2 * Copyright 2011 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
22 * Authors: Alex Deucher
24 #include <linux/firmware.h>
25 #include <linux/platform_device.h>
26 #include <linux/slab.h>
27 #include <linux/module.h>
30 #include "radeon_asic.h"
31 #include "radeon_drm.h"
34 #include "si_blit_shaders.h"
36 #define SI_PFP_UCODE_SIZE 2144
37 #define SI_PM4_UCODE_SIZE 2144
38 #define SI_CE_UCODE_SIZE 2144
39 #define SI_RLC_UCODE_SIZE 2048
40 #define SI_MC_UCODE_SIZE 7769
42 MODULE_FIRMWARE("radeon/TAHITI_pfp.bin");
43 MODULE_FIRMWARE("radeon/TAHITI_me.bin");
44 MODULE_FIRMWARE("radeon/TAHITI_ce.bin");
45 MODULE_FIRMWARE("radeon/TAHITI_mc.bin");
46 MODULE_FIRMWARE("radeon/TAHITI_rlc.bin");
47 MODULE_FIRMWARE("radeon/PITCAIRN_pfp.bin");
48 MODULE_FIRMWARE("radeon/PITCAIRN_me.bin");
49 MODULE_FIRMWARE("radeon/PITCAIRN_ce.bin");
50 MODULE_FIRMWARE("radeon/PITCAIRN_mc.bin");
51 MODULE_FIRMWARE("radeon/PITCAIRN_rlc.bin");
52 MODULE_FIRMWARE("radeon/VERDE_pfp.bin");
53 MODULE_FIRMWARE("radeon/VERDE_me.bin");
54 MODULE_FIRMWARE("radeon/VERDE_ce.bin");
55 MODULE_FIRMWARE("radeon/VERDE_mc.bin");
56 MODULE_FIRMWARE("radeon/VERDE_rlc.bin");
58 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
59 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
60 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
62 /* get temperature in millidegrees */
63 int si_get_temp(struct radeon_device *rdev)
68 temp = (RREG32(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
74 actual_temp = temp & 0x1ff;
76 actual_temp = (actual_temp * 1000);
81 #define TAHITI_IO_MC_REGS_SIZE 36
83 static const u32 tahiti_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
84 {0x0000006f, 0x03044000},
85 {0x00000070, 0x0480c018},
86 {0x00000071, 0x00000040},
87 {0x00000072, 0x01000000},
88 {0x00000074, 0x000000ff},
89 {0x00000075, 0x00143400},
90 {0x00000076, 0x08ec0800},
91 {0x00000077, 0x040000cc},
92 {0x00000079, 0x00000000},
93 {0x0000007a, 0x21000409},
94 {0x0000007c, 0x00000000},
95 {0x0000007d, 0xe8000000},
96 {0x0000007e, 0x044408a8},
97 {0x0000007f, 0x00000003},
98 {0x00000080, 0x00000000},
99 {0x00000081, 0x01000000},
100 {0x00000082, 0x02000000},
101 {0x00000083, 0x00000000},
102 {0x00000084, 0xe3f3e4f4},
103 {0x00000085, 0x00052024},
104 {0x00000087, 0x00000000},
105 {0x00000088, 0x66036603},
106 {0x00000089, 0x01000000},
107 {0x0000008b, 0x1c0a0000},
108 {0x0000008c, 0xff010000},
109 {0x0000008e, 0xffffefff},
110 {0x0000008f, 0xfff3efff},
111 {0x00000090, 0xfff3efbf},
112 {0x00000094, 0x00101101},
113 {0x00000095, 0x00000fff},
114 {0x00000096, 0x00116fff},
115 {0x00000097, 0x60010000},
116 {0x00000098, 0x10010000},
117 {0x00000099, 0x00006000},
118 {0x0000009a, 0x00001000},
119 {0x0000009f, 0x00a77400}
122 static const u32 pitcairn_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
123 {0x0000006f, 0x03044000},
124 {0x00000070, 0x0480c018},
125 {0x00000071, 0x00000040},
126 {0x00000072, 0x01000000},
127 {0x00000074, 0x000000ff},
128 {0x00000075, 0x00143400},
129 {0x00000076, 0x08ec0800},
130 {0x00000077, 0x040000cc},
131 {0x00000079, 0x00000000},
132 {0x0000007a, 0x21000409},
133 {0x0000007c, 0x00000000},
134 {0x0000007d, 0xe8000000},
135 {0x0000007e, 0x044408a8},
136 {0x0000007f, 0x00000003},
137 {0x00000080, 0x00000000},
138 {0x00000081, 0x01000000},
139 {0x00000082, 0x02000000},
140 {0x00000083, 0x00000000},
141 {0x00000084, 0xe3f3e4f4},
142 {0x00000085, 0x00052024},
143 {0x00000087, 0x00000000},
144 {0x00000088, 0x66036603},
145 {0x00000089, 0x01000000},
146 {0x0000008b, 0x1c0a0000},
147 {0x0000008c, 0xff010000},
148 {0x0000008e, 0xffffefff},
149 {0x0000008f, 0xfff3efff},
150 {0x00000090, 0xfff3efbf},
151 {0x00000094, 0x00101101},
152 {0x00000095, 0x00000fff},
153 {0x00000096, 0x00116fff},
154 {0x00000097, 0x60010000},
155 {0x00000098, 0x10010000},
156 {0x00000099, 0x00006000},
157 {0x0000009a, 0x00001000},
158 {0x0000009f, 0x00a47400}
161 static const u32 verde_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
162 {0x0000006f, 0x03044000},
163 {0x00000070, 0x0480c018},
164 {0x00000071, 0x00000040},
165 {0x00000072, 0x01000000},
166 {0x00000074, 0x000000ff},
167 {0x00000075, 0x00143400},
168 {0x00000076, 0x08ec0800},
169 {0x00000077, 0x040000cc},
170 {0x00000079, 0x00000000},
171 {0x0000007a, 0x21000409},
172 {0x0000007c, 0x00000000},
173 {0x0000007d, 0xe8000000},
174 {0x0000007e, 0x044408a8},
175 {0x0000007f, 0x00000003},
176 {0x00000080, 0x00000000},
177 {0x00000081, 0x01000000},
178 {0x00000082, 0x02000000},
179 {0x00000083, 0x00000000},
180 {0x00000084, 0xe3f3e4f4},
181 {0x00000085, 0x00052024},
182 {0x00000087, 0x00000000},
183 {0x00000088, 0x66036603},
184 {0x00000089, 0x01000000},
185 {0x0000008b, 0x1c0a0000},
186 {0x0000008c, 0xff010000},
187 {0x0000008e, 0xffffefff},
188 {0x0000008f, 0xfff3efff},
189 {0x00000090, 0xfff3efbf},
190 {0x00000094, 0x00101101},
191 {0x00000095, 0x00000fff},
192 {0x00000096, 0x00116fff},
193 {0x00000097, 0x60010000},
194 {0x00000098, 0x10010000},
195 {0x00000099, 0x00006000},
196 {0x0000009a, 0x00001000},
197 {0x0000009f, 0x00a37400}
201 static int si_mc_load_microcode(struct radeon_device *rdev)
203 const __be32 *fw_data;
204 u32 running, blackout = 0;
206 int i, ucode_size, regs_size;
211 switch (rdev->family) {
213 io_mc_regs = (u32 *)&tahiti_io_mc_regs;
214 ucode_size = SI_MC_UCODE_SIZE;
215 regs_size = TAHITI_IO_MC_REGS_SIZE;
218 io_mc_regs = (u32 *)&pitcairn_io_mc_regs;
219 ucode_size = SI_MC_UCODE_SIZE;
220 regs_size = TAHITI_IO_MC_REGS_SIZE;
224 io_mc_regs = (u32 *)&verde_io_mc_regs;
225 ucode_size = SI_MC_UCODE_SIZE;
226 regs_size = TAHITI_IO_MC_REGS_SIZE;
230 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
234 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
235 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
238 /* reset the engine and set to writable */
239 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
240 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
242 /* load mc io regs */
243 for (i = 0; i < regs_size; i++) {
244 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
245 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
247 /* load the MC ucode */
248 fw_data = (const __be32 *)rdev->mc_fw->data;
249 for (i = 0; i < ucode_size; i++)
250 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
252 /* put the engine back into the active state */
253 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
254 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
255 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
257 /* wait for training to complete */
258 for (i = 0; i < rdev->usec_timeout; i++) {
259 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
263 for (i = 0; i < rdev->usec_timeout; i++) {
264 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
270 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
276 static int si_init_microcode(struct radeon_device *rdev)
278 struct platform_device *pdev;
279 const char *chip_name;
280 const char *rlc_chip_name;
281 size_t pfp_req_size, me_req_size, ce_req_size, rlc_req_size, mc_req_size;
287 pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
290 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
294 switch (rdev->family) {
296 chip_name = "TAHITI";
297 rlc_chip_name = "TAHITI";
298 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
299 me_req_size = SI_PM4_UCODE_SIZE * 4;
300 ce_req_size = SI_CE_UCODE_SIZE * 4;
301 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
302 mc_req_size = SI_MC_UCODE_SIZE * 4;
305 chip_name = "PITCAIRN";
306 rlc_chip_name = "PITCAIRN";
307 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
308 me_req_size = SI_PM4_UCODE_SIZE * 4;
309 ce_req_size = SI_CE_UCODE_SIZE * 4;
310 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
311 mc_req_size = SI_MC_UCODE_SIZE * 4;
315 rlc_chip_name = "VERDE";
316 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
317 me_req_size = SI_PM4_UCODE_SIZE * 4;
318 ce_req_size = SI_CE_UCODE_SIZE * 4;
319 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
320 mc_req_size = SI_MC_UCODE_SIZE * 4;
325 DRM_INFO("Loading %s Microcode\n", chip_name);
327 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
328 err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
331 if (rdev->pfp_fw->size != pfp_req_size) {
333 "si_cp: Bogus length %zu in firmware \"%s\"\n",
334 rdev->pfp_fw->size, fw_name);
339 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
340 err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
343 if (rdev->me_fw->size != me_req_size) {
345 "si_cp: Bogus length %zu in firmware \"%s\"\n",
346 rdev->me_fw->size, fw_name);
350 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
351 err = request_firmware(&rdev->ce_fw, fw_name, &pdev->dev);
354 if (rdev->ce_fw->size != ce_req_size) {
356 "si_cp: Bogus length %zu in firmware \"%s\"\n",
357 rdev->ce_fw->size, fw_name);
361 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
362 err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
365 if (rdev->rlc_fw->size != rlc_req_size) {
367 "si_rlc: Bogus length %zu in firmware \"%s\"\n",
368 rdev->rlc_fw->size, fw_name);
372 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
373 err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev);
376 if (rdev->mc_fw->size != mc_req_size) {
378 "si_mc: Bogus length %zu in firmware \"%s\"\n",
379 rdev->mc_fw->size, fw_name);
384 platform_device_unregister(pdev);
389 "si_cp: Failed to load firmware \"%s\"\n",
391 release_firmware(rdev->pfp_fw);
393 release_firmware(rdev->me_fw);
395 release_firmware(rdev->ce_fw);
397 release_firmware(rdev->rlc_fw);
399 release_firmware(rdev->mc_fw);
405 /* watermark setup */
406 static u32 dce6_line_buffer_adjust(struct radeon_device *rdev,
407 struct radeon_crtc *radeon_crtc,
408 struct drm_display_mode *mode,
409 struct drm_display_mode *other_mode)
414 * There are 3 line buffers, each one shared by 2 display controllers.
415 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
416 * the display controllers. The paritioning is done via one of four
417 * preset allocations specified in bits 21:20:
419 * 2 - whole lb, other crtc must be disabled
421 /* this can get tricky if we have two large displays on a paired group
422 * of crtcs. Ideally for multiple large displays we'd assign them to
423 * non-linked crtcs for maximum line buffer allocation.
425 if (radeon_crtc->base.enabled && mode) {
433 WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset,
434 DC_LB_MEMORY_CONFIG(tmp));
436 if (radeon_crtc->base.enabled && mode) {
446 /* controller not enabled, so no lb used */
450 static u32 dce6_get_number_of_dram_channels(struct radeon_device *rdev)
452 u32 tmp = RREG32(MC_SHARED_CHMAP);
454 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
477 struct dce6_wm_params {
478 u32 dram_channels; /* number of dram channels */
479 u32 yclk; /* bandwidth per dram data pin in kHz */
480 u32 sclk; /* engine clock in kHz */
481 u32 disp_clk; /* display clock in kHz */
482 u32 src_width; /* viewport width */
483 u32 active_time; /* active display time in ns */
484 u32 blank_time; /* blank time in ns */
485 bool interlaced; /* mode is interlaced */
486 fixed20_12 vsc; /* vertical scale ratio */
487 u32 num_heads; /* number of active crtcs */
488 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
489 u32 lb_size; /* line buffer allocated to pipe */
490 u32 vtaps; /* vertical scaler taps */
493 static u32 dce6_dram_bandwidth(struct dce6_wm_params *wm)
495 /* Calculate raw DRAM Bandwidth */
496 fixed20_12 dram_efficiency; /* 0.7 */
497 fixed20_12 yclk, dram_channels, bandwidth;
500 a.full = dfixed_const(1000);
501 yclk.full = dfixed_const(wm->yclk);
502 yclk.full = dfixed_div(yclk, a);
503 dram_channels.full = dfixed_const(wm->dram_channels * 4);
504 a.full = dfixed_const(10);
505 dram_efficiency.full = dfixed_const(7);
506 dram_efficiency.full = dfixed_div(dram_efficiency, a);
507 bandwidth.full = dfixed_mul(dram_channels, yclk);
508 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
510 return dfixed_trunc(bandwidth);
513 static u32 dce6_dram_bandwidth_for_display(struct dce6_wm_params *wm)
515 /* Calculate DRAM Bandwidth and the part allocated to display. */
516 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
517 fixed20_12 yclk, dram_channels, bandwidth;
520 a.full = dfixed_const(1000);
521 yclk.full = dfixed_const(wm->yclk);
522 yclk.full = dfixed_div(yclk, a);
523 dram_channels.full = dfixed_const(wm->dram_channels * 4);
524 a.full = dfixed_const(10);
525 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
526 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
527 bandwidth.full = dfixed_mul(dram_channels, yclk);
528 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
530 return dfixed_trunc(bandwidth);
533 static u32 dce6_data_return_bandwidth(struct dce6_wm_params *wm)
535 /* Calculate the display Data return Bandwidth */
536 fixed20_12 return_efficiency; /* 0.8 */
537 fixed20_12 sclk, bandwidth;
540 a.full = dfixed_const(1000);
541 sclk.full = dfixed_const(wm->sclk);
542 sclk.full = dfixed_div(sclk, a);
543 a.full = dfixed_const(10);
544 return_efficiency.full = dfixed_const(8);
545 return_efficiency.full = dfixed_div(return_efficiency, a);
546 a.full = dfixed_const(32);
547 bandwidth.full = dfixed_mul(a, sclk);
548 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
550 return dfixed_trunc(bandwidth);
553 static u32 dce6_get_dmif_bytes_per_request(struct dce6_wm_params *wm)
558 static u32 dce6_dmif_request_bandwidth(struct dce6_wm_params *wm)
560 /* Calculate the DMIF Request Bandwidth */
561 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
562 fixed20_12 disp_clk, sclk, bandwidth;
563 fixed20_12 a, b1, b2;
566 a.full = dfixed_const(1000);
567 disp_clk.full = dfixed_const(wm->disp_clk);
568 disp_clk.full = dfixed_div(disp_clk, a);
569 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm) / 2);
570 b1.full = dfixed_mul(a, disp_clk);
572 a.full = dfixed_const(1000);
573 sclk.full = dfixed_const(wm->sclk);
574 sclk.full = dfixed_div(sclk, a);
575 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm));
576 b2.full = dfixed_mul(a, sclk);
578 a.full = dfixed_const(10);
579 disp_clk_request_efficiency.full = dfixed_const(8);
580 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
582 min_bandwidth = min(dfixed_trunc(b1), dfixed_trunc(b2));
584 a.full = dfixed_const(min_bandwidth);
585 bandwidth.full = dfixed_mul(a, disp_clk_request_efficiency);
587 return dfixed_trunc(bandwidth);
590 static u32 dce6_available_bandwidth(struct dce6_wm_params *wm)
592 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
593 u32 dram_bandwidth = dce6_dram_bandwidth(wm);
594 u32 data_return_bandwidth = dce6_data_return_bandwidth(wm);
595 u32 dmif_req_bandwidth = dce6_dmif_request_bandwidth(wm);
597 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
600 static u32 dce6_average_bandwidth(struct dce6_wm_params *wm)
602 /* Calculate the display mode Average Bandwidth
603 * DisplayMode should contain the source and destination dimensions,
607 fixed20_12 line_time;
608 fixed20_12 src_width;
609 fixed20_12 bandwidth;
612 a.full = dfixed_const(1000);
613 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
614 line_time.full = dfixed_div(line_time, a);
615 bpp.full = dfixed_const(wm->bytes_per_pixel);
616 src_width.full = dfixed_const(wm->src_width);
617 bandwidth.full = dfixed_mul(src_width, bpp);
618 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
619 bandwidth.full = dfixed_div(bandwidth, line_time);
621 return dfixed_trunc(bandwidth);
624 static u32 dce6_latency_watermark(struct dce6_wm_params *wm)
626 /* First calcualte the latency in ns */
627 u32 mc_latency = 2000; /* 2000 ns. */
628 u32 available_bandwidth = dce6_available_bandwidth(wm);
629 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
630 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
631 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
632 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
633 (wm->num_heads * cursor_line_pair_return_time);
634 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
635 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
636 u32 tmp, dmif_size = 12288;
639 if (wm->num_heads == 0)
642 a.full = dfixed_const(2);
643 b.full = dfixed_const(1);
644 if ((wm->vsc.full > a.full) ||
645 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
647 ((wm->vsc.full >= a.full) && wm->interlaced))
648 max_src_lines_per_dst_line = 4;
650 max_src_lines_per_dst_line = 2;
652 a.full = dfixed_const(available_bandwidth);
653 b.full = dfixed_const(wm->num_heads);
654 a.full = dfixed_div(a, b);
656 b.full = dfixed_const(mc_latency + 512);
657 c.full = dfixed_const(wm->disp_clk);
658 b.full = dfixed_div(b, c);
660 c.full = dfixed_const(dmif_size);
661 b.full = dfixed_div(c, b);
663 tmp = min(dfixed_trunc(a), dfixed_trunc(b));
665 b.full = dfixed_const(1000);
666 c.full = dfixed_const(wm->disp_clk);
667 b.full = dfixed_div(c, b);
668 c.full = dfixed_const(wm->bytes_per_pixel);
669 b.full = dfixed_mul(b, c);
671 lb_fill_bw = min(tmp, dfixed_trunc(b));
673 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
674 b.full = dfixed_const(1000);
675 c.full = dfixed_const(lb_fill_bw);
676 b.full = dfixed_div(c, b);
677 a.full = dfixed_div(a, b);
678 line_fill_time = dfixed_trunc(a);
680 if (line_fill_time < wm->active_time)
683 return latency + (line_fill_time - wm->active_time);
687 static bool dce6_average_bandwidth_vs_dram_bandwidth_for_display(struct dce6_wm_params *wm)
689 if (dce6_average_bandwidth(wm) <=
690 (dce6_dram_bandwidth_for_display(wm) / wm->num_heads))
696 static bool dce6_average_bandwidth_vs_available_bandwidth(struct dce6_wm_params *wm)
698 if (dce6_average_bandwidth(wm) <=
699 (dce6_available_bandwidth(wm) / wm->num_heads))
705 static bool dce6_check_latency_hiding(struct dce6_wm_params *wm)
707 u32 lb_partitions = wm->lb_size / wm->src_width;
708 u32 line_time = wm->active_time + wm->blank_time;
709 u32 latency_tolerant_lines;
713 a.full = dfixed_const(1);
714 if (wm->vsc.full > a.full)
715 latency_tolerant_lines = 1;
717 if (lb_partitions <= (wm->vtaps + 1))
718 latency_tolerant_lines = 1;
720 latency_tolerant_lines = 2;
723 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
725 if (dce6_latency_watermark(wm) <= latency_hiding)
731 static void dce6_program_watermarks(struct radeon_device *rdev,
732 struct radeon_crtc *radeon_crtc,
733 u32 lb_size, u32 num_heads)
735 struct drm_display_mode *mode = &radeon_crtc->base.mode;
736 struct dce6_wm_params wm;
739 u32 latency_watermark_a = 0, latency_watermark_b = 0;
740 u32 priority_a_mark = 0, priority_b_mark = 0;
741 u32 priority_a_cnt = PRIORITY_OFF;
742 u32 priority_b_cnt = PRIORITY_OFF;
743 u32 tmp, arb_control3;
746 if (radeon_crtc->base.enabled && num_heads && mode) {
747 pixel_period = 1000000 / (u32)mode->clock;
748 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
752 wm.yclk = rdev->pm.current_mclk * 10;
753 wm.sclk = rdev->pm.current_sclk * 10;
754 wm.disp_clk = mode->clock;
755 wm.src_width = mode->crtc_hdisplay;
756 wm.active_time = mode->crtc_hdisplay * pixel_period;
757 wm.blank_time = line_time - wm.active_time;
758 wm.interlaced = false;
759 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
760 wm.interlaced = true;
761 wm.vsc = radeon_crtc->vsc;
763 if (radeon_crtc->rmx_type != RMX_OFF)
765 wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
766 wm.lb_size = lb_size;
767 wm.dram_channels = dce6_get_number_of_dram_channels(rdev);
768 wm.num_heads = num_heads;
770 /* set for high clocks */
771 latency_watermark_a = min(dce6_latency_watermark(&wm), (u32)65535);
772 /* set for low clocks */
773 /* wm.yclk = low clk; wm.sclk = low clk */
774 latency_watermark_b = min(dce6_latency_watermark(&wm), (u32)65535);
776 /* possibly force display priority to high */
777 /* should really do this at mode validation time... */
778 if (!dce6_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
779 !dce6_average_bandwidth_vs_available_bandwidth(&wm) ||
780 !dce6_check_latency_hiding(&wm) ||
781 (rdev->disp_priority == 2)) {
782 DRM_DEBUG_KMS("force priority to high\n");
783 priority_a_cnt |= PRIORITY_ALWAYS_ON;
784 priority_b_cnt |= PRIORITY_ALWAYS_ON;
787 a.full = dfixed_const(1000);
788 b.full = dfixed_const(mode->clock);
789 b.full = dfixed_div(b, a);
790 c.full = dfixed_const(latency_watermark_a);
791 c.full = dfixed_mul(c, b);
792 c.full = dfixed_mul(c, radeon_crtc->hsc);
793 c.full = dfixed_div(c, a);
794 a.full = dfixed_const(16);
795 c.full = dfixed_div(c, a);
796 priority_a_mark = dfixed_trunc(c);
797 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
799 a.full = dfixed_const(1000);
800 b.full = dfixed_const(mode->clock);
801 b.full = dfixed_div(b, a);
802 c.full = dfixed_const(latency_watermark_b);
803 c.full = dfixed_mul(c, b);
804 c.full = dfixed_mul(c, radeon_crtc->hsc);
805 c.full = dfixed_div(c, a);
806 a.full = dfixed_const(16);
807 c.full = dfixed_div(c, a);
808 priority_b_mark = dfixed_trunc(c);
809 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
813 arb_control3 = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
815 tmp &= ~LATENCY_WATERMARK_MASK(3);
816 tmp |= LATENCY_WATERMARK_MASK(1);
817 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
818 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
819 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
820 LATENCY_HIGH_WATERMARK(line_time)));
822 tmp = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
823 tmp &= ~LATENCY_WATERMARK_MASK(3);
824 tmp |= LATENCY_WATERMARK_MASK(2);
825 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
826 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
827 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
828 LATENCY_HIGH_WATERMARK(line_time)));
829 /* restore original selection */
830 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, arb_control3);
832 /* write the priority marks */
833 WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
834 WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
838 void dce6_bandwidth_update(struct radeon_device *rdev)
840 struct drm_display_mode *mode0 = NULL;
841 struct drm_display_mode *mode1 = NULL;
842 u32 num_heads = 0, lb_size;
845 radeon_update_display_priority(rdev);
847 for (i = 0; i < rdev->num_crtc; i++) {
848 if (rdev->mode_info.crtcs[i]->base.enabled)
851 for (i = 0; i < rdev->num_crtc; i += 2) {
852 mode0 = &rdev->mode_info.crtcs[i]->base.mode;
853 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
854 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
855 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
856 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
857 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
864 static u32 si_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
866 u32 num_backends_per_asic,
867 u32 *backend_disable_mask_per_asic,
868 u32 num_shader_engines)
871 u32 enabled_backends_mask = 0;
872 u32 enabled_backends_count = 0;
873 u32 num_backends_per_se;
875 u32 swizzle_pipe[SI_MAX_PIPES];
878 bool force_no_swizzle;
880 /* force legal values */
881 if (num_tile_pipes < 1)
883 if (num_tile_pipes > rdev->config.si.max_tile_pipes)
884 num_tile_pipes = rdev->config.si.max_tile_pipes;
885 if (num_shader_engines < 1)
886 num_shader_engines = 1;
887 if (num_shader_engines > rdev->config.si.max_shader_engines)
888 num_shader_engines = rdev->config.si.max_shader_engines;
889 if (num_backends_per_asic < num_shader_engines)
890 num_backends_per_asic = num_shader_engines;
891 if (num_backends_per_asic > (rdev->config.si.max_backends_per_se * num_shader_engines))
892 num_backends_per_asic = rdev->config.si.max_backends_per_se * num_shader_engines;
894 /* make sure we have the same number of backends per se */
895 num_backends_per_asic = ALIGN(num_backends_per_asic, num_shader_engines);
896 /* set up the number of backends per se */
897 num_backends_per_se = num_backends_per_asic / num_shader_engines;
898 if (num_backends_per_se > rdev->config.si.max_backends_per_se) {
899 num_backends_per_se = rdev->config.si.max_backends_per_se;
900 num_backends_per_asic = num_backends_per_se * num_shader_engines;
903 /* create enable mask and count for enabled backends */
904 for (i = 0; i < SI_MAX_BACKENDS; ++i) {
905 if (((*backend_disable_mask_per_asic >> i) & 1) == 0) {
906 enabled_backends_mask |= (1 << i);
907 ++enabled_backends_count;
909 if (enabled_backends_count == num_backends_per_asic)
913 /* force the backends mask to match the current number of backends */
914 if (enabled_backends_count != num_backends_per_asic) {
915 u32 this_backend_enabled;
919 enabled_backends_mask = 0;
920 enabled_backends_count = 0;
921 *backend_disable_mask_per_asic = SI_MAX_BACKENDS_MASK;
922 for (i = 0; i < SI_MAX_BACKENDS; ++i) {
923 /* calc the current se */
924 shader_engine = i / rdev->config.si.max_backends_per_se;
925 /* calc the backend per se */
926 backend_per_se = i % rdev->config.si.max_backends_per_se;
927 /* default to not enabled */
928 this_backend_enabled = 0;
929 if ((shader_engine < num_shader_engines) &&
930 (backend_per_se < num_backends_per_se))
931 this_backend_enabled = 1;
932 if (this_backend_enabled) {
933 enabled_backends_mask |= (1 << i);
934 *backend_disable_mask_per_asic &= ~(1 << i);
935 ++enabled_backends_count;
941 memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * SI_MAX_PIPES);
942 switch (rdev->family) {
946 force_no_swizzle = true;
949 force_no_swizzle = false;
952 if (force_no_swizzle) {
953 bool last_backend_enabled = false;
955 force_no_swizzle = false;
956 for (i = 0; i < SI_MAX_BACKENDS; ++i) {
957 if (((enabled_backends_mask >> i) & 1) == 1) {
958 if (last_backend_enabled)
959 force_no_swizzle = true;
960 last_backend_enabled = true;
962 last_backend_enabled = false;
966 switch (num_tile_pipes) {
971 DRM_ERROR("odd number of pipes!\n");
978 if (force_no_swizzle) {
991 if (force_no_swizzle) {
1000 swizzle_pipe[1] = 2;
1001 swizzle_pipe[2] = 4;
1002 swizzle_pipe[3] = 1;
1003 swizzle_pipe[4] = 3;
1004 swizzle_pipe[5] = 5;
1008 if (force_no_swizzle) {
1009 swizzle_pipe[0] = 0;
1010 swizzle_pipe[1] = 1;
1011 swizzle_pipe[2] = 2;
1012 swizzle_pipe[3] = 3;
1013 swizzle_pipe[4] = 4;
1014 swizzle_pipe[5] = 5;
1015 swizzle_pipe[6] = 6;
1016 swizzle_pipe[7] = 7;
1018 swizzle_pipe[0] = 0;
1019 swizzle_pipe[1] = 2;
1020 swizzle_pipe[2] = 4;
1021 swizzle_pipe[3] = 6;
1022 swizzle_pipe[4] = 1;
1023 swizzle_pipe[5] = 3;
1024 swizzle_pipe[6] = 5;
1025 swizzle_pipe[7] = 7;
1030 for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
1031 while (((1 << cur_backend) & enabled_backends_mask) == 0)
1032 cur_backend = (cur_backend + 1) % SI_MAX_BACKENDS;
1034 backend_map |= (((cur_backend & 0xf) << (swizzle_pipe[cur_pipe] * 4)));
1036 cur_backend = (cur_backend + 1) % SI_MAX_BACKENDS;
1042 static u32 si_get_disable_mask_per_asic(struct radeon_device *rdev,
1043 u32 disable_mask_per_se,
1044 u32 max_disable_mask_per_se,
1045 u32 num_shader_engines)
1047 u32 disable_field_width_per_se = r600_count_pipe_bits(disable_mask_per_se);
1048 u32 disable_mask_per_asic = disable_mask_per_se & max_disable_mask_per_se;
1050 if (num_shader_engines == 1)
1051 return disable_mask_per_asic;
1052 else if (num_shader_engines == 2)
1053 return disable_mask_per_asic | (disable_mask_per_asic << disable_field_width_per_se);
1058 static void si_tiling_mode_table_init(struct radeon_device *rdev)
1060 const u32 num_tile_mode_states = 32;
1061 u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
1063 switch (rdev->config.si.mem_row_size_in_kb) {
1065 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
1069 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
1072 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
1076 if ((rdev->family == CHIP_TAHITI) ||
1077 (rdev->family == CHIP_PITCAIRN)) {
1078 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1079 switch (reg_offset) {
1080 case 0: /* non-AA compressed depth or any compressed stencil */
1081 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1082 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1083 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1084 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1085 NUM_BANKS(ADDR_SURF_16_BANK) |
1086 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1087 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1088 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1090 case 1: /* 2xAA/4xAA compressed depth only */
1091 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1092 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1093 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1094 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1095 NUM_BANKS(ADDR_SURF_16_BANK) |
1096 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1097 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1098 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1100 case 2: /* 8xAA compressed depth only */
1101 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1102 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1103 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1104 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1105 NUM_BANKS(ADDR_SURF_16_BANK) |
1106 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1107 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1108 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1110 case 3: /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1111 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1112 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1113 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1114 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1115 NUM_BANKS(ADDR_SURF_16_BANK) |
1116 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1117 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1118 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1120 case 4: /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1121 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1122 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1123 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1124 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1125 NUM_BANKS(ADDR_SURF_16_BANK) |
1126 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1127 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1128 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1130 case 5: /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1131 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1132 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1133 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1134 TILE_SPLIT(split_equal_to_row_size) |
1135 NUM_BANKS(ADDR_SURF_16_BANK) |
1136 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1137 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1138 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1140 case 6: /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1141 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1142 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1143 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1144 TILE_SPLIT(split_equal_to_row_size) |
1145 NUM_BANKS(ADDR_SURF_16_BANK) |
1146 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1147 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1148 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1150 case 7: /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1151 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1152 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1153 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1154 TILE_SPLIT(split_equal_to_row_size) |
1155 NUM_BANKS(ADDR_SURF_16_BANK) |
1156 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1157 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1158 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1160 case 8: /* 1D and 1D Array Surfaces */
1161 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1162 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1163 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1164 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1165 NUM_BANKS(ADDR_SURF_16_BANK) |
1166 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1167 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1168 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1170 case 9: /* Displayable maps. */
1171 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1172 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1173 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1174 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1175 NUM_BANKS(ADDR_SURF_16_BANK) |
1176 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1177 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1178 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1180 case 10: /* Display 8bpp. */
1181 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1182 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1183 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1184 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1185 NUM_BANKS(ADDR_SURF_16_BANK) |
1186 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1187 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1188 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1190 case 11: /* Display 16bpp. */
1191 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1192 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1193 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1194 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1195 NUM_BANKS(ADDR_SURF_16_BANK) |
1196 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1197 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1198 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1200 case 12: /* Display 32bpp. */
1201 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1202 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1203 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1204 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1205 NUM_BANKS(ADDR_SURF_16_BANK) |
1206 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1207 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1208 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1210 case 13: /* Thin. */
1211 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1212 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1213 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1214 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1215 NUM_BANKS(ADDR_SURF_16_BANK) |
1216 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1217 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1218 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1220 case 14: /* Thin 8 bpp. */
1221 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1222 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1223 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1224 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1225 NUM_BANKS(ADDR_SURF_16_BANK) |
1226 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1227 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1228 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1230 case 15: /* Thin 16 bpp. */
1231 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1232 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1233 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1234 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1235 NUM_BANKS(ADDR_SURF_16_BANK) |
1236 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1237 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1238 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1240 case 16: /* Thin 32 bpp. */
1241 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1242 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1243 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1244 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1245 NUM_BANKS(ADDR_SURF_16_BANK) |
1246 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1247 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1248 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1250 case 17: /* Thin 64 bpp. */
1251 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1252 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1253 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1254 TILE_SPLIT(split_equal_to_row_size) |
1255 NUM_BANKS(ADDR_SURF_16_BANK) |
1256 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1257 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1258 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1260 case 21: /* 8 bpp PRT. */
1261 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1262 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1263 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1264 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1265 NUM_BANKS(ADDR_SURF_16_BANK) |
1266 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1267 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1268 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1270 case 22: /* 16 bpp PRT */
1271 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1272 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1273 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1274 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1275 NUM_BANKS(ADDR_SURF_16_BANK) |
1276 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1277 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1278 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1280 case 23: /* 32 bpp PRT */
1281 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1282 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1283 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1284 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1285 NUM_BANKS(ADDR_SURF_16_BANK) |
1286 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1287 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1288 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1290 case 24: /* 64 bpp PRT */
1291 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1292 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1293 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1294 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1295 NUM_BANKS(ADDR_SURF_16_BANK) |
1296 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1297 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1298 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1300 case 25: /* 128 bpp PRT */
1301 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1302 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1303 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1304 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1305 NUM_BANKS(ADDR_SURF_8_BANK) |
1306 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1307 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1308 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1314 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1316 } else if (rdev->family == CHIP_VERDE) {
1317 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1318 switch (reg_offset) {
1319 case 0: /* non-AA compressed depth or any compressed stencil */
1320 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1321 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1322 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1323 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1324 NUM_BANKS(ADDR_SURF_16_BANK) |
1325 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1326 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1327 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1329 case 1: /* 2xAA/4xAA compressed depth only */
1330 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1331 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1332 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1333 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1334 NUM_BANKS(ADDR_SURF_16_BANK) |
1335 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1336 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1337 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1339 case 2: /* 8xAA compressed depth only */
1340 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1341 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1342 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1343 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1344 NUM_BANKS(ADDR_SURF_16_BANK) |
1345 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1346 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1347 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1349 case 3: /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1350 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1351 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1352 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1353 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1354 NUM_BANKS(ADDR_SURF_16_BANK) |
1355 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1356 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1357 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1359 case 4: /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1360 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1361 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1362 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1363 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1364 NUM_BANKS(ADDR_SURF_16_BANK) |
1365 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1366 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1367 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1369 case 5: /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1370 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1371 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1372 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1373 TILE_SPLIT(split_equal_to_row_size) |
1374 NUM_BANKS(ADDR_SURF_16_BANK) |
1375 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1376 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1377 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1379 case 6: /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1380 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1381 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1382 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1383 TILE_SPLIT(split_equal_to_row_size) |
1384 NUM_BANKS(ADDR_SURF_16_BANK) |
1385 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1386 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1387 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1389 case 7: /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1390 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1391 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1392 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1393 TILE_SPLIT(split_equal_to_row_size) |
1394 NUM_BANKS(ADDR_SURF_16_BANK) |
1395 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1396 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1397 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1399 case 8: /* 1D and 1D Array Surfaces */
1400 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1401 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1402 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1403 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1404 NUM_BANKS(ADDR_SURF_16_BANK) |
1405 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1406 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1407 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1409 case 9: /* Displayable maps. */
1410 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1411 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1412 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1413 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1414 NUM_BANKS(ADDR_SURF_16_BANK) |
1415 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1416 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1417 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1419 case 10: /* Display 8bpp. */
1420 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1421 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1422 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1423 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1424 NUM_BANKS(ADDR_SURF_16_BANK) |
1425 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1426 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1427 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1429 case 11: /* Display 16bpp. */
1430 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1431 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1432 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1433 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1434 NUM_BANKS(ADDR_SURF_16_BANK) |
1435 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1436 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1437 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1439 case 12: /* Display 32bpp. */
1440 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1441 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1442 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1443 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1444 NUM_BANKS(ADDR_SURF_16_BANK) |
1445 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1446 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1447 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1449 case 13: /* Thin. */
1450 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1451 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1452 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1453 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1454 NUM_BANKS(ADDR_SURF_16_BANK) |
1455 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1456 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1457 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1459 case 14: /* Thin 8 bpp. */
1460 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1461 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1462 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1463 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1464 NUM_BANKS(ADDR_SURF_16_BANK) |
1465 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1466 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1467 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1469 case 15: /* Thin 16 bpp. */
1470 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1471 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1472 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1473 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1474 NUM_BANKS(ADDR_SURF_16_BANK) |
1475 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1476 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1477 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1479 case 16: /* Thin 32 bpp. */
1480 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1481 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1482 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1483 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1484 NUM_BANKS(ADDR_SURF_16_BANK) |
1485 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1486 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1487 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1489 case 17: /* Thin 64 bpp. */
1490 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1491 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1492 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1493 TILE_SPLIT(split_equal_to_row_size) |
1494 NUM_BANKS(ADDR_SURF_16_BANK) |
1495 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1496 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1497 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1499 case 21: /* 8 bpp PRT. */
1500 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1501 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1502 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1503 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1504 NUM_BANKS(ADDR_SURF_16_BANK) |
1505 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1506 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1507 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1509 case 22: /* 16 bpp PRT */
1510 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1511 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1512 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1513 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1514 NUM_BANKS(ADDR_SURF_16_BANK) |
1515 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1516 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1517 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1519 case 23: /* 32 bpp PRT */
1520 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1521 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1522 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1523 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1524 NUM_BANKS(ADDR_SURF_16_BANK) |
1525 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1526 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1527 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1529 case 24: /* 64 bpp PRT */
1530 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1531 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1532 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1533 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1534 NUM_BANKS(ADDR_SURF_16_BANK) |
1535 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1536 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1537 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1539 case 25: /* 128 bpp PRT */
1540 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1541 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1542 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1543 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1544 NUM_BANKS(ADDR_SURF_8_BANK) |
1545 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1546 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1547 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1553 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1556 DRM_ERROR("unknown asic: 0x%x\n", rdev->family);
1559 static void si_gpu_init(struct radeon_device *rdev)
1561 u32 cc_rb_backend_disable = 0;
1562 u32 cc_gc_shader_array_config;
1563 u32 gb_addr_config = 0;
1564 u32 mc_shared_chmap, mc_arb_ramcfg;
1566 u32 cgts_tcc_disable;
1568 u32 gc_user_shader_array_config;
1569 u32 gc_user_rb_backend_disable;
1570 u32 cgts_user_tcc_disable;
1571 u32 hdp_host_path_cntl;
1575 switch (rdev->family) {
1577 rdev->config.si.max_shader_engines = 2;
1578 rdev->config.si.max_pipes_per_simd = 4;
1579 rdev->config.si.max_tile_pipes = 12;
1580 rdev->config.si.max_simds_per_se = 8;
1581 rdev->config.si.max_backends_per_se = 4;
1582 rdev->config.si.max_texture_channel_caches = 12;
1583 rdev->config.si.max_gprs = 256;
1584 rdev->config.si.max_gs_threads = 32;
1585 rdev->config.si.max_hw_contexts = 8;
1587 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1588 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1589 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1590 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1593 rdev->config.si.max_shader_engines = 2;
1594 rdev->config.si.max_pipes_per_simd = 4;
1595 rdev->config.si.max_tile_pipes = 8;
1596 rdev->config.si.max_simds_per_se = 5;
1597 rdev->config.si.max_backends_per_se = 4;
1598 rdev->config.si.max_texture_channel_caches = 8;
1599 rdev->config.si.max_gprs = 256;
1600 rdev->config.si.max_gs_threads = 32;
1601 rdev->config.si.max_hw_contexts = 8;
1603 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1604 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1605 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1606 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1610 rdev->config.si.max_shader_engines = 1;
1611 rdev->config.si.max_pipes_per_simd = 4;
1612 rdev->config.si.max_tile_pipes = 4;
1613 rdev->config.si.max_simds_per_se = 2;
1614 rdev->config.si.max_backends_per_se = 4;
1615 rdev->config.si.max_texture_channel_caches = 4;
1616 rdev->config.si.max_gprs = 256;
1617 rdev->config.si.max_gs_threads = 32;
1618 rdev->config.si.max_hw_contexts = 8;
1620 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1621 rdev->config.si.sc_prim_fifo_size_backend = 0x40;
1622 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1623 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1627 /* Initialize HDP */
1628 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1629 WREG32((0x2c14 + j), 0x00000000);
1630 WREG32((0x2c18 + j), 0x00000000);
1631 WREG32((0x2c1c + j), 0x00000000);
1632 WREG32((0x2c20 + j), 0x00000000);
1633 WREG32((0x2c24 + j), 0x00000000);
1636 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1638 evergreen_fix_pci_max_read_req_size(rdev);
1640 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
1642 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1643 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1645 cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE);
1646 cc_gc_shader_array_config = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
1647 cgts_tcc_disable = 0xffff0000;
1648 for (i = 0; i < rdev->config.si.max_texture_channel_caches; i++)
1649 cgts_tcc_disable &= ~(1 << (16 + i));
1650 gc_user_rb_backend_disable = RREG32(GC_USER_RB_BACKEND_DISABLE);
1651 gc_user_shader_array_config = RREG32(GC_USER_SHADER_ARRAY_CONFIG);
1652 cgts_user_tcc_disable = RREG32(CGTS_USER_TCC_DISABLE);
1654 rdev->config.si.num_shader_engines = rdev->config.si.max_shader_engines;
1655 rdev->config.si.num_tile_pipes = rdev->config.si.max_tile_pipes;
1656 tmp = ((~gc_user_rb_backend_disable) & BACKEND_DISABLE_MASK) >> BACKEND_DISABLE_SHIFT;
1657 rdev->config.si.num_backends_per_se = r600_count_pipe_bits(tmp);
1658 tmp = (gc_user_rb_backend_disable & BACKEND_DISABLE_MASK) >> BACKEND_DISABLE_SHIFT;
1659 rdev->config.si.backend_disable_mask_per_asic =
1660 si_get_disable_mask_per_asic(rdev, tmp, SI_MAX_BACKENDS_PER_SE_MASK,
1661 rdev->config.si.num_shader_engines);
1662 rdev->config.si.backend_map =
1663 si_get_tile_pipe_to_backend_map(rdev, rdev->config.si.num_tile_pipes,
1664 rdev->config.si.num_backends_per_se *
1665 rdev->config.si.num_shader_engines,
1666 &rdev->config.si.backend_disable_mask_per_asic,
1667 rdev->config.si.num_shader_engines);
1668 tmp = ((~cgts_user_tcc_disable) & TCC_DISABLE_MASK) >> TCC_DISABLE_SHIFT;
1669 rdev->config.si.num_texture_channel_caches = r600_count_pipe_bits(tmp);
1670 rdev->config.si.mem_max_burst_length_bytes = 256;
1671 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
1672 rdev->config.si.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1673 if (rdev->config.si.mem_row_size_in_kb > 4)
1674 rdev->config.si.mem_row_size_in_kb = 4;
1675 /* XXX use MC settings? */
1676 rdev->config.si.shader_engine_tile_size = 32;
1677 rdev->config.si.num_gpus = 1;
1678 rdev->config.si.multi_gpu_tile_size = 64;
1681 switch (rdev->config.si.num_tile_pipes) {
1683 gb_addr_config |= NUM_PIPES(0);
1686 gb_addr_config |= NUM_PIPES(1);
1689 gb_addr_config |= NUM_PIPES(2);
1693 gb_addr_config |= NUM_PIPES(3);
1697 tmp = (rdev->config.si.mem_max_burst_length_bytes / 256) - 1;
1698 gb_addr_config |= PIPE_INTERLEAVE_SIZE(tmp);
1699 gb_addr_config |= NUM_SHADER_ENGINES(rdev->config.si.num_shader_engines - 1);
1700 tmp = (rdev->config.si.shader_engine_tile_size / 16) - 1;
1701 gb_addr_config |= SHADER_ENGINE_TILE_SIZE(tmp);
1702 switch (rdev->config.si.num_gpus) {
1705 gb_addr_config |= NUM_GPUS(0);
1708 gb_addr_config |= NUM_GPUS(1);
1711 gb_addr_config |= NUM_GPUS(2);
1714 switch (rdev->config.si.multi_gpu_tile_size) {
1716 gb_addr_config |= MULTI_GPU_TILE_SIZE(0);
1720 gb_addr_config |= MULTI_GPU_TILE_SIZE(1);
1723 gb_addr_config |= MULTI_GPU_TILE_SIZE(2);
1726 gb_addr_config |= MULTI_GPU_TILE_SIZE(3);
1729 switch (rdev->config.si.mem_row_size_in_kb) {
1732 gb_addr_config |= ROW_SIZE(0);
1735 gb_addr_config |= ROW_SIZE(1);
1738 gb_addr_config |= ROW_SIZE(2);
1742 tmp = (gb_addr_config & NUM_PIPES_MASK) >> NUM_PIPES_SHIFT;
1743 rdev->config.si.num_tile_pipes = (1 << tmp);
1744 tmp = (gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT;
1745 rdev->config.si.mem_max_burst_length_bytes = (tmp + 1) * 256;
1746 tmp = (gb_addr_config & NUM_SHADER_ENGINES_MASK) >> NUM_SHADER_ENGINES_SHIFT;
1747 rdev->config.si.num_shader_engines = tmp + 1;
1748 tmp = (gb_addr_config & NUM_GPUS_MASK) >> NUM_GPUS_SHIFT;
1749 rdev->config.si.num_gpus = tmp + 1;
1750 tmp = (gb_addr_config & MULTI_GPU_TILE_SIZE_MASK) >> MULTI_GPU_TILE_SIZE_SHIFT;
1751 rdev->config.si.multi_gpu_tile_size = 1 << tmp;
1752 tmp = (gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT;
1753 rdev->config.si.mem_row_size_in_kb = 1 << tmp;
1756 si_get_tile_pipe_to_backend_map(rdev, rdev->config.si.num_tile_pipes,
1757 rdev->config.si.num_backends_per_se *
1758 rdev->config.si.num_shader_engines,
1759 &rdev->config.si.backend_disable_mask_per_asic,
1760 rdev->config.si.num_shader_engines);
1762 /* setup tiling info dword. gb_addr_config is not adequate since it does
1763 * not have bank info, so create a custom tiling dword.
1764 * bits 3:0 num_pipes
1765 * bits 7:4 num_banks
1766 * bits 11:8 group_size
1767 * bits 15:12 row_size
1769 rdev->config.si.tile_config = 0;
1770 switch (rdev->config.si.num_tile_pipes) {
1772 rdev->config.si.tile_config |= (0 << 0);
1775 rdev->config.si.tile_config |= (1 << 0);
1778 rdev->config.si.tile_config |= (2 << 0);
1782 /* XXX what about 12? */
1783 rdev->config.si.tile_config |= (3 << 0);
1786 rdev->config.si.tile_config |=
1787 ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) << 4;
1788 rdev->config.si.tile_config |=
1789 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
1790 rdev->config.si.tile_config |=
1791 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
1793 rdev->config.si.backend_map = gb_backend_map;
1794 WREG32(GB_ADDR_CONFIG, gb_addr_config);
1795 WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
1796 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
1798 /* primary versions */
1799 WREG32(CC_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1800 WREG32(CC_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1801 WREG32(CC_GC_SHADER_ARRAY_CONFIG, cc_gc_shader_array_config);
1803 WREG32(CGTS_TCC_DISABLE, cgts_tcc_disable);
1806 WREG32(GC_USER_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1807 WREG32(GC_USER_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1808 WREG32(GC_USER_SHADER_ARRAY_CONFIG, cc_gc_shader_array_config);
1810 WREG32(CGTS_USER_TCC_DISABLE, cgts_tcc_disable);
1812 si_tiling_mode_table_init(rdev);
1814 /* set HW defaults for 3D engine */
1815 WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
1816 ROQ_IB2_START(0x2b)));
1817 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
1819 sx_debug_1 = RREG32(SX_DEBUG_1);
1820 WREG32(SX_DEBUG_1, sx_debug_1);
1822 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
1824 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_frontend) |
1825 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_backend) |
1826 SC_HIZ_TILE_FIFO_SIZE(rdev->config.si.sc_hiz_tile_fifo_size) |
1827 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.si.sc_earlyz_tile_fifo_size)));
1829 WREG32(VGT_NUM_INSTANCES, 1);
1831 WREG32(CP_PERFMON_CNTL, 0);
1833 WREG32(SQ_CONFIG, 0);
1835 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
1836 FORCE_EOV_MAX_REZ_CNT(255)));
1838 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
1839 AUTO_INVLD_EN(ES_AND_GS_AUTO));
1841 WREG32(VGT_GS_VERTEX_REUSE, 16);
1842 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1844 WREG32(CB_PERFCOUNTER0_SELECT0, 0);
1845 WREG32(CB_PERFCOUNTER0_SELECT1, 0);
1846 WREG32(CB_PERFCOUNTER1_SELECT0, 0);
1847 WREG32(CB_PERFCOUNTER1_SELECT1, 0);
1848 WREG32(CB_PERFCOUNTER2_SELECT0, 0);
1849 WREG32(CB_PERFCOUNTER2_SELECT1, 0);
1850 WREG32(CB_PERFCOUNTER3_SELECT0, 0);
1851 WREG32(CB_PERFCOUNTER3_SELECT1, 0);
1853 tmp = RREG32(HDP_MISC_CNTL);
1854 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1855 WREG32(HDP_MISC_CNTL, tmp);
1857 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1858 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1860 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1866 * GPU scratch registers helpers function.
1868 static void si_scratch_init(struct radeon_device *rdev)
1872 rdev->scratch.num_reg = 7;
1873 rdev->scratch.reg_base = SCRATCH_REG0;
1874 for (i = 0; i < rdev->scratch.num_reg; i++) {
1875 rdev->scratch.free[i] = true;
1876 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
1880 void si_fence_ring_emit(struct radeon_device *rdev,
1881 struct radeon_fence *fence)
1883 struct radeon_ring *ring = &rdev->ring[fence->ring];
1884 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1886 /* flush read cache over gart */
1887 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1888 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1889 radeon_ring_write(ring, 0);
1890 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1891 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1892 PACKET3_TC_ACTION_ENA |
1893 PACKET3_SH_KCACHE_ACTION_ENA |
1894 PACKET3_SH_ICACHE_ACTION_ENA);
1895 radeon_ring_write(ring, 0xFFFFFFFF);
1896 radeon_ring_write(ring, 0);
1897 radeon_ring_write(ring, 10); /* poll interval */
1898 /* EVENT_WRITE_EOP - flush caches, send int */
1899 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1900 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5));
1901 radeon_ring_write(ring, addr & 0xffffffff);
1902 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1903 radeon_ring_write(ring, fence->seq);
1904 radeon_ring_write(ring, 0);
1910 void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1912 struct radeon_ring *ring = &rdev->ring[ib->fence->ring];
1915 if (ib->is_const_ib)
1916 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
1918 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
1920 radeon_ring_write(ring, header);
1921 radeon_ring_write(ring,
1925 (ib->gpu_addr & 0xFFFFFFFC));
1926 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
1927 radeon_ring_write(ring, ib->length_dw | (ib->vm_id << 24));
1929 /* flush read cache over gart for this vmid */
1930 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1931 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1932 radeon_ring_write(ring, ib->vm_id);
1933 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1934 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1935 PACKET3_TC_ACTION_ENA |
1936 PACKET3_SH_KCACHE_ACTION_ENA |
1937 PACKET3_SH_ICACHE_ACTION_ENA);
1938 radeon_ring_write(ring, 0xFFFFFFFF);
1939 radeon_ring_write(ring, 0);
1940 radeon_ring_write(ring, 10); /* poll interval */
1946 static void si_cp_enable(struct radeon_device *rdev, bool enable)
1949 WREG32(CP_ME_CNTL, 0);
1951 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1952 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
1953 WREG32(SCRATCH_UMSK, 0);
1958 static int si_cp_load_microcode(struct radeon_device *rdev)
1960 const __be32 *fw_data;
1963 if (!rdev->me_fw || !rdev->pfp_fw)
1966 si_cp_enable(rdev, false);
1969 fw_data = (const __be32 *)rdev->pfp_fw->data;
1970 WREG32(CP_PFP_UCODE_ADDR, 0);
1971 for (i = 0; i < SI_PFP_UCODE_SIZE; i++)
1972 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1973 WREG32(CP_PFP_UCODE_ADDR, 0);
1976 fw_data = (const __be32 *)rdev->ce_fw->data;
1977 WREG32(CP_CE_UCODE_ADDR, 0);
1978 for (i = 0; i < SI_CE_UCODE_SIZE; i++)
1979 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
1980 WREG32(CP_CE_UCODE_ADDR, 0);
1983 fw_data = (const __be32 *)rdev->me_fw->data;
1984 WREG32(CP_ME_RAM_WADDR, 0);
1985 for (i = 0; i < SI_PM4_UCODE_SIZE; i++)
1986 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1987 WREG32(CP_ME_RAM_WADDR, 0);
1989 WREG32(CP_PFP_UCODE_ADDR, 0);
1990 WREG32(CP_CE_UCODE_ADDR, 0);
1991 WREG32(CP_ME_RAM_WADDR, 0);
1992 WREG32(CP_ME_RAM_RADDR, 0);
1996 static int si_cp_start(struct radeon_device *rdev)
1998 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2001 r = radeon_ring_lock(rdev, ring, 7 + 4);
2003 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2007 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
2008 radeon_ring_write(ring, 0x1);
2009 radeon_ring_write(ring, 0x0);
2010 radeon_ring_write(ring, rdev->config.si.max_hw_contexts - 1);
2011 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
2012 radeon_ring_write(ring, 0);
2013 radeon_ring_write(ring, 0);
2015 /* init the CE partitions */
2016 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2017 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2018 radeon_ring_write(ring, 0xc000);
2019 radeon_ring_write(ring, 0xe000);
2020 radeon_ring_unlock_commit(rdev, ring);
2022 si_cp_enable(rdev, true);
2024 r = radeon_ring_lock(rdev, ring, si_default_size + 10);
2026 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2030 /* setup clear context state */
2031 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2032 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2034 for (i = 0; i < si_default_size; i++)
2035 radeon_ring_write(ring, si_default_state[i]);
2037 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2038 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2040 /* set clear context state */
2041 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2042 radeon_ring_write(ring, 0);
2044 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2045 radeon_ring_write(ring, 0x00000316);
2046 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2047 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2049 radeon_ring_unlock_commit(rdev, ring);
2051 for (i = RADEON_RING_TYPE_GFX_INDEX; i <= CAYMAN_RING_TYPE_CP2_INDEX; ++i) {
2052 ring = &rdev->ring[i];
2053 r = radeon_ring_lock(rdev, ring, 2);
2055 /* clear the compute context state */
2056 radeon_ring_write(ring, PACKET3_COMPUTE(PACKET3_CLEAR_STATE, 0));
2057 radeon_ring_write(ring, 0);
2059 radeon_ring_unlock_commit(rdev, ring);
2065 static void si_cp_fini(struct radeon_device *rdev)
2067 si_cp_enable(rdev, false);
2068 radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
2069 radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]);
2070 radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]);
2073 static int si_cp_resume(struct radeon_device *rdev)
2075 struct radeon_ring *ring;
2080 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
2081 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
2086 RREG32(GRBM_SOFT_RESET);
2088 WREG32(GRBM_SOFT_RESET, 0);
2089 RREG32(GRBM_SOFT_RESET);
2091 WREG32(CP_SEM_WAIT_TIMER, 0x0);
2092 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2094 /* Set the write pointer delay */
2095 WREG32(CP_RB_WPTR_DELAY, 0);
2097 WREG32(CP_DEBUG, 0);
2098 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
2100 /* ring 0 - compute and gfx */
2101 /* Set ring buffer size */
2102 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2103 rb_bufsz = drm_order(ring->ring_size / 8);
2104 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2106 tmp |= BUF_SWAP_32BIT;
2108 WREG32(CP_RB0_CNTL, tmp);
2110 /* Initialize the ring buffer's read and write pointers */
2111 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
2113 WREG32(CP_RB0_WPTR, ring->wptr);
2115 /* set the wb address wether it's enabled or not */
2116 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
2117 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
2119 if (rdev->wb.enabled)
2120 WREG32(SCRATCH_UMSK, 0xff);
2122 tmp |= RB_NO_UPDATE;
2123 WREG32(SCRATCH_UMSK, 0);
2127 WREG32(CP_RB0_CNTL, tmp);
2129 WREG32(CP_RB0_BASE, ring->gpu_addr >> 8);
2131 ring->rptr = RREG32(CP_RB0_RPTR);
2133 /* ring1 - compute only */
2134 /* Set ring buffer size */
2135 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
2136 rb_bufsz = drm_order(ring->ring_size / 8);
2137 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2139 tmp |= BUF_SWAP_32BIT;
2141 WREG32(CP_RB1_CNTL, tmp);
2143 /* Initialize the ring buffer's read and write pointers */
2144 WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA);
2146 WREG32(CP_RB1_WPTR, ring->wptr);
2148 /* set the wb address wether it's enabled or not */
2149 WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC);
2150 WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF);
2153 WREG32(CP_RB1_CNTL, tmp);
2155 WREG32(CP_RB1_BASE, ring->gpu_addr >> 8);
2157 ring->rptr = RREG32(CP_RB1_RPTR);
2159 /* ring2 - compute only */
2160 /* Set ring buffer size */
2161 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
2162 rb_bufsz = drm_order(ring->ring_size / 8);
2163 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2165 tmp |= BUF_SWAP_32BIT;
2167 WREG32(CP_RB2_CNTL, tmp);
2169 /* Initialize the ring buffer's read and write pointers */
2170 WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA);
2172 WREG32(CP_RB2_WPTR, ring->wptr);
2174 /* set the wb address wether it's enabled or not */
2175 WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC);
2176 WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF);
2179 WREG32(CP_RB2_CNTL, tmp);
2181 WREG32(CP_RB2_BASE, ring->gpu_addr >> 8);
2183 ring->rptr = RREG32(CP_RB2_RPTR);
2185 /* start the rings */
2187 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
2188 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = true;
2189 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = true;
2190 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
2192 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
2193 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2194 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2197 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP1_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]);
2199 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2201 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP2_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]);
2203 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2209 bool si_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2212 u32 grbm_status, grbm_status2;
2213 u32 grbm_status_se0, grbm_status_se1;
2214 struct r100_gpu_lockup *lockup = &rdev->config.si.lockup;
2217 srbm_status = RREG32(SRBM_STATUS);
2218 grbm_status = RREG32(GRBM_STATUS);
2219 grbm_status2 = RREG32(GRBM_STATUS2);
2220 grbm_status_se0 = RREG32(GRBM_STATUS_SE0);
2221 grbm_status_se1 = RREG32(GRBM_STATUS_SE1);
2222 if (!(grbm_status & GUI_ACTIVE)) {
2223 r100_gpu_lockup_update(lockup, ring);
2226 /* force CP activities */
2227 r = radeon_ring_lock(rdev, ring, 2);
2230 radeon_ring_write(ring, 0x80000000);
2231 radeon_ring_write(ring, 0x80000000);
2232 radeon_ring_unlock_commit(rdev, ring);
2234 /* XXX deal with CP0,1,2 */
2235 ring->rptr = RREG32(ring->rptr_reg);
2236 return r100_gpu_cp_is_lockup(rdev, lockup, ring);
2239 static int si_gpu_soft_reset(struct radeon_device *rdev)
2241 struct evergreen_mc_save save;
2244 if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
2247 dev_info(rdev->dev, "GPU softreset \n");
2248 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
2249 RREG32(GRBM_STATUS));
2250 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
2251 RREG32(GRBM_STATUS2));
2252 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
2253 RREG32(GRBM_STATUS_SE0));
2254 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
2255 RREG32(GRBM_STATUS_SE1));
2256 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
2257 RREG32(SRBM_STATUS));
2258 evergreen_mc_stop(rdev, &save);
2259 if (radeon_mc_wait_for_idle(rdev)) {
2260 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2262 /* Disable CP parsing/prefetching */
2263 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
2265 /* reset all the gfx blocks */
2266 grbm_reset = (SOFT_RESET_CP |
2279 dev_info(rdev->dev, " GRBM_SOFT_RESET=0x%08X\n", grbm_reset);
2280 WREG32(GRBM_SOFT_RESET, grbm_reset);
2281 (void)RREG32(GRBM_SOFT_RESET);
2283 WREG32(GRBM_SOFT_RESET, 0);
2284 (void)RREG32(GRBM_SOFT_RESET);
2285 /* Wait a little for things to settle down */
2287 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
2288 RREG32(GRBM_STATUS));
2289 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
2290 RREG32(GRBM_STATUS2));
2291 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
2292 RREG32(GRBM_STATUS_SE0));
2293 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
2294 RREG32(GRBM_STATUS_SE1));
2295 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
2296 RREG32(SRBM_STATUS));
2297 evergreen_mc_resume(rdev, &save);
2301 int si_asic_reset(struct radeon_device *rdev)
2303 return si_gpu_soft_reset(rdev);
2307 static void si_mc_program(struct radeon_device *rdev)
2309 struct evergreen_mc_save save;
2313 /* Initialize HDP */
2314 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
2315 WREG32((0x2c14 + j), 0x00000000);
2316 WREG32((0x2c18 + j), 0x00000000);
2317 WREG32((0x2c1c + j), 0x00000000);
2318 WREG32((0x2c20 + j), 0x00000000);
2319 WREG32((0x2c24 + j), 0x00000000);
2321 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
2323 evergreen_mc_stop(rdev, &save);
2324 if (radeon_mc_wait_for_idle(rdev)) {
2325 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2327 /* Lockout access through VGA aperture*/
2328 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
2329 /* Update configuration */
2330 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2331 rdev->mc.vram_start >> 12);
2332 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2333 rdev->mc.vram_end >> 12);
2334 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
2335 rdev->vram_scratch.gpu_addr >> 12);
2336 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
2337 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
2338 WREG32(MC_VM_FB_LOCATION, tmp);
2339 /* XXX double check these! */
2340 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
2341 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
2342 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
2343 WREG32(MC_VM_AGP_BASE, 0);
2344 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
2345 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
2346 if (radeon_mc_wait_for_idle(rdev)) {
2347 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2349 evergreen_mc_resume(rdev, &save);
2350 /* we need to own VRAM, so turn off the VGA renderer here
2351 * to stop it overwriting our objects */
2352 rv515_vga_render_disable(rdev);
2355 /* SI MC address space is 40 bits */
2356 static void si_vram_location(struct radeon_device *rdev,
2357 struct radeon_mc *mc, u64 base)
2359 mc->vram_start = base;
2360 if (mc->mc_vram_size > (0xFFFFFFFFFFULL - base + 1)) {
2361 dev_warn(rdev->dev, "limiting VRAM to PCI aperture size\n");
2362 mc->real_vram_size = mc->aper_size;
2363 mc->mc_vram_size = mc->aper_size;
2365 mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
2366 dev_info(rdev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
2367 mc->mc_vram_size >> 20, mc->vram_start,
2368 mc->vram_end, mc->real_vram_size >> 20);
2371 static void si_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
2373 u64 size_af, size_bf;
2375 size_af = ((0xFFFFFFFFFFULL - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align;
2376 size_bf = mc->vram_start & ~mc->gtt_base_align;
2377 if (size_bf > size_af) {
2378 if (mc->gtt_size > size_bf) {
2379 dev_warn(rdev->dev, "limiting GTT\n");
2380 mc->gtt_size = size_bf;
2382 mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size;
2384 if (mc->gtt_size > size_af) {
2385 dev_warn(rdev->dev, "limiting GTT\n");
2386 mc->gtt_size = size_af;
2388 mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align;
2390 mc->gtt_end = mc->gtt_start + mc->gtt_size - 1;
2391 dev_info(rdev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n",
2392 mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end);
2395 static void si_vram_gtt_location(struct radeon_device *rdev,
2396 struct radeon_mc *mc)
2398 if (mc->mc_vram_size > 0xFFC0000000ULL) {
2399 /* leave room for at least 1024M GTT */
2400 dev_warn(rdev->dev, "limiting VRAM\n");
2401 mc->real_vram_size = 0xFFC0000000ULL;
2402 mc->mc_vram_size = 0xFFC0000000ULL;
2404 si_vram_location(rdev, &rdev->mc, 0);
2405 rdev->mc.gtt_base_align = 0;
2406 si_gtt_location(rdev, mc);
2409 static int si_mc_init(struct radeon_device *rdev)
2412 int chansize, numchan;
2414 /* Get VRAM informations */
2415 rdev->mc.vram_is_ddr = true;
2416 tmp = RREG32(MC_ARB_RAMCFG);
2417 if (tmp & CHANSIZE_OVERRIDE) {
2419 } else if (tmp & CHANSIZE_MASK) {
2424 tmp = RREG32(MC_SHARED_CHMAP);
2425 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
2455 rdev->mc.vram_width = numchan * chansize;
2456 /* Could aper size report 0 ? */
2457 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2458 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2459 /* size in MB on si */
2460 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2461 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2462 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2463 si_vram_gtt_location(rdev, &rdev->mc);
2464 radeon_update_bandwidth_info(rdev);
2472 void si_pcie_gart_tlb_flush(struct radeon_device *rdev)
2474 /* flush hdp cache */
2475 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2477 /* bits 0-15 are the VM contexts0-15 */
2478 WREG32(VM_INVALIDATE_REQUEST, 1);
2481 int si_pcie_gart_enable(struct radeon_device *rdev)
2485 if (rdev->gart.robj == NULL) {
2486 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
2489 r = radeon_gart_table_vram_pin(rdev);
2492 radeon_gart_restore(rdev);
2493 /* Setup TLB control */
2494 WREG32(MC_VM_MX_L1_TLB_CNTL,
2497 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2498 ENABLE_ADVANCED_DRIVER_MODEL |
2499 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2500 /* Setup L2 cache */
2501 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
2502 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2503 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2504 EFFECTIVE_L2_QUEUE_SIZE(7) |
2505 CONTEXT1_IDENTITY_ACCESS_MODE(1));
2506 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
2507 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2508 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2509 /* setup context0 */
2510 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
2511 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
2512 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
2513 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
2514 (u32)(rdev->dummy_page.addr >> 12));
2515 WREG32(VM_CONTEXT0_CNTL2, 0);
2516 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
2517 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
2523 /* empty context1-15 */
2524 /* FIXME start with 1G, once using 2 level pt switch to full
2527 /* set vm size, must be a multiple of 4 */
2528 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
2529 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, (1 << 30) / RADEON_GPU_PAGE_SIZE);
2530 for (i = 1; i < 16; i++) {
2532 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
2533 rdev->gart.table_addr >> 12);
2535 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
2536 rdev->gart.table_addr >> 12);
2539 /* enable context1-15 */
2540 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
2541 (u32)(rdev->dummy_page.addr >> 12));
2542 WREG32(VM_CONTEXT1_CNTL2, 0);
2543 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
2544 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
2546 si_pcie_gart_tlb_flush(rdev);
2547 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
2548 (unsigned)(rdev->mc.gtt_size >> 20),
2549 (unsigned long long)rdev->gart.table_addr);
2550 rdev->gart.ready = true;
2554 void si_pcie_gart_disable(struct radeon_device *rdev)
2556 /* Disable all tables */
2557 WREG32(VM_CONTEXT0_CNTL, 0);
2558 WREG32(VM_CONTEXT1_CNTL, 0);
2559 /* Setup TLB control */
2560 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2561 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2562 /* Setup L2 cache */
2563 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2564 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2565 EFFECTIVE_L2_QUEUE_SIZE(7) |
2566 CONTEXT1_IDENTITY_ACCESS_MODE(1));
2567 WREG32(VM_L2_CNTL2, 0);
2568 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2569 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2570 radeon_gart_table_vram_unpin(rdev);
2573 void si_pcie_gart_fini(struct radeon_device *rdev)
2575 si_pcie_gart_disable(rdev);
2576 radeon_gart_table_vram_free(rdev);
2577 radeon_gart_fini(rdev);
2581 static bool si_vm_reg_valid(u32 reg)
2583 /* context regs are fine */
2587 /* check config regs */
2589 case GRBM_GFX_INDEX:
2590 case VGT_VTX_VECT_EJECT_REG:
2591 case VGT_CACHE_INVALIDATION:
2592 case VGT_ESGS_RING_SIZE:
2593 case VGT_GSVS_RING_SIZE:
2594 case VGT_GS_VERTEX_REUSE:
2595 case VGT_PRIMITIVE_TYPE:
2596 case VGT_INDEX_TYPE:
2597 case VGT_NUM_INDICES:
2598 case VGT_NUM_INSTANCES:
2599 case VGT_TF_RING_SIZE:
2600 case VGT_HS_OFFCHIP_PARAM:
2601 case VGT_TF_MEMORY_BASE:
2603 case PA_SU_LINE_STIPPLE_VALUE:
2604 case PA_SC_LINE_STIPPLE_STATE:
2607 case SPI_STATIC_THREAD_MGMT_1:
2608 case SPI_STATIC_THREAD_MGMT_2:
2609 case SPI_STATIC_THREAD_MGMT_3:
2610 case SPI_PS_MAX_WAVE_ID:
2611 case SPI_CONFIG_CNTL:
2612 case SPI_CONFIG_CNTL_1:
2616 DRM_ERROR("Invalid register 0x%x in CS\n", reg);
2621 static int si_vm_packet3_ce_check(struct radeon_device *rdev,
2622 u32 *ib, struct radeon_cs_packet *pkt)
2624 switch (pkt->opcode) {
2626 case PACKET3_SET_BASE:
2627 case PACKET3_SET_CE_DE_COUNTERS:
2628 case PACKET3_LOAD_CONST_RAM:
2629 case PACKET3_WRITE_CONST_RAM:
2630 case PACKET3_WRITE_CONST_RAM_OFFSET:
2631 case PACKET3_DUMP_CONST_RAM:
2632 case PACKET3_INCREMENT_CE_COUNTER:
2633 case PACKET3_WAIT_ON_DE_COUNTER:
2634 case PACKET3_CE_WRITE:
2637 DRM_ERROR("Invalid CE packet3: 0x%x\n", pkt->opcode);
2643 static int si_vm_packet3_gfx_check(struct radeon_device *rdev,
2644 u32 *ib, struct radeon_cs_packet *pkt)
2646 u32 idx = pkt->idx + 1;
2647 u32 idx_value = ib[idx];
2648 u32 start_reg, end_reg, reg, i;
2650 switch (pkt->opcode) {
2652 case PACKET3_SET_BASE:
2653 case PACKET3_CLEAR_STATE:
2654 case PACKET3_INDEX_BUFFER_SIZE:
2655 case PACKET3_DISPATCH_DIRECT:
2656 case PACKET3_DISPATCH_INDIRECT:
2657 case PACKET3_ALLOC_GDS:
2658 case PACKET3_WRITE_GDS_RAM:
2659 case PACKET3_ATOMIC_GDS:
2660 case PACKET3_ATOMIC:
2661 case PACKET3_OCCLUSION_QUERY:
2662 case PACKET3_SET_PREDICATION:
2663 case PACKET3_COND_EXEC:
2664 case PACKET3_PRED_EXEC:
2665 case PACKET3_DRAW_INDIRECT:
2666 case PACKET3_DRAW_INDEX_INDIRECT:
2667 case PACKET3_INDEX_BASE:
2668 case PACKET3_DRAW_INDEX_2:
2669 case PACKET3_CONTEXT_CONTROL:
2670 case PACKET3_INDEX_TYPE:
2671 case PACKET3_DRAW_INDIRECT_MULTI:
2672 case PACKET3_DRAW_INDEX_AUTO:
2673 case PACKET3_DRAW_INDEX_IMMD:
2674 case PACKET3_NUM_INSTANCES:
2675 case PACKET3_DRAW_INDEX_MULTI_AUTO:
2676 case PACKET3_STRMOUT_BUFFER_UPDATE:
2677 case PACKET3_DRAW_INDEX_OFFSET_2:
2678 case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
2679 case PACKET3_DRAW_INDEX_INDIRECT_MULTI:
2680 case PACKET3_MPEG_INDEX:
2681 case PACKET3_WAIT_REG_MEM:
2682 case PACKET3_MEM_WRITE:
2683 case PACKET3_PFP_SYNC_ME:
2684 case PACKET3_SURFACE_SYNC:
2685 case PACKET3_EVENT_WRITE:
2686 case PACKET3_EVENT_WRITE_EOP:
2687 case PACKET3_EVENT_WRITE_EOS:
2688 case PACKET3_SET_CONTEXT_REG:
2689 case PACKET3_SET_CONTEXT_REG_INDIRECT:
2690 case PACKET3_SET_SH_REG:
2691 case PACKET3_SET_SH_REG_OFFSET:
2692 case PACKET3_INCREMENT_DE_COUNTER:
2693 case PACKET3_WAIT_ON_CE_COUNTER:
2694 case PACKET3_WAIT_ON_AVAIL_BUFFER:
2695 case PACKET3_ME_WRITE:
2697 case PACKET3_COPY_DATA:
2698 if ((idx_value & 0xf00) == 0) {
2699 reg = ib[idx + 3] * 4;
2700 if (!si_vm_reg_valid(reg))
2704 case PACKET3_WRITE_DATA:
2705 if ((idx_value & 0xf00) == 0) {
2706 start_reg = ib[idx + 1] * 4;
2707 if (idx_value & 0x10000) {
2708 if (!si_vm_reg_valid(start_reg))
2711 for (i = 0; i < (pkt->count - 2); i++) {
2712 reg = start_reg + (4 * i);
2713 if (!si_vm_reg_valid(reg))
2719 case PACKET3_COND_WRITE:
2720 if (idx_value & 0x100) {
2721 reg = ib[idx + 5] * 4;
2722 if (!si_vm_reg_valid(reg))
2726 case PACKET3_COPY_DW:
2727 if (idx_value & 0x2) {
2728 reg = ib[idx + 3] * 4;
2729 if (!si_vm_reg_valid(reg))
2733 case PACKET3_SET_CONFIG_REG:
2734 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2735 end_reg = 4 * pkt->count + start_reg - 4;
2736 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2737 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2738 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2739 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2742 for (i = 0; i < pkt->count; i++) {
2743 reg = start_reg + (4 * i);
2744 if (!si_vm_reg_valid(reg))
2749 DRM_ERROR("Invalid GFX packet3: 0x%x\n", pkt->opcode);
2755 static int si_vm_packet3_compute_check(struct radeon_device *rdev,
2756 u32 *ib, struct radeon_cs_packet *pkt)
2758 u32 idx = pkt->idx + 1;
2759 u32 idx_value = ib[idx];
2760 u32 start_reg, reg, i;
2762 switch (pkt->opcode) {
2764 case PACKET3_SET_BASE:
2765 case PACKET3_CLEAR_STATE:
2766 case PACKET3_DISPATCH_DIRECT:
2767 case PACKET3_DISPATCH_INDIRECT:
2768 case PACKET3_ALLOC_GDS:
2769 case PACKET3_WRITE_GDS_RAM:
2770 case PACKET3_ATOMIC_GDS:
2771 case PACKET3_ATOMIC:
2772 case PACKET3_OCCLUSION_QUERY:
2773 case PACKET3_SET_PREDICATION:
2774 case PACKET3_COND_EXEC:
2775 case PACKET3_PRED_EXEC:
2776 case PACKET3_CONTEXT_CONTROL:
2777 case PACKET3_STRMOUT_BUFFER_UPDATE:
2778 case PACKET3_WAIT_REG_MEM:
2779 case PACKET3_MEM_WRITE:
2780 case PACKET3_PFP_SYNC_ME:
2781 case PACKET3_SURFACE_SYNC:
2782 case PACKET3_EVENT_WRITE:
2783 case PACKET3_EVENT_WRITE_EOP:
2784 case PACKET3_EVENT_WRITE_EOS:
2785 case PACKET3_SET_CONTEXT_REG:
2786 case PACKET3_SET_CONTEXT_REG_INDIRECT:
2787 case PACKET3_SET_SH_REG:
2788 case PACKET3_SET_SH_REG_OFFSET:
2789 case PACKET3_INCREMENT_DE_COUNTER:
2790 case PACKET3_WAIT_ON_CE_COUNTER:
2791 case PACKET3_WAIT_ON_AVAIL_BUFFER:
2792 case PACKET3_ME_WRITE:
2794 case PACKET3_COPY_DATA:
2795 if ((idx_value & 0xf00) == 0) {
2796 reg = ib[idx + 3] * 4;
2797 if (!si_vm_reg_valid(reg))
2801 case PACKET3_WRITE_DATA:
2802 if ((idx_value & 0xf00) == 0) {
2803 start_reg = ib[idx + 1] * 4;
2804 if (idx_value & 0x10000) {
2805 if (!si_vm_reg_valid(start_reg))
2808 for (i = 0; i < (pkt->count - 2); i++) {
2809 reg = start_reg + (4 * i);
2810 if (!si_vm_reg_valid(reg))
2816 case PACKET3_COND_WRITE:
2817 if (idx_value & 0x100) {
2818 reg = ib[idx + 5] * 4;
2819 if (!si_vm_reg_valid(reg))
2823 case PACKET3_COPY_DW:
2824 if (idx_value & 0x2) {
2825 reg = ib[idx + 3] * 4;
2826 if (!si_vm_reg_valid(reg))
2831 DRM_ERROR("Invalid Compute packet3: 0x%x\n", pkt->opcode);
2837 int si_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
2841 struct radeon_cs_packet pkt;
2845 pkt.type = CP_PACKET_GET_TYPE(ib->ptr[idx]);
2846 pkt.count = CP_PACKET_GET_COUNT(ib->ptr[idx]);
2850 dev_err(rdev->dev, "Packet0 not allowed!\n");
2857 pkt.opcode = CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
2858 if (ib->is_const_ib)
2859 ret = si_vm_packet3_ce_check(rdev, ib->ptr, &pkt);
2861 switch (ib->fence->ring) {
2862 case RADEON_RING_TYPE_GFX_INDEX:
2863 ret = si_vm_packet3_gfx_check(rdev, ib->ptr, &pkt);
2865 case CAYMAN_RING_TYPE_CP1_INDEX:
2866 case CAYMAN_RING_TYPE_CP2_INDEX:
2867 ret = si_vm_packet3_compute_check(rdev, ib->ptr, &pkt);
2870 dev_err(rdev->dev, "Non-PM4 ring %d !\n", ib->fence->ring);
2875 idx += pkt.count + 2;
2878 dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
2884 } while (idx < ib->length_dw);
2892 int si_vm_init(struct radeon_device *rdev)
2895 rdev->vm_manager.nvm = 16;
2896 /* base offset of vram pages */
2897 rdev->vm_manager.vram_base_offset = 0;
2902 void si_vm_fini(struct radeon_device *rdev)
2906 int si_vm_bind(struct radeon_device *rdev, struct radeon_vm *vm, int id)
2909 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (id << 2), vm->pt_gpu_addr >> 12);
2911 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((id - 8) << 2),
2912 vm->pt_gpu_addr >> 12);
2913 /* flush hdp cache */
2914 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2915 /* bits 0-15 are the VM contexts0-15 */
2916 WREG32(VM_INVALIDATE_REQUEST, 1 << id);
2920 void si_vm_unbind(struct radeon_device *rdev, struct radeon_vm *vm)
2923 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2), 0);
2925 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2), 0);
2926 /* flush hdp cache */
2927 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2928 /* bits 0-15 are the VM contexts0-15 */
2929 WREG32(VM_INVALIDATE_REQUEST, 1 << vm->id);
2932 void si_vm_tlb_flush(struct radeon_device *rdev, struct radeon_vm *vm)
2937 /* flush hdp cache */
2938 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2939 /* bits 0-15 are the VM contexts0-15 */
2940 WREG32(VM_INVALIDATE_REQUEST, 1 << vm->id);
2946 static void si_rlc_fini(struct radeon_device *rdev)
2950 /* save restore block */
2951 if (rdev->rlc.save_restore_obj) {
2952 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
2953 if (unlikely(r != 0))
2954 dev_warn(rdev->dev, "(%d) reserve RLC sr bo failed\n", r);
2955 radeon_bo_unpin(rdev->rlc.save_restore_obj);
2956 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
2958 radeon_bo_unref(&rdev->rlc.save_restore_obj);
2959 rdev->rlc.save_restore_obj = NULL;
2962 /* clear state block */
2963 if (rdev->rlc.clear_state_obj) {
2964 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
2965 if (unlikely(r != 0))
2966 dev_warn(rdev->dev, "(%d) reserve RLC c bo failed\n", r);
2967 radeon_bo_unpin(rdev->rlc.clear_state_obj);
2968 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
2970 radeon_bo_unref(&rdev->rlc.clear_state_obj);
2971 rdev->rlc.clear_state_obj = NULL;
2975 static int si_rlc_init(struct radeon_device *rdev)
2979 /* save restore block */
2980 if (rdev->rlc.save_restore_obj == NULL) {
2981 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
2982 RADEON_GEM_DOMAIN_VRAM, &rdev->rlc.save_restore_obj);
2984 dev_warn(rdev->dev, "(%d) create RLC sr bo failed\n", r);
2989 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
2990 if (unlikely(r != 0)) {
2994 r = radeon_bo_pin(rdev->rlc.save_restore_obj, RADEON_GEM_DOMAIN_VRAM,
2995 &rdev->rlc.save_restore_gpu_addr);
2997 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
2998 dev_warn(rdev->dev, "(%d) pin RLC sr bo failed\n", r);
3003 /* clear state block */
3004 if (rdev->rlc.clear_state_obj == NULL) {
3005 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3006 RADEON_GEM_DOMAIN_VRAM, &rdev->rlc.clear_state_obj);
3008 dev_warn(rdev->dev, "(%d) create RLC c bo failed\n", r);
3013 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3014 if (unlikely(r != 0)) {
3018 r = radeon_bo_pin(rdev->rlc.clear_state_obj, RADEON_GEM_DOMAIN_VRAM,
3019 &rdev->rlc.clear_state_gpu_addr);
3022 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3023 dev_warn(rdev->dev, "(%d) pin RLC c bo failed\n", r);
3031 static void si_rlc_stop(struct radeon_device *rdev)
3033 WREG32(RLC_CNTL, 0);
3036 static void si_rlc_start(struct radeon_device *rdev)
3038 WREG32(RLC_CNTL, RLC_ENABLE);
3041 static int si_rlc_resume(struct radeon_device *rdev)
3044 const __be32 *fw_data;
3051 WREG32(RLC_RL_BASE, 0);
3052 WREG32(RLC_RL_SIZE, 0);
3053 WREG32(RLC_LB_CNTL, 0);
3054 WREG32(RLC_LB_CNTR_MAX, 0xffffffff);
3055 WREG32(RLC_LB_CNTR_INIT, 0);
3057 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
3058 WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8);
3060 WREG32(RLC_MC_CNTL, 0);
3061 WREG32(RLC_UCODE_CNTL, 0);
3063 fw_data = (const __be32 *)rdev->rlc_fw->data;
3064 for (i = 0; i < SI_RLC_UCODE_SIZE; i++) {
3065 WREG32(RLC_UCODE_ADDR, i);
3066 WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3068 WREG32(RLC_UCODE_ADDR, 0);