2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
24 * Authors: Dave Airlie
28 #include <linux/seq_file.h>
29 #include <linux/slab.h>
32 #include "radeon_drm.h"
33 #include "radeon_reg.h"
35 #include "radeon_asic.h"
42 #include <linux/firmware.h>
43 #include <linux/platform_device.h>
45 #include "r100_reg_safe.h"
46 #include "rn50_reg_safe.h"
49 #define FIRMWARE_R100 "radeon/R100_cp.bin"
50 #define FIRMWARE_R200 "radeon/R200_cp.bin"
51 #define FIRMWARE_R300 "radeon/R300_cp.bin"
52 #define FIRMWARE_R420 "radeon/R420_cp.bin"
53 #define FIRMWARE_RS690 "radeon/RS690_cp.bin"
54 #define FIRMWARE_RS600 "radeon/RS600_cp.bin"
55 #define FIRMWARE_R520 "radeon/R520_cp.bin"
57 MODULE_FIRMWARE(FIRMWARE_R100);
58 MODULE_FIRMWARE(FIRMWARE_R200);
59 MODULE_FIRMWARE(FIRMWARE_R300);
60 MODULE_FIRMWARE(FIRMWARE_R420);
61 MODULE_FIRMWARE(FIRMWARE_RS690);
62 MODULE_FIRMWARE(FIRMWARE_RS600);
63 MODULE_FIRMWARE(FIRMWARE_R520);
65 #include "r100_track.h"
67 /* This files gather functions specifics to:
68 * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
71 void r100_pre_page_flip(struct radeon_device *rdev, int crtc)
73 struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc];
76 /* make sure flip is at vb rather than hb */
77 tmp = RREG32(RADEON_CRTC_OFFSET_CNTL + radeon_crtc->crtc_offset);
78 tmp &= ~RADEON_CRTC_OFFSET_FLIP_CNTL;
79 /* make sure pending bit is asserted */
80 tmp |= RADEON_CRTC_GUI_TRIG_OFFSET_LEFT_EN;
81 WREG32(RADEON_CRTC_OFFSET_CNTL + radeon_crtc->crtc_offset, tmp);
83 /* set pageflip to happen as late as possible in the vblank interval.
84 * same field for crtc1/2
86 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
87 tmp &= ~RADEON_CRTC_VSTAT_MODE_MASK;
88 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
90 /* enable the pflip int */
91 radeon_irq_kms_pflip_irq_get(rdev, crtc);
94 void r100_post_page_flip(struct radeon_device *rdev, int crtc)
96 /* disable the pflip int */
97 radeon_irq_kms_pflip_irq_put(rdev, crtc);
100 u32 r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
102 struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
103 u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
105 /* Lock the graphics update lock */
106 /* update the scanout addresses */
107 WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
109 /* Wait for update_pending to go high. */
110 while (!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET));
111 DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
113 /* Unlock the lock, so double-buffering can take place inside vblank */
114 tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
115 WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
117 /* Return current update_pending status: */
118 return RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET;
121 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
124 rdev->pm.dynpm_can_upclock = true;
125 rdev->pm.dynpm_can_downclock = true;
127 switch (rdev->pm.dynpm_planned_action) {
128 case DYNPM_ACTION_MINIMUM:
129 rdev->pm.requested_power_state_index = 0;
130 rdev->pm.dynpm_can_downclock = false;
132 case DYNPM_ACTION_DOWNCLOCK:
133 if (rdev->pm.current_power_state_index == 0) {
134 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
135 rdev->pm.dynpm_can_downclock = false;
137 if (rdev->pm.active_crtc_count > 1) {
138 for (i = 0; i < rdev->pm.num_power_states; i++) {
139 if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
141 else if (i >= rdev->pm.current_power_state_index) {
142 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
145 rdev->pm.requested_power_state_index = i;
150 rdev->pm.requested_power_state_index =
151 rdev->pm.current_power_state_index - 1;
153 /* don't use the power state if crtcs are active and no display flag is set */
154 if ((rdev->pm.active_crtc_count > 0) &&
155 (rdev->pm.power_state[rdev->pm.requested_power_state_index].clock_info[0].flags &
156 RADEON_PM_MODE_NO_DISPLAY)) {
157 rdev->pm.requested_power_state_index++;
160 case DYNPM_ACTION_UPCLOCK:
161 if (rdev->pm.current_power_state_index == (rdev->pm.num_power_states - 1)) {
162 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
163 rdev->pm.dynpm_can_upclock = false;
165 if (rdev->pm.active_crtc_count > 1) {
166 for (i = (rdev->pm.num_power_states - 1); i >= 0; i--) {
167 if (rdev->pm.power_state[i].flags & RADEON_PM_STATE_SINGLE_DISPLAY_ONLY)
169 else if (i <= rdev->pm.current_power_state_index) {
170 rdev->pm.requested_power_state_index = rdev->pm.current_power_state_index;
173 rdev->pm.requested_power_state_index = i;
178 rdev->pm.requested_power_state_index =
179 rdev->pm.current_power_state_index + 1;
182 case DYNPM_ACTION_DEFAULT:
183 rdev->pm.requested_power_state_index = rdev->pm.default_power_state_index;
184 rdev->pm.dynpm_can_upclock = false;
186 case DYNPM_ACTION_NONE:
188 DRM_ERROR("Requested mode for not defined action\n");
191 /* only one clock mode per power state */
192 rdev->pm.requested_clock_mode_index = 0;
194 DRM_DEBUG_DRIVER("Requested: e: %d m: %d p: %d\n",
195 rdev->pm.power_state[rdev->pm.requested_power_state_index].
196 clock_info[rdev->pm.requested_clock_mode_index].sclk,
197 rdev->pm.power_state[rdev->pm.requested_power_state_index].
198 clock_info[rdev->pm.requested_clock_mode_index].mclk,
199 rdev->pm.power_state[rdev->pm.requested_power_state_index].
203 void r100_pm_init_profile(struct radeon_device *rdev)
206 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
207 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
208 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
209 rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
211 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
212 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
213 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
214 rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
216 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
217 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 0;
218 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
219 rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
221 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
222 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
223 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
224 rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
226 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
227 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
228 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
229 rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
231 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
232 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
233 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
234 rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
236 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
237 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
238 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
239 rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
242 void r100_pm_misc(struct radeon_device *rdev)
244 int requested_index = rdev->pm.requested_power_state_index;
245 struct radeon_power_state *ps = &rdev->pm.power_state[requested_index];
246 struct radeon_voltage *voltage = &ps->clock_info[0].voltage;
247 u32 tmp, sclk_cntl, sclk_cntl2, sclk_more_cntl;
249 if ((voltage->type == VOLTAGE_GPIO) && (voltage->gpio.valid)) {
250 if (ps->misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) {
251 tmp = RREG32(voltage->gpio.reg);
252 if (voltage->active_high)
253 tmp |= voltage->gpio.mask;
255 tmp &= ~(voltage->gpio.mask);
256 WREG32(voltage->gpio.reg, tmp);
258 udelay(voltage->delay);
260 tmp = RREG32(voltage->gpio.reg);
261 if (voltage->active_high)
262 tmp &= ~voltage->gpio.mask;
264 tmp |= voltage->gpio.mask;
265 WREG32(voltage->gpio.reg, tmp);
267 udelay(voltage->delay);
271 sclk_cntl = RREG32_PLL(SCLK_CNTL);
272 sclk_cntl2 = RREG32_PLL(SCLK_CNTL2);
273 sclk_cntl2 &= ~REDUCED_SPEED_SCLK_SEL(3);
274 sclk_more_cntl = RREG32_PLL(SCLK_MORE_CNTL);
275 sclk_more_cntl &= ~VOLTAGE_DELAY_SEL(3);
276 if (ps->misc & ATOM_PM_MISCINFO_ASIC_REDUCED_SPEED_SCLK_EN) {
277 sclk_more_cntl |= REDUCED_SPEED_SCLK_EN;
278 if (ps->misc & ATOM_PM_MISCINFO_DYN_CLK_3D_IDLE)
279 sclk_cntl2 |= REDUCED_SPEED_SCLK_MODE;
281 sclk_cntl2 &= ~REDUCED_SPEED_SCLK_MODE;
282 if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_2)
283 sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(0);
284 else if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_CLOCK_DIVIDER_BY_4)
285 sclk_cntl2 |= REDUCED_SPEED_SCLK_SEL(2);
287 sclk_more_cntl &= ~REDUCED_SPEED_SCLK_EN;
289 if (ps->misc & ATOM_PM_MISCINFO_ASIC_DYNAMIC_VOLTAGE_EN) {
290 sclk_more_cntl |= IO_CG_VOLTAGE_DROP;
291 if (voltage->delay) {
292 sclk_more_cntl |= VOLTAGE_DROP_SYNC;
293 switch (voltage->delay) {
295 sclk_more_cntl |= VOLTAGE_DELAY_SEL(0);
298 sclk_more_cntl |= VOLTAGE_DELAY_SEL(1);
301 sclk_more_cntl |= VOLTAGE_DELAY_SEL(2);
304 sclk_more_cntl |= VOLTAGE_DELAY_SEL(3);
308 sclk_more_cntl &= ~VOLTAGE_DROP_SYNC;
310 sclk_more_cntl &= ~IO_CG_VOLTAGE_DROP;
312 if (ps->misc & ATOM_PM_MISCINFO_DYNAMIC_HDP_BLOCK_EN)
313 sclk_cntl &= ~FORCE_HDP;
315 sclk_cntl |= FORCE_HDP;
317 WREG32_PLL(SCLK_CNTL, sclk_cntl);
318 WREG32_PLL(SCLK_CNTL2, sclk_cntl2);
319 WREG32_PLL(SCLK_MORE_CNTL, sclk_more_cntl);
322 if ((rdev->flags & RADEON_IS_PCIE) &&
323 !(rdev->flags & RADEON_IS_IGP) &&
324 rdev->asic->set_pcie_lanes &&
326 rdev->pm.power_state[rdev->pm.current_power_state_index].pcie_lanes)) {
327 radeon_set_pcie_lanes(rdev,
329 DRM_DEBUG_DRIVER("Setting: p: %d\n", ps->pcie_lanes);
333 void r100_pm_prepare(struct radeon_device *rdev)
335 struct drm_device *ddev = rdev->ddev;
336 struct drm_crtc *crtc;
337 struct radeon_crtc *radeon_crtc;
340 /* disable any active CRTCs */
341 list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
342 radeon_crtc = to_radeon_crtc(crtc);
343 if (radeon_crtc->enabled) {
344 if (radeon_crtc->crtc_id) {
345 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
346 tmp |= RADEON_CRTC2_DISP_REQ_EN_B;
347 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
349 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
350 tmp |= RADEON_CRTC_DISP_REQ_EN_B;
351 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
357 void r100_pm_finish(struct radeon_device *rdev)
359 struct drm_device *ddev = rdev->ddev;
360 struct drm_crtc *crtc;
361 struct radeon_crtc *radeon_crtc;
364 /* enable any active CRTCs */
365 list_for_each_entry(crtc, &ddev->mode_config.crtc_list, head) {
366 radeon_crtc = to_radeon_crtc(crtc);
367 if (radeon_crtc->enabled) {
368 if (radeon_crtc->crtc_id) {
369 tmp = RREG32(RADEON_CRTC2_GEN_CNTL);
370 tmp &= ~RADEON_CRTC2_DISP_REQ_EN_B;
371 WREG32(RADEON_CRTC2_GEN_CNTL, tmp);
373 tmp = RREG32(RADEON_CRTC_GEN_CNTL);
374 tmp &= ~RADEON_CRTC_DISP_REQ_EN_B;
375 WREG32(RADEON_CRTC_GEN_CNTL, tmp);
381 bool r100_gui_idle(struct radeon_device *rdev)
383 if (RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_ACTIVE)
389 /* hpd for digital panel detect/disconnect */
390 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
392 bool connected = false;
396 if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
400 if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
409 void r100_hpd_set_polarity(struct radeon_device *rdev,
410 enum radeon_hpd_id hpd)
413 bool connected = r100_hpd_sense(rdev, hpd);
417 tmp = RREG32(RADEON_FP_GEN_CNTL);
419 tmp &= ~RADEON_FP_DETECT_INT_POL;
421 tmp |= RADEON_FP_DETECT_INT_POL;
422 WREG32(RADEON_FP_GEN_CNTL, tmp);
425 tmp = RREG32(RADEON_FP2_GEN_CNTL);
427 tmp &= ~RADEON_FP2_DETECT_INT_POL;
429 tmp |= RADEON_FP2_DETECT_INT_POL;
430 WREG32(RADEON_FP2_GEN_CNTL, tmp);
437 void r100_hpd_init(struct radeon_device *rdev)
439 struct drm_device *dev = rdev->ddev;
440 struct drm_connector *connector;
442 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
443 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
444 switch (radeon_connector->hpd.hpd) {
446 rdev->irq.hpd[0] = true;
449 rdev->irq.hpd[1] = true;
455 if (rdev->irq.installed)
459 void r100_hpd_fini(struct radeon_device *rdev)
461 struct drm_device *dev = rdev->ddev;
462 struct drm_connector *connector;
464 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
465 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
466 switch (radeon_connector->hpd.hpd) {
468 rdev->irq.hpd[0] = false;
471 rdev->irq.hpd[1] = false;
482 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
484 /* TODO: can we do somethings here ? */
485 /* It seems hw only cache one entry so we should discard this
486 * entry otherwise if first GPU GART read hit this entry it
487 * could end up in wrong address. */
490 int r100_pci_gart_init(struct radeon_device *rdev)
494 if (rdev->gart.table.ram.ptr) {
495 WARN(1, "R100 PCI GART already initialized\n");
498 /* Initialize common gart structure */
499 r = radeon_gart_init(rdev);
502 rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
503 rdev->asic->gart_tlb_flush = &r100_pci_gart_tlb_flush;
504 rdev->asic->gart_set_page = &r100_pci_gart_set_page;
505 return radeon_gart_table_ram_alloc(rdev);
508 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
509 void r100_enable_bm(struct radeon_device *rdev)
512 /* Enable bus mastering */
513 tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
514 WREG32(RADEON_BUS_CNTL, tmp);
517 int r100_pci_gart_enable(struct radeon_device *rdev)
521 radeon_gart_restore(rdev);
522 /* discard memory request outside of configured range */
523 tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
524 WREG32(RADEON_AIC_CNTL, tmp);
525 /* set address range for PCI address translate */
526 WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
527 WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
528 /* set PCI GART page-table base address */
529 WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
530 tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
531 WREG32(RADEON_AIC_CNTL, tmp);
532 r100_pci_gart_tlb_flush(rdev);
533 rdev->gart.ready = true;
537 void r100_pci_gart_disable(struct radeon_device *rdev)
541 /* discard memory request outside of configured range */
542 tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
543 WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
544 WREG32(RADEON_AIC_LO_ADDR, 0);
545 WREG32(RADEON_AIC_HI_ADDR, 0);
548 int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
550 if (i < 0 || i > rdev->gart.num_gpu_pages) {
553 rdev->gart.table.ram.ptr[i] = cpu_to_le32(lower_32_bits(addr));
557 void r100_pci_gart_fini(struct radeon_device *rdev)
559 radeon_gart_fini(rdev);
560 r100_pci_gart_disable(rdev);
561 radeon_gart_table_ram_free(rdev);
564 int r100_irq_set(struct radeon_device *rdev)
568 if (!rdev->irq.installed) {
569 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
570 WREG32(R_000040_GEN_INT_CNTL, 0);
573 if (rdev->irq.sw_int) {
574 tmp |= RADEON_SW_INT_ENABLE;
576 if (rdev->irq.gui_idle) {
577 tmp |= RADEON_GUI_IDLE_MASK;
579 if (rdev->irq.crtc_vblank_int[0] ||
580 rdev->irq.pflip[0]) {
581 tmp |= RADEON_CRTC_VBLANK_MASK;
583 if (rdev->irq.crtc_vblank_int[1] ||
584 rdev->irq.pflip[1]) {
585 tmp |= RADEON_CRTC2_VBLANK_MASK;
587 if (rdev->irq.hpd[0]) {
588 tmp |= RADEON_FP_DETECT_MASK;
590 if (rdev->irq.hpd[1]) {
591 tmp |= RADEON_FP2_DETECT_MASK;
593 WREG32(RADEON_GEN_INT_CNTL, tmp);
597 void r100_irq_disable(struct radeon_device *rdev)
601 WREG32(R_000040_GEN_INT_CNTL, 0);
602 /* Wait and acknowledge irq */
604 tmp = RREG32(R_000044_GEN_INT_STATUS);
605 WREG32(R_000044_GEN_INT_STATUS, tmp);
608 static inline uint32_t r100_irq_ack(struct radeon_device *rdev)
610 uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
611 uint32_t irq_mask = RADEON_SW_INT_TEST |
612 RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
613 RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
615 /* the interrupt works, but the status bit is permanently asserted */
616 if (rdev->irq.gui_idle && radeon_gui_idle(rdev)) {
617 if (!rdev->irq.gui_idle_acked)
618 irq_mask |= RADEON_GUI_IDLE_STAT;
622 WREG32(RADEON_GEN_INT_STATUS, irqs);
624 return irqs & irq_mask;
627 int r100_irq_process(struct radeon_device *rdev)
629 uint32_t status, msi_rearm;
630 bool queue_hotplug = false;
632 /* reset gui idle ack. the status bit is broken */
633 rdev->irq.gui_idle_acked = false;
635 status = r100_irq_ack(rdev);
639 if (rdev->shutdown) {
644 if (status & RADEON_SW_INT_TEST) {
645 radeon_fence_process(rdev);
647 /* gui idle interrupt */
648 if (status & RADEON_GUI_IDLE_STAT) {
649 rdev->irq.gui_idle_acked = true;
650 rdev->pm.gui_idle = true;
651 wake_up(&rdev->irq.idle_queue);
653 /* Vertical blank interrupts */
654 if (status & RADEON_CRTC_VBLANK_STAT) {
655 if (rdev->irq.crtc_vblank_int[0]) {
656 drm_handle_vblank(rdev->ddev, 0);
657 rdev->pm.vblank_sync = true;
658 wake_up(&rdev->irq.vblank_queue);
660 if (rdev->irq.pflip[0])
661 radeon_crtc_handle_flip(rdev, 0);
663 if (status & RADEON_CRTC2_VBLANK_STAT) {
664 if (rdev->irq.crtc_vblank_int[1]) {
665 drm_handle_vblank(rdev->ddev, 1);
666 rdev->pm.vblank_sync = true;
667 wake_up(&rdev->irq.vblank_queue);
669 if (rdev->irq.pflip[1])
670 radeon_crtc_handle_flip(rdev, 1);
672 if (status & RADEON_FP_DETECT_STAT) {
673 queue_hotplug = true;
676 if (status & RADEON_FP2_DETECT_STAT) {
677 queue_hotplug = true;
680 status = r100_irq_ack(rdev);
682 /* reset gui idle ack. the status bit is broken */
683 rdev->irq.gui_idle_acked = false;
685 schedule_work(&rdev->hotplug_work);
686 if (rdev->msi_enabled) {
687 switch (rdev->family) {
690 msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
691 WREG32(RADEON_AIC_CNTL, msi_rearm);
692 WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
695 msi_rearm = RREG32(RADEON_MSI_REARM_EN) & ~RV370_MSI_REARM_EN;
696 WREG32(RADEON_MSI_REARM_EN, msi_rearm);
697 WREG32(RADEON_MSI_REARM_EN, msi_rearm | RV370_MSI_REARM_EN);
704 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
707 return RREG32(RADEON_CRTC_CRNT_FRAME);
709 return RREG32(RADEON_CRTC2_CRNT_FRAME);
712 /* Who ever call radeon_fence_emit should call ring_lock and ask
713 * for enough space (today caller are ib schedule and buffer move) */
714 void r100_fence_ring_emit(struct radeon_device *rdev,
715 struct radeon_fence *fence)
717 /* We have to make sure that caches are flushed before
718 * CPU might read something from VRAM. */
719 radeon_ring_write(rdev, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
720 radeon_ring_write(rdev, RADEON_RB3D_DC_FLUSH_ALL);
721 radeon_ring_write(rdev, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
722 radeon_ring_write(rdev, RADEON_RB3D_ZC_FLUSH_ALL);
723 /* Wait until IDLE & CLEAN */
724 radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
725 radeon_ring_write(rdev, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
726 radeon_ring_write(rdev, PACKET0(RADEON_HOST_PATH_CNTL, 0));
727 radeon_ring_write(rdev, rdev->config.r100.hdp_cntl |
728 RADEON_HDP_READ_BUFFER_INVALIDATE);
729 radeon_ring_write(rdev, PACKET0(RADEON_HOST_PATH_CNTL, 0));
730 radeon_ring_write(rdev, rdev->config.r100.hdp_cntl);
731 /* Emit fence sequence & fire IRQ */
732 radeon_ring_write(rdev, PACKET0(rdev->fence_drv.scratch_reg, 0));
733 radeon_ring_write(rdev, fence->seq);
734 radeon_ring_write(rdev, PACKET0(RADEON_GEN_INT_STATUS, 0));
735 radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
738 int r100_copy_blit(struct radeon_device *rdev,
742 struct radeon_fence *fence)
745 uint32_t stride_bytes = PAGE_SIZE;
747 uint32_t stride_pixels;
752 /* radeon limited to 16k stride */
753 stride_bytes &= 0x3fff;
754 /* radeon pitch is /64 */
755 pitch = stride_bytes / 64;
756 stride_pixels = stride_bytes / 4;
757 num_loops = DIV_ROUND_UP(num_pages, 8191);
759 /* Ask for enough room for blit + flush + fence */
760 ndw = 64 + (10 * num_loops);
761 r = radeon_ring_lock(rdev, ndw);
763 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
766 while (num_pages > 0) {
767 cur_pages = num_pages;
768 if (cur_pages > 8191) {
771 num_pages -= cur_pages;
773 /* pages are in Y direction - height
774 page width in X direction - width */
775 radeon_ring_write(rdev, PACKET3(PACKET3_BITBLT_MULTI, 8));
776 radeon_ring_write(rdev,
777 RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
778 RADEON_GMC_DST_PITCH_OFFSET_CNTL |
779 RADEON_GMC_SRC_CLIPPING |
780 RADEON_GMC_DST_CLIPPING |
781 RADEON_GMC_BRUSH_NONE |
782 (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
783 RADEON_GMC_SRC_DATATYPE_COLOR |
785 RADEON_DP_SRC_SOURCE_MEMORY |
786 RADEON_GMC_CLR_CMP_CNTL_DIS |
787 RADEON_GMC_WR_MSK_DIS);
788 radeon_ring_write(rdev, (pitch << 22) | (src_offset >> 10));
789 radeon_ring_write(rdev, (pitch << 22) | (dst_offset >> 10));
790 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
791 radeon_ring_write(rdev, 0);
792 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
793 radeon_ring_write(rdev, num_pages);
794 radeon_ring_write(rdev, num_pages);
795 radeon_ring_write(rdev, cur_pages | (stride_pixels << 16));
797 radeon_ring_write(rdev, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
798 radeon_ring_write(rdev, RADEON_RB2D_DC_FLUSH_ALL);
799 radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
800 radeon_ring_write(rdev,
801 RADEON_WAIT_2D_IDLECLEAN |
802 RADEON_WAIT_HOST_IDLECLEAN |
803 RADEON_WAIT_DMA_GUI_IDLE);
805 r = radeon_fence_emit(rdev, fence);
807 radeon_ring_unlock_commit(rdev);
811 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
816 for (i = 0; i < rdev->usec_timeout; i++) {
817 tmp = RREG32(R_000E40_RBBM_STATUS);
818 if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
826 void r100_ring_start(struct radeon_device *rdev)
830 r = radeon_ring_lock(rdev, 2);
834 radeon_ring_write(rdev, PACKET0(RADEON_ISYNC_CNTL, 0));
835 radeon_ring_write(rdev,
836 RADEON_ISYNC_ANY2D_IDLE3D |
837 RADEON_ISYNC_ANY3D_IDLE2D |
838 RADEON_ISYNC_WAIT_IDLEGUI |
839 RADEON_ISYNC_CPSCRATCH_IDLEGUI);
840 radeon_ring_unlock_commit(rdev);
844 /* Load the microcode for the CP */
845 static int r100_cp_init_microcode(struct radeon_device *rdev)
847 struct platform_device *pdev;
848 const char *fw_name = NULL;
853 pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
856 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
859 if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
860 (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
861 (rdev->family == CHIP_RS200)) {
862 DRM_INFO("Loading R100 Microcode\n");
863 fw_name = FIRMWARE_R100;
864 } else if ((rdev->family == CHIP_R200) ||
865 (rdev->family == CHIP_RV250) ||
866 (rdev->family == CHIP_RV280) ||
867 (rdev->family == CHIP_RS300)) {
868 DRM_INFO("Loading R200 Microcode\n");
869 fw_name = FIRMWARE_R200;
870 } else if ((rdev->family == CHIP_R300) ||
871 (rdev->family == CHIP_R350) ||
872 (rdev->family == CHIP_RV350) ||
873 (rdev->family == CHIP_RV380) ||
874 (rdev->family == CHIP_RS400) ||
875 (rdev->family == CHIP_RS480)) {
876 DRM_INFO("Loading R300 Microcode\n");
877 fw_name = FIRMWARE_R300;
878 } else if ((rdev->family == CHIP_R420) ||
879 (rdev->family == CHIP_R423) ||
880 (rdev->family == CHIP_RV410)) {
881 DRM_INFO("Loading R400 Microcode\n");
882 fw_name = FIRMWARE_R420;
883 } else if ((rdev->family == CHIP_RS690) ||
884 (rdev->family == CHIP_RS740)) {
885 DRM_INFO("Loading RS690/RS740 Microcode\n");
886 fw_name = FIRMWARE_RS690;
887 } else if (rdev->family == CHIP_RS600) {
888 DRM_INFO("Loading RS600 Microcode\n");
889 fw_name = FIRMWARE_RS600;
890 } else if ((rdev->family == CHIP_RV515) ||
891 (rdev->family == CHIP_R520) ||
892 (rdev->family == CHIP_RV530) ||
893 (rdev->family == CHIP_R580) ||
894 (rdev->family == CHIP_RV560) ||
895 (rdev->family == CHIP_RV570)) {
896 DRM_INFO("Loading R500 Microcode\n");
897 fw_name = FIRMWARE_R520;
900 err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
901 platform_device_unregister(pdev);
903 printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
905 } else if (rdev->me_fw->size % 8) {
907 "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
908 rdev->me_fw->size, fw_name);
910 release_firmware(rdev->me_fw);
916 static void r100_cp_load_microcode(struct radeon_device *rdev)
918 const __be32 *fw_data;
921 if (r100_gui_wait_for_idle(rdev)) {
922 printk(KERN_WARNING "Failed to wait GUI idle while "
923 "programming pipes. Bad things might happen.\n");
927 size = rdev->me_fw->size / 4;
928 fw_data = (const __be32 *)&rdev->me_fw->data[0];
929 WREG32(RADEON_CP_ME_RAM_ADDR, 0);
930 for (i = 0; i < size; i += 2) {
931 WREG32(RADEON_CP_ME_RAM_DATAH,
932 be32_to_cpup(&fw_data[i]));
933 WREG32(RADEON_CP_ME_RAM_DATAL,
934 be32_to_cpup(&fw_data[i + 1]));
939 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
944 unsigned pre_write_timer;
945 unsigned pre_write_limit;
946 unsigned indirect2_start;
947 unsigned indirect1_start;
951 if (r100_debugfs_cp_init(rdev)) {
952 DRM_ERROR("Failed to register debugfs file for CP !\n");
955 r = r100_cp_init_microcode(rdev);
957 DRM_ERROR("Failed to load firmware!\n");
962 /* Align ring size */
963 rb_bufsz = drm_order(ring_size / 8);
964 ring_size = (1 << (rb_bufsz + 1)) * 4;
965 r100_cp_load_microcode(rdev);
966 r = radeon_ring_init(rdev, ring_size);
970 /* Each time the cp read 1024 bytes (16 dword/quadword) update
971 * the rptr copy in system ram */
973 /* cp will read 128bytes at a time (4 dwords) */
975 rdev->cp.align_mask = 16 - 1;
976 /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
977 pre_write_timer = 64;
978 /* Force CP_RB_WPTR write if written more than one time before the
982 /* Setup the cp cache like this (cache size is 96 dwords) :
986 * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
987 * indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
988 * indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
989 * Idea being that most of the gpu cmd will be through indirect1 buffer
990 * so it gets the bigger cache.
992 indirect2_start = 80;
993 indirect1_start = 16;
995 WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
996 tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
997 REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
998 REG_SET(RADEON_MAX_FETCH, max_fetch));
1000 tmp |= RADEON_BUF_SWAP_32BIT;
1002 WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
1004 /* Set ring address */
1005 DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)rdev->cp.gpu_addr);
1006 WREG32(RADEON_CP_RB_BASE, rdev->cp.gpu_addr);
1007 /* Force read & write ptr to 0 */
1008 WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
1009 WREG32(RADEON_CP_RB_RPTR_WR, 0);
1010 WREG32(RADEON_CP_RB_WPTR, 0);
1012 /* set the wb address whether it's enabled or not */
1013 WREG32(R_00070C_CP_RB_RPTR_ADDR,
1014 S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
1015 WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
1017 if (rdev->wb.enabled)
1018 WREG32(R_000770_SCRATCH_UMSK, 0xff);
1020 tmp |= RADEON_RB_NO_UPDATE;
1021 WREG32(R_000770_SCRATCH_UMSK, 0);
1024 WREG32(RADEON_CP_RB_CNTL, tmp);
1026 rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
1027 rdev->cp.wptr = RREG32(RADEON_CP_RB_WPTR);
1028 /* protect against crazy HW on resume */
1029 rdev->cp.wptr &= rdev->cp.ptr_mask;
1030 /* Set cp mode to bus mastering & enable cp*/
1031 WREG32(RADEON_CP_CSQ_MODE,
1032 REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
1033 REG_SET(RADEON_INDIRECT1_START, indirect1_start));
1035 WREG32(0x744, 0x00004D4D);
1036 WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
1037 radeon_ring_start(rdev);
1038 r = radeon_ring_test(rdev);
1040 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
1043 rdev->cp.ready = true;
1044 rdev->mc.active_vram_size = rdev->mc.real_vram_size;
1048 void r100_cp_fini(struct radeon_device *rdev)
1050 if (r100_cp_wait_for_idle(rdev)) {
1051 DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
1054 r100_cp_disable(rdev);
1055 radeon_ring_fini(rdev);
1056 DRM_INFO("radeon: cp finalized\n");
1059 void r100_cp_disable(struct radeon_device *rdev)
1062 rdev->mc.active_vram_size = rdev->mc.visible_vram_size;
1063 rdev->cp.ready = false;
1064 WREG32(RADEON_CP_CSQ_MODE, 0);
1065 WREG32(RADEON_CP_CSQ_CNTL, 0);
1066 WREG32(R_000770_SCRATCH_UMSK, 0);
1067 if (r100_gui_wait_for_idle(rdev)) {
1068 printk(KERN_WARNING "Failed to wait GUI idle while "
1069 "programming pipes. Bad things might happen.\n");
1073 void r100_cp_commit(struct radeon_device *rdev)
1075 WREG32(RADEON_CP_RB_WPTR, rdev->cp.wptr);
1076 (void)RREG32(RADEON_CP_RB_WPTR);
1083 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
1084 struct radeon_cs_packet *pkt,
1085 const unsigned *auth, unsigned n,
1086 radeon_packet0_check_t check)
1095 /* Check that register fall into register range
1096 * determined by the number of entry (n) in the
1097 * safe register bitmap.
1099 if (pkt->one_reg_wr) {
1100 if ((reg >> 7) > n) {
1104 if (((reg + (pkt->count << 2)) >> 7) > n) {
1108 for (i = 0; i <= pkt->count; i++, idx++) {
1110 m = 1 << ((reg >> 2) & 31);
1112 r = check(p, pkt, idx, reg);
1117 if (pkt->one_reg_wr) {
1118 if (!(auth[j] & m)) {
1128 void r100_cs_dump_packet(struct radeon_cs_parser *p,
1129 struct radeon_cs_packet *pkt)
1131 volatile uint32_t *ib;
1137 for (i = 0; i <= (pkt->count + 1); i++, idx++) {
1138 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
1143 * r100_cs_packet_parse() - parse cp packet and point ib index to next packet
1144 * @parser: parser structure holding parsing context.
1145 * @pkt: where to store packet informations
1147 * Assume that chunk_ib_index is properly set. Will return -EINVAL
1148 * if packet is bigger than remaining ib size. or if packets is unknown.
1150 int r100_cs_packet_parse(struct radeon_cs_parser *p,
1151 struct radeon_cs_packet *pkt,
1154 struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
1157 if (idx >= ib_chunk->length_dw) {
1158 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
1159 idx, ib_chunk->length_dw);
1162 header = radeon_get_ib_value(p, idx);
1164 pkt->type = CP_PACKET_GET_TYPE(header);
1165 pkt->count = CP_PACKET_GET_COUNT(header);
1166 switch (pkt->type) {
1168 pkt->reg = CP_PACKET0_GET_REG(header);
1169 pkt->one_reg_wr = CP_PACKET0_GET_ONE_REG_WR(header);
1172 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
1178 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
1181 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
1182 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
1183 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
1190 * r100_cs_packet_next_vline() - parse userspace VLINE packet
1191 * @parser: parser structure holding parsing context.
1193 * Userspace sends a special sequence for VLINE waits.
1194 * PACKET0 - VLINE_START_END + value
1195 * PACKET0 - WAIT_UNTIL +_value
1196 * RELOC (P3) - crtc_id in reloc.
1198 * This function parses this and relocates the VLINE START END
1199 * and WAIT UNTIL packets to the correct crtc.
1200 * It also detects a switched off crtc and nulls out the
1201 * wait in that case.
1203 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
1205 struct drm_mode_object *obj;
1206 struct drm_crtc *crtc;
1207 struct radeon_crtc *radeon_crtc;
1208 struct radeon_cs_packet p3reloc, waitreloc;
1211 uint32_t header, h_idx, reg;
1212 volatile uint32_t *ib;
1216 /* parse the wait until */
1217 r = r100_cs_packet_parse(p, &waitreloc, p->idx);
1221 /* check its a wait until and only 1 count */
1222 if (waitreloc.reg != RADEON_WAIT_UNTIL ||
1223 waitreloc.count != 0) {
1224 DRM_ERROR("vline wait had illegal wait until segment\n");
1229 if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
1230 DRM_ERROR("vline wait had illegal wait until\n");
1235 /* jump over the NOP */
1236 r = r100_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
1241 p->idx += waitreloc.count + 2;
1242 p->idx += p3reloc.count + 2;
1244 header = radeon_get_ib_value(p, h_idx);
1245 crtc_id = radeon_get_ib_value(p, h_idx + 5);
1246 reg = CP_PACKET0_GET_REG(header);
1247 obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
1249 DRM_ERROR("cannot find crtc %d\n", crtc_id);
1253 crtc = obj_to_crtc(obj);
1254 radeon_crtc = to_radeon_crtc(crtc);
1255 crtc_id = radeon_crtc->crtc_id;
1257 if (!crtc->enabled) {
1258 /* if the CRTC isn't enabled - we need to nop out the wait until */
1259 ib[h_idx + 2] = PACKET2(0);
1260 ib[h_idx + 3] = PACKET2(0);
1261 } else if (crtc_id == 1) {
1263 case AVIVO_D1MODE_VLINE_START_END:
1264 header &= ~R300_CP_PACKET0_REG_MASK;
1265 header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1267 case RADEON_CRTC_GUI_TRIG_VLINE:
1268 header &= ~R300_CP_PACKET0_REG_MASK;
1269 header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1272 DRM_ERROR("unknown crtc reloc\n");
1277 ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1284 * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
1285 * @parser: parser structure holding parsing context.
1286 * @data: pointer to relocation data
1287 * @offset_start: starting offset
1288 * @offset_mask: offset mask (to align start offset on)
1289 * @reloc: reloc informations
1291 * Check next packet is relocation packet3, do bo validation and compute
1292 * GPU offset using the provided start.
1294 int r100_cs_packet_next_reloc(struct radeon_cs_parser *p,
1295 struct radeon_cs_reloc **cs_reloc)
1297 struct radeon_cs_chunk *relocs_chunk;
1298 struct radeon_cs_packet p3reloc;
1302 if (p->chunk_relocs_idx == -1) {
1303 DRM_ERROR("No relocation chunk !\n");
1307 relocs_chunk = &p->chunks[p->chunk_relocs_idx];
1308 r = r100_cs_packet_parse(p, &p3reloc, p->idx);
1312 p->idx += p3reloc.count + 2;
1313 if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
1314 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
1316 r100_cs_dump_packet(p, &p3reloc);
1319 idx = radeon_get_ib_value(p, p3reloc.idx + 1);
1320 if (idx >= relocs_chunk->length_dw) {
1321 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
1322 idx, relocs_chunk->length_dw);
1323 r100_cs_dump_packet(p, &p3reloc);
1326 /* FIXME: we assume reloc size is 4 dwords */
1327 *cs_reloc = p->relocs_ptr[(idx / 4)];
1331 static int r100_get_vtx_size(uint32_t vtx_fmt)
1335 /* ordered according to bits in spec */
1336 if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1338 if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1340 if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1342 if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1344 if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1346 if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1348 if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1350 if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1352 if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1354 if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1356 if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1358 if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1360 if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1362 if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1364 if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1367 if (vtx_fmt & (0x7 << 15))
1368 vtx_size += (vtx_fmt >> 15) & 0x7;
1369 if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1371 if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1373 if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1375 if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1377 if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1379 if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1384 static int r100_packet0_check(struct radeon_cs_parser *p,
1385 struct radeon_cs_packet *pkt,
1386 unsigned idx, unsigned reg)
1388 struct radeon_cs_reloc *reloc;
1389 struct r100_cs_track *track;
1390 volatile uint32_t *ib;
1398 track = (struct r100_cs_track *)p->track;
1400 idx_value = radeon_get_ib_value(p, idx);
1403 case RADEON_CRTC_GUI_TRIG_VLINE:
1404 r = r100_cs_packet_parse_vline(p);
1406 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1408 r100_cs_dump_packet(p, pkt);
1412 /* FIXME: only allow PACKET3 blit? easier to check for out of
1414 case RADEON_DST_PITCH_OFFSET:
1415 case RADEON_SRC_PITCH_OFFSET:
1416 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1420 case RADEON_RB3D_DEPTHOFFSET:
1421 r = r100_cs_packet_next_reloc(p, &reloc);
1423 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1425 r100_cs_dump_packet(p, pkt);
1428 track->zb.robj = reloc->robj;
1429 track->zb.offset = idx_value;
1430 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1432 case RADEON_RB3D_COLOROFFSET:
1433 r = r100_cs_packet_next_reloc(p, &reloc);
1435 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1437 r100_cs_dump_packet(p, pkt);
1440 track->cb[0].robj = reloc->robj;
1441 track->cb[0].offset = idx_value;
1442 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1444 case RADEON_PP_TXOFFSET_0:
1445 case RADEON_PP_TXOFFSET_1:
1446 case RADEON_PP_TXOFFSET_2:
1447 i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1448 r = r100_cs_packet_next_reloc(p, &reloc);
1450 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1452 r100_cs_dump_packet(p, pkt);
1455 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1456 track->textures[i].robj = reloc->robj;
1458 case RADEON_PP_CUBIC_OFFSET_T0_0:
1459 case RADEON_PP_CUBIC_OFFSET_T0_1:
1460 case RADEON_PP_CUBIC_OFFSET_T0_2:
1461 case RADEON_PP_CUBIC_OFFSET_T0_3:
1462 case RADEON_PP_CUBIC_OFFSET_T0_4:
1463 i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1464 r = r100_cs_packet_next_reloc(p, &reloc);
1466 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1468 r100_cs_dump_packet(p, pkt);
1471 track->textures[0].cube_info[i].offset = idx_value;
1472 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1473 track->textures[0].cube_info[i].robj = reloc->robj;
1475 case RADEON_PP_CUBIC_OFFSET_T1_0:
1476 case RADEON_PP_CUBIC_OFFSET_T1_1:
1477 case RADEON_PP_CUBIC_OFFSET_T1_2:
1478 case RADEON_PP_CUBIC_OFFSET_T1_3:
1479 case RADEON_PP_CUBIC_OFFSET_T1_4:
1480 i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1481 r = r100_cs_packet_next_reloc(p, &reloc);
1483 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1485 r100_cs_dump_packet(p, pkt);
1488 track->textures[1].cube_info[i].offset = idx_value;
1489 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1490 track->textures[1].cube_info[i].robj = reloc->robj;
1492 case RADEON_PP_CUBIC_OFFSET_T2_0:
1493 case RADEON_PP_CUBIC_OFFSET_T2_1:
1494 case RADEON_PP_CUBIC_OFFSET_T2_2:
1495 case RADEON_PP_CUBIC_OFFSET_T2_3:
1496 case RADEON_PP_CUBIC_OFFSET_T2_4:
1497 i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1498 r = r100_cs_packet_next_reloc(p, &reloc);
1500 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1502 r100_cs_dump_packet(p, pkt);
1505 track->textures[2].cube_info[i].offset = idx_value;
1506 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1507 track->textures[2].cube_info[i].robj = reloc->robj;
1509 case RADEON_RE_WIDTH_HEIGHT:
1510 track->maxy = ((idx_value >> 16) & 0x7FF);
1512 case RADEON_RB3D_COLORPITCH:
1513 r = r100_cs_packet_next_reloc(p, &reloc);
1515 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1517 r100_cs_dump_packet(p, pkt);
1521 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1522 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1523 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1524 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1526 tmp = idx_value & ~(0x7 << 16);
1530 track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1532 case RADEON_RB3D_DEPTHPITCH:
1533 track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1535 case RADEON_RB3D_CNTL:
1536 switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1542 track->cb[0].cpp = 1;
1547 track->cb[0].cpp = 2;
1550 track->cb[0].cpp = 4;
1553 DRM_ERROR("Invalid color buffer format (%d) !\n",
1554 ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1557 track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1559 case RADEON_RB3D_ZSTENCILCNTL:
1560 switch (idx_value & 0xf) {
1576 case RADEON_RB3D_ZPASS_ADDR:
1577 r = r100_cs_packet_next_reloc(p, &reloc);
1579 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1581 r100_cs_dump_packet(p, pkt);
1584 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1586 case RADEON_PP_CNTL:
1588 uint32_t temp = idx_value >> 4;
1589 for (i = 0; i < track->num_texture; i++)
1590 track->textures[i].enabled = !!(temp & (1 << i));
1593 case RADEON_SE_VF_CNTL:
1594 track->vap_vf_cntl = idx_value;
1596 case RADEON_SE_VTX_FMT:
1597 track->vtx_size = r100_get_vtx_size(idx_value);
1599 case RADEON_PP_TEX_SIZE_0:
1600 case RADEON_PP_TEX_SIZE_1:
1601 case RADEON_PP_TEX_SIZE_2:
1602 i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1603 track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1604 track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1606 case RADEON_PP_TEX_PITCH_0:
1607 case RADEON_PP_TEX_PITCH_1:
1608 case RADEON_PP_TEX_PITCH_2:
1609 i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1610 track->textures[i].pitch = idx_value + 32;
1612 case RADEON_PP_TXFILTER_0:
1613 case RADEON_PP_TXFILTER_1:
1614 case RADEON_PP_TXFILTER_2:
1615 i = (reg - RADEON_PP_TXFILTER_0) / 24;
1616 track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1617 >> RADEON_MAX_MIP_LEVEL_SHIFT);
1618 tmp = (idx_value >> 23) & 0x7;
1619 if (tmp == 2 || tmp == 6)
1620 track->textures[i].roundup_w = false;
1621 tmp = (idx_value >> 27) & 0x7;
1622 if (tmp == 2 || tmp == 6)
1623 track->textures[i].roundup_h = false;
1625 case RADEON_PP_TXFORMAT_0:
1626 case RADEON_PP_TXFORMAT_1:
1627 case RADEON_PP_TXFORMAT_2:
1628 i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1629 if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1630 track->textures[i].use_pitch = 1;
1632 track->textures[i].use_pitch = 0;
1633 track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1634 track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1636 if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1637 track->textures[i].tex_coord_type = 2;
1638 switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1639 case RADEON_TXFORMAT_I8:
1640 case RADEON_TXFORMAT_RGB332:
1641 case RADEON_TXFORMAT_Y8:
1642 track->textures[i].cpp = 1;
1643 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1645 case RADEON_TXFORMAT_AI88:
1646 case RADEON_TXFORMAT_ARGB1555:
1647 case RADEON_TXFORMAT_RGB565:
1648 case RADEON_TXFORMAT_ARGB4444:
1649 case RADEON_TXFORMAT_VYUY422:
1650 case RADEON_TXFORMAT_YVYU422:
1651 case RADEON_TXFORMAT_SHADOW16:
1652 case RADEON_TXFORMAT_LDUDV655:
1653 case RADEON_TXFORMAT_DUDV88:
1654 track->textures[i].cpp = 2;
1655 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1657 case RADEON_TXFORMAT_ARGB8888:
1658 case RADEON_TXFORMAT_RGBA8888:
1659 case RADEON_TXFORMAT_SHADOW32:
1660 case RADEON_TXFORMAT_LDUDUV8888:
1661 track->textures[i].cpp = 4;
1662 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
1664 case RADEON_TXFORMAT_DXT1:
1665 track->textures[i].cpp = 1;
1666 track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1668 case RADEON_TXFORMAT_DXT23:
1669 case RADEON_TXFORMAT_DXT45:
1670 track->textures[i].cpp = 1;
1671 track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1674 track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1675 track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1677 case RADEON_PP_CUBIC_FACES_0:
1678 case RADEON_PP_CUBIC_FACES_1:
1679 case RADEON_PP_CUBIC_FACES_2:
1681 i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1682 for (face = 0; face < 4; face++) {
1683 track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1684 track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1688 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1695 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1696 struct radeon_cs_packet *pkt,
1697 struct radeon_bo *robj)
1702 value = radeon_get_ib_value(p, idx + 2);
1703 if ((value + 1) > radeon_bo_size(robj)) {
1704 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1705 "(need %u have %lu) !\n",
1707 radeon_bo_size(robj));
1713 static int r100_packet3_check(struct radeon_cs_parser *p,
1714 struct radeon_cs_packet *pkt)
1716 struct radeon_cs_reloc *reloc;
1717 struct r100_cs_track *track;
1719 volatile uint32_t *ib;
1724 track = (struct r100_cs_track *)p->track;
1725 switch (pkt->opcode) {
1726 case PACKET3_3D_LOAD_VBPNTR:
1727 r = r100_packet3_load_vbpntr(p, pkt, idx);
1731 case PACKET3_INDX_BUFFER:
1732 r = r100_cs_packet_next_reloc(p, &reloc);
1734 DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1735 r100_cs_dump_packet(p, pkt);
1738 ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
1739 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1745 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1746 r = r100_cs_packet_next_reloc(p, &reloc);
1748 DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1749 r100_cs_dump_packet(p, pkt);
1752 ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
1753 track->num_arrays = 1;
1754 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1756 track->arrays[0].robj = reloc->robj;
1757 track->arrays[0].esize = track->vtx_size;
1759 track->max_indx = radeon_get_ib_value(p, idx+1);
1761 track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1762 track->immd_dwords = pkt->count - 1;
1763 r = r100_cs_track_check(p->rdev, track);
1767 case PACKET3_3D_DRAW_IMMD:
1768 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1769 DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1772 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1773 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1774 track->immd_dwords = pkt->count - 1;
1775 r = r100_cs_track_check(p->rdev, track);
1779 /* triggers drawing using in-packet vertex data */
1780 case PACKET3_3D_DRAW_IMMD_2:
1781 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1782 DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1785 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1786 track->immd_dwords = pkt->count;
1787 r = r100_cs_track_check(p->rdev, track);
1791 /* triggers drawing using in-packet vertex data */
1792 case PACKET3_3D_DRAW_VBUF_2:
1793 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1794 r = r100_cs_track_check(p->rdev, track);
1798 /* triggers drawing of vertex buffers setup elsewhere */
1799 case PACKET3_3D_DRAW_INDX_2:
1800 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1801 r = r100_cs_track_check(p->rdev, track);
1805 /* triggers drawing using indices to vertex buffer */
1806 case PACKET3_3D_DRAW_VBUF:
1807 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1808 r = r100_cs_track_check(p->rdev, track);
1812 /* triggers drawing of vertex buffers setup elsewhere */
1813 case PACKET3_3D_DRAW_INDX:
1814 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1815 r = r100_cs_track_check(p->rdev, track);
1819 /* triggers drawing using indices to vertex buffer */
1820 case PACKET3_3D_CLEAR_HIZ:
1821 case PACKET3_3D_CLEAR_ZMASK:
1822 if (p->rdev->hyperz_filp != p->filp)
1828 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1834 int r100_cs_parse(struct radeon_cs_parser *p)
1836 struct radeon_cs_packet pkt;
1837 struct r100_cs_track *track;
1840 track = kzalloc(sizeof(*track), GFP_KERNEL);
1841 r100_cs_track_clear(p->rdev, track);
1844 r = r100_cs_packet_parse(p, &pkt, p->idx);
1848 p->idx += pkt.count + 2;
1851 if (p->rdev->family >= CHIP_R200)
1852 r = r100_cs_parse_packet0(p, &pkt,
1853 p->rdev->config.r100.reg_safe_bm,
1854 p->rdev->config.r100.reg_safe_bm_size,
1855 &r200_packet0_check);
1857 r = r100_cs_parse_packet0(p, &pkt,
1858 p->rdev->config.r100.reg_safe_bm,
1859 p->rdev->config.r100.reg_safe_bm_size,
1860 &r100_packet0_check);
1865 r = r100_packet3_check(p, &pkt);
1868 DRM_ERROR("Unknown packet type %d !\n",
1875 } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1881 * Global GPU functions
1883 void r100_errata(struct radeon_device *rdev)
1885 rdev->pll_errata = 0;
1887 if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
1888 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
1891 if (rdev->family == CHIP_RV100 ||
1892 rdev->family == CHIP_RS100 ||
1893 rdev->family == CHIP_RS200) {
1894 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
1898 /* Wait for vertical sync on primary CRTC */
1899 void r100_gpu_wait_for_vsync(struct radeon_device *rdev)
1901 uint32_t crtc_gen_cntl, tmp;
1904 crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
1905 if ((crtc_gen_cntl & RADEON_CRTC_DISP_REQ_EN_B) ||
1906 !(crtc_gen_cntl & RADEON_CRTC_EN)) {
1909 /* Clear the CRTC_VBLANK_SAVE bit */
1910 WREG32(RADEON_CRTC_STATUS, RADEON_CRTC_VBLANK_SAVE_CLEAR);
1911 for (i = 0; i < rdev->usec_timeout; i++) {
1912 tmp = RREG32(RADEON_CRTC_STATUS);
1913 if (tmp & RADEON_CRTC_VBLANK_SAVE) {
1920 /* Wait for vertical sync on secondary CRTC */
1921 void r100_gpu_wait_for_vsync2(struct radeon_device *rdev)
1923 uint32_t crtc2_gen_cntl, tmp;
1926 crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
1927 if ((crtc2_gen_cntl & RADEON_CRTC2_DISP_REQ_EN_B) ||
1928 !(crtc2_gen_cntl & RADEON_CRTC2_EN))
1931 /* Clear the CRTC_VBLANK_SAVE bit */
1932 WREG32(RADEON_CRTC2_STATUS, RADEON_CRTC2_VBLANK_SAVE_CLEAR);
1933 for (i = 0; i < rdev->usec_timeout; i++) {
1934 tmp = RREG32(RADEON_CRTC2_STATUS);
1935 if (tmp & RADEON_CRTC2_VBLANK_SAVE) {
1942 int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
1947 for (i = 0; i < rdev->usec_timeout; i++) {
1948 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
1957 int r100_gui_wait_for_idle(struct radeon_device *rdev)
1962 if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
1963 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
1964 " Bad things might happen.\n");
1966 for (i = 0; i < rdev->usec_timeout; i++) {
1967 tmp = RREG32(RADEON_RBBM_STATUS);
1968 if (!(tmp & RADEON_RBBM_ACTIVE)) {
1976 int r100_mc_wait_for_idle(struct radeon_device *rdev)
1981 for (i = 0; i < rdev->usec_timeout; i++) {
1982 /* read MC_STATUS */
1983 tmp = RREG32(RADEON_MC_STATUS);
1984 if (tmp & RADEON_MC_IDLE) {
1992 void r100_gpu_lockup_update(struct r100_gpu_lockup *lockup, struct radeon_cp *cp)
1994 lockup->last_cp_rptr = cp->rptr;
1995 lockup->last_jiffies = jiffies;
1999 * r100_gpu_cp_is_lockup() - check if CP is lockup by recording information
2000 * @rdev: radeon device structure
2001 * @lockup: r100_gpu_lockup structure holding CP lockup tracking informations
2002 * @cp: radeon_cp structure holding CP information
2004 * We don't need to initialize the lockup tracking information as we will either
2005 * have CP rptr to a different value of jiffies wrap around which will force
2006 * initialization of the lockup tracking informations.
2008 * A possible false positivie is if we get call after while and last_cp_rptr ==
2009 * the current CP rptr, even if it's unlikely it might happen. To avoid this
2010 * if the elapsed time since last call is bigger than 2 second than we return
2011 * false and update the tracking information. Due to this the caller must call
2012 * r100_gpu_cp_is_lockup several time in less than 2sec for lockup to be reported
2013 * the fencing code should be cautious about that.
2015 * Caller should write to the ring to force CP to do something so we don't get
2016 * false positive when CP is just gived nothing to do.
2019 bool r100_gpu_cp_is_lockup(struct radeon_device *rdev, struct r100_gpu_lockup *lockup, struct radeon_cp *cp)
2021 unsigned long cjiffies, elapsed;
2024 if (!time_after(cjiffies, lockup->last_jiffies)) {
2025 /* likely a wrap around */
2026 lockup->last_cp_rptr = cp->rptr;
2027 lockup->last_jiffies = jiffies;
2030 if (cp->rptr != lockup->last_cp_rptr) {
2031 /* CP is still working no lockup */
2032 lockup->last_cp_rptr = cp->rptr;
2033 lockup->last_jiffies = jiffies;
2036 elapsed = jiffies_to_msecs(cjiffies - lockup->last_jiffies);
2037 if (elapsed >= 10000) {
2038 dev_err(rdev->dev, "GPU lockup CP stall for more than %lumsec\n", elapsed);
2041 /* give a chance to the GPU ... */
2045 bool r100_gpu_is_lockup(struct radeon_device *rdev)
2050 rbbm_status = RREG32(R_000E40_RBBM_STATUS);
2051 if (!G_000E40_GUI_ACTIVE(rbbm_status)) {
2052 r100_gpu_lockup_update(&rdev->config.r100.lockup, &rdev->cp);
2055 /* force CP activities */
2056 r = radeon_ring_lock(rdev, 2);
2059 radeon_ring_write(rdev, 0x80000000);
2060 radeon_ring_write(rdev, 0x80000000);
2061 radeon_ring_unlock_commit(rdev);
2063 rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
2064 return r100_gpu_cp_is_lockup(rdev, &rdev->config.r100.lockup, &rdev->cp);
2067 void r100_bm_disable(struct radeon_device *rdev)
2071 /* disable bus mastering */
2072 tmp = RREG32(R_000030_BUS_CNTL);
2073 WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000044);
2075 WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000042);
2077 WREG32(R_000030_BUS_CNTL, (tmp & 0xFFFFFFFF) | 0x00000040);
2078 tmp = RREG32(RADEON_BUS_CNTL);
2080 pci_read_config_word(rdev->pdev, 0x4, (u16*)&tmp);
2081 pci_write_config_word(rdev->pdev, 0x4, tmp & 0xFFFB);
2085 int r100_asic_reset(struct radeon_device *rdev)
2087 struct r100_mc_save save;
2090 r100_mc_stop(rdev, &save);
2091 status = RREG32(R_000E40_RBBM_STATUS);
2092 if (!G_000E40_GUI_ACTIVE(status)) {
2095 status = RREG32(R_000E40_RBBM_STATUS);
2096 dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2098 WREG32(RADEON_CP_CSQ_CNTL, 0);
2099 tmp = RREG32(RADEON_CP_RB_CNTL);
2100 WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
2101 WREG32(RADEON_CP_RB_RPTR_WR, 0);
2102 WREG32(RADEON_CP_RB_WPTR, 0);
2103 WREG32(RADEON_CP_RB_CNTL, tmp);
2104 /* save PCI state */
2105 pci_save_state(rdev->pdev);
2106 /* disable bus mastering */
2107 r100_bm_disable(rdev);
2108 WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_SE(1) |
2109 S_0000F0_SOFT_RESET_RE(1) |
2110 S_0000F0_SOFT_RESET_PP(1) |
2111 S_0000F0_SOFT_RESET_RB(1));
2112 RREG32(R_0000F0_RBBM_SOFT_RESET);
2114 WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2116 status = RREG32(R_000E40_RBBM_STATUS);
2117 dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2119 WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
2120 RREG32(R_0000F0_RBBM_SOFT_RESET);
2122 WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
2124 status = RREG32(R_000E40_RBBM_STATUS);
2125 dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
2126 /* restore PCI & busmastering */
2127 pci_restore_state(rdev->pdev);
2128 r100_enable_bm(rdev);
2129 /* Check if GPU is idle */
2130 if (G_000E40_SE_BUSY(status) || G_000E40_RE_BUSY(status) ||
2131 G_000E40_TAM_BUSY(status) || G_000E40_PB_BUSY(status)) {
2132 dev_err(rdev->dev, "failed to reset GPU\n");
2133 rdev->gpu_lockup = true;
2136 r100_mc_resume(rdev, &save);
2137 dev_info(rdev->dev, "GPU reset succeed\n");
2141 void r100_set_common_regs(struct radeon_device *rdev)
2143 struct drm_device *dev = rdev->ddev;
2144 bool force_dac2 = false;
2147 /* set these so they don't interfere with anything */
2148 WREG32(RADEON_OV0_SCALE_CNTL, 0);
2149 WREG32(RADEON_SUBPIC_CNTL, 0);
2150 WREG32(RADEON_VIPH_CONTROL, 0);
2151 WREG32(RADEON_I2C_CNTL_1, 0);
2152 WREG32(RADEON_DVI_I2C_CNTL_1, 0);
2153 WREG32(RADEON_CAP0_TRIG_CNTL, 0);
2154 WREG32(RADEON_CAP1_TRIG_CNTL, 0);
2156 /* always set up dac2 on rn50 and some rv100 as lots
2157 * of servers seem to wire it up to a VGA port but
2158 * don't report it in the bios connector
2161 switch (dev->pdev->device) {
2170 /* DELL triple head servers */
2171 if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
2172 ((dev->pdev->subsystem_device == 0x016c) ||
2173 (dev->pdev->subsystem_device == 0x016d) ||
2174 (dev->pdev->subsystem_device == 0x016e) ||
2175 (dev->pdev->subsystem_device == 0x016f) ||
2176 (dev->pdev->subsystem_device == 0x0170) ||
2177 (dev->pdev->subsystem_device == 0x017d) ||
2178 (dev->pdev->subsystem_device == 0x017e) ||
2179 (dev->pdev->subsystem_device == 0x0183) ||
2180 (dev->pdev->subsystem_device == 0x018a) ||
2181 (dev->pdev->subsystem_device == 0x019a)))
2187 u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
2188 u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
2189 u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
2191 /* For CRT on DAC2, don't turn it on if BIOS didn't
2192 enable it, even it's detected.
2195 /* force it to crtc0 */
2196 dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
2197 dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
2198 disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
2200 /* set up the TV DAC */
2201 tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
2202 RADEON_TV_DAC_STD_MASK |
2203 RADEON_TV_DAC_RDACPD |
2204 RADEON_TV_DAC_GDACPD |
2205 RADEON_TV_DAC_BDACPD |
2206 RADEON_TV_DAC_BGADJ_MASK |
2207 RADEON_TV_DAC_DACADJ_MASK);
2208 tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
2209 RADEON_TV_DAC_NHOLD |
2210 RADEON_TV_DAC_STD_PS2 |
2213 WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
2214 WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
2215 WREG32(RADEON_DAC_CNTL2, dac2_cntl);
2218 /* switch PM block to ACPI mode */
2219 tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
2220 tmp &= ~RADEON_PM_MODE_SEL;
2221 WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
2228 static void r100_vram_get_type(struct radeon_device *rdev)
2232 rdev->mc.vram_is_ddr = false;
2233 if (rdev->flags & RADEON_IS_IGP)
2234 rdev->mc.vram_is_ddr = true;
2235 else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
2236 rdev->mc.vram_is_ddr = true;
2237 if ((rdev->family == CHIP_RV100) ||
2238 (rdev->family == CHIP_RS100) ||
2239 (rdev->family == CHIP_RS200)) {
2240 tmp = RREG32(RADEON_MEM_CNTL);
2241 if (tmp & RV100_HALF_MODE) {
2242 rdev->mc.vram_width = 32;
2244 rdev->mc.vram_width = 64;
2246 if (rdev->flags & RADEON_SINGLE_CRTC) {
2247 rdev->mc.vram_width /= 4;
2248 rdev->mc.vram_is_ddr = true;
2250 } else if (rdev->family <= CHIP_RV280) {
2251 tmp = RREG32(RADEON_MEM_CNTL);
2252 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
2253 rdev->mc.vram_width = 128;
2255 rdev->mc.vram_width = 64;
2259 rdev->mc.vram_width = 128;
2263 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
2268 aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2270 /* Set HDP_APER_CNTL only on cards that are known not to be broken,
2271 * that is has the 2nd generation multifunction PCI interface
2273 if (rdev->family == CHIP_RV280 ||
2274 rdev->family >= CHIP_RV350) {
2275 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
2276 ~RADEON_HDP_APER_CNTL);
2277 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
2278 return aper_size * 2;
2281 /* Older cards have all sorts of funny issues to deal with. First
2282 * check if it's a multifunction card by reading the PCI config
2283 * header type... Limit those to one aperture size
2285 pci_read_config_byte(rdev->pdev, 0xe, &byte);
2287 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
2288 DRM_INFO("Limiting VRAM to one aperture\n");
2292 /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
2293 * have set it up. We don't write this as it's broken on some ASICs but
2294 * we expect the BIOS to have done the right thing (might be too optimistic...)
2296 if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
2297 return aper_size * 2;
2301 void r100_vram_init_sizes(struct radeon_device *rdev)
2303 u64 config_aper_size;
2305 /* work out accessible VRAM */
2306 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2307 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2308 rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
2309 /* FIXME we don't use the second aperture yet when we could use it */
2310 if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
2311 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2312 rdev->mc.active_vram_size = rdev->mc.visible_vram_size;
2313 config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
2314 if (rdev->flags & RADEON_IS_IGP) {
2316 /* read NB_TOM to get the amount of ram stolen for the GPU */
2317 tom = RREG32(RADEON_NB_TOM);
2318 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
2319 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2320 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2322 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
2323 /* Some production boards of m6 will report 0
2326 if (rdev->mc.real_vram_size == 0) {
2327 rdev->mc.real_vram_size = 8192 * 1024;
2328 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
2330 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
2331 * Novell bug 204882 + along with lots of ubuntu ones
2333 if (rdev->mc.aper_size > config_aper_size)
2334 config_aper_size = rdev->mc.aper_size;
2336 if (config_aper_size > rdev->mc.real_vram_size)
2337 rdev->mc.mc_vram_size = config_aper_size;
2339 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2343 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2347 temp = RREG32(RADEON_CONFIG_CNTL);
2348 if (state == false) {
2354 WREG32(RADEON_CONFIG_CNTL, temp);
2357 void r100_mc_init(struct radeon_device *rdev)
2361 r100_vram_get_type(rdev);
2362 r100_vram_init_sizes(rdev);
2363 base = rdev->mc.aper_base;
2364 if (rdev->flags & RADEON_IS_IGP)
2365 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2366 radeon_vram_location(rdev, &rdev->mc, base);
2367 rdev->mc.gtt_base_align = 0;
2368 if (!(rdev->flags & RADEON_IS_AGP))
2369 radeon_gtt_location(rdev, &rdev->mc);
2370 radeon_update_bandwidth_info(rdev);
2375 * Indirect registers accessor
2377 void r100_pll_errata_after_index(struct radeon_device *rdev)
2379 if (rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS) {
2380 (void)RREG32(RADEON_CLOCK_CNTL_DATA);
2381 (void)RREG32(RADEON_CRTC_GEN_CNTL);
2385 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2387 /* This workarounds is necessary on RV100, RS100 and RS200 chips
2388 * or the chip could hang on a subsequent access
2390 if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2394 /* This function is required to workaround a hardware bug in some (all?)
2395 * revisions of the R300. This workaround should be called after every
2396 * CLOCK_CNTL_INDEX register access. If not, register reads afterward
2397 * may not be correct.
2399 if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2402 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2403 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2404 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2405 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2406 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2410 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2414 WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2415 r100_pll_errata_after_index(rdev);
2416 data = RREG32(RADEON_CLOCK_CNTL_DATA);
2417 r100_pll_errata_after_data(rdev);
2421 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2423 WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2424 r100_pll_errata_after_index(rdev);
2425 WREG32(RADEON_CLOCK_CNTL_DATA, v);
2426 r100_pll_errata_after_data(rdev);
2429 void r100_set_safe_registers(struct radeon_device *rdev)
2431 if (ASIC_IS_RN50(rdev)) {
2432 rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2433 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2434 } else if (rdev->family < CHIP_R200) {
2435 rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2436 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2438 r200_set_safe_registers(rdev);
2445 #if defined(CONFIG_DEBUG_FS)
2446 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2448 struct drm_info_node *node = (struct drm_info_node *) m->private;
2449 struct drm_device *dev = node->minor->dev;
2450 struct radeon_device *rdev = dev->dev_private;
2451 uint32_t reg, value;
2454 seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2455 seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2456 seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2457 for (i = 0; i < 64; i++) {
2458 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2459 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2460 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2461 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2462 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2467 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2469 struct drm_info_node *node = (struct drm_info_node *) m->private;
2470 struct drm_device *dev = node->minor->dev;
2471 struct radeon_device *rdev = dev->dev_private;
2473 unsigned count, i, j;
2475 radeon_ring_free_size(rdev);
2476 rdp = RREG32(RADEON_CP_RB_RPTR);
2477 wdp = RREG32(RADEON_CP_RB_WPTR);
2478 count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask;
2479 seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2480 seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2481 seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2482 seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
2483 seq_printf(m, "%u dwords in ring\n", count);
2484 for (j = 0; j <= count; j++) {
2485 i = (rdp + j) & rdev->cp.ptr_mask;
2486 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
2492 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2494 struct drm_info_node *node = (struct drm_info_node *) m->private;
2495 struct drm_device *dev = node->minor->dev;
2496 struct radeon_device *rdev = dev->dev_private;
2497 uint32_t csq_stat, csq2_stat, tmp;
2498 unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2501 seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2502 seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2503 csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2504 csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2505 r_rptr = (csq_stat >> 0) & 0x3ff;
2506 r_wptr = (csq_stat >> 10) & 0x3ff;
2507 ib1_rptr = (csq_stat >> 20) & 0x3ff;
2508 ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2509 ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2510 ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2511 seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2512 seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2513 seq_printf(m, "Ring rptr %u\n", r_rptr);
2514 seq_printf(m, "Ring wptr %u\n", r_wptr);
2515 seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2516 seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2517 seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2518 seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2519 /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2520 * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2521 seq_printf(m, "Ring fifo:\n");
2522 for (i = 0; i < 256; i++) {
2523 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2524 tmp = RREG32(RADEON_CP_CSQ_DATA);
2525 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
2527 seq_printf(m, "Indirect1 fifo:\n");
2528 for (i = 256; i <= 512; i++) {
2529 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2530 tmp = RREG32(RADEON_CP_CSQ_DATA);
2531 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
2533 seq_printf(m, "Indirect2 fifo:\n");
2534 for (i = 640; i < ib1_wptr; i++) {
2535 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2536 tmp = RREG32(RADEON_CP_CSQ_DATA);
2537 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
2542 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
2544 struct drm_info_node *node = (struct drm_info_node *) m->private;
2545 struct drm_device *dev = node->minor->dev;
2546 struct radeon_device *rdev = dev->dev_private;
2549 tmp = RREG32(RADEON_CONFIG_MEMSIZE);
2550 seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
2551 tmp = RREG32(RADEON_MC_FB_LOCATION);
2552 seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
2553 tmp = RREG32(RADEON_BUS_CNTL);
2554 seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
2555 tmp = RREG32(RADEON_MC_AGP_LOCATION);
2556 seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
2557 tmp = RREG32(RADEON_AGP_BASE);
2558 seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
2559 tmp = RREG32(RADEON_HOST_PATH_CNTL);
2560 seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
2561 tmp = RREG32(0x01D0);
2562 seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
2563 tmp = RREG32(RADEON_AIC_LO_ADDR);
2564 seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
2565 tmp = RREG32(RADEON_AIC_HI_ADDR);
2566 seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
2567 tmp = RREG32(0x01E4);
2568 seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
2572 static struct drm_info_list r100_debugfs_rbbm_list[] = {
2573 {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
2576 static struct drm_info_list r100_debugfs_cp_list[] = {
2577 {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
2578 {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
2581 static struct drm_info_list r100_debugfs_mc_info_list[] = {
2582 {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
2586 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
2588 #if defined(CONFIG_DEBUG_FS)
2589 return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
2595 int r100_debugfs_cp_init(struct radeon_device *rdev)
2597 #if defined(CONFIG_DEBUG_FS)
2598 return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
2604 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
2606 #if defined(CONFIG_DEBUG_FS)
2607 return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
2613 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
2614 uint32_t tiling_flags, uint32_t pitch,
2615 uint32_t offset, uint32_t obj_size)
2617 int surf_index = reg * 16;
2620 if (rdev->family <= CHIP_RS200) {
2621 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
2622 == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
2623 flags |= RADEON_SURF_TILE_COLOR_BOTH;
2624 if (tiling_flags & RADEON_TILING_MACRO)
2625 flags |= RADEON_SURF_TILE_COLOR_MACRO;
2626 } else if (rdev->family <= CHIP_RV280) {
2627 if (tiling_flags & (RADEON_TILING_MACRO))
2628 flags |= R200_SURF_TILE_COLOR_MACRO;
2629 if (tiling_flags & RADEON_TILING_MICRO)
2630 flags |= R200_SURF_TILE_COLOR_MICRO;
2632 if (tiling_flags & RADEON_TILING_MACRO)
2633 flags |= R300_SURF_TILE_MACRO;
2634 if (tiling_flags & RADEON_TILING_MICRO)
2635 flags |= R300_SURF_TILE_MICRO;
2638 if (tiling_flags & RADEON_TILING_SWAP_16BIT)
2639 flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
2640 if (tiling_flags & RADEON_TILING_SWAP_32BIT)
2641 flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
2643 /* when we aren't tiling the pitch seems to needs to be furtherdivided down. - tested on power5 + rn50 server */
2644 if (tiling_flags & (RADEON_TILING_SWAP_16BIT | RADEON_TILING_SWAP_32BIT)) {
2645 if (!(tiling_flags & (RADEON_TILING_MACRO | RADEON_TILING_MICRO)))
2646 if (ASIC_IS_RN50(rdev))
2650 /* r100/r200 divide by 16 */
2651 if (rdev->family < CHIP_R300)
2652 flags |= pitch / 16;
2657 DRM_DEBUG_KMS("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
2658 WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
2659 WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
2660 WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
2664 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
2666 int surf_index = reg * 16;
2667 WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
2670 void r100_bandwidth_update(struct radeon_device *rdev)
2672 fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
2673 fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
2674 fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
2675 uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
2676 fixed20_12 memtcas_ff[8] = {
2681 dfixed_init_half(1),
2682 dfixed_init_half(2),
2685 fixed20_12 memtcas_rs480_ff[8] = {
2691 dfixed_init_half(1),
2692 dfixed_init_half(2),
2693 dfixed_init_half(3),
2695 fixed20_12 memtcas2_ff[8] = {
2705 fixed20_12 memtrbs[8] = {
2707 dfixed_init_half(1),
2709 dfixed_init_half(2),
2711 dfixed_init_half(3),
2715 fixed20_12 memtrbs_r4xx[8] = {
2725 fixed20_12 min_mem_eff;
2726 fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
2727 fixed20_12 cur_latency_mclk, cur_latency_sclk;
2728 fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
2729 disp_drain_rate2, read_return_rate;
2730 fixed20_12 time_disp1_drop_priority;
2732 int cur_size = 16; /* in octawords */
2733 int critical_point = 0, critical_point2;
2734 /* uint32_t read_return_rate, time_disp1_drop_priority; */
2735 int stop_req, max_stop_req;
2736 struct drm_display_mode *mode1 = NULL;
2737 struct drm_display_mode *mode2 = NULL;
2738 uint32_t pixel_bytes1 = 0;
2739 uint32_t pixel_bytes2 = 0;
2741 radeon_update_display_priority(rdev);
2743 if (rdev->mode_info.crtcs[0]->base.enabled) {
2744 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
2745 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
2747 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
2748 if (rdev->mode_info.crtcs[1]->base.enabled) {
2749 mode2 = &rdev->mode_info.crtcs[1]->base.mode;
2750 pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
2754 min_mem_eff.full = dfixed_const_8(0);
2756 if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
2757 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
2758 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
2759 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
2760 /* check crtc enables */
2762 mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
2764 mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
2765 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
2769 * determine is there is enough bw for current mode
2771 sclk_ff = rdev->pm.sclk;
2772 mclk_ff = rdev->pm.mclk;
2774 temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
2775 temp_ff.full = dfixed_const(temp);
2776 mem_bw.full = dfixed_mul(mclk_ff, temp_ff);
2780 peak_disp_bw.full = 0;
2782 temp_ff.full = dfixed_const(1000);
2783 pix_clk.full = dfixed_const(mode1->clock); /* convert to fixed point */
2784 pix_clk.full = dfixed_div(pix_clk, temp_ff);
2785 temp_ff.full = dfixed_const(pixel_bytes1);
2786 peak_disp_bw.full += dfixed_mul(pix_clk, temp_ff);
2789 temp_ff.full = dfixed_const(1000);
2790 pix_clk2.full = dfixed_const(mode2->clock); /* convert to fixed point */
2791 pix_clk2.full = dfixed_div(pix_clk2, temp_ff);
2792 temp_ff.full = dfixed_const(pixel_bytes2);
2793 peak_disp_bw.full += dfixed_mul(pix_clk2, temp_ff);
2796 mem_bw.full = dfixed_mul(mem_bw, min_mem_eff);
2797 if (peak_disp_bw.full >= mem_bw.full) {
2798 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
2799 "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
2802 /* Get values from the EXT_MEM_CNTL register...converting its contents. */
2803 temp = RREG32(RADEON_MEM_TIMING_CNTL);
2804 if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
2805 mem_trcd = ((temp >> 2) & 0x3) + 1;
2806 mem_trp = ((temp & 0x3)) + 1;
2807 mem_tras = ((temp & 0x70) >> 4) + 1;
2808 } else if (rdev->family == CHIP_R300 ||
2809 rdev->family == CHIP_R350) { /* r300, r350 */
2810 mem_trcd = (temp & 0x7) + 1;
2811 mem_trp = ((temp >> 8) & 0x7) + 1;
2812 mem_tras = ((temp >> 11) & 0xf) + 4;
2813 } else if (rdev->family == CHIP_RV350 ||
2814 rdev->family <= CHIP_RV380) {
2816 mem_trcd = (temp & 0x7) + 3;
2817 mem_trp = ((temp >> 8) & 0x7) + 3;
2818 mem_tras = ((temp >> 11) & 0xf) + 6;
2819 } else if (rdev->family == CHIP_R420 ||
2820 rdev->family == CHIP_R423 ||
2821 rdev->family == CHIP_RV410) {
2823 mem_trcd = (temp & 0xf) + 3;
2826 mem_trp = ((temp >> 8) & 0xf) + 3;
2829 mem_tras = ((temp >> 12) & 0x1f) + 6;
2832 } else { /* RV200, R200 */
2833 mem_trcd = (temp & 0x7) + 1;
2834 mem_trp = ((temp >> 8) & 0x7) + 1;
2835 mem_tras = ((temp >> 12) & 0xf) + 4;
2838 trcd_ff.full = dfixed_const(mem_trcd);
2839 trp_ff.full = dfixed_const(mem_trp);
2840 tras_ff.full = dfixed_const(mem_tras);
2842 /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
2843 temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
2844 data = (temp & (7 << 20)) >> 20;
2845 if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
2846 if (rdev->family == CHIP_RS480) /* don't think rs400 */
2847 tcas_ff = memtcas_rs480_ff[data];
2849 tcas_ff = memtcas_ff[data];
2851 tcas_ff = memtcas2_ff[data];
2853 if (rdev->family == CHIP_RS400 ||
2854 rdev->family == CHIP_RS480) {
2855 /* extra cas latency stored in bits 23-25 0-4 clocks */
2856 data = (temp >> 23) & 0x7;
2858 tcas_ff.full += dfixed_const(data);
2861 if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
2862 /* on the R300, Tcas is included in Trbs.
2864 temp = RREG32(RADEON_MEM_CNTL);
2865 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
2867 if (R300_MEM_USE_CD_CH_ONLY & temp) {
2868 temp = RREG32(R300_MC_IND_INDEX);
2869 temp &= ~R300_MC_IND_ADDR_MASK;
2870 temp |= R300_MC_READ_CNTL_CD_mcind;
2871 WREG32(R300_MC_IND_INDEX, temp);
2872 temp = RREG32(R300_MC_IND_DATA);
2873 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
2875 temp = RREG32(R300_MC_READ_CNTL_AB);
2876 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2879 temp = RREG32(R300_MC_READ_CNTL_AB);
2880 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2882 if (rdev->family == CHIP_RV410 ||
2883 rdev->family == CHIP_R420 ||
2884 rdev->family == CHIP_R423)
2885 trbs_ff = memtrbs_r4xx[data];
2887 trbs_ff = memtrbs[data];
2888 tcas_ff.full += trbs_ff.full;
2891 sclk_eff_ff.full = sclk_ff.full;
2893 if (rdev->flags & RADEON_IS_AGP) {
2894 fixed20_12 agpmode_ff;
2895 agpmode_ff.full = dfixed_const(radeon_agpmode);
2896 temp_ff.full = dfixed_const_666(16);
2897 sclk_eff_ff.full -= dfixed_mul(agpmode_ff, temp_ff);
2899 /* TODO PCIE lanes may affect this - agpmode == 16?? */
2901 if (ASIC_IS_R300(rdev)) {
2902 sclk_delay_ff.full = dfixed_const(250);
2904 if ((rdev->family == CHIP_RV100) ||
2905 rdev->flags & RADEON_IS_IGP) {
2906 if (rdev->mc.vram_is_ddr)
2907 sclk_delay_ff.full = dfixed_const(41);
2909 sclk_delay_ff.full = dfixed_const(33);
2911 if (rdev->mc.vram_width == 128)
2912 sclk_delay_ff.full = dfixed_const(57);
2914 sclk_delay_ff.full = dfixed_const(41);
2918 mc_latency_sclk.full = dfixed_div(sclk_delay_ff, sclk_eff_ff);
2920 if (rdev->mc.vram_is_ddr) {
2921 if (rdev->mc.vram_width == 32) {
2922 k1.full = dfixed_const(40);
2925 k1.full = dfixed_const(20);
2929 k1.full = dfixed_const(40);
2933 temp_ff.full = dfixed_const(2);
2934 mc_latency_mclk.full = dfixed_mul(trcd_ff, temp_ff);
2935 temp_ff.full = dfixed_const(c);
2936 mc_latency_mclk.full += dfixed_mul(tcas_ff, temp_ff);
2937 temp_ff.full = dfixed_const(4);
2938 mc_latency_mclk.full += dfixed_mul(tras_ff, temp_ff);
2939 mc_latency_mclk.full += dfixed_mul(trp_ff, temp_ff);
2940 mc_latency_mclk.full += k1.full;
2942 mc_latency_mclk.full = dfixed_div(mc_latency_mclk, mclk_ff);
2943 mc_latency_mclk.full += dfixed_div(temp_ff, sclk_eff_ff);
2946 HW cursor time assuming worst case of full size colour cursor.
2948 temp_ff.full = dfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
2949 temp_ff.full += trcd_ff.full;
2950 if (temp_ff.full < tras_ff.full)
2951 temp_ff.full = tras_ff.full;
2952 cur_latency_mclk.full = dfixed_div(temp_ff, mclk_ff);
2954 temp_ff.full = dfixed_const(cur_size);
2955 cur_latency_sclk.full = dfixed_div(temp_ff, sclk_eff_ff);
2957 Find the total latency for the display data.
2959 disp_latency_overhead.full = dfixed_const(8);
2960 disp_latency_overhead.full = dfixed_div(disp_latency_overhead, sclk_ff);
2961 mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
2962 mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
2964 if (mc_latency_mclk.full > mc_latency_sclk.full)
2965 disp_latency.full = mc_latency_mclk.full;
2967 disp_latency.full = mc_latency_sclk.full;
2969 /* setup Max GRPH_STOP_REQ default value */
2970 if (ASIC_IS_RV100(rdev))
2971 max_stop_req = 0x5c;
2973 max_stop_req = 0x7c;
2977 Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
2978 GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
2980 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
2982 if (stop_req > max_stop_req)
2983 stop_req = max_stop_req;
2986 Find the drain rate of the display buffer.
2988 temp_ff.full = dfixed_const((16/pixel_bytes1));
2989 disp_drain_rate.full = dfixed_div(pix_clk, temp_ff);
2992 Find the critical point of the display buffer.
2994 crit_point_ff.full = dfixed_mul(disp_drain_rate, disp_latency);
2995 crit_point_ff.full += dfixed_const_half(0);
2997 critical_point = dfixed_trunc(crit_point_ff);
2999 if (rdev->disp_priority == 2) {
3004 The critical point should never be above max_stop_req-4. Setting
3005 GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
3007 if (max_stop_req - critical_point < 4)
3010 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
3011 /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
3012 critical_point = 0x10;
3015 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
3016 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
3017 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3018 temp &= ~(RADEON_GRPH_START_REQ_MASK);
3019 if ((rdev->family == CHIP_R350) &&
3020 (stop_req > 0x15)) {
3023 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3024 temp |= RADEON_GRPH_BUFFER_SIZE;
3025 temp &= ~(RADEON_GRPH_CRITICAL_CNTL |
3026 RADEON_GRPH_CRITICAL_AT_SOF |
3027 RADEON_GRPH_STOP_CNTL);
3029 Write the result into the register.
3031 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3032 (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3035 if ((rdev->family == CHIP_RS400) ||
3036 (rdev->family == CHIP_RS480)) {
3037 /* attempt to program RS400 disp regs correctly ??? */
3038 temp = RREG32(RS400_DISP1_REG_CNTL);
3039 temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
3040 RS400_DISP1_STOP_REQ_LEVEL_MASK);
3041 WREG32(RS400_DISP1_REQ_CNTL1, (temp |
3042 (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3043 (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3044 temp = RREG32(RS400_DMIF_MEM_CNTL1);
3045 temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
3046 RS400_DISP1_CRITICAL_POINT_STOP_MASK);
3047 WREG32(RS400_DMIF_MEM_CNTL1, (temp |
3048 (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
3049 (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
3053 DRM_DEBUG_KMS("GRPH_BUFFER_CNTL from to %x\n",
3054 /* (unsigned int)info->SavedReg->grph_buffer_cntl, */
3055 (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
3060 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
3062 if (stop_req > max_stop_req)
3063 stop_req = max_stop_req;
3066 Find the drain rate of the display buffer.
3068 temp_ff.full = dfixed_const((16/pixel_bytes2));
3069 disp_drain_rate2.full = dfixed_div(pix_clk2, temp_ff);
3071 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
3072 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
3073 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
3074 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
3075 if ((rdev->family == CHIP_R350) &&
3076 (stop_req > 0x15)) {
3079 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
3080 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
3081 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL |
3082 RADEON_GRPH_CRITICAL_AT_SOF |
3083 RADEON_GRPH_STOP_CNTL);
3085 if ((rdev->family == CHIP_RS100) ||
3086 (rdev->family == CHIP_RS200))
3087 critical_point2 = 0;
3089 temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
3090 temp_ff.full = dfixed_const(temp);
3091 temp_ff.full = dfixed_mul(mclk_ff, temp_ff);
3092 if (sclk_ff.full < temp_ff.full)
3093 temp_ff.full = sclk_ff.full;
3095 read_return_rate.full = temp_ff.full;
3098 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
3099 time_disp1_drop_priority.full = dfixed_div(crit_point_ff, temp_ff);
3101 time_disp1_drop_priority.full = 0;
3103 crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
3104 crit_point_ff.full = dfixed_mul(crit_point_ff, disp_drain_rate2);
3105 crit_point_ff.full += dfixed_const_half(0);
3107 critical_point2 = dfixed_trunc(crit_point_ff);
3109 if (rdev->disp_priority == 2) {
3110 critical_point2 = 0;
3113 if (max_stop_req - critical_point2 < 4)
3114 critical_point2 = 0;
3118 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
3119 /* some R300 cards have problem with this set to 0 */
3120 critical_point2 = 0x10;
3123 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
3124 (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
3126 if ((rdev->family == CHIP_RS400) ||
3127 (rdev->family == CHIP_RS480)) {
3129 /* attempt to program RS400 disp2 regs correctly ??? */
3130 temp = RREG32(RS400_DISP2_REQ_CNTL1);
3131 temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
3132 RS400_DISP2_STOP_REQ_LEVEL_MASK);
3133 WREG32(RS400_DISP2_REQ_CNTL1, (temp |
3134 (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
3135 (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
3136 temp = RREG32(RS400_DISP2_REQ_CNTL2);
3137 temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
3138 RS400_DISP2_CRITICAL_POINT_STOP_MASK);
3139 WREG32(RS400_DISP2_REQ_CNTL2, (temp |
3140 (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
3141 (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
3143 WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
3144 WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
3145 WREG32(RS400_DMIF_MEM_CNTL1, 0x29CA71DC);
3146 WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
3149 DRM_DEBUG_KMS("GRPH2_BUFFER_CNTL from to %x\n",
3150 (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
3154 static inline void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
3156 DRM_ERROR("pitch %d\n", t->pitch);
3157 DRM_ERROR("use_pitch %d\n", t->use_pitch);
3158 DRM_ERROR("width %d\n", t->width);
3159 DRM_ERROR("width_11 %d\n", t->width_11);
3160 DRM_ERROR("height %d\n", t->height);
3161 DRM_ERROR("height_11 %d\n", t->height_11);
3162 DRM_ERROR("num levels %d\n", t->num_levels);
3163 DRM_ERROR("depth %d\n", t->txdepth);
3164 DRM_ERROR("bpp %d\n", t->cpp);
3165 DRM_ERROR("coordinate type %d\n", t->tex_coord_type);
3166 DRM_ERROR("width round to power of 2 %d\n", t->roundup_w);
3167 DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
3168 DRM_ERROR("compress format %d\n", t->compress_format);
3171 static int r100_track_compress_size(int compress_format, int w, int h)
3173 int block_width, block_height, block_bytes;
3174 int wblocks, hblocks;
3181 switch (compress_format) {
3182 case R100_TRACK_COMP_DXT1:
3187 case R100_TRACK_COMP_DXT35:
3193 hblocks = (h + block_height - 1) / block_height;
3194 wblocks = (w + block_width - 1) / block_width;
3195 if (wblocks < min_wblocks)
3196 wblocks = min_wblocks;
3197 sz = wblocks * hblocks * block_bytes;
3201 static int r100_cs_track_cube(struct radeon_device *rdev,
3202 struct r100_cs_track *track, unsigned idx)
3204 unsigned face, w, h;
3205 struct radeon_bo *cube_robj;
3207 unsigned compress_format = track->textures[idx].compress_format;
3209 for (face = 0; face < 5; face++) {
3210 cube_robj = track->textures[idx].cube_info[face].robj;
3211 w = track->textures[idx].cube_info[face].width;
3212 h = track->textures[idx].cube_info[face].height;
3214 if (compress_format) {
3215 size = r100_track_compress_size(compress_format, w, h);
3218 size *= track->textures[idx].cpp;
3220 size += track->textures[idx].cube_info[face].offset;
3222 if (size > radeon_bo_size(cube_robj)) {
3223 DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
3224 size, radeon_bo_size(cube_robj));
3225 r100_cs_track_texture_print(&track->textures[idx]);
3232 static int r100_cs_track_texture_check(struct radeon_device *rdev,
3233 struct r100_cs_track *track)
3235 struct radeon_bo *robj;
3237 unsigned u, i, w, h, d;
3240 for (u = 0; u < track->num_texture; u++) {
3241 if (!track->textures[u].enabled)
3243 if (track->textures[u].lookup_disable)
3245 robj = track->textures[u].robj;
3247 DRM_ERROR("No texture bound to unit %u\n", u);
3251 for (i = 0; i <= track->textures[u].num_levels; i++) {
3252 if (track->textures[u].use_pitch) {
3253 if (rdev->family < CHIP_R300)
3254 w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
3256 w = track->textures[u].pitch / (1 << i);
3258 w = track->textures[u].width;
3259 if (rdev->family >= CHIP_RV515)
3260 w |= track->textures[u].width_11;
3262 if (track->textures[u].roundup_w)
3263 w = roundup_pow_of_two(w);
3265 h = track->textures[u].height;
3266 if (rdev->family >= CHIP_RV515)
3267 h |= track->textures[u].height_11;
3269 if (track->textures[u].roundup_h)
3270 h = roundup_pow_of_two(h);
3271 if (track->textures[u].tex_coord_type == 1) {
3272 d = (1 << track->textures[u].txdepth) / (1 << i);
3278 if (track->textures[u].compress_format) {
3280 size += r100_track_compress_size(track->textures[u].compress_format, w, h) * d;
3281 /* compressed textures are block based */
3285 size *= track->textures[u].cpp;
3287 switch (track->textures[u].tex_coord_type) {
3292 if (track->separate_cube) {
3293 ret = r100_cs_track_cube(rdev, track, u);
3300 DRM_ERROR("Invalid texture coordinate type %u for unit "
3301 "%u\n", track->textures[u].tex_coord_type, u);
3304 if (size > radeon_bo_size(robj)) {
3305 DRM_ERROR("Texture of unit %u needs %lu bytes but is "
3306 "%lu\n", u, size, radeon_bo_size(robj));
3307 r100_cs_track_texture_print(&track->textures[u]);
3314 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
3320 unsigned num_cb = track->num_cb;
3322 if (!track->zb_cb_clear && !track->color_channel_mask &&
3323 !track->blend_read_enable)
3326 for (i = 0; i < num_cb; i++) {
3327 if (track->cb[i].robj == NULL) {
3328 DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
3331 size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
3332 size += track->cb[i].offset;
3333 if (size > radeon_bo_size(track->cb[i].robj)) {
3334 DRM_ERROR("[drm] Buffer too small for color buffer %d "
3335 "(need %lu have %lu) !\n", i, size,
3336 radeon_bo_size(track->cb[i].robj));
3337 DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
3338 i, track->cb[i].pitch, track->cb[i].cpp,
3339 track->cb[i].offset, track->maxy);
3343 if (track->z_enabled) {
3344 if (track->zb.robj == NULL) {
3345 DRM_ERROR("[drm] No buffer for z buffer !\n");
3348 size = track->zb.pitch * track->zb.cpp * track->maxy;
3349 size += track->zb.offset;
3350 if (size > radeon_bo_size(track->zb.robj)) {
3351 DRM_ERROR("[drm] Buffer too small for z buffer "
3352 "(need %lu have %lu) !\n", size,
3353 radeon_bo_size(track->zb.robj));
3354 DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
3355 track->zb.pitch, track->zb.cpp,
3356 track->zb.offset, track->maxy);
3360 prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
3361 if (track->vap_vf_cntl & (1 << 14)) {
3362 nverts = track->vap_alt_nverts;
3364 nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
3366 switch (prim_walk) {
3368 for (i = 0; i < track->num_arrays; i++) {
3369 size = track->arrays[i].esize * track->max_indx * 4;
3370 if (track->arrays[i].robj == NULL) {
3371 DRM_ERROR("(PW %u) Vertex array %u no buffer "
3372 "bound\n", prim_walk, i);
3375 if (size > radeon_bo_size(track->arrays[i].robj)) {
3376 dev_err(rdev->dev, "(PW %u) Vertex array %u "
3377 "need %lu dwords have %lu dwords\n",
3378 prim_walk, i, size >> 2,
3379 radeon_bo_size(track->arrays[i].robj)
3381 DRM_ERROR("Max indices %u\n", track->max_indx);
3387 for (i = 0; i < track->num_arrays; i++) {
3388 size = track->arrays[i].esize * (nverts - 1) * 4;
3389 if (track->arrays[i].robj == NULL) {
3390 DRM_ERROR("(PW %u) Vertex array %u no buffer "
3391 "bound\n", prim_walk, i);
3394 if (size > radeon_bo_size(track->arrays[i].robj)) {
3395 dev_err(rdev->dev, "(PW %u) Vertex array %u "
3396 "need %lu dwords have %lu dwords\n",
3397 prim_walk, i, size >> 2,
3398 radeon_bo_size(track->arrays[i].robj)
3405 size = track->vtx_size * nverts;
3406 if (size != track->immd_dwords) {
3407 DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
3408 track->immd_dwords, size);
3409 DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
3410 nverts, track->vtx_size);
3415 DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
3419 return r100_cs_track_texture_check(rdev, track);
3422 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
3426 if (rdev->family < CHIP_R300) {
3428 if (rdev->family <= CHIP_RS200)
3429 track->num_texture = 3;
3431 track->num_texture = 6;
3433 track->separate_cube = 1;
3436 track->num_texture = 16;
3438 track->separate_cube = 0;
3441 for (i = 0; i < track->num_cb; i++) {
3442 track->cb[i].robj = NULL;
3443 track->cb[i].pitch = 8192;
3444 track->cb[i].cpp = 16;
3445 track->cb[i].offset = 0;
3447 track->z_enabled = true;
3448 track->zb.robj = NULL;
3449 track->zb.pitch = 8192;
3451 track->zb.offset = 0;
3452 track->vtx_size = 0x7F;
3453 track->immd_dwords = 0xFFFFFFFFUL;
3454 track->num_arrays = 11;
3455 track->max_indx = 0x00FFFFFFUL;
3456 for (i = 0; i < track->num_arrays; i++) {
3457 track->arrays[i].robj = NULL;
3458 track->arrays[i].esize = 0x7F;
3460 for (i = 0; i < track->num_texture; i++) {
3461 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
3462 track->textures[i].pitch = 16536;
3463 track->textures[i].width = 16536;
3464 track->textures[i].height = 16536;
3465 track->textures[i].width_11 = 1 << 11;
3466 track->textures[i].height_11 = 1 << 11;
3467 track->textures[i].num_levels = 12;
3468 if (rdev->family <= CHIP_RS200) {
3469 track->textures[i].tex_coord_type = 0;
3470 track->textures[i].txdepth = 0;
3472 track->textures[i].txdepth = 16;
3473 track->textures[i].tex_coord_type = 1;
3475 track->textures[i].cpp = 64;
3476 track->textures[i].robj = NULL;
3477 /* CS IB emission code makes sure texture unit are disabled */
3478 track->textures[i].enabled = false;
3479 track->textures[i].lookup_disable = false;
3480 track->textures[i].roundup_w = true;
3481 track->textures[i].roundup_h = true;
3482 if (track->separate_cube)
3483 for (face = 0; face < 5; face++) {
3484 track->textures[i].cube_info[face].robj = NULL;
3485 track->textures[i].cube_info[face].width = 16536;
3486 track->textures[i].cube_info[face].height = 16536;
3487 track->textures[i].cube_info[face].offset = 0;
3492 int r100_ring_test(struct radeon_device *rdev)
3499 r = radeon_scratch_get(rdev, &scratch);
3501 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3504 WREG32(scratch, 0xCAFEDEAD);
3505 r = radeon_ring_lock(rdev, 2);
3507 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3508 radeon_scratch_free(rdev, scratch);
3511 radeon_ring_write(rdev, PACKET0(scratch, 0));
3512 radeon_ring_write(rdev, 0xDEADBEEF);
3513 radeon_ring_unlock_commit(rdev);
3514 for (i = 0; i < rdev->usec_timeout; i++) {
3515 tmp = RREG32(scratch);
3516 if (tmp == 0xDEADBEEF) {
3521 if (i < rdev->usec_timeout) {
3522 DRM_INFO("ring test succeeded in %d usecs\n", i);
3524 DRM_ERROR("radeon: ring test failed (sracth(0x%04X)=0x%08X)\n",
3528 radeon_scratch_free(rdev, scratch);
3532 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3534 radeon_ring_write(rdev, PACKET0(RADEON_CP_IB_BASE, 1));
3535 radeon_ring_write(rdev, ib->gpu_addr);
3536 radeon_ring_write(rdev, ib->length_dw);
3539 int r100_ib_test(struct radeon_device *rdev)
3541 struct radeon_ib *ib;
3547 r = radeon_scratch_get(rdev, &scratch);
3549 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3552 WREG32(scratch, 0xCAFEDEAD);
3553 r = radeon_ib_get(rdev, &ib);
3557 ib->ptr[0] = PACKET0(scratch, 0);
3558 ib->ptr[1] = 0xDEADBEEF;
3559 ib->ptr[2] = PACKET2(0);
3560 ib->ptr[3] = PACKET2(0);
3561 ib->ptr[4] = PACKET2(0);
3562 ib->ptr[5] = PACKET2(0);
3563 ib->ptr[6] = PACKET2(0);
3564 ib->ptr[7] = PACKET2(0);
3566 r = radeon_ib_schedule(rdev, ib);
3568 radeon_scratch_free(rdev, scratch);
3569 radeon_ib_free(rdev, &ib);
3572 r = radeon_fence_wait(ib->fence, false);
3576 for (i = 0; i < rdev->usec_timeout; i++) {
3577 tmp = RREG32(scratch);
3578 if (tmp == 0xDEADBEEF) {
3583 if (i < rdev->usec_timeout) {
3584 DRM_INFO("ib test succeeded in %u usecs\n", i);
3586 DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n",
3590 radeon_scratch_free(rdev, scratch);
3591 radeon_ib_free(rdev, &ib);
3595 void r100_ib_fini(struct radeon_device *rdev)
3597 radeon_ib_pool_fini(rdev);
3600 int r100_ib_init(struct radeon_device *rdev)
3604 r = radeon_ib_pool_init(rdev);
3606 dev_err(rdev->dev, "failled initializing IB pool (%d).\n", r);
3610 r = r100_ib_test(rdev);
3612 dev_err(rdev->dev, "failled testing IB (%d).\n", r);
3619 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3621 /* Shutdown CP we shouldn't need to do that but better be safe than
3624 rdev->cp.ready = false;
3625 WREG32(R_000740_CP_CSQ_CNTL, 0);
3627 /* Save few CRTC registers */
3628 save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3629 save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3630 save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3631 save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3632 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3633 save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3634 save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3637 /* Disable VGA aperture access */
3638 WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3639 /* Disable cursor, overlay, crtc */
3640 WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3641 WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3642 S_000054_CRTC_DISPLAY_DIS(1));
3643 WREG32(R_000050_CRTC_GEN_CNTL,
3644 (C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3645 S_000050_CRTC_DISP_REQ_EN_B(1));
3646 WREG32(R_000420_OV0_SCALE_CNTL,
3647 C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3648 WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3649 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3650 WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3651 S_000360_CUR2_LOCK(1));
3652 WREG32(R_0003F8_CRTC2_GEN_CNTL,
3653 (C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3654 S_0003F8_CRTC2_DISPLAY_DIS(1) |
3655 S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3656 WREG32(R_000360_CUR2_OFFSET,
3657 C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3661 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3663 /* Update base address for crtc */
3664 WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3665 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3666 WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3668 /* Restore CRTC registers */
3669 WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3670 WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3671 WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3672 if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3673 WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3677 void r100_vga_render_disable(struct radeon_device *rdev)
3681 tmp = RREG8(R_0003C2_GENMO_WT);
3682 WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3685 static void r100_debugfs(struct radeon_device *rdev)
3689 r = r100_debugfs_mc_info_init(rdev);
3691 dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3694 static void r100_mc_program(struct radeon_device *rdev)
3696 struct r100_mc_save save;
3698 /* Stops all mc clients */
3699 r100_mc_stop(rdev, &save);
3700 if (rdev->flags & RADEON_IS_AGP) {
3701 WREG32(R_00014C_MC_AGP_LOCATION,
3702 S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3703 S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3704 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3705 if (rdev->family > CHIP_RV200)
3706 WREG32(R_00015C_AGP_BASE_2,
3707 upper_32_bits(rdev->mc.agp_base) & 0xff);
3709 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3710 WREG32(R_000170_AGP_BASE, 0);
3711 if (rdev->family > CHIP_RV200)
3712 WREG32(R_00015C_AGP_BASE_2, 0);
3714 /* Wait for mc idle */
3715 if (r100_mc_wait_for_idle(rdev))
3716 dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3717 /* Program MC, should be a 32bits limited address space */
3718 WREG32(R_000148_MC_FB_LOCATION,
3719 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3720 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3721 r100_mc_resume(rdev, &save);
3724 void r100_clock_startup(struct radeon_device *rdev)
3728 if (radeon_dynclks != -1 && radeon_dynclks)
3729 radeon_legacy_set_clock_gating(rdev, 1);
3730 /* We need to force on some of the block */
3731 tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3732 tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3733 if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3734 tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3735 WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3738 static int r100_startup(struct radeon_device *rdev)
3742 /* set common regs */
3743 r100_set_common_regs(rdev);
3745 r100_mc_program(rdev);
3747 r100_clock_startup(rdev);
3748 /* Initialize GPU configuration (# pipes, ...) */
3749 // r100_gpu_init(rdev);
3750 /* Initialize GART (initialize after TTM so we can allocate
3751 * memory through TTM but finalize after TTM) */
3752 r100_enable_bm(rdev);
3753 if (rdev->flags & RADEON_IS_PCI) {
3754 r = r100_pci_gart_enable(rdev);
3759 /* allocate wb buffer */
3760 r = radeon_wb_init(rdev);
3766 rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3767 /* 1M ring buffer */
3768 r = r100_cp_init(rdev, 1024 * 1024);
3770 dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
3773 r = r100_ib_init(rdev);
3775 dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
3781 int r100_resume(struct radeon_device *rdev)
3783 /* Make sur GART are not working */
3784 if (rdev->flags & RADEON_IS_PCI)
3785 r100_pci_gart_disable(rdev);
3786 /* Resume clock before doing reset */
3787 r100_clock_startup(rdev);
3788 /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3789 if (radeon_asic_reset(rdev)) {
3790 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3791 RREG32(R_000E40_RBBM_STATUS),
3792 RREG32(R_0007C0_CP_STAT));
3795 radeon_combios_asic_init(rdev->ddev);
3796 /* Resume clock after posting */
3797 r100_clock_startup(rdev);
3798 /* Initialize surface registers */
3799 radeon_surface_init(rdev);
3800 return r100_startup(rdev);
3803 int r100_suspend(struct radeon_device *rdev)
3805 r100_cp_disable(rdev);
3806 radeon_wb_disable(rdev);
3807 r100_irq_disable(rdev);
3808 if (rdev->flags & RADEON_IS_PCI)
3809 r100_pci_gart_disable(rdev);
3813 void r100_fini(struct radeon_device *rdev)
3816 radeon_wb_fini(rdev);
3818 radeon_gem_fini(rdev);
3819 if (rdev->flags & RADEON_IS_PCI)
3820 r100_pci_gart_fini(rdev);
3821 radeon_agp_fini(rdev);
3822 radeon_irq_kms_fini(rdev);
3823 radeon_fence_driver_fini(rdev);
3824 radeon_bo_fini(rdev);
3825 radeon_atombios_fini(rdev);
3831 * Due to how kexec works, it can leave the hw fully initialised when it
3832 * boots the new kernel. However doing our init sequence with the CP and
3833 * WB stuff setup causes GPU hangs on the RN50 at least. So at startup
3834 * do some quick sanity checks and restore sane values to avoid this
3837 void r100_restore_sanity(struct radeon_device *rdev)
3841 tmp = RREG32(RADEON_CP_CSQ_CNTL);
3843 WREG32(RADEON_CP_CSQ_CNTL, 0);
3845 tmp = RREG32(RADEON_CP_RB_CNTL);
3847 WREG32(RADEON_CP_RB_CNTL, 0);
3849 tmp = RREG32(RADEON_SCRATCH_UMSK);
3851 WREG32(RADEON_SCRATCH_UMSK, 0);
3855 int r100_init(struct radeon_device *rdev)
3859 /* Register debugfs file specific to this group of asics */
3862 r100_vga_render_disable(rdev);
3863 /* Initialize scratch registers */
3864 radeon_scratch_init(rdev);
3865 /* Initialize surface registers */
3866 radeon_surface_init(rdev);
3867 /* sanity check some register to avoid hangs like after kexec */
3868 r100_restore_sanity(rdev);
3869 /* TODO: disable VGA need to use VGA request */
3871 if (!radeon_get_bios(rdev)) {
3872 if (ASIC_IS_AVIVO(rdev))
3875 if (rdev->is_atom_bios) {
3876 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
3879 r = radeon_combios_init(rdev);
3883 /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3884 if (radeon_asic_reset(rdev)) {
3886 "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3887 RREG32(R_000E40_RBBM_STATUS),
3888 RREG32(R_0007C0_CP_STAT));
3890 /* check if cards are posted or not */
3891 if (radeon_boot_test_post_card(rdev) == false)
3893 /* Set asic errata */
3895 /* Initialize clocks */
3896 radeon_get_clock_info(rdev->ddev);
3897 /* initialize AGP */
3898 if (rdev->flags & RADEON_IS_AGP) {
3899 r = radeon_agp_init(rdev);
3901 radeon_agp_disable(rdev);
3904 /* initialize VRAM */
3907 r = radeon_fence_driver_init(rdev);
3910 r = radeon_irq_kms_init(rdev);
3913 /* Memory manager */
3914 r = radeon_bo_init(rdev);
3917 if (rdev->flags & RADEON_IS_PCI) {
3918 r = r100_pci_gart_init(rdev);
3922 r100_set_safe_registers(rdev);
3923 rdev->accel_working = true;
3924 r = r100_startup(rdev);
3926 /* Somethings want wront with the accel init stop accel */
3927 dev_err(rdev->dev, "Disabling GPU acceleration\n");
3929 radeon_wb_fini(rdev);
3931 radeon_irq_kms_fini(rdev);
3932 if (rdev->flags & RADEON_IS_PCI)
3933 r100_pci_gart_fini(rdev);
3934 rdev->accel_working = false;