Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-btrfs-devel.git] / drivers / gpu / drm / radeon / evergreen.c
blobc4ffa14fb2f45ed09733cca219c54873f48d5a82
1 /*
2 * Copyright 2010 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
22 * Authors: Alex Deucher
24 #include <linux/firmware.h>
25 #include <linux/platform_device.h>
26 #include <linux/slab.h>
27 #include "drmP.h"
28 #include "radeon.h"
29 #include "radeon_asic.h"
30 #include "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"
37 #define EVERGREEN_PFP_UCODE_SIZE 1120
38 #define EVERGREEN_PM4_UCODE_SIZE 1376
40 static void evergreen_gpu_init(struct radeon_device *rdev);
41 void evergreen_fini(struct radeon_device *rdev);
42 static void evergreen_pcie_gen2_enable(struct radeon_device *rdev);
44 void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev)
46 u16 ctl, v;
47 int cap, err;
49 cap = pci_pcie_cap(rdev->pdev);
50 if (!cap)
51 return;
53 err = pci_read_config_word(rdev->pdev, cap + PCI_EXP_DEVCTL, &ctl);
54 if (err)
55 return;
57 v = (ctl & PCI_EXP_DEVCTL_READRQ) >> 12;
59 /* if bios or OS sets MAX_READ_REQUEST_SIZE to an invalid value, fix it
60 * to avoid hangs or perfomance issues
62 if ((v == 0) || (v == 6) || (v == 7)) {
63 ctl &= ~PCI_EXP_DEVCTL_READRQ;
64 ctl |= (2 << 12);
65 pci_write_config_word(rdev->pdev, cap + PCI_EXP_DEVCTL, ctl);
69 void evergreen_pre_page_flip(struct radeon_device *rdev, int crtc)
71 /* enable the pflip int */
72 radeon_irq_kms_pflip_irq_get(rdev, crtc);
75 void evergreen_post_page_flip(struct radeon_device *rdev, int crtc)
77 /* disable the pflip int */
78 radeon_irq_kms_pflip_irq_put(rdev, crtc);
81 u32 evergreen_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
83 struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
84 u32 tmp = RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset);
86 /* Lock the graphics update lock */
87 tmp |= EVERGREEN_GRPH_UPDATE_LOCK;
88 WREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
90 /* update the scanout addresses */
91 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
92 upper_32_bits(crtc_base));
93 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
94 (u32)crtc_base);
96 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
97 upper_32_bits(crtc_base));
98 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
99 (u32)crtc_base);
101 /* Wait for update_pending to go high. */
102 while (!(RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING));
103 DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
105 /* Unlock the lock, so double-buffering can take place inside vblank */
106 tmp &= ~EVERGREEN_GRPH_UPDATE_LOCK;
107 WREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
109 /* Return current update_pending status: */
110 return RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING;
113 /* get temperature in millidegrees */
114 int evergreen_get_temp(struct radeon_device *rdev)
116 u32 temp, toffset;
117 int actual_temp = 0;
119 if (rdev->family == CHIP_JUNIPER) {
120 toffset = (RREG32(CG_THERMAL_CTRL) & TOFFSET_MASK) >>
121 TOFFSET_SHIFT;
122 temp = (RREG32(CG_TS0_STATUS) & TS0_ADC_DOUT_MASK) >>
123 TS0_ADC_DOUT_SHIFT;
125 if (toffset & 0x100)
126 actual_temp = temp / 2 - (0x200 - toffset);
127 else
128 actual_temp = temp / 2 + toffset;
130 actual_temp = actual_temp * 1000;
132 } else {
133 temp = (RREG32(CG_MULT_THERMAL_STATUS) & ASIC_T_MASK) >>
134 ASIC_T_SHIFT;
136 if (temp & 0x400)
137 actual_temp = -256;
138 else if (temp & 0x200)
139 actual_temp = 255;
140 else if (temp & 0x100) {
141 actual_temp = temp & 0x1ff;
142 actual_temp |= ~0x1ff;
143 } else
144 actual_temp = temp & 0xff;
146 actual_temp = (actual_temp * 1000) / 2;
149 return actual_temp;
152 int sumo_get_temp(struct radeon_device *rdev)
154 u32 temp = RREG32(CG_THERMAL_STATUS) & 0xff;
155 int actual_temp = temp - 49;
157 return actual_temp * 1000;
160 void evergreen_pm_misc(struct radeon_device *rdev)
162 int req_ps_idx = rdev->pm.requested_power_state_index;
163 int req_cm_idx = rdev->pm.requested_clock_mode_index;
164 struct radeon_power_state *ps = &rdev->pm.power_state[req_ps_idx];
165 struct radeon_voltage *voltage = &ps->clock_info[req_cm_idx].voltage;
167 if (voltage->type == VOLTAGE_SW) {
168 /* 0xff01 is a flag rather then an actual voltage */
169 if (voltage->voltage == 0xff01)
170 return;
171 if (voltage->voltage && (voltage->voltage != rdev->pm.current_vddc)) {
172 radeon_atom_set_voltage(rdev, voltage->voltage, SET_VOLTAGE_TYPE_ASIC_VDDC);
173 rdev->pm.current_vddc = voltage->voltage;
174 DRM_DEBUG("Setting: vddc: %d\n", voltage->voltage);
176 /* 0xff01 is a flag rather then an actual voltage */
177 if (voltage->vddci == 0xff01)
178 return;
179 if (voltage->vddci && (voltage->vddci != rdev->pm.current_vddci)) {
180 radeon_atom_set_voltage(rdev, voltage->vddci, SET_VOLTAGE_TYPE_ASIC_VDDCI);
181 rdev->pm.current_vddci = voltage->vddci;
182 DRM_DEBUG("Setting: vddci: %d\n", voltage->vddci);
187 void evergreen_pm_prepare(struct radeon_device *rdev)
189 struct drm_device *ddev = rdev->ddev;
190 struct drm_crtc *crtc;
191 struct radeon_crtc *radeon_crtc;
192 u32 tmp;
194 /* disable any active CRTCs */
195 list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
196 radeon_crtc = to_radeon_crtc(crtc);
197 if (radeon_crtc->enabled) {
198 tmp = RREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset);
199 tmp |= EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
200 WREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset, tmp);
205 void evergreen_pm_finish(struct radeon_device *rdev)
207 struct drm_device *ddev = rdev->ddev;
208 struct drm_crtc *crtc;
209 struct radeon_crtc *radeon_crtc;
210 u32 tmp;
212 /* enable any active CRTCs */
213 list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
214 radeon_crtc = to_radeon_crtc(crtc);
215 if (radeon_crtc->enabled) {
216 tmp = RREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset);
217 tmp &= ~EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE;
218 WREG32(EVERGREEN_CRTC_CONTROL + radeon_crtc->crtc_offset, tmp);
223 bool evergreen_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
225 bool connected = false;
227 switch (hpd) {
228 case RADEON_HPD_1:
229 if (RREG32(DC_HPD1_INT_STATUS) & DC_HPDx_SENSE)
230 connected = true;
231 break;
232 case RADEON_HPD_2:
233 if (RREG32(DC_HPD2_INT_STATUS) & DC_HPDx_SENSE)
234 connected = true;
235 break;
236 case RADEON_HPD_3:
237 if (RREG32(DC_HPD3_INT_STATUS) & DC_HPDx_SENSE)
238 connected = true;
239 break;
240 case RADEON_HPD_4:
241 if (RREG32(DC_HPD4_INT_STATUS) & DC_HPDx_SENSE)
242 connected = true;
243 break;
244 case RADEON_HPD_5:
245 if (RREG32(DC_HPD5_INT_STATUS) & DC_HPDx_SENSE)
246 connected = true;
247 break;
248 case RADEON_HPD_6:
249 if (RREG32(DC_HPD6_INT_STATUS) & DC_HPDx_SENSE)
250 connected = true;
251 break;
252 default:
253 break;
256 return connected;
259 void evergreen_hpd_set_polarity(struct radeon_device *rdev,
260 enum radeon_hpd_id hpd)
262 u32 tmp;
263 bool connected = evergreen_hpd_sense(rdev, hpd);
265 switch (hpd) {
266 case RADEON_HPD_1:
267 tmp = RREG32(DC_HPD1_INT_CONTROL);
268 if (connected)
269 tmp &= ~DC_HPDx_INT_POLARITY;
270 else
271 tmp |= DC_HPDx_INT_POLARITY;
272 WREG32(DC_HPD1_INT_CONTROL, tmp);
273 break;
274 case RADEON_HPD_2:
275 tmp = RREG32(DC_HPD2_INT_CONTROL);
276 if (connected)
277 tmp &= ~DC_HPDx_INT_POLARITY;
278 else
279 tmp |= DC_HPDx_INT_POLARITY;
280 WREG32(DC_HPD2_INT_CONTROL, tmp);
281 break;
282 case RADEON_HPD_3:
283 tmp = RREG32(DC_HPD3_INT_CONTROL);
284 if (connected)
285 tmp &= ~DC_HPDx_INT_POLARITY;
286 else
287 tmp |= DC_HPDx_INT_POLARITY;
288 WREG32(DC_HPD3_INT_CONTROL, tmp);
289 break;
290 case RADEON_HPD_4:
291 tmp = RREG32(DC_HPD4_INT_CONTROL);
292 if (connected)
293 tmp &= ~DC_HPDx_INT_POLARITY;
294 else
295 tmp |= DC_HPDx_INT_POLARITY;
296 WREG32(DC_HPD4_INT_CONTROL, tmp);
297 break;
298 case RADEON_HPD_5:
299 tmp = RREG32(DC_HPD5_INT_CONTROL);
300 if (connected)
301 tmp &= ~DC_HPDx_INT_POLARITY;
302 else
303 tmp |= DC_HPDx_INT_POLARITY;
304 WREG32(DC_HPD5_INT_CONTROL, tmp);
305 break;
306 case RADEON_HPD_6:
307 tmp = RREG32(DC_HPD6_INT_CONTROL);
308 if (connected)
309 tmp &= ~DC_HPDx_INT_POLARITY;
310 else
311 tmp |= DC_HPDx_INT_POLARITY;
312 WREG32(DC_HPD6_INT_CONTROL, tmp);
313 break;
314 default:
315 break;
319 void evergreen_hpd_init(struct radeon_device *rdev)
321 struct drm_device *dev = rdev->ddev;
322 struct drm_connector *connector;
323 u32 tmp = DC_HPDx_CONNECTION_TIMER(0x9c4) |
324 DC_HPDx_RX_INT_TIMER(0xfa) | DC_HPDx_EN;
326 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
327 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
328 switch (radeon_connector->hpd.hpd) {
329 case RADEON_HPD_1:
330 WREG32(DC_HPD1_CONTROL, tmp);
331 rdev->irq.hpd[0] = true;
332 break;
333 case RADEON_HPD_2:
334 WREG32(DC_HPD2_CONTROL, tmp);
335 rdev->irq.hpd[1] = true;
336 break;
337 case RADEON_HPD_3:
338 WREG32(DC_HPD3_CONTROL, tmp);
339 rdev->irq.hpd[2] = true;
340 break;
341 case RADEON_HPD_4:
342 WREG32(DC_HPD4_CONTROL, tmp);
343 rdev->irq.hpd[3] = true;
344 break;
345 case RADEON_HPD_5:
346 WREG32(DC_HPD5_CONTROL, tmp);
347 rdev->irq.hpd[4] = true;
348 break;
349 case RADEON_HPD_6:
350 WREG32(DC_HPD6_CONTROL, tmp);
351 rdev->irq.hpd[5] = true;
352 break;
353 default:
354 break;
357 if (rdev->irq.installed)
358 evergreen_irq_set(rdev);
361 void evergreen_hpd_fini(struct radeon_device *rdev)
363 struct drm_device *dev = rdev->ddev;
364 struct drm_connector *connector;
366 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
367 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
368 switch (radeon_connector->hpd.hpd) {
369 case RADEON_HPD_1:
370 WREG32(DC_HPD1_CONTROL, 0);
371 rdev->irq.hpd[0] = false;
372 break;
373 case RADEON_HPD_2:
374 WREG32(DC_HPD2_CONTROL, 0);
375 rdev->irq.hpd[1] = false;
376 break;
377 case RADEON_HPD_3:
378 WREG32(DC_HPD3_CONTROL, 0);
379 rdev->irq.hpd[2] = false;
380 break;
381 case RADEON_HPD_4:
382 WREG32(DC_HPD4_CONTROL, 0);
383 rdev->irq.hpd[3] = false;
384 break;
385 case RADEON_HPD_5:
386 WREG32(DC_HPD5_CONTROL, 0);
387 rdev->irq.hpd[4] = false;
388 break;
389 case RADEON_HPD_6:
390 WREG32(DC_HPD6_CONTROL, 0);
391 rdev->irq.hpd[5] = false;
392 break;
393 default:
394 break;
399 /* watermark setup */
401 static u32 evergreen_line_buffer_adjust(struct radeon_device *rdev,
402 struct radeon_crtc *radeon_crtc,
403 struct drm_display_mode *mode,
404 struct drm_display_mode *other_mode)
406 u32 tmp;
408 * Line Buffer Setup
409 * There are 3 line buffers, each one shared by 2 display controllers.
410 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
411 * the display controllers. The paritioning is done via one of four
412 * preset allocations specified in bits 2:0:
413 * first display controller
414 * 0 - first half of lb (3840 * 2)
415 * 1 - first 3/4 of lb (5760 * 2)
416 * 2 - whole lb (7680 * 2), other crtc must be disabled
417 * 3 - first 1/4 of lb (1920 * 2)
418 * second display controller
419 * 4 - second half of lb (3840 * 2)
420 * 5 - second 3/4 of lb (5760 * 2)
421 * 6 - whole lb (7680 * 2), other crtc must be disabled
422 * 7 - last 1/4 of lb (1920 * 2)
424 /* this can get tricky if we have two large displays on a paired group
425 * of crtcs. Ideally for multiple large displays we'd assign them to
426 * non-linked crtcs for maximum line buffer allocation.
428 if (radeon_crtc->base.enabled && mode) {
429 if (other_mode)
430 tmp = 0; /* 1/2 */
431 else
432 tmp = 2; /* whole */
433 } else
434 tmp = 0;
436 /* second controller of the pair uses second half of the lb */
437 if (radeon_crtc->crtc_id % 2)
438 tmp += 4;
439 WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset, tmp);
441 if (radeon_crtc->base.enabled && mode) {
442 switch (tmp) {
443 case 0:
444 case 4:
445 default:
446 if (ASIC_IS_DCE5(rdev))
447 return 4096 * 2;
448 else
449 return 3840 * 2;
450 case 1:
451 case 5:
452 if (ASIC_IS_DCE5(rdev))
453 return 6144 * 2;
454 else
455 return 5760 * 2;
456 case 2:
457 case 6:
458 if (ASIC_IS_DCE5(rdev))
459 return 8192 * 2;
460 else
461 return 7680 * 2;
462 case 3:
463 case 7:
464 if (ASIC_IS_DCE5(rdev))
465 return 2048 * 2;
466 else
467 return 1920 * 2;
471 /* controller not enabled, so no lb used */
472 return 0;
475 static u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev)
477 u32 tmp = RREG32(MC_SHARED_CHMAP);
479 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
480 case 0:
481 default:
482 return 1;
483 case 1:
484 return 2;
485 case 2:
486 return 4;
487 case 3:
488 return 8;
492 struct evergreen_wm_params {
493 u32 dram_channels; /* number of dram channels */
494 u32 yclk; /* bandwidth per dram data pin in kHz */
495 u32 sclk; /* engine clock in kHz */
496 u32 disp_clk; /* display clock in kHz */
497 u32 src_width; /* viewport width */
498 u32 active_time; /* active display time in ns */
499 u32 blank_time; /* blank time in ns */
500 bool interlaced; /* mode is interlaced */
501 fixed20_12 vsc; /* vertical scale ratio */
502 u32 num_heads; /* number of active crtcs */
503 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
504 u32 lb_size; /* line buffer allocated to pipe */
505 u32 vtaps; /* vertical scaler taps */
508 static u32 evergreen_dram_bandwidth(struct evergreen_wm_params *wm)
510 /* Calculate DRAM Bandwidth and the part allocated to display. */
511 fixed20_12 dram_efficiency; /* 0.7 */
512 fixed20_12 yclk, dram_channels, bandwidth;
513 fixed20_12 a;
515 a.full = dfixed_const(1000);
516 yclk.full = dfixed_const(wm->yclk);
517 yclk.full = dfixed_div(yclk, a);
518 dram_channels.full = dfixed_const(wm->dram_channels * 4);
519 a.full = dfixed_const(10);
520 dram_efficiency.full = dfixed_const(7);
521 dram_efficiency.full = dfixed_div(dram_efficiency, a);
522 bandwidth.full = dfixed_mul(dram_channels, yclk);
523 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
525 return dfixed_trunc(bandwidth);
528 static u32 evergreen_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
530 /* Calculate DRAM Bandwidth and the part allocated to display. */
531 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
532 fixed20_12 yclk, dram_channels, bandwidth;
533 fixed20_12 a;
535 a.full = dfixed_const(1000);
536 yclk.full = dfixed_const(wm->yclk);
537 yclk.full = dfixed_div(yclk, a);
538 dram_channels.full = dfixed_const(wm->dram_channels * 4);
539 a.full = dfixed_const(10);
540 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
541 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
542 bandwidth.full = dfixed_mul(dram_channels, yclk);
543 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
545 return dfixed_trunc(bandwidth);
548 static u32 evergreen_data_return_bandwidth(struct evergreen_wm_params *wm)
550 /* Calculate the display Data return Bandwidth */
551 fixed20_12 return_efficiency; /* 0.8 */
552 fixed20_12 sclk, bandwidth;
553 fixed20_12 a;
555 a.full = dfixed_const(1000);
556 sclk.full = dfixed_const(wm->sclk);
557 sclk.full = dfixed_div(sclk, a);
558 a.full = dfixed_const(10);
559 return_efficiency.full = dfixed_const(8);
560 return_efficiency.full = dfixed_div(return_efficiency, a);
561 a.full = dfixed_const(32);
562 bandwidth.full = dfixed_mul(a, sclk);
563 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
565 return dfixed_trunc(bandwidth);
568 static u32 evergreen_dmif_request_bandwidth(struct evergreen_wm_params *wm)
570 /* Calculate the DMIF Request Bandwidth */
571 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
572 fixed20_12 disp_clk, bandwidth;
573 fixed20_12 a;
575 a.full = dfixed_const(1000);
576 disp_clk.full = dfixed_const(wm->disp_clk);
577 disp_clk.full = dfixed_div(disp_clk, a);
578 a.full = dfixed_const(10);
579 disp_clk_request_efficiency.full = dfixed_const(8);
580 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
581 a.full = dfixed_const(32);
582 bandwidth.full = dfixed_mul(a, disp_clk);
583 bandwidth.full = dfixed_mul(bandwidth, disp_clk_request_efficiency);
585 return dfixed_trunc(bandwidth);
588 static u32 evergreen_available_bandwidth(struct evergreen_wm_params *wm)
590 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
591 u32 dram_bandwidth = evergreen_dram_bandwidth(wm);
592 u32 data_return_bandwidth = evergreen_data_return_bandwidth(wm);
593 u32 dmif_req_bandwidth = evergreen_dmif_request_bandwidth(wm);
595 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
598 static u32 evergreen_average_bandwidth(struct evergreen_wm_params *wm)
600 /* Calculate the display mode Average Bandwidth
601 * DisplayMode should contain the source and destination dimensions,
602 * timing, etc.
604 fixed20_12 bpp;
605 fixed20_12 line_time;
606 fixed20_12 src_width;
607 fixed20_12 bandwidth;
608 fixed20_12 a;
610 a.full = dfixed_const(1000);
611 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
612 line_time.full = dfixed_div(line_time, a);
613 bpp.full = dfixed_const(wm->bytes_per_pixel);
614 src_width.full = dfixed_const(wm->src_width);
615 bandwidth.full = dfixed_mul(src_width, bpp);
616 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
617 bandwidth.full = dfixed_div(bandwidth, line_time);
619 return dfixed_trunc(bandwidth);
622 static u32 evergreen_latency_watermark(struct evergreen_wm_params *wm)
624 /* First calcualte the latency in ns */
625 u32 mc_latency = 2000; /* 2000 ns. */
626 u32 available_bandwidth = evergreen_available_bandwidth(wm);
627 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
628 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
629 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
630 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
631 (wm->num_heads * cursor_line_pair_return_time);
632 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
633 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
634 fixed20_12 a, b, c;
636 if (wm->num_heads == 0)
637 return 0;
639 a.full = dfixed_const(2);
640 b.full = dfixed_const(1);
641 if ((wm->vsc.full > a.full) ||
642 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
643 (wm->vtaps >= 5) ||
644 ((wm->vsc.full >= a.full) && wm->interlaced))
645 max_src_lines_per_dst_line = 4;
646 else
647 max_src_lines_per_dst_line = 2;
649 a.full = dfixed_const(available_bandwidth);
650 b.full = dfixed_const(wm->num_heads);
651 a.full = dfixed_div(a, b);
653 b.full = dfixed_const(1000);
654 c.full = dfixed_const(wm->disp_clk);
655 b.full = dfixed_div(c, b);
656 c.full = dfixed_const(wm->bytes_per_pixel);
657 b.full = dfixed_mul(b, c);
659 lb_fill_bw = min(dfixed_trunc(a), dfixed_trunc(b));
661 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
662 b.full = dfixed_const(1000);
663 c.full = dfixed_const(lb_fill_bw);
664 b.full = dfixed_div(c, b);
665 a.full = dfixed_div(a, b);
666 line_fill_time = dfixed_trunc(a);
668 if (line_fill_time < wm->active_time)
669 return latency;
670 else
671 return latency + (line_fill_time - wm->active_time);
675 static bool evergreen_average_bandwidth_vs_dram_bandwidth_for_display(struct evergreen_wm_params *wm)
677 if (evergreen_average_bandwidth(wm) <=
678 (evergreen_dram_bandwidth_for_display(wm) / wm->num_heads))
679 return true;
680 else
681 return false;
684 static bool evergreen_average_bandwidth_vs_available_bandwidth(struct evergreen_wm_params *wm)
686 if (evergreen_average_bandwidth(wm) <=
687 (evergreen_available_bandwidth(wm) / wm->num_heads))
688 return true;
689 else
690 return false;
693 static bool evergreen_check_latency_hiding(struct evergreen_wm_params *wm)
695 u32 lb_partitions = wm->lb_size / wm->src_width;
696 u32 line_time = wm->active_time + wm->blank_time;
697 u32 latency_tolerant_lines;
698 u32 latency_hiding;
699 fixed20_12 a;
701 a.full = dfixed_const(1);
702 if (wm->vsc.full > a.full)
703 latency_tolerant_lines = 1;
704 else {
705 if (lb_partitions <= (wm->vtaps + 1))
706 latency_tolerant_lines = 1;
707 else
708 latency_tolerant_lines = 2;
711 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
713 if (evergreen_latency_watermark(wm) <= latency_hiding)
714 return true;
715 else
716 return false;
719 static void evergreen_program_watermarks(struct radeon_device *rdev,
720 struct radeon_crtc *radeon_crtc,
721 u32 lb_size, u32 num_heads)
723 struct drm_display_mode *mode = &radeon_crtc->base.mode;
724 struct evergreen_wm_params wm;
725 u32 pixel_period;
726 u32 line_time = 0;
727 u32 latency_watermark_a = 0, latency_watermark_b = 0;
728 u32 priority_a_mark = 0, priority_b_mark = 0;
729 u32 priority_a_cnt = PRIORITY_OFF;
730 u32 priority_b_cnt = PRIORITY_OFF;
731 u32 pipe_offset = radeon_crtc->crtc_id * 16;
732 u32 tmp, arb_control3;
733 fixed20_12 a, b, c;
735 if (radeon_crtc->base.enabled && num_heads && mode) {
736 pixel_period = 1000000 / (u32)mode->clock;
737 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
738 priority_a_cnt = 0;
739 priority_b_cnt = 0;
741 wm.yclk = rdev->pm.current_mclk * 10;
742 wm.sclk = rdev->pm.current_sclk * 10;
743 wm.disp_clk = mode->clock;
744 wm.src_width = mode->crtc_hdisplay;
745 wm.active_time = mode->crtc_hdisplay * pixel_period;
746 wm.blank_time = line_time - wm.active_time;
747 wm.interlaced = false;
748 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
749 wm.interlaced = true;
750 wm.vsc = radeon_crtc->vsc;
751 wm.vtaps = 1;
752 if (radeon_crtc->rmx_type != RMX_OFF)
753 wm.vtaps = 2;
754 wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
755 wm.lb_size = lb_size;
756 wm.dram_channels = evergreen_get_number_of_dram_channels(rdev);
757 wm.num_heads = num_heads;
759 /* set for high clocks */
760 latency_watermark_a = min(evergreen_latency_watermark(&wm), (u32)65535);
761 /* set for low clocks */
762 /* wm.yclk = low clk; wm.sclk = low clk */
763 latency_watermark_b = min(evergreen_latency_watermark(&wm), (u32)65535);
765 /* possibly force display priority to high */
766 /* should really do this at mode validation time... */
767 if (!evergreen_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
768 !evergreen_average_bandwidth_vs_available_bandwidth(&wm) ||
769 !evergreen_check_latency_hiding(&wm) ||
770 (rdev->disp_priority == 2)) {
771 DRM_DEBUG_KMS("force priority to high\n");
772 priority_a_cnt |= PRIORITY_ALWAYS_ON;
773 priority_b_cnt |= PRIORITY_ALWAYS_ON;
776 a.full = dfixed_const(1000);
777 b.full = dfixed_const(mode->clock);
778 b.full = dfixed_div(b, a);
779 c.full = dfixed_const(latency_watermark_a);
780 c.full = dfixed_mul(c, b);
781 c.full = dfixed_mul(c, radeon_crtc->hsc);
782 c.full = dfixed_div(c, a);
783 a.full = dfixed_const(16);
784 c.full = dfixed_div(c, a);
785 priority_a_mark = dfixed_trunc(c);
786 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
788 a.full = dfixed_const(1000);
789 b.full = dfixed_const(mode->clock);
790 b.full = dfixed_div(b, a);
791 c.full = dfixed_const(latency_watermark_b);
792 c.full = dfixed_mul(c, b);
793 c.full = dfixed_mul(c, radeon_crtc->hsc);
794 c.full = dfixed_div(c, a);
795 a.full = dfixed_const(16);
796 c.full = dfixed_div(c, a);
797 priority_b_mark = dfixed_trunc(c);
798 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
801 /* select wm A */
802 arb_control3 = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
803 tmp = arb_control3;
804 tmp &= ~LATENCY_WATERMARK_MASK(3);
805 tmp |= LATENCY_WATERMARK_MASK(1);
806 WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
807 WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
808 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
809 LATENCY_HIGH_WATERMARK(line_time)));
810 /* select wm B */
811 tmp = RREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset);
812 tmp &= ~LATENCY_WATERMARK_MASK(3);
813 tmp |= LATENCY_WATERMARK_MASK(2);
814 WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, tmp);
815 WREG32(PIPE0_LATENCY_CONTROL + pipe_offset,
816 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
817 LATENCY_HIGH_WATERMARK(line_time)));
818 /* restore original selection */
819 WREG32(PIPE0_ARBITRATION_CONTROL3 + pipe_offset, arb_control3);
821 /* write the priority marks */
822 WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
823 WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
827 void evergreen_bandwidth_update(struct radeon_device *rdev)
829 struct drm_display_mode *mode0 = NULL;
830 struct drm_display_mode *mode1 = NULL;
831 u32 num_heads = 0, lb_size;
832 int i;
834 radeon_update_display_priority(rdev);
836 for (i = 0; i < rdev->num_crtc; i++) {
837 if (rdev->mode_info.crtcs[i]->base.enabled)
838 num_heads++;
840 for (i = 0; i < rdev->num_crtc; i += 2) {
841 mode0 = &rdev->mode_info.crtcs[i]->base.mode;
842 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
843 lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
844 evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
845 lb_size = evergreen_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
846 evergreen_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
850 int evergreen_mc_wait_for_idle(struct radeon_device *rdev)
852 unsigned i;
853 u32 tmp;
855 for (i = 0; i < rdev->usec_timeout; i++) {
856 /* read MC_STATUS */
857 tmp = RREG32(SRBM_STATUS) & 0x1F00;
858 if (!tmp)
859 return 0;
860 udelay(1);
862 return -1;
866 * GART
868 void evergreen_pcie_gart_tlb_flush(struct radeon_device *rdev)
870 unsigned i;
871 u32 tmp;
873 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
875 WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
876 for (i = 0; i < rdev->usec_timeout; i++) {
877 /* read MC_STATUS */
878 tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
879 tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
880 if (tmp == 2) {
881 printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
882 return;
884 if (tmp) {
885 return;
887 udelay(1);
891 int evergreen_pcie_gart_enable(struct radeon_device *rdev)
893 u32 tmp;
894 int r;
896 if (rdev->gart.table.vram.robj == NULL) {
897 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
898 return -EINVAL;
900 r = radeon_gart_table_vram_pin(rdev);
901 if (r)
902 return r;
903 radeon_gart_restore(rdev);
904 /* Setup L2 cache */
905 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
906 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
907 EFFECTIVE_L2_QUEUE_SIZE(7));
908 WREG32(VM_L2_CNTL2, 0);
909 WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
910 /* Setup TLB control */
911 tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
912 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
913 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
914 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
915 if (rdev->flags & RADEON_IS_IGP) {
916 WREG32(FUS_MC_VM_MD_L1_TLB0_CNTL, tmp);
917 WREG32(FUS_MC_VM_MD_L1_TLB1_CNTL, tmp);
918 WREG32(FUS_MC_VM_MD_L1_TLB2_CNTL, tmp);
919 } else {
920 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
921 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
922 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
924 WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
925 WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
926 WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
927 WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
928 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
929 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
930 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
931 WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
932 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
933 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
934 (u32)(rdev->dummy_page.addr >> 12));
935 WREG32(VM_CONTEXT1_CNTL, 0);
937 evergreen_pcie_gart_tlb_flush(rdev);
938 rdev->gart.ready = true;
939 return 0;
942 void evergreen_pcie_gart_disable(struct radeon_device *rdev)
944 u32 tmp;
945 int r;
947 /* Disable all tables */
948 WREG32(VM_CONTEXT0_CNTL, 0);
949 WREG32(VM_CONTEXT1_CNTL, 0);
951 /* Setup L2 cache */
952 WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
953 EFFECTIVE_L2_QUEUE_SIZE(7));
954 WREG32(VM_L2_CNTL2, 0);
955 WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
956 /* Setup TLB control */
957 tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
958 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
959 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
960 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
961 WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
962 WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
963 WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
964 WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
965 if (rdev->gart.table.vram.robj) {
966 r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
967 if (likely(r == 0)) {
968 radeon_bo_kunmap(rdev->gart.table.vram.robj);
969 radeon_bo_unpin(rdev->gart.table.vram.robj);
970 radeon_bo_unreserve(rdev->gart.table.vram.robj);
975 void evergreen_pcie_gart_fini(struct radeon_device *rdev)
977 evergreen_pcie_gart_disable(rdev);
978 radeon_gart_table_vram_free(rdev);
979 radeon_gart_fini(rdev);
983 void evergreen_agp_enable(struct radeon_device *rdev)
985 u32 tmp;
987 /* Setup L2 cache */
988 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
989 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
990 EFFECTIVE_L2_QUEUE_SIZE(7));
991 WREG32(VM_L2_CNTL2, 0);
992 WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2));
993 /* Setup TLB control */
994 tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
995 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
996 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU |
997 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5);
998 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp);
999 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp);
1000 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp);
1001 WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp);
1002 WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp);
1003 WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp);
1004 WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp);
1005 WREG32(VM_CONTEXT0_CNTL, 0);
1006 WREG32(VM_CONTEXT1_CNTL, 0);
1009 void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save)
1011 save->vga_control[0] = RREG32(D1VGA_CONTROL);
1012 save->vga_control[1] = RREG32(D2VGA_CONTROL);
1013 save->vga_render_control = RREG32(VGA_RENDER_CONTROL);
1014 save->vga_hdp_control = RREG32(VGA_HDP_CONTROL);
1015 save->crtc_control[0] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET);
1016 save->crtc_control[1] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET);
1017 if (rdev->num_crtc >= 4) {
1018 save->vga_control[2] = RREG32(EVERGREEN_D3VGA_CONTROL);
1019 save->vga_control[3] = RREG32(EVERGREEN_D4VGA_CONTROL);
1020 save->crtc_control[2] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET);
1021 save->crtc_control[3] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET);
1023 if (rdev->num_crtc >= 6) {
1024 save->vga_control[4] = RREG32(EVERGREEN_D5VGA_CONTROL);
1025 save->vga_control[5] = RREG32(EVERGREEN_D6VGA_CONTROL);
1026 save->crtc_control[4] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET);
1027 save->crtc_control[5] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET);
1030 /* Stop all video */
1031 WREG32(VGA_RENDER_CONTROL, 0);
1032 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 1);
1033 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 1);
1034 if (rdev->num_crtc >= 4) {
1035 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 1);
1036 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 1);
1038 if (rdev->num_crtc >= 6) {
1039 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 1);
1040 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 1);
1042 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
1043 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
1044 if (rdev->num_crtc >= 4) {
1045 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
1046 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
1048 if (rdev->num_crtc >= 6) {
1049 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
1050 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
1052 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
1053 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
1054 if (rdev->num_crtc >= 4) {
1055 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
1056 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
1058 if (rdev->num_crtc >= 6) {
1059 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
1060 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
1063 WREG32(D1VGA_CONTROL, 0);
1064 WREG32(D2VGA_CONTROL, 0);
1065 if (rdev->num_crtc >= 4) {
1066 WREG32(EVERGREEN_D3VGA_CONTROL, 0);
1067 WREG32(EVERGREEN_D4VGA_CONTROL, 0);
1069 if (rdev->num_crtc >= 6) {
1070 WREG32(EVERGREEN_D5VGA_CONTROL, 0);
1071 WREG32(EVERGREEN_D6VGA_CONTROL, 0);
1075 void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save)
1077 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC0_REGISTER_OFFSET,
1078 upper_32_bits(rdev->mc.vram_start));
1079 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC0_REGISTER_OFFSET,
1080 upper_32_bits(rdev->mc.vram_start));
1081 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC0_REGISTER_OFFSET,
1082 (u32)rdev->mc.vram_start);
1083 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC0_REGISTER_OFFSET,
1084 (u32)rdev->mc.vram_start);
1086 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC1_REGISTER_OFFSET,
1087 upper_32_bits(rdev->mc.vram_start));
1088 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC1_REGISTER_OFFSET,
1089 upper_32_bits(rdev->mc.vram_start));
1090 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC1_REGISTER_OFFSET,
1091 (u32)rdev->mc.vram_start);
1092 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC1_REGISTER_OFFSET,
1093 (u32)rdev->mc.vram_start);
1095 if (rdev->num_crtc >= 4) {
1096 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC2_REGISTER_OFFSET,
1097 upper_32_bits(rdev->mc.vram_start));
1098 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC2_REGISTER_OFFSET,
1099 upper_32_bits(rdev->mc.vram_start));
1100 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC2_REGISTER_OFFSET,
1101 (u32)rdev->mc.vram_start);
1102 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC2_REGISTER_OFFSET,
1103 (u32)rdev->mc.vram_start);
1105 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC3_REGISTER_OFFSET,
1106 upper_32_bits(rdev->mc.vram_start));
1107 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC3_REGISTER_OFFSET,
1108 upper_32_bits(rdev->mc.vram_start));
1109 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC3_REGISTER_OFFSET,
1110 (u32)rdev->mc.vram_start);
1111 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC3_REGISTER_OFFSET,
1112 (u32)rdev->mc.vram_start);
1114 if (rdev->num_crtc >= 6) {
1115 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC4_REGISTER_OFFSET,
1116 upper_32_bits(rdev->mc.vram_start));
1117 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC4_REGISTER_OFFSET,
1118 upper_32_bits(rdev->mc.vram_start));
1119 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC4_REGISTER_OFFSET,
1120 (u32)rdev->mc.vram_start);
1121 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC4_REGISTER_OFFSET,
1122 (u32)rdev->mc.vram_start);
1124 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC5_REGISTER_OFFSET,
1125 upper_32_bits(rdev->mc.vram_start));
1126 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC5_REGISTER_OFFSET,
1127 upper_32_bits(rdev->mc.vram_start));
1128 WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC5_REGISTER_OFFSET,
1129 (u32)rdev->mc.vram_start);
1130 WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC5_REGISTER_OFFSET,
1131 (u32)rdev->mc.vram_start);
1134 WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS_HIGH, upper_32_bits(rdev->mc.vram_start));
1135 WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS, (u32)rdev->mc.vram_start);
1136 /* Unlock host access */
1137 WREG32(VGA_HDP_CONTROL, save->vga_hdp_control);
1138 mdelay(1);
1139 /* Restore video state */
1140 WREG32(D1VGA_CONTROL, save->vga_control[0]);
1141 WREG32(D2VGA_CONTROL, save->vga_control[1]);
1142 if (rdev->num_crtc >= 4) {
1143 WREG32(EVERGREEN_D3VGA_CONTROL, save->vga_control[2]);
1144 WREG32(EVERGREEN_D4VGA_CONTROL, save->vga_control[3]);
1146 if (rdev->num_crtc >= 6) {
1147 WREG32(EVERGREEN_D5VGA_CONTROL, save->vga_control[4]);
1148 WREG32(EVERGREEN_D6VGA_CONTROL, save->vga_control[5]);
1150 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 1);
1151 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 1);
1152 if (rdev->num_crtc >= 4) {
1153 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 1);
1154 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 1);
1156 if (rdev->num_crtc >= 6) {
1157 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 1);
1158 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 1);
1160 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, save->crtc_control[0]);
1161 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, save->crtc_control[1]);
1162 if (rdev->num_crtc >= 4) {
1163 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, save->crtc_control[2]);
1164 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, save->crtc_control[3]);
1166 if (rdev->num_crtc >= 6) {
1167 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, save->crtc_control[4]);
1168 WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, save->crtc_control[5]);
1170 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
1171 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
1172 if (rdev->num_crtc >= 4) {
1173 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
1174 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
1176 if (rdev->num_crtc >= 6) {
1177 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
1178 WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
1180 WREG32(VGA_RENDER_CONTROL, save->vga_render_control);
1183 void evergreen_mc_program(struct radeon_device *rdev)
1185 struct evergreen_mc_save save;
1186 u32 tmp;
1187 int i, j;
1189 /* Initialize HDP */
1190 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1191 WREG32((0x2c14 + j), 0x00000000);
1192 WREG32((0x2c18 + j), 0x00000000);
1193 WREG32((0x2c1c + j), 0x00000000);
1194 WREG32((0x2c20 + j), 0x00000000);
1195 WREG32((0x2c24 + j), 0x00000000);
1197 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
1199 evergreen_mc_stop(rdev, &save);
1200 if (evergreen_mc_wait_for_idle(rdev)) {
1201 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1203 /* Lockout access through VGA aperture*/
1204 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
1205 /* Update configuration */
1206 if (rdev->flags & RADEON_IS_AGP) {
1207 if (rdev->mc.vram_start < rdev->mc.gtt_start) {
1208 /* VRAM before AGP */
1209 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1210 rdev->mc.vram_start >> 12);
1211 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1212 rdev->mc.gtt_end >> 12);
1213 } else {
1214 /* VRAM after AGP */
1215 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1216 rdev->mc.gtt_start >> 12);
1217 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1218 rdev->mc.vram_end >> 12);
1220 } else {
1221 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
1222 rdev->mc.vram_start >> 12);
1223 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
1224 rdev->mc.vram_end >> 12);
1226 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
1227 if (rdev->flags & RADEON_IS_IGP) {
1228 tmp = RREG32(MC_FUS_VM_FB_OFFSET) & 0x000FFFFF;
1229 tmp |= ((rdev->mc.vram_end >> 20) & 0xF) << 24;
1230 tmp |= ((rdev->mc.vram_start >> 20) & 0xF) << 20;
1231 WREG32(MC_FUS_VM_FB_OFFSET, tmp);
1233 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
1234 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
1235 WREG32(MC_VM_FB_LOCATION, tmp);
1236 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
1237 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
1238 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
1239 if (rdev->flags & RADEON_IS_AGP) {
1240 WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 16);
1241 WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 16);
1242 WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
1243 } else {
1244 WREG32(MC_VM_AGP_BASE, 0);
1245 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
1246 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
1248 if (evergreen_mc_wait_for_idle(rdev)) {
1249 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
1251 evergreen_mc_resume(rdev, &save);
1252 /* we need to own VRAM, so turn off the VGA renderer here
1253 * to stop it overwriting our objects */
1254 rv515_vga_render_disable(rdev);
1258 * CP.
1260 void evergreen_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1262 /* set to DX10/11 mode */
1263 radeon_ring_write(rdev, PACKET3(PACKET3_MODE_CONTROL, 0));
1264 radeon_ring_write(rdev, 1);
1265 /* FIXME: implement */
1266 radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
1267 radeon_ring_write(rdev,
1268 #ifdef __BIG_ENDIAN
1269 (2 << 0) |
1270 #endif
1271 (ib->gpu_addr & 0xFFFFFFFC));
1272 radeon_ring_write(rdev, upper_32_bits(ib->gpu_addr) & 0xFF);
1273 radeon_ring_write(rdev, ib->length_dw);
1277 static int evergreen_cp_load_microcode(struct radeon_device *rdev)
1279 const __be32 *fw_data;
1280 int i;
1282 if (!rdev->me_fw || !rdev->pfp_fw)
1283 return -EINVAL;
1285 r700_cp_stop(rdev);
1286 WREG32(CP_RB_CNTL,
1287 #ifdef __BIG_ENDIAN
1288 BUF_SWAP_32BIT |
1289 #endif
1290 RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
1292 fw_data = (const __be32 *)rdev->pfp_fw->data;
1293 WREG32(CP_PFP_UCODE_ADDR, 0);
1294 for (i = 0; i < EVERGREEN_PFP_UCODE_SIZE; i++)
1295 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1296 WREG32(CP_PFP_UCODE_ADDR, 0);
1298 fw_data = (const __be32 *)rdev->me_fw->data;
1299 WREG32(CP_ME_RAM_WADDR, 0);
1300 for (i = 0; i < EVERGREEN_PM4_UCODE_SIZE; i++)
1301 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1303 WREG32(CP_PFP_UCODE_ADDR, 0);
1304 WREG32(CP_ME_RAM_WADDR, 0);
1305 WREG32(CP_ME_RAM_RADDR, 0);
1306 return 0;
1309 static int evergreen_cp_start(struct radeon_device *rdev)
1311 int r, i;
1312 uint32_t cp_me;
1314 r = radeon_ring_lock(rdev, 7);
1315 if (r) {
1316 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1317 return r;
1319 radeon_ring_write(rdev, PACKET3(PACKET3_ME_INITIALIZE, 5));
1320 radeon_ring_write(rdev, 0x1);
1321 radeon_ring_write(rdev, 0x0);
1322 radeon_ring_write(rdev, rdev->config.evergreen.max_hw_contexts - 1);
1323 radeon_ring_write(rdev, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1324 radeon_ring_write(rdev, 0);
1325 radeon_ring_write(rdev, 0);
1326 radeon_ring_unlock_commit(rdev);
1328 cp_me = 0xff;
1329 WREG32(CP_ME_CNTL, cp_me);
1331 r = radeon_ring_lock(rdev, evergreen_default_size + 19);
1332 if (r) {
1333 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1334 return r;
1337 /* setup clear context state */
1338 radeon_ring_write(rdev, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1339 radeon_ring_write(rdev, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1341 for (i = 0; i < evergreen_default_size; i++)
1342 radeon_ring_write(rdev, evergreen_default_state[i]);
1344 radeon_ring_write(rdev, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1345 radeon_ring_write(rdev, PACKET3_PREAMBLE_END_CLEAR_STATE);
1347 /* set clear context state */
1348 radeon_ring_write(rdev, PACKET3(PACKET3_CLEAR_STATE, 0));
1349 radeon_ring_write(rdev, 0);
1351 /* SQ_VTX_BASE_VTX_LOC */
1352 radeon_ring_write(rdev, 0xc0026f00);
1353 radeon_ring_write(rdev, 0x00000000);
1354 radeon_ring_write(rdev, 0x00000000);
1355 radeon_ring_write(rdev, 0x00000000);
1357 /* Clear consts */
1358 radeon_ring_write(rdev, 0xc0036f00);
1359 radeon_ring_write(rdev, 0x00000bc4);
1360 radeon_ring_write(rdev, 0xffffffff);
1361 radeon_ring_write(rdev, 0xffffffff);
1362 radeon_ring_write(rdev, 0xffffffff);
1364 radeon_ring_write(rdev, 0xc0026900);
1365 radeon_ring_write(rdev, 0x00000316);
1366 radeon_ring_write(rdev, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
1367 radeon_ring_write(rdev, 0x00000010); /* */
1369 radeon_ring_unlock_commit(rdev);
1371 return 0;
1374 int evergreen_cp_resume(struct radeon_device *rdev)
1376 u32 tmp;
1377 u32 rb_bufsz;
1378 int r;
1380 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
1381 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
1382 SOFT_RESET_PA |
1383 SOFT_RESET_SH |
1384 SOFT_RESET_VGT |
1385 SOFT_RESET_SPI |
1386 SOFT_RESET_SX));
1387 RREG32(GRBM_SOFT_RESET);
1388 mdelay(15);
1389 WREG32(GRBM_SOFT_RESET, 0);
1390 RREG32(GRBM_SOFT_RESET);
1392 /* Set ring buffer size */
1393 rb_bufsz = drm_order(rdev->cp.ring_size / 8);
1394 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1395 #ifdef __BIG_ENDIAN
1396 tmp |= BUF_SWAP_32BIT;
1397 #endif
1398 WREG32(CP_RB_CNTL, tmp);
1399 WREG32(CP_SEM_WAIT_TIMER, 0x4);
1401 /* Set the write pointer delay */
1402 WREG32(CP_RB_WPTR_DELAY, 0);
1404 /* Initialize the ring buffer's read and write pointers */
1405 WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
1406 WREG32(CP_RB_RPTR_WR, 0);
1407 rdev->cp.wptr = 0;
1408 WREG32(CP_RB_WPTR, rdev->cp.wptr);
1410 /* set the wb address wether it's enabled or not */
1411 WREG32(CP_RB_RPTR_ADDR,
1412 ((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC));
1413 WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
1414 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
1416 if (rdev->wb.enabled)
1417 WREG32(SCRATCH_UMSK, 0xff);
1418 else {
1419 tmp |= RB_NO_UPDATE;
1420 WREG32(SCRATCH_UMSK, 0);
1423 mdelay(1);
1424 WREG32(CP_RB_CNTL, tmp);
1426 WREG32(CP_RB_BASE, rdev->cp.gpu_addr >> 8);
1427 WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
1429 rdev->cp.rptr = RREG32(CP_RB_RPTR);
1431 evergreen_cp_start(rdev);
1432 rdev->cp.ready = true;
1433 r = radeon_ring_test(rdev);
1434 if (r) {
1435 rdev->cp.ready = false;
1436 return r;
1438 return 0;
1442 * Core functions
1444 static u32 evergreen_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
1445 u32 num_tile_pipes,
1446 u32 num_backends,
1447 u32 backend_disable_mask)
1449 u32 backend_map = 0;
1450 u32 enabled_backends_mask = 0;
1451 u32 enabled_backends_count = 0;
1452 u32 cur_pipe;
1453 u32 swizzle_pipe[EVERGREEN_MAX_PIPES];
1454 u32 cur_backend = 0;
1455 u32 i;
1456 bool force_no_swizzle;
1458 if (num_tile_pipes > EVERGREEN_MAX_PIPES)
1459 num_tile_pipes = EVERGREEN_MAX_PIPES;
1460 if (num_tile_pipes < 1)
1461 num_tile_pipes = 1;
1462 if (num_backends > EVERGREEN_MAX_BACKENDS)
1463 num_backends = EVERGREEN_MAX_BACKENDS;
1464 if (num_backends < 1)
1465 num_backends = 1;
1467 for (i = 0; i < EVERGREEN_MAX_BACKENDS; ++i) {
1468 if (((backend_disable_mask >> i) & 1) == 0) {
1469 enabled_backends_mask |= (1 << i);
1470 ++enabled_backends_count;
1472 if (enabled_backends_count == num_backends)
1473 break;
1476 if (enabled_backends_count == 0) {
1477 enabled_backends_mask = 1;
1478 enabled_backends_count = 1;
1481 if (enabled_backends_count != num_backends)
1482 num_backends = enabled_backends_count;
1484 memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * EVERGREEN_MAX_PIPES);
1485 switch (rdev->family) {
1486 case CHIP_CEDAR:
1487 case CHIP_REDWOOD:
1488 case CHIP_PALM:
1489 case CHIP_SUMO:
1490 case CHIP_SUMO2:
1491 case CHIP_TURKS:
1492 case CHIP_CAICOS:
1493 force_no_swizzle = false;
1494 break;
1495 case CHIP_CYPRESS:
1496 case CHIP_HEMLOCK:
1497 case CHIP_JUNIPER:
1498 case CHIP_BARTS:
1499 default:
1500 force_no_swizzle = true;
1501 break;
1503 if (force_no_swizzle) {
1504 bool last_backend_enabled = false;
1506 force_no_swizzle = false;
1507 for (i = 0; i < EVERGREEN_MAX_BACKENDS; ++i) {
1508 if (((enabled_backends_mask >> i) & 1) == 1) {
1509 if (last_backend_enabled)
1510 force_no_swizzle = true;
1511 last_backend_enabled = true;
1512 } else
1513 last_backend_enabled = false;
1517 switch (num_tile_pipes) {
1518 case 1:
1519 case 3:
1520 case 5:
1521 case 7:
1522 DRM_ERROR("odd number of pipes!\n");
1523 break;
1524 case 2:
1525 swizzle_pipe[0] = 0;
1526 swizzle_pipe[1] = 1;
1527 break;
1528 case 4:
1529 if (force_no_swizzle) {
1530 swizzle_pipe[0] = 0;
1531 swizzle_pipe[1] = 1;
1532 swizzle_pipe[2] = 2;
1533 swizzle_pipe[3] = 3;
1534 } else {
1535 swizzle_pipe[0] = 0;
1536 swizzle_pipe[1] = 2;
1537 swizzle_pipe[2] = 1;
1538 swizzle_pipe[3] = 3;
1540 break;
1541 case 6:
1542 if (force_no_swizzle) {
1543 swizzle_pipe[0] = 0;
1544 swizzle_pipe[1] = 1;
1545 swizzle_pipe[2] = 2;
1546 swizzle_pipe[3] = 3;
1547 swizzle_pipe[4] = 4;
1548 swizzle_pipe[5] = 5;
1549 } else {
1550 swizzle_pipe[0] = 0;
1551 swizzle_pipe[1] = 2;
1552 swizzle_pipe[2] = 4;
1553 swizzle_pipe[3] = 1;
1554 swizzle_pipe[4] = 3;
1555 swizzle_pipe[5] = 5;
1557 break;
1558 case 8:
1559 if (force_no_swizzle) {
1560 swizzle_pipe[0] = 0;
1561 swizzle_pipe[1] = 1;
1562 swizzle_pipe[2] = 2;
1563 swizzle_pipe[3] = 3;
1564 swizzle_pipe[4] = 4;
1565 swizzle_pipe[5] = 5;
1566 swizzle_pipe[6] = 6;
1567 swizzle_pipe[7] = 7;
1568 } else {
1569 swizzle_pipe[0] = 0;
1570 swizzle_pipe[1] = 2;
1571 swizzle_pipe[2] = 4;
1572 swizzle_pipe[3] = 6;
1573 swizzle_pipe[4] = 1;
1574 swizzle_pipe[5] = 3;
1575 swizzle_pipe[6] = 5;
1576 swizzle_pipe[7] = 7;
1578 break;
1581 for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
1582 while (((1 << cur_backend) & enabled_backends_mask) == 0)
1583 cur_backend = (cur_backend + 1) % EVERGREEN_MAX_BACKENDS;
1585 backend_map |= (((cur_backend & 0xf) << (swizzle_pipe[cur_pipe] * 4)));
1587 cur_backend = (cur_backend + 1) % EVERGREEN_MAX_BACKENDS;
1590 return backend_map;
1593 static void evergreen_gpu_init(struct radeon_device *rdev)
1595 u32 cc_rb_backend_disable = 0;
1596 u32 cc_gc_shader_pipe_config;
1597 u32 gb_addr_config = 0;
1598 u32 mc_shared_chmap, mc_arb_ramcfg;
1599 u32 gb_backend_map;
1600 u32 grbm_gfx_index;
1601 u32 sx_debug_1;
1602 u32 smx_dc_ctl0;
1603 u32 sq_config;
1604 u32 sq_lds_resource_mgmt;
1605 u32 sq_gpr_resource_mgmt_1;
1606 u32 sq_gpr_resource_mgmt_2;
1607 u32 sq_gpr_resource_mgmt_3;
1608 u32 sq_thread_resource_mgmt;
1609 u32 sq_thread_resource_mgmt_2;
1610 u32 sq_stack_resource_mgmt_1;
1611 u32 sq_stack_resource_mgmt_2;
1612 u32 sq_stack_resource_mgmt_3;
1613 u32 vgt_cache_invalidation;
1614 u32 hdp_host_path_cntl, tmp;
1615 int i, j, num_shader_engines, ps_thread_count;
1617 switch (rdev->family) {
1618 case CHIP_CYPRESS:
1619 case CHIP_HEMLOCK:
1620 rdev->config.evergreen.num_ses = 2;
1621 rdev->config.evergreen.max_pipes = 4;
1622 rdev->config.evergreen.max_tile_pipes = 8;
1623 rdev->config.evergreen.max_simds = 10;
1624 rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
1625 rdev->config.evergreen.max_gprs = 256;
1626 rdev->config.evergreen.max_threads = 248;
1627 rdev->config.evergreen.max_gs_threads = 32;
1628 rdev->config.evergreen.max_stack_entries = 512;
1629 rdev->config.evergreen.sx_num_of_sets = 4;
1630 rdev->config.evergreen.sx_max_export_size = 256;
1631 rdev->config.evergreen.sx_max_export_pos_size = 64;
1632 rdev->config.evergreen.sx_max_export_smx_size = 192;
1633 rdev->config.evergreen.max_hw_contexts = 8;
1634 rdev->config.evergreen.sq_num_cf_insts = 2;
1636 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1637 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1638 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1639 break;
1640 case CHIP_JUNIPER:
1641 rdev->config.evergreen.num_ses = 1;
1642 rdev->config.evergreen.max_pipes = 4;
1643 rdev->config.evergreen.max_tile_pipes = 4;
1644 rdev->config.evergreen.max_simds = 10;
1645 rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
1646 rdev->config.evergreen.max_gprs = 256;
1647 rdev->config.evergreen.max_threads = 248;
1648 rdev->config.evergreen.max_gs_threads = 32;
1649 rdev->config.evergreen.max_stack_entries = 512;
1650 rdev->config.evergreen.sx_num_of_sets = 4;
1651 rdev->config.evergreen.sx_max_export_size = 256;
1652 rdev->config.evergreen.sx_max_export_pos_size = 64;
1653 rdev->config.evergreen.sx_max_export_smx_size = 192;
1654 rdev->config.evergreen.max_hw_contexts = 8;
1655 rdev->config.evergreen.sq_num_cf_insts = 2;
1657 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1658 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1659 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1660 break;
1661 case CHIP_REDWOOD:
1662 rdev->config.evergreen.num_ses = 1;
1663 rdev->config.evergreen.max_pipes = 4;
1664 rdev->config.evergreen.max_tile_pipes = 4;
1665 rdev->config.evergreen.max_simds = 5;
1666 rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
1667 rdev->config.evergreen.max_gprs = 256;
1668 rdev->config.evergreen.max_threads = 248;
1669 rdev->config.evergreen.max_gs_threads = 32;
1670 rdev->config.evergreen.max_stack_entries = 256;
1671 rdev->config.evergreen.sx_num_of_sets = 4;
1672 rdev->config.evergreen.sx_max_export_size = 256;
1673 rdev->config.evergreen.sx_max_export_pos_size = 64;
1674 rdev->config.evergreen.sx_max_export_smx_size = 192;
1675 rdev->config.evergreen.max_hw_contexts = 8;
1676 rdev->config.evergreen.sq_num_cf_insts = 2;
1678 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1679 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1680 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1681 break;
1682 case CHIP_CEDAR:
1683 default:
1684 rdev->config.evergreen.num_ses = 1;
1685 rdev->config.evergreen.max_pipes = 2;
1686 rdev->config.evergreen.max_tile_pipes = 2;
1687 rdev->config.evergreen.max_simds = 2;
1688 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1689 rdev->config.evergreen.max_gprs = 256;
1690 rdev->config.evergreen.max_threads = 192;
1691 rdev->config.evergreen.max_gs_threads = 16;
1692 rdev->config.evergreen.max_stack_entries = 256;
1693 rdev->config.evergreen.sx_num_of_sets = 4;
1694 rdev->config.evergreen.sx_max_export_size = 128;
1695 rdev->config.evergreen.sx_max_export_pos_size = 32;
1696 rdev->config.evergreen.sx_max_export_smx_size = 96;
1697 rdev->config.evergreen.max_hw_contexts = 4;
1698 rdev->config.evergreen.sq_num_cf_insts = 1;
1700 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1701 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1702 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1703 break;
1704 case CHIP_PALM:
1705 rdev->config.evergreen.num_ses = 1;
1706 rdev->config.evergreen.max_pipes = 2;
1707 rdev->config.evergreen.max_tile_pipes = 2;
1708 rdev->config.evergreen.max_simds = 2;
1709 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1710 rdev->config.evergreen.max_gprs = 256;
1711 rdev->config.evergreen.max_threads = 192;
1712 rdev->config.evergreen.max_gs_threads = 16;
1713 rdev->config.evergreen.max_stack_entries = 256;
1714 rdev->config.evergreen.sx_num_of_sets = 4;
1715 rdev->config.evergreen.sx_max_export_size = 128;
1716 rdev->config.evergreen.sx_max_export_pos_size = 32;
1717 rdev->config.evergreen.sx_max_export_smx_size = 96;
1718 rdev->config.evergreen.max_hw_contexts = 4;
1719 rdev->config.evergreen.sq_num_cf_insts = 1;
1721 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1722 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1723 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1724 break;
1725 case CHIP_SUMO:
1726 rdev->config.evergreen.num_ses = 1;
1727 rdev->config.evergreen.max_pipes = 4;
1728 rdev->config.evergreen.max_tile_pipes = 2;
1729 if (rdev->pdev->device == 0x9648)
1730 rdev->config.evergreen.max_simds = 3;
1731 else if ((rdev->pdev->device == 0x9647) ||
1732 (rdev->pdev->device == 0x964a))
1733 rdev->config.evergreen.max_simds = 4;
1734 else
1735 rdev->config.evergreen.max_simds = 5;
1736 rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
1737 rdev->config.evergreen.max_gprs = 256;
1738 rdev->config.evergreen.max_threads = 248;
1739 rdev->config.evergreen.max_gs_threads = 32;
1740 rdev->config.evergreen.max_stack_entries = 256;
1741 rdev->config.evergreen.sx_num_of_sets = 4;
1742 rdev->config.evergreen.sx_max_export_size = 256;
1743 rdev->config.evergreen.sx_max_export_pos_size = 64;
1744 rdev->config.evergreen.sx_max_export_smx_size = 192;
1745 rdev->config.evergreen.max_hw_contexts = 8;
1746 rdev->config.evergreen.sq_num_cf_insts = 2;
1748 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1749 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1750 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1751 break;
1752 case CHIP_SUMO2:
1753 rdev->config.evergreen.num_ses = 1;
1754 rdev->config.evergreen.max_pipes = 4;
1755 rdev->config.evergreen.max_tile_pipes = 4;
1756 rdev->config.evergreen.max_simds = 2;
1757 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1758 rdev->config.evergreen.max_gprs = 256;
1759 rdev->config.evergreen.max_threads = 248;
1760 rdev->config.evergreen.max_gs_threads = 32;
1761 rdev->config.evergreen.max_stack_entries = 512;
1762 rdev->config.evergreen.sx_num_of_sets = 4;
1763 rdev->config.evergreen.sx_max_export_size = 256;
1764 rdev->config.evergreen.sx_max_export_pos_size = 64;
1765 rdev->config.evergreen.sx_max_export_smx_size = 192;
1766 rdev->config.evergreen.max_hw_contexts = 8;
1767 rdev->config.evergreen.sq_num_cf_insts = 2;
1769 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1770 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1771 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1772 break;
1773 case CHIP_BARTS:
1774 rdev->config.evergreen.num_ses = 2;
1775 rdev->config.evergreen.max_pipes = 4;
1776 rdev->config.evergreen.max_tile_pipes = 8;
1777 rdev->config.evergreen.max_simds = 7;
1778 rdev->config.evergreen.max_backends = 4 * rdev->config.evergreen.num_ses;
1779 rdev->config.evergreen.max_gprs = 256;
1780 rdev->config.evergreen.max_threads = 248;
1781 rdev->config.evergreen.max_gs_threads = 32;
1782 rdev->config.evergreen.max_stack_entries = 512;
1783 rdev->config.evergreen.sx_num_of_sets = 4;
1784 rdev->config.evergreen.sx_max_export_size = 256;
1785 rdev->config.evergreen.sx_max_export_pos_size = 64;
1786 rdev->config.evergreen.sx_max_export_smx_size = 192;
1787 rdev->config.evergreen.max_hw_contexts = 8;
1788 rdev->config.evergreen.sq_num_cf_insts = 2;
1790 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1791 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1792 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1793 break;
1794 case CHIP_TURKS:
1795 rdev->config.evergreen.num_ses = 1;
1796 rdev->config.evergreen.max_pipes = 4;
1797 rdev->config.evergreen.max_tile_pipes = 4;
1798 rdev->config.evergreen.max_simds = 6;
1799 rdev->config.evergreen.max_backends = 2 * rdev->config.evergreen.num_ses;
1800 rdev->config.evergreen.max_gprs = 256;
1801 rdev->config.evergreen.max_threads = 248;
1802 rdev->config.evergreen.max_gs_threads = 32;
1803 rdev->config.evergreen.max_stack_entries = 256;
1804 rdev->config.evergreen.sx_num_of_sets = 4;
1805 rdev->config.evergreen.sx_max_export_size = 256;
1806 rdev->config.evergreen.sx_max_export_pos_size = 64;
1807 rdev->config.evergreen.sx_max_export_smx_size = 192;
1808 rdev->config.evergreen.max_hw_contexts = 8;
1809 rdev->config.evergreen.sq_num_cf_insts = 2;
1811 rdev->config.evergreen.sc_prim_fifo_size = 0x100;
1812 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1813 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1814 break;
1815 case CHIP_CAICOS:
1816 rdev->config.evergreen.num_ses = 1;
1817 rdev->config.evergreen.max_pipes = 4;
1818 rdev->config.evergreen.max_tile_pipes = 2;
1819 rdev->config.evergreen.max_simds = 2;
1820 rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
1821 rdev->config.evergreen.max_gprs = 256;
1822 rdev->config.evergreen.max_threads = 192;
1823 rdev->config.evergreen.max_gs_threads = 16;
1824 rdev->config.evergreen.max_stack_entries = 256;
1825 rdev->config.evergreen.sx_num_of_sets = 4;
1826 rdev->config.evergreen.sx_max_export_size = 128;
1827 rdev->config.evergreen.sx_max_export_pos_size = 32;
1828 rdev->config.evergreen.sx_max_export_smx_size = 96;
1829 rdev->config.evergreen.max_hw_contexts = 4;
1830 rdev->config.evergreen.sq_num_cf_insts = 1;
1832 rdev->config.evergreen.sc_prim_fifo_size = 0x40;
1833 rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
1834 rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
1835 break;
1838 /* Initialize HDP */
1839 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1840 WREG32((0x2c14 + j), 0x00000000);
1841 WREG32((0x2c18 + j), 0x00000000);
1842 WREG32((0x2c1c + j), 0x00000000);
1843 WREG32((0x2c20 + j), 0x00000000);
1844 WREG32((0x2c24 + j), 0x00000000);
1847 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1849 evergreen_fix_pci_max_read_req_size(rdev);
1851 cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & ~2;
1853 cc_gc_shader_pipe_config |=
1854 INACTIVE_QD_PIPES((EVERGREEN_MAX_PIPES_MASK << rdev->config.evergreen.max_pipes)
1855 & EVERGREEN_MAX_PIPES_MASK);
1856 cc_gc_shader_pipe_config |=
1857 INACTIVE_SIMDS((EVERGREEN_MAX_SIMDS_MASK << rdev->config.evergreen.max_simds)
1858 & EVERGREEN_MAX_SIMDS_MASK);
1860 cc_rb_backend_disable =
1861 BACKEND_DISABLE((EVERGREEN_MAX_BACKENDS_MASK << rdev->config.evergreen.max_backends)
1862 & EVERGREEN_MAX_BACKENDS_MASK);
1865 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1866 if (rdev->flags & RADEON_IS_IGP)
1867 mc_arb_ramcfg = RREG32(FUS_MC_ARB_RAMCFG);
1868 else
1869 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1871 switch (rdev->config.evergreen.max_tile_pipes) {
1872 case 1:
1873 default:
1874 gb_addr_config |= NUM_PIPES(0);
1875 break;
1876 case 2:
1877 gb_addr_config |= NUM_PIPES(1);
1878 break;
1879 case 4:
1880 gb_addr_config |= NUM_PIPES(2);
1881 break;
1882 case 8:
1883 gb_addr_config |= NUM_PIPES(3);
1884 break;
1887 gb_addr_config |= PIPE_INTERLEAVE_SIZE((mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT);
1888 gb_addr_config |= BANK_INTERLEAVE_SIZE(0);
1889 gb_addr_config |= NUM_SHADER_ENGINES(rdev->config.evergreen.num_ses - 1);
1890 gb_addr_config |= SHADER_ENGINE_TILE_SIZE(1);
1891 gb_addr_config |= NUM_GPUS(0); /* Hemlock? */
1892 gb_addr_config |= MULTI_GPU_TILE_SIZE(2);
1894 if (((mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT) > 2)
1895 gb_addr_config |= ROW_SIZE(2);
1896 else
1897 gb_addr_config |= ROW_SIZE((mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT);
1899 if (rdev->ddev->pdev->device == 0x689e) {
1900 u32 efuse_straps_4;
1901 u32 efuse_straps_3;
1902 u8 efuse_box_bit_131_124;
1904 WREG32(RCU_IND_INDEX, 0x204);
1905 efuse_straps_4 = RREG32(RCU_IND_DATA);
1906 WREG32(RCU_IND_INDEX, 0x203);
1907 efuse_straps_3 = RREG32(RCU_IND_DATA);
1908 efuse_box_bit_131_124 = (u8)(((efuse_straps_4 & 0xf) << 4) | ((efuse_straps_3 & 0xf0000000) >> 28));
1910 switch(efuse_box_bit_131_124) {
1911 case 0x00:
1912 gb_backend_map = 0x76543210;
1913 break;
1914 case 0x55:
1915 gb_backend_map = 0x77553311;
1916 break;
1917 case 0x56:
1918 gb_backend_map = 0x77553300;
1919 break;
1920 case 0x59:
1921 gb_backend_map = 0x77552211;
1922 break;
1923 case 0x66:
1924 gb_backend_map = 0x77443300;
1925 break;
1926 case 0x99:
1927 gb_backend_map = 0x66552211;
1928 break;
1929 case 0x5a:
1930 gb_backend_map = 0x77552200;
1931 break;
1932 case 0xaa:
1933 gb_backend_map = 0x66442200;
1934 break;
1935 case 0x95:
1936 gb_backend_map = 0x66553311;
1937 break;
1938 default:
1939 DRM_ERROR("bad backend map, using default\n");
1940 gb_backend_map =
1941 evergreen_get_tile_pipe_to_backend_map(rdev,
1942 rdev->config.evergreen.max_tile_pipes,
1943 rdev->config.evergreen.max_backends,
1944 ((EVERGREEN_MAX_BACKENDS_MASK <<
1945 rdev->config.evergreen.max_backends) &
1946 EVERGREEN_MAX_BACKENDS_MASK));
1947 break;
1949 } else if (rdev->ddev->pdev->device == 0x68b9) {
1950 u32 efuse_straps_3;
1951 u8 efuse_box_bit_127_124;
1953 WREG32(RCU_IND_INDEX, 0x203);
1954 efuse_straps_3 = RREG32(RCU_IND_DATA);
1955 efuse_box_bit_127_124 = (u8)((efuse_straps_3 & 0xF0000000) >> 28);
1957 switch(efuse_box_bit_127_124) {
1958 case 0x0:
1959 gb_backend_map = 0x00003210;
1960 break;
1961 case 0x5:
1962 case 0x6:
1963 case 0x9:
1964 case 0xa:
1965 gb_backend_map = 0x00003311;
1966 break;
1967 default:
1968 DRM_ERROR("bad backend map, using default\n");
1969 gb_backend_map =
1970 evergreen_get_tile_pipe_to_backend_map(rdev,
1971 rdev->config.evergreen.max_tile_pipes,
1972 rdev->config.evergreen.max_backends,
1973 ((EVERGREEN_MAX_BACKENDS_MASK <<
1974 rdev->config.evergreen.max_backends) &
1975 EVERGREEN_MAX_BACKENDS_MASK));
1976 break;
1978 } else {
1979 switch (rdev->family) {
1980 case CHIP_CYPRESS:
1981 case CHIP_HEMLOCK:
1982 case CHIP_BARTS:
1983 gb_backend_map = 0x66442200;
1984 break;
1985 case CHIP_JUNIPER:
1986 gb_backend_map = 0x00002200;
1987 break;
1988 default:
1989 gb_backend_map =
1990 evergreen_get_tile_pipe_to_backend_map(rdev,
1991 rdev->config.evergreen.max_tile_pipes,
1992 rdev->config.evergreen.max_backends,
1993 ((EVERGREEN_MAX_BACKENDS_MASK <<
1994 rdev->config.evergreen.max_backends) &
1995 EVERGREEN_MAX_BACKENDS_MASK));
1999 /* setup tiling info dword. gb_addr_config is not adequate since it does
2000 * not have bank info, so create a custom tiling dword.
2001 * bits 3:0 num_pipes
2002 * bits 7:4 num_banks
2003 * bits 11:8 group_size
2004 * bits 15:12 row_size
2006 rdev->config.evergreen.tile_config = 0;
2007 switch (rdev->config.evergreen.max_tile_pipes) {
2008 case 1:
2009 default:
2010 rdev->config.evergreen.tile_config |= (0 << 0);
2011 break;
2012 case 2:
2013 rdev->config.evergreen.tile_config |= (1 << 0);
2014 break;
2015 case 4:
2016 rdev->config.evergreen.tile_config |= (2 << 0);
2017 break;
2018 case 8:
2019 rdev->config.evergreen.tile_config |= (3 << 0);
2020 break;
2022 /* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */
2023 if (rdev->flags & RADEON_IS_IGP)
2024 rdev->config.evergreen.tile_config |= 1 << 4;
2025 else
2026 rdev->config.evergreen.tile_config |=
2027 ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) << 4;
2028 rdev->config.evergreen.tile_config |=
2029 ((mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT) << 8;
2030 rdev->config.evergreen.tile_config |=
2031 ((gb_addr_config & 0x30000000) >> 28) << 12;
2033 rdev->config.evergreen.backend_map = gb_backend_map;
2034 WREG32(GB_BACKEND_MAP, gb_backend_map);
2035 WREG32(GB_ADDR_CONFIG, gb_addr_config);
2036 WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
2037 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
2039 num_shader_engines = ((RREG32(GB_ADDR_CONFIG) & NUM_SHADER_ENGINES(3)) >> 12) + 1;
2040 grbm_gfx_index = INSTANCE_BROADCAST_WRITES;
2042 for (i = 0; i < rdev->config.evergreen.num_ses; i++) {
2043 u32 rb = cc_rb_backend_disable | (0xf0 << 16);
2044 u32 sp = cc_gc_shader_pipe_config;
2045 u32 gfx = grbm_gfx_index | SE_INDEX(i);
2047 if (i == num_shader_engines) {
2048 rb |= BACKEND_DISABLE(EVERGREEN_MAX_BACKENDS_MASK);
2049 sp |= INACTIVE_SIMDS(EVERGREEN_MAX_SIMDS_MASK);
2052 WREG32(GRBM_GFX_INDEX, gfx);
2053 WREG32(RLC_GFX_INDEX, gfx);
2055 WREG32(CC_RB_BACKEND_DISABLE, rb);
2056 WREG32(CC_SYS_RB_BACKEND_DISABLE, rb);
2057 WREG32(GC_USER_RB_BACKEND_DISABLE, rb);
2058 WREG32(CC_GC_SHADER_PIPE_CONFIG, sp);
2061 grbm_gfx_index |= SE_BROADCAST_WRITES;
2062 WREG32(GRBM_GFX_INDEX, grbm_gfx_index);
2063 WREG32(RLC_GFX_INDEX, grbm_gfx_index);
2065 WREG32(CGTS_SYS_TCC_DISABLE, 0);
2066 WREG32(CGTS_TCC_DISABLE, 0);
2067 WREG32(CGTS_USER_SYS_TCC_DISABLE, 0);
2068 WREG32(CGTS_USER_TCC_DISABLE, 0);
2070 /* set HW defaults for 3D engine */
2071 WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
2072 ROQ_IB2_START(0x2b)));
2074 WREG32(CP_MEQ_THRESHOLDS, STQ_SPLIT(0x30));
2076 WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO |
2077 SYNC_GRADIENT |
2078 SYNC_WALKER |
2079 SYNC_ALIGNER));
2081 sx_debug_1 = RREG32(SX_DEBUG_1);
2082 sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS;
2083 WREG32(SX_DEBUG_1, sx_debug_1);
2086 smx_dc_ctl0 = RREG32(SMX_DC_CTL0);
2087 smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff);
2088 smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.evergreen.sx_num_of_sets);
2089 WREG32(SMX_DC_CTL0, smx_dc_ctl0);
2091 WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_size / 4) - 1) |
2092 POSITION_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_pos_size / 4) - 1) |
2093 SMX_BUFFER_SIZE((rdev->config.evergreen.sx_max_export_smx_size / 4) - 1)));
2095 WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.evergreen.sc_prim_fifo_size) |
2096 SC_HIZ_TILE_FIFO_SIZE(rdev->config.evergreen.sc_hiz_tile_fifo_size) |
2097 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.evergreen.sc_earlyz_tile_fifo_size)));
2099 WREG32(VGT_NUM_INSTANCES, 1);
2100 WREG32(SPI_CONFIG_CNTL, 0);
2101 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
2102 WREG32(CP_PERFMON_CNTL, 0);
2104 WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.evergreen.sq_num_cf_insts) |
2105 FETCH_FIFO_HIWATER(0x4) |
2106 DONE_FIFO_HIWATER(0xe0) |
2107 ALU_UPDATE_FIFO_HIWATER(0x8)));
2109 sq_config = RREG32(SQ_CONFIG);
2110 sq_config &= ~(PS_PRIO(3) |
2111 VS_PRIO(3) |
2112 GS_PRIO(3) |
2113 ES_PRIO(3));
2114 sq_config |= (VC_ENABLE |
2115 EXPORT_SRC_C |
2116 PS_PRIO(0) |
2117 VS_PRIO(1) |
2118 GS_PRIO(2) |
2119 ES_PRIO(3));
2121 switch (rdev->family) {
2122 case CHIP_CEDAR:
2123 case CHIP_PALM:
2124 case CHIP_SUMO:
2125 case CHIP_SUMO2:
2126 case CHIP_CAICOS:
2127 /* no vertex cache */
2128 sq_config &= ~VC_ENABLE;
2129 break;
2130 default:
2131 break;
2134 sq_lds_resource_mgmt = RREG32(SQ_LDS_RESOURCE_MGMT);
2136 sq_gpr_resource_mgmt_1 = NUM_PS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2))* 12 / 32);
2137 sq_gpr_resource_mgmt_1 |= NUM_VS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 6 / 32);
2138 sq_gpr_resource_mgmt_1 |= NUM_CLAUSE_TEMP_GPRS(4);
2139 sq_gpr_resource_mgmt_2 = NUM_GS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 4 / 32);
2140 sq_gpr_resource_mgmt_2 |= NUM_ES_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 4 / 32);
2141 sq_gpr_resource_mgmt_3 = NUM_HS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
2142 sq_gpr_resource_mgmt_3 |= NUM_LS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
2144 switch (rdev->family) {
2145 case CHIP_CEDAR:
2146 case CHIP_PALM:
2147 case CHIP_SUMO:
2148 case CHIP_SUMO2:
2149 ps_thread_count = 96;
2150 break;
2151 default:
2152 ps_thread_count = 128;
2153 break;
2156 sq_thread_resource_mgmt = NUM_PS_THREADS(ps_thread_count);
2157 sq_thread_resource_mgmt |= NUM_VS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2158 sq_thread_resource_mgmt |= NUM_GS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2159 sq_thread_resource_mgmt |= NUM_ES_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2160 sq_thread_resource_mgmt_2 = NUM_HS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2161 sq_thread_resource_mgmt_2 |= NUM_LS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
2163 sq_stack_resource_mgmt_1 = NUM_PS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2164 sq_stack_resource_mgmt_1 |= NUM_VS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2165 sq_stack_resource_mgmt_2 = NUM_GS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2166 sq_stack_resource_mgmt_2 |= NUM_ES_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2167 sq_stack_resource_mgmt_3 = NUM_HS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2168 sq_stack_resource_mgmt_3 |= NUM_LS_STACK_ENTRIES((rdev->config.evergreen.max_stack_entries * 1) / 6);
2170 WREG32(SQ_CONFIG, sq_config);
2171 WREG32(SQ_GPR_RESOURCE_MGMT_1, sq_gpr_resource_mgmt_1);
2172 WREG32(SQ_GPR_RESOURCE_MGMT_2, sq_gpr_resource_mgmt_2);
2173 WREG32(SQ_GPR_RESOURCE_MGMT_3, sq_gpr_resource_mgmt_3);
2174 WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
2175 WREG32(SQ_THREAD_RESOURCE_MGMT_2, sq_thread_resource_mgmt_2);
2176 WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
2177 WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
2178 WREG32(SQ_STACK_RESOURCE_MGMT_3, sq_stack_resource_mgmt_3);
2179 WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, 0);
2180 WREG32(SQ_LDS_RESOURCE_MGMT, sq_lds_resource_mgmt);
2182 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
2183 FORCE_EOV_MAX_REZ_CNT(255)));
2185 switch (rdev->family) {
2186 case CHIP_CEDAR:
2187 case CHIP_PALM:
2188 case CHIP_SUMO:
2189 case CHIP_SUMO2:
2190 case CHIP_CAICOS:
2191 vgt_cache_invalidation = CACHE_INVALIDATION(TC_ONLY);
2192 break;
2193 default:
2194 vgt_cache_invalidation = CACHE_INVALIDATION(VC_AND_TC);
2195 break;
2197 vgt_cache_invalidation |= AUTO_INVLD_EN(ES_AND_GS_AUTO);
2198 WREG32(VGT_CACHE_INVALIDATION, vgt_cache_invalidation);
2200 WREG32(VGT_GS_VERTEX_REUSE, 16);
2201 WREG32(PA_SU_LINE_STIPPLE_VALUE, 0);
2202 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
2204 WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, 14);
2205 WREG32(VGT_OUT_DEALLOC_CNTL, 16);
2207 WREG32(CB_PERF_CTR0_SEL_0, 0);
2208 WREG32(CB_PERF_CTR0_SEL_1, 0);
2209 WREG32(CB_PERF_CTR1_SEL_0, 0);
2210 WREG32(CB_PERF_CTR1_SEL_1, 0);
2211 WREG32(CB_PERF_CTR2_SEL_0, 0);
2212 WREG32(CB_PERF_CTR2_SEL_1, 0);
2213 WREG32(CB_PERF_CTR3_SEL_0, 0);
2214 WREG32(CB_PERF_CTR3_SEL_1, 0);
2216 /* clear render buffer base addresses */
2217 WREG32(CB_COLOR0_BASE, 0);
2218 WREG32(CB_COLOR1_BASE, 0);
2219 WREG32(CB_COLOR2_BASE, 0);
2220 WREG32(CB_COLOR3_BASE, 0);
2221 WREG32(CB_COLOR4_BASE, 0);
2222 WREG32(CB_COLOR5_BASE, 0);
2223 WREG32(CB_COLOR6_BASE, 0);
2224 WREG32(CB_COLOR7_BASE, 0);
2225 WREG32(CB_COLOR8_BASE, 0);
2226 WREG32(CB_COLOR9_BASE, 0);
2227 WREG32(CB_COLOR10_BASE, 0);
2228 WREG32(CB_COLOR11_BASE, 0);
2230 /* set the shader const cache sizes to 0 */
2231 for (i = SQ_ALU_CONST_BUFFER_SIZE_PS_0; i < 0x28200; i += 4)
2232 WREG32(i, 0);
2233 for (i = SQ_ALU_CONST_BUFFER_SIZE_HS_0; i < 0x29000; i += 4)
2234 WREG32(i, 0);
2236 tmp = RREG32(HDP_MISC_CNTL);
2237 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
2238 WREG32(HDP_MISC_CNTL, tmp);
2240 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
2241 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
2243 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
2245 udelay(50);
2249 int evergreen_mc_init(struct radeon_device *rdev)
2251 u32 tmp;
2252 int chansize, numchan;
2254 /* Get VRAM informations */
2255 rdev->mc.vram_is_ddr = true;
2256 if (rdev->flags & RADEON_IS_IGP)
2257 tmp = RREG32(FUS_MC_ARB_RAMCFG);
2258 else
2259 tmp = RREG32(MC_ARB_RAMCFG);
2260 if (tmp & CHANSIZE_OVERRIDE) {
2261 chansize = 16;
2262 } else if (tmp & CHANSIZE_MASK) {
2263 chansize = 64;
2264 } else {
2265 chansize = 32;
2267 tmp = RREG32(MC_SHARED_CHMAP);
2268 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
2269 case 0:
2270 default:
2271 numchan = 1;
2272 break;
2273 case 1:
2274 numchan = 2;
2275 break;
2276 case 2:
2277 numchan = 4;
2278 break;
2279 case 3:
2280 numchan = 8;
2281 break;
2283 rdev->mc.vram_width = numchan * chansize;
2284 /* Could aper size report 0 ? */
2285 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2286 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2287 /* Setup GPU memory space */
2288 if (rdev->flags & RADEON_IS_IGP) {
2289 /* size in bytes on fusion */
2290 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
2291 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
2292 } else {
2293 /* size in MB on evergreen */
2294 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2295 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2297 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2298 r700_vram_gtt_location(rdev, &rdev->mc);
2299 radeon_update_bandwidth_info(rdev);
2301 return 0;
2304 bool evergreen_gpu_is_lockup(struct radeon_device *rdev)
2306 u32 srbm_status;
2307 u32 grbm_status;
2308 u32 grbm_status_se0, grbm_status_se1;
2309 struct r100_gpu_lockup *lockup = &rdev->config.evergreen.lockup;
2310 int r;
2312 srbm_status = RREG32(SRBM_STATUS);
2313 grbm_status = RREG32(GRBM_STATUS);
2314 grbm_status_se0 = RREG32(GRBM_STATUS_SE0);
2315 grbm_status_se1 = RREG32(GRBM_STATUS_SE1);
2316 if (!(grbm_status & GUI_ACTIVE)) {
2317 r100_gpu_lockup_update(lockup, &rdev->cp);
2318 return false;
2320 /* force CP activities */
2321 r = radeon_ring_lock(rdev, 2);
2322 if (!r) {
2323 /* PACKET2 NOP */
2324 radeon_ring_write(rdev, 0x80000000);
2325 radeon_ring_write(rdev, 0x80000000);
2326 radeon_ring_unlock_commit(rdev);
2328 rdev->cp.rptr = RREG32(CP_RB_RPTR);
2329 return r100_gpu_cp_is_lockup(rdev, lockup, &rdev->cp);
2332 static int evergreen_gpu_soft_reset(struct radeon_device *rdev)
2334 struct evergreen_mc_save save;
2335 u32 grbm_reset = 0;
2337 if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
2338 return 0;
2340 dev_info(rdev->dev, "GPU softreset \n");
2341 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
2342 RREG32(GRBM_STATUS));
2343 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
2344 RREG32(GRBM_STATUS_SE0));
2345 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
2346 RREG32(GRBM_STATUS_SE1));
2347 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
2348 RREG32(SRBM_STATUS));
2349 evergreen_mc_stop(rdev, &save);
2350 if (evergreen_mc_wait_for_idle(rdev)) {
2351 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2353 /* Disable CP parsing/prefetching */
2354 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT);
2356 /* reset all the gfx blocks */
2357 grbm_reset = (SOFT_RESET_CP |
2358 SOFT_RESET_CB |
2359 SOFT_RESET_DB |
2360 SOFT_RESET_PA |
2361 SOFT_RESET_SC |
2362 SOFT_RESET_SPI |
2363 SOFT_RESET_SH |
2364 SOFT_RESET_SX |
2365 SOFT_RESET_TC |
2366 SOFT_RESET_TA |
2367 SOFT_RESET_VC |
2368 SOFT_RESET_VGT);
2370 dev_info(rdev->dev, " GRBM_SOFT_RESET=0x%08X\n", grbm_reset);
2371 WREG32(GRBM_SOFT_RESET, grbm_reset);
2372 (void)RREG32(GRBM_SOFT_RESET);
2373 udelay(50);
2374 WREG32(GRBM_SOFT_RESET, 0);
2375 (void)RREG32(GRBM_SOFT_RESET);
2376 /* Wait a little for things to settle down */
2377 udelay(50);
2378 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
2379 RREG32(GRBM_STATUS));
2380 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
2381 RREG32(GRBM_STATUS_SE0));
2382 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
2383 RREG32(GRBM_STATUS_SE1));
2384 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
2385 RREG32(SRBM_STATUS));
2386 evergreen_mc_resume(rdev, &save);
2387 return 0;
2390 int evergreen_asic_reset(struct radeon_device *rdev)
2392 return evergreen_gpu_soft_reset(rdev);
2395 /* Interrupts */
2397 u32 evergreen_get_vblank_counter(struct radeon_device *rdev, int crtc)
2399 switch (crtc) {
2400 case 0:
2401 return RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC0_REGISTER_OFFSET);
2402 case 1:
2403 return RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC1_REGISTER_OFFSET);
2404 case 2:
2405 return RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC2_REGISTER_OFFSET);
2406 case 3:
2407 return RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC3_REGISTER_OFFSET);
2408 case 4:
2409 return RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC4_REGISTER_OFFSET);
2410 case 5:
2411 return RREG32(CRTC_STATUS_FRAME_COUNT + EVERGREEN_CRTC5_REGISTER_OFFSET);
2412 default:
2413 return 0;
2417 void evergreen_disable_interrupt_state(struct radeon_device *rdev)
2419 u32 tmp;
2421 WREG32(CP_INT_CNTL, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
2422 WREG32(GRBM_INT_CNTL, 0);
2423 WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
2424 WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
2425 if (rdev->num_crtc >= 4) {
2426 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
2427 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
2429 if (rdev->num_crtc >= 6) {
2430 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
2431 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
2434 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
2435 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
2436 if (rdev->num_crtc >= 4) {
2437 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
2438 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
2440 if (rdev->num_crtc >= 6) {
2441 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
2442 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
2445 WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
2446 WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
2448 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2449 WREG32(DC_HPD1_INT_CONTROL, tmp);
2450 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2451 WREG32(DC_HPD2_INT_CONTROL, tmp);
2452 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2453 WREG32(DC_HPD3_INT_CONTROL, tmp);
2454 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2455 WREG32(DC_HPD4_INT_CONTROL, tmp);
2456 tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2457 WREG32(DC_HPD5_INT_CONTROL, tmp);
2458 tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2459 WREG32(DC_HPD6_INT_CONTROL, tmp);
2463 int evergreen_irq_set(struct radeon_device *rdev)
2465 u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
2466 u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
2467 u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
2468 u32 grbm_int_cntl = 0;
2469 u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
2471 if (!rdev->irq.installed) {
2472 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
2473 return -EINVAL;
2475 /* don't enable anything if the ih is disabled */
2476 if (!rdev->ih.enabled) {
2477 r600_disable_interrupts(rdev);
2478 /* force the active interrupt state to all disabled */
2479 evergreen_disable_interrupt_state(rdev);
2480 return 0;
2483 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2484 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2485 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2486 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
2487 hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
2488 hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
2490 if (rdev->irq.sw_int) {
2491 DRM_DEBUG("evergreen_irq_set: sw int\n");
2492 cp_int_cntl |= RB_INT_ENABLE;
2493 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
2495 if (rdev->irq.crtc_vblank_int[0] ||
2496 rdev->irq.pflip[0]) {
2497 DRM_DEBUG("evergreen_irq_set: vblank 0\n");
2498 crtc1 |= VBLANK_INT_MASK;
2500 if (rdev->irq.crtc_vblank_int[1] ||
2501 rdev->irq.pflip[1]) {
2502 DRM_DEBUG("evergreen_irq_set: vblank 1\n");
2503 crtc2 |= VBLANK_INT_MASK;
2505 if (rdev->irq.crtc_vblank_int[2] ||
2506 rdev->irq.pflip[2]) {
2507 DRM_DEBUG("evergreen_irq_set: vblank 2\n");
2508 crtc3 |= VBLANK_INT_MASK;
2510 if (rdev->irq.crtc_vblank_int[3] ||
2511 rdev->irq.pflip[3]) {
2512 DRM_DEBUG("evergreen_irq_set: vblank 3\n");
2513 crtc4 |= VBLANK_INT_MASK;
2515 if (rdev->irq.crtc_vblank_int[4] ||
2516 rdev->irq.pflip[4]) {
2517 DRM_DEBUG("evergreen_irq_set: vblank 4\n");
2518 crtc5 |= VBLANK_INT_MASK;
2520 if (rdev->irq.crtc_vblank_int[5] ||
2521 rdev->irq.pflip[5]) {
2522 DRM_DEBUG("evergreen_irq_set: vblank 5\n");
2523 crtc6 |= VBLANK_INT_MASK;
2525 if (rdev->irq.hpd[0]) {
2526 DRM_DEBUG("evergreen_irq_set: hpd 1\n");
2527 hpd1 |= DC_HPDx_INT_EN;
2529 if (rdev->irq.hpd[1]) {
2530 DRM_DEBUG("evergreen_irq_set: hpd 2\n");
2531 hpd2 |= DC_HPDx_INT_EN;
2533 if (rdev->irq.hpd[2]) {
2534 DRM_DEBUG("evergreen_irq_set: hpd 3\n");
2535 hpd3 |= DC_HPDx_INT_EN;
2537 if (rdev->irq.hpd[3]) {
2538 DRM_DEBUG("evergreen_irq_set: hpd 4\n");
2539 hpd4 |= DC_HPDx_INT_EN;
2541 if (rdev->irq.hpd[4]) {
2542 DRM_DEBUG("evergreen_irq_set: hpd 5\n");
2543 hpd5 |= DC_HPDx_INT_EN;
2545 if (rdev->irq.hpd[5]) {
2546 DRM_DEBUG("evergreen_irq_set: hpd 6\n");
2547 hpd6 |= DC_HPDx_INT_EN;
2549 if (rdev->irq.gui_idle) {
2550 DRM_DEBUG("gui idle\n");
2551 grbm_int_cntl |= GUI_IDLE_INT_ENABLE;
2554 WREG32(CP_INT_CNTL, cp_int_cntl);
2555 WREG32(GRBM_INT_CNTL, grbm_int_cntl);
2557 WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
2558 WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
2559 if (rdev->num_crtc >= 4) {
2560 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
2561 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
2563 if (rdev->num_crtc >= 6) {
2564 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
2565 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
2568 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, grph1);
2569 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, grph2);
2570 if (rdev->num_crtc >= 4) {
2571 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, grph3);
2572 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, grph4);
2574 if (rdev->num_crtc >= 6) {
2575 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, grph5);
2576 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, grph6);
2579 WREG32(DC_HPD1_INT_CONTROL, hpd1);
2580 WREG32(DC_HPD2_INT_CONTROL, hpd2);
2581 WREG32(DC_HPD3_INT_CONTROL, hpd3);
2582 WREG32(DC_HPD4_INT_CONTROL, hpd4);
2583 WREG32(DC_HPD5_INT_CONTROL, hpd5);
2584 WREG32(DC_HPD6_INT_CONTROL, hpd6);
2586 return 0;
2589 static inline void evergreen_irq_ack(struct radeon_device *rdev)
2591 u32 tmp;
2593 rdev->irq.stat_regs.evergreen.disp_int = RREG32(DISP_INTERRUPT_STATUS);
2594 rdev->irq.stat_regs.evergreen.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
2595 rdev->irq.stat_regs.evergreen.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
2596 rdev->irq.stat_regs.evergreen.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
2597 rdev->irq.stat_regs.evergreen.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
2598 rdev->irq.stat_regs.evergreen.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
2599 rdev->irq.stat_regs.evergreen.d1grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
2600 rdev->irq.stat_regs.evergreen.d2grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
2601 if (rdev->num_crtc >= 4) {
2602 rdev->irq.stat_regs.evergreen.d3grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
2603 rdev->irq.stat_regs.evergreen.d4grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
2605 if (rdev->num_crtc >= 6) {
2606 rdev->irq.stat_regs.evergreen.d5grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
2607 rdev->irq.stat_regs.evergreen.d6grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
2610 if (rdev->irq.stat_regs.evergreen.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
2611 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2612 if (rdev->irq.stat_regs.evergreen.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
2613 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2614 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT)
2615 WREG32(VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
2616 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT)
2617 WREG32(VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
2618 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
2619 WREG32(VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
2620 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT)
2621 WREG32(VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
2623 if (rdev->num_crtc >= 4) {
2624 if (rdev->irq.stat_regs.evergreen.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
2625 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2626 if (rdev->irq.stat_regs.evergreen.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
2627 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2628 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
2629 WREG32(VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
2630 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
2631 WREG32(VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
2632 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
2633 WREG32(VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
2634 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
2635 WREG32(VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
2638 if (rdev->num_crtc >= 6) {
2639 if (rdev->irq.stat_regs.evergreen.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
2640 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2641 if (rdev->irq.stat_regs.evergreen.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
2642 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
2643 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
2644 WREG32(VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
2645 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
2646 WREG32(VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
2647 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
2648 WREG32(VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
2649 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
2650 WREG32(VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
2653 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
2654 tmp = RREG32(DC_HPD1_INT_CONTROL);
2655 tmp |= DC_HPDx_INT_ACK;
2656 WREG32(DC_HPD1_INT_CONTROL, tmp);
2658 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
2659 tmp = RREG32(DC_HPD2_INT_CONTROL);
2660 tmp |= DC_HPDx_INT_ACK;
2661 WREG32(DC_HPD2_INT_CONTROL, tmp);
2663 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
2664 tmp = RREG32(DC_HPD3_INT_CONTROL);
2665 tmp |= DC_HPDx_INT_ACK;
2666 WREG32(DC_HPD3_INT_CONTROL, tmp);
2668 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
2669 tmp = RREG32(DC_HPD4_INT_CONTROL);
2670 tmp |= DC_HPDx_INT_ACK;
2671 WREG32(DC_HPD4_INT_CONTROL, tmp);
2673 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
2674 tmp = RREG32(DC_HPD5_INT_CONTROL);
2675 tmp |= DC_HPDx_INT_ACK;
2676 WREG32(DC_HPD5_INT_CONTROL, tmp);
2678 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
2679 tmp = RREG32(DC_HPD5_INT_CONTROL);
2680 tmp |= DC_HPDx_INT_ACK;
2681 WREG32(DC_HPD6_INT_CONTROL, tmp);
2685 void evergreen_irq_disable(struct radeon_device *rdev)
2687 r600_disable_interrupts(rdev);
2688 /* Wait and acknowledge irq */
2689 mdelay(1);
2690 evergreen_irq_ack(rdev);
2691 evergreen_disable_interrupt_state(rdev);
2694 void evergreen_irq_suspend(struct radeon_device *rdev)
2696 evergreen_irq_disable(rdev);
2697 r600_rlc_stop(rdev);
2700 static inline u32 evergreen_get_ih_wptr(struct radeon_device *rdev)
2702 u32 wptr, tmp;
2704 if (rdev->wb.enabled)
2705 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
2706 else
2707 wptr = RREG32(IH_RB_WPTR);
2709 if (wptr & RB_OVERFLOW) {
2710 /* When a ring buffer overflow happen start parsing interrupt
2711 * from the last not overwritten vector (wptr + 16). Hopefully
2712 * this should allow us to catchup.
2714 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
2715 wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
2716 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
2717 tmp = RREG32(IH_RB_CNTL);
2718 tmp |= IH_WPTR_OVERFLOW_CLEAR;
2719 WREG32(IH_RB_CNTL, tmp);
2721 return (wptr & rdev->ih.ptr_mask);
2724 int evergreen_irq_process(struct radeon_device *rdev)
2726 u32 wptr;
2727 u32 rptr;
2728 u32 src_id, src_data;
2729 u32 ring_index;
2730 unsigned long flags;
2731 bool queue_hotplug = false;
2733 if (!rdev->ih.enabled || rdev->shutdown)
2734 return IRQ_NONE;
2736 wptr = evergreen_get_ih_wptr(rdev);
2737 rptr = rdev->ih.rptr;
2738 DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
2740 spin_lock_irqsave(&rdev->ih.lock, flags);
2741 if (rptr == wptr) {
2742 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2743 return IRQ_NONE;
2745 restart_ih:
2746 /* Order reading of wptr vs. reading of IH ring data */
2747 rmb();
2749 /* display interrupts */
2750 evergreen_irq_ack(rdev);
2752 rdev->ih.wptr = wptr;
2753 while (rptr != wptr) {
2754 /* wptr/rptr are in bytes! */
2755 ring_index = rptr / 4;
2756 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
2757 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
2759 switch (src_id) {
2760 case 1: /* D1 vblank/vline */
2761 switch (src_data) {
2762 case 0: /* D1 vblank */
2763 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT) {
2764 if (rdev->irq.crtc_vblank_int[0]) {
2765 drm_handle_vblank(rdev->ddev, 0);
2766 rdev->pm.vblank_sync = true;
2767 wake_up(&rdev->irq.vblank_queue);
2769 if (rdev->irq.pflip[0])
2770 radeon_crtc_handle_flip(rdev, 0);
2771 rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
2772 DRM_DEBUG("IH: D1 vblank\n");
2774 break;
2775 case 1: /* D1 vline */
2776 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT) {
2777 rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VLINE_INTERRUPT;
2778 DRM_DEBUG("IH: D1 vline\n");
2780 break;
2781 default:
2782 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2783 break;
2785 break;
2786 case 2: /* D2 vblank/vline */
2787 switch (src_data) {
2788 case 0: /* D2 vblank */
2789 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
2790 if (rdev->irq.crtc_vblank_int[1]) {
2791 drm_handle_vblank(rdev->ddev, 1);
2792 rdev->pm.vblank_sync = true;
2793 wake_up(&rdev->irq.vblank_queue);
2795 if (rdev->irq.pflip[1])
2796 radeon_crtc_handle_flip(rdev, 1);
2797 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
2798 DRM_DEBUG("IH: D2 vblank\n");
2800 break;
2801 case 1: /* D2 vline */
2802 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
2803 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
2804 DRM_DEBUG("IH: D2 vline\n");
2806 break;
2807 default:
2808 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2809 break;
2811 break;
2812 case 3: /* D3 vblank/vline */
2813 switch (src_data) {
2814 case 0: /* D3 vblank */
2815 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
2816 if (rdev->irq.crtc_vblank_int[2]) {
2817 drm_handle_vblank(rdev->ddev, 2);
2818 rdev->pm.vblank_sync = true;
2819 wake_up(&rdev->irq.vblank_queue);
2821 if (rdev->irq.pflip[2])
2822 radeon_crtc_handle_flip(rdev, 2);
2823 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
2824 DRM_DEBUG("IH: D3 vblank\n");
2826 break;
2827 case 1: /* D3 vline */
2828 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
2829 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
2830 DRM_DEBUG("IH: D3 vline\n");
2832 break;
2833 default:
2834 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2835 break;
2837 break;
2838 case 4: /* D4 vblank/vline */
2839 switch (src_data) {
2840 case 0: /* D4 vblank */
2841 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
2842 if (rdev->irq.crtc_vblank_int[3]) {
2843 drm_handle_vblank(rdev->ddev, 3);
2844 rdev->pm.vblank_sync = true;
2845 wake_up(&rdev->irq.vblank_queue);
2847 if (rdev->irq.pflip[3])
2848 radeon_crtc_handle_flip(rdev, 3);
2849 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
2850 DRM_DEBUG("IH: D4 vblank\n");
2852 break;
2853 case 1: /* D4 vline */
2854 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
2855 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
2856 DRM_DEBUG("IH: D4 vline\n");
2858 break;
2859 default:
2860 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2861 break;
2863 break;
2864 case 5: /* D5 vblank/vline */
2865 switch (src_data) {
2866 case 0: /* D5 vblank */
2867 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
2868 if (rdev->irq.crtc_vblank_int[4]) {
2869 drm_handle_vblank(rdev->ddev, 4);
2870 rdev->pm.vblank_sync = true;
2871 wake_up(&rdev->irq.vblank_queue);
2873 if (rdev->irq.pflip[4])
2874 radeon_crtc_handle_flip(rdev, 4);
2875 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
2876 DRM_DEBUG("IH: D5 vblank\n");
2878 break;
2879 case 1: /* D5 vline */
2880 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
2881 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
2882 DRM_DEBUG("IH: D5 vline\n");
2884 break;
2885 default:
2886 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2887 break;
2889 break;
2890 case 6: /* D6 vblank/vline */
2891 switch (src_data) {
2892 case 0: /* D6 vblank */
2893 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
2894 if (rdev->irq.crtc_vblank_int[5]) {
2895 drm_handle_vblank(rdev->ddev, 5);
2896 rdev->pm.vblank_sync = true;
2897 wake_up(&rdev->irq.vblank_queue);
2899 if (rdev->irq.pflip[5])
2900 radeon_crtc_handle_flip(rdev, 5);
2901 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
2902 DRM_DEBUG("IH: D6 vblank\n");
2904 break;
2905 case 1: /* D6 vline */
2906 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
2907 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
2908 DRM_DEBUG("IH: D6 vline\n");
2910 break;
2911 default:
2912 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2913 break;
2915 break;
2916 case 42: /* HPD hotplug */
2917 switch (src_data) {
2918 case 0:
2919 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
2920 rdev->irq.stat_regs.evergreen.disp_int &= ~DC_HPD1_INTERRUPT;
2921 queue_hotplug = true;
2922 DRM_DEBUG("IH: HPD1\n");
2924 break;
2925 case 1:
2926 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
2927 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~DC_HPD2_INTERRUPT;
2928 queue_hotplug = true;
2929 DRM_DEBUG("IH: HPD2\n");
2931 break;
2932 case 2:
2933 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
2934 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
2935 queue_hotplug = true;
2936 DRM_DEBUG("IH: HPD3\n");
2938 break;
2939 case 3:
2940 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
2941 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
2942 queue_hotplug = true;
2943 DRM_DEBUG("IH: HPD4\n");
2945 break;
2946 case 4:
2947 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
2948 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
2949 queue_hotplug = true;
2950 DRM_DEBUG("IH: HPD5\n");
2952 break;
2953 case 5:
2954 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
2955 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
2956 queue_hotplug = true;
2957 DRM_DEBUG("IH: HPD6\n");
2959 break;
2960 default:
2961 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2962 break;
2964 break;
2965 case 176: /* CP_INT in ring buffer */
2966 case 177: /* CP_INT in IB1 */
2967 case 178: /* CP_INT in IB2 */
2968 DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
2969 radeon_fence_process(rdev);
2970 break;
2971 case 181: /* CP EOP event */
2972 DRM_DEBUG("IH: CP EOP\n");
2973 radeon_fence_process(rdev);
2974 break;
2975 case 233: /* GUI IDLE */
2976 DRM_DEBUG("IH: GUI idle\n");
2977 rdev->pm.gui_idle = true;
2978 wake_up(&rdev->irq.idle_queue);
2979 break;
2980 default:
2981 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2982 break;
2985 /* wptr/rptr are in bytes! */
2986 rptr += 16;
2987 rptr &= rdev->ih.ptr_mask;
2989 /* make sure wptr hasn't changed while processing */
2990 wptr = evergreen_get_ih_wptr(rdev);
2991 if (wptr != rdev->ih.wptr)
2992 goto restart_ih;
2993 if (queue_hotplug)
2994 schedule_work(&rdev->hotplug_work);
2995 rdev->ih.rptr = rptr;
2996 WREG32(IH_RB_RPTR, rdev->ih.rptr);
2997 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2998 return IRQ_HANDLED;
3001 static int evergreen_startup(struct radeon_device *rdev)
3003 int r;
3005 /* enable pcie gen2 link */
3006 if (!ASIC_IS_DCE5(rdev))
3007 evergreen_pcie_gen2_enable(rdev);
3009 if (ASIC_IS_DCE5(rdev)) {
3010 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) {
3011 r = ni_init_microcode(rdev);
3012 if (r) {
3013 DRM_ERROR("Failed to load firmware!\n");
3014 return r;
3017 r = ni_mc_load_microcode(rdev);
3018 if (r) {
3019 DRM_ERROR("Failed to load MC firmware!\n");
3020 return r;
3022 } else {
3023 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
3024 r = r600_init_microcode(rdev);
3025 if (r) {
3026 DRM_ERROR("Failed to load firmware!\n");
3027 return r;
3032 evergreen_mc_program(rdev);
3033 if (rdev->flags & RADEON_IS_AGP) {
3034 evergreen_agp_enable(rdev);
3035 } else {
3036 r = evergreen_pcie_gart_enable(rdev);
3037 if (r)
3038 return r;
3040 evergreen_gpu_init(rdev);
3042 r = evergreen_blit_init(rdev);
3043 if (r) {
3044 evergreen_blit_fini(rdev);
3045 rdev->asic->copy = NULL;
3046 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
3049 /* allocate wb buffer */
3050 r = radeon_wb_init(rdev);
3051 if (r)
3052 return r;
3054 /* Enable IRQ */
3055 r = r600_irq_init(rdev);
3056 if (r) {
3057 DRM_ERROR("radeon: IH init failed (%d).\n", r);
3058 radeon_irq_kms_fini(rdev);
3059 return r;
3061 evergreen_irq_set(rdev);
3063 r = radeon_ring_init(rdev, rdev->cp.ring_size);
3064 if (r)
3065 return r;
3066 r = evergreen_cp_load_microcode(rdev);
3067 if (r)
3068 return r;
3069 r = evergreen_cp_resume(rdev);
3070 if (r)
3071 return r;
3073 return 0;
3076 int evergreen_resume(struct radeon_device *rdev)
3078 int r;
3080 /* reset the asic, the gfx blocks are often in a bad state
3081 * after the driver is unloaded or after a resume
3083 if (radeon_asic_reset(rdev))
3084 dev_warn(rdev->dev, "GPU reset failed !\n");
3085 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
3086 * posting will perform necessary task to bring back GPU into good
3087 * shape.
3089 /* post card */
3090 atom_asic_init(rdev->mode_info.atom_context);
3092 r = evergreen_startup(rdev);
3093 if (r) {
3094 DRM_ERROR("evergreen startup failed on resume\n");
3095 return r;
3098 r = r600_ib_test(rdev);
3099 if (r) {
3100 DRM_ERROR("radeon: failed testing IB (%d).\n", r);
3101 return r;
3104 return r;
3108 int evergreen_suspend(struct radeon_device *rdev)
3110 int r;
3112 /* FIXME: we should wait for ring to be empty */
3113 r700_cp_stop(rdev);
3114 rdev->cp.ready = false;
3115 evergreen_irq_suspend(rdev);
3116 radeon_wb_disable(rdev);
3117 evergreen_pcie_gart_disable(rdev);
3119 /* unpin shaders bo */
3120 r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
3121 if (likely(r == 0)) {
3122 radeon_bo_unpin(rdev->r600_blit.shader_obj);
3123 radeon_bo_unreserve(rdev->r600_blit.shader_obj);
3126 return 0;
3129 int evergreen_copy_blit(struct radeon_device *rdev,
3130 uint64_t src_offset,
3131 uint64_t dst_offset,
3132 unsigned num_gpu_pages,
3133 struct radeon_fence *fence)
3135 int r;
3137 mutex_lock(&rdev->r600_blit.mutex);
3138 rdev->r600_blit.vb_ib = NULL;
3139 r = evergreen_blit_prepare_copy(rdev, num_gpu_pages * RADEON_GPU_PAGE_SIZE);
3140 if (r) {
3141 if (rdev->r600_blit.vb_ib)
3142 radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
3143 mutex_unlock(&rdev->r600_blit.mutex);
3144 return r;
3146 evergreen_kms_blit_copy(rdev, src_offset, dst_offset, num_gpu_pages * RADEON_GPU_PAGE_SIZE);
3147 evergreen_blit_done_copy(rdev, fence);
3148 mutex_unlock(&rdev->r600_blit.mutex);
3149 return 0;
3152 /* Plan is to move initialization in that function and use
3153 * helper function so that radeon_device_init pretty much
3154 * do nothing more than calling asic specific function. This
3155 * should also allow to remove a bunch of callback function
3156 * like vram_info.
3158 int evergreen_init(struct radeon_device *rdev)
3160 int r;
3162 /* This don't do much */
3163 r = radeon_gem_init(rdev);
3164 if (r)
3165 return r;
3166 /* Read BIOS */
3167 if (!radeon_get_bios(rdev)) {
3168 if (ASIC_IS_AVIVO(rdev))
3169 return -EINVAL;
3171 /* Must be an ATOMBIOS */
3172 if (!rdev->is_atom_bios) {
3173 dev_err(rdev->dev, "Expecting atombios for evergreen GPU\n");
3174 return -EINVAL;
3176 r = radeon_atombios_init(rdev);
3177 if (r)
3178 return r;
3179 /* reset the asic, the gfx blocks are often in a bad state
3180 * after the driver is unloaded or after a resume
3182 if (radeon_asic_reset(rdev))
3183 dev_warn(rdev->dev, "GPU reset failed !\n");
3184 /* Post card if necessary */
3185 if (!radeon_card_posted(rdev)) {
3186 if (!rdev->bios) {
3187 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
3188 return -EINVAL;
3190 DRM_INFO("GPU not posted. posting now...\n");
3191 atom_asic_init(rdev->mode_info.atom_context);
3193 /* Initialize scratch registers */
3194 r600_scratch_init(rdev);
3195 /* Initialize surface registers */
3196 radeon_surface_init(rdev);
3197 /* Initialize clocks */
3198 radeon_get_clock_info(rdev->ddev);
3199 /* Fence driver */
3200 r = radeon_fence_driver_init(rdev);
3201 if (r)
3202 return r;
3203 /* initialize AGP */
3204 if (rdev->flags & RADEON_IS_AGP) {
3205 r = radeon_agp_init(rdev);
3206 if (r)
3207 radeon_agp_disable(rdev);
3209 /* initialize memory controller */
3210 r = evergreen_mc_init(rdev);
3211 if (r)
3212 return r;
3213 /* Memory manager */
3214 r = radeon_bo_init(rdev);
3215 if (r)
3216 return r;
3218 r = radeon_irq_kms_init(rdev);
3219 if (r)
3220 return r;
3222 rdev->cp.ring_obj = NULL;
3223 r600_ring_init(rdev, 1024 * 1024);
3225 rdev->ih.ring_obj = NULL;
3226 r600_ih_ring_init(rdev, 64 * 1024);
3228 r = r600_pcie_gart_init(rdev);
3229 if (r)
3230 return r;
3232 rdev->accel_working = true;
3233 r = evergreen_startup(rdev);
3234 if (r) {
3235 dev_err(rdev->dev, "disabling GPU acceleration\n");
3236 r700_cp_fini(rdev);
3237 r600_irq_fini(rdev);
3238 radeon_wb_fini(rdev);
3239 radeon_irq_kms_fini(rdev);
3240 evergreen_pcie_gart_fini(rdev);
3241 rdev->accel_working = false;
3243 if (rdev->accel_working) {
3244 r = radeon_ib_pool_init(rdev);
3245 if (r) {
3246 DRM_ERROR("radeon: failed initializing IB pool (%d).\n", r);
3247 rdev->accel_working = false;
3249 r = r600_ib_test(rdev);
3250 if (r) {
3251 DRM_ERROR("radeon: failed testing IB (%d).\n", r);
3252 rdev->accel_working = false;
3255 return 0;
3258 void evergreen_fini(struct radeon_device *rdev)
3260 evergreen_blit_fini(rdev);
3261 r700_cp_fini(rdev);
3262 r600_irq_fini(rdev);
3263 radeon_wb_fini(rdev);
3264 radeon_ib_pool_fini(rdev);
3265 radeon_irq_kms_fini(rdev);
3266 evergreen_pcie_gart_fini(rdev);
3267 radeon_gem_fini(rdev);
3268 radeon_fence_driver_fini(rdev);
3269 radeon_agp_fini(rdev);
3270 radeon_bo_fini(rdev);
3271 radeon_atombios_fini(rdev);
3272 kfree(rdev->bios);
3273 rdev->bios = NULL;
3276 static void evergreen_pcie_gen2_enable(struct radeon_device *rdev)
3278 u32 link_width_cntl, speed_cntl;
3280 if (radeon_pcie_gen2 == 0)
3281 return;
3283 if (rdev->flags & RADEON_IS_IGP)
3284 return;
3286 if (!(rdev->flags & RADEON_IS_PCIE))
3287 return;
3289 /* x2 cards have a special sequence */
3290 if (ASIC_IS_X2(rdev))
3291 return;
3293 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
3294 if ((speed_cntl & LC_OTHER_SIDE_EVER_SENT_GEN2) ||
3295 (speed_cntl & LC_OTHER_SIDE_SUPPORTS_GEN2)) {
3297 link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL);
3298 link_width_cntl &= ~LC_UPCONFIGURE_DIS;
3299 WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
3301 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
3302 speed_cntl &= ~LC_TARGET_LINK_SPEED_OVERRIDE_EN;
3303 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
3305 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
3306 speed_cntl |= LC_CLR_FAILED_SPD_CHANGE_CNT;
3307 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
3309 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
3310 speed_cntl &= ~LC_CLR_FAILED_SPD_CHANGE_CNT;
3311 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
3313 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL);
3314 speed_cntl |= LC_GEN2_EN_STRAP;
3315 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl);
3317 } else {
3318 link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL);
3319 /* XXX: only disable it if gen1 bridge vendor == 0x111d or 0x1106 */
3320 if (1)
3321 link_width_cntl |= LC_UPCONFIGURE_DIS;
3322 else
3323 link_width_cntl &= ~LC_UPCONFIGURE_DIS;
3324 WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);