]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - drivers/gpu/drm/radeon/evergreen.c
drm/radeon: switch to a finer grained reset for evergreen
[~shefty/rdma-dev.git] / drivers / gpu / drm / radeon / evergreen.c
1 /*
2  * Copyright 2010 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  * Authors: Alex Deucher
23  */
24 #include <linux/firmware.h>
25 #include <linux/platform_device.h>
26 #include <linux/slab.h>
27 #include <drm/drmP.h>
28 #include "radeon.h"
29 #include "radeon_asic.h"
30 #include <drm/radeon_drm.h>
31 #include "evergreend.h"
32 #include "atom.h"
33 #include "avivod.h"
34 #include "evergreen_reg.h"
35 #include "evergreen_blit_shaders.h"
36
37 #define EVERGREEN_PFP_UCODE_SIZE 1120
38 #define EVERGREEN_PM4_UCODE_SIZE 1376
39
40 static const u32 crtc_offsets[6] =
41 {
42         EVERGREEN_CRTC0_REGISTER_OFFSET,
43         EVERGREEN_CRTC1_REGISTER_OFFSET,
44         EVERGREEN_CRTC2_REGISTER_OFFSET,
45         EVERGREEN_CRTC3_REGISTER_OFFSET,
46         EVERGREEN_CRTC4_REGISTER_OFFSET,
47         EVERGREEN_CRTC5_REGISTER_OFFSET
48 };
49
50 static void evergreen_gpu_init(struct radeon_device *rdev);
51 void evergreen_fini(struct radeon_device *rdev);
52 void evergreen_pcie_gen2_enable(struct radeon_device *rdev);
53 extern void cayman_cp_int_cntl_setup(struct radeon_device *rdev,
54                                      int ring, u32 cp_int_cntl);
55
56 void evergreen_tiling_fields(unsigned tiling_flags, unsigned *bankw,
57                              unsigned *bankh, unsigned *mtaspect,
58                              unsigned *tile_split)
59 {
60         *bankw = (tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
61         *bankh = (tiling_flags >> RADEON_TILING_EG_BANKH_SHIFT) & RADEON_TILING_EG_BANKH_MASK;
62         *mtaspect = (tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
63         *tile_split = (tiling_flags >> RADEON_TILING_EG_TILE_SPLIT_SHIFT) & RADEON_TILING_EG_TILE_SPLIT_MASK;
64         switch (*bankw) {
65         default:
66         case 1: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_1; break;
67         case 2: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_2; break;
68         case 4: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_4; break;
69         case 8: *bankw = EVERGREEN_ADDR_SURF_BANK_WIDTH_8; break;
70         }
71         switch (*bankh) {
72         default:
73         case 1: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_1; break;
74         case 2: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_2; break;
75         case 4: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_4; break;
76         case 8: *bankh = EVERGREEN_ADDR_SURF_BANK_HEIGHT_8; break;
77         }
78         switch (*mtaspect) {
79         default:
80         case 1: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_1; break;
81         case 2: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_2; break;
82         case 4: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_4; break;
83         case 8: *mtaspect = EVERGREEN_ADDR_SURF_MACRO_TILE_ASPECT_8; break;
84         }
85 }
86
87 void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev)
88 {
89         u16 ctl, v;
90         int err;
91
92         err = pcie_capability_read_word(rdev->pdev, PCI_EXP_DEVCTL, &ctl);
93         if (err)
94                 return;
95
96         v = (ctl & PCI_EXP_DEVCTL_READRQ) >> 12;
97
98         /* if bios or OS sets MAX_READ_REQUEST_SIZE to an invalid value, fix it
99          * to avoid hangs or perfomance issues
100          */
101         if ((v == 0) || (v == 6) || (v == 7)) {
102                 ctl &= ~PCI_EXP_DEVCTL_READRQ;
103                 ctl |= (2 << 12);
104                 pcie_capability_write_word(rdev->pdev, PCI_EXP_DEVCTL, ctl);
105         }
106 }
107
108 /**
109  * dce4_wait_for_vblank - vblank wait asic callback.
110  *
111  * @rdev: radeon_device pointer
112  * @crtc: crtc to wait for vblank on
113  *
114  * Wait for vblank on the requested crtc (evergreen+).
115  */
116 void dce4_wait_for_vblank(struct radeon_device *rdev, int crtc)
117 {
118         int i;
119
120         if (crtc >= rdev->num_crtc)
121                 return;
122
123         if (RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[crtc]) & EVERGREEN_CRTC_MASTER_EN) {
124                 for (i = 0; i < rdev->usec_timeout; i++) {
125                         if (!(RREG32(EVERGREEN_CRTC_STATUS + crtc_offsets[crtc]) & EVERGREEN_CRTC_V_BLANK))
126                                 break;
127                         udelay(1);
128                 }
129                 for (i = 0; i < rdev->usec_timeout; i++) {
130                         if (RREG32(EVERGREEN_CRTC_STATUS + crtc_offsets[crtc]) & EVERGREEN_CRTC_V_BLANK)
131                                 break;
132                         udelay(1);
133                 }
134         }
135 }
136
137 /**
138  * radeon_irq_kms_pflip_irq_get - pre-pageflip callback.
139  *
140  * @rdev: radeon_device pointer
141  * @crtc: crtc to prepare for pageflip on
142  *
143  * Pre-pageflip callback (evergreen+).
144  * Enables the pageflip irq (vblank irq).
145  */
146 void evergreen_pre_page_flip(struct radeon_device *rdev, int crtc)
147 {
148         /* enable the pflip int */
149         radeon_irq_kms_pflip_irq_get(rdev, crtc);
150 }
151
152 /**
153  * evergreen_post_page_flip - pos-pageflip callback.
154  *
155  * @rdev: radeon_device pointer
156  * @crtc: crtc to cleanup pageflip on
157  *
158  * Post-pageflip callback (evergreen+).
159  * Disables the pageflip irq (vblank irq).
160  */
161 void evergreen_post_page_flip(struct radeon_device *rdev, int crtc)
162 {
163         /* disable the pflip int */
164         radeon_irq_kms_pflip_irq_put(rdev, crtc);
165 }
166
167 /**
168  * evergreen_page_flip - pageflip callback.
169  *
170  * @rdev: radeon_device pointer
171  * @crtc_id: crtc to cleanup pageflip on
172  * @crtc_base: new address of the crtc (GPU MC address)
173  *
174  * Does the actual pageflip (evergreen+).
175  * During vblank we take the crtc lock and wait for the update_pending
176  * bit to go high, when it does, we release the lock, and allow the
177  * double buffered update to take place.
178  * Returns the current update pending status.
179  */
180 u32 evergreen_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
181 {
182         struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
183         u32 tmp = RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset);
184         int i;
185
186         /* Lock the graphics update lock */
187         tmp |= EVERGREEN_GRPH_UPDATE_LOCK;
188         WREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
189
190         /* update the scanout addresses */
191         WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
192                upper_32_bits(crtc_base));
193         WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
194                (u32)crtc_base);
195
196         WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
197                upper_32_bits(crtc_base));
198         WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
199                (u32)crtc_base);
200
201         /* Wait for update_pending to go high. */
202         for (i = 0; i < rdev->usec_timeout; i++) {
203                 if (RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING)
204                         break;
205                 udelay(1);
206         }
207         DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
208
209         /* Unlock the lock, so double-buffering can take place inside vblank */
210         tmp &= ~EVERGREEN_GRPH_UPDATE_LOCK;
211         WREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
212
213         /* Return current update_pending status: */
214         return RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING;
215 }
216
217 /* get temperature in millidegrees */
218 int evergreen_get_temp(struct radeon_device *rdev)
219 {
220         u32 temp, toffset;
221         int actual_temp = 0;
222
223         if (rdev->family == CHIP_JUNIPER) {
224                 toffset = (RREG32(CG_THERMAL_CTRL) & TOFFSET_MASK) >>
225                         TOFFSET_SHIFT;
226                 temp = (RREG32(CG_TS0_STATUS) & TS0_ADC_DOUT_MASK) >>
227                         TS0_ADC_DOUT_SHIFT;
228
229                 if (toffset & 0x100)
230                         actual_temp = temp / 2 - (0x200 - toffset);
231                 else
232                         actual_temp = temp / 2 + toffset;
233
234                 actual_temp = actual_temp * 1000;
235
236         } else {
237                 temp = (RREG32(CG_MULT_THERMAL_STATUS) & ASIC_T_MASK) >>
238                         ASIC_T_SHIFT;
239
240                 if (temp & 0x400)
241                         actual_temp = -256;
242                 else if (temp & 0x200)
243                         actual_temp = 255;
244                 else if (temp & 0x100) {
245                         actual_temp = temp & 0x1ff;
246                         actual_temp |= ~0x1ff;
247                 } else
248                         actual_temp = temp & 0xff;
249
250                 actual_temp = (actual_temp * 1000) / 2;
251         }
252
253         return actual_temp;
254 }
255
256 int sumo_get_temp(struct radeon_device *rdev)
257 {
258         u32 temp = RREG32(CG_THERMAL_STATUS) & 0xff;
259         int actual_temp = temp - 49;
260
261         return actual_temp * 1000;
262 }
263
264 /**
265  * sumo_pm_init_profile - Initialize power profiles callback.
266  *
267  * @rdev: radeon_device pointer
268  *
269  * Initialize the power states used in profile mode
270  * (sumo, trinity, SI).
271  * Used for profile mode only.
272  */
273 void sumo_pm_init_profile(struct radeon_device *rdev)
274 {
275         int idx;
276
277         /* default */
278         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
279         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
280         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
281         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
282
283         /* low,mid sh/mh */
284         if (rdev->flags & RADEON_IS_MOBILITY)
285                 idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_BATTERY, 0);
286         else
287                 idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
288
289         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = idx;
290         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = idx;
291         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
292         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
293
294         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = idx;
295         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = idx;
296         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
297         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
298
299         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = idx;
300         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = idx;
301         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
302         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
303
304         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = idx;
305         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = idx;
306         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
307         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
308
309         /* high sh/mh */
310         idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
311         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = idx;
312         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = idx;
313         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
314         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx =
315                 rdev->pm.power_state[idx].num_clock_modes - 1;
316
317         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = idx;
318         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = idx;
319         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
320         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx =
321                 rdev->pm.power_state[idx].num_clock_modes - 1;
322 }
323
324 /**
325  * btc_pm_init_profile - Initialize power profiles callback.
326  *
327  * @rdev: radeon_device pointer
328  *
329  * Initialize the power states used in profile mode
330  * (BTC, cayman).
331  * Used for profile mode only.
332  */
333 void btc_pm_init_profile(struct radeon_device *rdev)
334 {
335         int idx;
336
337         /* default */
338         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
339         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
340         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
341         rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 2;
342         /* starting with BTC, there is one state that is used for both
343          * MH and SH.  Difference is that we always use the high clock index for
344          * mclk.
345          */
346         if (rdev->flags & RADEON_IS_MOBILITY)
347                 idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_BATTERY, 0);
348         else
349                 idx = radeon_pm_get_type_index(rdev, POWER_STATE_TYPE_PERFORMANCE, 0);
350         /* low sh */
351         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = idx;
352         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = idx;
353         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
354         rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
355         /* mid sh */
356         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = idx;
357         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = idx;
358         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
359         rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 1;
360         /* high sh */
361         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = idx;
362         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = idx;
363         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
364         rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 2;
365         /* low mh */
366         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = idx;
367         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = idx;
368         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
369         rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
370         /* mid mh */
371         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = idx;
372         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = idx;
373         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
374         rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 1;
375         /* high mh */
376         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = idx;
377         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = idx;
378         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
379         rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 2;
380 }
381
382 /**
383  * evergreen_pm_misc - set additional pm hw parameters callback.
384  *
385  * @rdev: radeon_device pointer
386  *
387  * Set non-clock parameters associated with a power state
388  * (voltage, etc.) (evergreen+).
389  */
390 void evergreen_pm_misc(struct radeon_device *rdev)
391 {
392         int req_ps_idx = rdev->pm.requested_power_state_index;
393         int req_cm_idx = rdev->pm.requested_clock_mode_index;
394         struct radeon_power_state *ps = &rdev->pm.power_state[req_ps_idx];
395         struct radeon_voltage *voltage = &ps->clock_info[req_cm_idx].voltage;
396
397         if (voltage->type == VOLTAGE_SW) {
398                 /* 0xff01 is a flag rather then an actual voltage */
399                 if (voltage->voltage == 0xff01)
400                         return;
401                 if (voltage->voltage && (voltage->voltage != rdev->pm.current_vddc)) {
402                         radeon_atom_set_voltage(rdev, voltage->voltage, SET_VOLTAGE_TYPE_ASIC_VDDC);
403                         rdev->pm.current_vddc = voltage->voltage;
404                         DRM_DEBUG("Setting: vddc: %d\n", voltage->voltage);
405                 }
406                 /* 0xff01 is a flag rather then an actual voltage */
407                 if (voltage->vddci == 0xff01)
408                         return;
409                 if (voltage->vddci && (voltage->vddci != rdev->pm.current_vddci)) {
410                         radeon_atom_set_voltage(rdev, voltage->vddci, SET_VOLTAGE_TYPE_ASIC_VDDCI);
411                         rdev->pm.current_vddci = voltage->vddci;
412                         DRM_DEBUG("Setting: vddci: %d\n", voltage->vddci);
413                 }
414         }
415 }
416
417 /**
418  * evergreen_pm_prepare - pre-power state change callback.
419  *
420  * @rdev: radeon_device pointer
421  *
422  * Prepare for a power state change (evergreen+).
423  */
424 void evergreen_pm_prepare(struct radeon_device *rdev)
425 {
426         struct drm_device *ddev = rdev->ddev;
427         struct drm_crtc *crtc;
428         struct radeon_crtc *radeon_crtc;
429         u32 tmp;
430
431         /* disable any active CRTCs */
432         list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
433                 radeon_crtc = to_radeon_crtc(crtc);
434                 if (radeon_crtc->enabled) {
435                         tmp = RREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset);
436                         tmp |= EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
437                         WREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset, tmp);
438                 }
439         }
440 }
441
442 /**
443  * evergreen_pm_finish - post-power state change callback.
444  *
445  * @rdev: radeon_device pointer
446  *
447  * Clean up after a power state change (evergreen+).
448  */
449 void evergreen_pm_finish(struct radeon_device *rdev)
450 {
451         struct drm_device *ddev = rdev->ddev;
452         struct drm_crtc *crtc;
453         struct radeon_crtc *radeon_crtc;
454         u32 tmp;
455
456         /* enable any active CRTCs */
457         list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
458                 radeon_crtc = to_radeon_crtc(crtc);
459                 if (radeon_crtc->enabled) {
460                         tmp = RREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset);
461                         tmp &= ~EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
462                         WREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset, tmp);
463                 }
464         }
465 }
466
467 /**
468  * evergreen_hpd_sense - hpd sense callback.
469  *
470  * @rdev: radeon_device pointer
471  * @hpd: hpd (hotplug detect) pin
472  *
473  * Checks if a digital monitor is connected (evergreen+).
474  * Returns true if connected, false if not connected.
475  */
476 bool evergreen_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
477 {
478         bool connected = false;
479
480         switch (hpd) {
481         case RADEON_HPD_1:
482                 if (RREG32(DC_HPD1_INT_STATUS) & DC_HPDx_SENSE)
483                         connected = true;
484                 break;
485         case RADEON_HPD_2:
486                 if (RREG32(DC_HPD2_INT_STATUS) & DC_HPDx_SENSE)
487                         connected = true;
488                 break;
489         case RADEON_HPD_3:
490                 if (RREG32(DC_HPD3_INT_STATUS) & DC_HPDx_SENSE)
491                         connected = true;
492                 break;
493         case RADEON_HPD_4:
494                 if (RREG32(DC_HPD4_INT_STATUS) & DC_HPDx_SENSE)
495                         connected = true;
496                 break;
497         case RADEON_HPD_5:
498                 if (RREG32(DC_HPD5_INT_STATUS) & DC_HPDx_SENSE)
499                         connected = true;
500                 break;
501         case RADEON_HPD_6:
502                 if (RREG32(DC_HPD6_INT_STATUS) & DC_HPDx_SENSE)
503                         connected = true;
504                         break;
505         default:
506                 break;
507         }
508
509         return connected;
510 }
511
512 /**
513  * evergreen_hpd_set_polarity - hpd set polarity callback.
514  *
515  * @rdev: radeon_device pointer
516  * @hpd: hpd (hotplug detect) pin
517  *
518  * Set the polarity of the hpd pin (evergreen+).
519  */
520 void evergreen_hpd_set_polarity(struct radeon_device *rdev,
521                                 enum radeon_hpd_id hpd)
522 {
523         u32 tmp;
524         bool connected = evergreen_hpd_sense(rdev, hpd);
525
526         switch (hpd) {
527         case RADEON_HPD_1:
528                 tmp = RREG32(DC_HPD1_INT_CONTROL);
529                 if (connected)
530                         tmp &= ~DC_HPDx_INT_POLARITY;
531                 else
532                         tmp |= DC_HPDx_INT_POLARITY;
533                 WREG32(DC_HPD1_INT_CONTROL, tmp);
534                 break;
535         case RADEON_HPD_2:
536                 tmp = RREG32(DC_HPD2_INT_CONTROL);
537                 if (connected)
538                         tmp &= ~DC_HPDx_INT_POLARITY;
539                 else
540                         tmp |= DC_HPDx_INT_POLARITY;
541                 WREG32(DC_HPD2_INT_CONTROL, tmp);
542                 break;
543         case RADEON_HPD_3:
544                 tmp = RREG32(DC_HPD3_INT_CONTROL);
545                 if (connected)
546                         tmp &= ~DC_HPDx_INT_POLARITY;
547                 else
548                         tmp |= DC_HPDx_INT_POLARITY;
549                 WREG32(DC_HPD3_INT_CONTROL, tmp);
550                 break;
551         case RADEON_HPD_4:
552                 tmp = RREG32(DC_HPD4_INT_CONTROL);
553                 if (connected)
554                         tmp &= ~DC_HPDx_INT_POLARITY;
555                 else
556                         tmp |= DC_HPDx_INT_POLARITY;
557                 WREG32(DC_HPD4_INT_CONTROL, tmp);
558                 break;
559         case RADEON_HPD_5:
560                 tmp = RREG32(DC_HPD5_INT_CONTROL);
561                 if (connected)
562                         tmp &= ~DC_HPDx_INT_POLARITY;
563                 else
564                         tmp |= DC_HPDx_INT_POLARITY;
565                 WREG32(DC_HPD5_INT_CONTROL, tmp);
566                         break;
567         case RADEON_HPD_6:
568                 tmp = RREG32(DC_HPD6_INT_CONTROL);
569                 if (connected)
570                         tmp &= ~DC_HPDx_INT_POLARITY;
571                 else
572                         tmp |= DC_HPDx_INT_POLARITY;
573                 WREG32(DC_HPD6_INT_CONTROL, tmp);
574                 break;
575         default:
576                 break;
577         }
578 }
579
580 /**
581  * evergreen_hpd_init - hpd setup callback.
582  *
583  * @rdev: radeon_device pointer
584  *
585  * Setup the hpd pins used by the card (evergreen+).
586  * Enable the pin, set the polarity, and enable the hpd interrupts.
587  */
588 void evergreen_hpd_init(struct radeon_device *rdev)
589 {
590         struct drm_device *dev = rdev->ddev;
591         struct drm_connector *connector;
592         unsigned enabled = 0;
593         u32 tmp = DC_HPDx_CONNECTION_TIMER(0x9c4) |
594                 DC_HPDx_RX_INT_TIMER(0xfa) | DC_HPDx_EN;
595
596         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
597                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
598                 switch (radeon_connector->hpd.hpd) {
599                 case RADEON_HPD_1:
600                         WREG32(DC_HPD1_CONTROL, tmp);
601                         break;
602                 case RADEON_HPD_2:
603                         WREG32(DC_HPD2_CONTROL, tmp);
604                         break;
605                 case RADEON_HPD_3:
606                         WREG32(DC_HPD3_CONTROL, tmp);
607                         break;
608                 case RADEON_HPD_4:
609                         WREG32(DC_HPD4_CONTROL, tmp);
610                         break;
611                 case RADEON_HPD_5:
612                         WREG32(DC_HPD5_CONTROL, tmp);
613                         break;
614                 case RADEON_HPD_6:
615                         WREG32(DC_HPD6_CONTROL, tmp);
616                         break;
617                 default:
618                         break;
619                 }
620                 radeon_hpd_set_polarity(rdev, radeon_connector->hpd.hpd);
621                 enabled |= 1 << radeon_connector->hpd.hpd;
622         }
623         radeon_irq_kms_enable_hpd(rdev, enabled);
624 }
625
626 /**
627  * evergreen_hpd_fini - hpd tear down callback.
628  *
629  * @rdev: radeon_device pointer
630  *
631  * Tear down the hpd pins used by the card (evergreen+).
632  * Disable the hpd interrupts.
633  */
634 void evergreen_hpd_fini(struct radeon_device *rdev)
635 {
636         struct drm_device *dev = rdev->ddev;
637         struct drm_connector *connector;
638         unsigned disabled = 0;
639
640         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
641                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
642                 switch (radeon_connector->hpd.hpd) {
643                 case RADEON_HPD_1:
644                         WREG32(DC_HPD1_CONTROL, 0);
645                         break;
646                 case RADEON_HPD_2:
647                         WREG32(DC_HPD2_CONTROL, 0);
648                         break;
649                 case RADEON_HPD_3:
650                         WREG32(DC_HPD3_CONTROL, 0);
651                         break;
652                 case RADEON_HPD_4:
653                         WREG32(DC_HPD4_CONTROL, 0);
654                         break;
655                 case RADEON_HPD_5:
656                         WREG32(DC_HPD5_CONTROL, 0);
657                         break;
658                 case RADEON_HPD_6:
659                         WREG32(DC_HPD6_CONTROL, 0);
660                         break;
661                 default:
662                         break;
663                 }
664                 disabled |= 1 << radeon_connector->hpd.hpd;
665         }
666         radeon_irq_kms_disable_hpd(rdev, disabled);
667 }
668
669 /* watermark setup */
670
671 static u32 evergreen_line_buffer_adjust(struct radeon_device *rdev,
672                                         struct radeon_crtc *radeon_crtc,
673                                         struct drm_display_mode *mode,
674                                         struct drm_display_mode *other_mode)
675 {
676         u32 tmp;
677         /*
678          * Line Buffer Setup
679          * There are 3 line buffers, each one shared by 2 display controllers.
680          * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
681          * the display controllers.  The paritioning is done via one of four
682          * preset allocations specified in bits 2:0:
683          * first display controller
684          *  0 - first half of lb (3840 * 2)
685          *  1 - first 3/4 of lb (5760 * 2)
686          *  2 - whole lb (7680 * 2), other crtc must be disabled
687          *  3 - first 1/4 of lb (1920 * 2)
688          * second display controller
689          *  4 - second half of lb (3840 * 2)
690          *  5 - second 3/4 of lb (5760 * 2)
691          *  6 - whole lb (7680 * 2), other crtc must be disabled
692          *  7 - last 1/4 of lb (1920 * 2)
693          */
694         /* this can get tricky if we have two large displays on a paired group
695          * of crtcs.  Ideally for multiple large displays we'd assign them to
696          * non-linked crtcs for maximum line buffer allocation.
697          */
698         if (radeon_crtc->base.enabled && mode) {
699                 if (other_mode)
700                         tmp = 0; /* 1/2 */
701                 else
702                         tmp = 2; /* whole */
703         } else
704                 tmp = 0;
705
706         /* second controller of the pair uses second half of the lb */
707         if (radeon_crtc->crtc_id % 2)
708                 tmp += 4;
709         WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset, tmp);
710
711         if (radeon_crtc->base.enabled && mode) {
712                 switch (tmp) {
713                 case 0:
714                 case 4:
715                 default:
716                         if (ASIC_IS_DCE5(rdev))
717                                 return 4096 * 2;
718                         else
719                                 return 3840 * 2;
720                 case 1:
721                 case 5:
722                         if (ASIC_IS_DCE5(rdev))
723                                 return 6144 * 2;
724                         else
725                                 return 5760 * 2;
726                 case 2:
727                 case 6:
728                         if (ASIC_IS_DCE5(rdev))
729                                 return 8192 * 2;
730                         else
731                                 return 7680 * 2;
732                 case 3:
733                 case 7:
734                         if (ASIC_IS_DCE5(rdev))
735                                 return 2048 * 2;
736                         else
737                                 return 1920 * 2;
738                 }
739         }
740
741         /* controller not enabled, so no lb used */
742         return 0;
743 }
744
745 u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev)
746 {
747         u32 tmp = RREG32(MC_SHARED_CHMAP);
748
749         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
750         case 0:
751         default:
752                 return 1;
753         case 1:
754                 return 2;
755         case 2:
756                 return 4;
757         case 3:
758                 return 8;
759         }
760 }
761
762 struct evergreen_wm_params {
763         u32 dram_channels; /* number of dram channels */
764         u32 yclk;          /* bandwidth per dram data pin in kHz */
765         u32 sclk;          /* engine clock in kHz */
766         u32 disp_clk;      /* display clock in kHz */
767         u32 src_width;     /* viewport width */
768         u32 active_time;   /* active display time in ns */
769         u32 blank_time;    /* blank time in ns */
770         bool interlaced;    /* mode is interlaced */
771         fixed20_12 vsc;    /* vertical scale ratio */
772         u32 num_heads;     /* number of active crtcs */
773         u32 bytes_per_pixel; /* bytes per pixel display + overlay */
774         u32 lb_size;       /* line buffer allocated to pipe */
775         u32 vtaps;         /* vertical scaler taps */
776 };
777
778 static u32 evergreen_dram_bandwidth(struct evergreen_wm_params *wm)
779 {
780         /* Calculate DRAM Bandwidth and the part allocated to display. */
781         fixed20_12 dram_efficiency; /* 0.7 */
782         fixed20_12 yclk, dram_channels, bandwidth;
783         fixed20_12 a;
784
785         a.full = dfixed_const(1000);
786         yclk.full = dfixed_const(wm->yclk);
787         yclk.full = dfixed_div(yclk, a);
788         dram_channels.full = dfixed_const(wm->dram_channels * 4);
789         a.full = dfixed_const(10);
790         dram_efficiency.full = dfixed_const(7);
791         dram_efficiency.full = dfixed_div(dram_efficiency, a);
792         bandwidth.full = dfixed_mul(dram_channels, yclk);
793         bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
794
795         return dfixed_trunc(bandwidth);
796 }
797
798 static u32 evergreen_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
799 {
800         /* Calculate DRAM Bandwidth and the part allocated to display. */
801         fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
802         fixed20_12 yclk, dram_channels, bandwidth;
803         fixed20_12 a;
804
805         a.full = dfixed_const(1000);
806         yclk.full = dfixed_const(wm->yclk);
807         yclk.full = dfixed_div(yclk, a);
808         dram_channels.full = dfixed_const(wm->dram_channels * 4);
809         a.full = dfixed_const(10);
810         disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
811         disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
812         bandwidth.full = dfixed_mul(dram_channels, yclk);
813         bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
814
815         return dfixed_trunc(bandwidth);
816 }
817
818 static u32 evergreen_data_return_bandwidth(struct evergreen_wm_params *wm)
819 {
820         /* Calculate the display Data return Bandwidth */
821         fixed20_12 return_efficiency; /* 0.8 */
822         fixed20_12 sclk, bandwidth;
823         fixed20_12 a;
824
825         a.full = dfixed_const(1000);
826         sclk.full = dfixed_const(wm->sclk);
827         sclk.full = dfixed_div(sclk, a);
828         a.full = dfixed_const(10);
829         return_efficiency.full = dfixed_const(8);
830         return_efficiency.full = dfixed_div(return_efficiency, a);
831         a.full = dfixed_const(32);
832         bandwidth.full = dfixed_mul(a, sclk);
833         bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
834
835         return dfixed_trunc(bandwidth);
836 }
837
838 static u32 evergreen_dmif_request_bandwidth(struct evergreen_wm_params *wm)
839 {
840         /* Calculate the DMIF Request Bandwidth */
841         fixed20_12 disp_clk_request_efficiency; /* 0.8 */
842         fixed20_12 disp_clk, bandwidth;
843         fixed20_12 a;
844
845         a.full = dfixed_const(1000);
846         disp_clk.full = dfixed_const(wm->disp_clk);
847         disp_clk.full = dfixed_div(disp_clk, a);
848         a.full = dfixed_const(10);
849         disp_clk_request_efficiency.full = dfixed_const(8);
850         disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
851         a.full = dfixed_const(32);
852         bandwidth.full = dfixed_mul(a, disp_clk);
853         bandwidth.full = dfixed_mul(bandwidth, disp_clk_request_efficiency);
854
855         return dfixed_trunc(bandwidth);
856 }
857
858 static u32 evergreen_available_bandwidth(struct evergreen_wm_params *wm)
859 {
860         /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
861         u32 dram_bandwidth = evergreen_dram_bandwidth(wm);
862         u32 data_return_bandwidth = evergreen_data_return_bandwidth(wm);
863         u32 dmif_req_bandwidth = evergreen_dmif_request_bandwidth(wm);
864
865         return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
866 }
867
868 static u32 evergreen_average_bandwidth(struct evergreen_wm_params *wm)
869 {
870         /* Calculate the display mode Average Bandwidth
871          * DisplayMode should contain the source and destination dimensions,
872          * timing, etc.
873          */
874         fixed20_12 bpp;
875         fixed20_12 line_time;
876         fixed20_12 src_width;
877         fixed20_12 bandwidth;
878         fixed20_12 a;
879
880         a.full = dfixed_const(1000);
881         line_time.full = dfixed_const(wm->active_time + wm->blank_time);
882         line_time.full = dfixed_div(line_time, a);
883         bpp.full = dfixed_const(wm->bytes_per_pixel);
884         src_width.full = dfixed_const(wm->src_width);
885         bandwidth.full = dfixed_mul(src_width, bpp);
886         bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
887         bandwidth.full = dfixed_div(bandwidth, line_time);
888
889         return dfixed_trunc(bandwidth);
890 }
891
892 static u32 evergreen_latency_watermark(struct evergreen_wm_params *wm)
893 {
894         /* First calcualte the latency in ns */
895         u32 mc_latency = 2000; /* 2000 ns. */
896         u32 available_bandwidth = evergreen_available_bandwidth(wm);
897         u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
898         u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
899         u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
900         u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
901                 (wm->num_heads * cursor_line_pair_return_time);
902         u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
903         u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
904         fixed20_12 a, b, c;
905
906         if (wm->num_heads == 0)
907                 return 0;
908
909         a.full = dfixed_const(2);
910         b.full = dfixed_const(1);
911         if ((wm->vsc.full > a.full) ||
912             ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
913             (wm->vtaps >= 5) ||
914             ((wm->vsc.full >= a.full) && wm->interlaced))
915                 max_src_lines_per_dst_line = 4;
916         else
917                 max_src_lines_per_dst_line = 2;
918
919         a.full = dfixed_const(available_bandwidth);
920         b.full = dfixed_const(wm->num_heads);
921         a.full = dfixed_div(a, b);
922
923         b.full = dfixed_const(1000);
924         c.full = dfixed_const(wm->disp_clk);
925         b.full = dfixed_div(c, b);
926         c.full = dfixed_const(wm->bytes_per_pixel);
927         b.full = dfixed_mul(b, c);
928
929         lb_fill_bw = min(dfixed_trunc(a), dfixed_trunc(b));
930
931         a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
932         b.full = dfixed_const(1000);
933         c.full = dfixed_const(lb_fill_bw);
934         b.full = dfixed_div(c, b);
935         a.full = dfixed_div(a, b);
936         line_fill_time = dfixed_trunc(a);
937
938         if (line_fill_time < wm->active_time)
939                 return latency;
940         else
941                 return latency + (line_fill_time - wm->active_time);
942
943 }
944
945 static bool evergreen_average_bandwidth_vs_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
946 {
947         if (evergreen_average_bandwidth(wm) <=
948             (evergreen_dram_bandwidth_for_display(wm) / wm->num_heads))
949                 return true;
950         else
951                 return false;
952 };
953
954 static bool evergreen_average_bandwidth_vs_available_bandwidth(struct evergreen_wm_params *wm)
955 {
956         if (evergreen_average_bandwidth(wm) <=
957             (evergreen_available_bandwidth(wm) / wm->num_heads))
958                 return true;
959         else
960                 return false;
961 };
962
963 static bool evergreen_check_latency_hiding(struct evergreen_wm_params *wm)
964 {
965         u32 lb_partitions = wm->lb_size / wm->src_width;
966         u32 line_time = wm->active_time + wm->blank_time;
967         u32 latency_tolerant_lines;
968         u32 latency_hiding;
969         fixed20_12 a;
970
971         a.full = dfixed_const(1);
972         if (wm->vsc.full > a.full)
973                 latency_tolerant_lines = 1;
974         else {
975                 if (lb_partitions <= (wm->vtaps + 1))
976                         latency_tolerant_lines = 1;
977                 else
978                         latency_tolerant_lines = 2;
979         }
980
981         latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
982
983         if (evergreen_latency_watermark(wm) <= latency_hiding)
984                 return true;
985         else
986                 return false;
987 }
988
989 static void evergreen_program_watermarks(struct radeon_device *rdev,
990                                          struct radeon_crtc *radeon_crtc,
991                                          u32 lb_size, u32 num_heads)
992 {
993         struct drm_display_mode *mode = &radeon_crtc->base.mode;
994         struct evergreen_wm_params wm;
995         u32 pixel_period;
996         u32 line_time = 0;
997         u32 latency_watermark_a = 0, latency_watermark_b = 0;
998         u32 priority_a_mark = 0, priority_b_mark = 0;
999         u32 priority_a_cnt = PRIORITY_OFF;
1000         u32 priority_b_cnt = PRIORITY_OFF;
1001         u32 pipe_offset = radeon_crtc->crtc_id * 16;
1002         u32 tmp, arb_control3;
1003         fixed20_12 a, b, c;
1004
1005         if (radeon_crtc->base.enabled && num_heads && mode) {
1006                 pixel_period = 1000000 / (u32)mode->clock;
1007                 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
1008                 priority_a_cnt = 0;
1009                 priority_b_cnt = 0;
1010
1011                 wm.yclk = rdev->pm.current_mclk * 10;
1012                 wm.sclk = rdev->pm.current_sclk * 10;
1013                 wm.disp_clk = mode->clock;
1014                 wm.src_width = mode->crtc_hdisplay;
1015                 wm.active_time = mode->crtc_hdisplay * pixel_period;
1016                 wm.blank_time = line_time - wm.active_time;
1017                 wm.interlaced = false;
1018                 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
1019                         wm.interlaced = true;
1020                 wm.vsc = radeon_crtc->vsc;
1021                 wm.vtaps = 1;
1022                 if (radeon_crtc->rmx_type != RMX_OFF)
1023                         wm.vtaps = 2;
1024                 wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
1025                 wm.lb_size = lb_size;
1026                 wm.dram_channels = evergreen_get_number_of_dram_channels(rdev);
1027                 wm.num_heads = num_heads;
1028
1029                 /* set for high clocks */
1030                 latency_watermark_a = min(evergreen_latency_watermark(&wm), (u32)65535);
1031                 /* set for low clocks */
1032                 /* wm.yclk = low clk; wm.sclk = low clk */
1033                 latency_watermark_b = min(evergreen_latency_watermark(&wm), (u32)65535);
1034
1035                 /* possibly force display priority to high */
1036                 /* should really do this at mode validation time... */
1037                 if (!evergreen_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
1038                     !evergreen_average_bandwidth_vs_available_bandwidth(&wm) ||
1039                     !evergreen_check_latency_hiding(&wm) ||
1040                     (rdev->disp_priority == 2)) {
1041                         DRM_DEBUG_KMS("force priority to high\n");
1042                         priority_a_cnt |= PRIORITY_ALWAYS_ON;
1043                         priority_b_cnt |= PRIORITY_ALWAYS_ON;
1044                 }
1045
1046                 a.full = dfixed_const(1000);
1047                 b.full = dfixed_const(mode->clock);
1048                 b.full = dfixed_div(b, a);
1049                 c.full = dfixed_const(latency_watermark_a);
1050                 c.full = dfixed_mul(c, b);
1051                 c.full = dfixed_mul(c, radeon_crtc->hsc);
1052                 c.full = dfixed_div(c, a);
1053                 a.full = dfixed_const(16);
1054                 c.full = dfixed_div(c, a);
1055                 priority_a_mark = dfixed_trunc(c);
1056                 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
1057
1058                 a.full = dfixed_const(1000);
1059                 b.full = dfixed_const(mode->clock);
1060                 b.full = dfixed_div(b, a);
1061                 c.full = dfixed_const(latency_watermark_b);
1062                 c.full = dfixed_mul(c, b);
1063                 c.full = dfixed_mul(c, radeon_crtc->hsc);
1064                 c.full = dfixed_div(c, a);
1065                 a.full = dfixed_const(16);
1066                 c.full = dfixed_div(c, a);
1067                 priority_b_mark = dfixed_trunc(c);
1068                 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
1069         }
1070
1071         /* select wm A */
1072         arb_control3 = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
1073         tmp = arb_control3;
1074         tmp &= ~LATENCY_WATERMARK_MASK(3);
1075         tmp |= LATENCY_WATERMARK_MASK(1);
1076         WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
1077         WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
1078                (LATENCY_LOW_WATERMARK(latency_watermark_a) |
1079                 LATENCY_HIGH_WATERMARK(line_time)));
1080         /* select wm B */
1081         tmp = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
1082         tmp &= ~LATENCY_WATERMARK_MASK(3);
1083         tmp |= LATENCY_WATERMARK_MASK(2);
1084         WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
1085         WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
1086                (LATENCY_LOW_WATERMARK(latency_watermark_b) |
1087                 LATENCY_HIGH_WATERMARK(line_time)));
1088         /* restore original selection */
1089         WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, arb_control3);
1090
1091         /* write the priority marks */
1092         WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
1093         WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
1094
1095 }
1096
1097 /**
1098  * evergreen_bandwidth_update - update display watermarks callback.
1099  *
1100  * @rdev: radeon_device pointer
1101  *
1102  * Update the display watermarks based on the requested mode(s)
1103  * (evergreen+).
1104  */
1105 void evergreen_bandwidth_update(struct radeon_device *rdev)
1106 {
1107         struct drm_display_mode *mode0 = NULL;
1108         struct drm_display_mode *mode1 = NULL;
1109         u32 num_heads = 0, lb_size;
1110         int i;
1111
1112         radeon_update_display_priority(rdev);
1113
1114         for (i = 0; i < rdev->num_crtc; i++) {
1115                 if (rdev->mode_info.crtcs[i]->base.enabled)
1116                         num_heads++;
1117         }
1118         for (i = 0; i < rdev->num_crtc; i += 2) {
1119                 mode0 = &rdev->mode_info.crtcs[i]->base.mode;
1120                 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
1121                 lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
1122                 evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
1123                 lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
1124                 evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
1125         }
1126 }
1127
1128 /**
1129  * evergreen_mc_wait_for_idle - wait for MC idle callback.
1130  *
1131  * @rdev: radeon_device pointer
1132  *
1133  * Wait for the MC (memory controller) to be idle.
1134  * (evergreen+).
1135  * Returns 0 if the MC is idle, -1 if not.
1136  */
1137 int evergreen_mc_wait_for_idle(struct radeon_device *rdev)
1138 {
1139         unsigned i;
1140         u32 tmp;
1141
1142         for (i = 0; i < rdev->usec_timeout; i++) {
1143                 /* read MC_STATUS */
1144                 tmp = RREG32(SRBM_STATUS) & 0x1F00;
1145                 if (!tmp)
1146                         return 0;
1147                 udelay(1);
1148         }
1149         return -1;
1150 }
1151
1152 /*
1153  * GART
1154  */
1155 void evergreen_pcie_gart_tlb_flush(struct radeon_device *rdev)
1156 {
1157         unsigned i;
1158         u32 tmp;
1159
1160         WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1161
1162         WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
1163         for (i = 0; i < rdev->usec_timeout; i++) {
1164                 /* read MC_STATUS */
1165                 tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
1166                 tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
1167                 if (tmp == 2) {
1168                         printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
1169                         return;
1170                 }
1171                 if (tmp) {
1172                         return;
1173                 }
1174                 udelay(1);
1175         }
1176 }
1177
1178 static int evergreen_pcie_gart_enable(struct radeon_device *rdev)
1179 {
1180         u32 tmp;
1181         int r;
1182
1183         if (rdev->gart.robj == NULL) {
1184                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
1185                 return -EINVAL;
1186         }
1187         r = radeon_gart_table_vram_pin(rdev);
1188         if (r)
1189                 return r;
1190         radeon_gart_restore(rdev);
1191         /* Setup L2 cache */
1192         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
1193                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1194                                 EFFECTIVE_L2_QUEUE_SIZE(7));
1195         WREG32(VM_L2_CNTL2, 0);
1196         WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
1197         /* Setup TLB control */
1198         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
1199                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1200                 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
1201                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
1202         if (rdev->flags & RADEON_IS_IGP) {
1203                 WREG32(FUS_MC_VM_MD_L1_TLB0_CNTL, tmp);
1204                 WREG32(FUS_MC_VM_MD_L1_TLB1_CNTL, tmp);
1205                 WREG32(FUS_MC_VM_MD_L1_TLB2_CNTL, tmp);
1206         } else {
1207                 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
1208                 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
1209                 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
1210                 if ((rdev->family == CHIP_JUNIPER) ||
1211                     (rdev->family == CHIP_CYPRESS) ||
1212                     (rdev->family == CHIP_HEMLOCK) ||
1213                     (rdev->family == CHIP_BARTS))
1214                         WREG32(MC_VM_MD_L1_TLB3_CNTL, tmp);
1215         }
1216         WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
1217         WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
1218         WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
1219         WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
1220         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
1221         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
1222         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
1223         WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
1224                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
1225         WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
1226                         (u32)(rdev->dummy_page.addr >> 12));
1227         WREG32(VM_CONTEXT1_CNTL, 0);
1228
1229         evergreen_pcie_gart_tlb_flush(rdev);
1230         DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
1231                  (unsigned)(rdev->mc.gtt_size >> 20),
1232                  (unsigned long long)rdev->gart.table_addr);
1233         rdev->gart.ready = true;
1234         return 0;
1235 }
1236
1237 static void evergreen_pcie_gart_disable(struct radeon_device *rdev)
1238 {
1239         u32 tmp;
1240
1241         /* Disable all tables */
1242         WREG32(VM_CONTEXT0_CNTL, 0);
1243         WREG32(VM_CONTEXT1_CNTL, 0);
1244
1245         /* Setup L2 cache */
1246         WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
1247                                 EFFECTIVE_L2_QUEUE_SIZE(7));
1248         WREG32(VM_L2_CNTL2, 0);
1249         WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
1250         /* Setup TLB control */
1251         tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
1252         WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
1253         WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
1254         WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
1255         WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
1256         WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
1257         WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
1258         WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
1259         radeon_gart_table_vram_unpin(rdev);
1260 }
1261
1262 static void evergreen_pcie_gart_fini(struct radeon_device *rdev)
1263 {
1264         evergreen_pcie_gart_disable(rdev);
1265         radeon_gart_table_vram_free(rdev);
1266         radeon_gart_fini(rdev);
1267 }
1268
1269
1270 static void evergreen_agp_enable(struct radeon_device *rdev)
1271 {
1272         u32 tmp;
1273
1274         /* Setup L2 cache */
1275         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
1276                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
1277                                 EFFECTIVE_L2_QUEUE_SIZE(7));
1278         WREG32(VM_L2_CNTL2, 0);
1279         WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
1280         /* Setup TLB control */
1281         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
1282                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
1283                 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
1284                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
1285         WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
1286         WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
1287         WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
1288         WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
1289         WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
1290         WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
1291         WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
1292         WREG32(VM_CONTEXT0_CNTL, 0);
1293         WREG32(VM_CONTEXT1_CNTL, 0);
1294 }
1295
1296 void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save)
1297 {
1298         u32 crtc_enabled, tmp, frame_count, blackout;
1299         int i, j;
1300
1301         save->vga_render_control = RREG32(VGA_RENDER_CONTROL);
1302         save->vga_hdp_control = RREG32(VGA_HDP_CONTROL);
1303
1304         /* disable VGA render */
1305         WREG32(VGA_RENDER_CONTROL, 0);
1306         /* blank the display controllers */
1307         for (i = 0; i < rdev->num_crtc; i++) {
1308                 crtc_enabled = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]) & EVERGREEN_CRTC_MASTER_EN;
1309                 if (crtc_enabled) {
1310                         save->crtc_enabled[i] = true;
1311                         if (ASIC_IS_DCE6(rdev)) {
1312                                 tmp = RREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i]);
1313                                 if (!(tmp & EVERGREEN_CRTC_BLANK_DATA_EN)) {
1314                                         radeon_wait_for_vblank(rdev, i);
1315                                         tmp |= EVERGREEN_CRTC_BLANK_DATA_EN;
1316                                         WREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i], tmp);
1317                                 }
1318                         } else {
1319                                 tmp = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]);
1320                                 if (!(tmp & EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE)) {
1321                                         radeon_wait_for_vblank(rdev, i);
1322                                         tmp |= EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
1323                                         WREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i], tmp);
1324                                 }
1325                         }
1326                         /* wait for the next frame */
1327                         frame_count = radeon_get_vblank_counter(rdev, i);
1328                         for (j = 0; j < rdev->usec_timeout; j++) {
1329                                 if (radeon_get_vblank_counter(rdev, i) != frame_count)
1330                                         break;
1331                                 udelay(1);
1332                         }
1333                 } else {
1334                         save->crtc_enabled[i] = false;
1335                 }
1336         }
1337
1338         radeon_mc_wait_for_idle(rdev);
1339
1340         blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
1341         if ((blackout & BLACKOUT_MODE_MASK) != 1) {
1342                 /* Block CPU access */
1343                 WREG32(BIF_FB_EN, 0);
1344                 /* blackout the MC */
1345                 blackout &= ~BLACKOUT_MODE_MASK;
1346                 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
1347         }
1348 }
1349
1350 void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save)
1351 {
1352         u32 tmp, frame_count;
1353         int i, j;
1354
1355         /* update crtc base addresses */
1356         for (i = 0; i < rdev->num_crtc; i++) {
1357                 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + crtc_offsets[i],
1358                        upper_32_bits(rdev->mc.vram_start));
1359                 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + crtc_offsets[i],
1360                        upper_32_bits(rdev->mc.vram_start));
1361                 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + crtc_offsets[i],
1362                        (u32)rdev->mc.vram_start);
1363                 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + crtc_offsets[i],
1364                        (u32)rdev->mc.vram_start);
1365         }
1366         WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS_HIGH, upper_32_bits(rdev->mc.vram_start));
1367         WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS, (u32)rdev->mc.vram_start);
1368
1369         /* unblackout the MC */
1370         tmp = RREG32(MC_SHARED_BLACKOUT_CNTL);
1371         tmp &= ~BLACKOUT_MODE_MASK;
1372         WREG32(MC_SHARED_BLACKOUT_CNTL, tmp);
1373         /* allow CPU access */
1374         WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
1375
1376         for (i = 0; i < rdev->num_crtc; i++) {
1377                 if (save->crtc_enabled[i]) {
1378                         if (ASIC_IS_DCE6(rdev)) {
1379                                 tmp = RREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i]);
1380                                 tmp |= EVERGREEN_CRTC_BLANK_DATA_EN;
1381                                 WREG32(EVERGREEN_CRTC_BLANK_CONTROL + crtc_offsets[i], tmp);
1382                         } else {
1383                                 tmp = RREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i]);
1384                                 tmp &= ~EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
1385                                 WREG32(EVERGREEN_CRTC_CONTROL + crtc_offsets[i], tmp);
1386                         }
1387                         /* wait for the next frame */
1388                         frame_count = radeon_get_vblank_counter(rdev, i);
1389                         for (j = 0; j < rdev->usec_timeout; j++) {
1390                                 if (radeon_get_vblank_counter(rdev, i) != frame_count)
1391                                         break;
1392                                 udelay(1);
1393                         }
1394                 }
1395         }
1396         /* Unlock vga access */
1397         WREG32(VGA_HDP_CONTROL, save->vga_hdp_control);
1398         mdelay(1);
1399         WREG32(VGA_RENDER_CONTROL, save->vga_render_control);
1400 }
1401
1402 void evergreen_mc_program(struct radeon_device *rdev)
1403 {
1404         struct evergreen_mc_save save;
1405         u32 tmp;
1406         int i, j;
1407
1408         /* Initialize HDP */
1409         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1410                 WREG32((0x2c14 + j), 0x00000000);
1411                 WREG32((0x2c18 + j), 0x00000000);
1412                 WREG32((0x2c1c + j), 0x00000000);
1413                 WREG32((0x2c20 + j), 0x00000000);
1414                 WREG32((0x2c24 + j), 0x00000000);
1415         }
1416         WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
1417
1418         evergreen_mc_stop(rdev, &save);
1419         if (evergreen_mc_wait_for_idle(rdev)) {
1420                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1421         }
1422         /* Lockout access through VGA aperture*/
1423         WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
1424         /* Update configuration */
1425         if (rdev->flags & RADEON_IS_AGP) {
1426                 if (rdev->mc.vram_start < rdev->mc.gtt_start) {
1427                         /* VRAM before AGP */
1428                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1429                                 rdev->mc.vram_start >> 12);
1430                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1431                                 rdev->mc.gtt_end >> 12);
1432                 } else {
1433                         /* VRAM after AGP */
1434                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1435                                 rdev->mc.gtt_start >> 12);
1436                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1437                                 rdev->mc.vram_end >> 12);
1438                 }
1439         } else {
1440                 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1441                         rdev->mc.vram_start >> 12);
1442                 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1443                         rdev->mc.vram_end >> 12);
1444         }
1445         WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, rdev->vram_scratch.gpu_addr >> 12);
1446         /* llano/ontario only */
1447         if ((rdev->family == CHIP_PALM) ||
1448             (rdev->family == CHIP_SUMO) ||
1449             (rdev->family == CHIP_SUMO2)) {
1450                 tmp = RREG32(MC_FUS_VM_FB_OFFSET) & 0x000FFFFF;
1451                 tmp |= ((rdev->mc.vram_end >> 20) & 0xF) << 24;
1452                 tmp |= ((rdev->mc.vram_start >> 20) & 0xF) << 20;
1453                 WREG32(MC_FUS_VM_FB_OFFSET, tmp);
1454         }
1455         tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
1456         tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
1457         WREG32(MC_VM_FB_LOCATION, tmp);
1458         WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
1459         WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
1460         WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
1461         if (rdev->flags & RADEON_IS_AGP) {
1462                 WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 16);
1463                 WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 16);
1464                 WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
1465         } else {
1466                 WREG32(MC_VM_AGP_BASE, 0);
1467                 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
1468                 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
1469         }
1470         if (evergreen_mc_wait_for_idle(rdev)) {
1471                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1472         }
1473         evergreen_mc_resume(rdev, &save);
1474         /* we need to own VRAM, so turn off the VGA renderer here
1475          * to stop it overwriting our objects */
1476         rv515_vga_render_disable(rdev);
1477 }
1478
1479 /*
1480  * CP.
1481  */
1482 void evergreen_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1483 {
1484         struct radeon_ring *ring = &rdev->ring[ib->ring];
1485         u32 next_rptr;
1486
1487         /* set to DX10/11 mode */
1488         radeon_ring_write(ring, PACKET3(PACKET3_MODE_CONTROL, 0));
1489         radeon_ring_write(ring, 1);
1490
1491         if (ring->rptr_save_reg) {
1492                 next_rptr = ring->wptr + 3 + 4;
1493                 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1494                 radeon_ring_write(ring, ((ring->rptr_save_reg - 
1495                                           PACKET3_SET_CONFIG_REG_START) >> 2));
1496                 radeon_ring_write(ring, next_rptr);
1497         } else if (rdev->wb.enabled) {
1498                 next_rptr = ring->wptr + 5 + 4;
1499                 radeon_ring_write(ring, PACKET3(PACKET3_MEM_WRITE, 3));
1500                 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
1501                 radeon_ring_write(ring, (upper_32_bits(ring->next_rptr_gpu_addr) & 0xff) | (1 << 18));
1502                 radeon_ring_write(ring, next_rptr);
1503                 radeon_ring_write(ring, 0);
1504         }
1505
1506         radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
1507         radeon_ring_write(ring,
1508 #ifdef __BIG_ENDIAN
1509                           (2 << 0) |
1510 #endif
1511                           (ib->gpu_addr & 0xFFFFFFFC));
1512         radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF);
1513         radeon_ring_write(ring, ib->length_dw);
1514 }
1515
1516
1517 static int evergreen_cp_load_microcode(struct radeon_device *rdev)
1518 {
1519         const __be32 *fw_data;
1520         int i;
1521
1522         if (!rdev->me_fw || !rdev->pfp_fw)
1523                 return -EINVAL;
1524
1525         r700_cp_stop(rdev);
1526         WREG32(CP_RB_CNTL,
1527 #ifdef __BIG_ENDIAN
1528                BUF_SWAP_32BIT |
1529 #endif
1530                RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
1531
1532         fw_data = (const __be32 *)rdev->pfp_fw->data;
1533         WREG32(CP_PFP_UCODE_ADDR, 0);
1534         for (i = 0; i < EVERGREEN_PFP_UCODE_SIZE; i++)
1535                 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1536         WREG32(CP_PFP_UCODE_ADDR, 0);
1537
1538         fw_data = (const __be32 *)rdev->me_fw->data;
1539         WREG32(CP_ME_RAM_WADDR, 0);
1540         for (i = 0; i < EVERGREEN_PM4_UCODE_SIZE; i++)
1541                 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1542
1543         WREG32(CP_PFP_UCODE_ADDR, 0);
1544         WREG32(CP_ME_RAM_WADDR, 0);
1545         WREG32(CP_ME_RAM_RADDR, 0);
1546         return 0;
1547 }
1548
1549 static int evergreen_cp_start(struct radeon_device *rdev)
1550 {
1551         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1552         int r, i;
1553         uint32_t cp_me;
1554
1555         r = radeon_ring_lock(rdev, ring, 7);
1556         if (r) {
1557                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1558                 return r;
1559         }
1560         radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
1561         radeon_ring_write(ring, 0x1);
1562         radeon_ring_write(ring, 0x0);
1563         radeon_ring_write(ring, rdev->config.evergreen.max_hw_contexts - 1);
1564         radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1565         radeon_ring_write(ring, 0);
1566         radeon_ring_write(ring, 0);
1567         radeon_ring_unlock_commit(rdev, ring);
1568
1569         cp_me = 0xff;
1570         WREG32(CP_ME_CNTL, cp_me);
1571
1572         r = radeon_ring_lock(rdev, ring, evergreen_default_size + 19);
1573         if (r) {
1574                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1575                 return r;
1576         }
1577
1578         /* setup clear context state */
1579         radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1580         radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1581
1582         for (i = 0; i < evergreen_default_size; i++)
1583                 radeon_ring_write(ring, evergreen_default_state[i]);
1584
1585         radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1586         radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
1587
1588         /* set clear context state */
1589         radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
1590         radeon_ring_write(ring, 0);
1591
1592         /* SQ_VTX_BASE_VTX_LOC */
1593         radeon_ring_write(ring, 0xc0026f00);
1594         radeon_ring_write(ring, 0x00000000);
1595         radeon_ring_write(ring, 0x00000000);
1596         radeon_ring_write(ring, 0x00000000);
1597
1598         /* Clear consts */
1599         radeon_ring_write(ring, 0xc0036f00);
1600         radeon_ring_write(ring, 0x00000bc4);
1601         radeon_ring_write(ring, 0xffffffff);
1602         radeon_ring_write(ring, 0xffffffff);
1603         radeon_ring_write(ring, 0xffffffff);
1604
1605         radeon_ring_write(ring, 0xc0026900);
1606         radeon_ring_write(ring, 0x00000316);
1607         radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
1608         radeon_ring_write(ring, 0x00000010); /*  */
1609
1610         radeon_ring_unlock_commit(rdev, ring);
1611
1612         return 0;
1613 }
1614
1615 static int evergreen_cp_resume(struct radeon_device *rdev)
1616 {
1617         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1618         u32 tmp;
1619         u32 rb_bufsz;
1620         int r;
1621
1622         /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
1623         WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
1624                                  SOFT_RESET_PA |
1625                                  SOFT_RESET_SH |
1626                                  SOFT_RESET_VGT |
1627                                  SOFT_RESET_SPI |
1628                                  SOFT_RESET_SX));
1629         RREG32(GRBM_SOFT_RESET);
1630         mdelay(15);
1631         WREG32(GRBM_SOFT_RESET, 0);
1632         RREG32(GRBM_SOFT_RESET);
1633
1634         /* Set ring buffer size */
1635         rb_bufsz = drm_order(ring->ring_size / 8);
1636         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1637 #ifdef __BIG_ENDIAN
1638         tmp |= BUF_SWAP_32BIT;
1639 #endif
1640         WREG32(CP_RB_CNTL, tmp);
1641         WREG32(CP_SEM_WAIT_TIMER, 0x0);
1642         WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
1643
1644         /* Set the write pointer delay */
1645         WREG32(CP_RB_WPTR_DELAY, 0);
1646
1647         /* Initialize the ring buffer's read and write pointers */
1648         WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
1649         WREG32(CP_RB_RPTR_WR, 0);
1650         ring->wptr = 0;
1651         WREG32(CP_RB_WPTR, ring->wptr);
1652
1653         /* set the wb address whether it's enabled or not */
1654         WREG32(CP_RB_RPTR_ADDR,
1655                ((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC));
1656         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
1657         WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
1658
1659         if (rdev->wb.enabled)
1660                 WREG32(SCRATCH_UMSK, 0xff);
1661         else {
1662                 tmp |= RB_NO_UPDATE;
1663                 WREG32(SCRATCH_UMSK, 0);
1664         }
1665
1666         mdelay(1);
1667         WREG32(CP_RB_CNTL, tmp);
1668
1669         WREG32(CP_RB_BASE, ring->gpu_addr >> 8);
1670         WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
1671
1672         ring->rptr = RREG32(CP_RB_RPTR);
1673
1674         evergreen_cp_start(rdev);
1675         ring->ready = true;
1676         r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, ring);
1677         if (r) {
1678                 ring->ready = false;
1679                 return r;
1680         }
1681         return 0;
1682 }
1683
1684 /*
1685  * Core functions
1686  */
1687 static void evergreen_gpu_init(struct radeon_device *rdev)
1688 {
1689         u32 gb_addr_config;
1690         u32 mc_shared_chmap, mc_arb_ramcfg;
1691         u32 sx_debug_1;
1692         u32 smx_dc_ctl0;
1693         u32 sq_config;
1694         u32 sq_lds_resource_mgmt;
1695         u32 sq_gpr_resource_mgmt_1;
1696         u32 sq_gpr_resource_mgmt_2;
1697         u32 sq_gpr_resource_mgmt_3;
1698         u32 sq_thread_resource_mgmt;
1699         u32 sq_thread_resource_mgmt_2;
1700         u32 sq_stack_resource_mgmt_1;
1701         u32 sq_stack_resource_mgmt_2;
1702         u32 sq_stack_resource_mgmt_3;
1703         u32 vgt_cache_invalidation;
1704         u32 hdp_host_path_cntl, tmp;
1705         u32 disabled_rb_mask;
1706         int i, j, num_shader_engines, ps_thread_count;
1707
1708         switch (rdev->family) {
1709         case CHIP_CYPRESS:
1710         case CHIP_HEMLOCK:
1711                 rdev->config.evergreen.num_ses = 2;
1712                 rdev->config.evergreen.max_pipes = 4;
1713                 rdev->config.evergreen.max_tile_pipes = 8;
1714                 rdev->config.evergreen.max_simds = 10;
1715                 rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
1716                 rdev->config.evergreen.max_gprs = 256;
1717                 rdev->config.evergreen.max_threads = 248;
1718                 rdev->config.evergreen.max_gs_threads = 32;
1719                 rdev->config.evergreen.max_stack_entries = 512;
1720                 rdev->config.evergreen.sx_num_of_sets = 4;
1721                 rdev->config.evergreen.sx_max_export_size = 256;
1722                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1723                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1724                 rdev->config.evergreen.max_hw_contexts = 8;
1725                 rdev->config.evergreen.sq_num_cf_insts = 2;
1726
1727                 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1728                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1729                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1730                 gb_addr_config = CYPRESS_GB_ADDR_CONFIG_GOLDEN;
1731                 break;
1732         case CHIP_JUNIPER:
1733                 rdev->config.evergreen.num_ses = 1;
1734                 rdev->config.evergreen.max_pipes = 4;
1735                 rdev->config.evergreen.max_tile_pipes = 4;
1736                 rdev->config.evergreen.max_simds = 10;
1737                 rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
1738                 rdev->config.evergreen.max_gprs = 256;
1739                 rdev->config.evergreen.max_threads = 248;
1740                 rdev->config.evergreen.max_gs_threads = 32;
1741                 rdev->config.evergreen.max_stack_entries = 512;
1742                 rdev->config.evergreen.sx_num_of_sets = 4;
1743                 rdev->config.evergreen.sx_max_export_size = 256;
1744                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1745                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1746                 rdev->config.evergreen.max_hw_contexts = 8;
1747                 rdev->config.evergreen.sq_num_cf_insts = 2;
1748
1749                 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1750                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1751                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1752                 gb_addr_config = JUNIPER_GB_ADDR_CONFIG_GOLDEN;
1753                 break;
1754         case CHIP_REDWOOD:
1755                 rdev->config.evergreen.num_ses = 1;
1756                 rdev->config.evergreen.max_pipes = 4;
1757                 rdev->config.evergreen.max_tile_pipes = 4;
1758                 rdev->config.evergreen.max_simds = 5;
1759                 rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
1760                 rdev->config.evergreen.max_gprs = 256;
1761                 rdev->config.evergreen.max_threads = 248;
1762                 rdev->config.evergreen.max_gs_threads = 32;
1763                 rdev->config.evergreen.max_stack_entries = 256;
1764                 rdev->config.evergreen.sx_num_of_sets = 4;
1765                 rdev->config.evergreen.sx_max_export_size = 256;
1766                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1767                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1768                 rdev->config.evergreen.max_hw_contexts = 8;
1769                 rdev->config.evergreen.sq_num_cf_insts = 2;
1770
1771                 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1772                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1773                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1774                 gb_addr_config = REDWOOD_GB_ADDR_CONFIG_GOLDEN;
1775                 break;
1776         case CHIP_CEDAR:
1777         default:
1778                 rdev->config.evergreen.num_ses = 1;
1779                 rdev->config.evergreen.max_pipes = 2;
1780                 rdev->config.evergreen.max_tile_pipes = 2;
1781                 rdev->config.evergreen.max_simds = 2;
1782                 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1783                 rdev->config.evergreen.max_gprs = 256;
1784                 rdev->config.evergreen.max_threads = 192;
1785                 rdev->config.evergreen.max_gs_threads = 16;
1786                 rdev->config.evergreen.max_stack_entries = 256;
1787                 rdev->config.evergreen.sx_num_of_sets = 4;
1788                 rdev->config.evergreen.sx_max_export_size = 128;
1789                 rdev->config.evergreen.sx_max_export_pos_size = 32;
1790                 rdev->config.evergreen.sx_max_export_smx_size = 96;
1791                 rdev->config.evergreen.max_hw_contexts = 4;
1792                 rdev->config.evergreen.sq_num_cf_insts = 1;
1793
1794                 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1795                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1796                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1797                 gb_addr_config = CEDAR_GB_ADDR_CONFIG_GOLDEN;
1798                 break;
1799         case CHIP_PALM:
1800                 rdev->config.evergreen.num_ses = 1;
1801                 rdev->config.evergreen.max_pipes = 2;
1802                 rdev->config.evergreen.max_tile_pipes = 2;
1803                 rdev->config.evergreen.max_simds = 2;
1804                 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1805                 rdev->config.evergreen.max_gprs = 256;
1806                 rdev->config.evergreen.max_threads = 192;
1807                 rdev->config.evergreen.max_gs_threads = 16;
1808                 rdev->config.evergreen.max_stack_entries = 256;
1809                 rdev->config.evergreen.sx_num_of_sets = 4;
1810                 rdev->config.evergreen.sx_max_export_size = 128;
1811                 rdev->config.evergreen.sx_max_export_pos_size = 32;
1812                 rdev->config.evergreen.sx_max_export_smx_size = 96;
1813                 rdev->config.evergreen.max_hw_contexts = 4;
1814                 rdev->config.evergreen.sq_num_cf_insts = 1;
1815
1816                 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1817                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1818                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1819                 gb_addr_config = CEDAR_GB_ADDR_CONFIG_GOLDEN;
1820                 break;
1821         case CHIP_SUMO:
1822                 rdev->config.evergreen.num_ses = 1;
1823                 rdev->config.evergreen.max_pipes = 4;
1824                 rdev->config.evergreen.max_tile_pipes = 4;
1825                 if (rdev->pdev->device == 0x9648)
1826                         rdev->config.evergreen.max_simds = 3;
1827                 else if ((rdev->pdev->device == 0x9647) ||
1828                          (rdev->pdev->device == 0x964a))
1829                         rdev->config.evergreen.max_simds = 4;
1830                 else
1831                         rdev->config.evergreen.max_simds = 5;
1832                 rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
1833                 rdev->config.evergreen.max_gprs = 256;
1834                 rdev->config.evergreen.max_threads = 248;
1835                 rdev->config.evergreen.max_gs_threads = 32;
1836                 rdev->config.evergreen.max_stack_entries = 256;
1837                 rdev->config.evergreen.sx_num_of_sets = 4;
1838                 rdev->config.evergreen.sx_max_export_size = 256;
1839                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1840                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1841                 rdev->config.evergreen.max_hw_contexts = 8;
1842                 rdev->config.evergreen.sq_num_cf_insts = 2;
1843
1844                 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1845                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1846                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1847                 gb_addr_config = SUMO_GB_ADDR_CONFIG_GOLDEN;
1848                 break;
1849         case CHIP_SUMO2:
1850                 rdev->config.evergreen.num_ses = 1;
1851                 rdev->config.evergreen.max_pipes = 4;
1852                 rdev->config.evergreen.max_tile_pipes = 4;
1853                 rdev->config.evergreen.max_simds = 2;
1854                 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1855                 rdev->config.evergreen.max_gprs = 256;
1856                 rdev->config.evergreen.max_threads = 248;
1857                 rdev->config.evergreen.max_gs_threads = 32;
1858                 rdev->config.evergreen.max_stack_entries = 512;
1859                 rdev->config.evergreen.sx_num_of_sets = 4;
1860                 rdev->config.evergreen.sx_max_export_size = 256;
1861                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1862                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1863                 rdev->config.evergreen.max_hw_contexts = 8;
1864                 rdev->config.evergreen.sq_num_cf_insts = 2;
1865
1866                 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1867                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1868                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1869                 gb_addr_config = SUMO2_GB_ADDR_CONFIG_GOLDEN;
1870                 break;
1871         case CHIP_BARTS:
1872                 rdev->config.evergreen.num_ses = 2;
1873                 rdev->config.evergreen.max_pipes = 4;
1874                 rdev->config.evergreen.max_tile_pipes = 8;
1875                 rdev->config.evergreen.max_simds = 7;
1876                 rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
1877                 rdev->config.evergreen.max_gprs = 256;
1878                 rdev->config.evergreen.max_threads = 248;
1879                 rdev->config.evergreen.max_gs_threads = 32;
1880                 rdev->config.evergreen.max_stack_entries = 512;
1881                 rdev->config.evergreen.sx_num_of_sets = 4;
1882                 rdev->config.evergreen.sx_max_export_size = 256;
1883                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1884                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1885                 rdev->config.evergreen.max_hw_contexts = 8;
1886                 rdev->config.evergreen.sq_num_cf_insts = 2;
1887
1888                 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1889                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1890                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1891                 gb_addr_config = BARTS_GB_ADDR_CONFIG_GOLDEN;
1892                 break;
1893         case CHIP_TURKS:
1894                 rdev->config.evergreen.num_ses = 1;
1895                 rdev->config.evergreen.max_pipes = 4;
1896                 rdev->config.evergreen.max_tile_pipes = 4;
1897                 rdev->config.evergreen.max_simds = 6;
1898                 rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
1899                 rdev->config.evergreen.max_gprs = 256;
1900                 rdev->config.evergreen.max_threads = 248;
1901                 rdev->config.evergreen.max_gs_threads = 32;
1902                 rdev->config.evergreen.max_stack_entries = 256;
1903                 rdev->config.evergreen.sx_num_of_sets = 4;
1904                 rdev->config.evergreen.sx_max_export_size = 256;
1905                 rdev->config.evergreen.sx_max_export_pos_size = 64;
1906                 rdev->config.evergreen.sx_max_export_smx_size = 192;
1907                 rdev->config.evergreen.max_hw_contexts = 8;
1908                 rdev->config.evergreen.sq_num_cf_insts = 2;
1909
1910                 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1911                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1912                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1913                 gb_addr_config = TURKS_GB_ADDR_CONFIG_GOLDEN;
1914                 break;
1915         case CHIP_CAICOS:
1916                 rdev->config.evergreen.num_ses = 1;
1917                 rdev->config.evergreen.max_pipes = 2;
1918                 rdev->config.evergreen.max_tile_pipes = 2;
1919                 rdev->config.evergreen.max_simds = 2;
1920                 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1921                 rdev->config.evergreen.max_gprs = 256;
1922                 rdev->config.evergreen.max_threads = 192;
1923                 rdev->config.evergreen.max_gs_threads = 16;
1924                 rdev->config.evergreen.max_stack_entries = 256;
1925                 rdev->config.evergreen.sx_num_of_sets = 4;
1926                 rdev->config.evergreen.sx_max_export_size = 128;
1927                 rdev->config.evergreen.sx_max_export_pos_size = 32;
1928                 rdev->config.evergreen.sx_max_export_smx_size = 96;
1929                 rdev->config.evergreen.max_hw_contexts = 4;
1930                 rdev->config.evergreen.sq_num_cf_insts = 1;
1931
1932                 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1933                 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1934                 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1935                 gb_addr_config = CAICOS_GB_ADDR_CONFIG_GOLDEN;
1936                 break;
1937         }
1938
1939         /* Initialize HDP */
1940         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1941                 WREG32((0x2c14 + j), 0x00000000);
1942                 WREG32((0x2c18 + j), 0x00000000);
1943                 WREG32((0x2c1c + j), 0x00000000);
1944                 WREG32((0x2c20 + j), 0x00000000);
1945                 WREG32((0x2c24 + j), 0x00000000);
1946         }
1947
1948         WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1949
1950         evergreen_fix_pci_max_read_req_size(rdev);
1951
1952         mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1953         if ((rdev->family == CHIP_PALM) ||
1954             (rdev->family == CHIP_SUMO) ||
1955             (rdev->family == CHIP_SUMO2))
1956                 mc_arb_ramcfg = RREG32(FUS_MC_ARB_RAMCFG);
1957         else
1958                 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1959
1960         /* setup tiling info dword.  gb_addr_config is not adequate since it does
1961          * not have bank info, so create a custom tiling dword.
1962          * bits 3:0   num_pipes
1963          * bits 7:4   num_banks
1964          * bits 11:8  group_size
1965          * bits 15:12 row_size
1966          */
1967         rdev->config.evergreen.tile_config = 0;
1968         switch (rdev->config.evergreen.max_tile_pipes) {
1969         case 1:
1970         default:
1971                 rdev->config.evergreen.tile_config |= (0 << 0);
1972                 break;
1973         case 2:
1974                 rdev->config.evergreen.tile_config |= (1 << 0);
1975                 break;
1976         case 4:
1977                 rdev->config.evergreen.tile_config |= (2 << 0);
1978                 break;
1979         case 8:
1980                 rdev->config.evergreen.tile_config |= (3 << 0);
1981                 break;
1982         }
1983         /* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */
1984         if (rdev->flags & RADEON_IS_IGP)
1985                 rdev->config.evergreen.tile_config |= 1 << 4;
1986         else {
1987                 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
1988                 case 0: /* four banks */
1989                         rdev->config.evergreen.tile_config |= 0 << 4;
1990                         break;
1991                 case 1: /* eight banks */
1992                         rdev->config.evergreen.tile_config |= 1 << 4;
1993                         break;
1994                 case 2: /* sixteen banks */
1995                 default:
1996                         rdev->config.evergreen.tile_config |= 2 << 4;
1997                         break;
1998                 }
1999         }
2000         rdev->config.evergreen.tile_config |= 0 << 8;
2001         rdev->config.evergreen.tile_config |=
2002                 ((gb_addr_config & 0x30000000) >> 28) << 12;
2003
2004         num_shader_engines = (gb_addr_config & NUM_SHADER_ENGINES(3) >> 12) + 1;
2005
2006         if ((rdev->family >= CHIP_CEDAR) && (rdev->family <= CHIP_HEMLOCK)) {
2007                 u32 efuse_straps_4;
2008                 u32 efuse_straps_3;
2009
2010                 WREG32(RCU_IND_INDEX, 0x204);
2011                 efuse_straps_4 = RREG32(RCU_IND_DATA);
2012                 WREG32(RCU_IND_INDEX, 0x203);
2013                 efuse_straps_3 = RREG32(RCU_IND_DATA);
2014                 tmp = (((efuse_straps_4 & 0xf) << 4) |
2015                       ((efuse_straps_3 & 0xf0000000) >> 28));
2016         } else {
2017                 tmp = 0;
2018                 for (i = (rdev->config.evergreen.num_ses - 1); i >= 0; i--) {
2019                         u32 rb_disable_bitmap;
2020
2021                         WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
2022                         WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i));
2023                         rb_disable_bitmap = (RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000) >> 16;
2024                         tmp <<= 4;
2025                         tmp |= rb_disable_bitmap;
2026                 }
2027         }
2028         /* enabled rb are just the one not disabled :) */
2029         disabled_rb_mask = tmp;
2030
2031         WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
2032         WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES);
2033
2034         WREG32(GB_ADDR_CONFIG, gb_addr_config);
2035         WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
2036         WREG32(HDP_ADDR_CONFIG, gb_addr_config);
2037         WREG32(DMA_TILING_CONFIG, gb_addr_config);
2038
2039         tmp = gb_addr_config & NUM_PIPES_MASK;
2040         tmp = r6xx_remap_render_backend(rdev, tmp, rdev->config.evergreen.max_backends,
2041                                         EVERGREEN_MAX_BACKENDS, disabled_rb_mask);
2042         WREG32(GB_BACKEND_MAP, tmp);
2043
2044         WREG32(CGTS_SYS_TCC_DISABLE, 0);
2045         WREG32(CGTS_TCC_DISABLE, 0);
2046         WREG32(CGTS_USER_SYS_TCC_DISABLE, 0);
2047         WREG32(CGTS_USER_TCC_DISABLE, 0);
2048
2049         /* set HW defaults for 3D engine */
2050         WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
2051                                      ROQ_IB2_START(0x2b)));
2052
2053         WREG32(CP_MEQ_THRESHOLDS, STQ_SPLIT(0x30));
2054
2055         WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO |
2056                              SYNC_GRADIENT |
2057                              SYNC_WALKER |
2058                              SYNC_ALIGNER));
2059
2060         sx_debug_1 = RREG32(SX_DEBUG_1);
2061         sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS;
2062         WREG32(SX_DEBUG_1, sx_debug_1);
2063
2064
2065         smx_dc_ctl0 = RREG32(SMX_DC_CTL0);
2066         smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff);
2067         smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.evergreen.sx_num_of_sets);
2068         WREG32(SMX_DC_CTL0, smx_dc_ctl0);
2069
2070         if (rdev->family <= CHIP_SUMO2)
2071                 WREG32(SMX_SAR_CTL0, 0x00010000);
2072
2073         WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_size / 4) - 1) |
2074                                         POSITION_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_pos_size / 4) - 1) |
2075                                         SMX_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_smx_size / 4) - 1)));
2076
2077         WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.evergreen.sc_prim_fifo_size) |
2078                                  SC_HIZ_TILE_FIFO_SIZE(rdev->config.evergreen.sc_hiz_tile_fifo_size) |
2079                                  SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.evergreen.sc_earlyz_tile_fifo_size)));
2080
2081         WREG32(VGT_NUM_INSTANCES, 1);
2082         WREG32(SPI_CONFIG_CNTL, 0);
2083         WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
2084         WREG32(CP_PERFMON_CNTL, 0);
2085
2086         WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.evergreen.sq_num_cf_insts) |
2087                                   FETCH_FIFO_HIWATER(0x4) |
2088                                   DONE_FIFO_HIWATER(0xe0) |
2089                                   ALU_UPDATE_FIFO_HIWATER(0x8)));
2090
2091         sq_config = RREG32(SQ_CONFIG);
2092         sq_config &= ~(PS_PRIO(3) |
2093                        VS_PRIO(3) |
2094                        GS_PRIO(3) |
2095                        ES_PRIO(3));
2096         sq_config |= (VC_ENABLE |
2097                       EXPORT_SRC_C |
2098                       PS_PRIO(0) |
2099                       VS_PRIO(1) |
2100                       GS_PRIO(2) |
2101                       ES_PRIO(3));
2102
2103         switch (rdev->family) {
2104         case CHIP_CEDAR:
2105         case CHIP_PALM:
2106         case CHIP_SUMO:
2107         case CHIP_SUMO2:
2108         case CHIP_CAICOS:
2109                 /* no vertex cache */
2110                 sq_config &= ~VC_ENABLE;
2111                 break;
2112         default:
2113                 break;
2114         }
2115
2116         sq_lds_resource_mgmt = RREG32(SQ_LDS_RESOURCE_MGMT);
2117
2118         sq_gpr_resource_mgmt_1 = NUM_PS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2))* 12 / 32);
2119         sq_gpr_resource_mgmt_1 |= NUM_VS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 6 / 32);
2120         sq_gpr_resource_mgmt_1 |= NUM_CLAUSE_TEMP_GPRS(4);
2121         sq_gpr_resource_mgmt_2 = NUM_GS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 4 / 32);
2122         sq_gpr_resource_mgmt_2 |= NUM_ES_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 4 / 32);
2123         sq_gpr_resource_mgmt_3 = NUM_HS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
2124         sq_gpr_resource_mgmt_3 |= NUM_LS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
2125
2126         switch (rdev->family) {
2127         case CHIP_CEDAR:
2128         case CHIP_PALM:
2129         case CHIP_SUMO:
2130         case CHIP_SUMO2:
2131                 ps_thread_count = 96;
2132                 break;
2133         default:
2134                 ps_thread_count = 128;
2135                 break;
2136         }
2137
2138         sq_thread_resource_mgmt = NUM_PS_THREADS(ps_thread_count);
2139         sq_thread_resource_mgmt |= NUM_VS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2140         sq_thread_resource_mgmt |= NUM_GS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2141         sq_thread_resource_mgmt |= NUM_ES_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2142         sq_thread_resource_mgmt_2 = NUM_HS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2143         sq_thread_resource_mgmt_2 |= NUM_LS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2144
2145         sq_stack_resource_mgmt_1 = NUM_PS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2146         sq_stack_resource_mgmt_1 |= NUM_VS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2147         sq_stack_resource_mgmt_2 = NUM_GS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2148         sq_stack_resource_mgmt_2 |= NUM_ES_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2149         sq_stack_resource_mgmt_3 = NUM_HS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2150         sq_stack_resource_mgmt_3 |= NUM_LS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2151
2152         WREG32(SQ_CONFIG, sq_config);
2153         WREG32(SQ_GPR_RESOURCE_MGMT_1, sq_gpr_resource_mgmt_1);
2154         WREG32(SQ_GPR_RESOURCE_MGMT_2, sq_gpr_resource_mgmt_2);
2155         WREG32(SQ_GPR_RESOURCE_MGMT_3, sq_gpr_resource_mgmt_3);
2156         WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
2157         WREG32(SQ_THREAD_RESOURCE_MGMT_2, sq_thread_resource_mgmt_2);
2158         WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
2159         WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
2160         WREG32(SQ_STACK_RESOURCE_MGMT_3, sq_stack_resource_mgmt_3);
2161         WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0);
2162         WREG32(SQ_LDS_RESOURCE_MGMT, sq_lds_resource_mgmt);
2163
2164         WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
2165                                           FORCE_EOV_MAX_REZ_CNT(255)));
2166
2167         switch (rdev->family) {
2168         case CHIP_CEDAR:
2169         case CHIP_PALM:
2170         case CHIP_SUMO:
2171         case CHIP_SUMO2:
2172         case CHIP_CAICOS:
2173                 vgt_cache_invalidation = CACHE_INVALIDATION(TC_ONLY);
2174                 break;
2175         default:
2176                 vgt_cache_invalidation = CACHE_INVALIDATION(VC_AND_TC);
2177                 break;
2178         }
2179         vgt_cache_invalidation |= AUTO_INVLD_EN(ES_AND_GS_AUTO);
2180         WREG32(VGT_CACHE_INVALIDATION, vgt_cache_invalidation);
2181
2182         WREG32(VGT_GS_VERTEX_REUSE, 16);
2183         WREG32(PA_SU_LINE_STIPPLE_VALUE, 0);
2184         WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
2185
2186         WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
2187         WREG32(VGT_OUT_DEALLOC_CNTL, 16);
2188
2189         WREG32(CB_PERF_CTR0_SEL_0, 0);
2190         WREG32(CB_PERF_CTR0_SEL_1, 0);
2191         WREG32(CB_PERF_CTR1_SEL_0, 0);
2192         WREG32(CB_PERF_CTR1_SEL_1, 0);
2193         WREG32(CB_PERF_CTR2_SEL_0, 0);
2194         WREG32(CB_PERF_CTR2_SEL_1, 0);
2195         WREG32(CB_PERF_CTR3_SEL_0, 0);
2196         WREG32(CB_PERF_CTR3_SEL_1, 0);
2197
2198         /* clear render buffer base addresses */
2199         WREG32(CB_COLOR0_BASE, 0);
2200         WREG32(CB_COLOR1_BASE, 0);
2201         WREG32(CB_COLOR2_BASE, 0);
2202         WREG32(CB_COLOR3_BASE, 0);
2203         WREG32(CB_COLOR4_BASE, 0);
2204         WREG32(CB_COLOR5_BASE, 0);
2205         WREG32(CB_COLOR6_BASE, 0);
2206         WREG32(CB_COLOR7_BASE, 0);
2207         WREG32(CB_COLOR8_BASE, 0);
2208         WREG32(CB_COLOR9_BASE, 0);
2209         WREG32(CB_COLOR10_BASE, 0);
2210         WREG32(CB_COLOR11_BASE, 0);
2211
2212         /* set the shader const cache sizes to 0 */
2213         for (i = SQ_ALU_CONST_BUFFER_SIZE_PS_0; i < 0x28200; i += 4)
2214                 WREG32(i, 0);
2215         for (i = SQ_ALU_CONST_BUFFER_SIZE_HS_0; i < 0x29000; i += 4)
2216                 WREG32(i, 0);
2217
2218         tmp = RREG32(HDP_MISC_CNTL);
2219         tmp |= HDP_FLUSH_INVALIDATE_CACHE;
2220         WREG32(HDP_MISC_CNTL, tmp);
2221
2222         hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
2223         WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
2224
2225         WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
2226
2227         udelay(50);
2228
2229 }
2230
2231 int evergreen_mc_init(struct radeon_device *rdev)
2232 {
2233         u32 tmp;
2234         int chansize, numchan;
2235
2236         /* Get VRAM informations */
2237         rdev->mc.vram_is_ddr = true;
2238         if ((rdev->family == CHIP_PALM) ||
2239             (rdev->family == CHIP_SUMO) ||
2240             (rdev->family == CHIP_SUMO2))
2241                 tmp = RREG32(FUS_MC_ARB_RAMCFG);
2242         else
2243                 tmp = RREG32(MC_ARB_RAMCFG);
2244         if (tmp & CHANSIZE_OVERRIDE) {
2245                 chansize = 16;
2246         } else if (tmp & CHANSIZE_MASK) {
2247                 chansize = 64;
2248         } else {
2249                 chansize = 32;
2250         }
2251         tmp = RREG32(MC_SHARED_CHMAP);
2252         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
2253         case 0:
2254         default:
2255                 numchan = 1;
2256                 break;
2257         case 1:
2258                 numchan = 2;
2259                 break;
2260         case 2:
2261                 numchan = 4;
2262                 break;
2263         case 3:
2264                 numchan = 8;
2265                 break;
2266         }
2267         rdev->mc.vram_width = numchan * chansize;
2268         /* Could aper size report 0 ? */
2269         rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2270         rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2271         /* Setup GPU memory space */
2272         if ((rdev->family == CHIP_PALM) ||
2273             (rdev->family == CHIP_SUMO) ||
2274             (rdev->family == CHIP_SUMO2)) {
2275                 /* size in bytes on fusion */
2276                 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
2277                 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
2278         } else {
2279                 /* size in MB on evergreen/cayman/tn */
2280                 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2281                 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2282         }
2283         rdev->mc.visible_vram_size = rdev->mc.aper_size;
2284         r700_vram_gtt_location(rdev, &rdev->mc);
2285         radeon_update_bandwidth_info(rdev);
2286
2287         return 0;
2288 }
2289
2290 bool evergreen_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2291 {
2292         u32 srbm_status;
2293         u32 grbm_status;
2294         u32 grbm_status_se0, grbm_status_se1;
2295
2296         srbm_status = RREG32(SRBM_STATUS);
2297         grbm_status = RREG32(GRBM_STATUS);
2298         grbm_status_se0 = RREG32(GRBM_STATUS_SE0);
2299         grbm_status_se1 = RREG32(GRBM_STATUS_SE1);
2300         if (!(grbm_status & GUI_ACTIVE)) {
2301                 radeon_ring_lockup_update(ring);
2302                 return false;
2303         }
2304         /* force CP activities */
2305         radeon_ring_force_activity(rdev, ring);
2306         return radeon_ring_test_lockup(rdev, ring);
2307 }
2308
2309 static void evergreen_gpu_soft_reset_gfx(struct radeon_device *rdev)
2310 {
2311         u32 grbm_reset = 0;
2312
2313         if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
2314                 return;
2315
2316         dev_info(rdev->dev, "  GRBM_STATUS               = 0x%08X\n",
2317                 RREG32(GRBM_STATUS));
2318         dev_info(rdev->dev, "  GRBM_STATUS_SE0           = 0x%08X\n",
2319                 RREG32(GRBM_STATUS_SE0));
2320         dev_info(rdev->dev, "  GRBM_STATUS_SE1           = 0x%08X\n",
2321                 RREG32(GRBM_STATUS_SE1));
2322         dev_info(rdev->dev, "  SRBM_STATUS               = 0x%08X\n",
2323                 RREG32(SRBM_STATUS));
2324         dev_info(rdev->dev, "  R_008674_CP_STALLED_STAT1 = 0x%08X\n",
2325                 RREG32(CP_STALLED_STAT1));
2326         dev_info(rdev->dev, "  R_008678_CP_STALLED_STAT2 = 0x%08X\n",
2327                 RREG32(CP_STALLED_STAT2));
2328         dev_info(rdev->dev, "  R_00867C_CP_BUSY_STAT     = 0x%08X\n",
2329                 RREG32(CP_BUSY_STAT));
2330         dev_info(rdev->dev, "  R_008680_CP_STAT          = 0x%08X\n",
2331                 RREG32(CP_STAT));
2332
2333         /* Disable CP parsing/prefetching */
2334         WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
2335
2336         /* reset all the gfx blocks */
2337         grbm_reset = (SOFT_RESET_CP |
2338                       SOFT_RESET_CB |
2339                       SOFT_RESET_DB |
2340                       SOFT_RESET_PA |
2341                       SOFT_RESET_SC |
2342                       SOFT_RESET_SPI |
2343                       SOFT_RESET_SH |
2344                       SOFT_RESET_SX |
2345                       SOFT_RESET_TC |
2346                       SOFT_RESET_TA |
2347                       SOFT_RESET_VC |
2348                       SOFT_RESET_VGT);
2349
2350         dev_info(rdev->dev, "  GRBM_SOFT_RESET=0x%08X\n", grbm_reset);
2351         WREG32(GRBM_SOFT_RESET, grbm_reset);
2352         (void)RREG32(GRBM_SOFT_RESET);
2353         udelay(50);
2354         WREG32(GRBM_SOFT_RESET, 0);
2355         (void)RREG32(GRBM_SOFT_RESET);
2356
2357         dev_info(rdev->dev, "  GRBM_STATUS               = 0x%08X\n",
2358                 RREG32(GRBM_STATUS));
2359         dev_info(rdev->dev, "  GRBM_STATUS_SE0           = 0x%08X\n",
2360                 RREG32(GRBM_STATUS_SE0));
2361         dev_info(rdev->dev, "  GRBM_STATUS_SE1           = 0x%08X\n",
2362                 RREG32(GRBM_STATUS_SE1));
2363         dev_info(rdev->dev, "  SRBM_STATUS               = 0x%08X\n",
2364                 RREG32(SRBM_STATUS));
2365         dev_info(rdev->dev, "  R_008674_CP_STALLED_STAT1 = 0x%08X\n",
2366                 RREG32(CP_STALLED_STAT1));
2367         dev_info(rdev->dev, "  R_008678_CP_STALLED_STAT2 = 0x%08X\n",
2368                 RREG32(CP_STALLED_STAT2));
2369         dev_info(rdev->dev, "  R_00867C_CP_BUSY_STAT     = 0x%08X\n",
2370                 RREG32(CP_BUSY_STAT));
2371         dev_info(rdev->dev, "  R_008680_CP_STAT          = 0x%08X\n",
2372                 RREG32(CP_STAT));
2373 }
2374
2375 static void evergreen_gpu_soft_reset_dma(struct radeon_device *rdev)
2376 {
2377         u32 tmp;
2378
2379         if (RREG32(DMA_STATUS_REG) & DMA_IDLE)
2380                 return;
2381
2382         dev_info(rdev->dev, "  R_00D034_DMA_STATUS_REG   = 0x%08X\n",
2383                 RREG32(DMA_STATUS_REG));
2384
2385         /* Disable DMA */
2386         tmp = RREG32(DMA_RB_CNTL);
2387         tmp &= ~DMA_RB_ENABLE;
2388         WREG32(DMA_RB_CNTL, tmp);
2389
2390         /* Reset dma */
2391         WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA);
2392         RREG32(SRBM_SOFT_RESET);
2393         udelay(50);
2394         WREG32(SRBM_SOFT_RESET, 0);
2395
2396         dev_info(rdev->dev, "  R_00D034_DMA_STATUS_REG   = 0x%08X\n",
2397                 RREG32(DMA_STATUS_REG));
2398 }
2399
2400 static int evergreen_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
2401 {
2402         struct evergreen_mc_save save;
2403
2404         if (reset_mask == 0)
2405                 return 0;
2406
2407         dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
2408
2409         evergreen_mc_stop(rdev, &save);
2410         if (evergreen_mc_wait_for_idle(rdev)) {
2411                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2412         }
2413
2414         if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE))
2415                 evergreen_gpu_soft_reset_gfx(rdev);
2416
2417         if (reset_mask & RADEON_RESET_DMA)
2418                 evergreen_gpu_soft_reset_dma(rdev);
2419
2420         /* Wait a little for things to settle down */
2421         udelay(50);
2422
2423         evergreen_mc_resume(rdev, &save);
2424         return 0;
2425 }
2426
2427 int evergreen_asic_reset(struct radeon_device *rdev)
2428 {
2429         return evergreen_gpu_soft_reset(rdev, (RADEON_RESET_GFX |
2430                                                RADEON_RESET_COMPUTE |
2431                                                RADEON_RESET_DMA));
2432 }
2433
2434 /* Interrupts */
2435
2436 u32 evergreen_get_vblank_counter(struct radeon_device *rdev, int crtc)
2437 {
2438         if (crtc >= rdev->num_crtc)
2439                 return 0;
2440         else
2441                 return RREG32(CRTC_STATUS_FRAME_COUNT + crtc_offsets[crtc]);
2442 }
2443
2444 void evergreen_disable_interrupt_state(struct radeon_device *rdev)
2445 {
2446         u32 tmp;
2447
2448         if (rdev->family >= CHIP_CAYMAN) {
2449                 cayman_cp_int_cntl_setup(rdev, 0,
2450                                          CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
2451                 cayman_cp_int_cntl_setup(rdev, 1, 0);
2452                 cayman_cp_int_cntl_setup(rdev, 2, 0);
2453                 tmp = RREG32(CAYMAN_DMA1_CNTL) & ~TRAP_ENABLE;
2454                 WREG32(CAYMAN_DMA1_CNTL, tmp);
2455         } else
2456                 WREG32(CP_INT_CNTL, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
2457         tmp = RREG32(DMA_CNTL) & ~TRAP_ENABLE;
2458         WREG32(DMA_CNTL, tmp);
2459         WREG32(GRBM_INT_CNTL, 0);
2460         WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
2461         WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
2462         if (rdev->num_crtc >= 4) {
2463                 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
2464                 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
2465         }
2466         if (rdev->num_crtc >= 6) {
2467                 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
2468                 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
2469         }
2470
2471         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
2472         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
2473         if (rdev->num_crtc >= 4) {
2474                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
2475                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
2476         }
2477         if (rdev->num_crtc >= 6) {
2478                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
2479                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
2480         }
2481
2482         /* only one DAC on DCE6 */
2483         if (!ASIC_IS_DCE6(rdev))
2484                 WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
2485         WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
2486
2487         tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2488         WREG32(DC_HPD1_INT_CONTROL, tmp);
2489         tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2490         WREG32(DC_HPD2_INT_CONTROL, tmp);
2491         tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2492         WREG32(DC_HPD3_INT_CONTROL, tmp);
2493         tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2494         WREG32(DC_HPD4_INT_CONTROL, tmp);
2495         tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2496         WREG32(DC_HPD5_INT_CONTROL, tmp);
2497         tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2498         WREG32(DC_HPD6_INT_CONTROL, tmp);
2499
2500 }
2501
2502 int evergreen_irq_set(struct radeon_device *rdev)
2503 {
2504         u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
2505         u32 cp_int_cntl1 = 0, cp_int_cntl2 = 0;
2506         u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
2507         u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
2508         u32 grbm_int_cntl = 0;
2509         u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
2510         u32 afmt1 = 0, afmt2 = 0, afmt3 = 0, afmt4 = 0, afmt5 = 0, afmt6 = 0;
2511         u32 dma_cntl, dma_cntl1 = 0;
2512
2513         if (!rdev->irq.installed) {
2514                 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
2515                 return -EINVAL;
2516         }
2517         /* don't enable anything if the ih is disabled */
2518         if (!rdev->ih.enabled) {
2519                 r600_disable_interrupts(rdev);
2520                 /* force the active interrupt state to all disabled */
2521                 evergreen_disable_interrupt_state(rdev);
2522                 return 0;
2523         }
2524
2525         hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2526         hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2527         hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2528         hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
2529         hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
2530         hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
2531
2532         afmt1 = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
2533         afmt2 = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
2534         afmt3 = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
2535         afmt4 = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
2536         afmt5 = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
2537         afmt6 = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET) & ~AFMT_AZ_FORMAT_WTRIG_MASK;
2538
2539         dma_cntl = RREG32(DMA_CNTL) & ~TRAP_ENABLE;
2540
2541         if (rdev->family >= CHIP_CAYMAN) {
2542                 /* enable CP interrupts on all rings */
2543                 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
2544                         DRM_DEBUG("evergreen_irq_set: sw int gfx\n");
2545                         cp_int_cntl |= TIME_STAMP_INT_ENABLE;
2546                 }
2547                 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
2548                         DRM_DEBUG("evergreen_irq_set: sw int cp1\n");
2549                         cp_int_cntl1 |= TIME_STAMP_INT_ENABLE;
2550                 }
2551                 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
2552                         DRM_DEBUG("evergreen_irq_set: sw int cp2\n");
2553                         cp_int_cntl2 |= TIME_STAMP_INT_ENABLE;
2554                 }
2555         } else {
2556                 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
2557                         DRM_DEBUG("evergreen_irq_set: sw int gfx\n");
2558                         cp_int_cntl |= RB_INT_ENABLE;
2559                         cp_int_cntl |= TIME_STAMP_INT_ENABLE;
2560                 }
2561         }
2562
2563         if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
2564                 DRM_DEBUG("r600_irq_set: sw int dma\n");
2565                 dma_cntl |= TRAP_ENABLE;
2566         }
2567
2568         if (rdev->family >= CHIP_CAYMAN) {
2569                 dma_cntl1 = RREG32(CAYMAN_DMA1_CNTL) & ~TRAP_ENABLE;
2570                 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
2571                         DRM_DEBUG("r600_irq_set: sw int dma1\n");
2572                         dma_cntl1 |= TRAP_ENABLE;
2573                 }
2574         }
2575
2576         if (rdev->irq.crtc_vblank_int[0] ||
2577             atomic_read(&rdev->irq.pflip[0])) {
2578                 DRM_DEBUG("evergreen_irq_set: vblank 0\n");
2579                 crtc1 |= VBLANK_INT_MASK;
2580         }
2581         if (rdev->irq.crtc_vblank_int[1] ||
2582             atomic_read(&rdev->irq.pflip[1])) {
2583                 DRM_DEBUG("evergreen_irq_set: vblank 1\n");
2584                 crtc2 |= VBLANK_INT_MASK;
2585         }
2586         if (rdev->irq.crtc_vblank_int[2] ||
2587             atomic_read(&rdev->irq.pflip[2])) {
2588                 DRM_DEBUG("evergreen_irq_set: vblank 2\n");
2589                 crtc3 |= VBLANK_INT_MASK;
2590         }
2591         if (rdev->irq.crtc_vblank_int[3] ||
2592             atomic_read(&rdev->irq.pflip[3])) {
2593                 DRM_DEBUG("evergreen_irq_set: vblank 3\n");
2594                 crtc4 |= VBLANK_INT_MASK;
2595         }
2596         if (rdev->irq.crtc_vblank_int[4] ||
2597             atomic_read(&rdev->irq.pflip[4])) {
2598                 DRM_DEBUG("evergreen_irq_set: vblank 4\n");
2599                 crtc5 |= VBLANK_INT_MASK;
2600         }
2601         if (rdev->irq.crtc_vblank_int[5] ||
2602             atomic_read(&rdev->irq.pflip[5])) {
2603                 DRM_DEBUG("evergreen_irq_set: vblank 5\n");
2604                 crtc6 |= VBLANK_INT_MASK;
2605         }
2606         if (rdev->irq.hpd[0]) {
2607                 DRM_DEBUG("evergreen_irq_set: hpd 1\n");
2608                 hpd1 |= DC_HPDx_INT_EN;
2609         }
2610         if (rdev->irq.hpd[1]) {
2611                 DRM_DEBUG("evergreen_irq_set: hpd 2\n");
2612                 hpd2 |= DC_HPDx_INT_EN;
2613         }
2614         if (rdev->irq.hpd[2]) {
2615                 DRM_DEBUG("evergreen_irq_set: hpd 3\n");
2616                 hpd3 |= DC_HPDx_INT_EN;
2617         }
2618         if (rdev->irq.hpd[3]) {
2619                 DRM_DEBUG("evergreen_irq_set: hpd 4\n");
2620                 hpd4 |= DC_HPDx_INT_EN;
2621         }
2622         if (rdev->irq.hpd[4]) {
2623                 DRM_DEBUG("evergreen_irq_set: hpd 5\n");
2624                 hpd5 |= DC_HPDx_INT_EN;
2625         }
2626         if (rdev->irq.hpd[5]) {
2627                 DRM_DEBUG("evergreen_irq_set: hpd 6\n");
2628                 hpd6 |= DC_HPDx_INT_EN;
2629         }
2630         if (rdev->irq.afmt[0]) {
2631                 DRM_DEBUG("evergreen_irq_set: hdmi 0\n");
2632                 afmt1 |= AFMT_AZ_FORMAT_WTRIG_MASK;
2633         }
2634         if (rdev->irq.afmt[1]) {
2635                 DRM_DEBUG("evergreen_irq_set: hdmi 1\n");
2636                 afmt2 |= AFMT_AZ_FORMAT_WTRIG_MASK;
2637         }
2638         if (rdev->irq.afmt[2]) {
2639                 DRM_DEBUG("evergreen_irq_set: hdmi 2\n");
2640                 afmt3 |= AFMT_AZ_FORMAT_WTRIG_MASK;
2641         }
2642         if (rdev->irq.afmt[3]) {
2643                 DRM_DEBUG("evergreen_irq_set: hdmi 3\n");
2644                 afmt4 |= AFMT_AZ_FORMAT_WTRIG_MASK;
2645         }
2646         if (rdev->irq.afmt[4]) {
2647                 DRM_DEBUG("evergreen_irq_set: hdmi 4\n");
2648                 afmt5 |= AFMT_AZ_FORMAT_WTRIG_MASK;
2649         }
2650         if (rdev->irq.afmt[5]) {
2651                 DRM_DEBUG("evergreen_irq_set: hdmi 5\n");
2652                 afmt6 |= AFMT_AZ_FORMAT_WTRIG_MASK;
2653         }
2654
2655         if (rdev->family >= CHIP_CAYMAN) {
2656                 cayman_cp_int_cntl_setup(rdev, 0, cp_int_cntl);
2657                 cayman_cp_int_cntl_setup(rdev, 1, cp_int_cntl1);
2658                 cayman_cp_int_cntl_setup(rdev, 2, cp_int_cntl2);
2659         } else
2660                 WREG32(CP_INT_CNTL, cp_int_cntl);
2661
2662         WREG32(DMA_CNTL, dma_cntl);
2663
2664         if (rdev->family >= CHIP_CAYMAN)
2665                 WREG32(CAYMAN_DMA1_CNTL, dma_cntl1);
2666
2667         WREG32(GRBM_INT_CNTL, grbm_int_cntl);
2668
2669         WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
2670         WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
2671         if (rdev->num_crtc >= 4) {
2672                 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
2673                 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
2674         }
2675         if (rdev->num_crtc >= 6) {
2676                 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
2677                 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
2678         }
2679
2680         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, grph1);
2681         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, grph2);
2682         if (rdev->num_crtc >= 4) {
2683                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, grph3);
2684                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, grph4);
2685         }
2686         if (rdev->num_crtc >= 6) {
2687                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, grph5);
2688                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, grph6);
2689         }
2690
2691         WREG32(DC_HPD1_INT_CONTROL, hpd1);
2692         WREG32(DC_HPD2_INT_CONTROL, hpd2);
2693         WREG32(DC_HPD3_INT_CONTROL, hpd3);
2694         WREG32(DC_HPD4_INT_CONTROL, hpd4);
2695         WREG32(DC_HPD5_INT_CONTROL, hpd5);
2696         WREG32(DC_HPD6_INT_CONTROL, hpd6);
2697
2698         WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, afmt1);
2699         WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, afmt2);
2700         WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, afmt3);
2701         WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, afmt4);
2702         WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, afmt5);
2703         WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, afmt6);
2704
2705         return 0;
2706 }
2707
2708 static void evergreen_irq_ack(struct radeon_device *rdev)
2709 {
2710         u32 tmp;
2711
2712         rdev->irq.stat_regs.evergreen.disp_int = RREG32(DISP_INTERRUPT_STATUS);
2713         rdev->irq.stat_regs.evergreen.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
2714         rdev->irq.stat_regs.evergreen.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
2715         rdev->irq.stat_regs.evergreen.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
2716         rdev->irq.stat_regs.evergreen.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
2717         rdev->irq.stat_regs.evergreen.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
2718         rdev->irq.stat_regs.evergreen.d1grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
2719         rdev->irq.stat_regs.evergreen.d2grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
2720         if (rdev->num_crtc >= 4) {
2721                 rdev->irq.stat_regs.evergreen.d3grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
2722                 rdev->irq.stat_regs.evergreen.d4grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
2723         }
2724         if (rdev->num_crtc >= 6) {
2725                 rdev->irq.stat_regs.evergreen.d5grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
2726                 rdev->irq.stat_regs.evergreen.d6grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
2727         }
2728
2729         rdev->irq.stat_regs.evergreen.afmt_status1 = RREG32(AFMT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
2730         rdev->irq.stat_regs.evergreen.afmt_status2 = RREG32(AFMT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
2731         rdev->irq.stat_regs.evergreen.afmt_status3 = RREG32(AFMT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
2732         rdev->irq.stat_regs.evergreen.afmt_status4 = RREG32(AFMT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
2733         rdev->irq.stat_regs.evergreen.afmt_status5 = RREG32(AFMT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
2734         rdev->irq.stat_regs.evergreen.afmt_status6 = RREG32(AFMT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
2735
2736         if (rdev->irq.stat_regs.evergreen.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
2737                 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2738         if (rdev->irq.stat_regs.evergreen.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
2739                 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2740         if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT)
2741                 WREG32(VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
2742         if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT)
2743                 WREG32(VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
2744         if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
2745                 WREG32(VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
2746         if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT)
2747                 WREG32(VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
2748
2749         if (rdev->num_crtc >= 4) {
2750                 if (rdev->irq.stat_regs.evergreen.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
2751                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2752                 if (rdev->irq.stat_regs.evergreen.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
2753                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2754                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
2755                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
2756                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
2757                         WREG32(VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
2758                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
2759                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
2760                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
2761                         WREG32(VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
2762         }
2763
2764         if (rdev->num_crtc >= 6) {
2765                 if (rdev->irq.stat_regs.evergreen.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
2766                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2767                 if (rdev->irq.stat_regs.evergreen.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
2768                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2769                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
2770                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
2771                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
2772                         WREG32(VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
2773                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
2774                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
2775                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
2776                         WREG32(VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
2777         }
2778
2779         if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
2780                 tmp = RREG32(DC_HPD1_INT_CONTROL);
2781                 tmp |= DC_HPDx_INT_ACK;
2782                 WREG32(DC_HPD1_INT_CONTROL, tmp);
2783         }
2784         if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
2785                 tmp = RREG32(DC_HPD2_INT_CONTROL);
2786                 tmp |= DC_HPDx_INT_ACK;
2787                 WREG32(DC_HPD2_INT_CONTROL, tmp);
2788         }
2789         if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
2790                 tmp = RREG32(DC_HPD3_INT_CONTROL);
2791                 tmp |= DC_HPDx_INT_ACK;
2792                 WREG32(DC_HPD3_INT_CONTROL, tmp);
2793         }
2794         if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
2795                 tmp = RREG32(DC_HPD4_INT_CONTROL);
2796                 tmp |= DC_HPDx_INT_ACK;
2797                 WREG32(DC_HPD4_INT_CONTROL, tmp);
2798         }
2799         if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
2800                 tmp = RREG32(DC_HPD5_INT_CONTROL);
2801                 tmp |= DC_HPDx_INT_ACK;
2802                 WREG32(DC_HPD5_INT_CONTROL, tmp);
2803         }
2804         if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
2805                 tmp = RREG32(DC_HPD5_INT_CONTROL);
2806                 tmp |= DC_HPDx_INT_ACK;
2807                 WREG32(DC_HPD6_INT_CONTROL, tmp);
2808         }
2809         if (rdev->irq.stat_regs.evergreen.afmt_status1 & AFMT_AZ_FORMAT_WTRIG) {
2810                 tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET);
2811                 tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
2812                 WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, tmp);
2813         }
2814         if (rdev->irq.stat_regs.evergreen.afmt_status2 & AFMT_AZ_FORMAT_WTRIG) {
2815                 tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET);
2816                 tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
2817                 WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, tmp);
2818         }
2819         if (rdev->irq.stat_regs.evergreen.afmt_status3 & AFMT_AZ_FORMAT_WTRIG) {
2820                 tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET);
2821                 tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
2822                 WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, tmp);
2823         }
2824         if (rdev->irq.stat_regs.evergreen.afmt_status4 & AFMT_AZ_FORMAT_WTRIG) {
2825                 tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET);
2826                 tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
2827                 WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, tmp);
2828         }
2829         if (rdev->irq.stat_regs.evergreen.afmt_status5 & AFMT_AZ_FORMAT_WTRIG) {
2830                 tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET);
2831                 tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
2832                 WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, tmp);
2833         }
2834         if (rdev->irq.stat_regs.evergreen.afmt_status6 & AFMT_AZ_FORMAT_WTRIG) {
2835                 tmp = RREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET);
2836                 tmp |= AFMT_AZ_FORMAT_WTRIG_ACK;
2837                 WREG32(AFMT_AUDIO_PACKET_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, tmp);
2838         }
2839 }
2840
2841 static void evergreen_irq_disable(struct radeon_device *rdev)
2842 {
2843         r600_disable_interrupts(rdev);
2844         /* Wait and acknowledge irq */
2845         mdelay(1);
2846         evergreen_irq_ack(rdev);
2847         evergreen_disable_interrupt_state(rdev);
2848 }
2849
2850 void evergreen_irq_suspend(struct radeon_device *rdev)
2851 {
2852         evergreen_irq_disable(rdev);
2853         r600_rlc_stop(rdev);
2854 }
2855
2856 static u32 evergreen_get_ih_wptr(struct radeon_device *rdev)
2857 {
2858         u32 wptr, tmp;
2859
2860         if (rdev->wb.enabled)
2861                 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
2862         else
2863                 wptr = RREG32(IH_RB_WPTR);
2864
2865         if (wptr & RB_OVERFLOW) {
2866                 /* When a ring buffer overflow happen start parsing interrupt
2867                  * from the last not overwritten vector (wptr + 16). Hopefully
2868                  * this should allow us to catchup.
2869                  */
2870                 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
2871                         wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
2872                 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
2873                 tmp = RREG32(IH_RB_CNTL);
2874                 tmp |= IH_WPTR_OVERFLOW_CLEAR;
2875                 WREG32(IH_RB_CNTL, tmp);
2876         }
2877         return (wptr & rdev->ih.ptr_mask);
2878 }
2879
2880 int evergreen_irq_process(struct radeon_device *rdev)
2881 {
2882         u32 wptr;
2883         u32 rptr;
2884         u32 src_id, src_data;
2885         u32 ring_index;
2886         bool queue_hotplug = false;
2887         bool queue_hdmi = false;
2888
2889         if (!rdev->ih.enabled || rdev->shutdown)
2890                 return IRQ_NONE;
2891
2892         wptr = evergreen_get_ih_wptr(rdev);
2893
2894 restart_ih:
2895         /* is somebody else already processing irqs? */
2896         if (atomic_xchg(&rdev->ih.lock, 1))
2897                 return IRQ_NONE;
2898
2899         rptr = rdev->ih.rptr;
2900         DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
2901
2902         /* Order reading of wptr vs. reading of IH ring data */
2903         rmb();
2904
2905         /* display interrupts */
2906         evergreen_irq_ack(rdev);
2907
2908         while (rptr != wptr) {
2909                 /* wptr/rptr are in bytes! */
2910                 ring_index = rptr / 4;
2911                 src_id =  le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
2912                 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
2913
2914                 switch (src_id) {
2915                 case 1: /* D1 vblank/vline */
2916                         switch (src_data) {
2917                         case 0: /* D1 vblank */
2918                                 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT) {
2919                                         if (rdev->irq.crtc_vblank_int[0]) {
2920                                                 drm_handle_vblank(rdev->ddev, 0);
2921                                                 rdev->pm.vblank_sync = true;
2922                                                 wake_up(&rdev->irq.vblank_queue);
2923                                         }
2924                                         if (atomic_read(&rdev->irq.pflip[0]))
2925                                                 radeon_crtc_handle_flip(rdev, 0);
2926                                         rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
2927                                         DRM_DEBUG("IH: D1 vblank\n");
2928                                 }
2929                                 break;
2930                         case 1: /* D1 vline */
2931                                 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT) {
2932                                         rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VLINE_INTERRUPT;
2933                                         DRM_DEBUG("IH: D1 vline\n");
2934                                 }
2935                                 break;
2936                         default:
2937                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2938                                 break;
2939                         }
2940                         break;
2941                 case 2: /* D2 vblank/vline */
2942                         switch (src_data) {
2943                         case 0: /* D2 vblank */
2944                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
2945                                         if (rdev->irq.crtc_vblank_int[1]) {
2946                                                 drm_handle_vblank(rdev->ddev, 1);
2947                                                 rdev->pm.vblank_sync = true;
2948                                                 wake_up(&rdev->irq.vblank_queue);
2949                                         }
2950                                         if (atomic_read(&rdev->irq.pflip[1]))
2951                                                 radeon_crtc_handle_flip(rdev, 1);
2952                                         rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
2953                                         DRM_DEBUG("IH: D2 vblank\n");
2954                                 }
2955                                 break;
2956                         case 1: /* D2 vline */
2957                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
2958                                         rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
2959                                         DRM_DEBUG("IH: D2 vline\n");
2960                                 }
2961                                 break;
2962                         default:
2963                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2964                                 break;
2965                         }
2966                         break;
2967                 case 3: /* D3 vblank/vline */
2968                         switch (src_data) {
2969                         case 0: /* D3 vblank */
2970                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
2971                                         if (rdev->irq.crtc_vblank_int[2]) {
2972                                                 drm_handle_vblank(rdev->ddev, 2);
2973                                                 rdev->pm.vblank_sync = true;
2974                                                 wake_up(&rdev->irq.vblank_queue);
2975                                         }
2976                                         if (atomic_read(&rdev->irq.pflip[2]))
2977                                                 radeon_crtc_handle_flip(rdev, 2);
2978                                         rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
2979                                         DRM_DEBUG("IH: D3 vblank\n");
2980                                 }
2981                                 break;
2982                         case 1: /* D3 vline */
2983                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
2984                                         rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
2985                                         DRM_DEBUG("IH: D3 vline\n");
2986                                 }
2987                                 break;
2988                         default:
2989                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2990                                 break;
2991                         }
2992                         break;
2993                 case 4: /* D4 vblank/vline */
2994                         switch (src_data) {
2995                         case 0: /* D4 vblank */
2996                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
2997                                         if (rdev->irq.crtc_vblank_int[3]) {
2998                                                 drm_handle_vblank(rdev->ddev, 3);
2999                                                 rdev->pm.vblank_sync = true;
3000                                                 wake_up(&rdev->irq.vblank_queue);
3001                                         }
3002                                         if (atomic_read(&rdev->irq.pflip[3]))
3003                                                 radeon_crtc_handle_flip(rdev, 3);
3004                                         rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
3005                                         DRM_DEBUG("IH: D4 vblank\n");
3006                                 }
3007                                 break;
3008                         case 1: /* D4 vline */
3009                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
3010                                         rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
3011                                         DRM_DEBUG("IH: D4 vline\n");
3012                                 }
3013                                 break;
3014                         default:
3015                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3016                                 break;
3017                         }
3018                         break;
3019                 case 5: /* D5 vblank/vline */
3020                         switch (src_data) {
3021                         case 0: /* D5 vblank */
3022                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
3023                                         if (rdev->irq.crtc_vblank_int[4]) {
3024                                                 drm_handle_vblank(rdev->ddev, 4);
3025                                                 rdev->pm.vblank_sync = true;
3026                                                 wake_up(&rdev->irq.vblank_queue);
3027                                         }
3028                                         if (atomic_read(&rdev->irq.pflip[4]))
3029                                                 radeon_crtc_handle_flip(rdev, 4);
3030                                         rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
3031                                         DRM_DEBUG("IH: D5 vblank\n");
3032                                 }
3033                                 break;
3034                         case 1: /* D5 vline */
3035                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
3036                                         rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
3037                                         DRM_DEBUG("IH: D5 vline\n");
3038                                 }
3039                                 break;
3040                         default:
3041                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3042                                 break;
3043                         }
3044                         break;
3045                 case 6: /* D6 vblank/vline */
3046                         switch (src_data) {
3047                         case 0: /* D6 vblank */
3048                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
3049                                         if (rdev->irq.crtc_vblank_int[5]) {
3050                                                 drm_handle_vblank(rdev->ddev, 5);
3051                                                 rdev->pm.vblank_sync = true;
3052                                                 wake_up(&rdev->irq.vblank_queue);
3053                                         }
3054                                         if (atomic_read(&rdev->irq.pflip[5]))
3055                                                 radeon_crtc_handle_flip(rdev, 5);
3056                                         rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
3057                                         DRM_DEBUG("IH: D6 vblank\n");
3058                                 }
3059                                 break;
3060                         case 1: /* D6 vline */
3061                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
3062                                         rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
3063                            &nbs