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_share.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 /* Initialize common gart structure */
53 r
= radeon_gart_init(rdev
);
57 rdev
->gart
.table_size
= rdev
->gart
.num_gpu_pages
* 8;
58 r
= radeon_gart_table_vram_alloc(rdev
);
62 for (i
= 0; i
< rdev
->gart
.num_gpu_pages
; i
++)
63 r600_gart_clear_page(rdev
, i
);
65 WREG32(VM_L2_CNTL
, ENABLE_L2_CACHE
| ENABLE_L2_FRAGMENT_PROCESSING
|
66 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE
|
67 EFFECTIVE_L2_QUEUE_SIZE(7));
68 WREG32(VM_L2_CNTL2
, 0);
69 WREG32(VM_L2_CNTL3
, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
70 /* Setup TLB control */
71 tmp
= ENABLE_L1_TLB
| ENABLE_L1_FRAGMENT_PROCESSING
|
72 SYSTEM_ACCESS_MODE_NOT_IN_SYS
|
73 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU
|
74 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
75 WREG32(MC_VM_MD_L1_TLB0_CNTL
, tmp
);
76 WREG32(MC_VM_MD_L1_TLB1_CNTL
, tmp
);
77 WREG32(MC_VM_MD_L1_TLB2_CNTL
, tmp
);
78 WREG32(MC_VM_MB_L1_TLB0_CNTL
, tmp
);
79 WREG32(MC_VM_MB_L1_TLB1_CNTL
, tmp
);
80 WREG32(MC_VM_MB_L1_TLB2_CNTL
, tmp
);
81 WREG32(MC_VM_MB_L1_TLB3_CNTL
, tmp
);
82 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR
, rdev
->mc
.gtt_start
>> 12);
83 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR
, (rdev
->mc
.gtt_end
- 1) >> 12);
84 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR
, rdev
->gart
.table_addr
>> 12);
85 WREG32(VM_CONTEXT0_CNTL
, ENABLE_CONTEXT
| PAGE_TABLE_DEPTH(0) |
86 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT
);
87 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR
,
88 (u32
)(rdev
->dummy_page
.addr
>> 12));
89 for (i
= 1; i
< 7; i
++)
90 WREG32(VM_CONTEXT0_CNTL
+ (i
* 4), 0);
92 r600_pcie_gart_tlb_flush(rdev
);
93 rdev
->gart
.ready
= true;
97 void rv770_pcie_gart_disable(struct radeon_device
*rdev
)
103 for (i
= 0; i
< rdev
->gart
.num_gpu_pages
; i
++)
104 r600_gart_clear_page(rdev
, i
);
105 r600_pcie_gart_tlb_flush(rdev
);
106 /* Disable all tables */
107 for (i
= 0; i
< 7; i
++)
108 WREG32(VM_CONTEXT0_CNTL
+ (i
* 4), 0);
111 WREG32(VM_L2_CNTL
, ENABLE_L2_FRAGMENT_PROCESSING
|
112 EFFECTIVE_L2_QUEUE_SIZE(7));
113 WREG32(VM_L2_CNTL2
, 0);
114 WREG32(VM_L2_CNTL3
, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
115 /* Setup TLB control */
116 tmp
= EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
117 WREG32(MC_VM_MD_L1_TLB0_CNTL
, tmp
);
118 WREG32(MC_VM_MD_L1_TLB1_CNTL
, tmp
);
119 WREG32(MC_VM_MD_L1_TLB2_CNTL
, tmp
);
120 WREG32(MC_VM_MB_L1_TLB0_CNTL
, tmp
);
121 WREG32(MC_VM_MB_L1_TLB1_CNTL
, tmp
);
122 WREG32(MC_VM_MB_L1_TLB2_CNTL
, tmp
);
123 WREG32(MC_VM_MB_L1_TLB3_CNTL
, tmp
);
130 static void rv770_mc_resume(struct radeon_device
*rdev
)
132 u32 d1vga_control
, d2vga_control
;
133 u32 vga_render_control
, vga_hdp_control
;
134 u32 d1crtc_control
, d2crtc_control
;
135 u32 new_d1grph_primary
, new_d1grph_secondary
;
136 u32 new_d2grph_primary
, new_d2grph_secondary
;
142 for (i
= 0, j
= 0; i
< 32; i
++, j
+= 0x18) {
143 WREG32((0x2c14 + j
), 0x00000000);
144 WREG32((0x2c18 + j
), 0x00000000);
145 WREG32((0x2c1c + j
), 0x00000000);
146 WREG32((0x2c20 + j
), 0x00000000);
147 WREG32((0x2c24 + j
), 0x00000000);
149 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL
, 0);
151 d1vga_control
= RREG32(D1VGA_CONTROL
);
152 d2vga_control
= RREG32(D2VGA_CONTROL
);
153 vga_render_control
= RREG32(VGA_RENDER_CONTROL
);
154 vga_hdp_control
= RREG32(VGA_HDP_CONTROL
);
155 d1crtc_control
= RREG32(D1CRTC_CONTROL
);
156 d2crtc_control
= RREG32(D2CRTC_CONTROL
);
157 old_vram_start
= (u64
)(RREG32(MC_VM_FB_LOCATION
) & 0xFFFF) << 24;
158 new_d1grph_primary
= RREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS
);
159 new_d1grph_secondary
= RREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS
);
160 new_d1grph_primary
+= rdev
->mc
.vram_start
- old_vram_start
;
161 new_d1grph_secondary
+= rdev
->mc
.vram_start
- old_vram_start
;
162 new_d2grph_primary
= RREG32(D2GRPH_PRIMARY_SURFACE_ADDRESS
);
163 new_d2grph_secondary
= RREG32(D2GRPH_SECONDARY_SURFACE_ADDRESS
);
164 new_d2grph_primary
+= rdev
->mc
.vram_start
- old_vram_start
;
165 new_d2grph_secondary
+= rdev
->mc
.vram_start
- old_vram_start
;
168 WREG32(D1VGA_CONTROL
, 0);
169 WREG32(D2VGA_CONTROL
, 0);
170 WREG32(VGA_RENDER_CONTROL
, 0);
171 WREG32(D1CRTC_UPDATE_LOCK
, 1);
172 WREG32(D2CRTC_UPDATE_LOCK
, 1);
173 WREG32(D1CRTC_CONTROL
, 0);
174 WREG32(D2CRTC_CONTROL
, 0);
175 WREG32(D1CRTC_UPDATE_LOCK
, 0);
176 WREG32(D2CRTC_UPDATE_LOCK
, 0);
179 if (r600_mc_wait_for_idle(rdev
)) {
180 printk(KERN_WARNING
"[drm] MC not idle !\n");
183 /* Lockout access through VGA aperture*/
184 WREG32(VGA_HDP_CONTROL
, VGA_MEMORY_DISABLE
);
186 /* Update configuration */
187 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR
, rdev
->mc
.vram_start
>> 12);
188 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR
, (rdev
->mc
.vram_end
- 1) >> 12);
189 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR
, 0);
190 tmp
= (((rdev
->mc
.vram_end
- 1) >> 24) & 0xFFFF) << 16;
191 tmp
|= ((rdev
->mc
.vram_start
>> 24) & 0xFFFF);
192 WREG32(MC_VM_FB_LOCATION
, tmp
);
193 WREG32(HDP_NONSURFACE_BASE
, (rdev
->mc
.vram_start
>> 8));
194 WREG32(HDP_NONSURFACE_INFO
, (2 << 7));
195 WREG32(HDP_NONSURFACE_SIZE
, (rdev
->mc
.mc_vram_size
- 1) | 0x3FF);
196 if (rdev
->flags
& RADEON_IS_AGP
) {
197 WREG32(MC_VM_AGP_TOP
, (rdev
->mc
.gtt_end
- 1) >> 16);
198 WREG32(MC_VM_AGP_BOT
, rdev
->mc
.gtt_start
>> 16);
199 WREG32(MC_VM_AGP_BASE
, rdev
->mc
.agp_base
>> 22);
201 WREG32(MC_VM_AGP_BASE
, 0);
202 WREG32(MC_VM_AGP_TOP
, 0x0FFFFFFF);
203 WREG32(MC_VM_AGP_BOT
, 0x0FFFFFFF);
205 WREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS
, new_d1grph_primary
);
206 WREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS
, new_d1grph_secondary
);
207 WREG32(D2GRPH_PRIMARY_SURFACE_ADDRESS
, new_d2grph_primary
);
208 WREG32(D2GRPH_SECONDARY_SURFACE_ADDRESS
, new_d2grph_secondary
);
209 WREG32(VGA_MEMORY_BASE_ADDRESS
, rdev
->mc
.vram_start
);
211 /* Unlock host access */
212 WREG32(VGA_HDP_CONTROL
, vga_hdp_control
);
215 if (r600_mc_wait_for_idle(rdev
)) {
216 printk(KERN_WARNING
"[drm] MC not idle !\n");
219 /* Restore video state */
220 WREG32(D1CRTC_UPDATE_LOCK
, 1);
221 WREG32(D2CRTC_UPDATE_LOCK
, 1);
222 WREG32(D1CRTC_CONTROL
, d1crtc_control
);
223 WREG32(D2CRTC_CONTROL
, d2crtc_control
);
224 WREG32(D1CRTC_UPDATE_LOCK
, 0);
225 WREG32(D2CRTC_UPDATE_LOCK
, 0);
226 WREG32(D1VGA_CONTROL
, d1vga_control
);
227 WREG32(D2VGA_CONTROL
, d2vga_control
);
228 WREG32(VGA_RENDER_CONTROL
, vga_render_control
);
235 void r700_cp_stop(struct radeon_device
*rdev
)
237 WREG32(CP_ME_CNTL
, (CP_ME_HALT
| CP_PFP_HALT
));
241 static int rv770_cp_load_microcode(struct radeon_device
*rdev
)
243 const __be32
*fw_data
;
246 if (!rdev
->me_fw
|| !rdev
->pfp_fw
)
250 WREG32(CP_RB_CNTL
, RB_NO_UPDATE
| (15 << 8) | (3 << 0));
253 WREG32(GRBM_SOFT_RESET
, SOFT_RESET_CP
);
254 RREG32(GRBM_SOFT_RESET
);
256 WREG32(GRBM_SOFT_RESET
, 0);
258 fw_data
= (const __be32
*)rdev
->pfp_fw
->data
;
259 WREG32(CP_PFP_UCODE_ADDR
, 0);
260 for (i
= 0; i
< R700_PFP_UCODE_SIZE
; i
++)
261 WREG32(CP_PFP_UCODE_DATA
, be32_to_cpup(fw_data
++));
262 WREG32(CP_PFP_UCODE_ADDR
, 0);
264 fw_data
= (const __be32
*)rdev
->me_fw
->data
;
265 WREG32(CP_ME_RAM_WADDR
, 0);
266 for (i
= 0; i
< R700_PM4_UCODE_SIZE
; i
++)
267 WREG32(CP_ME_RAM_DATA
, be32_to_cpup(fw_data
++));
269 WREG32(CP_PFP_UCODE_ADDR
, 0);
270 WREG32(CP_ME_RAM_WADDR
, 0);
271 WREG32(CP_ME_RAM_RADDR
, 0);
279 static u32
r700_get_tile_pipe_to_backend_map(u32 num_tile_pipes
,
281 u32 backend_disable_mask
)
284 u32 enabled_backends_mask
;
285 u32 enabled_backends_count
;
287 u32 swizzle_pipe
[R7XX_MAX_PIPES
];
291 if (num_tile_pipes
> R7XX_MAX_PIPES
)
292 num_tile_pipes
= R7XX_MAX_PIPES
;
293 if (num_tile_pipes
< 1)
295 if (num_backends
> R7XX_MAX_BACKENDS
)
296 num_backends
= R7XX_MAX_BACKENDS
;
297 if (num_backends
< 1)
300 enabled_backends_mask
= 0;
301 enabled_backends_count
= 0;
302 for (i
= 0; i
< R7XX_MAX_BACKENDS
; ++i
) {
303 if (((backend_disable_mask
>> i
) & 1) == 0) {
304 enabled_backends_mask
|= (1 << i
);
305 ++enabled_backends_count
;
307 if (enabled_backends_count
== num_backends
)
311 if (enabled_backends_count
== 0) {
312 enabled_backends_mask
= 1;
313 enabled_backends_count
= 1;
316 if (enabled_backends_count
!= num_backends
)
317 num_backends
= enabled_backends_count
;
319 memset((uint8_t *)&swizzle_pipe
[0], 0, sizeof(u32
) * R7XX_MAX_PIPES
);
320 switch (num_tile_pipes
) {
376 for (cur_pipe
= 0; cur_pipe
< num_tile_pipes
; ++cur_pipe
) {
377 while (((1 << cur_backend
) & enabled_backends_mask
) == 0)
378 cur_backend
= (cur_backend
+ 1) % R7XX_MAX_BACKENDS
;
380 backend_map
|= (u32
)(((cur_backend
& 3) << (swizzle_pipe
[cur_pipe
] * 2)));
382 cur_backend
= (cur_backend
+ 1) % R7XX_MAX_BACKENDS
;
388 static void rv770_gpu_init(struct radeon_device
*rdev
)
390 int i
, j
, num_qd_pipes
;
393 u32 num_gs_verts_per_thread
;
395 u32 gs_prim_buffer_depth
= 0;
396 u32 sq_ms_fifo_sizes
;
398 u32 sq_thread_resource_mgmt
;
399 u32 hdp_host_path_cntl
;
400 u32 sq_dyn_gpr_size_simd_ab_0
;
402 u32 gb_tiling_config
= 0;
403 u32 cc_rb_backend_disable
= 0;
404 u32 cc_gc_shader_pipe_config
= 0;
408 /* setup chip specs */
409 switch (rdev
->family
) {
411 rdev
->config
.rv770
.max_pipes
= 4;
412 rdev
->config
.rv770
.max_tile_pipes
= 8;
413 rdev
->config
.rv770
.max_simds
= 10;
414 rdev
->config
.rv770
.max_backends
= 4;
415 rdev
->config
.rv770
.max_gprs
= 256;
416 rdev
->config
.rv770
.max_threads
= 248;
417 rdev
->config
.rv770
.max_stack_entries
= 512;
418 rdev
->config
.rv770
.max_hw_contexts
= 8;
419 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
420 rdev
->config
.rv770
.sx_max_export_size
= 128;
421 rdev
->config
.rv770
.sx_max_export_pos_size
= 16;
422 rdev
->config
.rv770
.sx_max_export_smx_size
= 112;
423 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
425 rdev
->config
.rv770
.sx_num_of_sets
= 7;
426 rdev
->config
.rv770
.sc_prim_fifo_size
= 0xF9;
427 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
428 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
431 rdev
->config
.rv770
.max_pipes
= 2;
432 rdev
->config
.rv770
.max_tile_pipes
= 4;
433 rdev
->config
.rv770
.max_simds
= 8;
434 rdev
->config
.rv770
.max_backends
= 2;
435 rdev
->config
.rv770
.max_gprs
= 128;
436 rdev
->config
.rv770
.max_threads
= 248;
437 rdev
->config
.rv770
.max_stack_entries
= 256;
438 rdev
->config
.rv770
.max_hw_contexts
= 8;
439 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
440 rdev
->config
.rv770
.sx_max_export_size
= 256;
441 rdev
->config
.rv770
.sx_max_export_pos_size
= 32;
442 rdev
->config
.rv770
.sx_max_export_smx_size
= 224;
443 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
445 rdev
->config
.rv770
.sx_num_of_sets
= 7;
446 rdev
->config
.rv770
.sc_prim_fifo_size
= 0xf9;
447 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
448 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
449 if (rdev
->config
.rv770
.sx_max_export_pos_size
> 16) {
450 rdev
->config
.rv770
.sx_max_export_pos_size
-= 16;
451 rdev
->config
.rv770
.sx_max_export_smx_size
+= 16;
455 rdev
->config
.rv770
.max_pipes
= 2;
456 rdev
->config
.rv770
.max_tile_pipes
= 2;
457 rdev
->config
.rv770
.max_simds
= 2;
458 rdev
->config
.rv770
.max_backends
= 1;
459 rdev
->config
.rv770
.max_gprs
= 256;
460 rdev
->config
.rv770
.max_threads
= 192;
461 rdev
->config
.rv770
.max_stack_entries
= 256;
462 rdev
->config
.rv770
.max_hw_contexts
= 4;
463 rdev
->config
.rv770
.max_gs_threads
= 8 * 2;
464 rdev
->config
.rv770
.sx_max_export_size
= 128;
465 rdev
->config
.rv770
.sx_max_export_pos_size
= 16;
466 rdev
->config
.rv770
.sx_max_export_smx_size
= 112;
467 rdev
->config
.rv770
.sq_num_cf_insts
= 1;
469 rdev
->config
.rv770
.sx_num_of_sets
= 7;
470 rdev
->config
.rv770
.sc_prim_fifo_size
= 0x40;
471 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
472 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
475 rdev
->config
.rv770
.max_pipes
= 4;
476 rdev
->config
.rv770
.max_tile_pipes
= 4;
477 rdev
->config
.rv770
.max_simds
= 8;
478 rdev
->config
.rv770
.max_backends
= 4;
479 rdev
->config
.rv770
.max_gprs
= 256;
480 rdev
->config
.rv770
.max_threads
= 248;
481 rdev
->config
.rv770
.max_stack_entries
= 512;
482 rdev
->config
.rv770
.max_hw_contexts
= 8;
483 rdev
->config
.rv770
.max_gs_threads
= 16 * 2;
484 rdev
->config
.rv770
.sx_max_export_size
= 256;
485 rdev
->config
.rv770
.sx_max_export_pos_size
= 32;
486 rdev
->config
.rv770
.sx_max_export_smx_size
= 224;
487 rdev
->config
.rv770
.sq_num_cf_insts
= 2;
489 rdev
->config
.rv770
.sx_num_of_sets
= 7;
490 rdev
->config
.rv770
.sc_prim_fifo_size
= 0x100;
491 rdev
->config
.rv770
.sc_hiz_tile_fifo_size
= 0x30;
492 rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
= 0x130;
494 if (rdev
->config
.rv770
.sx_max_export_pos_size
> 16) {
495 rdev
->config
.rv770
.sx_max_export_pos_size
-= 16;
496 rdev
->config
.rv770
.sx_max_export_smx_size
+= 16;
505 for (i
= 0; i
< 32; i
++) {
506 WREG32((0x2c14 + j
), 0x00000000);
507 WREG32((0x2c18 + j
), 0x00000000);
508 WREG32((0x2c1c + j
), 0x00000000);
509 WREG32((0x2c20 + j
), 0x00000000);
510 WREG32((0x2c24 + j
), 0x00000000);
514 WREG32(GRBM_CNTL
, GRBM_READ_TIMEOUT(0xff));
516 /* setup tiling, simd, pipe config */
517 mc_arb_ramcfg
= RREG32(MC_ARB_RAMCFG
);
519 switch (rdev
->config
.rv770
.max_tile_pipes
) {
521 gb_tiling_config
|= PIPE_TILING(0);
524 gb_tiling_config
|= PIPE_TILING(1);
527 gb_tiling_config
|= PIPE_TILING(2);
530 gb_tiling_config
|= PIPE_TILING(3);
536 if (rdev
->family
== CHIP_RV770
)
537 gb_tiling_config
|= BANK_TILING(1);
539 gb_tiling_config
|= BANK_TILING((mc_arb_ramcfg
& NOOFBANK_SHIFT
) >> NOOFBANK_MASK
);
541 gb_tiling_config
|= GROUP_SIZE(0);
543 if (((mc_arb_ramcfg
& NOOFROWS_MASK
) & NOOFROWS_SHIFT
) > 3) {
544 gb_tiling_config
|= ROW_TILING(3);
545 gb_tiling_config
|= SAMPLE_SPLIT(3);
548 ROW_TILING(((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
));
550 SAMPLE_SPLIT(((mc_arb_ramcfg
& NOOFROWS_MASK
) >> NOOFROWS_SHIFT
));
553 gb_tiling_config
|= BANK_SWAPS(1);
555 backend_map
= r700_get_tile_pipe_to_backend_map(rdev
->config
.rv770
.max_tile_pipes
,
556 rdev
->config
.rv770
.max_backends
,
557 (0xff << rdev
->config
.rv770
.max_backends
) & 0xff);
558 gb_tiling_config
|= BACKEND_MAP(backend_map
);
560 cc_gc_shader_pipe_config
=
561 INACTIVE_QD_PIPES((R7XX_MAX_PIPES_MASK
<< rdev
->config
.rv770
.max_pipes
) & R7XX_MAX_PIPES_MASK
);
562 cc_gc_shader_pipe_config
|=
563 INACTIVE_SIMDS((R7XX_MAX_SIMDS_MASK
<< rdev
->config
.rv770
.max_simds
) & R7XX_MAX_SIMDS_MASK
);
565 cc_rb_backend_disable
=
566 BACKEND_DISABLE((R7XX_MAX_BACKENDS_MASK
<< rdev
->config
.rv770
.max_backends
) & R7XX_MAX_BACKENDS_MASK
);
568 WREG32(GB_TILING_CONFIG
, gb_tiling_config
);
569 WREG32(DCP_TILING_CONFIG
, (gb_tiling_config
& 0xffff));
570 WREG32(HDP_TILING_CONFIG
, (gb_tiling_config
& 0xffff));
572 WREG32(CC_RB_BACKEND_DISABLE
, cc_rb_backend_disable
);
573 WREG32(CC_GC_SHADER_PIPE_CONFIG
, cc_gc_shader_pipe_config
);
574 WREG32(GC_USER_SHADER_PIPE_CONFIG
, cc_gc_shader_pipe_config
);
576 WREG32(CC_SYS_RB_BACKEND_DISABLE
, cc_rb_backend_disable
);
577 WREG32(CGTS_SYS_TCC_DISABLE
, 0);
578 WREG32(CGTS_TCC_DISABLE
, 0);
579 WREG32(CGTS_USER_SYS_TCC_DISABLE
, 0);
580 WREG32(CGTS_USER_TCC_DISABLE
, 0);
583 R7XX_MAX_BACKENDS
- r600_count_pipe_bits(cc_gc_shader_pipe_config
& INACTIVE_QD_PIPES_MASK
);
584 WREG32(VGT_OUT_DEALLOC_CNTL
, (num_qd_pipes
* 4) & DEALLOC_DIST_MASK
);
585 WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL
, ((num_qd_pipes
* 4) - 2) & VTX_REUSE_DEPTH_MASK
);
587 /* set HW defaults for 3D engine */
588 WREG32(CP_QUEUE_THRESHOLDS
, (ROQ_IB1_START(0x16) |
589 ROQ_IB2_START(0x2b)));
591 WREG32(CP_MEQ_THRESHOLDS
, STQ_SPLIT(0x30));
593 WREG32(TA_CNTL_AUX
, (DISABLE_CUBE_ANISO
|
598 sx_debug_1
= RREG32(SX_DEBUG_1
);
599 sx_debug_1
|= ENABLE_NEW_SMX_ADDRESS
;
600 WREG32(SX_DEBUG_1
, sx_debug_1
);
602 smx_dc_ctl0
= RREG32(SMX_DC_CTL0
);
603 smx_dc_ctl0
&= ~CACHE_DEPTH(0x1ff);
604 smx_dc_ctl0
|= CACHE_DEPTH((rdev
->config
.rv770
.sx_num_of_sets
* 64) - 1);
605 WREG32(SMX_DC_CTL0
, smx_dc_ctl0
);
607 WREG32(SMX_EVENT_CTL
, (ES_FLUSH_CTL(4) |
612 if (rdev
->family
== CHIP_RV770
)
613 WREG32(DB_DEBUG3
, DB_CLK_OFF_DELAY(0x1f));
615 db_debug4
= RREG32(DB_DEBUG4
);
616 db_debug4
|= DISABLE_TILE_COVERED_FOR_PS_ITER
;
617 WREG32(DB_DEBUG4
, db_debug4
);
620 WREG32(SX_EXPORT_BUFFER_SIZES
, (COLOR_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_size
/ 4) - 1) |
621 POSITION_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_pos_size
/ 4) - 1) |
622 SMX_BUFFER_SIZE((rdev
->config
.rv770
.sx_max_export_smx_size
/ 4) - 1)));
624 WREG32(PA_SC_FIFO_SIZE
, (SC_PRIM_FIFO_SIZE(rdev
->config
.rv770
.sc_prim_fifo_size
) |
625 SC_HIZ_TILE_FIFO_SIZE(rdev
->config
.rv770
.sc_hiz_tile_fifo_size
) |
626 SC_EARLYZ_TILE_FIFO_SIZE(rdev
->config
.rv770
.sc_earlyz_tile_fifo_fize
)));
628 WREG32(PA_SC_MULTI_CHIP_CNTL
, 0);
630 WREG32(VGT_NUM_INSTANCES
, 1);
632 WREG32(SPI_CONFIG_CNTL
, GPR_WRITE_PRIORITY(0));
634 WREG32(SPI_CONFIG_CNTL_1
, VTX_DONE_DELAY(4));
636 WREG32(CP_PERFMON_CNTL
, 0);
638 sq_ms_fifo_sizes
= (CACHE_FIFO_SIZE(16 * rdev
->config
.rv770
.sq_num_cf_insts
) |
639 DONE_FIFO_HIWATER(0xe0) |
640 ALU_UPDATE_FIFO_HIWATER(0x8));
641 switch (rdev
->family
) {
643 sq_ms_fifo_sizes
|= FETCH_FIFO_HIWATER(0x1);
649 sq_ms_fifo_sizes
|= FETCH_FIFO_HIWATER(0x4);
652 WREG32(SQ_MS_FIFO_SIZES
, sq_ms_fifo_sizes
);
654 /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
655 * should be adjusted as needed by the 2D/3D drivers. This just sets default values
657 sq_config
= RREG32(SQ_CONFIG
);
658 sq_config
&= ~(PS_PRIO(3) |
662 sq_config
|= (DX9_CONSTS
|
669 if (rdev
->family
== CHIP_RV710
)
670 /* no vertex cache */
671 sq_config
&= ~VC_ENABLE
;
673 WREG32(SQ_CONFIG
, sq_config
);
675 WREG32(SQ_GPR_RESOURCE_MGMT_1
, (NUM_PS_GPRS((rdev
->config
.rv770
.max_gprs
* 24)/64) |
676 NUM_VS_GPRS((rdev
->config
.rv770
.max_gprs
* 24)/64) |
677 NUM_CLAUSE_TEMP_GPRS(((rdev
->config
.rv770
.max_gprs
* 24)/64)/2)));
679 WREG32(SQ_GPR_RESOURCE_MGMT_2
, (NUM_GS_GPRS((rdev
->config
.rv770
.max_gprs
* 7)/64) |
680 NUM_ES_GPRS((rdev
->config
.rv770
.max_gprs
* 7)/64)));
682 sq_thread_resource_mgmt
= (NUM_PS_THREADS((rdev
->config
.rv770
.max_threads
* 4)/8) |
683 NUM_VS_THREADS((rdev
->config
.rv770
.max_threads
* 2)/8) |
684 NUM_ES_THREADS((rdev
->config
.rv770
.max_threads
* 1)/8));
685 if (((rdev
->config
.rv770
.max_threads
* 1) / 8) > rdev
->config
.rv770
.max_gs_threads
)
686 sq_thread_resource_mgmt
|= NUM_GS_THREADS(rdev
->config
.rv770
.max_gs_threads
);
688 sq_thread_resource_mgmt
|= NUM_GS_THREADS((rdev
->config
.rv770
.max_gs_threads
* 1)/8);
689 WREG32(SQ_THREAD_RESOURCE_MGMT
, sq_thread_resource_mgmt
);
691 WREG32(SQ_STACK_RESOURCE_MGMT_1
, (NUM_PS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4) |
692 NUM_VS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4)));
694 WREG32(SQ_STACK_RESOURCE_MGMT_2
, (NUM_GS_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4) |
695 NUM_ES_STACK_ENTRIES((rdev
->config
.rv770
.max_stack_entries
* 1)/4)));
697 sq_dyn_gpr_size_simd_ab_0
= (SIMDA_RING0((rdev
->config
.rv770
.max_gprs
* 38)/64) |
698 SIMDA_RING1((rdev
->config
.rv770
.max_gprs
* 38)/64) |
699 SIMDB_RING0((rdev
->config
.rv770
.max_gprs
* 38)/64) |
700 SIMDB_RING1((rdev
->config
.rv770
.max_gprs
* 38)/64));
702 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_0
, sq_dyn_gpr_size_simd_ab_0
);
703 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_1
, sq_dyn_gpr_size_simd_ab_0
);
704 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_2
, sq_dyn_gpr_size_simd_ab_0
);
705 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_3
, sq_dyn_gpr_size_simd_ab_0
);
706 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_4
, sq_dyn_gpr_size_simd_ab_0
);
707 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_5
, sq_dyn_gpr_size_simd_ab_0
);
708 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_6
, sq_dyn_gpr_size_simd_ab_0
);
709 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_7
, sq_dyn_gpr_size_simd_ab_0
);
711 WREG32(PA_SC_FORCE_EOV_MAX_CNTS
, (FORCE_EOV_MAX_CLK_CNT(4095) |
712 FORCE_EOV_MAX_REZ_CNT(255)));
714 if (rdev
->family
== CHIP_RV710
)
715 WREG32(VGT_CACHE_INVALIDATION
, (CACHE_INVALIDATION(TC_ONLY
) |
716 AUTO_INVLD_EN(ES_AND_GS_AUTO
)));
718 WREG32(VGT_CACHE_INVALIDATION
, (CACHE_INVALIDATION(VC_AND_TC
) |
719 AUTO_INVLD_EN(ES_AND_GS_AUTO
)));
721 switch (rdev
->family
) {
725 gs_prim_buffer_depth
= 384;
728 gs_prim_buffer_depth
= 128;
734 num_gs_verts_per_thread
= rdev
->config
.rv770
.max_pipes
* 16;
735 vgt_gs_per_es
= gs_prim_buffer_depth
+ num_gs_verts_per_thread
;
736 /* Max value for this is 256 */
737 if (vgt_gs_per_es
> 256)
740 WREG32(VGT_ES_PER_GS
, 128);
741 WREG32(VGT_GS_PER_ES
, vgt_gs_per_es
);
742 WREG32(VGT_GS_PER_VS
, 2);
744 /* more default values. 2D/3D driver should adjust as needed */
745 WREG32(VGT_GS_VERTEX_REUSE
, 16);
746 WREG32(PA_SC_LINE_STIPPLE_STATE
, 0);
747 WREG32(VGT_STRMOUT_EN
, 0);
749 WREG32(PA_SC_MODE_CNTL
, 0);
750 WREG32(PA_SC_EDGERULE
, 0xaaaaaaaa);
751 WREG32(PA_SC_AA_CONFIG
, 0);
752 WREG32(PA_SC_CLIPRECT_RULE
, 0xffff);
753 WREG32(PA_SC_LINE_STIPPLE
, 0);
754 WREG32(SPI_INPUT_Z
, 0);
755 WREG32(SPI_PS_IN_CONTROL_0
, NUM_INTERP(2));
756 WREG32(CB_COLOR7_FRAG
, 0);
758 /* clear render buffer base addresses */
759 WREG32(CB_COLOR0_BASE
, 0);
760 WREG32(CB_COLOR1_BASE
, 0);
761 WREG32(CB_COLOR2_BASE
, 0);
762 WREG32(CB_COLOR3_BASE
, 0);
763 WREG32(CB_COLOR4_BASE
, 0);
764 WREG32(CB_COLOR5_BASE
, 0);
765 WREG32(CB_COLOR6_BASE
, 0);
766 WREG32(CB_COLOR7_BASE
, 0);
770 hdp_host_path_cntl
= RREG32(HDP_HOST_PATH_CNTL
);
771 WREG32(HDP_HOST_PATH_CNTL
, hdp_host_path_cntl
);
773 WREG32(PA_SC_MULTI_CHIP_CNTL
, 0);
775 WREG32(PA_CL_ENHANCE
, (CLIP_VTX_REORDER_ENA
|
780 int rv770_mc_init(struct radeon_device
*rdev
)
786 /* Get VRAM informations */
787 /* FIXME: Don't know how to determine vram width, need to check
790 rdev
->mc
.vram_width
= 128;
791 rdev
->mc
.vram_is_ddr
= true;
792 /* Could aper size report 0 ? */
793 rdev
->mc
.aper_base
= drm_get_resource_start(rdev
->ddev
, 0);
794 rdev
->mc
.aper_size
= drm_get_resource_len(rdev
->ddev
, 0);
795 /* Setup GPU memory space */
796 rdev
->mc
.mc_vram_size
= RREG32(CONFIG_MEMSIZE
);
797 rdev
->mc
.real_vram_size
= RREG32(CONFIG_MEMSIZE
);
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
;
831 rdev
->mc
.gtt_start
= rdev
->mc
.gtt_location
;
832 rdev
->mc
.gtt_end
= rdev
->mc
.gtt_location
+ rdev
->mc
.gtt_size
;
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 */
847 int rv770_resume(struct radeon_device
*rdev
)
851 rv770_mc_resume(rdev
);
852 r
= rv770_pcie_gart_enable(rdev
);
855 rv770_gpu_init(rdev
);
856 r
= radeon_ring_init(rdev
, rdev
->cp
.ring_size
);
859 r
= rv770_cp_load_microcode(rdev
);
862 r
= r600_cp_resume(rdev
);
865 r
= r600_wb_init(rdev
);
871 int rv770_suspend(struct radeon_device
*rdev
)
873 /* FIXME: we should wait for ring to be empty */
878 /* Plan is to move initialization in that function and use
879 * helper function so that radeon_device_init pretty much
880 * do nothing more than calling asic specific function. This
881 * should also allow to remove a bunch of callback function
884 int rv770_init(struct radeon_device
*rdev
)
888 rdev
->new_init_path
= true;
889 r
= radeon_dummy_page_init(rdev
);
892 /* This don't do much */
893 r
= radeon_gem_init(rdev
);
897 if (!radeon_get_bios(rdev
)) {
898 if (ASIC_IS_AVIVO(rdev
))
901 /* Must be an ATOMBIOS */
902 if (!rdev
->is_atom_bios
)
904 r
= radeon_atombios_init(rdev
);
907 /* Post card if necessary */
908 if (!r600_card_posted(rdev
) && rdev
->bios
) {
909 DRM_INFO("GPU not posted. posting now...\n");
910 atom_asic_init(rdev
->mode_info
.atom_context
);
912 /* Initialize scratch registers */
913 r600_scratch_init(rdev
);
914 /* Initialize surface registers */
915 radeon_surface_init(rdev
);
916 r
= radeon_clocks_init(rdev
);
920 r
= radeon_fence_driver_init(rdev
);
923 r
= rv770_mc_init(rdev
);
925 if (rdev
->flags
& RADEON_IS_AGP
) {
926 /* Retry with disabling AGP */
928 rdev
->flags
&= ~RADEON_IS_AGP
;
929 return rv770_init(rdev
);
934 r
= radeon_object_init(rdev
);
937 rdev
->cp
.ring_obj
= NULL
;
938 r600_ring_init(rdev
, 1024 * 1024);
940 if (!rdev
->me_fw
|| !rdev
->pfp_fw
) {
941 r
= r600_cp_init_microcode(rdev
);
943 DRM_ERROR("Failed to load firmware!\n");
948 r
= rv770_resume(rdev
);
950 if (rdev
->flags
& RADEON_IS_AGP
) {
951 /* Retry with disabling AGP */
953 rdev
->flags
&= ~RADEON_IS_AGP
;
954 return rv770_init(rdev
);
958 r
= r600_blit_init(rdev
);
960 DRM_ERROR("radeon: failled blitter (%d).\n", r
);
963 r
= radeon_ib_pool_init(rdev
);
965 DRM_ERROR("radeon: failled initializing IB pool (%d).\n", r
);
968 r
= radeon_ib_test(rdev
);
970 DRM_ERROR("radeon: failled testing IB (%d).\n", r
);
976 void rv770_fini(struct radeon_device
*rdev
)
978 r600_blit_fini(rdev
);
979 radeon_ring_fini(rdev
);
980 rv770_pcie_gart_disable(rdev
);
981 radeon_gart_table_vram_free(rdev
);
982 radeon_gart_fini(rdev
);
983 radeon_gem_fini(rdev
);
984 radeon_fence_driver_fini(rdev
);
985 radeon_clocks_fini(rdev
);
987 if (rdev
->flags
& RADEON_IS_AGP
)
988 radeon_agp_fini(rdev
);
990 radeon_object_fini(rdev
);
991 if (rdev
->is_atom_bios
) {
992 radeon_atombios_fini(rdev
);
994 radeon_combios_fini(rdev
);
998 radeon_dummy_page_fini(rdev
);