xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c (revision 2596e07a)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 #include <linux/firmware.h>
24 #include "drmP.h"
25 #include "amdgpu.h"
26 #include "gmc_v8_0.h"
27 #include "amdgpu_ucode.h"
28 
29 #include "gmc/gmc_8_1_d.h"
30 #include "gmc/gmc_8_1_sh_mask.h"
31 
32 #include "bif/bif_5_0_d.h"
33 #include "bif/bif_5_0_sh_mask.h"
34 
35 #include "oss/oss_3_0_d.h"
36 #include "oss/oss_3_0_sh_mask.h"
37 
38 #include "vid.h"
39 #include "vi.h"
40 
41 
42 static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev);
43 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
44 
45 MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
46 
47 static const u32 golden_settings_tonga_a11[] =
48 {
49 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
50 	mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
51 	mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
52 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
53 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
54 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
55 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
56 };
57 
58 static const u32 tonga_mgcg_cgcg_init[] =
59 {
60 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
61 };
62 
63 static const u32 golden_settings_fiji_a10[] =
64 {
65 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
66 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
67 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
68 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
69 };
70 
71 static const u32 fiji_mgcg_cgcg_init[] =
72 {
73 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
74 };
75 
76 static const u32 cz_mgcg_cgcg_init[] =
77 {
78 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
79 };
80 
81 static const u32 stoney_mgcg_cgcg_init[] =
82 {
83 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
84 };
85 
86 
87 static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
88 {
89 	switch (adev->asic_type) {
90 	case CHIP_FIJI:
91 		amdgpu_program_register_sequence(adev,
92 						 fiji_mgcg_cgcg_init,
93 						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
94 		amdgpu_program_register_sequence(adev,
95 						 golden_settings_fiji_a10,
96 						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
97 		break;
98 	case CHIP_TONGA:
99 		amdgpu_program_register_sequence(adev,
100 						 tonga_mgcg_cgcg_init,
101 						 (const u32)ARRAY_SIZE(tonga_mgcg_cgcg_init));
102 		amdgpu_program_register_sequence(adev,
103 						 golden_settings_tonga_a11,
104 						 (const u32)ARRAY_SIZE(golden_settings_tonga_a11));
105 		break;
106 	case CHIP_CARRIZO:
107 		amdgpu_program_register_sequence(adev,
108 						 cz_mgcg_cgcg_init,
109 						 (const u32)ARRAY_SIZE(cz_mgcg_cgcg_init));
110 		break;
111 	case CHIP_STONEY:
112 		amdgpu_program_register_sequence(adev,
113 						 stoney_mgcg_cgcg_init,
114 						 (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init));
115 		break;
116 	default:
117 		break;
118 	}
119 }
120 
121 /**
122  * gmc8_mc_wait_for_idle - wait for MC idle callback.
123  *
124  * @adev: amdgpu_device pointer
125  *
126  * Wait for the MC (memory controller) to be idle.
127  * (evergreen+).
128  * Returns 0 if the MC is idle, -1 if not.
129  */
130 int gmc_v8_0_mc_wait_for_idle(struct amdgpu_device *adev)
131 {
132 	unsigned i;
133 	u32 tmp;
134 
135 	for (i = 0; i < adev->usec_timeout; i++) {
136 		/* read MC_STATUS */
137 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__VMC_BUSY_MASK |
138 					       SRBM_STATUS__MCB_BUSY_MASK |
139 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
140 					       SRBM_STATUS__MCC_BUSY_MASK |
141 					       SRBM_STATUS__MCD_BUSY_MASK |
142 					       SRBM_STATUS__VMC1_BUSY_MASK);
143 		if (!tmp)
144 			return 0;
145 		udelay(1);
146 	}
147 	return -1;
148 }
149 
150 void gmc_v8_0_mc_stop(struct amdgpu_device *adev,
151 		      struct amdgpu_mode_mc_save *save)
152 {
153 	u32 blackout;
154 
155 	if (adev->mode_info.num_crtc)
156 		amdgpu_display_stop_mc_access(adev, save);
157 
158 	amdgpu_asic_wait_for_mc_idle(adev);
159 
160 	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
161 	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
162 		/* Block CPU access */
163 		WREG32(mmBIF_FB_EN, 0);
164 		/* blackout the MC */
165 		blackout = REG_SET_FIELD(blackout,
166 					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
167 		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
168 	}
169 	/* wait for the MC to settle */
170 	udelay(100);
171 }
172 
173 void gmc_v8_0_mc_resume(struct amdgpu_device *adev,
174 			struct amdgpu_mode_mc_save *save)
175 {
176 	u32 tmp;
177 
178 	/* unblackout the MC */
179 	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
180 	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
181 	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
182 	/* allow CPU access */
183 	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
184 	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
185 	WREG32(mmBIF_FB_EN, tmp);
186 
187 	if (adev->mode_info.num_crtc)
188 		amdgpu_display_resume_mc_access(adev, save);
189 }
190 
191 /**
192  * gmc_v8_0_init_microcode - load ucode images from disk
193  *
194  * @adev: amdgpu_device pointer
195  *
196  * Use the firmware interface to load the ucode images into
197  * the driver (not loaded into hw).
198  * Returns 0 on success, error on failure.
199  */
200 static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
201 {
202 	const char *chip_name;
203 	char fw_name[30];
204 	int err;
205 
206 	DRM_DEBUG("\n");
207 
208 	switch (adev->asic_type) {
209 	case CHIP_TONGA:
210 		chip_name = "tonga";
211 		break;
212 	case CHIP_FIJI:
213 	case CHIP_CARRIZO:
214 	case CHIP_STONEY:
215 		return 0;
216 	default: BUG();
217 	}
218 
219 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
220 	err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
221 	if (err)
222 		goto out;
223 	err = amdgpu_ucode_validate(adev->mc.fw);
224 
225 out:
226 	if (err) {
227 		printk(KERN_ERR
228 		       "mc: Failed to load firmware \"%s\"\n",
229 		       fw_name);
230 		release_firmware(adev->mc.fw);
231 		adev->mc.fw = NULL;
232 	}
233 	return err;
234 }
235 
236 /**
237  * gmc_v8_0_mc_load_microcode - load MC ucode into the hw
238  *
239  * @adev: amdgpu_device pointer
240  *
241  * Load the GDDR MC ucode into the hw (CIK).
242  * Returns 0 on success, error on failure.
243  */
244 static int gmc_v8_0_mc_load_microcode(struct amdgpu_device *adev)
245 {
246 	const struct mc_firmware_header_v1_0 *hdr;
247 	const __le32 *fw_data = NULL;
248 	const __le32 *io_mc_regs = NULL;
249 	u32 running, blackout = 0;
250 	int i, ucode_size, regs_size;
251 
252 	if (!adev->mc.fw)
253 		return -EINVAL;
254 
255 	hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
256 	amdgpu_ucode_print_mc_hdr(&hdr->header);
257 
258 	adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
259 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
260 	io_mc_regs = (const __le32 *)
261 		(adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
262 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
263 	fw_data = (const __le32 *)
264 		(adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
265 
266 	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
267 
268 	if (running == 0) {
269 		if (running) {
270 			blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
271 			WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout | 1);
272 		}
273 
274 		/* reset the engine and set to writable */
275 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
276 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
277 
278 		/* load mc io regs */
279 		for (i = 0; i < regs_size; i++) {
280 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
281 			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
282 		}
283 		/* load the MC ucode */
284 		for (i = 0; i < ucode_size; i++)
285 			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
286 
287 		/* put the engine back into the active state */
288 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
289 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
290 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
291 
292 		/* wait for training to complete */
293 		for (i = 0; i < adev->usec_timeout; i++) {
294 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
295 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
296 				break;
297 			udelay(1);
298 		}
299 		for (i = 0; i < adev->usec_timeout; i++) {
300 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
301 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
302 				break;
303 			udelay(1);
304 		}
305 
306 		if (running)
307 			WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
308 	}
309 
310 	return 0;
311 }
312 
313 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
314 				       struct amdgpu_mc *mc)
315 {
316 	if (mc->mc_vram_size > 0xFFC0000000ULL) {
317 		/* leave room for at least 1024M GTT */
318 		dev_warn(adev->dev, "limiting VRAM\n");
319 		mc->real_vram_size = 0xFFC0000000ULL;
320 		mc->mc_vram_size = 0xFFC0000000ULL;
321 	}
322 	amdgpu_vram_location(adev, &adev->mc, 0);
323 	adev->mc.gtt_base_align = 0;
324 	amdgpu_gtt_location(adev, mc);
325 }
326 
327 /**
328  * gmc_v8_0_mc_program - program the GPU memory controller
329  *
330  * @adev: amdgpu_device pointer
331  *
332  * Set the location of vram, gart, and AGP in the GPU's
333  * physical address space (CIK).
334  */
335 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
336 {
337 	struct amdgpu_mode_mc_save save;
338 	u32 tmp;
339 	int i, j;
340 
341 	/* Initialize HDP */
342 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
343 		WREG32((0xb05 + j), 0x00000000);
344 		WREG32((0xb06 + j), 0x00000000);
345 		WREG32((0xb07 + j), 0x00000000);
346 		WREG32((0xb08 + j), 0x00000000);
347 		WREG32((0xb09 + j), 0x00000000);
348 	}
349 	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
350 
351 	if (adev->mode_info.num_crtc)
352 		amdgpu_display_set_vga_render_state(adev, false);
353 
354 	gmc_v8_0_mc_stop(adev, &save);
355 	if (amdgpu_asic_wait_for_mc_idle(adev)) {
356 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
357 	}
358 	/* Update configuration */
359 	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
360 	       adev->mc.vram_start >> 12);
361 	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
362 	       adev->mc.vram_end >> 12);
363 	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
364 	       adev->vram_scratch.gpu_addr >> 12);
365 	tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
366 	tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
367 	WREG32(mmMC_VM_FB_LOCATION, tmp);
368 	/* XXX double check these! */
369 	WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
370 	WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
371 	WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
372 	WREG32(mmMC_VM_AGP_BASE, 0);
373 	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
374 	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
375 	if (amdgpu_asic_wait_for_mc_idle(adev)) {
376 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
377 	}
378 	gmc_v8_0_mc_resume(adev, &save);
379 
380 	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
381 
382 	tmp = RREG32(mmHDP_MISC_CNTL);
383 	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1);
384 	WREG32(mmHDP_MISC_CNTL, tmp);
385 
386 	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
387 	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
388 }
389 
390 /**
391  * gmc_v8_0_mc_init - initialize the memory controller driver params
392  *
393  * @adev: amdgpu_device pointer
394  *
395  * Look up the amount of vram, vram width, and decide how to place
396  * vram and gart within the GPU's physical address space (CIK).
397  * Returns 0 for success.
398  */
399 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
400 {
401 	u32 tmp;
402 	int chansize, numchan;
403 
404 	/* Get VRAM informations */
405 	tmp = RREG32(mmMC_ARB_RAMCFG);
406 	if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
407 		chansize = 64;
408 	} else {
409 		chansize = 32;
410 	}
411 	tmp = RREG32(mmMC_SHARED_CHMAP);
412 	switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
413 	case 0:
414 	default:
415 		numchan = 1;
416 		break;
417 	case 1:
418 		numchan = 2;
419 		break;
420 	case 2:
421 		numchan = 4;
422 		break;
423 	case 3:
424 		numchan = 8;
425 		break;
426 	case 4:
427 		numchan = 3;
428 		break;
429 	case 5:
430 		numchan = 6;
431 		break;
432 	case 6:
433 		numchan = 10;
434 		break;
435 	case 7:
436 		numchan = 12;
437 		break;
438 	case 8:
439 		numchan = 16;
440 		break;
441 	}
442 	adev->mc.vram_width = numchan * chansize;
443 	/* Could aper size report 0 ? */
444 	adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
445 	adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
446 	/* size in MB on si */
447 	adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
448 	adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
449 	adev->mc.visible_vram_size = adev->mc.aper_size;
450 
451 	/* In case the PCI BAR is larger than the actual amount of vram */
452 	if (adev->mc.visible_vram_size > adev->mc.real_vram_size)
453 		adev->mc.visible_vram_size = adev->mc.real_vram_size;
454 
455 	/* unless the user had overridden it, set the gart
456 	 * size equal to the 1024 or vram, whichever is larger.
457 	 */
458 	if (amdgpu_gart_size == -1)
459 		adev->mc.gtt_size = max((1024ULL << 20), adev->mc.mc_vram_size);
460 	else
461 		adev->mc.gtt_size = (uint64_t)amdgpu_gart_size << 20;
462 
463 	gmc_v8_0_vram_gtt_location(adev, &adev->mc);
464 
465 	return 0;
466 }
467 
468 /*
469  * GART
470  * VMID 0 is the physical GPU addresses as used by the kernel.
471  * VMIDs 1-15 are used for userspace clients and are handled
472  * by the amdgpu vm/hsa code.
473  */
474 
475 /**
476  * gmc_v8_0_gart_flush_gpu_tlb - gart tlb flush callback
477  *
478  * @adev: amdgpu_device pointer
479  * @vmid: vm instance to flush
480  *
481  * Flush the TLB for the requested page table (CIK).
482  */
483 static void gmc_v8_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
484 					uint32_t vmid)
485 {
486 	/* flush hdp cache */
487 	WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
488 
489 	/* bits 0-15 are the VM contexts0-15 */
490 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
491 }
492 
493 /**
494  * gmc_v8_0_gart_set_pte_pde - update the page tables using MMIO
495  *
496  * @adev: amdgpu_device pointer
497  * @cpu_pt_addr: cpu address of the page table
498  * @gpu_page_idx: entry in the page table to update
499  * @addr: dst addr to write into pte/pde
500  * @flags: access flags
501  *
502  * Update the page tables using the CPU.
503  */
504 static int gmc_v8_0_gart_set_pte_pde(struct amdgpu_device *adev,
505 				     void *cpu_pt_addr,
506 				     uint32_t gpu_page_idx,
507 				     uint64_t addr,
508 				     uint32_t flags)
509 {
510 	void __iomem *ptr = (void *)cpu_pt_addr;
511 	uint64_t value;
512 
513 	/*
514 	 * PTE format on VI:
515 	 * 63:40 reserved
516 	 * 39:12 4k physical page base address
517 	 * 11:7 fragment
518 	 * 6 write
519 	 * 5 read
520 	 * 4 exe
521 	 * 3 reserved
522 	 * 2 snooped
523 	 * 1 system
524 	 * 0 valid
525 	 *
526 	 * PDE format on VI:
527 	 * 63:59 block fragment size
528 	 * 58:40 reserved
529 	 * 39:1 physical base address of PTE
530 	 * bits 5:1 must be 0.
531 	 * 0 valid
532 	 */
533 	value = addr & 0x000000FFFFFFF000ULL;
534 	value |= flags;
535 	writeq(value, ptr + (gpu_page_idx * 8));
536 
537 	return 0;
538 }
539 
540 /**
541  * gmc_v8_0_set_fault_enable_default - update VM fault handling
542  *
543  * @adev: amdgpu_device pointer
544  * @value: true redirects VM faults to the default page
545  */
546 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
547 					      bool value)
548 {
549 	u32 tmp;
550 
551 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
552 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
553 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
554 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
555 			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
556 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
557 			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
558 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
559 			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
560 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
561 			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
562 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
563 			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
564 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
565 			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
566 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
567 }
568 
569 /**
570  * gmc_v8_0_gart_enable - gart enable
571  *
572  * @adev: amdgpu_device pointer
573  *
574  * This sets up the TLBs, programs the page tables for VMID0,
575  * sets up the hw for VMIDs 1-15 which are allocated on
576  * demand, and sets up the global locations for the LDS, GDS,
577  * and GPUVM for FSA64 clients (CIK).
578  * Returns 0 for success, errors for failure.
579  */
580 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
581 {
582 	int r, i;
583 	u32 tmp;
584 
585 	if (adev->gart.robj == NULL) {
586 		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
587 		return -EINVAL;
588 	}
589 	r = amdgpu_gart_table_vram_pin(adev);
590 	if (r)
591 		return r;
592 	/* Setup TLB control */
593 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
594 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
595 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
596 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
597 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
598 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
599 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
600 	/* Setup L2 cache */
601 	tmp = RREG32(mmVM_L2_CNTL);
602 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
603 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
604 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
605 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
606 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
607 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
608 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
609 	WREG32(mmVM_L2_CNTL, tmp);
610 	tmp = RREG32(mmVM_L2_CNTL2);
611 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
612 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
613 	WREG32(mmVM_L2_CNTL2, tmp);
614 	tmp = RREG32(mmVM_L2_CNTL3);
615 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
616 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 4);
617 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, 4);
618 	WREG32(mmVM_L2_CNTL3, tmp);
619 	/* XXX: set to enable PTE/PDE in system memory */
620 	tmp = RREG32(mmVM_L2_CNTL4);
621 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
622 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
623 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
624 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
625 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
626 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
627 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
628 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
629 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
630 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
631 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
632 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
633 	WREG32(mmVM_L2_CNTL4, tmp);
634 	/* setup context0 */
635 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12);
636 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12);
637 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
638 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
639 			(u32)(adev->dummy_page.addr >> 12));
640 	WREG32(mmVM_CONTEXT0_CNTL2, 0);
641 	tmp = RREG32(mmVM_CONTEXT0_CNTL);
642 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
643 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
644 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
645 	WREG32(mmVM_CONTEXT0_CNTL, tmp);
646 
647 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
648 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
649 	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
650 
651 	/* empty context1-15 */
652 	/* FIXME start with 4G, once using 2 level pt switch to full
653 	 * vm size space
654 	 */
655 	/* set vm size, must be a multiple of 4 */
656 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
657 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
658 	for (i = 1; i < 16; i++) {
659 		if (i < 8)
660 			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
661 			       adev->gart.table_addr >> 12);
662 		else
663 			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
664 			       adev->gart.table_addr >> 12);
665 	}
666 
667 	/* enable context1-15 */
668 	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
669 	       (u32)(adev->dummy_page.addr >> 12));
670 	WREG32(mmVM_CONTEXT1_CNTL2, 4);
671 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
672 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
673 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
674 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
675 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
676 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
677 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
678 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
679 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
680 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
681 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
682 			    amdgpu_vm_block_size - 9);
683 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
684 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
685 		gmc_v8_0_set_fault_enable_default(adev, false);
686 	else
687 		gmc_v8_0_set_fault_enable_default(adev, true);
688 
689 	gmc_v8_0_gart_flush_gpu_tlb(adev, 0);
690 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
691 		 (unsigned)(adev->mc.gtt_size >> 20),
692 		 (unsigned long long)adev->gart.table_addr);
693 	adev->gart.ready = true;
694 	return 0;
695 }
696 
697 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
698 {
699 	int r;
700 
701 	if (adev->gart.robj) {
702 		WARN(1, "R600 PCIE GART already initialized\n");
703 		return 0;
704 	}
705 	/* Initialize common gart structure */
706 	r = amdgpu_gart_init(adev);
707 	if (r)
708 		return r;
709 	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
710 	return amdgpu_gart_table_vram_alloc(adev);
711 }
712 
713 /**
714  * gmc_v8_0_gart_disable - gart disable
715  *
716  * @adev: amdgpu_device pointer
717  *
718  * This disables all VM page table (CIK).
719  */
720 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
721 {
722 	u32 tmp;
723 
724 	/* Disable all tables */
725 	WREG32(mmVM_CONTEXT0_CNTL, 0);
726 	WREG32(mmVM_CONTEXT1_CNTL, 0);
727 	/* Setup TLB control */
728 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
729 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
730 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
731 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
732 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
733 	/* Setup L2 cache */
734 	tmp = RREG32(mmVM_L2_CNTL);
735 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
736 	WREG32(mmVM_L2_CNTL, tmp);
737 	WREG32(mmVM_L2_CNTL2, 0);
738 	amdgpu_gart_table_vram_unpin(adev);
739 }
740 
741 /**
742  * gmc_v8_0_gart_fini - vm fini callback
743  *
744  * @adev: amdgpu_device pointer
745  *
746  * Tears down the driver GART/VM setup (CIK).
747  */
748 static void gmc_v8_0_gart_fini(struct amdgpu_device *adev)
749 {
750 	amdgpu_gart_table_vram_free(adev);
751 	amdgpu_gart_fini(adev);
752 }
753 
754 /*
755  * vm
756  * VMID 0 is the physical GPU addresses as used by the kernel.
757  * VMIDs 1-15 are used for userspace clients and are handled
758  * by the amdgpu vm/hsa code.
759  */
760 /**
761  * gmc_v8_0_vm_init - cik vm init callback
762  *
763  * @adev: amdgpu_device pointer
764  *
765  * Inits cik specific vm parameters (number of VMs, base of vram for
766  * VMIDs 1-15) (CIK).
767  * Returns 0 for success.
768  */
769 static int gmc_v8_0_vm_init(struct amdgpu_device *adev)
770 {
771 	/*
772 	 * number of VMs
773 	 * VMID 0 is reserved for System
774 	 * amdgpu graphics/compute will use VMIDs 1-7
775 	 * amdkfd will use VMIDs 8-15
776 	 */
777 	adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS;
778 
779 	/* base offset of vram pages */
780 	if (adev->flags & AMD_IS_APU) {
781 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
782 		tmp <<= 22;
783 		adev->vm_manager.vram_base_offset = tmp;
784 	} else
785 		adev->vm_manager.vram_base_offset = 0;
786 
787 	return 0;
788 }
789 
790 /**
791  * gmc_v8_0_vm_fini - cik vm fini callback
792  *
793  * @adev: amdgpu_device pointer
794  *
795  * Tear down any asic specific VM setup (CIK).
796  */
797 static void gmc_v8_0_vm_fini(struct amdgpu_device *adev)
798 {
799 }
800 
801 /**
802  * gmc_v8_0_vm_decode_fault - print human readable fault info
803  *
804  * @adev: amdgpu_device pointer
805  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
806  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
807  *
808  * Print human readable fault information (CIK).
809  */
810 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev,
811 				     u32 status, u32 addr, u32 mc_client)
812 {
813 	u32 mc_id;
814 	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
815 	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
816 					PROTECTIONS);
817 	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
818 		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
819 
820 	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
821 			      MEMORY_CLIENT_ID);
822 
823 	printk("VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
824 	       protections, vmid, addr,
825 	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
826 			     MEMORY_CLIENT_RW) ?
827 	       "write" : "read", block, mc_client, mc_id);
828 }
829 
830 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
831 {
832 	switch (mc_seq_vram_type) {
833 	case MC_SEQ_MISC0__MT__GDDR1:
834 		return AMDGPU_VRAM_TYPE_GDDR1;
835 	case MC_SEQ_MISC0__MT__DDR2:
836 		return AMDGPU_VRAM_TYPE_DDR2;
837 	case MC_SEQ_MISC0__MT__GDDR3:
838 		return AMDGPU_VRAM_TYPE_GDDR3;
839 	case MC_SEQ_MISC0__MT__GDDR4:
840 		return AMDGPU_VRAM_TYPE_GDDR4;
841 	case MC_SEQ_MISC0__MT__GDDR5:
842 		return AMDGPU_VRAM_TYPE_GDDR5;
843 	case MC_SEQ_MISC0__MT__HBM:
844 		return AMDGPU_VRAM_TYPE_HBM;
845 	case MC_SEQ_MISC0__MT__DDR3:
846 		return AMDGPU_VRAM_TYPE_DDR3;
847 	default:
848 		return AMDGPU_VRAM_TYPE_UNKNOWN;
849 	}
850 }
851 
852 static int gmc_v8_0_early_init(void *handle)
853 {
854 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
855 
856 	gmc_v8_0_set_gart_funcs(adev);
857 	gmc_v8_0_set_irq_funcs(adev);
858 
859 	if (adev->flags & AMD_IS_APU) {
860 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
861 	} else {
862 		u32 tmp = RREG32(mmMC_SEQ_MISC0);
863 		tmp &= MC_SEQ_MISC0__MT__MASK;
864 		adev->mc.vram_type = gmc_v8_0_convert_vram_type(tmp);
865 	}
866 
867 	return 0;
868 }
869 
870 static int gmc_v8_0_late_init(void *handle)
871 {
872 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
873 
874 	return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
875 }
876 
877 static int gmc_v8_0_sw_init(void *handle)
878 {
879 	int r;
880 	int dma_bits;
881 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
882 
883 	r = amdgpu_gem_init(adev);
884 	if (r)
885 		return r;
886 
887 	r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault);
888 	if (r)
889 		return r;
890 
891 	r = amdgpu_irq_add_id(adev, 147, &adev->mc.vm_fault);
892 	if (r)
893 		return r;
894 
895 	/* Adjust VM size here.
896 	 * Currently set to 4GB ((1 << 20) 4k pages).
897 	 * Max GPUVM size for cayman and SI is 40 bits.
898 	 */
899 	adev->vm_manager.max_pfn = amdgpu_vm_size << 18;
900 
901 	/* Set the internal MC address mask
902 	 * This is the max address of the GPU's
903 	 * internal address space.
904 	 */
905 	adev->mc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
906 
907 	/* set DMA mask + need_dma32 flags.
908 	 * PCIE - can handle 40-bits.
909 	 * IGP - can handle 40-bits
910 	 * PCI - dma32 for legacy pci gart, 40 bits on newer asics
911 	 */
912 	adev->need_dma32 = false;
913 	dma_bits = adev->need_dma32 ? 32 : 40;
914 	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
915 	if (r) {
916 		adev->need_dma32 = true;
917 		dma_bits = 32;
918 		printk(KERN_WARNING "amdgpu: No suitable DMA available.\n");
919 	}
920 	r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
921 	if (r) {
922 		pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
923 		printk(KERN_WARNING "amdgpu: No coherent DMA available.\n");
924 	}
925 
926 	r = gmc_v8_0_init_microcode(adev);
927 	if (r) {
928 		DRM_ERROR("Failed to load mc firmware!\n");
929 		return r;
930 	}
931 
932 	r = gmc_v8_0_mc_init(adev);
933 	if (r)
934 		return r;
935 
936 	/* Memory manager */
937 	r = amdgpu_bo_init(adev);
938 	if (r)
939 		return r;
940 
941 	r = gmc_v8_0_gart_init(adev);
942 	if (r)
943 		return r;
944 
945 	if (!adev->vm_manager.enabled) {
946 		r = gmc_v8_0_vm_init(adev);
947 		if (r) {
948 			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
949 			return r;
950 		}
951 		adev->vm_manager.enabled = true;
952 	}
953 
954 	return r;
955 }
956 
957 static int gmc_v8_0_sw_fini(void *handle)
958 {
959 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
960 
961 	if (adev->vm_manager.enabled) {
962 		amdgpu_vm_manager_fini(adev);
963 		gmc_v8_0_vm_fini(adev);
964 		adev->vm_manager.enabled = false;
965 	}
966 	gmc_v8_0_gart_fini(adev);
967 	amdgpu_gem_fini(adev);
968 	amdgpu_bo_fini(adev);
969 
970 	return 0;
971 }
972 
973 static int gmc_v8_0_hw_init(void *handle)
974 {
975 	int r;
976 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
977 
978 	gmc_v8_0_init_golden_registers(adev);
979 
980 	gmc_v8_0_mc_program(adev);
981 
982 	if (adev->asic_type == CHIP_TONGA) {
983 		r = gmc_v8_0_mc_load_microcode(adev);
984 		if (r) {
985 			DRM_ERROR("Failed to load MC firmware!\n");
986 			return r;
987 		}
988 	}
989 
990 	r = gmc_v8_0_gart_enable(adev);
991 	if (r)
992 		return r;
993 
994 	return r;
995 }
996 
997 static int gmc_v8_0_hw_fini(void *handle)
998 {
999 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1000 
1001 	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
1002 	gmc_v8_0_gart_disable(adev);
1003 
1004 	return 0;
1005 }
1006 
1007 static int gmc_v8_0_suspend(void *handle)
1008 {
1009 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1010 
1011 	if (adev->vm_manager.enabled) {
1012 		gmc_v8_0_vm_fini(adev);
1013 		adev->vm_manager.enabled = false;
1014 	}
1015 	gmc_v8_0_hw_fini(adev);
1016 
1017 	return 0;
1018 }
1019 
1020 static int gmc_v8_0_resume(void *handle)
1021 {
1022 	int r;
1023 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1024 
1025 	r = gmc_v8_0_hw_init(adev);
1026 	if (r)
1027 		return r;
1028 
1029 	if (!adev->vm_manager.enabled) {
1030 		r = gmc_v8_0_vm_init(adev);
1031 		if (r) {
1032 			dev_err(adev->dev, "vm manager initialization failed (%d).\n", r);
1033 			return r;
1034 		}
1035 		adev->vm_manager.enabled = true;
1036 	}
1037 
1038 	return r;
1039 }
1040 
1041 static bool gmc_v8_0_is_idle(void *handle)
1042 {
1043 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1044 	u32 tmp = RREG32(mmSRBM_STATUS);
1045 
1046 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1047 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1048 		return false;
1049 
1050 	return true;
1051 }
1052 
1053 static int gmc_v8_0_wait_for_idle(void *handle)
1054 {
1055 	unsigned i;
1056 	u32 tmp;
1057 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1058 
1059 	for (i = 0; i < adev->usec_timeout; i++) {
1060 		/* read MC_STATUS */
1061 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1062 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1063 					       SRBM_STATUS__MCC_BUSY_MASK |
1064 					       SRBM_STATUS__MCD_BUSY_MASK |
1065 					       SRBM_STATUS__VMC_BUSY_MASK |
1066 					       SRBM_STATUS__VMC1_BUSY_MASK);
1067 		if (!tmp)
1068 			return 0;
1069 		udelay(1);
1070 	}
1071 	return -ETIMEDOUT;
1072 
1073 }
1074 
1075 static void gmc_v8_0_print_status(void *handle)
1076 {
1077 	int i, j;
1078 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1079 
1080 	dev_info(adev->dev, "GMC 8.x registers\n");
1081 	dev_info(adev->dev, "  SRBM_STATUS=0x%08X\n",
1082 		RREG32(mmSRBM_STATUS));
1083 	dev_info(adev->dev, "  SRBM_STATUS2=0x%08X\n",
1084 		RREG32(mmSRBM_STATUS2));
1085 
1086 	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1087 		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR));
1088 	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1089 		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS));
1090 	dev_info(adev->dev, "  MC_VM_MX_L1_TLB_CNTL=0x%08X\n",
1091 		 RREG32(mmMC_VM_MX_L1_TLB_CNTL));
1092 	dev_info(adev->dev, "  VM_L2_CNTL=0x%08X\n",
1093 		 RREG32(mmVM_L2_CNTL));
1094 	dev_info(adev->dev, "  VM_L2_CNTL2=0x%08X\n",
1095 		 RREG32(mmVM_L2_CNTL2));
1096 	dev_info(adev->dev, "  VM_L2_CNTL3=0x%08X\n",
1097 		 RREG32(mmVM_L2_CNTL3));
1098 	dev_info(adev->dev, "  VM_L2_CNTL4=0x%08X\n",
1099 		 RREG32(mmVM_L2_CNTL4));
1100 	dev_info(adev->dev, "  VM_CONTEXT0_PAGE_TABLE_START_ADDR=0x%08X\n",
1101 		 RREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR));
1102 	dev_info(adev->dev, "  VM_CONTEXT0_PAGE_TABLE_END_ADDR=0x%08X\n",
1103 		 RREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR));
1104 	dev_info(adev->dev, "  VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR=0x%08X\n",
1105 		 RREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR));
1106 	dev_info(adev->dev, "  VM_CONTEXT0_CNTL2=0x%08X\n",
1107 		 RREG32(mmVM_CONTEXT0_CNTL2));
1108 	dev_info(adev->dev, "  VM_CONTEXT0_CNTL=0x%08X\n",
1109 		 RREG32(mmVM_CONTEXT0_CNTL));
1110 	dev_info(adev->dev, "  VM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR=0x%08X\n",
1111 		 RREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR));
1112 	dev_info(adev->dev, "  VM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR=0x%08X\n",
1113 		 RREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR));
1114 	dev_info(adev->dev, "  mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET=0x%08X\n",
1115 		 RREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET));
1116 	dev_info(adev->dev, "  VM_CONTEXT1_PAGE_TABLE_START_ADDR=0x%08X\n",
1117 		 RREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR));
1118 	dev_info(adev->dev, "  VM_CONTEXT1_PAGE_TABLE_END_ADDR=0x%08X\n",
1119 		 RREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR));
1120 	dev_info(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR=0x%08X\n",
1121 		 RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR));
1122 	dev_info(adev->dev, "  VM_CONTEXT1_CNTL2=0x%08X\n",
1123 		 RREG32(mmVM_CONTEXT1_CNTL2));
1124 	dev_info(adev->dev, "  VM_CONTEXT1_CNTL=0x%08X\n",
1125 		 RREG32(mmVM_CONTEXT1_CNTL));
1126 	for (i = 0; i < 16; i++) {
1127 		if (i < 8)
1128 			dev_info(adev->dev, "  VM_CONTEXT%d_PAGE_TABLE_BASE_ADDR=0x%08X\n",
1129 				 i, RREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i));
1130 		else
1131 			dev_info(adev->dev, "  VM_CONTEXT%d_PAGE_TABLE_BASE_ADDR=0x%08X\n",
1132 				 i, RREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8));
1133 	}
1134 	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_LOW_ADDR=0x%08X\n",
1135 		 RREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR));
1136 	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_HIGH_ADDR=0x%08X\n",
1137 		 RREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR));
1138 	dev_info(adev->dev, "  MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR=0x%08X\n",
1139 		 RREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR));
1140 	dev_info(adev->dev, "  MC_VM_FB_LOCATION=0x%08X\n",
1141 		 RREG32(mmMC_VM_FB_LOCATION));
1142 	dev_info(adev->dev, "  MC_VM_AGP_BASE=0x%08X\n",
1143 		 RREG32(mmMC_VM_AGP_BASE));
1144 	dev_info(adev->dev, "  MC_VM_AGP_TOP=0x%08X\n",
1145 		 RREG32(mmMC_VM_AGP_TOP));
1146 	dev_info(adev->dev, "  MC_VM_AGP_BOT=0x%08X\n",
1147 		 RREG32(mmMC_VM_AGP_BOT));
1148 
1149 	dev_info(adev->dev, "  HDP_REG_COHERENCY_FLUSH_CNTL=0x%08X\n",
1150 		 RREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL));
1151 	dev_info(adev->dev, "  HDP_NONSURFACE_BASE=0x%08X\n",
1152 		 RREG32(mmHDP_NONSURFACE_BASE));
1153 	dev_info(adev->dev, "  HDP_NONSURFACE_INFO=0x%08X\n",
1154 		 RREG32(mmHDP_NONSURFACE_INFO));
1155 	dev_info(adev->dev, "  HDP_NONSURFACE_SIZE=0x%08X\n",
1156 		 RREG32(mmHDP_NONSURFACE_SIZE));
1157 	dev_info(adev->dev, "  HDP_MISC_CNTL=0x%08X\n",
1158 		 RREG32(mmHDP_MISC_CNTL));
1159 	dev_info(adev->dev, "  HDP_HOST_PATH_CNTL=0x%08X\n",
1160 		 RREG32(mmHDP_HOST_PATH_CNTL));
1161 
1162 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
1163 		dev_info(adev->dev, "  %d:\n", i);
1164 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1165 			 0xb05 + j, RREG32(0xb05 + j));
1166 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1167 			 0xb06 + j, RREG32(0xb06 + j));
1168 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1169 			 0xb07 + j, RREG32(0xb07 + j));
1170 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1171 			 0xb08 + j, RREG32(0xb08 + j));
1172 		dev_info(adev->dev, "  0x%04X=0x%08X\n",
1173 			 0xb09 + j, RREG32(0xb09 + j));
1174 	}
1175 
1176 	dev_info(adev->dev, "  BIF_FB_EN=0x%08X\n",
1177 		 RREG32(mmBIF_FB_EN));
1178 }
1179 
1180 static int gmc_v8_0_soft_reset(void *handle)
1181 {
1182 	struct amdgpu_mode_mc_save save;
1183 	u32 srbm_soft_reset = 0;
1184 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1185 	u32 tmp = RREG32(mmSRBM_STATUS);
1186 
1187 	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1188 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1189 						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1190 
1191 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1192 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1193 		if (!(adev->flags & AMD_IS_APU))
1194 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1195 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1196 	}
1197 
1198 	if (srbm_soft_reset) {
1199 		gmc_v8_0_print_status((void *)adev);
1200 
1201 		gmc_v8_0_mc_stop(adev, &save);
1202 		if (gmc_v8_0_wait_for_idle(adev)) {
1203 			dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1204 		}
1205 
1206 
1207 		tmp = RREG32(mmSRBM_SOFT_RESET);
1208 		tmp |= srbm_soft_reset;
1209 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1210 		WREG32(mmSRBM_SOFT_RESET, tmp);
1211 		tmp = RREG32(mmSRBM_SOFT_RESET);
1212 
1213 		udelay(50);
1214 
1215 		tmp &= ~srbm_soft_reset;
1216 		WREG32(mmSRBM_SOFT_RESET, tmp);
1217 		tmp = RREG32(mmSRBM_SOFT_RESET);
1218 
1219 		/* Wait a little for things to settle down */
1220 		udelay(50);
1221 
1222 		gmc_v8_0_mc_resume(adev, &save);
1223 		udelay(50);
1224 
1225 		gmc_v8_0_print_status((void *)adev);
1226 	}
1227 
1228 	return 0;
1229 }
1230 
1231 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1232 					     struct amdgpu_irq_src *src,
1233 					     unsigned type,
1234 					     enum amdgpu_interrupt_state state)
1235 {
1236 	u32 tmp;
1237 	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1238 		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1239 		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1240 		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1241 		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1242 		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1243 		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1244 
1245 	switch (state) {
1246 	case AMDGPU_IRQ_STATE_DISABLE:
1247 		/* system context */
1248 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1249 		tmp &= ~bits;
1250 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1251 		/* VMs */
1252 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1253 		tmp &= ~bits;
1254 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1255 		break;
1256 	case AMDGPU_IRQ_STATE_ENABLE:
1257 		/* system context */
1258 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1259 		tmp |= bits;
1260 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1261 		/* VMs */
1262 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1263 		tmp |= bits;
1264 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1265 		break;
1266 	default:
1267 		break;
1268 	}
1269 
1270 	return 0;
1271 }
1272 
1273 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1274 				      struct amdgpu_irq_src *source,
1275 				      struct amdgpu_iv_entry *entry)
1276 {
1277 	u32 addr, status, mc_client;
1278 
1279 	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1280 	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1281 	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1282 	/* reset addr and status */
1283 	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1284 
1285 	if (!addr && !status)
1286 		return 0;
1287 
1288 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1289 		gmc_v8_0_set_fault_enable_default(adev, false);
1290 
1291 	dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1292 		entry->src_id, entry->src_data);
1293 	dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1294 		addr);
1295 	dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1296 		status);
1297 	gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client);
1298 
1299 	return 0;
1300 }
1301 
1302 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1303 		bool enable)
1304 {
1305 	uint32_t data;
1306 
1307 	if (enable) {
1308 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1309 		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1310 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1311 
1312 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1313 		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1314 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1315 
1316 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1317 		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1318 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1319 
1320 		data = RREG32(mmMC_XPB_CLK_GAT);
1321 		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1322 		WREG32(mmMC_XPB_CLK_GAT, data);
1323 
1324 		data = RREG32(mmATC_MISC_CG);
1325 		data |= ATC_MISC_CG__ENABLE_MASK;
1326 		WREG32(mmATC_MISC_CG, data);
1327 
1328 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1329 		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1330 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1331 
1332 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1333 		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1334 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1335 
1336 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1337 		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1338 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1339 
1340 		data = RREG32(mmVM_L2_CG);
1341 		data |= VM_L2_CG__ENABLE_MASK;
1342 		WREG32(mmVM_L2_CG, data);
1343 	} else {
1344 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1345 		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1346 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1347 
1348 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1349 		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1350 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1351 
1352 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1353 		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1354 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1355 
1356 		data = RREG32(mmMC_XPB_CLK_GAT);
1357 		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1358 		WREG32(mmMC_XPB_CLK_GAT, data);
1359 
1360 		data = RREG32(mmATC_MISC_CG);
1361 		data &= ~ATC_MISC_CG__ENABLE_MASK;
1362 		WREG32(mmATC_MISC_CG, data);
1363 
1364 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1365 		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1366 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1367 
1368 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1369 		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1370 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1371 
1372 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1373 		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1374 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1375 
1376 		data = RREG32(mmVM_L2_CG);
1377 		data &= ~VM_L2_CG__ENABLE_MASK;
1378 		WREG32(mmVM_L2_CG, data);
1379 	}
1380 }
1381 
1382 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1383 		bool enable)
1384 {
1385 	uint32_t data;
1386 
1387 	if (enable) {
1388 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1389 		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1390 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1391 
1392 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1393 		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1394 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1395 
1396 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1397 		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1398 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1399 
1400 		data = RREG32(mmMC_XPB_CLK_GAT);
1401 		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1402 		WREG32(mmMC_XPB_CLK_GAT, data);
1403 
1404 		data = RREG32(mmATC_MISC_CG);
1405 		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1406 		WREG32(mmATC_MISC_CG, data);
1407 
1408 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1409 		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1410 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1411 
1412 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1413 		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1414 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1415 
1416 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1417 		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1418 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1419 
1420 		data = RREG32(mmVM_L2_CG);
1421 		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1422 		WREG32(mmVM_L2_CG, data);
1423 	} else {
1424 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1425 		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1426 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1427 
1428 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1429 		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1430 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1431 
1432 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1433 		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1434 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1435 
1436 		data = RREG32(mmMC_XPB_CLK_GAT);
1437 		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1438 		WREG32(mmMC_XPB_CLK_GAT, data);
1439 
1440 		data = RREG32(mmATC_MISC_CG);
1441 		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1442 		WREG32(mmATC_MISC_CG, data);
1443 
1444 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1445 		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1446 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1447 
1448 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1449 		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1450 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1451 
1452 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1453 		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1454 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1455 
1456 		data = RREG32(mmVM_L2_CG);
1457 		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1458 		WREG32(mmVM_L2_CG, data);
1459 	}
1460 }
1461 
1462 static int gmc_v8_0_set_clockgating_state(void *handle,
1463 					  enum amd_clockgating_state state)
1464 {
1465 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1466 
1467 	switch (adev->asic_type) {
1468 	case CHIP_FIJI:
1469 		fiji_update_mc_medium_grain_clock_gating(adev,
1470 				state == AMD_CG_STATE_GATE ? true : false);
1471 		fiji_update_mc_light_sleep(adev,
1472 				state == AMD_CG_STATE_GATE ? true : false);
1473 		break;
1474 	default:
1475 		break;
1476 	}
1477 	return 0;
1478 }
1479 
1480 static int gmc_v8_0_set_powergating_state(void *handle,
1481 					  enum amd_powergating_state state)
1482 {
1483 	return 0;
1484 }
1485 
1486 const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1487 	.early_init = gmc_v8_0_early_init,
1488 	.late_init = gmc_v8_0_late_init,
1489 	.sw_init = gmc_v8_0_sw_init,
1490 	.sw_fini = gmc_v8_0_sw_fini,
1491 	.hw_init = gmc_v8_0_hw_init,
1492 	.hw_fini = gmc_v8_0_hw_fini,
1493 	.suspend = gmc_v8_0_suspend,
1494 	.resume = gmc_v8_0_resume,
1495 	.is_idle = gmc_v8_0_is_idle,
1496 	.wait_for_idle = gmc_v8_0_wait_for_idle,
1497 	.soft_reset = gmc_v8_0_soft_reset,
1498 	.print_status = gmc_v8_0_print_status,
1499 	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1500 	.set_powergating_state = gmc_v8_0_set_powergating_state,
1501 };
1502 
1503 static const struct amdgpu_gart_funcs gmc_v8_0_gart_funcs = {
1504 	.flush_gpu_tlb = gmc_v8_0_gart_flush_gpu_tlb,
1505 	.set_pte_pde = gmc_v8_0_gart_set_pte_pde,
1506 };
1507 
1508 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1509 	.set = gmc_v8_0_vm_fault_interrupt_state,
1510 	.process = gmc_v8_0_process_interrupt,
1511 };
1512 
1513 static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev)
1514 {
1515 	if (adev->gart.gart_funcs == NULL)
1516 		adev->gart.gart_funcs = &gmc_v8_0_gart_funcs;
1517 }
1518 
1519 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1520 {
1521 	adev->mc.vm_fault.num_types = 1;
1522 	adev->mc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1523 }
1524