xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c (revision 711aab1d)
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 <drm/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 "dce/dce_10_0_d.h"
39 #include "dce/dce_10_0_sh_mask.h"
40 
41 #include "vid.h"
42 #include "vi.h"
43 
44 #include "amdgpu_atombios.h"
45 
46 
47 static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev);
48 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
49 static int gmc_v8_0_wait_for_idle(void *handle);
50 
51 MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
52 MODULE_FIRMWARE("amdgpu/polaris11_mc.bin");
53 MODULE_FIRMWARE("amdgpu/polaris10_mc.bin");
54 MODULE_FIRMWARE("amdgpu/polaris12_mc.bin");
55 
56 static const u32 golden_settings_tonga_a11[] =
57 {
58 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
59 	mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
60 	mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
61 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
62 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
63 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
64 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
65 };
66 
67 static const u32 tonga_mgcg_cgcg_init[] =
68 {
69 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
70 };
71 
72 static const u32 golden_settings_fiji_a10[] =
73 {
74 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
75 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
76 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
77 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
78 };
79 
80 static const u32 fiji_mgcg_cgcg_init[] =
81 {
82 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
83 };
84 
85 static const u32 golden_settings_polaris11_a11[] =
86 {
87 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
88 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
89 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
90 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
91 };
92 
93 static const u32 golden_settings_polaris10_a11[] =
94 {
95 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
96 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
97 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
98 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
99 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
100 };
101 
102 static const u32 cz_mgcg_cgcg_init[] =
103 {
104 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
105 };
106 
107 static const u32 stoney_mgcg_cgcg_init[] =
108 {
109 	mmATC_MISC_CG, 0xffffffff, 0x000c0200,
110 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
111 };
112 
113 static const u32 golden_settings_stoney_common[] =
114 {
115 	mmMC_HUB_RDREQ_UVD, MC_HUB_RDREQ_UVD__PRESCALE_MASK, 0x00000004,
116 	mmMC_RD_GRP_OTH, MC_RD_GRP_OTH__UVD_MASK, 0x00600000
117 };
118 
119 static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
120 {
121 	switch (adev->asic_type) {
122 	case CHIP_FIJI:
123 		amdgpu_program_register_sequence(adev,
124 						 fiji_mgcg_cgcg_init,
125 						 (const u32)ARRAY_SIZE(fiji_mgcg_cgcg_init));
126 		amdgpu_program_register_sequence(adev,
127 						 golden_settings_fiji_a10,
128 						 (const u32)ARRAY_SIZE(golden_settings_fiji_a10));
129 		break;
130 	case CHIP_TONGA:
131 		amdgpu_program_register_sequence(adev,
132 						 tonga_mgcg_cgcg_init,
133 						 (const u32)ARRAY_SIZE(tonga_mgcg_cgcg_init));
134 		amdgpu_program_register_sequence(adev,
135 						 golden_settings_tonga_a11,
136 						 (const u32)ARRAY_SIZE(golden_settings_tonga_a11));
137 		break;
138 	case CHIP_POLARIS11:
139 	case CHIP_POLARIS12:
140 		amdgpu_program_register_sequence(adev,
141 						 golden_settings_polaris11_a11,
142 						 (const u32)ARRAY_SIZE(golden_settings_polaris11_a11));
143 		break;
144 	case CHIP_POLARIS10:
145 		amdgpu_program_register_sequence(adev,
146 						 golden_settings_polaris10_a11,
147 						 (const u32)ARRAY_SIZE(golden_settings_polaris10_a11));
148 		break;
149 	case CHIP_CARRIZO:
150 		amdgpu_program_register_sequence(adev,
151 						 cz_mgcg_cgcg_init,
152 						 (const u32)ARRAY_SIZE(cz_mgcg_cgcg_init));
153 		break;
154 	case CHIP_STONEY:
155 		amdgpu_program_register_sequence(adev,
156 						 stoney_mgcg_cgcg_init,
157 						 (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init));
158 		amdgpu_program_register_sequence(adev,
159 						 golden_settings_stoney_common,
160 						 (const u32)ARRAY_SIZE(golden_settings_stoney_common));
161 		break;
162 	default:
163 		break;
164 	}
165 }
166 
167 static void gmc_v8_0_mc_stop(struct amdgpu_device *adev)
168 {
169 	u32 blackout;
170 
171 	gmc_v8_0_wait_for_idle(adev);
172 
173 	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
174 	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
175 		/* Block CPU access */
176 		WREG32(mmBIF_FB_EN, 0);
177 		/* blackout the MC */
178 		blackout = REG_SET_FIELD(blackout,
179 					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
180 		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
181 	}
182 	/* wait for the MC to settle */
183 	udelay(100);
184 }
185 
186 static void gmc_v8_0_mc_resume(struct amdgpu_device *adev)
187 {
188 	u32 tmp;
189 
190 	/* unblackout the MC */
191 	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
192 	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
193 	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
194 	/* allow CPU access */
195 	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
196 	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
197 	WREG32(mmBIF_FB_EN, tmp);
198 }
199 
200 /**
201  * gmc_v8_0_init_microcode - load ucode images from disk
202  *
203  * @adev: amdgpu_device pointer
204  *
205  * Use the firmware interface to load the ucode images into
206  * the driver (not loaded into hw).
207  * Returns 0 on success, error on failure.
208  */
209 static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
210 {
211 	const char *chip_name;
212 	char fw_name[30];
213 	int err;
214 
215 	DRM_DEBUG("\n");
216 
217 	switch (adev->asic_type) {
218 	case CHIP_TONGA:
219 		chip_name = "tonga";
220 		break;
221 	case CHIP_POLARIS11:
222 		chip_name = "polaris11";
223 		break;
224 	case CHIP_POLARIS10:
225 		chip_name = "polaris10";
226 		break;
227 	case CHIP_POLARIS12:
228 		chip_name = "polaris12";
229 		break;
230 	case CHIP_FIJI:
231 	case CHIP_CARRIZO:
232 	case CHIP_STONEY:
233 		return 0;
234 	default: BUG();
235 	}
236 
237 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
238 	err = request_firmware(&adev->mc.fw, fw_name, adev->dev);
239 	if (err)
240 		goto out;
241 	err = amdgpu_ucode_validate(adev->mc.fw);
242 
243 out:
244 	if (err) {
245 		pr_err("mc: Failed to load firmware \"%s\"\n", fw_name);
246 		release_firmware(adev->mc.fw);
247 		adev->mc.fw = NULL;
248 	}
249 	return err;
250 }
251 
252 /**
253  * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
254  *
255  * @adev: amdgpu_device pointer
256  *
257  * Load the GDDR MC ucode into the hw (CIK).
258  * Returns 0 on success, error on failure.
259  */
260 static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
261 {
262 	const struct mc_firmware_header_v1_0 *hdr;
263 	const __le32 *fw_data = NULL;
264 	const __le32 *io_mc_regs = NULL;
265 	u32 running;
266 	int i, ucode_size, regs_size;
267 
268 	/* Skip MC ucode loading on SR-IOV capable boards.
269 	 * vbios does this for us in asic_init in that case.
270 	 * Skip MC ucode loading on VF, because hypervisor will do that
271 	 * for this adaptor.
272 	 */
273 	if (amdgpu_sriov_bios(adev))
274 		return 0;
275 
276 	if (!adev->mc.fw)
277 		return -EINVAL;
278 
279 	hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
280 	amdgpu_ucode_print_mc_hdr(&hdr->header);
281 
282 	adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
283 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
284 	io_mc_regs = (const __le32 *)
285 		(adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
286 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
287 	fw_data = (const __le32 *)
288 		(adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
289 
290 	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
291 
292 	if (running == 0) {
293 		/* reset the engine and set to writable */
294 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
295 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
296 
297 		/* load mc io regs */
298 		for (i = 0; i < regs_size; i++) {
299 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
300 			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
301 		}
302 		/* load the MC ucode */
303 		for (i = 0; i < ucode_size; i++)
304 			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
305 
306 		/* put the engine back into the active state */
307 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
308 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
309 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
310 
311 		/* wait for training to complete */
312 		for (i = 0; i < adev->usec_timeout; i++) {
313 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
314 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
315 				break;
316 			udelay(1);
317 		}
318 		for (i = 0; i < adev->usec_timeout; i++) {
319 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
320 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
321 				break;
322 			udelay(1);
323 		}
324 	}
325 
326 	return 0;
327 }
328 
329 static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
330 {
331 	const struct mc_firmware_header_v1_0 *hdr;
332 	const __le32 *fw_data = NULL;
333 	const __le32 *io_mc_regs = NULL;
334 	u32 data, vbios_version;
335 	int i, ucode_size, regs_size;
336 
337 	/* Skip MC ucode loading on SR-IOV capable boards.
338 	 * vbios does this for us in asic_init in that case.
339 	 * Skip MC ucode loading on VF, because hypervisor will do that
340 	 * for this adaptor.
341 	 */
342 	if (amdgpu_sriov_bios(adev))
343 		return 0;
344 
345 	WREG32(mmMC_SEQ_IO_DEBUG_INDEX, 0x9F);
346 	data = RREG32(mmMC_SEQ_IO_DEBUG_DATA);
347 	vbios_version = data & 0xf;
348 
349 	if (vbios_version == 0)
350 		return 0;
351 
352 	if (!adev->mc.fw)
353 		return -EINVAL;
354 
355 	hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data;
356 	amdgpu_ucode_print_mc_hdr(&hdr->header);
357 
358 	adev->mc.fw_version = le32_to_cpu(hdr->header.ucode_version);
359 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
360 	io_mc_regs = (const __le32 *)
361 		(adev->mc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
362 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
363 	fw_data = (const __le32 *)
364 		(adev->mc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
365 
366 	data = RREG32(mmMC_SEQ_MISC0);
367 	data &= ~(0x40);
368 	WREG32(mmMC_SEQ_MISC0, data);
369 
370 	/* load mc io regs */
371 	for (i = 0; i < regs_size; i++) {
372 		WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
373 		WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
374 	}
375 
376 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
377 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
378 
379 	/* load the MC ucode */
380 	for (i = 0; i < ucode_size; i++)
381 		WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
382 
383 	/* put the engine back into the active state */
384 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
385 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
386 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
387 
388 	/* wait for training to complete */
389 	for (i = 0; i < adev->usec_timeout; i++) {
390 		data = RREG32(mmMC_SEQ_MISC0);
391 		if (data & 0x80)
392 			break;
393 		udelay(1);
394 	}
395 
396 	return 0;
397 }
398 
399 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
400 				       struct amdgpu_mc *mc)
401 {
402 	u64 base = 0;
403 
404 	if (!amdgpu_sriov_vf(adev))
405 		base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
406 	base <<= 24;
407 
408 	if (mc->mc_vram_size > 0xFFC0000000ULL) {
409 		/* leave room for at least 1024M GTT */
410 		dev_warn(adev->dev, "limiting VRAM\n");
411 		mc->real_vram_size = 0xFFC0000000ULL;
412 		mc->mc_vram_size = 0xFFC0000000ULL;
413 	}
414 	amdgpu_vram_location(adev, &adev->mc, base);
415 	amdgpu_gart_location(adev, mc);
416 }
417 
418 /**
419  * gmc_v8_0_mc_program - program the GPU memory controller
420  *
421  * @adev: amdgpu_device pointer
422  *
423  * Set the location of vram, gart, and AGP in the GPU's
424  * physical address space (CIK).
425  */
426 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
427 {
428 	u32 tmp;
429 	int i, j;
430 
431 	/* Initialize HDP */
432 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
433 		WREG32((0xb05 + j), 0x00000000);
434 		WREG32((0xb06 + j), 0x00000000);
435 		WREG32((0xb07 + j), 0x00000000);
436 		WREG32((0xb08 + j), 0x00000000);
437 		WREG32((0xb09 + j), 0x00000000);
438 	}
439 	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
440 
441 	if (gmc_v8_0_wait_for_idle((void *)adev)) {
442 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
443 	}
444 	if (adev->mode_info.num_crtc) {
445 		/* Lockout access through VGA aperture*/
446 		tmp = RREG32(mmVGA_HDP_CONTROL);
447 		tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
448 		WREG32(mmVGA_HDP_CONTROL, tmp);
449 
450 		/* disable VGA render */
451 		tmp = RREG32(mmVGA_RENDER_CONTROL);
452 		tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
453 		WREG32(mmVGA_RENDER_CONTROL, tmp);
454 	}
455 	/* Update configuration */
456 	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
457 	       adev->mc.vram_start >> 12);
458 	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
459 	       adev->mc.vram_end >> 12);
460 	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
461 	       adev->vram_scratch.gpu_addr >> 12);
462 
463 	if (amdgpu_sriov_vf(adev)) {
464 		tmp = ((adev->mc.vram_end >> 24) & 0xFFFF) << 16;
465 		tmp |= ((adev->mc.vram_start >> 24) & 0xFFFF);
466 		WREG32(mmMC_VM_FB_LOCATION, tmp);
467 		/* XXX double check these! */
468 		WREG32(mmHDP_NONSURFACE_BASE, (adev->mc.vram_start >> 8));
469 		WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
470 		WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
471 	}
472 
473 	WREG32(mmMC_VM_AGP_BASE, 0);
474 	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
475 	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
476 	if (gmc_v8_0_wait_for_idle((void *)adev)) {
477 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
478 	}
479 
480 	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
481 
482 	tmp = RREG32(mmHDP_MISC_CNTL);
483 	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
484 	WREG32(mmHDP_MISC_CNTL, tmp);
485 
486 	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
487 	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
488 }
489 
490 /**
491  * gmc_v8_0_mc_init - initialize the memory controller driver params
492  *
493  * @adev: amdgpu_device pointer
494  *
495  * Look up the amount of vram, vram width, and decide how to place
496  * vram and gart within the GPU's physical address space (CIK).
497  * Returns 0 for success.
498  */
499 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
500 {
501 	adev->mc.vram_width = amdgpu_atombios_get_vram_width(adev);
502 	if (!adev->mc.vram_width) {
503 		u32 tmp;
504 		int chansize, numchan;
505 
506 		/* Get VRAM informations */
507 		tmp = RREG32(mmMC_ARB_RAMCFG);
508 		if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
509 			chansize = 64;
510 		} else {
511 			chansize = 32;
512 		}
513 		tmp = RREG32(mmMC_SHARED_CHMAP);
514 		switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
515 		case 0:
516 		default:
517 			numchan = 1;
518 			break;
519 		case 1:
520 			numchan = 2;
521 			break;
522 		case 2:
523 			numchan = 4;
524 			break;
525 		case 3:
526 			numchan = 8;
527 			break;
528 		case 4:
529 			numchan = 3;
530 			break;
531 		case 5:
532 			numchan = 6;
533 			break;
534 		case 6:
535 			numchan = 10;
536 			break;
537 		case 7:
538 			numchan = 12;
539 			break;
540 		case 8:
541 			numchan = 16;
542 			break;
543 		}
544 		adev->mc.vram_width = numchan * chansize;
545 	}
546 	/* Could aper size report 0 ? */
547 	adev->mc.aper_base = pci_resource_start(adev->pdev, 0);
548 	adev->mc.aper_size = pci_resource_len(adev->pdev, 0);
549 	/* size in MB on si */
550 	adev->mc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
551 	adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
552 
553 #ifdef CONFIG_X86_64
554 	if (adev->flags & AMD_IS_APU) {
555 		adev->mc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
556 		adev->mc.aper_size = adev->mc.real_vram_size;
557 	}
558 #endif
559 
560 	/* In case the PCI BAR is larger than the actual amount of vram */
561 	adev->mc.visible_vram_size = adev->mc.aper_size;
562 	if (adev->mc.visible_vram_size > adev->mc.real_vram_size)
563 		adev->mc.visible_vram_size = adev->mc.real_vram_size;
564 
565 	amdgpu_gart_set_defaults(adev);
566 	gmc_v8_0_vram_gtt_location(adev, &adev->mc);
567 
568 	return 0;
569 }
570 
571 /*
572  * GART
573  * VMID 0 is the physical GPU addresses as used by the kernel.
574  * VMIDs 1-15 are used for userspace clients and are handled
575  * by the amdgpu vm/hsa code.
576  */
577 
578 /**
579  * gmc_v8_0_gart_flush_gpu_tlb - gart tlb flush callback
580  *
581  * @adev: amdgpu_device pointer
582  * @vmid: vm instance to flush
583  *
584  * Flush the TLB for the requested page table (CIK).
585  */
586 static void gmc_v8_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
587 					uint32_t vmid)
588 {
589 	/* flush hdp cache */
590 	WREG32(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
591 
592 	/* bits 0-15 are the VM contexts0-15 */
593 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
594 }
595 
596 /**
597  * gmc_v8_0_gart_set_pte_pde - update the page tables using MMIO
598  *
599  * @adev: amdgpu_device pointer
600  * @cpu_pt_addr: cpu address of the page table
601  * @gpu_page_idx: entry in the page table to update
602  * @addr: dst addr to write into pte/pde
603  * @flags: access flags
604  *
605  * Update the page tables using the CPU.
606  */
607 static int gmc_v8_0_gart_set_pte_pde(struct amdgpu_device *adev,
608 				     void *cpu_pt_addr,
609 				     uint32_t gpu_page_idx,
610 				     uint64_t addr,
611 				     uint64_t flags)
612 {
613 	void __iomem *ptr = (void *)cpu_pt_addr;
614 	uint64_t value;
615 
616 	/*
617 	 * PTE format on VI:
618 	 * 63:40 reserved
619 	 * 39:12 4k physical page base address
620 	 * 11:7 fragment
621 	 * 6 write
622 	 * 5 read
623 	 * 4 exe
624 	 * 3 reserved
625 	 * 2 snooped
626 	 * 1 system
627 	 * 0 valid
628 	 *
629 	 * PDE format on VI:
630 	 * 63:59 block fragment size
631 	 * 58:40 reserved
632 	 * 39:1 physical base address of PTE
633 	 * bits 5:1 must be 0.
634 	 * 0 valid
635 	 */
636 	value = addr & 0x000000FFFFFFF000ULL;
637 	value |= flags;
638 	writeq(value, ptr + (gpu_page_idx * 8));
639 
640 	return 0;
641 }
642 
643 static uint64_t gmc_v8_0_get_vm_pte_flags(struct amdgpu_device *adev,
644 					  uint32_t flags)
645 {
646 	uint64_t pte_flag = 0;
647 
648 	if (flags & AMDGPU_VM_PAGE_EXECUTABLE)
649 		pte_flag |= AMDGPU_PTE_EXECUTABLE;
650 	if (flags & AMDGPU_VM_PAGE_READABLE)
651 		pte_flag |= AMDGPU_PTE_READABLE;
652 	if (flags & AMDGPU_VM_PAGE_WRITEABLE)
653 		pte_flag |= AMDGPU_PTE_WRITEABLE;
654 	if (flags & AMDGPU_VM_PAGE_PRT)
655 		pte_flag |= AMDGPU_PTE_PRT;
656 
657 	return pte_flag;
658 }
659 
660 static uint64_t gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, uint64_t addr)
661 {
662 	BUG_ON(addr & 0xFFFFFF0000000FFFULL);
663 	return addr;
664 }
665 
666 /**
667  * gmc_v8_0_set_fault_enable_default - update VM fault handling
668  *
669  * @adev: amdgpu_device pointer
670  * @value: true redirects VM faults to the default page
671  */
672 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
673 					      bool value)
674 {
675 	u32 tmp;
676 
677 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
678 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
679 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
680 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
681 			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
682 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
683 			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
684 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
685 			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
686 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
687 			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
688 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
689 			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
690 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
691 			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
692 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
693 }
694 
695 /**
696  * gmc_v8_0_set_prt - set PRT VM fault
697  *
698  * @adev: amdgpu_device pointer
699  * @enable: enable/disable VM fault handling for PRT
700 */
701 static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
702 {
703 	u32 tmp;
704 
705 	if (enable && !adev->mc.prt_warning) {
706 		dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
707 		adev->mc.prt_warning = true;
708 	}
709 
710 	tmp = RREG32(mmVM_PRT_CNTL);
711 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
712 			    CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
713 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
714 			    CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
715 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
716 			    TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
717 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
718 			    TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
719 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
720 			    L2_CACHE_STORE_INVALID_ENTRIES, enable);
721 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
722 			    L1_TLB_STORE_INVALID_ENTRIES, enable);
723 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
724 			    MASK_PDE0_FAULT, enable);
725 	WREG32(mmVM_PRT_CNTL, tmp);
726 
727 	if (enable) {
728 		uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
729 		uint32_t high = adev->vm_manager.max_pfn;
730 
731 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
732 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
733 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
734 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
735 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
736 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
737 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
738 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
739 	} else {
740 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
741 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
742 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
743 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
744 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
745 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
746 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
747 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
748 	}
749 }
750 
751 /**
752  * gmc_v8_0_gart_enable - gart enable
753  *
754  * @adev: amdgpu_device pointer
755  *
756  * This sets up the TLBs, programs the page tables for VMID0,
757  * sets up the hw for VMIDs 1-15 which are allocated on
758  * demand, and sets up the global locations for the LDS, GDS,
759  * and GPUVM for FSA64 clients (CIK).
760  * Returns 0 for success, errors for failure.
761  */
762 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
763 {
764 	int r, i;
765 	u32 tmp, field;
766 
767 	if (adev->gart.robj == NULL) {
768 		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
769 		return -EINVAL;
770 	}
771 	r = amdgpu_gart_table_vram_pin(adev);
772 	if (r)
773 		return r;
774 	/* Setup TLB control */
775 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
776 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
777 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
778 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
779 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
780 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
781 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
782 	/* Setup L2 cache */
783 	tmp = RREG32(mmVM_L2_CNTL);
784 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
785 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
786 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
787 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
788 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
789 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
790 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
791 	WREG32(mmVM_L2_CNTL, tmp);
792 	tmp = RREG32(mmVM_L2_CNTL2);
793 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
794 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
795 	WREG32(mmVM_L2_CNTL2, tmp);
796 
797 	field = adev->vm_manager.fragment_size;
798 	tmp = RREG32(mmVM_L2_CNTL3);
799 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
800 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
801 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
802 	WREG32(mmVM_L2_CNTL3, tmp);
803 	/* XXX: set to enable PTE/PDE in system memory */
804 	tmp = RREG32(mmVM_L2_CNTL4);
805 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
806 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
807 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
808 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
809 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
810 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
811 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
812 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
813 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
814 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
815 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
816 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
817 	WREG32(mmVM_L2_CNTL4, tmp);
818 	/* setup context0 */
819 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gart_start >> 12);
820 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gart_end >> 12);
821 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12);
822 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
823 			(u32)(adev->dummy_page.addr >> 12));
824 	WREG32(mmVM_CONTEXT0_CNTL2, 0);
825 	tmp = RREG32(mmVM_CONTEXT0_CNTL);
826 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
827 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
828 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
829 	WREG32(mmVM_CONTEXT0_CNTL, tmp);
830 
831 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
832 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
833 	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
834 
835 	/* empty context1-15 */
836 	/* FIXME start with 4G, once using 2 level pt switch to full
837 	 * vm size space
838 	 */
839 	/* set vm size, must be a multiple of 4 */
840 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
841 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
842 	for (i = 1; i < 16; i++) {
843 		if (i < 8)
844 			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
845 			       adev->gart.table_addr >> 12);
846 		else
847 			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
848 			       adev->gart.table_addr >> 12);
849 	}
850 
851 	/* enable context1-15 */
852 	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
853 	       (u32)(adev->dummy_page.addr >> 12));
854 	WREG32(mmVM_CONTEXT1_CNTL2, 4);
855 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
856 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
857 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
858 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
859 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
860 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
861 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
862 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
863 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
864 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
865 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
866 			    adev->vm_manager.block_size - 9);
867 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
868 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
869 		gmc_v8_0_set_fault_enable_default(adev, false);
870 	else
871 		gmc_v8_0_set_fault_enable_default(adev, true);
872 
873 	gmc_v8_0_gart_flush_gpu_tlb(adev, 0);
874 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
875 		 (unsigned)(adev->mc.gart_size >> 20),
876 		 (unsigned long long)adev->gart.table_addr);
877 	adev->gart.ready = true;
878 	return 0;
879 }
880 
881 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
882 {
883 	int r;
884 
885 	if (adev->gart.robj) {
886 		WARN(1, "R600 PCIE GART already initialized\n");
887 		return 0;
888 	}
889 	/* Initialize common gart structure */
890 	r = amdgpu_gart_init(adev);
891 	if (r)
892 		return r;
893 	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
894 	adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
895 	return amdgpu_gart_table_vram_alloc(adev);
896 }
897 
898 /**
899  * gmc_v8_0_gart_disable - gart disable
900  *
901  * @adev: amdgpu_device pointer
902  *
903  * This disables all VM page table (CIK).
904  */
905 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
906 {
907 	u32 tmp;
908 
909 	/* Disable all tables */
910 	WREG32(mmVM_CONTEXT0_CNTL, 0);
911 	WREG32(mmVM_CONTEXT1_CNTL, 0);
912 	/* Setup TLB control */
913 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
914 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
915 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
916 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
917 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
918 	/* Setup L2 cache */
919 	tmp = RREG32(mmVM_L2_CNTL);
920 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
921 	WREG32(mmVM_L2_CNTL, tmp);
922 	WREG32(mmVM_L2_CNTL2, 0);
923 	amdgpu_gart_table_vram_unpin(adev);
924 }
925 
926 /**
927  * gmc_v8_0_gart_fini - vm fini callback
928  *
929  * @adev: amdgpu_device pointer
930  *
931  * Tears down the driver GART/VM setup (CIK).
932  */
933 static void gmc_v8_0_gart_fini(struct amdgpu_device *adev)
934 {
935 	amdgpu_gart_table_vram_free(adev);
936 	amdgpu_gart_fini(adev);
937 }
938 
939 /**
940  * gmc_v8_0_vm_decode_fault - print human readable fault info
941  *
942  * @adev: amdgpu_device pointer
943  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
944  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
945  *
946  * Print human readable fault information (CIK).
947  */
948 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev,
949 				     u32 status, u32 addr, u32 mc_client)
950 {
951 	u32 mc_id;
952 	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
953 	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
954 					PROTECTIONS);
955 	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
956 		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
957 
958 	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
959 			      MEMORY_CLIENT_ID);
960 
961 	dev_err(adev->dev, "VM fault (0x%02x, vmid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
962 	       protections, vmid, addr,
963 	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
964 			     MEMORY_CLIENT_RW) ?
965 	       "write" : "read", block, mc_client, mc_id);
966 }
967 
968 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
969 {
970 	switch (mc_seq_vram_type) {
971 	case MC_SEQ_MISC0__MT__GDDR1:
972 		return AMDGPU_VRAM_TYPE_GDDR1;
973 	case MC_SEQ_MISC0__MT__DDR2:
974 		return AMDGPU_VRAM_TYPE_DDR2;
975 	case MC_SEQ_MISC0__MT__GDDR3:
976 		return AMDGPU_VRAM_TYPE_GDDR3;
977 	case MC_SEQ_MISC0__MT__GDDR4:
978 		return AMDGPU_VRAM_TYPE_GDDR4;
979 	case MC_SEQ_MISC0__MT__GDDR5:
980 		return AMDGPU_VRAM_TYPE_GDDR5;
981 	case MC_SEQ_MISC0__MT__HBM:
982 		return AMDGPU_VRAM_TYPE_HBM;
983 	case MC_SEQ_MISC0__MT__DDR3:
984 		return AMDGPU_VRAM_TYPE_DDR3;
985 	default:
986 		return AMDGPU_VRAM_TYPE_UNKNOWN;
987 	}
988 }
989 
990 static int gmc_v8_0_early_init(void *handle)
991 {
992 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
993 
994 	gmc_v8_0_set_gart_funcs(adev);
995 	gmc_v8_0_set_irq_funcs(adev);
996 
997 	adev->mc.shared_aperture_start = 0x2000000000000000ULL;
998 	adev->mc.shared_aperture_end =
999 		adev->mc.shared_aperture_start + (4ULL << 30) - 1;
1000 	adev->mc.private_aperture_start =
1001 		adev->mc.shared_aperture_end + 1;
1002 	adev->mc.private_aperture_end =
1003 		adev->mc.private_aperture_start + (4ULL << 30) - 1;
1004 
1005 	return 0;
1006 }
1007 
1008 static int gmc_v8_0_late_init(void *handle)
1009 {
1010 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1011 
1012 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1013 		return amdgpu_irq_get(adev, &adev->mc.vm_fault, 0);
1014 	else
1015 		return 0;
1016 }
1017 
1018 #define mmMC_SEQ_MISC0_FIJI 0xA71
1019 
1020 static int gmc_v8_0_sw_init(void *handle)
1021 {
1022 	int r;
1023 	int dma_bits;
1024 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1025 
1026 	if (adev->flags & AMD_IS_APU) {
1027 		adev->mc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1028 	} else {
1029 		u32 tmp;
1030 
1031 		if (adev->asic_type == CHIP_FIJI)
1032 			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1033 		else
1034 			tmp = RREG32(mmMC_SEQ_MISC0);
1035 		tmp &= MC_SEQ_MISC0__MT__MASK;
1036 		adev->mc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1037 	}
1038 
1039 	r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 146, &adev->mc.vm_fault);
1040 	if (r)
1041 		return r;
1042 
1043 	r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 147, &adev->mc.vm_fault);
1044 	if (r)
1045 		return r;
1046 
1047 	/* Adjust VM size here.
1048 	 * Currently set to 4GB ((1 << 20) 4k pages).
1049 	 * Max GPUVM size for cayman and SI is 40 bits.
1050 	 */
1051 	amdgpu_vm_adjust_size(adev, 64, 4);
1052 	adev->vm_manager.max_pfn = adev->vm_manager.vm_size << 18;
1053 
1054 	/* Set the internal MC address mask
1055 	 * This is the max address of the GPU's
1056 	 * internal address space.
1057 	 */
1058 	adev->mc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1059 
1060 	adev->mc.stolen_size = 256 * 1024;
1061 
1062 	/* set DMA mask + need_dma32 flags.
1063 	 * PCIE - can handle 40-bits.
1064 	 * IGP - can handle 40-bits
1065 	 * PCI - dma32 for legacy pci gart, 40 bits on newer asics
1066 	 */
1067 	adev->need_dma32 = false;
1068 	dma_bits = adev->need_dma32 ? 32 : 40;
1069 	r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
1070 	if (r) {
1071 		adev->need_dma32 = true;
1072 		dma_bits = 32;
1073 		pr_warn("amdgpu: No suitable DMA available\n");
1074 	}
1075 	r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
1076 	if (r) {
1077 		pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
1078 		pr_warn("amdgpu: No coherent DMA available\n");
1079 	}
1080 
1081 	r = gmc_v8_0_init_microcode(adev);
1082 	if (r) {
1083 		DRM_ERROR("Failed to load mc firmware!\n");
1084 		return r;
1085 	}
1086 
1087 	r = gmc_v8_0_mc_init(adev);
1088 	if (r)
1089 		return r;
1090 
1091 	/* Memory manager */
1092 	r = amdgpu_bo_init(adev);
1093 	if (r)
1094 		return r;
1095 
1096 	r = gmc_v8_0_gart_init(adev);
1097 	if (r)
1098 		return r;
1099 
1100 	/*
1101 	 * number of VMs
1102 	 * VMID 0 is reserved for System
1103 	 * amdgpu graphics/compute will use VMIDs 1-7
1104 	 * amdkfd will use VMIDs 8-15
1105 	 */
1106 	adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
1107 	adev->vm_manager.num_level = 1;
1108 	amdgpu_vm_manager_init(adev);
1109 
1110 	/* base offset of vram pages */
1111 	if (adev->flags & AMD_IS_APU) {
1112 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1113 
1114 		tmp <<= 22;
1115 		adev->vm_manager.vram_base_offset = tmp;
1116 	} else {
1117 		adev->vm_manager.vram_base_offset = 0;
1118 	}
1119 
1120 	return 0;
1121 }
1122 
1123 static int gmc_v8_0_sw_fini(void *handle)
1124 {
1125 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1126 
1127 	amdgpu_vm_manager_fini(adev);
1128 	gmc_v8_0_gart_fini(adev);
1129 	amdgpu_gem_force_release(adev);
1130 	amdgpu_bo_fini(adev);
1131 
1132 	return 0;
1133 }
1134 
1135 static int gmc_v8_0_hw_init(void *handle)
1136 {
1137 	int r;
1138 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1139 
1140 	gmc_v8_0_init_golden_registers(adev);
1141 
1142 	gmc_v8_0_mc_program(adev);
1143 
1144 	if (adev->asic_type == CHIP_TONGA) {
1145 		r = gmc_v8_0_tonga_mc_load_microcode(adev);
1146 		if (r) {
1147 			DRM_ERROR("Failed to load MC firmware!\n");
1148 			return r;
1149 		}
1150 	} else if (adev->asic_type == CHIP_POLARIS11 ||
1151 			adev->asic_type == CHIP_POLARIS10 ||
1152 			adev->asic_type == CHIP_POLARIS12) {
1153 		r = gmc_v8_0_polaris_mc_load_microcode(adev);
1154 		if (r) {
1155 			DRM_ERROR("Failed to load MC firmware!\n");
1156 			return r;
1157 		}
1158 	}
1159 
1160 	r = gmc_v8_0_gart_enable(adev);
1161 	if (r)
1162 		return r;
1163 
1164 	return r;
1165 }
1166 
1167 static int gmc_v8_0_hw_fini(void *handle)
1168 {
1169 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1170 
1171 	amdgpu_irq_put(adev, &adev->mc.vm_fault, 0);
1172 	gmc_v8_0_gart_disable(adev);
1173 
1174 	return 0;
1175 }
1176 
1177 static int gmc_v8_0_suspend(void *handle)
1178 {
1179 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1180 
1181 	gmc_v8_0_hw_fini(adev);
1182 
1183 	return 0;
1184 }
1185 
1186 static int gmc_v8_0_resume(void *handle)
1187 {
1188 	int r;
1189 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1190 
1191 	r = gmc_v8_0_hw_init(adev);
1192 	if (r)
1193 		return r;
1194 
1195 	amdgpu_vm_reset_all_ids(adev);
1196 
1197 	return 0;
1198 }
1199 
1200 static bool gmc_v8_0_is_idle(void *handle)
1201 {
1202 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1203 	u32 tmp = RREG32(mmSRBM_STATUS);
1204 
1205 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1206 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1207 		return false;
1208 
1209 	return true;
1210 }
1211 
1212 static int gmc_v8_0_wait_for_idle(void *handle)
1213 {
1214 	unsigned i;
1215 	u32 tmp;
1216 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1217 
1218 	for (i = 0; i < adev->usec_timeout; i++) {
1219 		/* read MC_STATUS */
1220 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1221 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1222 					       SRBM_STATUS__MCC_BUSY_MASK |
1223 					       SRBM_STATUS__MCD_BUSY_MASK |
1224 					       SRBM_STATUS__VMC_BUSY_MASK |
1225 					       SRBM_STATUS__VMC1_BUSY_MASK);
1226 		if (!tmp)
1227 			return 0;
1228 		udelay(1);
1229 	}
1230 	return -ETIMEDOUT;
1231 
1232 }
1233 
1234 static bool gmc_v8_0_check_soft_reset(void *handle)
1235 {
1236 	u32 srbm_soft_reset = 0;
1237 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1238 	u32 tmp = RREG32(mmSRBM_STATUS);
1239 
1240 	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1241 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1242 						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1243 
1244 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1245 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1246 		if (!(adev->flags & AMD_IS_APU))
1247 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1248 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1249 	}
1250 	if (srbm_soft_reset) {
1251 		adev->mc.srbm_soft_reset = srbm_soft_reset;
1252 		return true;
1253 	} else {
1254 		adev->mc.srbm_soft_reset = 0;
1255 		return false;
1256 	}
1257 }
1258 
1259 static int gmc_v8_0_pre_soft_reset(void *handle)
1260 {
1261 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1262 
1263 	if (!adev->mc.srbm_soft_reset)
1264 		return 0;
1265 
1266 	gmc_v8_0_mc_stop(adev);
1267 	if (gmc_v8_0_wait_for_idle(adev)) {
1268 		dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1269 	}
1270 
1271 	return 0;
1272 }
1273 
1274 static int gmc_v8_0_soft_reset(void *handle)
1275 {
1276 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1277 	u32 srbm_soft_reset;
1278 
1279 	if (!adev->mc.srbm_soft_reset)
1280 		return 0;
1281 	srbm_soft_reset = adev->mc.srbm_soft_reset;
1282 
1283 	if (srbm_soft_reset) {
1284 		u32 tmp;
1285 
1286 		tmp = RREG32(mmSRBM_SOFT_RESET);
1287 		tmp |= srbm_soft_reset;
1288 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1289 		WREG32(mmSRBM_SOFT_RESET, tmp);
1290 		tmp = RREG32(mmSRBM_SOFT_RESET);
1291 
1292 		udelay(50);
1293 
1294 		tmp &= ~srbm_soft_reset;
1295 		WREG32(mmSRBM_SOFT_RESET, tmp);
1296 		tmp = RREG32(mmSRBM_SOFT_RESET);
1297 
1298 		/* Wait a little for things to settle down */
1299 		udelay(50);
1300 	}
1301 
1302 	return 0;
1303 }
1304 
1305 static int gmc_v8_0_post_soft_reset(void *handle)
1306 {
1307 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1308 
1309 	if (!adev->mc.srbm_soft_reset)
1310 		return 0;
1311 
1312 	gmc_v8_0_mc_resume(adev);
1313 	return 0;
1314 }
1315 
1316 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1317 					     struct amdgpu_irq_src *src,
1318 					     unsigned type,
1319 					     enum amdgpu_interrupt_state state)
1320 {
1321 	u32 tmp;
1322 	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1323 		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1324 		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1325 		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1326 		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1327 		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1328 		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1329 
1330 	switch (state) {
1331 	case AMDGPU_IRQ_STATE_DISABLE:
1332 		/* system context */
1333 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1334 		tmp &= ~bits;
1335 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1336 		/* VMs */
1337 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1338 		tmp &= ~bits;
1339 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1340 		break;
1341 	case AMDGPU_IRQ_STATE_ENABLE:
1342 		/* system context */
1343 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1344 		tmp |= bits;
1345 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1346 		/* VMs */
1347 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1348 		tmp |= bits;
1349 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1350 		break;
1351 	default:
1352 		break;
1353 	}
1354 
1355 	return 0;
1356 }
1357 
1358 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1359 				      struct amdgpu_irq_src *source,
1360 				      struct amdgpu_iv_entry *entry)
1361 {
1362 	u32 addr, status, mc_client;
1363 
1364 	if (amdgpu_sriov_vf(adev)) {
1365 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1366 			entry->src_id, entry->src_data[0]);
1367 		dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1368 		return 0;
1369 	}
1370 
1371 	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1372 	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1373 	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1374 	/* reset addr and status */
1375 	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1376 
1377 	if (!addr && !status)
1378 		return 0;
1379 
1380 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1381 		gmc_v8_0_set_fault_enable_default(adev, false);
1382 
1383 	if (printk_ratelimit()) {
1384 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1385 			entry->src_id, entry->src_data[0]);
1386 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1387 			addr);
1388 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1389 			status);
1390 		gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client);
1391 	}
1392 
1393 	return 0;
1394 }
1395 
1396 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1397 						     bool enable)
1398 {
1399 	uint32_t data;
1400 
1401 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1402 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1403 		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1404 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1405 
1406 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1407 		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1408 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1409 
1410 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1411 		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1412 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1413 
1414 		data = RREG32(mmMC_XPB_CLK_GAT);
1415 		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1416 		WREG32(mmMC_XPB_CLK_GAT, data);
1417 
1418 		data = RREG32(mmATC_MISC_CG);
1419 		data |= ATC_MISC_CG__ENABLE_MASK;
1420 		WREG32(mmATC_MISC_CG, data);
1421 
1422 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1423 		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1424 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1425 
1426 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1427 		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1428 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1429 
1430 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1431 		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1432 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1433 
1434 		data = RREG32(mmVM_L2_CG);
1435 		data |= VM_L2_CG__ENABLE_MASK;
1436 		WREG32(mmVM_L2_CG, data);
1437 	} else {
1438 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1439 		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1440 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1441 
1442 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1443 		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1444 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1445 
1446 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1447 		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1448 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1449 
1450 		data = RREG32(mmMC_XPB_CLK_GAT);
1451 		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1452 		WREG32(mmMC_XPB_CLK_GAT, data);
1453 
1454 		data = RREG32(mmATC_MISC_CG);
1455 		data &= ~ATC_MISC_CG__ENABLE_MASK;
1456 		WREG32(mmATC_MISC_CG, data);
1457 
1458 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1459 		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1460 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1461 
1462 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1463 		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1464 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1465 
1466 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1467 		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1468 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1469 
1470 		data = RREG32(mmVM_L2_CG);
1471 		data &= ~VM_L2_CG__ENABLE_MASK;
1472 		WREG32(mmVM_L2_CG, data);
1473 	}
1474 }
1475 
1476 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1477 				       bool enable)
1478 {
1479 	uint32_t data;
1480 
1481 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1482 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1483 		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1484 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1485 
1486 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1487 		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1488 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1489 
1490 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1491 		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1492 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1493 
1494 		data = RREG32(mmMC_XPB_CLK_GAT);
1495 		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1496 		WREG32(mmMC_XPB_CLK_GAT, data);
1497 
1498 		data = RREG32(mmATC_MISC_CG);
1499 		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1500 		WREG32(mmATC_MISC_CG, data);
1501 
1502 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1503 		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1504 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1505 
1506 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1507 		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1508 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1509 
1510 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1511 		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1512 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1513 
1514 		data = RREG32(mmVM_L2_CG);
1515 		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1516 		WREG32(mmVM_L2_CG, data);
1517 	} else {
1518 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1519 		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1520 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1521 
1522 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1523 		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1524 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1525 
1526 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1527 		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1528 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1529 
1530 		data = RREG32(mmMC_XPB_CLK_GAT);
1531 		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1532 		WREG32(mmMC_XPB_CLK_GAT, data);
1533 
1534 		data = RREG32(mmATC_MISC_CG);
1535 		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1536 		WREG32(mmATC_MISC_CG, data);
1537 
1538 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1539 		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1540 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1541 
1542 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1543 		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1544 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1545 
1546 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1547 		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1548 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1549 
1550 		data = RREG32(mmVM_L2_CG);
1551 		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1552 		WREG32(mmVM_L2_CG, data);
1553 	}
1554 }
1555 
1556 static int gmc_v8_0_set_clockgating_state(void *handle,
1557 					  enum amd_clockgating_state state)
1558 {
1559 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1560 
1561 	if (amdgpu_sriov_vf(adev))
1562 		return 0;
1563 
1564 	switch (adev->asic_type) {
1565 	case CHIP_FIJI:
1566 		fiji_update_mc_medium_grain_clock_gating(adev,
1567 				state == AMD_CG_STATE_GATE);
1568 		fiji_update_mc_light_sleep(adev,
1569 				state == AMD_CG_STATE_GATE);
1570 		break;
1571 	default:
1572 		break;
1573 	}
1574 	return 0;
1575 }
1576 
1577 static int gmc_v8_0_set_powergating_state(void *handle,
1578 					  enum amd_powergating_state state)
1579 {
1580 	return 0;
1581 }
1582 
1583 static void gmc_v8_0_get_clockgating_state(void *handle, u32 *flags)
1584 {
1585 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1586 	int data;
1587 
1588 	if (amdgpu_sriov_vf(adev))
1589 		*flags = 0;
1590 
1591 	/* AMD_CG_SUPPORT_MC_MGCG */
1592 	data = RREG32(mmMC_HUB_MISC_HUB_CG);
1593 	if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1594 		*flags |= AMD_CG_SUPPORT_MC_MGCG;
1595 
1596 	/* AMD_CG_SUPPORT_MC_LS */
1597 	if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1598 		*flags |= AMD_CG_SUPPORT_MC_LS;
1599 }
1600 
1601 static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1602 	.name = "gmc_v8_0",
1603 	.early_init = gmc_v8_0_early_init,
1604 	.late_init = gmc_v8_0_late_init,
1605 	.sw_init = gmc_v8_0_sw_init,
1606 	.sw_fini = gmc_v8_0_sw_fini,
1607 	.hw_init = gmc_v8_0_hw_init,
1608 	.hw_fini = gmc_v8_0_hw_fini,
1609 	.suspend = gmc_v8_0_suspend,
1610 	.resume = gmc_v8_0_resume,
1611 	.is_idle = gmc_v8_0_is_idle,
1612 	.wait_for_idle = gmc_v8_0_wait_for_idle,
1613 	.check_soft_reset = gmc_v8_0_check_soft_reset,
1614 	.pre_soft_reset = gmc_v8_0_pre_soft_reset,
1615 	.soft_reset = gmc_v8_0_soft_reset,
1616 	.post_soft_reset = gmc_v8_0_post_soft_reset,
1617 	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1618 	.set_powergating_state = gmc_v8_0_set_powergating_state,
1619 	.get_clockgating_state = gmc_v8_0_get_clockgating_state,
1620 };
1621 
1622 static const struct amdgpu_gart_funcs gmc_v8_0_gart_funcs = {
1623 	.flush_gpu_tlb = gmc_v8_0_gart_flush_gpu_tlb,
1624 	.set_pte_pde = gmc_v8_0_gart_set_pte_pde,
1625 	.set_prt = gmc_v8_0_set_prt,
1626 	.get_vm_pte_flags = gmc_v8_0_get_vm_pte_flags,
1627 	.get_vm_pde = gmc_v8_0_get_vm_pde
1628 };
1629 
1630 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1631 	.set = gmc_v8_0_vm_fault_interrupt_state,
1632 	.process = gmc_v8_0_process_interrupt,
1633 };
1634 
1635 static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev)
1636 {
1637 	if (adev->gart.gart_funcs == NULL)
1638 		adev->gart.gart_funcs = &gmc_v8_0_gart_funcs;
1639 }
1640 
1641 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1642 {
1643 	adev->mc.vm_fault.num_types = 1;
1644 	adev->mc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1645 }
1646 
1647 const struct amdgpu_ip_block_version gmc_v8_0_ip_block =
1648 {
1649 	.type = AMD_IP_BLOCK_TYPE_GMC,
1650 	.major = 8,
1651 	.minor = 0,
1652 	.rev = 0,
1653 	.funcs = &gmc_v8_0_ip_funcs,
1654 };
1655 
1656 const struct amdgpu_ip_block_version gmc_v8_1_ip_block =
1657 {
1658 	.type = AMD_IP_BLOCK_TYPE_GMC,
1659 	.major = 8,
1660 	.minor = 1,
1661 	.rev = 0,
1662 	.funcs = &gmc_v8_0_ip_funcs,
1663 };
1664 
1665 const struct amdgpu_ip_block_version gmc_v8_5_ip_block =
1666 {
1667 	.type = AMD_IP_BLOCK_TYPE_GMC,
1668 	.major = 8,
1669 	.minor = 5,
1670 	.rev = 0,
1671 	.funcs = &gmc_v8_0_ip_funcs,
1672 };
1673