |
705 FORCE_EOV_MAX_REZ_CNT(255)));
707 if (rdev
->family
== CHIP_RV710
)
708 WREG32(VGT_CACHE_INVALIDATION
, (CACHE_INVALIDATION(TC_ONLY
) |
709 AUTO_INVLD_EN(ES_AND_GS_AUTO
)));
711 WREG32(VGT_CACHE_INVALIDATION
, (CACHE_INVALIDATION(VC_AND_TC
) |
712 AUTO_INVLD_EN(ES_AND_GS_AUTO
)));
714 switch (rdev
->family
) {
718 gs_prim_buffer_depth
= 384;
721 gs_prim_buffer_depth
= 128;
727 num_gs_verts_per_thread
= rdev
->config
.rv770
.max_pipes
* 16;
728 vgt_gs_per_es
= gs_prim_buffer_depth
+ num_gs_verts_per_thread
;
729 /* Max value for this is 256 */
730 if (vgt_gs_per_es
> 256)
733 WREG32(VGT_ES_PER_GS
, 128);
734 WREG32(VGT_GS_PER_ES
, vgt_gs_per_es
);
735 WREG32(VGT_GS_PER_VS
, 2);
737 /* more default values. 2D/3D driver should adjust as needed */
738 WREG32(VGT_GS_VERTEX_REUSE
, 16);
739 WREG32(PA_SC_LINE_STIPPLE_STATE
, 0);
740 WREG32(VGT_STRMOUT_EN
, 0);
742 WREG32(PA_SC_MODE_CNTL
, 0);
743 WREG32(PA_SC_EDGERULE
, 0xaaaaaaaa);
744 WREG32(PA_SC_AA_CONFIG
, 0);
745 WREG32(PA_SC_CLIPRECT_RULE
, 0xffff);
746 WREG32(PA_SC_LINE_STIPPLE
, 0);
747 WREG32(SPI_INPUT_Z
, 0);
748 WREG32(SPI_PS_IN_CONTROL_0
, NUM_INTERP(2));
749 WREG32(CB_COLOR7_FRAG
, 0);
751 /* clear render buffer base addresses */
752 WREG32(CB_COLOR0_BASE
, 0);
753 WREG32(CB_COLOR1_BASE
, 0);
754 WREG32(CB_COLOR2_BASE
, 0);
755 WREG32(CB_COLOR3_BASE
, 0);
756 WREG32(CB_COLOR4_BASE
, 0);
757 WREG32(CB_COLOR5_BASE
, 0);
758 WREG32(CB_COLOR6_BASE
, 0);
759 WREG32(CB_COLOR7_BASE
, 0);
763 hdp_host_path_cntl
= RREG32(HDP_HOST_PATH_CNTL
);
764 WREG32(HDP_HOST_PATH_CNTL
, hdp_host_path_cntl
);
766 WREG32(PA_SC_MULTI_CHIP_CNTL
, 0);
768 WREG32(PA_CL_ENHANCE
, (CLIP_VTX_REORDER_ENA
|
773 int rv770_mc_init(struct radeon_device
*rdev
)
779 /* Get VRAM informations */
780 /* FIXME: Don't know how to determine vram width, need to check
783 rdev
->mc
.vram_width
= 128;
784 rdev
->mc
.vram_is_ddr
= true;
785 /* Could aper size report 0 ? */
786 rdev
->mc
.aper_base
= drm_get_resource_start(rdev
->ddev
, 0);
787 rdev
->mc
.aper_size
= drm_get_resource_len(rdev
->ddev
, 0);
788 /* Setup GPU memory space */
789 rdev
->mc
.mc_vram_size
= RREG32(CONFIG_MEMSIZE
);
790 rdev
->mc
.real_vram_size
= RREG32(CONFIG_MEMSIZE
);
792 if (rdev
->mc
.mc_vram_size
> rdev
->mc
.aper_size
)
793 rdev
->mc
.mc_vram_size
= rdev
->mc
.aper_size
;
795 if (rdev
->mc
.real_vram_size
> rdev
->mc
.aper_size
)
796 rdev
->mc
.real_vram_size
= rdev
->mc
.aper_size
;
798 if (rdev
->flags
& RADEON_IS_AGP
) {
799 r
= radeon_agp_init(rdev
);
802 /* gtt_size is setup by radeon_agp_init */
803 rdev
->mc
.gtt_location
= rdev
->mc
.agp_base
;
804 tmp
= 0xFFFFFFFFUL
- rdev
->mc
.agp_base
- rdev
->mc
.gtt_size
;
805 /* Try to put vram before or after AGP because we
806 * we want SYSTEM_APERTURE to cover both VRAM and
807 * AGP so that GPU can catch out of VRAM/AGP access
809 if (rdev
->mc
.gtt_location
> rdev
->mc
.mc_vram_size
) {
810 /* Enought place before */
811 rdev
->mc
.vram_location
= rdev
->mc
.gtt_location
-
812 rdev
->mc
.mc_vram_size
;
813 } else if (tmp
> rdev
->mc
.mc_vram_size
) {
814 /* Enought place after */
815 rdev
->mc
.vram_location
= rdev
->mc
.gtt_location
+
818 /* Try to setup VRAM then AGP might not
819 * not work on some card
821 rdev
->mc
.vram_location
= 0x00000000UL
;
822 rdev
->mc
.gtt_location
= rdev
->mc
.mc_vram_size
;
825 rdev
->mc
.vram_location
= 0x00000000UL
;
826 rdev
->mc
.gtt_location
= rdev
->mc
.mc_vram_size
;
827 rdev
->mc
.gtt_size
= radeon_gart_size
* 1024 * 1024;
829 rdev
->mc
.vram_start
= rdev
->mc
.vram_location
;
830 rdev
->mc
.vram_end
= rdev
->mc
.vram_location
+ rdev
->mc
.mc_vram_size
- 1;
831 rdev
->mc
.gtt_start
= rdev
->mc
.gtt_location
;
832 rdev
->mc
.gtt_end
= rdev
->mc
.gtt_location
+ rdev
->mc
.gtt_size
- 1;
833 /* FIXME: we should enforce default clock in case GPU is not in
836 a
.full
= rfixed_const(100);
837 rdev
->pm
.sclk
.full
= rfixed_const(rdev
->clock
.default_sclk
);
838 rdev
->pm
.sclk
.full
= rfixed_div(rdev
->pm
.sclk
, a
);
841 int rv770_gpu_reset(struct radeon_device
*rdev
)
843 /* FIXME: implement any rv770 specific bits */
844 return r600_gpu_reset(rdev
);
847 static int rv770_startup(struct radeon_device
*rdev
)
851 rv770_mc_program(rdev
);
852 if (rdev
->flags
& RADEON_IS_AGP
) {
853 rv770_agp_enable(rdev
);
855 r
= rv770_pcie_gart_enable(rdev
);
859 rv770_gpu_init(rdev
);
861 r
= radeon_object_pin(rdev
->r600_blit
.shader_obj
, RADEON_GEM_DOMAIN_VRAM
,
862 &rdev
->r600_blit
.shader_gpu_addr
);
864 DRM_ERROR("failed to pin blit object %d\n", r
);
868 r
= radeon_ring_init(rdev
, rdev
->cp
.ring_size
);
871 r
= rv770_cp_load_microcode(rdev
);
874 r
= r600_cp_resume(rdev
);
877 /* write back buffer are not vital so don't worry about failure */
878 r600_wb_enable(rdev
);
882 int rv770_resume(struct radeon_device
*rdev
)
886 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
887 * posting will perform necessary task to bring back GPU into good
891 atom_asic_init(rdev
->mode_info
.atom_context
);
892 /* Initialize clocks */
893 r
= radeon_clocks_init(rdev
);
898 r
= rv770_startup(rdev
);
900 DRM_ERROR("r600 startup failed on resume\n");
904 r
= r600_ib_test(rdev
);
906 DRM_ERROR("radeon: failled testing IB (%d).\n", r
);
913 int rv770_suspend(struct radeon_device
*rdev
)
915 /* FIXME: we should wait for ring to be empty */
917 rdev
->cp
.ready
= false;
918 r600_wb_disable(rdev
);
919 rv770_pcie_gart_disable(rdev
);
920 /* unpin shaders bo */
921 radeon_object_unpin(rdev
->r600_blit
.shader_obj
);
925 /* Plan is to move initialization in that function and use
926 * helper function so that radeon_device_init pretty much
927 * do nothing more than calling asic specific function. This
928 * should also allow to remove a bunch of callback function
931 int rv770_init(struct radeon_device
*rdev
)
935 r
= radeon_dummy_page_init(rdev
);
938 /* This don't do much */
939 r
= radeon_gem_init(rdev
);
943 if (!radeon_get_bios(rdev
)) {
944 if (ASIC_IS_AVIVO(rdev
))
947 /* Must be an ATOMBIOS */
948 if (!rdev
->is_atom_bios
) {
949 dev_err(rdev
->dev
, "Expecting atombios for R600 GPU\n");
952 r
= radeon_atombios_init(rdev
);
955 /* Post card if necessary */
956 if (!r600_card_posted(rdev
) && rdev
->bios
) {
957 DRM_INFO("GPU not posted. posting now...\n");
958 atom_asic_init(rdev
->mode_info
.atom_context
);
960 /* Initialize scratch registers */
961 r600_scratch_init(rdev
);
962 /* Initialize surface registers */
963 radeon_surface_init(rdev
);
964 radeon_get_clock_info(rdev
->ddev
);
965 r
= radeon_clocks_init(rdev
);
969 r
= radeon_fence_driver_init(rdev
);
972 r
= rv770_mc_init(rdev
);
976 r
= radeon_object_init(rdev
);
979 rdev
->cp
.ring_obj
= NULL
;
980 r600_ring_init(rdev
, 1024 * 1024);
982 if (!rdev
->me_fw
|| !rdev
->pfp_fw
) {
983 r
= r600_cp_init_microcode(rdev
);
985 DRM_ERROR("Failed to load firmware!\n");
990 r
= r600_pcie_gart_init(rdev
);
994 rdev
->accel_working
= true;
995 r
= r600_blit_init(rdev
);
997 DRM_ERROR("radeon: failled blitter (%d).\n", r
);
998 rdev
->accel_working
= false;
1001 r
= rv770_startup(rdev
);
1003 rv770_suspend(rdev
);
1005 radeon_ring_fini(rdev
);
1006 rv770_pcie_gart_fini(rdev
);
1007 rdev
->accel_working
= false;
1009 if (rdev
->accel_working
) {
1010 r
= radeon_ib_pool_init(rdev
);
1012 DRM_ERROR("radeon: failled initializing IB pool (%d).\n", r
);
1013 rdev
->accel_working
= false;
1015 r
= r600_ib_test(rdev
);
1017 DRM_ERROR("radeon: failled testing IB (%d).\n", r
);
1018 rdev
->accel_working
= false;
1024 void rv770_fini(struct radeon_device
*rdev
)
1026 rv770_suspend(rdev
);
1028 r600_blit_fini(rdev
);
1029 radeon_ring_fini(rdev
);
1031 rv770_pcie_gart_fini(rdev
);
1032 radeon_gem_fini(rdev
);
1033 radeon_fence_driver_fini(rdev
);
1034 radeon_clocks_fini(rdev
);
1035 if (rdev
->flags
& RADEON_IS_AGP
)
1036 radeon_agp_fini(rdev
);
1037 radeon_object_fini(rdev
);
1038 radeon_atombios_fini(rdev
);
1041 radeon_dummy_page_fini(rdev
);