drm/radeon: add support for RV790.
[linux-2.6.git] / 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/RV790 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/RV790 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                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV740)) {
494                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
495                 DRM_INFO("Loading RV730/RV740 PFP Microcode\n");
496                 for (i = 0; i < R700_PFP_UCODE_SIZE; i++)
497                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV730_pfp_microcode[i]);
498                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
499
500                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
501                 DRM_INFO("Loading RV730/RV740 CP Microcode\n");
502                 for (i = 0; i < R700_PM4_UCODE_SIZE; i++)
503                         RADEON_WRITE(R600_CP_ME_RAM_DATA, RV730_cp_microcode[i]);
504                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
505
506         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV710)) {
507                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
508                 DRM_INFO("Loading RV710 PFP Microcode\n");
509                 for (i = 0; i < R700_PFP_UCODE_SIZE; i++)
510                         RADEON_WRITE(R600_CP_PFP_UCODE_DATA, RV710_pfp_microcode[i]);
511                 RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
512
513                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
514                 DRM_INFO("Loading RV710 CP Microcode\n");
515                 for (i = 0; i < R700_PM4_UCODE_SIZE; i++)
516                         RADEON_WRITE(R600_CP_ME_RAM_DATA, RV710_cp_microcode[i]);
517                 RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
518
519         }
520         RADEON_WRITE(R600_CP_PFP_UCODE_ADDR, 0);
521         RADEON_WRITE(R600_CP_ME_RAM_WADDR, 0);
522         RADEON_WRITE(R600_CP_ME_RAM_RADDR, 0);
523
524 }
525
526 static void r600_test_writeback(drm_radeon_private_t *dev_priv)
527 {
528         u32 tmp;
529
530         /* Start with assuming that writeback doesn't work */
531         dev_priv->writeback_works = 0;
532
533         /* Writeback doesn't seem to work everywhere, test it here and possibly
534          * enable it if it appears to work
535          */
536         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(1), 0);
537
538         RADEON_WRITE(R600_SCRATCH_REG1, 0xdeadbeef);
539
540         for (tmp = 0; tmp < dev_priv->usec_timeout; tmp++) {
541                 u32 val;
542
543                 val = radeon_read_ring_rptr(dev_priv, R600_SCRATCHOFF(1));
544                 if (val == 0xdeadbeef)
545                         break;
546                 DRM_UDELAY(1);
547         }
548
549         if (tmp < dev_priv->usec_timeout) {
550                 dev_priv->writeback_works = 1;
551                 DRM_INFO("writeback test succeeded in %d usecs\n", tmp);
552         } else {
553                 dev_priv->writeback_works = 0;
554                 DRM_INFO("writeback test failed\n");
555         }
556         if (radeon_no_wb == 1) {
557                 dev_priv->writeback_works = 0;
558                 DRM_INFO("writeback forced off\n");
559         }
560
561         if (!dev_priv->writeback_works) {
562                 /* Disable writeback to avoid unnecessary bus master transfer */
563                 RADEON_WRITE(R600_CP_RB_CNTL, RADEON_READ(R600_CP_RB_CNTL) |
564                              RADEON_RB_NO_UPDATE);
565                 RADEON_WRITE(R600_SCRATCH_UMSK, 0);
566         }
567 }
568
569 int r600_do_engine_reset(struct drm_device *dev)
570 {
571         drm_radeon_private_t *dev_priv = dev->dev_private;
572         u32 cp_ptr, cp_me_cntl, cp_rb_cntl;
573
574         DRM_INFO("Resetting GPU\n");
575
576         cp_ptr = RADEON_READ(R600_CP_RB_WPTR);
577         cp_me_cntl = RADEON_READ(R600_CP_ME_CNTL);
578         RADEON_WRITE(R600_CP_ME_CNTL, R600_CP_ME_HALT);
579
580         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0x7fff);
581         RADEON_READ(R600_GRBM_SOFT_RESET);
582         DRM_UDELAY(50);
583         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0);
584         RADEON_READ(R600_GRBM_SOFT_RESET);
585
586         RADEON_WRITE(R600_CP_RB_WPTR_DELAY, 0);
587         cp_rb_cntl = RADEON_READ(R600_CP_RB_CNTL);
588         RADEON_WRITE(R600_CP_RB_CNTL, R600_RB_RPTR_WR_ENA);
589
590         RADEON_WRITE(R600_CP_RB_RPTR_WR, cp_ptr);
591         RADEON_WRITE(R600_CP_RB_WPTR, cp_ptr);
592         RADEON_WRITE(R600_CP_RB_CNTL, cp_rb_cntl);
593         RADEON_WRITE(R600_CP_ME_CNTL, cp_me_cntl);
594
595         /* Reset the CP ring */
596         r600_do_cp_reset(dev_priv);
597
598         /* The CP is no longer running after an engine reset */
599         dev_priv->cp_running = 0;
600
601         /* Reset any pending vertex, indirect buffers */
602         radeon_freelist_reset(dev);
603
604         return 0;
605
606 }
607
608 static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
609                                              u32 num_backends,
610                                              u32 backend_disable_mask)
611 {
612         u32 backend_map = 0;
613         u32 enabled_backends_mask;
614         u32 enabled_backends_count;
615         u32 cur_pipe;
616         u32 swizzle_pipe[R6XX_MAX_PIPES];
617         u32 cur_backend;
618         u32 i;
619
620         if (num_tile_pipes > R6XX_MAX_PIPES)
621                 num_tile_pipes = R6XX_MAX_PIPES;
622         if (num_tile_pipes < 1)
623                 num_tile_pipes = 1;
624         if (num_backends > R6XX_MAX_BACKENDS)
625                 num_backends = R6XX_MAX_BACKENDS;
626         if (num_backends < 1)
627                 num_backends = 1;
628
629         enabled_backends_mask = 0;
630         enabled_backends_count = 0;
631         for (i = 0; i < R6XX_MAX_BACKENDS; ++i) {
632                 if (((backend_disable_mask >> i) & 1) == 0) {
633                         enabled_backends_mask |= (1 << i);
634                         ++enabled_backends_count;
635                 }
636                 if (enabled_backends_count == num_backends)
637                         break;
638         }
639
640         if (enabled_backends_count == 0) {
641                 enabled_backends_mask = 1;
642                 enabled_backends_count = 1;
643         }
644
645         if (enabled_backends_count != num_backends)
646                 num_backends = enabled_backends_count;
647
648         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES);
649         switch (num_tile_pipes) {
650         case 1:
651                 swizzle_pipe[0] = 0;
652                 break;
653         case 2:
654                 swizzle_pipe[0] = 0;
655                 swizzle_pipe[1] = 1;
656                 break;
657         case 3:
658                 swizzle_pipe[0] = 0;
659                 swizzle_pipe[1] = 1;
660                 swizzle_pipe[2] = 2;
661                 break;
662         case 4:
663                 swizzle_pipe[0] = 0;
664                 swizzle_pipe[1] = 1;
665                 swizzle_pipe[2] = 2;
666                 swizzle_pipe[3] = 3;
667                 break;
668         case 5:
669                 swizzle_pipe[0] = 0;
670                 swizzle_pipe[1] = 1;
671                 swizzle_pipe[2] = 2;
672                 swizzle_pipe[3] = 3;
673                 swizzle_pipe[4] = 4;
674                 break;
675         case 6:
676                 swizzle_pipe[0] = 0;
677                 swizzle_pipe[1] = 2;
678                 swizzle_pipe[2] = 4;
679                 swizzle_pipe[3] = 5;
680                 swizzle_pipe[4] = 1;
681                 swizzle_pipe[5] = 3;
682                 break;
683         case 7:
684                 swizzle_pipe[0] = 0;
685                 swizzle_pipe[1] = 2;
686                 swizzle_pipe[2] = 4;
687                 swizzle_pipe[3] = 6;
688                 swizzle_pipe[4] = 1;
689                 swizzle_pipe[5] = 3;
690                 swizzle_pipe[6] = 5;
691                 break;
692         case 8:
693                 swizzle_pipe[0] = 0;
694                 swizzle_pipe[1] = 2;
695                 swizzle_pipe[2] = 4;
696                 swizzle_pipe[3] = 6;
697                 swizzle_pipe[4] = 1;
698                 swizzle_pipe[5] = 3;
699                 swizzle_pipe[6] = 5;
700                 swizzle_pipe[7] = 7;
701                 break;
702         }
703
704         cur_backend = 0;
705         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
706                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
707                         cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
708
709                 backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
710
711                 cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
712         }
713
714         return backend_map;
715 }
716
717 static int r600_count_pipe_bits(uint32_t val)
718 {
719         int i, ret = 0;
720         for (i = 0; i < 32; i++) {
721                 ret += val & 1;
722                 val >>= 1;
723         }
724         return ret;
725 }
726
727 static void r600_gfx_init(struct drm_device *dev,
728                           drm_radeon_private_t *dev_priv)
729 {
730         int i, j, num_qd_pipes;
731         u32 sx_debug_1;
732         u32 tc_cntl;
733         u32 arb_pop;
734         u32 num_gs_verts_per_thread;
735         u32 vgt_gs_per_es;
736         u32 gs_prim_buffer_depth = 0;
737         u32 sq_ms_fifo_sizes;
738         u32 sq_config;
739         u32 sq_gpr_resource_mgmt_1 = 0;
740         u32 sq_gpr_resource_mgmt_2 = 0;
741         u32 sq_thread_resource_mgmt = 0;
742         u32 sq_stack_resource_mgmt_1 = 0;
743         u32 sq_stack_resource_mgmt_2 = 0;
744         u32 hdp_host_path_cntl;
745         u32 backend_map;
746         u32 gb_tiling_config = 0;
747         u32 cc_rb_backend_disable = 0;
748         u32 cc_gc_shader_pipe_config = 0;
749         u32 ramcfg;
750
751         /* setup chip specs */
752         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
753         case CHIP_R600:
754                 dev_priv->r600_max_pipes = 4;
755                 dev_priv->r600_max_tile_pipes = 8;
756                 dev_priv->r600_max_simds = 4;
757                 dev_priv->r600_max_backends = 4;
758                 dev_priv->r600_max_gprs = 256;
759                 dev_priv->r600_max_threads = 192;
760                 dev_priv->r600_max_stack_entries = 256;
761                 dev_priv->r600_max_hw_contexts = 8;
762                 dev_priv->r600_max_gs_threads = 16;
763                 dev_priv->r600_sx_max_export_size = 128;
764                 dev_priv->r600_sx_max_export_pos_size = 16;
765                 dev_priv->r600_sx_max_export_smx_size = 128;
766                 dev_priv->r600_sq_num_cf_insts = 2;
767                 break;
768         case CHIP_RV630:
769         case CHIP_RV635:
770                 dev_priv->r600_max_pipes = 2;
771                 dev_priv->r600_max_tile_pipes = 2;
772                 dev_priv->r600_max_simds = 3;
773                 dev_priv->r600_max_backends = 1;
774                 dev_priv->r600_max_gprs = 128;
775                 dev_priv->r600_max_threads = 192;
776                 dev_priv->r600_max_stack_entries = 128;
777                 dev_priv->r600_max_hw_contexts = 8;
778                 dev_priv->r600_max_gs_threads = 4;
779                 dev_priv->r600_sx_max_export_size = 128;
780                 dev_priv->r600_sx_max_export_pos_size = 16;
781                 dev_priv->r600_sx_max_export_smx_size = 128;
782                 dev_priv->r600_sq_num_cf_insts = 2;
783                 break;
784         case CHIP_RV610:
785         case CHIP_RS780:
786         case CHIP_RV620:
787                 dev_priv->r600_max_pipes = 1;
788                 dev_priv->r600_max_tile_pipes = 1;
789                 dev_priv->r600_max_simds = 2;
790                 dev_priv->r600_max_backends = 1;
791                 dev_priv->r600_max_gprs = 128;
792                 dev_priv->r600_max_threads = 192;
793                 dev_priv->r600_max_stack_entries = 128;
794                 dev_priv->r600_max_hw_contexts = 4;
795                 dev_priv->r600_max_gs_threads = 4;
796                 dev_priv->r600_sx_max_export_size = 128;
797                 dev_priv->r600_sx_max_export_pos_size = 16;
798                 dev_priv->r600_sx_max_export_smx_size = 128;
799                 dev_priv->r600_sq_num_cf_insts = 1;
800                 break;
801         case CHIP_RV670:
802                 dev_priv->r600_max_pipes = 4;
803                 dev_priv->r600_max_tile_pipes = 4;
804                 dev_priv->r600_max_simds = 4;
805                 dev_priv->r600_max_backends = 4;
806                 dev_priv->r600_max_gprs = 192;
807                 dev_priv->r600_max_threads = 192;
808                 dev_priv->r600_max_stack_entries = 256;
809                 dev_priv->r600_max_hw_contexts = 8;
810                 dev_priv->r600_max_gs_threads = 16;
811                 dev_priv->r600_sx_max_export_size = 128;
812                 dev_priv->r600_sx_max_export_pos_size = 16;
813                 dev_priv->r600_sx_max_export_smx_size = 128;
814                 dev_priv->r600_sq_num_cf_insts = 2;
815                 break;
816         default:
817                 break;
818         }
819
820         /* Initialize HDP */
821         j = 0;
822         for (i = 0; i < 32; i++) {
823                 RADEON_WRITE((0x2c14 + j), 0x00000000);
824                 RADEON_WRITE((0x2c18 + j), 0x00000000);
825                 RADEON_WRITE((0x2c1c + j), 0x00000000);
826                 RADEON_WRITE((0x2c20 + j), 0x00000000);
827                 RADEON_WRITE((0x2c24 + j), 0x00000000);
828                 j += 0x18;
829         }
830
831         RADEON_WRITE(R600_GRBM_CNTL, R600_GRBM_READ_TIMEOUT(0xff));
832
833         /* setup tiling, simd, pipe config */
834         ramcfg = RADEON_READ(R600_RAMCFG);
835
836         switch (dev_priv->r600_max_tile_pipes) {
837         case 1:
838                 gb_tiling_config |= R600_PIPE_TILING(0);
839                 break;
840         case 2:
841                 gb_tiling_config |= R600_PIPE_TILING(1);
842                 break;
843         case 4:
844                 gb_tiling_config |= R600_PIPE_TILING(2);
845                 break;
846         case 8:
847                 gb_tiling_config |= R600_PIPE_TILING(3);
848                 break;
849         default:
850                 break;
851         }
852
853         gb_tiling_config |= R600_BANK_TILING((ramcfg >> R600_NOOFBANK_SHIFT) & R600_NOOFBANK_MASK);
854
855         gb_tiling_config |= R600_GROUP_SIZE(0);
856
857         if (((ramcfg >> R600_NOOFROWS_SHIFT) & R600_NOOFROWS_MASK) > 3) {
858                 gb_tiling_config |= R600_ROW_TILING(3);
859                 gb_tiling_config |= R600_SAMPLE_SPLIT(3);
860         } else {
861                 gb_tiling_config |=
862                         R600_ROW_TILING(((ramcfg >> R600_NOOFROWS_SHIFT) & R600_NOOFROWS_MASK));
863                 gb_tiling_config |=
864                         R600_SAMPLE_SPLIT(((ramcfg >> R600_NOOFROWS_SHIFT) & R600_NOOFROWS_MASK));
865         }
866
867         gb_tiling_config |= R600_BANK_SWAPS(1);
868
869         backend_map = r600_get_tile_pipe_to_backend_map(dev_priv->r600_max_tile_pipes,
870                                                         dev_priv->r600_max_backends,
871                                                         (0xff << dev_priv->r600_max_backends) & 0xff);
872         gb_tiling_config |= R600_BACKEND_MAP(backend_map);
873
874         cc_gc_shader_pipe_config =
875                 R600_INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << dev_priv->r600_max_pipes) & R6XX_MAX_PIPES_MASK);
876         cc_gc_shader_pipe_config |=
877                 R600_INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << dev_priv->r600_max_simds) & R6XX_MAX_SIMDS_MASK);
878
879         cc_rb_backend_disable =
880                 R600_BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << dev_priv->r600_max_backends) & R6XX_MAX_BACKENDS_MASK);
881
882         RADEON_WRITE(R600_GB_TILING_CONFIG,      gb_tiling_config);
883         RADEON_WRITE(R600_DCP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
884         RADEON_WRITE(R600_HDP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
885
886         RADEON_WRITE(R600_CC_RB_BACKEND_DISABLE,      cc_rb_backend_disable);
887         RADEON_WRITE(R600_CC_GC_SHADER_PIPE_CONFIG,   cc_gc_shader_pipe_config);
888         RADEON_WRITE(R600_GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
889
890         num_qd_pipes =
891                 R6XX_MAX_BACKENDS - r600_count_pipe_bits(cc_gc_shader_pipe_config & R600_INACTIVE_QD_PIPES_MASK);
892         RADEON_WRITE(R600_VGT_OUT_DEALLOC_CNTL, (num_qd_pipes * 4) & R600_DEALLOC_DIST_MASK);
893         RADEON_WRITE(R600_VGT_VERTEX_REUSE_BLOCK_CNTL, ((num_qd_pipes * 4) - 2) & R600_VTX_REUSE_DEPTH_MASK);
894
895         /* set HW defaults for 3D engine */
896         RADEON_WRITE(R600_CP_QUEUE_THRESHOLDS, (R600_ROQ_IB1_START(0x16) |
897                                                 R600_ROQ_IB2_START(0x2b)));
898
899         RADEON_WRITE(R600_CP_MEQ_THRESHOLDS, (R600_MEQ_END(0x40) |
900                                               R600_ROQ_END(0x40)));
901
902         RADEON_WRITE(R600_TA_CNTL_AUX, (R600_DISABLE_CUBE_ANISO |
903                                         R600_SYNC_GRADIENT |
904                                         R600_SYNC_WALKER |
905                                         R600_SYNC_ALIGNER));
906
907         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV670)
908                 RADEON_WRITE(R600_ARB_GDEC_RD_CNTL, 0x00000021);
909
910         sx_debug_1 = RADEON_READ(R600_SX_DEBUG_1);
911         sx_debug_1 |= R600_SMX_EVENT_RELEASE;
912         if (((dev_priv->flags & RADEON_FAMILY_MASK) > CHIP_R600))
913                 sx_debug_1 |= R600_ENABLE_NEW_SMX_ADDRESS;
914         RADEON_WRITE(R600_SX_DEBUG_1, sx_debug_1);
915
916         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600) ||
917             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630) ||
918             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
919             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
920             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780))
921                 RADEON_WRITE(R600_DB_DEBUG, R600_PREZ_MUST_WAIT_FOR_POSTZ_DONE);
922         else
923                 RADEON_WRITE(R600_DB_DEBUG, 0);
924
925         RADEON_WRITE(R600_DB_WATERMARKS, (R600_DEPTH_FREE(4) |
926                                           R600_DEPTH_FLUSH(16) |
927                                           R600_DEPTH_PENDING_FREE(4) |
928                                           R600_DEPTH_CACHELINE_FREE(16)));
929         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
930         RADEON_WRITE(R600_VGT_NUM_INSTANCES, 0);
931
932         RADEON_WRITE(R600_SPI_CONFIG_CNTL, R600_GPR_WRITE_PRIORITY(0));
933         RADEON_WRITE(R600_SPI_CONFIG_CNTL_1, R600_VTX_DONE_DELAY(0));
934
935         sq_ms_fifo_sizes = RADEON_READ(R600_SQ_MS_FIFO_SIZES);
936         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
937             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
938             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780)) {
939                 sq_ms_fifo_sizes = (R600_CACHE_FIFO_SIZE(0xa) |
940                                     R600_FETCH_FIFO_HIWATER(0xa) |
941                                     R600_DONE_FIFO_HIWATER(0xe0) |
942                                     R600_ALU_UPDATE_FIFO_HIWATER(0x8));
943         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600) ||
944                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630)) {
945                 sq_ms_fifo_sizes &= ~R600_DONE_FIFO_HIWATER(0xff);
946                 sq_ms_fifo_sizes |= R600_DONE_FIFO_HIWATER(0x4);
947         }
948         RADEON_WRITE(R600_SQ_MS_FIFO_SIZES, sq_ms_fifo_sizes);
949
950         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
951          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
952          */
953         sq_config = RADEON_READ(R600_SQ_CONFIG);
954         sq_config &= ~(R600_PS_PRIO(3) |
955                        R600_VS_PRIO(3) |
956                        R600_GS_PRIO(3) |
957                        R600_ES_PRIO(3));
958         sq_config |= (R600_DX9_CONSTS |
959                       R600_VC_ENABLE |
960                       R600_PS_PRIO(0) |
961                       R600_VS_PRIO(1) |
962                       R600_GS_PRIO(2) |
963                       R600_ES_PRIO(3));
964
965         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_R600) {
966                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(124) |
967                                           R600_NUM_VS_GPRS(124) |
968                                           R600_NUM_CLAUSE_TEMP_GPRS(4));
969                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(0) |
970                                           R600_NUM_ES_GPRS(0));
971                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(136) |
972                                            R600_NUM_VS_THREADS(48) |
973                                            R600_NUM_GS_THREADS(4) |
974                                            R600_NUM_ES_THREADS(4));
975                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(128) |
976                                             R600_NUM_VS_STACK_ENTRIES(128));
977                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(0) |
978                                             R600_NUM_ES_STACK_ENTRIES(0));
979         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
980                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
981                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780)) {
982                 /* no vertex cache */
983                 sq_config &= ~R600_VC_ENABLE;
984
985                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(44) |
986                                           R600_NUM_VS_GPRS(44) |
987                                           R600_NUM_CLAUSE_TEMP_GPRS(2));
988                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(17) |
989                                           R600_NUM_ES_GPRS(17));
990                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(79) |
991                                            R600_NUM_VS_THREADS(78) |
992                                            R600_NUM_GS_THREADS(4) |
993                                            R600_NUM_ES_THREADS(31));
994                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(40) |
995                                             R600_NUM_VS_STACK_ENTRIES(40));
996                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(32) |
997                                             R600_NUM_ES_STACK_ENTRIES(16));
998         } else if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV630) ||
999                    ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV635)) {
1000                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(44) |
1001                                           R600_NUM_VS_GPRS(44) |
1002                                           R600_NUM_CLAUSE_TEMP_GPRS(2));
1003                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(18) |
1004                                           R600_NUM_ES_GPRS(18));
1005                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(79) |
1006                                            R600_NUM_VS_THREADS(78) |
1007                                            R600_NUM_GS_THREADS(4) |
1008                                            R600_NUM_ES_THREADS(31));
1009                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(40) |
1010                                             R600_NUM_VS_STACK_ENTRIES(40));
1011                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(32) |
1012                                             R600_NUM_ES_STACK_ENTRIES(16));
1013         } else if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV670) {
1014                 sq_gpr_resource_mgmt_1 = (R600_NUM_PS_GPRS(44) |
1015                                           R600_NUM_VS_GPRS(44) |
1016                                           R600_NUM_CLAUSE_TEMP_GPRS(2));
1017                 sq_gpr_resource_mgmt_2 = (R600_NUM_GS_GPRS(17) |
1018                                           R600_NUM_ES_GPRS(17));
1019                 sq_thread_resource_mgmt = (R600_NUM_PS_THREADS(79) |
1020                                            R600_NUM_VS_THREADS(78) |
1021                                            R600_NUM_GS_THREADS(4) |
1022                                            R600_NUM_ES_THREADS(31));
1023                 sq_stack_resource_mgmt_1 = (R600_NUM_PS_STACK_ENTRIES(64) |
1024                                             R600_NUM_VS_STACK_ENTRIES(64));
1025                 sq_stack_resource_mgmt_2 = (R600_NUM_GS_STACK_ENTRIES(64) |
1026                                             R600_NUM_ES_STACK_ENTRIES(64));
1027         }
1028
1029         RADEON_WRITE(R600_SQ_CONFIG, sq_config);
1030         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
1031         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
1032         RADEON_WRITE(R600_SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1033         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
1034         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
1035
1036         if (((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV610) ||
1037             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV620) ||
1038             ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RS780))
1039                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, R600_CACHE_INVALIDATION(R600_TC_ONLY));
1040         else
1041                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, R600_CACHE_INVALIDATION(R600_VC_AND_TC));
1042
1043         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_2S, (R600_S0_X(0xc) |
1044                                                     R600_S0_Y(0x4) |
1045                                                     R600_S1_X(0x4) |
1046                                                     R600_S1_Y(0xc)));
1047         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_4S, (R600_S0_X(0xe) |
1048                                                     R600_S0_Y(0xe) |
1049                                                     R600_S1_X(0x2) |
1050                                                     R600_S1_Y(0x2) |
1051                                                     R600_S2_X(0xa) |
1052                                                     R600_S2_Y(0x6) |
1053                                                     R600_S3_X(0x6) |
1054                                                     R600_S3_Y(0xa)));
1055         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_8S_WD0, (R600_S0_X(0xe) |
1056                                                         R600_S0_Y(0xb) |
1057                                                         R600_S1_X(0x4) |
1058                                                         R600_S1_Y(0xc) |
1059                                                         R600_S2_X(0x1) |
1060                                                         R600_S2_Y(0x6) |
1061                                                         R600_S3_X(0xa) |
1062                                                         R600_S3_Y(0xe)));
1063         RADEON_WRITE(R600_PA_SC_AA_SAMPLE_LOCS_8S_WD1, (R600_S4_X(0x6) |
1064                                                         R600_S4_Y(0x1) |
1065                                                         R600_S5_X(0x0) |
1066                                                         R600_S5_Y(0x0) |
1067                                                         R600_S6_X(0xb) |
1068                                                         R600_S6_Y(0x4) |
1069                                                         R600_S7_X(0x7) |
1070                                                         R600_S7_Y(0x8)));
1071
1072
1073         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1074         case CHIP_R600:
1075         case CHIP_RV630:
1076         case CHIP_RV635:
1077                 gs_prim_buffer_depth = 0;
1078                 break;
1079         case CHIP_RV610:
1080         case CHIP_RS780:
1081         case CHIP_RV620:
1082                 gs_prim_buffer_depth = 32;
1083                 break;
1084         case CHIP_RV670:
1085                 gs_prim_buffer_depth = 128;
1086                 break;
1087         default:
1088                 break;
1089         }
1090
1091         num_gs_verts_per_thread = dev_priv->r600_max_pipes * 16;
1092         vgt_gs_per_es = gs_prim_buffer_depth + num_gs_verts_per_thread;
1093         /* Max value for this is 256 */
1094         if (vgt_gs_per_es > 256)
1095                 vgt_gs_per_es = 256;
1096
1097         RADEON_WRITE(R600_VGT_ES_PER_GS, 128);
1098         RADEON_WRITE(R600_VGT_GS_PER_ES, vgt_gs_per_es);
1099         RADEON_WRITE(R600_VGT_GS_PER_VS, 2);
1100         RADEON_WRITE(R600_VGT_GS_VERTEX_REUSE, 16);
1101
1102         /* more default values. 2D/3D driver should adjust as needed */
1103         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE_STATE, 0);
1104         RADEON_WRITE(R600_VGT_STRMOUT_EN, 0);
1105         RADEON_WRITE(R600_SX_MISC, 0);
1106         RADEON_WRITE(R600_PA_SC_MODE_CNTL, 0);
1107         RADEON_WRITE(R600_PA_SC_AA_CONFIG, 0);
1108         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE, 0);
1109         RADEON_WRITE(R600_SPI_INPUT_Z, 0);
1110         RADEON_WRITE(R600_SPI_PS_IN_CONTROL_0, R600_NUM_INTERP(2));
1111         RADEON_WRITE(R600_CB_COLOR7_FRAG, 0);
1112
1113         /* clear render buffer base addresses */
1114         RADEON_WRITE(R600_CB_COLOR0_BASE, 0);
1115         RADEON_WRITE(R600_CB_COLOR1_BASE, 0);
1116         RADEON_WRITE(R600_CB_COLOR2_BASE, 0);
1117         RADEON_WRITE(R600_CB_COLOR3_BASE, 0);
1118         RADEON_WRITE(R600_CB_COLOR4_BASE, 0);
1119         RADEON_WRITE(R600_CB_COLOR5_BASE, 0);
1120         RADEON_WRITE(R600_CB_COLOR6_BASE, 0);
1121         RADEON_WRITE(R600_CB_COLOR7_BASE, 0);
1122
1123         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1124         case CHIP_RV610:
1125         case CHIP_RS780:
1126         case CHIP_RV620:
1127                 tc_cntl = R600_TC_L2_SIZE(8);
1128                 break;
1129         case CHIP_RV630:
1130         case CHIP_RV635:
1131                 tc_cntl = R600_TC_L2_SIZE(4);
1132                 break;
1133         case CHIP_R600:
1134                 tc_cntl = R600_TC_L2_SIZE(0) | R600_L2_DISABLE_LATE_HIT;
1135                 break;
1136         default:
1137                 tc_cntl = R600_TC_L2_SIZE(0);
1138                 break;
1139         }
1140
1141         RADEON_WRITE(R600_TC_CNTL, tc_cntl);
1142
1143         hdp_host_path_cntl = RADEON_READ(R600_HDP_HOST_PATH_CNTL);
1144         RADEON_WRITE(R600_HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1145
1146         arb_pop = RADEON_READ(R600_ARB_POP);
1147         arb_pop |= R600_ENABLE_TC128;
1148         RADEON_WRITE(R600_ARB_POP, arb_pop);
1149
1150         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
1151         RADEON_WRITE(R600_PA_CL_ENHANCE, (R600_CLIP_VTX_REORDER_ENA |
1152                                           R600_NUM_CLIP_SEQ(3)));
1153         RADEON_WRITE(R600_PA_SC_ENHANCE, R600_FORCE_EOV_MAX_CLK_CNT(4095));
1154
1155 }
1156
1157 static u32 r700_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
1158                                              u32 num_backends,
1159                                              u32 backend_disable_mask)
1160 {
1161         u32 backend_map = 0;
1162         u32 enabled_backends_mask;
1163         u32 enabled_backends_count;
1164         u32 cur_pipe;
1165         u32 swizzle_pipe[R7XX_MAX_PIPES];
1166         u32 cur_backend;
1167         u32 i;
1168
1169         if (num_tile_pipes > R7XX_MAX_PIPES)
1170                 num_tile_pipes = R7XX_MAX_PIPES;
1171         if (num_tile_pipes < 1)
1172                 num_tile_pipes = 1;
1173         if (num_backends > R7XX_MAX_BACKENDS)
1174                 num_backends = R7XX_MAX_BACKENDS;
1175         if (num_backends < 1)
1176                 num_backends = 1;
1177
1178         enabled_backends_mask = 0;
1179         enabled_backends_count = 0;
1180         for (i = 0; i < R7XX_MAX_BACKENDS; ++i) {
1181                 if (((backend_disable_mask >> i) & 1) == 0) {
1182                         enabled_backends_mask |= (1 << i);
1183                         ++enabled_backends_count;
1184                 }
1185                 if (enabled_backends_count == num_backends)
1186                         break;
1187         }
1188
1189         if (enabled_backends_count == 0) {
1190                 enabled_backends_mask = 1;
1191                 enabled_backends_count = 1;
1192         }
1193
1194         if (enabled_backends_count != num_backends)
1195                 num_backends = enabled_backends_count;
1196
1197         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R7XX_MAX_PIPES);
1198         switch (num_tile_pipes) {
1199         case 1:
1200                 swizzle_pipe[0] = 0;
1201                 break;
1202         case 2:
1203                 swizzle_pipe[0] = 0;
1204                 swizzle_pipe[1] = 1;
1205                 break;
1206         case 3:
1207                 swizzle_pipe[0] = 0;
1208                 swizzle_pipe[1] = 2;
1209                 swizzle_pipe[2] = 1;
1210                 break;
1211         case 4:
1212                 swizzle_pipe[0] = 0;
1213                 swizzle_pipe[1] = 2;
1214                 swizzle_pipe[2] = 3;
1215                 swizzle_pipe[3] = 1;
1216                 break;
1217         case 5:
1218                 swizzle_pipe[0] = 0;
1219                 swizzle_pipe[1] = 2;
1220                 swizzle_pipe[2] = 4;
1221                 swizzle_pipe[3] = 1;
1222                 swizzle_pipe[4] = 3;
1223                 break;
1224         case 6:
1225                 swizzle_pipe[0] = 0;
1226                 swizzle_pipe[1] = 2;
1227                 swizzle_pipe[2] = 4;
1228                 swizzle_pipe[3] = 5;
1229                 swizzle_pipe[4] = 3;
1230                 swizzle_pipe[5] = 1;
1231                 break;
1232         case 7:
1233                 swizzle_pipe[0] = 0;
1234                 swizzle_pipe[1] = 2;
1235                 swizzle_pipe[2] = 4;
1236                 swizzle_pipe[3] = 6;
1237                 swizzle_pipe[4] = 3;
1238                 swizzle_pipe[5] = 1;
1239                 swizzle_pipe[6] = 5;
1240                 break;
1241         case 8:
1242                 swizzle_pipe[0] = 0;
1243                 swizzle_pipe[1] = 2;
1244                 swizzle_pipe[2] = 4;
1245                 swizzle_pipe[3] = 6;
1246                 swizzle_pipe[4] = 3;
1247                 swizzle_pipe[5] = 1;
1248                 swizzle_pipe[6] = 7;
1249                 swizzle_pipe[7] = 5;
1250                 break;
1251         }
1252
1253         cur_backend = 0;
1254         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
1255                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
1256                         cur_backend = (cur_backend + 1) % R7XX_MAX_BACKENDS;
1257
1258                 backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
1259
1260                 cur_backend = (cur_backend + 1) % R7XX_MAX_BACKENDS;
1261         }
1262
1263         return backend_map;
1264 }
1265
1266 static void r700_gfx_init(struct drm_device *dev,
1267                           drm_radeon_private_t *dev_priv)
1268 {
1269         int i, j, num_qd_pipes;
1270         u32 sx_debug_1;
1271         u32 smx_dc_ctl0;
1272         u32 num_gs_verts_per_thread;
1273         u32 vgt_gs_per_es;
1274         u32 gs_prim_buffer_depth = 0;
1275         u32 sq_ms_fifo_sizes;
1276         u32 sq_config;
1277         u32 sq_thread_resource_mgmt;
1278         u32 hdp_host_path_cntl;
1279         u32 sq_dyn_gpr_size_simd_ab_0;
1280         u32 backend_map;
1281         u32 gb_tiling_config = 0;
1282         u32 cc_rb_backend_disable = 0;
1283         u32 cc_gc_shader_pipe_config = 0;
1284         u32 mc_arb_ramcfg;
1285         u32 db_debug4;
1286
1287         /* setup chip specs */
1288         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1289         case CHIP_RV770:
1290                 dev_priv->r600_max_pipes = 4;
1291                 dev_priv->r600_max_tile_pipes = 8;
1292                 dev_priv->r600_max_simds = 10;
1293                 dev_priv->r600_max_backends = 4;
1294                 dev_priv->r600_max_gprs = 256;
1295                 dev_priv->r600_max_threads = 248;
1296                 dev_priv->r600_max_stack_entries = 512;
1297                 dev_priv->r600_max_hw_contexts = 8;
1298                 dev_priv->r600_max_gs_threads = 16 * 2;
1299                 dev_priv->r600_sx_max_export_size = 128;
1300                 dev_priv->r600_sx_max_export_pos_size = 16;
1301                 dev_priv->r600_sx_max_export_smx_size = 112;
1302                 dev_priv->r600_sq_num_cf_insts = 2;
1303
1304                 dev_priv->r700_sx_num_of_sets = 7;
1305                 dev_priv->r700_sc_prim_fifo_size = 0xF9;
1306                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1307                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1308                 break;
1309         case CHIP_RV730:
1310                 dev_priv->r600_max_pipes = 2;
1311                 dev_priv->r600_max_tile_pipes = 4;
1312                 dev_priv->r600_max_simds = 8;
1313                 dev_priv->r600_max_backends = 2;
1314                 dev_priv->r600_max_gprs = 128;
1315                 dev_priv->r600_max_threads = 248;
1316                 dev_priv->r600_max_stack_entries = 256;
1317                 dev_priv->r600_max_hw_contexts = 8;
1318                 dev_priv->r600_max_gs_threads = 16 * 2;
1319                 dev_priv->r600_sx_max_export_size = 256;
1320                 dev_priv->r600_sx_max_export_pos_size = 32;
1321                 dev_priv->r600_sx_max_export_smx_size = 224;
1322                 dev_priv->r600_sq_num_cf_insts = 2;
1323
1324                 dev_priv->r700_sx_num_of_sets = 7;
1325                 dev_priv->r700_sc_prim_fifo_size = 0xf9;
1326                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1327                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1328                 if (dev_priv->r600_sx_max_export_pos_size > 16) {
1329                         dev_priv->r600_sx_max_export_pos_size -= 16;
1330                         dev_priv->r600_sx_max_export_smx_size += 16;
1331                 }
1332                 break;
1333         case CHIP_RV710:
1334                 dev_priv->r600_max_pipes = 2;
1335                 dev_priv->r600_max_tile_pipes = 2;
1336                 dev_priv->r600_max_simds = 2;
1337                 dev_priv->r600_max_backends = 1;
1338                 dev_priv->r600_max_gprs = 256;
1339                 dev_priv->r600_max_threads = 192;
1340                 dev_priv->r600_max_stack_entries = 256;
1341                 dev_priv->r600_max_hw_contexts = 4;
1342                 dev_priv->r600_max_gs_threads = 8 * 2;
1343                 dev_priv->r600_sx_max_export_size = 128;
1344                 dev_priv->r600_sx_max_export_pos_size = 16;
1345                 dev_priv->r600_sx_max_export_smx_size = 112;
1346                 dev_priv->r600_sq_num_cf_insts = 1;
1347
1348                 dev_priv->r700_sx_num_of_sets = 7;
1349                 dev_priv->r700_sc_prim_fifo_size = 0x40;
1350                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1351                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1352                 break;
1353         case CHIP_RV740:
1354                 dev_priv->r600_max_pipes = 4;
1355                 dev_priv->r600_max_tile_pipes = 4;
1356                 dev_priv->r600_max_simds = 8;
1357                 dev_priv->r600_max_backends = 4;
1358                 dev_priv->r600_max_gprs = 256;
1359                 dev_priv->r600_max_threads = 248;
1360                 dev_priv->r600_max_stack_entries = 512;
1361                 dev_priv->r600_max_hw_contexts = 8;
1362                 dev_priv->r600_max_gs_threads = 16 * 2;
1363                 dev_priv->r600_sx_max_export_size = 256;
1364                 dev_priv->r600_sx_max_export_pos_size = 32;
1365                 dev_priv->r600_sx_max_export_smx_size = 224;
1366                 dev_priv->r600_sq_num_cf_insts = 2;
1367
1368                 dev_priv->r700_sx_num_of_sets = 7;
1369                 dev_priv->r700_sc_prim_fifo_size = 0x100;
1370                 dev_priv->r700_sc_hiz_tile_fifo_size = 0x30;
1371                 dev_priv->r700_sc_earlyz_tile_fifo_fize = 0x130;
1372
1373                 if (dev_priv->r600_sx_max_export_pos_size > 16) {
1374                         dev_priv->r600_sx_max_export_pos_size -= 16;
1375                         dev_priv->r600_sx_max_export_smx_size += 16;
1376                 }
1377                 break;
1378         default:
1379                 break;
1380         }
1381
1382         /* Initialize HDP */
1383         j = 0;
1384         for (i = 0; i < 32; i++) {
1385                 RADEON_WRITE((0x2c14 + j), 0x00000000);
1386                 RADEON_WRITE((0x2c18 + j), 0x00000000);
1387                 RADEON_WRITE((0x2c1c + j), 0x00000000);
1388                 RADEON_WRITE((0x2c20 + j), 0x00000000);
1389                 RADEON_WRITE((0x2c24 + j), 0x00000000);
1390                 j += 0x18;
1391         }
1392
1393         RADEON_WRITE(R600_GRBM_CNTL, R600_GRBM_READ_TIMEOUT(0xff));
1394
1395         /* setup tiling, simd, pipe config */
1396         mc_arb_ramcfg = RADEON_READ(R700_MC_ARB_RAMCFG);
1397
1398         switch (dev_priv->r600_max_tile_pipes) {
1399         case 1:
1400                 gb_tiling_config |= R600_PIPE_TILING(0);
1401                 break;
1402         case 2:
1403                 gb_tiling_config |= R600_PIPE_TILING(1);
1404                 break;
1405         case 4:
1406                 gb_tiling_config |= R600_PIPE_TILING(2);
1407                 break;
1408         case 8:
1409                 gb_tiling_config |= R600_PIPE_TILING(3);
1410                 break;
1411         default:
1412                 break;
1413         }
1414
1415         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV770)
1416                 gb_tiling_config |= R600_BANK_TILING(1);
1417         else
1418                 gb_tiling_config |= R600_BANK_TILING((mc_arb_ramcfg >> R700_NOOFBANK_SHIFT) & R700_NOOFBANK_MASK);
1419
1420         gb_tiling_config |= R600_GROUP_SIZE(0);
1421
1422         if (((mc_arb_ramcfg >> R700_NOOFROWS_SHIFT) & R700_NOOFROWS_MASK) > 3) {
1423                 gb_tiling_config |= R600_ROW_TILING(3);
1424                 gb_tiling_config |= R600_SAMPLE_SPLIT(3);
1425         } else {
1426                 gb_tiling_config |=
1427                         R600_ROW_TILING(((mc_arb_ramcfg >> R700_NOOFROWS_SHIFT) & R700_NOOFROWS_MASK));
1428                 gb_tiling_config |=
1429                         R600_SAMPLE_SPLIT(((mc_arb_ramcfg >> R700_NOOFROWS_SHIFT) & R700_NOOFROWS_MASK));
1430         }
1431
1432         gb_tiling_config |= R600_BANK_SWAPS(1);
1433
1434         backend_map = r700_get_tile_pipe_to_backend_map(dev_priv->r600_max_tile_pipes,
1435                                                         dev_priv->r600_max_backends,
1436                                                         (0xff << dev_priv->r600_max_backends) & 0xff);
1437         gb_tiling_config |= R600_BACKEND_MAP(backend_map);
1438
1439         cc_gc_shader_pipe_config =
1440                 R600_INACTIVE_QD_PIPES((R7XX_MAX_PIPES_MASK << dev_priv->r600_max_pipes) & R7XX_MAX_PIPES_MASK);
1441         cc_gc_shader_pipe_config |=
1442                 R600_INACTIVE_SIMDS((R7XX_MAX_SIMDS_MASK << dev_priv->r600_max_simds) & R7XX_MAX_SIMDS_MASK);
1443
1444         cc_rb_backend_disable =
1445                 R600_BACKEND_DISABLE((R7XX_MAX_BACKENDS_MASK << dev_priv->r600_max_backends) & R7XX_MAX_BACKENDS_MASK);
1446
1447         RADEON_WRITE(R600_GB_TILING_CONFIG,      gb_tiling_config);
1448         RADEON_WRITE(R600_DCP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
1449         RADEON_WRITE(R600_HDP_TILING_CONFIG,    (gb_tiling_config & 0xffff));
1450
1451         RADEON_WRITE(R600_CC_RB_BACKEND_DISABLE,      cc_rb_backend_disable);
1452         RADEON_WRITE(R600_CC_GC_SHADER_PIPE_CONFIG,   cc_gc_shader_pipe_config);
1453         RADEON_WRITE(R600_GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
1454
1455         RADEON_WRITE(R700_CC_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1456         RADEON_WRITE(R700_CGTS_SYS_TCC_DISABLE, 0);
1457         RADEON_WRITE(R700_CGTS_TCC_DISABLE, 0);
1458         RADEON_WRITE(R700_CGTS_USER_SYS_TCC_DISABLE, 0);
1459         RADEON_WRITE(R700_CGTS_USER_TCC_DISABLE, 0);
1460
1461         num_qd_pipes =
1462                 R7XX_MAX_BACKENDS - r600_count_pipe_bits(cc_gc_shader_pipe_config & R600_INACTIVE_QD_PIPES_MASK);
1463         RADEON_WRITE(R600_VGT_OUT_DEALLOC_CNTL, (num_qd_pipes * 4) & R600_DEALLOC_DIST_MASK);
1464         RADEON_WRITE(R600_VGT_VERTEX_REUSE_BLOCK_CNTL, ((num_qd_pipes * 4) - 2) & R600_VTX_REUSE_DEPTH_MASK);
1465
1466         /* set HW defaults for 3D engine */
1467         RADEON_WRITE(R600_CP_QUEUE_THRESHOLDS, (R600_ROQ_IB1_START(0x16) |
1468                                                 R600_ROQ_IB2_START(0x2b)));
1469
1470         RADEON_WRITE(R600_CP_MEQ_THRESHOLDS, R700_STQ_SPLIT(0x30));
1471
1472         RADEON_WRITE(R600_TA_CNTL_AUX, (R600_DISABLE_CUBE_ANISO |
1473                                         R600_SYNC_GRADIENT |
1474                                         R600_SYNC_WALKER |
1475                                         R600_SYNC_ALIGNER));
1476
1477         sx_debug_1 = RADEON_READ(R700_SX_DEBUG_1);
1478         sx_debug_1 |= R700_ENABLE_NEW_SMX_ADDRESS;
1479         RADEON_WRITE(R700_SX_DEBUG_1, sx_debug_1);
1480
1481         smx_dc_ctl0 = RADEON_READ(R600_SMX_DC_CTL0);
1482         smx_dc_ctl0 &= ~R700_CACHE_DEPTH(0x1ff);
1483         smx_dc_ctl0 |= R700_CACHE_DEPTH((dev_priv->r700_sx_num_of_sets * 64) - 1);
1484         RADEON_WRITE(R600_SMX_DC_CTL0, smx_dc_ctl0);
1485
1486         RADEON_WRITE(R700_SMX_EVENT_CTL, (R700_ES_FLUSH_CTL(4) |
1487                                           R700_GS_FLUSH_CTL(4) |
1488                                           R700_ACK_FLUSH_CTL(3) |
1489                                           R700_SYNC_FLUSH_CTL));
1490
1491         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV770)
1492                 RADEON_WRITE(R700_DB_DEBUG3, R700_DB_CLK_OFF_DELAY(0x1f));
1493         else {
1494                 db_debug4 = RADEON_READ(RV700_DB_DEBUG4);
1495                 db_debug4 |= RV700_DISABLE_TILE_COVERED_FOR_PS_ITER;
1496                 RADEON_WRITE(RV700_DB_DEBUG4, db_debug4);
1497         }
1498
1499         RADEON_WRITE(R600_SX_EXPORT_BUFFER_SIZES, (R600_COLOR_BUFFER_SIZE((dev_priv->r600_sx_max_export_size / 4) - 1) |
1500                                                    R600_POSITION_BUFFER_SIZE((dev_priv->r600_sx_max_export_pos_size / 4) - 1) |
1501                                                    R600_SMX_BUFFER_SIZE((dev_priv->r600_sx_max_export_smx_size / 4) - 1)));
1502
1503         RADEON_WRITE(R700_PA_SC_FIFO_SIZE_R7XX, (R700_SC_PRIM_FIFO_SIZE(dev_priv->r700_sc_prim_fifo_size) |
1504                                                  R700_SC_HIZ_TILE_FIFO_SIZE(dev_priv->r700_sc_hiz_tile_fifo_size) |
1505                                                  R700_SC_EARLYZ_TILE_FIFO_SIZE(dev_priv->r700_sc_earlyz_tile_fifo_fize)));
1506
1507         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
1508
1509         RADEON_WRITE(R600_VGT_NUM_INSTANCES, 1);
1510
1511         RADEON_WRITE(R600_SPI_CONFIG_CNTL, R600_GPR_WRITE_PRIORITY(0));
1512
1513         RADEON_WRITE(R600_SPI_CONFIG_CNTL_1, R600_VTX_DONE_DELAY(4));
1514
1515         RADEON_WRITE(R600_CP_PERFMON_CNTL, 0);
1516
1517         sq_ms_fifo_sizes = (R600_CACHE_FIFO_SIZE(16 * dev_priv->r600_sq_num_cf_insts) |
1518                             R600_DONE_FIFO_HIWATER(0xe0) |
1519                             R600_ALU_UPDATE_FIFO_HIWATER(0x8));
1520         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1521         case CHIP_RV770:
1522                 sq_ms_fifo_sizes |= R600_FETCH_FIFO_HIWATER(0x1);
1523                 break;
1524         case CHIP_RV730:
1525         case CHIP_RV710:
1526         case CHIP_RV740:
1527         default:
1528                 sq_ms_fifo_sizes |= R600_FETCH_FIFO_HIWATER(0x4);
1529                 break;
1530         }
1531         RADEON_WRITE(R600_SQ_MS_FIFO_SIZES, sq_ms_fifo_sizes);
1532
1533         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
1534          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
1535          */
1536         sq_config = RADEON_READ(R600_SQ_CONFIG);
1537         sq_config &= ~(R600_PS_PRIO(3) |
1538                        R600_VS_PRIO(3) |
1539                        R600_GS_PRIO(3) |
1540                        R600_ES_PRIO(3));
1541         sq_config |= (R600_DX9_CONSTS |
1542                       R600_VC_ENABLE |
1543                       R600_EXPORT_SRC_C |
1544                       R600_PS_PRIO(0) |
1545                       R600_VS_PRIO(1) |
1546                       R600_GS_PRIO(2) |
1547                       R600_ES_PRIO(3));
1548         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV710)
1549                 /* no vertex cache */
1550                 sq_config &= ~R600_VC_ENABLE;
1551
1552         RADEON_WRITE(R600_SQ_CONFIG, sq_config);
1553
1554         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_1,  (R600_NUM_PS_GPRS((dev_priv->r600_max_gprs * 24)/64) |
1555                                                     R600_NUM_VS_GPRS((dev_priv->r600_max_gprs * 24)/64) |
1556                                                     R600_NUM_CLAUSE_TEMP_GPRS(((dev_priv->r600_max_gprs * 24)/64)/2)));
1557
1558         RADEON_WRITE(R600_SQ_GPR_RESOURCE_MGMT_2,  (R600_NUM_GS_GPRS((dev_priv->r600_max_gprs * 7)/64) |
1559                                                     R600_NUM_ES_GPRS((dev_priv->r600_max_gprs * 7)/64)));
1560
1561         sq_thread_resource_mgmt = (R600_NUM_PS_THREADS((dev_priv->r600_max_threads * 4)/8) |
1562                                    R600_NUM_VS_THREADS((dev_priv->r600_max_threads * 2)/8) |
1563                                    R600_NUM_ES_THREADS((dev_priv->r600_max_threads * 1)/8));
1564         if (((dev_priv->r600_max_threads * 1) / 8) > dev_priv->r600_max_gs_threads)
1565                 sq_thread_resource_mgmt |= R600_NUM_GS_THREADS(dev_priv->r600_max_gs_threads);
1566         else
1567                 sq_thread_resource_mgmt |= R600_NUM_GS_THREADS((dev_priv->r600_max_gs_threads * 1)/8);
1568         RADEON_WRITE(R600_SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1569
1570         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_1, (R600_NUM_PS_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4) |
1571                                                      R600_NUM_VS_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4)));
1572
1573         RADEON_WRITE(R600_SQ_STACK_RESOURCE_MGMT_2, (R600_NUM_GS_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4) |
1574                                                      R600_NUM_ES_STACK_ENTRIES((dev_priv->r600_max_stack_entries * 1)/4)));
1575
1576         sq_dyn_gpr_size_simd_ab_0 = (R700_SIMDA_RING0((dev_priv->r600_max_gprs * 38)/64) |
1577                                      R700_SIMDA_RING1((dev_priv->r600_max_gprs * 38)/64) |
1578                                      R700_SIMDB_RING0((dev_priv->r600_max_gprs * 38)/64) |
1579                                      R700_SIMDB_RING1((dev_priv->r600_max_gprs * 38)/64));
1580
1581         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_0, sq_dyn_gpr_size_simd_ab_0);
1582         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_1, sq_dyn_gpr_size_simd_ab_0);
1583         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_2, sq_dyn_gpr_size_simd_ab_0);
1584         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_3, sq_dyn_gpr_size_simd_ab_0);
1585         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_4, sq_dyn_gpr_size_simd_ab_0);
1586         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_5, sq_dyn_gpr_size_simd_ab_0);
1587         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_6, sq_dyn_gpr_size_simd_ab_0);
1588         RADEON_WRITE(R700_SQ_DYN_GPR_SIZE_SIMD_AB_7, sq_dyn_gpr_size_simd_ab_0);
1589
1590         RADEON_WRITE(R700_PA_SC_FORCE_EOV_MAX_CNTS, (R700_FORCE_EOV_MAX_CLK_CNT(4095) |
1591                                                      R700_FORCE_EOV_MAX_REZ_CNT(255)));
1592
1593         if ((dev_priv->flags & RADEON_FAMILY_MASK) == CHIP_RV710)
1594                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, (R600_CACHE_INVALIDATION(R600_TC_ONLY) |
1595                                                            R700_AUTO_INVLD_EN(R700_ES_AND_GS_AUTO)));
1596         else
1597                 RADEON_WRITE(R600_VGT_CACHE_INVALIDATION, (R600_CACHE_INVALIDATION(R600_VC_AND_TC) |
1598                                                            R700_AUTO_INVLD_EN(R700_ES_AND_GS_AUTO)));
1599
1600         switch (dev_priv->flags & RADEON_FAMILY_MASK) {
1601         case CHIP_RV770:
1602         case CHIP_RV730:
1603         case CHIP_RV740:
1604                 gs_prim_buffer_depth = 384;
1605                 break;
1606         case CHIP_RV710:
1607                 gs_prim_buffer_depth = 128;
1608                 break;
1609         default:
1610                 break;
1611         }
1612
1613         num_gs_verts_per_thread = dev_priv->r600_max_pipes * 16;
1614         vgt_gs_per_es = gs_prim_buffer_depth + num_gs_verts_per_thread;
1615         /* Max value for this is 256 */
1616         if (vgt_gs_per_es > 256)
1617                 vgt_gs_per_es = 256;
1618
1619         RADEON_WRITE(R600_VGT_ES_PER_GS, 128);
1620         RADEON_WRITE(R600_VGT_GS_PER_ES, vgt_gs_per_es);
1621         RADEON_WRITE(R600_VGT_GS_PER_VS, 2);
1622
1623         /* more default values. 2D/3D driver should adjust as needed */
1624         RADEON_WRITE(R600_VGT_GS_VERTEX_REUSE, 16);
1625         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE_STATE, 0);
1626         RADEON_WRITE(R600_VGT_STRMOUT_EN, 0);
1627         RADEON_WRITE(R600_SX_MISC, 0);
1628         RADEON_WRITE(R600_PA_SC_MODE_CNTL, 0);
1629         RADEON_WRITE(R700_PA_SC_EDGERULE, 0xaaaaaaaa);
1630         RADEON_WRITE(R600_PA_SC_AA_CONFIG, 0);
1631         RADEON_WRITE(R600_PA_SC_CLIPRECT_RULE, 0xffff);
1632         RADEON_WRITE(R600_PA_SC_LINE_STIPPLE, 0);
1633         RADEON_WRITE(R600_SPI_INPUT_Z, 0);
1634         RADEON_WRITE(R600_SPI_PS_IN_CONTROL_0, R600_NUM_INTERP(2));
1635         RADEON_WRITE(R600_CB_COLOR7_FRAG, 0);
1636
1637         /* clear render buffer base addresses */
1638         RADEON_WRITE(R600_CB_COLOR0_BASE, 0);
1639         RADEON_WRITE(R600_CB_COLOR1_BASE, 0);
1640         RADEON_WRITE(R600_CB_COLOR2_BASE, 0);
1641         RADEON_WRITE(R600_CB_COLOR3_BASE, 0);
1642         RADEON_WRITE(R600_CB_COLOR4_BASE, 0);
1643         RADEON_WRITE(R600_CB_COLOR5_BASE, 0);
1644         RADEON_WRITE(R600_CB_COLOR6_BASE, 0);
1645         RADEON_WRITE(R600_CB_COLOR7_BASE, 0);
1646
1647         RADEON_WRITE(R700_TCP_CNTL, 0);
1648
1649         hdp_host_path_cntl = RADEON_READ(R600_HDP_HOST_PATH_CNTL);
1650         RADEON_WRITE(R600_HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1651
1652         RADEON_WRITE(R600_PA_SC_MULTI_CHIP_CNTL, 0);
1653
1654         RADEON_WRITE(R600_PA_CL_ENHANCE, (R600_CLIP_VTX_REORDER_ENA |
1655                                           R600_NUM_CLIP_SEQ(3)));
1656
1657 }
1658
1659 static void r600_cp_init_ring_buffer(struct drm_device *dev,
1660                                        drm_radeon_private_t *dev_priv,
1661                                        struct drm_file *file_priv)
1662 {
1663         struct drm_radeon_master_private *master_priv;
1664         u32 ring_start;
1665         u64 rptr_addr;
1666
1667         if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770))
1668                 r700_gfx_init(dev, dev_priv);
1669         else
1670                 r600_gfx_init(dev, dev_priv);
1671
1672         RADEON_WRITE(R600_GRBM_SOFT_RESET, R600_SOFT_RESET_CP);
1673         RADEON_READ(R600_GRBM_SOFT_RESET);
1674         DRM_UDELAY(15000);
1675         RADEON_WRITE(R600_GRBM_SOFT_RESET, 0);
1676
1677
1678         /* Set ring buffer size */
1679 #ifdef __BIG_ENDIAN
1680         RADEON_WRITE(R600_CP_RB_CNTL,
1681                      RADEON_BUF_SWAP_32BIT |
1682                      RADEON_RB_NO_UPDATE |
1683                      (dev_priv->ring.rptr_update_l2qw << 8) |
1684                      dev_priv->ring.size_l2qw);
1685 #else
1686         RADEON_WRITE(R600_CP_RB_CNTL,
1687                      RADEON_RB_NO_UPDATE |
1688                      (dev_priv->ring.rptr_update_l2qw << 8) |
1689                      dev_priv->ring.size_l2qw);
1690 #endif
1691
1692         RADEON_WRITE(R600_CP_SEM_WAIT_TIMER, 0x4);
1693
1694         /* Set the write pointer delay */
1695         RADEON_WRITE(R600_CP_RB_WPTR_DELAY, 0);
1696
1697 #ifdef __BIG_ENDIAN
1698         RADEON_WRITE(R600_CP_RB_CNTL,
1699                      RADEON_BUF_SWAP_32BIT |
1700                      RADEON_RB_NO_UPDATE |
1701                      RADEON_RB_RPTR_WR_ENA |
1702                      (dev_priv->ring.rptr_update_l2qw << 8) |
1703                      dev_priv->ring.size_l2qw);
1704 #else
1705         RADEON_WRITE(R600_CP_RB_CNTL,
1706                      RADEON_RB_NO_UPDATE |
1707                      RADEON_RB_RPTR_WR_ENA |
1708                      (dev_priv->ring.rptr_update_l2qw << 8) |
1709                      dev_priv->ring.size_l2qw);
1710 #endif
1711
1712         /* Initialize the ring buffer's read and write pointers */
1713         RADEON_WRITE(R600_CP_RB_RPTR_WR, 0);
1714         RADEON_WRITE(R600_CP_RB_WPTR, 0);
1715         SET_RING_HEAD(dev_priv, 0);
1716         dev_priv->ring.tail = 0;
1717
1718 #if __OS_HAS_AGP
1719         if (dev_priv->flags & RADEON_IS_AGP) {
1720                 rptr_addr = dev_priv->ring_rptr->offset
1721                         - dev->agp->base +
1722                         dev_priv->gart_vm_start;
1723         } else
1724 #endif
1725         {
1726                 rptr_addr = dev_priv->ring_rptr->offset
1727                         - ((unsigned long) dev->sg->virtual)
1728                         + dev_priv->gart_vm_start;
1729         }
1730         RADEON_WRITE(R600_CP_RB_RPTR_ADDR,
1731                      rptr_addr & 0xffffffff);
1732         RADEON_WRITE(R600_CP_RB_RPTR_ADDR_HI,
1733                      upper_32_bits(rptr_addr));
1734
1735 #ifdef __BIG_ENDIAN
1736         RADEON_WRITE(R600_CP_RB_CNTL,
1737                      RADEON_BUF_SWAP_32BIT |
1738                      (dev_priv->ring.rptr_update_l2qw << 8) |
1739                      dev_priv->ring.size_l2qw);
1740 #else
1741         RADEON_WRITE(R600_CP_RB_CNTL,
1742                      (dev_priv->ring.rptr_update_l2qw << 8) |
1743                      dev_priv->ring.size_l2qw);
1744 #endif
1745
1746 #if __OS_HAS_AGP
1747         if (dev_priv->flags & RADEON_IS_AGP) {
1748                 /* XXX */
1749                 radeon_write_agp_base(dev_priv, dev->agp->base);
1750
1751                 /* XXX */
1752                 radeon_write_agp_location(dev_priv,
1753                              (((dev_priv->gart_vm_start - 1 +
1754                                 dev_priv->gart_size) & 0xffff0000) |
1755                               (dev_priv->gart_vm_start >> 16)));
1756
1757                 ring_start = (dev_priv->cp_ring->offset
1758                               - dev->agp->base
1759                               + dev_priv->gart_vm_start);
1760         } else
1761 #endif
1762                 ring_start = (dev_priv->cp_ring->offset
1763                               - (unsigned long)dev->sg->virtual
1764                               + dev_priv->gart_vm_start);
1765
1766         RADEON_WRITE(R600_CP_RB_BASE, ring_start >> 8);
1767
1768         RADEON_WRITE(R600_CP_ME_CNTL, 0xff);
1769
1770         RADEON_WRITE(R600_CP_DEBUG, (1 << 27) | (1 << 28));
1771
1772         /* Initialize the scratch register pointer.  This will cause
1773          * the scratch register values to be written out to memory
1774          * whenever they are updated.
1775          *
1776          * We simply put this behind the ring read pointer, this works
1777          * with PCI GART as well as (whatever kind of) AGP GART
1778          */
1779         {
1780                 u64 scratch_addr;
1781
1782                 scratch_addr = RADEON_READ(R600_CP_RB_RPTR_ADDR);
1783                 scratch_addr |= ((u64)RADEON_READ(R600_CP_RB_RPTR_ADDR_HI)) << 32;
1784                 scratch_addr += R600_SCRATCH_REG_OFFSET;
1785                 scratch_addr >>= 8;
1786                 scratch_addr &= 0xffffffff;
1787
1788                 RADEON_WRITE(R600_SCRATCH_ADDR, (uint32_t)scratch_addr);
1789         }
1790
1791         RADEON_WRITE(R600_SCRATCH_UMSK, 0x7);
1792
1793         /* Turn on bus mastering */
1794         radeon_enable_bm(dev_priv);
1795
1796         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(0), 0);
1797         RADEON_WRITE(R600_LAST_FRAME_REG, 0);
1798
1799         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(1), 0);
1800         RADEON_WRITE(R600_LAST_DISPATCH_REG, 0);
1801
1802         radeon_write_ring_rptr(dev_priv, R600_SCRATCHOFF(2), 0);
1803         RADEON_WRITE(R600_LAST_CLEAR_REG, 0);
1804
1805         /* reset sarea copies of these */
1806         master_priv = file_priv->master->driver_priv;
1807         if (master_priv->sarea_priv) {
1808                 master_priv->sarea_priv->last_frame = 0;
1809                 master_priv->sarea_priv->last_dispatch = 0;
1810                 master_priv->sarea_priv->last_clear = 0;
1811         }
1812
1813         r600_do_wait_for_idle(dev_priv);
1814
1815 }
1816
1817 int r600_do_cleanup_cp(struct drm_device *dev)
1818 {
1819         drm_radeon_private_t *dev_priv = dev->dev_private;
1820         DRM_DEBUG("\n");
1821
1822         /* Make sure interrupts are disabled here because the uninstall ioctl
1823          * may not have been called from userspace and after dev_private
1824          * is freed, it's too late.
1825          */
1826         if (dev->irq_enabled)
1827                 drm_irq_uninstall(dev);
1828
1829 #if __OS_HAS_AGP
1830         if (dev_priv->flags & RADEON_IS_AGP) {
1831                 if (dev_priv->cp_ring != NULL) {
1832                         drm_core_ioremapfree(dev_priv->cp_ring, dev);
1833                         dev_priv->cp_ring = NULL;
1834                 }
1835                 if (dev_priv->ring_rptr != NULL) {
1836                         drm_core_ioremapfree(dev_priv->ring_rptr, dev);
1837                         dev_priv->ring_rptr = NULL;
1838                 }
1839                 if (dev->agp_buffer_map != NULL) {
1840                         drm_core_ioremapfree(dev->agp_buffer_map, dev);
1841                         dev->agp_buffer_map = NULL;
1842                 }
1843         } else
1844 #endif
1845         {
1846
1847                 if (dev_priv->gart_info.bus_addr)
1848                         r600_page_table_cleanup(dev, &dev_priv->gart_info);
1849
1850                 if (dev_priv->gart_info.gart_table_location == DRM_ATI_GART_FB) {
1851                         drm_core_ioremapfree(&dev_priv->gart_info.mapping, dev);
1852                         dev_priv->gart_info.addr = NULL;
1853                 }
1854         }
1855         /* only clear to the start of flags */
1856         memset(dev_priv, 0, offsetof(drm_radeon_private_t, flags));
1857
1858         return 0;
1859 }
1860
1861 int r600_do_init_cp(struct drm_device *dev, drm_radeon_init_t *init,
1862                     struct drm_file *file_priv)
1863 {
1864         drm_radeon_private_t *dev_priv = dev->dev_private;
1865         struct drm_radeon_master_private *master_priv = file_priv->master->driver_priv;
1866
1867         DRM_DEBUG("\n");
1868
1869         /* if we require new memory map but we don't have it fail */
1870         if ((dev_priv->flags & RADEON_NEW_MEMMAP) && !dev_priv->new_memmap) {
1871                 DRM_ERROR("Cannot initialise DRM on this card\nThis card requires a new X.org DDX for 3D\n");
1872                 r600_do_cleanup_cp(dev);
1873                 return -EINVAL;
1874         }
1875
1876         if (init->is_pci && (dev_priv->flags & RADEON_IS_AGP)) {
1877                 DRM_DEBUG("Forcing AGP card to PCI mode\n");
1878                 dev_priv->flags &= ~RADEON_IS_AGP;
1879                 /* The writeback test succeeds, but when writeback is enabled,
1880                  * the ring buffer read ptr update fails after first 128 bytes.
1881                  */
1882                 radeon_no_wb = 1;
1883         } else if (!(dev_priv->flags & (RADEON_IS_AGP | RADEON_IS_PCI | RADEON_IS_PCIE))
1884                  && !init->is_pci) {
1885                 DRM_DEBUG("Restoring AGP flag\n");
1886                 dev_priv->flags |= RADEON_IS_AGP;
1887         }
1888
1889         dev_priv->usec_timeout = init->usec_timeout;
1890         if (dev_priv->usec_timeout < 1 ||
1891             dev_priv->usec_timeout > RADEON_MAX_USEC_TIMEOUT) {
1892                 DRM_DEBUG("TIMEOUT problem!\n");
1893                 r600_do_cleanup_cp(dev);
1894                 return -EINVAL;
1895         }
1896
1897         /* Enable vblank on CRTC1 for older X servers
1898          */
1899         dev_priv->vblank_crtc = DRM_RADEON_VBLANK_CRTC1;
1900
1901         dev_priv->cp_mode = init->cp_mode;
1902
1903         /* We don't support anything other than bus-mastering ring mode,
1904          * but the ring can be in either AGP or PCI space for the ring
1905          * read pointer.
1906          */
1907         if ((init->cp_mode != RADEON_CSQ_PRIBM_INDDIS) &&
1908             (init->cp_mode != RADEON_CSQ_PRIBM_INDBM)) {
1909                 DRM_DEBUG("BAD cp_mode (%x)!\n", init->cp_mode);
1910                 r600_do_cleanup_cp(dev);
1911                 return -EINVAL;
1912         }
1913
1914         switch (init->fb_bpp) {
1915         case 16:
1916                 dev_priv->color_fmt = RADEON_COLOR_FORMAT_RGB565;
1917                 break;
1918         case 32:
1919         default:
1920                 dev_priv->color_fmt = RADEON_COLOR_FORMAT_ARGB8888;
1921                 break;
1922         }
1923         dev_priv->front_offset = init->front_offset;
1924         dev_priv->front_pitch = init->front_pitch;
1925         dev_priv->back_offset = init->back_offset;
1926         dev_priv->back_pitch = init->back_pitch;
1927
1928         dev_priv->ring_offset = init->ring_offset;
1929         dev_priv->ring_rptr_offset = init->ring_rptr_offset;
1930         dev_priv->buffers_offset = init->buffers_offset;
1931         dev_priv->gart_textures_offset = init->gart_textures_offset;
1932
1933         master_priv->sarea = drm_getsarea(dev);
1934         if (!master_priv->sarea) {
1935                 DRM_ERROR("could not find sarea!\n");
1936                 r600_do_cleanup_cp(dev);
1937                 return -EINVAL;
1938         }
1939
1940         dev_priv->cp_ring = drm_core_findmap(dev, init->ring_offset);
1941         if (!dev_priv->cp_ring) {
1942                 DRM_ERROR("could not find cp ring region!\n");
1943                 r600_do_cleanup_cp(dev);
1944                 return -EINVAL;
1945         }
1946         dev_priv->ring_rptr = drm_core_findmap(dev, init->ring_rptr_offset);
1947         if (!dev_priv->ring_rptr) {
1948                 DRM_ERROR("could not find ring read pointer!\n");
1949                 r600_do_cleanup_cp(dev);
1950                 return -EINVAL;
1951         }
1952         dev->agp_buffer_token = init->buffers_offset;
1953         dev->agp_buffer_map = drm_core_findmap(dev, init->buffers_offset);
1954         if (!dev->agp_buffer_map) {
1955                 DRM_ERROR("could not find dma buffer region!\n");
1956                 r600_do_cleanup_cp(dev);
1957                 return -EINVAL;
1958         }
1959
1960         if (init->gart_textures_offset) {
1961                 dev_priv->gart_textures =
1962                     drm_core_findmap(dev, init->gart_textures_offset);
1963                 if (!dev_priv->gart_textures) {
1964                         DRM_ERROR("could not find GART texture region!\n");
1965                         r600_do_cleanup_cp(dev);
1966                         return -EINVAL;
1967                 }
1968         }
1969
1970 #if __OS_HAS_AGP
1971         /* XXX */
1972         if (dev_priv->flags & RADEON_IS_AGP) {
1973                 drm_core_ioremap_wc(dev_priv->cp_ring, dev);
1974                 drm_core_ioremap_wc(dev_priv->ring_rptr, dev);
1975                 drm_core_ioremap_wc(dev->agp_buffer_map, dev);
1976                 if (!dev_priv->cp_ring->handle ||
1977                     !dev_priv->ring_rptr->handle ||
1978                     !dev->agp_buffer_map->handle) {
1979                         DRM_ERROR("could not find ioremap agp regions!\n");
1980                         r600_do_cleanup_cp(dev);
1981                         return -EINVAL;
1982                 }
1983         } else
1984 #endif
1985         {
1986                 dev_priv->cp_ring->handle = (void *)dev_priv->cp_ring->offset;
1987                 dev_priv->ring_rptr->handle =
1988                     (void *)dev_priv->ring_rptr->offset;
1989                 dev->agp_buffer_map->handle =
1990                     (void *)dev->agp_buffer_map->offset;
1991
1992                 DRM_DEBUG("dev_priv->cp_ring->handle %p\n",
1993                           dev_priv->cp_ring->handle);
1994                 DRM_DEBUG("dev_priv->ring_rptr->handle %p\n",
1995                           dev_priv->ring_rptr->handle);
1996                 DRM_DEBUG("dev->agp_buffer_map->handle %p\n",
1997                           dev->agp_buffer_map->handle);
1998         }
1999
2000         dev_priv->fb_location = (radeon_read_fb_location(dev_priv) & 0xffff) << 24;
2001         dev_priv->fb_size =
2002                 (((radeon_read_fb_location(dev_priv) & 0xffff0000u) << 8) + 0x1000000)
2003                 - dev_priv->fb_location;
2004
2005         dev_priv->front_pitch_offset = (((dev_priv->front_pitch / 64) << 22) |
2006                                         ((dev_priv->front_offset
2007                                           + dev_priv->fb_location) >> 10));
2008
2009         dev_priv->back_pitch_offset = (((dev_priv->back_pitch / 64) << 22) |
2010                                        ((dev_priv->back_offset
2011                                          + dev_priv->fb_location) >> 10));
2012
2013         dev_priv->depth_pitch_offset = (((dev_priv->depth_pitch / 64) << 22) |
2014                                         ((dev_priv->depth_offset
2015                                           + dev_priv->fb_location) >> 10));
2016
2017         dev_priv->gart_size = init->gart_size;
2018
2019         /* New let's set the memory map ... */
2020         if (dev_priv->new_memmap) {
2021                 u32 base = 0;
2022
2023                 DRM_INFO("Setting GART location based on new memory map\n");
2024
2025                 /* If using AGP, try to locate the AGP aperture at the same
2026                  * location in the card and on the bus, though we have to
2027                  * align it down.
2028                  */
2029 #if __OS_HAS_AGP
2030                 /* XXX */
2031                 if (dev_priv->flags & RADEON_IS_AGP) {
2032                         base = dev->agp->base;
2033                         /* Check if valid */
2034                         if ((base + dev_priv->gart_size - 1) >= dev_priv->fb_location &&
2035                             base < (dev_priv->fb_location + dev_priv->fb_size - 1)) {
2036                                 DRM_INFO("Can't use AGP base @0x%08lx, won't fit\n",
2037                                          dev->agp->base);
2038                                 base = 0;
2039                         }
2040                 }
2041 #endif
2042                 /* If not or if AGP is at 0 (Macs), try to put it elsewhere */
2043                 if (base == 0) {
2044                         base = dev_priv->fb_location + dev_priv->fb_size;
2045                         if (base < dev_priv->fb_location ||
2046                             ((base + dev_priv->gart_size) & 0xfffffffful) < base)
2047                                 base = dev_priv->fb_location
2048                                         - dev_priv->gart_size;
2049                 }
2050                 dev_priv->gart_vm_start = base & 0xffc00000u;
2051                 if (dev_priv->gart_vm_start != base)
2052                         DRM_INFO("GART aligned down from 0x%08x to 0x%08x\n",
2053                                  base, dev_priv->gart_vm_start);
2054         }
2055
2056 #if __OS_HAS_AGP
2057         /* XXX */
2058         if (dev_priv->flags & RADEON_IS_AGP)
2059                 dev_priv->gart_buffers_offset = (dev->agp_buffer_map->offset
2060                                                  - dev->agp->base
2061                                                  + dev_priv->gart_vm_start);
2062         else
2063 #endif
2064                 dev_priv->gart_buffers_offset = (dev->agp_buffer_map->offset
2065                                                  - (unsigned long)dev->sg->virtual
2066                                                  + dev_priv->gart_vm_start);
2067
2068         DRM_DEBUG("fb 0x%08x size %d\n",
2069                   (unsigned int) dev_priv->fb_location,
2070                   (unsigned int) dev_priv->fb_size);
2071         DRM_DEBUG("dev_priv->gart_size %d\n", dev_priv->gart_size);
2072         DRM_DEBUG("dev_priv->gart_vm_start 0x%08x\n",
2073                   (unsigned int) dev_priv->gart_vm_start);
2074         DRM_DEBUG("dev_priv->gart_buffers_offset 0x%08lx\n",
2075                   dev_priv->gart_buffers_offset);
2076
2077         dev_priv->ring.start = (u32 *) dev_priv->cp_ring->handle;
2078         dev_priv->ring.end = ((u32 *) dev_priv->cp_ring->handle
2079                               + init->ring_size / sizeof(u32));
2080         dev_priv->ring.size = init->ring_size;
2081         dev_priv->ring.size_l2qw = drm_order(init->ring_size / 8);
2082
2083         dev_priv->ring.rptr_update = /* init->rptr_update */ 4096;
2084         dev_priv->ring.rptr_update_l2qw = drm_order(/* init->rptr_update */ 4096 / 8);
2085
2086         dev_priv->ring.fetch_size = /* init->fetch_size */ 32;
2087         dev_priv->ring.fetch_size_l2ow = drm_order(/* init->fetch_size */ 32 / 16);
2088
2089         dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
2090
2091         dev_priv->ring.high_mark = RADEON_RING_HIGH_MARK;
2092
2093 #if __OS_HAS_AGP
2094         if (dev_priv->flags & RADEON_IS_AGP) {
2095                 /* XXX turn off pcie gart */
2096         } else
2097 #endif
2098         {
2099                 dev_priv->gart_info.table_mask = DMA_BIT_MASK(32);
2100                 /* if we have an offset set from userspace */
2101                 if (!dev_priv->pcigart_offset_set) {
2102                         DRM_ERROR("Need gart offset from userspace\n");
2103                         r600_do_cleanup_cp(dev);
2104                         return -EINVAL;
2105                 }
2106
2107                 DRM_DEBUG("Using gart offset 0x%08lx\n", dev_priv->pcigart_offset);
2108
2109                 dev_priv->gart_info.bus_addr =
2110                         dev_priv->pcigart_offset + dev_priv->fb_location;
2111                 dev_priv->gart_info.mapping.offset =
2112                         dev_priv->pcigart_offset + dev_priv->fb_aper_offset;
2113                 dev_priv->gart_info.mapping.size =
2114                         dev_priv->gart_info.table_size;
2115
2116                 drm_core_ioremap_wc(&dev_priv->gart_info.mapping, dev);
2117                 if (!dev_priv->gart_info.mapping.handle) {
2118                         DRM_ERROR("ioremap failed.\n");
2119                         r600_do_cleanup_cp(dev);
2120                         return -EINVAL;
2121                 }
2122
2123                 dev_priv->gart_info.addr =
2124                         dev_priv->gart_info.mapping.handle;
2125
2126                 DRM_DEBUG("Setting phys_pci_gart to %p %08lX\n",
2127                           dev_priv->gart_info.addr,
2128                           dev_priv->pcigart_offset);
2129
2130                 if (!r600_page_table_init(dev)) {
2131                         DRM_ERROR("Failed to init GART table\n");
2132                         r600_do_cleanup_cp(dev);
2133                         return -EINVAL;
2134                 }
2135
2136                 if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770))
2137                         r700_vm_init(dev);
2138                 else
2139                         r600_vm_init(dev);
2140         }
2141
2142         if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770))
2143                 r700_cp_load_microcode(dev_priv);
2144         else
2145                 r600_cp_load_microcode(dev_priv);
2146
2147         r600_cp_init_ring_buffer(dev, dev_priv, file_priv);
2148
2149         dev_priv->last_buf = 0;
2150
2151         r600_do_engine_reset(dev);
2152         r600_test_writeback(dev_priv);
2153
2154         return 0;
2155 }
2156
2157 int r600_do_resume_cp(struct drm_device *dev, struct drm_file *file_priv)
2158 {
2159         drm_radeon_private_t *dev_priv = dev->dev_private;
2160
2161         DRM_DEBUG("\n");
2162         if (((dev_priv->flags & RADEON_FAMILY_MASK) >= CHIP_RV770)) {
2163                 r700_vm_init(dev);
2164                 r700_cp_load_microcode(dev_priv);
2165         } else {
2166                 r600_vm_init(dev);
2167                 r600_cp_load_microcode(dev_priv);
2168         }
2169         r600_cp_init_ring_buffer(dev, dev_priv, file_priv);
2170         r600_do_engine_reset(dev);
2171
2172         return 0;
2173 }
2174
2175 /* Wait for the CP to go idle.
2176  */
2177 int r600_do_cp_idle(drm_radeon_private_t *dev_priv)
2178 {
2179         RING_LOCALS;
2180         DRM_DEBUG("\n");
2181
2182         BEGIN_RING(5);
2183         OUT_RING(CP_PACKET3(R600_IT_EVENT_WRITE, 0));
2184         OUT_RING(R600_CACHE_FLUSH_AND_INV_EVENT);
2185         /* wait for 3D idle clean */
2186         OUT_RING(CP_PACKET3(R600_IT_SET_CONFIG_REG, 1));
2187         OUT_RING((R600_WAIT_UNTIL - R600_SET_CONFIG_REG_OFFSET) >> 2);
2188         OUT_RING(RADEON_WAIT_3D_IDLE | RADEON_WAIT_3D_IDLECLEAN);
2189
2190         ADVANCE_RING();
2191         COMMIT_RING();
2192
2193         return r600_do_wait_for_idle(dev_priv);
2194 }
2195
2196 /* Start the Command Processor.
2197  */
2198 void r600_do_cp_start(drm_radeon_private_t *dev_priv)
2199 {
2200         u32 cp_me;
2201         RING_LOCALS;
2202         DRM_DEBUG("\n");
2203
2204         BEGIN_RING(7);
2205         OUT_RING(CP_PACKET3(R600_IT_ME_INITIALIZE, 5));
2206         OUT_RING(0x00000001);
2207         if (((dev_priv->flags & RADEON_FAMILY_MASK) < CHIP_RV770))
2208                 OUT_RING(0x00000003);
2209         else
2210                 OUT_RING(0x00000000);
2211         OUT_RING((dev_priv->r600_max_hw_contexts - 1));
2212         OUT_RING(R600_ME_INITIALIZE_DEVICE_ID(1));
2213         OUT_RING(0x00000000);
2214         OUT_RING(0x00000000);
2215         ADVANCE_RING();
2216         COMMIT_RING();
2217
2218         /* set the mux and reset the halt bit */
2219         cp_me = 0xff;
2220         RADEON_WRITE(R600_CP_ME_CNTL, cp_me);
2221
2222         dev_priv->cp_running = 1;
2223
2224 }
2225
2226 void r600_do_cp_reset(drm_radeon_private_t *dev_priv)
2227 {
2228         u32 cur_read_ptr;
2229         DRM_DEBUG("\n");
2230
2231         cur_read_ptr = RADEON_READ(R600_CP_RB_RPTR);
2232         RADEON_WRITE(R600_CP_RB_WPTR, cur_read_ptr);
2233         SET_RING_HEAD(dev_priv, cur_read_ptr);
2234         dev_priv->ring.tail = cur_read_ptr;
2235 }
2236
2237 void r600_do_cp_stop(drm_radeon_private_t *dev_priv)
2238 {
2239         uint32_t cp_me;
2240
2241         DRM_DEBUG("\n");
2242
2243         cp_me = 0xff | R600_CP_ME_HALT;
2244
2245         RADEON_WRITE(R600_CP_ME_CNTL, cp_me);
2246
2247         dev_priv->cp_running = 0;
2248 }
2249
2250 int r600_cp_dispatch_indirect(struct drm_device *dev,
2251                               struct drm_buf *buf, int start, int end)
2252 {
2253         drm_radeon_private_t *dev_priv = dev->dev_private;
2254         RING_LOCALS;
2255
2256         if (start != end) {
2257                 unsigned long offset = (dev_priv->gart_buffers_offset
2258                                         + buf->offset + start);
2259                 int dwords = (end - start + 3) / sizeof(u32);
2260
2261                 DRM_DEBUG("dwords:%d\n", dwords);
2262                 DRM_DEBUG("offset 0x%lx\n", offset);
2263
2264
2265                 /* Indirect buffer data must be a multiple of 16 dwords.
2266                  * pad the data with a Type-2 CP packet.
2267                  */
2268                 while (dwords & 0xf) {
2269                         u32 *data = (u32 *)
2270                             ((char *)dev->agp_buffer_map->handle
2271                              + buf->offset + start);
2272                         data[dwords++] = RADEON_CP_PACKET2;
2273                 }
2274
2275                 /* Fire off the indirect buffer */
2276                 BEGIN_RING(4);
2277                 OUT_RING(CP_PACKET3(R600_IT_INDIRECT_BUFFER, 2));
2278                 OUT_RING((offset & 0xfffffffc));
2279                 OUT_RING((upper_32_bits(offset) & 0xff));
2280                 OUT_RING(dwords);
2281                 ADVANCE_RING();
2282         }
2283
2284         return 0;
2285 }