xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c (revision 11c416e3)
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 
24 #include <linux/firmware.h>
25 #include <linux/module.h>
26 #include <linux/pci.h>
27 
28 #include <drm/drm_cache.h>
29 #include "amdgpu.h"
30 #include "gmc_v8_0.h"
31 #include "amdgpu_ucode.h"
32 #include "amdgpu_amdkfd.h"
33 #include "amdgpu_gem.h"
34 
35 #include "gmc/gmc_8_1_d.h"
36 #include "gmc/gmc_8_1_sh_mask.h"
37 
38 #include "bif/bif_5_0_d.h"
39 #include "bif/bif_5_0_sh_mask.h"
40 
41 #include "oss/oss_3_0_d.h"
42 #include "oss/oss_3_0_sh_mask.h"
43 
44 #include "dce/dce_10_0_d.h"
45 #include "dce/dce_10_0_sh_mask.h"
46 
47 #include "vid.h"
48 #include "vi.h"
49 
50 #include "amdgpu_atombios.h"
51 
52 #include "ivsrcid/ivsrcid_vislands30.h"
53 
54 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev);
55 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
56 static int gmc_v8_0_wait_for_idle(void *handle);
57 
58 MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
59 MODULE_FIRMWARE("amdgpu/polaris11_mc.bin");
60 MODULE_FIRMWARE("amdgpu/polaris10_mc.bin");
61 MODULE_FIRMWARE("amdgpu/polaris12_mc.bin");
62 MODULE_FIRMWARE("amdgpu/polaris11_k_mc.bin");
63 MODULE_FIRMWARE("amdgpu/polaris10_k_mc.bin");
64 MODULE_FIRMWARE("amdgpu/polaris12_k_mc.bin");
65 
66 static const u32 golden_settings_tonga_a11[] =
67 {
68 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
69 	mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
70 	mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
71 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
72 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
73 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
74 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
75 };
76 
77 static const u32 tonga_mgcg_cgcg_init[] =
78 {
79 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
80 };
81 
82 static const u32 golden_settings_fiji_a10[] =
83 {
84 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
85 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
86 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
87 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
88 };
89 
90 static const u32 fiji_mgcg_cgcg_init[] =
91 {
92 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
93 };
94 
95 static const u32 golden_settings_polaris11_a11[] =
96 {
97 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
98 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
99 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
100 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
101 };
102 
103 static const u32 golden_settings_polaris10_a11[] =
104 {
105 	mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
106 	mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
107 	mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
108 	mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
109 	mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
110 };
111 
112 static const u32 cz_mgcg_cgcg_init[] =
113 {
114 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
115 };
116 
117 static const u32 stoney_mgcg_cgcg_init[] =
118 {
119 	mmATC_MISC_CG, 0xffffffff, 0x000c0200,
120 	mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
121 };
122 
123 static const u32 golden_settings_stoney_common[] =
124 {
125 	mmMC_HUB_RDREQ_UVD, MC_HUB_RDREQ_UVD__PRESCALE_MASK, 0x00000004,
126 	mmMC_RD_GRP_OTH, MC_RD_GRP_OTH__UVD_MASK, 0x00600000
127 };
128 
129 static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
130 {
131 	switch (adev->asic_type) {
132 	case CHIP_FIJI:
133 		amdgpu_device_program_register_sequence(adev,
134 							fiji_mgcg_cgcg_init,
135 							ARRAY_SIZE(fiji_mgcg_cgcg_init));
136 		amdgpu_device_program_register_sequence(adev,
137 							golden_settings_fiji_a10,
138 							ARRAY_SIZE(golden_settings_fiji_a10));
139 		break;
140 	case CHIP_TONGA:
141 		amdgpu_device_program_register_sequence(adev,
142 							tonga_mgcg_cgcg_init,
143 							ARRAY_SIZE(tonga_mgcg_cgcg_init));
144 		amdgpu_device_program_register_sequence(adev,
145 							golden_settings_tonga_a11,
146 							ARRAY_SIZE(golden_settings_tonga_a11));
147 		break;
148 	case CHIP_POLARIS11:
149 	case CHIP_POLARIS12:
150 	case CHIP_VEGAM:
151 		amdgpu_device_program_register_sequence(adev,
152 							golden_settings_polaris11_a11,
153 							ARRAY_SIZE(golden_settings_polaris11_a11));
154 		break;
155 	case CHIP_POLARIS10:
156 		amdgpu_device_program_register_sequence(adev,
157 							golden_settings_polaris10_a11,
158 							ARRAY_SIZE(golden_settings_polaris10_a11));
159 		break;
160 	case CHIP_CARRIZO:
161 		amdgpu_device_program_register_sequence(adev,
162 							cz_mgcg_cgcg_init,
163 							ARRAY_SIZE(cz_mgcg_cgcg_init));
164 		break;
165 	case CHIP_STONEY:
166 		amdgpu_device_program_register_sequence(adev,
167 							stoney_mgcg_cgcg_init,
168 							ARRAY_SIZE(stoney_mgcg_cgcg_init));
169 		amdgpu_device_program_register_sequence(adev,
170 							golden_settings_stoney_common,
171 							ARRAY_SIZE(golden_settings_stoney_common));
172 		break;
173 	default:
174 		break;
175 	}
176 }
177 
178 static void gmc_v8_0_mc_stop(struct amdgpu_device *adev)
179 {
180 	u32 blackout;
181 
182 	gmc_v8_0_wait_for_idle(adev);
183 
184 	blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
185 	if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
186 		/* Block CPU access */
187 		WREG32(mmBIF_FB_EN, 0);
188 		/* blackout the MC */
189 		blackout = REG_SET_FIELD(blackout,
190 					 MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
191 		WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
192 	}
193 	/* wait for the MC to settle */
194 	udelay(100);
195 }
196 
197 static void gmc_v8_0_mc_resume(struct amdgpu_device *adev)
198 {
199 	u32 tmp;
200 
201 	/* unblackout the MC */
202 	tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
203 	tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
204 	WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
205 	/* allow CPU access */
206 	tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
207 	tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
208 	WREG32(mmBIF_FB_EN, tmp);
209 }
210 
211 /**
212  * gmc_v8_0_init_microcode - load ucode images from disk
213  *
214  * @adev: amdgpu_device pointer
215  *
216  * Use the firmware interface to load the ucode images into
217  * the driver (not loaded into hw).
218  * Returns 0 on success, error on failure.
219  */
220 static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
221 {
222 	const char *chip_name;
223 	char fw_name[30];
224 	int err;
225 
226 	DRM_DEBUG("\n");
227 
228 	switch (adev->asic_type) {
229 	case CHIP_TONGA:
230 		chip_name = "tonga";
231 		break;
232 	case CHIP_POLARIS11:
233 		if (((adev->pdev->device == 0x67ef) &&
234 		     ((adev->pdev->revision == 0xe0) ||
235 		      (adev->pdev->revision == 0xe5))) ||
236 		    ((adev->pdev->device == 0x67ff) &&
237 		     ((adev->pdev->revision == 0xcf) ||
238 		      (adev->pdev->revision == 0xef) ||
239 		      (adev->pdev->revision == 0xff))))
240 			chip_name = "polaris11_k";
241 		else if ((adev->pdev->device == 0x67ef) &&
242 			 (adev->pdev->revision == 0xe2))
243 			chip_name = "polaris11_k";
244 		else
245 			chip_name = "polaris11";
246 		break;
247 	case CHIP_POLARIS10:
248 		if ((adev->pdev->device == 0x67df) &&
249 		    ((adev->pdev->revision == 0xe1) ||
250 		     (adev->pdev->revision == 0xf7)))
251 			chip_name = "polaris10_k";
252 		else
253 			chip_name = "polaris10";
254 		break;
255 	case CHIP_POLARIS12:
256 		if (((adev->pdev->device == 0x6987) &&
257 		     ((adev->pdev->revision == 0xc0) ||
258 		      (adev->pdev->revision == 0xc3))) ||
259 		    ((adev->pdev->device == 0x6981) &&
260 		     ((adev->pdev->revision == 0x00) ||
261 		      (adev->pdev->revision == 0x01) ||
262 		      (adev->pdev->revision == 0x10))))
263 			chip_name = "polaris12_k";
264 		else
265 			chip_name = "polaris12";
266 		break;
267 	case CHIP_FIJI:
268 	case CHIP_CARRIZO:
269 	case CHIP_STONEY:
270 	case CHIP_VEGAM:
271 		return 0;
272 	default: BUG();
273 	}
274 
275 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
276 	err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
277 	if (err)
278 		goto out;
279 	err = amdgpu_ucode_validate(adev->gmc.fw);
280 
281 out:
282 	if (err) {
283 		pr_err("mc: Failed to load firmware \"%s\"\n", fw_name);
284 		release_firmware(adev->gmc.fw);
285 		adev->gmc.fw = NULL;
286 	}
287 	return err;
288 }
289 
290 /**
291  * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
292  *
293  * @adev: amdgpu_device pointer
294  *
295  * Load the GDDR MC ucode into the hw (VI).
296  * Returns 0 on success, error on failure.
297  */
298 static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
299 {
300 	const struct mc_firmware_header_v1_0 *hdr;
301 	const __le32 *fw_data = NULL;
302 	const __le32 *io_mc_regs = NULL;
303 	u32 running;
304 	int i, ucode_size, regs_size;
305 
306 	/* Skip MC ucode loading on SR-IOV capable boards.
307 	 * vbios does this for us in asic_init in that case.
308 	 * Skip MC ucode loading on VF, because hypervisor will do that
309 	 * for this adaptor.
310 	 */
311 	if (amdgpu_sriov_bios(adev))
312 		return 0;
313 
314 	if (!adev->gmc.fw)
315 		return -EINVAL;
316 
317 	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
318 	amdgpu_ucode_print_mc_hdr(&hdr->header);
319 
320 	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
321 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
322 	io_mc_regs = (const __le32 *)
323 		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
324 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
325 	fw_data = (const __le32 *)
326 		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
327 
328 	running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
329 
330 	if (running == 0) {
331 		/* reset the engine and set to writable */
332 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
333 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
334 
335 		/* load mc io regs */
336 		for (i = 0; i < regs_size; i++) {
337 			WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
338 			WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
339 		}
340 		/* load the MC ucode */
341 		for (i = 0; i < ucode_size; i++)
342 			WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
343 
344 		/* put the engine back into the active state */
345 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
346 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
347 		WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
348 
349 		/* wait for training to complete */
350 		for (i = 0; i < adev->usec_timeout; i++) {
351 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
352 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
353 				break;
354 			udelay(1);
355 		}
356 		for (i = 0; i < adev->usec_timeout; i++) {
357 			if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
358 					  MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
359 				break;
360 			udelay(1);
361 		}
362 	}
363 
364 	return 0;
365 }
366 
367 static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
368 {
369 	const struct mc_firmware_header_v1_0 *hdr;
370 	const __le32 *fw_data = NULL;
371 	const __le32 *io_mc_regs = NULL;
372 	u32 data;
373 	int i, ucode_size, regs_size;
374 
375 	/* Skip MC ucode loading on SR-IOV capable boards.
376 	 * vbios does this for us in asic_init in that case.
377 	 * Skip MC ucode loading on VF, because hypervisor will do that
378 	 * for this adaptor.
379 	 */
380 	if (amdgpu_sriov_bios(adev))
381 		return 0;
382 
383 	if (!adev->gmc.fw)
384 		return -EINVAL;
385 
386 	hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
387 	amdgpu_ucode_print_mc_hdr(&hdr->header);
388 
389 	adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
390 	regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
391 	io_mc_regs = (const __le32 *)
392 		(adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
393 	ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
394 	fw_data = (const __le32 *)
395 		(adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
396 
397 	data = RREG32(mmMC_SEQ_MISC0);
398 	data &= ~(0x40);
399 	WREG32(mmMC_SEQ_MISC0, data);
400 
401 	/* load mc io regs */
402 	for (i = 0; i < regs_size; i++) {
403 		WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
404 		WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
405 	}
406 
407 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
408 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
409 
410 	/* load the MC ucode */
411 	for (i = 0; i < ucode_size; i++)
412 		WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
413 
414 	/* put the engine back into the active state */
415 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
416 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
417 	WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
418 
419 	/* wait for training to complete */
420 	for (i = 0; i < adev->usec_timeout; i++) {
421 		data = RREG32(mmMC_SEQ_MISC0);
422 		if (data & 0x80)
423 			break;
424 		udelay(1);
425 	}
426 
427 	return 0;
428 }
429 
430 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
431 				       struct amdgpu_gmc *mc)
432 {
433 	u64 base = 0;
434 
435 	if (!amdgpu_sriov_vf(adev))
436 		base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
437 	base <<= 24;
438 
439 	amdgpu_gmc_vram_location(adev, mc, base);
440 	amdgpu_gmc_gart_location(adev, mc);
441 }
442 
443 /**
444  * gmc_v8_0_mc_program - program the GPU memory controller
445  *
446  * @adev: amdgpu_device pointer
447  *
448  * Set the location of vram, gart, and AGP in the GPU's
449  * physical address space (VI).
450  */
451 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
452 {
453 	u32 tmp;
454 	int i, j;
455 
456 	/* Initialize HDP */
457 	for (i = 0, j = 0; i < 32; i++, j += 0x6) {
458 		WREG32((0xb05 + j), 0x00000000);
459 		WREG32((0xb06 + j), 0x00000000);
460 		WREG32((0xb07 + j), 0x00000000);
461 		WREG32((0xb08 + j), 0x00000000);
462 		WREG32((0xb09 + j), 0x00000000);
463 	}
464 	WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
465 
466 	if (gmc_v8_0_wait_for_idle((void *)adev)) {
467 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
468 	}
469 	if (adev->mode_info.num_crtc) {
470 		/* Lockout access through VGA aperture*/
471 		tmp = RREG32(mmVGA_HDP_CONTROL);
472 		tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
473 		WREG32(mmVGA_HDP_CONTROL, tmp);
474 
475 		/* disable VGA render */
476 		tmp = RREG32(mmVGA_RENDER_CONTROL);
477 		tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
478 		WREG32(mmVGA_RENDER_CONTROL, tmp);
479 	}
480 	/* Update configuration */
481 	WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
482 	       adev->gmc.vram_start >> 12);
483 	WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
484 	       adev->gmc.vram_end >> 12);
485 	WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
486 	       adev->vram_scratch.gpu_addr >> 12);
487 
488 	if (amdgpu_sriov_vf(adev)) {
489 		tmp = ((adev->gmc.vram_end >> 24) & 0xFFFF) << 16;
490 		tmp |= ((adev->gmc.vram_start >> 24) & 0xFFFF);
491 		WREG32(mmMC_VM_FB_LOCATION, tmp);
492 		/* XXX double check these! */
493 		WREG32(mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
494 		WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
495 		WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
496 	}
497 
498 	WREG32(mmMC_VM_AGP_BASE, 0);
499 	WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
500 	WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
501 	if (gmc_v8_0_wait_for_idle((void *)adev)) {
502 		dev_warn(adev->dev, "Wait for MC idle timedout !\n");
503 	}
504 
505 	WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
506 
507 	tmp = RREG32(mmHDP_MISC_CNTL);
508 	tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
509 	WREG32(mmHDP_MISC_CNTL, tmp);
510 
511 	tmp = RREG32(mmHDP_HOST_PATH_CNTL);
512 	WREG32(mmHDP_HOST_PATH_CNTL, tmp);
513 }
514 
515 /**
516  * gmc_v8_0_mc_init - initialize the memory controller driver params
517  *
518  * @adev: amdgpu_device pointer
519  *
520  * Look up the amount of vram, vram width, and decide how to place
521  * vram and gart within the GPU's physical address space (VI).
522  * Returns 0 for success.
523  */
524 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
525 {
526 	int r;
527 
528 	adev->gmc.vram_width = amdgpu_atombios_get_vram_width(adev);
529 	if (!adev->gmc.vram_width) {
530 		u32 tmp;
531 		int chansize, numchan;
532 
533 		/* Get VRAM informations */
534 		tmp = RREG32(mmMC_ARB_RAMCFG);
535 		if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
536 			chansize = 64;
537 		} else {
538 			chansize = 32;
539 		}
540 		tmp = RREG32(mmMC_SHARED_CHMAP);
541 		switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
542 		case 0:
543 		default:
544 			numchan = 1;
545 			break;
546 		case 1:
547 			numchan = 2;
548 			break;
549 		case 2:
550 			numchan = 4;
551 			break;
552 		case 3:
553 			numchan = 8;
554 			break;
555 		case 4:
556 			numchan = 3;
557 			break;
558 		case 5:
559 			numchan = 6;
560 			break;
561 		case 6:
562 			numchan = 10;
563 			break;
564 		case 7:
565 			numchan = 12;
566 			break;
567 		case 8:
568 			numchan = 16;
569 			break;
570 		}
571 		adev->gmc.vram_width = numchan * chansize;
572 	}
573 	/* size in MB on si */
574 	adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
575 	adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
576 
577 	if (!(adev->flags & AMD_IS_APU)) {
578 		r = amdgpu_device_resize_fb_bar(adev);
579 		if (r)
580 			return r;
581 	}
582 	adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
583 	adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
584 
585 #ifdef CONFIG_X86_64
586 	if (adev->flags & AMD_IS_APU) {
587 		adev->gmc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
588 		adev->gmc.aper_size = adev->gmc.real_vram_size;
589 	}
590 #endif
591 
592 	/* In case the PCI BAR is larger than the actual amount of vram */
593 	adev->gmc.visible_vram_size = adev->gmc.aper_size;
594 	if (adev->gmc.visible_vram_size > adev->gmc.real_vram_size)
595 		adev->gmc.visible_vram_size = adev->gmc.real_vram_size;
596 
597 	/* set the gart size */
598 	if (amdgpu_gart_size == -1) {
599 		switch (adev->asic_type) {
600 		case CHIP_POLARIS10: /* all engines support GPUVM */
601 		case CHIP_POLARIS11: /* all engines support GPUVM */
602 		case CHIP_POLARIS12: /* all engines support GPUVM */
603 		case CHIP_VEGAM:     /* all engines support GPUVM */
604 		default:
605 			adev->gmc.gart_size = 256ULL << 20;
606 			break;
607 		case CHIP_TONGA:   /* UVD, VCE do not support GPUVM */
608 		case CHIP_FIJI:    /* UVD, VCE do not support GPUVM */
609 		case CHIP_CARRIZO: /* UVD, VCE do not support GPUVM, DCE SG support */
610 		case CHIP_STONEY:  /* UVD does not support GPUVM, DCE SG support */
611 			adev->gmc.gart_size = 1024ULL << 20;
612 			break;
613 		}
614 	} else {
615 		adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
616 	}
617 
618 	gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
619 
620 	return 0;
621 }
622 
623 /**
624  * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
625  *
626  * @adev: amdgpu_device pointer
627  * @pasid: pasid to be flush
628  *
629  * Flush the TLB for the requested pasid.
630  */
631 static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
632 					uint16_t pasid, uint32_t flush_type,
633 					bool all_hub)
634 {
635 	int vmid;
636 	unsigned int tmp;
637 
638 	if (adev->in_gpu_reset)
639 		return -EIO;
640 
641 	for (vmid = 1; vmid < 16; vmid++) {
642 
643 		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
644 		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
645 			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
646 			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
647 			RREG32(mmVM_INVALIDATE_RESPONSE);
648 			break;
649 		}
650 	}
651 
652 	return 0;
653 
654 }
655 
656 /*
657  * GART
658  * VMID 0 is the physical GPU addresses as used by the kernel.
659  * VMIDs 1-15 are used for userspace clients and are handled
660  * by the amdgpu vm/hsa code.
661  */
662 
663 /**
664  * gmc_v8_0_flush_gpu_tlb - gart tlb flush callback
665  *
666  * @adev: amdgpu_device pointer
667  * @vmid: vm instance to flush
668  *
669  * Flush the TLB for the requested page table (VI).
670  */
671 static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
672 					uint32_t vmhub, uint32_t flush_type)
673 {
674 	/* bits 0-15 are the VM contexts0-15 */
675 	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
676 }
677 
678 static uint64_t gmc_v8_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
679 					    unsigned vmid, uint64_t pd_addr)
680 {
681 	uint32_t reg;
682 
683 	if (vmid < 8)
684 		reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
685 	else
686 		reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
687 	amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
688 
689 	/* bits 0-15 are the VM contexts0-15 */
690 	amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
691 
692 	return pd_addr;
693 }
694 
695 static void gmc_v8_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned vmid,
696 					unsigned pasid)
697 {
698 	amdgpu_ring_emit_wreg(ring, mmIH_VMID_0_LUT + vmid, pasid);
699 }
700 
701 /*
702  * PTE format on VI:
703  * 63:40 reserved
704  * 39:12 4k physical page base address
705  * 11:7 fragment
706  * 6 write
707  * 5 read
708  * 4 exe
709  * 3 reserved
710  * 2 snooped
711  * 1 system
712  * 0 valid
713  *
714  * PDE format on VI:
715  * 63:59 block fragment size
716  * 58:40 reserved
717  * 39:1 physical base address of PTE
718  * bits 5:1 must be 0.
719  * 0 valid
720  */
721 
722 static void gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, int level,
723 				uint64_t *addr, uint64_t *flags)
724 {
725 	BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
726 }
727 
728 static void gmc_v8_0_get_vm_pte(struct amdgpu_device *adev,
729 				struct amdgpu_bo_va_mapping *mapping,
730 				uint64_t *flags)
731 {
732 	*flags &= ~AMDGPU_PTE_EXECUTABLE;
733 	*flags |= mapping->flags & AMDGPU_PTE_EXECUTABLE;
734 	*flags &= ~AMDGPU_PTE_PRT;
735 }
736 
737 /**
738  * gmc_v8_0_set_fault_enable_default - update VM fault handling
739  *
740  * @adev: amdgpu_device pointer
741  * @value: true redirects VM faults to the default page
742  */
743 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
744 					      bool value)
745 {
746 	u32 tmp;
747 
748 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
749 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
750 			    RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
751 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
752 			    DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
753 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
754 			    PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
755 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
756 			    VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
757 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
758 			    READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
759 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
760 			    WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
761 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
762 			    EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
763 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
764 }
765 
766 /**
767  * gmc_v8_0_set_prt - set PRT VM fault
768  *
769  * @adev: amdgpu_device pointer
770  * @enable: enable/disable VM fault handling for PRT
771 */
772 static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
773 {
774 	u32 tmp;
775 
776 	if (enable && !adev->gmc.prt_warning) {
777 		dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
778 		adev->gmc.prt_warning = true;
779 	}
780 
781 	tmp = RREG32(mmVM_PRT_CNTL);
782 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
783 			    CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
784 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
785 			    CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
786 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
787 			    TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
788 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
789 			    TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
790 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
791 			    L2_CACHE_STORE_INVALID_ENTRIES, enable);
792 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
793 			    L1_TLB_STORE_INVALID_ENTRIES, enable);
794 	tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
795 			    MASK_PDE0_FAULT, enable);
796 	WREG32(mmVM_PRT_CNTL, tmp);
797 
798 	if (enable) {
799 		uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
800 		uint32_t high = adev->vm_manager.max_pfn -
801 			(AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
802 
803 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
804 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
805 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
806 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
807 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
808 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
809 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
810 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
811 	} else {
812 		WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
813 		WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
814 		WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
815 		WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
816 		WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
817 		WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
818 		WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
819 		WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
820 	}
821 }
822 
823 /**
824  * gmc_v8_0_gart_enable - gart enable
825  *
826  * @adev: amdgpu_device pointer
827  *
828  * This sets up the TLBs, programs the page tables for VMID0,
829  * sets up the hw for VMIDs 1-15 which are allocated on
830  * demand, and sets up the global locations for the LDS, GDS,
831  * and GPUVM for FSA64 clients (VI).
832  * Returns 0 for success, errors for failure.
833  */
834 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
835 {
836 	uint64_t table_addr;
837 	int r, i;
838 	u32 tmp, field;
839 
840 	if (adev->gart.bo == NULL) {
841 		dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
842 		return -EINVAL;
843 	}
844 	r = amdgpu_gart_table_vram_pin(adev);
845 	if (r)
846 		return r;
847 
848 	table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
849 
850 	/* Setup TLB control */
851 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
852 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
853 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
854 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
855 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
856 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
857 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
858 	/* Setup L2 cache */
859 	tmp = RREG32(mmVM_L2_CNTL);
860 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
861 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
862 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
863 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
864 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
865 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
866 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
867 	WREG32(mmVM_L2_CNTL, tmp);
868 	tmp = RREG32(mmVM_L2_CNTL2);
869 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
870 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
871 	WREG32(mmVM_L2_CNTL2, tmp);
872 
873 	field = adev->vm_manager.fragment_size;
874 	tmp = RREG32(mmVM_L2_CNTL3);
875 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
876 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
877 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
878 	WREG32(mmVM_L2_CNTL3, tmp);
879 	/* XXX: set to enable PTE/PDE in system memory */
880 	tmp = RREG32(mmVM_L2_CNTL4);
881 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
882 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
883 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
884 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
885 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
886 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
887 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
888 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
889 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
890 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
891 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
892 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
893 	WREG32(mmVM_L2_CNTL4, tmp);
894 	/* setup context0 */
895 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
896 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
897 	WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
898 	WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
899 			(u32)(adev->dummy_page_addr >> 12));
900 	WREG32(mmVM_CONTEXT0_CNTL2, 0);
901 	tmp = RREG32(mmVM_CONTEXT0_CNTL);
902 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
903 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
904 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
905 	WREG32(mmVM_CONTEXT0_CNTL, tmp);
906 
907 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
908 	WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
909 	WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
910 
911 	/* empty context1-15 */
912 	/* FIXME start with 4G, once using 2 level pt switch to full
913 	 * vm size space
914 	 */
915 	/* set vm size, must be a multiple of 4 */
916 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
917 	WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
918 	for (i = 1; i < 16; i++) {
919 		if (i < 8)
920 			WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
921 			       table_addr >> 12);
922 		else
923 			WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
924 			       table_addr >> 12);
925 	}
926 
927 	/* enable context1-15 */
928 	WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
929 	       (u32)(adev->dummy_page_addr >> 12));
930 	WREG32(mmVM_CONTEXT1_CNTL2, 4);
931 	tmp = RREG32(mmVM_CONTEXT1_CNTL);
932 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
933 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
934 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
935 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
936 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
937 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
938 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
939 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
940 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
941 	tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
942 			    adev->vm_manager.block_size - 9);
943 	WREG32(mmVM_CONTEXT1_CNTL, tmp);
944 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
945 		gmc_v8_0_set_fault_enable_default(adev, false);
946 	else
947 		gmc_v8_0_set_fault_enable_default(adev, true);
948 
949 	gmc_v8_0_flush_gpu_tlb(adev, 0, 0, 0);
950 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
951 		 (unsigned)(adev->gmc.gart_size >> 20),
952 		 (unsigned long long)table_addr);
953 	adev->gart.ready = true;
954 	return 0;
955 }
956 
957 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
958 {
959 	int r;
960 
961 	if (adev->gart.bo) {
962 		WARN(1, "R600 PCIE GART already initialized\n");
963 		return 0;
964 	}
965 	/* Initialize common gart structure */
966 	r = amdgpu_gart_init(adev);
967 	if (r)
968 		return r;
969 	adev->gart.table_size = adev->gart.num_gpu_pages * 8;
970 	adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
971 	return amdgpu_gart_table_vram_alloc(adev);
972 }
973 
974 /**
975  * gmc_v8_0_gart_disable - gart disable
976  *
977  * @adev: amdgpu_device pointer
978  *
979  * This disables all VM page table (VI).
980  */
981 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
982 {
983 	u32 tmp;
984 
985 	/* Disable all tables */
986 	WREG32(mmVM_CONTEXT0_CNTL, 0);
987 	WREG32(mmVM_CONTEXT1_CNTL, 0);
988 	/* Setup TLB control */
989 	tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
990 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
991 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
992 	tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
993 	WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
994 	/* Setup L2 cache */
995 	tmp = RREG32(mmVM_L2_CNTL);
996 	tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
997 	WREG32(mmVM_L2_CNTL, tmp);
998 	WREG32(mmVM_L2_CNTL2, 0);
999 	amdgpu_gart_table_vram_unpin(adev);
1000 }
1001 
1002 /**
1003  * gmc_v8_0_vm_decode_fault - print human readable fault info
1004  *
1005  * @adev: amdgpu_device pointer
1006  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
1007  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
1008  * @mc_client: VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT register value
1009  *
1010  * Print human readable fault information (VI).
1011  */
1012 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev, u32 status,
1013 				     u32 addr, u32 mc_client, unsigned pasid)
1014 {
1015 	u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
1016 	u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1017 					PROTECTIONS);
1018 	char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
1019 		(mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
1020 	u32 mc_id;
1021 
1022 	mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1023 			      MEMORY_CLIENT_ID);
1024 
1025 	dev_err(adev->dev, "VM fault (0x%02x, vmid %d, pasid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
1026 	       protections, vmid, pasid, addr,
1027 	       REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1028 			     MEMORY_CLIENT_RW) ?
1029 	       "write" : "read", block, mc_client, mc_id);
1030 }
1031 
1032 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
1033 {
1034 	switch (mc_seq_vram_type) {
1035 	case MC_SEQ_MISC0__MT__GDDR1:
1036 		return AMDGPU_VRAM_TYPE_GDDR1;
1037 	case MC_SEQ_MISC0__MT__DDR2:
1038 		return AMDGPU_VRAM_TYPE_DDR2;
1039 	case MC_SEQ_MISC0__MT__GDDR3:
1040 		return AMDGPU_VRAM_TYPE_GDDR3;
1041 	case MC_SEQ_MISC0__MT__GDDR4:
1042 		return AMDGPU_VRAM_TYPE_GDDR4;
1043 	case MC_SEQ_MISC0__MT__GDDR5:
1044 		return AMDGPU_VRAM_TYPE_GDDR5;
1045 	case MC_SEQ_MISC0__MT__HBM:
1046 		return AMDGPU_VRAM_TYPE_HBM;
1047 	case MC_SEQ_MISC0__MT__DDR3:
1048 		return AMDGPU_VRAM_TYPE_DDR3;
1049 	default:
1050 		return AMDGPU_VRAM_TYPE_UNKNOWN;
1051 	}
1052 }
1053 
1054 static int gmc_v8_0_early_init(void *handle)
1055 {
1056 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1057 
1058 	gmc_v8_0_set_gmc_funcs(adev);
1059 	gmc_v8_0_set_irq_funcs(adev);
1060 
1061 	adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
1062 	adev->gmc.shared_aperture_end =
1063 		adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
1064 	adev->gmc.private_aperture_start =
1065 		adev->gmc.shared_aperture_end + 1;
1066 	adev->gmc.private_aperture_end =
1067 		adev->gmc.private_aperture_start + (4ULL << 30) - 1;
1068 
1069 	return 0;
1070 }
1071 
1072 static int gmc_v8_0_late_init(void *handle)
1073 {
1074 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1075 
1076 	amdgpu_bo_late_init(adev);
1077 
1078 	if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1079 		return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
1080 	else
1081 		return 0;
1082 }
1083 
1084 static unsigned gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
1085 {
1086 	u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
1087 	unsigned size;
1088 
1089 	if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
1090 		size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
1091 	} else {
1092 		u32 viewport = RREG32(mmVIEWPORT_SIZE);
1093 		size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
1094 			REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
1095 			4);
1096 	}
1097 	/* return 0 if the pre-OS buffer uses up most of vram */
1098 	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
1099 		return 0;
1100 	return size;
1101 }
1102 
1103 #define mmMC_SEQ_MISC0_FIJI 0xA71
1104 
1105 static int gmc_v8_0_sw_init(void *handle)
1106 {
1107 	int r;
1108 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1109 
1110 	adev->num_vmhubs = 1;
1111 
1112 	if (adev->flags & AMD_IS_APU) {
1113 		adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1114 	} else {
1115 		u32 tmp;
1116 
1117 		if ((adev->asic_type == CHIP_FIJI) ||
1118 		    (adev->asic_type == CHIP_VEGAM))
1119 			tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1120 		else
1121 			tmp = RREG32(mmMC_SEQ_MISC0);
1122 		tmp &= MC_SEQ_MISC0__MT__MASK;
1123 		adev->gmc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1124 	}
1125 
1126 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_PAGE_INV_FAULT, &adev->gmc.vm_fault);
1127 	if (r)
1128 		return r;
1129 
1130 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_MEM_PROT_FAULT, &adev->gmc.vm_fault);
1131 	if (r)
1132 		return r;
1133 
1134 	/* Adjust VM size here.
1135 	 * Currently set to 4GB ((1 << 20) 4k pages).
1136 	 * Max GPUVM size for cayman and SI is 40 bits.
1137 	 */
1138 	amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
1139 
1140 	/* Set the internal MC address mask
1141 	 * This is the max address of the GPU's
1142 	 * internal address space.
1143 	 */
1144 	adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1145 
1146 	r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(40));
1147 	if (r) {
1148 		pr_warn("No suitable DMA available\n");
1149 		return r;
1150 	}
1151 	adev->need_swiotlb = drm_need_swiotlb(40);
1152 
1153 	r = gmc_v8_0_init_microcode(adev);
1154 	if (r) {
1155 		DRM_ERROR("Failed to load mc firmware!\n");
1156 		return r;
1157 	}
1158 
1159 	r = gmc_v8_0_mc_init(adev);
1160 	if (r)
1161 		return r;
1162 
1163 	adev->gmc.stolen_size = gmc_v8_0_get_vbios_fb_size(adev);
1164 
1165 	/* Memory manager */
1166 	r = amdgpu_bo_init(adev);
1167 	if (r)
1168 		return r;
1169 
1170 	r = gmc_v8_0_gart_init(adev);
1171 	if (r)
1172 		return r;
1173 
1174 	/*
1175 	 * number of VMs
1176 	 * VMID 0 is reserved for System
1177 	 * amdgpu graphics/compute will use VMIDs 1-7
1178 	 * amdkfd will use VMIDs 8-15
1179 	 */
1180 	adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
1181 	amdgpu_vm_manager_init(adev);
1182 
1183 	/* base offset of vram pages */
1184 	if (adev->flags & AMD_IS_APU) {
1185 		u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1186 
1187 		tmp <<= 22;
1188 		adev->vm_manager.vram_base_offset = tmp;
1189 	} else {
1190 		adev->vm_manager.vram_base_offset = 0;
1191 	}
1192 
1193 	adev->gmc.vm_fault_info = kmalloc(sizeof(struct kfd_vm_fault_info),
1194 					GFP_KERNEL);
1195 	if (!adev->gmc.vm_fault_info)
1196 		return -ENOMEM;
1197 	atomic_set(&adev->gmc.vm_fault_info_updated, 0);
1198 
1199 	return 0;
1200 }
1201 
1202 static int gmc_v8_0_sw_fini(void *handle)
1203 {
1204 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1205 
1206 	amdgpu_gem_force_release(adev);
1207 	amdgpu_vm_manager_fini(adev);
1208 	kfree(adev->gmc.vm_fault_info);
1209 	amdgpu_gart_table_vram_free(adev);
1210 	amdgpu_bo_fini(adev);
1211 	amdgpu_gart_fini(adev);
1212 	release_firmware(adev->gmc.fw);
1213 	adev->gmc.fw = NULL;
1214 
1215 	return 0;
1216 }
1217 
1218 static int gmc_v8_0_hw_init(void *handle)
1219 {
1220 	int r;
1221 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1222 
1223 	gmc_v8_0_init_golden_registers(adev);
1224 
1225 	gmc_v8_0_mc_program(adev);
1226 
1227 	if (adev->asic_type == CHIP_TONGA) {
1228 		r = gmc_v8_0_tonga_mc_load_microcode(adev);
1229 		if (r) {
1230 			DRM_ERROR("Failed to load MC firmware!\n");
1231 			return r;
1232 		}
1233 	} else if (adev->asic_type == CHIP_POLARIS11 ||
1234 			adev->asic_type == CHIP_POLARIS10 ||
1235 			adev->asic_type == CHIP_POLARIS12) {
1236 		r = gmc_v8_0_polaris_mc_load_microcode(adev);
1237 		if (r) {
1238 			DRM_ERROR("Failed to load MC firmware!\n");
1239 			return r;
1240 		}
1241 	}
1242 
1243 	r = gmc_v8_0_gart_enable(adev);
1244 	if (r)
1245 		return r;
1246 
1247 	return r;
1248 }
1249 
1250 static int gmc_v8_0_hw_fini(void *handle)
1251 {
1252 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1253 
1254 	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
1255 	gmc_v8_0_gart_disable(adev);
1256 
1257 	return 0;
1258 }
1259 
1260 static int gmc_v8_0_suspend(void *handle)
1261 {
1262 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1263 
1264 	gmc_v8_0_hw_fini(adev);
1265 
1266 	return 0;
1267 }
1268 
1269 static int gmc_v8_0_resume(void *handle)
1270 {
1271 	int r;
1272 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1273 
1274 	r = gmc_v8_0_hw_init(adev);
1275 	if (r)
1276 		return r;
1277 
1278 	amdgpu_vmid_reset_all(adev);
1279 
1280 	return 0;
1281 }
1282 
1283 static bool gmc_v8_0_is_idle(void *handle)
1284 {
1285 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1286 	u32 tmp = RREG32(mmSRBM_STATUS);
1287 
1288 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1289 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1290 		return false;
1291 
1292 	return true;
1293 }
1294 
1295 static int gmc_v8_0_wait_for_idle(void *handle)
1296 {
1297 	unsigned i;
1298 	u32 tmp;
1299 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1300 
1301 	for (i = 0; i < adev->usec_timeout; i++) {
1302 		/* read MC_STATUS */
1303 		tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1304 					       SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1305 					       SRBM_STATUS__MCC_BUSY_MASK |
1306 					       SRBM_STATUS__MCD_BUSY_MASK |
1307 					       SRBM_STATUS__VMC_BUSY_MASK |
1308 					       SRBM_STATUS__VMC1_BUSY_MASK);
1309 		if (!tmp)
1310 			return 0;
1311 		udelay(1);
1312 	}
1313 	return -ETIMEDOUT;
1314 
1315 }
1316 
1317 static bool gmc_v8_0_check_soft_reset(void *handle)
1318 {
1319 	u32 srbm_soft_reset = 0;
1320 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1321 	u32 tmp = RREG32(mmSRBM_STATUS);
1322 
1323 	if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1324 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1325 						SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1326 
1327 	if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1328 		   SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1329 		if (!(adev->flags & AMD_IS_APU))
1330 			srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1331 							SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1332 	}
1333 	if (srbm_soft_reset) {
1334 		adev->gmc.srbm_soft_reset = srbm_soft_reset;
1335 		return true;
1336 	} else {
1337 		adev->gmc.srbm_soft_reset = 0;
1338 		return false;
1339 	}
1340 }
1341 
1342 static int gmc_v8_0_pre_soft_reset(void *handle)
1343 {
1344 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1345 
1346 	if (!adev->gmc.srbm_soft_reset)
1347 		return 0;
1348 
1349 	gmc_v8_0_mc_stop(adev);
1350 	if (gmc_v8_0_wait_for_idle(adev)) {
1351 		dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1352 	}
1353 
1354 	return 0;
1355 }
1356 
1357 static int gmc_v8_0_soft_reset(void *handle)
1358 {
1359 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1360 	u32 srbm_soft_reset;
1361 
1362 	if (!adev->gmc.srbm_soft_reset)
1363 		return 0;
1364 	srbm_soft_reset = adev->gmc.srbm_soft_reset;
1365 
1366 	if (srbm_soft_reset) {
1367 		u32 tmp;
1368 
1369 		tmp = RREG32(mmSRBM_SOFT_RESET);
1370 		tmp |= srbm_soft_reset;
1371 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1372 		WREG32(mmSRBM_SOFT_RESET, tmp);
1373 		tmp = RREG32(mmSRBM_SOFT_RESET);
1374 
1375 		udelay(50);
1376 
1377 		tmp &= ~srbm_soft_reset;
1378 		WREG32(mmSRBM_SOFT_RESET, tmp);
1379 		tmp = RREG32(mmSRBM_SOFT_RESET);
1380 
1381 		/* Wait a little for things to settle down */
1382 		udelay(50);
1383 	}
1384 
1385 	return 0;
1386 }
1387 
1388 static int gmc_v8_0_post_soft_reset(void *handle)
1389 {
1390 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1391 
1392 	if (!adev->gmc.srbm_soft_reset)
1393 		return 0;
1394 
1395 	gmc_v8_0_mc_resume(adev);
1396 	return 0;
1397 }
1398 
1399 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1400 					     struct amdgpu_irq_src *src,
1401 					     unsigned type,
1402 					     enum amdgpu_interrupt_state state)
1403 {
1404 	u32 tmp;
1405 	u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1406 		    VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1407 		    VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1408 		    VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1409 		    VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1410 		    VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1411 		    VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1412 
1413 	switch (state) {
1414 	case AMDGPU_IRQ_STATE_DISABLE:
1415 		/* system context */
1416 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1417 		tmp &= ~bits;
1418 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1419 		/* VMs */
1420 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1421 		tmp &= ~bits;
1422 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1423 		break;
1424 	case AMDGPU_IRQ_STATE_ENABLE:
1425 		/* system context */
1426 		tmp = RREG32(mmVM_CONTEXT0_CNTL);
1427 		tmp |= bits;
1428 		WREG32(mmVM_CONTEXT0_CNTL, tmp);
1429 		/* VMs */
1430 		tmp = RREG32(mmVM_CONTEXT1_CNTL);
1431 		tmp |= bits;
1432 		WREG32(mmVM_CONTEXT1_CNTL, tmp);
1433 		break;
1434 	default:
1435 		break;
1436 	}
1437 
1438 	return 0;
1439 }
1440 
1441 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1442 				      struct amdgpu_irq_src *source,
1443 				      struct amdgpu_iv_entry *entry)
1444 {
1445 	u32 addr, status, mc_client, vmid;
1446 
1447 	if (amdgpu_sriov_vf(adev)) {
1448 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1449 			entry->src_id, entry->src_data[0]);
1450 		dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1451 		return 0;
1452 	}
1453 
1454 	addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1455 	status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1456 	mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1457 	/* reset addr and status */
1458 	WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1459 
1460 	if (!addr && !status)
1461 		return 0;
1462 
1463 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1464 		gmc_v8_0_set_fault_enable_default(adev, false);
1465 
1466 	if (printk_ratelimit()) {
1467 		struct amdgpu_task_info task_info;
1468 
1469 		memset(&task_info, 0, sizeof(struct amdgpu_task_info));
1470 		amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
1471 
1472 		dev_err(adev->dev, "GPU fault detected: %d 0x%08x for process %s pid %d thread %s pid %d\n",
1473 			entry->src_id, entry->src_data[0], task_info.process_name,
1474 			task_info.tgid, task_info.task_name, task_info.pid);
1475 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1476 			addr);
1477 		dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1478 			status);
1479 		gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client,
1480 					 entry->pasid);
1481 	}
1482 
1483 	vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1484 			     VMID);
1485 	if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
1486 		&& !atomic_read(&adev->gmc.vm_fault_info_updated)) {
1487 		struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
1488 		u32 protections = REG_GET_FIELD(status,
1489 					VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1490 					PROTECTIONS);
1491 
1492 		info->vmid = vmid;
1493 		info->mc_id = REG_GET_FIELD(status,
1494 					    VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1495 					    MEMORY_CLIENT_ID);
1496 		info->status = status;
1497 		info->page_addr = addr;
1498 		info->prot_valid = protections & 0x7 ? true : false;
1499 		info->prot_read = protections & 0x8 ? true : false;
1500 		info->prot_write = protections & 0x10 ? true : false;
1501 		info->prot_exec = protections & 0x20 ? true : false;
1502 		mb();
1503 		atomic_set(&adev->gmc.vm_fault_info_updated, 1);
1504 	}
1505 
1506 	return 0;
1507 }
1508 
1509 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1510 						     bool enable)
1511 {
1512 	uint32_t data;
1513 
1514 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1515 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1516 		data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1517 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1518 
1519 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1520 		data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1521 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1522 
1523 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1524 		data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1525 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1526 
1527 		data = RREG32(mmMC_XPB_CLK_GAT);
1528 		data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1529 		WREG32(mmMC_XPB_CLK_GAT, data);
1530 
1531 		data = RREG32(mmATC_MISC_CG);
1532 		data |= ATC_MISC_CG__ENABLE_MASK;
1533 		WREG32(mmATC_MISC_CG, data);
1534 
1535 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1536 		data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1537 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1538 
1539 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1540 		data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1541 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1542 
1543 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1544 		data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1545 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1546 
1547 		data = RREG32(mmVM_L2_CG);
1548 		data |= VM_L2_CG__ENABLE_MASK;
1549 		WREG32(mmVM_L2_CG, data);
1550 	} else {
1551 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1552 		data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1553 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1554 
1555 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1556 		data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1557 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1558 
1559 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1560 		data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1561 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1562 
1563 		data = RREG32(mmMC_XPB_CLK_GAT);
1564 		data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1565 		WREG32(mmMC_XPB_CLK_GAT, data);
1566 
1567 		data = RREG32(mmATC_MISC_CG);
1568 		data &= ~ATC_MISC_CG__ENABLE_MASK;
1569 		WREG32(mmATC_MISC_CG, data);
1570 
1571 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1572 		data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1573 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1574 
1575 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1576 		data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1577 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1578 
1579 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1580 		data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1581 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1582 
1583 		data = RREG32(mmVM_L2_CG);
1584 		data &= ~VM_L2_CG__ENABLE_MASK;
1585 		WREG32(mmVM_L2_CG, data);
1586 	}
1587 }
1588 
1589 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1590 				       bool enable)
1591 {
1592 	uint32_t data;
1593 
1594 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1595 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1596 		data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1597 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1598 
1599 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1600 		data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1601 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1602 
1603 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1604 		data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1605 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1606 
1607 		data = RREG32(mmMC_XPB_CLK_GAT);
1608 		data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1609 		WREG32(mmMC_XPB_CLK_GAT, data);
1610 
1611 		data = RREG32(mmATC_MISC_CG);
1612 		data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1613 		WREG32(mmATC_MISC_CG, data);
1614 
1615 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1616 		data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1617 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1618 
1619 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1620 		data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1621 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1622 
1623 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1624 		data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1625 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1626 
1627 		data = RREG32(mmVM_L2_CG);
1628 		data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1629 		WREG32(mmVM_L2_CG, data);
1630 	} else {
1631 		data = RREG32(mmMC_HUB_MISC_HUB_CG);
1632 		data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1633 		WREG32(mmMC_HUB_MISC_HUB_CG, data);
1634 
1635 		data = RREG32(mmMC_HUB_MISC_SIP_CG);
1636 		data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1637 		WREG32(mmMC_HUB_MISC_SIP_CG, data);
1638 
1639 		data = RREG32(mmMC_HUB_MISC_VM_CG);
1640 		data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1641 		WREG32(mmMC_HUB_MISC_VM_CG, data);
1642 
1643 		data = RREG32(mmMC_XPB_CLK_GAT);
1644 		data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1645 		WREG32(mmMC_XPB_CLK_GAT, data);
1646 
1647 		data = RREG32(mmATC_MISC_CG);
1648 		data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1649 		WREG32(mmATC_MISC_CG, data);
1650 
1651 		data = RREG32(mmMC_CITF_MISC_WR_CG);
1652 		data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1653 		WREG32(mmMC_CITF_MISC_WR_CG, data);
1654 
1655 		data = RREG32(mmMC_CITF_MISC_RD_CG);
1656 		data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1657 		WREG32(mmMC_CITF_MISC_RD_CG, data);
1658 
1659 		data = RREG32(mmMC_CITF_MISC_VM_CG);
1660 		data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1661 		WREG32(mmMC_CITF_MISC_VM_CG, data);
1662 
1663 		data = RREG32(mmVM_L2_CG);
1664 		data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1665 		WREG32(mmVM_L2_CG, data);
1666 	}
1667 }
1668 
1669 static int gmc_v8_0_set_clockgating_state(void *handle,
1670 					  enum amd_clockgating_state state)
1671 {
1672 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1673 
1674 	if (amdgpu_sriov_vf(adev))
1675 		return 0;
1676 
1677 	switch (adev->asic_type) {
1678 	case CHIP_FIJI:
1679 		fiji_update_mc_medium_grain_clock_gating(adev,
1680 				state == AMD_CG_STATE_GATE);
1681 		fiji_update_mc_light_sleep(adev,
1682 				state == AMD_CG_STATE_GATE);
1683 		break;
1684 	default:
1685 		break;
1686 	}
1687 	return 0;
1688 }
1689 
1690 static int gmc_v8_0_set_powergating_state(void *handle,
1691 					  enum amd_powergating_state state)
1692 {
1693 	return 0;
1694 }
1695 
1696 static void gmc_v8_0_get_clockgating_state(void *handle, u32 *flags)
1697 {
1698 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1699 	int data;
1700 
1701 	if (amdgpu_sriov_vf(adev))
1702 		*flags = 0;
1703 
1704 	/* AMD_CG_SUPPORT_MC_MGCG */
1705 	data = RREG32(mmMC_HUB_MISC_HUB_CG);
1706 	if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1707 		*flags |= AMD_CG_SUPPORT_MC_MGCG;
1708 
1709 	/* AMD_CG_SUPPORT_MC_LS */
1710 	if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1711 		*flags |= AMD_CG_SUPPORT_MC_LS;
1712 }
1713 
1714 static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1715 	.name = "gmc_v8_0",
1716 	.early_init = gmc_v8_0_early_init,
1717 	.late_init = gmc_v8_0_late_init,
1718 	.sw_init = gmc_v8_0_sw_init,
1719 	.sw_fini = gmc_v8_0_sw_fini,
1720 	.hw_init = gmc_v8_0_hw_init,
1721 	.hw_fini = gmc_v8_0_hw_fini,
1722 	.suspend = gmc_v8_0_suspend,
1723 	.resume = gmc_v8_0_resume,
1724 	.is_idle = gmc_v8_0_is_idle,
1725 	.wait_for_idle = gmc_v8_0_wait_for_idle,
1726 	.check_soft_reset = gmc_v8_0_check_soft_reset,
1727 	.pre_soft_reset = gmc_v8_0_pre_soft_reset,
1728 	.soft_reset = gmc_v8_0_soft_reset,
1729 	.post_soft_reset = gmc_v8_0_post_soft_reset,
1730 	.set_clockgating_state = gmc_v8_0_set_clockgating_state,
1731 	.set_powergating_state = gmc_v8_0_set_powergating_state,
1732 	.get_clockgating_state = gmc_v8_0_get_clockgating_state,
1733 };
1734 
1735 static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
1736 	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
1737 	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
1738 	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
1739 	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
1740 	.set_prt = gmc_v8_0_set_prt,
1741 	.get_vm_pde = gmc_v8_0_get_vm_pde,
1742 	.get_vm_pte = gmc_v8_0_get_vm_pte
1743 };
1744 
1745 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1746 	.set = gmc_v8_0_vm_fault_interrupt_state,
1747 	.process = gmc_v8_0_process_interrupt,
1748 };
1749 
1750 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev)
1751 {
1752 	adev->gmc.gmc_funcs = &gmc_v8_0_gmc_funcs;
1753 }
1754 
1755 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1756 {
1757 	adev->gmc.vm_fault.num_types = 1;
1758 	adev->gmc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1759 }
1760 
1761 const struct amdgpu_ip_block_version gmc_v8_0_ip_block =
1762 {
1763 	.type = AMD_IP_BLOCK_TYPE_GMC,
1764 	.major = 8,
1765 	.minor = 0,
1766 	.rev = 0,
1767 	.funcs = &gmc_v8_0_ip_funcs,
1768 };
1769 
1770 const struct amdgpu_ip_block_version gmc_v8_1_ip_block =
1771 {
1772 	.type = AMD_IP_BLOCK_TYPE_GMC,
1773 	.major = 8,
1774 	.minor = 1,
1775 	.rev = 0,
1776 	.funcs = &gmc_v8_0_ip_funcs,
1777 };
1778 
1779 const struct amdgpu_ip_block_version gmc_v8_5_ip_block =
1780 {
1781 	.type = AMD_IP_BLOCK_TYPE_GMC,
1782 	.major = 8,
1783 	.minor = 5,
1784 	.rev = 0,
1785 	.funcs = &gmc_v8_0_ip_funcs,
1786 };
1787