xref: /openbmc/linux/drivers/gpu/drm/radeon/r600.c (revision 90aca4d2740255bd130ea71a91530b9920c70abe)
1 /*
2  * Copyright 2008 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 #include <linux/seq_file.h>
29 #include <linux/firmware.h>
30 #include <linux/platform_device.h>
31 #include "drmP.h"
32 #include "radeon_drm.h"
33 #include "radeon.h"
34 #include "radeon_asic.h"
35 #include "radeon_mode.h"
36 #include "r600d.h"
37 #include "atom.h"
38 #include "avivod.h"
39 
40 #define PFP_UCODE_SIZE 576
41 #define PM4_UCODE_SIZE 1792
42 #define RLC_UCODE_SIZE 768
43 #define R700_PFP_UCODE_SIZE 848
44 #define R700_PM4_UCODE_SIZE 1360
45 #define R700_RLC_UCODE_SIZE 1024
46 
47 /* Firmware Names */
48 MODULE_FIRMWARE("radeon/R600_pfp.bin");
49 MODULE_FIRMWARE("radeon/R600_me.bin");
50 MODULE_FIRMWARE("radeon/RV610_pfp.bin");
51 MODULE_FIRMWARE("radeon/RV610_me.bin");
52 MODULE_FIRMWARE("radeon/RV630_pfp.bin");
53 MODULE_FIRMWARE("radeon/RV630_me.bin");
54 MODULE_FIRMWARE("radeon/RV620_pfp.bin");
55 MODULE_FIRMWARE("radeon/RV620_me.bin");
56 MODULE_FIRMWARE("radeon/RV635_pfp.bin");
57 MODULE_FIRMWARE("radeon/RV635_me.bin");
58 MODULE_FIRMWARE("radeon/RV670_pfp.bin");
59 MODULE_FIRMWARE("radeon/RV670_me.bin");
60 MODULE_FIRMWARE("radeon/RS780_pfp.bin");
61 MODULE_FIRMWARE("radeon/RS780_me.bin");
62 MODULE_FIRMWARE("radeon/RV770_pfp.bin");
63 MODULE_FIRMWARE("radeon/RV770_me.bin");
64 MODULE_FIRMWARE("radeon/RV730_pfp.bin");
65 MODULE_FIRMWARE("radeon/RV730_me.bin");
66 MODULE_FIRMWARE("radeon/RV710_pfp.bin");
67 MODULE_FIRMWARE("radeon/RV710_me.bin");
68 MODULE_FIRMWARE("radeon/R600_rlc.bin");
69 MODULE_FIRMWARE("radeon/R700_rlc.bin");
70 
71 int r600_debugfs_mc_info_init(struct radeon_device *rdev);
72 
73 /* r600,rv610,rv630,rv620,rv635,rv670 */
74 int r600_mc_wait_for_idle(struct radeon_device *rdev);
75 void r600_gpu_init(struct radeon_device *rdev);
76 void r600_fini(struct radeon_device *rdev);
77 
78 /* hpd for digital panel detect/disconnect */
79 bool r600_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
80 {
81 	bool connected = false;
82 
83 	if (ASIC_IS_DCE3(rdev)) {
84 		switch (hpd) {
85 		case RADEON_HPD_1:
86 			if (RREG32(DC_HPD1_INT_STATUS) & DC_HPDx_SENSE)
87 				connected = true;
88 			break;
89 		case RADEON_HPD_2:
90 			if (RREG32(DC_HPD2_INT_STATUS) & DC_HPDx_SENSE)
91 				connected = true;
92 			break;
93 		case RADEON_HPD_3:
94 			if (RREG32(DC_HPD3_INT_STATUS) & DC_HPDx_SENSE)
95 				connected = true;
96 			break;
97 		case RADEON_HPD_4:
98 			if (RREG32(DC_HPD4_INT_STATUS) & DC_HPDx_SENSE)
99 				connected = true;
100 			break;
101 			/* DCE 3.2 */
102 		case RADEON_HPD_5:
103 			if (RREG32(DC_HPD5_INT_STATUS) & DC_HPDx_SENSE)
104 				connected = true;
105 			break;
106 		case RADEON_HPD_6:
107 			if (RREG32(DC_HPD6_INT_STATUS) & DC_HPDx_SENSE)
108 				connected = true;
109 			break;
110 		default:
111 			break;
112 		}
113 	} else {
114 		switch (hpd) {
115 		case RADEON_HPD_1:
116 			if (RREG32(DC_HOT_PLUG_DETECT1_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
117 				connected = true;
118 			break;
119 		case RADEON_HPD_2:
120 			if (RREG32(DC_HOT_PLUG_DETECT2_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
121 				connected = true;
122 			break;
123 		case RADEON_HPD_3:
124 			if (RREG32(DC_HOT_PLUG_DETECT3_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
125 				connected = true;
126 			break;
127 		default:
128 			break;
129 		}
130 	}
131 	return connected;
132 }
133 
134 void r600_hpd_set_polarity(struct radeon_device *rdev,
135 			   enum radeon_hpd_id hpd)
136 {
137 	u32 tmp;
138 	bool connected = r600_hpd_sense(rdev, hpd);
139 
140 	if (ASIC_IS_DCE3(rdev)) {
141 		switch (hpd) {
142 		case RADEON_HPD_1:
143 			tmp = RREG32(DC_HPD1_INT_CONTROL);
144 			if (connected)
145 				tmp &= ~DC_HPDx_INT_POLARITY;
146 			else
147 				tmp |= DC_HPDx_INT_POLARITY;
148 			WREG32(DC_HPD1_INT_CONTROL, tmp);
149 			break;
150 		case RADEON_HPD_2:
151 			tmp = RREG32(DC_HPD2_INT_CONTROL);
152 			if (connected)
153 				tmp &= ~DC_HPDx_INT_POLARITY;
154 			else
155 				tmp |= DC_HPDx_INT_POLARITY;
156 			WREG32(DC_HPD2_INT_CONTROL, tmp);
157 			break;
158 		case RADEON_HPD_3:
159 			tmp = RREG32(DC_HPD3_INT_CONTROL);
160 			if (connected)
161 				tmp &= ~DC_HPDx_INT_POLARITY;
162 			else
163 				tmp |= DC_HPDx_INT_POLARITY;
164 			WREG32(DC_HPD3_INT_CONTROL, tmp);
165 			break;
166 		case RADEON_HPD_4:
167 			tmp = RREG32(DC_HPD4_INT_CONTROL);
168 			if (connected)
169 				tmp &= ~DC_HPDx_INT_POLARITY;
170 			else
171 				tmp |= DC_HPDx_INT_POLARITY;
172 			WREG32(DC_HPD4_INT_CONTROL, tmp);
173 			break;
174 		case RADEON_HPD_5:
175 			tmp = RREG32(DC_HPD5_INT_CONTROL);
176 			if (connected)
177 				tmp &= ~DC_HPDx_INT_POLARITY;
178 			else
179 				tmp |= DC_HPDx_INT_POLARITY;
180 			WREG32(DC_HPD5_INT_CONTROL, tmp);
181 			break;
182 			/* DCE 3.2 */
183 		case RADEON_HPD_6:
184 			tmp = RREG32(DC_HPD6_INT_CONTROL);
185 			if (connected)
186 				tmp &= ~DC_HPDx_INT_POLARITY;
187 			else
188 				tmp |= DC_HPDx_INT_POLARITY;
189 			WREG32(DC_HPD6_INT_CONTROL, tmp);
190 			break;
191 		default:
192 			break;
193 		}
194 	} else {
195 		switch (hpd) {
196 		case RADEON_HPD_1:
197 			tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
198 			if (connected)
199 				tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
200 			else
201 				tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
202 			WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
203 			break;
204 		case RADEON_HPD_2:
205 			tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
206 			if (connected)
207 				tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
208 			else
209 				tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
210 			WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
211 			break;
212 		case RADEON_HPD_3:
213 			tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
214 			if (connected)
215 				tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
216 			else
217 				tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
218 			WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
219 			break;
220 		default:
221 			break;
222 		}
223 	}
224 }
225 
226 void r600_hpd_init(struct radeon_device *rdev)
227 {
228 	struct drm_device *dev = rdev->ddev;
229 	struct drm_connector *connector;
230 
231 	if (ASIC_IS_DCE3(rdev)) {
232 		u32 tmp = DC_HPDx_CONNECTION_TIMER(0x9c4) | DC_HPDx_RX_INT_TIMER(0xfa);
233 		if (ASIC_IS_DCE32(rdev))
234 			tmp |= DC_HPDx_EN;
235 
236 		list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
237 			struct radeon_connector *radeon_connector = to_radeon_connector(connector);
238 			switch (radeon_connector->hpd.hpd) {
239 			case RADEON_HPD_1:
240 				WREG32(DC_HPD1_CONTROL, tmp);
241 				rdev->irq.hpd[0] = true;
242 				break;
243 			case RADEON_HPD_2:
244 				WREG32(DC_HPD2_CONTROL, tmp);
245 				rdev->irq.hpd[1] = true;
246 				break;
247 			case RADEON_HPD_3:
248 				WREG32(DC_HPD3_CONTROL, tmp);
249 				rdev->irq.hpd[2] = true;
250 				break;
251 			case RADEON_HPD_4:
252 				WREG32(DC_HPD4_CONTROL, tmp);
253 				rdev->irq.hpd[3] = true;
254 				break;
255 				/* DCE 3.2 */
256 			case RADEON_HPD_5:
257 				WREG32(DC_HPD5_CONTROL, tmp);
258 				rdev->irq.hpd[4] = true;
259 				break;
260 			case RADEON_HPD_6:
261 				WREG32(DC_HPD6_CONTROL, tmp);
262 				rdev->irq.hpd[5] = true;
263 				break;
264 			default:
265 				break;
266 			}
267 		}
268 	} else {
269 		list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
270 			struct radeon_connector *radeon_connector = to_radeon_connector(connector);
271 			switch (radeon_connector->hpd.hpd) {
272 			case RADEON_HPD_1:
273 				WREG32(DC_HOT_PLUG_DETECT1_CONTROL, DC_HOT_PLUG_DETECTx_EN);
274 				rdev->irq.hpd[0] = true;
275 				break;
276 			case RADEON_HPD_2:
277 				WREG32(DC_HOT_PLUG_DETECT2_CONTROL, DC_HOT_PLUG_DETECTx_EN);
278 				rdev->irq.hpd[1] = true;
279 				break;
280 			case RADEON_HPD_3:
281 				WREG32(DC_HOT_PLUG_DETECT3_CONTROL, DC_HOT_PLUG_DETECTx_EN);
282 				rdev->irq.hpd[2] = true;
283 				break;
284 			default:
285 				break;
286 			}
287 		}
288 	}
289 	if (rdev->irq.installed)
290 		r600_irq_set(rdev);
291 }
292 
293 void r600_hpd_fini(struct radeon_device *rdev)
294 {
295 	struct drm_device *dev = rdev->ddev;
296 	struct drm_connector *connector;
297 
298 	if (ASIC_IS_DCE3(rdev)) {
299 		list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
300 			struct radeon_connector *radeon_connector = to_radeon_connector(connector);
301 			switch (radeon_connector->hpd.hpd) {
302 			case RADEON_HPD_1:
303 				WREG32(DC_HPD1_CONTROL, 0);
304 				rdev->irq.hpd[0] = false;
305 				break;
306 			case RADEON_HPD_2:
307 				WREG32(DC_HPD2_CONTROL, 0);
308 				rdev->irq.hpd[1] = false;
309 				break;
310 			case RADEON_HPD_3:
311 				WREG32(DC_HPD3_CONTROL, 0);
312 				rdev->irq.hpd[2] = false;
313 				break;
314 			case RADEON_HPD_4:
315 				WREG32(DC_HPD4_CONTROL, 0);
316 				rdev->irq.hpd[3] = false;
317 				break;
318 				/* DCE 3.2 */
319 			case RADEON_HPD_5:
320 				WREG32(DC_HPD5_CONTROL, 0);
321 				rdev->irq.hpd[4] = false;
322 				break;
323 			case RADEON_HPD_6:
324 				WREG32(DC_HPD6_CONTROL, 0);
325 				rdev->irq.hpd[5] = false;
326 				break;
327 			default:
328 				break;
329 			}
330 		}
331 	} else {
332 		list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
333 			struct radeon_connector *radeon_connector = to_radeon_connector(connector);
334 			switch (radeon_connector->hpd.hpd) {
335 			case RADEON_HPD_1:
336 				WREG32(DC_HOT_PLUG_DETECT1_CONTROL, 0);
337 				rdev->irq.hpd[0] = false;
338 				break;
339 			case RADEON_HPD_2:
340 				WREG32(DC_HOT_PLUG_DETECT2_CONTROL, 0);
341 				rdev->irq.hpd[1] = false;
342 				break;
343 			case RADEON_HPD_3:
344 				WREG32(DC_HOT_PLUG_DETECT3_CONTROL, 0);
345 				rdev->irq.hpd[2] = false;
346 				break;
347 			default:
348 				break;
349 			}
350 		}
351 	}
352 }
353 
354 /*
355  * R600 PCIE GART
356  */
357 void r600_pcie_gart_tlb_flush(struct radeon_device *rdev)
358 {
359 	unsigned i;
360 	u32 tmp;
361 
362 	/* flush hdp cache so updates hit vram */
363 	WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
364 
365 	WREG32(VM_CONTEXT0_INVALIDATION_LOW_ADDR, rdev->mc.gtt_start >> 12);
366 	WREG32(VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (rdev->mc.gtt_end - 1) >> 12);
367 	WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
368 	for (i = 0; i < rdev->usec_timeout; i++) {
369 		/* read MC_STATUS */
370 		tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
371 		tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
372 		if (tmp == 2) {
373 			printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
374 			return;
375 		}
376 		if (tmp) {
377 			return;
378 		}
379 		udelay(1);
380 	}
381 }
382 
383 int r600_pcie_gart_init(struct radeon_device *rdev)
384 {
385 	int r;
386 
387 	if (rdev->gart.table.vram.robj) {
388 		WARN(1, "R600 PCIE GART already initialized.\n");
389 		return 0;
390 	}
391 	/* Initialize common gart structure */
392 	r = radeon_gart_init(rdev);
393 	if (r)
394 		return r;
395 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 8;
396 	return radeon_gart_table_vram_alloc(rdev);
397 }
398 
399 int r600_pcie_gart_enable(struct radeon_device *rdev)
400 {
401 	u32 tmp;
402 	int r, i;
403 
404 	if (rdev->gart.table.vram.robj == NULL) {
405 		dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
406 		return -EINVAL;
407 	}
408 	r = radeon_gart_table_vram_pin(rdev);
409 	if (r)
410 		return r;
411 	radeon_gart_restore(rdev);
412 
413 	/* Setup L2 cache */
414 	WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
415 				ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
416 				EFFECTIVE_L2_QUEUE_SIZE(7));
417 	WREG32(VM_L2_CNTL2, 0);
418 	WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
419 	/* Setup TLB control */
420 	tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
421 		SYSTEM_ACCESS_MODE_NOT_IN_SYS |
422 		EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
423 		ENABLE_WAIT_L2_QUERY;
424 	WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
425 	WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
426 	WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
427 	WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
428 	WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
429 	WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
430 	WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
431 	WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
432 	WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
433 	WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
434 	WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
435 	WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
436 	WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
437 	WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
438 	WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
439 	WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
440 	WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
441 	WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
442 				RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
443 	WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
444 			(u32)(rdev->dummy_page.addr >> 12));
445 	for (i = 1; i < 7; i++)
446 		WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
447 
448 	r600_pcie_gart_tlb_flush(rdev);
449 	rdev->gart.ready = true;
450 	return 0;
451 }
452 
453 void r600_pcie_gart_disable(struct radeon_device *rdev)
454 {
455 	u32 tmp;
456 	int i, r;
457 
458 	/* Disable all tables */
459 	for (i = 0; i < 7; i++)
460 		WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
461 
462 	/* Disable L2 cache */
463 	WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
464 				EFFECTIVE_L2_QUEUE_SIZE(7));
465 	WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
466 	/* Setup L1 TLB control */
467 	tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
468 		ENABLE_WAIT_L2_QUERY;
469 	WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
470 	WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
471 	WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
472 	WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
473 	WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
474 	WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
475 	WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
476 	WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
477 	WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp);
478 	WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp);
479 	WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
480 	WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
481 	WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp);
482 	WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
483 	if (rdev->gart.table.vram.robj) {
484 		r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
485 		if (likely(r == 0)) {
486 			radeon_bo_kunmap(rdev->gart.table.vram.robj);
487 			radeon_bo_unpin(rdev->gart.table.vram.robj);
488 			radeon_bo_unreserve(rdev->gart.table.vram.robj);
489 		}
490 	}
491 }
492 
493 void r600_pcie_gart_fini(struct radeon_device *rdev)
494 {
495 	radeon_gart_fini(rdev);
496 	r600_pcie_gart_disable(rdev);
497 	radeon_gart_table_vram_free(rdev);
498 }
499 
500 void r600_agp_enable(struct radeon_device *rdev)
501 {
502 	u32 tmp;
503 	int i;
504 
505 	/* Setup L2 cache */
506 	WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
507 				ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
508 				EFFECTIVE_L2_QUEUE_SIZE(7));
509 	WREG32(VM_L2_CNTL2, 0);
510 	WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
511 	/* Setup TLB control */
512 	tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
513 		SYSTEM_ACCESS_MODE_NOT_IN_SYS |
514 		EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
515 		ENABLE_WAIT_L2_QUERY;
516 	WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
517 	WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
518 	WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
519 	WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
520 	WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
521 	WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
522 	WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
523 	WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
524 	WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
525 	WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
526 	WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
527 	WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
528 	WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
529 	WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
530 	for (i = 0; i < 7; i++)
531 		WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
532 }
533 
534 int r600_mc_wait_for_idle(struct radeon_device *rdev)
535 {
536 	unsigned i;
537 	u32 tmp;
538 
539 	for (i = 0; i < rdev->usec_timeout; i++) {
540 		/* read MC_STATUS */
541 		tmp = RREG32(R_000E50_SRBM_STATUS) & 0x3F00;
542 		if (!tmp)
543 			return 0;
544 		udelay(1);
545 	}
546 	return -1;
547 }
548 
549 static void r600_mc_program(struct radeon_device *rdev)
550 {
551 	struct rv515_mc_save save;
552 	u32 tmp;
553 	int i, j;
554 
555 	/* Initialize HDP */
556 	for (i = 0, j = 0; i < 32; i++, j += 0x18) {
557 		WREG32((0x2c14 + j), 0x00000000);
558 		WREG32((0x2c18 + j), 0x00000000);
559 		WREG32((0x2c1c + j), 0x00000000);
560 		WREG32((0x2c20 + j), 0x00000000);
561 		WREG32((0x2c24 + j), 0x00000000);
562 	}
563 	WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
564 
565 	rv515_mc_stop(rdev, &save);
566 	if (r600_mc_wait_for_idle(rdev)) {
567 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
568 	}
569 	/* Lockout access through VGA aperture (doesn't exist before R600) */
570 	WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
571 	/* Update configuration */
572 	if (rdev->flags & RADEON_IS_AGP) {
573 		if (rdev->mc.vram_start < rdev->mc.gtt_start) {
574 			/* VRAM before AGP */
575 			WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
576 				rdev->mc.vram_start >> 12);
577 			WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
578 				rdev->mc.gtt_end >> 12);
579 		} else {
580 			/* VRAM after AGP */
581 			WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
582 				rdev->mc.gtt_start >> 12);
583 			WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
584 				rdev->mc.vram_end >> 12);
585 		}
586 	} else {
587 		WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, rdev->mc.vram_start >> 12);
588 		WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, rdev->mc.vram_end >> 12);
589 	}
590 	WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
591 	tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
592 	tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
593 	WREG32(MC_VM_FB_LOCATION, tmp);
594 	WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
595 	WREG32(HDP_NONSURFACE_INFO, (2 << 7));
596 	WREG32(HDP_NONSURFACE_SIZE, rdev->mc.mc_vram_size | 0x3FF);
597 	if (rdev->flags & RADEON_IS_AGP) {
598 		WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 22);
599 		WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 22);
600 		WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
601 	} else {
602 		WREG32(MC_VM_AGP_BASE, 0);
603 		WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
604 		WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
605 	}
606 	if (r600_mc_wait_for_idle(rdev)) {
607 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
608 	}
609 	rv515_mc_resume(rdev, &save);
610 	/* we need to own VRAM, so turn off the VGA renderer here
611 	 * to stop it overwriting our objects */
612 	rv515_vga_render_disable(rdev);
613 }
614 
615 /**
616  * r600_vram_gtt_location - try to find VRAM & GTT location
617  * @rdev: radeon device structure holding all necessary informations
618  * @mc: memory controller structure holding memory informations
619  *
620  * Function will place try to place VRAM at same place as in CPU (PCI)
621  * address space as some GPU seems to have issue when we reprogram at
622  * different address space.
623  *
624  * If there is not enough space to fit the unvisible VRAM after the
625  * aperture then we limit the VRAM size to the aperture.
626  *
627  * If we are using AGP then place VRAM adjacent to AGP aperture are we need
628  * them to be in one from GPU point of view so that we can program GPU to
629  * catch access outside them (weird GPU policy see ??).
630  *
631  * This function will never fails, worst case are limiting VRAM or GTT.
632  *
633  * Note: GTT start, end, size should be initialized before calling this
634  * function on AGP platform.
635  */
636 void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
637 {
638 	u64 size_bf, size_af;
639 
640 	if (mc->mc_vram_size > 0xE0000000) {
641 		/* leave room for at least 512M GTT */
642 		dev_warn(rdev->dev, "limiting VRAM\n");
643 		mc->real_vram_size = 0xE0000000;
644 		mc->mc_vram_size = 0xE0000000;
645 	}
646 	if (rdev->flags & RADEON_IS_AGP) {
647 		size_bf = mc->gtt_start;
648 		size_af = 0xFFFFFFFF - mc->gtt_end + 1;
649 		if (size_bf > size_af) {
650 			if (mc->mc_vram_size > size_bf) {
651 				dev_warn(rdev->dev, "limiting VRAM\n");
652 				mc->real_vram_size = size_bf;
653 				mc->mc_vram_size = size_bf;
654 			}
655 			mc->vram_start = mc->gtt_start - mc->mc_vram_size;
656 		} else {
657 			if (mc->mc_vram_size > size_af) {
658 				dev_warn(rdev->dev, "limiting VRAM\n");
659 				mc->real_vram_size = size_af;
660 				mc->mc_vram_size = size_af;
661 			}
662 			mc->vram_start = mc->gtt_end;
663 		}
664 		mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
665 		dev_info(rdev->dev, "VRAM: %lluM 0x%08llX - 0x%08llX (%lluM used)\n",
666 				mc->mc_vram_size >> 20, mc->vram_start,
667 				mc->vram_end, mc->real_vram_size >> 20);
668 	} else {
669 		u64 base = 0;
670 		if (rdev->flags & RADEON_IS_IGP)
671 			base = (RREG32(MC_VM_FB_LOCATION) & 0xFFFF) << 24;
672 		radeon_vram_location(rdev, &rdev->mc, base);
673 		radeon_gtt_location(rdev, mc);
674 	}
675 }
676 
677 int r600_mc_init(struct radeon_device *rdev)
678 {
679 	u32 tmp;
680 	int chansize, numchan;
681 
682 	/* Get VRAM informations */
683 	rdev->mc.vram_is_ddr = true;
684 	tmp = RREG32(RAMCFG);
685 	if (tmp & CHANSIZE_OVERRIDE) {
686 		chansize = 16;
687 	} else if (tmp & CHANSIZE_MASK) {
688 		chansize = 64;
689 	} else {
690 		chansize = 32;
691 	}
692 	tmp = RREG32(CHMAP);
693 	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
694 	case 0:
695 	default:
696 		numchan = 1;
697 		break;
698 	case 1:
699 		numchan = 2;
700 		break;
701 	case 2:
702 		numchan = 4;
703 		break;
704 	case 3:
705 		numchan = 8;
706 		break;
707 	}
708 	rdev->mc.vram_width = numchan * chansize;
709 	/* Could aper size report 0 ? */
710 	rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
711 	rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
712 	/* Setup GPU memory space */
713 	rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
714 	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
715 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
716 	/* FIXME remove this once we support unmappable VRAM */
717 	if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
718 		rdev->mc.mc_vram_size = rdev->mc.aper_size;
719 		rdev->mc.real_vram_size = rdev->mc.aper_size;
720 	}
721 	r600_vram_gtt_location(rdev, &rdev->mc);
722 
723 	if (rdev->flags & RADEON_IS_IGP)
724 		rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
725 	radeon_update_bandwidth_info(rdev);
726 	return 0;
727 }
728 
729 /* We doesn't check that the GPU really needs a reset we simply do the
730  * reset, it's up to the caller to determine if the GPU needs one. We
731  * might add an helper function to check that.
732  */
733 int r600_gpu_soft_reset(struct radeon_device *rdev)
734 {
735 	struct rv515_mc_save save;
736 	u32 grbm_busy_mask = S_008010_VC_BUSY(1) | S_008010_VGT_BUSY_NO_DMA(1) |
737 				S_008010_VGT_BUSY(1) | S_008010_TA03_BUSY(1) |
738 				S_008010_TC_BUSY(1) | S_008010_SX_BUSY(1) |
739 				S_008010_SH_BUSY(1) | S_008010_SPI03_BUSY(1) |
740 				S_008010_SMX_BUSY(1) | S_008010_SC_BUSY(1) |
741 				S_008010_PA_BUSY(1) | S_008010_DB03_BUSY(1) |
742 				S_008010_CR_BUSY(1) | S_008010_CB03_BUSY(1) |
743 				S_008010_GUI_ACTIVE(1);
744 	u32 grbm2_busy_mask = S_008014_SPI0_BUSY(1) | S_008014_SPI1_BUSY(1) |
745 			S_008014_SPI2_BUSY(1) | S_008014_SPI3_BUSY(1) |
746 			S_008014_TA0_BUSY(1) | S_008014_TA1_BUSY(1) |
747 			S_008014_TA2_BUSY(1) | S_008014_TA3_BUSY(1) |
748 			S_008014_DB0_BUSY(1) | S_008014_DB1_BUSY(1) |
749 			S_008014_DB2_BUSY(1) | S_008014_DB3_BUSY(1) |
750 			S_008014_CB0_BUSY(1) | S_008014_CB1_BUSY(1) |
751 			S_008014_CB2_BUSY(1) | S_008014_CB3_BUSY(1);
752 	u32 tmp;
753 
754 	dev_info(rdev->dev, "GPU softreset \n");
755 	dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
756 		RREG32(R_008010_GRBM_STATUS));
757 	dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
758 		RREG32(R_008014_GRBM_STATUS2));
759 	dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
760 		RREG32(R_000E50_SRBM_STATUS));
761 	rv515_mc_stop(rdev, &save);
762 	if (r600_mc_wait_for_idle(rdev)) {
763 		dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
764 	}
765 	/* Disable CP parsing/prefetching */
766 	WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
767 	/* Check if any of the rendering block is busy and reset it */
768 	if ((RREG32(R_008010_GRBM_STATUS) & grbm_busy_mask) ||
769 	    (RREG32(R_008014_GRBM_STATUS2) & grbm2_busy_mask)) {
770 		tmp = S_008020_SOFT_RESET_CR(1) |
771 			S_008020_SOFT_RESET_DB(1) |
772 			S_008020_SOFT_RESET_CB(1) |
773 			S_008020_SOFT_RESET_PA(1) |
774 			S_008020_SOFT_RESET_SC(1) |
775 			S_008020_SOFT_RESET_SMX(1) |
776 			S_008020_SOFT_RESET_SPI(1) |
777 			S_008020_SOFT_RESET_SX(1) |
778 			S_008020_SOFT_RESET_SH(1) |
779 			S_008020_SOFT_RESET_TC(1) |
780 			S_008020_SOFT_RESET_TA(1) |
781 			S_008020_SOFT_RESET_VC(1) |
782 			S_008020_SOFT_RESET_VGT(1);
783 		dev_info(rdev->dev, "  R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
784 		WREG32(R_008020_GRBM_SOFT_RESET, tmp);
785 		RREG32(R_008020_GRBM_SOFT_RESET);
786 		mdelay(15);
787 		WREG32(R_008020_GRBM_SOFT_RESET, 0);
788 	}
789 	/* Reset CP (we always reset CP) */
790 	tmp = S_008020_SOFT_RESET_CP(1);
791 	dev_info(rdev->dev, "R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
792 	WREG32(R_008020_GRBM_SOFT_RESET, tmp);
793 	RREG32(R_008020_GRBM_SOFT_RESET);
794 	mdelay(15);
795 	WREG32(R_008020_GRBM_SOFT_RESET, 0);
796 	/* Wait a little for things to settle down */
797 	mdelay(1);
798 	dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
799 		RREG32(R_008010_GRBM_STATUS));
800 	dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
801 		RREG32(R_008014_GRBM_STATUS2));
802 	dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
803 		RREG32(R_000E50_SRBM_STATUS));
804 	rv515_mc_resume(rdev, &save);
805 	return 0;
806 }
807 
808 bool r600_gpu_is_lockup(struct radeon_device *rdev)
809 {
810 	u32 srbm_status;
811 	u32 grbm_status;
812 	u32 grbm_status2;
813 	int r;
814 
815 	srbm_status = RREG32(R_000E50_SRBM_STATUS);
816 	grbm_status = RREG32(R_008010_GRBM_STATUS);
817 	grbm_status2 = RREG32(R_008014_GRBM_STATUS2);
818 	if (!G_008010_GUI_ACTIVE(grbm_status)) {
819 		r100_gpu_lockup_update(&rdev->config.r300.lockup, &rdev->cp);
820 		return false;
821 	}
822 	/* force CP activities */
823 	r = radeon_ring_lock(rdev, 2);
824 	if (!r) {
825 		/* PACKET2 NOP */
826 		radeon_ring_write(rdev, 0x80000000);
827 		radeon_ring_write(rdev, 0x80000000);
828 		radeon_ring_unlock_commit(rdev);
829 	}
830 	rdev->cp.rptr = RREG32(R600_CP_RB_RPTR);
831 	return r100_gpu_cp_is_lockup(rdev, &rdev->config.r300.lockup, &rdev->cp);
832 }
833 
834 int r600_asic_reset(struct radeon_device *rdev)
835 {
836 	return r600_gpu_soft_reset(rdev);
837 }
838 
839 static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
840 					     u32 num_backends,
841 					     u32 backend_disable_mask)
842 {
843 	u32 backend_map = 0;
844 	u32 enabled_backends_mask;
845 	u32 enabled_backends_count;
846 	u32 cur_pipe;
847 	u32 swizzle_pipe[R6XX_MAX_PIPES];
848 	u32 cur_backend;
849 	u32 i;
850 
851 	if (num_tile_pipes > R6XX_MAX_PIPES)
852 		num_tile_pipes = R6XX_MAX_PIPES;
853 	if (num_tile_pipes < 1)
854 		num_tile_pipes = 1;
855 	if (num_backends > R6XX_MAX_BACKENDS)
856 		num_backends = R6XX_MAX_BACKENDS;
857 	if (num_backends < 1)
858 		num_backends = 1;
859 
860 	enabled_backends_mask = 0;
861 	enabled_backends_count = 0;
862 	for (i = 0; i < R6XX_MAX_BACKENDS; ++i) {
863 		if (((backend_disable_mask >> i) & 1) == 0) {
864 			enabled_backends_mask |= (1 << i);
865 			++enabled_backends_count;
866 		}
867 		if (enabled_backends_count == num_backends)
868 			break;
869 	}
870 
871 	if (enabled_backends_count == 0) {
872 		enabled_backends_mask = 1;
873 		enabled_backends_count = 1;
874 	}
875 
876 	if (enabled_backends_count != num_backends)
877 		num_backends = enabled_backends_count;
878 
879 	memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES);
880 	switch (num_tile_pipes) {
881 	case 1:
882 		swizzle_pipe[0] = 0;
883 		break;
884 	case 2:
885 		swizzle_pipe[0] = 0;
886 		swizzle_pipe[1] = 1;
887 		break;
888 	case 3:
889 		swizzle_pipe[0] = 0;
890 		swizzle_pipe[1] = 1;
891 		swizzle_pipe[2] = 2;
892 		break;
893 	case 4:
894 		swizzle_pipe[0] = 0;
895 		swizzle_pipe[1] = 1;
896 		swizzle_pipe[2] = 2;
897 		swizzle_pipe[3] = 3;
898 		break;
899 	case 5:
900 		swizzle_pipe[0] = 0;
901 		swizzle_pipe[1] = 1;
902 		swizzle_pipe[2] = 2;
903 		swizzle_pipe[3] = 3;
904 		swizzle_pipe[4] = 4;
905 		break;
906 	case 6:
907 		swizzle_pipe[0] = 0;
908 		swizzle_pipe[1] = 2;
909 		swizzle_pipe[2] = 4;
910 		swizzle_pipe[3] = 5;
911 		swizzle_pipe[4] = 1;
912 		swizzle_pipe[5] = 3;
913 		break;
914 	case 7:
915 		swizzle_pipe[0] = 0;
916 		swizzle_pipe[1] = 2;
917 		swizzle_pipe[2] = 4;
918 		swizzle_pipe[3] = 6;
919 		swizzle_pipe[4] = 1;
920 		swizzle_pipe[5] = 3;
921 		swizzle_pipe[6] = 5;
922 		break;
923 	case 8:
924 		swizzle_pipe[0] = 0;
925 		swizzle_pipe[1] = 2;
926 		swizzle_pipe[2] = 4;
927 		swizzle_pipe[3] = 6;
928 		swizzle_pipe[4] = 1;
929 		swizzle_pipe[5] = 3;
930 		swizzle_pipe[6] = 5;
931 		swizzle_pipe[7] = 7;
932 		break;
933 	}
934 
935 	cur_backend = 0;
936 	for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
937 		while (((1 << cur_backend) & enabled_backends_mask) == 0)
938 			cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
939 
940 		backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
941 
942 		cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
943 	}
944 
945 	return backend_map;
946 }
947 
948 int r600_count_pipe_bits(uint32_t val)
949 {
950 	int i, ret = 0;
951 
952 	for (i = 0; i < 32; i++) {
953 		ret += val & 1;
954 		val >>= 1;
955 	}
956 	return ret;
957 }
958 
959 void r600_gpu_init(struct radeon_device *rdev)
960 {
961 	u32 tiling_config;
962 	u32 ramcfg;
963 	u32 backend_map;
964 	u32 cc_rb_backend_disable;
965 	u32 cc_gc_shader_pipe_config;
966 	u32 tmp;
967 	int i, j;
968 	u32 sq_config;
969 	u32 sq_gpr_resource_mgmt_1 = 0;
970 	u32 sq_gpr_resource_mgmt_2 = 0;
971 	u32 sq_thread_resource_mgmt = 0;
972 	u32 sq_stack_resource_mgmt_1 = 0;
973 	u32 sq_stack_resource_mgmt_2 = 0;
974 
975 	/* FIXME: implement */
976 	switch (rdev->family) {
977 	case CHIP_R600:
978 		rdev->config.r600.max_pipes = 4;
979 		rdev->config.r600.max_tile_pipes = 8;
980 		rdev->config.r600.max_simds = 4;
981 		rdev->config.r600.max_backends = 4;
982 		rdev->config.r600.max_gprs = 256;
983 		rdev->config.r600.max_threads = 192;
984 		rdev->config.r600.max_stack_entries = 256;
985 		rdev->config.r600.max_hw_contexts = 8;
986 		rdev->config.r600.max_gs_threads = 16;
987 		rdev->config.r600.sx_max_export_size = 128;
988 		rdev->config.r600.sx_max_export_pos_size = 16;
989 		rdev->config.r600.sx_max_export_smx_size = 128;
990 		rdev->config.r600.sq_num_cf_insts = 2;
991 		break;
992 	case CHIP_RV630:
993 	case CHIP_RV635:
994 		rdev->config.r600.max_pipes = 2;
995 		rdev->config.r600.max_tile_pipes = 2;
996 		rdev->config.r600.max_simds = 3;
997 		rdev->config.r600.max_backends = 1;
998 		rdev->config.r600.max_gprs = 128;
999 		rdev->config.r600.max_threads = 192;
1000 		rdev->config.r600.max_stack_entries = 128;
1001 		rdev->config.r600.max_hw_contexts = 8;
1002 		rdev->config.r600.max_gs_threads = 4;
1003 		rdev->config.r600.sx_max_export_size = 128;
1004 		rdev->config.r600.sx_max_export_pos_size = 16;
1005 		rdev->config.r600.sx_max_export_smx_size = 128;
1006 		rdev->config.r600.sq_num_cf_insts = 2;
1007 		break;
1008 	case CHIP_RV610:
1009 	case CHIP_RV620:
1010 	case CHIP_RS780:
1011 	case CHIP_RS880:
1012 		rdev->config.r600.max_pipes = 1;
1013 		rdev->config.r600.max_tile_pipes = 1;
1014 		rdev->config.r600.max_simds = 2;
1015 		rdev->config.r600.max_backends = 1;
1016 		rdev->config.r600.max_gprs = 128;
1017 		rdev->config.r600.max_threads = 192;
1018 		rdev->config.r600.max_stack_entries = 128;
1019 		rdev->config.r600.max_hw_contexts = 4;
1020 		rdev->config.r600.max_gs_threads = 4;
1021 		rdev->config.r600.sx_max_export_size = 128;
1022 		rdev->config.r600.sx_max_export_pos_size = 16;
1023 		rdev->config.r600.sx_max_export_smx_size = 128;
1024 		rdev->config.r600.sq_num_cf_insts = 1;
1025 		break;
1026 	case CHIP_RV670:
1027 		rdev->config.r600.max_pipes = 4;
1028 		rdev->config.r600.max_tile_pipes = 4;
1029 		rdev->config.r600.max_simds = 4;
1030 		rdev->config.r600.max_backends = 4;
1031 		rdev->config.r600.max_gprs = 192;
1032 		rdev->config.r600.max_threads = 192;
1033 		rdev->config.r600.max_stack_entries = 256;
1034 		rdev->config.r600.max_hw_contexts = 8;
1035 		rdev->config.r600.max_gs_threads = 16;
1036 		rdev->config.r600.sx_max_export_size = 128;
1037 		rdev->config.r600.sx_max_export_pos_size = 16;
1038 		rdev->config.r600.sx_max_export_smx_size = 128;
1039 		rdev->config.r600.sq_num_cf_insts = 2;
1040 		break;
1041 	default:
1042 		break;
1043 	}
1044 
1045 	/* Initialize HDP */
1046 	for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1047 		WREG32((0x2c14 + j), 0x00000000);
1048 		WREG32((0x2c18 + j), 0x00000000);
1049 		WREG32((0x2c1c + j), 0x00000000);
1050 		WREG32((0x2c20 + j), 0x00000000);
1051 		WREG32((0x2c24 + j), 0x00000000);
1052 	}
1053 
1054 	WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1055 
1056 	/* Setup tiling */
1057 	tiling_config = 0;
1058 	ramcfg = RREG32(RAMCFG);
1059 	switch (rdev->config.r600.max_tile_pipes) {
1060 	case 1:
1061 		tiling_config |= PIPE_TILING(0);
1062 		break;
1063 	case 2:
1064 		tiling_config |= PIPE_TILING(1);
1065 		break;
1066 	case 4:
1067 		tiling_config |= PIPE_TILING(2);
1068 		break;
1069 	case 8:
1070 		tiling_config |= PIPE_TILING(3);
1071 		break;
1072 	default:
1073 		break;
1074 	}
1075 	rdev->config.r600.tiling_npipes = rdev->config.r600.max_tile_pipes;
1076 	rdev->config.r600.tiling_nbanks = 4 << ((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1077 	tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1078 	tiling_config |= GROUP_SIZE(0);
1079 	rdev->config.r600.tiling_group_size = 256;
1080 	tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT;
1081 	if (tmp > 3) {
1082 		tiling_config |= ROW_TILING(3);
1083 		tiling_config |= SAMPLE_SPLIT(3);
1084 	} else {
1085 		tiling_config |= ROW_TILING(tmp);
1086 		tiling_config |= SAMPLE_SPLIT(tmp);
1087 	}
1088 	tiling_config |= BANK_SWAPS(1);
1089 
1090 	cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000;
1091 	cc_rb_backend_disable |=
1092 		BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << rdev->config.r600.max_backends) & R6XX_MAX_BACKENDS_MASK);
1093 
1094 	cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffffff00;
1095 	cc_gc_shader_pipe_config |=
1096 		INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << rdev->config.r600.max_pipes) & R6XX_MAX_PIPES_MASK);
1097 	cc_gc_shader_pipe_config |=
1098 		INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << rdev->config.r600.max_simds) & R6XX_MAX_SIMDS_MASK);
1099 
1100 	backend_map = r600_get_tile_pipe_to_backend_map(rdev->config.r600.max_tile_pipes,
1101 							(R6XX_MAX_BACKENDS -
1102 							 r600_count_pipe_bits((cc_rb_backend_disable &
1103 									       R6XX_MAX_BACKENDS_MASK) >> 16)),
1104 							(cc_rb_backend_disable >> 16));
1105 
1106 	tiling_config |= BACKEND_MAP(backend_map);
1107 	WREG32(GB_TILING_CONFIG, tiling_config);
1108 	WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff);
1109 	WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff);
1110 
1111 	/* Setup pipes */
1112 	WREG32(CC_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1113 	WREG32(CC_GC_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
1114 	WREG32(GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
1115 
1116 	tmp = R6XX_MAX_PIPES - r600_count_pipe_bits((cc_gc_shader_pipe_config & INACTIVE_QD_PIPES_MASK) >> 8);
1117 	WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK);
1118 	WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK);
1119 
1120 	/* Setup some CP states */
1121 	WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | ROQ_IB2_START(0x2b)));
1122 	WREG32(CP_MEQ_THRESHOLDS, (MEQ_END(0x40) | ROQ_END(0x40)));
1123 
1124 	WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO | SYNC_GRADIENT |
1125 			     SYNC_WALKER | SYNC_ALIGNER));
1126 	/* Setup various GPU states */
1127 	if (rdev->family == CHIP_RV670)
1128 		WREG32(ARB_GDEC_RD_CNTL, 0x00000021);
1129 
1130 	tmp = RREG32(SX_DEBUG_1);
1131 	tmp |= SMX_EVENT_RELEASE;
1132 	if ((rdev->family > CHIP_R600))
1133 		tmp |= ENABLE_NEW_SMX_ADDRESS;
1134 	WREG32(SX_DEBUG_1, tmp);
1135 
1136 	if (((rdev->family) == CHIP_R600) ||
1137 	    ((rdev->family) == CHIP_RV630) ||
1138 	    ((rdev->family) == CHIP_RV610) ||
1139 	    ((rdev->family) == CHIP_RV620) ||
1140 	    ((rdev->family) == CHIP_RS780) ||
1141 	    ((rdev->family) == CHIP_RS880)) {
1142 		WREG32(DB_DEBUG, PREZ_MUST_WAIT_FOR_POSTZ_DONE);
1143 	} else {
1144 		WREG32(DB_DEBUG, 0);
1145 	}
1146 	WREG32(DB_WATERMARKS, (DEPTH_FREE(4) | DEPTH_CACHELINE_FREE(16) |
1147 			       DEPTH_FLUSH(16) | DEPTH_PENDING_FREE(4)));
1148 
1149 	WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1150 	WREG32(VGT_NUM_INSTANCES, 0);
1151 
1152 	WREG32(SPI_CONFIG_CNTL, GPR_WRITE_PRIORITY(0));
1153 	WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(0));
1154 
1155 	tmp = RREG32(SQ_MS_FIFO_SIZES);
1156 	if (((rdev->family) == CHIP_RV610) ||
1157 	    ((rdev->family) == CHIP_RV620) ||
1158 	    ((rdev->family) == CHIP_RS780) ||
1159 	    ((rdev->family) == CHIP_RS880)) {
1160 		tmp = (CACHE_FIFO_SIZE(0xa) |
1161 		       FETCH_FIFO_HIWATER(0xa) |
1162 		       DONE_FIFO_HIWATER(0xe0) |
1163 		       ALU_UPDATE_FIFO_HIWATER(0x8));
1164 	} else if (((rdev->family) == CHIP_R600) ||
1165 		   ((rdev->family) == CHIP_RV630)) {
1166 		tmp &= ~DONE_FIFO_HIWATER(0xff);
1167 		tmp |= DONE_FIFO_HIWATER(0x4);
1168 	}
1169 	WREG32(SQ_MS_FIFO_SIZES, tmp);
1170 
1171 	/* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
1172 	 * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
1173 	 */
1174 	sq_config = RREG32(SQ_CONFIG);
1175 	sq_config &= ~(PS_PRIO(3) |
1176 		       VS_PRIO(3) |
1177 		       GS_PRIO(3) |
1178 		       ES_PRIO(3));
1179 	sq_config |= (DX9_CONSTS |
1180 		      VC_ENABLE |
1181 		      PS_PRIO(0) |
1182 		      VS_PRIO(1) |
1183 		      GS_PRIO(2) |
1184 		      ES_PRIO(3));
1185 
1186 	if ((rdev->family) == CHIP_R600) {
1187 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(124) |
1188 					  NUM_VS_GPRS(124) |
1189 					  NUM_CLAUSE_TEMP_GPRS(4));
1190 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(0) |
1191 					  NUM_ES_GPRS(0));
1192 		sq_thread_resource_mgmt = (NUM_PS_THREADS(136) |
1193 					   NUM_VS_THREADS(48) |
1194 					   NUM_GS_THREADS(4) |
1195 					   NUM_ES_THREADS(4));
1196 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(128) |
1197 					    NUM_VS_STACK_ENTRIES(128));
1198 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(0) |
1199 					    NUM_ES_STACK_ENTRIES(0));
1200 	} else if (((rdev->family) == CHIP_RV610) ||
1201 		   ((rdev->family) == CHIP_RV620) ||
1202 		   ((rdev->family) == CHIP_RS780) ||
1203 		   ((rdev->family) == CHIP_RS880)) {
1204 		/* no vertex cache */
1205 		sq_config &= ~VC_ENABLE;
1206 
1207 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1208 					  NUM_VS_GPRS(44) |
1209 					  NUM_CLAUSE_TEMP_GPRS(2));
1210 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1211 					  NUM_ES_GPRS(17));
1212 		sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1213 					   NUM_VS_THREADS(78) |
1214 					   NUM_GS_THREADS(4) |
1215 					   NUM_ES_THREADS(31));
1216 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1217 					    NUM_VS_STACK_ENTRIES(40));
1218 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1219 					    NUM_ES_STACK_ENTRIES(16));
1220 	} else if (((rdev->family) == CHIP_RV630) ||
1221 		   ((rdev->family) == CHIP_RV635)) {
1222 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1223 					  NUM_VS_GPRS(44) |
1224 					  NUM_CLAUSE_TEMP_GPRS(2));
1225 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(18) |
1226 					  NUM_ES_GPRS(18));
1227 		sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1228 					   NUM_VS_THREADS(78) |
1229 					   NUM_GS_THREADS(4) |
1230 					   NUM_ES_THREADS(31));
1231 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1232 					    NUM_VS_STACK_ENTRIES(40));
1233 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1234 					    NUM_ES_STACK_ENTRIES(16));
1235 	} else if ((rdev->family) == CHIP_RV670) {
1236 		sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1237 					  NUM_VS_GPRS(44) |
1238 					  NUM_CLAUSE_TEMP_GPRS(2));
1239 		sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1240 					  NUM_ES_GPRS(17));
1241 		sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1242 					   NUM_VS_THREADS(78) |
1243 					   NUM_GS_THREADS(4) |
1244 					   NUM_ES_THREADS(31));
1245 		sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(64) |
1246 					    NUM_VS_STACK_ENTRIES(64));
1247 		sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(64) |
1248 					    NUM_ES_STACK_ENTRIES(64));
1249 	}
1250 
1251 	WREG32(SQ_CONFIG, sq_config);
1252 	WREG32(SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
1253 	WREG32(SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
1254 	WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1255 	WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
1256 	WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
1257 
1258 	if (((rdev->family) == CHIP_RV610) ||
1259 	    ((rdev->family) == CHIP_RV620) ||
1260 	    ((rdev->family) == CHIP_RS780) ||
1261 	    ((rdev->family) == CHIP_RS880)) {
1262 		WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(TC_ONLY));
1263 	} else {
1264 		WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC));
1265 	}
1266 
1267 	/* More default values. 2D/3D driver should adjust as needed */
1268 	WREG32(PA_SC_AA_SAMPLE_LOCS_2S, (S0_X(0xc) | S0_Y(0x4) |
1269 					 S1_X(0x4) | S1_Y(0xc)));
1270 	WREG32(PA_SC_AA_SAMPLE_LOCS_4S, (S0_X(0xe) | S0_Y(0xe) |
1271 					 S1_X(0x2) | S1_Y(0x2) |
1272 					 S2_X(0xa) | S2_Y(0x6) |
1273 					 S3_X(0x6) | S3_Y(0xa)));
1274 	WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD0, (S0_X(0xe) | S0_Y(0xb) |
1275 					     S1_X(0x4) | S1_Y(0xc) |
1276 					     S2_X(0x1) | S2_Y(0x6) |
1277 					     S3_X(0xa) | S3_Y(0xe)));
1278 	WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD1, (S4_X(0x6) | S4_Y(0x1) |
1279 					     S5_X(0x0) | S5_Y(0x0) |
1280 					     S6_X(0xb) | S6_Y(0x4) |
1281 					     S7_X(0x7) | S7_Y(0x8)));
1282 
1283 	WREG32(VGT_STRMOUT_EN, 0);
1284 	tmp = rdev->config.r600.max_pipes * 16;
1285 	switch (rdev->family) {
1286 	case CHIP_RV610:
1287 	case CHIP_RV620:
1288 	case CHIP_RS780:
1289 	case CHIP_RS880:
1290 		tmp += 32;
1291 		break;
1292 	case CHIP_RV670:
1293 		tmp += 128;
1294 		break;
1295 	default:
1296 		break;
1297 	}
1298 	if (tmp > 256) {
1299 		tmp = 256;
1300 	}
1301 	WREG32(VGT_ES_PER_GS, 128);
1302 	WREG32(VGT_GS_PER_ES, tmp);
1303 	WREG32(VGT_GS_PER_VS, 2);
1304 	WREG32(VGT_GS_VERTEX_REUSE, 16);
1305 
1306 	/* more default values. 2D/3D driver should adjust as needed */
1307 	WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1308 	WREG32(VGT_STRMOUT_EN, 0);
1309 	WREG32(SX_MISC, 0);
1310 	WREG32(PA_SC_MODE_CNTL, 0);
1311 	WREG32(PA_SC_AA_CONFIG, 0);
1312 	WREG32(PA_SC_LINE_STIPPLE, 0);
1313 	WREG32(SPI_INPUT_Z, 0);
1314 	WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2));
1315 	WREG32(CB_COLOR7_FRAG, 0);
1316 
1317 	/* Clear render buffer base addresses */
1318 	WREG32(CB_COLOR0_BASE, 0);
1319 	WREG32(CB_COLOR1_BASE, 0);
1320 	WREG32(CB_COLOR2_BASE, 0);
1321 	WREG32(CB_COLOR3_BASE, 0);
1322 	WREG32(CB_COLOR4_BASE, 0);
1323 	WREG32(CB_COLOR5_BASE, 0);
1324 	WREG32(CB_COLOR6_BASE, 0);
1325 	WREG32(CB_COLOR7_BASE, 0);
1326 	WREG32(CB_COLOR7_FRAG, 0);
1327 
1328 	switch (rdev->family) {
1329 	case CHIP_RV610:
1330 	case CHIP_RV620:
1331 	case CHIP_RS780:
1332 	case CHIP_RS880:
1333 		tmp = TC_L2_SIZE(8);
1334 		break;
1335 	case CHIP_RV630:
1336 	case CHIP_RV635:
1337 		tmp = TC_L2_SIZE(4);
1338 		break;
1339 	case CHIP_R600:
1340 		tmp = TC_L2_SIZE(0) | L2_DISABLE_LATE_HIT;
1341 		break;
1342 	default:
1343 		tmp = TC_L2_SIZE(0);
1344 		break;
1345 	}
1346 	WREG32(TC_CNTL, tmp);
1347 
1348 	tmp = RREG32(HDP_HOST_PATH_CNTL);
1349 	WREG32(HDP_HOST_PATH_CNTL, tmp);
1350 
1351 	tmp = RREG32(ARB_POP);
1352 	tmp |= ENABLE_TC128;
1353 	WREG32(ARB_POP, tmp);
1354 
1355 	WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1356 	WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA |
1357 			       NUM_CLIP_SEQ(3)));
1358 	WREG32(PA_SC_ENHANCE, FORCE_EOV_MAX_CLK_CNT(4095));
1359 }
1360 
1361 
1362 /*
1363  * Indirect registers accessor
1364  */
1365 u32 r600_pciep_rreg(struct radeon_device *rdev, u32 reg)
1366 {
1367 	u32 r;
1368 
1369 	WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1370 	(void)RREG32(PCIE_PORT_INDEX);
1371 	r = RREG32(PCIE_PORT_DATA);
1372 	return r;
1373 }
1374 
1375 void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
1376 {
1377 	WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1378 	(void)RREG32(PCIE_PORT_INDEX);
1379 	WREG32(PCIE_PORT_DATA, (v));
1380 	(void)RREG32(PCIE_PORT_DATA);
1381 }
1382 
1383 /*
1384  * CP & Ring
1385  */
1386 void r600_cp_stop(struct radeon_device *rdev)
1387 {
1388 	WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
1389 }
1390 
1391 int r600_init_microcode(struct radeon_device *rdev)
1392 {
1393 	struct platform_device *pdev;
1394 	const char *chip_name;
1395 	const char *rlc_chip_name;
1396 	size_t pfp_req_size, me_req_size, rlc_req_size;
1397 	char fw_name[30];
1398 	int err;
1399 
1400 	DRM_DEBUG("\n");
1401 
1402 	pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
1403 	err = IS_ERR(pdev);
1404 	if (err) {
1405 		printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
1406 		return -EINVAL;
1407 	}
1408 
1409 	switch (rdev->family) {
1410 	case CHIP_R600:
1411 		chip_name = "R600";
1412 		rlc_chip_name = "R600";
1413 		break;
1414 	case CHIP_RV610:
1415 		chip_name = "RV610";
1416 		rlc_chip_name = "R600";
1417 		break;
1418 	case CHIP_RV630:
1419 		chip_name = "RV630";
1420 		rlc_chip_name = "R600";
1421 		break;
1422 	case CHIP_RV620:
1423 		chip_name = "RV620";
1424 		rlc_chip_name = "R600";
1425 		break;
1426 	case CHIP_RV635:
1427 		chip_name = "RV635";
1428 		rlc_chip_name = "R600";
1429 		break;
1430 	case CHIP_RV670:
1431 		chip_name = "RV670";
1432 		rlc_chip_name = "R600";
1433 		break;
1434 	case CHIP_RS780:
1435 	case CHIP_RS880:
1436 		chip_name = "RS780";
1437 		rlc_chip_name = "R600";
1438 		break;
1439 	case CHIP_RV770:
1440 		chip_name = "RV770";
1441 		rlc_chip_name = "R700";
1442 		break;
1443 	case CHIP_RV730:
1444 	case CHIP_RV740:
1445 		chip_name = "RV730";
1446 		rlc_chip_name = "R700";
1447 		break;
1448 	case CHIP_RV710:
1449 		chip_name = "RV710";
1450 		rlc_chip_name = "R700";
1451 		break;
1452 	default: BUG();
1453 	}
1454 
1455 	if (rdev->family >= CHIP_RV770) {
1456 		pfp_req_size = R700_PFP_UCODE_SIZE * 4;
1457 		me_req_size = R700_PM4_UCODE_SIZE * 4;
1458 		rlc_req_size = R700_RLC_UCODE_SIZE * 4;
1459 	} else {
1460 		pfp_req_size = PFP_UCODE_SIZE * 4;
1461 		me_req_size = PM4_UCODE_SIZE * 12;
1462 		rlc_req_size = RLC_UCODE_SIZE * 4;
1463 	}
1464 
1465 	DRM_INFO("Loading %s Microcode\n", chip_name);
1466 
1467 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
1468 	err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
1469 	if (err)
1470 		goto out;
1471 	if (rdev->pfp_fw->size != pfp_req_size) {
1472 		printk(KERN_ERR
1473 		       "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1474 		       rdev->pfp_fw->size, fw_name);
1475 		err = -EINVAL;
1476 		goto out;
1477 	}
1478 
1479 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
1480 	err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
1481 	if (err)
1482 		goto out;
1483 	if (rdev->me_fw->size != me_req_size) {
1484 		printk(KERN_ERR
1485 		       "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1486 		       rdev->me_fw->size, fw_name);
1487 		err = -EINVAL;
1488 	}
1489 
1490 	snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
1491 	err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
1492 	if (err)
1493 		goto out;
1494 	if (rdev->rlc_fw->size != rlc_req_size) {
1495 		printk(KERN_ERR
1496 		       "r600_rlc: Bogus length %zu in firmware \"%s\"\n",
1497 		       rdev->rlc_fw->size, fw_name);
1498 		err = -EINVAL;
1499 	}
1500 
1501 out:
1502 	platform_device_unregister(pdev);
1503 
1504 	if (err) {
1505 		if (err != -EINVAL)
1506 			printk(KERN_ERR
1507 			       "r600_cp: Failed to load firmware \"%s\"\n",
1508 			       fw_name);
1509 		release_firmware(rdev->pfp_fw);
1510 		rdev->pfp_fw = NULL;
1511 		release_firmware(rdev->me_fw);
1512 		rdev->me_fw = NULL;
1513 		release_firmware(rdev->rlc_fw);
1514 		rdev->rlc_fw = NULL;
1515 	}
1516 	return err;
1517 }
1518 
1519 static int r600_cp_load_microcode(struct radeon_device *rdev)
1520 {
1521 	const __be32 *fw_data;
1522 	int i;
1523 
1524 	if (!rdev->me_fw || !rdev->pfp_fw)
1525 		return -EINVAL;
1526 
1527 	r600_cp_stop(rdev);
1528 
1529 	WREG32(CP_RB_CNTL, RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
1530 
1531 	/* Reset cp */
1532 	WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1533 	RREG32(GRBM_SOFT_RESET);
1534 	mdelay(15);
1535 	WREG32(GRBM_SOFT_RESET, 0);
1536 
1537 	WREG32(CP_ME_RAM_WADDR, 0);
1538 
1539 	fw_data = (const __be32 *)rdev->me_fw->data;
1540 	WREG32(CP_ME_RAM_WADDR, 0);
1541 	for (i = 0; i < PM4_UCODE_SIZE * 3; i++)
1542 		WREG32(CP_ME_RAM_DATA,
1543 		       be32_to_cpup(fw_data++));
1544 
1545 	fw_data = (const __be32 *)rdev->pfp_fw->data;
1546 	WREG32(CP_PFP_UCODE_ADDR, 0);
1547 	for (i = 0; i < PFP_UCODE_SIZE; i++)
1548 		WREG32(CP_PFP_UCODE_DATA,
1549 		       be32_to_cpup(fw_data++));
1550 
1551 	WREG32(CP_PFP_UCODE_ADDR, 0);
1552 	WREG32(CP_ME_RAM_WADDR, 0);
1553 	WREG32(CP_ME_RAM_RADDR, 0);
1554 	return 0;
1555 }
1556 
1557 int r600_cp_start(struct radeon_device *rdev)
1558 {
1559 	int r;
1560 	uint32_t cp_me;
1561 
1562 	r = radeon_ring_lock(rdev, 7);
1563 	if (r) {
1564 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1565 		return r;
1566 	}
1567 	radeon_ring_write(rdev, PACKET3(PACKET3_ME_INITIALIZE, 5));
1568 	radeon_ring_write(rdev, 0x1);
1569 	if (rdev->family < CHIP_RV770) {
1570 		radeon_ring_write(rdev, 0x3);
1571 		radeon_ring_write(rdev, rdev->config.r600.max_hw_contexts - 1);
1572 	} else {
1573 		radeon_ring_write(rdev, 0x0);
1574 		radeon_ring_write(rdev, rdev->config.rv770.max_hw_contexts - 1);
1575 	}
1576 	radeon_ring_write(rdev, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1577 	radeon_ring_write(rdev, 0);
1578 	radeon_ring_write(rdev, 0);
1579 	radeon_ring_unlock_commit(rdev);
1580 
1581 	cp_me = 0xff;
1582 	WREG32(R_0086D8_CP_ME_CNTL, cp_me);
1583 	return 0;
1584 }
1585 
1586 int r600_cp_resume(struct radeon_device *rdev)
1587 {
1588 	u32 tmp;
1589 	u32 rb_bufsz;
1590 	int r;
1591 
1592 	/* Reset cp */
1593 	WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1594 	RREG32(GRBM_SOFT_RESET);
1595 	mdelay(15);
1596 	WREG32(GRBM_SOFT_RESET, 0);
1597 
1598 	/* Set ring buffer size */
1599 	rb_bufsz = drm_order(rdev->cp.ring_size / 8);
1600 	tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1601 #ifdef __BIG_ENDIAN
1602 	tmp |= BUF_SWAP_32BIT;
1603 #endif
1604 	WREG32(CP_RB_CNTL, tmp);
1605 	WREG32(CP_SEM_WAIT_TIMER, 0x4);
1606 
1607 	/* Set the write pointer delay */
1608 	WREG32(CP_RB_WPTR_DELAY, 0);
1609 
1610 	/* Initialize the ring buffer's read and write pointers */
1611 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
1612 	WREG32(CP_RB_RPTR_WR, 0);
1613 	WREG32(CP_RB_WPTR, 0);
1614 	WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
1615 	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
1616 	mdelay(1);
1617 	WREG32(CP_RB_CNTL, tmp);
1618 
1619 	WREG32(CP_RB_BASE, rdev->cp.gpu_addr >> 8);
1620 	WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
1621 
1622 	rdev->cp.rptr = RREG32(CP_RB_RPTR);
1623 	rdev->cp.wptr = RREG32(CP_RB_WPTR);
1624 
1625 	r600_cp_start(rdev);
1626 	rdev->cp.ready = true;
1627 	r = radeon_ring_test(rdev);
1628 	if (r) {
1629 		rdev->cp.ready = false;
1630 		return r;
1631 	}
1632 	return 0;
1633 }
1634 
1635 void r600_cp_commit(struct radeon_device *rdev)
1636 {
1637 	WREG32(CP_RB_WPTR, rdev->cp.wptr);
1638 	(void)RREG32(CP_RB_WPTR);
1639 }
1640 
1641 void r600_ring_init(struct radeon_device *rdev, unsigned ring_size)
1642 {
1643 	u32 rb_bufsz;
1644 
1645 	/* Align ring size */
1646 	rb_bufsz = drm_order(ring_size / 8);
1647 	ring_size = (1 << (rb_bufsz + 1)) * 4;
1648 	rdev->cp.ring_size = ring_size;
1649 	rdev->cp.align_mask = 16 - 1;
1650 }
1651 
1652 void r600_cp_fini(struct radeon_device *rdev)
1653 {
1654 	r600_cp_stop(rdev);
1655 	radeon_ring_fini(rdev);
1656 }
1657 
1658 
1659 /*
1660  * GPU scratch registers helpers function.
1661  */
1662 void r600_scratch_init(struct radeon_device *rdev)
1663 {
1664 	int i;
1665 
1666 	rdev->scratch.num_reg = 7;
1667 	for (i = 0; i < rdev->scratch.num_reg; i++) {
1668 		rdev->scratch.free[i] = true;
1669 		rdev->scratch.reg[i] = SCRATCH_REG0 + (i * 4);
1670 	}
1671 }
1672 
1673 int r600_ring_test(struct radeon_device *rdev)
1674 {
1675 	uint32_t scratch;
1676 	uint32_t tmp = 0;
1677 	unsigned i;
1678 	int r;
1679 
1680 	r = radeon_scratch_get(rdev, &scratch);
1681 	if (r) {
1682 		DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
1683 		return r;
1684 	}
1685 	WREG32(scratch, 0xCAFEDEAD);
1686 	r = radeon_ring_lock(rdev, 3);
1687 	if (r) {
1688 		DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1689 		radeon_scratch_free(rdev, scratch);
1690 		return r;
1691 	}
1692 	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1693 	radeon_ring_write(rdev, ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1694 	radeon_ring_write(rdev, 0xDEADBEEF);
1695 	radeon_ring_unlock_commit(rdev);
1696 	for (i = 0; i < rdev->usec_timeout; i++) {
1697 		tmp = RREG32(scratch);
1698 		if (tmp == 0xDEADBEEF)
1699 			break;
1700 		DRM_UDELAY(1);
1701 	}
1702 	if (i < rdev->usec_timeout) {
1703 		DRM_INFO("ring test succeeded in %d usecs\n", i);
1704 	} else {
1705 		DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
1706 			  scratch, tmp);
1707 		r = -EINVAL;
1708 	}
1709 	radeon_scratch_free(rdev, scratch);
1710 	return r;
1711 }
1712 
1713 void r600_wb_disable(struct radeon_device *rdev)
1714 {
1715 	int r;
1716 
1717 	WREG32(SCRATCH_UMSK, 0);
1718 	if (rdev->wb.wb_obj) {
1719 		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
1720 		if (unlikely(r != 0))
1721 			return;
1722 		radeon_bo_kunmap(rdev->wb.wb_obj);
1723 		radeon_bo_unpin(rdev->wb.wb_obj);
1724 		radeon_bo_unreserve(rdev->wb.wb_obj);
1725 	}
1726 }
1727 
1728 void r600_wb_fini(struct radeon_device *rdev)
1729 {
1730 	r600_wb_disable(rdev);
1731 	if (rdev->wb.wb_obj) {
1732 		radeon_bo_unref(&rdev->wb.wb_obj);
1733 		rdev->wb.wb = NULL;
1734 		rdev->wb.wb_obj = NULL;
1735 	}
1736 }
1737 
1738 int r600_wb_enable(struct radeon_device *rdev)
1739 {
1740 	int r;
1741 
1742 	if (rdev->wb.wb_obj == NULL) {
1743 		r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
1744 				RADEON_GEM_DOMAIN_GTT, &rdev->wb.wb_obj);
1745 		if (r) {
1746 			dev_warn(rdev->dev, "(%d) create WB bo failed\n", r);
1747 			return r;
1748 		}
1749 		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
1750 		if (unlikely(r != 0)) {
1751 			r600_wb_fini(rdev);
1752 			return r;
1753 		}
1754 		r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
1755 				&rdev->wb.gpu_addr);
1756 		if (r) {
1757 			radeon_bo_unreserve(rdev->wb.wb_obj);
1758 			dev_warn(rdev->dev, "(%d) pin WB bo failed\n", r);
1759 			r600_wb_fini(rdev);
1760 			return r;
1761 		}
1762 		r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
1763 		radeon_bo_unreserve(rdev->wb.wb_obj);
1764 		if (r) {
1765 			dev_warn(rdev->dev, "(%d) map WB bo failed\n", r);
1766 			r600_wb_fini(rdev);
1767 			return r;
1768 		}
1769 	}
1770 	WREG32(SCRATCH_ADDR, (rdev->wb.gpu_addr >> 8) & 0xFFFFFFFF);
1771 	WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + 1024) & 0xFFFFFFFC);
1772 	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + 1024) & 0xFF);
1773 	WREG32(SCRATCH_UMSK, 0xff);
1774 	return 0;
1775 }
1776 
1777 void r600_fence_ring_emit(struct radeon_device *rdev,
1778 			  struct radeon_fence *fence)
1779 {
1780 	/* Also consider EVENT_WRITE_EOP.  it handles the interrupts + timestamps + events */
1781 
1782 	radeon_ring_write(rdev, PACKET3(PACKET3_EVENT_WRITE, 0));
1783 	radeon_ring_write(rdev, CACHE_FLUSH_AND_INV_EVENT);
1784 	/* wait for 3D idle clean */
1785 	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1786 	radeon_ring_write(rdev, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
1787 	radeon_ring_write(rdev, WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
1788 	/* Emit fence sequence & fire IRQ */
1789 	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1790 	radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1791 	radeon_ring_write(rdev, fence->seq);
1792 	/* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
1793 	radeon_ring_write(rdev, PACKET0(CP_INT_STATUS, 0));
1794 	radeon_ring_write(rdev, RB_INT_STAT);
1795 }
1796 
1797 int r600_copy_blit(struct radeon_device *rdev,
1798 		   uint64_t src_offset, uint64_t dst_offset,
1799 		   unsigned num_pages, struct radeon_fence *fence)
1800 {
1801 	int r;
1802 
1803 	mutex_lock(&rdev->r600_blit.mutex);
1804 	rdev->r600_blit.vb_ib = NULL;
1805 	r = r600_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
1806 	if (r) {
1807 		if (rdev->r600_blit.vb_ib)
1808 			radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
1809 		mutex_unlock(&rdev->r600_blit.mutex);
1810 		return r;
1811 	}
1812 	r600_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
1813 	r600_blit_done_copy(rdev, fence);
1814 	mutex_unlock(&rdev->r600_blit.mutex);
1815 	return 0;
1816 }
1817 
1818 int r600_set_surface_reg(struct radeon_device *rdev, int reg,
1819 			 uint32_t tiling_flags, uint32_t pitch,
1820 			 uint32_t offset, uint32_t obj_size)
1821 {
1822 	/* FIXME: implement */
1823 	return 0;
1824 }
1825 
1826 void r600_clear_surface_reg(struct radeon_device *rdev, int reg)
1827 {
1828 	/* FIXME: implement */
1829 }
1830 
1831 
1832 bool r600_card_posted(struct radeon_device *rdev)
1833 {
1834 	uint32_t reg;
1835 
1836 	/* first check CRTCs */
1837 	reg = RREG32(D1CRTC_CONTROL) |
1838 		RREG32(D2CRTC_CONTROL);
1839 	if (reg & CRTC_EN)
1840 		return true;
1841 
1842 	/* then check MEM_SIZE, in case the crtcs are off */
1843 	if (RREG32(CONFIG_MEMSIZE))
1844 		return true;
1845 
1846 	return false;
1847 }
1848 
1849 int r600_startup(struct radeon_device *rdev)
1850 {
1851 	int r;
1852 
1853 	if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
1854 		r = r600_init_microcode(rdev);
1855 		if (r) {
1856 			DRM_ERROR("Failed to load firmware!\n");
1857 			return r;
1858 		}
1859 	}
1860 
1861 	r600_mc_program(rdev);
1862 	if (rdev->flags & RADEON_IS_AGP) {
1863 		r600_agp_enable(rdev);
1864 	} else {
1865 		r = r600_pcie_gart_enable(rdev);
1866 		if (r)
1867 			return r;
1868 	}
1869 	r600_gpu_init(rdev);
1870 	r = r600_blit_init(rdev);
1871 	if (r) {
1872 		r600_blit_fini(rdev);
1873 		rdev->asic->copy = NULL;
1874 		dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
1875 	}
1876 	/* pin copy shader into vram */
1877 	if (rdev->r600_blit.shader_obj) {
1878 		r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
1879 		if (unlikely(r != 0))
1880 			return r;
1881 		r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
1882 				&rdev->r600_blit.shader_gpu_addr);
1883 		radeon_bo_unreserve(rdev->r600_blit.shader_obj);
1884 		if (r) {
1885 			dev_err(rdev->dev, "(%d) pin blit object failed\n", r);
1886 			return r;
1887 		}
1888 	}
1889 	/* Enable IRQ */
1890 	r = r600_irq_init(rdev);
1891 	if (r) {
1892 		DRM_ERROR("radeon: IH init failed (%d).\n", r);
1893 		radeon_irq_kms_fini(rdev);
1894 		return r;
1895 	}
1896 	r600_irq_set(rdev);
1897 
1898 	r = radeon_ring_init(rdev, rdev->cp.ring_size);
1899 	if (r)
1900 		return r;
1901 	r = r600_cp_load_microcode(rdev);
1902 	if (r)
1903 		return r;
1904 	r = r600_cp_resume(rdev);
1905 	if (r)
1906 		return r;
1907 	/* write back buffer are not vital so don't worry about failure */
1908 	r600_wb_enable(rdev);
1909 	return 0;
1910 }
1911 
1912 void r600_vga_set_state(struct radeon_device *rdev, bool state)
1913 {
1914 	uint32_t temp;
1915 
1916 	temp = RREG32(CONFIG_CNTL);
1917 	if (state == false) {
1918 		temp &= ~(1<<0);
1919 		temp |= (1<<1);
1920 	} else {
1921 		temp &= ~(1<<1);
1922 	}
1923 	WREG32(CONFIG_CNTL, temp);
1924 }
1925 
1926 int r600_resume(struct radeon_device *rdev)
1927 {
1928 	int r;
1929 
1930 	/* Do not reset GPU before posting, on r600 hw unlike on r500 hw,
1931 	 * posting will perform necessary task to bring back GPU into good
1932 	 * shape.
1933 	 */
1934 	/* post card */
1935 	atom_asic_init(rdev->mode_info.atom_context);
1936 	/* Initialize clocks */
1937 	r = radeon_clocks_init(rdev);
1938 	if (r) {
1939 		return r;
1940 	}
1941 
1942 	r = r600_startup(rdev);
1943 	if (r) {
1944 		DRM_ERROR("r600 startup failed on resume\n");
1945 		return r;
1946 	}
1947 
1948 	r = r600_ib_test(rdev);
1949 	if (r) {
1950 		DRM_ERROR("radeon: failled testing IB (%d).\n", r);
1951 		return r;
1952 	}
1953 
1954 	r = r600_audio_init(rdev);
1955 	if (r) {
1956 		DRM_ERROR("radeon: audio resume failed\n");
1957 		return r;
1958 	}
1959 
1960 	return r;
1961 }
1962 
1963 int r600_suspend(struct radeon_device *rdev)
1964 {
1965 	int r;
1966 
1967 	r600_audio_fini(rdev);
1968 	/* FIXME: we should wait for ring to be empty */
1969 	r600_cp_stop(rdev);
1970 	rdev->cp.ready = false;
1971 	r600_irq_suspend(rdev);
1972 	r600_wb_disable(rdev);
1973 	r600_pcie_gart_disable(rdev);
1974 	/* unpin shaders bo */
1975 	if (rdev->r600_blit.shader_obj) {
1976 		r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
1977 		if (!r) {
1978 			radeon_bo_unpin(rdev->r600_blit.shader_obj);
1979 			radeon_bo_unreserve(rdev->r600_blit.shader_obj);
1980 		}
1981 	}
1982 	return 0;
1983 }
1984 
1985 /* Plan is to move initialization in that function and use
1986  * helper function so that radeon_device_init pretty much
1987  * do nothing more than calling asic specific function. This
1988  * should also allow to remove a bunch of callback function
1989  * like vram_info.
1990  */
1991 int r600_init(struct radeon_device *rdev)
1992 {
1993 	int r;
1994 
1995 	r = radeon_dummy_page_init(rdev);
1996 	if (r)
1997 		return r;
1998 	if (r600_debugfs_mc_info_init(rdev)) {
1999 		DRM_ERROR("Failed to register debugfs file for mc !\n");
2000 	}
2001 	/* This don't do much */
2002 	r = radeon_gem_init(rdev);
2003 	if (r)
2004 		return r;
2005 	/* Read BIOS */
2006 	if (!radeon_get_bios(rdev)) {
2007 		if (ASIC_IS_AVIVO(rdev))
2008 			return -EINVAL;
2009 	}
2010 	/* Must be an ATOMBIOS */
2011 	if (!rdev->is_atom_bios) {
2012 		dev_err(rdev->dev, "Expecting atombios for R600 GPU\n");
2013 		return -EINVAL;
2014 	}
2015 	r = radeon_atombios_init(rdev);
2016 	if (r)
2017 		return r;
2018 	/* Post card if necessary */
2019 	if (!r600_card_posted(rdev)) {
2020 		if (!rdev->bios) {
2021 			dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
2022 			return -EINVAL;
2023 		}
2024 		DRM_INFO("GPU not posted. posting now...\n");
2025 		atom_asic_init(rdev->mode_info.atom_context);
2026 	}
2027 	/* Initialize scratch registers */
2028 	r600_scratch_init(rdev);
2029 	/* Initialize surface registers */
2030 	radeon_surface_init(rdev);
2031 	/* Initialize clocks */
2032 	radeon_get_clock_info(rdev->ddev);
2033 	r = radeon_clocks_init(rdev);
2034 	if (r)
2035 		return r;
2036 	/* Initialize power management */
2037 	radeon_pm_init(rdev);
2038 	/* Fence driver */
2039 	r = radeon_fence_driver_init(rdev);
2040 	if (r)
2041 		return r;
2042 	if (rdev->flags & RADEON_IS_AGP) {
2043 		r = radeon_agp_init(rdev);
2044 		if (r)
2045 			radeon_agp_disable(rdev);
2046 	}
2047 	r = r600_mc_init(rdev);
2048 	if (r)
2049 		return r;
2050 	/* Memory manager */
2051 	r = radeon_bo_init(rdev);
2052 	if (r)
2053 		return r;
2054 
2055 	r = radeon_irq_kms_init(rdev);
2056 	if (r)
2057 		return r;
2058 
2059 	rdev->cp.ring_obj = NULL;
2060 	r600_ring_init(rdev, 1024 * 1024);
2061 
2062 	rdev->ih.ring_obj = NULL;
2063 	r600_ih_ring_init(rdev, 64 * 1024);
2064 
2065 	r = r600_pcie_gart_init(rdev);
2066 	if (r)
2067 		return r;
2068 
2069 	rdev->accel_working = true;
2070 	r = r600_startup(rdev);
2071 	if (r) {
2072 		dev_err(rdev->dev, "disabling GPU acceleration\n");
2073 		r600_cp_fini(rdev);
2074 		r600_wb_fini(rdev);
2075 		r600_irq_fini(rdev);
2076 		radeon_irq_kms_fini(rdev);
2077 		r600_pcie_gart_fini(rdev);
2078 		rdev->accel_working = false;
2079 	}
2080 	if (rdev->accel_working) {
2081 		r = radeon_ib_pool_init(rdev);
2082 		if (r) {
2083 			dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
2084 			rdev->accel_working = false;
2085 		} else {
2086 			r = r600_ib_test(rdev);
2087 			if (r) {
2088 				dev_err(rdev->dev, "IB test failed (%d).\n", r);
2089 				rdev->accel_working = false;
2090 			}
2091 		}
2092 	}
2093 
2094 	r = r600_audio_init(rdev);
2095 	if (r)
2096 		return r; /* TODO error handling */
2097 	return 0;
2098 }
2099 
2100 void r600_fini(struct radeon_device *rdev)
2101 {
2102 	radeon_pm_fini(rdev);
2103 	r600_audio_fini(rdev);
2104 	r600_blit_fini(rdev);
2105 	r600_cp_fini(rdev);
2106 	r600_wb_fini(rdev);
2107 	r600_irq_fini(rdev);
2108 	radeon_irq_kms_fini(rdev);
2109 	r600_pcie_gart_fini(rdev);
2110 	radeon_agp_fini(rdev);
2111 	radeon_gem_fini(rdev);
2112 	radeon_fence_driver_fini(rdev);
2113 	radeon_clocks_fini(rdev);
2114 	radeon_bo_fini(rdev);
2115 	radeon_atombios_fini(rdev);
2116 	kfree(rdev->bios);
2117 	rdev->bios = NULL;
2118 	radeon_dummy_page_fini(rdev);
2119 }
2120 
2121 
2122 /*
2123  * CS stuff
2124  */
2125 void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
2126 {
2127 	/* FIXME: implement */
2128 	radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2129 	radeon_ring_write(rdev, ib->gpu_addr & 0xFFFFFFFC);
2130 	radeon_ring_write(rdev, upper_32_bits(ib->gpu_addr) & 0xFF);
2131 	radeon_ring_write(rdev, ib->length_dw);
2132 }
2133 
2134 int r600_ib_test(struct radeon_device *rdev)
2135 {
2136 	struct radeon_ib *ib;
2137 	uint32_t scratch;
2138 	uint32_t tmp = 0;
2139 	unsigned i;
2140 	int r;
2141 
2142 	r = radeon_scratch_get(rdev, &scratch);
2143 	if (r) {
2144 		DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
2145 		return r;
2146 	}
2147 	WREG32(scratch, 0xCAFEDEAD);
2148 	r = radeon_ib_get(rdev, &ib);
2149 	if (r) {
2150 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
2151 		return r;
2152 	}
2153 	ib->ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1);
2154 	ib->ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
2155 	ib->ptr[2] = 0xDEADBEEF;
2156 	ib->ptr[3] = PACKET2(0);
2157 	ib->ptr[4] = PACKET2(0);
2158 	ib->ptr[5] = PACKET2(0);
2159 	ib->ptr[6] = PACKET2(0);
2160 	ib->ptr[7] = PACKET2(0);
2161 	ib->ptr[8] = PACKET2(0);
2162 	ib->ptr[9] = PACKET2(0);
2163 	ib->ptr[10] = PACKET2(0);
2164 	ib->ptr[11] = PACKET2(0);
2165 	ib->ptr[12] = PACKET2(0);
2166 	ib->ptr[13] = PACKET2(0);
2167 	ib->ptr[14] = PACKET2(0);
2168 	ib->ptr[15] = PACKET2(0);
2169 	ib->length_dw = 16;
2170 	r = radeon_ib_schedule(rdev, ib);
2171 	if (r) {
2172 		radeon_scratch_free(rdev, scratch);
2173 		radeon_ib_free(rdev, &ib);
2174 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
2175 		return r;
2176 	}
2177 	r = radeon_fence_wait(ib->fence, false);
2178 	if (r) {
2179 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
2180 		return r;
2181 	}
2182 	for (i = 0; i < rdev->usec_timeout; i++) {
2183 		tmp = RREG32(scratch);
2184 		if (tmp == 0xDEADBEEF)
2185 			break;
2186 		DRM_UDELAY(1);
2187 	}
2188 	if (i < rdev->usec_timeout) {
2189 		DRM_INFO("ib test succeeded in %u usecs\n", i);
2190 	} else {
2191 		DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n",
2192 			  scratch, tmp);
2193 		r = -EINVAL;
2194 	}
2195 	radeon_scratch_free(rdev, scratch);
2196 	radeon_ib_free(rdev, &ib);
2197 	return r;
2198 }
2199 
2200 /*
2201  * Interrupts
2202  *
2203  * Interrupts use a ring buffer on r6xx/r7xx hardware.  It works pretty
2204  * the same as the CP ring buffer, but in reverse.  Rather than the CPU
2205  * writing to the ring and the GPU consuming, the GPU writes to the ring
2206  * and host consumes.  As the host irq handler processes interrupts, it
2207  * increments the rptr.  When the rptr catches up with the wptr, all the
2208  * current interrupts have been processed.
2209  */
2210 
2211 void r600_ih_ring_init(struct radeon_device *rdev, unsigned ring_size)
2212 {
2213 	u32 rb_bufsz;
2214 
2215 	/* Align ring size */
2216 	rb_bufsz = drm_order(ring_size / 4);
2217 	ring_size = (1 << rb_bufsz) * 4;
2218 	rdev->ih.ring_size = ring_size;
2219 	rdev->ih.ptr_mask = rdev->ih.ring_size - 1;
2220 	rdev->ih.rptr = 0;
2221 }
2222 
2223 static int r600_ih_ring_alloc(struct radeon_device *rdev)
2224 {
2225 	int r;
2226 
2227 	/* Allocate ring buffer */
2228 	if (rdev->ih.ring_obj == NULL) {
2229 		r = radeon_bo_create(rdev, NULL, rdev->ih.ring_size,
2230 				     true,
2231 				     RADEON_GEM_DOMAIN_GTT,
2232 				     &rdev->ih.ring_obj);
2233 		if (r) {
2234 			DRM_ERROR("radeon: failed to create ih ring buffer (%d).\n", r);
2235 			return r;
2236 		}
2237 		r = radeon_bo_reserve(rdev->ih.ring_obj, false);
2238 		if (unlikely(r != 0))
2239 			return r;
2240 		r = radeon_bo_pin(rdev->ih.ring_obj,
2241 				  RADEON_GEM_DOMAIN_GTT,
2242 				  &rdev->ih.gpu_addr);
2243 		if (r) {
2244 			radeon_bo_unreserve(rdev->ih.ring_obj);
2245 			DRM_ERROR("radeon: failed to pin ih ring buffer (%d).\n", r);
2246 			return r;
2247 		}
2248 		r = radeon_bo_kmap(rdev->ih.ring_obj,
2249 				   (void **)&rdev->ih.ring);
2250 		radeon_bo_unreserve(rdev->ih.ring_obj);
2251 		if (r) {
2252 			DRM_ERROR("radeon: failed to map ih ring buffer (%d).\n", r);
2253 			return r;
2254 		}
2255 	}
2256 	return 0;
2257 }
2258 
2259 static void r600_ih_ring_fini(struct radeon_device *rdev)
2260 {
2261 	int r;
2262 	if (rdev->ih.ring_obj) {
2263 		r = radeon_bo_reserve(rdev->ih.ring_obj, false);
2264 		if (likely(r == 0)) {
2265 			radeon_bo_kunmap(rdev->ih.ring_obj);
2266 			radeon_bo_unpin(rdev->ih.ring_obj);
2267 			radeon_bo_unreserve(rdev->ih.ring_obj);
2268 		}
2269 		radeon_bo_unref(&rdev->ih.ring_obj);
2270 		rdev->ih.ring = NULL;
2271 		rdev->ih.ring_obj = NULL;
2272 	}
2273 }
2274 
2275 static void r600_rlc_stop(struct radeon_device *rdev)
2276 {
2277 
2278 	if (rdev->family >= CHIP_RV770) {
2279 		/* r7xx asics need to soft reset RLC before halting */
2280 		WREG32(SRBM_SOFT_RESET, SOFT_RESET_RLC);
2281 		RREG32(SRBM_SOFT_RESET);
2282 		udelay(15000);
2283 		WREG32(SRBM_SOFT_RESET, 0);
2284 		RREG32(SRBM_SOFT_RESET);
2285 	}
2286 
2287 	WREG32(RLC_CNTL, 0);
2288 }
2289 
2290 static void r600_rlc_start(struct radeon_device *rdev)
2291 {
2292 	WREG32(RLC_CNTL, RLC_ENABLE);
2293 }
2294 
2295 static int r600_rlc_init(struct radeon_device *rdev)
2296 {
2297 	u32 i;
2298 	const __be32 *fw_data;
2299 
2300 	if (!rdev->rlc_fw)
2301 		return -EINVAL;
2302 
2303 	r600_rlc_stop(rdev);
2304 
2305 	WREG32(RLC_HB_BASE, 0);
2306 	WREG32(RLC_HB_CNTL, 0);
2307 	WREG32(RLC_HB_RPTR, 0);
2308 	WREG32(RLC_HB_WPTR, 0);
2309 	WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
2310 	WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
2311 	WREG32(RLC_MC_CNTL, 0);
2312 	WREG32(RLC_UCODE_CNTL, 0);
2313 
2314 	fw_data = (const __be32 *)rdev->rlc_fw->data;
2315 	if (rdev->family >= CHIP_RV770) {
2316 		for (i = 0; i < R700_RLC_UCODE_SIZE; i++) {
2317 			WREG32(RLC_UCODE_ADDR, i);
2318 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
2319 		}
2320 	} else {
2321 		for (i = 0; i < RLC_UCODE_SIZE; i++) {
2322 			WREG32(RLC_UCODE_ADDR, i);
2323 			WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
2324 		}
2325 	}
2326 	WREG32(RLC_UCODE_ADDR, 0);
2327 
2328 	r600_rlc_start(rdev);
2329 
2330 	return 0;
2331 }
2332 
2333 static void r600_enable_interrupts(struct radeon_device *rdev)
2334 {
2335 	u32 ih_cntl = RREG32(IH_CNTL);
2336 	u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2337 
2338 	ih_cntl |= ENABLE_INTR;
2339 	ih_rb_cntl |= IH_RB_ENABLE;
2340 	WREG32(IH_CNTL, ih_cntl);
2341 	WREG32(IH_RB_CNTL, ih_rb_cntl);
2342 	rdev->ih.enabled = true;
2343 }
2344 
2345 static void r600_disable_interrupts(struct radeon_device *rdev)
2346 {
2347 	u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2348 	u32 ih_cntl = RREG32(IH_CNTL);
2349 
2350 	ih_rb_cntl &= ~IH_RB_ENABLE;
2351 	ih_cntl &= ~ENABLE_INTR;
2352 	WREG32(IH_RB_CNTL, ih_rb_cntl);
2353 	WREG32(IH_CNTL, ih_cntl);
2354 	/* set rptr, wptr to 0 */
2355 	WREG32(IH_RB_RPTR, 0);
2356 	WREG32(IH_RB_WPTR, 0);
2357 	rdev->ih.enabled = false;
2358 	rdev->ih.wptr = 0;
2359 	rdev->ih.rptr = 0;
2360 }
2361 
2362 static void r600_disable_interrupt_state(struct radeon_device *rdev)
2363 {
2364 	u32 tmp;
2365 
2366 	WREG32(CP_INT_CNTL, 0);
2367 	WREG32(GRBM_INT_CNTL, 0);
2368 	WREG32(DxMODE_INT_MASK, 0);
2369 	if (ASIC_IS_DCE3(rdev)) {
2370 		WREG32(DCE3_DACA_AUTODETECT_INT_CONTROL, 0);
2371 		WREG32(DCE3_DACB_AUTODETECT_INT_CONTROL, 0);
2372 		tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2373 		WREG32(DC_HPD1_INT_CONTROL, tmp);
2374 		tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2375 		WREG32(DC_HPD2_INT_CONTROL, tmp);
2376 		tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2377 		WREG32(DC_HPD3_INT_CONTROL, tmp);
2378 		tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2379 		WREG32(DC_HPD4_INT_CONTROL, tmp);
2380 		if (ASIC_IS_DCE32(rdev)) {
2381 			tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2382 			WREG32(DC_HPD5_INT_CONTROL, tmp);
2383 			tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2384 			WREG32(DC_HPD6_INT_CONTROL, tmp);
2385 		}
2386 	} else {
2387 		WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
2388 		WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
2389 		tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2390 		WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
2391 		tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2392 		WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
2393 		tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2394 		WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
2395 	}
2396 }
2397 
2398 int r600_irq_init(struct radeon_device *rdev)
2399 {
2400 	int ret = 0;
2401 	int rb_bufsz;
2402 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
2403 
2404 	/* allocate ring */
2405 	ret = r600_ih_ring_alloc(rdev);
2406 	if (ret)
2407 		return ret;
2408 
2409 	/* disable irqs */
2410 	r600_disable_interrupts(rdev);
2411 
2412 	/* init rlc */
2413 	ret = r600_rlc_init(rdev);
2414 	if (ret) {
2415 		r600_ih_ring_fini(rdev);
2416 		return ret;
2417 	}
2418 
2419 	/* setup interrupt control */
2420 	/* set dummy read address to ring address */
2421 	WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
2422 	interrupt_cntl = RREG32(INTERRUPT_CNTL);
2423 	/* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
2424 	 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
2425 	 */
2426 	interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
2427 	/* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
2428 	interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
2429 	WREG32(INTERRUPT_CNTL, interrupt_cntl);
2430 
2431 	WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
2432 	rb_bufsz = drm_order(rdev->ih.ring_size / 4);
2433 
2434 	ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
2435 		      IH_WPTR_OVERFLOW_CLEAR |
2436 		      (rb_bufsz << 1));
2437 	/* WPTR writeback, not yet */
2438 	/*ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;*/
2439 	WREG32(IH_RB_WPTR_ADDR_LO, 0);
2440 	WREG32(IH_RB_WPTR_ADDR_HI, 0);
2441 
2442 	WREG32(IH_RB_CNTL, ih_rb_cntl);
2443 
2444 	/* set rptr, wptr to 0 */
2445 	WREG32(IH_RB_RPTR, 0);
2446 	WREG32(IH_RB_WPTR, 0);
2447 
2448 	/* Default settings for IH_CNTL (disabled at first) */
2449 	ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10);
2450 	/* RPTR_REARM only works if msi's are enabled */
2451 	if (rdev->msi_enabled)
2452 		ih_cntl |= RPTR_REARM;
2453 
2454 #ifdef __BIG_ENDIAN
2455 	ih_cntl |= IH_MC_SWAP(IH_MC_SWAP_32BIT);
2456 #endif
2457 	WREG32(IH_CNTL, ih_cntl);
2458 
2459 	/* force the active interrupt state to all disabled */
2460 	r600_disable_interrupt_state(rdev);
2461 
2462 	/* enable irqs */
2463 	r600_enable_interrupts(rdev);
2464 
2465 	return ret;
2466 }
2467 
2468 void r600_irq_suspend(struct radeon_device *rdev)
2469 {
2470 	r600_disable_interrupts(rdev);
2471 	r600_rlc_stop(rdev);
2472 }
2473 
2474 void r600_irq_fini(struct radeon_device *rdev)
2475 {
2476 	r600_irq_suspend(rdev);
2477 	r600_ih_ring_fini(rdev);
2478 }
2479 
2480 int r600_irq_set(struct radeon_device *rdev)
2481 {
2482 	u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
2483 	u32 mode_int = 0;
2484 	u32 hpd1, hpd2, hpd3, hpd4 = 0, hpd5 = 0, hpd6 = 0;
2485 
2486 	if (!rdev->irq.installed) {
2487 		WARN(1, "Can't enable IRQ/MSI because no handler is installed.\n");
2488 		return -EINVAL;
2489 	}
2490 	/* don't enable anything if the ih is disabled */
2491 	if (!rdev->ih.enabled) {
2492 		r600_disable_interrupts(rdev);
2493 		/* force the active interrupt state to all disabled */
2494 		r600_disable_interrupt_state(rdev);
2495 		return 0;
2496 	}
2497 
2498 	if (ASIC_IS_DCE3(rdev)) {
2499 		hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2500 		hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2501 		hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2502 		hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
2503 		if (ASIC_IS_DCE32(rdev)) {
2504 			hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
2505 			hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
2506 		}
2507 	} else {
2508 		hpd1 = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2509 		hpd2 = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2510 		hpd3 = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2511 	}
2512 
2513 	if (rdev->irq.sw_int) {
2514 		DRM_DEBUG("r600_irq_set: sw int\n");
2515 		cp_int_cntl |= RB_INT_ENABLE;
2516 	}
2517 	if (rdev->irq.crtc_vblank_int[0]) {
2518 		DRM_DEBUG("r600_irq_set: vblank 0\n");
2519 		mode_int |= D1MODE_VBLANK_INT_MASK;
2520 	}
2521 	if (rdev->irq.crtc_vblank_int[1]) {
2522 		DRM_DEBUG("r600_irq_set: vblank 1\n");
2523 		mode_int |= D2MODE_VBLANK_INT_MASK;
2524 	}
2525 	if (rdev->irq.hpd[0]) {
2526 		DRM_DEBUG("r600_irq_set: hpd 1\n");
2527 		hpd1 |= DC_HPDx_INT_EN;
2528 	}
2529 	if (rdev->irq.hpd[1]) {
2530 		DRM_DEBUG("r600_irq_set: hpd 2\n");
2531 		hpd2 |= DC_HPDx_INT_EN;
2532 	}
2533 	if (rdev->irq.hpd[2]) {
2534 		DRM_DEBUG("r600_irq_set: hpd 3\n");
2535 		hpd3 |= DC_HPDx_INT_EN;
2536 	}
2537 	if (rdev->irq.hpd[3]) {
2538 		DRM_DEBUG("r600_irq_set: hpd 4\n");
2539 		hpd4 |= DC_HPDx_INT_EN;
2540 	}
2541 	if (rdev->irq.hpd[4]) {
2542 		DRM_DEBUG("r600_irq_set: hpd 5\n");
2543 		hpd5 |= DC_HPDx_INT_EN;
2544 	}
2545 	if (rdev->irq.hpd[5]) {
2546 		DRM_DEBUG("r600_irq_set: hpd 6\n");
2547 		hpd6 |= DC_HPDx_INT_EN;
2548 	}
2549 
2550 	WREG32(CP_INT_CNTL, cp_int_cntl);
2551 	WREG32(DxMODE_INT_MASK, mode_int);
2552 	if (ASIC_IS_DCE3(rdev)) {
2553 		WREG32(DC_HPD1_INT_CONTROL, hpd1);
2554 		WREG32(DC_HPD2_INT_CONTROL, hpd2);
2555 		WREG32(DC_HPD3_INT_CONTROL, hpd3);
2556 		WREG32(DC_HPD4_INT_CONTROL, hpd4);
2557 		if (ASIC_IS_DCE32(rdev)) {
2558 			WREG32(DC_HPD5_INT_CONTROL, hpd5);
2559 			WREG32(DC_HPD6_INT_CONTROL, hpd6);
2560 		}
2561 	} else {
2562 		WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, hpd1);
2563 		WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, hpd2);
2564 		WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, hpd3);
2565 	}
2566 
2567 	return 0;
2568 }
2569 
2570 static inline void r600_irq_ack(struct radeon_device *rdev,
2571 				u32 *disp_int,
2572 				u32 *disp_int_cont,
2573 				u32 *disp_int_cont2)
2574 {
2575 	u32 tmp;
2576 
2577 	if (ASIC_IS_DCE3(rdev)) {
2578 		*disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
2579 		*disp_int_cont = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE);
2580 		*disp_int_cont2 = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE2);
2581 	} else {
2582 		*disp_int = RREG32(DISP_INTERRUPT_STATUS);
2583 		*disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
2584 		*disp_int_cont2 = 0;
2585 	}
2586 
2587 	if (*disp_int & LB_D1_VBLANK_INTERRUPT)
2588 		WREG32(D1MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2589 	if (*disp_int & LB_D1_VLINE_INTERRUPT)
2590 		WREG32(D1MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2591 	if (*disp_int & LB_D2_VBLANK_INTERRUPT)
2592 		WREG32(D2MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2593 	if (*disp_int & LB_D2_VLINE_INTERRUPT)
2594 		WREG32(D2MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2595 	if (*disp_int & DC_HPD1_INTERRUPT) {
2596 		if (ASIC_IS_DCE3(rdev)) {
2597 			tmp = RREG32(DC_HPD1_INT_CONTROL);
2598 			tmp |= DC_HPDx_INT_ACK;
2599 			WREG32(DC_HPD1_INT_CONTROL, tmp);
2600 		} else {
2601 			tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
2602 			tmp |= DC_HPDx_INT_ACK;
2603 			WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
2604 		}
2605 	}
2606 	if (*disp_int & DC_HPD2_INTERRUPT) {
2607 		if (ASIC_IS_DCE3(rdev)) {
2608 			tmp = RREG32(DC_HPD2_INT_CONTROL);
2609 			tmp |= DC_HPDx_INT_ACK;
2610 			WREG32(DC_HPD2_INT_CONTROL, tmp);
2611 		} else {
2612 			tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
2613 			tmp |= DC_HPDx_INT_ACK;
2614 			WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
2615 		}
2616 	}
2617 	if (*disp_int_cont & DC_HPD3_INTERRUPT) {
2618 		if (ASIC_IS_DCE3(rdev)) {
2619 			tmp = RREG32(DC_HPD3_INT_CONTROL);
2620 			tmp |= DC_HPDx_INT_ACK;
2621 			WREG32(DC_HPD3_INT_CONTROL, tmp);
2622 		} else {
2623 			tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
2624 			tmp |= DC_HPDx_INT_ACK;
2625 			WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
2626 		}
2627 	}
2628 	if (*disp_int_cont & DC_HPD4_INTERRUPT) {
2629 		tmp = RREG32(DC_HPD4_INT_CONTROL);
2630 		tmp |= DC_HPDx_INT_ACK;
2631 		WREG32(DC_HPD4_INT_CONTROL, tmp);
2632 	}
2633 	if (ASIC_IS_DCE32(rdev)) {
2634 		if (*disp_int_cont2 & DC_HPD5_INTERRUPT) {
2635 			tmp = RREG32(DC_HPD5_INT_CONTROL);
2636 			tmp |= DC_HPDx_INT_ACK;
2637 			WREG32(DC_HPD5_INT_CONTROL, tmp);
2638 		}
2639 		if (*disp_int_cont2 & DC_HPD6_INTERRUPT) {
2640 			tmp = RREG32(DC_HPD5_INT_CONTROL);
2641 			tmp |= DC_HPDx_INT_ACK;
2642 			WREG32(DC_HPD6_INT_CONTROL, tmp);
2643 		}
2644 	}
2645 }
2646 
2647 void r600_irq_disable(struct radeon_device *rdev)
2648 {
2649 	u32 disp_int, disp_int_cont, disp_int_cont2;
2650 
2651 	r600_disable_interrupts(rdev);
2652 	/* Wait and acknowledge irq */
2653 	mdelay(1);
2654 	r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
2655 	r600_disable_interrupt_state(rdev);
2656 }
2657 
2658 static inline u32 r600_get_ih_wptr(struct radeon_device *rdev)
2659 {
2660 	u32 wptr, tmp;
2661 
2662 	/* XXX use writeback */
2663 	wptr = RREG32(IH_RB_WPTR);
2664 
2665 	if (wptr & RB_OVERFLOW) {
2666 		/* When a ring buffer overflow happen start parsing interrupt
2667 		 * from the last not overwritten vector (wptr + 16). Hopefully
2668 		 * this should allow us to catchup.
2669 		 */
2670 		dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
2671 			wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
2672 		rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
2673 		tmp = RREG32(IH_RB_CNTL);
2674 		tmp |= IH_WPTR_OVERFLOW_CLEAR;
2675 		WREG32(IH_RB_CNTL, tmp);
2676 	}
2677 	return (wptr & rdev->ih.ptr_mask);
2678 }
2679 
2680 /*        r600 IV Ring
2681  * Each IV ring entry is 128 bits:
2682  * [7:0]    - interrupt source id
2683  * [31:8]   - reserved
2684  * [59:32]  - interrupt source data
2685  * [127:60]  - reserved
2686  *
2687  * The basic interrupt vector entries
2688  * are decoded as follows:
2689  * src_id  src_data  description
2690  *      1         0  D1 Vblank
2691  *      1         1  D1 Vline
2692  *      5         0  D2 Vblank
2693  *      5         1  D2 Vline
2694  *     19         0  FP Hot plug detection A
2695  *     19         1  FP Hot plug detection B
2696  *     19         2  DAC A auto-detection
2697  *     19         3  DAC B auto-detection
2698  *    176         -  CP_INT RB
2699  *    177         -  CP_INT IB1
2700  *    178         -  CP_INT IB2
2701  *    181         -  EOP Interrupt
2702  *    233         -  GUI Idle
2703  *
2704  * Note, these are based on r600 and may need to be
2705  * adjusted or added to on newer asics
2706  */
2707 
2708 int r600_irq_process(struct radeon_device *rdev)
2709 {
2710 	u32 wptr = r600_get_ih_wptr(rdev);
2711 	u32 rptr = rdev->ih.rptr;
2712 	u32 src_id, src_data;
2713 	u32 ring_index, disp_int, disp_int_cont, disp_int_cont2;
2714 	unsigned long flags;
2715 	bool queue_hotplug = false;
2716 
2717 	DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
2718 	if (!rdev->ih.enabled)
2719 		return IRQ_NONE;
2720 
2721 	spin_lock_irqsave(&rdev->ih.lock, flags);
2722 
2723 	if (rptr == wptr) {
2724 		spin_unlock_irqrestore(&rdev->ih.lock, flags);
2725 		return IRQ_NONE;
2726 	}
2727 	if (rdev->shutdown) {
2728 		spin_unlock_irqrestore(&rdev->ih.lock, flags);
2729 		return IRQ_NONE;
2730 	}
2731 
2732 restart_ih:
2733 	/* display interrupts */
2734 	r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
2735 
2736 	rdev->ih.wptr = wptr;
2737 	while (rptr != wptr) {
2738 		/* wptr/rptr are in bytes! */
2739 		ring_index = rptr / 4;
2740 		src_id =  rdev->ih.ring[ring_index] & 0xff;
2741 		src_data = rdev->ih.ring[ring_index + 1] & 0xfffffff;
2742 
2743 		switch (src_id) {
2744 		case 1: /* D1 vblank/vline */
2745 			switch (src_data) {
2746 			case 0: /* D1 vblank */
2747 				if (disp_int & LB_D1_VBLANK_INTERRUPT) {
2748 					drm_handle_vblank(rdev->ddev, 0);
2749 					rdev->pm.vblank_sync = true;
2750 					wake_up(&rdev->irq.vblank_queue);
2751 					disp_int &= ~LB_D1_VBLANK_INTERRUPT;
2752 					DRM_DEBUG("IH: D1 vblank\n");
2753 				}
2754 				break;
2755 			case 1: /* D1 vline */
2756 				if (disp_int & LB_D1_VLINE_INTERRUPT) {
2757 					disp_int &= ~LB_D1_VLINE_INTERRUPT;
2758 					DRM_DEBUG("IH: D1 vline\n");
2759 				}
2760 				break;
2761 			default:
2762 				DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2763 				break;
2764 			}
2765 			break;
2766 		case 5: /* D2 vblank/vline */
2767 			switch (src_data) {
2768 			case 0: /* D2 vblank */
2769 				if (disp_int & LB_D2_VBLANK_INTERRUPT) {
2770 					drm_handle_vblank(rdev->ddev, 1);
2771 					rdev->pm.vblank_sync = true;
2772 					wake_up(&rdev->irq.vblank_queue);
2773 					disp_int &= ~LB_D2_VBLANK_INTERRUPT;
2774 					DRM_DEBUG("IH: D2 vblank\n");
2775 				}
2776 				break;
2777 			case 1: /* D1 vline */
2778 				if (disp_int & LB_D2_VLINE_INTERRUPT) {
2779 					disp_int &= ~LB_D2_VLINE_INTERRUPT;
2780 					DRM_DEBUG("IH: D2 vline\n");
2781 				}
2782 				break;
2783 			default:
2784 				DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2785 				break;
2786 			}
2787 			break;
2788 		case 19: /* HPD/DAC hotplug */
2789 			switch (src_data) {
2790 			case 0:
2791 				if (disp_int & DC_HPD1_INTERRUPT) {
2792 					disp_int &= ~DC_HPD1_INTERRUPT;
2793 					queue_hotplug = true;
2794 					DRM_DEBUG("IH: HPD1\n");
2795 				}
2796 				break;
2797 			case 1:
2798 				if (disp_int & DC_HPD2_INTERRUPT) {
2799 					disp_int &= ~DC_HPD2_INTERRUPT;
2800 					queue_hotplug = true;
2801 					DRM_DEBUG("IH: HPD2\n");
2802 				}
2803 				break;
2804 			case 4:
2805 				if (disp_int_cont & DC_HPD3_INTERRUPT) {
2806 					disp_int_cont &= ~DC_HPD3_INTERRUPT;
2807 					queue_hotplug = true;
2808 					DRM_DEBUG("IH: HPD3\n");
2809 				}
2810 				break;
2811 			case 5:
2812 				if (disp_int_cont & DC_HPD4_INTERRUPT) {
2813 					disp_int_cont &= ~DC_HPD4_INTERRUPT;
2814 					queue_hotplug = true;
2815 					DRM_DEBUG("IH: HPD4\n");
2816 				}
2817 				break;
2818 			case 10:
2819 				if (disp_int_cont2 & DC_HPD5_INTERRUPT) {
2820 					disp_int_cont2 &= ~DC_HPD5_INTERRUPT;
2821 					queue_hotplug = true;
2822 					DRM_DEBUG("IH: HPD5\n");
2823 				}
2824 				break;
2825 			case 12:
2826 				if (disp_int_cont2 & DC_HPD6_INTERRUPT) {
2827 					disp_int_cont2 &= ~DC_HPD6_INTERRUPT;
2828 					queue_hotplug = true;
2829 					DRM_DEBUG("IH: HPD6\n");
2830 				}
2831 				break;
2832 			default:
2833 				DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2834 				break;
2835 			}
2836 			break;
2837 		case 176: /* CP_INT in ring buffer */
2838 		case 177: /* CP_INT in IB1 */
2839 		case 178: /* CP_INT in IB2 */
2840 			DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
2841 			radeon_fence_process(rdev);
2842 			break;
2843 		case 181: /* CP EOP event */
2844 			DRM_DEBUG("IH: CP EOP\n");
2845 			break;
2846 		default:
2847 			DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2848 			break;
2849 		}
2850 
2851 		/* wptr/rptr are in bytes! */
2852 		rptr += 16;
2853 		rptr &= rdev->ih.ptr_mask;
2854 	}
2855 	/* make sure wptr hasn't changed while processing */
2856 	wptr = r600_get_ih_wptr(rdev);
2857 	if (wptr != rdev->ih.wptr)
2858 		goto restart_ih;
2859 	if (queue_hotplug)
2860 		queue_work(rdev->wq, &rdev->hotplug_work);
2861 	rdev->ih.rptr = rptr;
2862 	WREG32(IH_RB_RPTR, rdev->ih.rptr);
2863 	spin_unlock_irqrestore(&rdev->ih.lock, flags);
2864 	return IRQ_HANDLED;
2865 }
2866 
2867 /*
2868  * Debugfs info
2869  */
2870 #if defined(CONFIG_DEBUG_FS)
2871 
2872 static int r600_debugfs_cp_ring_info(struct seq_file *m, void *data)
2873 {
2874 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2875 	struct drm_device *dev = node->minor->dev;
2876 	struct radeon_device *rdev = dev->dev_private;
2877 	unsigned count, i, j;
2878 
2879 	radeon_ring_free_size(rdev);
2880 	count = (rdev->cp.ring_size / 4) - rdev->cp.ring_free_dw;
2881 	seq_printf(m, "CP_STAT 0x%08x\n", RREG32(CP_STAT));
2882 	seq_printf(m, "CP_RB_WPTR 0x%08x\n", RREG32(CP_RB_WPTR));
2883 	seq_printf(m, "CP_RB_RPTR 0x%08x\n", RREG32(CP_RB_RPTR));
2884 	seq_printf(m, "driver's copy of the CP_RB_WPTR 0x%08x\n", rdev->cp.wptr);
2885 	seq_printf(m, "driver's copy of the CP_RB_RPTR 0x%08x\n", rdev->cp.rptr);
2886 	seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
2887 	seq_printf(m, "%u dwords in ring\n", count);
2888 	i = rdev->cp.rptr;
2889 	for (j = 0; j <= count; j++) {
2890 		seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
2891 		i = (i + 1) & rdev->cp.ptr_mask;
2892 	}
2893 	return 0;
2894 }
2895 
2896 static int r600_debugfs_mc_info(struct seq_file *m, void *data)
2897 {
2898 	struct drm_info_node *node = (struct drm_info_node *) m->private;
2899 	struct drm_device *dev = node->minor->dev;
2900 	struct radeon_device *rdev = dev->dev_private;
2901 
2902 	DREG32_SYS(m, rdev, R_000E50_SRBM_STATUS);
2903 	DREG32_SYS(m, rdev, VM_L2_STATUS);
2904 	return 0;
2905 }
2906 
2907 static struct drm_info_list r600_mc_info_list[] = {
2908 	{"r600_mc_info", r600_debugfs_mc_info, 0, NULL},
2909 	{"r600_ring_info", r600_debugfs_cp_ring_info, 0, NULL},
2910 };
2911 #endif
2912 
2913 int r600_debugfs_mc_info_init(struct radeon_device *rdev)
2914 {
2915 #if defined(CONFIG_DEBUG_FS)
2916 	return radeon_debugfs_add_files(rdev, r600_mc_info_list, ARRAY_SIZE(r600_mc_info_list));
2917 #else
2918 	return 0;
2919 #endif
2920 }
2921 
2922 /**
2923  * r600_ioctl_wait_idle - flush host path cache on wait idle ioctl
2924  * rdev: radeon device structure
2925  * bo: buffer object struct which userspace is waiting for idle
2926  *
2927  * Some R6XX/R7XX doesn't seems to take into account HDP flush performed
2928  * through ring buffer, this leads to corruption in rendering, see
2929  * http://bugzilla.kernel.org/show_bug.cgi?id=15186 to avoid this we
2930  * directly perform HDP flush by writing register through MMIO.
2931  */
2932 void r600_ioctl_wait_idle(struct radeon_device *rdev, struct radeon_bo *bo)
2933 {
2934 	WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2935 }
2936