2 * Copyright 2014 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 #include <linux/pci.h>
28 #include <drm/drm_cache.h>
31 #include "amdgpu_ucode.h"
32 #include "amdgpu_gem.h"
34 #include "bif/bif_3_0_d.h"
35 #include "bif/bif_3_0_sh_mask.h"
36 #include "oss/oss_1_0_d.h"
37 #include "oss/oss_1_0_sh_mask.h"
38 #include "gmc/gmc_6_0_d.h"
39 #include "gmc/gmc_6_0_sh_mask.h"
40 #include "dce/dce_6_0_d.h"
41 #include "dce/dce_6_0_sh_mask.h"
44 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device
*adev
);
45 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device
*adev
);
46 static int gmc_v6_0_wait_for_idle(void *handle
);
48 MODULE_FIRMWARE("amdgpu/tahiti_mc.bin");
49 MODULE_FIRMWARE("amdgpu/pitcairn_mc.bin");
50 MODULE_FIRMWARE("amdgpu/verde_mc.bin");
51 MODULE_FIRMWARE("amdgpu/oland_mc.bin");
52 MODULE_FIRMWARE("amdgpu/hainan_mc.bin");
53 MODULE_FIRMWARE("amdgpu/si58_mc.bin");
55 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
56 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000
57 #define MC_SEQ_MISC0__MT__DDR2 0x20000000
58 #define MC_SEQ_MISC0__MT__GDDR3 0x30000000
59 #define MC_SEQ_MISC0__MT__GDDR4 0x40000000
60 #define MC_SEQ_MISC0__MT__GDDR5 0x50000000
61 #define MC_SEQ_MISC0__MT__HBM 0x60000000
62 #define MC_SEQ_MISC0__MT__DDR3 0xB0000000
65 static const u32 crtc_offsets
[6] =
67 SI_CRTC0_REGISTER_OFFSET
,
68 SI_CRTC1_REGISTER_OFFSET
,
69 SI_CRTC2_REGISTER_OFFSET
,
70 SI_CRTC3_REGISTER_OFFSET
,
71 SI_CRTC4_REGISTER_OFFSET
,
72 SI_CRTC5_REGISTER_OFFSET
75 static void gmc_v6_0_mc_stop(struct amdgpu_device
*adev
)
79 gmc_v6_0_wait_for_idle((void *)adev
);
81 blackout
= RREG32(mmMC_SHARED_BLACKOUT_CNTL
);
82 if (REG_GET_FIELD(blackout
, MC_SHARED_BLACKOUT_CNTL
, BLACKOUT_MODE
) != 1) {
83 /* Block CPU access */
84 WREG32(mmBIF_FB_EN
, 0);
86 blackout
= REG_SET_FIELD(blackout
,
87 MC_SHARED_BLACKOUT_CNTL
, BLACKOUT_MODE
, 0);
88 WREG32(mmMC_SHARED_BLACKOUT_CNTL
, blackout
| 1);
90 /* wait for the MC to settle */
95 static void gmc_v6_0_mc_resume(struct amdgpu_device
*adev
)
99 /* unblackout the MC */
100 tmp
= RREG32(mmMC_SHARED_BLACKOUT_CNTL
);
101 tmp
= REG_SET_FIELD(tmp
, MC_SHARED_BLACKOUT_CNTL
, BLACKOUT_MODE
, 0);
102 WREG32(mmMC_SHARED_BLACKOUT_CNTL
, tmp
);
103 /* allow CPU access */
104 tmp
= REG_SET_FIELD(0, BIF_FB_EN
, FB_READ_EN
, 1);
105 tmp
= REG_SET_FIELD(tmp
, BIF_FB_EN
, FB_WRITE_EN
, 1);
106 WREG32(mmBIF_FB_EN
, tmp
);
109 static int gmc_v6_0_init_microcode(struct amdgpu_device
*adev
)
111 const char *chip_name
;
114 bool is_58_fw
= false;
118 switch (adev
->asic_type
) {
120 chip_name
= "tahiti";
123 chip_name
= "pitcairn";
132 chip_name
= "hainan";
137 /* this memory configuration requires special firmware */
138 if (((RREG32(mmMC_SEQ_MISC0
) & 0xff000000) >> 24) == 0x58)
142 snprintf(fw_name
, sizeof(fw_name
), "amdgpu/si58_mc.bin");
144 snprintf(fw_name
, sizeof(fw_name
), "amdgpu/%s_mc.bin", chip_name
);
145 err
= request_firmware(&adev
->gmc
.fw
, fw_name
, adev
->dev
);
149 err
= amdgpu_ucode_validate(adev
->gmc
.fw
);
154 "si_mc: Failed to load firmware \"%s\"\n",
156 release_firmware(adev
->gmc
.fw
);
162 static int gmc_v6_0_mc_load_microcode(struct amdgpu_device
*adev
)
164 const __le32
*new_fw_data
= NULL
;
166 const __le32
*new_io_mc_regs
= NULL
;
167 int i
, regs_size
, ucode_size
;
168 const struct mc_firmware_header_v1_0
*hdr
;
173 hdr
= (const struct mc_firmware_header_v1_0
*)adev
->gmc
.fw
->data
;
175 amdgpu_ucode_print_mc_hdr(&hdr
->header
);
177 adev
->gmc
.fw_version
= le32_to_cpu(hdr
->header
.ucode_version
);
178 regs_size
= le32_to_cpu(hdr
->io_debug_size_bytes
) / (4 * 2);
179 new_io_mc_regs
= (const __le32
*)
180 (adev
->gmc
.fw
->data
+ le32_to_cpu(hdr
->io_debug_array_offset_bytes
));
181 ucode_size
= le32_to_cpu(hdr
->header
.ucode_size_bytes
) / 4;
182 new_fw_data
= (const __le32
*)
183 (adev
->gmc
.fw
->data
+ le32_to_cpu(hdr
->header
.ucode_array_offset_bytes
));
185 running
= RREG32(mmMC_SEQ_SUP_CNTL
) & MC_SEQ_SUP_CNTL__RUN_MASK
;
189 /* reset the engine and set to writable */
190 WREG32(mmMC_SEQ_SUP_CNTL
, 0x00000008);
191 WREG32(mmMC_SEQ_SUP_CNTL
, 0x00000010);
193 /* load mc io regs */
194 for (i
= 0; i
< regs_size
; i
++) {
195 WREG32(mmMC_SEQ_IO_DEBUG_INDEX
, le32_to_cpup(new_io_mc_regs
++));
196 WREG32(mmMC_SEQ_IO_DEBUG_DATA
, le32_to_cpup(new_io_mc_regs
++));
198 /* load the MC ucode */
199 for (i
= 0; i
< ucode_size
; i
++) {
200 WREG32(mmMC_SEQ_SUP_PGM
, le32_to_cpup(new_fw_data
++));
203 /* put the engine back into the active state */
204 WREG32(mmMC_SEQ_SUP_CNTL
, 0x00000008);
205 WREG32(mmMC_SEQ_SUP_CNTL
, 0x00000004);
206 WREG32(mmMC_SEQ_SUP_CNTL
, 0x00000001);
208 /* wait for training to complete */
209 for (i
= 0; i
< adev
->usec_timeout
; i
++) {
210 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL
) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D0_MASK
)
214 for (i
= 0; i
< adev
->usec_timeout
; i
++) {
215 if (RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL
) & MC_SEQ_TRAIN_WAKEUP_CNTL__TRAIN_DONE_D1_MASK
)
225 static void gmc_v6_0_vram_gtt_location(struct amdgpu_device
*adev
,
226 struct amdgpu_gmc
*mc
)
228 u64 base
= RREG32(mmMC_VM_FB_LOCATION
) & 0xFFFF;
231 amdgpu_gmc_vram_location(adev
, mc
, base
);
232 amdgpu_gmc_gart_location(adev
, mc
);
235 static void gmc_v6_0_mc_program(struct amdgpu_device
*adev
)
240 for (i
= 0, j
= 0; i
< 32; i
++, j
+= 0x6) {
241 WREG32((0xb05 + j
), 0x00000000);
242 WREG32((0xb06 + j
), 0x00000000);
243 WREG32((0xb07 + j
), 0x00000000);
244 WREG32((0xb08 + j
), 0x00000000);
245 WREG32((0xb09 + j
), 0x00000000);
247 WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL
, 0);
249 if (gmc_v6_0_wait_for_idle((void *)adev
)) {
250 dev_warn(adev
->dev
, "Wait for MC idle timedout !\n");
253 if (adev
->mode_info
.num_crtc
) {
256 /* Lockout access through VGA aperture*/
257 tmp
= RREG32(mmVGA_HDP_CONTROL
);
258 tmp
|= VGA_HDP_CONTROL__VGA_MEMORY_DISABLE_MASK
;
259 WREG32(mmVGA_HDP_CONTROL
, tmp
);
261 /* disable VGA render */
262 tmp
= RREG32(mmVGA_RENDER_CONTROL
);
263 tmp
&= ~VGA_VSTATUS_CNTL
;
264 WREG32(mmVGA_RENDER_CONTROL
, tmp
);
266 /* Update configuration */
267 WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR
,
268 adev
->gmc
.vram_start
>> 12);
269 WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR
,
270 adev
->gmc
.vram_end
>> 12);
271 WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR
,
272 adev
->vram_scratch
.gpu_addr
>> 12);
273 WREG32(mmMC_VM_AGP_BASE
, 0);
274 WREG32(mmMC_VM_AGP_TOP
, 0x0FFFFFFF);
275 WREG32(mmMC_VM_AGP_BOT
, 0x0FFFFFFF);
277 if (gmc_v6_0_wait_for_idle((void *)adev
)) {
278 dev_warn(adev
->dev
, "Wait for MC idle timedout !\n");
282 static int gmc_v6_0_mc_init(struct amdgpu_device
*adev
)
286 int chansize
, numchan
;
289 tmp
= RREG32(mmMC_ARB_RAMCFG
);
290 if (tmp
& (1 << 11)) {
292 } else if (tmp
& MC_ARB_RAMCFG__CHANSIZE_MASK
) {
297 tmp
= RREG32(mmMC_SHARED_CHMAP
);
298 switch ((tmp
& MC_SHARED_CHMAP__NOOFCHAN_MASK
) >> MC_SHARED_CHMAP__NOOFCHAN__SHIFT
) {
328 adev
->gmc
.vram_width
= numchan
* chansize
;
329 /* size in MB on si */
330 adev
->gmc
.mc_vram_size
= RREG32(mmCONFIG_MEMSIZE
) * 1024ULL * 1024ULL;
331 adev
->gmc
.real_vram_size
= RREG32(mmCONFIG_MEMSIZE
) * 1024ULL * 1024ULL;
333 if (!(adev
->flags
& AMD_IS_APU
)) {
334 r
= amdgpu_device_resize_fb_bar(adev
);
338 adev
->gmc
.aper_base
= pci_resource_start(adev
->pdev
, 0);
339 adev
->gmc
.aper_size
= pci_resource_len(adev
->pdev
, 0);
340 adev
->gmc
.visible_vram_size
= adev
->gmc
.aper_size
;
342 /* set the gart size */
343 if (amdgpu_gart_size
== -1) {
344 switch (adev
->asic_type
) {
345 case CHIP_HAINAN
: /* no MM engines */
347 adev
->gmc
.gart_size
= 256ULL << 20;
349 case CHIP_VERDE
: /* UVD, VCE do not support GPUVM */
350 case CHIP_TAHITI
: /* UVD, VCE do not support GPUVM */
351 case CHIP_PITCAIRN
: /* UVD, VCE do not support GPUVM */
352 case CHIP_OLAND
: /* UVD, VCE do not support GPUVM */
353 adev
->gmc
.gart_size
= 1024ULL << 20;
357 adev
->gmc
.gart_size
= (u64
)amdgpu_gart_size
<< 20;
360 gmc_v6_0_vram_gtt_location(adev
, &adev
->gmc
);
365 static void gmc_v6_0_flush_gpu_tlb(struct amdgpu_device
*adev
, uint32_t vmid
,
366 uint32_t vmhub
, uint32_t flush_type
)
368 WREG32(mmVM_INVALIDATE_REQUEST
, 1 << vmid
);
371 static uint64_t gmc_v6_0_emit_flush_gpu_tlb(struct amdgpu_ring
*ring
,
372 unsigned vmid
, uint64_t pd_addr
)
376 /* write new base address */
378 reg
= mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR
+ vmid
;
380 reg
= mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR
+ (vmid
- 8);
381 amdgpu_ring_emit_wreg(ring
, reg
, pd_addr
>> 12);
383 /* bits 0-15 are the VM contexts0-15 */
384 amdgpu_ring_emit_wreg(ring
, mmVM_INVALIDATE_REQUEST
, 1 << vmid
);
389 static void gmc_v6_0_get_vm_pde(struct amdgpu_device
*adev
, int level
,
390 uint64_t *addr
, uint64_t *flags
)
392 BUG_ON(*addr
& 0xFFFFFF0000000FFFULL
);
395 static void gmc_v6_0_get_vm_pte(struct amdgpu_device
*adev
,
396 struct amdgpu_bo_va_mapping
*mapping
,
399 *flags
&= ~AMDGPU_PTE_EXECUTABLE
;
400 *flags
&= ~AMDGPU_PTE_PRT
;
403 static void gmc_v6_0_set_fault_enable_default(struct amdgpu_device
*adev
,
408 tmp
= RREG32(mmVM_CONTEXT1_CNTL
);
409 tmp
= REG_SET_FIELD(tmp
, VM_CONTEXT1_CNTL
,
410 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT
, value
);
411 tmp
= REG_SET_FIELD(tmp
, VM_CONTEXT1_CNTL
,
412 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT
, value
);
413 tmp
= REG_SET_FIELD(tmp
, VM_CONTEXT1_CNTL
,
414 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT
, value
);
415 tmp
= REG_SET_FIELD(tmp
, VM_CONTEXT1_CNTL
,
416 VALID_PROTECTION_FAULT_ENABLE_DEFAULT
, value
);
417 tmp
= REG_SET_FIELD(tmp
, VM_CONTEXT1_CNTL
,
418 READ_PROTECTION_FAULT_ENABLE_DEFAULT
, value
);
419 tmp
= REG_SET_FIELD(tmp
, VM_CONTEXT1_CNTL
,
420 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT
, value
);
421 WREG32(mmVM_CONTEXT1_CNTL
, tmp
);
425 + * gmc_v8_0_set_prt - set PRT VM fault
427 + * @adev: amdgpu_device pointer
428 + * @enable: enable/disable VM fault handling for PRT
430 static void gmc_v6_0_set_prt(struct amdgpu_device
*adev
, bool enable
)
434 if (enable
&& !adev
->gmc
.prt_warning
) {
435 dev_warn(adev
->dev
, "Disabling VM faults because of PRT request!\n");
436 adev
->gmc
.prt_warning
= true;
439 tmp
= RREG32(mmVM_PRT_CNTL
);
440 tmp
= REG_SET_FIELD(tmp
, VM_PRT_CNTL
,
441 CB_DISABLE_FAULT_ON_UNMAPPED_ACCESS
,
443 tmp
= REG_SET_FIELD(tmp
, VM_PRT_CNTL
,
444 TC_DISABLE_FAULT_ON_UNMAPPED_ACCESS
,
446 tmp
= REG_SET_FIELD(tmp
, VM_PRT_CNTL
,
447 L2_CACHE_STORE_INVALID_ENTRIES
,
449 tmp
= REG_SET_FIELD(tmp
, VM_PRT_CNTL
,
450 L1_TLB_STORE_INVALID_ENTRIES
,
452 WREG32(mmVM_PRT_CNTL
, tmp
);
455 uint32_t low
= AMDGPU_VA_RESERVED_SIZE
>> AMDGPU_GPU_PAGE_SHIFT
;
456 uint32_t high
= adev
->vm_manager
.max_pfn
-
457 (AMDGPU_VA_RESERVED_SIZE
>> AMDGPU_GPU_PAGE_SHIFT
);
459 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR
, low
);
460 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR
, low
);
461 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR
, low
);
462 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR
, low
);
463 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR
, high
);
464 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR
, high
);
465 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR
, high
);
466 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR
, high
);
468 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR
, 0xfffffff);
469 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR
, 0xfffffff);
470 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR
, 0xfffffff);
471 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR
, 0xfffffff);
472 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR
, 0x0);
473 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR
, 0x0);
474 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR
, 0x0);
475 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR
, 0x0);
479 static int gmc_v6_0_gart_enable(struct amdgpu_device
*adev
)
485 if (adev
->gart
.bo
== NULL
) {
486 dev_err(adev
->dev
, "No VRAM object for PCIE GART.\n");
489 r
= amdgpu_gart_table_vram_pin(adev
);
493 table_addr
= amdgpu_bo_gpu_offset(adev
->gart
.bo
);
495 /* Setup TLB control */
496 WREG32(mmMC_VM_MX_L1_TLB_CNTL
,
498 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_TLB_MASK
|
499 MC_VM_MX_L1_TLB_CNTL__ENABLE_L1_FRAGMENT_PROCESSING_MASK
|
500 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK
|
501 MC_VM_MX_L1_TLB_CNTL__ENABLE_ADVANCED_DRIVER_MODEL_MASK
|
502 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT
));
505 VM_L2_CNTL__ENABLE_L2_CACHE_MASK
|
506 VM_L2_CNTL__ENABLE_L2_FRAGMENT_PROCESSING_MASK
|
507 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK
|
508 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK
|
509 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT
) |
510 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT
));
511 WREG32(mmVM_L2_CNTL2
,
512 VM_L2_CNTL2__INVALIDATE_ALL_L1_TLBS_MASK
|
513 VM_L2_CNTL2__INVALIDATE_L2_CACHE_MASK
);
515 field
= adev
->vm_manager
.fragment_size
;
516 WREG32(mmVM_L2_CNTL3
,
517 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK
|
518 (field
<< VM_L2_CNTL3__BANK_SELECT__SHIFT
) |
519 (field
<< VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT
));
521 WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR
, adev
->gmc
.gart_start
>> 12);
522 WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR
, adev
->gmc
.gart_end
>> 12);
523 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR
, table_addr
>> 12);
524 WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR
,
525 (u32
)(adev
->dummy_page_addr
>> 12));
526 WREG32(mmVM_CONTEXT0_CNTL2
, 0);
527 WREG32(mmVM_CONTEXT0_CNTL
,
528 VM_CONTEXT0_CNTL__ENABLE_CONTEXT_MASK
|
529 (0UL << VM_CONTEXT0_CNTL__PAGE_TABLE_DEPTH__SHIFT
) |
530 VM_CONTEXT0_CNTL__RANGE_PROTECTION_FAULT_ENABLE_DEFAULT_MASK
);
536 /* empty context1-15 */
537 /* set vm size, must be a multiple of 4 */
538 WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR
, 0);
539 WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR
, adev
->vm_manager
.max_pfn
- 1);
540 /* Assign the pt base to something valid for now; the pts used for
541 * the VMs are determined by the application and setup and assigned
542 * on the fly in the vm part of radeon_gart.c
544 for (i
= 1; i
< 16; i
++) {
546 WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR
+ i
,
549 WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR
+ i
- 8,
553 /* enable context1-15 */
554 WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR
,
555 (u32
)(adev
->dummy_page_addr
>> 12));
556 WREG32(mmVM_CONTEXT1_CNTL2
, 4);
557 WREG32(mmVM_CONTEXT1_CNTL
,
558 VM_CONTEXT1_CNTL__ENABLE_CONTEXT_MASK
|
559 (1UL << VM_CONTEXT1_CNTL__PAGE_TABLE_DEPTH__SHIFT
) |
560 ((adev
->vm_manager
.block_size
- 9)
561 << VM_CONTEXT1_CNTL__PAGE_TABLE_BLOCK_SIZE__SHIFT
));
562 if (amdgpu_vm_fault_stop
== AMDGPU_VM_FAULT_STOP_ALWAYS
)
563 gmc_v6_0_set_fault_enable_default(adev
, false);
565 gmc_v6_0_set_fault_enable_default(adev
, true);
567 gmc_v6_0_flush_gpu_tlb(adev
, 0, 0, 0);
568 dev_info(adev
->dev
, "PCIE GART of %uM enabled (table at 0x%016llX).\n",
569 (unsigned)(adev
->gmc
.gart_size
>> 20),
570 (unsigned long long)table_addr
);
571 adev
->gart
.ready
= true;
575 static int gmc_v6_0_gart_init(struct amdgpu_device
*adev
)
580 dev_warn(adev
->dev
, "gmc_v6_0 PCIE GART already initialized\n");
583 r
= amdgpu_gart_init(adev
);
586 adev
->gart
.table_size
= adev
->gart
.num_gpu_pages
* 8;
587 adev
->gart
.gart_pte_flags
= 0;
588 return amdgpu_gart_table_vram_alloc(adev
);
591 static void gmc_v6_0_gart_disable(struct amdgpu_device
*adev
)
595 for (i = 1; i < 16; ++i) {
598 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i ;
600 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (i - 8);
601 adev->vm_manager.saved_table_addr[i] = RREG32(reg);
604 /* Disable all tables */
605 WREG32(mmVM_CONTEXT0_CNTL
, 0);
606 WREG32(mmVM_CONTEXT1_CNTL
, 0);
607 /* Setup TLB control */
608 WREG32(mmMC_VM_MX_L1_TLB_CNTL
,
609 MC_VM_MX_L1_TLB_CNTL__SYSTEM_ACCESS_MODE_MASK
|
610 (0UL << MC_VM_MX_L1_TLB_CNTL__SYSTEM_APERTURE_UNMAPPED_ACCESS__SHIFT
));
613 VM_L2_CNTL__ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE_MASK
|
614 VM_L2_CNTL__ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE_MASK
|
615 (7UL << VM_L2_CNTL__EFFECTIVE_L2_QUEUE_SIZE__SHIFT
) |
616 (1UL << VM_L2_CNTL__CONTEXT1_IDENTITY_ACCESS_MODE__SHIFT
));
617 WREG32(mmVM_L2_CNTL2
, 0);
618 WREG32(mmVM_L2_CNTL3
,
619 VM_L2_CNTL3__L2_CACHE_BIGK_ASSOCIATIVITY_MASK
|
620 (0UL << VM_L2_CNTL3__L2_CACHE_BIGK_FRAGMENT_SIZE__SHIFT
));
621 amdgpu_gart_table_vram_unpin(adev
);
624 static void gmc_v6_0_vm_decode_fault(struct amdgpu_device
*adev
,
625 u32 status
, u32 addr
, u32 mc_client
)
628 u32 vmid
= REG_GET_FIELD(status
, VM_CONTEXT1_PROTECTION_FAULT_STATUS
, VMID
);
629 u32 protections
= REG_GET_FIELD(status
, VM_CONTEXT1_PROTECTION_FAULT_STATUS
,
631 char block
[5] = { mc_client
>> 24, (mc_client
>> 16) & 0xff,
632 (mc_client
>> 8) & 0xff, mc_client
& 0xff, 0 };
634 mc_id
= REG_GET_FIELD(status
, VM_CONTEXT1_PROTECTION_FAULT_STATUS
,
637 dev_err(adev
->dev
, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
638 protections
, vmid
, addr
,
639 REG_GET_FIELD(status
, VM_CONTEXT1_PROTECTION_FAULT_STATUS
,
641 "write" : "read", block
, mc_client
, mc_id
);
645 static const u32 mc_cg_registers[] = {
657 static const u32 mc_cg_ls_en[] = {
658 MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK,
659 MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK,
660 MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK,
661 MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK,
662 ATC_MISC_CG__MEM_LS_ENABLE_MASK,
663 MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK,
664 MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK,
665 MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK,
666 VM_L2_CG__MEM_LS_ENABLE_MASK,
669 static const u32 mc_cg_en[] = {
670 MC_HUB_MISC_HUB_CG__ENABLE_MASK,
671 MC_HUB_MISC_SIP_CG__ENABLE_MASK,
672 MC_HUB_MISC_VM_CG__ENABLE_MASK,
673 MC_XPB_CLK_GAT__ENABLE_MASK,
674 ATC_MISC_CG__ENABLE_MASK,
675 MC_CITF_MISC_WR_CG__ENABLE_MASK,
676 MC_CITF_MISC_RD_CG__ENABLE_MASK,
677 MC_CITF_MISC_VM_CG__ENABLE_MASK,
678 VM_L2_CG__ENABLE_MASK,
681 static void gmc_v6_0_enable_mc_ls(struct amdgpu_device *adev,
687 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
688 orig = data = RREG32(mc_cg_registers[i]);
689 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS))
690 data |= mc_cg_ls_en[i];
692 data &= ~mc_cg_ls_en[i];
694 WREG32(mc_cg_registers[i], data);
698 static void gmc_v6_0_enable_mc_mgcg(struct amdgpu_device *adev,
704 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) {
705 orig = data = RREG32(mc_cg_registers[i]);
706 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG))
709 data &= ~mc_cg_en[i];
711 WREG32(mc_cg_registers[i], data);
715 static void gmc_v6_0_enable_bif_mgls(struct amdgpu_device *adev,
720 orig = data = RREG32_PCIE(ixPCIE_CNTL2);
722 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) {
723 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1);
724 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1);
725 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1);
726 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 1);
728 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 0);
729 data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 0);
730 data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 0);
731 data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_AGGRESSIVE_LS_EN, 0);
735 WREG32_PCIE(ixPCIE_CNTL2, data);
738 static void gmc_v6_0_enable_hdp_mgcg(struct amdgpu_device *adev,
743 orig = data = RREG32(mmHDP_HOST_PATH_CNTL);
745 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG))
746 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0);
748 data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1);
751 WREG32(mmHDP_HOST_PATH_CNTL, data);
754 static void gmc_v6_0_enable_hdp_ls(struct amdgpu_device *adev,
759 orig = data = RREG32(mmHDP_MEM_POWER_LS);
761 if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS))
762 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1);
764 data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0);
767 WREG32(mmHDP_MEM_POWER_LS, data);
771 static int gmc_v6_0_convert_vram_type(int mc_seq_vram_type
)
773 switch (mc_seq_vram_type
) {
774 case MC_SEQ_MISC0__MT__GDDR1
:
775 return AMDGPU_VRAM_TYPE_GDDR1
;
776 case MC_SEQ_MISC0__MT__DDR2
:
777 return AMDGPU_VRAM_TYPE_DDR2
;
778 case MC_SEQ_MISC0__MT__GDDR3
:
779 return AMDGPU_VRAM_TYPE_GDDR3
;
780 case MC_SEQ_MISC0__MT__GDDR4
:
781 return AMDGPU_VRAM_TYPE_GDDR4
;
782 case MC_SEQ_MISC0__MT__GDDR5
:
783 return AMDGPU_VRAM_TYPE_GDDR5
;
784 case MC_SEQ_MISC0__MT__DDR3
:
785 return AMDGPU_VRAM_TYPE_DDR3
;
787 return AMDGPU_VRAM_TYPE_UNKNOWN
;
791 static int gmc_v6_0_early_init(void *handle
)
793 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
795 gmc_v6_0_set_gmc_funcs(adev
);
796 gmc_v6_0_set_irq_funcs(adev
);
801 static int gmc_v6_0_late_init(void *handle
)
803 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
805 amdgpu_bo_late_init(adev
);
807 if (amdgpu_vm_fault_stop
!= AMDGPU_VM_FAULT_STOP_ALWAYS
)
808 return amdgpu_irq_get(adev
, &adev
->gmc
.vm_fault
, 0);
813 static unsigned gmc_v6_0_get_vbios_fb_size(struct amdgpu_device
*adev
)
815 u32 d1vga_control
= RREG32(mmD1VGA_CONTROL
);
818 if (REG_GET_FIELD(d1vga_control
, D1VGA_CONTROL
, D1VGA_MODE_ENABLE
)) {
819 size
= 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
821 u32 viewport
= RREG32(mmVIEWPORT_SIZE
);
822 size
= (REG_GET_FIELD(viewport
, VIEWPORT_SIZE
, VIEWPORT_HEIGHT
) *
823 REG_GET_FIELD(viewport
, VIEWPORT_SIZE
, VIEWPORT_WIDTH
) *
826 /* return 0 if the pre-OS buffer uses up most of vram */
827 if ((adev
->gmc
.real_vram_size
- size
) < (8 * 1024 * 1024))
832 static int gmc_v6_0_sw_init(void *handle
)
835 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
837 adev
->num_vmhubs
= 1;
839 if (adev
->flags
& AMD_IS_APU
) {
840 adev
->gmc
.vram_type
= AMDGPU_VRAM_TYPE_UNKNOWN
;
842 u32 tmp
= RREG32(mmMC_SEQ_MISC0
);
843 tmp
&= MC_SEQ_MISC0__MT__MASK
;
844 adev
->gmc
.vram_type
= gmc_v6_0_convert_vram_type(tmp
);
847 r
= amdgpu_irq_add_id(adev
, AMDGPU_IRQ_CLIENTID_LEGACY
, 146, &adev
->gmc
.vm_fault
);
851 r
= amdgpu_irq_add_id(adev
, AMDGPU_IRQ_CLIENTID_LEGACY
, 147, &adev
->gmc
.vm_fault
);
855 amdgpu_vm_adjust_size(adev
, 64, 9, 1, 40);
857 adev
->gmc
.mc_mask
= 0xffffffffffULL
;
859 r
= dma_set_mask_and_coherent(adev
->dev
, DMA_BIT_MASK(44));
861 dev_warn(adev
->dev
, "amdgpu: No suitable DMA available.\n");
864 adev
->need_swiotlb
= drm_need_swiotlb(44);
866 r
= gmc_v6_0_init_microcode(adev
);
868 dev_err(adev
->dev
, "Failed to load mc firmware!\n");
872 r
= gmc_v6_0_mc_init(adev
);
876 adev
->gmc
.stolen_size
= gmc_v6_0_get_vbios_fb_size(adev
);
878 r
= amdgpu_bo_init(adev
);
882 r
= gmc_v6_0_gart_init(adev
);
888 * VMID 0 is reserved for System
889 * amdgpu graphics/compute will use VMIDs 1-7
890 * amdkfd will use VMIDs 8-15
892 adev
->vm_manager
.id_mgr
[0].num_ids
= AMDGPU_NUM_OF_VMIDS
;
893 amdgpu_vm_manager_init(adev
);
895 /* base offset of vram pages */
896 if (adev
->flags
& AMD_IS_APU
) {
897 u64 tmp
= RREG32(mmMC_VM_FB_OFFSET
);
900 adev
->vm_manager
.vram_base_offset
= tmp
;
902 adev
->vm_manager
.vram_base_offset
= 0;
908 static int gmc_v6_0_sw_fini(void *handle
)
910 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
912 amdgpu_gem_force_release(adev
);
913 amdgpu_vm_manager_fini(adev
);
914 amdgpu_gart_table_vram_free(adev
);
915 amdgpu_bo_fini(adev
);
916 amdgpu_gart_fini(adev
);
917 release_firmware(adev
->gmc
.fw
);
923 static int gmc_v6_0_hw_init(void *handle
)
926 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
928 gmc_v6_0_mc_program(adev
);
930 if (!(adev
->flags
& AMD_IS_APU
)) {
931 r
= gmc_v6_0_mc_load_microcode(adev
);
933 dev_err(adev
->dev
, "Failed to load MC firmware!\n");
938 r
= gmc_v6_0_gart_enable(adev
);
945 static int gmc_v6_0_hw_fini(void *handle
)
947 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
949 amdgpu_irq_put(adev
, &adev
->gmc
.vm_fault
, 0);
950 gmc_v6_0_gart_disable(adev
);
955 static int gmc_v6_0_suspend(void *handle
)
957 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
959 gmc_v6_0_hw_fini(adev
);
964 static int gmc_v6_0_resume(void *handle
)
967 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
969 r
= gmc_v6_0_hw_init(adev
);
973 amdgpu_vmid_reset_all(adev
);
978 static bool gmc_v6_0_is_idle(void *handle
)
980 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
981 u32 tmp
= RREG32(mmSRBM_STATUS
);
983 if (tmp
& (SRBM_STATUS__MCB_BUSY_MASK
| SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK
|
984 SRBM_STATUS__MCC_BUSY_MASK
| SRBM_STATUS__MCD_BUSY_MASK
| SRBM_STATUS__VMC_BUSY_MASK
))
990 static int gmc_v6_0_wait_for_idle(void *handle
)
993 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
995 for (i
= 0; i
< adev
->usec_timeout
; i
++) {
996 if (gmc_v6_0_is_idle(handle
))
1004 static int gmc_v6_0_soft_reset(void *handle
)
1006 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
1007 u32 srbm_soft_reset
= 0;
1008 u32 tmp
= RREG32(mmSRBM_STATUS
);
1010 if (tmp
& SRBM_STATUS__VMC_BUSY_MASK
)
1011 srbm_soft_reset
= REG_SET_FIELD(srbm_soft_reset
,
1012 SRBM_SOFT_RESET
, SOFT_RESET_VMC
, 1);
1014 if (tmp
& (SRBM_STATUS__MCB_BUSY_MASK
| SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK
|
1015 SRBM_STATUS__MCC_BUSY_MASK
| SRBM_STATUS__MCD_BUSY_MASK
)) {
1016 if (!(adev
->flags
& AMD_IS_APU
))
1017 srbm_soft_reset
= REG_SET_FIELD(srbm_soft_reset
,
1018 SRBM_SOFT_RESET
, SOFT_RESET_MC
, 1);
1021 if (srbm_soft_reset
) {
1022 gmc_v6_0_mc_stop(adev
);
1023 if (gmc_v6_0_wait_for_idle(adev
)) {
1024 dev_warn(adev
->dev
, "Wait for GMC idle timed out !\n");
1028 tmp
= RREG32(mmSRBM_SOFT_RESET
);
1029 tmp
|= srbm_soft_reset
;
1030 dev_info(adev
->dev
, "SRBM_SOFT_RESET=0x%08X\n", tmp
);
1031 WREG32(mmSRBM_SOFT_RESET
, tmp
);
1032 tmp
= RREG32(mmSRBM_SOFT_RESET
);
1036 tmp
&= ~srbm_soft_reset
;
1037 WREG32(mmSRBM_SOFT_RESET
, tmp
);
1038 tmp
= RREG32(mmSRBM_SOFT_RESET
);
1042 gmc_v6_0_mc_resume(adev
);
1049 static int gmc_v6_0_vm_fault_interrupt_state(struct amdgpu_device
*adev
,
1050 struct amdgpu_irq_src
*src
,
1052 enum amdgpu_interrupt_state state
)
1055 u32 bits
= (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK
|
1056 VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK
|
1057 VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK
|
1058 VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK
|
1059 VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK
|
1060 VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK
);
1063 case AMDGPU_IRQ_STATE_DISABLE
:
1064 tmp
= RREG32(mmVM_CONTEXT0_CNTL
);
1066 WREG32(mmVM_CONTEXT0_CNTL
, tmp
);
1067 tmp
= RREG32(mmVM_CONTEXT1_CNTL
);
1069 WREG32(mmVM_CONTEXT1_CNTL
, tmp
);
1071 case AMDGPU_IRQ_STATE_ENABLE
:
1072 tmp
= RREG32(mmVM_CONTEXT0_CNTL
);
1074 WREG32(mmVM_CONTEXT0_CNTL
, tmp
);
1075 tmp
= RREG32(mmVM_CONTEXT1_CNTL
);
1077 WREG32(mmVM_CONTEXT1_CNTL
, tmp
);
1086 static int gmc_v6_0_process_interrupt(struct amdgpu_device
*adev
,
1087 struct amdgpu_irq_src
*source
,
1088 struct amdgpu_iv_entry
*entry
)
1092 addr
= RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR
);
1093 status
= RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS
);
1094 WREG32_P(mmVM_CONTEXT1_CNTL2
, 1, ~1);
1096 if (!addr
&& !status
)
1099 if (amdgpu_vm_fault_stop
== AMDGPU_VM_FAULT_STOP_FIRST
)
1100 gmc_v6_0_set_fault_enable_default(adev
, false);
1102 if (printk_ratelimit()) {
1103 dev_err(adev
->dev
, "GPU fault detected: %d 0x%08x\n",
1104 entry
->src_id
, entry
->src_data
[0]);
1105 dev_err(adev
->dev
, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
1107 dev_err(adev
->dev
, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1109 gmc_v6_0_vm_decode_fault(adev
, status
, addr
, 0);
1115 static int gmc_v6_0_set_clockgating_state(void *handle
,
1116 enum amd_clockgating_state state
)
1121 static int gmc_v6_0_set_powergating_state(void *handle
,
1122 enum amd_powergating_state state
)
1127 static const struct amd_ip_funcs gmc_v6_0_ip_funcs
= {
1129 .early_init
= gmc_v6_0_early_init
,
1130 .late_init
= gmc_v6_0_late_init
,
1131 .sw_init
= gmc_v6_0_sw_init
,
1132 .sw_fini
= gmc_v6_0_sw_fini
,
1133 .hw_init
= gmc_v6_0_hw_init
,
1134 .hw_fini
= gmc_v6_0_hw_fini
,
1135 .suspend
= gmc_v6_0_suspend
,
1136 .resume
= gmc_v6_0_resume
,
1137 .is_idle
= gmc_v6_0_is_idle
,
1138 .wait_for_idle
= gmc_v6_0_wait_for_idle
,
1139 .soft_reset
= gmc_v6_0_soft_reset
,
1140 .set_clockgating_state
= gmc_v6_0_set_clockgating_state
,
1141 .set_powergating_state
= gmc_v6_0_set_powergating_state
,
1144 static const struct amdgpu_gmc_funcs gmc_v6_0_gmc_funcs
= {
1145 .flush_gpu_tlb
= gmc_v6_0_flush_gpu_tlb
,
1146 .emit_flush_gpu_tlb
= gmc_v6_0_emit_flush_gpu_tlb
,
1147 .set_prt
= gmc_v6_0_set_prt
,
1148 .get_vm_pde
= gmc_v6_0_get_vm_pde
,
1149 .get_vm_pte
= gmc_v6_0_get_vm_pte
,
1152 static const struct amdgpu_irq_src_funcs gmc_v6_0_irq_funcs
= {
1153 .set
= gmc_v6_0_vm_fault_interrupt_state
,
1154 .process
= gmc_v6_0_process_interrupt
,
1157 static void gmc_v6_0_set_gmc_funcs(struct amdgpu_device
*adev
)
1159 adev
->gmc
.gmc_funcs
= &gmc_v6_0_gmc_funcs
;
1162 static void gmc_v6_0_set_irq_funcs(struct amdgpu_device
*adev
)
1164 adev
->gmc
.vm_fault
.num_types
= 1;
1165 adev
->gmc
.vm_fault
.funcs
= &gmc_v6_0_irq_funcs
;
1168 const struct amdgpu_ip_block_version gmc_v6_0_ip_block
=
1170 .type
= AMD_IP_BLOCK_TYPE_GMC
,
1174 .funcs
= &gmc_v6_0_ip_funcs
,