xref: /openbmc/linux/drivers/gpu/drm/radeon/r300.c (revision adb57164)
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 
29 #include <linux/pci.h>
30 #include <linux/seq_file.h>
31 #include <linux/slab.h>
32 
33 #include <drm/drm.h>
34 #include <drm/drm_crtc_helper.h>
35 #include <drm/drm_debugfs.h>
36 #include <drm/drm_device.h>
37 #include <drm/drm_file.h>
38 #include <drm/radeon_drm.h>
39 
40 #include "r100_track.h"
41 #include "r300_reg_safe.h"
42 #include "r300d.h"
43 #include "radeon.h"
44 #include "radeon_asic.h"
45 #include "radeon_reg.h"
46 #include "rv350d.h"
47 
48 /* This files gather functions specifics to: r300,r350,rv350,rv370,rv380
49  *
50  * GPU Errata:
51  * - HOST_PATH_CNTL: r300 family seems to dislike write to HOST_PATH_CNTL
52  *   using MMIO to flush host path read cache, this lead to HARDLOCKUP.
53  *   However, scheduling such write to the ring seems harmless, i suspect
54  *   the CP read collide with the flush somehow, or maybe the MC, hard to
55  *   tell. (Jerome Glisse)
56  */
57 
58 /*
59  * Indirect registers accessor
60  */
61 uint32_t rv370_pcie_rreg(struct radeon_device *rdev, uint32_t reg)
62 {
63 	unsigned long flags;
64 	uint32_t r;
65 
66 	spin_lock_irqsave(&rdev->pcie_idx_lock, flags);
67 	WREG32(RADEON_PCIE_INDEX, ((reg) & rdev->pcie_reg_mask));
68 	r = RREG32(RADEON_PCIE_DATA);
69 	spin_unlock_irqrestore(&rdev->pcie_idx_lock, flags);
70 	return r;
71 }
72 
73 void rv370_pcie_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
74 {
75 	unsigned long flags;
76 
77 	spin_lock_irqsave(&rdev->pcie_idx_lock, flags);
78 	WREG32(RADEON_PCIE_INDEX, ((reg) & rdev->pcie_reg_mask));
79 	WREG32(RADEON_PCIE_DATA, (v));
80 	spin_unlock_irqrestore(&rdev->pcie_idx_lock, flags);
81 }
82 
83 /*
84  * rv370,rv380 PCIE GART
85  */
86 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev);
87 
88 void rv370_pcie_gart_tlb_flush(struct radeon_device *rdev)
89 {
90 	uint32_t tmp;
91 	int i;
92 
93 	/* Workaround HW bug do flush 2 times */
94 	for (i = 0; i < 2; i++) {
95 		tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
96 		WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp | RADEON_PCIE_TX_GART_INVALIDATE_TLB);
97 		(void)RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
98 		WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
99 	}
100 	mb();
101 }
102 
103 #define R300_PTE_UNSNOOPED (1 << 0)
104 #define R300_PTE_WRITEABLE (1 << 2)
105 #define R300_PTE_READABLE  (1 << 3)
106 
107 uint64_t rv370_pcie_gart_get_page_entry(uint64_t addr, uint32_t flags)
108 {
109 	addr = (lower_32_bits(addr) >> 8) |
110 		((upper_32_bits(addr) & 0xff) << 24);
111 	if (flags & RADEON_GART_PAGE_READ)
112 		addr |= R300_PTE_READABLE;
113 	if (flags & RADEON_GART_PAGE_WRITE)
114 		addr |= R300_PTE_WRITEABLE;
115 	if (!(flags & RADEON_GART_PAGE_SNOOP))
116 		addr |= R300_PTE_UNSNOOPED;
117 	return addr;
118 }
119 
120 void rv370_pcie_gart_set_page(struct radeon_device *rdev, unsigned i,
121 			      uint64_t entry)
122 {
123 	void __iomem *ptr = rdev->gart.ptr;
124 
125 	/* on x86 we want this to be CPU endian, on powerpc
126 	 * on powerpc without HW swappers, it'll get swapped on way
127 	 * into VRAM - so no need for cpu_to_le32 on VRAM tables */
128 	writel(entry, ((void __iomem *)ptr) + (i * 4));
129 }
130 
131 int rv370_pcie_gart_init(struct radeon_device *rdev)
132 {
133 	int r;
134 
135 	if (rdev->gart.robj) {
136 		WARN(1, "RV370 PCIE GART already initialized\n");
137 		return 0;
138 	}
139 	/* Initialize common gart structure */
140 	r = radeon_gart_init(rdev);
141 	if (r)
142 		return r;
143 	r = rv370_debugfs_pcie_gart_info_init(rdev);
144 	if (r)
145 		DRM_ERROR("Failed to register debugfs file for PCIE gart !\n");
146 	rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
147 	rdev->asic->gart.tlb_flush = &rv370_pcie_gart_tlb_flush;
148 	rdev->asic->gart.get_page_entry = &rv370_pcie_gart_get_page_entry;
149 	rdev->asic->gart.set_page = &rv370_pcie_gart_set_page;
150 	return radeon_gart_table_vram_alloc(rdev);
151 }
152 
153 int rv370_pcie_gart_enable(struct radeon_device *rdev)
154 {
155 	uint32_t table_addr;
156 	uint32_t tmp;
157 	int r;
158 
159 	if (rdev->gart.robj == NULL) {
160 		dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
161 		return -EINVAL;
162 	}
163 	r = radeon_gart_table_vram_pin(rdev);
164 	if (r)
165 		return r;
166 	/* discard memory request outside of configured range */
167 	tmp = RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
168 	WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
169 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, rdev->mc.gtt_start);
170 	tmp = rdev->mc.gtt_end & ~RADEON_GPU_PAGE_MASK;
171 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, tmp);
172 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
173 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
174 	table_addr = rdev->gart.table_addr;
175 	WREG32_PCIE(RADEON_PCIE_TX_GART_BASE, table_addr);
176 	/* FIXME: setup default page */
177 	WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_LO, rdev->mc.vram_start);
178 	WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_HI, 0);
179 	/* Clear error */
180 	WREG32_PCIE(RADEON_PCIE_TX_GART_ERROR, 0);
181 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
182 	tmp |= RADEON_PCIE_TX_GART_EN;
183 	tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
184 	WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
185 	rv370_pcie_gart_tlb_flush(rdev);
186 	DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
187 		 (unsigned)(rdev->mc.gtt_size >> 20),
188 		 (unsigned long long)table_addr);
189 	rdev->gart.ready = true;
190 	return 0;
191 }
192 
193 void rv370_pcie_gart_disable(struct radeon_device *rdev)
194 {
195 	u32 tmp;
196 
197 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, 0);
198 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, 0);
199 	WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
200 	WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
201 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
202 	tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
203 	WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp & ~RADEON_PCIE_TX_GART_EN);
204 	radeon_gart_table_vram_unpin(rdev);
205 }
206 
207 void rv370_pcie_gart_fini(struct radeon_device *rdev)
208 {
209 	radeon_gart_fini(rdev);
210 	rv370_pcie_gart_disable(rdev);
211 	radeon_gart_table_vram_free(rdev);
212 }
213 
214 void r300_fence_ring_emit(struct radeon_device *rdev,
215 			  struct radeon_fence *fence)
216 {
217 	struct radeon_ring *ring = &rdev->ring[fence->ring];
218 
219 	/* Who ever call radeon_fence_emit should call ring_lock and ask
220 	 * for enough space (today caller are ib schedule and buffer move) */
221 	/* Write SC register so SC & US assert idle */
222 	radeon_ring_write(ring, PACKET0(R300_RE_SCISSORS_TL, 0));
223 	radeon_ring_write(ring, 0);
224 	radeon_ring_write(ring, PACKET0(R300_RE_SCISSORS_BR, 0));
225 	radeon_ring_write(ring, 0);
226 	/* Flush 3D cache */
227 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
228 	radeon_ring_write(ring, R300_RB3D_DC_FLUSH);
229 	radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
230 	radeon_ring_write(ring, R300_ZC_FLUSH);
231 	/* Wait until IDLE & CLEAN */
232 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
233 	radeon_ring_write(ring, (RADEON_WAIT_3D_IDLECLEAN |
234 				 RADEON_WAIT_2D_IDLECLEAN |
235 				 RADEON_WAIT_DMA_GUI_IDLE));
236 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
237 	radeon_ring_write(ring, rdev->config.r300.hdp_cntl |
238 				RADEON_HDP_READ_BUFFER_INVALIDATE);
239 	radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
240 	radeon_ring_write(ring, rdev->config.r300.hdp_cntl);
241 	/* Emit fence sequence & fire IRQ */
242 	radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
243 	radeon_ring_write(ring, fence->seq);
244 	radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
245 	radeon_ring_write(ring, RADEON_SW_INT_FIRE);
246 }
247 
248 void r300_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
249 {
250 	unsigned gb_tile_config;
251 	int r;
252 
253 	/* Sub pixel 1/12 so we can have 4K rendering according to doc */
254 	gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
255 	switch(rdev->num_gb_pipes) {
256 	case 2:
257 		gb_tile_config |= R300_PIPE_COUNT_R300;
258 		break;
259 	case 3:
260 		gb_tile_config |= R300_PIPE_COUNT_R420_3P;
261 		break;
262 	case 4:
263 		gb_tile_config |= R300_PIPE_COUNT_R420;
264 		break;
265 	case 1:
266 	default:
267 		gb_tile_config |= R300_PIPE_COUNT_RV350;
268 		break;
269 	}
270 
271 	r = radeon_ring_lock(rdev, ring, 64);
272 	if (r) {
273 		return;
274 	}
275 	radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
276 	radeon_ring_write(ring,
277 			  RADEON_ISYNC_ANY2D_IDLE3D |
278 			  RADEON_ISYNC_ANY3D_IDLE2D |
279 			  RADEON_ISYNC_WAIT_IDLEGUI |
280 			  RADEON_ISYNC_CPSCRATCH_IDLEGUI);
281 	radeon_ring_write(ring, PACKET0(R300_GB_TILE_CONFIG, 0));
282 	radeon_ring_write(ring, gb_tile_config);
283 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
284 	radeon_ring_write(ring,
285 			  RADEON_WAIT_2D_IDLECLEAN |
286 			  RADEON_WAIT_3D_IDLECLEAN);
287 	radeon_ring_write(ring, PACKET0(R300_DST_PIPE_CONFIG, 0));
288 	radeon_ring_write(ring, R300_PIPE_AUTO_CONFIG);
289 	radeon_ring_write(ring, PACKET0(R300_GB_SELECT, 0));
290 	radeon_ring_write(ring, 0);
291 	radeon_ring_write(ring, PACKET0(R300_GB_ENABLE, 0));
292 	radeon_ring_write(ring, 0);
293 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
294 	radeon_ring_write(ring, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
295 	radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
296 	radeon_ring_write(ring, R300_ZC_FLUSH | R300_ZC_FREE);
297 	radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
298 	radeon_ring_write(ring,
299 			  RADEON_WAIT_2D_IDLECLEAN |
300 			  RADEON_WAIT_3D_IDLECLEAN);
301 	radeon_ring_write(ring, PACKET0(R300_GB_AA_CONFIG, 0));
302 	radeon_ring_write(ring, 0);
303 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
304 	radeon_ring_write(ring, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
305 	radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
306 	radeon_ring_write(ring, R300_ZC_FLUSH | R300_ZC_FREE);
307 	radeon_ring_write(ring, PACKET0(R300_GB_MSPOS0, 0));
308 	radeon_ring_write(ring,
309 			  ((6 << R300_MS_X0_SHIFT) |
310 			   (6 << R300_MS_Y0_SHIFT) |
311 			   (6 << R300_MS_X1_SHIFT) |
312 			   (6 << R300_MS_Y1_SHIFT) |
313 			   (6 << R300_MS_X2_SHIFT) |
314 			   (6 << R300_MS_Y2_SHIFT) |
315 			   (6 << R300_MSBD0_Y_SHIFT) |
316 			   (6 << R300_MSBD0_X_SHIFT)));
317 	radeon_ring_write(ring, PACKET0(R300_GB_MSPOS1, 0));
318 	radeon_ring_write(ring,
319 			  ((6 << R300_MS_X3_SHIFT) |
320 			   (6 << R300_MS_Y3_SHIFT) |
321 			   (6 << R300_MS_X4_SHIFT) |
322 			   (6 << R300_MS_Y4_SHIFT) |
323 			   (6 << R300_MS_X5_SHIFT) |
324 			   (6 << R300_MS_Y5_SHIFT) |
325 			   (6 << R300_MSBD1_SHIFT)));
326 	radeon_ring_write(ring, PACKET0(R300_GA_ENHANCE, 0));
327 	radeon_ring_write(ring, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL);
328 	radeon_ring_write(ring, PACKET0(R300_GA_POLY_MODE, 0));
329 	radeon_ring_write(ring,
330 			  R300_FRONT_PTYPE_TRIANGE | R300_BACK_PTYPE_TRIANGE);
331 	radeon_ring_write(ring, PACKET0(R300_GA_ROUND_MODE, 0));
332 	radeon_ring_write(ring,
333 			  R300_GEOMETRY_ROUND_NEAREST |
334 			  R300_COLOR_ROUND_NEAREST);
335 	radeon_ring_unlock_commit(rdev, ring, false);
336 }
337 
338 static void r300_errata(struct radeon_device *rdev)
339 {
340 	rdev->pll_errata = 0;
341 
342 	if (rdev->family == CHIP_R300 &&
343 	    (RREG32(RADEON_CONFIG_CNTL) & RADEON_CFG_ATI_REV_ID_MASK) == RADEON_CFG_ATI_REV_A11) {
344 		rdev->pll_errata |= CHIP_ERRATA_R300_CG;
345 	}
346 }
347 
348 int r300_mc_wait_for_idle(struct radeon_device *rdev)
349 {
350 	unsigned i;
351 	uint32_t tmp;
352 
353 	for (i = 0; i < rdev->usec_timeout; i++) {
354 		/* read MC_STATUS */
355 		tmp = RREG32(RADEON_MC_STATUS);
356 		if (tmp & R300_MC_IDLE) {
357 			return 0;
358 		}
359 		udelay(1);
360 	}
361 	return -1;
362 }
363 
364 static void r300_gpu_init(struct radeon_device *rdev)
365 {
366 	uint32_t gb_tile_config, tmp;
367 
368 	if ((rdev->family == CHIP_R300 && rdev->pdev->device != 0x4144) ||
369 	    (rdev->family == CHIP_R350 && rdev->pdev->device != 0x4148)) {
370 		/* r300,r350 */
371 		rdev->num_gb_pipes = 2;
372 	} else {
373 		/* rv350,rv370,rv380,r300 AD, r350 AH */
374 		rdev->num_gb_pipes = 1;
375 	}
376 	rdev->num_z_pipes = 1;
377 	gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
378 	switch (rdev->num_gb_pipes) {
379 	case 2:
380 		gb_tile_config |= R300_PIPE_COUNT_R300;
381 		break;
382 	case 3:
383 		gb_tile_config |= R300_PIPE_COUNT_R420_3P;
384 		break;
385 	case 4:
386 		gb_tile_config |= R300_PIPE_COUNT_R420;
387 		break;
388 	default:
389 	case 1:
390 		gb_tile_config |= R300_PIPE_COUNT_RV350;
391 		break;
392 	}
393 	WREG32(R300_GB_TILE_CONFIG, gb_tile_config);
394 
395 	if (r100_gui_wait_for_idle(rdev)) {
396 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
397 	}
398 
399 	tmp = RREG32(R300_DST_PIPE_CONFIG);
400 	WREG32(R300_DST_PIPE_CONFIG, tmp | R300_PIPE_AUTO_CONFIG);
401 
402 	WREG32(R300_RB2D_DSTCACHE_MODE,
403 	       R300_DC_AUTOFLUSH_ENABLE |
404 	       R300_DC_DC_DISABLE_IGNORE_PE);
405 
406 	if (r100_gui_wait_for_idle(rdev)) {
407 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
408 	}
409 	if (r300_mc_wait_for_idle(rdev)) {
410 		pr_warn("Failed to wait MC idle while programming pipes. Bad things might happen.\n");
411 	}
412 	DRM_INFO("radeon: %d quad pipes, %d Z pipes initialized\n",
413 		 rdev->num_gb_pipes, rdev->num_z_pipes);
414 }
415 
416 int r300_asic_reset(struct radeon_device *rdev, bool hard)
417 {
418 	struct r100_mc_save save;
419 	u32 status, tmp;
420 	int ret = 0;
421 
422 	status = RREG32(R_000E40_RBBM_STATUS);
423 	if (!G_000E40_GUI_ACTIVE(status)) {
424 		return 0;
425 	}
426 	r100_mc_stop(rdev, &save);
427 	status = RREG32(R_000E40_RBBM_STATUS);
428 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
429 	/* stop CP */
430 	WREG32(RADEON_CP_CSQ_CNTL, 0);
431 	tmp = RREG32(RADEON_CP_RB_CNTL);
432 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
433 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
434 	WREG32(RADEON_CP_RB_WPTR, 0);
435 	WREG32(RADEON_CP_RB_CNTL, tmp);
436 	/* save PCI state */
437 	pci_save_state(rdev->pdev);
438 	/* disable bus mastering */
439 	r100_bm_disable(rdev);
440 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_VAP(1) |
441 					S_0000F0_SOFT_RESET_GA(1));
442 	RREG32(R_0000F0_RBBM_SOFT_RESET);
443 	mdelay(500);
444 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
445 	mdelay(1);
446 	status = RREG32(R_000E40_RBBM_STATUS);
447 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
448 	/* resetting the CP seems to be problematic sometimes it end up
449 	 * hard locking the computer, but it's necessary for successful
450 	 * reset more test & playing is needed on R3XX/R4XX to find a
451 	 * reliable (if any solution)
452 	 */
453 	WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
454 	RREG32(R_0000F0_RBBM_SOFT_RESET);
455 	mdelay(500);
456 	WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
457 	mdelay(1);
458 	status = RREG32(R_000E40_RBBM_STATUS);
459 	dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
460 	/* restore PCI & busmastering */
461 	pci_restore_state(rdev->pdev);
462 	r100_enable_bm(rdev);
463 	/* Check if GPU is idle */
464 	if (G_000E40_GA_BUSY(status) || G_000E40_VAP_BUSY(status)) {
465 		dev_err(rdev->dev, "failed to reset GPU\n");
466 		ret = -1;
467 	} else
468 		dev_info(rdev->dev, "GPU reset succeed\n");
469 	r100_mc_resume(rdev, &save);
470 	return ret;
471 }
472 
473 /*
474  * r300,r350,rv350,rv380 VRAM info
475  */
476 void r300_mc_init(struct radeon_device *rdev)
477 {
478 	u64 base;
479 	u32 tmp;
480 
481 	/* DDR for all card after R300 & IGP */
482 	rdev->mc.vram_is_ddr = true;
483 	tmp = RREG32(RADEON_MEM_CNTL);
484 	tmp &= R300_MEM_NUM_CHANNELS_MASK;
485 	switch (tmp) {
486 	case 0: rdev->mc.vram_width = 64; break;
487 	case 1: rdev->mc.vram_width = 128; break;
488 	case 2: rdev->mc.vram_width = 256; break;
489 	default:  rdev->mc.vram_width = 128; break;
490 	}
491 	r100_vram_init_sizes(rdev);
492 	base = rdev->mc.aper_base;
493 	if (rdev->flags & RADEON_IS_IGP)
494 		base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
495 	radeon_vram_location(rdev, &rdev->mc, base);
496 	rdev->mc.gtt_base_align = 0;
497 	if (!(rdev->flags & RADEON_IS_AGP))
498 		radeon_gtt_location(rdev, &rdev->mc);
499 	radeon_update_bandwidth_info(rdev);
500 }
501 
502 void rv370_set_pcie_lanes(struct radeon_device *rdev, int lanes)
503 {
504 	uint32_t link_width_cntl, mask;
505 
506 	if (rdev->flags & RADEON_IS_IGP)
507 		return;
508 
509 	if (!(rdev->flags & RADEON_IS_PCIE))
510 		return;
511 
512 	/* FIXME wait for idle */
513 
514 	switch (lanes) {
515 	case 0:
516 		mask = RADEON_PCIE_LC_LINK_WIDTH_X0;
517 		break;
518 	case 1:
519 		mask = RADEON_PCIE_LC_LINK_WIDTH_X1;
520 		break;
521 	case 2:
522 		mask = RADEON_PCIE_LC_LINK_WIDTH_X2;
523 		break;
524 	case 4:
525 		mask = RADEON_PCIE_LC_LINK_WIDTH_X4;
526 		break;
527 	case 8:
528 		mask = RADEON_PCIE_LC_LINK_WIDTH_X8;
529 		break;
530 	case 12:
531 		mask = RADEON_PCIE_LC_LINK_WIDTH_X12;
532 		break;
533 	case 16:
534 	default:
535 		mask = RADEON_PCIE_LC_LINK_WIDTH_X16;
536 		break;
537 	}
538 
539 	link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
540 
541 	if ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) ==
542 	    (mask << RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT))
543 		return;
544 
545 	link_width_cntl &= ~(RADEON_PCIE_LC_LINK_WIDTH_MASK |
546 			     RADEON_PCIE_LC_RECONFIG_NOW |
547 			     RADEON_PCIE_LC_RECONFIG_LATER |
548 			     RADEON_PCIE_LC_SHORT_RECONFIG_EN);
549 	link_width_cntl |= mask;
550 	WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
551 	WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, (link_width_cntl |
552 						     RADEON_PCIE_LC_RECONFIG_NOW));
553 
554 	/* wait for lane set to complete */
555 	link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
556 	while (link_width_cntl == 0xffffffff)
557 		link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
558 
559 }
560 
561 int rv370_get_pcie_lanes(struct radeon_device *rdev)
562 {
563 	u32 link_width_cntl;
564 
565 	if (rdev->flags & RADEON_IS_IGP)
566 		return 0;
567 
568 	if (!(rdev->flags & RADEON_IS_PCIE))
569 		return 0;
570 
571 	/* FIXME wait for idle */
572 
573 	link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
574 
575 	switch ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) >> RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT) {
576 	case RADEON_PCIE_LC_LINK_WIDTH_X0:
577 		return 0;
578 	case RADEON_PCIE_LC_LINK_WIDTH_X1:
579 		return 1;
580 	case RADEON_PCIE_LC_LINK_WIDTH_X2:
581 		return 2;
582 	case RADEON_PCIE_LC_LINK_WIDTH_X4:
583 		return 4;
584 	case RADEON_PCIE_LC_LINK_WIDTH_X8:
585 		return 8;
586 	case RADEON_PCIE_LC_LINK_WIDTH_X16:
587 	default:
588 		return 16;
589 	}
590 }
591 
592 #if defined(CONFIG_DEBUG_FS)
593 static int rv370_debugfs_pcie_gart_info(struct seq_file *m, void *data)
594 {
595 	struct drm_info_node *node = (struct drm_info_node *) m->private;
596 	struct drm_device *dev = node->minor->dev;
597 	struct radeon_device *rdev = dev->dev_private;
598 	uint32_t tmp;
599 
600 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
601 	seq_printf(m, "PCIE_TX_GART_CNTL 0x%08x\n", tmp);
602 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_BASE);
603 	seq_printf(m, "PCIE_TX_GART_BASE 0x%08x\n", tmp);
604 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_LO);
605 	seq_printf(m, "PCIE_TX_GART_START_LO 0x%08x\n", tmp);
606 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_HI);
607 	seq_printf(m, "PCIE_TX_GART_START_HI 0x%08x\n", tmp);
608 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_LO);
609 	seq_printf(m, "PCIE_TX_GART_END_LO 0x%08x\n", tmp);
610 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_HI);
611 	seq_printf(m, "PCIE_TX_GART_END_HI 0x%08x\n", tmp);
612 	tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_ERROR);
613 	seq_printf(m, "PCIE_TX_GART_ERROR 0x%08x\n", tmp);
614 	return 0;
615 }
616 
617 static struct drm_info_list rv370_pcie_gart_info_list[] = {
618 	{"rv370_pcie_gart_info", rv370_debugfs_pcie_gart_info, 0, NULL},
619 };
620 #endif
621 
622 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev)
623 {
624 #if defined(CONFIG_DEBUG_FS)
625 	return radeon_debugfs_add_files(rdev, rv370_pcie_gart_info_list, 1);
626 #else
627 	return 0;
628 #endif
629 }
630 
631 static int r300_packet0_check(struct radeon_cs_parser *p,
632 		struct radeon_cs_packet *pkt,
633 		unsigned idx, unsigned reg)
634 {
635 	struct radeon_bo_list *reloc;
636 	struct r100_cs_track *track;
637 	volatile uint32_t *ib;
638 	uint32_t tmp, tile_flags = 0;
639 	unsigned i;
640 	int r;
641 	u32 idx_value;
642 
643 	ib = p->ib.ptr;
644 	track = (struct r100_cs_track *)p->track;
645 	idx_value = radeon_get_ib_value(p, idx);
646 
647 	switch(reg) {
648 	case AVIVO_D1MODE_VLINE_START_END:
649 	case RADEON_CRTC_GUI_TRIG_VLINE:
650 		r = r100_cs_packet_parse_vline(p);
651 		if (r) {
652 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
653 					idx, reg);
654 			radeon_cs_dump_packet(p, pkt);
655 			return r;
656 		}
657 		break;
658 	case RADEON_DST_PITCH_OFFSET:
659 	case RADEON_SRC_PITCH_OFFSET:
660 		r = r100_reloc_pitch_offset(p, pkt, idx, reg);
661 		if (r)
662 			return r;
663 		break;
664 	case R300_RB3D_COLOROFFSET0:
665 	case R300_RB3D_COLOROFFSET1:
666 	case R300_RB3D_COLOROFFSET2:
667 	case R300_RB3D_COLOROFFSET3:
668 		i = (reg - R300_RB3D_COLOROFFSET0) >> 2;
669 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
670 		if (r) {
671 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
672 					idx, reg);
673 			radeon_cs_dump_packet(p, pkt);
674 			return r;
675 		}
676 		track->cb[i].robj = reloc->robj;
677 		track->cb[i].offset = idx_value;
678 		track->cb_dirty = true;
679 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
680 		break;
681 	case R300_ZB_DEPTHOFFSET:
682 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
683 		if (r) {
684 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
685 					idx, reg);
686 			radeon_cs_dump_packet(p, pkt);
687 			return r;
688 		}
689 		track->zb.robj = reloc->robj;
690 		track->zb.offset = idx_value;
691 		track->zb_dirty = true;
692 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
693 		break;
694 	case R300_TX_OFFSET_0:
695 	case R300_TX_OFFSET_0+4:
696 	case R300_TX_OFFSET_0+8:
697 	case R300_TX_OFFSET_0+12:
698 	case R300_TX_OFFSET_0+16:
699 	case R300_TX_OFFSET_0+20:
700 	case R300_TX_OFFSET_0+24:
701 	case R300_TX_OFFSET_0+28:
702 	case R300_TX_OFFSET_0+32:
703 	case R300_TX_OFFSET_0+36:
704 	case R300_TX_OFFSET_0+40:
705 	case R300_TX_OFFSET_0+44:
706 	case R300_TX_OFFSET_0+48:
707 	case R300_TX_OFFSET_0+52:
708 	case R300_TX_OFFSET_0+56:
709 	case R300_TX_OFFSET_0+60:
710 		i = (reg - R300_TX_OFFSET_0) >> 2;
711 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
712 		if (r) {
713 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
714 					idx, reg);
715 			radeon_cs_dump_packet(p, pkt);
716 			return r;
717 		}
718 
719 		if (p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) {
720 			ib[idx] = (idx_value & 31) | /* keep the 1st 5 bits */
721 				  ((idx_value & ~31) + (u32)reloc->gpu_offset);
722 		} else {
723 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
724 				tile_flags |= R300_TXO_MACRO_TILE;
725 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
726 				tile_flags |= R300_TXO_MICRO_TILE;
727 			else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
728 				tile_flags |= R300_TXO_MICRO_TILE_SQUARE;
729 
730 			tmp = idx_value + ((u32)reloc->gpu_offset);
731 			tmp |= tile_flags;
732 			ib[idx] = tmp;
733 		}
734 		track->textures[i].robj = reloc->robj;
735 		track->tex_dirty = true;
736 		break;
737 	/* Tracked registers */
738 	case 0x2084:
739 		/* VAP_VF_CNTL */
740 		track->vap_vf_cntl = idx_value;
741 		break;
742 	case 0x20B4:
743 		/* VAP_VTX_SIZE */
744 		track->vtx_size = idx_value & 0x7F;
745 		break;
746 	case 0x2134:
747 		/* VAP_VF_MAX_VTX_INDX */
748 		track->max_indx = idx_value & 0x00FFFFFFUL;
749 		break;
750 	case 0x2088:
751 		/* VAP_ALT_NUM_VERTICES - only valid on r500 */
752 		if (p->rdev->family < CHIP_RV515)
753 			goto fail;
754 		track->vap_alt_nverts = idx_value & 0xFFFFFF;
755 		break;
756 	case 0x43E4:
757 		/* SC_SCISSOR1 */
758 		track->maxy = ((idx_value >> 13) & 0x1FFF) + 1;
759 		if (p->rdev->family < CHIP_RV515) {
760 			track->maxy -= 1440;
761 		}
762 		track->cb_dirty = true;
763 		track->zb_dirty = true;
764 		break;
765 	case 0x4E00:
766 		/* RB3D_CCTL */
767 		if ((idx_value & (1 << 10)) && /* CMASK_ENABLE */
768 		    p->rdev->cmask_filp != p->filp) {
769 			DRM_ERROR("Invalid RB3D_CCTL: Cannot enable CMASK.\n");
770 			return -EINVAL;
771 		}
772 		track->num_cb = ((idx_value >> 5) & 0x3) + 1;
773 		track->cb_dirty = true;
774 		break;
775 	case 0x4E38:
776 	case 0x4E3C:
777 	case 0x4E40:
778 	case 0x4E44:
779 		/* RB3D_COLORPITCH0 */
780 		/* RB3D_COLORPITCH1 */
781 		/* RB3D_COLORPITCH2 */
782 		/* RB3D_COLORPITCH3 */
783 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
784 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
785 			if (r) {
786 				DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
787 					  idx, reg);
788 				radeon_cs_dump_packet(p, pkt);
789 				return r;
790 			}
791 
792 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
793 				tile_flags |= R300_COLOR_TILE_ENABLE;
794 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
795 				tile_flags |= R300_COLOR_MICROTILE_ENABLE;
796 			else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
797 				tile_flags |= R300_COLOR_MICROTILE_SQUARE_ENABLE;
798 
799 			tmp = idx_value & ~(0x7 << 16);
800 			tmp |= tile_flags;
801 			ib[idx] = tmp;
802 		}
803 		i = (reg - 0x4E38) >> 2;
804 		track->cb[i].pitch = idx_value & 0x3FFE;
805 		switch (((idx_value >> 21) & 0xF)) {
806 		case 9:
807 		case 11:
808 		case 12:
809 			track->cb[i].cpp = 1;
810 			break;
811 		case 3:
812 		case 4:
813 		case 13:
814 		case 15:
815 			track->cb[i].cpp = 2;
816 			break;
817 		case 5:
818 			if (p->rdev->family < CHIP_RV515) {
819 				DRM_ERROR("Invalid color buffer format (%d)!\n",
820 					  ((idx_value >> 21) & 0xF));
821 				return -EINVAL;
822 			}
823 			/* Fall through. */
824 		case 6:
825 			track->cb[i].cpp = 4;
826 			break;
827 		case 10:
828 			track->cb[i].cpp = 8;
829 			break;
830 		case 7:
831 			track->cb[i].cpp = 16;
832 			break;
833 		default:
834 			DRM_ERROR("Invalid color buffer format (%d) !\n",
835 				  ((idx_value >> 21) & 0xF));
836 			return -EINVAL;
837 		}
838 		track->cb_dirty = true;
839 		break;
840 	case 0x4F00:
841 		/* ZB_CNTL */
842 		if (idx_value & 2) {
843 			track->z_enabled = true;
844 		} else {
845 			track->z_enabled = false;
846 		}
847 		track->zb_dirty = true;
848 		break;
849 	case 0x4F10:
850 		/* ZB_FORMAT */
851 		switch ((idx_value & 0xF)) {
852 		case 0:
853 		case 1:
854 			track->zb.cpp = 2;
855 			break;
856 		case 2:
857 			track->zb.cpp = 4;
858 			break;
859 		default:
860 			DRM_ERROR("Invalid z buffer format (%d) !\n",
861 				  (idx_value & 0xF));
862 			return -EINVAL;
863 		}
864 		track->zb_dirty = true;
865 		break;
866 	case 0x4F24:
867 		/* ZB_DEPTHPITCH */
868 		if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
869 			r = radeon_cs_packet_next_reloc(p, &reloc, 0);
870 			if (r) {
871 				DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
872 					  idx, reg);
873 				radeon_cs_dump_packet(p, pkt);
874 				return r;
875 			}
876 
877 			if (reloc->tiling_flags & RADEON_TILING_MACRO)
878 				tile_flags |= R300_DEPTHMACROTILE_ENABLE;
879 			if (reloc->tiling_flags & RADEON_TILING_MICRO)
880 				tile_flags |= R300_DEPTHMICROTILE_TILED;
881 			else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
882 				tile_flags |= R300_DEPTHMICROTILE_TILED_SQUARE;
883 
884 			tmp = idx_value & ~(0x7 << 16);
885 			tmp |= tile_flags;
886 			ib[idx] = tmp;
887 		}
888 		track->zb.pitch = idx_value & 0x3FFC;
889 		track->zb_dirty = true;
890 		break;
891 	case 0x4104:
892 		/* TX_ENABLE */
893 		for (i = 0; i < 16; i++) {
894 			bool enabled;
895 
896 			enabled = !!(idx_value & (1 << i));
897 			track->textures[i].enabled = enabled;
898 		}
899 		track->tex_dirty = true;
900 		break;
901 	case 0x44C0:
902 	case 0x44C4:
903 	case 0x44C8:
904 	case 0x44CC:
905 	case 0x44D0:
906 	case 0x44D4:
907 	case 0x44D8:
908 	case 0x44DC:
909 	case 0x44E0:
910 	case 0x44E4:
911 	case 0x44E8:
912 	case 0x44EC:
913 	case 0x44F0:
914 	case 0x44F4:
915 	case 0x44F8:
916 	case 0x44FC:
917 		/* TX_FORMAT1_[0-15] */
918 		i = (reg - 0x44C0) >> 2;
919 		tmp = (idx_value >> 25) & 0x3;
920 		track->textures[i].tex_coord_type = tmp;
921 		switch ((idx_value & 0x1F)) {
922 		case R300_TX_FORMAT_X8:
923 		case R300_TX_FORMAT_Y4X4:
924 		case R300_TX_FORMAT_Z3Y3X2:
925 			track->textures[i].cpp = 1;
926 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
927 			break;
928 		case R300_TX_FORMAT_X16:
929 		case R300_TX_FORMAT_FL_I16:
930 		case R300_TX_FORMAT_Y8X8:
931 		case R300_TX_FORMAT_Z5Y6X5:
932 		case R300_TX_FORMAT_Z6Y5X5:
933 		case R300_TX_FORMAT_W4Z4Y4X4:
934 		case R300_TX_FORMAT_W1Z5Y5X5:
935 		case R300_TX_FORMAT_D3DMFT_CxV8U8:
936 		case R300_TX_FORMAT_B8G8_B8G8:
937 		case R300_TX_FORMAT_G8R8_G8B8:
938 			track->textures[i].cpp = 2;
939 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
940 			break;
941 		case R300_TX_FORMAT_Y16X16:
942 		case R300_TX_FORMAT_FL_I16A16:
943 		case R300_TX_FORMAT_Z11Y11X10:
944 		case R300_TX_FORMAT_Z10Y11X11:
945 		case R300_TX_FORMAT_W8Z8Y8X8:
946 		case R300_TX_FORMAT_W2Z10Y10X10:
947 		case 0x17:
948 		case R300_TX_FORMAT_FL_I32:
949 		case 0x1e:
950 			track->textures[i].cpp = 4;
951 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
952 			break;
953 		case R300_TX_FORMAT_W16Z16Y16X16:
954 		case R300_TX_FORMAT_FL_R16G16B16A16:
955 		case R300_TX_FORMAT_FL_I32A32:
956 			track->textures[i].cpp = 8;
957 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
958 			break;
959 		case R300_TX_FORMAT_FL_R32G32B32A32:
960 			track->textures[i].cpp = 16;
961 			track->textures[i].compress_format = R100_TRACK_COMP_NONE;
962 			break;
963 		case R300_TX_FORMAT_DXT1:
964 			track->textures[i].cpp = 1;
965 			track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
966 			break;
967 		case R300_TX_FORMAT_ATI2N:
968 			if (p->rdev->family < CHIP_R420) {
969 				DRM_ERROR("Invalid texture format %u\n",
970 					  (idx_value & 0x1F));
971 				return -EINVAL;
972 			}
973 			/* The same rules apply as for DXT3/5. */
974 			/* Fall through. */
975 		case R300_TX_FORMAT_DXT3:
976 		case R300_TX_FORMAT_DXT5:
977 			track->textures[i].cpp = 1;
978 			track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
979 			break;
980 		default:
981 			DRM_ERROR("Invalid texture format %u\n",
982 				  (idx_value & 0x1F));
983 			return -EINVAL;
984 		}
985 		track->tex_dirty = true;
986 		break;
987 	case 0x4400:
988 	case 0x4404:
989 	case 0x4408:
990 	case 0x440C:
991 	case 0x4410:
992 	case 0x4414:
993 	case 0x4418:
994 	case 0x441C:
995 	case 0x4420:
996 	case 0x4424:
997 	case 0x4428:
998 	case 0x442C:
999 	case 0x4430:
1000 	case 0x4434:
1001 	case 0x4438:
1002 	case 0x443C:
1003 		/* TX_FILTER0_[0-15] */
1004 		i = (reg - 0x4400) >> 2;
1005 		tmp = idx_value & 0x7;
1006 		if (tmp == 2 || tmp == 4 || tmp == 6) {
1007 			track->textures[i].roundup_w = false;
1008 		}
1009 		tmp = (idx_value >> 3) & 0x7;
1010 		if (tmp == 2 || tmp == 4 || tmp == 6) {
1011 			track->textures[i].roundup_h = false;
1012 		}
1013 		track->tex_dirty = true;
1014 		break;
1015 	case 0x4500:
1016 	case 0x4504:
1017 	case 0x4508:
1018 	case 0x450C:
1019 	case 0x4510:
1020 	case 0x4514:
1021 	case 0x4518:
1022 	case 0x451C:
1023 	case 0x4520:
1024 	case 0x4524:
1025 	case 0x4528:
1026 	case 0x452C:
1027 	case 0x4530:
1028 	case 0x4534:
1029 	case 0x4538:
1030 	case 0x453C:
1031 		/* TX_FORMAT2_[0-15] */
1032 		i = (reg - 0x4500) >> 2;
1033 		tmp = idx_value & 0x3FFF;
1034 		track->textures[i].pitch = tmp + 1;
1035 		if (p->rdev->family >= CHIP_RV515) {
1036 			tmp = ((idx_value >> 15) & 1) << 11;
1037 			track->textures[i].width_11 = tmp;
1038 			tmp = ((idx_value >> 16) & 1) << 11;
1039 			track->textures[i].height_11 = tmp;
1040 
1041 			/* ATI1N */
1042 			if (idx_value & (1 << 14)) {
1043 				/* The same rules apply as for DXT1. */
1044 				track->textures[i].compress_format =
1045 					R100_TRACK_COMP_DXT1;
1046 			}
1047 		} else if (idx_value & (1 << 14)) {
1048 			DRM_ERROR("Forbidden bit TXFORMAT_MSB\n");
1049 			return -EINVAL;
1050 		}
1051 		track->tex_dirty = true;
1052 		break;
1053 	case 0x4480:
1054 	case 0x4484:
1055 	case 0x4488:
1056 	case 0x448C:
1057 	case 0x4490:
1058 	case 0x4494:
1059 	case 0x4498:
1060 	case 0x449C:
1061 	case 0x44A0:
1062 	case 0x44A4:
1063 	case 0x44A8:
1064 	case 0x44AC:
1065 	case 0x44B0:
1066 	case 0x44B4:
1067 	case 0x44B8:
1068 	case 0x44BC:
1069 		/* TX_FORMAT0_[0-15] */
1070 		i = (reg - 0x4480) >> 2;
1071 		tmp = idx_value & 0x7FF;
1072 		track->textures[i].width = tmp + 1;
1073 		tmp = (idx_value >> 11) & 0x7FF;
1074 		track->textures[i].height = tmp + 1;
1075 		tmp = (idx_value >> 26) & 0xF;
1076 		track->textures[i].num_levels = tmp;
1077 		tmp = idx_value & (1 << 31);
1078 		track->textures[i].use_pitch = !!tmp;
1079 		tmp = (idx_value >> 22) & 0xF;
1080 		track->textures[i].txdepth = tmp;
1081 		track->tex_dirty = true;
1082 		break;
1083 	case R300_ZB_ZPASS_ADDR:
1084 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1085 		if (r) {
1086 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1087 					idx, reg);
1088 			radeon_cs_dump_packet(p, pkt);
1089 			return r;
1090 		}
1091 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1092 		break;
1093 	case 0x4e0c:
1094 		/* RB3D_COLOR_CHANNEL_MASK */
1095 		track->color_channel_mask = idx_value;
1096 		track->cb_dirty = true;
1097 		break;
1098 	case 0x43a4:
1099 		/* SC_HYPERZ_EN */
1100 		/* r300c emits this register - we need to disable hyperz for it
1101 		 * without complaining */
1102 		if (p->rdev->hyperz_filp != p->filp) {
1103 			if (idx_value & 0x1)
1104 				ib[idx] = idx_value & ~1;
1105 		}
1106 		break;
1107 	case 0x4f1c:
1108 		/* ZB_BW_CNTL */
1109 		track->zb_cb_clear = !!(idx_value & (1 << 5));
1110 		track->cb_dirty = true;
1111 		track->zb_dirty = true;
1112 		if (p->rdev->hyperz_filp != p->filp) {
1113 			if (idx_value & (R300_HIZ_ENABLE |
1114 					 R300_RD_COMP_ENABLE |
1115 					 R300_WR_COMP_ENABLE |
1116 					 R300_FAST_FILL_ENABLE))
1117 				goto fail;
1118 		}
1119 		break;
1120 	case 0x4e04:
1121 		/* RB3D_BLENDCNTL */
1122 		track->blend_read_enable = !!(idx_value & (1 << 2));
1123 		track->cb_dirty = true;
1124 		break;
1125 	case R300_RB3D_AARESOLVE_OFFSET:
1126 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1127 		if (r) {
1128 			DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1129 				  idx, reg);
1130 			radeon_cs_dump_packet(p, pkt);
1131 			return r;
1132 		}
1133 		track->aa.robj = reloc->robj;
1134 		track->aa.offset = idx_value;
1135 		track->aa_dirty = true;
1136 		ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1137 		break;
1138 	case R300_RB3D_AARESOLVE_PITCH:
1139 		track->aa.pitch = idx_value & 0x3FFE;
1140 		track->aa_dirty = true;
1141 		break;
1142 	case R300_RB3D_AARESOLVE_CTL:
1143 		track->aaresolve = idx_value & 0x1;
1144 		track->aa_dirty = true;
1145 		break;
1146 	case 0x4f30: /* ZB_MASK_OFFSET */
1147 	case 0x4f34: /* ZB_ZMASK_PITCH */
1148 	case 0x4f44: /* ZB_HIZ_OFFSET */
1149 	case 0x4f54: /* ZB_HIZ_PITCH */
1150 		if (idx_value && (p->rdev->hyperz_filp != p->filp))
1151 			goto fail;
1152 		break;
1153 	case 0x4028:
1154 		if (idx_value && (p->rdev->hyperz_filp != p->filp))
1155 			goto fail;
1156 		/* GB_Z_PEQ_CONFIG */
1157 		if (p->rdev->family >= CHIP_RV350)
1158 			break;
1159 		goto fail;
1160 		break;
1161 	case 0x4be8:
1162 		/* valid register only on RV530 */
1163 		if (p->rdev->family == CHIP_RV530)
1164 			break;
1165 		/* fallthrough do not move */
1166 	default:
1167 		goto fail;
1168 	}
1169 	return 0;
1170 fail:
1171 	pr_err("Forbidden register 0x%04X in cs at %d (val=%08x)\n",
1172 	       reg, idx, idx_value);
1173 	return -EINVAL;
1174 }
1175 
1176 static int r300_packet3_check(struct radeon_cs_parser *p,
1177 			      struct radeon_cs_packet *pkt)
1178 {
1179 	struct radeon_bo_list *reloc;
1180 	struct r100_cs_track *track;
1181 	volatile uint32_t *ib;
1182 	unsigned idx;
1183 	int r;
1184 
1185 	ib = p->ib.ptr;
1186 	idx = pkt->idx + 1;
1187 	track = (struct r100_cs_track *)p->track;
1188 	switch(pkt->opcode) {
1189 	case PACKET3_3D_LOAD_VBPNTR:
1190 		r = r100_packet3_load_vbpntr(p, pkt, idx);
1191 		if (r)
1192 			return r;
1193 		break;
1194 	case PACKET3_INDX_BUFFER:
1195 		r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1196 		if (r) {
1197 			DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1198 			radeon_cs_dump_packet(p, pkt);
1199 			return r;
1200 		}
1201 		ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1202 		r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1203 		if (r) {
1204 			return r;
1205 		}
1206 		break;
1207 	/* Draw packet */
1208 	case PACKET3_3D_DRAW_IMMD:
1209 		/* Number of dwords is vtx_size * (num_vertices - 1)
1210 		 * PRIM_WALK must be equal to 3 vertex data in embedded
1211 		 * in cmd stream */
1212 		if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1213 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1214 			return -EINVAL;
1215 		}
1216 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1217 		track->immd_dwords = pkt->count - 1;
1218 		r = r100_cs_track_check(p->rdev, track);
1219 		if (r) {
1220 			return r;
1221 		}
1222 		break;
1223 	case PACKET3_3D_DRAW_IMMD_2:
1224 		/* Number of dwords is vtx_size * (num_vertices - 1)
1225 		 * PRIM_WALK must be equal to 3 vertex data in embedded
1226 		 * in cmd stream */
1227 		if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1228 			DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1229 			return -EINVAL;
1230 		}
1231 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1232 		track->immd_dwords = pkt->count;
1233 		r = r100_cs_track_check(p->rdev, track);
1234 		if (r) {
1235 			return r;
1236 		}
1237 		break;
1238 	case PACKET3_3D_DRAW_VBUF:
1239 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1240 		r = r100_cs_track_check(p->rdev, track);
1241 		if (r) {
1242 			return r;
1243 		}
1244 		break;
1245 	case PACKET3_3D_DRAW_VBUF_2:
1246 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1247 		r = r100_cs_track_check(p->rdev, track);
1248 		if (r) {
1249 			return r;
1250 		}
1251 		break;
1252 	case PACKET3_3D_DRAW_INDX:
1253 		track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1254 		r = r100_cs_track_check(p->rdev, track);
1255 		if (r) {
1256 			return r;
1257 		}
1258 		break;
1259 	case PACKET3_3D_DRAW_INDX_2:
1260 		track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1261 		r = r100_cs_track_check(p->rdev, track);
1262 		if (r) {
1263 			return r;
1264 		}
1265 		break;
1266 	case PACKET3_3D_CLEAR_HIZ:
1267 	case PACKET3_3D_CLEAR_ZMASK:
1268 		if (p->rdev->hyperz_filp != p->filp)
1269 			return -EINVAL;
1270 		break;
1271 	case PACKET3_3D_CLEAR_CMASK:
1272 		if (p->rdev->cmask_filp != p->filp)
1273 			return -EINVAL;
1274 		break;
1275 	case PACKET3_NOP:
1276 		break;
1277 	default:
1278 		DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1279 		return -EINVAL;
1280 	}
1281 	return 0;
1282 }
1283 
1284 int r300_cs_parse(struct radeon_cs_parser *p)
1285 {
1286 	struct radeon_cs_packet pkt;
1287 	struct r100_cs_track *track;
1288 	int r;
1289 
1290 	track = kzalloc(sizeof(*track), GFP_KERNEL);
1291 	if (track == NULL)
1292 		return -ENOMEM;
1293 	r100_cs_track_clear(p->rdev, track);
1294 	p->track = track;
1295 	do {
1296 		r = radeon_cs_packet_parse(p, &pkt, p->idx);
1297 		if (r) {
1298 			return r;
1299 		}
1300 		p->idx += pkt.count + 2;
1301 		switch (pkt.type) {
1302 		case RADEON_PACKET_TYPE0:
1303 			r = r100_cs_parse_packet0(p, &pkt,
1304 						  p->rdev->config.r300.reg_safe_bm,
1305 						  p->rdev->config.r300.reg_safe_bm_size,
1306 						  &r300_packet0_check);
1307 			break;
1308 		case RADEON_PACKET_TYPE2:
1309 			break;
1310 		case RADEON_PACKET_TYPE3:
1311 			r = r300_packet3_check(p, &pkt);
1312 			break;
1313 		default:
1314 			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1315 			return -EINVAL;
1316 		}
1317 		if (r) {
1318 			return r;
1319 		}
1320 	} while (p->idx < p->chunk_ib->length_dw);
1321 	return 0;
1322 }
1323 
1324 void r300_set_reg_safe(struct radeon_device *rdev)
1325 {
1326 	rdev->config.r300.reg_safe_bm = r300_reg_safe_bm;
1327 	rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r300_reg_safe_bm);
1328 }
1329 
1330 void r300_mc_program(struct radeon_device *rdev)
1331 {
1332 	struct r100_mc_save save;
1333 	int r;
1334 
1335 	r = r100_debugfs_mc_info_init(rdev);
1336 	if (r) {
1337 		dev_err(rdev->dev, "Failed to create r100_mc debugfs file.\n");
1338 	}
1339 
1340 	/* Stops all mc clients */
1341 	r100_mc_stop(rdev, &save);
1342 	if (rdev->flags & RADEON_IS_AGP) {
1343 		WREG32(R_00014C_MC_AGP_LOCATION,
1344 			S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
1345 			S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
1346 		WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
1347 		WREG32(R_00015C_AGP_BASE_2,
1348 			upper_32_bits(rdev->mc.agp_base) & 0xff);
1349 	} else {
1350 		WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
1351 		WREG32(R_000170_AGP_BASE, 0);
1352 		WREG32(R_00015C_AGP_BASE_2, 0);
1353 	}
1354 	/* Wait for mc idle */
1355 	if (r300_mc_wait_for_idle(rdev))
1356 		DRM_INFO("Failed to wait MC idle before programming MC.\n");
1357 	/* Program MC, should be a 32bits limited address space */
1358 	WREG32(R_000148_MC_FB_LOCATION,
1359 		S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
1360 		S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
1361 	r100_mc_resume(rdev, &save);
1362 }
1363 
1364 void r300_clock_startup(struct radeon_device *rdev)
1365 {
1366 	u32 tmp;
1367 
1368 	if (radeon_dynclks != -1 && radeon_dynclks)
1369 		radeon_legacy_set_clock_gating(rdev, 1);
1370 	/* We need to force on some of the block */
1371 	tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
1372 	tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
1373 	if ((rdev->family == CHIP_RV350) || (rdev->family == CHIP_RV380))
1374 		tmp |= S_00000D_FORCE_VAP(1);
1375 	WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
1376 }
1377 
1378 static int r300_startup(struct radeon_device *rdev)
1379 {
1380 	int r;
1381 
1382 	/* set common regs */
1383 	r100_set_common_regs(rdev);
1384 	/* program mc */
1385 	r300_mc_program(rdev);
1386 	/* Resume clock */
1387 	r300_clock_startup(rdev);
1388 	/* Initialize GPU configuration (# pipes, ...) */
1389 	r300_gpu_init(rdev);
1390 	/* Initialize GART (initialize after TTM so we can allocate
1391 	 * memory through TTM but finalize after TTM) */
1392 	if (rdev->flags & RADEON_IS_PCIE) {
1393 		r = rv370_pcie_gart_enable(rdev);
1394 		if (r)
1395 			return r;
1396 	}
1397 
1398 	if (rdev->family == CHIP_R300 ||
1399 	    rdev->family == CHIP_R350 ||
1400 	    rdev->family == CHIP_RV350)
1401 		r100_enable_bm(rdev);
1402 
1403 	if (rdev->flags & RADEON_IS_PCI) {
1404 		r = r100_pci_gart_enable(rdev);
1405 		if (r)
1406 			return r;
1407 	}
1408 
1409 	/* allocate wb buffer */
1410 	r = radeon_wb_init(rdev);
1411 	if (r)
1412 		return r;
1413 
1414 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
1415 	if (r) {
1416 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
1417 		return r;
1418 	}
1419 
1420 	/* Enable IRQ */
1421 	if (!rdev->irq.installed) {
1422 		r = radeon_irq_kms_init(rdev);
1423 		if (r)
1424 			return r;
1425 	}
1426 
1427 	r100_irq_set(rdev);
1428 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
1429 	/* 1M ring buffer */
1430 	r = r100_cp_init(rdev, 1024 * 1024);
1431 	if (r) {
1432 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
1433 		return r;
1434 	}
1435 
1436 	r = radeon_ib_pool_init(rdev);
1437 	if (r) {
1438 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
1439 		return r;
1440 	}
1441 
1442 	return 0;
1443 }
1444 
1445 int r300_resume(struct radeon_device *rdev)
1446 {
1447 	int r;
1448 
1449 	/* Make sur GART are not working */
1450 	if (rdev->flags & RADEON_IS_PCIE)
1451 		rv370_pcie_gart_disable(rdev);
1452 	if (rdev->flags & RADEON_IS_PCI)
1453 		r100_pci_gart_disable(rdev);
1454 	/* Resume clock before doing reset */
1455 	r300_clock_startup(rdev);
1456 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
1457 	if (radeon_asic_reset(rdev)) {
1458 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1459 			RREG32(R_000E40_RBBM_STATUS),
1460 			RREG32(R_0007C0_CP_STAT));
1461 	}
1462 	/* post */
1463 	radeon_combios_asic_init(rdev->ddev);
1464 	/* Resume clock after posting */
1465 	r300_clock_startup(rdev);
1466 	/* Initialize surface registers */
1467 	radeon_surface_init(rdev);
1468 
1469 	rdev->accel_working = true;
1470 	r = r300_startup(rdev);
1471 	if (r) {
1472 		rdev->accel_working = false;
1473 	}
1474 	return r;
1475 }
1476 
1477 int r300_suspend(struct radeon_device *rdev)
1478 {
1479 	radeon_pm_suspend(rdev);
1480 	r100_cp_disable(rdev);
1481 	radeon_wb_disable(rdev);
1482 	r100_irq_disable(rdev);
1483 	if (rdev->flags & RADEON_IS_PCIE)
1484 		rv370_pcie_gart_disable(rdev);
1485 	if (rdev->flags & RADEON_IS_PCI)
1486 		r100_pci_gart_disable(rdev);
1487 	return 0;
1488 }
1489 
1490 void r300_fini(struct radeon_device *rdev)
1491 {
1492 	radeon_pm_fini(rdev);
1493 	r100_cp_fini(rdev);
1494 	radeon_wb_fini(rdev);
1495 	radeon_ib_pool_fini(rdev);
1496 	radeon_gem_fini(rdev);
1497 	if (rdev->flags & RADEON_IS_PCIE)
1498 		rv370_pcie_gart_fini(rdev);
1499 	if (rdev->flags & RADEON_IS_PCI)
1500 		r100_pci_gart_fini(rdev);
1501 	radeon_agp_fini(rdev);
1502 	radeon_irq_kms_fini(rdev);
1503 	radeon_fence_driver_fini(rdev);
1504 	radeon_bo_fini(rdev);
1505 	radeon_atombios_fini(rdev);
1506 	kfree(rdev->bios);
1507 	rdev->bios = NULL;
1508 }
1509 
1510 int r300_init(struct radeon_device *rdev)
1511 {
1512 	int r;
1513 
1514 	/* Disable VGA */
1515 	r100_vga_render_disable(rdev);
1516 	/* Initialize scratch registers */
1517 	radeon_scratch_init(rdev);
1518 	/* Initialize surface registers */
1519 	radeon_surface_init(rdev);
1520 	/* TODO: disable VGA need to use VGA request */
1521 	/* restore some register to sane defaults */
1522 	r100_restore_sanity(rdev);
1523 	/* BIOS*/
1524 	if (!radeon_get_bios(rdev)) {
1525 		if (ASIC_IS_AVIVO(rdev))
1526 			return -EINVAL;
1527 	}
1528 	if (rdev->is_atom_bios) {
1529 		dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
1530 		return -EINVAL;
1531 	} else {
1532 		r = radeon_combios_init(rdev);
1533 		if (r)
1534 			return r;
1535 	}
1536 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
1537 	if (radeon_asic_reset(rdev)) {
1538 		dev_warn(rdev->dev,
1539 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1540 			RREG32(R_000E40_RBBM_STATUS),
1541 			RREG32(R_0007C0_CP_STAT));
1542 	}
1543 	/* check if cards are posted or not */
1544 	if (radeon_boot_test_post_card(rdev) == false)
1545 		return -EINVAL;
1546 	/* Set asic errata */
1547 	r300_errata(rdev);
1548 	/* Initialize clocks */
1549 	radeon_get_clock_info(rdev->ddev);
1550 	/* initialize AGP */
1551 	if (rdev->flags & RADEON_IS_AGP) {
1552 		r = radeon_agp_init(rdev);
1553 		if (r) {
1554 			radeon_agp_disable(rdev);
1555 		}
1556 	}
1557 	/* initialize memory controller */
1558 	r300_mc_init(rdev);
1559 	/* Fence driver */
1560 	r = radeon_fence_driver_init(rdev);
1561 	if (r)
1562 		return r;
1563 	/* Memory manager */
1564 	r = radeon_bo_init(rdev);
1565 	if (r)
1566 		return r;
1567 	if (rdev->flags & RADEON_IS_PCIE) {
1568 		r = rv370_pcie_gart_init(rdev);
1569 		if (r)
1570 			return r;
1571 	}
1572 	if (rdev->flags & RADEON_IS_PCI) {
1573 		r = r100_pci_gart_init(rdev);
1574 		if (r)
1575 			return r;
1576 	}
1577 	r300_set_reg_safe(rdev);
1578 
1579 	/* Initialize power management */
1580 	radeon_pm_init(rdev);
1581 
1582 	rdev->accel_working = true;
1583 	r = r300_startup(rdev);
1584 	if (r) {
1585 		/* Something went wrong with the accel init, so stop accel */
1586 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
1587 		r100_cp_fini(rdev);
1588 		radeon_wb_fini(rdev);
1589 		radeon_ib_pool_fini(rdev);
1590 		radeon_irq_kms_fini(rdev);
1591 		if (rdev->flags & RADEON_IS_PCIE)
1592 			rv370_pcie_gart_fini(rdev);
1593 		if (rdev->flags & RADEON_IS_PCI)
1594 			r100_pci_gart_fini(rdev);
1595 		radeon_agp_fini(rdev);
1596 		rdev->accel_working = false;
1597 	}
1598 	return 0;
1599 }
1600