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
41 #define OLAND_MC_UCODE_SIZE 7863
43 MODULE_FIRMWARE("radeon/TAHITI_pfp.bin");
44 MODULE_FIRMWARE("radeon/TAHITI_me.bin");
45 MODULE_FIRMWARE("radeon/TAHITI_ce.bin");
46 MODULE_FIRMWARE("radeon/TAHITI_mc.bin");
47 MODULE_FIRMWARE("radeon/TAHITI_rlc.bin");
48 MODULE_FIRMWARE("radeon/PITCAIRN_pfp.bin");
49 MODULE_FIRMWARE("radeon/PITCAIRN_me.bin");
50 MODULE_FIRMWARE("radeon/PITCAIRN_ce.bin");
51 MODULE_FIRMWARE("radeon/PITCAIRN_mc.bin");
52 MODULE_FIRMWARE("radeon/PITCAIRN_rlc.bin");
53 MODULE_FIRMWARE("radeon/VERDE_pfp.bin");
54 MODULE_FIRMWARE("radeon/VERDE_me.bin");
55 MODULE_FIRMWARE("radeon/VERDE_ce.bin");
56 MODULE_FIRMWARE("radeon/VERDE_mc.bin");
57 MODULE_FIRMWARE("radeon/VERDE_rlc.bin");
58 MODULE_FIRMWARE("radeon/OLAND_pfp.bin");
59 MODULE_FIRMWARE("radeon/OLAND_me.bin");
60 MODULE_FIRMWARE("radeon/OLAND_ce.bin");
61 MODULE_FIRMWARE("radeon/OLAND_mc.bin");
62 MODULE_FIRMWARE("radeon/OLAND_rlc.bin");
64 extern int r600_ih_ring_alloc(struct radeon_device *rdev);
65 extern void r600_ih_ring_fini(struct radeon_device *rdev);
66 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
67 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
68 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
69 extern u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev);
70 extern void evergreen_print_gpu_status_regs(struct radeon_device *rdev);
71 extern bool evergreen_is_display_hung(struct radeon_device *rdev);
73 #define PCIE_BUS_CLK 10000
74 #define TCLK (PCIE_BUS_CLK / 10)
77 * si_get_xclk - get the xclk
79 * @rdev: radeon_device pointer
81 * Returns the reference clock used by the gfx engine
84 u32 si_get_xclk(struct radeon_device *rdev)
86 u32 reference_clock = rdev->clock.spll.reference_freq;
89 tmp = RREG32(CG_CLKPIN_CNTL_2);
90 if (tmp & MUX_TCLK_TO_XCLK)
93 tmp = RREG32(CG_CLKPIN_CNTL);
94 if (tmp & XTALIN_DIVIDE)
95 return reference_clock / 4;
97 return reference_clock;
100 /* get temperature in millidegrees */
101 int si_get_temp(struct radeon_device *rdev)
106 temp = (RREG32(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
112 actual_temp = temp & 0x1ff;
114 actual_temp = (actual_temp * 1000);
119 #define TAHITI_IO_MC_REGS_SIZE 36
121 static const u32 tahiti_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
122 {0x0000006f, 0x03044000},
123 {0x00000070, 0x0480c018},
124 {0x00000071, 0x00000040},
125 {0x00000072, 0x01000000},
126 {0x00000074, 0x000000ff},
127 {0x00000075, 0x00143400},
128 {0x00000076, 0x08ec0800},
129 {0x00000077, 0x040000cc},
130 {0x00000079, 0x00000000},
131 {0x0000007a, 0x21000409},
132 {0x0000007c, 0x00000000},
133 {0x0000007d, 0xe8000000},
134 {0x0000007e, 0x044408a8},
135 {0x0000007f, 0x00000003},
136 {0x00000080, 0x00000000},
137 {0x00000081, 0x01000000},
138 {0x00000082, 0x02000000},
139 {0x00000083, 0x00000000},
140 {0x00000084, 0xe3f3e4f4},
141 {0x00000085, 0x00052024},
142 {0x00000087, 0x00000000},
143 {0x00000088, 0x66036603},
144 {0x00000089, 0x01000000},
145 {0x0000008b, 0x1c0a0000},
146 {0x0000008c, 0xff010000},
147 {0x0000008e, 0xffffefff},
148 {0x0000008f, 0xfff3efff},
149 {0x00000090, 0xfff3efbf},
150 {0x00000094, 0x00101101},
151 {0x00000095, 0x00000fff},
152 {0x00000096, 0x00116fff},
153 {0x00000097, 0x60010000},
154 {0x00000098, 0x10010000},
155 {0x00000099, 0x00006000},
156 {0x0000009a, 0x00001000},
157 {0x0000009f, 0x00a77400}
160 static const u32 pitcairn_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
161 {0x0000006f, 0x03044000},
162 {0x00000070, 0x0480c018},
163 {0x00000071, 0x00000040},
164 {0x00000072, 0x01000000},
165 {0x00000074, 0x000000ff},
166 {0x00000075, 0x00143400},
167 {0x00000076, 0x08ec0800},
168 {0x00000077, 0x040000cc},
169 {0x00000079, 0x00000000},
170 {0x0000007a, 0x21000409},
171 {0x0000007c, 0x00000000},
172 {0x0000007d, 0xe8000000},
173 {0x0000007e, 0x044408a8},
174 {0x0000007f, 0x00000003},
175 {0x00000080, 0x00000000},
176 {0x00000081, 0x01000000},
177 {0x00000082, 0x02000000},
178 {0x00000083, 0x00000000},
179 {0x00000084, 0xe3f3e4f4},
180 {0x00000085, 0x00052024},
181 {0x00000087, 0x00000000},
182 {0x00000088, 0x66036603},
183 {0x00000089, 0x01000000},
184 {0x0000008b, 0x1c0a0000},
185 {0x0000008c, 0xff010000},
186 {0x0000008e, 0xffffefff},
187 {0x0000008f, 0xfff3efff},
188 {0x00000090, 0xfff3efbf},
189 {0x00000094, 0x00101101},
190 {0x00000095, 0x00000fff},
191 {0x00000096, 0x00116fff},
192 {0x00000097, 0x60010000},
193 {0x00000098, 0x10010000},
194 {0x00000099, 0x00006000},
195 {0x0000009a, 0x00001000},
196 {0x0000009f, 0x00a47400}
199 static const u32 verde_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
200 {0x0000006f, 0x03044000},
201 {0x00000070, 0x0480c018},
202 {0x00000071, 0x00000040},
203 {0x00000072, 0x01000000},
204 {0x00000074, 0x000000ff},
205 {0x00000075, 0x00143400},
206 {0x00000076, 0x08ec0800},
207 {0x00000077, 0x040000cc},
208 {0x00000079, 0x00000000},
209 {0x0000007a, 0x21000409},
210 {0x0000007c, 0x00000000},
211 {0x0000007d, 0xe8000000},
212 {0x0000007e, 0x044408a8},
213 {0x0000007f, 0x00000003},
214 {0x00000080, 0x00000000},
215 {0x00000081, 0x01000000},
216 {0x00000082, 0x02000000},
217 {0x00000083, 0x00000000},
218 {0x00000084, 0xe3f3e4f4},
219 {0x00000085, 0x00052024},
220 {0x00000087, 0x00000000},
221 {0x00000088, 0x66036603},
222 {0x00000089, 0x01000000},
223 {0x0000008b, 0x1c0a0000},
224 {0x0000008c, 0xff010000},
225 {0x0000008e, 0xffffefff},
226 {0x0000008f, 0xfff3efff},
227 {0x00000090, 0xfff3efbf},
228 {0x00000094, 0x00101101},
229 {0x00000095, 0x00000fff},
230 {0x00000096, 0x00116fff},
231 {0x00000097, 0x60010000},
232 {0x00000098, 0x10010000},
233 {0x00000099, 0x00006000},
234 {0x0000009a, 0x00001000},
235 {0x0000009f, 0x00a37400}
238 static const u32 oland_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
239 {0x0000006f, 0x03044000},
240 {0x00000070, 0x0480c018},
241 {0x00000071, 0x00000040},
242 {0x00000072, 0x01000000},
243 {0x00000074, 0x000000ff},
244 {0x00000075, 0x00143400},
245 {0x00000076, 0x08ec0800},
246 {0x00000077, 0x040000cc},
247 {0x00000079, 0x00000000},
248 {0x0000007a, 0x21000409},
249 {0x0000007c, 0x00000000},
250 {0x0000007d, 0xe8000000},
251 {0x0000007e, 0x044408a8},
252 {0x0000007f, 0x00000003},
253 {0x00000080, 0x00000000},
254 {0x00000081, 0x01000000},
255 {0x00000082, 0x02000000},
256 {0x00000083, 0x00000000},
257 {0x00000084, 0xe3f3e4f4},
258 {0x00000085, 0x00052024},
259 {0x00000087, 0x00000000},
260 {0x00000088, 0x66036603},
261 {0x00000089, 0x01000000},
262 {0x0000008b, 0x1c0a0000},
263 {0x0000008c, 0xff010000},
264 {0x0000008e, 0xffffefff},
265 {0x0000008f, 0xfff3efff},
266 {0x00000090, 0xfff3efbf},
267 {0x00000094, 0x00101101},
268 {0x00000095, 0x00000fff},
269 {0x00000096, 0x00116fff},
270 {0x00000097, 0x60010000},
271 {0x00000098, 0x10010000},
272 {0x00000099, 0x00006000},
273 {0x0000009a, 0x00001000},
274 {0x0000009f, 0x00a17730}
278 static int si_mc_load_microcode(struct radeon_device *rdev)
280 const __be32 *fw_data;
281 u32 running, blackout = 0;
283 int i, ucode_size, regs_size;
288 switch (rdev->family) {
290 io_mc_regs = (u32 *)&tahiti_io_mc_regs;
291 ucode_size = SI_MC_UCODE_SIZE;
292 regs_size = TAHITI_IO_MC_REGS_SIZE;
295 io_mc_regs = (u32 *)&pitcairn_io_mc_regs;
296 ucode_size = SI_MC_UCODE_SIZE;
297 regs_size = TAHITI_IO_MC_REGS_SIZE;
301 io_mc_regs = (u32 *)&verde_io_mc_regs;
302 ucode_size = SI_MC_UCODE_SIZE;
303 regs_size = TAHITI_IO_MC_REGS_SIZE;
306 io_mc_regs = (u32 *)&oland_io_mc_regs;
307 ucode_size = OLAND_MC_UCODE_SIZE;
308 regs_size = TAHITI_IO_MC_REGS_SIZE;
312 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
316 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
317 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
320 /* reset the engine and set to writable */
321 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
322 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
324 /* load mc io regs */
325 for (i = 0; i < regs_size; i++) {
326 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
327 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
329 /* load the MC ucode */
330 fw_data = (const __be32 *)rdev->mc_fw->data;
331 for (i = 0; i < ucode_size; i++)
332 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
334 /* put the engine back into the active state */
335 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
336 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
337 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
339 /* wait for training to complete */
340 for (i = 0; i < rdev->usec_timeout; i++) {
341 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
345 for (i = 0; i < rdev->usec_timeout; i++) {
346 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
352 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
358 static int si_init_microcode(struct radeon_device *rdev)
360 struct platform_device *pdev;
361 const char *chip_name;
362 const char *rlc_chip_name;
363 size_t pfp_req_size, me_req_size, ce_req_size, rlc_req_size, mc_req_size;
369 pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
372 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
376 switch (rdev->family) {
378 chip_name = "TAHITI";
379 rlc_chip_name = "TAHITI";
380 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
381 me_req_size = SI_PM4_UCODE_SIZE * 4;
382 ce_req_size = SI_CE_UCODE_SIZE * 4;
383 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
384 mc_req_size = SI_MC_UCODE_SIZE * 4;
387 chip_name = "PITCAIRN";
388 rlc_chip_name = "PITCAIRN";
389 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
390 me_req_size = SI_PM4_UCODE_SIZE * 4;
391 ce_req_size = SI_CE_UCODE_SIZE * 4;
392 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
393 mc_req_size = SI_MC_UCODE_SIZE * 4;
397 rlc_chip_name = "VERDE";
398 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
399 me_req_size = SI_PM4_UCODE_SIZE * 4;
400 ce_req_size = SI_CE_UCODE_SIZE * 4;
401 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
402 mc_req_size = SI_MC_UCODE_SIZE * 4;
406 rlc_chip_name = "OLAND";
407 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
408 me_req_size = SI_PM4_UCODE_SIZE * 4;
409 ce_req_size = SI_CE_UCODE_SIZE * 4;
410 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
411 mc_req_size = OLAND_MC_UCODE_SIZE * 4;
416 DRM_INFO("Loading %s Microcode\n", chip_name);
418 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
419 err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
422 if (rdev->pfp_fw->size != pfp_req_size) {
424 "si_cp: Bogus length %zu in firmware \"%s\"\n",
425 rdev->pfp_fw->size, fw_name);
430 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
431 err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
434 if (rdev->me_fw->size != me_req_size) {
436 "si_cp: Bogus length %zu in firmware \"%s\"\n",
437 rdev->me_fw->size, fw_name);
441 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
442 err = request_firmware(&rdev->ce_fw, fw_name, &pdev->dev);
445 if (rdev->ce_fw->size != ce_req_size) {
447 "si_cp: Bogus length %zu in firmware \"%s\"\n",
448 rdev->ce_fw->size, fw_name);
452 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
453 err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
456 if (rdev->rlc_fw->size != rlc_req_size) {
458 "si_rlc: Bogus length %zu in firmware \"%s\"\n",
459 rdev->rlc_fw->size, fw_name);
463 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
464 err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev);
467 if (rdev->mc_fw->size != mc_req_size) {
469 "si_mc: Bogus length %zu in firmware \"%s\"\n",
470 rdev->mc_fw->size, fw_name);
475 platform_device_unregister(pdev);
480 "si_cp: Failed to load firmware \"%s\"\n",
482 release_firmware(rdev->pfp_fw);
484 release_firmware(rdev->me_fw);
486 release_firmware(rdev->ce_fw);
488 release_firmware(rdev->rlc_fw);
490 release_firmware(rdev->mc_fw);
496 /* watermark setup */
497 static u32 dce6_line_buffer_adjust(struct radeon_device *rdev,
498 struct radeon_crtc *radeon_crtc,
499 struct drm_display_mode *mode,
500 struct drm_display_mode *other_mode)
505 * There are 3 line buffers, each one shared by 2 display controllers.
506 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
507 * the display controllers. The paritioning is done via one of four
508 * preset allocations specified in bits 21:20:
510 * 2 - whole lb, other crtc must be disabled
512 /* this can get tricky if we have two large displays on a paired group
513 * of crtcs. Ideally for multiple large displays we'd assign them to
514 * non-linked crtcs for maximum line buffer allocation.
516 if (radeon_crtc->base.enabled && mode) {
524 WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset,
525 DC_LB_MEMORY_CONFIG(tmp));
527 if (radeon_crtc->base.enabled && mode) {
537 /* controller not enabled, so no lb used */
541 static u32 si_get_number_of_dram_channels(struct radeon_device *rdev)
543 u32 tmp = RREG32(MC_SHARED_CHMAP);
545 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
568 struct dce6_wm_params {
569 u32 dram_channels; /* number of dram channels */
570 u32 yclk; /* bandwidth per dram data pin in kHz */
571 u32 sclk; /* engine clock in kHz */
572 u32 disp_clk; /* display clock in kHz */
573 u32 src_width; /* viewport width */
574 u32 active_time; /* active display time in ns */
575 u32 blank_time; /* blank time in ns */
576 bool interlaced; /* mode is interlaced */
577 fixed20_12 vsc; /* vertical scale ratio */
578 u32 num_heads; /* number of active crtcs */
579 u32 bytes_per_pixel; /* bytes per pixel display + overlay */
580 u32 lb_size; /* line buffer allocated to pipe */
581 u32 vtaps; /* vertical scaler taps */
584 static u32 dce6_dram_bandwidth(struct dce6_wm_params *wm)
586 /* Calculate raw DRAM Bandwidth */
587 fixed20_12 dram_efficiency; /* 0.7 */
588 fixed20_12 yclk, dram_channels, bandwidth;
591 a.full = dfixed_const(1000);
592 yclk.full = dfixed_const(wm->yclk);
593 yclk.full = dfixed_div(yclk, a);
594 dram_channels.full = dfixed_const(wm->dram_channels * 4);
595 a.full = dfixed_const(10);
596 dram_efficiency.full = dfixed_const(7);
597 dram_efficiency.full = dfixed_div(dram_efficiency, a);
598 bandwidth.full = dfixed_mul(dram_channels, yclk);
599 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
601 return dfixed_trunc(bandwidth);
604 static u32 dce6_dram_bandwidth_for_display(struct dce6_wm_params *wm)
606 /* Calculate DRAM Bandwidth and the part allocated to display. */
607 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
608 fixed20_12 yclk, dram_channels, bandwidth;
611 a.full = dfixed_const(1000);
612 yclk.full = dfixed_const(wm->yclk);
613 yclk.full = dfixed_div(yclk, a);
614 dram_channels.full = dfixed_const(wm->dram_channels * 4);
615 a.full = dfixed_const(10);
616 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
617 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
618 bandwidth.full = dfixed_mul(dram_channels, yclk);
619 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
621 return dfixed_trunc(bandwidth);
624 static u32 dce6_data_return_bandwidth(struct dce6_wm_params *wm)
626 /* Calculate the display Data return Bandwidth */
627 fixed20_12 return_efficiency; /* 0.8 */
628 fixed20_12 sclk, bandwidth;
631 a.full = dfixed_const(1000);
632 sclk.full = dfixed_const(wm->sclk);
633 sclk.full = dfixed_div(sclk, a);
634 a.full = dfixed_const(10);
635 return_efficiency.full = dfixed_const(8);
636 return_efficiency.full = dfixed_div(return_efficiency, a);
637 a.full = dfixed_const(32);
638 bandwidth.full = dfixed_mul(a, sclk);
639 bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
641 return dfixed_trunc(bandwidth);
644 static u32 dce6_get_dmif_bytes_per_request(struct dce6_wm_params *wm)
649 static u32 dce6_dmif_request_bandwidth(struct dce6_wm_params *wm)
651 /* Calculate the DMIF Request Bandwidth */
652 fixed20_12 disp_clk_request_efficiency; /* 0.8 */
653 fixed20_12 disp_clk, sclk, bandwidth;
654 fixed20_12 a, b1, b2;
657 a.full = dfixed_const(1000);
658 disp_clk.full = dfixed_const(wm->disp_clk);
659 disp_clk.full = dfixed_div(disp_clk, a);
660 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm) / 2);
661 b1.full = dfixed_mul(a, disp_clk);
663 a.full = dfixed_const(1000);
664 sclk.full = dfixed_const(wm->sclk);
665 sclk.full = dfixed_div(sclk, a);
666 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm));
667 b2.full = dfixed_mul(a, sclk);
669 a.full = dfixed_const(10);
670 disp_clk_request_efficiency.full = dfixed_const(8);
671 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
673 min_bandwidth = min(dfixed_trunc(b1), dfixed_trunc(b2));
675 a.full = dfixed_const(min_bandwidth);
676 bandwidth.full = dfixed_mul(a, disp_clk_request_efficiency);
678 return dfixed_trunc(bandwidth);
681 static u32 dce6_available_bandwidth(struct dce6_wm_params *wm)
683 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
684 u32 dram_bandwidth = dce6_dram_bandwidth(wm);
685 u32 data_return_bandwidth = dce6_data_return_bandwidth(wm);
686 u32 dmif_req_bandwidth = dce6_dmif_request_bandwidth(wm);
688 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
691 static u32 dce6_average_bandwidth(struct dce6_wm_params *wm)
693 /* Calculate the display mode Average Bandwidth
694 * DisplayMode should contain the source and destination dimensions,
698 fixed20_12 line_time;
699 fixed20_12 src_width;
700 fixed20_12 bandwidth;
703 a.full = dfixed_const(1000);
704 line_time.full = dfixed_const(wm->active_time + wm->blank_time);
705 line_time.full = dfixed_div(line_time, a);
706 bpp.full = dfixed_const(wm->bytes_per_pixel);
707 src_width.full = dfixed_const(wm->src_width);
708 bandwidth.full = dfixed_mul(src_width, bpp);
709 bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
710 bandwidth.full = dfixed_div(bandwidth, line_time);
712 return dfixed_trunc(bandwidth);
715 static u32 dce6_latency_watermark(struct dce6_wm_params *wm)
717 /* First calcualte the latency in ns */
718 u32 mc_latency = 2000; /* 2000 ns. */
719 u32 available_bandwidth = dce6_available_bandwidth(wm);
720 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
721 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
722 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
723 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
724 (wm->num_heads * cursor_line_pair_return_time);
725 u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
726 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
727 u32 tmp, dmif_size = 12288;
730 if (wm->num_heads == 0)
733 a.full = dfixed_const(2);
734 b.full = dfixed_const(1);
735 if ((wm->vsc.full > a.full) ||
736 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
738 ((wm->vsc.full >= a.full) && wm->interlaced))
739 max_src_lines_per_dst_line = 4;
741 max_src_lines_per_dst_line = 2;
743 a.full = dfixed_const(available_bandwidth);
744 b.full = dfixed_const(wm->num_heads);
745 a.full = dfixed_div(a, b);
747 b.full = dfixed_const(mc_latency + 512);
748 c.full = dfixed_const(wm->disp_clk);
749 b.full = dfixed_div(b, c);
751 c.full = dfixed_const(dmif_size);
752 b.full = dfixed_div(c, b);
754 tmp = min(dfixed_trunc(a), dfixed_trunc(b));
756 b.full = dfixed_const(1000);
757 c.full = dfixed_const(wm->disp_clk);
758 b.full = dfixed_div(c, b);
759 c.full = dfixed_const(wm->bytes_per_pixel);
760 b.full = dfixed_mul(b, c);
762 lb_fill_bw = min(tmp, dfixed_trunc(b));
764 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
765 b.full = dfixed_const(1000);
766 c.full = dfixed_const(lb_fill_bw);
767 b.full = dfixed_div(c, b);
768 a.full = dfixed_div(a, b);
769 line_fill_time = dfixed_trunc(a);
771 if (line_fill_time < wm->active_time)
774 return latency + (line_fill_time - wm->active_time);
778 static bool dce6_average_bandwidth_vs_dram_bandwidth_for_display(struct dce6_wm_params *wm)
780 if (dce6_average_bandwidth(wm) <=
781 (dce6_dram_bandwidth_for_display(wm) / wm->num_heads))
787 static bool dce6_average_bandwidth_vs_available_bandwidth(struct dce6_wm_params *wm)
789 if (dce6_average_bandwidth(wm) <=
790 (dce6_available_bandwidth(wm) / wm->num_heads))
796 static bool dce6_check_latency_hiding(struct dce6_wm_params *wm)
798 u32 lb_partitions = wm->lb_size / wm->src_width;
799 u32 line_time = wm->active_time + wm->blank_time;
800 u32 latency_tolerant_lines;
804 a.full = dfixed_const(1);
805 if (wm->vsc.full > a.full)
806 latency_tolerant_lines = 1;
808 if (lb_partitions <= (wm->vtaps + 1))
809 latency_tolerant_lines = 1;
811 latency_tolerant_lines = 2;
814 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
816 if (dce6_latency_watermark(wm) <= latency_hiding)
822 static void dce6_program_watermarks(struct radeon_device *rdev,
823 struct radeon_crtc *radeon_crtc,
824 u32 lb_size, u32 num_heads)
826 struct drm_display_mode *mode = &radeon_crtc->base.mode;
827 struct dce6_wm_params wm;
830 u32 latency_watermark_a = 0, latency_watermark_b = 0;
831 u32 priority_a_mark = 0, priority_b_mark = 0;
832 u32 priority_a_cnt = PRIORITY_OFF;
833 u32 priority_b_cnt = PRIORITY_OFF;
834 u32 tmp, arb_control3;
837 if (radeon_crtc->base.enabled && num_heads && mode) {
838 pixel_period = 1000000 / (u32)mode->clock;
839 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
843 wm.yclk = rdev->pm.current_mclk * 10;
844 wm.sclk = rdev->pm.current_sclk * 10;
845 wm.disp_clk = mode->clock;
846 wm.src_width = mode->crtc_hdisplay;
847 wm.active_time = mode->crtc_hdisplay * pixel_period;
848 wm.blank_time = line_time - wm.active_time;
849 wm.interlaced = false;
850 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
851 wm.interlaced = true;
852 wm.vsc = radeon_crtc->vsc;
854 if (radeon_crtc->rmx_type != RMX_OFF)
856 wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
857 wm.lb_size = lb_size;
858 if (rdev->family == CHIP_ARUBA)
859 wm.dram_channels = evergreen_get_number_of_dram_channels(rdev);
861 wm.dram_channels = si_get_number_of_dram_channels(rdev);
862 wm.num_heads = num_heads;
864 /* set for high clocks */
865 latency_watermark_a = min(dce6_latency_watermark(&wm), (u32)65535);
866 /* set for low clocks */
867 /* wm.yclk = low clk; wm.sclk = low clk */
868 latency_watermark_b = min(dce6_latency_watermark(&wm), (u32)65535);
870 /* possibly force display priority to high */
871 /* should really do this at mode validation time... */
872 if (!dce6_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
873 !dce6_average_bandwidth_vs_available_bandwidth(&wm) ||
874 !dce6_check_latency_hiding(&wm) ||
875 (rdev->disp_priority == 2)) {
876 DRM_DEBUG_KMS("force priority to high\n");
877 priority_a_cnt |= PRIORITY_ALWAYS_ON;
878 priority_b_cnt |= PRIORITY_ALWAYS_ON;
881 a.full = dfixed_const(1000);
882 b.full = dfixed_const(mode->clock);
883 b.full = dfixed_div(b, a);
884 c.full = dfixed_const(latency_watermark_a);
885 c.full = dfixed_mul(c, b);
886 c.full = dfixed_mul(c, radeon_crtc->hsc);
887 c.full = dfixed_div(c, a);
888 a.full = dfixed_const(16);
889 c.full = dfixed_div(c, a);
890 priority_a_mark = dfixed_trunc(c);
891 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
893 a.full = dfixed_const(1000);
894 b.full = dfixed_const(mode->clock);
895 b.full = dfixed_div(b, a);
896 c.full = dfixed_const(latency_watermark_b);
897 c.full = dfixed_mul(c, b);
898 c.full = dfixed_mul(c, radeon_crtc->hsc);
899 c.full = dfixed_div(c, a);
900 a.full = dfixed_const(16);
901 c.full = dfixed_div(c, a);
902 priority_b_mark = dfixed_trunc(c);
903 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
907 arb_control3 = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
909 tmp &= ~LATENCY_WATERMARK_MASK(3);
910 tmp |= LATENCY_WATERMARK_MASK(1);
911 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
912 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
913 (LATENCY_LOW_WATERMARK(latency_watermark_a) |
914 LATENCY_HIGH_WATERMARK(line_time)));
916 tmp = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
917 tmp &= ~LATENCY_WATERMARK_MASK(3);
918 tmp |= LATENCY_WATERMARK_MASK(2);
919 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
920 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
921 (LATENCY_LOW_WATERMARK(latency_watermark_b) |
922 LATENCY_HIGH_WATERMARK(line_time)));
923 /* restore original selection */
924 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, arb_control3);
926 /* write the priority marks */
927 WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
928 WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
932 void dce6_bandwidth_update(struct radeon_device *rdev)
934 struct drm_display_mode *mode0 = NULL;
935 struct drm_display_mode *mode1 = NULL;
936 u32 num_heads = 0, lb_size;
939 radeon_update_display_priority(rdev);
941 for (i = 0; i < rdev->num_crtc; i++) {
942 if (rdev->mode_info.crtcs[i]->base.enabled)
945 for (i = 0; i < rdev->num_crtc; i += 2) {
946 mode0 = &rdev->mode_info.crtcs[i]->base.mode;
947 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
948 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
949 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
950 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
951 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
958 static void si_tiling_mode_table_init(struct radeon_device *rdev)
960 const u32 num_tile_mode_states = 32;
961 u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
963 switch (rdev->config.si.mem_row_size_in_kb) {
965 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
969 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
972 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
976 if ((rdev->family == CHIP_TAHITI) ||
977 (rdev->family == CHIP_PITCAIRN)) {
978 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
979 switch (reg_offset) {
980 case 0: /* non-AA compressed depth or any compressed stencil */
981 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
982 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
983 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
984 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
985 NUM_BANKS(ADDR_SURF_16_BANK) |
986 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
987 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
988 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
990 case 1: /* 2xAA/4xAA compressed depth only */
991 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
992 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
993 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
994 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
995 NUM_BANKS(ADDR_SURF_16_BANK) |
996 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
997 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
998 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1000 case 2: /* 8xAA compressed depth only */
1001 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1002 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1003 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1004 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1005 NUM_BANKS(ADDR_SURF_16_BANK) |
1006 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1007 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1008 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1010 case 3: /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1011 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1012 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1013 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1014 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1015 NUM_BANKS(ADDR_SURF_16_BANK) |
1016 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1017 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1018 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1020 case 4: /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1021 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1022 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1023 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1024 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1025 NUM_BANKS(ADDR_SURF_16_BANK) |
1026 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1027 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1028 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1030 case 5: /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1031 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1032 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1033 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1034 TILE_SPLIT(split_equal_to_row_size) |
1035 NUM_BANKS(ADDR_SURF_16_BANK) |
1036 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1037 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1038 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1040 case 6: /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1041 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1042 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1043 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1044 TILE_SPLIT(split_equal_to_row_size) |
1045 NUM_BANKS(ADDR_SURF_16_BANK) |
1046 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1047 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1048 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1050 case 7: /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1051 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1052 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1053 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1054 TILE_SPLIT(split_equal_to_row_size) |
1055 NUM_BANKS(ADDR_SURF_16_BANK) |
1056 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1057 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1058 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1060 case 8: /* 1D and 1D Array Surfaces */
1061 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1062 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1063 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1064 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1065 NUM_BANKS(ADDR_SURF_16_BANK) |
1066 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1067 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1068 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1070 case 9: /* Displayable maps. */
1071 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1072 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1073 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1074 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1075 NUM_BANKS(ADDR_SURF_16_BANK) |
1076 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1077 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1078 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1080 case 10: /* Display 8bpp. */
1081 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1082 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1083 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1084 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1085 NUM_BANKS(ADDR_SURF_16_BANK) |
1086 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1087 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1088 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1090 case 11: /* Display 16bpp. */
1091 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1092 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1093 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1094 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1095 NUM_BANKS(ADDR_SURF_16_BANK) |
1096 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1097 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1098 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1100 case 12: /* Display 32bpp. */
1101 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1102 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1103 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1104 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1105 NUM_BANKS(ADDR_SURF_16_BANK) |
1106 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1107 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1108 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1110 case 13: /* Thin. */
1111 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1112 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1113 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1114 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1115 NUM_BANKS(ADDR_SURF_16_BANK) |
1116 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1117 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1118 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1120 case 14: /* Thin 8 bpp. */
1121 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1122 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1123 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1124 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1125 NUM_BANKS(ADDR_SURF_16_BANK) |
1126 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1127 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1128 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1130 case 15: /* Thin 16 bpp. */
1131 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1132 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1133 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1134 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1135 NUM_BANKS(ADDR_SURF_16_BANK) |
1136 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1137 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1138 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1140 case 16: /* Thin 32 bpp. */
1141 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1142 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1143 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1144 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1145 NUM_BANKS(ADDR_SURF_16_BANK) |
1146 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1147 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1148 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1150 case 17: /* Thin 64 bpp. */
1151 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1152 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1153 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1154 TILE_SPLIT(split_equal_to_row_size) |
1155 NUM_BANKS(ADDR_SURF_16_BANK) |
1156 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1157 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1158 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1160 case 21: /* 8 bpp PRT. */
1161 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1162 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1163 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1164 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1165 NUM_BANKS(ADDR_SURF_16_BANK) |
1166 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1167 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1168 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1170 case 22: /* 16 bpp PRT */
1171 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1172 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1173 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1174 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1175 NUM_BANKS(ADDR_SURF_16_BANK) |
1176 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1177 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1178 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1180 case 23: /* 32 bpp PRT */
1181 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1182 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1183 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1184 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1185 NUM_BANKS(ADDR_SURF_16_BANK) |
1186 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1187 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1188 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1190 case 24: /* 64 bpp PRT */
1191 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1192 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1193 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1194 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1195 NUM_BANKS(ADDR_SURF_16_BANK) |
1196 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1197 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1198 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1200 case 25: /* 128 bpp PRT */
1201 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1202 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1203 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1204 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1205 NUM_BANKS(ADDR_SURF_8_BANK) |
1206 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1207 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1208 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1214 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1216 } else if ((rdev->family == CHIP_VERDE) ||
1217 (rdev->family == CHIP_OLAND)) {
1218 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1219 switch (reg_offset) {
1220 case 0: /* non-AA compressed depth or any compressed stencil */
1221 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1222 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1223 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1224 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1225 NUM_BANKS(ADDR_SURF_16_BANK) |
1226 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1227 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1228 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1230 case 1: /* 2xAA/4xAA compressed depth only */
1231 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1232 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1233 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1234 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1235 NUM_BANKS(ADDR_SURF_16_BANK) |
1236 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1237 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1238 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1240 case 2: /* 8xAA compressed depth only */
1241 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1242 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1243 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1244 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1245 NUM_BANKS(ADDR_SURF_16_BANK) |
1246 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1247 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1248 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1250 case 3: /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1251 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1252 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1253 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1254 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1255 NUM_BANKS(ADDR_SURF_16_BANK) |
1256 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1257 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1258 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1260 case 4: /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1261 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1262 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1263 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1264 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1265 NUM_BANKS(ADDR_SURF_16_BANK) |
1266 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1267 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1268 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1270 case 5: /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1271 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1272 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1273 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1274 TILE_SPLIT(split_equal_to_row_size) |
1275 NUM_BANKS(ADDR_SURF_16_BANK) |
1276 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1277 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1278 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1280 case 6: /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1281 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1282 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1283 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1284 TILE_SPLIT(split_equal_to_row_size) |
1285 NUM_BANKS(ADDR_SURF_16_BANK) |
1286 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1287 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1288 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1290 case 7: /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1291 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1292 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1293 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1294 TILE_SPLIT(split_equal_to_row_size) |
1295 NUM_BANKS(ADDR_SURF_16_BANK) |
1296 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1297 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1298 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1300 case 8: /* 1D and 1D Array Surfaces */
1301 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1302 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1303 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1304 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1305 NUM_BANKS(ADDR_SURF_16_BANK) |
1306 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1307 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1308 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1310 case 9: /* Displayable maps. */
1311 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1312 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1313 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1314 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1315 NUM_BANKS(ADDR_SURF_16_BANK) |
1316 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1317 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1318 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1320 case 10: /* Display 8bpp. */
1321 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1322 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1323 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1324 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1325 NUM_BANKS(ADDR_SURF_16_BANK) |
1326 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1327 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1328 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1330 case 11: /* Display 16bpp. */
1331 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1332 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1333 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1334 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1335 NUM_BANKS(ADDR_SURF_16_BANK) |
1336 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1337 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1338 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1340 case 12: /* Display 32bpp. */
1341 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1342 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1343 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1344 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1345 NUM_BANKS(ADDR_SURF_16_BANK) |
1346 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1347 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1348 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1350 case 13: /* Thin. */
1351 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1352 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1353 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1354 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1355 NUM_BANKS(ADDR_SURF_16_BANK) |
1356 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1357 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1358 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1360 case 14: /* Thin 8 bpp. */
1361 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1362 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1363 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1364 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1365 NUM_BANKS(ADDR_SURF_16_BANK) |
1366 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1367 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1368 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1370 case 15: /* Thin 16 bpp. */
1371 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1372 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1373 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1374 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1375 NUM_BANKS(ADDR_SURF_16_BANK) |
1376 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1377 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1378 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1380 case 16: /* Thin 32 bpp. */
1381 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1382 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1383 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1384 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1385 NUM_BANKS(ADDR_SURF_16_BANK) |
1386 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1387 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1388 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1390 case 17: /* Thin 64 bpp. */
1391 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1392 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1393 PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1394 TILE_SPLIT(split_equal_to_row_size) |
1395 NUM_BANKS(ADDR_SURF_16_BANK) |
1396 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1397 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1398 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1400 case 21: /* 8 bpp PRT. */
1401 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1402 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1403 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1404 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1405 NUM_BANKS(ADDR_SURF_16_BANK) |
1406 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1407 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1408 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1410 case 22: /* 16 bpp PRT */
1411 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1412 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1413 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1414 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1415 NUM_BANKS(ADDR_SURF_16_BANK) |
1416 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1417 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1418 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1420 case 23: /* 32 bpp PRT */
1421 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1422 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1423 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1424 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1425 NUM_BANKS(ADDR_SURF_16_BANK) |
1426 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1427 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1428 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1430 case 24: /* 64 bpp PRT */
1431 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1432 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1433 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1434 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1435 NUM_BANKS(ADDR_SURF_16_BANK) |
1436 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1437 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1438 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1440 case 25: /* 128 bpp PRT */
1441 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1442 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1443 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1444 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1445 NUM_BANKS(ADDR_SURF_8_BANK) |
1446 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1447 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1448 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1454 WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1457 DRM_ERROR("unknown asic: 0x%x\n", rdev->family);
1460 static void si_select_se_sh(struct radeon_device *rdev,
1461 u32 se_num, u32 sh_num)
1463 u32 data = INSTANCE_BROADCAST_WRITES;
1465 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1466 data = SH_BROADCAST_WRITES | SE_BROADCAST_WRITES;
1467 else if (se_num == 0xffffffff)
1468 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num);
1469 else if (sh_num == 0xffffffff)
1470 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num);
1472 data |= SH_INDEX(sh_num) | SE_INDEX(se_num);
1473 WREG32(GRBM_GFX_INDEX, data);
1476 static u32 si_create_bitmask(u32 bit_width)
1480 for (i = 0; i < bit_width; i++) {
1487 static u32 si_get_cu_enabled(struct radeon_device *rdev, u32 cu_per_sh)
1491 data = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
1493 data &= INACTIVE_CUS_MASK;
1496 data |= RREG32(GC_USER_SHADER_ARRAY_CONFIG);
1498 data >>= INACTIVE_CUS_SHIFT;
1500 mask = si_create_bitmask(cu_per_sh);
1502 return ~data & mask;
1505 static void si_setup_spi(struct radeon_device *rdev,
1506 u32 se_num, u32 sh_per_se,
1510 u32 data, mask, active_cu;
1512 for (i = 0; i < se_num; i++) {
1513 for (j = 0; j < sh_per_se; j++) {
1514 si_select_se_sh(rdev, i, j);
1515 data = RREG32(SPI_STATIC_THREAD_MGMT_3);
1516 active_cu = si_get_cu_enabled(rdev, cu_per_sh);
1519 for (k = 0; k < 16; k++) {
1521 if (active_cu & mask) {
1523 WREG32(SPI_STATIC_THREAD_MGMT_3, data);
1529 si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1532 static u32 si_get_rb_disabled(struct radeon_device *rdev,
1533 u32 max_rb_num, u32 se_num,
1538 data = RREG32(CC_RB_BACKEND_DISABLE);
1540 data &= BACKEND_DISABLE_MASK;
1543 data |= RREG32(GC_USER_RB_BACKEND_DISABLE);
1545 data >>= BACKEND_DISABLE_SHIFT;
1547 mask = si_create_bitmask(max_rb_num / se_num / sh_per_se);
1552 static void si_setup_rb(struct radeon_device *rdev,
1553 u32 se_num, u32 sh_per_se,
1558 u32 disabled_rbs = 0;
1559 u32 enabled_rbs = 0;
1561 for (i = 0; i < se_num; i++) {
1562 for (j = 0; j < sh_per_se; j++) {
1563 si_select_se_sh(rdev, i, j);
1564 data = si_get_rb_disabled(rdev, max_rb_num, se_num, sh_per_se);
1565 disabled_rbs |= data << ((i * sh_per_se + j) * TAHITI_RB_BITMAP_WIDTH_PER_SH);
1568 si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1571 for (i = 0; i < max_rb_num; i++) {
1572 if (!(disabled_rbs & mask))
1573 enabled_rbs |= mask;
1577 for (i = 0; i < se_num; i++) {
1578 si_select_se_sh(rdev, i, 0xffffffff);
1580 for (j = 0; j < sh_per_se; j++) {
1581 switch (enabled_rbs & 3) {
1583 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2);
1586 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2);
1590 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2);
1595 WREG32(PA_SC_RASTER_CONFIG, data);
1597 si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1600 static void si_gpu_init(struct radeon_device *rdev)
1602 u32 gb_addr_config = 0;
1603 u32 mc_shared_chmap, mc_arb_ramcfg;
1605 u32 hdp_host_path_cntl;
1609 switch (rdev->family) {
1611 rdev->config.si.max_shader_engines = 2;
1612 rdev->config.si.max_tile_pipes = 12;
1613 rdev->config.si.max_cu_per_sh = 8;
1614 rdev->config.si.max_sh_per_se = 2;
1615 rdev->config.si.max_backends_per_se = 4;
1616 rdev->config.si.max_texture_channel_caches = 12;
1617 rdev->config.si.max_gprs = 256;
1618 rdev->config.si.max_gs_threads = 32;
1619 rdev->config.si.max_hw_contexts = 8;
1621 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1622 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1623 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1624 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1625 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN;
1628 rdev->config.si.max_shader_engines = 2;
1629 rdev->config.si.max_tile_pipes = 8;
1630 rdev->config.si.max_cu_per_sh = 5;
1631 rdev->config.si.max_sh_per_se = 2;
1632 rdev->config.si.max_backends_per_se = 4;
1633 rdev->config.si.max_texture_channel_caches = 8;
1634 rdev->config.si.max_gprs = 256;
1635 rdev->config.si.max_gs_threads = 32;
1636 rdev->config.si.max_hw_contexts = 8;
1638 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1639 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1640 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1641 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1642 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN;
1646 rdev->config.si.max_shader_engines = 1;
1647 rdev->config.si.max_tile_pipes = 4;
1648 rdev->config.si.max_cu_per_sh = 2;
1649 rdev->config.si.max_sh_per_se = 2;
1650 rdev->config.si.max_backends_per_se = 4;
1651 rdev->config.si.max_texture_channel_caches = 4;
1652 rdev->config.si.max_gprs = 256;
1653 rdev->config.si.max_gs_threads = 32;
1654 rdev->config.si.max_hw_contexts = 8;
1656 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1657 rdev->config.si.sc_prim_fifo_size_backend = 0x40;
1658 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1659 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1660 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN;
1663 rdev->config.si.max_shader_engines = 1;
1664 rdev->config.si.max_tile_pipes = 4;
1665 rdev->config.si.max_cu_per_sh = 6;
1666 rdev->config.si.max_sh_per_se = 1;
1667 rdev->config.si.max_backends_per_se = 2;
1668 rdev->config.si.max_texture_channel_caches = 4;
1669 rdev->config.si.max_gprs = 256;
1670 rdev->config.si.max_gs_threads = 16;
1671 rdev->config.si.max_hw_contexts = 8;
1673 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1674 rdev->config.si.sc_prim_fifo_size_backend = 0x40;
1675 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1676 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1677 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN;
1681 /* Initialize HDP */
1682 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1683 WREG32((0x2c14 + j), 0x00000000);
1684 WREG32((0x2c18 + j), 0x00000000);
1685 WREG32((0x2c1c + j), 0x00000000);
1686 WREG32((0x2c20 + j), 0x00000000);
1687 WREG32((0x2c24 + j), 0x00000000);
1690 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1692 evergreen_fix_pci_max_read_req_size(rdev);
1694 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
1696 mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1697 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1699 rdev->config.si.num_tile_pipes = rdev->config.si.max_tile_pipes;
1700 rdev->config.si.mem_max_burst_length_bytes = 256;
1701 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
1702 rdev->config.si.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1703 if (rdev->config.si.mem_row_size_in_kb > 4)
1704 rdev->config.si.mem_row_size_in_kb = 4;
1705 /* XXX use MC settings? */
1706 rdev->config.si.shader_engine_tile_size = 32;
1707 rdev->config.si.num_gpus = 1;
1708 rdev->config.si.multi_gpu_tile_size = 64;
1710 /* fix up row size */
1711 gb_addr_config &= ~ROW_SIZE_MASK;
1712 switch (rdev->config.si.mem_row_size_in_kb) {
1715 gb_addr_config |= ROW_SIZE(0);
1718 gb_addr_config |= ROW_SIZE(1);
1721 gb_addr_config |= ROW_SIZE(2);
1725 /* setup tiling info dword. gb_addr_config is not adequate since it does
1726 * not have bank info, so create a custom tiling dword.
1727 * bits 3:0 num_pipes
1728 * bits 7:4 num_banks
1729 * bits 11:8 group_size
1730 * bits 15:12 row_size
1732 rdev->config.si.tile_config = 0;
1733 switch (rdev->config.si.num_tile_pipes) {
1735 rdev->config.si.tile_config |= (0 << 0);
1738 rdev->config.si.tile_config |= (1 << 0);
1741 rdev->config.si.tile_config |= (2 << 0);
1745 /* XXX what about 12? */
1746 rdev->config.si.tile_config |= (3 << 0);
1749 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
1750 case 0: /* four banks */
1751 rdev->config.si.tile_config |= 0 << 4;
1753 case 1: /* eight banks */
1754 rdev->config.si.tile_config |= 1 << 4;
1756 case 2: /* sixteen banks */
1758 rdev->config.si.tile_config |= 2 << 4;
1761 rdev->config.si.tile_config |=
1762 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
1763 rdev->config.si.tile_config |=
1764 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
1766 WREG32(GB_ADDR_CONFIG, gb_addr_config);
1767 WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
1768 WREG32(HDP_ADDR_CONFIG, gb_addr_config);
1769 WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config);
1770 WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config);
1772 si_tiling_mode_table_init(rdev);
1774 si_setup_rb(rdev, rdev->config.si.max_shader_engines,
1775 rdev->config.si.max_sh_per_se,
1776 rdev->config.si.max_backends_per_se);
1778 si_setup_spi(rdev, rdev->config.si.max_shader_engines,
1779 rdev->config.si.max_sh_per_se,
1780 rdev->config.si.max_cu_per_sh);
1783 /* set HW defaults for 3D engine */
1784 WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
1785 ROQ_IB2_START(0x2b)));
1786 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
1788 sx_debug_1 = RREG32(SX_DEBUG_1);
1789 WREG32(SX_DEBUG_1, sx_debug_1);
1791 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
1793 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_frontend) |
1794 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_backend) |
1795 SC_HIZ_TILE_FIFO_SIZE(rdev->config.si.sc_hiz_tile_fifo_size) |
1796 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.si.sc_earlyz_tile_fifo_size)));
1798 WREG32(VGT_NUM_INSTANCES, 1);
1800 WREG32(CP_PERFMON_CNTL, 0);
1802 WREG32(SQ_CONFIG, 0);
1804 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
1805 FORCE_EOV_MAX_REZ_CNT(255)));
1807 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
1808 AUTO_INVLD_EN(ES_AND_GS_AUTO));
1810 WREG32(VGT_GS_VERTEX_REUSE, 16);
1811 WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1813 WREG32(CB_PERFCOUNTER0_SELECT0, 0);
1814 WREG32(CB_PERFCOUNTER0_SELECT1, 0);
1815 WREG32(CB_PERFCOUNTER1_SELECT0, 0);
1816 WREG32(CB_PERFCOUNTER1_SELECT1, 0);
1817 WREG32(CB_PERFCOUNTER2_SELECT0, 0);
1818 WREG32(CB_PERFCOUNTER2_SELECT1, 0);
1819 WREG32(CB_PERFCOUNTER3_SELECT0, 0);
1820 WREG32(CB_PERFCOUNTER3_SELECT1, 0);
1822 tmp = RREG32(HDP_MISC_CNTL);
1823 tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1824 WREG32(HDP_MISC_CNTL, tmp);
1826 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1827 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1829 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1835 * GPU scratch registers helpers function.
1837 static void si_scratch_init(struct radeon_device *rdev)
1841 rdev->scratch.num_reg = 7;
1842 rdev->scratch.reg_base = SCRATCH_REG0;
1843 for (i = 0; i < rdev->scratch.num_reg; i++) {
1844 rdev->scratch.free[i] = true;
1845 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
1849 void si_fence_ring_emit(struct radeon_device *rdev,
1850 struct radeon_fence *fence)
1852 struct radeon_ring *ring = &rdev->ring[fence->ring];
1853 u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1855 /* flush read cache over gart */
1856 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1857 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1858 radeon_ring_write(ring, 0);
1859 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1860 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1861 PACKET3_TC_ACTION_ENA |
1862 PACKET3_SH_KCACHE_ACTION_ENA |
1863 PACKET3_SH_ICACHE_ACTION_ENA);
1864 radeon_ring_write(ring, 0xFFFFFFFF);
1865 radeon_ring_write(ring, 0);
1866 radeon_ring_write(ring, 10); /* poll interval */
1867 /* EVENT_WRITE_EOP - flush caches, send int */
1868 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1869 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5));
1870 radeon_ring_write(ring, addr & 0xffffffff);
1871 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1872 radeon_ring_write(ring, fence->seq);
1873 radeon_ring_write(ring, 0);
1879 void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1881 struct radeon_ring *ring = &rdev->ring[ib->ring];
1884 if (ib->is_const_ib) {
1885 /* set switch buffer packet before const IB */
1886 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
1887 radeon_ring_write(ring, 0);
1889 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
1892 if (ring->rptr_save_reg) {
1893 next_rptr = ring->wptr + 3 + 4 + 8;
1894 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1895 radeon_ring_write(ring, ((ring->rptr_save_reg -
1896 PACKET3_SET_CONFIG_REG_START) >> 2));
1897 radeon_ring_write(ring, next_rptr);
1898 } else if (rdev->wb.enabled) {
1899 next_rptr = ring->wptr + 5 + 4 + 8;
1900 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
1901 radeon_ring_write(ring, (1 << 8));
1902 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
1903 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
1904 radeon_ring_write(ring, next_rptr);
1907 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
1910 radeon_ring_write(ring, header);
1911 radeon_ring_write(ring,
1915 (ib->gpu_addr & 0xFFFFFFFC));
1916 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
1917 radeon_ring_write(ring, ib->length_dw |
1918 (ib->vm ? (ib->vm->id << 24) : 0));
1920 if (!ib->is_const_ib) {
1921 /* flush read cache over gart for this vmid */
1922 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1923 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1924 radeon_ring_write(ring, ib->vm ? ib->vm->id : 0);
1925 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1926 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1927 PACKET3_TC_ACTION_ENA |
1928 PACKET3_SH_KCACHE_ACTION_ENA |
1929 PACKET3_SH_ICACHE_ACTION_ENA);
1930 radeon_ring_write(ring, 0xFFFFFFFF);
1931 radeon_ring_write(ring, 0);
1932 radeon_ring_write(ring, 10); /* poll interval */
1939 static void si_cp_enable(struct radeon_device *rdev, bool enable)
1942 WREG32(CP_ME_CNTL, 0);
1944 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1945 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
1946 WREG32(SCRATCH_UMSK, 0);
1947 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1948 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1949 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1954 static int si_cp_load_microcode(struct radeon_device *rdev)
1956 const __be32 *fw_data;
1959 if (!rdev->me_fw || !rdev->pfp_fw)
1962 si_cp_enable(rdev, false);
1965 fw_data = (const __be32 *)rdev->pfp_fw->data;
1966 WREG32(CP_PFP_UCODE_ADDR, 0);
1967 for (i = 0; i < SI_PFP_UCODE_SIZE; i++)
1968 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1969 WREG32(CP_PFP_UCODE_ADDR, 0);
1972 fw_data = (const __be32 *)rdev->ce_fw->data;
1973 WREG32(CP_CE_UCODE_ADDR, 0);
1974 for (i = 0; i < SI_CE_UCODE_SIZE; i++)
1975 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
1976 WREG32(CP_CE_UCODE_ADDR, 0);
1979 fw_data = (const __be32 *)rdev->me_fw->data;
1980 WREG32(CP_ME_RAM_WADDR, 0);
1981 for (i = 0; i < SI_PM4_UCODE_SIZE; i++)
1982 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1983 WREG32(CP_ME_RAM_WADDR, 0);
1985 WREG32(CP_PFP_UCODE_ADDR, 0);
1986 WREG32(CP_CE_UCODE_ADDR, 0);
1987 WREG32(CP_ME_RAM_WADDR, 0);
1988 WREG32(CP_ME_RAM_RADDR, 0);
1992 static int si_cp_start(struct radeon_device *rdev)
1994 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1997 r = radeon_ring_lock(rdev, ring, 7 + 4);
1999 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2003 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
2004 radeon_ring_write(ring, 0x1);
2005 radeon_ring_write(ring, 0x0);
2006 radeon_ring_write(ring, rdev->config.si.max_hw_contexts - 1);
2007 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
2008 radeon_ring_write(ring, 0);
2009 radeon_ring_write(ring, 0);
2011 /* init the CE partitions */
2012 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2013 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2014 radeon_ring_write(ring, 0xc000);
2015 radeon_ring_write(ring, 0xe000);
2016 radeon_ring_unlock_commit(rdev, ring);
2018 si_cp_enable(rdev, true);
2020 r = radeon_ring_lock(rdev, ring, si_default_size + 10);
2022 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2026 /* setup clear context state */
2027 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2028 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2030 for (i = 0; i < si_default_size; i++)
2031 radeon_ring_write(ring, si_default_state[i]);
2033 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2034 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2036 /* set clear context state */
2037 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2038 radeon_ring_write(ring, 0);
2040 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2041 radeon_ring_write(ring, 0x00000316);
2042 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2043 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2045 radeon_ring_unlock_commit(rdev, ring);
2047 for (i = RADEON_RING_TYPE_GFX_INDEX; i <= CAYMAN_RING_TYPE_CP2_INDEX; ++i) {
2048 ring = &rdev->ring[i];
2049 r = radeon_ring_lock(rdev, ring, 2);
2051 /* clear the compute context state */
2052 radeon_ring_write(ring, PACKET3_COMPUTE(PACKET3_CLEAR_STATE, 0));
2053 radeon_ring_write(ring, 0);
2055 radeon_ring_unlock_commit(rdev, ring);
2061 static void si_cp_fini(struct radeon_device *rdev)
2063 struct radeon_ring *ring;
2064 si_cp_enable(rdev, false);
2066 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2067 radeon_ring_fini(rdev, ring);
2068 radeon_scratch_free(rdev, ring->rptr_save_reg);
2070 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
2071 radeon_ring_fini(rdev, ring);
2072 radeon_scratch_free(rdev, ring->rptr_save_reg);
2074 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
2075 radeon_ring_fini(rdev, ring);
2076 radeon_scratch_free(rdev, ring->rptr_save_reg);
2079 static int si_cp_resume(struct radeon_device *rdev)
2081 struct radeon_ring *ring;
2086 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
2087 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
2092 RREG32(GRBM_SOFT_RESET);
2094 WREG32(GRBM_SOFT_RESET, 0);
2095 RREG32(GRBM_SOFT_RESET);
2097 WREG32(CP_SEM_WAIT_TIMER, 0x0);
2098 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2100 /* Set the write pointer delay */
2101 WREG32(CP_RB_WPTR_DELAY, 0);
2103 WREG32(CP_DEBUG, 0);
2104 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
2106 /* ring 0 - compute and gfx */
2107 /* Set ring buffer size */
2108 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2109 rb_bufsz = drm_order(ring->ring_size / 8);
2110 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2112 tmp |= BUF_SWAP_32BIT;
2114 WREG32(CP_RB0_CNTL, tmp);
2116 /* Initialize the ring buffer's read and write pointers */
2117 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
2119 WREG32(CP_RB0_WPTR, ring->wptr);
2121 /* set the wb address whether it's enabled or not */
2122 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
2123 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
2125 if (rdev->wb.enabled)
2126 WREG32(SCRATCH_UMSK, 0xff);
2128 tmp |= RB_NO_UPDATE;
2129 WREG32(SCRATCH_UMSK, 0);
2133 WREG32(CP_RB0_CNTL, tmp);
2135 WREG32(CP_RB0_BASE, ring->gpu_addr >> 8);
2137 ring->rptr = RREG32(CP_RB0_RPTR);
2139 /* ring1 - compute only */
2140 /* Set ring buffer size */
2141 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
2142 rb_bufsz = drm_order(ring->ring_size / 8);
2143 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2145 tmp |= BUF_SWAP_32BIT;
2147 WREG32(CP_RB1_CNTL, tmp);
2149 /* Initialize the ring buffer's read and write pointers */
2150 WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA);
2152 WREG32(CP_RB1_WPTR, ring->wptr);
2154 /* set the wb address whether it's enabled or not */
2155 WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC);
2156 WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF);
2159 WREG32(CP_RB1_CNTL, tmp);
2161 WREG32(CP_RB1_BASE, ring->gpu_addr >> 8);
2163 ring->rptr = RREG32(CP_RB1_RPTR);
2165 /* ring2 - compute only */
2166 /* Set ring buffer size */
2167 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
2168 rb_bufsz = drm_order(ring->ring_size / 8);
2169 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2171 tmp |= BUF_SWAP_32BIT;
2173 WREG32(CP_RB2_CNTL, tmp);
2175 /* Initialize the ring buffer's read and write pointers */
2176 WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA);
2178 WREG32(CP_RB2_WPTR, ring->wptr);
2180 /* set the wb address whether it's enabled or not */
2181 WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC);
2182 WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF);
2185 WREG32(CP_RB2_CNTL, tmp);
2187 WREG32(CP_RB2_BASE, ring->gpu_addr >> 8);
2189 ring->rptr = RREG32(CP_RB2_RPTR);
2191 /* start the rings */
2193 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
2194 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = true;
2195 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = true;
2196 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
2198 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
2199 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2200 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2203 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP1_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]);
2205 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2207 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP2_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]);
2209 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2215 static u32 si_gpu_check_soft_reset(struct radeon_device *rdev)
2221 tmp = RREG32(GRBM_STATUS);
2222 if (tmp & (PA_BUSY | SC_BUSY |
2223 BCI_BUSY | SX_BUSY |
2224 TA_BUSY | VGT_BUSY |
2226 GDS_BUSY | SPI_BUSY |
2227 IA_BUSY | IA_BUSY_NO_DMA))
2228 reset_mask |= RADEON_RESET_GFX;
2230 if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING |
2231 CP_BUSY | CP_COHERENCY_BUSY))
2232 reset_mask |= RADEON_RESET_CP;
2234 if (tmp & GRBM_EE_BUSY)
2235 reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP;
2238 tmp = RREG32(GRBM_STATUS2);
2239 if (tmp & (RLC_RQ_PENDING | RLC_BUSY))
2240 reset_mask |= RADEON_RESET_RLC;
2242 /* DMA_STATUS_REG 0 */
2243 tmp = RREG32(DMA_STATUS_REG + DMA0_REGISTER_OFFSET);
2244 if (!(tmp & DMA_IDLE))
2245 reset_mask |= RADEON_RESET_DMA;
2247 /* DMA_STATUS_REG 1 */
2248 tmp = RREG32(DMA_STATUS_REG + DMA1_REGISTER_OFFSET);
2249 if (!(tmp & DMA_IDLE))
2250 reset_mask |= RADEON_RESET_DMA1;
2253 tmp = RREG32(SRBM_STATUS2);
2255 reset_mask |= RADEON_RESET_DMA;
2257 if (tmp & DMA1_BUSY)
2258 reset_mask |= RADEON_RESET_DMA1;
2261 tmp = RREG32(SRBM_STATUS);
2264 reset_mask |= RADEON_RESET_IH;
2267 reset_mask |= RADEON_RESET_SEM;
2269 if (tmp & GRBM_RQ_PENDING)
2270 reset_mask |= RADEON_RESET_GRBM;
2273 reset_mask |= RADEON_RESET_VMC;
2275 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
2276 MCC_BUSY | MCD_BUSY))
2277 reset_mask |= RADEON_RESET_MC;
2279 if (evergreen_is_display_hung(rdev))
2280 reset_mask |= RADEON_RESET_DISPLAY;
2283 tmp = RREG32(VM_L2_STATUS);
2285 reset_mask |= RADEON_RESET_VMC;
2287 /* Skip MC reset as it's mostly likely not hung, just busy */
2288 if (reset_mask & RADEON_RESET_MC) {
2289 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask);
2290 reset_mask &= ~RADEON_RESET_MC;
2296 static void si_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
2298 struct evergreen_mc_save save;
2299 u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
2302 if (reset_mask == 0)
2305 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
2307 evergreen_print_gpu_status_regs(rdev);
2308 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
2309 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
2310 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
2311 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
2313 /* Disable CP parsing/prefetching */
2314 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
2316 if (reset_mask & RADEON_RESET_DMA) {
2318 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
2319 tmp &= ~DMA_RB_ENABLE;
2320 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp);
2322 if (reset_mask & RADEON_RESET_DMA1) {
2324 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
2325 tmp &= ~DMA_RB_ENABLE;
2326 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp);
2331 evergreen_mc_stop(rdev, &save);
2332 if (evergreen_mc_wait_for_idle(rdev)) {
2333 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2336 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP)) {
2337 grbm_soft_reset = SOFT_RESET_CB |
2351 if (reset_mask & RADEON_RESET_CP) {
2352 grbm_soft_reset |= SOFT_RESET_CP | SOFT_RESET_VGT;
2354 srbm_soft_reset |= SOFT_RESET_GRBM;
2357 if (reset_mask & RADEON_RESET_DMA)
2358 srbm_soft_reset |= SOFT_RESET_DMA;
2360 if (reset_mask & RADEON_RESET_DMA1)
2361 srbm_soft_reset |= SOFT_RESET_DMA1;
2363 if (reset_mask & RADEON_RESET_DISPLAY)
2364 srbm_soft_reset |= SOFT_RESET_DC;
2366 if (reset_mask & RADEON_RESET_RLC)
2367 grbm_soft_reset |= SOFT_RESET_RLC;
2369 if (reset_mask & RADEON_RESET_SEM)
2370 srbm_soft_reset |= SOFT_RESET_SEM;
2372 if (reset_mask & RADEON_RESET_IH)
2373 srbm_soft_reset |= SOFT_RESET_IH;
2375 if (reset_mask & RADEON_RESET_GRBM)
2376 srbm_soft_reset |= SOFT_RESET_GRBM;
2378 if (reset_mask & RADEON_RESET_VMC)
2379 srbm_soft_reset |= SOFT_RESET_VMC;
2381 if (reset_mask & RADEON_RESET_MC)
2382 srbm_soft_reset |= SOFT_RESET_MC;
2384 if (grbm_soft_reset) {
2385 tmp = RREG32(GRBM_SOFT_RESET);
2386 tmp |= grbm_soft_reset;
2387 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
2388 WREG32(GRBM_SOFT_RESET, tmp);
2389 tmp = RREG32(GRBM_SOFT_RESET);
2393 tmp &= ~grbm_soft_reset;
2394 WREG32(GRBM_SOFT_RESET, tmp);
2395 tmp = RREG32(GRBM_SOFT_RESET);
2398 if (srbm_soft_reset) {
2399 tmp = RREG32(SRBM_SOFT_RESET);
2400 tmp |= srbm_soft_reset;
2401 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
2402 WREG32(SRBM_SOFT_RESET, tmp);
2403 tmp = RREG32(SRBM_SOFT_RESET);
2407 tmp &= ~srbm_soft_reset;
2408 WREG32(SRBM_SOFT_RESET, tmp);
2409 tmp = RREG32(SRBM_SOFT_RESET);
2412 /* Wait a little for things to settle down */
2415 evergreen_mc_resume(rdev, &save);
2418 evergreen_print_gpu_status_regs(rdev);
2421 int si_asic_reset(struct radeon_device *rdev)
2425 reset_mask = si_gpu_check_soft_reset(rdev);
2428 r600_set_bios_scratch_engine_hung(rdev, true);
2430 si_gpu_soft_reset(rdev, reset_mask);
2432 reset_mask = si_gpu_check_soft_reset(rdev);
2435 r600_set_bios_scratch_engine_hung(rdev, false);
2441 * si_gfx_is_lockup - Check if the GFX engine is locked up
2443 * @rdev: radeon_device pointer
2444 * @ring: radeon_ring structure holding ring information
2446 * Check if the GFX engine is locked up.
2447 * Returns true if the engine appears to be locked up, false if not.
2449 bool si_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2451 u32 reset_mask = si_gpu_check_soft_reset(rdev);
2453 if (!(reset_mask & (RADEON_RESET_GFX |
2454 RADEON_RESET_COMPUTE |
2455 RADEON_RESET_CP))) {
2456 radeon_ring_lockup_update(ring);
2459 /* force CP activities */
2460 radeon_ring_force_activity(rdev, ring);
2461 return radeon_ring_test_lockup(rdev, ring);
2465 * si_dma_is_lockup - Check if the DMA engine is locked up
2467 * @rdev: radeon_device pointer
2468 * @ring: radeon_ring structure holding ring information
2470 * Check if the async DMA engine is locked up.
2471 * Returns true if the engine appears to be locked up, false if not.
2473 bool si_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2475 u32 reset_mask = si_gpu_check_soft_reset(rdev);
2478 if (ring->idx == R600_RING_TYPE_DMA_INDEX)
2479 mask = RADEON_RESET_DMA;
2481 mask = RADEON_RESET_DMA1;
2483 if (!(reset_mask & mask)) {
2484 radeon_ring_lockup_update(ring);
2487 /* force ring activities */
2488 radeon_ring_force_activity(rdev, ring);
2489 return radeon_ring_test_lockup(rdev, ring);
2493 static void si_mc_program(struct radeon_device *rdev)
2495 struct evergreen_mc_save save;
2499 /* Initialize HDP */
2500 for (i = 0, j = 0; i < 32; i++, j += 0x18) {
2501 WREG32((0x2c14 + j), 0x00000000);
2502 WREG32((0x2c18 + j), 0x00000000);
2503 WREG32((0x2c1c + j), 0x00000000);
2504 WREG32((0x2c20 + j), 0x00000000);
2505 WREG32((0x2c24 + j), 0x00000000);
2507 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
2509 evergreen_mc_stop(rdev, &save);
2510 if (radeon_mc_wait_for_idle(rdev)) {
2511 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2513 /* Lockout access through VGA aperture*/
2514 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
2515 /* Update configuration */
2516 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2517 rdev->mc.vram_start >> 12);
2518 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2519 rdev->mc.vram_end >> 12);
2520 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
2521 rdev->vram_scratch.gpu_addr >> 12);
2522 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
2523 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
2524 WREG32(MC_VM_FB_LOCATION, tmp);
2525 /* XXX double check these! */
2526 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
2527 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
2528 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
2529 WREG32(MC_VM_AGP_BASE, 0);
2530 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
2531 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
2532 if (radeon_mc_wait_for_idle(rdev)) {
2533 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2535 evergreen_mc_resume(rdev, &save);
2536 /* we need to own VRAM, so turn off the VGA renderer here
2537 * to stop it overwriting our objects */
2538 rv515_vga_render_disable(rdev);
2541 /* SI MC address space is 40 bits */
2542 static void si_vram_location(struct radeon_device *rdev,
2543 struct radeon_mc *mc, u64 base)
2545 mc->vram_start = base;
2546 if (mc->mc_vram_size > (0xFFFFFFFFFFULL - base + 1)) {
2547 dev_warn(rdev->dev, "limiting VRAM to PCI aperture size\n");
2548 mc->real_vram_size = mc->aper_size;
2549 mc->mc_vram_size = mc->aper_size;
2551 mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
2552 dev_info(rdev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
2553 mc->mc_vram_size >> 20, mc->vram_start,
2554 mc->vram_end, mc->real_vram_size >> 20);
2557 static void si_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
2559 u64 size_af, size_bf;
2561 size_af = ((0xFFFFFFFFFFULL - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align;
2562 size_bf = mc->vram_start & ~mc->gtt_base_align;
2563 if (size_bf > size_af) {
2564 if (mc->gtt_size > size_bf) {
2565 dev_warn(rdev->dev, "limiting GTT\n");
2566 mc->gtt_size = size_bf;
2568 mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size;
2570 if (mc->gtt_size > size_af) {
2571 dev_warn(rdev->dev, "limiting GTT\n");
2572 mc->gtt_size = size_af;
2574 mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align;
2576 mc->gtt_end = mc->gtt_start + mc->gtt_size - 1;
2577 dev_info(rdev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n",
2578 mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end);
2581 static void si_vram_gtt_location(struct radeon_device *rdev,
2582 struct radeon_mc *mc)
2584 if (mc->mc_vram_size > 0xFFC0000000ULL) {
2585 /* leave room for at least 1024M GTT */
2586 dev_warn(rdev->dev, "limiting VRAM\n");
2587 mc->real_vram_size = 0xFFC0000000ULL;
2588 mc->mc_vram_size = 0xFFC0000000ULL;
2590 si_vram_location(rdev, &rdev->mc, 0);
2591 rdev->mc.gtt_base_align = 0;
2592 si_gtt_location(rdev, mc);
2595 static int si_mc_init(struct radeon_device *rdev)
2598 int chansize, numchan;
2600 /* Get VRAM informations */
2601 rdev->mc.vram_is_ddr = true;
2602 tmp = RREG32(MC_ARB_RAMCFG);
2603 if (tmp & CHANSIZE_OVERRIDE) {
2605 } else if (tmp & CHANSIZE_MASK) {
2610 tmp = RREG32(MC_SHARED_CHMAP);
2611 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
2641 rdev->mc.vram_width = numchan * chansize;
2642 /* Could aper size report 0 ? */
2643 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2644 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2645 /* size in MB on si */
2646 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2647 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2648 rdev->mc.visible_vram_size = rdev->mc.aper_size;
2649 si_vram_gtt_location(rdev, &rdev->mc);
2650 radeon_update_bandwidth_info(rdev);
2658 void si_pcie_gart_tlb_flush(struct radeon_device *rdev)
2660 /* flush hdp cache */
2661 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2663 /* bits 0-15 are the VM contexts0-15 */
2664 WREG32(VM_INVALIDATE_REQUEST, 1);
2667 static int si_pcie_gart_enable(struct radeon_device *rdev)
2671 if (rdev->gart.robj == NULL) {
2672 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
2675 r = radeon_gart_table_vram_pin(rdev);
2678 radeon_gart_restore(rdev);
2679 /* Setup TLB control */
2680 WREG32(MC_VM_MX_L1_TLB_CNTL,
2683 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2684 ENABLE_ADVANCED_DRIVER_MODEL |
2685 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2686 /* Setup L2 cache */
2687 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
2688 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2689 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2690 EFFECTIVE_L2_QUEUE_SIZE(7) |
2691 CONTEXT1_IDENTITY_ACCESS_MODE(1));
2692 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
2693 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2694 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2695 /* setup context0 */
2696 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
2697 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
2698 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
2699 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
2700 (u32)(rdev->dummy_page.addr >> 12));
2701 WREG32(VM_CONTEXT0_CNTL2, 0);
2702 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
2703 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
2709 /* empty context1-15 */
2710 /* set vm size, must be a multiple of 4 */
2711 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
2712 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn);
2713 /* Assign the pt base to something valid for now; the pts used for
2714 * the VMs are determined by the application and setup and assigned
2715 * on the fly in the vm part of radeon_gart.c
2717 for (i = 1; i < 16; i++) {
2719 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
2720 rdev->gart.table_addr >> 12);
2722 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
2723 rdev->gart.table_addr >> 12);
2726 /* enable context1-15 */
2727 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
2728 (u32)(rdev->dummy_page.addr >> 12));
2729 WREG32(VM_CONTEXT1_CNTL2, 4);
2730 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
2731 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2732 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
2733 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2734 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
2735 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
2736 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
2737 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
2738 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
2739 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
2740 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
2741 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2742 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
2744 si_pcie_gart_tlb_flush(rdev);
2745 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
2746 (unsigned)(rdev->mc.gtt_size >> 20),
2747 (unsigned long long)rdev->gart.table_addr);
2748 rdev->gart.ready = true;
2752 static void si_pcie_gart_disable(struct radeon_device *rdev)
2754 /* Disable all tables */
2755 WREG32(VM_CONTEXT0_CNTL, 0);
2756 WREG32(VM_CONTEXT1_CNTL, 0);
2757 /* Setup TLB control */
2758 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2759 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2760 /* Setup L2 cache */
2761 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2762 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2763 EFFECTIVE_L2_QUEUE_SIZE(7) |
2764 CONTEXT1_IDENTITY_ACCESS_MODE(1));
2765 WREG32(VM_L2_CNTL2, 0);
2766 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2767 L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2768 radeon_gart_table_vram_unpin(rdev);
2771 static void si_pcie_gart_fini(struct radeon_device *rdev)
2773 si_pcie_gart_disable(rdev);
2774 radeon_gart_table_vram_free(rdev);
2775 radeon_gart_fini(rdev);
2779 static bool si_vm_reg_valid(u32 reg)
2781 /* context regs are fine */
2785 /* check config regs */
2787 case GRBM_GFX_INDEX:
2788 case CP_STRMOUT_CNTL:
2789 case VGT_VTX_VECT_EJECT_REG:
2790 case VGT_CACHE_INVALIDATION:
2791 case VGT_ESGS_RING_SIZE:
2792 case VGT_GSVS_RING_SIZE:
2793 case VGT_GS_VERTEX_REUSE:
2794 case VGT_PRIMITIVE_TYPE:
2795 case VGT_INDEX_TYPE:
2796 case VGT_NUM_INDICES:
2797 case VGT_NUM_INSTANCES:
2798 case VGT_TF_RING_SIZE:
2799 case VGT_HS_OFFCHIP_PARAM:
2800 case VGT_TF_MEMORY_BASE:
2802 case PA_SU_LINE_STIPPLE_VALUE:
2803 case PA_SC_LINE_STIPPLE_STATE:
2806 case SPI_STATIC_THREAD_MGMT_1:
2807 case SPI_STATIC_THREAD_MGMT_2:
2808 case SPI_STATIC_THREAD_MGMT_3:
2809 case SPI_PS_MAX_WAVE_ID:
2810 case SPI_CONFIG_CNTL:
2811 case SPI_CONFIG_CNTL_1:
2815 DRM_ERROR("Invalid register 0x%x in CS\n", reg);
2820 static int si_vm_packet3_ce_check(struct radeon_device *rdev,
2821 u32 *ib, struct radeon_cs_packet *pkt)
2823 switch (pkt->opcode) {
2825 case PACKET3_SET_BASE:
2826 case PACKET3_SET_CE_DE_COUNTERS:
2827 case PACKET3_LOAD_CONST_RAM:
2828 case PACKET3_WRITE_CONST_RAM:
2829 case PACKET3_WRITE_CONST_RAM_OFFSET:
2830 case PACKET3_DUMP_CONST_RAM:
2831 case PACKET3_INCREMENT_CE_COUNTER:
2832 case PACKET3_WAIT_ON_DE_COUNTER:
2833 case PACKET3_CE_WRITE:
2836 DRM_ERROR("Invalid CE packet3: 0x%x\n", pkt->opcode);
2842 static int si_vm_packet3_gfx_check(struct radeon_device *rdev,
2843 u32 *ib, struct radeon_cs_packet *pkt)
2845 u32 idx = pkt->idx + 1;
2846 u32 idx_value = ib[idx];
2847 u32 start_reg, end_reg, reg, i;
2850 switch (pkt->opcode) {
2852 case PACKET3_SET_BASE:
2853 case PACKET3_CLEAR_STATE:
2854 case PACKET3_INDEX_BUFFER_SIZE:
2855 case PACKET3_DISPATCH_DIRECT:
2856 case PACKET3_DISPATCH_INDIRECT:
2857 case PACKET3_ALLOC_GDS:
2858 case PACKET3_WRITE_GDS_RAM:
2859 case PACKET3_ATOMIC_GDS:
2860 case PACKET3_ATOMIC:
2861 case PACKET3_OCCLUSION_QUERY:
2862 case PACKET3_SET_PREDICATION:
2863 case PACKET3_COND_EXEC:
2864 case PACKET3_PRED_EXEC:
2865 case PACKET3_DRAW_INDIRECT:
2866 case PACKET3_DRAW_INDEX_INDIRECT:
2867 case PACKET3_INDEX_BASE:
2868 case PACKET3_DRAW_INDEX_2:
2869 case PACKET3_CONTEXT_CONTROL:
2870 case PACKET3_INDEX_TYPE:
2871 case PACKET3_DRAW_INDIRECT_MULTI:
2872 case PACKET3_DRAW_INDEX_AUTO:
2873 case PACKET3_DRAW_INDEX_IMMD:
2874 case PACKET3_NUM_INSTANCES:
2875 case PACKET3_DRAW_INDEX_MULTI_AUTO:
2876 case PACKET3_STRMOUT_BUFFER_UPDATE:
2877 case PACKET3_DRAW_INDEX_OFFSET_2:
2878 case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
2879 case PACKET3_DRAW_INDEX_INDIRECT_MULTI:
2880 case PACKET3_MPEG_INDEX:
2881 case PACKET3_WAIT_REG_MEM:
2882 case PACKET3_MEM_WRITE:
2883 case PACKET3_PFP_SYNC_ME:
2884 case PACKET3_SURFACE_SYNC:
2885 case PACKET3_EVENT_WRITE:
2886 case PACKET3_EVENT_WRITE_EOP:
2887 case PACKET3_EVENT_WRITE_EOS:
2888 case PACKET3_SET_CONTEXT_REG:
2889 case PACKET3_SET_CONTEXT_REG_INDIRECT:
2890 case PACKET3_SET_SH_REG:
2891 case PACKET3_SET_SH_REG_OFFSET:
2892 case PACKET3_INCREMENT_DE_COUNTER:
2893 case PACKET3_WAIT_ON_CE_COUNTER:
2894 case PACKET3_WAIT_ON_AVAIL_BUFFER:
2895 case PACKET3_ME_WRITE:
2897 case PACKET3_COPY_DATA:
2898 if ((idx_value & 0xf00) == 0) {
2899 reg = ib[idx + 3] * 4;
2900 if (!si_vm_reg_valid(reg))
2904 case PACKET3_WRITE_DATA:
2905 if ((idx_value & 0xf00) == 0) {
2906 start_reg = ib[idx + 1] * 4;
2907 if (idx_value & 0x10000) {
2908 if (!si_vm_reg_valid(start_reg))
2911 for (i = 0; i < (pkt->count - 2); i++) {
2912 reg = start_reg + (4 * i);
2913 if (!si_vm_reg_valid(reg))
2919 case PACKET3_COND_WRITE:
2920 if (idx_value & 0x100) {
2921 reg = ib[idx + 5] * 4;
2922 if (!si_vm_reg_valid(reg))
2926 case PACKET3_COPY_DW:
2927 if (idx_value & 0x2) {
2928 reg = ib[idx + 3] * 4;
2929 if (!si_vm_reg_valid(reg))
2933 case PACKET3_SET_CONFIG_REG:
2934 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2935 end_reg = 4 * pkt->count + start_reg - 4;
2936 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2937 (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2938 (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2939 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2942 for (i = 0; i < pkt->count; i++) {
2943 reg = start_reg + (4 * i);
2944 if (!si_vm_reg_valid(reg))
2948 case PACKET3_CP_DMA:
2949 command = ib[idx + 4];
2951 if (command & PACKET3_CP_DMA_CMD_SAS) {
2952 /* src address space is register */
2953 if (((info & 0x60000000) >> 29) == 0) {
2954 start_reg = idx_value << 2;
2955 if (command & PACKET3_CP_DMA_CMD_SAIC) {
2957 if (!si_vm_reg_valid(reg)) {
2958 DRM_ERROR("CP DMA Bad SRC register\n");
2962 for (i = 0; i < (command & 0x1fffff); i++) {
2963 reg = start_reg + (4 * i);
2964 if (!si_vm_reg_valid(reg)) {
2965 DRM_ERROR("CP DMA Bad SRC register\n");
2972 if (command & PACKET3_CP_DMA_CMD_DAS) {
2973 /* dst address space is register */
2974 if (((info & 0x00300000) >> 20) == 0) {
2975 start_reg = ib[idx + 2];
2976 if (command & PACKET3_CP_DMA_CMD_DAIC) {
2978 if (!si_vm_reg_valid(reg)) {
2979 DRM_ERROR("CP DMA Bad DST register\n");
2983 for (i = 0; i < (command & 0x1fffff); i++) {
2984 reg = start_reg + (4 * i);
2985 if (!si_vm_reg_valid(reg)) {
2986 DRM_ERROR("CP DMA Bad DST register\n");
2995 DRM_ERROR("Invalid GFX packet3: 0x%x\n", pkt->opcode);
3001 static int si_vm_packet3_compute_check(struct radeon_device *rdev,
3002 u32 *ib, struct radeon_cs_packet *pkt)
3004 u32 idx = pkt->idx + 1;
3005 u32 idx_value = ib[idx];
3006 u32 start_reg, reg, i;
3008 switch (pkt->opcode) {
3010 case PACKET3_SET_BASE:
3011 case PACKET3_CLEAR_STATE:
3012 case PACKET3_DISPATCH_DIRECT:
3013 case PACKET3_DISPATCH_INDIRECT:
3014 case PACKET3_ALLOC_GDS:
3015 case PACKET3_WRITE_GDS_RAM:
3016 case PACKET3_ATOMIC_GDS:
3017 case PACKET3_ATOMIC:
3018 case PACKET3_OCCLUSION_QUERY:
3019 case PACKET3_SET_PREDICATION:
3020 case PACKET3_COND_EXEC:
3021 case PACKET3_PRED_EXEC:
3022 case PACKET3_CONTEXT_CONTROL:
3023 case PACKET3_STRMOUT_BUFFER_UPDATE:
3024 case PACKET3_WAIT_REG_MEM:
3025 case PACKET3_MEM_WRITE:
3026 case PACKET3_PFP_SYNC_ME:
3027 case PACKET3_SURFACE_SYNC:
3028 case PACKET3_EVENT_WRITE:
3029 case PACKET3_EVENT_WRITE_EOP:
3030 case PACKET3_EVENT_WRITE_EOS:
3031 case PACKET3_SET_CONTEXT_REG:
3032 case PACKET3_SET_CONTEXT_REG_INDIRECT:
3033 case PACKET3_SET_SH_REG:
3034 case PACKET3_SET_SH_REG_OFFSET:
3035 case PACKET3_INCREMENT_DE_COUNTER:
3036 case PACKET3_WAIT_ON_CE_COUNTER:
3037 case PACKET3_WAIT_ON_AVAIL_BUFFER:
3038 case PACKET3_ME_WRITE:
3040 case PACKET3_COPY_DATA:
3041 if ((idx_value & 0xf00) == 0) {
3042 reg = ib[idx + 3] * 4;
3043 if (!si_vm_reg_valid(reg))
3047 case PACKET3_WRITE_DATA:
3048 if ((idx_value & 0xf00) == 0) {
3049 start_reg = ib[idx + 1] * 4;
3050 if (idx_value & 0x10000) {
3051 if (!si_vm_reg_valid(start_reg))
3054 for (i = 0; i < (pkt->count - 2); i++) {
3055 reg = start_reg + (4 * i);
3056 if (!si_vm_reg_valid(reg))
3062 case PACKET3_COND_WRITE:
3063 if (idx_value & 0x100) {
3064 reg = ib[idx + 5] * 4;
3065 if (!si_vm_reg_valid(reg))
3069 case PACKET3_COPY_DW:
3070 if (idx_value & 0x2) {
3071 reg = ib[idx + 3] * 4;
3072 if (!si_vm_reg_valid(reg))
3077 DRM_ERROR("Invalid Compute packet3: 0x%x\n", pkt->opcode);
3083 int si_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3087 struct radeon_cs_packet pkt;
3091 pkt.type = RADEON_CP_PACKET_GET_TYPE(ib->ptr[idx]);
3092 pkt.count = RADEON_CP_PACKET_GET_COUNT(ib->ptr[idx]);
3095 case RADEON_PACKET_TYPE0:
3096 dev_err(rdev->dev, "Packet0 not allowed!\n");
3099 case RADEON_PACKET_TYPE2:
3102 case RADEON_PACKET_TYPE3:
3103 pkt.opcode = RADEON_CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
3104 if (ib->is_const_ib)
3105 ret = si_vm_packet3_ce_check(rdev, ib->ptr, &pkt);
3108 case RADEON_RING_TYPE_GFX_INDEX:
3109 ret = si_vm_packet3_gfx_check(rdev, ib->ptr, &pkt);
3111 case CAYMAN_RING_TYPE_CP1_INDEX:
3112 case CAYMAN_RING_TYPE_CP2_INDEX:
3113 ret = si_vm_packet3_compute_check(rdev, ib->ptr, &pkt);
3116 dev_err(rdev->dev, "Non-PM4 ring %d !\n", ib->ring);
3121 idx += pkt.count + 2;
3124 dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
3130 } while (idx < ib->length_dw);
3138 int si_vm_init(struct radeon_device *rdev)
3141 rdev->vm_manager.nvm = 16;
3142 /* base offset of vram pages */
3143 rdev->vm_manager.vram_base_offset = 0;
3148 void si_vm_fini(struct radeon_device *rdev)
3153 * si_vm_set_page - update the page tables using the CP
3155 * @rdev: radeon_device pointer
3156 * @ib: indirect buffer to fill with commands
3157 * @pe: addr of the page entry
3158 * @addr: dst addr to write into pe
3159 * @count: number of page entries to update
3160 * @incr: increase next addr by incr bytes
3161 * @flags: access flags
3163 * Update the page tables using the CP (SI).
3165 void si_vm_set_page(struct radeon_device *rdev,
3166 struct radeon_ib *ib,
3168 uint64_t addr, unsigned count,
3169 uint32_t incr, uint32_t flags)
3171 uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
3175 if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) {
3177 ndw = 2 + count * 2;
3181 ib->ptr[ib->length_dw++] = PACKET3(PACKET3_WRITE_DATA, ndw);
3182 ib->ptr[ib->length_dw++] = (WRITE_DATA_ENGINE_SEL(0) |
3183 WRITE_DATA_DST_SEL(1));
3184 ib->ptr[ib->length_dw++] = pe;
3185 ib->ptr[ib->length_dw++] = upper_32_bits(pe);
3186 for (; ndw > 2; ndw -= 2, --count, pe += 8) {
3187 if (flags & RADEON_VM_PAGE_SYSTEM) {
3188 value = radeon_vm_map_gart(rdev, addr);
3189 value &= 0xFFFFFFFFFFFFF000ULL;
3190 } else if (flags & RADEON_VM_PAGE_VALID) {
3196 value |= r600_flags;
3197 ib->ptr[ib->length_dw++] = value;
3198 ib->ptr[ib->length_dw++] = upper_32_bits(value);
3203 if (flags & RADEON_VM_PAGE_SYSTEM) {
3209 /* for non-physically contiguous pages (system) */
3210 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw);
3211 ib->ptr[ib->length_dw++] = pe;
3212 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
3213 for (; ndw > 0; ndw -= 2, --count, pe += 8) {
3214 if (flags & RADEON_VM_PAGE_SYSTEM) {
3215 value = radeon_vm_map_gart(rdev, addr);
3216 value &= 0xFFFFFFFFFFFFF000ULL;
3217 } else if (flags & RADEON_VM_PAGE_VALID) {
3223 value |= r600_flags;
3224 ib->ptr[ib->length_dw++] = value;
3225 ib->ptr[ib->length_dw++] = upper_32_bits(value);
3234 if (flags & RADEON_VM_PAGE_VALID)
3238 /* for physically contiguous pages (vram) */
3239 ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw);
3240 ib->ptr[ib->length_dw++] = pe; /* dst addr */
3241 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
3242 ib->ptr[ib->length_dw++] = r600_flags; /* mask */
3243 ib->ptr[ib->length_dw++] = 0;
3244 ib->ptr[ib->length_dw++] = value; /* value */
3245 ib->ptr[ib->length_dw++] = upper_32_bits(value);
3246 ib->ptr[ib->length_dw++] = incr; /* increment size */
3247 ib->ptr[ib->length_dw++] = 0;
3249 addr += (ndw / 2) * incr;
3253 while (ib->length_dw & 0x7)
3254 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0);
3258 void si_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
3260 struct radeon_ring *ring = &rdev->ring[ridx];
3265 /* write new base address */
3266 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3267 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3268 WRITE_DATA_DST_SEL(0)));
3271 radeon_ring_write(ring,
3272 (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
3274 radeon_ring_write(ring,
3275 (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
3277 radeon_ring_write(ring, 0);
3278 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
3280 /* flush hdp cache */
3281 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3282 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3283 WRITE_DATA_DST_SEL(0)));
3284 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
3285 radeon_ring_write(ring, 0);
3286 radeon_ring_write(ring, 0x1);
3288 /* bits 0-15 are the VM contexts0-15 */
3289 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3290 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3291 WRITE_DATA_DST_SEL(0)));
3292 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
3293 radeon_ring_write(ring, 0);
3294 radeon_ring_write(ring, 1 << vm->id);
3296 /* sync PFP to ME, otherwise we might get invalid PFP reads */
3297 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3298 radeon_ring_write(ring, 0x0);
3301 void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
3303 struct radeon_ring *ring = &rdev->ring[ridx];
3308 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3310 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
3312 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2));
3314 radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
3316 /* flush hdp cache */
3317 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3318 radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
3319 radeon_ring_write(ring, 1);
3321 /* bits 0-7 are the VM contexts0-7 */
3322 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3323 radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
3324 radeon_ring_write(ring, 1 << vm->id);
3330 void si_rlc_fini(struct radeon_device *rdev)
3334 /* save restore block */
3335 if (rdev->rlc.save_restore_obj) {
3336 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
3337 if (unlikely(r != 0))
3338 dev_warn(rdev->dev, "(%d) reserve RLC sr bo failed\n", r);
3339 radeon_bo_unpin(rdev->rlc.save_restore_obj);
3340 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
3342 radeon_bo_unref(&rdev->rlc.save_restore_obj);
3343 rdev->rlc.save_restore_obj = NULL;
3346 /* clear state block */
3347 if (rdev->rlc.clear_state_obj) {
3348 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3349 if (unlikely(r != 0))
3350 dev_warn(rdev->dev, "(%d) reserve RLC c bo failed\n", r);
3351 radeon_bo_unpin(rdev->rlc.clear_state_obj);
3352 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3354 radeon_bo_unref(&rdev->rlc.clear_state_obj);
3355 rdev->rlc.clear_state_obj = NULL;
3359 int si_rlc_init(struct radeon_device *rdev)
3363 /* save restore block */
3364 if (rdev->rlc.save_restore_obj == NULL) {
3365 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3366 RADEON_GEM_DOMAIN_VRAM, NULL,
3367 &rdev->rlc.save_restore_obj);
3369 dev_warn(rdev->dev, "(%d) create RLC sr bo failed\n", r);
3374 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
3375 if (unlikely(r != 0)) {
3379 r = radeon_bo_pin(rdev->rlc.save_restore_obj, RADEON_GEM_DOMAIN_VRAM,
3380 &rdev->rlc.save_restore_gpu_addr);
3381 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
3383 dev_warn(rdev->dev, "(%d) pin RLC sr bo failed\n", r);
3388 /* clear state block */
3389 if (rdev->rlc.clear_state_obj == NULL) {
3390 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3391 RADEON_GEM_DOMAIN_VRAM, NULL,
3392 &rdev->rlc.clear_state_obj);
3394 dev_warn(rdev->dev, "(%d) create RLC c bo failed\n", r);
3399 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3400 if (unlikely(r != 0)) {
3404 r = radeon_bo_pin(rdev->rlc.clear_state_obj, RADEON_GEM_DOMAIN_VRAM,
3405 &rdev->rlc.clear_state_gpu_addr);
3406 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3408 dev_warn(rdev->dev, "(%d) pin RLC c bo failed\n", r);
3416 static void si_rlc_stop(struct radeon_device *rdev)
3418 WREG32(RLC_CNTL, 0);
3421 static void si_rlc_start(struct radeon_device *rdev)
3423 WREG32(RLC_CNTL, RLC_ENABLE);
3426 static int si_rlc_resume(struct radeon_device *rdev)
3429 const __be32 *fw_data;
3436 WREG32(RLC_RL_BASE, 0);
3437 WREG32(RLC_RL_SIZE, 0);
3438 WREG32(RLC_LB_CNTL, 0);
3439 WREG32(RLC_LB_CNTR_MAX, 0xffffffff);
3440 WREG32(RLC_LB_CNTR_INIT, 0);
3442 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
3443 WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8);
3445 WREG32(RLC_MC_CNTL, 0);
3446 WREG32(RLC_UCODE_CNTL, 0);
3448 fw_data = (const __be32 *)rdev->rlc_fw->data;
3449 for (i = 0; i < SI_RLC_UCODE_SIZE; i++) {
3450 WREG32(RLC_UCODE_ADDR, i);
3451 WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3453 WREG32(RLC_UCODE_ADDR, 0);
3460 static void si_enable_interrupts(struct radeon_device *rdev)
3462 u32 ih_cntl = RREG32(IH_CNTL);
3463 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3465 ih_cntl |= ENABLE_INTR;
3466 ih_rb_cntl |= IH_RB_ENABLE;
3467 WREG32(IH_CNTL, ih_cntl);
3468 WREG32(IH_RB_CNTL, ih_rb_cntl);
3469 rdev->ih.enabled = true;
3472 static void si_disable_interrupts(struct radeon_device *rdev)
3474 u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3475 u32 ih_cntl = RREG32(IH_CNTL);
3477 ih_rb_cntl &= ~IH_RB_ENABLE;
3478 ih_cntl &= ~ENABLE_INTR;
3479 WREG32(IH_RB_CNTL, ih_rb_cntl);
3480 WREG32(IH_CNTL, ih_cntl);
3481 /* set rptr, wptr to 0 */
3482 WREG32(IH_RB_RPTR, 0);
3483 WREG32(IH_RB_WPTR, 0);
3484 rdev->ih.enabled = false;
3488 static void si_disable_interrupt_state(struct radeon_device *rdev)
3492 WREG32(CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
3493 WREG32(CP_INT_CNTL_RING1, 0);
3494 WREG32(CP_INT_CNTL_RING2, 0);
3495 tmp = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
3496 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, tmp);
3497 tmp = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
3498 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, tmp);
3499 WREG32(GRBM_INT_CNTL, 0);
3500 WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
3501 WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
3502 if (rdev->num_crtc >= 4) {
3503 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
3504 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
3506 if (rdev->num_crtc >= 6) {
3507 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
3508 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
3511 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
3512 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
3513 if (rdev->num_crtc >= 4) {
3514 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
3515 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
3517 if (rdev->num_crtc >= 6) {
3518 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
3519 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
3522 WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
3524 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3525 WREG32(DC_HPD1_INT_CONTROL, tmp);
3526 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3527 WREG32(DC_HPD2_INT_CONTROL, tmp);
3528 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3529 WREG32(DC_HPD3_INT_CONTROL, tmp);
3530 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3531 WREG32(DC_HPD4_INT_CONTROL, tmp);
3532 tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3533 WREG32(DC_HPD5_INT_CONTROL, tmp);
3534 tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3535 WREG32(DC_HPD6_INT_CONTROL, tmp);
3539 static int si_irq_init(struct radeon_device *rdev)
3543 u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
3546 ret = r600_ih_ring_alloc(rdev);
3551 si_disable_interrupts(rdev);
3554 ret = si_rlc_resume(rdev);
3556 r600_ih_ring_fini(rdev);
3560 /* setup interrupt control */
3561 /* set dummy read address to ring address */
3562 WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
3563 interrupt_cntl = RREG32(INTERRUPT_CNTL);
3564 /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
3565 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
3567 interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
3568 /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
3569 interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
3570 WREG32(INTERRUPT_CNTL, interrupt_cntl);
3572 WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
3573 rb_bufsz = drm_order(rdev->ih.ring_size / 4);
3575 ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
3576 IH_WPTR_OVERFLOW_CLEAR |
3579 if (rdev->wb.enabled)
3580 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
3582 /* set the writeback address whether it's enabled or not */
3583 WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
3584 WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
3586 WREG32(IH_RB_CNTL, ih_rb_cntl);
3588 /* set rptr, wptr to 0 */
3589 WREG32(IH_RB_RPTR, 0);
3590 WREG32(IH_RB_WPTR, 0);
3592 /* Default settings for IH_CNTL (disabled at first) */
3593 ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0);
3594 /* RPTR_REARM only works if msi's are enabled */
3595 if (rdev->msi_enabled)
3596 ih_cntl |= RPTR_REARM;
3597 WREG32(IH_CNTL, ih_cntl);
3599 /* force the active interrupt state to all disabled */
3600 si_disable_interrupt_state(rdev);
3602 pci_set_master(rdev->pdev);
3605 si_enable_interrupts(rdev);
3610 int si_irq_set(struct radeon_device *rdev)
3612 u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
3613 u32 cp_int_cntl1 = 0, cp_int_cntl2 = 0;
3614 u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
3615 u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
3616 u32 grbm_int_cntl = 0;
3617 u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
3618 u32 dma_cntl, dma_cntl1;
3620 if (!rdev->irq.installed) {
3621 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
3624 /* don't enable anything if the ih is disabled */
3625 if (!rdev->ih.enabled) {
3626 si_disable_interrupts(rdev);
3627 /* force the active interrupt state to all disabled */
3628 si_disable_interrupt_state(rdev);
3632 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
3633 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
3634 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
3635 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
3636 hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
3637 hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
3639 dma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
3640 dma_cntl1 = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
3642 /* enable CP interrupts on all rings */
3643 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
3644 DRM_DEBUG("si_irq_set: sw int gfx\n");
3645 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
3647 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
3648 DRM_DEBUG("si_irq_set: sw int cp1\n");
3649 cp_int_cntl1 |= TIME_STAMP_INT_ENABLE;
3651 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
3652 DRM_DEBUG("si_irq_set: sw int cp2\n");
3653 cp_int_cntl2 |= TIME_STAMP_INT_ENABLE;
3655 if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
3656 DRM_DEBUG("si_irq_set: sw int dma\n");
3657 dma_cntl |= TRAP_ENABLE;
3660 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
3661 DRM_DEBUG("si_irq_set: sw int dma1\n");
3662 dma_cntl1 |= TRAP_ENABLE;
3664 if (rdev->irq.crtc_vblank_int[0] ||
3665 atomic_read(&rdev->irq.pflip[0])) {
3666 DRM_DEBUG("si_irq_set: vblank 0\n");
3667 crtc1 |= VBLANK_INT_MASK;
3669 if (rdev->irq.crtc_vblank_int[1] ||
3670 atomic_read(&rdev->irq.pflip[1])) {
3671 DRM_DEBUG("si_irq_set: vblank 1\n");
3672 crtc2 |= VBLANK_INT_MASK;
3674 if (rdev->irq.crtc_vblank_int[2] ||
3675 atomic_read(&rdev->irq.pflip[2])) {
3676 DRM_DEBUG("si_irq_set: vblank 2\n");
3677 crtc3 |= VBLANK_INT_MASK;
3679 if (rdev->irq.crtc_vblank_int[3] ||
3680 atomic_read(&rdev->irq.pflip[3])) {
3681 DRM_DEBUG("si_irq_set: vblank 3\n");
3682 crtc4 |= VBLANK_INT_MASK;
3684 if (rdev->irq.crtc_vblank_int[4] ||
3685 atomic_read(&rdev->irq.pflip[4])) {
3686 DRM_DEBUG("si_irq_set: vblank 4\n");
3687 crtc5 |= VBLANK_INT_MASK;
3689 if (rdev->irq.crtc_vblank_int[5] ||
3690 atomic_read(&rdev->irq.pflip[5])) {
3691 DRM_DEBUG("si_irq_set: vblank 5\n");
3692 crtc6 |= VBLANK_INT_MASK;
3694 if (rdev->irq.hpd[0]) {
3695 DRM_DEBUG("si_irq_set: hpd 1\n");
3696 hpd1 |= DC_HPDx_INT_EN;
3698 if (rdev->irq.hpd[1]) {
3699 DRM_DEBUG("si_irq_set: hpd 2\n");
3700 hpd2 |= DC_HPDx_INT_EN;
3702 if (rdev->irq.hpd[2]) {
3703 DRM_DEBUG("si_irq_set: hpd 3\n");
3704 hpd3 |= DC_HPDx_INT_EN;
3706 if (rdev->irq.hpd[3]) {
3707 DRM_DEBUG("si_irq_set: hpd 4\n");
3708 hpd4 |= DC_HPDx_INT_EN;
3710 if (rdev->irq.hpd[4]) {
3711 DRM_DEBUG("si_irq_set: hpd 5\n");
3712 hpd5 |= DC_HPDx_INT_EN;
3714 if (rdev->irq.hpd[5]) {
3715 DRM_DEBUG("si_irq_set: hpd 6\n");
3716 hpd6 |= DC_HPDx_INT_EN;
3719 WREG32(CP_INT_CNTL_RING0, cp_int_cntl);
3720 WREG32(CP_INT_CNTL_RING1, cp_int_cntl1);
3721 WREG32(CP_INT_CNTL_RING2, cp_int_cntl2);
3723 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, dma_cntl);
3724 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, dma_cntl1);
3726 WREG32(GRBM_INT_CNTL, grbm_int_cntl);
3728 WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
3729 WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
3730 if (rdev->num_crtc >= 4) {
3731 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
3732 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
3734 if (rdev->num_crtc >= 6) {
3735 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
3736 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
3739 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, grph1);
3740 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, grph2);
3741 if (rdev->num_crtc >= 4) {
3742 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, grph3);
3743 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, grph4);
3745 if (rdev->num_crtc >= 6) {
3746 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, grph5);
3747 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, grph6);
3750 WREG32(DC_HPD1_INT_CONTROL, hpd1);
3751 WREG32(DC_HPD2_INT_CONTROL, hpd2);
3752 WREG32(DC_HPD3_INT_CONTROL, hpd3);
3753 WREG32(DC_HPD4_INT_CONTROL, hpd4);
3754 WREG32(DC_HPD5_INT_CONTROL, hpd5);
3755 WREG32(DC_HPD6_INT_CONTROL, hpd6);
3760 static inline void si_irq_ack(struct radeon_device *rdev)
3764 rdev->irq.stat_regs.evergreen.disp_int = RREG32(DISP_INTERRUPT_STATUS);
3765 rdev->irq.stat_regs.evergreen.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
3766 rdev->irq.stat_regs.evergreen.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
3767 rdev->irq.stat_regs.evergreen.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
3768 rdev->irq.stat_regs.evergreen.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
3769 rdev->irq.stat_regs.evergreen.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
3770 rdev->irq.stat_regs.evergreen.d1grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
3771 rdev->irq.stat_regs.evergreen.d2grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
3772 if (rdev->num_crtc >= 4) {
3773 rdev->irq.stat_regs.evergreen.d3grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
3774 rdev->irq.stat_regs.evergreen.d4grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
3776 if (rdev->num_crtc >= 6) {
3777 rdev->irq.stat_regs.evergreen.d5grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
3778 rdev->irq.stat_regs.evergreen.d6grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
3781 if (rdev->irq.stat_regs.evergreen.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
3782 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3783 if (rdev->irq.stat_regs.evergreen.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
3784 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3785 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT)
3786 WREG32(VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
3787 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT)
3788 WREG32(VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
3789 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
3790 WREG32(VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
3791 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT)
3792 WREG32(VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
3794 if (rdev->num_crtc >= 4) {
3795 if (rdev->irq.stat_regs.evergreen.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
3796 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3797 if (rdev->irq.stat_regs.evergreen.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
3798 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3799 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
3800 WREG32(VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
3801 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
3802 WREG32(VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
3803 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
3804 WREG32(VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
3805 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
3806 WREG32(VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
3809 if (rdev->num_crtc >= 6) {
3810 if (rdev->irq.stat_regs.evergreen.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
3811 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3812 if (rdev->irq.stat_regs.evergreen.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
3813 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3814 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
3815 WREG32(VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
3816 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
3817 WREG32(VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
3818 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
3819 WREG32(VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
3820 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
3821 WREG32(VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
3824 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
3825 tmp = RREG32(DC_HPD1_INT_CONTROL);
3826 tmp |= DC_HPDx_INT_ACK;
3827 WREG32(DC_HPD1_INT_CONTROL, tmp);
3829 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
3830 tmp = RREG32(DC_HPD2_INT_CONTROL);
3831 tmp |= DC_HPDx_INT_ACK;
3832 WREG32(DC_HPD2_INT_CONTROL, tmp);
3834 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
3835 tmp = RREG32(DC_HPD3_INT_CONTROL);
3836 tmp |= DC_HPDx_INT_ACK;
3837 WREG32(DC_HPD3_INT_CONTROL, tmp);
3839 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
3840 tmp = RREG32(DC_HPD4_INT_CONTROL);
3841 tmp |= DC_HPDx_INT_ACK;
3842 WREG32(DC_HPD4_INT_CONTROL, tmp);
3844 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
3845 tmp = RREG32(DC_HPD5_INT_CONTROL);
3846 tmp |= DC_HPDx_INT_ACK;
3847 WREG32(DC_HPD5_INT_CONTROL, tmp);
3849 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
3850 tmp = RREG32(DC_HPD5_INT_CONTROL);
3851 tmp |= DC_HPDx_INT_ACK;
3852 WREG32(DC_HPD6_INT_CONTROL, tmp);
3856 static void si_irq_disable(struct radeon_device *rdev)
3858 si_disable_interrupts(rdev);
3859 /* Wait and acknowledge irq */
3862 si_disable_interrupt_state(rdev);
3865 static void si_irq_suspend(struct radeon_device *rdev)
3867 si_irq_disable(rdev);
3871 static void si_irq_fini(struct radeon_device *rdev)
3873 si_irq_suspend(rdev);
3874 r600_ih_ring_fini(rdev);
3877 static inline u32 si_get_ih_wptr(struct radeon_device *rdev)
3881 if (rdev->wb.enabled)
3882 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
3884 wptr = RREG32(IH_RB_WPTR);
3886 if (wptr & RB_OVERFLOW) {
3887 /* When a ring buffer overflow happen start parsing interrupt
3888 * from the last not overwritten vector (wptr + 16). Hopefully
3889 * this should allow us to catchup.
3891 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
3892 wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
3893 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
3894 tmp = RREG32(IH_RB_CNTL);
3895 tmp |= IH_WPTR_OVERFLOW_CLEAR;
3896 WREG32(IH_RB_CNTL, tmp);
3898 return (wptr & rdev->ih.ptr_mask);
3902 * Each IV ring entry is 128 bits:
3903 * [7:0] - interrupt source id
3905 * [59:32] - interrupt source data
3906 * [63:60] - reserved
3909 * [127:80] - reserved
3911 int si_irq_process(struct radeon_device *rdev)
3915 u32 src_id, src_data, ring_id;
3917 bool queue_hotplug = false;
3919 if (!rdev->ih.enabled || rdev->shutdown)
3922 wptr = si_get_ih_wptr(rdev);
3925 /* is somebody else already processing irqs? */
3926 if (atomic_xchg(&rdev->ih.lock, 1))
3929 rptr = rdev->ih.rptr;
3930 DRM_DEBUG("si_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
3932 /* Order reading of wptr vs. reading of IH ring data */
3935 /* display interrupts */
3938 while (rptr != wptr) {
3939 /* wptr/rptr are in bytes! */
3940 ring_index = rptr / 4;
3941 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
3942 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
3943 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
3946 case 1: /* D1 vblank/vline */
3948 case 0: /* D1 vblank */
3949 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT) {
3950 if (rdev->irq.crtc_vblank_int[0]) {
3951 drm_handle_vblank(rdev->ddev, 0);
3952 rdev->pm.vblank_sync = true;
3953 wake_up(&rdev->irq.vblank_queue);
3955 if (atomic_read(&rdev->irq.pflip[0]))
3956 radeon_crtc_handle_flip(rdev, 0);
3957 rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
3958 DRM_DEBUG("IH: D1 vblank\n");
3961 case 1: /* D1 vline */
3962 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT) {
3963 rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VLINE_INTERRUPT;
3964 DRM_DEBUG("IH: D1 vline\n");
3968 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3972 case 2: /* D2 vblank/vline */
3974 case 0: /* D2 vblank */
3975 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
3976 if (rdev->irq.crtc_vblank_int[1]) {
3977 drm_handle_vblank(rdev->ddev, 1);
3978 rdev->pm.vblank_sync = true;
3979 wake_up(&rdev->irq.vblank_queue);
3981 if (atomic_read(&rdev->irq.pflip[1]))
3982 radeon_crtc_handle_flip(rdev, 1);
3983 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
3984 DRM_DEBUG("IH: D2 vblank\n");
3987 case 1: /* D2 vline */
3988 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
3989 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
3990 DRM_DEBUG("IH: D2 vline\n");
3994 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3998 case 3: /* D3 vblank/vline */
4000 case 0: /* D3 vblank */
4001 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
4002 if (rdev->irq.crtc_vblank_int[2]) {
4003 drm_handle_vblank(rdev->ddev, 2);
4004 rdev->pm.vblank_sync = true;
4005 wake_up(&rdev->irq.vblank_queue);
4007 if (atomic_read(&rdev->irq.pflip[2]))
4008 radeon_crtc_handle_flip(rdev, 2);
4009 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
4010 DRM_DEBUG("IH: D3 vblank\n");
4013 case 1: /* D3 vline */
4014 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
4015 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
4016 DRM_DEBUG("IH: D3 vline\n");
4020 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4024 case 4: /* D4 vblank/vline */
4026 case 0: /* D4 vblank */
4027 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
4028 if (rdev->irq.crtc_vblank_int[3]) {
4029 drm_handle_vblank(rdev->ddev, 3);
4030 rdev->pm.vblank_sync = true;
4031 wake_up(&rdev->irq.vblank_queue);
4033 if (atomic_read(&rdev->irq.pflip[3]))
4034 radeon_crtc_handle_flip(rdev, 3);
4035 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
4036 DRM_DEBUG("IH: D4 vblank\n");
4039 case 1: /* D4 vline */
4040 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
4041 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
4042 DRM_DEBUG("IH: D4 vline\n");
4046 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4050 case 5: /* D5 vblank/vline */
4052 case 0: /* D5 vblank */
4053 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
4054 if (rdev->irq.crtc_vblank_int[4]) {
4055 drm_handle_vblank(rdev->ddev, 4);
4056 rdev->pm.vblank_sync = true;
4057 wake_up(&rdev->irq.vblank_queue);
4059 if (atomic_read(&rdev->irq.pflip[4]))
4060 radeon_crtc_handle_flip(rdev, 4);
4061 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
4062 DRM_DEBUG("IH: D5 vblank\n");
4065 case 1: /* D5 vline */
4066 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
4067 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
4068 DRM_DEBUG("IH: D5 vline\n");
4072 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4076 case 6: /* D6 vblank/vline */
4078 case 0: /* D6 vblank */
4079 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
4080 if (rdev->irq.crtc_vblank_int[5]) {
4081 drm_handle_vblank(rdev->ddev, 5);
4082 rdev->pm.vblank_sync = true;
4083 wake_up(&rdev->irq.vblank_queue);
4085 if (atomic_read(&rdev->irq.pflip[5]))
4086 radeon_crtc_handle_flip(rdev, 5);
4087 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
4088 DRM_DEBUG("IH: D6 vblank\n");
4091 case 1: /* D6 vline */
4092 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
4093 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
4094 DRM_DEBUG("IH: D6 vline\n");
4098 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4102 case 42: /* HPD hotplug */
4105 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
4106 rdev->irq.stat_regs.evergreen.disp_int &= ~DC_HPD1_INTERRUPT;
4107 queue_hotplug = true;
4108 DRM_DEBUG("IH: HPD1\n");
4112 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
4113 rdev->irq.stat_regs.evergreen.disp_int_cont &= ~DC_HPD2_INTERRUPT;
4114 queue_hotplug = true;
4115 DRM_DEBUG("IH: HPD2\n");
4119 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
4120 rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
4121 queue_hotplug = true;
4122 DRM_DEBUG("IH: HPD3\n");
4126 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
4127 rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
4128 queue_hotplug = true;
4129 DRM_DEBUG("IH: HPD4\n");
4133 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
4134 rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
4135 queue_hotplug = true;
4136 DRM_DEBUG("IH: HPD5\n");
4140 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
4141 rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
4142 queue_hotplug = true;
4143 DRM_DEBUG("IH: HPD6\n");
4147 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4153 dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
4154 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n",
4155 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
4156 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
4157 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
4158 /* reset addr and status */
4159 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
4161 case 176: /* RINGID0 CP_INT */
4162 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4164 case 177: /* RINGID1 CP_INT */
4165 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4167 case 178: /* RINGID2 CP_INT */
4168 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4170 case 181: /* CP EOP event */
4171 DRM_DEBUG("IH: CP EOP\n");
4174 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4177 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4180 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4184 case 224: /* DMA trap event */
4185 DRM_DEBUG("IH: DMA trap\n");
4186 radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
4188 case 233: /* GUI IDLE */
4189 DRM_DEBUG("IH: GUI idle\n");
4191 case 244: /* DMA trap event */
4192 DRM_DEBUG("IH: DMA1 trap\n");
4193 radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
4196 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4200 /* wptr/rptr are in bytes! */
4202 rptr &= rdev->ih.ptr_mask;
4205 schedule_work(&rdev->hotplug_work);
4206 rdev->ih.rptr = rptr;
4207 WREG32(IH_RB_RPTR, rdev->ih.rptr);
4208 atomic_set(&rdev->ih.lock, 0);
4210 /* make sure wptr hasn't changed while processing */
4211 wptr = si_get_ih_wptr(rdev);
4219 * si_copy_dma - copy pages using the DMA engine
4221 * @rdev: radeon_device pointer
4222 * @src_offset: src GPU address
4223 * @dst_offset: dst GPU address
4224 * @num_gpu_pages: number of GPU pages to xfer
4225 * @fence: radeon fence object
4227 * Copy GPU paging using the DMA engine (SI).
4228 * Used by the radeon ttm implementation to move pages if
4229 * registered as the asic copy callback.
4231 int si_copy_dma(struct radeon_device *rdev,
4232 uint64_t src_offset, uint64_t dst_offset,
4233 unsigned num_gpu_pages,
4234 struct radeon_fence **fence)
4236 struct radeon_semaphore *sem = NULL;
4237 int ring_index = rdev->asic->copy.dma_ring_index;
4238 struct radeon_ring *ring = &rdev->ring[ring_index];
4239 u32 size_in_bytes, cur_size_in_bytes;
4243 r = radeon_semaphore_create(rdev, &sem);
4245 DRM_ERROR("radeon: moving bo (%d).\n", r);
4249 size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
4250 num_loops = DIV_ROUND_UP(size_in_bytes, 0xfffff);
4251 r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
4253 DRM_ERROR("radeon: moving bo (%d).\n", r);
4254 radeon_semaphore_free(rdev, &sem, NULL);
4258 if (radeon_fence_need_sync(*fence, ring->idx)) {
4259 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
4261 radeon_fence_note_sync(*fence, ring->idx);
4263 radeon_semaphore_free(rdev, &sem, NULL);
4266 for (i = 0; i < num_loops; i++) {
4267 cur_size_in_bytes = size_in_bytes;
4268 if (cur_size_in_bytes > 0xFFFFF)
4269 cur_size_in_bytes = 0xFFFFF;
4270 size_in_bytes -= cur_size_in_bytes;
4271 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 1, 0, 0, cur_size_in_bytes));
4272 radeon_ring_write(ring, dst_offset & 0xffffffff);
4273 radeon_ring_write(ring, src_offset & 0xffffffff);
4274 radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
4275 radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
4276 src_offset += cur_size_in_bytes;
4277 dst_offset += cur_size_in_bytes;
4280 r = radeon_fence_emit(rdev, fence, ring->idx);
4282 radeon_ring_unlock_undo(rdev, ring);
4286 radeon_ring_unlock_commit(rdev, ring);
4287 radeon_semaphore_free(rdev, &sem, *fence);
4293 * startup/shutdown callbacks
4295 static int si_startup(struct radeon_device *rdev)
4297 struct radeon_ring *ring;
4300 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
4301 !rdev->rlc_fw || !rdev->mc_fw) {
4302 r = si_init_microcode(rdev);
4304 DRM_ERROR("Failed to load firmware!\n");
4309 r = si_mc_load_microcode(rdev);
4311 DRM_ERROR("Failed to load MC firmware!\n");
4315 r = r600_vram_scratch_init(rdev);
4319 si_mc_program(rdev);
4320 r = si_pcie_gart_enable(rdev);
4326 r = evergreen_blit_init(rdev);
4328 r600_blit_fini(rdev);
4329 rdev->asic->copy = NULL;
4330 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
4333 /* allocate rlc buffers */
4334 r = si_rlc_init(rdev);
4336 DRM_ERROR("Failed to init rlc BOs!\n");
4340 /* allocate wb buffer */
4341 r = radeon_wb_init(rdev);
4345 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
4347 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4351 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4353 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4357 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4359 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4363 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
4365 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
4369 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
4371 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
4376 r = si_irq_init(rdev);
4378 DRM_ERROR("radeon: IH init failed (%d).\n", r);
4379 radeon_irq_kms_fini(rdev);
4384 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4385 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
4386 CP_RB0_RPTR, CP_RB0_WPTR,
4387 0, 0xfffff, RADEON_CP_PACKET2);
4391 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
4392 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET,
4393 CP_RB1_RPTR, CP_RB1_WPTR,
4394 0, 0xfffff, RADEON_CP_PACKET2);
4398 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
4399 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET,
4400 CP_RB2_RPTR, CP_RB2_WPTR,
4401 0, 0xfffff, RADEON_CP_PACKET2);
4405 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4406 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
4407 DMA_RB_RPTR + DMA0_REGISTER_OFFSET,
4408 DMA_RB_WPTR + DMA0_REGISTER_OFFSET,
4409 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
4413 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4414 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
4415 DMA_RB_RPTR + DMA1_REGISTER_OFFSET,
4416 DMA_RB_WPTR + DMA1_REGISTER_OFFSET,
4417 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
4421 r = si_cp_load_microcode(rdev);
4424 r = si_cp_resume(rdev);
4428 r = cayman_dma_resume(rdev);
4432 r = radeon_ib_pool_init(rdev);
4434 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
4438 r = radeon_vm_manager_init(rdev);
4440 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
4447 int si_resume(struct radeon_device *rdev)
4451 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
4452 * posting will perform necessary task to bring back GPU into good
4456 atom_asic_init(rdev->mode_info.atom_context);
4458 rdev->accel_working = true;
4459 r = si_startup(rdev);
4461 DRM_ERROR("si startup failed on resume\n");
4462 rdev->accel_working = false;
4470 int si_suspend(struct radeon_device *rdev)
4472 radeon_vm_manager_fini(rdev);
4473 si_cp_enable(rdev, false);
4474 cayman_dma_stop(rdev);
4475 si_irq_suspend(rdev);
4476 radeon_wb_disable(rdev);
4477 si_pcie_gart_disable(rdev);
4481 /* Plan is to move initialization in that function and use
4482 * helper function so that radeon_device_init pretty much
4483 * do nothing more than calling asic specific function. This
4484 * should also allow to remove a bunch of callback function
4487 int si_init(struct radeon_device *rdev)
4489 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4493 if (!radeon_get_bios(rdev)) {
4494 if (ASIC_IS_AVIVO(rdev))
4497 /* Must be an ATOMBIOS */
4498 if (!rdev->is_atom_bios) {
4499 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
4502 r = radeon_atombios_init(rdev);
4506 /* Post card if necessary */
4507 if (!radeon_card_posted(rdev)) {
4509 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
4512 DRM_INFO("GPU not posted. posting now...\n");
4513 atom_asic_init(rdev->mode_info.atom_context);
4515 /* Initialize scratch registers */
4516 si_scratch_init(rdev);
4517 /* Initialize surface registers */
4518 radeon_surface_init(rdev);
4519 /* Initialize clocks */
4520 radeon_get_clock_info(rdev->ddev);
4523 r = radeon_fence_driver_init(rdev);
4527 /* initialize memory controller */
4528 r = si_mc_init(rdev);
4531 /* Memory manager */
4532 r = radeon_bo_init(rdev);
4536 r = radeon_irq_kms_init(rdev);
4540 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4541 ring->ring_obj = NULL;
4542 r600_ring_init(rdev, ring, 1024 * 1024);
4544 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
4545 ring->ring_obj = NULL;
4546 r600_ring_init(rdev, ring, 1024 * 1024);
4548 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
4549 ring->ring_obj = NULL;
4550 r600_ring_init(rdev, ring, 1024 * 1024);
4552 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4553 ring->ring_obj = NULL;
4554 r600_ring_init(rdev, ring, 64 * 1024);
4556 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4557 ring->ring_obj = NULL;
4558 r600_ring_init(rdev, ring, 64 * 1024);
4560 rdev->ih.ring_obj = NULL;
4561 r600_ih_ring_init(rdev, 64 * 1024);
4563 r = r600_pcie_gart_init(rdev);
4567 rdev->accel_working = true;
4568 r = si_startup(rdev);
4570 dev_err(rdev->dev, "disabling GPU acceleration\n");
4572 cayman_dma_fini(rdev);
4575 radeon_wb_fini(rdev);
4576 radeon_ib_pool_fini(rdev);
4577 radeon_vm_manager_fini(rdev);
4578 radeon_irq_kms_fini(rdev);
4579 si_pcie_gart_fini(rdev);
4580 rdev->accel_working = false;
4583 /* Don't start up if the MC ucode is missing.
4584 * The default clocks and voltages before the MC ucode
4585 * is loaded are not suffient for advanced operations.
4588 DRM_ERROR("radeon: MC ucode required for NI+.\n");
4595 void si_fini(struct radeon_device *rdev)
4598 r600_blit_fini(rdev);
4601 cayman_dma_fini(rdev);
4604 radeon_wb_fini(rdev);
4605 radeon_vm_manager_fini(rdev);
4606 radeon_ib_pool_fini(rdev);
4607 radeon_irq_kms_fini(rdev);
4608 si_pcie_gart_fini(rdev);
4609 r600_vram_scratch_fini(rdev);
4610 radeon_gem_fini(rdev);
4611 radeon_fence_driver_fini(rdev);
4612 radeon_bo_fini(rdev);
4613 radeon_atombios_fini(rdev);
4619 * si_get_gpu_clock_counter - return GPU clock counter snapshot
4621 * @rdev: radeon_device pointer
4623 * Fetches a GPU clock counter snapshot (SI).
4624 * Returns the 64 bit clock counter snapshot.
4626 uint64_t si_get_gpu_clock_counter(struct radeon_device *rdev)
4630 mutex_lock(&rdev->gpu_clock_mutex);
4631 WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4632 clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
4633 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4634 mutex_unlock(&rdev->gpu_clock_mutex);