xref: /linux/drivers/gpu/drm/radeon/r300.c (revision b15ba51207e54245409d6f46e20dab36f906eed1)
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/slab.h>
30 #include <drm/drmP.h>
31 #include <drm/drm.h>
32 #include <drm/drm_crtc_helper.h>
33 #include "radeon_reg.h"
34 #include "radeon.h"
35 #include "radeon_asic.h"
36 #include "radeon_drm.h"
37 #include "r100_track.h"
38 #include "r300d.h"
39 #include "rv350d.h"
40 #include "r300_reg_safe.h"
41 
42 /* This files gather functions specifics to: r300,r350,rv350,rv370,rv380
43  *
44  * GPU Errata:
45  * - HOST_PATH_CNTL: r300 family seems to dislike write to HOST_PATH_CNTL
46  *   using MMIO to flush host path read cache, this lead to HARDLOCKUP.
47  *   However, scheduling such write to the ring seems harmless, i suspect
48  *   the CP read collide with the flush somehow, or maybe the MC, hard to
49  *   tell. (Jerome Glisse)
50  */
51 
52 /*
53  * rv370,rv380 PCIE GART
54  */
55 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev);
56 
57 void rv370_pcie_gart_tlb_flush(struct radeon_device *rdev)
58 {
59 	uint32_t tmp;
60 	int i;
61 
62 	/* Workaround HW bug do flush 2 times */
63 	for (i = 0; i < 2; i++) {
64 		tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
65 		WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp | RADEON_PCIE_TX_GART_INVALIDATE_TLB);
66 		(void)RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
67 		WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
68 	}
69 	mb();
70 }
71 
72 #define R300_PTE_WRITEABLE (1 << 2)
73 #define R300_PTE_READABLE  (1 << 3)
74 
75 int rv370_pcie_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
76 {
77 	void __iomem *ptr = rdev->gart.ptr;
78 
79 	if (i < 0 || i > rdev->gart.num_gpu_pages) {
80 		return -EINVAL;
81 	}
82 	addr = (lower_32_bits(addr) >> 8) |
83 	       ((upper_32_bits(addr) & 0xff) << 24) |
84 	       R300_PTE_WRITEABLE | R300_PTE_READABLE;
85 	/* on x86 we want this to be CPU endian, on powerpc
86 	 * on powerpc without HW swappers, it'll get swapped on way
87 	 * into VRAM - so no need for cpu_to_le32 on VRAM tables */
88 	writel(addr, ((void __iomem *)ptr) + (i * 4));
89 	return 0;
90 }
91 
92 int rv370_pcie_gart_init(struct radeon_device *rdev)
93 {
94 	int r;
95 
96 	if (rdev->gart.robj) {
97 		WARN(1, "RV370 PCIE GART already initialized\n");
98 		return 0;
99 	}
100 	/* Initialize common gart structure */
101 	r = radeon_gart_init(rdev);
102 	if (r)
103 		return r;
104 	r = rv370_debugfs_pcie_gart_info_init(rdev);
105 	if (r)
106 		DRM_ERROR("Failed to register debugfs file for PCIE gart !\n");
107 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
108 	rdev->asic->gart_tlb_flush = &rv370_pcie_gart_tlb_flush;
109 	rdev->asic->gart_set_page = &rv370_pcie_gart_set_page;
110 	return radeon_gart_table_vram_alloc(rdev);
111 }
112 
113 int rv370_pcie_gart_enable(struct radeon_device *rdev)
114 {
115 	uint32_t table_addr;
116 	uint32_t tmp;
117 	int r;
118 
119 	if (rdev->gart.robj == NULL) {
120 		dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
121 		return -EINVAL;
122 	}
123 	r = radeon_gart_table_vram_pin(rdev);
124 	if (r)
125 		return r;
126 	radeon_gart_restore(rdev);
127 	/* discard memory request outside of configured range */
128 	tmp = RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
129 	WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
130 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, rdev->mc.gtt_start);
131 	tmp = rdev->mc.gtt_end & ~RADEON_GPU_PAGE_MASK;
132 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, tmp);
133 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
134 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
135 	table_addr = rdev->gart.table_addr;
136 	WREG32_PCIE(RADEON_PCIE_TX_GART_BASE, table_addr);
137 	/* FIXME: setup default page */
138 	WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_LO, rdev->mc.vram_start);
139 	WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_HI, 0);
140 	/* Clear error */
141 	WREG32_PCIE(RADEON_PCIE_TX_GART_ERROR, 0);
142 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
143 	tmp |= RADEON_PCIE_TX_GART_EN;
144 	tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
145 	WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
146 	rv370_pcie_gart_tlb_flush(rdev);
147 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
148 		 (unsigned)(rdev->mc.gtt_size >> 20),
149 		 (unsigned long long)table_addr);
150 	rdev->gart.ready = true;
151 	return 0;
152 }
153 
154 void rv370_pcie_gart_disable(struct radeon_device *rdev)
155 {
156 	u32 tmp;
157 
158 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, 0);
159 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, 0);
160 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
161 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
162 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
163 	tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
164 	WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp & ~RADEON_PCIE_TX_GART_EN);
165 	radeon_gart_table_vram_unpin(rdev);
166 }
167 
168 void rv370_pcie_gart_fini(struct radeon_device *rdev)
169 {
170 	radeon_gart_fini(rdev);
171 	rv370_pcie_gart_disable(rdev);
172 	radeon_gart_table_vram_free(rdev);
173 }
174 
175 void r300_fence_ring_emit(struct radeon_device *rdev,
176 			  struct radeon_fence *fence)
177 {
178 	struct radeon_ring *ring = &rdev->ring[fence->ring];
179 
180 	/* Who ever call radeon_fence_emit should call ring_lock and ask
181 	 * for enough space (today caller are ib schedule and buffer move) */
182 	/* Write SC register so SC & US assert idle */
183 	radeon_ring_write(ring, PACKET0(R300_RE_SCISSORS_TL, 0));
184 	radeon_ring_write(ring, 0);
185 	radeon_ring_write(ring, PACKET0(R300_RE_SCISSORS_BR, 0));
186 	radeon_ring_write(ring, 0);
187 	/* Flush 3D cache */
188 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
189 	radeon_ring_write(ring, R300_RB3D_DC_FLUSH);
190 	radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
191 	radeon_ring_write(ring, R300_ZC_FLUSH);
192 	/* Wait until IDLE & CLEAN */
193 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
194 	radeon_ring_write(ring, (RADEON_WAIT_3D_IDLECLEAN |
195 				 RADEON_WAIT_2D_IDLECLEAN |
196 				 RADEON_WAIT_DMA_GUI_IDLE));
197 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
198 	radeon_ring_write(ring, rdev->config.r300.hdp_cntl |
199 				RADEON_HDP_READ_BUFFER_INVALIDATE);
200 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
201 	radeon_ring_write(ring, rdev->config.r300.hdp_cntl);
202 	/* Emit fence sequence & fire IRQ */
203 	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
204 	radeon_ring_write(ring, fence->seq);
205 	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
206 	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
207 }
208 
209 void r300_ring_start(struct radeon_device *rdev)
210 {
211 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
212 	unsigned gb_tile_config;
213 	int r;
214 
215 	/* Sub pixel 1/12 so we can have 4K rendering according to doc */
216 	gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
217 	switch(rdev->num_gb_pipes) {
218 	case 2:
219 		gb_tile_config |= R300_PIPE_COUNT_R300;
220 		break;
221 	case 3:
222 		gb_tile_config |= R300_PIPE_COUNT_R420_3P;
223 		break;
224 	case 4:
225 		gb_tile_config |= R300_PIPE_COUNT_R420;
226 		break;
227 	case 1:
228 	default:
229 		gb_tile_config |= R300_PIPE_COUNT_RV350;
230 		break;
231 	}
232 
233 	r = radeon_ring_lock(rdev, ring, 64);
234 	if (r) {
235 		return;
236 	}
237 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
238 	radeon_ring_write(ring,
239 			  RADEON_ISYNC_ANY2D_IDLE3D |
240 			  RADEON_ISYNC_ANY3D_IDLE2D |
241 			  RADEON_ISYNC_WAIT_IDLEGUI |
242 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
243 	radeon_ring_write(ring, PACKET0(R300_GB_TILE_CONFIG, 0));
244 	radeon_ring_write(ring, gb_tile_config);
245 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
246 	radeon_ring_write(ring,
247 			  RADEON_WAIT_2D_IDLECLEAN |
248 			  RADEON_WAIT_3D_IDLECLEAN);
249 	radeon_ring_write(ring, PACKET0(R300_DST_PIPE_CONFIG, 0));
250 	radeon_ring_write(ring, R300_PIPE_AUTO_CONFIG);
251 	radeon_ring_write(ring, PACKET0(R300_GB_SELECT, 0));
252 	radeon_ring_write(ring, 0);
253 	radeon_ring_write(ring, PACKET0(R300_GB_ENABLE, 0));
254 	radeon_ring_write(ring, 0);
255 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
256 	radeon_ring_write(ring, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
257 	radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
258 	radeon_ring_write(ring, R300_ZC_FLUSH | R300_ZC_FREE);
259 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
260 	radeon_ring_write(ring,
261 			  RADEON_WAIT_2D_IDLECLEAN |
262 			  RADEON_WAIT_3D_IDLECLEAN);
263 	radeon_ring_write(ring, PACKET0(R300_GB_AA_CONFIG, 0));
264 	radeon_ring_write(ring, 0);
265 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
266 	radeon_ring_write(ring, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
267 	radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
268 	radeon_ring_write(ring, R300_ZC_FLUSH | R300_ZC_FREE);
269 	radeon_ring_write(ring, PACKET0(R300_GB_MSPOS0, 0));
270 	radeon_ring_write(ring,
271 			  ((6 << R300_MS_X0_SHIFT) |
272 			   (6 << R300_MS_Y0_SHIFT) |
273 			   (6 << R300_MS_X1_SHIFT) |
274 			   (6 << R300_MS_Y1_SHIFT) |
275 			   (6 << R300_MS_X2_SHIFT) |
276 			   (6 << R300_MS_Y2_SHIFT) |
277 			   (6 << R300_MSBD0_Y_SHIFT) |
278 			   (6 << R300_MSBD0_X_SHIFT)));
279 	radeon_ring_write(ring, PACKET0(R300_GB_MSPOS1, 0));
280 	radeon_ring_write(ring,
281 			  ((6 << R300_MS_X3_SHIFT) |
282 			   (6 << R300_MS_Y3_SHIFT) |
283 			   (6 << R300_MS_X4_SHIFT) |
284 			   (6 << R300_MS_Y4_SHIFT) |
285 			   (6 << R300_MS_X5_SHIFT) |
286 			   (6 << R300_MS_Y5_SHIFT) |
287 			   (6 << R300_MSBD1_SHIFT)));
288 	radeon_ring_write(ring, PACKET0(R300_GA_ENHANCE, 0));
289 	radeon_ring_write(ring, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL);
290 	radeon_ring_write(ring, PACKET0(R300_GA_POLY_MODE, 0));
291 	radeon_ring_write(ring,
292 			  R300_FRONT_PTYPE_TRIANGE | R300_BACK_PTYPE_TRIANGE);
293 	radeon_ring_write(ring, PACKET0(R300_GA_ROUND_MODE, 0));
294 	radeon_ring_write(ring,
295 			  R300_GEOMETRY_ROUND_NEAREST |
296 			  R300_COLOR_ROUND_NEAREST);
297 	radeon_ring_unlock_commit(rdev, ring);
298 }
299 
300 void r300_errata(struct radeon_device *rdev)
301 {
302 	rdev->pll_errata = 0;
303 
304 	if (rdev->family == CHIP_R300 &&
305 	    (RREG32(RADEON_CONFIG_CNTL) & RADEON_CFG_ATI_REV_ID_MASK) == RADEON_CFG_ATI_REV_A11) {
306 		rdev->pll_errata |= CHIP_ERRATA_R300_CG;
307 	}
308 }
309 
310 int r300_mc_wait_for_idle(struct radeon_device *rdev)
311 {
312 	unsigned i;
313 	uint32_t tmp;
314 
315 	for (i = 0; i < rdev->usec_timeout; i++) {
316 		/* read MC_STATUS */
317 		tmp = RREG32(RADEON_MC_STATUS);
318 		if (tmp & R300_MC_IDLE) {
319 			return 0;
320 		}
321 		DRM_UDELAY(1);
322 	}
323 	return -1;
324 }
325 
326 void r300_gpu_init(struct radeon_device *rdev)
327 {
328 	uint32_t gb_tile_config, tmp;
329 
330 	if ((rdev->family == CHIP_R300 && rdev->pdev->device != 0x4144) ||
331 	    (rdev->family == CHIP_R350 && rdev->pdev->device != 0x4148)) {
332 		/* r300,r350 */
333 		rdev->num_gb_pipes = 2;
334 	} else {
335 		/* rv350,rv370,rv380,r300 AD, r350 AH */
336 		rdev->num_gb_pipes = 1;
337 	}
338 	rdev->num_z_pipes = 1;
339 	gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
340 	switch (rdev->num_gb_pipes) {
341 	case 2:
342 		gb_tile_config |= R300_PIPE_COUNT_R300;
343 		break;
344 	case 3:
345 		gb_tile_config |= R300_PIPE_COUNT_R420_3P;
346 		break;
347 	case 4:
348 		gb_tile_config |= R300_PIPE_COUNT_R420;
349 		break;
350 	default:
351 	case 1:
352 		gb_tile_config |= R300_PIPE_COUNT_RV350;
353 		break;
354 	}
355 	WREG32(R300_GB_TILE_CONFIG, gb_tile_config);
356 
357 	if (r100_gui_wait_for_idle(rdev)) {
358 		printk(KERN_WARNING "Failed to wait GUI idle while "
359 		       "programming pipes. Bad things might happen.\n");
360 	}
361 
362 	tmp = RREG32(R300_DST_PIPE_CONFIG);
363 	WREG32(R300_DST_PIPE_CONFIG, tmp | R300_PIPE_AUTO_CONFIG);
364 
365 	WREG32(R300_RB2D_DSTCACHE_MODE,
366 	       R300_DC_AUTOFLUSH_ENABLE |
367 	       R300_DC_DC_DISABLE_IGNORE_PE);
368 
369 	if (r100_gui_wait_for_idle(rdev)) {
370 		printk(KERN_WARNING "Failed to wait GUI idle while "
371 		       "programming pipes. Bad things might happen.\n");
372 	}
373 	if (r300_mc_wait_for_idle(rdev)) {
374 		printk(KERN_WARNING "Failed to wait MC idle while "
375 		       "programming pipes. Bad things might happen.\n");
376 	}
377 	DRM_INFO("radeon: %d quad pipes, %d Z pipes initialized.\n",
378 		 rdev->num_gb_pipes, rdev->num_z_pipes);
379 }
380 
381 bool r300_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
382 {
383 	u32 rbbm_status;
384 	int r;
385 
386 	rbbm_status = RREG32(R_000E40_RBBM_STATUS);
387 	if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
388 		r100_gpu_lockup_update(&rdev->config.r300.lockup, ring);
389 		return false;
390 	}
391 	/* force CP activities */
392 	r = radeon_ring_lock(rdev, ring, 2);
393 	if (!r) {
394 		/* PACKET2 NOP */
395 		radeon_ring_write(ring, 0x80000000);
396 		radeon_ring_write(ring, 0x80000000);
397 		radeon_ring_unlock_commit(rdev, ring);
398 	}
399 	ring->rptr = RREG32(RADEON_CP_RB_RPTR);
400 	return r100_gpu_cp_is_lockup(rdev, &rdev->config.r300.lockup, ring);
401 }
402 
403 int r300_asic_reset(struct radeon_device *rdev)
404 {
405 	struct r100_mc_save save;
406 	u32 status, tmp;
407 	int ret = 0;
408 
409 	status = RREG32(R_000E40_RBBM_STATUS);
410 	if (!G_000E40_GUI_ACTIVE(status)) {
411 		return 0;
412 	}
413 	r100_mc_stop(rdev, &save);
414 	status = RREG32(R_000E40_RBBM_STATUS);
415 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
416 	/* stop CP */
417 	WREG32(RADEON_CP_CSQ_CNTL, 0);
418 	tmp = RREG32(RADEON_CP_RB_CNTL);
419 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
420 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
421 	WREG32(RADEON_CP_RB_WPTR, 0);
422 	WREG32(RADEON_CP_RB_CNTL, tmp);
423 	/* save PCI state */
424 	pci_save_state(rdev->pdev);
425 	/* disable bus mastering */
426 	r100_bm_disable(rdev);
427 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_VAP(1) |
428 					S_0000F0_SOFT_RESET_GA(1));
429 	RREG32(R_0000F0_RBBM_SOFT_RESET);
430 	mdelay(500);
431 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
432 	mdelay(1);
433 	status = RREG32(R_000E40_RBBM_STATUS);
434 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
435 	/* resetting the CP seems to be problematic sometimes it end up
436 	 * hard locking the computer, but it's necessary for successful
437 	 * reset more test & playing is needed on R3XX/R4XX to find a
438 	 * reliable (if any solution)
439 	 */
440 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
441 	RREG32(R_0000F0_RBBM_SOFT_RESET);
442 	mdelay(500);
443 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
444 	mdelay(1);
445 	status = RREG32(R_000E40_RBBM_STATUS);
446 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
447 	/* restore PCI & busmastering */
448 	pci_restore_state(rdev->pdev);
449 	r100_enable_bm(rdev);
450 	/* Check if GPU is idle */
451 	if (G_000E40_GA_BUSY(status) || G_000E40_VAP_BUSY(status)) {
452 		dev_err(rdev->dev, "failed to reset GPU\n");
453 		rdev->gpu_lockup = true;
454 		ret = -1;
455 	} else
456 		dev_info(rdev->dev, "GPU reset succeed\n");
457 	r100_mc_resume(rdev, &save);
458 	return ret;
459 }
460 
461 /*
462  * r300,r350,rv350,rv380 VRAM info
463  */
464 void r300_mc_init(struct radeon_device *rdev)
465 {
466 	u64 base;
467 	u32 tmp;
468 
469 	/* DDR for all card after R300 & IGP */
470 	rdev->mc.vram_is_ddr = true;
471 	tmp = RREG32(RADEON_MEM_CNTL);
472 	tmp &= R300_MEM_NUM_CHANNELS_MASK;
473 	switch (tmp) {
474 	case 0: rdev->mc.vram_width = 64; break;
475 	case 1: rdev->mc.vram_width = 128; break;
476 	case 2: rdev->mc.vram_width = 256; break;
477 	default:  rdev->mc.vram_width = 128; break;
478 	}
479 	r100_vram_init_sizes(rdev);
480 	base = rdev->mc.aper_base;
481 	if (rdev->flags & RADEON_IS_IGP)
482 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
483 	radeon_vram_location(rdev, &rdev->mc, base);
484 	rdev->mc.gtt_base_align = 0;
485 	if (!(rdev->flags & RADEON_IS_AGP))
486 		radeon_gtt_location(rdev, &rdev->mc);
487 	radeon_update_bandwidth_info(rdev);
488 }
489 
490 void rv370_set_pcie_lanes(struct radeon_device *rdev, int lanes)
491 {
492 	uint32_t link_width_cntl, mask;
493 
494 	if (rdev->flags & RADEON_IS_IGP)
495 		return;
496 
497 	if (!(rdev->flags & RADEON_IS_PCIE))
498 		return;
499 
500 	/* FIXME wait for idle */
501 
502 	switch (lanes) {
503 	case 0:
504 		mask = RADEON_PCIE_LC_LINK_WIDTH_X0;
505 		break;
506 	case 1:
507 		mask = RADEON_PCIE_LC_LINK_WIDTH_X1;
508 		break;
509 	case 2:
510 		mask = RADEON_PCIE_LC_LINK_WIDTH_X2;
511 		break;
512 	case 4:
513 		mask = RADEON_PCIE_LC_LINK_WIDTH_X4;
514 		break;
515 	case 8:
516 		mask = RADEON_PCIE_LC_LINK_WIDTH_X8;
517 		break;
518 	case 12:
519 		mask = RADEON_PCIE_LC_LINK_WIDTH_X12;
520 		break;
521 	case 16:
522 	default:
523 		mask = RADEON_PCIE_LC_LINK_WIDTH_X16;
524 		break;
525 	}
526 
527 	link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
528 
529 	if ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) ==
530 	    (mask << RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT))
531 		return;
532 
533 	link_width_cntl &= ~(RADEON_PCIE_LC_LINK_WIDTH_MASK |
534 			     RADEON_PCIE_LC_RECONFIG_NOW |
535 			     RADEON_PCIE_LC_RECONFIG_LATER |
536 			     RADEON_PCIE_LC_SHORT_RECONFIG_EN);
537 	link_width_cntl |= mask;
538 	WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
539 	WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, (link_width_cntl |
540 						     RADEON_PCIE_LC_RECONFIG_NOW));
541 
542 	/* wait for lane set to complete */
543 	link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
544 	while (link_width_cntl == 0xffffffff)
545 		link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
546 
547 }
548 
549 int rv370_get_pcie_lanes(struct radeon_device *rdev)
550 {
551 	u32 link_width_cntl;
552 
553 	if (rdev->flags & RADEON_IS_IGP)
554 		return 0;
555 
556 	if (!(rdev->flags & RADEON_IS_PCIE))
557 		return 0;
558 
559 	/* FIXME wait for idle */
560 
561 	link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
562 
563 	switch ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) >> RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT) {
564 	case RADEON_PCIE_LC_LINK_WIDTH_X0:
565 		return 0;
566 	case RADEON_PCIE_LC_LINK_WIDTH_X1:
567 		return 1;
568 	case RADEON_PCIE_LC_LINK_WIDTH_X2:
569 		return 2;
570 	case RADEON_PCIE_LC_LINK_WIDTH_X4:
571 		return 4;
572 	case RADEON_PCIE_LC_LINK_WIDTH_X8:
573 		return 8;
574 	case RADEON_PCIE_LC_LINK_WIDTH_X16:
575 	default:
576 		return 16;
577 	}
578 }
579 
580 #if defined(CONFIG_DEBUG_FS)
581 static int rv370_debugfs_pcie_gart_info(struct seq_file *m, void *data)
582 {
583 	struct drm_info_node *node = (struct drm_info_node *) m->private;
584 	struct drm_device *dev = node->minor->dev;
585 	struct radeon_device *rdev = dev->dev_private;
586 	uint32_t tmp;
587 
588 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
589 	seq_printf(m, "PCIE_TX_GART_CNTL 0x%08x\n", tmp);
590 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_BASE);
591 	seq_printf(m, "PCIE_TX_GART_BASE 0x%08x\n", tmp);
592 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_LO);
593 	seq_printf(m, "PCIE_TX_GART_START_LO 0x%08x\n", tmp);
594 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_HI);
595 	seq_printf(m, "PCIE_TX_GART_START_HI 0x%08x\n", tmp);
596 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_LO);
597 	seq_printf(m, "PCIE_TX_GART_END_LO 0x%08x\n", tmp);
598 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_HI);
599 	seq_printf(m, "PCIE_TX_GART_END_HI 0x%08x\n", tmp);
600 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_ERROR);
601 	seq_printf(m, "PCIE_TX_GART_ERROR 0x%08x\n", tmp);
602 	return 0;
603 }
604 
605 static struct drm_info_list rv370_pcie_gart_info_list[] = {
606 	{"rv370_pcie_gart_info", rv370_debugfs_pcie_gart_info, 0, NULL},
607 };
608 #endif
609 
610 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev)
611 {
612 #if defined(CONFIG_DEBUG_FS)
613 	return radeon_debugfs_add_files(rdev, rv370_pcie_gart_info_list, 1);
614 #else
615 	return 0;
616 #endif
617 }
618 
619 static int r300_packet0_check(struct radeon_cs_parser *p,
620 		struct radeon_cs_packet *pkt,
621 		unsigned idx, unsigned reg)
622 {
623 	struct radeon_cs_reloc *reloc;
624 	struct r100_cs_track *track;
625 	volatile uint32_t *ib;
626 	uint32_t tmp, tile_flags = 0;
627 	unsigned i;
628 	int r;
629 	u32 idx_value;
630 
631 	ib = p->ib->ptr;
632 	track = (struct r100_cs_track *)p->track;
633 	idx_value = radeon_get_ib_value(p, idx);
634 
635 	switch(reg) {
636 	case AVIVO_D1MODE_VLINE_START_END:
637 	case RADEON_CRTC_GUI_TRIG_VLINE:
638 		r = r100_cs_packet_parse_vline(p);
639 		if (r) {
640 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
641 					idx, reg);
642 			r100_cs_dump_packet(p, pkt);
643 			return r;
644 		}
645 		break;
646 	case RADEON_DST_PITCH_OFFSET:
647 	case RADEON_SRC_PITCH_OFFSET:
648 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
649 		if (r)
650 			return r;
651 		break;
652 	case R300_RB3D_COLOROFFSET0:
653 	case R300_RB3D_COLOROFFSET1:
654 	case R300_RB3D_COLOROFFSET2:
655 	case R300_RB3D_COLOROFFSET3:
656 		i = (reg - R300_RB3D_COLOROFFSET0) >> 2;
657 		r = r100_cs_packet_next_reloc(p, &reloc);
658 		if (r) {
659 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
660 					idx, reg);
661 			r100_cs_dump_packet(p, pkt);
662 			return r;
663 		}
664 		track->cb[i].robj = reloc->robj;
665 		track->cb[i].offset = idx_value;
666 		track->cb_dirty = true;
667 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
668 		break;
669 	case R300_ZB_DEPTHOFFSET:
670 		r = r100_cs_packet_next_reloc(p, &reloc);
671 		if (r) {
672 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
673 					idx, reg);
674 			r100_cs_dump_packet(p, pkt);
675 			return r;
676 		}
677 		track->zb.robj = reloc->robj;
678 		track->zb.offset = idx_value;
679 		track->zb_dirty = true;
680 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
681 		break;
682 	case R300_TX_OFFSET_0:
683 	case R300_TX_OFFSET_0+4:
684 	case R300_TX_OFFSET_0+8:
685 	case R300_TX_OFFSET_0+12:
686 	case R300_TX_OFFSET_0+16:
687 	case R300_TX_OFFSET_0+20:
688 	case R300_TX_OFFSET_0+24:
689 	case R300_TX_OFFSET_0+28:
690 	case R300_TX_OFFSET_0+32:
691 	case R300_TX_OFFSET_0+36:
692 	case R300_TX_OFFSET_0+40:
693 	case R300_TX_OFFSET_0+44:
694 	case R300_TX_OFFSET_0+48:
695 	case R300_TX_OFFSET_0+52:
696 	case R300_TX_OFFSET_0+56:
697 	case R300_TX_OFFSET_0+60:
698 		i = (reg - R300_TX_OFFSET_0) >> 2;
699 		r = r100_cs_packet_next_reloc(p, &reloc);
700 		if (r) {
701 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
702 					idx, reg);
703 			r100_cs_dump_packet(p, pkt);
704 			return r;
705 		}
706 
707 		if (p->keep_tiling_flags) {
708 			ib[idx] = (idx_value & 31) | /* keep the 1st 5 bits */
709 				  ((idx_value & ~31) + (u32)reloc->lobj.gpu_offset);
710 		} else {
711 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
712 				tile_flags |= R300_TXO_MACRO_TILE;
713 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
714 				tile_flags |= R300_TXO_MICRO_TILE;
715 			else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
716 				tile_flags |= R300_TXO_MICRO_TILE_SQUARE;
717 
718 			tmp = idx_value + ((u32)reloc->lobj.gpu_offset);
719 			tmp |= tile_flags;
720 			ib[idx] = tmp;
721 		}
722 		track->textures[i].robj = reloc->robj;
723 		track->tex_dirty = true;
724 		break;
725 	/* Tracked registers */
726 	case 0x2084:
727 		/* VAP_VF_CNTL */
728 		track->vap_vf_cntl = idx_value;
729 		break;
730 	case 0x20B4:
731 		/* VAP_VTX_SIZE */
732 		track->vtx_size = idx_value & 0x7F;
733 		break;
734 	case 0x2134:
735 		/* VAP_VF_MAX_VTX_INDX */
736 		track->max_indx = idx_value & 0x00FFFFFFUL;
737 		break;
738 	case 0x2088:
739 		/* VAP_ALT_NUM_VERTICES - only valid on r500 */
740 		if (p->rdev->family < CHIP_RV515)
741 			goto fail;
742 		track->vap_alt_nverts = idx_value & 0xFFFFFF;
743 		break;
744 	case 0x43E4:
745 		/* SC_SCISSOR1 */
746 		track->maxy = ((idx_value >> 13) & 0x1FFF) + 1;
747 		if (p->rdev->family < CHIP_RV515) {
748 			track->maxy -= 1440;
749 		}
750 		track->cb_dirty = true;
751 		track->zb_dirty = true;
752 		break;
753 	case 0x4E00:
754 		/* RB3D_CCTL */
755 		if ((idx_value & (1 << 10)) && /* CMASK_ENABLE */
756 		    p->rdev->cmask_filp != p->filp) {
757 			DRM_ERROR("Invalid RB3D_CCTL: Cannot enable CMASK.\n");
758 			return -EINVAL;
759 		}
760 		track->num_cb = ((idx_value >> 5) & 0x3) + 1;
761 		track->cb_dirty = true;
762 		break;
763 	case 0x4E38:
764 	case 0x4E3C:
765 	case 0x4E40:
766 	case 0x4E44:
767 		/* RB3D_COLORPITCH0 */
768 		/* RB3D_COLORPITCH1 */
769 		/* RB3D_COLORPITCH2 */
770 		/* RB3D_COLORPITCH3 */
771 		if (!p->keep_tiling_flags) {
772 			r = r100_cs_packet_next_reloc(p, &reloc);
773 			if (r) {
774 				DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
775 					  idx, reg);
776 				r100_cs_dump_packet(p, pkt);
777 				return r;
778 			}
779 
780 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
781 				tile_flags |= R300_COLOR_TILE_ENABLE;
782 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
783 				tile_flags |= R300_COLOR_MICROTILE_ENABLE;
784 			else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
785 				tile_flags |= R300_COLOR_MICROTILE_SQUARE_ENABLE;
786 
787 			tmp = idx_value & ~(0x7 << 16);
788 			tmp |= tile_flags;
789 			ib[idx] = tmp;
790 		}
791 		i = (reg - 0x4E38) >> 2;
792 		track->cb[i].pitch = idx_value & 0x3FFE;
793 		switch (((idx_value >> 21) & 0xF)) {
794 		case 9:
795 		case 11:
796 		case 12:
797 			track->cb[i].cpp = 1;
798 			break;
799 		case 3:
800 		case 4:
801 		case 13:
802 		case 15:
803 			track->cb[i].cpp = 2;
804 			break;
805 		case 5:
806 			if (p->rdev->family < CHIP_RV515) {
807 				DRM_ERROR("Invalid color buffer format (%d)!\n",
808 					  ((idx_value >> 21) & 0xF));
809 				return -EINVAL;
810 			}
811 			/* Pass through. */
812 		case 6:
813 			track->cb[i].cpp = 4;
814 			break;
815 		case 10:
816 			track->cb[i].cpp = 8;
817 			break;
818 		case 7:
819 			track->cb[i].cpp = 16;
820 			break;
821 		default:
822 			DRM_ERROR("Invalid color buffer format (%d) !\n",
823 				  ((idx_value >> 21) & 0xF));
824 			return -EINVAL;
825 		}
826 		track->cb_dirty = true;
827 		break;
828 	case 0x4F00:
829 		/* ZB_CNTL */
830 		if (idx_value & 2) {
831 			track->z_enabled = true;
832 		} else {
833 			track->z_enabled = false;
834 		}
835 		track->zb_dirty = true;
836 		break;
837 	case 0x4F10:
838 		/* ZB_FORMAT */
839 		switch ((idx_value & 0xF)) {
840 		case 0:
841 		case 1:
842 			track->zb.cpp = 2;
843 			break;
844 		case 2:
845 			track->zb.cpp = 4;
846 			break;
847 		default:
848 			DRM_ERROR("Invalid z buffer format (%d) !\n",
849 				  (idx_value & 0xF));
850 			return -EINVAL;
851 		}
852 		track->zb_dirty = true;
853 		break;
854 	case 0x4F24:
855 		/* ZB_DEPTHPITCH */
856 		if (!p->keep_tiling_flags) {
857 			r = r100_cs_packet_next_reloc(p, &reloc);
858 			if (r) {
859 				DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
860 					  idx, reg);
861 				r100_cs_dump_packet(p, pkt);
862 				return r;
863 			}
864 
865 			if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
866 				tile_flags |= R300_DEPTHMACROTILE_ENABLE;
867 			if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
868 				tile_flags |= R300_DEPTHMICROTILE_TILED;
869 			else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
870 				tile_flags |= R300_DEPTHMICROTILE_TILED_SQUARE;
871 
872 			tmp = idx_value & ~(0x7 << 16);
873 			tmp |= tile_flags;
874 			ib[idx] = tmp;
875 		}
876 		track->zb.pitch = idx_value & 0x3FFC;
877 		track->zb_dirty = true;
878 		break;
879 	case 0x4104:
880 		/* TX_ENABLE */
881 		for (i = 0; i < 16; i++) {
882 			bool enabled;
883 
884 			enabled = !!(idx_value & (1 << i));
885 			track->textures[i].enabled = enabled;
886 		}
887 		track->tex_dirty = true;
888 		break;
889 	case 0x44C0:
890 	case 0x44C4:
891 	case 0x44C8:
892 	case 0x44CC:
893 	case 0x44D0:
894 	case 0x44D4:
895 	case 0x44D8:
896 	case 0x44DC:
897 	case 0x44E0:
898 	case 0x44E4:
899 	case 0x44E8:
900 	case 0x44EC:
901 	case 0x44F0:
902 	case 0x44F4:
903 	case 0x44F8:
904 	case 0x44FC:
905 		/* TX_FORMAT1_[0-15] */
906 		i = (reg - 0x44C0) >> 2;
907 		tmp = (idx_value >> 25) & 0x3;
908 		track->textures[i].tex_coord_type = tmp;
909 		switch ((idx_value & 0x1F)) {
910 		case R300_TX_FORMAT_X8:
911 		case R300_TX_FORMAT_Y4X4:
912 		case R300_TX_FORMAT_Z3Y3X2:
913 			track->textures[i].cpp = 1;
914 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
915 			break;
916 		case R300_TX_FORMAT_X16:
917 		case R300_TX_FORMAT_FL_I16:
918 		case R300_TX_FORMAT_Y8X8:
919 		case R300_TX_FORMAT_Z5Y6X5:
920 		case R300_TX_FORMAT_Z6Y5X5:
921 		case R300_TX_FORMAT_W4Z4Y4X4:
922 		case R300_TX_FORMAT_W1Z5Y5X5:
923 		case R300_TX_FORMAT_D3DMFT_CxV8U8:
924 		case R300_TX_FORMAT_B8G8_B8G8:
925 		case R300_TX_FORMAT_G8R8_G8B8:
926 			track->textures[i].cpp = 2;
927 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
928 			break;
929 		case R300_TX_FORMAT_Y16X16:
930 		case R300_TX_FORMAT_FL_I16A16:
931 		case R300_TX_FORMAT_Z11Y11X10:
932 		case R300_TX_FORMAT_Z10Y11X11:
933 		case R300_TX_FORMAT_W8Z8Y8X8:
934 		case R300_TX_FORMAT_W2Z10Y10X10:
935 		case 0x17:
936 		case R300_TX_FORMAT_FL_I32:
937 		case 0x1e:
938 			track->textures[i].cpp = 4;
939 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
940 			break;
941 		case R300_TX_FORMAT_W16Z16Y16X16:
942 		case R300_TX_FORMAT_FL_R16G16B16A16:
943 		case R300_TX_FORMAT_FL_I32A32:
944 			track->textures[i].cpp = 8;
945 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
946 			break;
947 		case R300_TX_FORMAT_FL_R32G32B32A32:
948 			track->textures[i].cpp = 16;
949 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
950 			break;
951 		case R300_TX_FORMAT_DXT1:
952 			track->textures[i].cpp = 1;
953 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
954 			break;
955 		case R300_TX_FORMAT_ATI2N:
956 			if (p->rdev->family < CHIP_R420) {
957 				DRM_ERROR("Invalid texture format %u\n",
958 					  (idx_value & 0x1F));
959 				return -EINVAL;
960 			}
961 			/* The same rules apply as for DXT3/5. */
962 			/* Pass through. */
963 		case R300_TX_FORMAT_DXT3:
964 		case R300_TX_FORMAT_DXT5:
965 			track->textures[i].cpp = 1;
966 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
967 			break;
968 		default:
969 			DRM_ERROR("Invalid texture format %u\n",
970 				  (idx_value & 0x1F));
971 			return -EINVAL;
972 		}
973 		track->tex_dirty = true;
974 		break;
975 	case 0x4400:
976 	case 0x4404:
977 	case 0x4408:
978 	case 0x440C:
979 	case 0x4410:
980 	case 0x4414:
981 	case 0x4418:
982 	case 0x441C:
983 	case 0x4420:
984 	case 0x4424:
985 	case 0x4428:
986 	case 0x442C:
987 	case 0x4430:
988 	case 0x4434:
989 	case 0x4438:
990 	case 0x443C:
991 		/* TX_FILTER0_[0-15] */
992 		i = (reg - 0x4400) >> 2;
993 		tmp = idx_value & 0x7;
994 		if (tmp == 2 || tmp == 4 || tmp == 6) {
995 			track->textures[i].roundup_w = false;
996 		}
997 		tmp = (idx_value >> 3) & 0x7;
998 		if (tmp == 2 || tmp == 4 || tmp == 6) {
999 			track->textures[i].roundup_h = false;
1000 		}
1001 		track->tex_dirty = true;
1002 		break;
1003 	case 0x4500:
1004 	case 0x4504:
1005 	case 0x4508:
1006 	case 0x450C:
1007 	case 0x4510:
1008 	case 0x4514:
1009 	case 0x4518:
1010 	case 0x451C:
1011 	case 0x4520:
1012 	case 0x4524:
1013 	case 0x4528:
1014 	case 0x452C:
1015 	case 0x4530:
1016 	case 0x4534:
1017 	case 0x4538:
1018 	case 0x453C:
1019 		/* TX_FORMAT2_[0-15] */
1020 		i = (reg - 0x4500) >> 2;
1021 		tmp = idx_value & 0x3FFF;
1022 		track->textures[i].pitch = tmp + 1;
1023 		if (p->rdev->family >= CHIP_RV515) {
1024 			tmp = ((idx_value >> 15) & 1) << 11;
1025 			track->textures[i].width_11 = tmp;
1026 			tmp = ((idx_value >> 16) & 1) << 11;
1027 			track->textures[i].height_11 = tmp;
1028 
1029 			/* ATI1N */
1030 			if (idx_value & (1 << 14)) {
1031 				/* The same rules apply as for DXT1. */
1032 				track->textures[i].compress_format =
1033 					R100_TRACK_COMP_DXT1;
1034 			}
1035 		} else if (idx_value & (1 << 14)) {
1036 			DRM_ERROR("Forbidden bit TXFORMAT_MSB\n");
1037 			return -EINVAL;
1038 		}
1039 		track->tex_dirty = true;
1040 		break;
1041 	case 0x4480:
1042 	case 0x4484:
1043 	case 0x4488:
1044 	case 0x448C:
1045 	case 0x4490:
1046 	case 0x4494:
1047 	case 0x4498:
1048 	case 0x449C:
1049 	case 0x44A0:
1050 	case 0x44A4:
1051 	case 0x44A8:
1052 	case 0x44AC:
1053 	case 0x44B0:
1054 	case 0x44B4:
1055 	case 0x44B8:
1056 	case 0x44BC:
1057 		/* TX_FORMAT0_[0-15] */
1058 		i = (reg - 0x4480) >> 2;
1059 		tmp = idx_value & 0x7FF;
1060 		track->textures[i].width = tmp + 1;
1061 		tmp = (idx_value >> 11) & 0x7FF;
1062 		track->textures[i].height = tmp + 1;
1063 		tmp = (idx_value >> 26) & 0xF;
1064 		track->textures[i].num_levels = tmp;
1065 		tmp = idx_value & (1 << 31);
1066 		track->textures[i].use_pitch = !!tmp;
1067 		tmp = (idx_value >> 22) & 0xF;
1068 		track->textures[i].txdepth = tmp;
1069 		track->tex_dirty = true;
1070 		break;
1071 	case R300_ZB_ZPASS_ADDR:
1072 		r = r100_cs_packet_next_reloc(p, &reloc);
1073 		if (r) {
1074 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1075 					idx, reg);
1076 			r100_cs_dump_packet(p, pkt);
1077 			return r;
1078 		}
1079 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1080 		break;
1081 	case 0x4e0c:
1082 		/* RB3D_COLOR_CHANNEL_MASK */
1083 		track->color_channel_mask = idx_value;
1084 		track->cb_dirty = true;
1085 		break;
1086 	case 0x43a4:
1087 		/* SC_HYPERZ_EN */
1088 		/* r300c emits this register - we need to disable hyperz for it
1089 		 * without complaining */
1090 		if (p->rdev->hyperz_filp != p->filp) {
1091 			if (idx_value & 0x1)
1092 				ib[idx] = idx_value & ~1;
1093 		}
1094 		break;
1095 	case 0x4f1c:
1096 		/* ZB_BW_CNTL */
1097 		track->zb_cb_clear = !!(idx_value & (1 << 5));
1098 		track->cb_dirty = true;
1099 		track->zb_dirty = true;
1100 		if (p->rdev->hyperz_filp != p->filp) {
1101 			if (idx_value & (R300_HIZ_ENABLE |
1102 					 R300_RD_COMP_ENABLE |
1103 					 R300_WR_COMP_ENABLE |
1104 					 R300_FAST_FILL_ENABLE))
1105 				goto fail;
1106 		}
1107 		break;
1108 	case 0x4e04:
1109 		/* RB3D_BLENDCNTL */
1110 		track->blend_read_enable = !!(idx_value & (1 << 2));
1111 		track->cb_dirty = true;
1112 		break;
1113 	case R300_RB3D_AARESOLVE_OFFSET:
1114 		r = r100_cs_packet_next_reloc(p, &reloc);
1115 		if (r) {
1116 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1117 				  idx, reg);
1118 			r100_cs_dump_packet(p, pkt);
1119 			return r;
1120 		}
1121 		track->aa.robj = reloc->robj;
1122 		track->aa.offset = idx_value;
1123 		track->aa_dirty = true;
1124 		ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1125 		break;
1126 	case R300_RB3D_AARESOLVE_PITCH:
1127 		track->aa.pitch = idx_value & 0x3FFE;
1128 		track->aa_dirty = true;
1129 		break;
1130 	case R300_RB3D_AARESOLVE_CTL:
1131 		track->aaresolve = idx_value & 0x1;
1132 		track->aa_dirty = true;
1133 		break;
1134 	case 0x4f30: /* ZB_MASK_OFFSET */
1135 	case 0x4f34: /* ZB_ZMASK_PITCH */
1136 	case 0x4f44: /* ZB_HIZ_OFFSET */
1137 	case 0x4f54: /* ZB_HIZ_PITCH */
1138 		if (idx_value && (p->rdev->hyperz_filp != p->filp))
1139 			goto fail;
1140 		break;
1141 	case 0x4028:
1142 		if (idx_value && (p->rdev->hyperz_filp != p->filp))
1143 			goto fail;
1144 		/* GB_Z_PEQ_CONFIG */
1145 		if (p->rdev->family >= CHIP_RV350)
1146 			break;
1147 		goto fail;
1148 		break;
1149 	case 0x4be8:
1150 		/* valid register only on RV530 */
1151 		if (p->rdev->family == CHIP_RV530)
1152 			break;
1153 		/* fallthrough do not move */
1154 	default:
1155 		goto fail;
1156 	}
1157 	return 0;
1158 fail:
1159 	printk(KERN_ERR "Forbidden register 0x%04X in cs at %d (val=%08x)\n",
1160 	       reg, idx, idx_value);
1161 	return -EINVAL;
1162 }
1163 
1164 static int r300_packet3_check(struct radeon_cs_parser *p,
1165 			      struct radeon_cs_packet *pkt)
1166 {
1167 	struct radeon_cs_reloc *reloc;
1168 	struct r100_cs_track *track;
1169 	volatile uint32_t *ib;
1170 	unsigned idx;
1171 	int r;
1172 
1173 	ib = p->ib->ptr;
1174 	idx = pkt->idx + 1;
1175 	track = (struct r100_cs_track *)p->track;
1176 	switch(pkt->opcode) {
1177 	case PACKET3_3D_LOAD_VBPNTR:
1178 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1179 		if (r)
1180 			return r;
1181 		break;
1182 	case PACKET3_INDX_BUFFER:
1183 		r = r100_cs_packet_next_reloc(p, &reloc);
1184 		if (r) {
1185 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1186 			r100_cs_dump_packet(p, pkt);
1187 			return r;
1188 		}
1189 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1190 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1191 		if (r) {
1192 			return r;
1193 		}
1194 		break;
1195 	/* Draw packet */
1196 	case PACKET3_3D_DRAW_IMMD:
1197 		/* Number of dwords is vtx_size * (num_vertices - 1)
1198 		 * PRIM_WALK must be equal to 3 vertex data in embedded
1199 		 * in cmd stream */
1200 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1201 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1202 			return -EINVAL;
1203 		}
1204 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1205 		track->immd_dwords = pkt->count - 1;
1206 		r = r100_cs_track_check(p->rdev, track);
1207 		if (r) {
1208 			return r;
1209 		}
1210 		break;
1211 	case PACKET3_3D_DRAW_IMMD_2:
1212 		/* Number of dwords is vtx_size * (num_vertices - 1)
1213 		 * PRIM_WALK must be equal to 3 vertex data in embedded
1214 		 * in cmd stream */
1215 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1216 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1217 			return -EINVAL;
1218 		}
1219 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1220 		track->immd_dwords = pkt->count;
1221 		r = r100_cs_track_check(p->rdev, track);
1222 		if (r) {
1223 			return r;
1224 		}
1225 		break;
1226 	case PACKET3_3D_DRAW_VBUF:
1227 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1228 		r = r100_cs_track_check(p->rdev, track);
1229 		if (r) {
1230 			return r;
1231 		}
1232 		break;
1233 	case PACKET3_3D_DRAW_VBUF_2:
1234 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1235 		r = r100_cs_track_check(p->rdev, track);
1236 		if (r) {
1237 			return r;
1238 		}
1239 		break;
1240 	case PACKET3_3D_DRAW_INDX:
1241 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1242 		r = r100_cs_track_check(p->rdev, track);
1243 		if (r) {
1244 			return r;
1245 		}
1246 		break;
1247 	case PACKET3_3D_DRAW_INDX_2:
1248 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1249 		r = r100_cs_track_check(p->rdev, track);
1250 		if (r) {
1251 			return r;
1252 		}
1253 		break;
1254 	case PACKET3_3D_CLEAR_HIZ:
1255 	case PACKET3_3D_CLEAR_ZMASK:
1256 		if (p->rdev->hyperz_filp != p->filp)
1257 			return -EINVAL;
1258 		break;
1259 	case PACKET3_3D_CLEAR_CMASK:
1260 		if (p->rdev->cmask_filp != p->filp)
1261 			return -EINVAL;
1262 		break;
1263 	case PACKET3_NOP:
1264 		break;
1265 	default:
1266 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1267 		return -EINVAL;
1268 	}
1269 	return 0;
1270 }
1271 
1272 int r300_cs_parse(struct radeon_cs_parser *p)
1273 {
1274 	struct radeon_cs_packet pkt;
1275 	struct r100_cs_track *track;
1276 	int r;
1277 
1278 	track = kzalloc(sizeof(*track), GFP_KERNEL);
1279 	if (track == NULL)
1280 		return -ENOMEM;
1281 	r100_cs_track_clear(p->rdev, track);
1282 	p->track = track;
1283 	do {
1284 		r = r100_cs_packet_parse(p, &pkt, p->idx);
1285 		if (r) {
1286 			return r;
1287 		}
1288 		p->idx += pkt.count + 2;
1289 		switch (pkt.type) {
1290 		case PACKET_TYPE0:
1291 			r = r100_cs_parse_packet0(p, &pkt,
1292 						  p->rdev->config.r300.reg_safe_bm,
1293 						  p->rdev->config.r300.reg_safe_bm_size,
1294 						  &r300_packet0_check);
1295 			break;
1296 		case PACKET_TYPE2:
1297 			break;
1298 		case PACKET_TYPE3:
1299 			r = r300_packet3_check(p, &pkt);
1300 			break;
1301 		default:
1302 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1303 			return -EINVAL;
1304 		}
1305 		if (r) {
1306 			return r;
1307 		}
1308 	} while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1309 	return 0;
1310 }
1311 
1312 void r300_set_reg_safe(struct radeon_device *rdev)
1313 {
1314 	rdev->config.r300.reg_safe_bm = r300_reg_safe_bm;
1315 	rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r300_reg_safe_bm);
1316 }
1317 
1318 void r300_mc_program(struct radeon_device *rdev)
1319 {
1320 	struct r100_mc_save save;
1321 	int r;
1322 
1323 	r = r100_debugfs_mc_info_init(rdev);
1324 	if (r) {
1325 		dev_err(rdev->dev, "Failed to create r100_mc debugfs file.\n");
1326 	}
1327 
1328 	/* Stops all mc clients */
1329 	r100_mc_stop(rdev, &save);
1330 	if (rdev->flags & RADEON_IS_AGP) {
1331 		WREG32(R_00014C_MC_AGP_LOCATION,
1332 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
1333 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
1334 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
1335 		WREG32(R_00015C_AGP_BASE_2,
1336 			upper_32_bits(rdev->mc.agp_base) & 0xff);
1337 	} else {
1338 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
1339 		WREG32(R_000170_AGP_BASE, 0);
1340 		WREG32(R_00015C_AGP_BASE_2, 0);
1341 	}
1342 	/* Wait for mc idle */
1343 	if (r300_mc_wait_for_idle(rdev))
1344 		DRM_INFO("Failed to wait MC idle before programming MC.\n");
1345 	/* Program MC, should be a 32bits limited address space */
1346 	WREG32(R_000148_MC_FB_LOCATION,
1347 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
1348 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
1349 	r100_mc_resume(rdev, &save);
1350 }
1351 
1352 void r300_clock_startup(struct radeon_device *rdev)
1353 {
1354 	u32 tmp;
1355 
1356 	if (radeon_dynclks != -1 && radeon_dynclks)
1357 		radeon_legacy_set_clock_gating(rdev, 1);
1358 	/* We need to force on some of the block */
1359 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
1360 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
1361 	if ((rdev->family == CHIP_RV350) || (rdev->family == CHIP_RV380))
1362 		tmp |= S_00000D_FORCE_VAP(1);
1363 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
1364 }
1365 
1366 static int r300_startup(struct radeon_device *rdev)
1367 {
1368 	int r;
1369 
1370 	/* set common regs */
1371 	r100_set_common_regs(rdev);
1372 	/* program mc */
1373 	r300_mc_program(rdev);
1374 	/* Resume clock */
1375 	r300_clock_startup(rdev);
1376 	/* Initialize GPU configuration (# pipes, ...) */
1377 	r300_gpu_init(rdev);
1378 	/* Initialize GART (initialize after TTM so we can allocate
1379 	 * memory through TTM but finalize after TTM) */
1380 	if (rdev->flags & RADEON_IS_PCIE) {
1381 		r = rv370_pcie_gart_enable(rdev);
1382 		if (r)
1383 			return r;
1384 	}
1385 
1386 	if (rdev->family == CHIP_R300 ||
1387 	    rdev->family == CHIP_R350 ||
1388 	    rdev->family == CHIP_RV350)
1389 		r100_enable_bm(rdev);
1390 
1391 	if (rdev->flags & RADEON_IS_PCI) {
1392 		r = r100_pci_gart_enable(rdev);
1393 		if (r)
1394 			return r;
1395 	}
1396 
1397 	/* allocate wb buffer */
1398 	r = radeon_wb_init(rdev);
1399 	if (r)
1400 		return r;
1401 
1402 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
1403 	if (r) {
1404 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
1405 		return r;
1406 	}
1407 
1408 	/* Enable IRQ */
1409 	r100_irq_set(rdev);
1410 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
1411 	/* 1M ring buffer */
1412 	r = r100_cp_init(rdev, 1024 * 1024);
1413 	if (r) {
1414 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
1415 		return r;
1416 	}
1417 
1418 	r = radeon_ib_pool_start(rdev);
1419 	if (r)
1420 		return r;
1421 
1422 	r = r100_ib_test(rdev);
1423 	if (r) {
1424 		dev_err(rdev->dev, "failed testing IB (%d).\n", r);
1425 		rdev->accel_working = false;
1426 		return r;
1427 	}
1428 
1429 	return 0;
1430 }
1431 
1432 int r300_resume(struct radeon_device *rdev)
1433 {
1434 	/* Make sur GART are not working */
1435 	if (rdev->flags & RADEON_IS_PCIE)
1436 		rv370_pcie_gart_disable(rdev);
1437 	if (rdev->flags & RADEON_IS_PCI)
1438 		r100_pci_gart_disable(rdev);
1439 	/* Resume clock before doing reset */
1440 	r300_clock_startup(rdev);
1441 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
1442 	if (radeon_asic_reset(rdev)) {
1443 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1444 			RREG32(R_000E40_RBBM_STATUS),
1445 			RREG32(R_0007C0_CP_STAT));
1446 	}
1447 	/* post */
1448 	radeon_combios_asic_init(rdev->ddev);
1449 	/* Resume clock after posting */
1450 	r300_clock_startup(rdev);
1451 	/* Initialize surface registers */
1452 	radeon_surface_init(rdev);
1453 
1454 	rdev->accel_working = true;
1455 	return r300_startup(rdev);
1456 }
1457 
1458 int r300_suspend(struct radeon_device *rdev)
1459 {
1460 	radeon_ib_pool_suspend(rdev);
1461 	r100_cp_disable(rdev);
1462 	radeon_wb_disable(rdev);
1463 	r100_irq_disable(rdev);
1464 	if (rdev->flags & RADEON_IS_PCIE)
1465 		rv370_pcie_gart_disable(rdev);
1466 	if (rdev->flags & RADEON_IS_PCI)
1467 		r100_pci_gart_disable(rdev);
1468 	return 0;
1469 }
1470 
1471 void r300_fini(struct radeon_device *rdev)
1472 {
1473 	r100_cp_fini(rdev);
1474 	radeon_wb_fini(rdev);
1475 	r100_ib_fini(rdev);
1476 	radeon_gem_fini(rdev);
1477 	if (rdev->flags & RADEON_IS_PCIE)
1478 		rv370_pcie_gart_fini(rdev);
1479 	if (rdev->flags & RADEON_IS_PCI)
1480 		r100_pci_gart_fini(rdev);
1481 	radeon_agp_fini(rdev);
1482 	radeon_irq_kms_fini(rdev);
1483 	radeon_fence_driver_fini(rdev);
1484 	radeon_bo_fini(rdev);
1485 	radeon_atombios_fini(rdev);
1486 	kfree(rdev->bios);
1487 	rdev->bios = NULL;
1488 }
1489 
1490 int r300_init(struct radeon_device *rdev)
1491 {
1492 	int r;
1493 
1494 	/* Disable VGA */
1495 	r100_vga_render_disable(rdev);
1496 	/* Initialize scratch registers */
1497 	radeon_scratch_init(rdev);
1498 	/* Initialize surface registers */
1499 	radeon_surface_init(rdev);
1500 	/* TODO: disable VGA need to use VGA request */
1501 	/* restore some register to sane defaults */
1502 	r100_restore_sanity(rdev);
1503 	/* BIOS*/
1504 	if (!radeon_get_bios(rdev)) {
1505 		if (ASIC_IS_AVIVO(rdev))
1506 			return -EINVAL;
1507 	}
1508 	if (rdev->is_atom_bios) {
1509 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
1510 		return -EINVAL;
1511 	} else {
1512 		r = radeon_combios_init(rdev);
1513 		if (r)
1514 			return r;
1515 	}
1516 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
1517 	if (radeon_asic_reset(rdev)) {
1518 		dev_warn(rdev->dev,
1519 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1520 			RREG32(R_000E40_RBBM_STATUS),
1521 			RREG32(R_0007C0_CP_STAT));
1522 	}
1523 	/* check if cards are posted or not */
1524 	if (radeon_boot_test_post_card(rdev) == false)
1525 		return -EINVAL;
1526 	/* Set asic errata */
1527 	r300_errata(rdev);
1528 	/* Initialize clocks */
1529 	radeon_get_clock_info(rdev->ddev);
1530 	/* initialize AGP */
1531 	if (rdev->flags & RADEON_IS_AGP) {
1532 		r = radeon_agp_init(rdev);
1533 		if (r) {
1534 			radeon_agp_disable(rdev);
1535 		}
1536 	}
1537 	/* initialize memory controller */
1538 	r300_mc_init(rdev);
1539 	/* Fence driver */
1540 	r = radeon_fence_driver_init(rdev);
1541 	if (r)
1542 		return r;
1543 	r = radeon_irq_kms_init(rdev);
1544 	if (r)
1545 		return r;
1546 	/* Memory manager */
1547 	r = radeon_bo_init(rdev);
1548 	if (r)
1549 		return r;
1550 	if (rdev->flags & RADEON_IS_PCIE) {
1551 		r = rv370_pcie_gart_init(rdev);
1552 		if (r)
1553 			return r;
1554 	}
1555 	if (rdev->flags & RADEON_IS_PCI) {
1556 		r = r100_pci_gart_init(rdev);
1557 		if (r)
1558 			return r;
1559 	}
1560 	r300_set_reg_safe(rdev);
1561 
1562 	r = radeon_ib_pool_init(rdev);
1563 	rdev->accel_working = true;
1564 	if (r) {
1565 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
1566 		rdev->accel_working = false;
1567 	}
1568 
1569 	r = r300_startup(rdev);
1570 	if (r) {
1571 		/* Somethings want wront with the accel init stop accel */
1572 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
1573 		r100_cp_fini(rdev);
1574 		radeon_wb_fini(rdev);
1575 		r100_ib_fini(rdev);
1576 		radeon_irq_kms_fini(rdev);
1577 		if (rdev->flags & RADEON_IS_PCIE)
1578 			rv370_pcie_gart_fini(rdev);
1579 		if (rdev->flags & RADEON_IS_PCI)
1580 			r100_pci_gart_fini(rdev);
1581 		radeon_agp_fini(rdev);
1582 		rdev->accel_working = false;
1583 	}
1584 	return 0;
1585 }
1586