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 <drm/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 int r600_ih_ring_alloc(struct radeon_device *rdev);
59 extern void r600_ih_ring_fini(struct radeon_device *rdev);
60 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
61 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
62 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
63 extern u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev);
65 /* get temperature in millidegrees */
66 int si_get_temp(struct radeon_device *rdev)
71 temp = (RREG32(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
77 actual_temp = temp & 0x1ff;
79 actual_temp = (actual_temp * 1000);
84 #define TAHITI_IO_MC_REGS_SIZE 36
86 static const u32 tahiti_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
87 {0x0000006f, 0x03044000},
88 {0x00000070, 0x0480c018},
89 {0x00000071, 0x00000040},
90 {0x00000072, 0x01000000},
91 {0x00000074, 0x000000ff},
92 {0x00000075, 0x00143400},
93 {0x00000076, 0x08ec0800},
94 {0x00000077, 0x040000cc},
95 {0x00000079, 0x00000000},
96 {0x0000007a, 0x21000409},
97 {0x0000007c, 0x00000000},
98 {0x0000007d, 0xe8000000},
99 {0x0000007e, 0x044408a8},
100 {0x0000007f, 0x00000003},
101 {0x00000080, 0x00000000},
102 {0x00000081, 0x01000000},
103 {0x00000082, 0x02000000},
104 {0x00000083, 0x00000000},
105 {0x00000084, 0xe3f3e4f4},
106 {0x00000085, 0x00052024},
107 {0x00000087, 0x00000000},
108 {0x00000088, 0x66036603},
109 {0x00000089, 0x01000000},
110 {0x0000008b, 0x1c0a0000},
111 {0x0000008c, 0xff010000},
112 {0x0000008e, 0xffffefff},
113 {0x0000008f, 0xfff3efff},
114 {0x00000090, 0xfff3efbf},
115 {0x00000094, 0x00101101},
116 {0x00000095, 0x00000fff},
117 {0x00000096, 0x00116fff},
118 {0x00000097, 0x60010000},
119 {0x00000098, 0x10010000},
120 {0x00000099, 0x00006000},
121 {0x0000009a, 0x00001000},
122 {0x0000009f, 0x00a77400}
125 static const u32 pitcairn_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
126 {0x0000006f, 0x03044000},
127 {0x00000070, 0x0480c018},
128 {0x00000071, 0x00000040},
129 {0x00000072, 0x01000000},
130 {0x00000074, 0x000000ff},
131 {0x00000075, 0x00143400},
132 {0x00000076, 0x08ec0800},
133 {0x00000077, 0x040000cc},
134 {0x00000079, 0x00000000},
135 {0x0000007a, 0x21000409},
136 {0x0000007c, 0x00000000},
137 {0x0000007d, 0xe8000000},
138 {0x0000007e, 0x044408a8},
139 {0x0000007f, 0x00000003},
140 {0x00000080, 0x00000000},
141 {0x00000081, 0x01000000},
142 {0x00000082, 0x02000000},
143 {0x00000083, 0x00000000},
144 {0x00000084, 0xe3f3e4f4},
145 {0x00000085, 0x00052024},
146 {0x00000087, 0x00000000},
147 {0x00000088, 0x66036603},
148 {0x00000089, 0x01000000},
149 {0x0000008b, 0x1c0a0000},
150 {0x0000008c, 0xff010000},
151 {0x0000008e, 0xffffefff},
152 {0x0000008f, 0xfff3efff},
153 {0x00000090, 0xfff3efbf},
154 {0x00000094, 0x00101101},
155 {0x00000095, 0x00000fff},
156 {0x00000096, 0x00116fff},
157 {0x00000097, 0x60010000},
158 {0x00000098, 0x10010000},
159 {0x00000099, 0x00006000},
160 {0x0000009a, 0x00001000},
161 {0x0000009f, 0x00a47400}
164 static const u32 verde_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
165 {0x0000006f, 0x03044000},
166 {0x00000070, 0x0480c018},
167 {0x00000071, 0x00000040},
168 {0x00000072, 0x01000000},
169 {0x00000074, 0x000000ff},
170 {0x00000075, 0x00143400},
171 {0x00000076, 0x08ec0800},
172 {0x00000077, 0x040000cc},
173 {0x00000079, 0x00000000},
174 {0x0000007a, 0x21000409},
175 {0x0000007c, 0x00000000},
176 {0x0000007d, 0xe8000000},
177 {0x0000007e, 0x044408a8},
178 {0x0000007f, 0x00000003},
179 {0x00000080, 0x00000000},
180 {0x00000081, 0x01000000},
181 {0x00000082, 0x02000000},
182 {0x00000083, 0x00000000},
183 {0x00000084, 0xe3f3e4f4},
184 {0x00000085, 0x00052024},
185 {0x00000087, 0x00000000},
186 {0x00000088, 0x66036603},
187 {0x00000089, 0x01000000},
188 {0x0000008b, 0x1c0a0000},
189 {0x0000008c, 0xff010000},
190 {0x0000008e, 0xffffefff},
191 {0x0000008f, 0xfff3efff},
192 {0x00000090, 0xfff3efbf},
193 {0x00000094, 0x00101101},
194 {0x00000095, 0x00000fff},
195 {0x00000096, 0x00116fff},
196 {0x00000097, 0x60010000},
197 {0x00000098, 0x10010000},
198 {0x00000099, 0x00006000},
199 {0x0000009a, 0x00001000},
200 {0x0000009f, 0x00a37400}
204 static int si_mc_load_microcode(struct radeon_device *rdev)
206 const __be32 *fw_data;
207 u32 running, blackout = 0;
209 int i, ucode_size, regs_size;
214 switch (rdev->family) {
216 io_mc_regs = (u32 *)&tahiti_io_mc_regs;
217 ucode_size = SI_MC_UCODE_SIZE;
218 regs_size = TAHITI_IO_MC_REGS_SIZE;
221 io_mc_regs = (u32 *)&pitcairn_io_mc_regs;
222 ucode_size = SI_MC_UCODE_SIZE;
223 regs_size = TAHITI_IO_MC_REGS_SIZE;
227 io_mc_regs = (u32 *)&verde_io_mc_regs;
228 ucode_size = SI_MC_UCODE_SIZE;
229 regs_size = TAHITI_IO_MC_REGS_SIZE;
233 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
237 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
238 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
241 /* reset the engine and set to writable */
242 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
243 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
245 /* load mc io regs */
246 for (i = 0; i < regs_size; i++) {
247 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
248 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
250 /* load the MC ucode */
251 fw_data = (const __be32 *)rdev->mc_fw->data;
252 for (i = 0; i < ucode_size; i++)
253 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
255 /* put the engine back into the active state */
256 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
257 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
258 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
260 /* wait for training to complete */
261 for (i = 0; i < rdev->usec_timeout; i++) {
262 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
266 for (i = 0; i < rdev->usec_timeout; i++) {
267 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
273 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
279 static int si_init_microcode(struct radeon_device *rdev)
281 struct platform_device *pdev;
282 const char *chip_name;
283 const char *rlc_chip_name;
284 size_t pfp_req_size, me_req_size, ce_req_size, rlc_req_size, mc_req_size;
290 pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
293 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
297 switch (rdev->family) {
299 chip_name = "TAHITI";
300 rlc_chip_name = "TAHITI";
301 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
302 me_req_size = SI_PM4_UCODE_SIZE * 4;
303 ce_req_size = SI_CE_UCODE_SIZE * 4;
304 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
305 mc_req_size = SI_MC_UCODE_SIZE * 4;
308 chip_name = "PITCAIRN";
309 rlc_chip_name = "PITCAIRN";
310 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
311 me_req_size = SI_PM4_UCODE_SIZE * 4;
312 ce_req_size = SI_CE_UCODE_SIZE * 4;
313 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
314 mc_req_size = SI_MC_UCODE_SIZE * 4;
318 rlc_chip_name = "VERDE";
319 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
320 me_req_size = SI_PM4_UCODE_SIZE * 4;
321 ce_req_size = SI_CE_UCODE_SIZE * 4;
322 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
323 mc_req_size = SI_MC_UCODE_SIZE * 4;
328 DRM_INFO("Loading %s Microcode\n", chip_name);
330 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
331 err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
334 if (rdev->pfp_fw->size != pfp_req_size) {
336 "si_cp: Bogus length %zu in firmware \"%s\"\n",
337 rdev->pfp_fw->size, fw_name);
342 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
343 err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
346 if (rdev->me_fw->size != me_req_size) {
348 "si_cp: Bogus length %zu in firmware \"%s\"\n",
349 rdev->me_fw->size, fw_name);
353 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
354 err = request_firmware(&rdev->ce_fw, fw_name, &pdev->dev);
357 if (rdev->ce_fw->size != ce_req_size) {
359 "si_cp: Bogus length %zu in firmware \"%s\"\n",
360 rdev->ce_fw->size, fw_name);
364 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
365 err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
368 if (rdev->rlc_fw->size != rlc_req_size) {
370 "si_rlc: Bogus length %zu in firmware \"%s\"\n",
371 rdev->rlc_fw->size, fw_name);
375 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
376 err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev);
379 if (rdev->mc_fw->size != mc_req_size) {
381 "si_mc: Bogus length %zu in firmware \"%s\"\n",
382 rdev->mc_fw->size, fw_name);
387 platform_device_unregister(pdev);
392 "si_cp: Failed to load firmware \"%s\"\n",
394 release_firmware(rdev->pfp_fw);
396 release_firmware(rdev->me_fw);
398 release_firmware(rdev->ce_fw);
400 release_firmware(rdev->rlc_fw);
402 release_firmware(rdev->mc_fw);
408 /* watermark setup */
409 static u32 dce6_line_buffer_adjust(struct radeon_device *rdev,
410 struct radeon_crtc *radeon_crtc,
411 struct drm_display_mode *mode,
412 struct drm_display_mode *other_mode)
417 * There are 3 line buffers, each one shared by 2 display controllers.
418 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
419 * the display controllers. The paritioning is done via one of four
420 * preset allocations specified in bits 21:20:
422 * 2 - whole lb, other crtc must be disabled
424 /* this can get tricky if we have two large displays on a paired group
425 * of crtcs. Ideally for multiple large displays we'd assign them to
426 * non-linked crtcs for maximum line buffer allocation.
428 if (radeon_crtc->base.enabled && mode) {
436 WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset,
437 DC_LB_MEMORY_CONFIG(tmp));
439 if (radeon_crtc->base.enabled && mode) {
449 /* controller not enabled, so no lb used */
453 static u32 si_get_number_of_dram_channels(struct radeon_device *rdev)
455 u32 tmp = RREG32(MC_SHARED_CHMAP);
457 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
480 struct dce6_wm_params {
481 u32 dram_channels; /* number of dram channels */
482 u32 yclk; /* bandwidth per dram data pin in kHz */
483 u32 sclk; /* engine clock in kHz */
484 u32 disp_clk; /* display clock in kHz */
485 u32 src_width; /* viewport width */
486 u32 active_time; /* active display time in ns */
487 u32 blank_time; /* blank time in ns */
488 bool interlaced; /* mode is interlaced */
489 fixed20_12 vsc; /* vertical scale ratio */
490 u32 num_heads; /* number of active crtcs */
491 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
492 u32 lb_size; /* line buffer allocated to pipe */
493 u32 vtaps; /* vertical scaler taps */
496 static u32 dce6_dram_bandwidth(struct dce6_wm_params *wm)
498 /* Calculate raw DRAM Bandwidth */
499 fixed20_12 dram_efficiency; /* 0.7 */
500 fixed20_12 yclk, dram_channels, bandwidth;
503 a.full = dfixed_const(1000);
504 yclk.full = dfixed_const(wm->yclk);
505 yclk.full = dfixed_div(yclk, a);
506 dram_channels.full = dfixed_const(wm->dram_channels * 4);
507 a.full = dfixed_const(10);
508 dram_efficiency.full = dfixed_const(7);
509 dram_efficiency.full = dfixed_div(dram_efficiency, a);
510 bandwidth.full = dfixed_mul(dram_channels, yclk);
511 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
513 return dfixed_trunc(bandwidth);
516 static u32 dce6_dram_bandwidth_for_display(struct dce6_wm_params *wm)
518 /* Calculate DRAM Bandwidth and the part allocated to display. */
519 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
520 fixed20_12 yclk, dram_channels, bandwidth;
523 a.full = dfixed_const(1000);
524 yclk.full = dfixed_const(wm->yclk);
525 yclk.full = dfixed_div(yclk, a);
526 dram_channels.full = dfixed_const(wm->dram_channels * 4);
527 a.full = dfixed_const(10);
528 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
529 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
530 bandwidth.full = dfixed_mul(dram_channels, yclk);
531 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
533 return dfixed_trunc(bandwidth);
536 static u32 dce6_data_return_bandwidth(struct dce6_wm_params *wm)
538 /* Calculate the display Data return Bandwidth */
539 fixed20_12 return_efficiency; /* 0.8 */
540 fixed20_12 sclk, bandwidth;
543 a.full = dfixed_const(1000);
544 sclk.full = dfixed_const(wm->sclk);
545 sclk.full = dfixed_div(sclk, a);
546 a.full = dfixed_const(10);
547 return_efficiency.full = dfixed_const(8);
548 return_efficiency.full = dfixed_div(return_efficiency, a);
549 a.full = dfixed_const(32);
550 bandwidth.full = dfixed_mul(a, sclk);
551 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
553 return dfixed_trunc(bandwidth);
556 static u32 dce6_get_dmif_bytes_per_request(struct dce6_wm_params *wm)
561 static u32 dce6_dmif_request_bandwidth(struct dce6_wm_params *wm)
563 /* Calculate the DMIF Request Bandwidth */
564 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
565 fixed20_12 disp_clk, sclk, bandwidth;
566 fixed20_12 a, b1, b2;
569 a.full = dfixed_const(1000);
570 disp_clk.full = dfixed_const(wm->disp_clk);
571 disp_clk.full = dfixed_div(disp_clk, a);
572 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm) / 2);
573 b1.full = dfixed_mul(a, disp_clk);
575 a.full = dfixed_const(1000);
576 sclk.full = dfixed_const(wm->sclk);
577 sclk.full = dfixed_div(sclk, a);
578 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm));
579 b2.full = dfixed_mul(a, sclk);
581 a.full = dfixed_const(10);
582 disp_clk_request_efficiency.full = dfixed_const(8);
583 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
585 min_bandwidth = min(dfixed_trunc(b1), dfixed_trunc(b2));
587 a.full = dfixed_const(min_bandwidth);
588 bandwidth.full = dfixed_mul(a, disp_clk_request_efficiency);
590 return dfixed_trunc(bandwidth);
593 static u32 dce6_available_bandwidth(struct dce6_wm_params *wm)
595 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
596 u32 dram_bandwidth = dce6_dram_bandwidth(wm);
597 u32 data_return_bandwidth = dce6_data_return_bandwidth(wm);
598 u32 dmif_req_bandwidth = dce6_dmif_request_bandwidth(wm);
600 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
603 static u32 dce6_average_bandwidth(struct dce6_wm_params *wm)
605 /* Calculate the display mode Average Bandwidth
606 * DisplayMode should contain the source and destination dimensions,
610 fixed20_12 line_time;
611 fixed20_12 src_width;
612 fixed20_12 bandwidth;
615 a.full = dfixed_const(1000);
616 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
617 line_time.full = dfixed_div(line_time, a);
618 bpp.full = dfixed_const(wm->bytes_per_pixel);
619 src_width.full = dfixed_const(wm->src_width);
620 bandwidth.full = dfixed_mul(src_width, bpp);
621 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
622 bandwidth.full = dfixed_div(bandwidth, line_time);
624 return dfixed_trunc(bandwidth);
627 static u32 dce6_latency_watermark(struct dce6_wm_params *wm)
629 /* First calcualte the latency in ns */
630 u32 mc_latency = 2000; /* 2000 ns. */
631 u32 available_bandwidth = dce6_available_bandwidth(wm);
632 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
633 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
634 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
635 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
636 (wm->num_heads * cursor_line_pair_return_time);
637 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
638 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
639 u32 tmp, dmif_size = 12288;
642 if (wm->num_heads == 0)
645 a.full = dfixed_const(2);
646 b.full = dfixed_const(1);
647 if ((wm->vsc.full > a.full) ||
648 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
650 ((wm->vsc.full >= a.full) && wm->interlaced))
651 max_src_lines_per_dst_line = 4;
653 max_src_lines_per_dst_line = 2;
655 a.full = dfixed_const(available_bandwidth);
656 b.full = dfixed_const(wm->num_heads);
657 a.full = dfixed_div(a, b);
659 b.full = dfixed_const(mc_latency + 512);
660 c.full = dfixed_const(wm->disp_clk);
661 b.full = dfixed_div(b, c);
663 c.full = dfixed_const(dmif_size);
664 b.full = dfixed_div(c, b);
666 tmp = min(dfixed_trunc(a), dfixed_trunc(b));
668 b.full = dfixed_const(1000);
669 c.full = dfixed_const(wm->disp_clk);
670 b.full = dfixed_div(c, b);
671 c.full = dfixed_const(wm->bytes_per_pixel);
672 b.full = dfixed_mul(b, c);
674 lb_fill_bw = min(tmp, dfixed_trunc(b));
676 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
677 b.full = dfixed_const(1000);
678 c.full = dfixed_const(lb_fill_bw);
679 b.full = dfixed_div(c, b);
680 a.full = dfixed_div(a, b);
681 line_fill_time = dfixed_trunc(a);
683 if (line_fill_time < wm->active_time)
686 return latency + (line_fill_time - wm->active_time);
690 static bool dce6_average_bandwidth_vs_dram_bandwidth_for_display(struct dce6_wm_params *wm)
692 if (dce6_average_bandwidth(wm) <=
693 (dce6_dram_bandwidth_for_display(wm) / wm->num_heads))
699 static bool dce6_average_bandwidth_vs_available_bandwidth(struct dce6_wm_params *wm)
701 if (dce6_average_bandwidth(wm) <=
702 (dce6_available_bandwidth(wm) / wm->num_heads))
708 static bool dce6_check_latency_hiding(struct dce6_wm_params *wm)
710 u32 lb_partitions = wm->lb_size / wm->src_width;
711 u32 line_time = wm->active_time + wm->blank_time;
712 u32 latency_tolerant_lines;
716 a.full = dfixed_const(1);
717 if (wm->vsc.full > a.full)
718 latency_tolerant_lines = 1;
720 if (lb_partitions <= (wm->vtaps + 1))
721 latency_tolerant_lines = 1;
723 latency_tolerant_lines = 2;
726 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
728 if (dce6_latency_watermark(wm) <= latency_hiding)
734 static void dce6_program_watermarks(struct radeon_device *rdev,
735 struct radeon_crtc *radeon_crtc,
736 u32 lb_size, u32 num_heads)
738 struct drm_display_mode *mode = &radeon_crtc->base.mode;
739 struct dce6_wm_params wm;
742 u32 latency_watermark_a = 0, latency_watermark_b = 0;
743 u32 priority_a_mark = 0, priority_b_mark = 0;
744 u32 priority_a_cnt = PRIORITY_OFF;
745 u32 priority_b_cnt = PRIORITY_OFF;
746 u32 tmp, arb_control3;
749 if (radeon_crtc->base.enabled && num_heads && mode) {
750 pixel_period = 1000000 / (u32)mode->clock;
751 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
755 wm.yclk = rdev->pm.current_mclk * 10;
756 wm.sclk = rdev->pm.current_sclk * 10;
757 wm.disp_clk = mode->clock;
758 wm.src_width = mode->crtc_hdisplay;
759 wm.active_time = mode->crtc_hdisplay * pixel_period;
760 wm.blank_time = line_time - wm.active_time;
761 wm.interlaced = false;
762 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
763 wm.interlaced = true;
764 wm.vsc = radeon_crtc->vsc;
766 if (radeon_crtc->rmx_type != RMX_OFF)
768 wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
769 wm.lb_size = lb_size;
770 if (rdev->family == CHIP_ARUBA)
771 wm.dram_channels = evergreen_get_number_of_dram_channels(rdev);
773 wm.dram_channels = si_get_number_of_dram_channels(rdev);
774 wm.num_heads = num_heads;
776 /* set for high clocks */
777 latency_watermark_a = min(dce6_latency_watermark(&wm), (u32)65535);
778 /* set for low clocks */
779 /* wm.yclk = low clk; wm.sclk = low clk */
780 latency_watermark_b = min(dce6_latency_watermark(&wm), (u32)65535);
782 /* possibly force display priority to high */
783 /* should really do this at mode validation time... */
784 if (!dce6_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
785 !dce6_average_bandwidth_vs_available_bandwidth(&wm) ||
786 !dce6_check_latency_hiding(&wm) ||
787 (rdev->disp_priority == 2)) {
788 DRM_DEBUG_KMS("force priority to high\n");
789 priority_a_cnt |= PRIORITY_ALWAYS_ON;
790 priority_b_cnt |= PRIORITY_ALWAYS_ON;
793 a.full = dfixed_const(1000);
794 b.full = dfixed_const(mode->clock);
795 b.full = dfixed_div(b, a);
796 c.full = dfixed_const(latency_watermark_a);
797 c.full = dfixed_mul(c, b);
798 c.full = dfixed_mul(c, radeon_crtc->hsc);
799 c.full = dfixed_div(c, a);
800 a.full = dfixed_const(16);
801 c.full = dfixed_div(c, a);
802 priority_a_mark = dfixed_trunc(c);
803 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
805 a.full = dfixed_const(1000);
806 b.full = dfixed_const(mode->clock);
807 b.full = dfixed_div(b, a);
808 c.full = dfixed_const(latency_watermark_b);
809 c.full = dfixed_mul(c, b);
810 c.full = dfixed_mul(c, radeon_crtc->hsc);
811 c.full = dfixed_div(c, a);
812 a.full = dfixed_const(16);
813 c.full = dfixed_div(c, a);
814 priority_b_mark = dfixed_trunc(c);
815 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
819 arb_control3 = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
821 tmp &= ~LATENCY_WATERMARK_MASK(3);
822 tmp |= LATENCY_WATERMARK_MASK(1);
823 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
824 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
825 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
826 LATENCY_HIGH_WATERMARK(line_time)));
828 tmp = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
829 tmp &= ~LATENCY_WATERMARK_MASK(3);
830 tmp |= LATENCY_WATERMARK_MASK(2);
831 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
832 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
833 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
834 LATENCY_HIGH_WATERMARK(line_time)));
835 /* restore original selection */
836 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, arb_control3);
838 /* write the priority marks */
839 WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
840 WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
844 void dce6_bandwidth_update(struct radeon_device *rdev)
846 struct drm_display_mode *mode0 = NULL;
847 struct drm_display_mode *mode1 = NULL;
848 u32 num_heads = 0, lb_size;
851 radeon_update_display_priority(rdev);
853 for (i = 0; i < rdev->num_crtc; i++) {
854 if (rdev->mode_info.crtcs[i]->base.enabled)
857 for (i = 0; i < rdev->num_crtc; i += 2) {
858 mode0 = &rdev->mode_info.crtcs[i]->base.mode;
859 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
860 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
861 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
862 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
863 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
870 static void si_tiling_mode_table_init(struct radeon_device *rdev)
872 const u32 num_tile_mode_states = 32;
873 u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
875 switch (rdev->config.si.mem_row_size_in_kb) {
877 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
881 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
884 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
888 if ((rdev->family == CHIP_TAHITI) ||
889 (rdev->family == CHIP_PITCAIRN)) {
890 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
891 switch (reg_offset) {
892 case 0: /* non-AA compressed depth or any compressed stencil */
893 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
894 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
895 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
896 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
897 NUM_BANKS(ADDR_SURF_16_BANK) |
898 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
899 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
900 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
902 case 1: /* 2xAA/4xAA compressed depth only */
903 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
904 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
905 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
906 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
907 NUM_BANKS(ADDR_SURF_16_BANK) |
908 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
909 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
910 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
912 case 2: /* 8xAA compressed depth only */
913 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
914 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
915 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
916 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
917 NUM_BANKS(ADDR_SURF_16_BANK) |
918 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
919 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
920 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
922 case 3: /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
923 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
924 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
925 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
926 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
927 NUM_BANKS(ADDR_SURF_16_BANK) |
928 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
929 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
930 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
932 case 4: /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
933 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
934 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
935 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
936 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
937 NUM_BANKS(ADDR_SURF_16_BANK) |
938 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
939 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
940 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
942 case 5: /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
943 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
944 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
945 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
946 TILE_SPLIT(split_equal_to_row_size) |
947 NUM_BANKS(ADDR_SURF_16_BANK) |
948 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
949 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
950 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
952 case 6: /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
953 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
954 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
955 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
956 TILE_SPLIT(split_equal_to_row_size) |
957 NUM_BANKS(ADDR_SURF_16_BANK) |
958 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
959 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
960 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
962 case 7: /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
963 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
964 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
965 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
966 TILE_SPLIT(split_equal_to_row_size) |
967 NUM_BANKS(ADDR_SURF_16_BANK) |
968 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
969 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
970 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
972 case 8: /* 1D and 1D Array Surfaces */
973 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
974 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
975 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
976 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
977 NUM_BANKS(ADDR_SURF_16_BANK) |
978 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
979 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
980 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
982 case 9: /* Displayable maps. */
983 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
984 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
985 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
986 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
987 NUM_BANKS(ADDR_SURF_16_BANK) |
988 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
989 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
990 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
992 case 10: /* Display 8bpp. */
993 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
994 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
995 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
996 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
997 NUM_BANKS(ADDR_SURF_16_BANK) |
998 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
999 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1000 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1002 case 11: /* Display 16bpp. */
1003 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1004 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1005 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1006 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1007 NUM_BANKS(ADDR_SURF_16_BANK) |
1008 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1009 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1010 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1012 case 12: /* Display 32bpp. */
1013 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1014 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1015 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1016 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1017 NUM_BANKS(ADDR_SURF_16_BANK) |
1018 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1019 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1020 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1022 case 13: /* Thin. */
1023 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1024 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1025 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1026 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1027 NUM_BANKS(ADDR_SURF_16_BANK) |
1028 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1029 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1030 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1032 case 14: /* Thin 8 bpp. */
1033 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1034 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1035 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1036 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1037 NUM_BANKS(ADDR_SURF_16_BANK) |
1038 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1039 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1040 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1042 case 15: /* Thin 16 bpp. */
1043 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1044 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1045 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1046 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1047 NUM_BANKS(ADDR_SURF_16_BANK) |
1048 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1049 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1050 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1052 case 16: /* Thin 32 bpp. */
1053 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1054 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1055 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1056 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1057 NUM_BANKS(ADDR_SURF_16_BANK) |
1058 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1059 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1060 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1062 case 17: /* Thin 64 bpp. */
1063 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1064 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1065 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1066 TILE_SPLIT(split_equal_to_row_size) |
1067 NUM_BANKS(ADDR_SURF_16_BANK) |
1068 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1069 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1070 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1072 case 21: /* 8 bpp PRT. */
1073 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1074 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1075 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1076 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1077 NUM_BANKS(ADDR_SURF_16_BANK) |
1078 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1079 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1080 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1082 case 22: /* 16 bpp PRT */
1083 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1084 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1085 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1086 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1087 NUM_BANKS(ADDR_SURF_16_BANK) |
1088 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1089 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1090 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1092 case 23: /* 32 bpp PRT */
1093 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1094 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1095 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1096 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1097 NUM_BANKS(ADDR_SURF_16_BANK) |
1098 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1099 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1100 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1102 case 24: /* 64 bpp PRT */
1103 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1104 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1105 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1106 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1107 NUM_BANKS(ADDR_SURF_16_BANK) |
1108 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1109 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1110 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1112 case 25: /* 128 bpp PRT */
1113 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1114 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1115 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1116 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1117 NUM_BANKS(ADDR_SURF_8_BANK) |
1118 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1119 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1120 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1126 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1128 } else if (rdev->family == CHIP_VERDE) {
1129 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1130 switch (reg_offset) {
1131 case 0: /* non-AA compressed depth or any compressed stencil */
1132 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1133 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1134 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1135 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1136 NUM_BANKS(ADDR_SURF_16_BANK) |
1137 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1138 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1139 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1141 case 1: /* 2xAA/4xAA compressed depth only */
1142 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1143 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1144 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1145 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1146 NUM_BANKS(ADDR_SURF_16_BANK) |
1147 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1148 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1149 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1151 case 2: /* 8xAA compressed depth only */
1152 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1153 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1154 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1155 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1156 NUM_BANKS(ADDR_SURF_16_BANK) |
1157 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1158 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1159 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1161 case 3: /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1162 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1163 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1164 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1165 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1166 NUM_BANKS(ADDR_SURF_16_BANK) |
1167 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1168 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1169 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1171 case 4: /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1172 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1173 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1174 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1175 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1176 NUM_BANKS(ADDR_SURF_16_BANK) |
1177 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1178 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1179 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1181 case 5: /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1182 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1183 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1184 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1185 TILE_SPLIT(split_equal_to_row_size) |
1186 NUM_BANKS(ADDR_SURF_16_BANK) |
1187 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1188 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1189 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1191 case 6: /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1192 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1193 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1194 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1195 TILE_SPLIT(split_equal_to_row_size) |
1196 NUM_BANKS(ADDR_SURF_16_BANK) |
1197 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1198 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1199 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1201 case 7: /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1202 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1203 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1204 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1205 TILE_SPLIT(split_equal_to_row_size) |
1206 NUM_BANKS(ADDR_SURF_16_BANK) |
1207 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1208 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1209 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1211 case 8: /* 1D and 1D Array Surfaces */
1212 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1213 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1214 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1215 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1216 NUM_BANKS(ADDR_SURF_16_BANK) |
1217 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1218 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1219 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1221 case 9: /* Displayable maps. */
1222 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1223 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1224 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1225 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1226 NUM_BANKS(ADDR_SURF_16_BANK) |
1227 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1228 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1229 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1231 case 10: /* Display 8bpp. */
1232 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1233 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1234 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1235 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1236 NUM_BANKS(ADDR_SURF_16_BANK) |
1237 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1238 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1239 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1241 case 11: /* Display 16bpp. */
1242 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1243 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1244 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1245 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1246 NUM_BANKS(ADDR_SURF_16_BANK) |
1247 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1248 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1249 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1251 case 12: /* Display 32bpp. */
1252 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1253 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1254 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1255 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1256 NUM_BANKS(ADDR_SURF_16_BANK) |
1257 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1258 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1259 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1261 case 13: /* Thin. */
1262 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1263 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1264 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1265 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1266 NUM_BANKS(ADDR_SURF_16_BANK) |
1267 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1268 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1269 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1271 case 14: /* Thin 8 bpp. */
1272 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1273 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1274 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1275 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1276 NUM_BANKS(ADDR_SURF_16_BANK) |
1277 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1278 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1279 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1281 case 15: /* Thin 16 bpp. */
1282 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1283 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1284 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1285 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1286 NUM_BANKS(ADDR_SURF_16_BANK) |
1287 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1288 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1289 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1291 case 16: /* Thin 32 bpp. */
1292 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1293 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1294 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1295 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1296 NUM_BANKS(ADDR_SURF_16_BANK) |
1297 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1298 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1299 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1301 case 17: /* Thin 64 bpp. */
1302 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1303 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1304 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1305 TILE_SPLIT(split_equal_to_row_size) |
1306 NUM_BANKS(ADDR_SURF_16_BANK) |
1307 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1308 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1309 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1311 case 21: /* 8 bpp PRT. */
1312 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1313 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1314 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1315 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1316 NUM_BANKS(ADDR_SURF_16_BANK) |
1317 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1318 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1319 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1321 case 22: /* 16 bpp PRT */
1322 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1323 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1324 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1325 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1326 NUM_BANKS(ADDR_SURF_16_BANK) |
1327 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1328 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1329 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1331 case 23: /* 32 bpp PRT */
1332 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1333 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1334 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1335 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1336 NUM_BANKS(ADDR_SURF_16_BANK) |
1337 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1338 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1339 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1341 case 24: /* 64 bpp PRT */
1342 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1343 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1344 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1345 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1346 NUM_BANKS(ADDR_SURF_16_BANK) |
1347 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1348 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1349 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1351 case 25: /* 128 bpp PRT */
1352 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1353 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1354 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1355 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1356 NUM_BANKS(ADDR_SURF_8_BANK) |
1357 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1358 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1359 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1365 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1368 DRM_ERROR("unknown asic: 0x%x\n", rdev->family);
1371 static void si_select_se_sh(struct radeon_device *rdev,
1372 u32 se_num, u32 sh_num)
1374 u32 data = INSTANCE_BROADCAST_WRITES;
1376 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1377 data = SH_BROADCAST_WRITES | SE_BROADCAST_WRITES;
1378 else if (se_num == 0xffffffff)
1379 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num);
1380 else if (sh_num == 0xffffffff)
1381 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num);
1383 data |= SH_INDEX(sh_num) | SE_INDEX(se_num);
1384 WREG32(GRBM_GFX_INDEX, data);
1387 static u32 si_create_bitmask(u32 bit_width)
1391 for (i = 0; i < bit_width; i++) {
1398 static u32 si_get_cu_enabled(struct radeon_device *rdev, u32 cu_per_sh)
1402 data = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
1404 data &= INACTIVE_CUS_MASK;
1407 data |= RREG32(GC_USER_SHADER_ARRAY_CONFIG);
1409 data >>= INACTIVE_CUS_SHIFT;
1411 mask = si_create_bitmask(cu_per_sh);
1413 return ~data & mask;
1416 static void si_setup_spi(struct radeon_device *rdev,
1417 u32 se_num, u32 sh_per_se,
1421 u32 data, mask, active_cu;
1423 for (i = 0; i < se_num; i++) {
1424 for (j = 0; j < sh_per_se; j++) {
1425 si_select_se_sh(rdev, i, j);
1426 data = RREG32(SPI_STATIC_THREAD_MGMT_3);
1427 active_cu = si_get_cu_enabled(rdev, cu_per_sh);
1430 for (k = 0; k < 16; k++) {
1432 if (active_cu & mask) {
1434 WREG32(SPI_STATIC_THREAD_MGMT_3, data);
1440 si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1443 static u32 si_get_rb_disabled(struct radeon_device *rdev,
1444 u32 max_rb_num, u32 se_num,
1449 data = RREG32(CC_RB_BACKEND_DISABLE);
1451 data &= BACKEND_DISABLE_MASK;
1454 data |= RREG32(GC_USER_RB_BACKEND_DISABLE);
1456 data >>= BACKEND_DISABLE_SHIFT;
1458 mask = si_create_bitmask(max_rb_num / se_num / sh_per_se);
1463 static void si_setup_rb(struct radeon_device *rdev,
1464 u32 se_num, u32 sh_per_se,
1469 u32 disabled_rbs = 0;
1470 u32 enabled_rbs = 0;
1472 for (i = 0; i < se_num; i++) {
1473 for (j = 0; j < sh_per_se; j++) {
1474 si_select_se_sh(rdev, i, j);
1475 data = si_get_rb_disabled(rdev, max_rb_num, se_num, sh_per_se);
1476 disabled_rbs |= data << ((i * sh_per_se + j) * TAHITI_RB_BITMAP_WIDTH_PER_SH);
1479 si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1482 for (i = 0; i < max_rb_num; i++) {
1483 if (!(disabled_rbs & mask))
1484 enabled_rbs |= mask;
1488 for (i = 0; i < se_num; i++) {
1489 si_select_se_sh(rdev, i, 0xffffffff);
1491 for (j = 0; j < sh_per_se; j++) {
1492 switch (enabled_rbs & 3) {
1494 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2);
1497 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2);
1501 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2);
1506 WREG32(PA_SC_RASTER_CONFIG, data);
1508 si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1511 static void si_gpu_init(struct radeon_device *rdev)
1513 u32 gb_addr_config = 0;
1514 u32 mc_shared_chmap, mc_arb_ramcfg;
1516 u32 hdp_host_path_cntl;
1520 switch (rdev->family) {
1522 rdev->config.si.max_shader_engines = 2;
1523 rdev->config.si.max_tile_pipes = 12;
1524 rdev->config.si.max_cu_per_sh = 8;
1525 rdev->config.si.max_sh_per_se = 2;
1526 rdev->config.si.max_backends_per_se = 4;
1527 rdev->config.si.max_texture_channel_caches = 12;
1528 rdev->config.si.max_gprs = 256;
1529 rdev->config.si.max_gs_threads = 32;
1530 rdev->config.si.max_hw_contexts = 8;
1532 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1533 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1534 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1535 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1536 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN;
1539 rdev->config.si.max_shader_engines = 2;
1540 rdev->config.si.max_tile_pipes = 8;
1541 rdev->config.si.max_cu_per_sh = 5;
1542 rdev->config.si.max_sh_per_se = 2;
1543 rdev->config.si.max_backends_per_se = 4;
1544 rdev->config.si.max_texture_channel_caches = 8;
1545 rdev->config.si.max_gprs = 256;
1546 rdev->config.si.max_gs_threads = 32;
1547 rdev->config.si.max_hw_contexts = 8;
1549 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1550 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1551 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1552 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1553 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN;
1557 rdev->config.si.max_shader_engines = 1;
1558 rdev->config.si.max_tile_pipes = 4;
1559 rdev->config.si.max_cu_per_sh = 2;
1560 rdev->config.si.max_sh_per_se = 2;
1561 rdev->config.si.max_backends_per_se = 4;
1562 rdev->config.si.max_texture_channel_caches = 4;
1563 rdev->config.si.max_gprs = 256;
1564 rdev->config.si.max_gs_threads = 32;
1565 rdev->config.si.max_hw_contexts = 8;
1567 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1568 rdev->config.si.sc_prim_fifo_size_backend = 0x40;
1569 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1570 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1571 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN;
1575 /* Initialize HDP */
1576 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1577 WREG32((0x2c14 + j), 0x00000000);
1578 WREG32((0x2c18 + j), 0x00000000);
1579 WREG32((0x2c1c + j), 0x00000000);
1580 WREG32((0x2c20 + j), 0x00000000);
1581 WREG32((0x2c24 + j), 0x00000000);
1584 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1586 evergreen_fix_pci_max_read_req_size(rdev);
1588 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
1590 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1591 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1593 rdev->config.si.num_tile_pipes = rdev->config.si.max_tile_pipes;
1594 rdev->config.si.mem_max_burst_length_bytes = 256;
1595 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
1596 rdev->config.si.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1597 if (rdev->config.si.mem_row_size_in_kb > 4)
1598 rdev->config.si.mem_row_size_in_kb = 4;
1599 /* XXX use MC settings? */
1600 rdev->config.si.shader_engine_tile_size = 32;
1601 rdev->config.si.num_gpus = 1;
1602 rdev->config.si.multi_gpu_tile_size = 64;
1604 /* fix up row size */
1605 gb_addr_config &= ~ROW_SIZE_MASK;
1606 switch (rdev->config.si.mem_row_size_in_kb) {
1609 gb_addr_config |= ROW_SIZE(0);
1612 gb_addr_config |= ROW_SIZE(1);
1615 gb_addr_config |= ROW_SIZE(2);
1619 /* setup tiling info dword. gb_addr_config is not adequate since it does
1620 * not have bank info, so create a custom tiling dword.
1621 * bits 3:0 num_pipes
1622 * bits 7:4 num_banks
1623 * bits 11:8 group_size
1624 * bits 15:12 row_size
1626 rdev->config.si.tile_config = 0;
1627 switch (rdev->config.si.num_tile_pipes) {
1629 rdev->config.si.tile_config |= (0 << 0);
1632 rdev->config.si.tile_config |= (1 << 0);
1635 rdev->config.si.tile_config |= (2 << 0);
1639 /* XXX what about 12? */
1640 rdev->config.si.tile_config |= (3 << 0);
1643 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
1644 case 0: /* four banks */
1645 rdev->config.si.tile_config |= 0 << 4;
1647 case 1: /* eight banks */
1648 rdev->config.si.tile_config |= 1 << 4;
1650 case 2: /* sixteen banks */
1652 rdev->config.si.tile_config |= 2 << 4;
1655 rdev->config.si.tile_config |=
1656 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
1657 rdev->config.si.tile_config |=
1658 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
1660 WREG32(GB_ADDR_CONFIG, gb_addr_config);
1661 WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
1662 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
1663 WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config);
1664 WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config);
1666 si_tiling_mode_table_init(rdev);
1668 si_setup_rb(rdev, rdev->config.si.max_shader_engines,
1669 rdev->config.si.max_sh_per_se,
1670 rdev->config.si.max_backends_per_se);
1672 si_setup_spi(rdev, rdev->config.si.max_shader_engines,
1673 rdev->config.si.max_sh_per_se,
1674 rdev->config.si.max_cu_per_sh);
1677 /* set HW defaults for 3D engine */
1678 WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
1679 ROQ_IB2_START(0x2b)));
1680 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
1682 sx_debug_1 = RREG32(SX_DEBUG_1);
1683 WREG32(SX_DEBUG_1, sx_debug_1);
1685 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
1687 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_frontend) |
1688 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_backend) |
1689 SC_HIZ_TILE_FIFO_SIZE(rdev->config.si.sc_hiz_tile_fifo_size) |
1690 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.si.sc_earlyz_tile_fifo_size)));
1692 WREG32(VGT_NUM_INSTANCES, 1);
1694 WREG32(CP_PERFMON_CNTL, 0);
1696 WREG32(SQ_CONFIG, 0);
1698 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
1699 FORCE_EOV_MAX_REZ_CNT(255)));
1701 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
1702 AUTO_INVLD_EN(ES_AND_GS_AUTO));
1704 WREG32(VGT_GS_VERTEX_REUSE, 16);
1705 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1707 WREG32(CB_PERFCOUNTER0_SELECT0, 0);
1708 WREG32(CB_PERFCOUNTER0_SELECT1, 0);
1709 WREG32(CB_PERFCOUNTER1_SELECT0, 0);
1710 WREG32(CB_PERFCOUNTER1_SELECT1, 0);
1711 WREG32(CB_PERFCOUNTER2_SELECT0, 0);
1712 WREG32(CB_PERFCOUNTER2_SELECT1, 0);
1713 WREG32(CB_PERFCOUNTER3_SELECT0, 0);
1714 WREG32(CB_PERFCOUNTER3_SELECT1, 0);
1716 tmp = RREG32(HDP_MISC_CNTL);
1717 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1718 WREG32(HDP_MISC_CNTL, tmp);
1720 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1721 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1723 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1729 * GPU scratch registers helpers function.
1731 static void si_scratch_init(struct radeon_device *rdev)
1735 rdev->scratch.num_reg = 7;
1736 rdev->scratch.reg_base = SCRATCH_REG0;
1737 for (i = 0; i < rdev->scratch.num_reg; i++) {
1738 rdev->scratch.free[i] = true;
1739 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
1743 void si_fence_ring_emit(struct radeon_device *rdev,
1744 struct radeon_fence *fence)
1746 struct radeon_ring *ring = &rdev->ring[fence->ring];
1747 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1749 /* flush read cache over gart */
1750 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1751 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1752 radeon_ring_write(ring, 0);
1753 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1754 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1755 PACKET3_TC_ACTION_ENA |
1756 PACKET3_SH_KCACHE_ACTION_ENA |
1757 PACKET3_SH_ICACHE_ACTION_ENA);
1758 radeon_ring_write(ring, 0xFFFFFFFF);
1759 radeon_ring_write(ring, 0);
1760 radeon_ring_write(ring, 10); /* poll interval */
1761 /* EVENT_WRITE_EOP - flush caches, send int */
1762 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1763 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5));
1764 radeon_ring_write(ring, addr & 0xffffffff);
1765 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1766 radeon_ring_write(ring, fence->seq);
1767 radeon_ring_write(ring, 0);
1773 void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1775 struct radeon_ring *ring = &rdev->ring[ib->ring];
1778 if (ib->is_const_ib) {
1779 /* set switch buffer packet before const IB */
1780 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
1781 radeon_ring_write(ring, 0);
1783 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
1786 if (ring->rptr_save_reg) {
1787 next_rptr = ring->wptr + 3 + 4 + 8;
1788 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1789 radeon_ring_write(ring, ((ring->rptr_save_reg -
1790 PACKET3_SET_CONFIG_REG_START) >> 2));
1791 radeon_ring_write(ring, next_rptr);
1792 } else if (rdev->wb.enabled) {
1793 next_rptr = ring->wptr + 5 + 4 + 8;
1794 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
1795 radeon_ring_write(ring, (1 << 8));
1796 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
1797 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
1798 radeon_ring_write(ring, next_rptr);
1801 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
1804 radeon_ring_write(ring, header);
1805 radeon_ring_write(ring,
1809 (ib->gpu_addr & 0xFFFFFFFC));
1810 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
1811 radeon_ring_write(ring, ib->length_dw |
1812 (ib->vm ? (ib->vm->id << 24) : 0));
1814 if (!ib->is_const_ib) {
1815 /* flush read cache over gart for this vmid */
1816 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1817 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1818 radeon_ring_write(ring, ib->vm ? ib->vm->id : 0);
1819 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1820 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1821 PACKET3_TC_ACTION_ENA |
1822 PACKET3_SH_KCACHE_ACTION_ENA |
1823 PACKET3_SH_ICACHE_ACTION_ENA);
1824 radeon_ring_write(ring, 0xFFFFFFFF);
1825 radeon_ring_write(ring, 0);
1826 radeon_ring_write(ring, 10); /* poll interval */
1833 static void si_cp_enable(struct radeon_device *rdev, bool enable)
1836 WREG32(CP_ME_CNTL, 0);
1838 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1839 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
1840 WREG32(SCRATCH_UMSK, 0);
1841 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1842 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1843 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1848 static int si_cp_load_microcode(struct radeon_device *rdev)
1850 const __be32 *fw_data;
1853 if (!rdev->me_fw || !rdev->pfp_fw)
1856 si_cp_enable(rdev, false);
1859 fw_data = (const __be32 *)rdev->pfp_fw->data;
1860 WREG32(CP_PFP_UCODE_ADDR, 0);
1861 for (i = 0; i < SI_PFP_UCODE_SIZE; i++)
1862 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1863 WREG32(CP_PFP_UCODE_ADDR, 0);
1866 fw_data = (const __be32 *)rdev->ce_fw->data;
1867 WREG32(CP_CE_UCODE_ADDR, 0);
1868 for (i = 0; i < SI_CE_UCODE_SIZE; i++)
1869 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
1870 WREG32(CP_CE_UCODE_ADDR, 0);
1873 fw_data = (const __be32 *)rdev->me_fw->data;
1874 WREG32(CP_ME_RAM_WADDR, 0);
1875 for (i = 0; i < SI_PM4_UCODE_SIZE; i++)
1876 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1877 WREG32(CP_ME_RAM_WADDR, 0);
1879 WREG32(CP_PFP_UCODE_ADDR, 0);
1880 WREG32(CP_CE_UCODE_ADDR, 0);
1881 WREG32(CP_ME_RAM_WADDR, 0);
1882 WREG32(CP_ME_RAM_RADDR, 0);
1886 static int si_cp_start(struct radeon_device *rdev)
1888 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1891 r = radeon_ring_lock(rdev, ring, 7 + 4);
1893 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1897 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
1898 radeon_ring_write(ring, 0x1);
1899 radeon_ring_write(ring, 0x0);
1900 radeon_ring_write(ring, rdev->config.si.max_hw_contexts - 1);
1901 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1902 radeon_ring_write(ring, 0);
1903 radeon_ring_write(ring, 0);
1905 /* init the CE partitions */
1906 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
1907 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
1908 radeon_ring_write(ring, 0xc000);
1909 radeon_ring_write(ring, 0xe000);
1910 radeon_ring_unlock_commit(rdev, ring);
1912 si_cp_enable(rdev, true);
1914 r = radeon_ring_lock(rdev, ring, si_default_size + 10);
1916 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1920 /* setup clear context state */
1921 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1922 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1924 for (i = 0; i < si_default_size; i++)
1925 radeon_ring_write(ring, si_default_state[i]);
1927 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1928 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
1930 /* set clear context state */
1931 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
1932 radeon_ring_write(ring, 0);
1934 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1935 radeon_ring_write(ring, 0x00000316);
1936 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
1937 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
1939 radeon_ring_unlock_commit(rdev, ring);
1941 for (i = RADEON_RING_TYPE_GFX_INDEX; i <= CAYMAN_RING_TYPE_CP2_INDEX; ++i) {
1942 ring = &rdev->ring[i];
1943 r = radeon_ring_lock(rdev, ring, 2);
1945 /* clear the compute context state */
1946 radeon_ring_write(ring, PACKET3_COMPUTE(PACKET3_CLEAR_STATE, 0));
1947 radeon_ring_write(ring, 0);
1949 radeon_ring_unlock_commit(rdev, ring);
1955 static void si_cp_fini(struct radeon_device *rdev)
1957 struct radeon_ring *ring;
1958 si_cp_enable(rdev, false);
1960 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1961 radeon_ring_fini(rdev, ring);
1962 radeon_scratch_free(rdev, ring->rptr_save_reg);
1964 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
1965 radeon_ring_fini(rdev, ring);
1966 radeon_scratch_free(rdev, ring->rptr_save_reg);
1968 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
1969 radeon_ring_fini(rdev, ring);
1970 radeon_scratch_free(rdev, ring->rptr_save_reg);
1973 static int si_cp_resume(struct radeon_device *rdev)
1975 struct radeon_ring *ring;
1980 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
1981 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
1986 RREG32(GRBM_SOFT_RESET);
1988 WREG32(GRBM_SOFT_RESET, 0);
1989 RREG32(GRBM_SOFT_RESET);
1991 WREG32(CP_SEM_WAIT_TIMER, 0x0);
1992 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
1994 /* Set the write pointer delay */
1995 WREG32(CP_RB_WPTR_DELAY, 0);
1997 WREG32(CP_DEBUG, 0);
1998 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
2000 /* ring 0 - compute and gfx */
2001 /* Set ring buffer size */
2002 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2003 rb_bufsz = drm_order(ring->ring_size / 8);
2004 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2006 tmp |= BUF_SWAP_32BIT;
2008 WREG32(CP_RB0_CNTL, tmp);
2010 /* Initialize the ring buffer's read and write pointers */
2011 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
2013 WREG32(CP_RB0_WPTR, ring->wptr);
2015 /* set the wb address whether it's enabled or not */
2016 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
2017 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
2019 if (rdev->wb.enabled)
2020 WREG32(SCRATCH_UMSK, 0xff);
2022 tmp |= RB_NO_UPDATE;
2023 WREG32(SCRATCH_UMSK, 0);
2027 WREG32(CP_RB0_CNTL, tmp);
2029 WREG32(CP_RB0_BASE, ring->gpu_addr >> 8);
2031 ring->rptr = RREG32(CP_RB0_RPTR);
2033 /* ring1 - compute only */
2034 /* Set ring buffer size */
2035 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
2036 rb_bufsz = drm_order(ring->ring_size / 8);
2037 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2039 tmp |= BUF_SWAP_32BIT;
2041 WREG32(CP_RB1_CNTL, tmp);
2043 /* Initialize the ring buffer's read and write pointers */
2044 WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA);
2046 WREG32(CP_RB1_WPTR, ring->wptr);
2048 /* set the wb address whether it's enabled or not */
2049 WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC);
2050 WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF);
2053 WREG32(CP_RB1_CNTL, tmp);
2055 WREG32(CP_RB1_BASE, ring->gpu_addr >> 8);
2057 ring->rptr = RREG32(CP_RB1_RPTR);
2059 /* ring2 - compute only */
2060 /* Set ring buffer size */
2061 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
2062 rb_bufsz = drm_order(ring->ring_size / 8);
2063 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2065 tmp |= BUF_SWAP_32BIT;
2067 WREG32(CP_RB2_CNTL, tmp);
2069 /* Initialize the ring buffer's read and write pointers */
2070 WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA);
2072 WREG32(CP_RB2_WPTR, ring->wptr);
2074 /* set the wb address whether it's enabled or not */
2075 WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC);
2076 WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF);
2079 WREG32(CP_RB2_CNTL, tmp);
2081 WREG32(CP_RB2_BASE, ring->gpu_addr >> 8);
2083 ring->rptr = RREG32(CP_RB2_RPTR);
2085 /* start the rings */
2087 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
2088 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = true;
2089 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = true;
2090 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
2092 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
2093 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2094 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2097 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP1_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]);
2099 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2101 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP2_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]);
2103 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2109 bool si_gpu_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2112 u32 grbm_status, grbm_status2;
2113 u32 grbm_status_se0, grbm_status_se1;
2115 srbm_status = RREG32(SRBM_STATUS);
2116 grbm_status = RREG32(GRBM_STATUS);
2117 grbm_status2 = RREG32(GRBM_STATUS2);
2118 grbm_status_se0 = RREG32(GRBM_STATUS_SE0);
2119 grbm_status_se1 = RREG32(GRBM_STATUS_SE1);
2120 if (!(grbm_status & GUI_ACTIVE)) {
2121 radeon_ring_lockup_update(ring);
2124 /* force CP activities */
2125 radeon_ring_force_activity(rdev, ring);
2126 return radeon_ring_test_lockup(rdev, ring);
2129 static void si_gpu_soft_reset_gfx(struct radeon_device *rdev)
2133 if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
2136 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
2137 RREG32(GRBM_STATUS));
2138 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
2139 RREG32(GRBM_STATUS2));
2140 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
2141 RREG32(GRBM_STATUS_SE0));
2142 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
2143 RREG32(GRBM_STATUS_SE1));
2144 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
2145 RREG32(SRBM_STATUS));
2147 /* Disable CP parsing/prefetching */
2148 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
2150 /* reset all the gfx blocks */
2151 grbm_reset = (SOFT_RESET_CP |
2165 dev_info(rdev->dev, " GRBM_SOFT_RESET=0x%08X\n", grbm_reset);
2166 WREG32(GRBM_SOFT_RESET, grbm_reset);
2167 (void)RREG32(GRBM_SOFT_RESET);
2169 WREG32(GRBM_SOFT_RESET, 0);
2170 (void)RREG32(GRBM_SOFT_RESET);
2172 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n",
2173 RREG32(GRBM_STATUS));
2174 dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n",
2175 RREG32(GRBM_STATUS2));
2176 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n",
2177 RREG32(GRBM_STATUS_SE0));
2178 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n",
2179 RREG32(GRBM_STATUS_SE1));
2180 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n",
2181 RREG32(SRBM_STATUS));
2184 static void si_gpu_soft_reset_dma(struct radeon_device *rdev)
2188 if (RREG32(DMA_STATUS_REG) & DMA_IDLE)
2191 dev_info(rdev->dev, " DMA_STATUS_REG = 0x%08X\n",
2192 RREG32(DMA_STATUS_REG));
2195 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
2196 tmp &= ~DMA_RB_ENABLE;
2197 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp);
2200 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
2201 tmp &= ~DMA_RB_ENABLE;
2202 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp);
2205 WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA | SOFT_RESET_DMA1);
2206 RREG32(SRBM_SOFT_RESET);
2208 WREG32(SRBM_SOFT_RESET, 0);
2210 dev_info(rdev->dev, " DMA_STATUS_REG = 0x%08X\n",
2211 RREG32(DMA_STATUS_REG));
2214 static int si_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
2216 struct evergreen_mc_save save;
2218 if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE))
2219 reset_mask &= ~(RADEON_RESET_GFX | RADEON_RESET_COMPUTE);
2221 if (RREG32(DMA_STATUS_REG) & DMA_IDLE)
2222 reset_mask &= ~RADEON_RESET_DMA;
2224 if (reset_mask == 0)
2227 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
2229 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
2230 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
2231 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
2232 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
2234 evergreen_mc_stop(rdev, &save);
2235 if (radeon_mc_wait_for_idle(rdev)) {
2236 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2239 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE))
2240 si_gpu_soft_reset_gfx(rdev);
2242 if (reset_mask & RADEON_RESET_DMA)
2243 si_gpu_soft_reset_dma(rdev);
2245 /* Wait a little for things to settle down */
2248 evergreen_mc_resume(rdev, &save);
2252 int si_asic_reset(struct radeon_device *rdev)
2254 return si_gpu_soft_reset(rdev, (RADEON_RESET_GFX |
2255 RADEON_RESET_COMPUTE |
2260 static void si_mc_program(struct radeon_device *rdev)
2262 struct evergreen_mc_save save;
2266 /* Initialize HDP */
2267 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
2268 WREG32((0x2c14 + j), 0x00000000);
2269 WREG32((0x2c18 + j), 0x00000000);
2270 WREG32((0x2c1c + j), 0x00000000);
2271 WREG32((0x2c20 + j), 0x00000000);
2272 WREG32((0x2c24 + j), 0x00000000);
2274 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
2276 evergreen_mc_stop(rdev, &save);
2277 if (radeon_mc_wait_for_idle(rdev)) {
2278 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2280 /* Lockout access through VGA aperture*/
2281 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
2282 /* Update configuration */
2283 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2284 rdev->mc.vram_start >> 12);
2285 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2286 rdev->mc.vram_end >> 12);
2287 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
2288 rdev->vram_scratch.gpu_addr >> 12);
2289 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
2290 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
2291 WREG32(MC_VM_FB_LOCATION, tmp);
2292 /* XXX double check these! */
2293 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
2294 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
2295 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
2296 WREG32(MC_VM_AGP_BASE, 0);
2297 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
2298 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
2299 if (radeon_mc_wait_for_idle(rdev)) {
2300 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2302 evergreen_mc_resume(rdev, &save);
2303 /* we need to own VRAM, so turn off the VGA renderer here
2304 * to stop it overwriting our objects */
2305 rv515_vga_render_disable(rdev);
2308 /* SI MC address space is 40 bits */
2309 static void si_vram_location(struct radeon_device *rdev,
2310 struct radeon_mc *mc, u64 base)
2312 mc->vram_start = base;
2313 if (mc->mc_vram_size > (0xFFFFFFFFFFULL - base + 1)) {
2314 dev_warn(rdev->dev, "limiting VRAM to PCI aperture size\n");
2315 mc->real_vram_size = mc->aper_size;
2316 mc->mc_vram_size = mc->aper_size;
2318 mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
2319 dev_info(rdev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
2320 mc->mc_vram_size >> 20, mc->vram_start,
2321 mc->vram_end, mc->real_vram_size >> 20);
2324 static void si_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
2326 u64 size_af, size_bf;
2328 size_af = ((0xFFFFFFFFFFULL - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align;
2329 size_bf = mc->vram_start & ~mc->gtt_base_align;
2330 if (size_bf > size_af) {
2331 if (mc->gtt_size > size_bf) {
2332 dev_warn(rdev->dev, "limiting GTT\n");
2333 mc->gtt_size = size_bf;
2335 mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size;
2337 if (mc->gtt_size > size_af) {
2338 dev_warn(rdev->dev, "limiting GTT\n");
2339 mc->gtt_size = size_af;
2341 mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align;
2343 mc->gtt_end = mc->gtt_start + mc->gtt_size - 1;
2344 dev_info(rdev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n",
2345 mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end);
2348 static void si_vram_gtt_location(struct radeon_device *rdev,
2349 struct radeon_mc *mc)
2351 if (mc->mc_vram_size > 0xFFC0000000ULL) {
2352 /* leave room for at least 1024M GTT */
2353 dev_warn(rdev->dev, "limiting VRAM\n");
2354 mc->real_vram_size = 0xFFC0000000ULL;
2355 mc->mc_vram_size = 0xFFC0000000ULL;
2357 si_vram_location(rdev, &rdev->mc, 0);
2358 rdev->mc.gtt_base_align = 0;
2359 si_gtt_location(rdev, mc);
2362 static int si_mc_init(struct radeon_device *rdev)
2365 int chansize, numchan;
2367 /* Get VRAM informations */
2368 rdev->mc.vram_is_ddr = true;
2369 tmp = RREG32(MC_ARB_RAMCFG);
2370 if (tmp & CHANSIZE_OVERRIDE) {
2372 } else if (tmp & CHANSIZE_MASK) {
2377 tmp = RREG32(MC_SHARED_CHMAP);
2378 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
2408 rdev->mc.vram_width = numchan * chansize;
2409 /* Could aper size report 0 ? */
2410 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2411 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2412 /* size in MB on si */
2413 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2414 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2415 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2416 si_vram_gtt_location(rdev, &rdev->mc);
2417 radeon_update_bandwidth_info(rdev);
2425 void si_pcie_gart_tlb_flush(struct radeon_device *rdev)
2427 /* flush hdp cache */
2428 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2430 /* bits 0-15 are the VM contexts0-15 */
2431 WREG32(VM_INVALIDATE_REQUEST, 1);
2434 static int si_pcie_gart_enable(struct radeon_device *rdev)
2438 if (rdev->gart.robj == NULL) {
2439 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
2442 r = radeon_gart_table_vram_pin(rdev);
2445 radeon_gart_restore(rdev);
2446 /* Setup TLB control */
2447 WREG32(MC_VM_MX_L1_TLB_CNTL,
2450 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2451 ENABLE_ADVANCED_DRIVER_MODEL |
2452 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2453 /* Setup L2 cache */
2454 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
2455 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2456 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2457 EFFECTIVE_L2_QUEUE_SIZE(7) |
2458 CONTEXT1_IDENTITY_ACCESS_MODE(1));
2459 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
2460 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2461 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2462 /* setup context0 */
2463 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
2464 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
2465 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
2466 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
2467 (u32)(rdev->dummy_page.addr >> 12));
2468 WREG32(VM_CONTEXT0_CNTL2, 0);
2469 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
2470 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
2476 /* empty context1-15 */
2477 /* set vm size, must be a multiple of 4 */
2478 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
2479 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn);
2480 /* Assign the pt base to something valid for now; the pts used for
2481 * the VMs are determined by the application and setup and assigned
2482 * on the fly in the vm part of radeon_gart.c
2484 for (i = 1; i < 16; i++) {
2486 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
2487 rdev->gart.table_addr >> 12);
2489 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
2490 rdev->gart.table_addr >> 12);
2493 /* enable context1-15 */
2494 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
2495 (u32)(rdev->dummy_page.addr >> 12));
2496 WREG32(VM_CONTEXT1_CNTL2, 4);
2497 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
2498 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2499 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
2500 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2501 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
2502 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
2503 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
2504 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
2505 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
2506 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
2507 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
2508 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2509 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
2511 si_pcie_gart_tlb_flush(rdev);
2512 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
2513 (unsigned)(rdev->mc.gtt_size >> 20),
2514 (unsigned long long)rdev->gart.table_addr);
2515 rdev->gart.ready = true;
2519 static void si_pcie_gart_disable(struct radeon_device *rdev)
2521 /* Disable all tables */
2522 WREG32(VM_CONTEXT0_CNTL, 0);
2523 WREG32(VM_CONTEXT1_CNTL, 0);
2524 /* Setup TLB control */
2525 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2526 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2527 /* Setup L2 cache */
2528 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2529 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2530 EFFECTIVE_L2_QUEUE_SIZE(7) |
2531 CONTEXT1_IDENTITY_ACCESS_MODE(1));
2532 WREG32(VM_L2_CNTL2, 0);
2533 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2534 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2535 radeon_gart_table_vram_unpin(rdev);
2538 static void si_pcie_gart_fini(struct radeon_device *rdev)
2540 si_pcie_gart_disable(rdev);
2541 radeon_gart_table_vram_free(rdev);
2542 radeon_gart_fini(rdev);
2546 static bool si_vm_reg_valid(u32 reg)
2548 /* context regs are fine */
2552 /* check config regs */
2554 case GRBM_GFX_INDEX:
2555 case CP_STRMOUT_CNTL:
2556 case VGT_VTX_VECT_EJECT_REG:
2557 case VGT_CACHE_INVALIDATION:
2558 case VGT_ESGS_RING_SIZE:
2559 case VGT_GSVS_RING_SIZE:
2560 case VGT_GS_VERTEX_REUSE:
2561 case VGT_PRIMITIVE_TYPE:
2562 case VGT_INDEX_TYPE:
2563 case VGT_NUM_INDICES:
2564 case VGT_NUM_INSTANCES:
2565 case VGT_TF_RING_SIZE:
2566 case VGT_HS_OFFCHIP_PARAM:
2567 case VGT_TF_MEMORY_BASE:
2569 case PA_SU_LINE_STIPPLE_VALUE:
2570 case PA_SC_LINE_STIPPLE_STATE:
2573 case SPI_STATIC_THREAD_MGMT_1:
2574 case SPI_STATIC_THREAD_MGMT_2:
2575 case SPI_STATIC_THREAD_MGMT_3:
2576 case SPI_PS_MAX_WAVE_ID:
2577 case SPI_CONFIG_CNTL:
2578 case SPI_CONFIG_CNTL_1:
2582 DRM_ERROR("Invalid register 0x%x in CS\n", reg);
2587 static int si_vm_packet3_ce_check(struct radeon_device *rdev,
2588 u32 *ib, struct radeon_cs_packet *pkt)
2590 switch (pkt->opcode) {
2592 case PACKET3_SET_BASE:
2593 case PACKET3_SET_CE_DE_COUNTERS:
2594 case PACKET3_LOAD_CONST_RAM:
2595 case PACKET3_WRITE_CONST_RAM:
2596 case PACKET3_WRITE_CONST_RAM_OFFSET:
2597 case PACKET3_DUMP_CONST_RAM:
2598 case PACKET3_INCREMENT_CE_COUNTER:
2599 case PACKET3_WAIT_ON_DE_COUNTER:
2600 case PACKET3_CE_WRITE:
2603 DRM_ERROR("Invalid CE packet3: 0x%x\n", pkt->opcode);
2609 static int si_vm_packet3_gfx_check(struct radeon_device *rdev,
2610 u32 *ib, struct radeon_cs_packet *pkt)
2612 u32 idx = pkt->idx + 1;
2613 u32 idx_value = ib[idx];
2614 u32 start_reg, end_reg, reg, i;
2617 switch (pkt->opcode) {
2619 case PACKET3_SET_BASE:
2620 case PACKET3_CLEAR_STATE:
2621 case PACKET3_INDEX_BUFFER_SIZE:
2622 case PACKET3_DISPATCH_DIRECT:
2623 case PACKET3_DISPATCH_INDIRECT:
2624 case PACKET3_ALLOC_GDS:
2625 case PACKET3_WRITE_GDS_RAM:
2626 case PACKET3_ATOMIC_GDS:
2627 case PACKET3_ATOMIC:
2628 case PACKET3_OCCLUSION_QUERY:
2629 case PACKET3_SET_PREDICATION:
2630 case PACKET3_COND_EXEC:
2631 case PACKET3_PRED_EXEC:
2632 case PACKET3_DRAW_INDIRECT:
2633 case PACKET3_DRAW_INDEX_INDIRECT:
2634 case PACKET3_INDEX_BASE:
2635 case PACKET3_DRAW_INDEX_2:
2636 case PACKET3_CONTEXT_CONTROL:
2637 case PACKET3_INDEX_TYPE:
2638 case PACKET3_DRAW_INDIRECT_MULTI:
2639 case PACKET3_DRAW_INDEX_AUTO:
2640 case PACKET3_DRAW_INDEX_IMMD:
2641 case PACKET3_NUM_INSTANCES:
2642 case PACKET3_DRAW_INDEX_MULTI_AUTO:
2643 case PACKET3_STRMOUT_BUFFER_UPDATE:
2644 case PACKET3_DRAW_INDEX_OFFSET_2:
2645 case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
2646 case PACKET3_DRAW_INDEX_INDIRECT_MULTI:
2647 case PACKET3_MPEG_INDEX:
2648 case PACKET3_WAIT_REG_MEM:
2649 case PACKET3_MEM_WRITE:
2650 case PACKET3_PFP_SYNC_ME:
2651 case PACKET3_SURFACE_SYNC:
2652 case PACKET3_EVENT_WRITE:
2653 case PACKET3_EVENT_WRITE_EOP:
2654 case PACKET3_EVENT_WRITE_EOS:
2655 case PACKET3_SET_CONTEXT_REG:
2656 case PACKET3_SET_CONTEXT_REG_INDIRECT:
2657 case PACKET3_SET_SH_REG:
2658 case PACKET3_SET_SH_REG_OFFSET:
2659 case PACKET3_INCREMENT_DE_COUNTER:
2660 case PACKET3_WAIT_ON_CE_COUNTER:
2661 case PACKET3_WAIT_ON_AVAIL_BUFFER:
2662 case PACKET3_ME_WRITE:
2664 case PACKET3_COPY_DATA:
2665 if ((idx_value & 0xf00) == 0) {
2666 reg = ib[idx + 3] * 4;
2667 if (!si_vm_reg_valid(reg))
2671 case PACKET3_WRITE_DATA:
2672 if ((idx_value & 0xf00) == 0) {
2673 start_reg = ib[idx + 1] * 4;
2674 if (idx_value & 0x10000) {
2675 if (!si_vm_reg_valid(start_reg))
2678 for (i = 0; i < (pkt->count - 2); i++) {
2679 reg = start_reg + (4 * i);
2680 if (!si_vm_reg_valid(reg))
2686 case PACKET3_COND_WRITE:
2687 if (idx_value & 0x100) {
2688 reg = ib[idx + 5] * 4;
2689 if (!si_vm_reg_valid(reg))
2693 case PACKET3_COPY_DW:
2694 if (idx_value & 0x2) {
2695 reg = ib[idx + 3] * 4;
2696 if (!si_vm_reg_valid(reg))
2700 case PACKET3_SET_CONFIG_REG:
2701 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2702 end_reg = 4 * pkt->count + start_reg - 4;
2703 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2704 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2705 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2706 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2709 for (i = 0; i < pkt->count; i++) {
2710 reg = start_reg + (4 * i);
2711 if (!si_vm_reg_valid(reg))
2715 case PACKET3_CP_DMA:
2716 command = ib[idx + 4];
2718 if (command & PACKET3_CP_DMA_CMD_SAS) {
2719 /* src address space is register */
2720 if (((info & 0x60000000) >> 29) == 0) {
2721 start_reg = idx_value << 2;
2722 if (command & PACKET3_CP_DMA_CMD_SAIC) {
2724 if (!si_vm_reg_valid(reg)) {
2725 DRM_ERROR("CP DMA Bad SRC register\n");
2729 for (i = 0; i < (command & 0x1fffff); i++) {
2730 reg = start_reg + (4 * i);
2731 if (!si_vm_reg_valid(reg)) {
2732 DRM_ERROR("CP DMA Bad SRC register\n");
2739 if (command & PACKET3_CP_DMA_CMD_DAS) {
2740 /* dst address space is register */
2741 if (((info & 0x00300000) >> 20) == 0) {
2742 start_reg = ib[idx + 2];
2743 if (command & PACKET3_CP_DMA_CMD_DAIC) {
2745 if (!si_vm_reg_valid(reg)) {
2746 DRM_ERROR("CP DMA Bad DST register\n");
2750 for (i = 0; i < (command & 0x1fffff); i++) {
2751 reg = start_reg + (4 * i);
2752 if (!si_vm_reg_valid(reg)) {
2753 DRM_ERROR("CP DMA Bad DST register\n");
2762 DRM_ERROR("Invalid GFX packet3: 0x%x\n", pkt->opcode);
2768 static int si_vm_packet3_compute_check(struct radeon_device *rdev,
2769 u32 *ib, struct radeon_cs_packet *pkt)
2771 u32 idx = pkt->idx + 1;
2772 u32 idx_value = ib[idx];
2773 u32 start_reg, reg, i;
2775 switch (pkt->opcode) {
2777 case PACKET3_SET_BASE:
2778 case PACKET3_CLEAR_STATE:
2779 case PACKET3_DISPATCH_DIRECT:
2780 case PACKET3_DISPATCH_INDIRECT:
2781 case PACKET3_ALLOC_GDS:
2782 case PACKET3_WRITE_GDS_RAM:
2783 case PACKET3_ATOMIC_GDS:
2784 case PACKET3_ATOMIC:
2785 case PACKET3_OCCLUSION_QUERY:
2786 case PACKET3_SET_PREDICATION:
2787 case PACKET3_COND_EXEC:
2788 case PACKET3_PRED_EXEC:
2789 case PACKET3_CONTEXT_CONTROL:
2790 case PACKET3_STRMOUT_BUFFER_UPDATE:
2791 case PACKET3_WAIT_REG_MEM:
2792 case PACKET3_MEM_WRITE:
2793 case PACKET3_PFP_SYNC_ME:
2794 case PACKET3_SURFACE_SYNC:
2795 case PACKET3_EVENT_WRITE:
2796 case PACKET3_EVENT_WRITE_EOP:
2797 case PACKET3_EVENT_WRITE_EOS:
2798 case PACKET3_SET_CONTEXT_REG:
2799 case PACKET3_SET_CONTEXT_REG_INDIRECT:
2800 case PACKET3_SET_SH_REG:
2801 case PACKET3_SET_SH_REG_OFFSET:
2802 case PACKET3_INCREMENT_DE_COUNTER:
2803 case PACKET3_WAIT_ON_CE_COUNTER:
2804 case PACKET3_WAIT_ON_AVAIL_BUFFER:
2805 case PACKET3_ME_WRITE:
2807 case PACKET3_COPY_DATA:
2808 if ((idx_value & 0xf00) == 0) {
2809 reg = ib[idx + 3] * 4;
2810 if (!si_vm_reg_valid(reg))
2814 case PACKET3_WRITE_DATA:
2815 if ((idx_value & 0xf00) == 0) {
2816 start_reg = ib[idx + 1] * 4;
2817 if (idx_value & 0x10000) {
2818 if (!si_vm_reg_valid(start_reg))
2821 for (i = 0; i < (pkt->count - 2); i++) {
2822 reg = start_reg + (4 * i);
2823 if (!si_vm_reg_valid(reg))
2829 case PACKET3_COND_WRITE:
2830 if (idx_value & 0x100) {
2831 reg = ib[idx + 5] * 4;
2832 if (!si_vm_reg_valid(reg))
2836 case PACKET3_COPY_DW:
2837 if (idx_value & 0x2) {
2838 reg = ib[idx + 3] * 4;
2839 if (!si_vm_reg_valid(reg))
2844 DRM_ERROR("Invalid Compute packet3: 0x%x\n", pkt->opcode);
2850 int si_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
2854 struct radeon_cs_packet pkt;
2858 pkt.type = CP_PACKET_GET_TYPE(ib->ptr[idx]);
2859 pkt.count = CP_PACKET_GET_COUNT(ib->ptr[idx]);
2863 dev_err(rdev->dev, "Packet0 not allowed!\n");
2870 pkt.opcode = CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
2871 if (ib->is_const_ib)
2872 ret = si_vm_packet3_ce_check(rdev, ib->ptr, &pkt);
2875 case RADEON_RING_TYPE_GFX_INDEX:
2876 ret = si_vm_packet3_gfx_check(rdev, ib->ptr, &pkt);
2878 case CAYMAN_RING_TYPE_CP1_INDEX:
2879 case CAYMAN_RING_TYPE_CP2_INDEX:
2880 ret = si_vm_packet3_compute_check(rdev, ib->ptr, &pkt);
2883 dev_err(rdev->dev, "Non-PM4 ring %d !\n", ib->ring);
2888 idx += pkt.count + 2;
2891 dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
2897 } while (idx < ib->length_dw);
2905 int si_vm_init(struct radeon_device *rdev)
2908 rdev->vm_manager.nvm = 16;
2909 /* base offset of vram pages */
2910 rdev->vm_manager.vram_base_offset = 0;
2915 void si_vm_fini(struct radeon_device *rdev)
2920 * si_vm_set_page - update the page tables using the CP
2922 * @rdev: radeon_device pointer
2923 * @pe: addr of the page entry
2924 * @addr: dst addr to write into pe
2925 * @count: number of page entries to update
2926 * @incr: increase next addr by incr bytes
2927 * @flags: access flags
2929 * Update the page tables using the CP (cayman-si).
2931 void si_vm_set_page(struct radeon_device *rdev, uint64_t pe,
2932 uint64_t addr, unsigned count,
2933 uint32_t incr, uint32_t flags)
2935 struct radeon_ring *ring = &rdev->ring[rdev->asic->vm.pt_ring_index];
2936 uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
2940 if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) {
2942 ndw = 2 + count * 2;
2946 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, ndw));
2947 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
2948 WRITE_DATA_DST_SEL(1)));
2949 radeon_ring_write(ring, pe);
2950 radeon_ring_write(ring, upper_32_bits(pe));
2951 for (; ndw > 2; ndw -= 2, --count, pe += 8) {
2952 if (flags & RADEON_VM_PAGE_SYSTEM) {
2953 value = radeon_vm_map_gart(rdev, addr);
2954 value &= 0xFFFFFFFFFFFFF000ULL;
2955 } else if (flags & RADEON_VM_PAGE_VALID) {
2961 value |= r600_flags;
2962 radeon_ring_write(ring, value);
2963 radeon_ring_write(ring, upper_32_bits(value));
2968 if (flags & RADEON_VM_PAGE_SYSTEM) {
2974 /* for non-physically contiguous pages (system) */
2975 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw));
2976 radeon_ring_write(ring, pe);
2977 radeon_ring_write(ring, upper_32_bits(pe) & 0xff);
2978 for (; ndw > 0; ndw -= 2, --count, pe += 8) {
2979 if (flags & RADEON_VM_PAGE_SYSTEM) {
2980 value = radeon_vm_map_gart(rdev, addr);
2981 value &= 0xFFFFFFFFFFFFF000ULL;
2982 } else if (flags & RADEON_VM_PAGE_VALID) {
2988 value |= r600_flags;
2989 radeon_ring_write(ring, value);
2990 radeon_ring_write(ring, upper_32_bits(value));
2999 if (flags & RADEON_VM_PAGE_VALID)
3003 /* for physically contiguous pages (vram) */
3004 radeon_ring_write(ring, DMA_PTE_PDE_PACKET(ndw));
3005 radeon_ring_write(ring, pe); /* dst addr */
3006 radeon_ring_write(ring, upper_32_bits(pe) & 0xff);
3007 radeon_ring_write(ring, r600_flags); /* mask */
3008 radeon_ring_write(ring, 0);
3009 radeon_ring_write(ring, value); /* value */
3010 radeon_ring_write(ring, upper_32_bits(value));
3011 radeon_ring_write(ring, incr); /* increment size */
3012 radeon_ring_write(ring, 0);
3014 addr += (ndw / 2) * incr;
3021 void si_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
3023 struct radeon_ring *ring = &rdev->ring[ridx];
3028 /* write new base address */
3029 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3030 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3031 WRITE_DATA_DST_SEL(0)));
3034 radeon_ring_write(ring,
3035 (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
3037 radeon_ring_write(ring,
3038 (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
3040 radeon_ring_write(ring, 0);
3041 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
3043 /* flush hdp cache */
3044 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3045 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3046 WRITE_DATA_DST_SEL(0)));
3047 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
3048 radeon_ring_write(ring, 0);
3049 radeon_ring_write(ring, 0x1);
3051 /* bits 0-15 are the VM contexts0-15 */
3052 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3053 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3054 WRITE_DATA_DST_SEL(0)));
3055 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
3056 radeon_ring_write(ring, 0);
3057 radeon_ring_write(ring, 1 << vm->id);
3059 /* sync PFP to ME, otherwise we might get invalid PFP reads */
3060 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3061 radeon_ring_write(ring, 0x0);
3064 void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
3066 struct radeon_ring *ring = &rdev->ring[ridx];
3071 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3073 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
3075 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2));
3077 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
3079 /* flush hdp cache */
3080 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3081 radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
3082 radeon_ring_write(ring, 1);
3084 /* bits 0-7 are the VM contexts0-7 */
3085 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3086 radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
3087 radeon_ring_write(ring, 1 << vm->id);
3093 void si_rlc_fini(struct radeon_device *rdev)
3097 /* save restore block */
3098 if (rdev->rlc.save_restore_obj) {
3099 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
3100 if (unlikely(r != 0))
3101 dev_warn(rdev->dev, "(%d) reserve RLC sr bo failed\n", r);
3102 radeon_bo_unpin(rdev->rlc.save_restore_obj);
3103 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
3105 radeon_bo_unref(&rdev->rlc.save_restore_obj);
3106 rdev->rlc.save_restore_obj = NULL;
3109 /* clear state block */
3110 if (rdev->rlc.clear_state_obj) {
3111 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3112 if (unlikely(r != 0))
3113 dev_warn(rdev->dev, "(%d) reserve RLC c bo failed\n", r);
3114 radeon_bo_unpin(rdev->rlc.clear_state_obj);
3115 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3117 radeon_bo_unref(&rdev->rlc.clear_state_obj);
3118 rdev->rlc.clear_state_obj = NULL;
3122 int si_rlc_init(struct radeon_device *rdev)
3126 /* save restore block */
3127 if (rdev->rlc.save_restore_obj == NULL) {
3128 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3129 RADEON_GEM_DOMAIN_VRAM, NULL,
3130 &rdev->rlc.save_restore_obj);
3132 dev_warn(rdev->dev, "(%d) create RLC sr bo failed\n", r);
3137 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
3138 if (unlikely(r != 0)) {
3142 r = radeon_bo_pin(rdev->rlc.save_restore_obj, RADEON_GEM_DOMAIN_VRAM,
3143 &rdev->rlc.save_restore_gpu_addr);
3144 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
3146 dev_warn(rdev->dev, "(%d) pin RLC sr bo failed\n", r);
3151 /* clear state block */
3152 if (rdev->rlc.clear_state_obj == NULL) {
3153 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3154 RADEON_GEM_DOMAIN_VRAM, NULL,
3155 &rdev->rlc.clear_state_obj);
3157 dev_warn(rdev->dev, "(%d) create RLC c bo failed\n", r);
3162 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3163 if (unlikely(r != 0)) {
3167 r = radeon_bo_pin(rdev->rlc.clear_state_obj, RADEON_GEM_DOMAIN_VRAM,
3168 &rdev->rlc.clear_state_gpu_addr);
3169 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3171 dev_warn(rdev->dev, "(%d) pin RLC c bo failed\n", r);
3179 static void si_rlc_stop(struct radeon_device *rdev)
3181 WREG32(RLC_CNTL, 0);
3184 static void si_rlc_start(struct radeon_device *rdev)
3186 WREG32(RLC_CNTL, RLC_ENABLE);
3189 static int si_rlc_resume(struct radeon_device *rdev)
3192 const __be32 *fw_data;
3199 WREG32(RLC_RL_BASE, 0);
3200 WREG32(RLC_RL_SIZE, 0);
3201 WREG32(RLC_LB_CNTL, 0);
3202 WREG32(RLC_LB_CNTR_MAX, 0xffffffff);
3203 WREG32(RLC_LB_CNTR_INIT, 0);
3205 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
3206 WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8);
3208 WREG32(RLC_MC_CNTL, 0);
3209 WREG32(RLC_UCODE_CNTL, 0);
3211 fw_data = (const __be32 *)rdev->rlc_fw->data;
3212 for (i = 0; i < SI_RLC_UCODE_SIZE; i++) {
3213 WREG32(RLC_UCODE_ADDR, i);
3214 WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3216 WREG32(RLC_UCODE_ADDR, 0);
3223 static void si_enable_interrupts(struct radeon_device *rdev)
3225 u32 ih_cntl = RREG32(IH_CNTL);
3226 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3228 ih_cntl |= ENABLE_INTR;
3229 ih_rb_cntl |= IH_RB_ENABLE;
3230 WREG32(IH_CNTL, ih_cntl);
3231 WREG32(IH_RB_CNTL, ih_rb_cntl);
3232 rdev->ih.enabled = true;
3235 static void si_disable_interrupts(struct radeon_device *rdev)
3237 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3238 u32 ih_cntl = RREG32(IH_CNTL);
3240 ih_rb_cntl &= ~IH_RB_ENABLE;
3241 ih_cntl &= ~ENABLE_INTR;
3242 WREG32(IH_RB_CNTL, ih_rb_cntl);
3243 WREG32(IH_CNTL, ih_cntl);
3244 /* set rptr, wptr to 0 */
3245 WREG32(IH_RB_RPTR, 0);
3246 WREG32(IH_RB_WPTR, 0);
3247 rdev->ih.enabled = false;
3251 static void si_disable_interrupt_state(struct radeon_device *rdev)
3255 WREG32(CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
3256 WREG32(CP_INT_CNTL_RING1, 0);
3257 WREG32(CP_INT_CNTL_RING2, 0);
3258 tmp = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
3259 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, tmp);
3260 tmp = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
3261 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, tmp);
3262 WREG32(GRBM_INT_CNTL, 0);
3263 WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
3264 WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
3265 if (rdev->num_crtc >= 4) {
3266 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
3267 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
3269 if (rdev->num_crtc >= 6) {
3270 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
3271 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
3274 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
3275 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
3276 if (rdev->num_crtc >= 4) {
3277 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
3278 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
3280 if (rdev->num_crtc >= 6) {
3281 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
3282 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
3285 WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
3287 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3288 WREG32(DC_HPD1_INT_CONTROL, tmp);
3289 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3290 WREG32(DC_HPD2_INT_CONTROL, tmp);
3291 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3292 WREG32(DC_HPD3_INT_CONTROL, tmp);
3293 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3294 WREG32(DC_HPD4_INT_CONTROL, tmp);
3295 tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3296 WREG32(DC_HPD5_INT_CONTROL, tmp);
3297 tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3298 WREG32(DC_HPD6_INT_CONTROL, tmp);
3302 static int si_irq_init(struct radeon_device *rdev)
3306 u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
3309 ret = r600_ih_ring_alloc(rdev);
3314 si_disable_interrupts(rdev);
3317 ret = si_rlc_resume(rdev);
3319 r600_ih_ring_fini(rdev);
3323 /* setup interrupt control */
3324 /* set dummy read address to ring address */
3325 WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
3326 interrupt_cntl = RREG32(INTERRUPT_CNTL);
3327 /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
3328 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
3330 interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
3331 /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
3332 interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
3333 WREG32(INTERRUPT_CNTL, interrupt_cntl);
3335 WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
3336 rb_bufsz = drm_order(rdev->ih.ring_size / 4);
3338 ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
3339 IH_WPTR_OVERFLOW_CLEAR |
3342 if (rdev->wb.enabled)
3343 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
3345 /* set the writeback address whether it's enabled or not */
3346 WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
3347 WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
3349 WREG32(IH_RB_CNTL, ih_rb_cntl);
3351 /* set rptr, wptr to 0 */
3352 WREG32(IH_RB_RPTR, 0);
3353 WREG32(IH_RB_WPTR, 0);
3355 /* Default settings for IH_CNTL (disabled at first) */
3356 ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0);
3357 /* RPTR_REARM only works if msi's are enabled */
3358 if (rdev->msi_enabled)
3359 ih_cntl |= RPTR_REARM;
3360 WREG32(IH_CNTL, ih_cntl);
3362 /* force the active interrupt state to all disabled */
3363 si_disable_interrupt_state(rdev);
3365 pci_set_master(rdev->pdev);
3368 si_enable_interrupts(rdev);
3373 int si_irq_set(struct radeon_device *rdev)
3375 u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
3376 u32 cp_int_cntl1 = 0, cp_int_cntl2 = 0;
3377 u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
3378 u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
3379 u32 grbm_int_cntl = 0;
3380 u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
3381 u32 dma_cntl, dma_cntl1;
3383 if (!rdev->irq.installed) {
3384 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
3387 /* don't enable anything if the ih is disabled */
3388 if (!rdev->ih.enabled) {
3389 si_disable_interrupts(rdev);
3390 /* force the active interrupt state to all disabled */
3391 si_disable_interrupt_state(rdev);
3395 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
3396 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
3397 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
3398 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
3399 hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
3400 hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
3402 dma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
3403 dma_cntl1 = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
3405 /* enable CP interrupts on all rings */
3406 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
3407 DRM_DEBUG("si_irq_set: sw int gfx\n");
3408 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
3410 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
3411 DRM_DEBUG("si_irq_set: sw int cp1\n");
3412 cp_int_cntl1 |= TIME_STAMP_INT_ENABLE;
3414 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
3415 DRM_DEBUG("si_irq_set: sw int cp2\n");
3416 cp_int_cntl2 |= TIME_STAMP_INT_ENABLE;
3418 if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
3419 DRM_DEBUG("si_irq_set: sw int dma\n");
3420 dma_cntl |= TRAP_ENABLE;
3423 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
3424 DRM_DEBUG("si_irq_set: sw int dma1\n");
3425 dma_cntl1 |= TRAP_ENABLE;
3427 if (rdev->irq.crtc_vblank_int[0] ||
3428 atomic_read(&rdev->irq.pflip[0])) {
3429 DRM_DEBUG("si_irq_set: vblank 0\n");
3430 crtc1 |= VBLANK_INT_MASK;
3432 if (rdev->irq.crtc_vblank_int[1] ||
3433 atomic_read(&rdev->irq.pflip[1])) {
3434 DRM_DEBUG("si_irq_set: vblank 1\n");
3435 crtc2 |= VBLANK_INT_MASK;
3437 if (rdev->irq.crtc_vblank_int[2] ||
3438 atomic_read(&rdev->irq.pflip[2])) {
3439 DRM_DEBUG("si_irq_set: vblank 2\n");
3440 crtc3 |= VBLANK_INT_MASK;
3442 if (rdev->irq.crtc_vblank_int[3] ||
3443 atomic_read(&rdev->irq.pflip[3])) {
3444 DRM_DEBUG("si_irq_set: vblank 3\n");
3445 crtc4 |= VBLANK_INT_MASK;
3447 if (rdev->irq.crtc_vblank_int[4] ||
3448 atomic_read(&rdev->irq.pflip[4])) {
3449 DRM_DEBUG("si_irq_set: vblank 4\n");
3450 crtc5 |= VBLANK_INT_MASK;
3452 if (rdev->irq.crtc_vblank_int[5] ||
3453 atomic_read(&rdev->irq.pflip[5])) {
3454 DRM_DEBUG("si_irq_set: vblank 5\n");
3455 crtc6 |= VBLANK_INT_MASK;
3457 if (rdev->irq.hpd[0]) {
3458 DRM_DEBUG("si_irq_set: hpd 1\n");
3459 hpd1 |= DC_HPDx_INT_EN;
3461 if (rdev->irq.hpd[1]) {
3462 DRM_DEBUG("si_irq_set: hpd 2\n");
3463 hpd2 |= DC_HPDx_INT_EN;
3465 if (rdev->irq.hpd[2]) {
3466 DRM_DEBUG("si_irq_set: hpd 3\n");
3467 hpd3 |= DC_HPDx_INT_EN;
3469 if (rdev->irq.hpd[3]) {
3470 DRM_DEBUG("si_irq_set: hpd 4\n");
3471 hpd4 |= DC_HPDx_INT_EN;
3473 if (rdev->irq.hpd[4]) {
3474 DRM_DEBUG("si_irq_set: hpd 5\n");
3475 hpd5 |= DC_HPDx_INT_EN;
3477 if (rdev->irq.hpd[5]) {
3478 DRM_DEBUG("si_irq_set: hpd 6\n");
3479 hpd6 |= DC_HPDx_INT_EN;
3482 WREG32(CP_INT_CNTL_RING0, cp_int_cntl);
3483 WREG32(CP_INT_CNTL_RING1, cp_int_cntl1);
3484 WREG32(CP_INT_CNTL_RING2, cp_int_cntl2);
3486 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, dma_cntl);
3487 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, dma_cntl1);
3489 WREG32(GRBM_INT_CNTL, grbm_int_cntl);
3491 WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
3492 WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
3493 if (rdev->num_crtc >= 4) {
3494 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
3495 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
3497 if (rdev->num_crtc >= 6) {
3498 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
3499 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
3502 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, grph1);
3503 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, grph2);
3504 if (rdev->num_crtc >= 4) {
3505 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, grph3);
3506 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, grph4);
3508 if (rdev->num_crtc >= 6) {
3509 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, grph5);
3510 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, grph6);
3513 WREG32(DC_HPD1_INT_CONTROL, hpd1);
3514 WREG32(DC_HPD2_INT_CONTROL, hpd2);
3515 WREG32(DC_HPD3_INT_CONTROL, hpd3);
3516 WREG32(DC_HPD4_INT_CONTROL, hpd4);
3517 WREG32(DC_HPD5_INT_CONTROL, hpd5);
3518 WREG32(DC_HPD6_INT_CONTROL, hpd6);
3523 static inline void si_irq_ack(struct radeon_device *rdev)
3527 rdev->irq.stat_regs.evergreen.disp_int = RREG32(DISP_INTERRUPT_STATUS);
3528 rdev->irq.stat_regs.evergreen.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
3529 rdev->irq.stat_regs.evergreen.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
3530 rdev->irq.stat_regs.evergreen.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
3531 rdev->irq.stat_regs.evergreen.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
3532 rdev->irq.stat_regs.evergreen.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
3533 rdev->irq.stat_regs.evergreen.d1grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
3534 rdev->irq.stat_regs.evergreen.d2grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
3535 if (rdev->num_crtc >= 4) {
3536 rdev->irq.stat_regs.evergreen.d3grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
3537 rdev->irq.stat_regs.evergreen.d4grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
3539 if (rdev->num_crtc >= 6) {
3540 rdev->irq.stat_regs.evergreen.d5grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
3541 rdev->irq.stat_regs.evergreen.d6grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
3544 if (rdev->irq.stat_regs.evergreen.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
3545 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3546 if (rdev->irq.stat_regs.evergreen.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
3547 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3548 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT)
3549 WREG32(VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
3550 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT)
3551 WREG32(VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
3552 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
3553 WREG32(VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
3554 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT)
3555 WREG32(VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
3557 if (rdev->num_crtc >= 4) {
3558 if (rdev->irq.stat_regs.evergreen.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
3559 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3560 if (rdev->irq.stat_regs.evergreen.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
3561 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3562 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
3563 WREG32(VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
3564 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
3565 WREG32(VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
3566 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
3567 WREG32(VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
3568 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
3569 WREG32(VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
3572 if (rdev->num_crtc >= 6) {
3573 if (rdev->irq.stat_regs.evergreen.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
3574 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3575 if (rdev->irq.stat_regs.evergreen.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
3576 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3577 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
3578 WREG32(VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
3579 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
3580 WREG32(VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
3581 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
3582 WREG32(VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
3583 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
3584 WREG32(VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
3587 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
3588 tmp = RREG32(DC_HPD1_INT_CONTROL);
3589 tmp |= DC_HPDx_INT_ACK;
3590 WREG32(DC_HPD1_INT_CONTROL, tmp);
3592 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
3593 tmp = RREG32(DC_HPD2_INT_CONTROL);
3594 tmp |= DC_HPDx_INT_ACK;
3595 WREG32(DC_HPD2_INT_CONTROL, tmp);
3597 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
3598 tmp = RREG32(DC_HPD3_INT_CONTROL);
3599 tmp |= DC_HPDx_INT_ACK;
3600 WREG32(DC_HPD3_INT_CONTROL, tmp);
3602 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
3603 tmp = RREG32(DC_HPD4_INT_CONTROL);
3604 tmp |= DC_HPDx_INT_ACK;
3605 WREG32(DC_HPD4_INT_CONTROL, tmp);
3607 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
3608 tmp = RREG32(DC_HPD5_INT_CONTROL);
3609 tmp |= DC_HPDx_INT_ACK;
3610 WREG32(DC_HPD5_INT_CONTROL, tmp);
3612 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
3613 tmp = RREG32(DC_HPD5_INT_CONTROL);
3614 tmp |= DC_HPDx_INT_ACK;
3615 WREG32(DC_HPD6_INT_CONTROL, tmp);
3619 static void si_irq_disable(struct radeon_device *rdev)
3621 si_disable_interrupts(rdev);
3622 /* Wait and acknowledge irq */
3625 si_disable_interrupt_state(rdev);
3628 static void si_irq_suspend(struct radeon_device *rdev)
3630 si_irq_disable(rdev);
3634 static void si_irq_fini(struct radeon_device *rdev)
3636 si_irq_suspend(rdev);
3637 r600_ih_ring_fini(rdev);
3640 static inline u32 si_get_ih_wptr(struct radeon_device *rdev)
3644 if (rdev->wb.enabled)
3645 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
3647 wptr = RREG32(IH_RB_WPTR);
3649 if (wptr & RB_OVERFLOW) {
3650 /* When a ring buffer overflow happen start parsing interrupt
3651 * from the last not overwritten vector (wptr + 16). Hopefully
3652 * this should allow us to catchup.
3654 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
3655 wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
3656 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
3657 tmp = RREG32(IH_RB_CNTL);
3658 tmp |= IH_WPTR_OVERFLOW_CLEAR;
3659 WREG32(IH_RB_CNTL, tmp);
3661 return (wptr & rdev->ih.ptr_mask);
3665 * Each IV ring entry is 128 bits:
3666 * [7:0] - interrupt source id
3668 * [59:32] - interrupt source data
3669 * [63:60] - reserved
3672 * [127:80] - reserved
3674 int si_irq_process(struct radeon_device *rdev)
3678 u32 src_id, src_data, ring_id;
3680 bool queue_hotplug = false;
3682 if (!rdev->ih.enabled || rdev->shutdown)
3685 wptr = si_get_ih_wptr(rdev);
3688 /* is somebody else already processing irqs? */
3689 if (atomic_xchg(&rdev->ih.lock, 1))
3692 rptr = rdev->ih.rptr;
3693 DRM_DEBUG("si_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
3695 /* Order reading of wptr vs. reading of IH ring data */
3698 /* display interrupts */
3701 while (rptr != wptr) {
3702 /* wptr/rptr are in bytes! */
3703 ring_index = rptr / 4;
3704 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
3705 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
3706 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
3709 case 1: /* D1 vblank/vline */
3711 case 0: /* D1 vblank */
3712 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT) {
3713 if (rdev->irq.crtc_vblank_int[0]) {
3714 drm_handle_vblank(rdev->ddev, 0);
3715 rdev->pm.vblank_sync = true;
3716 wake_up(&rdev->irq.vblank_queue);
3718 if (atomic_read(&rdev->irq.pflip[0]))
3719 radeon_crtc_handle_flip(rdev, 0);
3720 rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
3721 DRM_DEBUG("IH: D1 vblank\n");
3724 case 1: /* D1 vline */
3725 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT) {
3726 rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VLINE_INTERRUPT;
3727 DRM_DEBUG("IH: D1 vline\n");
3731 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3735 case 2: /* D2 vblank/vline */
3737 case 0: /* D2 vblank */
3738 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
3739 if (rdev->irq.crtc_vblank_int[1]) {
3740 drm_handle_vblank(rdev->ddev, 1);
3741 rdev->pm.vblank_sync = true;
3742 wake_up(&rdev->irq.vblank_queue);
3744 if (atomic_read(&rdev->irq.pflip[1]))
3745 radeon_crtc_handle_flip(rdev, 1);
3746 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
3747 DRM_DEBUG("IH: D2 vblank\n");
3750 case 1: /* D2 vline */
3751 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
3752 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
3753 DRM_DEBUG("IH: D2 vline\n");
3757 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3761 case 3: /* D3 vblank/vline */
3763 case 0: /* D3 vblank */
3764 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
3765 if (rdev->irq.crtc_vblank_int[2]) {
3766 drm_handle_vblank(rdev->ddev, 2);
3767 rdev->pm.vblank_sync = true;
3768 wake_up(&rdev->irq.vblank_queue);
3770 if (atomic_read(&rdev->irq.pflip[2]))
3771 radeon_crtc_handle_flip(rdev, 2);
3772 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
3773 DRM_DEBUG("IH: D3 vblank\n");
3776 case 1: /* D3 vline */
3777 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
3778 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
3779 DRM_DEBUG("IH: D3 vline\n");
3783 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3787 case 4: /* D4 vblank/vline */
3789 case 0: /* D4 vblank */
3790 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
3791 if (rdev->irq.crtc_vblank_int[3]) {
3792 drm_handle_vblank(rdev->ddev, 3);
3793 rdev->pm.vblank_sync = true;
3794 wake_up(&rdev->irq.vblank_queue);
3796 if (atomic_read(&rdev->irq.pflip[3]))
3797 radeon_crtc_handle_flip(rdev, 3);
3798 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
3799 DRM_DEBUG("IH: D4 vblank\n");
3802 case 1: /* D4 vline */
3803 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
3804 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
3805 DRM_DEBUG("IH: D4 vline\n");
3809 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3813 case 5: /* D5 vblank/vline */
3815 case 0: /* D5 vblank */
3816 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
3817 if (rdev->irq.crtc_vblank_int[4]) {
3818 drm_handle_vblank(rdev->ddev, 4);
3819 rdev->pm.vblank_sync = true;
3820 wake_up(&rdev->irq.vblank_queue);
3822 if (atomic_read(&rdev->irq.pflip[4]))
3823 radeon_crtc_handle_flip(rdev, 4);
3824 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
3825 DRM_DEBUG("IH: D5 vblank\n");
3828 case 1: /* D5 vline */
3829 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
3830 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
3831 DRM_DEBUG("IH: D5 vline\n");
3835 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3839 case 6: /* D6 vblank/vline */
3841 case 0: /* D6 vblank */
3842 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
3843 if (rdev->irq.crtc_vblank_int[5]) {
3844 drm_handle_vblank(rdev->ddev, 5);
3845 rdev->pm.vblank_sync = true;
3846 wake_up(&rdev->irq.vblank_queue);
3848 if (atomic_read(&rdev->irq.pflip[5]))
3849 radeon_crtc_handle_flip(rdev, 5);
3850 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
3851 DRM_DEBUG("IH: D6 vblank\n");
3854 case 1: /* D6 vline */
3855 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
3856 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
3857 DRM_DEBUG("IH: D6 vline\n");
3861 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3865 case 42: /* HPD hotplug */
3868 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
3869 rdev->irq.stat_regs.evergreen.disp_int &= ~DC_HPD1_INTERRUPT;
3870 queue_hotplug = true;
3871 DRM_DEBUG("IH: HPD1\n");
3875 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
3876 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~DC_HPD2_INTERRUPT;
3877 queue_hotplug = true;
3878 DRM_DEBUG("IH: HPD2\n");
3882 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
3883 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
3884 queue_hotplug = true;
3885 DRM_DEBUG("IH: HPD3\n");
3889 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
3890 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
3891 queue_hotplug = true;
3892 DRM_DEBUG("IH: HPD4\n");
3896 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
3897 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
3898 queue_hotplug = true;
3899 DRM_DEBUG("IH: HPD5\n");
3903 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
3904 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
3905 queue_hotplug = true;
3906 DRM_DEBUG("IH: HPD6\n");
3910 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3916 dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
3917 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
3918 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
3919 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
3920 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
3921 /* reset addr and status */
3922 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
3924 case 176: /* RINGID0 CP_INT */
3925 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
3927 case 177: /* RINGID1 CP_INT */
3928 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
3930 case 178: /* RINGID2 CP_INT */
3931 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
3933 case 181: /* CP EOP event */
3934 DRM_DEBUG("IH: CP EOP\n");
3937 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
3940 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
3943 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
3947 case 224: /* DMA trap event */
3948 DRM_DEBUG("IH: DMA trap\n");
3949 radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
3951 case 233: /* GUI IDLE */
3952 DRM_DEBUG("IH: GUI idle\n");
3954 case 244: /* DMA trap event */
3955 DRM_DEBUG("IH: DMA1 trap\n");
3956 radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
3959 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3963 /* wptr/rptr are in bytes! */
3965 rptr &= rdev->ih.ptr_mask;
3968 schedule_work(&rdev->hotplug_work);
3969 rdev->ih.rptr = rptr;
3970 WREG32(IH_RB_RPTR, rdev->ih.rptr);
3971 atomic_set(&rdev->ih.lock, 0);
3973 /* make sure wptr hasn't changed while processing */
3974 wptr = si_get_ih_wptr(rdev);
3982 * si_copy_dma - copy pages using the DMA engine
3984 * @rdev: radeon_device pointer
3985 * @src_offset: src GPU address
3986 * @dst_offset: dst GPU address
3987 * @num_gpu_pages: number of GPU pages to xfer
3988 * @fence: radeon fence object
3990 * Copy GPU paging using the DMA engine (SI).
3991 * Used by the radeon ttm implementation to move pages if
3992 * registered as the asic copy callback.
3994 int si_copy_dma(struct radeon_device *rdev,
3995 uint64_t src_offset, uint64_t dst_offset,
3996 unsigned num_gpu_pages,
3997 struct radeon_fence **fence)
3999 struct radeon_semaphore *sem = NULL;
4000 int ring_index = rdev->asic->copy.dma_ring_index;
4001 struct radeon_ring *ring = &rdev->ring[ring_index];
4002 u32 size_in_bytes, cur_size_in_bytes;
4006 r = radeon_semaphore_create(rdev, &sem);
4008 DRM_ERROR("radeon: moving bo (%d).\n", r);
4012 size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
4013 num_loops = DIV_ROUND_UP(size_in_bytes, 0xfffff);
4014 r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
4016 DRM_ERROR("radeon: moving bo (%d).\n", r);
4017 radeon_semaphore_free(rdev, &sem, NULL);
4021 if (radeon_fence_need_sync(*fence, ring->idx)) {
4022 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
4024 radeon_fence_note_sync(*fence, ring->idx);
4026 radeon_semaphore_free(rdev, &sem, NULL);
4029 for (i = 0; i < num_loops; i++) {
4030 cur_size_in_bytes = size_in_bytes;
4031 if (cur_size_in_bytes > 0xFFFFF)
4032 cur_size_in_bytes = 0xFFFFF;
4033 size_in_bytes -= cur_size_in_bytes;
4034 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 1, 0, 0, cur_size_in_bytes));
4035 radeon_ring_write(ring, dst_offset & 0xffffffff);
4036 radeon_ring_write(ring, src_offset & 0xffffffff);
4037 radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
4038 radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
4039 src_offset += cur_size_in_bytes;
4040 dst_offset += cur_size_in_bytes;
4043 r = radeon_fence_emit(rdev, fence, ring->idx);
4045 radeon_ring_unlock_undo(rdev, ring);
4049 radeon_ring_unlock_commit(rdev, ring);
4050 radeon_semaphore_free(rdev, &sem, *fence);
4056 * startup/shutdown callbacks
4058 static int si_startup(struct radeon_device *rdev)
4060 struct radeon_ring *ring;
4063 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
4064 !rdev->rlc_fw || !rdev->mc_fw) {
4065 r = si_init_microcode(rdev);
4067 DRM_ERROR("Failed to load firmware!\n");
4072 r = si_mc_load_microcode(rdev);
4074 DRM_ERROR("Failed to load MC firmware!\n");
4078 r = r600_vram_scratch_init(rdev);
4082 si_mc_program(rdev);
4083 r = si_pcie_gart_enable(rdev);
4089 r = evergreen_blit_init(rdev);
4091 r600_blit_fini(rdev);
4092 rdev->asic->copy = NULL;
4093 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
4096 /* allocate rlc buffers */
4097 r = si_rlc_init(rdev);
4099 DRM_ERROR("Failed to init rlc BOs!\n");
4103 /* allocate wb buffer */
4104 r = radeon_wb_init(rdev);
4108 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
4110 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4114 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4116 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4120 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4122 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4126 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
4128 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
4132 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
4134 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
4139 r = si_irq_init(rdev);
4141 DRM_ERROR("radeon: IH init failed (%d).\n", r);
4142 radeon_irq_kms_fini(rdev);
4147 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4148 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
4149 CP_RB0_RPTR, CP_RB0_WPTR,
4150 0, 0xfffff, RADEON_CP_PACKET2);
4154 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
4155 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET,
4156 CP_RB1_RPTR, CP_RB1_WPTR,
4157 0, 0xfffff, RADEON_CP_PACKET2);
4161 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
4162 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET,
4163 CP_RB2_RPTR, CP_RB2_WPTR,
4164 0, 0xfffff, RADEON_CP_PACKET2);
4168 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4169 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
4170 DMA_RB_RPTR + DMA0_REGISTER_OFFSET,
4171 DMA_RB_WPTR + DMA0_REGISTER_OFFSET,
4172 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
4176 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4177 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
4178 DMA_RB_RPTR + DMA1_REGISTER_OFFSET,
4179 DMA_RB_WPTR + DMA1_REGISTER_OFFSET,
4180 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
4184 r = si_cp_load_microcode(rdev);
4187 r = si_cp_resume(rdev);
4191 r = cayman_dma_resume(rdev);
4195 r = radeon_ib_pool_init(rdev);
4197 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
4201 r = radeon_vm_manager_init(rdev);
4203 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
4210 int si_resume(struct radeon_device *rdev)
4214 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
4215 * posting will perform necessary task to bring back GPU into good
4219 atom_asic_init(rdev->mode_info.atom_context);
4221 rdev->accel_working = true;
4222 r = si_startup(rdev);
4224 DRM_ERROR("si startup failed on resume\n");
4225 rdev->accel_working = false;
4233 int si_suspend(struct radeon_device *rdev)
4235 si_cp_enable(rdev, false);
4236 cayman_dma_stop(rdev);
4237 si_irq_suspend(rdev);
4238 radeon_wb_disable(rdev);
4239 si_pcie_gart_disable(rdev);
4243 /* Plan is to move initialization in that function and use
4244 * helper function so that radeon_device_init pretty much
4245 * do nothing more than calling asic specific function. This
4246 * should also allow to remove a bunch of callback function
4249 int si_init(struct radeon_device *rdev)
4251 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4255 if (!radeon_get_bios(rdev)) {
4256 if (ASIC_IS_AVIVO(rdev))
4259 /* Must be an ATOMBIOS */
4260 if (!rdev->is_atom_bios) {
4261 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
4264 r = radeon_atombios_init(rdev);
4268 /* Post card if necessary */
4269 if (!radeon_card_posted(rdev)) {
4271 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
4274 DRM_INFO("GPU not posted. posting now...\n");
4275 atom_asic_init(rdev->mode_info.atom_context);
4277 /* Initialize scratch registers */
4278 si_scratch_init(rdev);
4279 /* Initialize surface registers */
4280 radeon_surface_init(rdev);
4281 /* Initialize clocks */
4282 radeon_get_clock_info(rdev->ddev);
4285 r = radeon_fence_driver_init(rdev);
4289 /* initialize memory controller */
4290 r = si_mc_init(rdev);
4293 /* Memory manager */
4294 r = radeon_bo_init(rdev);
4298 r = radeon_irq_kms_init(rdev);
4302 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4303 ring->ring_obj = NULL;
4304 r600_ring_init(rdev, ring, 1024 * 1024);
4306 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
4307 ring->ring_obj = NULL;
4308 r600_ring_init(rdev, ring, 1024 * 1024);
4310 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
4311 ring->ring_obj = NULL;
4312 r600_ring_init(rdev, ring, 1024 * 1024);
4314 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4315 ring->ring_obj = NULL;
4316 r600_ring_init(rdev, ring, 64 * 1024);
4318 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4319 ring->ring_obj = NULL;
4320 r600_ring_init(rdev, ring, 64 * 1024);
4322 rdev->ih.ring_obj = NULL;
4323 r600_ih_ring_init(rdev, 64 * 1024);
4325 r = r600_pcie_gart_init(rdev);
4329 rdev->accel_working = true;
4330 r = si_startup(rdev);
4332 dev_err(rdev->dev, "disabling GPU acceleration\n");
4334 cayman_dma_fini(rdev);
4337 radeon_wb_fini(rdev);
4338 radeon_ib_pool_fini(rdev);
4339 radeon_vm_manager_fini(rdev);
4340 radeon_irq_kms_fini(rdev);
4341 si_pcie_gart_fini(rdev);
4342 rdev->accel_working = false;
4345 /* Don't start up if the MC ucode is missing.
4346 * The default clocks and voltages before the MC ucode
4347 * is loaded are not suffient for advanced operations.
4350 DRM_ERROR("radeon: MC ucode required for NI+.\n");
4357 void si_fini(struct radeon_device *rdev)
4360 r600_blit_fini(rdev);
4363 cayman_dma_fini(rdev);
4366 radeon_wb_fini(rdev);
4367 radeon_vm_manager_fini(rdev);
4368 radeon_ib_pool_fini(rdev);
4369 radeon_irq_kms_fini(rdev);
4370 si_pcie_gart_fini(rdev);
4371 r600_vram_scratch_fini(rdev);
4372 radeon_gem_fini(rdev);
4373 radeon_fence_driver_fini(rdev);
4374 radeon_bo_fini(rdev);
4375 radeon_atombios_fini(rdev);
4381 * si_get_gpu_clock - return GPU clock counter snapshot
4383 * @rdev: radeon_device pointer
4385 * Fetches a GPU clock counter snapshot (SI).
4386 * Returns the 64 bit clock counter snapshot.
4388 uint64_t si_get_gpu_clock(struct radeon_device *rdev)
4392 mutex_lock(&rdev->gpu_clock_mutex);
4393 WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4394 clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
4395 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4396 mutex_unlock(&rdev->gpu_clock_mutex);