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