Merge branch 'for-2.6.30' into for-2.6.31
[linux-2.6] / drivers / gpu / drm / radeon / r600_cp.c
1 /*
2  * Copyright 2008-2009 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the next
13  * paragraph) shall be included in all copies or substantial portions of the
14  * 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) AND/OR ITS SUPPLIERS 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 OTHER
22  * DEALINGS IN THE SOFTWARE.
23  *
24  * Authors:
25  *     Dave Airlie <airlied@redhat.com>
26  *     Alex Deucher <alexander.deucher@amd.com>
27  */
28
29 #include "drmP.h"
30 #include "drm.h"
31 #include "radeon_drm.h"
32 #include "radeon_drv.h"
33
34 #include "r600_microcode.h"
35
36 # define ATI_PCIGART_PAGE_SIZE          4096    /**< PCI GART page size */
37 # define ATI_PCIGART_PAGE_MASK          (~(ATI_PCIGART_PAGE_SIZE-1))
38
39 #define R600_PTE_VALID     (1 << 0)
40 #define R600_PTE_SYSTEM    (1 << 1)
41 #define R600_PTE_SNOOPED   (1 << 2)
42 #define R600_PTE_READABLE  (1 << 5)
43 #define R600_PTE_WRITEABLE (1 << 6)
44
45 /* MAX values used for gfx init */
46 #define R6XX_MAX_SH_GPRS           256
47 #define R6XX_MAX_TEMP_GPRS         16
48 #define R6XX_MAX_SH_THREADS        256
49 #define R6XX_MAX_SH_STACK_ENTRIES  4096
50 #define R6XX_MAX_BACKENDS          8
51 #define R6XX_MAX_BACKENDS_MASK     0xff
52 #define R6XX_MAX_SIMDS             8
53 #define R6XX_MAX_SIMDS_MASK        0xff
54 #define R6XX_MAX_PIPES             8
55 #define R6XX_MAX_PIPES_MASK        0xff
56
57 #define R7XX_MAX_SH_GPRS           256
58 #define R7XX_MAX_TEMP_GPRS         16
59 #define R7XX_MAX_SH_THREADS        256
60 #define R7XX_MAX_SH_STACK_ENTRIES  4096
61 #define R7XX_MAX_BACKENDS          8
62 #define R7XX_MAX_BACKENDS_MASK     0xff
63 #define R7XX_MAX_SIMDS             16
64 #define R7XX_MAX_SIMDS_MASK        0xffff
65 #define R7XX_MAX_PIPES             8
66 #define R7XX_MAX_PIPES_MASK        0xff
67
68 static int r600_do_wait_for_fifo(drm_radeon_private_t *dev_priv, int entries)
69 {
70         int i;
71
72         dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
73
74         for (i = 0; i < dev_priv->usec_timeout; i++) {
75                 int slots;
76                 if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770)
77                         slots = (RADEON_READ(R600_GRBM_STATUS)
78                                  & R700_CMDFIFO_AVAIL_MASK);
79                 else
80                         slots = (RADEON_READ(R600_GRBM_STATUS)
81                                  & R600_CMDFIFO_AVAIL_MASK);
82                 if (slots >= entries)
83                         return 0;
84                 DRM_UDELAY(1);
85         }
86         DRM_INFO("wait for fifo failed status : 0x%08X 0x%08X\n",
87                  RADEON_READ(R600_GRBM_STATUS),
88                  RADEON_READ(R600_GRBM_STATUS2));
89
90         return -EBUSY;
91 }
92
93 static int r600_do_wait_for_idle(drm_radeon_private_t *dev_priv)
94 {
95         int i, ret;
96
97         dev_priv->stats.boxes |= RADEON_BOX_WAIT_IDLE;
98
99         if ((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770)
100                 ret = r600_do_wait_for_fifo(dev_priv, 8);
101         else
102                 ret = r600_do_wait_for_fifo(dev_priv, 16);
103         if (ret)
104                 return ret;
105         for (i = 0; i < dev_priv->usec_timeout; i++) {
106                 if (!(RADEON_READ(R600_GRBM_STATUS) & R600_GUI_ACTIVE))
107                         return 0;
108                 DRM_UDELAY(1);
109         }
110         DRM_INFO("wait idle failed status : 0x%08X 0x%08X\n",
111                  RADEON_READ(R600_GRBM_STATUS),
112                  RADEON_READ(R600_GRBM_STATUS2));
113
114         return -EBUSY;
115 }
116
117 void r600_page_table_cleanup(struct drm_device *dev, struct drm_ati_pcigart_info *gart_info)
118 {
119         struct drm_sg_mem *entry = dev->sg;
120         int max_pages;
121         int pages;
122         int i;
123
124         if (!entry)
125                 return;
126
127         if (gart_info->bus_addr) {
128                 max_pages = (gart_info->table_size / sizeof(u64));
129                 pages = (entry->pages <= max_pages)
130                   ? entry->pages : max_pages;
131
132                 for (i = 0; i < pages; i++) {
133                         if (!entry->busaddr[i])
134                                 break;
135                         pci_unmap_page(dev->pdev, entry->busaddr[i],
136                                        PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
137                 }
138                 if (gart_info->gart_table_location == DRM_ATI_GART_MAIN)
139                         gart_info->bus_addr = 0;
140         }
141 }
142
143 /* R600 has page table setup */
144 int r600_page_table_init(struct drm_device *dev)
145 {
146         drm_radeon_private_t *dev_priv = dev->dev_private;
147         struct drm_ati_pcigart_info *gart_info = &dev_priv->gart_info;
148         struct drm_local_map *map = &gart_info->mapping;
149         struct drm_sg_mem *entry = dev->sg;
150         int ret = 0;
151         int i, j;
152         int pages;
153         u64 page_base;
154         dma_addr_t entry_addr;
155         int max_ati_pages, max_real_pages, gart_idx;
156
157         /* okay page table is available - lets rock */
158         max_ati_pages = (gart_info->table_size / sizeof(u64));
159         max_real_pages = max_ati_pages / (PAGE_SIZE / ATI_PCIGART_PAGE_SIZE);
160
161         pages = (entry->pages <= max_real_pages) ?
162                 entry->pages : max_real_pages;
163
164         memset_io((void __iomem *)map->handle, 0, max_ati_pages * sizeof(u64));
165
166         gart_idx = 0;
167         for (i = 0; i < pages; i++) {
168                 entry->busaddr[i] = pci_map_page(dev->pdev,
169                                                  entry->pagelist[i], 0,
170                                                  PAGE_SIZE,
171                                                  PCI_DMA_BIDIRECTIONAL);
172                 if (entry->busaddr[i] == 0) {
173                         DRM_ERROR("unable to map PCIGART pages!\n");
174                         r600_page_table_cleanup(dev, gart_info);
175                         goto done;
176                 }
177                 entry_addr = entry->busaddr[i];
178                 for (j = 0; j < (PAGE_SIZE / ATI_PCIGART_PAGE_SIZE); j++) {
179                         page_base = (u64) entry_addr & ATI_PCIGART_PAGE_MASK;
180                         page_base |= R600_PTE_VALID | R600_PTE_SYSTEM | R600_PTE_SNOOPED;
181                         page_base |= R600_PTE_READABLE | R600_PTE_WRITEABLE;
182
183                         DRM_WRITE64(map, gart_idx * sizeof(u64), page_base);
184
185                         gart_idx++;
186
187                         if ((i % 128) == 0)
188                                 DRM_DEBUG("page entry %d: 0x%016llx\n",
189                                     i, (unsigned long long)page_base);
190                         entry_addr += ATI_PCIGART_PAGE_SIZE;
191                 }
192         }
193         ret = 1;
194 done:
195         return ret;
196 }
197
198 static void r600_vm_flush_gart_range(struct drm_device *dev)
199 {
200         drm_radeon_private_t *dev_priv = dev->dev_private;
201         u32 resp, countdown = 1000;
202         RADEON_WRITE(R600_VM_CONTEXT0_INVALIDATION_LOW_ADDR, dev_priv->gart_vm_start >> 12);
203         RADEON_WRITE(R600_VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (dev_priv->gart_vm_start + dev_priv->gart_size - 1) >> 12);
204         RADEON_WRITE(R600_VM_CONTEXT0_REQUEST_RESPONSE, 2);
205
206         do {
207                 resp = RADEON_READ(R600_VM_CONTEXT0_REQUEST_RESPONSE);
208                 countdown--;
209                 DRM_UDELAY(1);
210         } while (((resp & 0xf0) == 0) && countdown);
211 }
212
213 static void r600_vm_init(struct drm_device *dev)
214 {
215         drm_radeon_private_t *dev_priv = dev->dev_private;
216         /* initialise the VM to use the page table we constructed up there */
217         u32 vm_c0, i;
218         u32 mc_rd_a;
219         u32 vm_l2_cntl, vm_l2_cntl3;
220         /* okay set up the PCIE aperture type thingo */
221         RADEON_WRITE(R600_MC_VM_SYSTEM_APERTURE_LOW_ADDR, dev_priv->gart_vm_start >> 12);
222         RADEON_WRITE(R600_MC_VM_SYSTEM_APERTURE_HIGH_ADDR, (dev_priv->gart_vm_start + dev_priv->gart_size - 1) >> 12);
223         RADEON_WRITE(R600_MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
224
225         /* setup MC RD a */
226         mc_rd_a = R600_MCD_L1_TLB | R600_MCD_L1_FRAG_PROC | R600_MCD_SYSTEM_ACCESS_MODE_IN_SYS |
227                 R600_MCD_SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU | R600_MCD_EFFECTIVE_L1_TLB_SIZE(5) |
228                 R600_MCD_EFFECTIVE_L1_QUEUE_SIZE(5) | R600_MCD_WAIT_L2_QUERY;
229
230         RADEON_WRITE(R600_MCD_RD_A_CNTL, mc_rd_a);
231         RADEON_WRITE(R600_MCD_RD_B_CNTL, mc_rd_a);
232
233         RADEON_WRITE(R600_MCD_WR_A_CNTL, mc_rd_a);
234         RADEON_WRITE(R600_MCD_WR_B_CNTL, mc_rd_a);
235
236         RADEON_WRITE(R600_MCD_RD_GFX_CNTL, mc_rd_a);
237         RADEON_WRITE(R600_MCD_WR_GFX_CNTL, mc_rd_a);
238
239         RADEON_WRITE(R600_MCD_RD_SYS_CNTL, mc_rd_a);
240         RADEON_WRITE(R600_MCD_WR_SYS_CNTL, mc_rd_a);
241
242         RADEON_WRITE(R600_MCD_RD_HDP_CNTL, mc_rd_a | R600_MCD_L1_STRICT_ORDERING);
243         RADEON_WRITE(R600_MCD_WR_HDP_CNTL, mc_rd_a /*| R600_MCD_L1_STRICT_ORDERING*/);
244
245         RADEON_WRITE(R600_MCD_RD_PDMA_CNTL, mc_rd_a);
246         RADEON_WRITE(R600_MCD_WR_PDMA_CNTL, mc_rd_a);
247
248         RADEON_WRITE(R600_MCD_RD_SEM_CNTL, mc_rd_a | R600_MCD_SEMAPHORE_MODE);
249         RADEON_WRITE(R600_MCD_WR_SEM_CNTL, mc_rd_a);
250
251         vm_l2_cntl = R600_VM_L2_CACHE_EN | R600_VM_L2_FRAG_PROC | R600_VM_ENABLE_PTE_CACHE_LRU_W;
252         vm_l2_cntl |= R600_VM_L2_CNTL_QUEUE_SIZE(7);
253         RADEON_WRITE(R600_VM_L2_CNTL, vm_l2_cntl);
254
255         RADEON_WRITE(R600_VM_L2_CNTL2, 0);
256         vm_l2_cntl3 = (R600_VM_L2_CNTL3_BANK_SELECT_0(0) |
257                        R600_VM_L2_CNTL3_BANK_SELECT_1(1) |
258                        R600_VM_L2_CNTL3_CACHE_UPDATE_MODE(2));
259         RADEON_WRITE(R600_VM_L2_CNTL3, vm_l2_cntl3);
260
261         vm_c0 = R600_VM_ENABLE_CONTEXT | R600_VM_PAGE_TABLE_DEPTH_FLAT;
262
263         RADEON_WRITE(R600_VM_CONTEXT0_CNTL, vm_c0);
264
265         vm_c0 &= ~R600_VM_ENABLE_CONTEXT;
266
267         /* disable all other contexts */
268         for (i = 1; i < 8; i++)
269                 RADEON_WRITE(R600_VM_CONTEXT0_CNTL + (i * 4), vm_c0);
270
271         RADEON_WRITE(R600_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, dev_priv->gart_info.bus_addr >> 12);
272         RADEON_WRITE(R600_VM_CONTEXT0_PAGE_TABLE_START_ADDR, dev_priv->gart_vm_start >> 12);
273         RADEON_WRITE(R600_VM_CONTEXT0_PAGE_TABLE_END_ADDR, (dev_priv->gart_vm_start + dev_priv->gart_size - 1) >> 12);
274
275         r600_vm_flush_gart_range(dev);
276 }
277
278 /* load r600 microcode */
279 static void r600_cp_load_microcode(drm_radeon_private_t *dev_priv)
280 {
281         int i;
282
283         r600_do_cp_stop(dev_priv);
284
285         RADEON_WRITE(R600_CP_RB_CNTL,
286                      R600_RB_NO_UPDATE |
287                      R600_RB_BLKSZ(15) |
288                      R600_RB_BUFSZ(3));
289
290         RADEON_WRITE(R600_GRBM_SOFT_RESET, R600_SOFT_RESET_CP);
291         RADEON_READ(R600_GRBM_SOFT_RESET);
292         DRM_UDELAY(15000);
293         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0);
294
295         RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
296
297         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600)) {
298                 DRM_INFO("Loading R600 CP Microcode\n");
299                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
300                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
301                                      R600_cp_microcode[i][0]);
302                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
303                                      R600_cp_microcode[i][1]);
304                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
305                                      R600_cp_microcode[i][2]);
306                 }
307
308                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
309                 DRM_INFO("Loading R600 PFP Microcode\n");
310                 for (i = 0; i < PFP_UCODE_SIZE; i++)
311                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, R600_pfp_microcode[i]);
312         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610)) {
313                 DRM_INFO("Loading RV610 CP Microcode\n");
314                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
315                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
316                                      RV610_cp_microcode[i][0]);
317                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
318                                      RV610_cp_microcode[i][1]);
319                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
320                                      RV610_cp_microcode[i][2]);
321                 }
322
323                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
324                 DRM_INFO("Loading RV610 PFP Microcode\n");
325                 for (i = 0; i < PFP_UCODE_SIZE; i++)
326                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV610_pfp_microcode[i]);
327         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630)) {
328                 DRM_INFO("Loading RV630 CP Microcode\n");
329                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
330                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
331                                      RV630_cp_microcode[i][0]);
332                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
333                                      RV630_cp_microcode[i][1]);
334                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
335                                      RV630_cp_microcode[i][2]);
336                 }
337
338                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
339                 DRM_INFO("Loading RV630 PFP Microcode\n");
340                 for (i = 0; i < PFP_UCODE_SIZE; i++)
341                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV630_pfp_microcode[i]);
342         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620)) {
343                 DRM_INFO("Loading RV620 CP Microcode\n");
344                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
345                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
346                                      RV620_cp_microcode[i][0]);
347                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
348                                      RV620_cp_microcode[i][1]);
349                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
350                                      RV620_cp_microcode[i][2]);
351                 }
352
353                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
354                 DRM_INFO("Loading RV620 PFP Microcode\n");
355                 for (i = 0; i < PFP_UCODE_SIZE; i++)
356                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV620_pfp_microcode[i]);
357         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV635)) {
358                 DRM_INFO("Loading RV635 CP Microcode\n");
359                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
360                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
361                                      RV635_cp_microcode[i][0]);
362                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
363                                      RV635_cp_microcode[i][1]);
364                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
365                                      RV635_cp_microcode[i][2]);
366                 }
367
368                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
369                 DRM_INFO("Loading RV635 PFP Microcode\n");
370                 for (i = 0; i < PFP_UCODE_SIZE; i++)
371                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV635_pfp_microcode[i]);
372         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV670)) {
373                 DRM_INFO("Loading RV670 CP Microcode\n");
374                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
375                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
376                                      RV670_cp_microcode[i][0]);
377                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
378                                      RV670_cp_microcode[i][1]);
379                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
380                                      RV670_cp_microcode[i][2]);
381                 }
382
383                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
384                 DRM_INFO("Loading RV670 PFP Microcode\n");
385                 for (i = 0; i < PFP_UCODE_SIZE; i++)
386                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV670_pfp_microcode[i]);
387         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780)) {
388                 DRM_INFO("Loading RS780 CP Microcode\n");
389                 for (i = 0; i < PM4_UCODE_SIZE; i++) {
390                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
391                                      RS780_cp_microcode[i][0]);
392                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
393                                      RS780_cp_microcode[i][1]);
394                         RADEON_WRITE(R600_CP_ME_RAM_DATA,
395                                      RS780_cp_microcode[i][2]);
396                 }
397
398                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
399                 DRM_INFO("Loading RS780 PFP Microcode\n");
400                 for (i = 0; i < PFP_UCODE_SIZE; i++)
401                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RS780_pfp_microcode[i]);
402         }
403         RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
404         RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
405         RADEON_WRITE(R600_CP_ME_RAM_RADDR, 0);
406
407 }
408
409 static void r700_vm_init(struct drm_device *dev)
410 {
411         drm_radeon_private_t *dev_priv = dev->dev_private;
412         /* initialise the VM to use the page table we constructed up there */
413         u32 vm_c0, i;
414         u32 mc_vm_md_l1;
415         u32 vm_l2_cntl, vm_l2_cntl3;
416         /* okay set up the PCIE aperture type thingo */
417         RADEON_WRITE(R700_MC_VM_SYSTEM_APERTURE_LOW_ADDR, dev_priv->gart_vm_start >> 12);
418         RADEON_WRITE(R700_MC_VM_SYSTEM_APERTURE_HIGH_ADDR, (dev_priv->gart_vm_start + dev_priv->gart_size - 1) >> 12);
419         RADEON_WRITE(R700_MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
420
421         mc_vm_md_l1 = R700_ENABLE_L1_TLB |
422             R700_ENABLE_L1_FRAGMENT_PROCESSING |
423             R700_SYSTEM_ACCESS_MODE_IN_SYS |
424             R700_SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
425             R700_EFFECTIVE_L1_TLB_SIZE(5) |
426             R700_EFFECTIVE_L1_QUEUE_SIZE(5);
427
428         RADEON_WRITE(R700_MC_VM_MD_L1_TLB0_CNTL, mc_vm_md_l1);
429         RADEON_WRITE(R700_MC_VM_MD_L1_TLB1_CNTL, mc_vm_md_l1);
430         RADEON_WRITE(R700_MC_VM_MD_L1_TLB2_CNTL, mc_vm_md_l1);
431         RADEON_WRITE(R700_MC_VM_MB_L1_TLB0_CNTL, mc_vm_md_l1);
432         RADEON_WRITE(R700_MC_VM_MB_L1_TLB1_CNTL, mc_vm_md_l1);
433         RADEON_WRITE(R700_MC_VM_MB_L1_TLB2_CNTL, mc_vm_md_l1);
434         RADEON_WRITE(R700_MC_VM_MB_L1_TLB3_CNTL, mc_vm_md_l1);
435
436         vm_l2_cntl = R600_VM_L2_CACHE_EN | R600_VM_L2_FRAG_PROC | R600_VM_ENABLE_PTE_CACHE_LRU_W;
437         vm_l2_cntl |= R700_VM_L2_CNTL_QUEUE_SIZE(7);
438         RADEON_WRITE(R600_VM_L2_CNTL, vm_l2_cntl);
439
440         RADEON_WRITE(R600_VM_L2_CNTL2, 0);
441         vm_l2_cntl3 = R700_VM_L2_CNTL3_BANK_SELECT(0) | R700_VM_L2_CNTL3_CACHE_UPDATE_MODE(2);
442         RADEON_WRITE(R600_VM_L2_CNTL3, vm_l2_cntl3);
443
444         vm_c0 = R600_VM_ENABLE_CONTEXT | R600_VM_PAGE_TABLE_DEPTH_FLAT;
445
446         RADEON_WRITE(R600_VM_CONTEXT0_CNTL, vm_c0);
447
448         vm_c0 &= ~R600_VM_ENABLE_CONTEXT;
449
450         /* disable all other contexts */
451         for (i = 1; i < 8; i++)
452                 RADEON_WRITE(R600_VM_CONTEXT0_CNTL + (i * 4), vm_c0);
453
454         RADEON_WRITE(R700_VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, dev_priv->gart_info.bus_addr >> 12);
455         RADEON_WRITE(R700_VM_CONTEXT0_PAGE_TABLE_START_ADDR, dev_priv->gart_vm_start >> 12);
456         RADEON_WRITE(R700_VM_CONTEXT0_PAGE_TABLE_END_ADDR, (dev_priv->gart_vm_start + dev_priv->gart_size - 1) >> 12);
457
458         r600_vm_flush_gart_range(dev);
459 }
460
461 /* load r600 microcode */
462 static void r700_cp_load_microcode(drm_radeon_private_t *dev_priv)
463 {
464         int i;
465
466         r600_do_cp_stop(dev_priv);
467
468         RADEON_WRITE(R600_CP_RB_CNTL,
469                      R600_RB_NO_UPDATE |
470                      (15 << 8) |
471                      (3 << 0));
472
473         RADEON_WRITE(R600_GRBM_SOFT_RESET, R600_SOFT_RESET_CP);
474         RADEON_READ(R600_GRBM_SOFT_RESET);
475         DRM_UDELAY(15000);
476         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0);
477
478
479         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV770)) {
480                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
481                 DRM_INFO("Loading RV770 PFP Microcode\n");
482                 for (i = 0; i < R700_PFP_UCODE_SIZE; i++)
483                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV770_pfp_microcode[i]);
484                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
485
486                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
487                 DRM_INFO("Loading RV770 CP Microcode\n");
488                 for (i = 0; i < R700_PM4_UCODE_SIZE; i++)
489                         RADEON_WRITE(R600_CP_ME_RAM_DATA, RV770_cp_microcode[i]);
490                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
491
492         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV730)) {
493                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
494                 DRM_INFO("Loading RV730 PFP Microcode\n");
495                 for (i = 0; i < R700_PFP_UCODE_SIZE; i++)
496                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV730_pfp_microcode[i]);
497                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
498
499                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
500                 DRM_INFO("Loading RV730 CP Microcode\n");
501                 for (i = 0; i < R700_PM4_UCODE_SIZE; i++)
502                         RADEON_WRITE(R600_CP_ME_RAM_DATA, RV730_cp_microcode[i]);
503                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
504
505         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV710)) {
506                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
507                 DRM_INFO("Loading RV710 PFP Microcode\n");
508                 for (i = 0; i < R700_PFP_UCODE_SIZE; i++)
509                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV710_pfp_microcode[i]);
510                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
511
512                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
513                 DRM_INFO("Loading RV710 CP Microcode\n");
514                 for (i = 0; i < R700_PM4_UCODE_SIZE; i++)
515                         RADEON_WRITE(R600_CP_ME_RAM_DATA, RV710_cp_microcode[i]);
516                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
517
518         }
519         RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
520         RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
521         RADEON_WRITE(R600_CP_ME_RAM_RADDR, 0);
522
523 }
524
525 static void r600_test_writeback(drm_radeon_private_t *dev_priv)
526 {
527         u32 tmp;
528
529         /* Start with assuming that writeback doesn't work */
530         dev_priv->writeback_works = 0;
531
532         /* Writeback doesn't seem to work everywhere, test it here and possibly
533          * enable it if it appears to work
534          */
535         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(1), 0);
536
537         RADEON_WRITE(R600_SCRATCH_REG1, 0xdeadbeef);
538
539         for (tmp = 0; tmp < dev_priv->usec_timeout; tmp++) {
540                 u32 val;
541
542                 val = radeon_read_ring_rptr(dev_priv, R600_SCRATCHOFF(1));
543                 if (val == 0xdeadbeef)
544                         break;
545                 DRM_UDELAY(1);
546         }
547
548         if (tmp < dev_priv->usec_timeout) {
549                 dev_priv->writeback_works = 1;
550                 DRM_INFO("writeback test succeeded in %d usecs\n", tmp);
551         } else {
552                 dev_priv->writeback_works = 0;
553                 DRM_INFO("writeback test failed\n");
554         }
555         if (radeon_no_wb == 1) {
556                 dev_priv->writeback_works = 0;
557                 DRM_INFO("writeback forced off\n");
558         }
559
560         if (!dev_priv->writeback_works) {
561                 /* Disable writeback to avoid unnecessary bus master transfer */
562                 RADEON_WRITE(R600_CP_RB_CNTL, RADEON_READ(R600_CP_RB_CNTL) |
563                              RADEON_RB_NO_UPDATE);
564                 RADEON_WRITE(R600_SCRATCH_UMSK, 0);
565         }
566 }
567
568 int r600_do_engine_reset(struct drm_device *dev)
569 {
570         drm_radeon_private_t *dev_priv = dev->dev_private;
571         u32 cp_ptr, cp_me_cntl, cp_rb_cntl;
572
573         DRM_INFO("Resetting GPU\n");
574
575         cp_ptr = RADEON_READ(R600_CP_RB_WPTR);
576         cp_me_cntl = RADEON_READ(R600_CP_ME_CNTL);
577         RADEON_WRITE(R600_CP_ME_CNTL, R600_CP_ME_HALT);
578
579         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0x7fff);
580         RADEON_READ(R600_GRBM_SOFT_RESET);
581         DRM_UDELAY(50);
582         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0);
583         RADEON_READ(R600_GRBM_SOFT_RESET);
584
585         RADEON_WRITE(R600_CP_RB_WPTR_DELAY, 0);
586         cp_rb_cntl = RADEON_READ(R600_CP_RB_CNTL);
587         RADEON_WRITE(R600_CP_RB_CNTL, R600_RB_RPTR_WR_ENA);
588
589         RADEON_WRITE(R600_CP_RB_RPTR_WR, cp_ptr);
590         RADEON_WRITE(R600_CP_RB_WPTR, cp_ptr);
591         RADEON_WRITE(R600_CP_RB_CNTL, cp_rb_cntl);
592         RADEON_WRITE(R600_CP_ME_CNTL, cp_me_cntl);
593
594         /* Reset the CP ring */
595         r600_do_cp_reset(dev_priv);
596
597         /* The CP is no longer running after an engine reset */
598         dev_priv->cp_running = 0;
599
600         /* Reset any pending vertex, indirect buffers */
601         radeon_freelist_reset(dev);
602
603         return 0;
604
605 }
606
607 static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
608                                              u32 num_backends,
609                                              u32 backend_disable_mask)
610 {
611         u32 backend_map = 0;
612         u32 enabled_backends_mask;
613         u32 enabled_backends_count;
614         u32 cur_pipe;
615         u32 swizzle_pipe[R6XX_MAX_PIPES];
616         u32 cur_backend;
617         u32 i;
618
619         if (num_tile_pipes > R6XX_MAX_PIPES)
620                 num_tile_pipes = R6XX_MAX_PIPES;
621         if (num_tile_pipes < 1)
622                 num_tile_pipes = 1;
623         if (num_backends > R6XX_MAX_BACKENDS)
624                 num_backends = R6XX_MAX_BACKENDS;
625         if (num_backends < 1)
626                 num_backends = 1;
627
628         enabled_backends_mask = 0;
629         enabled_backends_count = 0;
630         for (i = 0; i < R6XX_MAX_BACKENDS; ++i) {
631                 if (((backend_disable_mask >> i) & 1) == 0) {
632                         enabled_backends_mask |= (1 << i);
633                         ++enabled_backends_count;
634                 }
635                 if (enabled_backends_count == num_backends)
636                         break;
637         }
638
639         if (enabled_backends_count == 0) {
640                 enabled_backends_mask = 1;
641                 enabled_backends_count = 1;
642         }
643
644         if (enabled_backends_count != num_backends)
645                 num_backends = enabled_backends_count;
646
647         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES);
648         switch (num_tile_pipes) {
649         case 1:
650                 swizzle_pipe[0] = 0;
651                 break;
652         case 2:
653                 swizzle_pipe[0] = 0;
654                 swizzle_pipe[1] = 1;
655                 break;
656         case 3:
657                 swizzle_pipe[0] = 0;
658                 swizzle_pipe[1] = 1;
659                 swizzle_pipe[2] = 2;
660                 break;
661         case 4:
662                 swizzle_pipe[0] = 0;
663                 swizzle_pipe[1] = 1;
664                 swizzle_pipe[2] = 2;
665                 swizzle_pipe[3] = 3;
666                 break;
667         case 5:
668                 swizzle_pipe[0] = 0;
669                 swizzle_pipe[1] = 1;
670                 swizzle_pipe[2] = 2;
671                 swizzle_pipe[3] = 3;
672                 swizzle_pipe[4] = 4;
673                 break;
674         case 6:
675                 swizzle_pipe[0] = 0;
676                 swizzle_pipe[1] = 2;
677                 swizzle_pipe[2] = 4;
678                 swizzle_pipe[3] = 5;
679                 swizzle_pipe[4] = 1;
680                 swizzle_pipe[5] = 3;
681                 break;
682         case 7:
683                 swizzle_pipe[0] = 0;
684                 swizzle_pipe[1] = 2;
685                 swizzle_pipe[2] = 4;
686                 swizzle_pipe[3] = 6;
687                 swizzle_pipe[4] = 1;
688                 swizzle_pipe[5] = 3;
689                 swizzle_pipe[6] = 5;
690                 break;
691         case 8:
692                 swizzle_pipe[0] = 0;
693                 swizzle_pipe[1] = 2;
694                 swizzle_pipe[2] = 4;
695                 swizzle_pipe[3] = 6;
696                 swizzle_pipe[4] = 1;
697                 swizzle_pipe[5] = 3;
698                 swizzle_pipe[6] = 5;
699                 swizzle_pipe[7] = 7;
700                 break;
701         }
702
703         cur_backend = 0;
704         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
705                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
706                         cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
707
708                 backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
709
710                 cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
711         }
712
713         return backend_map;
714 }
715
716 static int r600_count_pipe_bits(uint32_t val)
717 {
718         int i, ret = 0;
719         for (i = 0; i < 32; i++) {
720                 ret += val & 1;
721                 val >>= 1;
722         }
723         return ret;
724 }
725
726 static void r600_gfx_init(struct drm_device *dev,
727                           drm_radeon_private_t *dev_priv)
728 {
729         int i, j, num_qd_pipes;
730         u32 sx_debug_1;
731         u32 tc_cntl;
732         u32 arb_pop;
733         u32 num_gs_verts_per_thread;
734         u32 vgt_gs_per_es;
735         u32 gs_prim_buffer_depth = 0;
736         u32 sq_ms_fifo_sizes;
737         u32 sq_config;
738         u32 sq_gpr_resource_mgmt_1 = 0;
739         u32 sq_gpr_resource_mgmt_2 = 0;
740         u32 sq_thread_resource_mgmt = 0;
741         u32 sq_stack_resource_mgmt_1 = 0;
742         u32 sq_stack_resource_mgmt_2 = 0;
743         u32 hdp_host_path_cntl;
744         u32 backend_map;
745         u32 gb_tiling_config = 0;
746         u32 cc_rb_backend_disable = 0;
747         u32 cc_gc_shader_pipe_config = 0;
748         u32 ramcfg;
749
750         /* setup chip specs */
751         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
752         case CHIP_R600:
753                 dev_priv->r600_max_pipes = 4;
754                 dev_priv->r600_max_tile_pipes = 8;
755                 dev_priv->r600_max_simds = 4;
756                 dev_priv->r600_max_backends = 4;
757                 dev_priv->r600_max_gprs = 256;
758                 dev_priv->r600_max_threads = 192;
759                 dev_priv->r600_max_stack_entries = 256;
760                 dev_priv->r600_max_hw_contexts = 8;
761                 dev_priv->r600_max_gs_threads = 16;
762                 dev_priv->r600_sx_max_export_size = 128;
763                 dev_priv->r600_sx_max_export_pos_size = 16;
764                 dev_priv->r600_sx_max_export_smx_size = 128;
765                 dev_priv->r600_sq_num_cf_insts = 2;
766                 break;
767         case CHIP_RV630:
768         case CHIP_RV635:
769                 dev_priv->r600_max_pipes = 2;
770                 dev_priv->r600_max_tile_pipes = 2;
771                 dev_priv->r600_max_simds = 3;
772                 dev_priv->r600_max_backends = 1;
773                 dev_priv->r600_max_gprs = 128;
774                 dev_priv->r600_max_threads = 192;
775                 dev_priv->r600_max_stack_entries = 128;
776                 dev_priv->r600_max_hw_contexts = 8;
777                 dev_priv->r600_max_gs_threads = 4;
778                 dev_priv->r600_sx_max_export_size = 128;
779                 dev_priv->r600_sx_max_export_pos_size = 16;
780                 dev_priv->r600_sx_max_export_smx_size = 128;
781                 dev_priv->r600_sq_num_cf_insts = 2;
782                 break;
783         case CHIP_RV610:
784         case CHIP_RS780:
785         case CHIP_RV620:
786                 dev_priv->r600_max_pipes = 1;
787                 dev_priv->r600_max_tile_pipes = 1;
788                 dev_priv->r600_max_simds = 2;
789                 dev_priv->r600_max_backends = 1;
790                 dev_priv->r600_max_gprs = 128;
791                 dev_priv->r600_max_threads = 192;
792                 dev_priv->r600_max_stack_entries = 128;
793                 dev_priv->r600_max_hw_contexts = 4;
794                 dev_priv->r600_max_gs_threads = 4;
795                 dev_priv->r600_sx_max_export_size = 128;
796                 dev_priv->r600_sx_max_export_pos_size = 16;
797                 dev_priv->r600_sx_max_export_smx_size = 128;
798                 dev_priv->r600_sq_num_cf_insts = 1;
799                 break;
800         case CHIP_RV670:
801                 dev_priv->r600_max_pipes = 4;
802                 dev_priv->r600_max_tile_pipes = 4;
803                 dev_priv->r600_max_simds = 4;
804                 dev_priv->r600_max_backends = 4;
805                 dev_priv->r600_max_gprs = 192;
806                 dev_priv->r600_max_threads = 192;
807                 dev_priv->r600_max_stack_entries = 256;
808                 dev_priv->r600_max_hw_contexts = 8;
809                 dev_priv->r600_max_gs_threads = 16;
810                 dev_priv->r600_sx_max_export_size = 128;
811                 dev_priv->r600_sx_max_export_pos_size = 16;
812                 dev_priv->r600_sx_max_export_smx_size = 128;
813                 dev_priv->r600_sq_num_cf_insts = 2;
814                 break;
815         default:
816                 break;
817         }
818
819         /* Initialize HDP */
820         j = 0;
821         for (i = 0; i < 32; i++) {
822                 RADEON_WRITE((0x2c14 + j), 0x00000000);
823                 RADEON_WRITE((0x2c18 + j), 0x00000000);
824                 RADEON_WRITE((0x2c1c + j), 0x00000000);
825                 RADEON_WRITE((0x2c20 + j), 0x00000000);
826                 RADEON_WRITE((0x2c24 + j), 0x00000000);
827                 j += 0x18;
828         }
829
830         RADEON_WRITE(R600_GRBM_CNTL, R600_GRBM_READ_TIMEOUT(0xff));
831
832         /* setup tiling, simd, pipe config */
833         ramcfg = RADEON_READ(R600_RAMCFG);
834
835         switch (dev_priv->r600_max_tile_pipes) {
836         case 1:
837                 gb_tiling_config |= R600_PIPE_TILING(0);
838                 break;
839         case 2:
840                 gb_tiling_config |= R600_PIPE_TILING(1);
841                 break;
842         case 4:
843                 gb_tiling_config |= R600_PIPE_TILING(2);
844                 break;
845         case 8:
846                 gb_tiling_config |= R600_PIPE_TILING(3);
847                 break;
848         default:
849                 break;
850         }
851
852         gb_tiling_config |= R600_BANK_TILING((ramcfg >> R600_NOOFBANK_SHIFT) & R600_NOOFBANK_MASK);
853
854         gb_tiling_config |= R600_GROUP_SIZE(0);
855
856         if (((ramcfg >> R600_NOOFROWS_SHIFT) & R600_NOOFROWS_MASK) > 3) {
857                 gb_tiling_config |= R600_ROW_TILING(3);
858                 gb_tiling_config |= R600_SAMPLE_SPLIT(3);
859         } else {
860                 gb_tiling_config |=
861                         R600_ROW_TILING(((ramcfg >> R600_NOOFROWS_SHIFT) & R600_NOOFROWS_MASK));
862                 gb_tiling_config |=
863                         R600_SAMPLE_SPLIT(((ramcfg >> R600_NOOFROWS_SHIFT) & R600_NOOFROWS_MASK));
864         }
865
866         gb_tiling_config |= R600_BANK_SWAPS(1);
867
868         backend_map = r600_get_tile_pipe_to_backend_map(dev_priv->r600_max_tile_pipes,
869                                                         dev_priv->r600_max_backends,
870                                                         (0xff << dev_priv->r600_max_backends) & 0xff);
871         gb_tiling_config |= R600_BACKEND_MAP(backend_map);
872
873         cc_gc_shader_pipe_config =
874                 R600_INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << dev_priv->r600_max_pipes) & R6XX_MAX_PIPES_MASK);
875         cc_gc_shader_pipe_config |=
876                 R600_INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << dev_priv->r600_max_simds) & R6XX_MAX_SIMDS_MASK);
877
878         cc_rb_backend_disable =
879                 R600_BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << dev_priv->r600_max_backends) & R6XX_MAX_BACKENDS_MASK);
880
881         RADEON_WRITE(R600_GB_TILING_CONFIG,      gb_tiling_config);
882         RADEON_WRITE(R600_DCP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
883         RADEON_WRITE(R600_HDP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
884
885         RADEON_WRITE(R600_CC_RB_BACKEND_DISABLE,      cc_rb_backend_disable);
886         RADEON_WRITE(R600_CC_GC_SHADER_PIPE_CONFIG,   cc_gc_shader_pipe_config);
887         RADEON_WRITE(R600_GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
888
889         num_qd_pipes =
890                 R6XX_MAX_BACKENDS - r600_count_pipe_bits(cc_gc_shader_pipe_config & R600_INACTIVE_QD_PIPES_MASK);
891         RADEON_WRITE(R600_VGT_OUT_DEALLOC_CNTL, (num_qd_pipes * 4) & R600_DEALLOC_DIST_MASK);
892         RADEON_WRITE(R600_VGT_VERTEX_REUSE_BLOCK_CNTL, ((num_qd_pipes * 4) - 2) & R600_VTX_REUSE_DEPTH_MASK);
893
894         /* set HW defaults for 3D engine */
895         RADEON_WRITE(R600_CP_QUEUE_THRESHOLDS, (R600_ROQ_IB1_START(0x16) |
896                                                 R600_ROQ_IB2_START(0x2b)));
897
898         RADEON_WRITE(R600_CP_MEQ_THRESHOLDS, (R600_MEQ_END(0x40) |
899                                               R600_ROQ_END(0x40)));
900
901         RADEON_WRITE(R600_TA_CNTL_AUX, (R600_DISABLE_CUBE_ANISO |
902                                         R600_SYNC_GRADIENT |
903                                         R600_SYNC_WALKER |
904                                         R600_SYNC_ALIGNER));
905
906         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV670)
907                 RADEON_WRITE(R600_ARB_GDEC_RD_CNTL, 0x00000021);
908
909         sx_debug_1 = RADEON_READ(R600_SX_DEBUG_1);
910         sx_debug_1 |= R600_SMX_EVENT_RELEASE;
911         if (((dev_priv->flags & RADEON_FAMILY_MASK) > CHIP_R600))
912                 sx_debug_1 |= R600_ENABLE_NEW_SMX_ADDRESS;
913         RADEON_WRITE(R600_SX_DEBUG_1, sx_debug_1);
914
915         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600) ||
916             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630) ||
917             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
918             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
919             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780))
920                 RADEON_WRITE(R600_DB_DEBUG, R600_PREZ_MUST_WAIT_FOR_POSTZ_DONE);
921         else
922                 RADEON_WRITE(R600_DB_DEBUG, 0);
923
924         RADEON_WRITE(R600_DB_WATERMARKS, (R600_DEPTH_FREE(4) |
925                                           R600_DEPTH_FLUSH(16) |
926                                           R600_DEPTH_PENDING_FREE(4) |
927                                           R600_DEPTH_CACHELINE_FREE(16)));
928         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
929         RADEON_WRITE(R600_VGT_NUM_INSTANCES, 0);
930
931         RADEON_WRITE(R600_SPI_CONFIG_CNTL, R600_GPR_WRITE_PRIORITY(0));
932         RADEON_WRITE(R600_SPI_CONFIG_CNTL_1, R600_VTX_DONE_DELAY(0));
933
934         sq_ms_fifo_sizes = RADEON_READ(R600_SQ_MS_FIFO_SIZES);
935         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
936             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
937             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780)) {
938                 sq_ms_fifo_sizes = (R600_CACHE_FIFO_SIZE(0xa) |
939                                     R600_FETCH_FIFO_HIWATER(0xa) |
940                                     R600_DONE_FIFO_HIWATER(0xe0) |
941                                     R600_ALU_UPDATE_FIFO_HIWATER(0x8));
942         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600) ||
943                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630)) {
944                 sq_ms_fifo_sizes &= ~R600_DONE_FIFO_HIWATER(0xff);
945                 sq_ms_fifo_sizes |= R600_DONE_FIFO_HIWATER(0x4);
946         }
947         RADEON_WRITE(R600_SQ_MS_FIFO_SIZES, sq_ms_fifo_sizes);
948
949         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
950          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
951          */
952         sq_config = RADEON_READ(R600_SQ_CONFIG);
953         sq_config &= ~(R600_PS_PRIO(3) |
954                        R600_VS_PRIO(3) |
955                        R600_GS_PRIO(3) |
956                        R600_ES_PRIO(3));
957         sq_config |= (R600_DX9_CONSTS |
958                       R600_VC_ENABLE |
959                       R600_PS_PRIO(0) |
960                       R600_VS_PRIO(1) |
961                       R600_GS_PRIO(2) |
962                       R600_ES_PRIO(3));
963
964         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600) {
965                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(124) |
966                                           R600_NUM_VS_GPRS(124) |
967                                           R600_NUM_CLAUSE_TEMP_GPRS(4));
968                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(0) |
969                                           R600_NUM_ES_GPRS(0));
970                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(136) |
971                                            R600_NUM_VS_THREADS(48) |
972                                            R600_NUM_GS_THREADS(4) |
973                                            R600_NUM_ES_THREADS(4));
974                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(128) |
975                                             R600_NUM_VS_STACK_ENTRIES(128));
976                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(0) |
977                                             R600_NUM_ES_STACK_ENTRIES(0));
978         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
979                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
980                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780)) {
981                 /* no vertex cache */
982                 sq_config &= ~R600_VC_ENABLE;
983
984                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(44) |
985                                           R600_NUM_VS_GPRS(44) |
986                                           R600_NUM_CLAUSE_TEMP_GPRS(2));
987                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(17) |
988                                           R600_NUM_ES_GPRS(17));
989                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(79) |
990                                            R600_NUM_VS_THREADS(78) |
991                                            R600_NUM_GS_THREADS(4) |
992                                            R600_NUM_ES_THREADS(31));
993                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(40) |
994                                             R600_NUM_VS_STACK_ENTRIES(40));
995                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(32) |
996                                             R600_NUM_ES_STACK_ENTRIES(16));
997         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630) ||
998                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV635)) {
999                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(44) |
1000                                           R600_NUM_VS_GPRS(44) |
1001                                           R600_NUM_CLAUSE_TEMP_GPRS(2));
1002                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(18) |
1003                                           R600_NUM_ES_GPRS(18));
1004                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(79) |
1005                                            R600_NUM_VS_THREADS(78) |
1006                                            R600_NUM_GS_THREADS(4) |
1007                                            R600_NUM_ES_THREADS(31));
1008                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(40) |
1009                                             R600_NUM_VS_STACK_ENTRIES(40));
1010                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(32) |
1011                                             R600_NUM_ES_STACK_ENTRIES(16));
1012         } else if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV670) {
1013                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(44) |
1014                                           R600_NUM_VS_GPRS(44) |
1015                                           R600_NUM_CLAUSE_TEMP_GPRS(2));
1016                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(17) |
1017                                           R600_NUM_ES_GPRS(17));
1018                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(79) |
1019                                            R600_NUM_VS_THREADS(78) |
1020                                            R600_NUM_GS_THREADS(4) |
1021                                            R600_NUM_ES_THREADS(31));
1022                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(64) |
1023                                             R600_NUM_VS_STACK_ENTRIES(64));
1024                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(64) |
1025                                             R600_NUM_ES_STACK_ENTRIES(64));
1026         }
1027
1028         RADEON_WRITE(R600_SQ_CONFIG, sq_config);
1029         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
1030         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
1031         RADEON_WRITE(R600_SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1032         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
1033         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
1034
1035         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
1036             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
1037             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780))
1038                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, R600_CACHE_INVALIDATION(R600_TC_ONLY));
1039         else
1040                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, R600_CACHE_INVALIDATION(R600_VC_AND_TC));
1041
1042         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_2S, (R600_S0_X(0xc) |
1043                                                     R600_S0_Y(0x4) |
1044                                                     R600_S1_X(0x4) |
1045                                                     R600_S1_Y(0xc)));
1046         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_4S, (R600_S0_X(0xe) |
1047                                                     R600_S0_Y(0xe) |
1048                                                     R600_S1_X(0x2) |
1049                                                     R600_S1_Y(0x2) |
1050                                                     R600_S2_X(0xa) |
1051                                                     R600_S2_Y(0x6) |
1052                                                     R600_S3_X(0x6) |
1053                                                     R600_S3_Y(0xa)));
1054         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_8S_WD0, (R600_S0_X(0xe) |
1055                                                         R600_S0_Y(0xb) |
1056                                                         R600_S1_X(0x4) |
1057                                                         R600_S1_Y(0xc) |
1058                                                         R600_S2_X(0x1) |
1059                                                         R600_S2_Y(0x6) |
1060                                                         R600_S3_X(0xa) |
1061                                                         R600_S3_Y(0xe)));
1062         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_8S_WD1, (R600_S4_X(0x6) |
1063                                                         R600_S4_Y(0x1) |
1064                                                         R600_S5_X(0x0) |
1065                                                         R600_S5_Y(0x0) |
1066                                                         R600_S6_X(0xb) |
1067                                                         R600_S6_Y(0x4) |
1068                                                         R600_S7_X(0x7) |
1069                                                         R600_S7_Y(0x8)));
1070
1071
1072         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1073         case CHIP_R600:
1074         case CHIP_RV630:
1075         case CHIP_RV635:
1076                 gs_prim_buffer_depth = 0;
1077                 break;
1078         case CHIP_RV610:
1079         case CHIP_RS780:
1080         case CHIP_RV620:
1081                 gs_prim_buffer_depth = 32;
1082                 break;
1083         case CHIP_RV670:
1084                 gs_prim_buffer_depth = 128;
1085                 break;
1086         default:
1087                 break;
1088         }
1089
1090         num_gs_verts_per_thread = dev_priv->r600_max_pipes * 16;
1091         vgt_gs_per_es = gs_prim_buffer_depth + num_gs_verts_per_thread;
1092         /* Max value for this is 256 */
1093         if (vgt_gs_per_es > 256)
1094                 vgt_gs_per_es = 256;
1095
1096         RADEON_WRITE(R600_VGT_ES_PER_GS, 128);
1097         RADEON_WRITE(R600_VGT_GS_PER_ES, vgt_gs_per_es);
1098         RADEON_WRITE(R600_VGT_GS_PER_VS, 2);
1099         RADEON_WRITE(R600_VGT_GS_VERTEX_REUSE, 16);
1100
1101         /* more default values. 2D/3D driver should adjust as needed */
1102         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE_STATE, 0);
1103         RADEON_WRITE(R600_VGT_STRMOUT_EN, 0);
1104         RADEON_WRITE(R600_SX_MISC, 0);
1105         RADEON_WRITE(R600_PA_SC_MODE_CNTL, 0);
1106         RADEON_WRITE(R600_PA_SC_AA_CONFIG, 0);
1107         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE, 0);
1108         RADEON_WRITE(R600_SPI_INPUT_Z, 0);
1109         RADEON_WRITE(R600_SPI_PS_IN_CONTROL_0, R600_NUM_INTERP(2));
1110         RADEON_WRITE(R600_CB_COLOR7_FRAG, 0);
1111
1112         /* clear render buffer base addresses */
1113         RADEON_WRITE(R600_CB_COLOR0_BASE, 0);
1114         RADEON_WRITE(R600_CB_COLOR1_BASE, 0);
1115         RADEON_WRITE(R600_CB_COLOR2_BASE, 0);
1116         RADEON_WRITE(R600_CB_COLOR3_BASE, 0);
1117         RADEON_WRITE(R600_CB_COLOR4_BASE, 0);
1118         RADEON_WRITE(R600_CB_COLOR5_BASE, 0);
1119         RADEON_WRITE(R600_CB_COLOR6_BASE, 0);
1120         RADEON_WRITE(R600_CB_COLOR7_BASE, 0);
1121
1122         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1123         case CHIP_RV610:
1124         case CHIP_RS780:
1125         case CHIP_RV620:
1126                 tc_cntl = R600_TC_L2_SIZE(8);
1127                 break;
1128         case CHIP_RV630:
1129         case CHIP_RV635:
1130                 tc_cntl = R600_TC_L2_SIZE(4);
1131                 break;
1132         case CHIP_R600:
1133                 tc_cntl = R600_TC_L2_SIZE(0) | R600_L2_DISABLE_LATE_HIT;
1134                 break;
1135         default:
1136                 tc_cntl = R600_TC_L2_SIZE(0);
1137                 break;
1138         }
1139
1140         RADEON_WRITE(R600_TC_CNTL, tc_cntl);
1141
1142         hdp_host_path_cntl = RADEON_READ(R600_HDP_HOST_PATH_CNTL);
1143         RADEON_WRITE(R600_HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1144
1145         arb_pop = RADEON_READ(R600_ARB_POP);
1146         arb_pop |= R600_ENABLE_TC128;
1147         RADEON_WRITE(R600_ARB_POP, arb_pop);
1148
1149         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
1150         RADEON_WRITE(R600_PA_CL_ENHANCE, (R600_CLIP_VTX_REORDER_ENA |
1151                                           R600_NUM_CLIP_SEQ(3)));
1152         RADEON_WRITE(R600_PA_SC_ENHANCE, R600_FORCE_EOV_MAX_CLK_CNT(4095));
1153
1154 }
1155
1156 static u32 r700_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
1157                                              u32 num_backends,
1158                                              u32 backend_disable_mask)
1159 {
1160         u32 backend_map = 0;
1161         u32 enabled_backends_mask;
1162         u32 enabled_backends_count;
1163         u32 cur_pipe;
1164         u32 swizzle_pipe[R7XX_MAX_PIPES];
1165         u32 cur_backend;
1166         u32 i;
1167
1168         if (num_tile_pipes > R7XX_MAX_PIPES)
1169                 num_tile_pipes = R7XX_MAX_PIPES;
1170         if (num_tile_pipes < 1)
1171                 num_tile_pipes = 1;
1172         if (num_backends > R7XX_MAX_BACKENDS)
1173                 num_backends = R7XX_MAX_BACKENDS;
1174         if (num_backends < 1)
1175                 num_backends = 1;
1176
1177         enabled_backends_mask = 0;
1178         enabled_backends_count = 0;
1179         for (i = 0; i < R7XX_MAX_BACKENDS; ++i) {
1180                 if (((backend_disable_mask >> i) & 1) == 0) {
1181                         enabled_backends_mask |= (1 << i);
1182                         ++enabled_backends_count;
1183                 }
1184                 if (enabled_backends_count == num_backends)
1185                         break;
1186         }
1187
1188         if (enabled_backends_count == 0) {
1189                 enabled_backends_mask = 1;
1190                 enabled_backends_count = 1;
1191         }
1192
1193         if (enabled_backends_count != num_backends)
1194                 num_backends = enabled_backends_count;
1195
1196         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R7XX_MAX_PIPES);
1197         switch (num_tile_pipes) {
1198         case 1:
1199                 swizzle_pipe[0] = 0;
1200                 break;
1201         case 2:
1202                 swizzle_pipe[0] = 0;
1203                 swizzle_pipe[1] = 1;
1204                 break;
1205         case 3:
1206                 swizzle_pipe[0] = 0;
1207                 swizzle_pipe[1] = 2;
1208                 swizzle_pipe[2] = 1;
1209                 break;
1210         case 4:
1211                 swizzle_pipe[0] = 0;
1212                 swizzle_pipe[1] = 2;
1213                 swizzle_pipe[2] = 3;
1214                 swizzle_pipe[3] = 1;
1215                 break;
1216         case 5:
1217                 swizzle_pipe[0] = 0;
1218                 swizzle_pipe[1] = 2;
1219                 swizzle_pipe[2] = 4;
1220                 swizzle_pipe[3] = 1;
1221                 swizzle_pipe[4] = 3;
1222                 break;
1223         case 6:
1224                 swizzle_pipe[0] = 0;
1225                 swizzle_pipe[1] = 2;
1226                 swizzle_pipe[2] = 4;
1227                 swizzle_pipe[3] = 5;
1228                 swizzle_pipe[4] = 3;
1229                 swizzle_pipe[5] = 1;
1230                 break;
1231         case 7:
1232                 swizzle_pipe[0] = 0;
1233                 swizzle_pipe[1] = 2;
1234                 swizzle_pipe[2] = 4;
1235                 swizzle_pipe[3] = 6;
1236                 swizzle_pipe[4] = 3;
1237                 swizzle_pipe[5] = 1;
1238                 swizzle_pipe[6] = 5;
1239                 break;
1240         case 8:
1241                 swizzle_pipe[0] = 0;
1242                 swizzle_pipe[1] = 2;
1243                 swizzle_pipe[2] = 4;
1244                 swizzle_pipe[3] = 6;
1245                 swizzle_pipe[4] = 3;
1246                 swizzle_pipe[5] = 1;
1247                 swizzle_pipe[6] = 7;
1248                 swizzle_pipe[7] = 5;
1249                 break;
1250         }
1251
1252         cur_backend = 0;
1253         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
1254                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
1255                         cur_backend = (cur_backend + 1) % R7XX_MAX_BACKENDS;
1256
1257                 backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
1258
1259                 cur_backend = (cur_backend + 1) % R7XX_MAX_BACKENDS;
1260         }
1261
1262         return backend_map;
1263 }
1264
1265 static void r700_gfx_init(struct drm_device *dev,
1266                           drm_radeon_private_t *dev_priv)
1267 {
1268         int i, j, num_qd_pipes;
1269         u32 sx_debug_1;
1270         u32 smx_dc_ctl0;
1271         u32 num_gs_verts_per_thread;
1272         u32 vgt_gs_per_es;
1273         u32 gs_prim_buffer_depth = 0;
1274         u32 sq_ms_fifo_sizes;
1275         u32 sq_config;
1276         u32 sq_thread_resource_mgmt;
1277         u32 hdp_host_path_cntl;
1278         u32 sq_dyn_gpr_size_simd_ab_0;
1279         u32 backend_map;
1280         u32 gb_tiling_config = 0;
1281         u32 cc_rb_backend_disable = 0;
1282         u32 cc_gc_shader_pipe_config = 0;
1283         u32 mc_arb_ramcfg;
1284         u32 db_debug4;
1285
1286         /* setup chip specs */
1287         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1288         case CHIP_RV770:
1289                 dev_priv->r600_max_pipes = 4;
1290                 dev_priv->r600_max_tile_pipes = 8;
1291                 dev_priv->r600_max_simds = 10;
1292                 dev_priv->r600_max_backends = 4;
1293                 dev_priv->r600_max_gprs = 256;
1294                 dev_priv->r600_max_threads = 248;
1295                 dev_priv->r600_max_stack_entries = 512;
1296                 dev_priv->r600_max_hw_contexts = 8;
1297                 dev_priv->r600_max_gs_threads = 16 * 2;
1298                 dev_priv->r600_sx_max_export_size = 128;
1299                 dev_priv->r600_sx_max_export_pos_size = 16;
1300                 dev_priv->r600_sx_max_export_smx_size = 112;
1301                 dev_priv->r600_sq_num_cf_insts = 2;
1302
1303                 dev_priv->r700_sx_num_of_sets = 7;
1304                 dev_priv->r700_sc_prim_fifo_size = 0xF9;
1305                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1306                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1307                 break;
1308         case CHIP_RV730:
1309                 dev_priv->r600_max_pipes = 2;
1310                 dev_priv->r600_max_tile_pipes = 4;
1311                 dev_priv->r600_max_simds = 8;
1312                 dev_priv->r600_max_backends = 2;
1313                 dev_priv->r600_max_gprs = 128;
1314                 dev_priv->r600_max_threads = 248;
1315                 dev_priv->r600_max_stack_entries = 256;
1316                 dev_priv->r600_max_hw_contexts = 8;
1317                 dev_priv->r600_max_gs_threads = 16 * 2;
1318                 dev_priv->r600_sx_max_export_size = 256;
1319                 dev_priv->r600_sx_max_export_pos_size = 32;
1320                 dev_priv->r600_sx_max_export_smx_size = 224;
1321                 dev_priv->r600_sq_num_cf_insts = 2;
1322
1323                 dev_priv->r700_sx_num_of_sets = 7;
1324                 dev_priv->r700_sc_prim_fifo_size = 0xf9;
1325                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1326                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1327                 break;
1328         case CHIP_RV710:
1329                 dev_priv->r600_max_pipes = 2;
1330                 dev_priv->r600_max_tile_pipes = 2;
1331                 dev_priv->r600_max_simds = 2;
1332                 dev_priv->r600_max_backends = 1;
1333                 dev_priv->r600_max_gprs = 256;
1334                 dev_priv->r600_max_threads = 192;
1335                 dev_priv->r600_max_stack_entries = 256;
1336                 dev_priv->r600_max_hw_contexts = 4;
1337                 dev_priv->r600_max_gs_threads = 8 * 2;
1338                 dev_priv->r600_sx_max_export_size = 128;
1339                 dev_priv->r600_sx_max_export_pos_size = 16;
1340                 dev_priv->r600_sx_max_export_smx_size = 112;
1341                 dev_priv->r600_sq_num_cf_insts = 1;
1342
1343                 dev_priv->r700_sx_num_of_sets = 7;
1344                 dev_priv->r700_sc_prim_fifo_size = 0x40;
1345                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1346                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1347                 break;
1348         default:
1349                 break;
1350         }
1351
1352         /* Initialize HDP */
1353         j = 0;
1354         for (i = 0; i < 32; i++) {
1355                 RADEON_WRITE((0x2c14 + j), 0x00000000);
1356                 RADEON_WRITE((0x2c18 + j), 0x00000000);
1357                 RADEON_WRITE((0x2c1c + j), 0x00000000);
1358                 RADEON_WRITE((0x2c20 + j), 0x00000000);
1359                 RADEON_WRITE((0x2c24 + j), 0x00000000);
1360                 j += 0x18;
1361         }
1362
1363         RADEON_WRITE(R600_GRBM_CNTL, R600_GRBM_READ_TIMEOUT(0xff));
1364
1365         /* setup tiling, simd, pipe config */
1366         mc_arb_ramcfg = RADEON_READ(R700_MC_ARB_RAMCFG);
1367
1368         switch (dev_priv->r600_max_tile_pipes) {
1369         case 1:
1370                 gb_tiling_config |= R600_PIPE_TILING(0);
1371                 break;
1372         case 2:
1373                 gb_tiling_config |= R600_PIPE_TILING(1);
1374                 break;
1375         case 4:
1376                 gb_tiling_config |= R600_PIPE_TILING(2);
1377                 break;
1378         case 8:
1379                 gb_tiling_config |= R600_PIPE_TILING(3);
1380                 break;
1381         default:
1382                 break;
1383         }
1384
1385         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV770)
1386                 gb_tiling_config |= R600_BANK_TILING(1);
1387         else
1388                 gb_tiling_config |= R600_BANK_TILING((mc_arb_ramcfg >> R700_NOOFBANK_SHIFT) & R700_NOOFBANK_MASK);
1389
1390         gb_tiling_config |= R600_GROUP_SIZE(0);
1391
1392         if (((mc_arb_ramcfg >> R700_NOOFROWS_SHIFT) & R700_NOOFROWS_MASK) > 3) {
1393                 gb_tiling_config |= R600_ROW_TILING(3);
1394                 gb_tiling_config |= R600_SAMPLE_SPLIT(3);
1395         } else {
1396                 gb_tiling_config |=
1397                         R600_ROW_TILING(((mc_arb_ramcfg >> R700_NOOFROWS_SHIFT) & R700_NOOFROWS_MASK));
1398                 gb_tiling_config |=
1399                         R600_SAMPLE_SPLIT(((mc_arb_ramcfg >> R700_NOOFROWS_SHIFT) & R700_NOOFROWS_MASK));
1400         }
1401
1402         gb_tiling_config |= R600_BANK_SWAPS(1);
1403
1404         backend_map = r700_get_tile_pipe_to_backend_map(dev_priv->r600_max_tile_pipes,
1405                                                         dev_priv->r600_max_backends,
1406                                                         (0xff << dev_priv->r600_max_backends) & 0xff);
1407         gb_tiling_config |= R600_BACKEND_MAP(backend_map);
1408
1409         cc_gc_shader_pipe_config =
1410                 R600_INACTIVE_QD_PIPES((R7XX_MAX_PIPES_MASK << dev_priv->r600_max_pipes) & R7XX_MAX_PIPES_MASK);
1411         cc_gc_shader_pipe_config |=
1412                 R600_INACTIVE_SIMDS((R7XX_MAX_SIMDS_MASK << dev_priv->r600_max_simds) & R7XX_MAX_SIMDS_MASK);
1413
1414         cc_rb_backend_disable =
1415                 R600_BACKEND_DISABLE((R7XX_MAX_BACKENDS_MASK << dev_priv->r600_max_backends) & R7XX_MAX_BACKENDS_MASK);
1416
1417         RADEON_WRITE(R600_GB_TILING_CONFIG,      gb_tiling_config);
1418         RADEON_WRITE(R600_DCP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
1419         RADEON_WRITE(R600_HDP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
1420
1421         RADEON_WRITE(R600_CC_RB_BACKEND_DISABLE,      cc_rb_backend_disable);
1422         RADEON_WRITE(R600_CC_GC_SHADER_PIPE_CONFIG,   cc_gc_shader_pipe_config);
1423         RADEON_WRITE(R600_GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
1424
1425         RADEON_WRITE(R700_CC_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1426         RADEON_WRITE(R700_CGTS_SYS_TCC_DISABLE, 0);
1427         RADEON_WRITE(R700_CGTS_TCC_DISABLE, 0);
1428         RADEON_WRITE(R700_CGTS_USER_SYS_TCC_DISABLE, 0);
1429         RADEON_WRITE(R700_CGTS_USER_TCC_DISABLE, 0);
1430
1431         num_qd_pipes =
1432                 R7XX_MAX_BACKENDS - r600_count_pipe_bits(cc_gc_shader_pipe_config & R600_INACTIVE_QD_PIPES_MASK);
1433         RADEON_WRITE(R600_VGT_OUT_DEALLOC_CNTL, (num_qd_pipes * 4) & R600_DEALLOC_DIST_MASK);
1434         RADEON_WRITE(R600_VGT_VERTEX_REUSE_BLOCK_CNTL, ((num_qd_pipes * 4) - 2) & R600_VTX_REUSE_DEPTH_MASK);
1435
1436         /* set HW defaults for 3D engine */
1437         RADEON_WRITE(R600_CP_QUEUE_THRESHOLDS, (R600_ROQ_IB1_START(0x16) |
1438                                                 R600_ROQ_IB2_START(0x2b)));
1439
1440         RADEON_WRITE(R600_CP_MEQ_THRESHOLDS, R700_STQ_SPLIT(0x30));
1441
1442         RADEON_WRITE(R600_TA_CNTL_AUX, (R600_DISABLE_CUBE_ANISO |
1443                                         R600_SYNC_GRADIENT |
1444                                         R600_SYNC_WALKER |
1445                                         R600_SYNC_ALIGNER));
1446
1447         sx_debug_1 = RADEON_READ(R700_SX_DEBUG_1);
1448         sx_debug_1 |= R700_ENABLE_NEW_SMX_ADDRESS;
1449         RADEON_WRITE(R700_SX_DEBUG_1, sx_debug_1);
1450
1451         smx_dc_ctl0 = RADEON_READ(R600_SMX_DC_CTL0);
1452         smx_dc_ctl0 &= ~R700_CACHE_DEPTH(0x1ff);
1453         smx_dc_ctl0 |= R700_CACHE_DEPTH((dev_priv->r700_sx_num_of_sets * 64) - 1);
1454         RADEON_WRITE(R600_SMX_DC_CTL0, smx_dc_ctl0);
1455
1456         RADEON_WRITE(R700_SMX_EVENT_CTL, (R700_ES_FLUSH_CTL(4) |
1457                                           R700_GS_FLUSH_CTL(4) |
1458                                           R700_ACK_FLUSH_CTL(3) |
1459                                           R700_SYNC_FLUSH_CTL));
1460
1461         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV770)
1462                 RADEON_WRITE(R700_DB_DEBUG3, R700_DB_CLK_OFF_DELAY(0x1f));
1463         else {
1464                 db_debug4 = RADEON_READ(RV700_DB_DEBUG4);
1465                 db_debug4 |= RV700_DISABLE_TILE_COVERED_FOR_PS_ITER;
1466                 RADEON_WRITE(RV700_DB_DEBUG4, db_debug4);
1467         }
1468
1469         RADEON_WRITE(R600_SX_EXPORT_BUFFER_SIZES, (R600_COLOR_BUFFER_SIZE((dev_priv->r600_sx_max_export_size / 4) - 1) |
1470                                                    R600_POSITION_BUFFER_SIZE((dev_priv->r600_sx_max_export_pos_size / 4) - 1) |
1471                                                    R600_SMX_BUFFER_SIZE((dev_priv->r600_sx_max_export_smx_size / 4) - 1)));
1472
1473         RADEON_WRITE(R700_PA_SC_FIFO_SIZE_R7XX, (R700_SC_PRIM_FIFO_SIZE(dev_priv->r700_sc_prim_fifo_size) |
1474                                                  R700_SC_HIZ_TILE_FIFO_SIZE(dev_priv->r700_sc_hiz_tile_fifo_size) |
1475                                                  R700_SC_EARLYZ_TILE_FIFO_SIZE(dev_priv->r700_sc_earlyz_tile_fifo_fize)));
1476
1477         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
1478
1479         RADEON_WRITE(R600_VGT_NUM_INSTANCES, 1);
1480
1481         RADEON_WRITE(R600_SPI_CONFIG_CNTL, R600_GPR_WRITE_PRIORITY(0));
1482
1483         RADEON_WRITE(R600_SPI_CONFIG_CNTL_1, R600_VTX_DONE_DELAY(4));
1484
1485         RADEON_WRITE(R600_CP_PERFMON_CNTL, 0);
1486
1487         sq_ms_fifo_sizes = (R600_CACHE_FIFO_SIZE(16 * dev_priv->r600_sq_num_cf_insts) |
1488                             R600_DONE_FIFO_HIWATER(0xe0) |
1489                             R600_ALU_UPDATE_FIFO_HIWATER(0x8));
1490         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1491         case CHIP_RV770:
1492                 sq_ms_fifo_sizes |= R600_FETCH_FIFO_HIWATER(0x1);
1493                 break;
1494         case CHIP_RV730:
1495         case CHIP_RV710:
1496         default:
1497                 sq_ms_fifo_sizes |= R600_FETCH_FIFO_HIWATER(0x4);
1498                 break;
1499         }
1500         RADEON_WRITE(R600_SQ_MS_FIFO_SIZES, sq_ms_fifo_sizes);
1501
1502         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
1503          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
1504          */
1505         sq_config = RADEON_READ(R600_SQ_CONFIG);
1506         sq_config &= ~(R600_PS_PRIO(3) |
1507                        R600_VS_PRIO(3) |
1508                        R600_GS_PRIO(3) |
1509                        R600_ES_PRIO(3));
1510         sq_config |= (R600_DX9_CONSTS |
1511                       R600_VC_ENABLE |
1512                       R600_EXPORT_SRC_C |
1513                       R600_PS_PRIO(0) |
1514                       R600_VS_PRIO(1) |
1515                       R600_GS_PRIO(2) |
1516                       R600_ES_PRIO(3));
1517         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV710)
1518                 /* no vertex cache */
1519                 sq_config &= ~R600_VC_ENABLE;
1520
1521         RADEON_WRITE(R600_SQ_CONFIG, sq_config);
1522
1523         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_1,  (R600_NUM_PS_GPRS((dev_priv->r600_max_gprs * 24)/64) |
1524                                                     R600_NUM_VS_GPRS((dev_priv->r600_max_gprs * 24)/64) |
1525                                                     R600_NUM_CLAUSE_TEMP_GPRS(((dev_priv->r600_max_gprs * 24)/64)/2)));
1526
1527         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_2,  (R600_NUM_GS_GPRS((dev_priv->r600_max_gprs * 7)/64) |
1528                                                     R600_NUM_ES_GPRS((dev_priv->r600_max_gprs * 7)/64)));
1529
1530         sq_thread_resource_mgmt = (R600_NUM_PS_THREADS((dev_priv->r600_max_threads * 4)/8) |
1531                                    R600_NUM_VS_THREADS((dev_priv->r600_max_threads * 2)/8) |
1532                                    R600_NUM_ES_THREADS((dev_priv->r600_max_threads * 1)/8));
1533         if (((dev_priv->r600_max_threads * 1) / 8) > dev_priv->r600_max_gs_threads)
1534                 sq_thread_resource_mgmt |= R600_NUM_GS_THREADS(dev_priv->r600_max_gs_threads);
1535         else
1536                 sq_thread_resource_mgmt |= R600_NUM_GS_THREADS((dev_priv->r600_max_gs_threads * 1)/8);
1537         RADEON_WRITE(R600_SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1538
1539         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_1, (R600_NUM_PS_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4) |
1540                                                      R600_NUM_VS_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4)));
1541
1542         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_2, (R600_NUM_GS_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4) |
1543                                                      R600_NUM_ES_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4)));
1544
1545         sq_dyn_gpr_size_simd_ab_0 = (R700_SIMDA_RING0((dev_priv->r600_max_gprs * 38)/64) |
1546                                      R700_SIMDA_RING1((dev_priv->r600_max_gprs * 38)/64) |
1547                                      R700_SIMDB_RING0((dev_priv->r600_max_gprs * 38)/64) |
1548                                      R700_SIMDB_RING1((dev_priv->r600_max_gprs * 38)/64));
1549
1550         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_0, sq_dyn_gpr_size_simd_ab_0);
1551         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_1, sq_dyn_gpr_size_simd_ab_0);
1552         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_2, sq_dyn_gpr_size_simd_ab_0);
1553         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_3, sq_dyn_gpr_size_simd_ab_0);
1554         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_4, sq_dyn_gpr_size_simd_ab_0);
1555         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_5, sq_dyn_gpr_size_simd_ab_0);
1556         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_6, sq_dyn_gpr_size_simd_ab_0);
1557         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_7, sq_dyn_gpr_size_simd_ab_0);
1558
1559         RADEON_WRITE(R700_PA_SC_FORCE_EOV_MAX_CNTS, (R700_FORCE_EOV_MAX_CLK_CNT(4095) |
1560                                                      R700_FORCE_EOV_MAX_REZ_CNT(255)));
1561
1562         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV710)
1563                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, (R600_CACHE_INVALIDATION(R600_TC_ONLY) |
1564                                                            R700_AUTO_INVLD_EN(R700_ES_AND_GS_AUTO)));
1565         else
1566                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, (R600_CACHE_INVALIDATION(R600_VC_AND_TC) |
1567                                                            R700_AUTO_INVLD_EN(R700_ES_AND_GS_AUTO)));
1568
1569         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1570         case CHIP_RV770:
1571         case CHIP_RV730:
1572                 gs_prim_buffer_depth = 384;
1573                 break;
1574         case CHIP_RV710:
1575                 gs_prim_buffer_depth = 128;
1576                 break;
1577         default:
1578                 break;
1579         }
1580
1581         num_gs_verts_per_thread = dev_priv->r600_max_pipes * 16;
1582         vgt_gs_per_es = gs_prim_buffer_depth + num_gs_verts_per_thread;
1583         /* Max value for this is 256 */
1584         if (vgt_gs_per_es > 256)
1585                 vgt_gs_per_es = 256;
1586
1587         RADEON_WRITE(R600_VGT_ES_PER_GS, 128);
1588         RADEON_WRITE(R600_VGT_GS_PER_ES, vgt_gs_per_es);
1589         RADEON_WRITE(R600_VGT_GS_PER_VS, 2);
1590
1591         /* more default values. 2D/3D driver should adjust as needed */
1592         RADEON_WRITE(R600_VGT_GS_VERTEX_REUSE, 16);
1593         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE_STATE, 0);
1594         RADEON_WRITE(R600_VGT_STRMOUT_EN, 0);
1595         RADEON_WRITE(R600_SX_MISC, 0);
1596         RADEON_WRITE(R600_PA_SC_MODE_CNTL, 0);
1597         RADEON_WRITE(R700_PA_SC_EDGERULE, 0xaaaaaaaa);
1598         RADEON_WRITE(R600_PA_SC_AA_CONFIG, 0);
1599         RADEON_WRITE(R600_PA_SC_CLIPRECT_RULE, 0xffff);
1600         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE, 0);
1601         RADEON_WRITE(R600_SPI_INPUT_Z, 0);
1602         RADEON_WRITE(R600_SPI_PS_IN_CONTROL_0, R600_NUM_INTERP(2));
1603         RADEON_WRITE(R600_CB_COLOR7_FRAG, 0);
1604
1605         /* clear render buffer base addresses */
1606         RADEON_WRITE(R600_CB_COLOR0_BASE, 0);
1607         RADEON_WRITE(R600_CB_COLOR1_BASE, 0);
1608         RADEON_WRITE(R600_CB_COLOR2_BASE, 0);
1609         RADEON_WRITE(R600_CB_COLOR3_BASE, 0);
1610         RADEON_WRITE(R600_CB_COLOR4_BASE, 0);
1611         RADEON_WRITE(R600_CB_COLOR5_BASE, 0);
1612         RADEON_WRITE(R600_CB_COLOR6_BASE, 0);
1613         RADEON_WRITE(R600_CB_COLOR7_BASE, 0);
1614
1615         RADEON_WRITE(R700_TCP_CNTL, 0);
1616
1617         hdp_host_path_cntl = RADEON_READ(R600_HDP_HOST_PATH_CNTL);
1618         RADEON_WRITE(R600_HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1619
1620         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
1621
1622         RADEON_WRITE(R600_PA_CL_ENHANCE, (R600_CLIP_VTX_REORDER_ENA |
1623                                           R600_NUM_CLIP_SEQ(3)));
1624
1625 }
1626
1627 static void r600_cp_init_ring_buffer(struct drm_device *dev,
1628                                        drm_radeon_private_t *dev_priv,
1629                                        struct drm_file *file_priv)
1630 {
1631         struct drm_radeon_master_private *master_priv;
1632         u32 ring_start;
1633         u64 rptr_addr;
1634
1635         if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770))
1636                 r700_gfx_init(dev, dev_priv);
1637         else
1638                 r600_gfx_init(dev, dev_priv);
1639
1640         RADEON_WRITE(R600_GRBM_SOFT_RESET, R600_SOFT_RESET_CP);
1641         RADEON_READ(R600_GRBM_SOFT_RESET);
1642         DRM_UDELAY(15000);
1643         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0);
1644
1645
1646         /* Set ring buffer size */
1647 #ifdef __BIG_ENDIAN
1648         RADEON_WRITE(R600_CP_RB_CNTL,
1649                      RADEON_BUF_SWAP_32BIT |
1650                      RADEON_RB_NO_UPDATE |
1651                      (dev_priv->ring.rptr_update_l2qw << 8) |
1652                      dev_priv->ring.size_l2qw);
1653 #else
1654         RADEON_WRITE(R600_CP_RB_CNTL,
1655                      RADEON_RB_NO_UPDATE |
1656                      (dev_priv->ring.rptr_update_l2qw << 8) |
1657                      dev_priv->ring.size_l2qw);
1658 #endif
1659
1660         RADEON_WRITE(R600_CP_SEM_WAIT_TIMER, 0x4);
1661
1662         /* Set the write pointer delay */
1663         RADEON_WRITE(R600_CP_RB_WPTR_DELAY, 0);
1664
1665 #ifdef __BIG_ENDIAN
1666         RADEON_WRITE(R600_CP_RB_CNTL,
1667                      RADEON_BUF_SWAP_32BIT |
1668                      RADEON_RB_NO_UPDATE |
1669                      RADEON_RB_RPTR_WR_ENA |
1670                      (dev_priv->ring.rptr_update_l2qw << 8) |
1671                      dev_priv->ring.size_l2qw);
1672 #else
1673         RADEON_WRITE(R600_CP_RB_CNTL,
1674                      RADEON_RB_NO_UPDATE |
1675                      RADEON_RB_RPTR_WR_ENA |
1676                      (dev_priv->ring.rptr_update_l2qw << 8) |
1677                      dev_priv->ring.size_l2qw);
1678 #endif
1679
1680         /* Initialize the ring buffer's read and write pointers */
1681         RADEON_WRITE(R600_CP_RB_RPTR_WR, 0);
1682         RADEON_WRITE(R600_CP_RB_WPTR, 0);
1683         SET_RING_HEAD(dev_priv, 0);
1684         dev_priv->ring.tail = 0;
1685
1686 #if __OS_HAS_AGP
1687         if (dev_priv->flags & RADEON_IS_AGP) {
1688                 rptr_addr = dev_priv->ring_rptr->offset
1689                         - dev->agp->base +
1690                         dev_priv->gart_vm_start;
1691         } else
1692 #endif
1693         {
1694                 rptr_addr = dev_priv->ring_rptr->offset
1695                         - ((unsigned long) dev->sg->virtual)
1696                         + dev_priv->gart_vm_start;
1697         }
1698         RADEON_WRITE(R600_CP_RB_RPTR_ADDR,
1699                      rptr_addr & 0xffffffff);
1700         RADEON_WRITE(R600_CP_RB_RPTR_ADDR_HI,
1701                      upper_32_bits(rptr_addr));
1702
1703 #ifdef __BIG_ENDIAN
1704         RADEON_WRITE(R600_CP_RB_CNTL,
1705                      RADEON_BUF_SWAP_32BIT |
1706                      (dev_priv->ring.rptr_update_l2qw << 8) |
1707                      dev_priv->ring.size_l2qw);
1708 #else
1709         RADEON_WRITE(R600_CP_RB_CNTL,
1710                      (dev_priv->ring.rptr_update_l2qw << 8) |
1711                      dev_priv->ring.size_l2qw);
1712 #endif
1713
1714 #if __OS_HAS_AGP
1715         if (dev_priv->flags & RADEON_IS_AGP) {
1716                 /* XXX */
1717                 radeon_write_agp_base(dev_priv, dev->agp->base);
1718
1719                 /* XXX */
1720                 radeon_write_agp_location(dev_priv,
1721                              (((dev_priv->gart_vm_start - 1 +
1722                                 dev_priv->gart_size) & 0xffff0000) |
1723                               (dev_priv->gart_vm_start >> 16)));
1724
1725                 ring_start = (dev_priv->cp_ring->offset
1726                               - dev->agp->base
1727                               + dev_priv->gart_vm_start);
1728         } else
1729 #endif
1730                 ring_start = (dev_priv->cp_ring->offset
1731                               - (unsigned long)dev->sg->virtual
1732                               + dev_priv->gart_vm_start);
1733
1734         RADEON_WRITE(R600_CP_RB_BASE, ring_start >> 8);
1735
1736         RADEON_WRITE(R600_CP_ME_CNTL, 0xff);
1737
1738         RADEON_WRITE(R600_CP_DEBUG, (1 << 27) | (1 << 28));
1739
1740         /* Initialize the scratch register pointer.  This will cause
1741          * the scratch register values to be written out to memory
1742          * whenever they are updated.
1743          *
1744          * We simply put this behind the ring read pointer, this works
1745          * with PCI GART as well as (whatever kind of) AGP GART
1746          */
1747         {
1748                 u64 scratch_addr;
1749
1750                 scratch_addr = RADEON_READ(R600_CP_RB_RPTR_ADDR);
1751                 scratch_addr |= ((u64)RADEON_READ(R600_CP_RB_RPTR_ADDR_HI)) << 32;
1752                 scratch_addr += R600_SCRATCH_REG_OFFSET;
1753                 scratch_addr >>= 8;
1754                 scratch_addr &= 0xffffffff;
1755
1756                 RADEON_WRITE(R600_SCRATCH_ADDR, (uint32_t)scratch_addr);
1757         }
1758
1759         RADEON_WRITE(R600_SCRATCH_UMSK, 0x7);
1760
1761         /* Turn on bus mastering */
1762         radeon_enable_bm(dev_priv);
1763
1764         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(0), 0);
1765         RADEON_WRITE(R600_LAST_FRAME_REG, 0);
1766
1767         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(1), 0);
1768         RADEON_WRITE(R600_LAST_DISPATCH_REG, 0);
1769
1770         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(2), 0);
1771         RADEON_WRITE(R600_LAST_CLEAR_REG, 0);
1772
1773         /* reset sarea copies of these */
1774         master_priv = file_priv->master->driver_priv;
1775         if (master_priv->sarea_priv) {
1776                 master_priv->sarea_priv->last_frame = 0;
1777                 master_priv->sarea_priv->last_dispatch = 0;
1778                 master_priv->sarea_priv->last_clear = 0;
1779         }
1780
1781         r600_do_wait_for_idle(dev_priv);
1782
1783 }
1784
1785 int r600_do_cleanup_cp(struct drm_device *dev)
1786 {
1787         drm_radeon_private_t *dev_priv = dev->dev_private;
1788         DRM_DEBUG("\n");
1789
1790         /* Make sure interrupts are disabled here because the uninstall ioctl
1791          * may not have been called from userspace and after dev_private
1792          * is freed, it's too late.
1793          */
1794         if (dev->irq_enabled)
1795                 drm_irq_uninstall(dev);
1796
1797 #if __OS_HAS_AGP
1798         if (dev_priv->flags & RADEON_IS_AGP) {
1799                 if (dev_priv->cp_ring != NULL) {
1800                         drm_core_ioremapfree(dev_priv->cp_ring, dev);
1801                         dev_priv->cp_ring = NULL;
1802                 }
1803                 if (dev_priv->ring_rptr != NULL) {
1804                         drm_core_ioremapfree(dev_priv->ring_rptr, dev);
1805                         dev_priv->ring_rptr = NULL;
1806                 }
1807                 if (dev->agp_buffer_map != NULL) {
1808                         drm_core_ioremapfree(dev->agp_buffer_map, dev);
1809                         dev->agp_buffer_map = NULL;
1810                 }
1811         } else
1812 #endif
1813         {
1814
1815                 if (dev_priv->gart_info.bus_addr)
1816                         r600_page_table_cleanup(dev, &dev_priv->gart_info);
1817
1818                 if (dev_priv->gart_info.gart_table_location == DRM_ATI_GART_FB) {
1819                         drm_core_ioremapfree(&dev_priv->gart_info.mapping, dev);
1820                         dev_priv->gart_info.addr = NULL;
1821                 }
1822         }
1823         /* only clear to the start of flags */
1824         memset(dev_priv, 0, offsetof(drm_radeon_private_t, flags));
1825
1826         return 0;
1827 }
1828
1829 int r600_do_init_cp(struct drm_device *dev, drm_radeon_init_t *init,
1830                     struct drm_file *file_priv)
1831 {
1832         drm_radeon_private_t *dev_priv = dev->dev_private;
1833         struct drm_radeon_master_private *master_priv = file_priv->master->driver_priv;
1834
1835         DRM_DEBUG("\n");
1836
1837         /* if we require new memory map but we don't have it fail */
1838         if ((dev_priv->flags & RADEON_NEW_MEMMAP) && !dev_priv->new_memmap) {
1839                 DRM_ERROR("Cannot initialise DRM on this card\nThis card requires a new X.org DDX for 3D\n");
1840                 r600_do_cleanup_cp(dev);
1841                 return -EINVAL;
1842         }
1843
1844         if (init->is_pci && (dev_priv->flags & RADEON_IS_AGP)) {
1845                 DRM_DEBUG("Forcing AGP card to PCI mode\n");
1846                 dev_priv->flags &= ~RADEON_IS_AGP;
1847                 /* The writeback test succeeds, but when writeback is enabled,
1848                  * the ring buffer read ptr update fails after first 128 bytes.
1849                  */
1850                 radeon_no_wb = 1;
1851         } else if (!(dev_priv->flags & (RADEON_IS_AGP | RADEON_IS_PCI | RADEON_IS_PCIE))
1852                  && !init->is_pci) {
1853                 DRM_DEBUG("Restoring AGP flag\n");
1854                 dev_priv->flags |= RADEON_IS_AGP;
1855         }
1856
1857         dev_priv->usec_timeout = init->usec_timeout;
1858         if (dev_priv->usec_timeout < 1 ||
1859             dev_priv->usec_timeout > RADEON_MAX_USEC_TIMEOUT) {
1860                 DRM_DEBUG("TIMEOUT problem!\n");
1861                 r600_do_cleanup_cp(dev);
1862                 return -EINVAL;
1863         }
1864
1865         /* Enable vblank on CRTC1 for older X servers
1866          */
1867         dev_priv->vblank_crtc = DRM_RADEON_VBLANK_CRTC1;
1868
1869         dev_priv->cp_mode = init->cp_mode;
1870
1871         /* We don't support anything other than bus-mastering ring mode,
1872          * but the ring can be in either AGP or PCI space for the ring
1873          * read pointer.
1874          */
1875         if ((init->cp_mode != RADEON_CSQ_PRIBM_INDDIS) &&
1876             (init->cp_mode != RADEON_CSQ_PRIBM_INDBM)) {
1877                 DRM_DEBUG("BAD cp_mode (%x)!\n", init->cp_mode);
1878                 r600_do_cleanup_cp(dev);
1879                 return -EINVAL;
1880         }
1881
1882         switch (init->fb_bpp) {
1883         case 16:
1884                 dev_priv->color_fmt = RADEON_COLOR_FORMAT_RGB565;
1885                 break;
1886         case 32:
1887         default:
1888                 dev_priv->color_fmt = RADEON_COLOR_FORMAT_ARGB8888;
1889                 break;
1890         }
1891         dev_priv->front_offset = init->front_offset;
1892         dev_priv->front_pitch = init->front_pitch;
1893         dev_priv->back_offset = init->back_offset;
1894         dev_priv->back_pitch = init->back_pitch;
1895
1896         dev_priv->ring_offset = init->ring_offset;
1897         dev_priv->ring_rptr_offset = init->ring_rptr_offset;
1898         dev_priv->buffers_offset = init->buffers_offset;
1899         dev_priv->gart_textures_offset = init->gart_textures_offset;
1900
1901         master_priv->sarea = drm_getsarea(dev);
1902         if (!master_priv->sarea) {
1903                 DRM_ERROR("could not find sarea!\n");
1904                 r600_do_cleanup_cp(dev);
1905                 return -EINVAL;
1906         }
1907
1908         dev_priv->cp_ring = drm_core_findmap(dev, init->ring_offset);
1909         if (!dev_priv->cp_ring) {
1910                 DRM_ERROR("could not find cp ring region!\n");
1911                 r600_do_cleanup_cp(dev);
1912                 return -EINVAL;
1913         }
1914         dev_priv->ring_rptr = drm_core_findmap(dev, init->ring_rptr_offset);
1915         if (!dev_priv->ring_rptr) {
1916                 DRM_ERROR("could not find ring read pointer!\n");
1917                 r600_do_cleanup_cp(dev);
1918                 return -EINVAL;
1919         }
1920         dev->agp_buffer_token = init->buffers_offset;
1921         dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
1922         if (!dev->agp_buffer_map) {
1923                 DRM_ERROR("could not find dma buffer region!\n");
1924                 r600_do_cleanup_cp(dev);
1925                 return -EINVAL;
1926         }
1927
1928         if (init->gart_textures_offset) {
1929                 dev_priv->gart_textures =
1930                     drm_core_findmap(dev, init->gart_textures_offset);
1931                 if (!dev_priv->gart_textures) {
1932                         DRM_ERROR("could not find GART texture region!\n");
1933                         r600_do_cleanup_cp(dev);
1934                         return -EINVAL;
1935                 }
1936         }
1937
1938 #if __OS_HAS_AGP
1939         /* XXX */
1940         if (dev_priv->flags & RADEON_IS_AGP) {
1941                 drm_core_ioremap_wc(dev_priv->cp_ring, dev);
1942                 drm_core_ioremap_wc(dev_priv->ring_rptr, dev);
1943                 drm_core_ioremap_wc(dev->agp_buffer_map, dev);
1944                 if (!dev_priv->cp_ring->handle ||
1945                     !dev_priv->ring_rptr->handle ||
1946                     !dev->agp_buffer_map->handle) {
1947                         DRM_ERROR("could not find ioremap agp regions!\n");
1948                         r600_do_cleanup_cp(dev);
1949                         return -EINVAL;
1950                 }
1951         } else
1952 #endif
1953         {
1954                 dev_priv->cp_ring->handle = (void *)dev_priv->cp_ring->offset;
1955                 dev_priv->ring_rptr->handle =
1956                     (void *)dev_priv->ring_rptr->offset;
1957                 dev->agp_buffer_map->handle =
1958                     (void *)dev->agp_buffer_map->offset;
1959
1960                 DRM_DEBUG("dev_priv->cp_ring->handle %p\n",
1961                           dev_priv->cp_ring->handle);
1962                 DRM_DEBUG("dev_priv->ring_rptr->handle %p\n",
1963                           dev_priv->ring_rptr->handle);
1964                 DRM_DEBUG("dev->agp_buffer_map->handle %p\n",
1965                           dev->agp_buffer_map->handle);
1966         }
1967
1968         dev_priv->fb_location = (radeon_read_fb_location(dev_priv) & 0xffff) << 24;
1969         dev_priv->fb_size =
1970                 (((radeon_read_fb_location(dev_priv) & 0xffff0000u) << 8) + 0x1000000)
1971                 - dev_priv->fb_location;
1972
1973         dev_priv->front_pitch_offset = (((dev_priv->front_pitch / 64) << 22) |
1974                                         ((dev_priv->front_offset
1975                                           + dev_priv->fb_location) >> 10));
1976
1977         dev_priv->back_pitch_offset = (((dev_priv->back_pitch / 64) << 22) |
1978                                        ((dev_priv->back_offset
1979                                          + dev_priv->fb_location) >> 10));
1980
1981         dev_priv->depth_pitch_offset = (((dev_priv->depth_pitch / 64) << 22) |
1982                                         ((dev_priv->depth_offset
1983                                           + dev_priv->fb_location) >> 10));
1984
1985         dev_priv->gart_size = init->gart_size;
1986
1987         /* New let's set the memory map ... */
1988         if (dev_priv->new_memmap) {
1989                 u32 base = 0;
1990
1991                 DRM_INFO("Setting GART location based on new memory map\n");
1992
1993                 /* If using AGP, try to locate the AGP aperture at the same
1994                  * location in the card and on the bus, though we have to
1995                  * align it down.
1996                  */
1997 #if __OS_HAS_AGP
1998                 /* XXX */
1999                 if (dev_priv->flags & RADEON_IS_AGP) {
2000                         base = dev->agp->base;
2001                         /* Check if valid */
2002                         if ((base + dev_priv->gart_size - 1) >= dev_priv->fb_location &&
2003                             base < (dev_priv->fb_location + dev_priv->fb_size - 1)) {
2004                                 DRM_INFO("Can't use AGP base @0x%08lx, won't fit\n",
2005                                          dev->agp->base);
2006                                 base = 0;
2007                         }
2008                 }
2009 #endif
2010                 /* If not or if AGP is at 0 (Macs), try to put it elsewhere */
2011                 if (base == 0) {
2012                         base = dev_priv->fb_location + dev_priv->fb_size;
2013                         if (base < dev_priv->fb_location ||
2014                             ((base + dev_priv->gart_size) & 0xfffffffful) < base)
2015                                 base = dev_priv->fb_location
2016                                         - dev_priv->gart_size;
2017                 }
2018                 dev_priv->gart_vm_start = base & 0xffc00000u;
2019                 if (dev_priv->gart_vm_start != base)
2020                         DRM_INFO("GART aligned down from 0x%08x to 0x%08x\n",
2021                                  base, dev_priv->gart_vm_start);
2022         }
2023
2024 #if __OS_HAS_AGP
2025         /* XXX */
2026         if (dev_priv->flags & RADEON_IS_AGP)
2027                 dev_priv->gart_buffers_offset = (dev->agp_buffer_map->offset
2028                                                  - dev->agp->base
2029                                                  + dev_priv->gart_vm_start);
2030         else
2031 #endif
2032                 dev_priv->gart_buffers_offset = (dev->agp_buffer_map->offset
2033                                                  - (unsigned long)dev->sg->virtual
2034                                                  + dev_priv->gart_vm_start);
2035
2036         DRM_DEBUG("fb 0x%08x size %d\n",
2037                   (unsigned int) dev_priv->fb_location,
2038                   (unsigned int) dev_priv->fb_size);
2039         DRM_DEBUG("dev_priv->gart_size %d\n", dev_priv->gart_size);
2040         DRM_DEBUG("dev_priv->gart_vm_start 0x%08x\n",
2041                   (unsigned int) dev_priv->gart_vm_start);
2042         DRM_DEBUG("dev_priv->gart_buffers_offset 0x%08lx\n",
2043                   dev_priv->gart_buffers_offset);
2044
2045         dev_priv->ring.start = (u32 *) dev_priv->cp_ring->handle;
2046         dev_priv->ring.end = ((u32 *) dev_priv->cp_ring->handle
2047                               + init->ring_size / sizeof(u32));
2048         dev_priv->ring.size = init->ring_size;
2049         dev_priv->ring.size_l2qw = drm_order(init->ring_size / 8);
2050
2051         dev_priv->ring.rptr_update = /* init->rptr_update */ 4096;
2052         dev_priv->ring.rptr_update_l2qw = drm_order(/* init->rptr_update */ 4096 / 8);
2053
2054         dev_priv->ring.fetch_size = /* init->fetch_size */ 32;
2055         dev_priv->ring.fetch_size_l2ow = drm_order(/* init->fetch_size */ 32 / 16);
2056
2057         dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
2058
2059         dev_priv->ring.high_mark = RADEON_RING_HIGH_MARK;
2060
2061 #if __OS_HAS_AGP
2062         if (dev_priv->flags & RADEON_IS_AGP) {
2063                 /* XXX turn off pcie gart */
2064         } else
2065 #endif
2066         {
2067                 dev_priv->gart_info.table_mask = DMA_BIT_MASK(32);
2068                 /* if we have an offset set from userspace */
2069                 if (!dev_priv->pcigart_offset_set) {
2070                         DRM_ERROR("Need gart offset from userspace\n");
2071                         r600_do_cleanup_cp(dev);
2072                         return -EINVAL;
2073                 }
2074
2075                 DRM_DEBUG("Using gart offset 0x%08lx\n", dev_priv->pcigart_offset);
2076
2077                 dev_priv->gart_info.bus_addr =
2078                         dev_priv->pcigart_offset + dev_priv->fb_location;
2079                 dev_priv->gart_info.mapping.offset =
2080                         dev_priv->pcigart_offset + dev_priv->fb_aper_offset;
2081                 dev_priv->gart_info.mapping.size =
2082                         dev_priv->gart_info.table_size;
2083
2084                 drm_core_ioremap_wc(&dev_priv->gart_info.mapping, dev);
2085                 if (!dev_priv->gart_info.mapping.handle) {
2086                         DRM_ERROR("ioremap failed.\n");
2087                         r600_do_cleanup_cp(dev);
2088                         return -EINVAL;
2089                 }
2090
2091                 dev_priv->gart_info.addr =
2092                         dev_priv->gart_info.mapping.handle;
2093
2094                 DRM_DEBUG("Setting phys_pci_gart to %p %08lX\n",
2095                           dev_priv->gart_info.addr,
2096                           dev_priv->pcigart_offset);
2097
2098                 if (!r600_page_table_init(dev)) {
2099                         DRM_ERROR("Failed to init GART table\n");
2100                         r600_do_cleanup_cp(dev);
2101                         return -EINVAL;
2102                 }
2103
2104                 if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770))
2105                         r700_vm_init(dev);
2106                 else
2107                         r600_vm_init(dev);
2108         }
2109
2110         if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770))
2111                 r700_cp_load_microcode(dev_priv);
2112         else
2113                 r600_cp_load_microcode(dev_priv);
2114
2115         r600_cp_init_ring_buffer(dev, dev_priv, file_priv);
2116
2117         dev_priv->last_buf = 0;
2118
2119         r600_do_engine_reset(dev);
2120         r600_test_writeback(dev_priv);
2121
2122         return 0;
2123 }
2124
2125 int r600_do_resume_cp(struct drm_device *dev, struct drm_file *file_priv)
2126 {
2127         drm_radeon_private_t *dev_priv = dev->dev_private;
2128
2129         DRM_DEBUG("\n");
2130         if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770)) {
2131                 r700_vm_init(dev);
2132                 r700_cp_load_microcode(dev_priv);
2133         } else {
2134                 r600_vm_init(dev);
2135                 r600_cp_load_microcode(dev_priv);
2136         }
2137         r600_cp_init_ring_buffer(dev, dev_priv, file_priv);
2138         r600_do_engine_reset(dev);
2139
2140         return 0;
2141 }
2142
2143 /* Wait for the CP to go idle.
2144  */
2145 int r600_do_cp_idle(drm_radeon_private_t *dev_priv)
2146 {
2147         RING_LOCALS;
2148         DRM_DEBUG("\n");
2149
2150         BEGIN_RING(5);
2151         OUT_RING(CP_PACKET3(R600_IT_EVENT_WRITE, 0));
2152         OUT_RING(R600_CACHE_FLUSH_AND_INV_EVENT);
2153         /* wait for 3D idle clean */
2154         OUT_RING(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
2155         OUT_RING((R600_WAIT_UNTIL - R600_SET_CONFIG_REG_OFFSET) >> 2);
2156         OUT_RING(RADEON_WAIT_3D_IDLE | RADEON_WAIT_3D_IDLECLEAN);
2157
2158         ADVANCE_RING();
2159         COMMIT_RING();
2160
2161         return r600_do_wait_for_idle(dev_priv);
2162 }
2163
2164 /* Start the Command Processor.
2165  */
2166 void r600_do_cp_start(drm_radeon_private_t *dev_priv)
2167 {
2168         u32 cp_me;
2169         RING_LOCALS;
2170         DRM_DEBUG("\n");
2171
2172         BEGIN_RING(7);
2173         OUT_RING(CP_PACKET3(R600_IT_ME_INITIALIZE, 5));
2174         OUT_RING(0x00000001);
2175         if (((dev_priv->flags & RADEON_FAMILY_MASK) < CHIP_RV770))
2176                 OUT_RING(0x00000003);
2177         else
2178                 OUT_RING(0x00000000);
2179         OUT_RING((dev_priv->r600_max_hw_contexts - 1));
2180         OUT_RING(R600_ME_INITIALIZE_DEVICE_ID(1));
2181         OUT_RING(0x00000000);
2182         OUT_RING(0x00000000);
2183         ADVANCE_RING();
2184         COMMIT_RING();
2185
2186         /* set the mux and reset the halt bit */
2187         cp_me = 0xff;
2188         RADEON_WRITE(R600_CP_ME_CNTL, cp_me);
2189
2190         dev_priv->cp_running = 1;
2191
2192 }
2193
2194 void r600_do_cp_reset(drm_radeon_private_t *dev_priv)
2195 {
2196         u32 cur_read_ptr;
2197         DRM_DEBUG("\n");
2198
2199         cur_read_ptr = RADEON_READ(R600_CP_RB_RPTR);
2200         RADEON_WRITE(R600_CP_RB_WPTR, cur_read_ptr);
2201         SET_RING_HEAD(dev_priv, cur_read_ptr);
2202         dev_priv->ring.tail = cur_read_ptr;
2203 }
2204
2205 void r600_do_cp_stop(drm_radeon_private_t *dev_priv)
2206 {
2207         uint32_t cp_me;
2208
2209         DRM_DEBUG("\n");
2210
2211         cp_me = 0xff | R600_CP_ME_HALT;
2212
2213         RADEON_WRITE(R600_CP_ME_CNTL, cp_me);
2214
2215         dev_priv->cp_running = 0;
2216 }
2217
2218 int r600_cp_dispatch_indirect(struct drm_device *dev,
2219                               struct drm_buf *buf, int start, int end)
2220 {
2221         drm_radeon_private_t *dev_priv = dev->dev_private;
2222         RING_LOCALS;
2223
2224         if (start != end) {
2225                 unsigned long offset = (dev_priv->gart_buffers_offset
2226                                         + buf->offset + start);
2227                 int dwords = (end - start + 3) / sizeof(u32);
2228
2229                 DRM_DEBUG("dwords:%d\n", dwords);
2230                 DRM_DEBUG("offset 0x%lx\n", offset);
2231
2232
2233                 /* Indirect buffer data must be a multiple of 16 dwords.
2234                  * pad the data with a Type-2 CP packet.
2235                  */
2236                 while (dwords & 0xf) {
2237                         u32 *data = (u32 *)
2238                             ((char *)dev->agp_buffer_map->handle
2239                              + buf->offset + start);
2240                         data[dwords++] = RADEON_CP_PACKET2;
2241                 }
2242
2243                 /* Fire off the indirect buffer */
2244                 BEGIN_RING(4);
2245                 OUT_RING(CP_PACKET3(R600_IT_INDIRECT_BUFFER, 2));
2246                 OUT_RING((offset & 0xfffffffc));
2247                 OUT_RING((upper_32_bits(offset) & 0xff));
2248                 OUT_RING(dwords);
2249                 ADVANCE_RING();
2250         }
2251
2252         return 0;
2253 }