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 r
= radeon_bo_reserve(rdev
->gart
.table
.vram
.robj
, false);
117 if (likely(r
== 0)) {
118 radeon_bo_kunmap(rdev
->gart
.table
.vram
.robj
);
119 radeon_bo_unpin(rdev
->gart
.table
.vram
.robj
);
120 radeon_bo_unreserve(rdev
->gart
.table
.vram
.robj
);
125 void rv770_pcie_gart_fini(struct radeon_device
*rdev
)
127 rv770_pcie_gart_disable(rdev
);
128 radeon_gart_table_vram_free(rdev
);
129 radeon_gart_fini(rdev
);
133 void rv770_agp_enable(struct radeon_device
*rdev
)
139 WREG32(VM_L2_CNTL
, ENABLE_L2_CACHE
| ENABLE_L2_FRAGMENT_PROCESSING
|
140 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE
|
141 EFFECTIVE_L2_QUEUE_SIZE(7));
142 WREG32(VM_L2_CNTL2
, 0);
143 WREG32(VM_L2_CNTL3
, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
144 /* Setup TLB control */
145 tmp
= ENABLE_L1_TLB
| ENABLE_L1_FRAGMENT_PROCESSING
|
146 SYSTEM_ACCESS_MODE_NOT_IN_SYS
|
147 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU
|
148 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
149 WREG32(MC_VM_MD_L1_TLB0_CNTL
, tmp
);
150 WREG32(MC_VM_MD_L1_TLB1_CNTL
, tmp
);
151 WREG32(MC_VM_MD_L1_TLB2_CNTL
, tmp
);
152 WREG32(MC_VM_MB_L1_TLB0_CNTL
, tmp
);
153 WREG32(MC_VM_MB_L1_TLB1_CNTL
, tmp
);
154 WREG32(MC_VM_MB_L1_TLB2_CNTL
, tmp
);
155 WREG32(MC_VM_MB_L1_TLB3_CNTL
, tmp
);
156 for (i
= 0; i
< 7; i
++)
157 WREG32(VM_CONTEXT0_CNTL
+ (i
* 4), 0);
160 static void rv770_mc_program(struct radeon_device
*rdev
)
162 struct rv515_mc_save save
;
167 for (i
= 0, j
= 0; i
< 32; i
++, j
+= 0x18) {
168 WREG32((0x2c14 + j
), 0x00000000);
169 WREG32((0x2c18 + j
), 0x00000000);
170 WREG32((0x2c1c + j
), 0x00000000);
171 WREG32((0x2c20 + j
), 0x00000000);
172 WREG32((0x2c24 + j
), 0x00000000);
174 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL
, 0);
176 rv515_mc_stop(rdev
, &save
);
177 if (r600_mc_wait_for_idle(rdev
)) {
178 dev_warn(rdev
->dev
, "Wait for MC idle timedout !\n");
180 /* Lockout access through VGA aperture*/
181 WREG32(VGA_HDP_CONTROL
, VGA_MEMORY_DISABLE
);
182 /* Update configuration */
183 if (rdev
->flags
& RADEON_IS_AGP
) {
184 if (rdev
->mc
.vram_start
< rdev
->mc
.gtt_start
) {
185 /* VRAM before AGP */
186 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
,
187 rdev
->mc
.vram_start
>> 12);
188 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
189 rdev
->mc
.gtt_end
>> 12);
192 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
,
193 rdev
->mc
.gtt_start
>> 12);
194 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
195 rdev
->mc
.vram_end
>> 12);
198 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
,
199 rdev
->mc
.vram_start
>> 12);
200 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
201 rdev
->mc
.vram_end
>> 12);
203 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR
, 0);
204 tmp
= ((rdev
->mc
.vram_end
>> 24) & 0xFFFF) << 16;
205 tmp
|= ((rdev
->mc
.vram_start
>> 24) & 0xFFFF);
206 WREG32(MC_VM_FB_LOCATION
, tmp
);
207 WREG32(HDP_NONSURFACE_BASE
, (rdev
->mc
.vram_start
>> 8));
208 WREG32(HDP_NONSURFACE_INFO
, (2 << 7));
209 WREG32(HDP_NONSURFACE_SIZE
, (rdev
->mc
.mc_vram_size
- 1) | 0x3FF);
210 if (rdev
->flags
& RADEON_IS_AGP
) {
211 WREG32(MC_VM_AGP_TOP
, rdev
->mc
.gtt_end
>> 16);
212 WREG32(MC_VM_AGP_BOT
, rdev
->mc
.gtt_start
>> 16);
213 WREG32(MC_VM_AGP_BASE
, rdev
->mc
.agp_base
>> 22);
215 WREG32(MC_VM_AGP_BASE
, 0);
216 WREG32(MC_VM_AGP_TOP
, 0x0FFFFFFF);
217 WREG32(MC_VM_AGP_BOT
, 0x0FFFFFFF);
219 if (r600_mc_wait_for_idle(rdev
)) {
220 dev_warn(rdev
->dev
, "Wait for MC idle timedout !\n");
222 rv515_mc_resume(rdev
, &save
);
223 /* we need to own VRAM, so turn off the VGA renderer here
224 * to stop it overwriting our objects */
225 rv515_vga_render_disable(rdev
);
232 void r700_cp_stop(struct radeon_device
*rdev
)
234 WREG32(CP_ME_CNTL
, (CP_ME_HALT
| CP_PFP_HALT
));
238 static int rv770_cp_load_microcode(struct radeon_device
*rdev
)
240 const __be32
*fw_data
;
243 if (!rdev
->me_fw
|| !rdev
->pfp_fw
)
247 WREG32(CP_RB_CNTL
, RB_NO_UPDATE
| (15 << 8) | (3 << 0));
250 WREG32(GRBM_SOFT_RESET
, SOFT_RESET_CP
);
251 RREG32(GRBM_SOFT_RESET
);
253 WREG32(GRBM_SOFT_RESET
, 0);
255 fw_data
= (const __be32
*)rdev
->pfp_fw
->data
;
256 WREG32(CP_PFP_UCODE_ADDR
, 0);
257 for (i
= 0; i
< R700_PFP_UCODE_SIZE
; i
++)
258 WREG32(CP_PFP_UCODE_DATA
, be32_to_cpup(fw_data
++));
259 WREG32(CP_PFP_UCODE_ADDR
, 0);
261 fw_data
= (const __be32
*)rdev
->me_fw
->data
;
262 WREG32(CP_ME_RAM_WADDR
, 0);
263 for (i
= 0; i
< R700_PM4_UCODE_SIZE
; i
++)
264 WREG32(CP_ME_RAM_DATA
, be32_to_cpup(fw_data
++));
266 WREG32(CP_PFP_UCODE_ADDR
, 0);
267 WREG32(CP_ME_RAM_WADDR
, 0);
268 WREG32(CP_ME_RAM_RADDR
, 0);
276 static u32
r700_get_tile_pipe_to_backend_map(u32 num_tile_pipes
,
278 u32 backend_disable_mask
)
281 u32 enabled_backends_mask
;
282 u32 enabled_backends_count
;
284 u32 swizzle_pipe
[R7XX_MAX_PIPES
];
288 if (num_tile_pipes
> R7XX_MAX_PIPES
)
289 num_tile_pipes
= R7XX_MAX_PIPES
;
290 if (num_tile_pipes
< 1)
292 if (num_backends
> R7XX_MAX_BACKENDS
)
293 num_backends
= R7XX_MAX_BACKENDS
;
294 if (num_backends
< 1)
297 enabled_backends_mask
= 0;
298 enabled_backends_count
= 0;
299 for (i
= 0; i
< R7XX_MAX_BACKENDS
; ++i
) {
300 if (((backend_disable_mask
>> i
) & 1) == 0) {
301 enabled_backends_mask
|= (1 << i
);
302 ++enabled_backends_count
;
304 if (enabled_backends_count
== num_backends
)
308 if (enabled_backends_count
== 0) {
309 enabled_backends_mask
= 1;
310 enabled_backends_count
= 1;
313 if (enabled_backends_count
!= num_backends
)
314 num_backends
= enabled_backends_count
;
316 memset((uint8_t *)&swizzle_pipe
[0], 0, sizeof(u32
) * R7XX_MAX_PIPES
);
317 switch (num_tile_pipes
) {
373 for (cur_pipe
= 0; cur_pipe
< num_tile_pipes
; ++cur_pipe
) {
374 while (((1 << cur_backend
) & enabled_backends_mask
) == 0)
375 cur_backend
= (cur_backend
+ 1) % R7XX_MAX_BACKENDS
;
377 backend_map
|= (u32
)(((cur_backend
& 3) << (swizzle_pipe
[cur_pipe
] * 2)));
379 cur_backend
= (cur_backend
+ 1) % R7XX_MAX_BACKENDS
;
385 static void rv770_gpu_init(struct radeon_device
*rdev
)
387 int i
, j
, num_qd_pipes
;
390 u32 num_gs_verts_per_thread
;
392 u32 gs_prim_buffer_depth
= 0;
393 u32 sq_ms_fifo_sizes
;
395 u32 sq_thread_resource_mgmt
;
396 u32 hdp_host_path_cntl
;
397 u32 sq_dyn_gpr_size_simd_ab_0
;
399 u32 gb_tiling_config
= 0;
400 u32 cc_rb_backend_disable
= 0;
401 u32 cc_gc_shader_pipe_config
= 0;
405 /* setup chip specs */
406 switch (rdev
->family
) {
408 rdev
->config
.rv770
.max_pipes
= 4;
409 rdev
->config
.rv770
.max_tile_pipes
= 8;
410 rdev
->config
.rv770
.max_simds
= 10;
411 rdev
->config
.rv770
.max_backends
= 4;
412 rdev
->config
.rv770
.max_gprs
= 256;
413 rdev
->config
.rv770
.max_threads
= 248;
414 rdev
->config
.rv770
.max_stack_entries
= 512;
415 rdev
->config
.rv770
.max_hw_contexts
= 8;
416 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
417 rdev
->config
.rv770
.sx_max_export_size
= 128;
418 rdev
->config
.rv770
.sx_max_export_pos_size
= 16;
419 rdev
->config
.rv770
.sx_max_export_smx_size
= 112;
420 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
422 rdev
->config
.rv770
.sx_num_of_sets
= 7;
423 rdev
->config
.rv770
.sc_prim_fifo_size
= 0xF9;
424 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
425 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
428 rdev
->config
.rv770
.max_pipes
= 2;
429 rdev
->config
.rv770
.max_tile_pipes
= 4;
430 rdev
->config
.rv770
.max_simds
= 8;
431 rdev
->config
.rv770
.max_backends
= 2;
432 rdev
->config
.rv770
.max_gprs
= 128;
433 rdev
->config
.rv770
.max_threads
= 248;
434 rdev
->config
.rv770
.max_stack_entries
= 256;
435 rdev
->config
.rv770
.max_hw_contexts
= 8;
436 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
437 rdev
->config
.rv770
.sx_max_export_size
= 256;
438 rdev
->config
.rv770
.sx_max_export_pos_size
= 32;
439 rdev
->config
.rv770
.sx_max_export_smx_size
= 224;
440 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
442 rdev
->config
.rv770
.sx_num_of_sets
= 7;
443 rdev
->config
.rv770
.sc_prim_fifo_size
= 0xf9;
444 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
445 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
446 if (rdev
->config
.rv770
.sx_max_export_pos_size
> 16) {
447 rdev
->config
.rv770
.sx_max_export_pos_size
-= 16;
448 rdev
->config
.rv770
.sx_max_export_smx_size
+= 16;
452 rdev
->config
.rv770
.max_pipes
= 2;
453 rdev
->config
.rv770
.max_tile_pipes
= 2;
454 rdev
->config
.rv770
.max_simds
= 2;
455 rdev
->config
.rv770
.max_backends
= 1;
456 rdev
->config
.rv770
.max_gprs
= 256;
457 rdev
->config
.rv770
.max_threads
= 192;
458 rdev
->config
.rv770
.max_stack_entries
= 256;
459 rdev
->config
.rv770
.max_hw_contexts
= 4;
460 rdev
->config
.rv770
.max_gs_threads
= 8 * 2;
461 rdev
->config
.rv770
.sx_max_export_size
= 128;
462 rdev
->config
.rv770
.sx_max_export_pos_size
= 16;
463 rdev
->config
.rv770
.sx_max_export_smx_size
= 112;
464 rdev
->config
.rv770
.sq_num_cf_insts
= 1;
466 rdev
->config
.rv770
.sx_num_of_sets
= 7;
467 rdev
->config
.rv770
.sc_prim_fifo_size
= 0x40;
468 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
469 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
472 rdev
->config
.rv770
.max_pipes
= 4;
473 rdev
->config
.rv770
.max_tile_pipes
= 4;
474 rdev
->config
.rv770
.max_simds
= 8;
475 rdev
->config
.rv770
.max_backends
= 4;
476 rdev
->config
.rv770
.max_gprs
= 256;
477 rdev
->config
.rv770
.max_threads
= 248;
478 rdev
->config
.rv770
.max_stack_entries
= 512;
479 rdev
->config
.rv770
.max_hw_contexts
= 8;
480 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
481 rdev
->config
.rv770
.sx_max_export_size
= 256;
482 rdev
->config
.rv770
.sx_max_export_pos_size
= 32;
483 rdev
->config
.rv770
.sx_max_export_smx_size
= 224;
484 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
486 rdev
->config
.rv770
.sx_num_of_sets
= 7;
487 rdev
->config
.rv770
.sc_prim_fifo_size
= 0x100;
488 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
489 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
491 if (rdev
->config
.rv770
.sx_max_export_pos_size
> 16) {
492 rdev
->config
.rv770
.sx_max_export_pos_size
-= 16;
493 rdev
->config
.rv770
.sx_max_export_smx_size
+= 16;
502 for (i
= 0; i
< 32; i
++) {
503 WREG32((0x2c14 + j
), 0x00000000);
504 WREG32((0x2c18 + j
), 0x00000000);
505 WREG32((0x2c1c + j
), 0x00000000);
506 WREG32((0x2c20 + j
), 0x00000000);
507 WREG32((0x2c24 + j
), 0x00000000);
511 WREG32(GRBM_CNTL
, GRBM_READ_TIMEOUT(0xff));
513 /* setup tiling, simd, pipe config */
514 mc_arb_ramcfg
= RREG32(MC_ARB_RAMCFG
);
516 switch (rdev
->config
.rv770
.max_tile_pipes
) {
518 gb_tiling_config
|= PIPE_TILING(0);
521 gb_tiling_config
|= PIPE_TILING(1);
524 gb_tiling_config
|= PIPE_TILING(2);
527 gb_tiling_config
|= PIPE_TILING(3);
533 if (rdev
->family
== CHIP_RV770
)
534 gb_tiling_config
|= BANK_TILING(1);
536 gb_tiling_config
|= BANK_TILING((mc_arb_ramcfg
& NOOFBANK_MASK
) >> NOOFBANK_SHIFT
);
538 gb_tiling_config
|= GROUP_SIZE(0);
540 if (((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
) > 3) {
541 gb_tiling_config
|= ROW_TILING(3);
542 gb_tiling_config
|= SAMPLE_SPLIT(3);
545 ROW_TILING(((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
));
547 SAMPLE_SPLIT(((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
));
550 gb_tiling_config
|= BANK_SWAPS(1);
552 backend_map
= r700_get_tile_pipe_to_backend_map(rdev
->config
.rv770
.max_tile_pipes
,
553 rdev
->config
.rv770
.max_backends
,
554 (0xff << rdev
->config
.rv770
.max_backends
) & 0xff);
555 gb_tiling_config
|= BACKEND_MAP(backend_map
);
557 cc_gc_shader_pipe_config
=
558 INACTIVE_QD_PIPES((R7XX_MAX_PIPES_MASK
<< rdev
->config
.rv770
.max_pipes
) & R7XX_MAX_PIPES_MASK
);
559 cc_gc_shader_pipe_config
|=
560 INACTIVE_SIMDS((R7XX_MAX_SIMDS_MASK
<< rdev
->config
.rv770
.max_simds
) & R7XX_MAX_SIMDS_MASK
);
562 cc_rb_backend_disable
=
563 BACKEND_DISABLE((R7XX_MAX_BACKENDS_MASK
<< rdev
->config
.rv770
.max_backends
) & R7XX_MAX_BACKENDS_MASK
);
565 WREG32(GB_TILING_CONFIG
, gb_tiling_config
);
566 WREG32(DCP_TILING_CONFIG
, (gb_tiling_config
& 0xffff));
567 WREG32(HDP_TILING_CONFIG
, (gb_tiling_config
& 0xffff));
569 WREG32(CC_RB_BACKEND_DISABLE
, cc_rb_backend_disable
);
570 WREG32(CC_GC_SHADER_PIPE_CONFIG
, cc_gc_shader_pipe_config
);
571 WREG32(GC_USER_SHADER_PIPE_CONFIG
, cc_gc_shader_pipe_config
);
573 WREG32(CC_SYS_RB_BACKEND_DISABLE
, cc_rb_backend_disable
);
574 WREG32(CGTS_SYS_TCC_DISABLE
, 0);
575 WREG32(CGTS_TCC_DISABLE
, 0);
576 WREG32(CGTS_USER_SYS_TCC_DISABLE
, 0);
577 WREG32(CGTS_USER_TCC_DISABLE
, 0);
580 R7XX_MAX_BACKENDS
- r600_count_pipe_bits(cc_gc_shader_pipe_config
& INACTIVE_QD_PIPES_MASK
);
581 WREG32(VGT_OUT_DEALLOC_CNTL
, (num_qd_pipes
* 4) & DEALLOC_DIST_MASK
);
582 WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL
, ((num_qd_pipes
* 4) - 2) & VTX_REUSE_DEPTH_MASK
);
584 /* set HW defaults for 3D engine */
585 WREG32(CP_QUEUE_THRESHOLDS
, (ROQ_IB1_START(0x16) |
586 ROQ_IB2_START(0x2b)));
588 WREG32(CP_MEQ_THRESHOLDS
, STQ_SPLIT(0x30));
590 WREG32(TA_CNTL_AUX
, (DISABLE_CUBE_ANISO
|
595 sx_debug_1
= RREG32(SX_DEBUG_1
);
596 sx_debug_1
|= ENABLE_NEW_SMX_ADDRESS
;
597 WREG32(SX_DEBUG_1
, sx_debug_1
);
599 smx_dc_ctl0
= RREG32(SMX_DC_CTL0
);
600 smx_dc_ctl0
&= ~CACHE_DEPTH(0x1ff);
601 smx_dc_ctl0
|= CACHE_DEPTH((rdev
->config
.rv770
.sx_num_of_sets
* 64) - 1);
602 WREG32(SMX_DC_CTL0
, smx_dc_ctl0
);
604 WREG32(SMX_EVENT_CTL
, (ES_FLUSH_CTL(4) |
609 if (rdev
->family
== CHIP_RV770
)
610 WREG32(DB_DEBUG3
, DB_CLK_OFF_DELAY(0x1f));
612 db_debug4
= RREG32(DB_DEBUG4
);
613 db_debug4
|= DISABLE_TILE_COVERED_FOR_PS_ITER
;
614 WREG32(DB_DEBUG4
, db_debug4
);
617 WREG32(SX_EXPORT_BUFFER_SIZES
, (COLOR_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_size
/ 4) - 1) |
618 POSITION_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_pos_size
/ 4) - 1) |
619 SMX_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_smx_size
/ 4) - 1)));
621 WREG32(PA_SC_FIFO_SIZE
, (SC_PRIM_FIFO_SIZE(rdev
->config
.rv770
.sc_prim_fifo_size
) |
622 SC_HIZ_TILE_FIFO_SIZE(rdev
->config
.rv770
.sc_hiz_tile_fifo_size
) |
623 SC_EARLYZ_TILE_FIFO_SIZE(rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
)));
625 WREG32(PA_SC_MULTI_CHIP_CNTL
, 0);
627 WREG32(VGT_NUM_INSTANCES
, 1);
629 WREG32(SPI_CONFIG_CNTL
, GPR_WRITE_PRIORITY(0));
631 WREG32(SPI_CONFIG_CNTL_1
, VTX_DONE_DELAY(4));
633 WREG32(CP_PERFMON_CNTL
, 0);
635 sq_ms_fifo_sizes
= (CACHE_FIFO_SIZE(16 * rdev
->config
.rv770
.sq_num_cf_insts
) |
636 DONE_FIFO_HIWATER(0xe0) |
637 ALU_UPDATE_FIFO_HIWATER(0x8));
638 switch (rdev
->family
) {
640 sq_ms_fifo_sizes
|= FETCH_FIFO_HIWATER(0x1);
646 sq_ms_fifo_sizes
|= FETCH_FIFO_HIWATER(0x4);
649 WREG32(SQ_MS_FIFO_SIZES
, sq_ms_fifo_sizes
);
651 /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
652 * should be adjusted as needed by the 2D/3D drivers. This just sets default values
654 sq_config
= RREG32(SQ_CONFIG
);
655 sq_config
&= ~(PS_PRIO(3) |
659 sq_config
|= (DX9_CONSTS
|
666 if (rdev
->family
== CHIP_RV710
)
667 /* no vertex cache */
668 sq_config
&= ~VC_ENABLE
;
670 WREG32(SQ_CONFIG
, sq_config
);
672 WREG32(SQ_GPR_RESOURCE_MGMT_1
, (NUM_PS_GPRS((rdev
->config
.rv770
.max_gprs
* 24)/64) |
673 NUM_VS_GPRS((rdev
->config
.rv770
.max_gprs
* 24)/64) |
674 NUM_CLAUSE_TEMP_GPRS(((rdev
->config
.rv770
.max_gprs
* 24)/64)/2)));
676 WREG32(SQ_GPR_RESOURCE_MGMT_2
, (NUM_GS_GPRS((rdev
->config
.rv770
.max_gprs
* 7)/64) |
677 NUM_ES_GPRS((rdev
->config
.rv770
.max_gprs
* 7)/64)));
679 sq_thread_resource_mgmt
= (NUM_PS_THREADS((rdev
->config
.rv770
.max_threads
* 4)/8) |
680 NUM_VS_THREADS((rdev
->config
.rv770
.max_threads
* 2)/8) |
681 NUM_ES_THREADS((rdev
->config
.rv770
.max_threads
* 1)/8));
682 if (((rdev
->config
.rv770
.max_threads
* 1) / 8) > rdev
->config
.rv770
.max_gs_threads
)
683 sq_thread_resource_mgmt
|= NUM_GS_THREADS(rdev
->config
.rv770
.max_gs_threads
);
685 sq_thread_resource_mgmt
|= NUM_GS_THREADS((rdev
->config
.rv770
.max_gs_threads
* 1)/8);
686 WREG32(SQ_THREAD_RESOURCE_MGMT
, sq_thread_resource_mgmt
);
688 WREG32(SQ_STACK_RESOURCE_MGMT_1
, (NUM_PS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4) |
689 NUM_VS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4)));
691 WREG32(SQ_STACK_RESOURCE_MGMT_2
, (NUM_GS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4) |
692 NUM_ES_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4)));
694 sq_dyn_gpr_size_simd_ab_0
= (SIMDA_RING0((rdev
->config
.rv770
.max_gprs
* 38)/64) |
695 SIMDA_RING1((rdev
->config
.rv770
.max_gprs
* 38)/64) |
696 SIMDB_RING0((rdev
->config
.rv770
.max_gprs
* 38)/64) |
697 SIMDB_RING1((rdev
->config
.rv770
.max_gprs
* 38)/64));
699 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_0
, sq_dyn_gpr_size_simd_ab_0
);
700 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_1
, sq_dyn_gpr_size_simd_ab_0
);
701 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_2
, sq_dyn_gpr_size_simd_ab_0
);
702 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_3
, sq_dyn_gpr_size_simd_ab_0
);
703 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_4
, sq_dyn_gpr_size_simd_ab_0
);
704 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_5
, sq_dyn_gpr_size_simd_ab_0
);
705 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_6
, sq_dyn_gpr_size_simd_ab_0
);
706 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_7
, sq_dyn_gpr_size_simd_ab_0
);
708 WREG32(PA_SC_FORCE_EOV_MAX_CNTS
, (FORCE_EOV_MAX_CLK_CNT(4095) |
709 FORCE_EOV_MAX_REZ_CNT(255)));
711 if (rdev
->family
== CHIP_RV710
)
712 WREG32(VGT_CACHE_INVALIDATION
, (CACHE_INVALIDATION(TC_ONLY
) |
713 AUTO_INVLD_EN(ES_AND_GS_AUTO
)));
715 WREG32(VGT_CACHE_INVALIDATION
, (CACHE_INVALIDATION(VC_AND_TC
) |
716 AUTO_INVLD_EN(ES_AND_GS_AUTO
)));
718 switch (rdev
->family
) {
722 gs_prim_buffer_depth
= 384;
725 gs_prim_buffer_depth
= 128;
731 num_gs_verts_per_thread
= rdev
->config
.rv770
.max_pipes
* 16;
732 vgt_gs_per_es
= gs_prim_buffer_depth
+ num_gs_verts_per_thread
;
733 /* Max value for this is 256 */
734 if (vgt_gs_per_es
> 256)
737 WREG32(VGT_ES_PER_GS
, 128);
738 WREG32(VGT_GS_PER_ES
, vgt_gs_per_es
);
739 WREG32(VGT_GS_PER_VS
, 2);
741 /* more default values. 2D/3D driver should adjust as needed */
742 WREG32(VGT_GS_VERTEX_REUSE
, 16);
743 WREG32(PA_SC_LINE_STIPPLE_STATE
, 0);
744 WREG32(VGT_STRMOUT_EN
, 0);
746 WREG32(PA_SC_MODE_CNTL
, 0);
747 WREG32(PA_SC_EDGERULE
, 0xaaaaaaaa);
748 WREG32(PA_SC_AA_CONFIG
, 0);
749 WREG32(PA_SC_CLIPRECT_RULE
, 0xffff);
750 WREG32(PA_SC_LINE_STIPPLE
, 0);
751 WREG32(SPI_INPUT_Z
, 0);
752 WREG32(SPI_PS_IN_CONTROL_0
, NUM_INTERP(2));
753 WREG32(CB_COLOR7_FRAG
, 0);
755 /* clear render buffer base addresses */
756 WREG32(CB_COLOR0_BASE
, 0);
757 WREG32(CB_COLOR1_BASE
, 0);
758 WREG32(CB_COLOR2_BASE
, 0);
759 WREG32(CB_COLOR3_BASE
, 0);
760 WREG32(CB_COLOR4_BASE
, 0);
761 WREG32(CB_COLOR5_BASE
, 0);
762 WREG32(CB_COLOR6_BASE
, 0);
763 WREG32(CB_COLOR7_BASE
, 0);
767 hdp_host_path_cntl
= RREG32(HDP_HOST_PATH_CNTL
);
768 WREG32(HDP_HOST_PATH_CNTL
, hdp_host_path_cntl
);
770 WREG32(PA_SC_MULTI_CHIP_CNTL
, 0);
772 WREG32(PA_CL_ENHANCE
, (CLIP_VTX_REORDER_ENA
|
777 int rv770_mc_init(struct radeon_device
*rdev
)
781 int chansize
, numchan
;
783 /* Get VRAM informations */
784 rdev
->mc
.vram_is_ddr
= true;
785 tmp
= RREG32(MC_ARB_RAMCFG
);
786 if (tmp
& CHANSIZE_OVERRIDE
) {
788 } else if (tmp
& CHANSIZE_MASK
) {
793 tmp
= RREG32(MC_SHARED_CHMAP
);
794 switch ((tmp
& NOOFCHAN_MASK
) >> NOOFCHAN_SHIFT
) {
809 rdev
->mc
.vram_width
= numchan
* chansize
;
810 /* Could aper size report 0 ? */
811 rdev
->mc
.aper_base
= drm_get_resource_start(rdev
->ddev
, 0);
812 rdev
->mc
.aper_size
= drm_get_resource_len(rdev
->ddev
, 0);
813 /* Setup GPU memory space */
814 rdev
->mc
.mc_vram_size
= RREG32(CONFIG_MEMSIZE
);
815 rdev
->mc
.real_vram_size
= RREG32(CONFIG_MEMSIZE
);
817 if (rdev
->mc
.mc_vram_size
> rdev
->mc
.aper_size
)
818 rdev
->mc
.mc_vram_size
= rdev
->mc
.aper_size
;
820 if (rdev
->mc
.real_vram_size
> rdev
->mc
.aper_size
)
821 rdev
->mc
.real_vram_size
= rdev
->mc
.aper_size
;
823 if (rdev
->flags
& RADEON_IS_AGP
) {
824 /* gtt_size is setup by radeon_agp_init */
825 rdev
->mc
.gtt_location
= rdev
->mc
.agp_base
;
826 tmp
= 0xFFFFFFFFUL
- rdev
->mc
.agp_base
- rdev
->mc
.gtt_size
;
827 /* Try to put vram before or after AGP because we
828 * we want SYSTEM_APERTURE to cover both VRAM and
829 * AGP so that GPU can catch out of VRAM/AGP access
831 if (rdev
->mc
.gtt_location
> rdev
->mc
.mc_vram_size
) {
832 /* Enough place before */
833 rdev
->mc
.vram_location
= rdev
->mc
.gtt_location
-
834 rdev
->mc
.mc_vram_size
;
835 } else if (tmp
> rdev
->mc
.mc_vram_size
) {
836 /* Enough place after */
837 rdev
->mc
.vram_location
= rdev
->mc
.gtt_location
+
840 /* Try to setup VRAM then AGP might not
841 * not work on some card
843 rdev
->mc
.vram_location
= 0x00000000UL
;
844 rdev
->mc
.gtt_location
= rdev
->mc
.mc_vram_size
;
847 rdev
->mc
.vram_location
= 0x00000000UL
;
848 rdev
->mc
.gtt_location
= rdev
->mc
.mc_vram_size
;
849 rdev
->mc
.gtt_size
= radeon_gart_size
* 1024 * 1024;
851 rdev
->mc
.vram_start
= rdev
->mc
.vram_location
;
852 rdev
->mc
.vram_end
= rdev
->mc
.vram_location
+ rdev
->mc
.mc_vram_size
- 1;
853 rdev
->mc
.gtt_start
= rdev
->mc
.gtt_location
;
854 rdev
->mc
.gtt_end
= rdev
->mc
.gtt_location
+ rdev
->mc
.gtt_size
- 1;
855 /* FIXME: we should enforce default clock in case GPU is not in
858 a
.full
= rfixed_const(100);
859 rdev
->pm
.sclk
.full
= rfixed_const(rdev
->clock
.default_sclk
);
860 rdev
->pm
.sclk
.full
= rfixed_div(rdev
->pm
.sclk
, a
);
863 int rv770_gpu_reset(struct radeon_device
*rdev
)
865 /* FIXME: implement any rv770 specific bits */
866 return r600_gpu_reset(rdev
);
869 static int rv770_startup(struct radeon_device
*rdev
)
873 if (!rdev
->me_fw
|| !rdev
->pfp_fw
|| !rdev
->rlc_fw
) {
874 r
= r600_init_microcode(rdev
);
876 DRM_ERROR("Failed to load firmware!\n");
881 rv770_mc_program(rdev
);
882 if (rdev
->flags
& RADEON_IS_AGP
) {
883 rv770_agp_enable(rdev
);
885 r
= rv770_pcie_gart_enable(rdev
);
889 rv770_gpu_init(rdev
);
891 if (!rdev
->r600_blit
.shader_obj
) {
892 r
= r600_blit_init(rdev
);
894 DRM_ERROR("radeon: failed blitter (%d).\n", r
);
899 r
= radeon_bo_reserve(rdev
->r600_blit
.shader_obj
, false);
900 if (unlikely(r
!= 0))
902 r
= radeon_bo_pin(rdev
->r600_blit
.shader_obj
, RADEON_GEM_DOMAIN_VRAM
,
903 &rdev
->r600_blit
.shader_gpu_addr
);
904 radeon_bo_unreserve(rdev
->r600_blit
.shader_obj
);
906 DRM_ERROR("failed to pin blit object %d\n", r
);
911 r
= r600_irq_init(rdev
);
913 DRM_ERROR("radeon: IH init failed (%d).\n", r
);
914 radeon_irq_kms_fini(rdev
);
919 r
= radeon_ring_init(rdev
, rdev
->cp
.ring_size
);
922 r
= rv770_cp_load_microcode(rdev
);
925 r
= r600_cp_resume(rdev
);
928 /* write back buffer are not vital so don't worry about failure */
929 r600_wb_enable(rdev
);
933 int rv770_resume(struct radeon_device
*rdev
)
937 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
938 * posting will perform necessary task to bring back GPU into good
942 atom_asic_init(rdev
->mode_info
.atom_context
);
943 /* Initialize clocks */
944 r
= radeon_clocks_init(rdev
);
949 r
= rv770_startup(rdev
);
951 DRM_ERROR("r600 startup failed on resume\n");
955 r
= r600_ib_test(rdev
);
957 DRM_ERROR("radeon: failled testing IB (%d).\n", r
);
964 int rv770_suspend(struct radeon_device
*rdev
)
968 /* FIXME: we should wait for ring to be empty */
970 rdev
->cp
.ready
= false;
971 r600_irq_suspend(rdev
);
972 r600_wb_disable(rdev
);
973 rv770_pcie_gart_disable(rdev
);
974 /* unpin shaders bo */
975 if (rdev
->r600_blit
.shader_obj
) {
976 r
= radeon_bo_reserve(rdev
->r600_blit
.shader_obj
, false);
977 if (likely(r
== 0)) {
978 radeon_bo_unpin(rdev
->r600_blit
.shader_obj
);
979 radeon_bo_unreserve(rdev
->r600_blit
.shader_obj
);
985 /* Plan is to move initialization in that function and use
986 * helper function so that radeon_device_init pretty much
987 * do nothing more than calling asic specific function. This
988 * should also allow to remove a bunch of callback function
991 int rv770_init(struct radeon_device
*rdev
)
995 r
= radeon_dummy_page_init(rdev
);
998 /* This don't do much */
999 r
= radeon_gem_init(rdev
);
1003 if (!radeon_get_bios(rdev
)) {
1004 if (ASIC_IS_AVIVO(rdev
))
1007 /* Must be an ATOMBIOS */
1008 if (!rdev
->is_atom_bios
) {
1009 dev_err(rdev
->dev
, "Expecting atombios for R600 GPU\n");
1012 r
= radeon_atombios_init(rdev
);
1015 /* Post card if necessary */
1016 if (!r600_card_posted(rdev
)) {
1018 dev_err(rdev
->dev
, "Card not posted and no BIOS - ignoring\n");
1021 DRM_INFO("GPU not posted. posting now...\n");
1022 atom_asic_init(rdev
->mode_info
.atom_context
);
1024 /* Initialize scratch registers */
1025 r600_scratch_init(rdev
);
1026 /* Initialize surface registers */
1027 radeon_surface_init(rdev
);
1028 /* Initialize clocks */
1029 radeon_get_clock_info(rdev
->ddev
);
1030 r
= radeon_clocks_init(rdev
);
1033 /* Initialize power management */
1034 radeon_pm_init(rdev
);
1036 r
= radeon_fence_driver_init(rdev
);
1039 if (rdev
->flags
& RADEON_IS_AGP
) {
1040 r
= radeon_agp_init(rdev
);
1042 radeon_agp_disable(rdev
);
1044 r
= rv770_mc_init(rdev
);
1047 /* Memory manager */
1048 r
= radeon_bo_init(rdev
);
1052 r
= radeon_irq_kms_init(rdev
);
1056 rdev
->cp
.ring_obj
= NULL
;
1057 r600_ring_init(rdev
, 1024 * 1024);
1059 rdev
->ih
.ring_obj
= NULL
;
1060 r600_ih_ring_init(rdev
, 64 * 1024);
1062 r
= r600_pcie_gart_init(rdev
);
1066 rdev
->accel_working
= true;
1067 r
= rv770_startup(rdev
);
1069 rv770_suspend(rdev
);
1071 radeon_ring_fini(rdev
);
1072 rv770_pcie_gart_fini(rdev
);
1073 rdev
->accel_working
= false;
1075 if (rdev
->accel_working
) {
1076 r
= radeon_ib_pool_init(rdev
);
1078 dev_err(rdev
->dev
, "IB initialization failed (%d).\n", r
);
1079 rdev
->accel_working
= false;
1081 r
= r600_ib_test(rdev
);
1083 dev_err(rdev
->dev
, "IB test failed (%d).\n", r
);
1084 rdev
->accel_working
= false;
1091 void rv770_fini(struct radeon_device
*rdev
)
1093 rv770_suspend(rdev
);
1095 r600_blit_fini(rdev
);
1096 r600_irq_fini(rdev
);
1097 radeon_irq_kms_fini(rdev
);
1098 radeon_ring_fini(rdev
);
1100 rv770_pcie_gart_fini(rdev
);
1101 radeon_gem_fini(rdev
);
1102 radeon_fence_driver_fini(rdev
);
1103 radeon_clocks_fini(rdev
);
1104 radeon_agp_fini(rdev
);
1105 radeon_bo_fini(rdev
);
1106 radeon_atombios_fini(rdev
);
1109 radeon_dummy_page_fini(rdev
);