Commit | Line | Data |
---|---|---|
8cc1a532 AD |
1 | /* |
2 | * Copyright 2012 Advanced Micro Devices, Inc. | |
3 | * | |
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: | |
10 | * | |
11 | * The above copyright notice and this permission notice shall be included in | |
12 | * all copies or substantial portions of the Software. | |
13 | * | |
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. | |
21 | * | |
22 | * Authors: Alex Deucher | |
23 | */ | |
24 | #include <linux/firmware.h> | |
25 | #include <linux/platform_device.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/module.h> | |
28 | #include "drmP.h" | |
29 | #include "radeon.h" | |
6f2043ce | 30 | #include "radeon_asic.h" |
8cc1a532 AD |
31 | #include "cikd.h" |
32 | #include "atom.h" | |
841cf442 | 33 | #include "cik_blit_shaders.h" |
8cc1a532 | 34 | |
02c81327 AD |
35 | /* GFX */ |
36 | #define CIK_PFP_UCODE_SIZE 2144 | |
37 | #define CIK_ME_UCODE_SIZE 2144 | |
38 | #define CIK_CE_UCODE_SIZE 2144 | |
39 | /* compute */ | |
40 | #define CIK_MEC_UCODE_SIZE 4192 | |
41 | /* interrupts */ | |
42 | #define BONAIRE_RLC_UCODE_SIZE 2048 | |
43 | #define KB_RLC_UCODE_SIZE 2560 | |
44 | #define KV_RLC_UCODE_SIZE 2560 | |
45 | /* gddr controller */ | |
46 | #define CIK_MC_UCODE_SIZE 7866 | |
21a93e13 AD |
47 | /* sdma */ |
48 | #define CIK_SDMA_UCODE_SIZE 1050 | |
49 | #define CIK_SDMA_UCODE_VERSION 64 | |
02c81327 AD |
50 | |
51 | MODULE_FIRMWARE("radeon/BONAIRE_pfp.bin"); | |
52 | MODULE_FIRMWARE("radeon/BONAIRE_me.bin"); | |
53 | MODULE_FIRMWARE("radeon/BONAIRE_ce.bin"); | |
54 | MODULE_FIRMWARE("radeon/BONAIRE_mec.bin"); | |
55 | MODULE_FIRMWARE("radeon/BONAIRE_mc.bin"); | |
56 | MODULE_FIRMWARE("radeon/BONAIRE_rlc.bin"); | |
21a93e13 | 57 | MODULE_FIRMWARE("radeon/BONAIRE_sdma.bin"); |
02c81327 AD |
58 | MODULE_FIRMWARE("radeon/KAVERI_pfp.bin"); |
59 | MODULE_FIRMWARE("radeon/KAVERI_me.bin"); | |
60 | MODULE_FIRMWARE("radeon/KAVERI_ce.bin"); | |
61 | MODULE_FIRMWARE("radeon/KAVERI_mec.bin"); | |
62 | MODULE_FIRMWARE("radeon/KAVERI_rlc.bin"); | |
21a93e13 | 63 | MODULE_FIRMWARE("radeon/KAVERI_sdma.bin"); |
02c81327 AD |
64 | MODULE_FIRMWARE("radeon/KABINI_pfp.bin"); |
65 | MODULE_FIRMWARE("radeon/KABINI_me.bin"); | |
66 | MODULE_FIRMWARE("radeon/KABINI_ce.bin"); | |
67 | MODULE_FIRMWARE("radeon/KABINI_mec.bin"); | |
68 | MODULE_FIRMWARE("radeon/KABINI_rlc.bin"); | |
21a93e13 | 69 | MODULE_FIRMWARE("radeon/KABINI_sdma.bin"); |
02c81327 | 70 | |
a59781bb AD |
71 | extern int r600_ih_ring_alloc(struct radeon_device *rdev); |
72 | extern void r600_ih_ring_fini(struct radeon_device *rdev); | |
6f2043ce AD |
73 | extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save); |
74 | extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save); | |
cc066715 | 75 | extern bool evergreen_is_display_hung(struct radeon_device *rdev); |
1c49165d | 76 | extern void si_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc); |
7bf94a2c AD |
77 | extern void si_rlc_fini(struct radeon_device *rdev); |
78 | extern int si_rlc_init(struct radeon_device *rdev); | |
cc066715 | 79 | static void cik_rlc_stop(struct radeon_device *rdev); |
6f2043ce | 80 | |
6e2c3c0a AD |
81 | /* |
82 | * Indirect registers accessor | |
83 | */ | |
84 | u32 cik_pciep_rreg(struct radeon_device *rdev, u32 reg) | |
85 | { | |
86 | u32 r; | |
87 | ||
88 | WREG32(PCIE_INDEX, reg); | |
89 | (void)RREG32(PCIE_INDEX); | |
90 | r = RREG32(PCIE_DATA); | |
91 | return r; | |
92 | } | |
93 | ||
94 | void cik_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v) | |
95 | { | |
96 | WREG32(PCIE_INDEX, reg); | |
97 | (void)RREG32(PCIE_INDEX); | |
98 | WREG32(PCIE_DATA, v); | |
99 | (void)RREG32(PCIE_DATA); | |
100 | } | |
101 | ||
2c67912c AD |
102 | /** |
103 | * cik_get_xclk - get the xclk | |
104 | * | |
105 | * @rdev: radeon_device pointer | |
106 | * | |
107 | * Returns the reference clock used by the gfx engine | |
108 | * (CIK). | |
109 | */ | |
110 | u32 cik_get_xclk(struct radeon_device *rdev) | |
111 | { | |
112 | u32 reference_clock = rdev->clock.spll.reference_freq; | |
113 | ||
114 | if (rdev->flags & RADEON_IS_IGP) { | |
115 | if (RREG32_SMC(GENERAL_PWRMGT) & GPU_COUNTER_CLK) | |
116 | return reference_clock / 2; | |
117 | } else { | |
118 | if (RREG32_SMC(CG_CLKPIN_CNTL) & XTALIN_DIVIDE) | |
119 | return reference_clock / 4; | |
120 | } | |
121 | return reference_clock; | |
122 | } | |
123 | ||
75efdee1 AD |
124 | /** |
125 | * cik_mm_rdoorbell - read a doorbell dword | |
126 | * | |
127 | * @rdev: radeon_device pointer | |
128 | * @offset: byte offset into the aperture | |
129 | * | |
130 | * Returns the value in the doorbell aperture at the | |
131 | * requested offset (CIK). | |
132 | */ | |
133 | u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset) | |
134 | { | |
135 | if (offset < rdev->doorbell.size) { | |
136 | return readl(((void __iomem *)rdev->doorbell.ptr) + offset); | |
137 | } else { | |
138 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", offset); | |
139 | return 0; | |
140 | } | |
141 | } | |
142 | ||
143 | /** | |
144 | * cik_mm_wdoorbell - write a doorbell dword | |
145 | * | |
146 | * @rdev: radeon_device pointer | |
147 | * @offset: byte offset into the aperture | |
148 | * @v: value to write | |
149 | * | |
150 | * Writes @v to the doorbell aperture at the | |
151 | * requested offset (CIK). | |
152 | */ | |
153 | void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v) | |
154 | { | |
155 | if (offset < rdev->doorbell.size) { | |
156 | writel(v, ((void __iomem *)rdev->doorbell.ptr) + offset); | |
157 | } else { | |
158 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", offset); | |
159 | } | |
160 | } | |
161 | ||
bc8273fe AD |
162 | #define BONAIRE_IO_MC_REGS_SIZE 36 |
163 | ||
164 | static const u32 bonaire_io_mc_regs[BONAIRE_IO_MC_REGS_SIZE][2] = | |
165 | { | |
166 | {0x00000070, 0x04400000}, | |
167 | {0x00000071, 0x80c01803}, | |
168 | {0x00000072, 0x00004004}, | |
169 | {0x00000073, 0x00000100}, | |
170 | {0x00000074, 0x00ff0000}, | |
171 | {0x00000075, 0x34000000}, | |
172 | {0x00000076, 0x08000014}, | |
173 | {0x00000077, 0x00cc08ec}, | |
174 | {0x00000078, 0x00000400}, | |
175 | {0x00000079, 0x00000000}, | |
176 | {0x0000007a, 0x04090000}, | |
177 | {0x0000007c, 0x00000000}, | |
178 | {0x0000007e, 0x4408a8e8}, | |
179 | {0x0000007f, 0x00000304}, | |
180 | {0x00000080, 0x00000000}, | |
181 | {0x00000082, 0x00000001}, | |
182 | {0x00000083, 0x00000002}, | |
183 | {0x00000084, 0xf3e4f400}, | |
184 | {0x00000085, 0x052024e3}, | |
185 | {0x00000087, 0x00000000}, | |
186 | {0x00000088, 0x01000000}, | |
187 | {0x0000008a, 0x1c0a0000}, | |
188 | {0x0000008b, 0xff010000}, | |
189 | {0x0000008d, 0xffffefff}, | |
190 | {0x0000008e, 0xfff3efff}, | |
191 | {0x0000008f, 0xfff3efbf}, | |
192 | {0x00000092, 0xf7ffffff}, | |
193 | {0x00000093, 0xffffff7f}, | |
194 | {0x00000095, 0x00101101}, | |
195 | {0x00000096, 0x00000fff}, | |
196 | {0x00000097, 0x00116fff}, | |
197 | {0x00000098, 0x60010000}, | |
198 | {0x00000099, 0x10010000}, | |
199 | {0x0000009a, 0x00006000}, | |
200 | {0x0000009b, 0x00001000}, | |
201 | {0x0000009f, 0x00b48000} | |
202 | }; | |
203 | ||
b556b12e AD |
204 | /** |
205 | * cik_srbm_select - select specific register instances | |
206 | * | |
207 | * @rdev: radeon_device pointer | |
208 | * @me: selected ME (micro engine) | |
209 | * @pipe: pipe | |
210 | * @queue: queue | |
211 | * @vmid: VMID | |
212 | * | |
213 | * Switches the currently active registers instances. Some | |
214 | * registers are instanced per VMID, others are instanced per | |
215 | * me/pipe/queue combination. | |
216 | */ | |
217 | static void cik_srbm_select(struct radeon_device *rdev, | |
218 | u32 me, u32 pipe, u32 queue, u32 vmid) | |
219 | { | |
220 | u32 srbm_gfx_cntl = (PIPEID(pipe & 0x3) | | |
221 | MEID(me & 0x3) | | |
222 | VMID(vmid & 0xf) | | |
223 | QUEUEID(queue & 0x7)); | |
224 | WREG32(SRBM_GFX_CNTL, srbm_gfx_cntl); | |
225 | } | |
226 | ||
bc8273fe AD |
227 | /* ucode loading */ |
228 | /** | |
229 | * ci_mc_load_microcode - load MC ucode into the hw | |
230 | * | |
231 | * @rdev: radeon_device pointer | |
232 | * | |
233 | * Load the GDDR MC ucode into the hw (CIK). | |
234 | * Returns 0 on success, error on failure. | |
235 | */ | |
236 | static int ci_mc_load_microcode(struct radeon_device *rdev) | |
237 | { | |
238 | const __be32 *fw_data; | |
239 | u32 running, blackout = 0; | |
240 | u32 *io_mc_regs; | |
241 | int i, ucode_size, regs_size; | |
242 | ||
243 | if (!rdev->mc_fw) | |
244 | return -EINVAL; | |
245 | ||
246 | switch (rdev->family) { | |
247 | case CHIP_BONAIRE: | |
248 | default: | |
249 | io_mc_regs = (u32 *)&bonaire_io_mc_regs; | |
250 | ucode_size = CIK_MC_UCODE_SIZE; | |
251 | regs_size = BONAIRE_IO_MC_REGS_SIZE; | |
252 | break; | |
253 | } | |
254 | ||
255 | running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK; | |
256 | ||
257 | if (running == 0) { | |
258 | if (running) { | |
259 | blackout = RREG32(MC_SHARED_BLACKOUT_CNTL); | |
260 | WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1); | |
261 | } | |
262 | ||
263 | /* reset the engine and set to writable */ | |
264 | WREG32(MC_SEQ_SUP_CNTL, 0x00000008); | |
265 | WREG32(MC_SEQ_SUP_CNTL, 0x00000010); | |
266 | ||
267 | /* load mc io regs */ | |
268 | for (i = 0; i < regs_size; i++) { | |
269 | WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]); | |
270 | WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]); | |
271 | } | |
272 | /* load the MC ucode */ | |
273 | fw_data = (const __be32 *)rdev->mc_fw->data; | |
274 | for (i = 0; i < ucode_size; i++) | |
275 | WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++)); | |
276 | ||
277 | /* put the engine back into the active state */ | |
278 | WREG32(MC_SEQ_SUP_CNTL, 0x00000008); | |
279 | WREG32(MC_SEQ_SUP_CNTL, 0x00000004); | |
280 | WREG32(MC_SEQ_SUP_CNTL, 0x00000001); | |
281 | ||
282 | /* wait for training to complete */ | |
283 | for (i = 0; i < rdev->usec_timeout; i++) { | |
284 | if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0) | |
285 | break; | |
286 | udelay(1); | |
287 | } | |
288 | for (i = 0; i < rdev->usec_timeout; i++) { | |
289 | if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1) | |
290 | break; | |
291 | udelay(1); | |
292 | } | |
293 | ||
294 | if (running) | |
295 | WREG32(MC_SHARED_BLACKOUT_CNTL, blackout); | |
296 | } | |
297 | ||
298 | return 0; | |
299 | } | |
300 | ||
02c81327 AD |
301 | /** |
302 | * cik_init_microcode - load ucode images from disk | |
303 | * | |
304 | * @rdev: radeon_device pointer | |
305 | * | |
306 | * Use the firmware interface to load the ucode images into | |
307 | * the driver (not loaded into hw). | |
308 | * Returns 0 on success, error on failure. | |
309 | */ | |
310 | static int cik_init_microcode(struct radeon_device *rdev) | |
311 | { | |
312 | struct platform_device *pdev; | |
313 | const char *chip_name; | |
314 | size_t pfp_req_size, me_req_size, ce_req_size, | |
21a93e13 AD |
315 | mec_req_size, rlc_req_size, mc_req_size, |
316 | sdma_req_size; | |
02c81327 AD |
317 | char fw_name[30]; |
318 | int err; | |
319 | ||
320 | DRM_DEBUG("\n"); | |
321 | ||
322 | pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0); | |
323 | err = IS_ERR(pdev); | |
324 | if (err) { | |
325 | printk(KERN_ERR "radeon_cp: Failed to register firmware\n"); | |
326 | return -EINVAL; | |
327 | } | |
328 | ||
329 | switch (rdev->family) { | |
330 | case CHIP_BONAIRE: | |
331 | chip_name = "BONAIRE"; | |
332 | pfp_req_size = CIK_PFP_UCODE_SIZE * 4; | |
333 | me_req_size = CIK_ME_UCODE_SIZE * 4; | |
334 | ce_req_size = CIK_CE_UCODE_SIZE * 4; | |
335 | mec_req_size = CIK_MEC_UCODE_SIZE * 4; | |
336 | rlc_req_size = BONAIRE_RLC_UCODE_SIZE * 4; | |
337 | mc_req_size = CIK_MC_UCODE_SIZE * 4; | |
21a93e13 | 338 | sdma_req_size = CIK_SDMA_UCODE_SIZE * 4; |
02c81327 AD |
339 | break; |
340 | case CHIP_KAVERI: | |
341 | chip_name = "KAVERI"; | |
342 | pfp_req_size = CIK_PFP_UCODE_SIZE * 4; | |
343 | me_req_size = CIK_ME_UCODE_SIZE * 4; | |
344 | ce_req_size = CIK_CE_UCODE_SIZE * 4; | |
345 | mec_req_size = CIK_MEC_UCODE_SIZE * 4; | |
346 | rlc_req_size = KV_RLC_UCODE_SIZE * 4; | |
21a93e13 | 347 | sdma_req_size = CIK_SDMA_UCODE_SIZE * 4; |
02c81327 AD |
348 | break; |
349 | case CHIP_KABINI: | |
350 | chip_name = "KABINI"; | |
351 | pfp_req_size = CIK_PFP_UCODE_SIZE * 4; | |
352 | me_req_size = CIK_ME_UCODE_SIZE * 4; | |
353 | ce_req_size = CIK_CE_UCODE_SIZE * 4; | |
354 | mec_req_size = CIK_MEC_UCODE_SIZE * 4; | |
355 | rlc_req_size = KB_RLC_UCODE_SIZE * 4; | |
21a93e13 | 356 | sdma_req_size = CIK_SDMA_UCODE_SIZE * 4; |
02c81327 AD |
357 | break; |
358 | default: BUG(); | |
359 | } | |
360 | ||
361 | DRM_INFO("Loading %s Microcode\n", chip_name); | |
362 | ||
363 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name); | |
364 | err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev); | |
365 | if (err) | |
366 | goto out; | |
367 | if (rdev->pfp_fw->size != pfp_req_size) { | |
368 | printk(KERN_ERR | |
369 | "cik_cp: Bogus length %zu in firmware \"%s\"\n", | |
370 | rdev->pfp_fw->size, fw_name); | |
371 | err = -EINVAL; | |
372 | goto out; | |
373 | } | |
374 | ||
375 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name); | |
376 | err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev); | |
377 | if (err) | |
378 | goto out; | |
379 | if (rdev->me_fw->size != me_req_size) { | |
380 | printk(KERN_ERR | |
381 | "cik_cp: Bogus length %zu in firmware \"%s\"\n", | |
382 | rdev->me_fw->size, fw_name); | |
383 | err = -EINVAL; | |
384 | } | |
385 | ||
386 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name); | |
387 | err = request_firmware(&rdev->ce_fw, fw_name, &pdev->dev); | |
388 | if (err) | |
389 | goto out; | |
390 | if (rdev->ce_fw->size != ce_req_size) { | |
391 | printk(KERN_ERR | |
392 | "cik_cp: Bogus length %zu in firmware \"%s\"\n", | |
393 | rdev->ce_fw->size, fw_name); | |
394 | err = -EINVAL; | |
395 | } | |
396 | ||
397 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_mec.bin", chip_name); | |
398 | err = request_firmware(&rdev->mec_fw, fw_name, &pdev->dev); | |
399 | if (err) | |
400 | goto out; | |
401 | if (rdev->mec_fw->size != mec_req_size) { | |
402 | printk(KERN_ERR | |
403 | "cik_cp: Bogus length %zu in firmware \"%s\"\n", | |
404 | rdev->mec_fw->size, fw_name); | |
405 | err = -EINVAL; | |
406 | } | |
407 | ||
408 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name); | |
409 | err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev); | |
410 | if (err) | |
411 | goto out; | |
412 | if (rdev->rlc_fw->size != rlc_req_size) { | |
413 | printk(KERN_ERR | |
414 | "cik_rlc: Bogus length %zu in firmware \"%s\"\n", | |
415 | rdev->rlc_fw->size, fw_name); | |
416 | err = -EINVAL; | |
417 | } | |
418 | ||
21a93e13 AD |
419 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name); |
420 | err = request_firmware(&rdev->sdma_fw, fw_name, &pdev->dev); | |
421 | if (err) | |
422 | goto out; | |
423 | if (rdev->sdma_fw->size != sdma_req_size) { | |
424 | printk(KERN_ERR | |
425 | "cik_sdma: Bogus length %zu in firmware \"%s\"\n", | |
426 | rdev->sdma_fw->size, fw_name); | |
427 | err = -EINVAL; | |
428 | } | |
429 | ||
02c81327 AD |
430 | /* No MC ucode on APUs */ |
431 | if (!(rdev->flags & RADEON_IS_IGP)) { | |
432 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name); | |
433 | err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev); | |
434 | if (err) | |
435 | goto out; | |
436 | if (rdev->mc_fw->size != mc_req_size) { | |
437 | printk(KERN_ERR | |
438 | "cik_mc: Bogus length %zu in firmware \"%s\"\n", | |
439 | rdev->mc_fw->size, fw_name); | |
440 | err = -EINVAL; | |
441 | } | |
442 | } | |
443 | ||
444 | out: | |
445 | platform_device_unregister(pdev); | |
446 | ||
447 | if (err) { | |
448 | if (err != -EINVAL) | |
449 | printk(KERN_ERR | |
450 | "cik_cp: Failed to load firmware \"%s\"\n", | |
451 | fw_name); | |
452 | release_firmware(rdev->pfp_fw); | |
453 | rdev->pfp_fw = NULL; | |
454 | release_firmware(rdev->me_fw); | |
455 | rdev->me_fw = NULL; | |
456 | release_firmware(rdev->ce_fw); | |
457 | rdev->ce_fw = NULL; | |
458 | release_firmware(rdev->rlc_fw); | |
459 | rdev->rlc_fw = NULL; | |
460 | release_firmware(rdev->mc_fw); | |
461 | rdev->mc_fw = NULL; | |
462 | } | |
463 | return err; | |
464 | } | |
465 | ||
8cc1a532 AD |
466 | /* |
467 | * Core functions | |
468 | */ | |
469 | /** | |
470 | * cik_tiling_mode_table_init - init the hw tiling table | |
471 | * | |
472 | * @rdev: radeon_device pointer | |
473 | * | |
474 | * Starting with SI, the tiling setup is done globally in a | |
475 | * set of 32 tiling modes. Rather than selecting each set of | |
476 | * parameters per surface as on older asics, we just select | |
477 | * which index in the tiling table we want to use, and the | |
478 | * surface uses those parameters (CIK). | |
479 | */ | |
480 | static void cik_tiling_mode_table_init(struct radeon_device *rdev) | |
481 | { | |
482 | const u32 num_tile_mode_states = 32; | |
483 | const u32 num_secondary_tile_mode_states = 16; | |
484 | u32 reg_offset, gb_tile_moden, split_equal_to_row_size; | |
485 | u32 num_pipe_configs; | |
486 | u32 num_rbs = rdev->config.cik.max_backends_per_se * | |
487 | rdev->config.cik.max_shader_engines; | |
488 | ||
489 | switch (rdev->config.cik.mem_row_size_in_kb) { | |
490 | case 1: | |
491 | split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB; | |
492 | break; | |
493 | case 2: | |
494 | default: | |
495 | split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB; | |
496 | break; | |
497 | case 4: | |
498 | split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB; | |
499 | break; | |
500 | } | |
501 | ||
502 | num_pipe_configs = rdev->config.cik.max_tile_pipes; | |
503 | if (num_pipe_configs > 8) | |
504 | num_pipe_configs = 8; /* ??? */ | |
505 | ||
506 | if (num_pipe_configs == 8) { | |
507 | for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { | |
508 | switch (reg_offset) { | |
509 | case 0: | |
510 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
511 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
512 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
513 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B)); | |
514 | break; | |
515 | case 1: | |
516 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
517 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
518 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
519 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B)); | |
520 | break; | |
521 | case 2: | |
522 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
523 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
524 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
525 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
526 | break; | |
527 | case 3: | |
528 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
529 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
530 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
531 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B)); | |
532 | break; | |
533 | case 4: | |
534 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
535 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
536 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
537 | TILE_SPLIT(split_equal_to_row_size)); | |
538 | break; | |
539 | case 5: | |
540 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
541 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); | |
542 | break; | |
543 | case 6: | |
544 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
545 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
546 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
547 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
548 | break; | |
549 | case 7: | |
550 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
551 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
552 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
553 | TILE_SPLIT(split_equal_to_row_size)); | |
554 | break; | |
555 | case 8: | |
556 | gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | | |
557 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16)); | |
558 | break; | |
559 | case 9: | |
560 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
561 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); | |
562 | break; | |
563 | case 10: | |
564 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
565 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
566 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
567 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
568 | break; | |
569 | case 11: | |
570 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
571 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
572 | PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | | |
573 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
574 | break; | |
575 | case 12: | |
576 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
577 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
578 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
579 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
580 | break; | |
581 | case 13: | |
582 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
583 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); | |
584 | break; | |
585 | case 14: | |
586 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
587 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
588 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
589 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
590 | break; | |
591 | case 16: | |
592 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
593 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
594 | PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | | |
595 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
596 | break; | |
597 | case 17: | |
598 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
599 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
600 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
601 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
602 | break; | |
603 | case 27: | |
604 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
605 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); | |
606 | break; | |
607 | case 28: | |
608 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
609 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
610 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
611 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
612 | break; | |
613 | case 29: | |
614 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
615 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
616 | PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | | |
617 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
618 | break; | |
619 | case 30: | |
620 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
621 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
622 | PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | | |
623 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
624 | break; | |
625 | default: | |
626 | gb_tile_moden = 0; | |
627 | break; | |
628 | } | |
629 | WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
630 | } | |
631 | for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { | |
632 | switch (reg_offset) { | |
633 | case 0: | |
634 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
635 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
636 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
637 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
638 | break; | |
639 | case 1: | |
640 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
641 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
642 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
643 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
644 | break; | |
645 | case 2: | |
646 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
647 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
648 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
649 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
650 | break; | |
651 | case 3: | |
652 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
653 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
654 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
655 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
656 | break; | |
657 | case 4: | |
658 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
659 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
660 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
661 | NUM_BANKS(ADDR_SURF_8_BANK)); | |
662 | break; | |
663 | case 5: | |
664 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
665 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
666 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
667 | NUM_BANKS(ADDR_SURF_4_BANK)); | |
668 | break; | |
669 | case 6: | |
670 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
671 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
672 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
673 | NUM_BANKS(ADDR_SURF_2_BANK)); | |
674 | break; | |
675 | case 8: | |
676 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
677 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | | |
678 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
679 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
680 | break; | |
681 | case 9: | |
682 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
683 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
684 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
685 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
686 | break; | |
687 | case 10: | |
688 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
689 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
690 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
691 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
692 | break; | |
693 | case 11: | |
694 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
695 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
696 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
697 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
698 | break; | |
699 | case 12: | |
700 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
701 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
702 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
703 | NUM_BANKS(ADDR_SURF_8_BANK)); | |
704 | break; | |
705 | case 13: | |
706 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
707 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
708 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
709 | NUM_BANKS(ADDR_SURF_4_BANK)); | |
710 | break; | |
711 | case 14: | |
712 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
713 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
714 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
715 | NUM_BANKS(ADDR_SURF_2_BANK)); | |
716 | break; | |
717 | default: | |
718 | gb_tile_moden = 0; | |
719 | break; | |
720 | } | |
721 | WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
722 | } | |
723 | } else if (num_pipe_configs == 4) { | |
724 | if (num_rbs == 4) { | |
725 | for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { | |
726 | switch (reg_offset) { | |
727 | case 0: | |
728 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
729 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
730 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
731 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B)); | |
732 | break; | |
733 | case 1: | |
734 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
735 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
736 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
737 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B)); | |
738 | break; | |
739 | case 2: | |
740 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
741 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
742 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
743 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
744 | break; | |
745 | case 3: | |
746 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
747 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
748 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
749 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B)); | |
750 | break; | |
751 | case 4: | |
752 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
753 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
754 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
755 | TILE_SPLIT(split_equal_to_row_size)); | |
756 | break; | |
757 | case 5: | |
758 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
759 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); | |
760 | break; | |
761 | case 6: | |
762 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
763 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
764 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
765 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
766 | break; | |
767 | case 7: | |
768 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
769 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
770 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
771 | TILE_SPLIT(split_equal_to_row_size)); | |
772 | break; | |
773 | case 8: | |
774 | gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | | |
775 | PIPE_CONFIG(ADDR_SURF_P4_16x16)); | |
776 | break; | |
777 | case 9: | |
778 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
779 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); | |
780 | break; | |
781 | case 10: | |
782 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
783 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
784 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
785 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
786 | break; | |
787 | case 11: | |
788 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
789 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
790 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
791 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
792 | break; | |
793 | case 12: | |
794 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
795 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
796 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
797 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
798 | break; | |
799 | case 13: | |
800 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
801 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); | |
802 | break; | |
803 | case 14: | |
804 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
805 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
806 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
807 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
808 | break; | |
809 | case 16: | |
810 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
811 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
812 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
813 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
814 | break; | |
815 | case 17: | |
816 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
817 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
818 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
819 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
820 | break; | |
821 | case 27: | |
822 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
823 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); | |
824 | break; | |
825 | case 28: | |
826 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
827 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
828 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
829 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
830 | break; | |
831 | case 29: | |
832 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
833 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
834 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
835 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
836 | break; | |
837 | case 30: | |
838 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
839 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
840 | PIPE_CONFIG(ADDR_SURF_P4_16x16) | | |
841 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
842 | break; | |
843 | default: | |
844 | gb_tile_moden = 0; | |
845 | break; | |
846 | } | |
847 | WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
848 | } | |
849 | } else if (num_rbs < 4) { | |
850 | for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { | |
851 | switch (reg_offset) { | |
852 | case 0: | |
853 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
854 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
855 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
856 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B)); | |
857 | break; | |
858 | case 1: | |
859 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
860 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
861 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
862 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B)); | |
863 | break; | |
864 | case 2: | |
865 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
866 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
867 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
868 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
869 | break; | |
870 | case 3: | |
871 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
872 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
873 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
874 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B)); | |
875 | break; | |
876 | case 4: | |
877 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
878 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
879 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
880 | TILE_SPLIT(split_equal_to_row_size)); | |
881 | break; | |
882 | case 5: | |
883 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
884 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); | |
885 | break; | |
886 | case 6: | |
887 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
888 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
889 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
890 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
891 | break; | |
892 | case 7: | |
893 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
894 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
895 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
896 | TILE_SPLIT(split_equal_to_row_size)); | |
897 | break; | |
898 | case 8: | |
899 | gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | | |
900 | PIPE_CONFIG(ADDR_SURF_P4_8x16)); | |
901 | break; | |
902 | case 9: | |
903 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
904 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); | |
905 | break; | |
906 | case 10: | |
907 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
908 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
909 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
910 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
911 | break; | |
912 | case 11: | |
913 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
914 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
915 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
916 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
917 | break; | |
918 | case 12: | |
919 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
920 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
921 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
922 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
923 | break; | |
924 | case 13: | |
925 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
926 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); | |
927 | break; | |
928 | case 14: | |
929 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
930 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
931 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
932 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
933 | break; | |
934 | case 16: | |
935 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
936 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
937 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
938 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
939 | break; | |
940 | case 17: | |
941 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
942 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
943 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
944 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
945 | break; | |
946 | case 27: | |
947 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
948 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); | |
949 | break; | |
950 | case 28: | |
951 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
952 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
953 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
954 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
955 | break; | |
956 | case 29: | |
957 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
958 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
959 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
960 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
961 | break; | |
962 | case 30: | |
963 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
964 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
965 | PIPE_CONFIG(ADDR_SURF_P4_8x16) | | |
966 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
967 | break; | |
968 | default: | |
969 | gb_tile_moden = 0; | |
970 | break; | |
971 | } | |
972 | WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
973 | } | |
974 | } | |
975 | for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { | |
976 | switch (reg_offset) { | |
977 | case 0: | |
978 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
979 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
980 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
981 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
982 | break; | |
983 | case 1: | |
984 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
985 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
986 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
987 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
988 | break; | |
989 | case 2: | |
990 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
991 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
992 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
993 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
994 | break; | |
995 | case 3: | |
996 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
997 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
998 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
999 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1000 | break; | |
1001 | case 4: | |
1002 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1003 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1004 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
1005 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1006 | break; | |
1007 | case 5: | |
1008 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1009 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1010 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
1011 | NUM_BANKS(ADDR_SURF_8_BANK)); | |
1012 | break; | |
1013 | case 6: | |
1014 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1015 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1016 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
1017 | NUM_BANKS(ADDR_SURF_4_BANK)); | |
1018 | break; | |
1019 | case 8: | |
1020 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | | |
1021 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | | |
1022 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1023 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1024 | break; | |
1025 | case 9: | |
1026 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | | |
1027 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
1028 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1029 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1030 | break; | |
1031 | case 10: | |
1032 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1033 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
1034 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1035 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1036 | break; | |
1037 | case 11: | |
1038 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1039 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
1040 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1041 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1042 | break; | |
1043 | case 12: | |
1044 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1045 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1046 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
1047 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1048 | break; | |
1049 | case 13: | |
1050 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1051 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1052 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
1053 | NUM_BANKS(ADDR_SURF_8_BANK)); | |
1054 | break; | |
1055 | case 14: | |
1056 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1057 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1058 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | | |
1059 | NUM_BANKS(ADDR_SURF_4_BANK)); | |
1060 | break; | |
1061 | default: | |
1062 | gb_tile_moden = 0; | |
1063 | break; | |
1064 | } | |
1065 | WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
1066 | } | |
1067 | } else if (num_pipe_configs == 2) { | |
1068 | for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { | |
1069 | switch (reg_offset) { | |
1070 | case 0: | |
1071 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1072 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1073 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1074 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B)); | |
1075 | break; | |
1076 | case 1: | |
1077 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1078 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1079 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1080 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B)); | |
1081 | break; | |
1082 | case 2: | |
1083 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1084 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1085 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1086 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
1087 | break; | |
1088 | case 3: | |
1089 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1090 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1091 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1092 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B)); | |
1093 | break; | |
1094 | case 4: | |
1095 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1096 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1097 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1098 | TILE_SPLIT(split_equal_to_row_size)); | |
1099 | break; | |
1100 | case 5: | |
1101 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
1102 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); | |
1103 | break; | |
1104 | case 6: | |
1105 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
1106 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1107 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1108 | TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B)); | |
1109 | break; | |
1110 | case 7: | |
1111 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
1112 | MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | | |
1113 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1114 | TILE_SPLIT(split_equal_to_row_size)); | |
1115 | break; | |
1116 | case 8: | |
1117 | gb_tile_moden = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); | |
1118 | break; | |
1119 | case 9: | |
1120 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
1121 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); | |
1122 | break; | |
1123 | case 10: | |
1124 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1125 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
1126 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1127 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1128 | break; | |
1129 | case 11: | |
1130 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
1131 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
1132 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1133 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1134 | break; | |
1135 | case 12: | |
1136 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
1137 | MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | | |
1138 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1139 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1140 | break; | |
1141 | case 13: | |
1142 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
1143 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); | |
1144 | break; | |
1145 | case 14: | |
1146 | gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | | |
1147 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
1148 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1149 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1150 | break; | |
1151 | case 16: | |
1152 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
1153 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
1154 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1155 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1156 | break; | |
1157 | case 17: | |
1158 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
1159 | MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | | |
1160 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1161 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1162 | break; | |
1163 | case 27: | |
1164 | gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | | |
1165 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); | |
1166 | break; | |
1167 | case 28: | |
1168 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
1169 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
1170 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1171 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1172 | break; | |
1173 | case 29: | |
1174 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | | |
1175 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
1176 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1177 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1178 | break; | |
1179 | case 30: | |
1180 | gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | | |
1181 | MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | | |
1182 | PIPE_CONFIG(ADDR_SURF_P2) | | |
1183 | SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); | |
1184 | break; | |
1185 | default: | |
1186 | gb_tile_moden = 0; | |
1187 | break; | |
1188 | } | |
1189 | WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
1190 | } | |
1191 | for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { | |
1192 | switch (reg_offset) { | |
1193 | case 0: | |
1194 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | | |
1195 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
1196 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1197 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1198 | break; | |
1199 | case 1: | |
1200 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | | |
1201 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
1202 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1203 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1204 | break; | |
1205 | case 2: | |
1206 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1207 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
1208 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1209 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1210 | break; | |
1211 | case 3: | |
1212 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1213 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1214 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1215 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1216 | break; | |
1217 | case 4: | |
1218 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1219 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1220 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1221 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1222 | break; | |
1223 | case 5: | |
1224 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1225 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1226 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1227 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1228 | break; | |
1229 | case 6: | |
1230 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1231 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1232 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
1233 | NUM_BANKS(ADDR_SURF_8_BANK)); | |
1234 | break; | |
1235 | case 8: | |
1236 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | | |
1237 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | | |
1238 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1239 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1240 | break; | |
1241 | case 9: | |
1242 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | | |
1243 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
1244 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1245 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1246 | break; | |
1247 | case 10: | |
1248 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | | |
1249 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | | |
1250 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1251 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1252 | break; | |
1253 | case 11: | |
1254 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | | |
1255 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
1256 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1257 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1258 | break; | |
1259 | case 12: | |
1260 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1261 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | | |
1262 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1263 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1264 | break; | |
1265 | case 13: | |
1266 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1267 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1268 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | | |
1269 | NUM_BANKS(ADDR_SURF_16_BANK)); | |
1270 | break; | |
1271 | case 14: | |
1272 | gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | | |
1273 | BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | | |
1274 | MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | | |
1275 | NUM_BANKS(ADDR_SURF_8_BANK)); | |
1276 | break; | |
1277 | default: | |
1278 | gb_tile_moden = 0; | |
1279 | break; | |
1280 | } | |
1281 | WREG32(GB_MACROTILE_MODE0 + (reg_offset * 4), gb_tile_moden); | |
1282 | } | |
1283 | } else | |
1284 | DRM_ERROR("unknown num pipe config: 0x%x\n", num_pipe_configs); | |
1285 | } | |
1286 | ||
1287 | /** | |
1288 | * cik_select_se_sh - select which SE, SH to address | |
1289 | * | |
1290 | * @rdev: radeon_device pointer | |
1291 | * @se_num: shader engine to address | |
1292 | * @sh_num: sh block to address | |
1293 | * | |
1294 | * Select which SE, SH combinations to address. Certain | |
1295 | * registers are instanced per SE or SH. 0xffffffff means | |
1296 | * broadcast to all SEs or SHs (CIK). | |
1297 | */ | |
1298 | static void cik_select_se_sh(struct radeon_device *rdev, | |
1299 | u32 se_num, u32 sh_num) | |
1300 | { | |
1301 | u32 data = INSTANCE_BROADCAST_WRITES; | |
1302 | ||
1303 | if ((se_num == 0xffffffff) && (sh_num == 0xffffffff)) | |
1304 | data = SH_BROADCAST_WRITES | SE_BROADCAST_WRITES; | |
1305 | else if (se_num == 0xffffffff) | |
1306 | data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num); | |
1307 | else if (sh_num == 0xffffffff) | |
1308 | data |= SH_BROADCAST_WRITES | SE_INDEX(se_num); | |
1309 | else | |
1310 | data |= SH_INDEX(sh_num) | SE_INDEX(se_num); | |
1311 | WREG32(GRBM_GFX_INDEX, data); | |
1312 | } | |
1313 | ||
1314 | /** | |
1315 | * cik_create_bitmask - create a bitmask | |
1316 | * | |
1317 | * @bit_width: length of the mask | |
1318 | * | |
1319 | * create a variable length bit mask (CIK). | |
1320 | * Returns the bitmask. | |
1321 | */ | |
1322 | static u32 cik_create_bitmask(u32 bit_width) | |
1323 | { | |
1324 | u32 i, mask = 0; | |
1325 | ||
1326 | for (i = 0; i < bit_width; i++) { | |
1327 | mask <<= 1; | |
1328 | mask |= 1; | |
1329 | } | |
1330 | return mask; | |
1331 | } | |
1332 | ||
1333 | /** | |
1334 | * cik_select_se_sh - select which SE, SH to address | |
1335 | * | |
1336 | * @rdev: radeon_device pointer | |
1337 | * @max_rb_num: max RBs (render backends) for the asic | |
1338 | * @se_num: number of SEs (shader engines) for the asic | |
1339 | * @sh_per_se: number of SH blocks per SE for the asic | |
1340 | * | |
1341 | * Calculates the bitmask of disabled RBs (CIK). | |
1342 | * Returns the disabled RB bitmask. | |
1343 | */ | |
1344 | static u32 cik_get_rb_disabled(struct radeon_device *rdev, | |
1345 | u32 max_rb_num, u32 se_num, | |
1346 | u32 sh_per_se) | |
1347 | { | |
1348 | u32 data, mask; | |
1349 | ||
1350 | data = RREG32(CC_RB_BACKEND_DISABLE); | |
1351 | if (data & 1) | |
1352 | data &= BACKEND_DISABLE_MASK; | |
1353 | else | |
1354 | data = 0; | |
1355 | data |= RREG32(GC_USER_RB_BACKEND_DISABLE); | |
1356 | ||
1357 | data >>= BACKEND_DISABLE_SHIFT; | |
1358 | ||
1359 | mask = cik_create_bitmask(max_rb_num / se_num / sh_per_se); | |
1360 | ||
1361 | return data & mask; | |
1362 | } | |
1363 | ||
1364 | /** | |
1365 | * cik_setup_rb - setup the RBs on the asic | |
1366 | * | |
1367 | * @rdev: radeon_device pointer | |
1368 | * @se_num: number of SEs (shader engines) for the asic | |
1369 | * @sh_per_se: number of SH blocks per SE for the asic | |
1370 | * @max_rb_num: max RBs (render backends) for the asic | |
1371 | * | |
1372 | * Configures per-SE/SH RB registers (CIK). | |
1373 | */ | |
1374 | static void cik_setup_rb(struct radeon_device *rdev, | |
1375 | u32 se_num, u32 sh_per_se, | |
1376 | u32 max_rb_num) | |
1377 | { | |
1378 | int i, j; | |
1379 | u32 data, mask; | |
1380 | u32 disabled_rbs = 0; | |
1381 | u32 enabled_rbs = 0; | |
1382 | ||
1383 | for (i = 0; i < se_num; i++) { | |
1384 | for (j = 0; j < sh_per_se; j++) { | |
1385 | cik_select_se_sh(rdev, i, j); | |
1386 | data = cik_get_rb_disabled(rdev, max_rb_num, se_num, sh_per_se); | |
1387 | disabled_rbs |= data << ((i * sh_per_se + j) * CIK_RB_BITMAP_WIDTH_PER_SH); | |
1388 | } | |
1389 | } | |
1390 | cik_select_se_sh(rdev, 0xffffffff, 0xffffffff); | |
1391 | ||
1392 | mask = 1; | |
1393 | for (i = 0; i < max_rb_num; i++) { | |
1394 | if (!(disabled_rbs & mask)) | |
1395 | enabled_rbs |= mask; | |
1396 | mask <<= 1; | |
1397 | } | |
1398 | ||
1399 | for (i = 0; i < se_num; i++) { | |
1400 | cik_select_se_sh(rdev, i, 0xffffffff); | |
1401 | data = 0; | |
1402 | for (j = 0; j < sh_per_se; j++) { | |
1403 | switch (enabled_rbs & 3) { | |
1404 | case 1: | |
1405 | data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2); | |
1406 | break; | |
1407 | case 2: | |
1408 | data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2); | |
1409 | break; | |
1410 | case 3: | |
1411 | default: | |
1412 | data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2); | |
1413 | break; | |
1414 | } | |
1415 | enabled_rbs >>= 2; | |
1416 | } | |
1417 | WREG32(PA_SC_RASTER_CONFIG, data); | |
1418 | } | |
1419 | cik_select_se_sh(rdev, 0xffffffff, 0xffffffff); | |
1420 | } | |
1421 | ||
1422 | /** | |
1423 | * cik_gpu_init - setup the 3D engine | |
1424 | * | |
1425 | * @rdev: radeon_device pointer | |
1426 | * | |
1427 | * Configures the 3D engine and tiling configuration | |
1428 | * registers so that the 3D engine is usable. | |
1429 | */ | |
1430 | static void cik_gpu_init(struct radeon_device *rdev) | |
1431 | { | |
1432 | u32 gb_addr_config = RREG32(GB_ADDR_CONFIG); | |
1433 | u32 mc_shared_chmap, mc_arb_ramcfg; | |
1434 | u32 hdp_host_path_cntl; | |
1435 | u32 tmp; | |
1436 | int i, j; | |
1437 | ||
1438 | switch (rdev->family) { | |
1439 | case CHIP_BONAIRE: | |
1440 | rdev->config.cik.max_shader_engines = 2; | |
1441 | rdev->config.cik.max_tile_pipes = 4; | |
1442 | rdev->config.cik.max_cu_per_sh = 7; | |
1443 | rdev->config.cik.max_sh_per_se = 1; | |
1444 | rdev->config.cik.max_backends_per_se = 2; | |
1445 | rdev->config.cik.max_texture_channel_caches = 4; | |
1446 | rdev->config.cik.max_gprs = 256; | |
1447 | rdev->config.cik.max_gs_threads = 32; | |
1448 | rdev->config.cik.max_hw_contexts = 8; | |
1449 | ||
1450 | rdev->config.cik.sc_prim_fifo_size_frontend = 0x20; | |
1451 | rdev->config.cik.sc_prim_fifo_size_backend = 0x100; | |
1452 | rdev->config.cik.sc_hiz_tile_fifo_size = 0x30; | |
1453 | rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130; | |
1454 | gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; | |
1455 | break; | |
1456 | case CHIP_KAVERI: | |
1457 | /* TODO */ | |
1458 | break; | |
1459 | case CHIP_KABINI: | |
1460 | default: | |
1461 | rdev->config.cik.max_shader_engines = 1; | |
1462 | rdev->config.cik.max_tile_pipes = 2; | |
1463 | rdev->config.cik.max_cu_per_sh = 2; | |
1464 | rdev->config.cik.max_sh_per_se = 1; | |
1465 | rdev->config.cik.max_backends_per_se = 1; | |
1466 | rdev->config.cik.max_texture_channel_caches = 2; | |
1467 | rdev->config.cik.max_gprs = 256; | |
1468 | rdev->config.cik.max_gs_threads = 16; | |
1469 | rdev->config.cik.max_hw_contexts = 8; | |
1470 | ||
1471 | rdev->config.cik.sc_prim_fifo_size_frontend = 0x20; | |
1472 | rdev->config.cik.sc_prim_fifo_size_backend = 0x100; | |
1473 | rdev->config.cik.sc_hiz_tile_fifo_size = 0x30; | |
1474 | rdev->config.cik.sc_earlyz_tile_fifo_size = 0x130; | |
1475 | gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; | |
1476 | break; | |
1477 | } | |
1478 | ||
1479 | /* Initialize HDP */ | |
1480 | for (i = 0, j = 0; i < 32; i++, j += 0x18) { | |
1481 | WREG32((0x2c14 + j), 0x00000000); | |
1482 | WREG32((0x2c18 + j), 0x00000000); | |
1483 | WREG32((0x2c1c + j), 0x00000000); | |
1484 | WREG32((0x2c20 + j), 0x00000000); | |
1485 | WREG32((0x2c24 + j), 0x00000000); | |
1486 | } | |
1487 | ||
1488 | WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff)); | |
1489 | ||
1490 | WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN); | |
1491 | ||
1492 | mc_shared_chmap = RREG32(MC_SHARED_CHMAP); | |
1493 | mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG); | |
1494 | ||
1495 | rdev->config.cik.num_tile_pipes = rdev->config.cik.max_tile_pipes; | |
1496 | rdev->config.cik.mem_max_burst_length_bytes = 256; | |
1497 | tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT; | |
1498 | rdev->config.cik.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; | |
1499 | if (rdev->config.cik.mem_row_size_in_kb > 4) | |
1500 | rdev->config.cik.mem_row_size_in_kb = 4; | |
1501 | /* XXX use MC settings? */ | |
1502 | rdev->config.cik.shader_engine_tile_size = 32; | |
1503 | rdev->config.cik.num_gpus = 1; | |
1504 | rdev->config.cik.multi_gpu_tile_size = 64; | |
1505 | ||
1506 | /* fix up row size */ | |
1507 | gb_addr_config &= ~ROW_SIZE_MASK; | |
1508 | switch (rdev->config.cik.mem_row_size_in_kb) { | |
1509 | case 1: | |
1510 | default: | |
1511 | gb_addr_config |= ROW_SIZE(0); | |
1512 | break; | |
1513 | case 2: | |
1514 | gb_addr_config |= ROW_SIZE(1); | |
1515 | break; | |
1516 | case 4: | |
1517 | gb_addr_config |= ROW_SIZE(2); | |
1518 | break; | |
1519 | } | |
1520 | ||
1521 | /* setup tiling info dword. gb_addr_config is not adequate since it does | |
1522 | * not have bank info, so create a custom tiling dword. | |
1523 | * bits 3:0 num_pipes | |
1524 | * bits 7:4 num_banks | |
1525 | * bits 11:8 group_size | |
1526 | * bits 15:12 row_size | |
1527 | */ | |
1528 | rdev->config.cik.tile_config = 0; | |
1529 | switch (rdev->config.cik.num_tile_pipes) { | |
1530 | case 1: | |
1531 | rdev->config.cik.tile_config |= (0 << 0); | |
1532 | break; | |
1533 | case 2: | |
1534 | rdev->config.cik.tile_config |= (1 << 0); | |
1535 | break; | |
1536 | case 4: | |
1537 | rdev->config.cik.tile_config |= (2 << 0); | |
1538 | break; | |
1539 | case 8: | |
1540 | default: | |
1541 | /* XXX what about 12? */ | |
1542 | rdev->config.cik.tile_config |= (3 << 0); | |
1543 | break; | |
1544 | } | |
1545 | if ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) | |
1546 | rdev->config.cik.tile_config |= 1 << 4; | |
1547 | else | |
1548 | rdev->config.cik.tile_config |= 0 << 4; | |
1549 | rdev->config.cik.tile_config |= | |
1550 | ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8; | |
1551 | rdev->config.cik.tile_config |= | |
1552 | ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12; | |
1553 | ||
1554 | WREG32(GB_ADDR_CONFIG, gb_addr_config); | |
1555 | WREG32(HDP_ADDR_CONFIG, gb_addr_config); | |
1556 | WREG32(DMIF_ADDR_CALC, gb_addr_config); | |
21a93e13 AD |
1557 | WREG32(SDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET, gb_addr_config & 0x70); |
1558 | WREG32(SDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET, gb_addr_config & 0x70); | |
87167bb1 CK |
1559 | WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config); |
1560 | WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); | |
1561 | WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); | |
8cc1a532 AD |
1562 | |
1563 | cik_tiling_mode_table_init(rdev); | |
1564 | ||
1565 | cik_setup_rb(rdev, rdev->config.cik.max_shader_engines, | |
1566 | rdev->config.cik.max_sh_per_se, | |
1567 | rdev->config.cik.max_backends_per_se); | |
1568 | ||
1569 | /* set HW defaults for 3D engine */ | |
1570 | WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60)); | |
1571 | ||
1572 | WREG32(SX_DEBUG_1, 0x20); | |
1573 | ||
1574 | WREG32(TA_CNTL_AUX, 0x00010000); | |
1575 | ||
1576 | tmp = RREG32(SPI_CONFIG_CNTL); | |
1577 | tmp |= 0x03000000; | |
1578 | WREG32(SPI_CONFIG_CNTL, tmp); | |
1579 | ||
1580 | WREG32(SQ_CONFIG, 1); | |
1581 | ||
1582 | WREG32(DB_DEBUG, 0); | |
1583 | ||
1584 | tmp = RREG32(DB_DEBUG2) & ~0xf00fffff; | |
1585 | tmp |= 0x00000400; | |
1586 | WREG32(DB_DEBUG2, tmp); | |
1587 | ||
1588 | tmp = RREG32(DB_DEBUG3) & ~0x0002021c; | |
1589 | tmp |= 0x00020200; | |
1590 | WREG32(DB_DEBUG3, tmp); | |
1591 | ||
1592 | tmp = RREG32(CB_HW_CONTROL) & ~0x00010000; | |
1593 | tmp |= 0x00018208; | |
1594 | WREG32(CB_HW_CONTROL, tmp); | |
1595 | ||
1596 | WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4)); | |
1597 | ||
1598 | WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_frontend) | | |
1599 | SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.cik.sc_prim_fifo_size_backend) | | |
1600 | SC_HIZ_TILE_FIFO_SIZE(rdev->config.cik.sc_hiz_tile_fifo_size) | | |
1601 | SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cik.sc_earlyz_tile_fifo_size))); | |
1602 | ||
1603 | WREG32(VGT_NUM_INSTANCES, 1); | |
1604 | ||
1605 | WREG32(CP_PERFMON_CNTL, 0); | |
1606 | ||
1607 | WREG32(SQ_CONFIG, 0); | |
1608 | ||
1609 | WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) | | |
1610 | FORCE_EOV_MAX_REZ_CNT(255))); | |
1611 | ||
1612 | WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) | | |
1613 | AUTO_INVLD_EN(ES_AND_GS_AUTO)); | |
1614 | ||
1615 | WREG32(VGT_GS_VERTEX_REUSE, 16); | |
1616 | WREG32(PA_SC_LINE_STIPPLE_STATE, 0); | |
1617 | ||
1618 | tmp = RREG32(HDP_MISC_CNTL); | |
1619 | tmp |= HDP_FLUSH_INVALIDATE_CACHE; | |
1620 | WREG32(HDP_MISC_CNTL, tmp); | |
1621 | ||
1622 | hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL); | |
1623 | WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl); | |
1624 | ||
1625 | WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3)); | |
1626 | WREG32(PA_SC_ENHANCE, ENABLE_PA_SC_OUT_OF_ORDER); | |
1627 | ||
1628 | udelay(50); | |
1629 | } | |
1630 | ||
2cae3bc3 AD |
1631 | /* |
1632 | * GPU scratch registers helpers function. | |
1633 | */ | |
1634 | /** | |
1635 | * cik_scratch_init - setup driver info for CP scratch regs | |
1636 | * | |
1637 | * @rdev: radeon_device pointer | |
1638 | * | |
1639 | * Set up the number and offset of the CP scratch registers. | |
1640 | * NOTE: use of CP scratch registers is a legacy inferface and | |
1641 | * is not used by default on newer asics (r6xx+). On newer asics, | |
1642 | * memory buffers are used for fences rather than scratch regs. | |
1643 | */ | |
1644 | static void cik_scratch_init(struct radeon_device *rdev) | |
1645 | { | |
1646 | int i; | |
1647 | ||
1648 | rdev->scratch.num_reg = 7; | |
1649 | rdev->scratch.reg_base = SCRATCH_REG0; | |
1650 | for (i = 0; i < rdev->scratch.num_reg; i++) { | |
1651 | rdev->scratch.free[i] = true; | |
1652 | rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4); | |
1653 | } | |
1654 | } | |
1655 | ||
fbc832c7 AD |
1656 | /** |
1657 | * cik_ring_test - basic gfx ring test | |
1658 | * | |
1659 | * @rdev: radeon_device pointer | |
1660 | * @ring: radeon_ring structure holding ring information | |
1661 | * | |
1662 | * Allocate a scratch register and write to it using the gfx ring (CIK). | |
1663 | * Provides a basic gfx ring test to verify that the ring is working. | |
1664 | * Used by cik_cp_gfx_resume(); | |
1665 | * Returns 0 on success, error on failure. | |
1666 | */ | |
1667 | int cik_ring_test(struct radeon_device *rdev, struct radeon_ring *ring) | |
1668 | { | |
1669 | uint32_t scratch; | |
1670 | uint32_t tmp = 0; | |
1671 | unsigned i; | |
1672 | int r; | |
1673 | ||
1674 | r = radeon_scratch_get(rdev, &scratch); | |
1675 | if (r) { | |
1676 | DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r); | |
1677 | return r; | |
1678 | } | |
1679 | WREG32(scratch, 0xCAFEDEAD); | |
1680 | r = radeon_ring_lock(rdev, ring, 3); | |
1681 | if (r) { | |
1682 | DRM_ERROR("radeon: cp failed to lock ring %d (%d).\n", ring->idx, r); | |
1683 | radeon_scratch_free(rdev, scratch); | |
1684 | return r; | |
1685 | } | |
1686 | radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); | |
1687 | radeon_ring_write(ring, ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2)); | |
1688 | radeon_ring_write(ring, 0xDEADBEEF); | |
1689 | radeon_ring_unlock_commit(rdev, ring); | |
1690 | for (i = 0; i < rdev->usec_timeout; i++) { | |
1691 | tmp = RREG32(scratch); | |
1692 | if (tmp == 0xDEADBEEF) | |
1693 | break; | |
1694 | DRM_UDELAY(1); | |
1695 | } | |
1696 | if (i < rdev->usec_timeout) { | |
1697 | DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i); | |
1698 | } else { | |
1699 | DRM_ERROR("radeon: ring %d test failed (scratch(0x%04X)=0x%08X)\n", | |
1700 | ring->idx, scratch, tmp); | |
1701 | r = -EINVAL; | |
1702 | } | |
1703 | radeon_scratch_free(rdev, scratch); | |
1704 | return r; | |
1705 | } | |
1706 | ||
2cae3bc3 AD |
1707 | /** |
1708 | * cik_fence_ring_emit - emit a fence on the gfx ring | |
1709 | * | |
1710 | * @rdev: radeon_device pointer | |
1711 | * @fence: radeon fence object | |
1712 | * | |
1713 | * Emits a fence sequnce number on the gfx ring and flushes | |
1714 | * GPU caches. | |
1715 | */ | |
1716 | void cik_fence_ring_emit(struct radeon_device *rdev, | |
1717 | struct radeon_fence *fence) | |
1718 | { | |
1719 | struct radeon_ring *ring = &rdev->ring[fence->ring]; | |
1720 | u64 addr = rdev->fence_drv[fence->ring].gpu_addr; | |
1721 | ||
1722 | /* EVENT_WRITE_EOP - flush caches, send int */ | |
1723 | radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); | |
1724 | radeon_ring_write(ring, (EOP_TCL1_ACTION_EN | | |
1725 | EOP_TC_ACTION_EN | | |
1726 | EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | | |
1727 | EVENT_INDEX(5))); | |
1728 | radeon_ring_write(ring, addr & 0xfffffffc); | |
1729 | radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | DATA_SEL(1) | INT_SEL(2)); | |
1730 | radeon_ring_write(ring, fence->seq); | |
1731 | radeon_ring_write(ring, 0); | |
1732 | /* HDP flush */ | |
1733 | /* We should be using the new WAIT_REG_MEM special op packet here | |
1734 | * but it causes the CP to hang | |
1735 | */ | |
1736 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
1737 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
1738 | WRITE_DATA_DST_SEL(0))); | |
1739 | radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2); | |
1740 | radeon_ring_write(ring, 0); | |
1741 | radeon_ring_write(ring, 0); | |
1742 | } | |
1743 | ||
1744 | void cik_semaphore_ring_emit(struct radeon_device *rdev, | |
1745 | struct radeon_ring *ring, | |
1746 | struct radeon_semaphore *semaphore, | |
1747 | bool emit_wait) | |
1748 | { | |
1749 | uint64_t addr = semaphore->gpu_addr; | |
1750 | unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL; | |
1751 | ||
1752 | radeon_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1)); | |
1753 | radeon_ring_write(ring, addr & 0xffffffff); | |
1754 | radeon_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel); | |
1755 | } | |
1756 | ||
1757 | /* | |
1758 | * IB stuff | |
1759 | */ | |
1760 | /** | |
1761 | * cik_ring_ib_execute - emit an IB (Indirect Buffer) on the gfx ring | |
1762 | * | |
1763 | * @rdev: radeon_device pointer | |
1764 | * @ib: radeon indirect buffer object | |
1765 | * | |
1766 | * Emits an DE (drawing engine) or CE (constant engine) IB | |
1767 | * on the gfx ring. IBs are usually generated by userspace | |
1768 | * acceleration drivers and submitted to the kernel for | |
1769 | * sheduling on the ring. This function schedules the IB | |
1770 | * on the gfx ring for execution by the GPU. | |
1771 | */ | |
1772 | void cik_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) | |
1773 | { | |
1774 | struct radeon_ring *ring = &rdev->ring[ib->ring]; | |
1775 | u32 header, control = INDIRECT_BUFFER_VALID; | |
1776 | ||
1777 | if (ib->is_const_ib) { | |
1778 | /* set switch buffer packet before const IB */ | |
1779 | radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); | |
1780 | radeon_ring_write(ring, 0); | |
1781 | ||
1782 | header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); | |
1783 | } else { | |
1784 | u32 next_rptr; | |
1785 | if (ring->rptr_save_reg) { | |
1786 | next_rptr = ring->wptr + 3 + 4; | |
1787 | radeon_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); | |
1788 | radeon_ring_write(ring, ((ring->rptr_save_reg - | |
1789 | PACKET3_SET_UCONFIG_REG_START) >> 2)); | |
1790 | radeon_ring_write(ring, next_rptr); | |
1791 | } else if (rdev->wb.enabled) { | |
1792 | next_rptr = ring->wptr + 5 + 4; | |
1793 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
1794 | radeon_ring_write(ring, WRITE_DATA_DST_SEL(1)); | |
1795 | radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc); | |
1796 | radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff); | |
1797 | radeon_ring_write(ring, next_rptr); | |
1798 | } | |
1799 | ||
1800 | header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); | |
1801 | } | |
1802 | ||
1803 | control |= ib->length_dw | | |
1804 | (ib->vm ? (ib->vm->id << 24) : 0); | |
1805 | ||
1806 | radeon_ring_write(ring, header); | |
1807 | radeon_ring_write(ring, | |
1808 | #ifdef __BIG_ENDIAN | |
1809 | (2 << 0) | | |
1810 | #endif | |
1811 | (ib->gpu_addr & 0xFFFFFFFC)); | |
1812 | radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); | |
1813 | radeon_ring_write(ring, control); | |
1814 | } | |
1815 | ||
fbc832c7 AD |
1816 | /** |
1817 | * cik_ib_test - basic gfx ring IB test | |
1818 | * | |
1819 | * @rdev: radeon_device pointer | |
1820 | * @ring: radeon_ring structure holding ring information | |
1821 | * | |
1822 | * Allocate an IB and execute it on the gfx ring (CIK). | |
1823 | * Provides a basic gfx ring test to verify that IBs are working. | |
1824 | * Returns 0 on success, error on failure. | |
1825 | */ | |
1826 | int cik_ib_test(struct radeon_device *rdev, struct radeon_ring *ring) | |
1827 | { | |
1828 | struct radeon_ib ib; | |
1829 | uint32_t scratch; | |
1830 | uint32_t tmp = 0; | |
1831 | unsigned i; | |
1832 | int r; | |
1833 | ||
1834 | r = radeon_scratch_get(rdev, &scratch); | |
1835 | if (r) { | |
1836 | DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r); | |
1837 | return r; | |
1838 | } | |
1839 | WREG32(scratch, 0xCAFEDEAD); | |
1840 | r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256); | |
1841 | if (r) { | |
1842 | DRM_ERROR("radeon: failed to get ib (%d).\n", r); | |
1843 | return r; | |
1844 | } | |
1845 | ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); | |
1846 | ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START) >> 2); | |
1847 | ib.ptr[2] = 0xDEADBEEF; | |
1848 | ib.length_dw = 3; | |
1849 | r = radeon_ib_schedule(rdev, &ib, NULL); | |
1850 | if (r) { | |
1851 | radeon_scratch_free(rdev, scratch); | |
1852 | radeon_ib_free(rdev, &ib); | |
1853 | DRM_ERROR("radeon: failed to schedule ib (%d).\n", r); | |
1854 | return r; | |
1855 | } | |
1856 | r = radeon_fence_wait(ib.fence, false); | |
1857 | if (r) { | |
1858 | DRM_ERROR("radeon: fence wait failed (%d).\n", r); | |
1859 | return r; | |
1860 | } | |
1861 | for (i = 0; i < rdev->usec_timeout; i++) { | |
1862 | tmp = RREG32(scratch); | |
1863 | if (tmp == 0xDEADBEEF) | |
1864 | break; | |
1865 | DRM_UDELAY(1); | |
1866 | } | |
1867 | if (i < rdev->usec_timeout) { | |
1868 | DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i); | |
1869 | } else { | |
1870 | DRM_ERROR("radeon: ib test failed (scratch(0x%04X)=0x%08X)\n", | |
1871 | scratch, tmp); | |
1872 | r = -EINVAL; | |
1873 | } | |
1874 | radeon_scratch_free(rdev, scratch); | |
1875 | radeon_ib_free(rdev, &ib); | |
1876 | return r; | |
1877 | } | |
1878 | ||
841cf442 AD |
1879 | /* |
1880 | * CP. | |
1881 | * On CIK, gfx and compute now have independant command processors. | |
1882 | * | |
1883 | * GFX | |
1884 | * Gfx consists of a single ring and can process both gfx jobs and | |
1885 | * compute jobs. The gfx CP consists of three microengines (ME): | |
1886 | * PFP - Pre-Fetch Parser | |
1887 | * ME - Micro Engine | |
1888 | * CE - Constant Engine | |
1889 | * The PFP and ME make up what is considered the Drawing Engine (DE). | |
1890 | * The CE is an asynchronous engine used for updating buffer desciptors | |
1891 | * used by the DE so that they can be loaded into cache in parallel | |
1892 | * while the DE is processing state update packets. | |
1893 | * | |
1894 | * Compute | |
1895 | * The compute CP consists of two microengines (ME): | |
1896 | * MEC1 - Compute MicroEngine 1 | |
1897 | * MEC2 - Compute MicroEngine 2 | |
1898 | * Each MEC supports 4 compute pipes and each pipe supports 8 queues. | |
1899 | * The queues are exposed to userspace and are programmed directly | |
1900 | * by the compute runtime. | |
1901 | */ | |
1902 | /** | |
1903 | * cik_cp_gfx_enable - enable/disable the gfx CP MEs | |
1904 | * | |
1905 | * @rdev: radeon_device pointer | |
1906 | * @enable: enable or disable the MEs | |
1907 | * | |
1908 | * Halts or unhalts the gfx MEs. | |
1909 | */ | |
1910 | static void cik_cp_gfx_enable(struct radeon_device *rdev, bool enable) | |
1911 | { | |
1912 | if (enable) | |
1913 | WREG32(CP_ME_CNTL, 0); | |
1914 | else { | |
1915 | WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT)); | |
1916 | rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; | |
1917 | } | |
1918 | udelay(50); | |
1919 | } | |
1920 | ||
1921 | /** | |
1922 | * cik_cp_gfx_load_microcode - load the gfx CP ME ucode | |
1923 | * | |
1924 | * @rdev: radeon_device pointer | |
1925 | * | |
1926 | * Loads the gfx PFP, ME, and CE ucode. | |
1927 | * Returns 0 for success, -EINVAL if the ucode is not available. | |
1928 | */ | |
1929 | static int cik_cp_gfx_load_microcode(struct radeon_device *rdev) | |
1930 | { | |
1931 | const __be32 *fw_data; | |
1932 | int i; | |
1933 | ||
1934 | if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw) | |
1935 | return -EINVAL; | |
1936 | ||
1937 | cik_cp_gfx_enable(rdev, false); | |
1938 | ||
1939 | /* PFP */ | |
1940 | fw_data = (const __be32 *)rdev->pfp_fw->data; | |
1941 | WREG32(CP_PFP_UCODE_ADDR, 0); | |
1942 | for (i = 0; i < CIK_PFP_UCODE_SIZE; i++) | |
1943 | WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++)); | |
1944 | WREG32(CP_PFP_UCODE_ADDR, 0); | |
1945 | ||
1946 | /* CE */ | |
1947 | fw_data = (const __be32 *)rdev->ce_fw->data; | |
1948 | WREG32(CP_CE_UCODE_ADDR, 0); | |
1949 | for (i = 0; i < CIK_CE_UCODE_SIZE; i++) | |
1950 | WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++)); | |
1951 | WREG32(CP_CE_UCODE_ADDR, 0); | |
1952 | ||
1953 | /* ME */ | |
1954 | fw_data = (const __be32 *)rdev->me_fw->data; | |
1955 | WREG32(CP_ME_RAM_WADDR, 0); | |
1956 | for (i = 0; i < CIK_ME_UCODE_SIZE; i++) | |
1957 | WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++)); | |
1958 | WREG32(CP_ME_RAM_WADDR, 0); | |
1959 | ||
1960 | WREG32(CP_PFP_UCODE_ADDR, 0); | |
1961 | WREG32(CP_CE_UCODE_ADDR, 0); | |
1962 | WREG32(CP_ME_RAM_WADDR, 0); | |
1963 | WREG32(CP_ME_RAM_RADDR, 0); | |
1964 | return 0; | |
1965 | } | |
1966 | ||
1967 | /** | |
1968 | * cik_cp_gfx_start - start the gfx ring | |
1969 | * | |
1970 | * @rdev: radeon_device pointer | |
1971 | * | |
1972 | * Enables the ring and loads the clear state context and other | |
1973 | * packets required to init the ring. | |
1974 | * Returns 0 for success, error for failure. | |
1975 | */ | |
1976 | static int cik_cp_gfx_start(struct radeon_device *rdev) | |
1977 | { | |
1978 | struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; | |
1979 | int r, i; | |
1980 | ||
1981 | /* init the CP */ | |
1982 | WREG32(CP_MAX_CONTEXT, rdev->config.cik.max_hw_contexts - 1); | |
1983 | WREG32(CP_ENDIAN_SWAP, 0); | |
1984 | WREG32(CP_DEVICE_ID, 1); | |
1985 | ||
1986 | cik_cp_gfx_enable(rdev, true); | |
1987 | ||
1988 | r = radeon_ring_lock(rdev, ring, cik_default_size + 17); | |
1989 | if (r) { | |
1990 | DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); | |
1991 | return r; | |
1992 | } | |
1993 | ||
1994 | /* init the CE partitions. CE only used for gfx on CIK */ | |
1995 | radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); | |
1996 | radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); | |
1997 | radeon_ring_write(ring, 0xc000); | |
1998 | radeon_ring_write(ring, 0xc000); | |
1999 | ||
2000 | /* setup clear context state */ | |
2001 | radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); | |
2002 | radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); | |
2003 | ||
2004 | radeon_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); | |
2005 | radeon_ring_write(ring, 0x80000000); | |
2006 | radeon_ring_write(ring, 0x80000000); | |
2007 | ||
2008 | for (i = 0; i < cik_default_size; i++) | |
2009 | radeon_ring_write(ring, cik_default_state[i]); | |
2010 | ||
2011 | radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); | |
2012 | radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); | |
2013 | ||
2014 | /* set clear context state */ | |
2015 | radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); | |
2016 | radeon_ring_write(ring, 0); | |
2017 | ||
2018 | radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); | |
2019 | radeon_ring_write(ring, 0x00000316); | |
2020 | radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */ | |
2021 | radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */ | |
2022 | ||
2023 | radeon_ring_unlock_commit(rdev, ring); | |
2024 | ||
2025 | return 0; | |
2026 | } | |
2027 | ||
2028 | /** | |
2029 | * cik_cp_gfx_fini - stop the gfx ring | |
2030 | * | |
2031 | * @rdev: radeon_device pointer | |
2032 | * | |
2033 | * Stop the gfx ring and tear down the driver ring | |
2034 | * info. | |
2035 | */ | |
2036 | static void cik_cp_gfx_fini(struct radeon_device *rdev) | |
2037 | { | |
2038 | cik_cp_gfx_enable(rdev, false); | |
2039 | radeon_ring_fini(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]); | |
2040 | } | |
2041 | ||
2042 | /** | |
2043 | * cik_cp_gfx_resume - setup the gfx ring buffer registers | |
2044 | * | |
2045 | * @rdev: radeon_device pointer | |
2046 | * | |
2047 | * Program the location and size of the gfx ring buffer | |
2048 | * and test it to make sure it's working. | |
2049 | * Returns 0 for success, error for failure. | |
2050 | */ | |
2051 | static int cik_cp_gfx_resume(struct radeon_device *rdev) | |
2052 | { | |
2053 | struct radeon_ring *ring; | |
2054 | u32 tmp; | |
2055 | u32 rb_bufsz; | |
2056 | u64 rb_addr; | |
2057 | int r; | |
2058 | ||
2059 | WREG32(CP_SEM_WAIT_TIMER, 0x0); | |
2060 | WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0); | |
2061 | ||
2062 | /* Set the write pointer delay */ | |
2063 | WREG32(CP_RB_WPTR_DELAY, 0); | |
2064 | ||
2065 | /* set the RB to use vmid 0 */ | |
2066 | WREG32(CP_RB_VMID, 0); | |
2067 | ||
2068 | WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF); | |
2069 | ||
2070 | /* ring 0 - compute and gfx */ | |
2071 | /* Set ring buffer size */ | |
2072 | ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; | |
2073 | rb_bufsz = drm_order(ring->ring_size / 8); | |
2074 | tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; | |
2075 | #ifdef __BIG_ENDIAN | |
2076 | tmp |= BUF_SWAP_32BIT; | |
2077 | #endif | |
2078 | WREG32(CP_RB0_CNTL, tmp); | |
2079 | ||
2080 | /* Initialize the ring buffer's read and write pointers */ | |
2081 | WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA); | |
2082 | ring->wptr = 0; | |
2083 | WREG32(CP_RB0_WPTR, ring->wptr); | |
2084 | ||
2085 | /* set the wb address wether it's enabled or not */ | |
2086 | WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC); | |
2087 | WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF); | |
2088 | ||
2089 | /* scratch register shadowing is no longer supported */ | |
2090 | WREG32(SCRATCH_UMSK, 0); | |
2091 | ||
2092 | if (!rdev->wb.enabled) | |
2093 | tmp |= RB_NO_UPDATE; | |
2094 | ||
2095 | mdelay(1); | |
2096 | WREG32(CP_RB0_CNTL, tmp); | |
2097 | ||
2098 | rb_addr = ring->gpu_addr >> 8; | |
2099 | WREG32(CP_RB0_BASE, rb_addr); | |
2100 | WREG32(CP_RB0_BASE_HI, upper_32_bits(rb_addr)); | |
2101 | ||
2102 | ring->rptr = RREG32(CP_RB0_RPTR); | |
2103 | ||
2104 | /* start the ring */ | |
2105 | cik_cp_gfx_start(rdev); | |
2106 | rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true; | |
2107 | r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]); | |
2108 | if (r) { | |
2109 | rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; | |
2110 | return r; | |
2111 | } | |
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | /** | |
2116 | * cik_cp_compute_enable - enable/disable the compute CP MEs | |
2117 | * | |
2118 | * @rdev: radeon_device pointer | |
2119 | * @enable: enable or disable the MEs | |
2120 | * | |
2121 | * Halts or unhalts the compute MEs. | |
2122 | */ | |
2123 | static void cik_cp_compute_enable(struct radeon_device *rdev, bool enable) | |
2124 | { | |
2125 | if (enable) | |
2126 | WREG32(CP_MEC_CNTL, 0); | |
2127 | else | |
2128 | WREG32(CP_MEC_CNTL, (MEC_ME1_HALT | MEC_ME2_HALT)); | |
2129 | udelay(50); | |
2130 | } | |
2131 | ||
2132 | /** | |
2133 | * cik_cp_compute_load_microcode - load the compute CP ME ucode | |
2134 | * | |
2135 | * @rdev: radeon_device pointer | |
2136 | * | |
2137 | * Loads the compute MEC1&2 ucode. | |
2138 | * Returns 0 for success, -EINVAL if the ucode is not available. | |
2139 | */ | |
2140 | static int cik_cp_compute_load_microcode(struct radeon_device *rdev) | |
2141 | { | |
2142 | const __be32 *fw_data; | |
2143 | int i; | |
2144 | ||
2145 | if (!rdev->mec_fw) | |
2146 | return -EINVAL; | |
2147 | ||
2148 | cik_cp_compute_enable(rdev, false); | |
2149 | ||
2150 | /* MEC1 */ | |
2151 | fw_data = (const __be32 *)rdev->mec_fw->data; | |
2152 | WREG32(CP_MEC_ME1_UCODE_ADDR, 0); | |
2153 | for (i = 0; i < CIK_MEC_UCODE_SIZE; i++) | |
2154 | WREG32(CP_MEC_ME1_UCODE_DATA, be32_to_cpup(fw_data++)); | |
2155 | WREG32(CP_MEC_ME1_UCODE_ADDR, 0); | |
2156 | ||
2157 | if (rdev->family == CHIP_KAVERI) { | |
2158 | /* MEC2 */ | |
2159 | fw_data = (const __be32 *)rdev->mec_fw->data; | |
2160 | WREG32(CP_MEC_ME2_UCODE_ADDR, 0); | |
2161 | for (i = 0; i < CIK_MEC_UCODE_SIZE; i++) | |
2162 | WREG32(CP_MEC_ME2_UCODE_DATA, be32_to_cpup(fw_data++)); | |
2163 | WREG32(CP_MEC_ME2_UCODE_ADDR, 0); | |
2164 | } | |
2165 | ||
2166 | return 0; | |
2167 | } | |
2168 | ||
2169 | /** | |
2170 | * cik_cp_compute_start - start the compute queues | |
2171 | * | |
2172 | * @rdev: radeon_device pointer | |
2173 | * | |
2174 | * Enable the compute queues. | |
2175 | * Returns 0 for success, error for failure. | |
2176 | */ | |
2177 | static int cik_cp_compute_start(struct radeon_device *rdev) | |
2178 | { | |
2179 | //todo | |
2180 | return 0; | |
2181 | } | |
2182 | ||
2183 | /** | |
2184 | * cik_cp_compute_fini - stop the compute queues | |
2185 | * | |
2186 | * @rdev: radeon_device pointer | |
2187 | * | |
2188 | * Stop the compute queues and tear down the driver queue | |
2189 | * info. | |
2190 | */ | |
2191 | static void cik_cp_compute_fini(struct radeon_device *rdev) | |
2192 | { | |
2193 | cik_cp_compute_enable(rdev, false); | |
2194 | //todo | |
2195 | } | |
2196 | ||
2197 | /** | |
2198 | * cik_cp_compute_resume - setup the compute queue registers | |
2199 | * | |
2200 | * @rdev: radeon_device pointer | |
2201 | * | |
2202 | * Program the compute queues and test them to make sure they | |
2203 | * are working. | |
2204 | * Returns 0 for success, error for failure. | |
2205 | */ | |
2206 | static int cik_cp_compute_resume(struct radeon_device *rdev) | |
2207 | { | |
2208 | int r; | |
2209 | ||
2210 | //todo | |
2211 | r = cik_cp_compute_start(rdev); | |
2212 | if (r) | |
2213 | return r; | |
2214 | return 0; | |
2215 | } | |
2216 | ||
2217 | /* XXX temporary wrappers to handle both compute and gfx */ | |
2218 | /* XXX */ | |
2219 | static void cik_cp_enable(struct radeon_device *rdev, bool enable) | |
2220 | { | |
2221 | cik_cp_gfx_enable(rdev, enable); | |
2222 | cik_cp_compute_enable(rdev, enable); | |
2223 | } | |
2224 | ||
2225 | /* XXX */ | |
2226 | static int cik_cp_load_microcode(struct radeon_device *rdev) | |
2227 | { | |
2228 | int r; | |
2229 | ||
2230 | r = cik_cp_gfx_load_microcode(rdev); | |
2231 | if (r) | |
2232 | return r; | |
2233 | r = cik_cp_compute_load_microcode(rdev); | |
2234 | if (r) | |
2235 | return r; | |
2236 | ||
2237 | return 0; | |
2238 | } | |
2239 | ||
2240 | /* XXX */ | |
2241 | static void cik_cp_fini(struct radeon_device *rdev) | |
2242 | { | |
2243 | cik_cp_gfx_fini(rdev); | |
2244 | cik_cp_compute_fini(rdev); | |
2245 | } | |
2246 | ||
2247 | /* XXX */ | |
2248 | static int cik_cp_resume(struct radeon_device *rdev) | |
2249 | { | |
2250 | int r; | |
2251 | ||
2252 | /* Reset all cp blocks */ | |
2253 | WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP); | |
2254 | RREG32(GRBM_SOFT_RESET); | |
2255 | mdelay(15); | |
2256 | WREG32(GRBM_SOFT_RESET, 0); | |
2257 | RREG32(GRBM_SOFT_RESET); | |
2258 | ||
2259 | r = cik_cp_load_microcode(rdev); | |
2260 | if (r) | |
2261 | return r; | |
2262 | ||
2263 | r = cik_cp_gfx_resume(rdev); | |
2264 | if (r) | |
2265 | return r; | |
2266 | r = cik_cp_compute_resume(rdev); | |
2267 | if (r) | |
2268 | return r; | |
2269 | ||
2270 | return 0; | |
2271 | } | |
2272 | ||
21a93e13 AD |
2273 | /* |
2274 | * sDMA - System DMA | |
2275 | * Starting with CIK, the GPU has new asynchronous | |
2276 | * DMA engines. These engines are used for compute | |
2277 | * and gfx. There are two DMA engines (SDMA0, SDMA1) | |
2278 | * and each one supports 1 ring buffer used for gfx | |
2279 | * and 2 queues used for compute. | |
2280 | * | |
2281 | * The programming model is very similar to the CP | |
2282 | * (ring buffer, IBs, etc.), but sDMA has it's own | |
2283 | * packet format that is different from the PM4 format | |
2284 | * used by the CP. sDMA supports copying data, writing | |
2285 | * embedded data, solid fills, and a number of other | |
2286 | * things. It also has support for tiling/detiling of | |
2287 | * buffers. | |
2288 | */ | |
2289 | /** | |
2290 | * cik_sdma_ring_ib_execute - Schedule an IB on the DMA engine | |
2291 | * | |
2292 | * @rdev: radeon_device pointer | |
2293 | * @ib: IB object to schedule | |
2294 | * | |
2295 | * Schedule an IB in the DMA ring (CIK). | |
2296 | */ | |
2297 | void cik_sdma_ring_ib_execute(struct radeon_device *rdev, | |
2298 | struct radeon_ib *ib) | |
2299 | { | |
2300 | struct radeon_ring *ring = &rdev->ring[ib->ring]; | |
2301 | u32 extra_bits = (ib->vm ? ib->vm->id : 0) & 0xf; | |
2302 | ||
2303 | if (rdev->wb.enabled) { | |
2304 | u32 next_rptr = ring->wptr + 5; | |
2305 | while ((next_rptr & 7) != 4) | |
2306 | next_rptr++; | |
2307 | next_rptr += 4; | |
2308 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0)); | |
2309 | radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc); | |
2310 | radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff); | |
2311 | radeon_ring_write(ring, 1); /* number of DWs to follow */ | |
2312 | radeon_ring_write(ring, next_rptr); | |
2313 | } | |
2314 | ||
2315 | /* IB packet must end on a 8 DW boundary */ | |
2316 | while ((ring->wptr & 7) != 4) | |
2317 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0)); | |
2318 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_INDIRECT_BUFFER, 0, extra_bits)); | |
2319 | radeon_ring_write(ring, ib->gpu_addr & 0xffffffe0); /* base must be 32 byte aligned */ | |
2320 | radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xffffffff); | |
2321 | radeon_ring_write(ring, ib->length_dw); | |
2322 | ||
2323 | } | |
2324 | ||
2325 | /** | |
2326 | * cik_sdma_fence_ring_emit - emit a fence on the DMA ring | |
2327 | * | |
2328 | * @rdev: radeon_device pointer | |
2329 | * @fence: radeon fence object | |
2330 | * | |
2331 | * Add a DMA fence packet to the ring to write | |
2332 | * the fence seq number and DMA trap packet to generate | |
2333 | * an interrupt if needed (CIK). | |
2334 | */ | |
2335 | void cik_sdma_fence_ring_emit(struct radeon_device *rdev, | |
2336 | struct radeon_fence *fence) | |
2337 | { | |
2338 | struct radeon_ring *ring = &rdev->ring[fence->ring]; | |
2339 | u64 addr = rdev->fence_drv[fence->ring].gpu_addr; | |
2340 | u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) | | |
2341 | SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */ | |
2342 | u32 ref_and_mask; | |
2343 | ||
2344 | if (fence->ring == R600_RING_TYPE_DMA_INDEX) | |
2345 | ref_and_mask = SDMA0; | |
2346 | else | |
2347 | ref_and_mask = SDMA1; | |
2348 | ||
2349 | /* write the fence */ | |
2350 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_FENCE, 0, 0)); | |
2351 | radeon_ring_write(ring, addr & 0xffffffff); | |
2352 | radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff); | |
2353 | radeon_ring_write(ring, fence->seq); | |
2354 | /* generate an interrupt */ | |
2355 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_TRAP, 0, 0)); | |
2356 | /* flush HDP */ | |
2357 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits)); | |
2358 | radeon_ring_write(ring, GPU_HDP_FLUSH_DONE); | |
2359 | radeon_ring_write(ring, GPU_HDP_FLUSH_REQ); | |
2360 | radeon_ring_write(ring, ref_and_mask); /* REFERENCE */ | |
2361 | radeon_ring_write(ring, ref_and_mask); /* MASK */ | |
2362 | radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */ | |
2363 | } | |
2364 | ||
2365 | /** | |
2366 | * cik_sdma_semaphore_ring_emit - emit a semaphore on the dma ring | |
2367 | * | |
2368 | * @rdev: radeon_device pointer | |
2369 | * @ring: radeon_ring structure holding ring information | |
2370 | * @semaphore: radeon semaphore object | |
2371 | * @emit_wait: wait or signal semaphore | |
2372 | * | |
2373 | * Add a DMA semaphore packet to the ring wait on or signal | |
2374 | * other rings (CIK). | |
2375 | */ | |
2376 | void cik_sdma_semaphore_ring_emit(struct radeon_device *rdev, | |
2377 | struct radeon_ring *ring, | |
2378 | struct radeon_semaphore *semaphore, | |
2379 | bool emit_wait) | |
2380 | { | |
2381 | u64 addr = semaphore->gpu_addr; | |
2382 | u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S; | |
2383 | ||
2384 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits)); | |
2385 | radeon_ring_write(ring, addr & 0xfffffff8); | |
2386 | radeon_ring_write(ring, upper_32_bits(addr) & 0xffffffff); | |
2387 | } | |
2388 | ||
2389 | /** | |
2390 | * cik_sdma_gfx_stop - stop the gfx async dma engines | |
2391 | * | |
2392 | * @rdev: radeon_device pointer | |
2393 | * | |
2394 | * Stop the gfx async dma ring buffers (CIK). | |
2395 | */ | |
2396 | static void cik_sdma_gfx_stop(struct radeon_device *rdev) | |
2397 | { | |
2398 | u32 rb_cntl, reg_offset; | |
2399 | int i; | |
2400 | ||
2401 | radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size); | |
2402 | ||
2403 | for (i = 0; i < 2; i++) { | |
2404 | if (i == 0) | |
2405 | reg_offset = SDMA0_REGISTER_OFFSET; | |
2406 | else | |
2407 | reg_offset = SDMA1_REGISTER_OFFSET; | |
2408 | rb_cntl = RREG32(SDMA0_GFX_RB_CNTL + reg_offset); | |
2409 | rb_cntl &= ~SDMA_RB_ENABLE; | |
2410 | WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl); | |
2411 | WREG32(SDMA0_GFX_IB_CNTL + reg_offset, 0); | |
2412 | } | |
2413 | } | |
2414 | ||
2415 | /** | |
2416 | * cik_sdma_rlc_stop - stop the compute async dma engines | |
2417 | * | |
2418 | * @rdev: radeon_device pointer | |
2419 | * | |
2420 | * Stop the compute async dma queues (CIK). | |
2421 | */ | |
2422 | static void cik_sdma_rlc_stop(struct radeon_device *rdev) | |
2423 | { | |
2424 | /* XXX todo */ | |
2425 | } | |
2426 | ||
2427 | /** | |
2428 | * cik_sdma_enable - stop the async dma engines | |
2429 | * | |
2430 | * @rdev: radeon_device pointer | |
2431 | * @enable: enable/disable the DMA MEs. | |
2432 | * | |
2433 | * Halt or unhalt the async dma engines (CIK). | |
2434 | */ | |
2435 | static void cik_sdma_enable(struct radeon_device *rdev, bool enable) | |
2436 | { | |
2437 | u32 me_cntl, reg_offset; | |
2438 | int i; | |
2439 | ||
2440 | for (i = 0; i < 2; i++) { | |
2441 | if (i == 0) | |
2442 | reg_offset = SDMA0_REGISTER_OFFSET; | |
2443 | else | |
2444 | reg_offset = SDMA1_REGISTER_OFFSET; | |
2445 | me_cntl = RREG32(SDMA0_ME_CNTL + reg_offset); | |
2446 | if (enable) | |
2447 | me_cntl &= ~SDMA_HALT; | |
2448 | else | |
2449 | me_cntl |= SDMA_HALT; | |
2450 | WREG32(SDMA0_ME_CNTL + reg_offset, me_cntl); | |
2451 | } | |
2452 | } | |
2453 | ||
2454 | /** | |
2455 | * cik_sdma_gfx_resume - setup and start the async dma engines | |
2456 | * | |
2457 | * @rdev: radeon_device pointer | |
2458 | * | |
2459 | * Set up the gfx DMA ring buffers and enable them (CIK). | |
2460 | * Returns 0 for success, error for failure. | |
2461 | */ | |
2462 | static int cik_sdma_gfx_resume(struct radeon_device *rdev) | |
2463 | { | |
2464 | struct radeon_ring *ring; | |
2465 | u32 rb_cntl, ib_cntl; | |
2466 | u32 rb_bufsz; | |
2467 | u32 reg_offset, wb_offset; | |
2468 | int i, r; | |
2469 | ||
2470 | for (i = 0; i < 2; i++) { | |
2471 | if (i == 0) { | |
2472 | ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; | |
2473 | reg_offset = SDMA0_REGISTER_OFFSET; | |
2474 | wb_offset = R600_WB_DMA_RPTR_OFFSET; | |
2475 | } else { | |
2476 | ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; | |
2477 | reg_offset = SDMA1_REGISTER_OFFSET; | |
2478 | wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET; | |
2479 | } | |
2480 | ||
2481 | WREG32(SDMA0_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0); | |
2482 | WREG32(SDMA0_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0); | |
2483 | ||
2484 | /* Set ring buffer size in dwords */ | |
2485 | rb_bufsz = drm_order(ring->ring_size / 4); | |
2486 | rb_cntl = rb_bufsz << 1; | |
2487 | #ifdef __BIG_ENDIAN | |
2488 | rb_cntl |= SDMA_RB_SWAP_ENABLE | SDMA_RPTR_WRITEBACK_SWAP_ENABLE; | |
2489 | #endif | |
2490 | WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl); | |
2491 | ||
2492 | /* Initialize the ring buffer's read and write pointers */ | |
2493 | WREG32(SDMA0_GFX_RB_RPTR + reg_offset, 0); | |
2494 | WREG32(SDMA0_GFX_RB_WPTR + reg_offset, 0); | |
2495 | ||
2496 | /* set the wb address whether it's enabled or not */ | |
2497 | WREG32(SDMA0_GFX_RB_RPTR_ADDR_HI + reg_offset, | |
2498 | upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF); | |
2499 | WREG32(SDMA0_GFX_RB_RPTR_ADDR_LO + reg_offset, | |
2500 | ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC)); | |
2501 | ||
2502 | if (rdev->wb.enabled) | |
2503 | rb_cntl |= SDMA_RPTR_WRITEBACK_ENABLE; | |
2504 | ||
2505 | WREG32(SDMA0_GFX_RB_BASE + reg_offset, ring->gpu_addr >> 8); | |
2506 | WREG32(SDMA0_GFX_RB_BASE_HI + reg_offset, ring->gpu_addr >> 40); | |
2507 | ||
2508 | ring->wptr = 0; | |
2509 | WREG32(SDMA0_GFX_RB_WPTR + reg_offset, ring->wptr << 2); | |
2510 | ||
2511 | ring->rptr = RREG32(SDMA0_GFX_RB_RPTR + reg_offset) >> 2; | |
2512 | ||
2513 | /* enable DMA RB */ | |
2514 | WREG32(SDMA0_GFX_RB_CNTL + reg_offset, rb_cntl | SDMA_RB_ENABLE); | |
2515 | ||
2516 | ib_cntl = SDMA_IB_ENABLE; | |
2517 | #ifdef __BIG_ENDIAN | |
2518 | ib_cntl |= SDMA_IB_SWAP_ENABLE; | |
2519 | #endif | |
2520 | /* enable DMA IBs */ | |
2521 | WREG32(SDMA0_GFX_IB_CNTL + reg_offset, ib_cntl); | |
2522 | ||
2523 | ring->ready = true; | |
2524 | ||
2525 | r = radeon_ring_test(rdev, ring->idx, ring); | |
2526 | if (r) { | |
2527 | ring->ready = false; | |
2528 | return r; | |
2529 | } | |
2530 | } | |
2531 | ||
2532 | radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size); | |
2533 | ||
2534 | return 0; | |
2535 | } | |
2536 | ||
2537 | /** | |
2538 | * cik_sdma_rlc_resume - setup and start the async dma engines | |
2539 | * | |
2540 | * @rdev: radeon_device pointer | |
2541 | * | |
2542 | * Set up the compute DMA queues and enable them (CIK). | |
2543 | * Returns 0 for success, error for failure. | |
2544 | */ | |
2545 | static int cik_sdma_rlc_resume(struct radeon_device *rdev) | |
2546 | { | |
2547 | /* XXX todo */ | |
2548 | return 0; | |
2549 | } | |
2550 | ||
2551 | /** | |
2552 | * cik_sdma_load_microcode - load the sDMA ME ucode | |
2553 | * | |
2554 | * @rdev: radeon_device pointer | |
2555 | * | |
2556 | * Loads the sDMA0/1 ucode. | |
2557 | * Returns 0 for success, -EINVAL if the ucode is not available. | |
2558 | */ | |
2559 | static int cik_sdma_load_microcode(struct radeon_device *rdev) | |
2560 | { | |
2561 | const __be32 *fw_data; | |
2562 | int i; | |
2563 | ||
2564 | if (!rdev->sdma_fw) | |
2565 | return -EINVAL; | |
2566 | ||
2567 | /* stop the gfx rings and rlc compute queues */ | |
2568 | cik_sdma_gfx_stop(rdev); | |
2569 | cik_sdma_rlc_stop(rdev); | |
2570 | ||
2571 | /* halt the MEs */ | |
2572 | cik_sdma_enable(rdev, false); | |
2573 | ||
2574 | /* sdma0 */ | |
2575 | fw_data = (const __be32 *)rdev->sdma_fw->data; | |
2576 | WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0); | |
2577 | for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++) | |
2578 | WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, be32_to_cpup(fw_data++)); | |
2579 | WREG32(SDMA0_UCODE_DATA + SDMA0_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION); | |
2580 | ||
2581 | /* sdma1 */ | |
2582 | fw_data = (const __be32 *)rdev->sdma_fw->data; | |
2583 | WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0); | |
2584 | for (i = 0; i < CIK_SDMA_UCODE_SIZE; i++) | |
2585 | WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, be32_to_cpup(fw_data++)); | |
2586 | WREG32(SDMA0_UCODE_DATA + SDMA1_REGISTER_OFFSET, CIK_SDMA_UCODE_VERSION); | |
2587 | ||
2588 | WREG32(SDMA0_UCODE_ADDR + SDMA0_REGISTER_OFFSET, 0); | |
2589 | WREG32(SDMA0_UCODE_ADDR + SDMA1_REGISTER_OFFSET, 0); | |
2590 | return 0; | |
2591 | } | |
2592 | ||
2593 | /** | |
2594 | * cik_sdma_resume - setup and start the async dma engines | |
2595 | * | |
2596 | * @rdev: radeon_device pointer | |
2597 | * | |
2598 | * Set up the DMA engines and enable them (CIK). | |
2599 | * Returns 0 for success, error for failure. | |
2600 | */ | |
2601 | static int cik_sdma_resume(struct radeon_device *rdev) | |
2602 | { | |
2603 | int r; | |
2604 | ||
2605 | /* Reset dma */ | |
2606 | WREG32(SRBM_SOFT_RESET, SOFT_RESET_SDMA | SOFT_RESET_SDMA1); | |
2607 | RREG32(SRBM_SOFT_RESET); | |
2608 | udelay(50); | |
2609 | WREG32(SRBM_SOFT_RESET, 0); | |
2610 | RREG32(SRBM_SOFT_RESET); | |
2611 | ||
2612 | r = cik_sdma_load_microcode(rdev); | |
2613 | if (r) | |
2614 | return r; | |
2615 | ||
2616 | /* unhalt the MEs */ | |
2617 | cik_sdma_enable(rdev, true); | |
2618 | ||
2619 | /* start the gfx rings and rlc compute queues */ | |
2620 | r = cik_sdma_gfx_resume(rdev); | |
2621 | if (r) | |
2622 | return r; | |
2623 | r = cik_sdma_rlc_resume(rdev); | |
2624 | if (r) | |
2625 | return r; | |
2626 | ||
2627 | return 0; | |
2628 | } | |
2629 | ||
2630 | /** | |
2631 | * cik_sdma_fini - tear down the async dma engines | |
2632 | * | |
2633 | * @rdev: radeon_device pointer | |
2634 | * | |
2635 | * Stop the async dma engines and free the rings (CIK). | |
2636 | */ | |
2637 | static void cik_sdma_fini(struct radeon_device *rdev) | |
2638 | { | |
2639 | /* stop the gfx rings and rlc compute queues */ | |
2640 | cik_sdma_gfx_stop(rdev); | |
2641 | cik_sdma_rlc_stop(rdev); | |
2642 | /* halt the MEs */ | |
2643 | cik_sdma_enable(rdev, false); | |
2644 | radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]); | |
2645 | radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]); | |
2646 | /* XXX - compute dma queue tear down */ | |
2647 | } | |
2648 | ||
2649 | /** | |
2650 | * cik_copy_dma - copy pages using the DMA engine | |
2651 | * | |
2652 | * @rdev: radeon_device pointer | |
2653 | * @src_offset: src GPU address | |
2654 | * @dst_offset: dst GPU address | |
2655 | * @num_gpu_pages: number of GPU pages to xfer | |
2656 | * @fence: radeon fence object | |
2657 | * | |
2658 | * Copy GPU paging using the DMA engine (CIK). | |
2659 | * Used by the radeon ttm implementation to move pages if | |
2660 | * registered as the asic copy callback. | |
2661 | */ | |
2662 | int cik_copy_dma(struct radeon_device *rdev, | |
2663 | uint64_t src_offset, uint64_t dst_offset, | |
2664 | unsigned num_gpu_pages, | |
2665 | struct radeon_fence **fence) | |
2666 | { | |
2667 | struct radeon_semaphore *sem = NULL; | |
2668 | int ring_index = rdev->asic->copy.dma_ring_index; | |
2669 | struct radeon_ring *ring = &rdev->ring[ring_index]; | |
2670 | u32 size_in_bytes, cur_size_in_bytes; | |
2671 | int i, num_loops; | |
2672 | int r = 0; | |
2673 | ||
2674 | r = radeon_semaphore_create(rdev, &sem); | |
2675 | if (r) { | |
2676 | DRM_ERROR("radeon: moving bo (%d).\n", r); | |
2677 | return r; | |
2678 | } | |
2679 | ||
2680 | size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT); | |
2681 | num_loops = DIV_ROUND_UP(size_in_bytes, 0x1fffff); | |
2682 | r = radeon_ring_lock(rdev, ring, num_loops * 7 + 14); | |
2683 | if (r) { | |
2684 | DRM_ERROR("radeon: moving bo (%d).\n", r); | |
2685 | radeon_semaphore_free(rdev, &sem, NULL); | |
2686 | return r; | |
2687 | } | |
2688 | ||
2689 | if (radeon_fence_need_sync(*fence, ring->idx)) { | |
2690 | radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, | |
2691 | ring->idx); | |
2692 | radeon_fence_note_sync(*fence, ring->idx); | |
2693 | } else { | |
2694 | radeon_semaphore_free(rdev, &sem, NULL); | |
2695 | } | |
2696 | ||
2697 | for (i = 0; i < num_loops; i++) { | |
2698 | cur_size_in_bytes = size_in_bytes; | |
2699 | if (cur_size_in_bytes > 0x1fffff) | |
2700 | cur_size_in_bytes = 0x1fffff; | |
2701 | size_in_bytes -= cur_size_in_bytes; | |
2702 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0)); | |
2703 | radeon_ring_write(ring, cur_size_in_bytes); | |
2704 | radeon_ring_write(ring, 0); /* src/dst endian swap */ | |
2705 | radeon_ring_write(ring, src_offset & 0xffffffff); | |
2706 | radeon_ring_write(ring, upper_32_bits(src_offset) & 0xffffffff); | |
2707 | radeon_ring_write(ring, dst_offset & 0xfffffffc); | |
2708 | radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xffffffff); | |
2709 | src_offset += cur_size_in_bytes; | |
2710 | dst_offset += cur_size_in_bytes; | |
2711 | } | |
2712 | ||
2713 | r = radeon_fence_emit(rdev, fence, ring->idx); | |
2714 | if (r) { | |
2715 | radeon_ring_unlock_undo(rdev, ring); | |
2716 | return r; | |
2717 | } | |
2718 | ||
2719 | radeon_ring_unlock_commit(rdev, ring); | |
2720 | radeon_semaphore_free(rdev, &sem, *fence); | |
2721 | ||
2722 | return r; | |
2723 | } | |
2724 | ||
2725 | /** | |
2726 | * cik_sdma_ring_test - simple async dma engine test | |
2727 | * | |
2728 | * @rdev: radeon_device pointer | |
2729 | * @ring: radeon_ring structure holding ring information | |
2730 | * | |
2731 | * Test the DMA engine by writing using it to write an | |
2732 | * value to memory. (CIK). | |
2733 | * Returns 0 for success, error for failure. | |
2734 | */ | |
2735 | int cik_sdma_ring_test(struct radeon_device *rdev, | |
2736 | struct radeon_ring *ring) | |
2737 | { | |
2738 | unsigned i; | |
2739 | int r; | |
2740 | void __iomem *ptr = (void *)rdev->vram_scratch.ptr; | |
2741 | u32 tmp; | |
2742 | ||
2743 | if (!ptr) { | |
2744 | DRM_ERROR("invalid vram scratch pointer\n"); | |
2745 | return -EINVAL; | |
2746 | } | |
2747 | ||
2748 | tmp = 0xCAFEDEAD; | |
2749 | writel(tmp, ptr); | |
2750 | ||
2751 | r = radeon_ring_lock(rdev, ring, 4); | |
2752 | if (r) { | |
2753 | DRM_ERROR("radeon: dma failed to lock ring %d (%d).\n", ring->idx, r); | |
2754 | return r; | |
2755 | } | |
2756 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0)); | |
2757 | radeon_ring_write(ring, rdev->vram_scratch.gpu_addr & 0xfffffffc); | |
2758 | radeon_ring_write(ring, upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff); | |
2759 | radeon_ring_write(ring, 1); /* number of DWs to follow */ | |
2760 | radeon_ring_write(ring, 0xDEADBEEF); | |
2761 | radeon_ring_unlock_commit(rdev, ring); | |
2762 | ||
2763 | for (i = 0; i < rdev->usec_timeout; i++) { | |
2764 | tmp = readl(ptr); | |
2765 | if (tmp == 0xDEADBEEF) | |
2766 | break; | |
2767 | DRM_UDELAY(1); | |
2768 | } | |
2769 | ||
2770 | if (i < rdev->usec_timeout) { | |
2771 | DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i); | |
2772 | } else { | |
2773 | DRM_ERROR("radeon: ring %d test failed (0x%08X)\n", | |
2774 | ring->idx, tmp); | |
2775 | r = -EINVAL; | |
2776 | } | |
2777 | return r; | |
2778 | } | |
2779 | ||
2780 | /** | |
2781 | * cik_sdma_ib_test - test an IB on the DMA engine | |
2782 | * | |
2783 | * @rdev: radeon_device pointer | |
2784 | * @ring: radeon_ring structure holding ring information | |
2785 | * | |
2786 | * Test a simple IB in the DMA ring (CIK). | |
2787 | * Returns 0 on success, error on failure. | |
2788 | */ | |
2789 | int cik_sdma_ib_test(struct radeon_device *rdev, struct radeon_ring *ring) | |
2790 | { | |
2791 | struct radeon_ib ib; | |
2792 | unsigned i; | |
2793 | int r; | |
2794 | void __iomem *ptr = (void *)rdev->vram_scratch.ptr; | |
2795 | u32 tmp = 0; | |
2796 | ||
2797 | if (!ptr) { | |
2798 | DRM_ERROR("invalid vram scratch pointer\n"); | |
2799 | return -EINVAL; | |
2800 | } | |
2801 | ||
2802 | tmp = 0xCAFEDEAD; | |
2803 | writel(tmp, ptr); | |
2804 | ||
2805 | r = radeon_ib_get(rdev, ring->idx, &ib, NULL, 256); | |
2806 | if (r) { | |
2807 | DRM_ERROR("radeon: failed to get ib (%d).\n", r); | |
2808 | return r; | |
2809 | } | |
2810 | ||
2811 | ib.ptr[0] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0); | |
2812 | ib.ptr[1] = rdev->vram_scratch.gpu_addr & 0xfffffffc; | |
2813 | ib.ptr[2] = upper_32_bits(rdev->vram_scratch.gpu_addr) & 0xffffffff; | |
2814 | ib.ptr[3] = 1; | |
2815 | ib.ptr[4] = 0xDEADBEEF; | |
2816 | ib.length_dw = 5; | |
2817 | ||
2818 | r = radeon_ib_schedule(rdev, &ib, NULL); | |
2819 | if (r) { | |
2820 | radeon_ib_free(rdev, &ib); | |
2821 | DRM_ERROR("radeon: failed to schedule ib (%d).\n", r); | |
2822 | return r; | |
2823 | } | |
2824 | r = radeon_fence_wait(ib.fence, false); | |
2825 | if (r) { | |
2826 | DRM_ERROR("radeon: fence wait failed (%d).\n", r); | |
2827 | return r; | |
2828 | } | |
2829 | for (i = 0; i < rdev->usec_timeout; i++) { | |
2830 | tmp = readl(ptr); | |
2831 | if (tmp == 0xDEADBEEF) | |
2832 | break; | |
2833 | DRM_UDELAY(1); | |
2834 | } | |
2835 | if (i < rdev->usec_timeout) { | |
2836 | DRM_INFO("ib test on ring %d succeeded in %u usecs\n", ib.fence->ring, i); | |
2837 | } else { | |
2838 | DRM_ERROR("radeon: ib test failed (0x%08X)\n", tmp); | |
2839 | r = -EINVAL; | |
2840 | } | |
2841 | radeon_ib_free(rdev, &ib); | |
2842 | return r; | |
2843 | } | |
2844 | ||
6f2043ce | 2845 | |
cc066715 | 2846 | static void cik_print_gpu_status_regs(struct radeon_device *rdev) |
6f2043ce | 2847 | { |
6f2043ce AD |
2848 | dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n", |
2849 | RREG32(GRBM_STATUS)); | |
2850 | dev_info(rdev->dev, " GRBM_STATUS2=0x%08X\n", | |
2851 | RREG32(GRBM_STATUS2)); | |
2852 | dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n", | |
2853 | RREG32(GRBM_STATUS_SE0)); | |
2854 | dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n", | |
2855 | RREG32(GRBM_STATUS_SE1)); | |
2856 | dev_info(rdev->dev, " GRBM_STATUS_SE2=0x%08X\n", | |
2857 | RREG32(GRBM_STATUS_SE2)); | |
2858 | dev_info(rdev->dev, " GRBM_STATUS_SE3=0x%08X\n", | |
2859 | RREG32(GRBM_STATUS_SE3)); | |
2860 | dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n", | |
2861 | RREG32(SRBM_STATUS)); | |
2862 | dev_info(rdev->dev, " SRBM_STATUS2=0x%08X\n", | |
2863 | RREG32(SRBM_STATUS2)); | |
cc066715 AD |
2864 | dev_info(rdev->dev, " SDMA0_STATUS_REG = 0x%08X\n", |
2865 | RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET)); | |
2866 | dev_info(rdev->dev, " SDMA1_STATUS_REG = 0x%08X\n", | |
2867 | RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET)); | |
2868 | } | |
6f2043ce | 2869 | |
cc066715 AD |
2870 | /** |
2871 | * cik_gpu_check_soft_reset - check which blocks are busy | |
2872 | * | |
2873 | * @rdev: radeon_device pointer | |
2874 | * | |
2875 | * Check which blocks are busy and return the relevant reset | |
2876 | * mask to be used by cik_gpu_soft_reset(). | |
2877 | * Returns a mask of the blocks to be reset. | |
2878 | */ | |
2879 | static u32 cik_gpu_check_soft_reset(struct radeon_device *rdev) | |
2880 | { | |
2881 | u32 reset_mask = 0; | |
2882 | u32 tmp; | |
6f2043ce | 2883 | |
cc066715 AD |
2884 | /* GRBM_STATUS */ |
2885 | tmp = RREG32(GRBM_STATUS); | |
2886 | if (tmp & (PA_BUSY | SC_BUSY | | |
2887 | BCI_BUSY | SX_BUSY | | |
2888 | TA_BUSY | VGT_BUSY | | |
2889 | DB_BUSY | CB_BUSY | | |
2890 | GDS_BUSY | SPI_BUSY | | |
2891 | IA_BUSY | IA_BUSY_NO_DMA)) | |
2892 | reset_mask |= RADEON_RESET_GFX; | |
2893 | ||
2894 | if (tmp & (CP_BUSY | CP_COHERENCY_BUSY)) | |
2895 | reset_mask |= RADEON_RESET_CP; | |
2896 | ||
2897 | /* GRBM_STATUS2 */ | |
2898 | tmp = RREG32(GRBM_STATUS2); | |
2899 | if (tmp & RLC_BUSY) | |
2900 | reset_mask |= RADEON_RESET_RLC; | |
2901 | ||
2902 | /* SDMA0_STATUS_REG */ | |
2903 | tmp = RREG32(SDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET); | |
2904 | if (!(tmp & SDMA_IDLE)) | |
2905 | reset_mask |= RADEON_RESET_DMA; | |
2906 | ||
2907 | /* SDMA1_STATUS_REG */ | |
2908 | tmp = RREG32(SDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET); | |
2909 | if (!(tmp & SDMA_IDLE)) | |
2910 | reset_mask |= RADEON_RESET_DMA1; | |
2911 | ||
2912 | /* SRBM_STATUS2 */ | |
2913 | tmp = RREG32(SRBM_STATUS2); | |
2914 | if (tmp & SDMA_BUSY) | |
2915 | reset_mask |= RADEON_RESET_DMA; | |
2916 | ||
2917 | if (tmp & SDMA1_BUSY) | |
2918 | reset_mask |= RADEON_RESET_DMA1; | |
2919 | ||
2920 | /* SRBM_STATUS */ | |
2921 | tmp = RREG32(SRBM_STATUS); | |
2922 | ||
2923 | if (tmp & IH_BUSY) | |
2924 | reset_mask |= RADEON_RESET_IH; | |
2925 | ||
2926 | if (tmp & SEM_BUSY) | |
2927 | reset_mask |= RADEON_RESET_SEM; | |
2928 | ||
2929 | if (tmp & GRBM_RQ_PENDING) | |
2930 | reset_mask |= RADEON_RESET_GRBM; | |
2931 | ||
2932 | if (tmp & VMC_BUSY) | |
2933 | reset_mask |= RADEON_RESET_VMC; | |
2934 | ||
2935 | if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY | | |
2936 | MCC_BUSY | MCD_BUSY)) | |
2937 | reset_mask |= RADEON_RESET_MC; | |
2938 | ||
2939 | if (evergreen_is_display_hung(rdev)) | |
2940 | reset_mask |= RADEON_RESET_DISPLAY; | |
2941 | ||
2942 | /* Skip MC reset as it's mostly likely not hung, just busy */ | |
2943 | if (reset_mask & RADEON_RESET_MC) { | |
2944 | DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask); | |
2945 | reset_mask &= ~RADEON_RESET_MC; | |
2946 | } | |
2947 | ||
2948 | return reset_mask; | |
6f2043ce AD |
2949 | } |
2950 | ||
2951 | /** | |
cc066715 | 2952 | * cik_gpu_soft_reset - soft reset GPU |
6f2043ce AD |
2953 | * |
2954 | * @rdev: radeon_device pointer | |
cc066715 | 2955 | * @reset_mask: mask of which blocks to reset |
6f2043ce | 2956 | * |
cc066715 | 2957 | * Soft reset the blocks specified in @reset_mask. |
6f2043ce | 2958 | */ |
cc066715 | 2959 | static void cik_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask) |
6f2043ce AD |
2960 | { |
2961 | struct evergreen_mc_save save; | |
cc066715 AD |
2962 | u32 grbm_soft_reset = 0, srbm_soft_reset = 0; |
2963 | u32 tmp; | |
2964 | ||
2965 | if (reset_mask == 0) | |
2966 | return; | |
2967 | ||
2968 | dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask); | |
2969 | ||
2970 | cik_print_gpu_status_regs(rdev); | |
2971 | dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", | |
2972 | RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR)); | |
2973 | dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", | |
2974 | RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS)); | |
2975 | ||
2976 | /* stop the rlc */ | |
2977 | cik_rlc_stop(rdev); | |
2978 | ||
2979 | /* Disable GFX parsing/prefetching */ | |
2980 | WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT); | |
2981 | ||
2982 | /* Disable MEC parsing/prefetching */ | |
2983 | WREG32(CP_MEC_CNTL, MEC_ME1_HALT | MEC_ME2_HALT); | |
2984 | ||
2985 | if (reset_mask & RADEON_RESET_DMA) { | |
2986 | /* sdma0 */ | |
2987 | tmp = RREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET); | |
2988 | tmp |= SDMA_HALT; | |
2989 | WREG32(SDMA0_ME_CNTL + SDMA0_REGISTER_OFFSET, tmp); | |
2990 | } | |
2991 | if (reset_mask & RADEON_RESET_DMA1) { | |
2992 | /* sdma1 */ | |
2993 | tmp = RREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET); | |
2994 | tmp |= SDMA_HALT; | |
2995 | WREG32(SDMA0_ME_CNTL + SDMA1_REGISTER_OFFSET, tmp); | |
2996 | } | |
6f2043ce | 2997 | |
6f2043ce | 2998 | evergreen_mc_stop(rdev, &save); |
cc066715 | 2999 | if (evergreen_mc_wait_for_idle(rdev)) { |
6f2043ce AD |
3000 | dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); |
3001 | } | |
6f2043ce | 3002 | |
cc066715 AD |
3003 | if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP)) |
3004 | grbm_soft_reset = SOFT_RESET_CP | SOFT_RESET_GFX; | |
3005 | ||
3006 | if (reset_mask & RADEON_RESET_CP) { | |
3007 | grbm_soft_reset |= SOFT_RESET_CP; | |
3008 | ||
3009 | srbm_soft_reset |= SOFT_RESET_GRBM; | |
3010 | } | |
3011 | ||
3012 | if (reset_mask & RADEON_RESET_DMA) | |
3013 | srbm_soft_reset |= SOFT_RESET_SDMA; | |
3014 | ||
3015 | if (reset_mask & RADEON_RESET_DMA1) | |
3016 | srbm_soft_reset |= SOFT_RESET_SDMA1; | |
3017 | ||
3018 | if (reset_mask & RADEON_RESET_DISPLAY) | |
3019 | srbm_soft_reset |= SOFT_RESET_DC; | |
3020 | ||
3021 | if (reset_mask & RADEON_RESET_RLC) | |
3022 | grbm_soft_reset |= SOFT_RESET_RLC; | |
3023 | ||
3024 | if (reset_mask & RADEON_RESET_SEM) | |
3025 | srbm_soft_reset |= SOFT_RESET_SEM; | |
3026 | ||
3027 | if (reset_mask & RADEON_RESET_IH) | |
3028 | srbm_soft_reset |= SOFT_RESET_IH; | |
3029 | ||
3030 | if (reset_mask & RADEON_RESET_GRBM) | |
3031 | srbm_soft_reset |= SOFT_RESET_GRBM; | |
3032 | ||
3033 | if (reset_mask & RADEON_RESET_VMC) | |
3034 | srbm_soft_reset |= SOFT_RESET_VMC; | |
3035 | ||
3036 | if (!(rdev->flags & RADEON_IS_IGP)) { | |
3037 | if (reset_mask & RADEON_RESET_MC) | |
3038 | srbm_soft_reset |= SOFT_RESET_MC; | |
3039 | } | |
3040 | ||
3041 | if (grbm_soft_reset) { | |
3042 | tmp = RREG32(GRBM_SOFT_RESET); | |
3043 | tmp |= grbm_soft_reset; | |
3044 | dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); | |
3045 | WREG32(GRBM_SOFT_RESET, tmp); | |
3046 | tmp = RREG32(GRBM_SOFT_RESET); | |
3047 | ||
3048 | udelay(50); | |
3049 | ||
3050 | tmp &= ~grbm_soft_reset; | |
3051 | WREG32(GRBM_SOFT_RESET, tmp); | |
3052 | tmp = RREG32(GRBM_SOFT_RESET); | |
3053 | } | |
3054 | ||
3055 | if (srbm_soft_reset) { | |
3056 | tmp = RREG32(SRBM_SOFT_RESET); | |
3057 | tmp |= srbm_soft_reset; | |
3058 | dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); | |
3059 | WREG32(SRBM_SOFT_RESET, tmp); | |
3060 | tmp = RREG32(SRBM_SOFT_RESET); | |
3061 | ||
3062 | udelay(50); | |
3063 | ||
3064 | tmp &= ~srbm_soft_reset; | |
3065 | WREG32(SRBM_SOFT_RESET, tmp); | |
3066 | tmp = RREG32(SRBM_SOFT_RESET); | |
3067 | } | |
6f2043ce | 3068 | |
6f2043ce AD |
3069 | /* Wait a little for things to settle down */ |
3070 | udelay(50); | |
cc066715 | 3071 | |
6f2043ce | 3072 | evergreen_mc_resume(rdev, &save); |
cc066715 AD |
3073 | udelay(50); |
3074 | ||
3075 | cik_print_gpu_status_regs(rdev); | |
6f2043ce AD |
3076 | } |
3077 | ||
3078 | /** | |
cc066715 | 3079 | * cik_asic_reset - soft reset GPU |
6f2043ce AD |
3080 | * |
3081 | * @rdev: radeon_device pointer | |
3082 | * | |
cc066715 AD |
3083 | * Look up which blocks are hung and attempt |
3084 | * to reset them. | |
6f2043ce AD |
3085 | * Returns 0 for success. |
3086 | */ | |
3087 | int cik_asic_reset(struct radeon_device *rdev) | |
3088 | { | |
cc066715 | 3089 | u32 reset_mask; |
6f2043ce | 3090 | |
cc066715 AD |
3091 | reset_mask = cik_gpu_check_soft_reset(rdev); |
3092 | ||
3093 | if (reset_mask) | |
3094 | r600_set_bios_scratch_engine_hung(rdev, true); | |
3095 | ||
3096 | cik_gpu_soft_reset(rdev, reset_mask); | |
6f2043ce | 3097 | |
cc066715 AD |
3098 | reset_mask = cik_gpu_check_soft_reset(rdev); |
3099 | ||
3100 | if (!reset_mask) | |
3101 | r600_set_bios_scratch_engine_hung(rdev, false); | |
3102 | ||
3103 | return 0; | |
3104 | } | |
3105 | ||
3106 | /** | |
3107 | * cik_gfx_is_lockup - check if the 3D engine is locked up | |
3108 | * | |
3109 | * @rdev: radeon_device pointer | |
3110 | * @ring: radeon_ring structure holding ring information | |
3111 | * | |
3112 | * Check if the 3D engine is locked up (CIK). | |
3113 | * Returns true if the engine is locked, false if not. | |
3114 | */ | |
3115 | bool cik_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring) | |
3116 | { | |
3117 | u32 reset_mask = cik_gpu_check_soft_reset(rdev); | |
3118 | ||
3119 | if (!(reset_mask & (RADEON_RESET_GFX | | |
3120 | RADEON_RESET_COMPUTE | | |
3121 | RADEON_RESET_CP))) { | |
3122 | radeon_ring_lockup_update(ring); | |
3123 | return false; | |
3124 | } | |
3125 | /* force CP activities */ | |
3126 | radeon_ring_force_activity(rdev, ring); | |
3127 | return radeon_ring_test_lockup(rdev, ring); | |
6f2043ce | 3128 | } |
1c49165d | 3129 | |
21a93e13 AD |
3130 | /** |
3131 | * cik_sdma_is_lockup - Check if the DMA engine is locked up | |
3132 | * | |
3133 | * @rdev: radeon_device pointer | |
3134 | * @ring: radeon_ring structure holding ring information | |
3135 | * | |
3136 | * Check if the async DMA engine is locked up (CIK). | |
3137 | * Returns true if the engine appears to be locked up, false if not. | |
3138 | */ | |
3139 | bool cik_sdma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring) | |
3140 | { | |
cc066715 AD |
3141 | u32 reset_mask = cik_gpu_check_soft_reset(rdev); |
3142 | u32 mask; | |
21a93e13 AD |
3143 | |
3144 | if (ring->idx == R600_RING_TYPE_DMA_INDEX) | |
cc066715 | 3145 | mask = RADEON_RESET_DMA; |
21a93e13 | 3146 | else |
cc066715 AD |
3147 | mask = RADEON_RESET_DMA1; |
3148 | ||
3149 | if (!(reset_mask & mask)) { | |
21a93e13 AD |
3150 | radeon_ring_lockup_update(ring); |
3151 | return false; | |
3152 | } | |
3153 | /* force ring activities */ | |
3154 | radeon_ring_force_activity(rdev, ring); | |
3155 | return radeon_ring_test_lockup(rdev, ring); | |
3156 | } | |
3157 | ||
1c49165d AD |
3158 | /* MC */ |
3159 | /** | |
3160 | * cik_mc_program - program the GPU memory controller | |
3161 | * | |
3162 | * @rdev: radeon_device pointer | |
3163 | * | |
3164 | * Set the location of vram, gart, and AGP in the GPU's | |
3165 | * physical address space (CIK). | |
3166 | */ | |
3167 | static void cik_mc_program(struct radeon_device *rdev) | |
3168 | { | |
3169 | struct evergreen_mc_save save; | |
3170 | u32 tmp; | |
3171 | int i, j; | |
3172 | ||
3173 | /* Initialize HDP */ | |
3174 | for (i = 0, j = 0; i < 32; i++, j += 0x18) { | |
3175 | WREG32((0x2c14 + j), 0x00000000); | |
3176 | WREG32((0x2c18 + j), 0x00000000); | |
3177 | WREG32((0x2c1c + j), 0x00000000); | |
3178 | WREG32((0x2c20 + j), 0x00000000); | |
3179 | WREG32((0x2c24 + j), 0x00000000); | |
3180 | } | |
3181 | WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0); | |
3182 | ||
3183 | evergreen_mc_stop(rdev, &save); | |
3184 | if (radeon_mc_wait_for_idle(rdev)) { | |
3185 | dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); | |
3186 | } | |
3187 | /* Lockout access through VGA aperture*/ | |
3188 | WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE); | |
3189 | /* Update configuration */ | |
3190 | WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, | |
3191 | rdev->mc.vram_start >> 12); | |
3192 | WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, | |
3193 | rdev->mc.vram_end >> 12); | |
3194 | WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, | |
3195 | rdev->vram_scratch.gpu_addr >> 12); | |
3196 | tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16; | |
3197 | tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF); | |
3198 | WREG32(MC_VM_FB_LOCATION, tmp); | |
3199 | /* XXX double check these! */ | |
3200 | WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8)); | |
3201 | WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30)); | |
3202 | WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF); | |
3203 | WREG32(MC_VM_AGP_BASE, 0); | |
3204 | WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF); | |
3205 | WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF); | |
3206 | if (radeon_mc_wait_for_idle(rdev)) { | |
3207 | dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); | |
3208 | } | |
3209 | evergreen_mc_resume(rdev, &save); | |
3210 | /* we need to own VRAM, so turn off the VGA renderer here | |
3211 | * to stop it overwriting our objects */ | |
3212 | rv515_vga_render_disable(rdev); | |
3213 | } | |
3214 | ||
3215 | /** | |
3216 | * cik_mc_init - initialize the memory controller driver params | |
3217 | * | |
3218 | * @rdev: radeon_device pointer | |
3219 | * | |
3220 | * Look up the amount of vram, vram width, and decide how to place | |
3221 | * vram and gart within the GPU's physical address space (CIK). | |
3222 | * Returns 0 for success. | |
3223 | */ | |
3224 | static int cik_mc_init(struct radeon_device *rdev) | |
3225 | { | |
3226 | u32 tmp; | |
3227 | int chansize, numchan; | |
3228 | ||
3229 | /* Get VRAM informations */ | |
3230 | rdev->mc.vram_is_ddr = true; | |
3231 | tmp = RREG32(MC_ARB_RAMCFG); | |
3232 | if (tmp & CHANSIZE_MASK) { | |
3233 | chansize = 64; | |
3234 | } else { | |
3235 | chansize = 32; | |
3236 | } | |
3237 | tmp = RREG32(MC_SHARED_CHMAP); | |
3238 | switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) { | |
3239 | case 0: | |
3240 | default: | |
3241 | numchan = 1; | |
3242 | break; | |
3243 | case 1: | |
3244 | numchan = 2; | |
3245 | break; | |
3246 | case 2: | |
3247 | numchan = 4; | |
3248 | break; | |
3249 | case 3: | |
3250 | numchan = 8; | |
3251 | break; | |
3252 | case 4: | |
3253 | numchan = 3; | |
3254 | break; | |
3255 | case 5: | |
3256 | numchan = 6; | |
3257 | break; | |
3258 | case 6: | |
3259 | numchan = 10; | |
3260 | break; | |
3261 | case 7: | |
3262 | numchan = 12; | |
3263 | break; | |
3264 | case 8: | |
3265 | numchan = 16; | |
3266 | break; | |
3267 | } | |
3268 | rdev->mc.vram_width = numchan * chansize; | |
3269 | /* Could aper size report 0 ? */ | |
3270 | rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0); | |
3271 | rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0); | |
3272 | /* size in MB on si */ | |
3273 | rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024; | |
3274 | rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024; | |
3275 | rdev->mc.visible_vram_size = rdev->mc.aper_size; | |
3276 | si_vram_gtt_location(rdev, &rdev->mc); | |
3277 | radeon_update_bandwidth_info(rdev); | |
3278 | ||
3279 | return 0; | |
3280 | } | |
3281 | ||
3282 | /* | |
3283 | * GART | |
3284 | * VMID 0 is the physical GPU addresses as used by the kernel. | |
3285 | * VMIDs 1-15 are used for userspace clients and are handled | |
3286 | * by the radeon vm/hsa code. | |
3287 | */ | |
3288 | /** | |
3289 | * cik_pcie_gart_tlb_flush - gart tlb flush callback | |
3290 | * | |
3291 | * @rdev: radeon_device pointer | |
3292 | * | |
3293 | * Flush the TLB for the VMID 0 page table (CIK). | |
3294 | */ | |
3295 | void cik_pcie_gart_tlb_flush(struct radeon_device *rdev) | |
3296 | { | |
3297 | /* flush hdp cache */ | |
3298 | WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0); | |
3299 | ||
3300 | /* bits 0-15 are the VM contexts0-15 */ | |
3301 | WREG32(VM_INVALIDATE_REQUEST, 0x1); | |
3302 | } | |
3303 | ||
3304 | /** | |
3305 | * cik_pcie_gart_enable - gart enable | |
3306 | * | |
3307 | * @rdev: radeon_device pointer | |
3308 | * | |
3309 | * This sets up the TLBs, programs the page tables for VMID0, | |
3310 | * sets up the hw for VMIDs 1-15 which are allocated on | |
3311 | * demand, and sets up the global locations for the LDS, GDS, | |
3312 | * and GPUVM for FSA64 clients (CIK). | |
3313 | * Returns 0 for success, errors for failure. | |
3314 | */ | |
3315 | static int cik_pcie_gart_enable(struct radeon_device *rdev) | |
3316 | { | |
3317 | int r, i; | |
3318 | ||
3319 | if (rdev->gart.robj == NULL) { | |
3320 | dev_err(rdev->dev, "No VRAM object for PCIE GART.\n"); | |
3321 | return -EINVAL; | |
3322 | } | |
3323 | r = radeon_gart_table_vram_pin(rdev); | |
3324 | if (r) | |
3325 | return r; | |
3326 | radeon_gart_restore(rdev); | |
3327 | /* Setup TLB control */ | |
3328 | WREG32(MC_VM_MX_L1_TLB_CNTL, | |
3329 | (0xA << 7) | | |
3330 | ENABLE_L1_TLB | | |
3331 | SYSTEM_ACCESS_MODE_NOT_IN_SYS | | |
3332 | ENABLE_ADVANCED_DRIVER_MODEL | | |
3333 | SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); | |
3334 | /* Setup L2 cache */ | |
3335 | WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | | |
3336 | ENABLE_L2_FRAGMENT_PROCESSING | | |
3337 | ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | | |
3338 | ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | | |
3339 | EFFECTIVE_L2_QUEUE_SIZE(7) | | |
3340 | CONTEXT1_IDENTITY_ACCESS_MODE(1)); | |
3341 | WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE); | |
3342 | WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | | |
3343 | L2_CACHE_BIGK_FRAGMENT_SIZE(6)); | |
3344 | /* setup context0 */ | |
3345 | WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12); | |
3346 | WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12); | |
3347 | WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12); | |
3348 | WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, | |
3349 | (u32)(rdev->dummy_page.addr >> 12)); | |
3350 | WREG32(VM_CONTEXT0_CNTL2, 0); | |
3351 | WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) | | |
3352 | RANGE_PROTECTION_FAULT_ENABLE_DEFAULT)); | |
3353 | ||
3354 | WREG32(0x15D4, 0); | |
3355 | WREG32(0x15D8, 0); | |
3356 | WREG32(0x15DC, 0); | |
3357 | ||
3358 | /* empty context1-15 */ | |
3359 | /* FIXME start with 4G, once using 2 level pt switch to full | |
3360 | * vm size space | |
3361 | */ | |
3362 | /* set vm size, must be a multiple of 4 */ | |
3363 | WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0); | |
3364 | WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn); | |
3365 | for (i = 1; i < 16; i++) { | |
3366 | if (i < 8) | |
3367 | WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2), | |
3368 | rdev->gart.table_addr >> 12); | |
3369 | else | |
3370 | WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2), | |
3371 | rdev->gart.table_addr >> 12); | |
3372 | } | |
3373 | ||
3374 | /* enable context1-15 */ | |
3375 | WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR, | |
3376 | (u32)(rdev->dummy_page.addr >> 12)); | |
a00024b0 | 3377 | WREG32(VM_CONTEXT1_CNTL2, 4); |
1c49165d | 3378 | WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) | |
a00024b0 AD |
3379 | RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT | |
3380 | RANGE_PROTECTION_FAULT_ENABLE_DEFAULT | | |
3381 | DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT | | |
3382 | DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT | | |
3383 | PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT | | |
3384 | PDE0_PROTECTION_FAULT_ENABLE_DEFAULT | | |
3385 | VALID_PROTECTION_FAULT_ENABLE_INTERRUPT | | |
3386 | VALID_PROTECTION_FAULT_ENABLE_DEFAULT | | |
3387 | READ_PROTECTION_FAULT_ENABLE_INTERRUPT | | |
3388 | READ_PROTECTION_FAULT_ENABLE_DEFAULT | | |
3389 | WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT | | |
3390 | WRITE_PROTECTION_FAULT_ENABLE_DEFAULT); | |
1c49165d AD |
3391 | |
3392 | /* TC cache setup ??? */ | |
3393 | WREG32(TC_CFG_L1_LOAD_POLICY0, 0); | |
3394 | WREG32(TC_CFG_L1_LOAD_POLICY1, 0); | |
3395 | WREG32(TC_CFG_L1_STORE_POLICY, 0); | |
3396 | ||
3397 | WREG32(TC_CFG_L2_LOAD_POLICY0, 0); | |
3398 | WREG32(TC_CFG_L2_LOAD_POLICY1, 0); | |
3399 | WREG32(TC_CFG_L2_STORE_POLICY0, 0); | |
3400 | WREG32(TC_CFG_L2_STORE_POLICY1, 0); | |
3401 | WREG32(TC_CFG_L2_ATOMIC_POLICY, 0); | |
3402 | ||
3403 | WREG32(TC_CFG_L1_VOLATILE, 0); | |
3404 | WREG32(TC_CFG_L2_VOLATILE, 0); | |
3405 | ||
3406 | if (rdev->family == CHIP_KAVERI) { | |
3407 | u32 tmp = RREG32(CHUB_CONTROL); | |
3408 | tmp &= ~BYPASS_VM; | |
3409 | WREG32(CHUB_CONTROL, tmp); | |
3410 | } | |
3411 | ||
3412 | /* XXX SH_MEM regs */ | |
3413 | /* where to put LDS, scratch, GPUVM in FSA64 space */ | |
3414 | for (i = 0; i < 16; i++) { | |
b556b12e | 3415 | cik_srbm_select(rdev, 0, 0, 0, i); |
21a93e13 | 3416 | /* CP and shaders */ |
1c49165d AD |
3417 | WREG32(SH_MEM_CONFIG, 0); |
3418 | WREG32(SH_MEM_APE1_BASE, 1); | |
3419 | WREG32(SH_MEM_APE1_LIMIT, 0); | |
3420 | WREG32(SH_MEM_BASES, 0); | |
21a93e13 AD |
3421 | /* SDMA GFX */ |
3422 | WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA0_REGISTER_OFFSET, 0); | |
3423 | WREG32(SDMA0_GFX_APE1_CNTL + SDMA0_REGISTER_OFFSET, 0); | |
3424 | WREG32(SDMA0_GFX_VIRTUAL_ADDR + SDMA1_REGISTER_OFFSET, 0); | |
3425 | WREG32(SDMA0_GFX_APE1_CNTL + SDMA1_REGISTER_OFFSET, 0); | |
3426 | /* XXX SDMA RLC - todo */ | |
1c49165d | 3427 | } |
b556b12e | 3428 | cik_srbm_select(rdev, 0, 0, 0, 0); |
1c49165d AD |
3429 | |
3430 | cik_pcie_gart_tlb_flush(rdev); | |
3431 | DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", | |
3432 | (unsigned)(rdev->mc.gtt_size >> 20), | |
3433 | (unsigned long long)rdev->gart.table_addr); | |
3434 | rdev->gart.ready = true; | |
3435 | return 0; | |
3436 | } | |
3437 | ||
3438 | /** | |
3439 | * cik_pcie_gart_disable - gart disable | |
3440 | * | |
3441 | * @rdev: radeon_device pointer | |
3442 | * | |
3443 | * This disables all VM page table (CIK). | |
3444 | */ | |
3445 | static void cik_pcie_gart_disable(struct radeon_device *rdev) | |
3446 | { | |
3447 | /* Disable all tables */ | |
3448 | WREG32(VM_CONTEXT0_CNTL, 0); | |
3449 | WREG32(VM_CONTEXT1_CNTL, 0); | |
3450 | /* Setup TLB control */ | |
3451 | WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS | | |
3452 | SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); | |
3453 | /* Setup L2 cache */ | |
3454 | WREG32(VM_L2_CNTL, | |
3455 | ENABLE_L2_FRAGMENT_PROCESSING | | |
3456 | ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | | |
3457 | ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | | |
3458 | EFFECTIVE_L2_QUEUE_SIZE(7) | | |
3459 | CONTEXT1_IDENTITY_ACCESS_MODE(1)); | |
3460 | WREG32(VM_L2_CNTL2, 0); | |
3461 | WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | | |
3462 | L2_CACHE_BIGK_FRAGMENT_SIZE(6)); | |
3463 | radeon_gart_table_vram_unpin(rdev); | |
3464 | } | |
3465 | ||
3466 | /** | |
3467 | * cik_pcie_gart_fini - vm fini callback | |
3468 | * | |
3469 | * @rdev: radeon_device pointer | |
3470 | * | |
3471 | * Tears down the driver GART/VM setup (CIK). | |
3472 | */ | |
3473 | static void cik_pcie_gart_fini(struct radeon_device *rdev) | |
3474 | { | |
3475 | cik_pcie_gart_disable(rdev); | |
3476 | radeon_gart_table_vram_free(rdev); | |
3477 | radeon_gart_fini(rdev); | |
3478 | } | |
3479 | ||
3480 | /* vm parser */ | |
3481 | /** | |
3482 | * cik_ib_parse - vm ib_parse callback | |
3483 | * | |
3484 | * @rdev: radeon_device pointer | |
3485 | * @ib: indirect buffer pointer | |
3486 | * | |
3487 | * CIK uses hw IB checking so this is a nop (CIK). | |
3488 | */ | |
3489 | int cik_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib) | |
3490 | { | |
3491 | return 0; | |
3492 | } | |
3493 | ||
3494 | /* | |
3495 | * vm | |
3496 | * VMID 0 is the physical GPU addresses as used by the kernel. | |
3497 | * VMIDs 1-15 are used for userspace clients and are handled | |
3498 | * by the radeon vm/hsa code. | |
3499 | */ | |
3500 | /** | |
3501 | * cik_vm_init - cik vm init callback | |
3502 | * | |
3503 | * @rdev: radeon_device pointer | |
3504 | * | |
3505 | * Inits cik specific vm parameters (number of VMs, base of vram for | |
3506 | * VMIDs 1-15) (CIK). | |
3507 | * Returns 0 for success. | |
3508 | */ | |
3509 | int cik_vm_init(struct radeon_device *rdev) | |
3510 | { | |
3511 | /* number of VMs */ | |
3512 | rdev->vm_manager.nvm = 16; | |
3513 | /* base offset of vram pages */ | |
3514 | if (rdev->flags & RADEON_IS_IGP) { | |
3515 | u64 tmp = RREG32(MC_VM_FB_OFFSET); | |
3516 | tmp <<= 22; | |
3517 | rdev->vm_manager.vram_base_offset = tmp; | |
3518 | } else | |
3519 | rdev->vm_manager.vram_base_offset = 0; | |
3520 | ||
3521 | return 0; | |
3522 | } | |
3523 | ||
3524 | /** | |
3525 | * cik_vm_fini - cik vm fini callback | |
3526 | * | |
3527 | * @rdev: radeon_device pointer | |
3528 | * | |
3529 | * Tear down any asic specific VM setup (CIK). | |
3530 | */ | |
3531 | void cik_vm_fini(struct radeon_device *rdev) | |
3532 | { | |
3533 | } | |
3534 | ||
f96ab484 AD |
3535 | /** |
3536 | * cik_vm_flush - cik vm flush using the CP | |
3537 | * | |
3538 | * @rdev: radeon_device pointer | |
3539 | * | |
3540 | * Update the page table base and flush the VM TLB | |
3541 | * using the CP (CIK). | |
3542 | */ | |
3543 | void cik_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm) | |
3544 | { | |
3545 | struct radeon_ring *ring = &rdev->ring[ridx]; | |
3546 | ||
3547 | if (vm == NULL) | |
3548 | return; | |
3549 | ||
3550 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
3551 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
3552 | WRITE_DATA_DST_SEL(0))); | |
3553 | if (vm->id < 8) { | |
3554 | radeon_ring_write(ring, | |
3555 | (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2); | |
3556 | } else { | |
3557 | radeon_ring_write(ring, | |
3558 | (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2); | |
3559 | } | |
3560 | radeon_ring_write(ring, 0); | |
3561 | radeon_ring_write(ring, vm->pd_gpu_addr >> 12); | |
3562 | ||
3563 | /* update SH_MEM_* regs */ | |
3564 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
3565 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
3566 | WRITE_DATA_DST_SEL(0))); | |
3567 | radeon_ring_write(ring, SRBM_GFX_CNTL >> 2); | |
3568 | radeon_ring_write(ring, 0); | |
3569 | radeon_ring_write(ring, VMID(vm->id)); | |
3570 | ||
3571 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 6)); | |
3572 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
3573 | WRITE_DATA_DST_SEL(0))); | |
3574 | radeon_ring_write(ring, SH_MEM_BASES >> 2); | |
3575 | radeon_ring_write(ring, 0); | |
3576 | ||
3577 | radeon_ring_write(ring, 0); /* SH_MEM_BASES */ | |
3578 | radeon_ring_write(ring, 0); /* SH_MEM_CONFIG */ | |
3579 | radeon_ring_write(ring, 1); /* SH_MEM_APE1_BASE */ | |
3580 | radeon_ring_write(ring, 0); /* SH_MEM_APE1_LIMIT */ | |
3581 | ||
3582 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
3583 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
3584 | WRITE_DATA_DST_SEL(0))); | |
3585 | radeon_ring_write(ring, SRBM_GFX_CNTL >> 2); | |
3586 | radeon_ring_write(ring, 0); | |
3587 | radeon_ring_write(ring, VMID(0)); | |
3588 | ||
3589 | /* HDP flush */ | |
3590 | /* We should be using the WAIT_REG_MEM packet here like in | |
3591 | * cik_fence_ring_emit(), but it causes the CP to hang in this | |
3592 | * context... | |
3593 | */ | |
3594 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
3595 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
3596 | WRITE_DATA_DST_SEL(0))); | |
3597 | radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2); | |
3598 | radeon_ring_write(ring, 0); | |
3599 | radeon_ring_write(ring, 0); | |
3600 | ||
3601 | /* bits 0-15 are the VM contexts0-15 */ | |
3602 | radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); | |
3603 | radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | | |
3604 | WRITE_DATA_DST_SEL(0))); | |
3605 | radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2); | |
3606 | radeon_ring_write(ring, 0); | |
3607 | radeon_ring_write(ring, 1 << vm->id); | |
3608 | ||
3609 | /* sync PFP to ME, otherwise we might get invalid PFP reads */ | |
3610 | radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); | |
3611 | radeon_ring_write(ring, 0x0); | |
3612 | } | |
3613 | ||
d0e092d9 AD |
3614 | /** |
3615 | * cik_vm_set_page - update the page tables using sDMA | |
3616 | * | |
3617 | * @rdev: radeon_device pointer | |
3618 | * @ib: indirect buffer to fill with commands | |
3619 | * @pe: addr of the page entry | |
3620 | * @addr: dst addr to write into pe | |
3621 | * @count: number of page entries to update | |
3622 | * @incr: increase next addr by incr bytes | |
3623 | * @flags: access flags | |
3624 | * | |
3625 | * Update the page tables using CP or sDMA (CIK). | |
3626 | */ | |
3627 | void cik_vm_set_page(struct radeon_device *rdev, | |
3628 | struct radeon_ib *ib, | |
3629 | uint64_t pe, | |
3630 | uint64_t addr, unsigned count, | |
3631 | uint32_t incr, uint32_t flags) | |
3632 | { | |
3633 | uint32_t r600_flags = cayman_vm_page_flags(rdev, flags); | |
3634 | uint64_t value; | |
3635 | unsigned ndw; | |
3636 | ||
3637 | if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) { | |
3638 | /* CP */ | |
3639 | while (count) { | |
3640 | ndw = 2 + count * 2; | |
3641 | if (ndw > 0x3FFE) | |
3642 | ndw = 0x3FFE; | |
3643 | ||
3644 | ib->ptr[ib->length_dw++] = PACKET3(PACKET3_WRITE_DATA, ndw); | |
3645 | ib->ptr[ib->length_dw++] = (WRITE_DATA_ENGINE_SEL(0) | | |
3646 | WRITE_DATA_DST_SEL(1)); | |
3647 | ib->ptr[ib->length_dw++] = pe; | |
3648 | ib->ptr[ib->length_dw++] = upper_32_bits(pe); | |
3649 | for (; ndw > 2; ndw -= 2, --count, pe += 8) { | |
3650 | if (flags & RADEON_VM_PAGE_SYSTEM) { | |
3651 | value = radeon_vm_map_gart(rdev, addr); | |
3652 | value &= 0xFFFFFFFFFFFFF000ULL; | |
3653 | } else if (flags & RADEON_VM_PAGE_VALID) { | |
3654 | value = addr; | |
3655 | } else { | |
3656 | value = 0; | |
3657 | } | |
3658 | addr += incr; | |
3659 | value |= r600_flags; | |
3660 | ib->ptr[ib->length_dw++] = value; | |
3661 | ib->ptr[ib->length_dw++] = upper_32_bits(value); | |
3662 | } | |
3663 | } | |
3664 | } else { | |
3665 | /* DMA */ | |
3666 | if (flags & RADEON_VM_PAGE_SYSTEM) { | |
3667 | while (count) { | |
3668 | ndw = count * 2; | |
3669 | if (ndw > 0xFFFFE) | |
3670 | ndw = 0xFFFFE; | |
3671 | ||
3672 | /* for non-physically contiguous pages (system) */ | |
3673 | ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_WRITE, SDMA_WRITE_SUB_OPCODE_LINEAR, 0); | |
3674 | ib->ptr[ib->length_dw++] = pe; | |
3675 | ib->ptr[ib->length_dw++] = upper_32_bits(pe); | |
3676 | ib->ptr[ib->length_dw++] = ndw; | |
3677 | for (; ndw > 0; ndw -= 2, --count, pe += 8) { | |
3678 | if (flags & RADEON_VM_PAGE_SYSTEM) { | |
3679 | value = radeon_vm_map_gart(rdev, addr); | |
3680 | value &= 0xFFFFFFFFFFFFF000ULL; | |
3681 | } else if (flags & RADEON_VM_PAGE_VALID) { | |
3682 | value = addr; | |
3683 | } else { | |
3684 | value = 0; | |
3685 | } | |
3686 | addr += incr; | |
3687 | value |= r600_flags; | |
3688 | ib->ptr[ib->length_dw++] = value; | |
3689 | ib->ptr[ib->length_dw++] = upper_32_bits(value); | |
3690 | } | |
3691 | } | |
3692 | } else { | |
3693 | while (count) { | |
3694 | ndw = count; | |
3695 | if (ndw > 0x7FFFF) | |
3696 | ndw = 0x7FFFF; | |
3697 | ||
3698 | if (flags & RADEON_VM_PAGE_VALID) | |
3699 | value = addr; | |
3700 | else | |
3701 | value = 0; | |
3702 | /* for physically contiguous pages (vram) */ | |
3703 | ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_GENERATE_PTE_PDE, 0, 0); | |
3704 | ib->ptr[ib->length_dw++] = pe; /* dst addr */ | |
3705 | ib->ptr[ib->length_dw++] = upper_32_bits(pe); | |
3706 | ib->ptr[ib->length_dw++] = r600_flags; /* mask */ | |
3707 | ib->ptr[ib->length_dw++] = 0; | |
3708 | ib->ptr[ib->length_dw++] = value; /* value */ | |
3709 | ib->ptr[ib->length_dw++] = upper_32_bits(value); | |
3710 | ib->ptr[ib->length_dw++] = incr; /* increment size */ | |
3711 | ib->ptr[ib->length_dw++] = 0; | |
3712 | ib->ptr[ib->length_dw++] = ndw; /* number of entries */ | |
3713 | pe += ndw * 8; | |
3714 | addr += ndw * incr; | |
3715 | count -= ndw; | |
3716 | } | |
3717 | } | |
3718 | while (ib->length_dw & 0x7) | |
3719 | ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0); | |
3720 | } | |
3721 | } | |
3722 | ||
605de6b9 AD |
3723 | /** |
3724 | * cik_dma_vm_flush - cik vm flush using sDMA | |
3725 | * | |
3726 | * @rdev: radeon_device pointer | |
3727 | * | |
3728 | * Update the page table base and flush the VM TLB | |
3729 | * using sDMA (CIK). | |
3730 | */ | |
3731 | void cik_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm) | |
3732 | { | |
3733 | struct radeon_ring *ring = &rdev->ring[ridx]; | |
3734 | u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) | | |
3735 | SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */ | |
3736 | u32 ref_and_mask; | |
3737 | ||
3738 | if (vm == NULL) | |
3739 | return; | |
3740 | ||
3741 | if (ridx == R600_RING_TYPE_DMA_INDEX) | |
3742 | ref_and_mask = SDMA0; | |
3743 | else | |
3744 | ref_and_mask = SDMA1; | |
3745 | ||
3746 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3747 | if (vm->id < 8) { | |
3748 | radeon_ring_write(ring, (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2); | |
3749 | } else { | |
3750 | radeon_ring_write(ring, (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2); | |
3751 | } | |
3752 | radeon_ring_write(ring, vm->pd_gpu_addr >> 12); | |
3753 | ||
3754 | /* update SH_MEM_* regs */ | |
3755 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3756 | radeon_ring_write(ring, SRBM_GFX_CNTL >> 2); | |
3757 | radeon_ring_write(ring, VMID(vm->id)); | |
3758 | ||
3759 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3760 | radeon_ring_write(ring, SH_MEM_BASES >> 2); | |
3761 | radeon_ring_write(ring, 0); | |
3762 | ||
3763 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3764 | radeon_ring_write(ring, SH_MEM_CONFIG >> 2); | |
3765 | radeon_ring_write(ring, 0); | |
3766 | ||
3767 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3768 | radeon_ring_write(ring, SH_MEM_APE1_BASE >> 2); | |
3769 | radeon_ring_write(ring, 1); | |
3770 | ||
3771 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3772 | radeon_ring_write(ring, SH_MEM_APE1_LIMIT >> 2); | |
3773 | radeon_ring_write(ring, 0); | |
3774 | ||
3775 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3776 | radeon_ring_write(ring, SRBM_GFX_CNTL >> 2); | |
3777 | radeon_ring_write(ring, VMID(0)); | |
3778 | ||
3779 | /* flush HDP */ | |
3780 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, extra_bits)); | |
3781 | radeon_ring_write(ring, GPU_HDP_FLUSH_DONE); | |
3782 | radeon_ring_write(ring, GPU_HDP_FLUSH_REQ); | |
3783 | radeon_ring_write(ring, ref_and_mask); /* REFERENCE */ | |
3784 | radeon_ring_write(ring, ref_and_mask); /* MASK */ | |
3785 | radeon_ring_write(ring, (4 << 16) | 10); /* RETRY_COUNT, POLL_INTERVAL */ | |
3786 | ||
3787 | /* flush TLB */ | |
3788 | radeon_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); | |
3789 | radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2); | |
3790 | radeon_ring_write(ring, 1 << vm->id); | |
3791 | } | |
3792 | ||
f6796cae AD |
3793 | /* |
3794 | * RLC | |
3795 | * The RLC is a multi-purpose microengine that handles a | |
3796 | * variety of functions, the most important of which is | |
3797 | * the interrupt controller. | |
3798 | */ | |
3799 | /** | |
3800 | * cik_rlc_stop - stop the RLC ME | |
3801 | * | |
3802 | * @rdev: radeon_device pointer | |
3803 | * | |
3804 | * Halt the RLC ME (MicroEngine) (CIK). | |
3805 | */ | |
3806 | static void cik_rlc_stop(struct radeon_device *rdev) | |
3807 | { | |
3808 | int i, j, k; | |
3809 | u32 mask, tmp; | |
3810 | ||
3811 | tmp = RREG32(CP_INT_CNTL_RING0); | |
3812 | tmp &= ~(CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); | |
3813 | WREG32(CP_INT_CNTL_RING0, tmp); | |
3814 | ||
3815 | RREG32(CB_CGTT_SCLK_CTRL); | |
3816 | RREG32(CB_CGTT_SCLK_CTRL); | |
3817 | RREG32(CB_CGTT_SCLK_CTRL); | |
3818 | RREG32(CB_CGTT_SCLK_CTRL); | |
3819 | ||
3820 | tmp = RREG32(RLC_CGCG_CGLS_CTRL) & 0xfffffffc; | |
3821 | WREG32(RLC_CGCG_CGLS_CTRL, tmp); | |
3822 | ||
3823 | WREG32(RLC_CNTL, 0); | |
3824 | ||
3825 | for (i = 0; i < rdev->config.cik.max_shader_engines; i++) { | |
3826 | for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) { | |
3827 | cik_select_se_sh(rdev, i, j); | |
3828 | for (k = 0; k < rdev->usec_timeout; k++) { | |
3829 | if (RREG32(RLC_SERDES_CU_MASTER_BUSY) == 0) | |
3830 | break; | |
3831 | udelay(1); | |
3832 | } | |
3833 | } | |
3834 | } | |
3835 | cik_select_se_sh(rdev, 0xffffffff, 0xffffffff); | |
3836 | ||
3837 | mask = SE_MASTER_BUSY_MASK | GC_MASTER_BUSY | TC0_MASTER_BUSY | TC1_MASTER_BUSY; | |
3838 | for (k = 0; k < rdev->usec_timeout; k++) { | |
3839 | if ((RREG32(RLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0) | |
3840 | break; | |
3841 | udelay(1); | |
3842 | } | |
3843 | } | |
3844 | ||
3845 | /** | |
3846 | * cik_rlc_start - start the RLC ME | |
3847 | * | |
3848 | * @rdev: radeon_device pointer | |
3849 | * | |
3850 | * Unhalt the RLC ME (MicroEngine) (CIK). | |
3851 | */ | |
3852 | static void cik_rlc_start(struct radeon_device *rdev) | |
3853 | { | |
3854 | u32 tmp; | |
3855 | ||
3856 | WREG32(RLC_CNTL, RLC_ENABLE); | |
3857 | ||
3858 | tmp = RREG32(CP_INT_CNTL_RING0); | |
3859 | tmp |= (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); | |
3860 | WREG32(CP_INT_CNTL_RING0, tmp); | |
3861 | ||
3862 | udelay(50); | |
3863 | } | |
3864 | ||
3865 | /** | |
3866 | * cik_rlc_resume - setup the RLC hw | |
3867 | * | |
3868 | * @rdev: radeon_device pointer | |
3869 | * | |
3870 | * Initialize the RLC registers, load the ucode, | |
3871 | * and start the RLC (CIK). | |
3872 | * Returns 0 for success, -EINVAL if the ucode is not available. | |
3873 | */ | |
3874 | static int cik_rlc_resume(struct radeon_device *rdev) | |
3875 | { | |
3876 | u32 i, size; | |
3877 | u32 clear_state_info[3]; | |
3878 | const __be32 *fw_data; | |
3879 | ||
3880 | if (!rdev->rlc_fw) | |
3881 | return -EINVAL; | |
3882 | ||
3883 | switch (rdev->family) { | |
3884 | case CHIP_BONAIRE: | |
3885 | default: | |
3886 | size = BONAIRE_RLC_UCODE_SIZE; | |
3887 | break; | |
3888 | case CHIP_KAVERI: | |
3889 | size = KV_RLC_UCODE_SIZE; | |
3890 | break; | |
3891 | case CHIP_KABINI: | |
3892 | size = KB_RLC_UCODE_SIZE; | |
3893 | break; | |
3894 | } | |
3895 | ||
3896 | cik_rlc_stop(rdev); | |
3897 | ||
3898 | WREG32(GRBM_SOFT_RESET, SOFT_RESET_RLC); | |
3899 | RREG32(GRBM_SOFT_RESET); | |
3900 | udelay(50); | |
3901 | WREG32(GRBM_SOFT_RESET, 0); | |
3902 | RREG32(GRBM_SOFT_RESET); | |
3903 | udelay(50); | |
3904 | ||
3905 | WREG32(RLC_LB_CNTR_INIT, 0); | |
3906 | WREG32(RLC_LB_CNTR_MAX, 0x00008000); | |
3907 | ||
3908 | cik_select_se_sh(rdev, 0xffffffff, 0xffffffff); | |
3909 | WREG32(RLC_LB_INIT_CU_MASK, 0xffffffff); | |
3910 | WREG32(RLC_LB_PARAMS, 0x00600408); | |
3911 | WREG32(RLC_LB_CNTL, 0x80000004); | |
3912 | ||
3913 | WREG32(RLC_MC_CNTL, 0); | |
3914 | WREG32(RLC_UCODE_CNTL, 0); | |
3915 | ||
3916 | fw_data = (const __be32 *)rdev->rlc_fw->data; | |
3917 | WREG32(RLC_GPM_UCODE_ADDR, 0); | |
3918 | for (i = 0; i < size; i++) | |
3919 | WREG32(RLC_GPM_UCODE_DATA, be32_to_cpup(fw_data++)); | |
3920 | WREG32(RLC_GPM_UCODE_ADDR, 0); | |
3921 | ||
3922 | /* XXX */ | |
3923 | clear_state_info[0] = 0;//upper_32_bits(rdev->rlc.save_restore_gpu_addr); | |
3924 | clear_state_info[1] = 0;//rdev->rlc.save_restore_gpu_addr; | |
3925 | clear_state_info[2] = 0;//cik_default_size; | |
3926 | WREG32(RLC_GPM_SCRATCH_ADDR, 0x3d); | |
3927 | for (i = 0; i < 3; i++) | |
3928 | WREG32(RLC_GPM_SCRATCH_DATA, clear_state_info[i]); | |
3929 | WREG32(RLC_DRIVER_DMA_STATUS, 0); | |
3930 | ||
3931 | cik_rlc_start(rdev); | |
3932 | ||
3933 | return 0; | |
3934 | } | |
a59781bb AD |
3935 | |
3936 | /* | |
3937 | * Interrupts | |
3938 | * Starting with r6xx, interrupts are handled via a ring buffer. | |
3939 | * Ring buffers are areas of GPU accessible memory that the GPU | |
3940 | * writes interrupt vectors into and the host reads vectors out of. | |
3941 | * There is a rptr (read pointer) that determines where the | |
3942 | * host is currently reading, and a wptr (write pointer) | |
3943 | * which determines where the GPU has written. When the | |
3944 | * pointers are equal, the ring is idle. When the GPU | |
3945 | * writes vectors to the ring buffer, it increments the | |
3946 | * wptr. When there is an interrupt, the host then starts | |
3947 | * fetching commands and processing them until the pointers are | |
3948 | * equal again at which point it updates the rptr. | |
3949 | */ | |
3950 | ||
3951 | /** | |
3952 | * cik_enable_interrupts - Enable the interrupt ring buffer | |
3953 | * | |
3954 | * @rdev: radeon_device pointer | |
3955 | * | |
3956 | * Enable the interrupt ring buffer (CIK). | |
3957 | */ | |
3958 | static void cik_enable_interrupts(struct radeon_device *rdev) | |
3959 | { | |
3960 | u32 ih_cntl = RREG32(IH_CNTL); | |
3961 | u32 ih_rb_cntl = RREG32(IH_RB_CNTL); | |
3962 | ||
3963 | ih_cntl |= ENABLE_INTR; | |
3964 | ih_rb_cntl |= IH_RB_ENABLE; | |
3965 | WREG32(IH_CNTL, ih_cntl); | |
3966 | WREG32(IH_RB_CNTL, ih_rb_cntl); | |
3967 | rdev->ih.enabled = true; | |
3968 | } | |
3969 | ||
3970 | /** | |
3971 | * cik_disable_interrupts - Disable the interrupt ring buffer | |
3972 | * | |
3973 | * @rdev: radeon_device pointer | |
3974 | * | |
3975 | * Disable the interrupt ring buffer (CIK). | |
3976 | */ | |
3977 | static void cik_disable_interrupts(struct radeon_device *rdev) | |
3978 | { | |
3979 | u32 ih_rb_cntl = RREG32(IH_RB_CNTL); | |
3980 | u32 ih_cntl = RREG32(IH_CNTL); | |
3981 | ||
3982 | ih_rb_cntl &= ~IH_RB_ENABLE; | |
3983 | ih_cntl &= ~ENABLE_INTR; | |
3984 | WREG32(IH_RB_CNTL, ih_rb_cntl); | |
3985 | WREG32(IH_CNTL, ih_cntl); | |
3986 | /* set rptr, wptr to 0 */ | |
3987 | WREG32(IH_RB_RPTR, 0); | |
3988 | WREG32(IH_RB_WPTR, 0); | |
3989 | rdev->ih.enabled = false; | |
3990 | rdev->ih.rptr = 0; | |
3991 | } | |
3992 | ||
3993 | /** | |
3994 | * cik_disable_interrupt_state - Disable all interrupt sources | |
3995 | * | |
3996 | * @rdev: radeon_device pointer | |
3997 | * | |
3998 | * Clear all interrupt enable bits used by the driver (CIK). | |
3999 | */ | |
4000 | static void cik_disable_interrupt_state(struct radeon_device *rdev) | |
4001 | { | |
4002 | u32 tmp; | |
4003 | ||
4004 | /* gfx ring */ | |
4005 | WREG32(CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); | |
21a93e13 AD |
4006 | /* sdma */ |
4007 | tmp = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE; | |
4008 | WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, tmp); | |
4009 | tmp = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE; | |
4010 | WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, tmp); | |
a59781bb AD |
4011 | /* compute queues */ |
4012 | WREG32(CP_ME1_PIPE0_INT_CNTL, 0); | |
4013 | WREG32(CP_ME1_PIPE1_INT_CNTL, 0); | |
4014 | WREG32(CP_ME1_PIPE2_INT_CNTL, 0); | |
4015 | WREG32(CP_ME1_PIPE3_INT_CNTL, 0); | |
4016 | WREG32(CP_ME2_PIPE0_INT_CNTL, 0); | |
4017 | WREG32(CP_ME2_PIPE1_INT_CNTL, 0); | |
4018 | WREG32(CP_ME2_PIPE2_INT_CNTL, 0); | |
4019 | WREG32(CP_ME2_PIPE3_INT_CNTL, 0); | |
4020 | /* grbm */ | |
4021 | WREG32(GRBM_INT_CNTL, 0); | |
4022 | /* vline/vblank, etc. */ | |
4023 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0); | |
4024 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0); | |
4025 | if (rdev->num_crtc >= 4) { | |
4026 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0); | |
4027 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0); | |
4028 | } | |
4029 | if (rdev->num_crtc >= 6) { | |
4030 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0); | |
4031 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0); | |
4032 | } | |
4033 | ||
4034 | /* dac hotplug */ | |
4035 | WREG32(DAC_AUTODETECT_INT_CONTROL, 0); | |
4036 | ||
4037 | /* digital hotplug */ | |
4038 | tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY; | |
4039 | WREG32(DC_HPD1_INT_CONTROL, tmp); | |
4040 | tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY; | |
4041 | WREG32(DC_HPD2_INT_CONTROL, tmp); | |
4042 | tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY; | |
4043 | WREG32(DC_HPD3_INT_CONTROL, tmp); | |
4044 | tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY; | |
4045 | WREG32(DC_HPD4_INT_CONTROL, tmp); | |
4046 | tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY; | |
4047 | WREG32(DC_HPD5_INT_CONTROL, tmp); | |
4048 | tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY; | |
4049 | WREG32(DC_HPD6_INT_CONTROL, tmp); | |
4050 | ||
4051 | } | |
4052 | ||
4053 | /** | |
4054 | * cik_irq_init - init and enable the interrupt ring | |
4055 | * | |
4056 | * @rdev: radeon_device pointer | |
4057 | * | |
4058 | * Allocate a ring buffer for the interrupt controller, | |
4059 | * enable the RLC, disable interrupts, enable the IH | |
4060 | * ring buffer and enable it (CIK). | |
4061 | * Called at device load and reume. | |
4062 | * Returns 0 for success, errors for failure. | |
4063 | */ | |
4064 | static int cik_irq_init(struct radeon_device *rdev) | |
4065 | { | |
4066 | int ret = 0; | |
4067 | int rb_bufsz; | |
4068 | u32 interrupt_cntl, ih_cntl, ih_rb_cntl; | |
4069 | ||
4070 | /* allocate ring */ | |
4071 | ret = r600_ih_ring_alloc(rdev); | |
4072 | if (ret) | |
4073 | return ret; | |
4074 | ||
4075 | /* disable irqs */ | |
4076 | cik_disable_interrupts(rdev); | |
4077 | ||
4078 | /* init rlc */ | |
4079 | ret = cik_rlc_resume(rdev); | |
4080 | if (ret) { | |
4081 | r600_ih_ring_fini(rdev); | |
4082 | return ret; | |
4083 | } | |
4084 | ||
4085 | /* setup interrupt control */ | |
4086 | /* XXX this should actually be a bus address, not an MC address. same on older asics */ | |
4087 | WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8); | |
4088 | interrupt_cntl = RREG32(INTERRUPT_CNTL); | |
4089 | /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi | |
4090 | * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN | |
4091 | */ | |
4092 | interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE; | |
4093 | /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */ | |
4094 | interrupt_cntl &= ~IH_REQ_NONSNOOP_EN; | |
4095 | WREG32(INTERRUPT_CNTL, interrupt_cntl); | |
4096 | ||
4097 | WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8); | |
4098 | rb_bufsz = drm_order(rdev->ih.ring_size / 4); | |
4099 | ||
4100 | ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE | | |
4101 | IH_WPTR_OVERFLOW_CLEAR | | |
4102 | (rb_bufsz << 1)); | |
4103 | ||
4104 | if (rdev->wb.enabled) | |
4105 | ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE; | |
4106 | ||
4107 | /* set the writeback address whether it's enabled or not */ | |
4108 | WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC); | |
4109 | WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF); | |
4110 | ||
4111 | WREG32(IH_RB_CNTL, ih_rb_cntl); | |
4112 | ||
4113 | /* set rptr, wptr to 0 */ | |
4114 | WREG32(IH_RB_RPTR, 0); | |
4115 | WREG32(IH_RB_WPTR, 0); | |
4116 | ||
4117 | /* Default settings for IH_CNTL (disabled at first) */ | |
4118 | ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0); | |
4119 | /* RPTR_REARM only works if msi's are enabled */ | |
4120 | if (rdev->msi_enabled) | |
4121 | ih_cntl |= RPTR_REARM; | |
4122 | WREG32(IH_CNTL, ih_cntl); | |
4123 | ||
4124 | /* force the active interrupt state to all disabled */ | |
4125 | cik_disable_interrupt_state(rdev); | |
4126 | ||
4127 | pci_set_master(rdev->pdev); | |
4128 | ||
4129 | /* enable irqs */ | |
4130 | cik_enable_interrupts(rdev); | |
4131 | ||
4132 | return ret; | |
4133 | } | |
4134 | ||
4135 | /** | |
4136 | * cik_irq_set - enable/disable interrupt sources | |
4137 | * | |
4138 | * @rdev: radeon_device pointer | |
4139 | * | |
4140 | * Enable interrupt sources on the GPU (vblanks, hpd, | |
4141 | * etc.) (CIK). | |
4142 | * Returns 0 for success, errors for failure. | |
4143 | */ | |
4144 | int cik_irq_set(struct radeon_device *rdev) | |
4145 | { | |
4146 | u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE | | |
4147 | PRIV_INSTR_INT_ENABLE | PRIV_REG_INT_ENABLE; | |
4148 | u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0; | |
4149 | u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6; | |
4150 | u32 grbm_int_cntl = 0; | |
21a93e13 | 4151 | u32 dma_cntl, dma_cntl1; |
a59781bb AD |
4152 | |
4153 | if (!rdev->irq.installed) { | |
4154 | WARN(1, "Can't enable IRQ/MSI because no handler is installed\n"); | |
4155 | return -EINVAL; | |
4156 | } | |
4157 | /* don't enable anything if the ih is disabled */ | |
4158 | if (!rdev->ih.enabled) { | |
4159 | cik_disable_interrupts(rdev); | |
4160 | /* force the active interrupt state to all disabled */ | |
4161 | cik_disable_interrupt_state(rdev); | |
4162 | return 0; | |
4163 | } | |
4164 | ||
4165 | hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN; | |
4166 | hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN; | |
4167 | hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN; | |
4168 | hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN; | |
4169 | hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN; | |
4170 | hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN; | |
4171 | ||
21a93e13 AD |
4172 | dma_cntl = RREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET) & ~TRAP_ENABLE; |
4173 | dma_cntl1 = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE; | |
4174 | ||
a59781bb AD |
4175 | /* enable CP interrupts on all rings */ |
4176 | if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) { | |
4177 | DRM_DEBUG("cik_irq_set: sw int gfx\n"); | |
4178 | cp_int_cntl |= TIME_STAMP_INT_ENABLE; | |
4179 | } | |
4180 | /* TODO: compute queues! */ | |
4181 | /* CP_ME[1-2]_PIPE[0-3]_INT_CNTL */ | |
4182 | ||
21a93e13 AD |
4183 | if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) { |
4184 | DRM_DEBUG("cik_irq_set: sw int dma\n"); | |
4185 | dma_cntl |= TRAP_ENABLE; | |
4186 | } | |
4187 | ||
4188 | if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) { | |
4189 | DRM_DEBUG("cik_irq_set: sw int dma1\n"); | |
4190 | dma_cntl1 |= TRAP_ENABLE; | |
4191 | } | |
4192 | ||
a59781bb AD |
4193 | if (rdev->irq.crtc_vblank_int[0] || |
4194 | atomic_read(&rdev->irq.pflip[0])) { | |
4195 | DRM_DEBUG("cik_irq_set: vblank 0\n"); | |
4196 | crtc1 |= VBLANK_INTERRUPT_MASK; | |
4197 | } | |
4198 | if (rdev->irq.crtc_vblank_int[1] || | |
4199 | atomic_read(&rdev->irq.pflip[1])) { | |
4200 | DRM_DEBUG("cik_irq_set: vblank 1\n"); | |
4201 | crtc2 |= VBLANK_INTERRUPT_MASK; | |
4202 | } | |
4203 | if (rdev->irq.crtc_vblank_int[2] || | |
4204 | atomic_read(&rdev->irq.pflip[2])) { | |
4205 | DRM_DEBUG("cik_irq_set: vblank 2\n"); | |
4206 | crtc3 |= VBLANK_INTERRUPT_MASK; | |
4207 | } | |
4208 | if (rdev->irq.crtc_vblank_int[3] || | |
4209 | atomic_read(&rdev->irq.pflip[3])) { | |
4210 | DRM_DEBUG("cik_irq_set: vblank 3\n"); | |
4211 | crtc4 |= VBLANK_INTERRUPT_MASK; | |
4212 | } | |
4213 | if (rdev->irq.crtc_vblank_int[4] || | |
4214 | atomic_read(&rdev->irq.pflip[4])) { | |
4215 | DRM_DEBUG("cik_irq_set: vblank 4\n"); | |
4216 | crtc5 |= VBLANK_INTERRUPT_MASK; | |
4217 | } | |
4218 | if (rdev->irq.crtc_vblank_int[5] || | |
4219 | atomic_read(&rdev->irq.pflip[5])) { | |
4220 | DRM_DEBUG("cik_irq_set: vblank 5\n"); | |
4221 | crtc6 |= VBLANK_INTERRUPT_MASK; | |
4222 | } | |
4223 | if (rdev->irq.hpd[0]) { | |
4224 | DRM_DEBUG("cik_irq_set: hpd 1\n"); | |
4225 | hpd1 |= DC_HPDx_INT_EN; | |
4226 | } | |
4227 | if (rdev->irq.hpd[1]) { | |
4228 | DRM_DEBUG("cik_irq_set: hpd 2\n"); | |
4229 | hpd2 |= DC_HPDx_INT_EN; | |
4230 | } | |
4231 | if (rdev->irq.hpd[2]) { | |
4232 | DRM_DEBUG("cik_irq_set: hpd 3\n"); | |
4233 | hpd3 |= DC_HPDx_INT_EN; | |
4234 | } | |
4235 | if (rdev->irq.hpd[3]) { | |
4236 | DRM_DEBUG("cik_irq_set: hpd 4\n"); | |
4237 | hpd4 |= DC_HPDx_INT_EN; | |
4238 | } | |
4239 | if (rdev->irq.hpd[4]) { | |
4240 | DRM_DEBUG("cik_irq_set: hpd 5\n"); | |
4241 | hpd5 |= DC_HPDx_INT_EN; | |
4242 | } | |
4243 | if (rdev->irq.hpd[5]) { | |
4244 | DRM_DEBUG("cik_irq_set: hpd 6\n"); | |
4245 | hpd6 |= DC_HPDx_INT_EN; | |
4246 | } | |
4247 | ||
4248 | WREG32(CP_INT_CNTL_RING0, cp_int_cntl); | |
4249 | ||
21a93e13 AD |
4250 | WREG32(SDMA0_CNTL + SDMA0_REGISTER_OFFSET, dma_cntl); |
4251 | WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, dma_cntl1); | |
4252 | ||
a59781bb AD |
4253 | WREG32(GRBM_INT_CNTL, grbm_int_cntl); |
4254 | ||
4255 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1); | |
4256 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2); | |
4257 | if (rdev->num_crtc >= 4) { | |
4258 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3); | |
4259 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4); | |
4260 | } | |
4261 | if (rdev->num_crtc >= 6) { | |
4262 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5); | |
4263 | WREG32(LB_INTERRUPT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6); | |
4264 | } | |
4265 | ||
4266 | WREG32(DC_HPD1_INT_CONTROL, hpd1); | |
4267 | WREG32(DC_HPD2_INT_CONTROL, hpd2); | |
4268 | WREG32(DC_HPD3_INT_CONTROL, hpd3); | |
4269 | WREG32(DC_HPD4_INT_CONTROL, hpd4); | |
4270 | WREG32(DC_HPD5_INT_CONTROL, hpd5); | |
4271 | WREG32(DC_HPD6_INT_CONTROL, hpd6); | |
4272 | ||
4273 | return 0; | |
4274 | } | |
4275 | ||
4276 | /** | |
4277 | * cik_irq_ack - ack interrupt sources | |
4278 | * | |
4279 | * @rdev: radeon_device pointer | |
4280 | * | |
4281 | * Ack interrupt sources on the GPU (vblanks, hpd, | |
4282 | * etc.) (CIK). Certain interrupts sources are sw | |
4283 | * generated and do not require an explicit ack. | |
4284 | */ | |
4285 | static inline void cik_irq_ack(struct radeon_device *rdev) | |
4286 | { | |
4287 | u32 tmp; | |
4288 | ||
4289 | rdev->irq.stat_regs.cik.disp_int = RREG32(DISP_INTERRUPT_STATUS); | |
4290 | rdev->irq.stat_regs.cik.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE); | |
4291 | rdev->irq.stat_regs.cik.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2); | |
4292 | rdev->irq.stat_regs.cik.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3); | |
4293 | rdev->irq.stat_regs.cik.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4); | |
4294 | rdev->irq.stat_regs.cik.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5); | |
4295 | rdev->irq.stat_regs.cik.disp_int_cont6 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE6); | |
4296 | ||
4297 | if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT) | |
4298 | WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK); | |
4299 | if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT) | |
4300 | WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK); | |
4301 | if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT) | |
4302 | WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK); | |
4303 | if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT) | |
4304 | WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK); | |
4305 | ||
4306 | if (rdev->num_crtc >= 4) { | |
4307 | if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) | |
4308 | WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK); | |
4309 | if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) | |
4310 | WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK); | |
4311 | if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) | |
4312 | WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK); | |
4313 | if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) | |
4314 | WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK); | |
4315 | } | |
4316 | ||
4317 | if (rdev->num_crtc >= 6) { | |
4318 | if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) | |
4319 | WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK); | |
4320 | if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) | |
4321 | WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK); | |
4322 | if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) | |
4323 | WREG32(LB_VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK); | |
4324 | if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) | |
4325 | WREG32(LB_VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK); | |
4326 | } | |
4327 | ||
4328 | if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) { | |
4329 | tmp = RREG32(DC_HPD1_INT_CONTROL); | |
4330 | tmp |= DC_HPDx_INT_ACK; | |
4331 | WREG32(DC_HPD1_INT_CONTROL, tmp); | |
4332 | } | |
4333 | if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) { | |
4334 | tmp = RREG32(DC_HPD2_INT_CONTROL); | |
4335 | tmp |= DC_HPDx_INT_ACK; | |
4336 | WREG32(DC_HPD2_INT_CONTROL, tmp); | |
4337 | } | |
4338 | if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) { | |
4339 | tmp = RREG32(DC_HPD3_INT_CONTROL); | |
4340 | tmp |= DC_HPDx_INT_ACK; | |
4341 | WREG32(DC_HPD3_INT_CONTROL, tmp); | |
4342 | } | |
4343 | if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) { | |
4344 | tmp = RREG32(DC_HPD4_INT_CONTROL); | |
4345 | tmp |= DC_HPDx_INT_ACK; | |
4346 | WREG32(DC_HPD4_INT_CONTROL, tmp); | |
4347 | } | |
4348 | if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) { | |
4349 | tmp = RREG32(DC_HPD5_INT_CONTROL); | |
4350 | tmp |= DC_HPDx_INT_ACK; | |
4351 | WREG32(DC_HPD5_INT_CONTROL, tmp); | |
4352 | } | |
4353 | if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) { | |
4354 | tmp = RREG32(DC_HPD5_INT_CONTROL); | |
4355 | tmp |= DC_HPDx_INT_ACK; | |
4356 | WREG32(DC_HPD6_INT_CONTROL, tmp); | |
4357 | } | |
4358 | } | |
4359 | ||
4360 | /** | |
4361 | * cik_irq_disable - disable interrupts | |
4362 | * | |
4363 | * @rdev: radeon_device pointer | |
4364 | * | |
4365 | * Disable interrupts on the hw (CIK). | |
4366 | */ | |
4367 | static void cik_irq_disable(struct radeon_device *rdev) | |
4368 | { | |
4369 | cik_disable_interrupts(rdev); | |
4370 | /* Wait and acknowledge irq */ | |
4371 | mdelay(1); | |
4372 | cik_irq_ack(rdev); | |
4373 | cik_disable_interrupt_state(rdev); | |
4374 | } | |
4375 | ||
4376 | /** | |
4377 | * cik_irq_disable - disable interrupts for suspend | |
4378 | * | |
4379 | * @rdev: radeon_device pointer | |
4380 | * | |
4381 | * Disable interrupts and stop the RLC (CIK). | |
4382 | * Used for suspend. | |
4383 | */ | |
4384 | static void cik_irq_suspend(struct radeon_device *rdev) | |
4385 | { | |
4386 | cik_irq_disable(rdev); | |
4387 | cik_rlc_stop(rdev); | |
4388 | } | |
4389 | ||
4390 | /** | |
4391 | * cik_irq_fini - tear down interrupt support | |
4392 | * | |
4393 | * @rdev: radeon_device pointer | |
4394 | * | |
4395 | * Disable interrupts on the hw and free the IH ring | |
4396 | * buffer (CIK). | |
4397 | * Used for driver unload. | |
4398 | */ | |
4399 | static void cik_irq_fini(struct radeon_device *rdev) | |
4400 | { | |
4401 | cik_irq_suspend(rdev); | |
4402 | r600_ih_ring_fini(rdev); | |
4403 | } | |
4404 | ||
4405 | /** | |
4406 | * cik_get_ih_wptr - get the IH ring buffer wptr | |
4407 | * | |
4408 | * @rdev: radeon_device pointer | |
4409 | * | |
4410 | * Get the IH ring buffer wptr from either the register | |
4411 | * or the writeback memory buffer (CIK). Also check for | |
4412 | * ring buffer overflow and deal with it. | |
4413 | * Used by cik_irq_process(). | |
4414 | * Returns the value of the wptr. | |
4415 | */ | |
4416 | static inline u32 cik_get_ih_wptr(struct radeon_device *rdev) | |
4417 | { | |
4418 | u32 wptr, tmp; | |
4419 | ||
4420 | if (rdev->wb.enabled) | |
4421 | wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]); | |
4422 | else | |
4423 | wptr = RREG32(IH_RB_WPTR); | |
4424 | ||
4425 | if (wptr & RB_OVERFLOW) { | |
4426 | /* When a ring buffer overflow happen start parsing interrupt | |
4427 | * from the last not overwritten vector (wptr + 16). Hopefully | |
4428 | * this should allow us to catchup. | |
4429 | */ | |
4430 | dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n", | |
4431 | wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask); | |
4432 | rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask; | |
4433 | tmp = RREG32(IH_RB_CNTL); | |
4434 | tmp |= IH_WPTR_OVERFLOW_CLEAR; | |
4435 | WREG32(IH_RB_CNTL, tmp); | |
4436 | } | |
4437 | return (wptr & rdev->ih.ptr_mask); | |
4438 | } | |
4439 | ||
4440 | /* CIK IV Ring | |
4441 | * Each IV ring entry is 128 bits: | |
4442 | * [7:0] - interrupt source id | |
4443 | * [31:8] - reserved | |
4444 | * [59:32] - interrupt source data | |
4445 | * [63:60] - reserved | |
21a93e13 AD |
4446 | * [71:64] - RINGID |
4447 | * CP: | |
4448 | * ME_ID [1:0], PIPE_ID[1:0], QUEUE_ID[2:0] | |
a59781bb AD |
4449 | * QUEUE_ID - for compute, which of the 8 queues owned by the dispatcher |
4450 | * - for gfx, hw shader state (0=PS...5=LS, 6=CS) | |
4451 | * ME_ID - 0 = gfx, 1 = first 4 CS pipes, 2 = second 4 CS pipes | |
4452 | * PIPE_ID - ME0 0=3D | |
4453 | * - ME1&2 compute dispatcher (4 pipes each) | |
21a93e13 AD |
4454 | * SDMA: |
4455 | * INSTANCE_ID [1:0], QUEUE_ID[1:0] | |
4456 | * INSTANCE_ID - 0 = sdma0, 1 = sdma1 | |
4457 | * QUEUE_ID - 0 = gfx, 1 = rlc0, 2 = rlc1 | |
a59781bb AD |
4458 | * [79:72] - VMID |
4459 | * [95:80] - PASID | |
4460 | * [127:96] - reserved | |
4461 | */ | |
4462 | /** | |
4463 | * cik_irq_process - interrupt handler | |
4464 | * | |
4465 | * @rdev: radeon_device pointer | |
4466 | * | |
4467 | * Interrupt hander (CIK). Walk the IH ring, | |
4468 | * ack interrupts and schedule work to handle | |
4469 | * interrupt events. | |
4470 | * Returns irq process return code. | |
4471 | */ | |
4472 | int cik_irq_process(struct radeon_device *rdev) | |
4473 | { | |
4474 | u32 wptr; | |
4475 | u32 rptr; | |
4476 | u32 src_id, src_data, ring_id; | |
4477 | u8 me_id, pipe_id, queue_id; | |
4478 | u32 ring_index; | |
4479 | bool queue_hotplug = false; | |
4480 | bool queue_reset = false; | |
4481 | ||
4482 | if (!rdev->ih.enabled || rdev->shutdown) | |
4483 | return IRQ_NONE; | |
4484 | ||
4485 | wptr = cik_get_ih_wptr(rdev); | |
4486 | ||
4487 | restart_ih: | |
4488 | /* is somebody else already processing irqs? */ | |
4489 | if (atomic_xchg(&rdev->ih.lock, 1)) | |
4490 | return IRQ_NONE; | |
4491 | ||
4492 | rptr = rdev->ih.rptr; | |
4493 | DRM_DEBUG("cik_irq_process start: rptr %d, wptr %d\n", rptr, wptr); | |
4494 | ||
4495 | /* Order reading of wptr vs. reading of IH ring data */ | |
4496 | rmb(); | |
4497 | ||
4498 | /* display interrupts */ | |
4499 | cik_irq_ack(rdev); | |
4500 | ||
4501 | while (rptr != wptr) { | |
4502 | /* wptr/rptr are in bytes! */ | |
4503 | ring_index = rptr / 4; | |
4504 | src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff; | |
4505 | src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff; | |
4506 | ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff; | |
a59781bb AD |
4507 | |
4508 | switch (src_id) { | |
4509 | case 1: /* D1 vblank/vline */ | |
4510 | switch (src_data) { | |
4511 | case 0: /* D1 vblank */ | |
4512 | if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VBLANK_INTERRUPT) { | |
4513 | if (rdev->irq.crtc_vblank_int[0]) { | |
4514 | drm_handle_vblank(rdev->ddev, 0); | |
4515 | rdev->pm.vblank_sync = true; | |
4516 | wake_up(&rdev->irq.vblank_queue); | |
4517 | } | |
4518 | if (atomic_read(&rdev->irq.pflip[0])) | |
4519 | radeon_crtc_handle_flip(rdev, 0); | |
4520 | rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VBLANK_INTERRUPT; | |
4521 | DRM_DEBUG("IH: D1 vblank\n"); | |
4522 | } | |
4523 | break; | |
4524 | case 1: /* D1 vline */ | |
4525 | if (rdev->irq.stat_regs.cik.disp_int & LB_D1_VLINE_INTERRUPT) { | |
4526 | rdev->irq.stat_regs.cik.disp_int &= ~LB_D1_VLINE_INTERRUPT; | |
4527 | DRM_DEBUG("IH: D1 vline\n"); | |
4528 | } | |
4529 | break; | |
4530 | default: | |
4531 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4532 | break; | |
4533 | } | |
4534 | break; | |
4535 | case 2: /* D2 vblank/vline */ | |
4536 | switch (src_data) { | |
4537 | case 0: /* D2 vblank */ | |
4538 | if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VBLANK_INTERRUPT) { | |
4539 | if (rdev->irq.crtc_vblank_int[1]) { | |
4540 | drm_handle_vblank(rdev->ddev, 1); | |
4541 | rdev->pm.vblank_sync = true; | |
4542 | wake_up(&rdev->irq.vblank_queue); | |
4543 | } | |
4544 | if (atomic_read(&rdev->irq.pflip[1])) | |
4545 | radeon_crtc_handle_flip(rdev, 1); | |
4546 | rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT; | |
4547 | DRM_DEBUG("IH: D2 vblank\n"); | |
4548 | } | |
4549 | break; | |
4550 | case 1: /* D2 vline */ | |
4551 | if (rdev->irq.stat_regs.cik.disp_int_cont & LB_D2_VLINE_INTERRUPT) { | |
4552 | rdev->irq.stat_regs.cik.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT; | |
4553 | DRM_DEBUG("IH: D2 vline\n"); | |
4554 | } | |
4555 | break; | |
4556 | default: | |
4557 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4558 | break; | |
4559 | } | |
4560 | break; | |
4561 | case 3: /* D3 vblank/vline */ | |
4562 | switch (src_data) { | |
4563 | case 0: /* D3 vblank */ | |
4564 | if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) { | |
4565 | if (rdev->irq.crtc_vblank_int[2]) { | |
4566 | drm_handle_vblank(rdev->ddev, 2); | |
4567 | rdev->pm.vblank_sync = true; | |
4568 | wake_up(&rdev->irq.vblank_queue); | |
4569 | } | |
4570 | if (atomic_read(&rdev->irq.pflip[2])) | |
4571 | radeon_crtc_handle_flip(rdev, 2); | |
4572 | rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT; | |
4573 | DRM_DEBUG("IH: D3 vblank\n"); | |
4574 | } | |
4575 | break; | |
4576 | case 1: /* D3 vline */ | |
4577 | if (rdev->irq.stat_regs.cik.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) { | |
4578 | rdev->irq.stat_regs.cik.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT; | |
4579 | DRM_DEBUG("IH: D3 vline\n"); | |
4580 | } | |
4581 | break; | |
4582 | default: | |
4583 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4584 | break; | |
4585 | } | |
4586 | break; | |
4587 | case 4: /* D4 vblank/vline */ | |
4588 | switch (src_data) { | |
4589 | case 0: /* D4 vblank */ | |
4590 | if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) { | |
4591 | if (rdev->irq.crtc_vblank_int[3]) { | |
4592 | drm_handle_vblank(rdev->ddev, 3); | |
4593 | rdev->pm.vblank_sync = true; | |
4594 | wake_up(&rdev->irq.vblank_queue); | |
4595 | } | |
4596 | if (atomic_read(&rdev->irq.pflip[3])) | |
4597 | radeon_crtc_handle_flip(rdev, 3); | |
4598 | rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT; | |
4599 | DRM_DEBUG("IH: D4 vblank\n"); | |
4600 | } | |
4601 | break; | |
4602 | case 1: /* D4 vline */ | |
4603 | if (rdev->irq.stat_regs.cik.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) { | |
4604 | rdev->irq.stat_regs.cik.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT; | |
4605 | DRM_DEBUG("IH: D4 vline\n"); | |
4606 | } | |
4607 | break; | |
4608 | default: | |
4609 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4610 | break; | |
4611 | } | |
4612 | break; | |
4613 | case 5: /* D5 vblank/vline */ | |
4614 | switch (src_data) { | |
4615 | case 0: /* D5 vblank */ | |
4616 | if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) { | |
4617 | if (rdev->irq.crtc_vblank_int[4]) { | |
4618 | drm_handle_vblank(rdev->ddev, 4); | |
4619 | rdev->pm.vblank_sync = true; | |
4620 | wake_up(&rdev->irq.vblank_queue); | |
4621 | } | |
4622 | if (atomic_read(&rdev->irq.pflip[4])) | |
4623 | radeon_crtc_handle_flip(rdev, 4); | |
4624 | rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT; | |
4625 | DRM_DEBUG("IH: D5 vblank\n"); | |
4626 | } | |
4627 | break; | |
4628 | case 1: /* D5 vline */ | |
4629 | if (rdev->irq.stat_regs.cik.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) { | |
4630 | rdev->irq.stat_regs.cik.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT; | |
4631 | DRM_DEBUG("IH: D5 vline\n"); | |
4632 | } | |
4633 | break; | |
4634 | default: | |
4635 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4636 | break; | |
4637 | } | |
4638 | break; | |
4639 | case 6: /* D6 vblank/vline */ | |
4640 | switch (src_data) { | |
4641 | case 0: /* D6 vblank */ | |
4642 | if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) { | |
4643 | if (rdev->irq.crtc_vblank_int[5]) { | |
4644 | drm_handle_vblank(rdev->ddev, 5); | |
4645 | rdev->pm.vblank_sync = true; | |
4646 | wake_up(&rdev->irq.vblank_queue); | |
4647 | } | |
4648 | if (atomic_read(&rdev->irq.pflip[5])) | |
4649 | radeon_crtc_handle_flip(rdev, 5); | |
4650 | rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT; | |
4651 | DRM_DEBUG("IH: D6 vblank\n"); | |
4652 | } | |
4653 | break; | |
4654 | case 1: /* D6 vline */ | |
4655 | if (rdev->irq.stat_regs.cik.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) { | |
4656 | rdev->irq.stat_regs.cik.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT; | |
4657 | DRM_DEBUG("IH: D6 vline\n"); | |
4658 | } | |
4659 | break; | |
4660 | default: | |
4661 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4662 | break; | |
4663 | } | |
4664 | break; | |
4665 | case 42: /* HPD hotplug */ | |
4666 | switch (src_data) { | |
4667 | case 0: | |
4668 | if (rdev->irq.stat_regs.cik.disp_int & DC_HPD1_INTERRUPT) { | |
4669 | rdev->irq.stat_regs.cik.disp_int &= ~DC_HPD1_INTERRUPT; | |
4670 | queue_hotplug = true; | |
4671 | DRM_DEBUG("IH: HPD1\n"); | |
4672 | } | |
4673 | break; | |
4674 | case 1: | |
4675 | if (rdev->irq.stat_regs.cik.disp_int_cont & DC_HPD2_INTERRUPT) { | |
4676 | rdev->irq.stat_regs.cik.disp_int_cont &= ~DC_HPD2_INTERRUPT; | |
4677 | queue_hotplug = true; | |
4678 | DRM_DEBUG("IH: HPD2\n"); | |
4679 | } | |
4680 | break; | |
4681 | case 2: | |
4682 | if (rdev->irq.stat_regs.cik.disp_int_cont2 & DC_HPD3_INTERRUPT) { | |
4683 | rdev->irq.stat_regs.cik.disp_int_cont2 &= ~DC_HPD3_INTERRUPT; | |
4684 | queue_hotplug = true; | |
4685 | DRM_DEBUG("IH: HPD3\n"); | |
4686 | } | |
4687 | break; | |
4688 | case 3: | |
4689 | if (rdev->irq.stat_regs.cik.disp_int_cont3 & DC_HPD4_INTERRUPT) { | |
4690 | rdev->irq.stat_regs.cik.disp_int_cont3 &= ~DC_HPD4_INTERRUPT; | |
4691 | queue_hotplug = true; | |
4692 | DRM_DEBUG("IH: HPD4\n"); | |
4693 | } | |
4694 | break; | |
4695 | case 4: | |
4696 | if (rdev->irq.stat_regs.cik.disp_int_cont4 & DC_HPD5_INTERRUPT) { | |
4697 | rdev->irq.stat_regs.cik.disp_int_cont4 &= ~DC_HPD5_INTERRUPT; | |
4698 | queue_hotplug = true; | |
4699 | DRM_DEBUG("IH: HPD5\n"); | |
4700 | } | |
4701 | break; | |
4702 | case 5: | |
4703 | if (rdev->irq.stat_regs.cik.disp_int_cont5 & DC_HPD6_INTERRUPT) { | |
4704 | rdev->irq.stat_regs.cik.disp_int_cont5 &= ~DC_HPD6_INTERRUPT; | |
4705 | queue_hotplug = true; | |
4706 | DRM_DEBUG("IH: HPD6\n"); | |
4707 | } | |
4708 | break; | |
4709 | default: | |
4710 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4711 | break; | |
4712 | } | |
4713 | break; | |
9d97c99b AD |
4714 | case 146: |
4715 | case 147: | |
4716 | dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data); | |
4717 | dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", | |
4718 | RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR)); | |
4719 | dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", | |
4720 | RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS)); | |
4721 | /* reset addr and status */ | |
4722 | WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1); | |
4723 | break; | |
a59781bb AD |
4724 | case 176: /* GFX RB CP_INT */ |
4725 | case 177: /* GFX IB CP_INT */ | |
4726 | radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX); | |
4727 | break; | |
4728 | case 181: /* CP EOP event */ | |
4729 | DRM_DEBUG("IH: CP EOP\n"); | |
21a93e13 AD |
4730 | /* XXX check the bitfield order! */ |
4731 | me_id = (ring_id & 0x60) >> 5; | |
4732 | pipe_id = (ring_id & 0x18) >> 3; | |
4733 | queue_id = (ring_id & 0x7) >> 0; | |
a59781bb AD |
4734 | switch (me_id) { |
4735 | case 0: | |
4736 | radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX); | |
4737 | break; | |
4738 | case 1: | |
4739 | /* XXX compute */ | |
4740 | break; | |
4741 | case 2: | |
4742 | /* XXX compute */ | |
4743 | break; | |
4744 | } | |
4745 | break; | |
4746 | case 184: /* CP Privileged reg access */ | |
4747 | DRM_ERROR("Illegal register access in command stream\n"); | |
4748 | /* XXX check the bitfield order! */ | |
4749 | me_id = (ring_id & 0x60) >> 5; | |
4750 | pipe_id = (ring_id & 0x18) >> 3; | |
4751 | queue_id = (ring_id & 0x7) >> 0; | |
4752 | switch (me_id) { | |
4753 | case 0: | |
4754 | /* This results in a full GPU reset, but all we need to do is soft | |
4755 | * reset the CP for gfx | |
4756 | */ | |
4757 | queue_reset = true; | |
4758 | break; | |
4759 | case 1: | |
4760 | /* XXX compute */ | |
4761 | break; | |
4762 | case 2: | |
4763 | /* XXX compute */ | |
4764 | break; | |
4765 | } | |
4766 | break; | |
4767 | case 185: /* CP Privileged inst */ | |
4768 | DRM_ERROR("Illegal instruction in command stream\n"); | |
21a93e13 AD |
4769 | /* XXX check the bitfield order! */ |
4770 | me_id = (ring_id & 0x60) >> 5; | |
4771 | pipe_id = (ring_id & 0x18) >> 3; | |
4772 | queue_id = (ring_id & 0x7) >> 0; | |
a59781bb AD |
4773 | switch (me_id) { |
4774 | case 0: | |
4775 | /* This results in a full GPU reset, but all we need to do is soft | |
4776 | * reset the CP for gfx | |
4777 | */ | |
4778 | queue_reset = true; | |
4779 | break; | |
4780 | case 1: | |
4781 | /* XXX compute */ | |
4782 | break; | |
4783 | case 2: | |
4784 | /* XXX compute */ | |
4785 | break; | |
4786 | } | |
4787 | break; | |
21a93e13 AD |
4788 | case 224: /* SDMA trap event */ |
4789 | /* XXX check the bitfield order! */ | |
4790 | me_id = (ring_id & 0x3) >> 0; | |
4791 | queue_id = (ring_id & 0xc) >> 2; | |
4792 | DRM_DEBUG("IH: SDMA trap\n"); | |
4793 | switch (me_id) { | |
4794 | case 0: | |
4795 | switch (queue_id) { | |
4796 | case 0: | |
4797 | radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX); | |
4798 | break; | |
4799 | case 1: | |
4800 | /* XXX compute */ | |
4801 | break; | |
4802 | case 2: | |
4803 | /* XXX compute */ | |
4804 | break; | |
4805 | } | |
4806 | break; | |
4807 | case 1: | |
4808 | switch (queue_id) { | |
4809 | case 0: | |
4810 | radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX); | |
4811 | break; | |
4812 | case 1: | |
4813 | /* XXX compute */ | |
4814 | break; | |
4815 | case 2: | |
4816 | /* XXX compute */ | |
4817 | break; | |
4818 | } | |
4819 | break; | |
4820 | } | |
4821 | break; | |
4822 | case 241: /* SDMA Privileged inst */ | |
4823 | case 247: /* SDMA Privileged inst */ | |
4824 | DRM_ERROR("Illegal instruction in SDMA command stream\n"); | |
4825 | /* XXX check the bitfield order! */ | |
4826 | me_id = (ring_id & 0x3) >> 0; | |
4827 | queue_id = (ring_id & 0xc) >> 2; | |
4828 | switch (me_id) { | |
4829 | case 0: | |
4830 | switch (queue_id) { | |
4831 | case 0: | |
4832 | queue_reset = true; | |
4833 | break; | |
4834 | case 1: | |
4835 | /* XXX compute */ | |
4836 | queue_reset = true; | |
4837 | break; | |
4838 | case 2: | |
4839 | /* XXX compute */ | |
4840 | queue_reset = true; | |
4841 | break; | |
4842 | } | |
4843 | break; | |
4844 | case 1: | |
4845 | switch (queue_id) { | |
4846 | case 0: | |
4847 | queue_reset = true; | |
4848 | break; | |
4849 | case 1: | |
4850 | /* XXX compute */ | |
4851 | queue_reset = true; | |
4852 | break; | |
4853 | case 2: | |
4854 | /* XXX compute */ | |
4855 | queue_reset = true; | |
4856 | break; | |
4857 | } | |
4858 | break; | |
4859 | } | |
4860 | break; | |
a59781bb AD |
4861 | case 233: /* GUI IDLE */ |
4862 | DRM_DEBUG("IH: GUI idle\n"); | |
4863 | break; | |
4864 | default: | |
4865 | DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); | |
4866 | break; | |
4867 | } | |
4868 | ||
4869 | /* wptr/rptr are in bytes! */ | |
4870 | rptr += 16; | |
4871 | rptr &= rdev->ih.ptr_mask; | |
4872 | } | |
4873 | if (queue_hotplug) | |
4874 | schedule_work(&rdev->hotplug_work); | |
4875 | if (queue_reset) | |
4876 | schedule_work(&rdev->reset_work); | |
4877 | rdev->ih.rptr = rptr; | |
4878 | WREG32(IH_RB_RPTR, rdev->ih.rptr); | |
4879 | atomic_set(&rdev->ih.lock, 0); | |
4880 | ||
4881 | /* make sure wptr hasn't changed while processing */ | |
4882 | wptr = cik_get_ih_wptr(rdev); | |
4883 | if (wptr != rptr) | |
4884 | goto restart_ih; | |
4885 | ||
4886 | return IRQ_HANDLED; | |
4887 | } | |
7bf94a2c AD |
4888 | |
4889 | /* | |
4890 | * startup/shutdown callbacks | |
4891 | */ | |
4892 | /** | |
4893 | * cik_startup - program the asic to a functional state | |
4894 | * | |
4895 | * @rdev: radeon_device pointer | |
4896 | * | |
4897 | * Programs the asic to a functional state (CIK). | |
4898 | * Called by cik_init() and cik_resume(). | |
4899 | * Returns 0 for success, error for failure. | |
4900 | */ | |
4901 | static int cik_startup(struct radeon_device *rdev) | |
4902 | { | |
4903 | struct radeon_ring *ring; | |
4904 | int r; | |
4905 | ||
4906 | if (rdev->flags & RADEON_IS_IGP) { | |
4907 | if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw || | |
4908 | !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw) { | |
4909 | r = cik_init_microcode(rdev); | |
4910 | if (r) { | |
4911 | DRM_ERROR("Failed to load firmware!\n"); | |
4912 | return r; | |
4913 | } | |
4914 | } | |
4915 | } else { | |
4916 | if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw || | |
4917 | !rdev->mec_fw || !rdev->sdma_fw || !rdev->rlc_fw || | |
4918 | !rdev->mc_fw) { | |
4919 | r = cik_init_microcode(rdev); | |
4920 | if (r) { | |
4921 | DRM_ERROR("Failed to load firmware!\n"); | |
4922 | return r; | |
4923 | } | |
4924 | } | |
4925 | ||
4926 | r = ci_mc_load_microcode(rdev); | |
4927 | if (r) { | |
4928 | DRM_ERROR("Failed to load MC firmware!\n"); | |
4929 | return r; | |
4930 | } | |
4931 | } | |
4932 | ||
4933 | r = r600_vram_scratch_init(rdev); | |
4934 | if (r) | |
4935 | return r; | |
4936 | ||
4937 | cik_mc_program(rdev); | |
4938 | r = cik_pcie_gart_enable(rdev); | |
4939 | if (r) | |
4940 | return r; | |
4941 | cik_gpu_init(rdev); | |
4942 | ||
4943 | /* allocate rlc buffers */ | |
4944 | r = si_rlc_init(rdev); | |
4945 | if (r) { | |
4946 | DRM_ERROR("Failed to init rlc BOs!\n"); | |
4947 | return r; | |
4948 | } | |
4949 | ||
4950 | /* allocate wb buffer */ | |
4951 | r = radeon_wb_init(rdev); | |
4952 | if (r) | |
4953 | return r; | |
4954 | ||
4955 | r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX); | |
4956 | if (r) { | |
4957 | dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); | |
4958 | return r; | |
4959 | } | |
4960 | ||
4961 | r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX); | |
4962 | if (r) { | |
4963 | dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); | |
4964 | return r; | |
4965 | } | |
4966 | ||
4967 | r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX); | |
4968 | if (r) { | |
4969 | dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); | |
4970 | return r; | |
4971 | } | |
4972 | ||
87167bb1 CK |
4973 | r = cik_uvd_resume(rdev); |
4974 | if (!r) { | |
4975 | r = radeon_fence_driver_start_ring(rdev, | |
4976 | R600_RING_TYPE_UVD_INDEX); | |
4977 | if (r) | |
4978 | dev_err(rdev->dev, "UVD fences init error (%d).\n", r); | |
4979 | } | |
4980 | if (r) | |
4981 | rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0; | |
4982 | ||
7bf94a2c AD |
4983 | /* Enable IRQ */ |
4984 | if (!rdev->irq.installed) { | |
4985 | r = radeon_irq_kms_init(rdev); | |
4986 | if (r) | |
4987 | return r; | |
4988 | } | |
4989 | ||
4990 | r = cik_irq_init(rdev); | |
4991 | if (r) { | |
4992 | DRM_ERROR("radeon: IH init failed (%d).\n", r); | |
4993 | radeon_irq_kms_fini(rdev); | |
4994 | return r; | |
4995 | } | |
4996 | cik_irq_set(rdev); | |
4997 | ||
4998 | ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; | |
4999 | r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET, | |
5000 | CP_RB0_RPTR, CP_RB0_WPTR, | |
5001 | 0, 0xfffff, RADEON_CP_PACKET2); | |
5002 | if (r) | |
5003 | return r; | |
5004 | ||
5005 | ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; | |
5006 | r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET, | |
5007 | SDMA0_GFX_RB_RPTR + SDMA0_REGISTER_OFFSET, | |
5008 | SDMA0_GFX_RB_WPTR + SDMA0_REGISTER_OFFSET, | |
5009 | 2, 0xfffffffc, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0)); | |
5010 | if (r) | |
5011 | return r; | |
5012 | ||
5013 | ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; | |
5014 | r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET, | |
5015 | SDMA0_GFX_RB_RPTR + SDMA1_REGISTER_OFFSET, | |
5016 | SDMA0_GFX_RB_WPTR + SDMA1_REGISTER_OFFSET, | |
5017 | 2, 0xfffffffc, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0)); | |
5018 | if (r) | |
5019 | return r; | |
5020 | ||
5021 | r = cik_cp_resume(rdev); | |
5022 | if (r) | |
5023 | return r; | |
5024 | ||
5025 | r = cik_sdma_resume(rdev); | |
5026 | if (r) | |
5027 | return r; | |
5028 | ||
87167bb1 CK |
5029 | ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX]; |
5030 | if (ring->ring_size) { | |
5031 | r = radeon_ring_init(rdev, ring, ring->ring_size, | |
5032 | R600_WB_UVD_RPTR_OFFSET, | |
5033 | UVD_RBC_RB_RPTR, UVD_RBC_RB_WPTR, | |
5034 | 0, 0xfffff, RADEON_CP_PACKET2); | |
5035 | if (!r) | |
5036 | r = r600_uvd_init(rdev); | |
5037 | if (r) | |
5038 | DRM_ERROR("radeon: failed initializing UVD (%d).\n", r); | |
5039 | } | |
5040 | ||
7bf94a2c AD |
5041 | r = radeon_ib_pool_init(rdev); |
5042 | if (r) { | |
5043 | dev_err(rdev->dev, "IB initialization failed (%d).\n", r); | |
5044 | return r; | |
5045 | } | |
5046 | ||
5047 | r = radeon_vm_manager_init(rdev); | |
5048 | if (r) { | |
5049 | dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r); | |
5050 | return r; | |
5051 | } | |
5052 | ||
5053 | return 0; | |
5054 | } | |
5055 | ||
5056 | /** | |
5057 | * cik_resume - resume the asic to a functional state | |
5058 | * | |
5059 | * @rdev: radeon_device pointer | |
5060 | * | |
5061 | * Programs the asic to a functional state (CIK). | |
5062 | * Called at resume. | |
5063 | * Returns 0 for success, error for failure. | |
5064 | */ | |
5065 | int cik_resume(struct radeon_device *rdev) | |
5066 | { | |
5067 | int r; | |
5068 | ||
5069 | /* post card */ | |
5070 | atom_asic_init(rdev->mode_info.atom_context); | |
5071 | ||
5072 | rdev->accel_working = true; | |
5073 | r = cik_startup(rdev); | |
5074 | if (r) { | |
5075 | DRM_ERROR("cik startup failed on resume\n"); | |
5076 | rdev->accel_working = false; | |
5077 | return r; | |
5078 | } | |
5079 | ||
5080 | return r; | |
5081 | ||
5082 | } | |
5083 | ||
5084 | /** | |
5085 | * cik_suspend - suspend the asic | |
5086 | * | |
5087 | * @rdev: radeon_device pointer | |
5088 | * | |
5089 | * Bring the chip into a state suitable for suspend (CIK). | |
5090 | * Called at suspend. | |
5091 | * Returns 0 for success. | |
5092 | */ | |
5093 | int cik_suspend(struct radeon_device *rdev) | |
5094 | { | |
5095 | radeon_vm_manager_fini(rdev); | |
5096 | cik_cp_enable(rdev, false); | |
5097 | cik_sdma_enable(rdev, false); | |
87167bb1 CK |
5098 | r600_uvd_rbc_stop(rdev); |
5099 | radeon_uvd_suspend(rdev); | |
7bf94a2c AD |
5100 | cik_irq_suspend(rdev); |
5101 | radeon_wb_disable(rdev); | |
5102 | cik_pcie_gart_disable(rdev); | |
5103 | return 0; | |
5104 | } | |
5105 | ||
5106 | /* Plan is to move initialization in that function and use | |
5107 | * helper function so that radeon_device_init pretty much | |
5108 | * do nothing more than calling asic specific function. This | |
5109 | * should also allow to remove a bunch of callback function | |
5110 | * like vram_info. | |
5111 | */ | |
5112 | /** | |
5113 | * cik_init - asic specific driver and hw init | |
5114 | * | |
5115 | * @rdev: radeon_device pointer | |
5116 | * | |
5117 | * Setup asic specific driver variables and program the hw | |
5118 | * to a functional state (CIK). | |
5119 | * Called at driver startup. | |
5120 | * Returns 0 for success, errors for failure. | |
5121 | */ | |
5122 | int cik_init(struct radeon_device *rdev) | |
5123 | { | |
5124 | struct radeon_ring *ring; | |
5125 | int r; | |
5126 | ||
5127 | /* Read BIOS */ | |
5128 | if (!radeon_get_bios(rdev)) { | |
5129 | if (ASIC_IS_AVIVO(rdev)) | |
5130 | return -EINVAL; | |
5131 | } | |
5132 | /* Must be an ATOMBIOS */ | |
5133 | if (!rdev->is_atom_bios) { | |
5134 | dev_err(rdev->dev, "Expecting atombios for cayman GPU\n"); | |
5135 | return -EINVAL; | |
5136 | } | |
5137 | r = radeon_atombios_init(rdev); | |
5138 | if (r) | |
5139 | return r; | |
5140 | ||
5141 | /* Post card if necessary */ | |
5142 | if (!radeon_card_posted(rdev)) { | |
5143 | if (!rdev->bios) { | |
5144 | dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n"); | |
5145 | return -EINVAL; | |
5146 | } | |
5147 | DRM_INFO("GPU not posted. posting now...\n"); | |
5148 | atom_asic_init(rdev->mode_info.atom_context); | |
5149 | } | |
5150 | /* Initialize scratch registers */ | |
5151 | cik_scratch_init(rdev); | |
5152 | /* Initialize surface registers */ | |
5153 | radeon_surface_init(rdev); | |
5154 | /* Initialize clocks */ | |
5155 | radeon_get_clock_info(rdev->ddev); | |
5156 | ||
5157 | /* Fence driver */ | |
5158 | r = radeon_fence_driver_init(rdev); | |
5159 | if (r) | |
5160 | return r; | |
5161 | ||
5162 | /* initialize memory controller */ | |
5163 | r = cik_mc_init(rdev); | |
5164 | if (r) | |
5165 | return r; | |
5166 | /* Memory manager */ | |
5167 | r = radeon_bo_init(rdev); | |
5168 | if (r) | |
5169 | return r; | |
5170 | ||
5171 | ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; | |
5172 | ring->ring_obj = NULL; | |
5173 | r600_ring_init(rdev, ring, 1024 * 1024); | |
5174 | ||
5175 | ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; | |
5176 | ring->ring_obj = NULL; | |
5177 | r600_ring_init(rdev, ring, 256 * 1024); | |
5178 | ||
5179 | ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; | |
5180 | ring->ring_obj = NULL; | |
5181 | r600_ring_init(rdev, ring, 256 * 1024); | |
5182 | ||
87167bb1 CK |
5183 | r = radeon_uvd_init(rdev); |
5184 | if (!r) { | |
5185 | ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX]; | |
5186 | ring->ring_obj = NULL; | |
5187 | r600_ring_init(rdev, ring, 4096); | |
5188 | } | |
5189 | ||
7bf94a2c AD |
5190 | rdev->ih.ring_obj = NULL; |
5191 | r600_ih_ring_init(rdev, 64 * 1024); | |
5192 | ||
5193 | r = r600_pcie_gart_init(rdev); | |
5194 | if (r) | |
5195 | return r; | |
5196 | ||
5197 | rdev->accel_working = true; | |
5198 | r = cik_startup(rdev); | |
5199 | if (r) { | |
5200 | dev_err(rdev->dev, "disabling GPU acceleration\n"); | |
5201 | cik_cp_fini(rdev); | |
5202 | cik_sdma_fini(rdev); | |
5203 | cik_irq_fini(rdev); | |
5204 | si_rlc_fini(rdev); | |
5205 | radeon_wb_fini(rdev); | |
5206 | radeon_ib_pool_fini(rdev); | |
5207 | radeon_vm_manager_fini(rdev); | |
5208 | radeon_irq_kms_fini(rdev); | |
5209 | cik_pcie_gart_fini(rdev); | |
5210 | rdev->accel_working = false; | |
5211 | } | |
5212 | ||
5213 | /* Don't start up if the MC ucode is missing. | |
5214 | * The default clocks and voltages before the MC ucode | |
5215 | * is loaded are not suffient for advanced operations. | |
5216 | */ | |
5217 | if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) { | |
5218 | DRM_ERROR("radeon: MC ucode required for NI+.\n"); | |
5219 | return -EINVAL; | |
5220 | } | |
5221 | ||
5222 | return 0; | |
5223 | } | |
5224 | ||
5225 | /** | |
5226 | * cik_fini - asic specific driver and hw fini | |
5227 | * | |
5228 | * @rdev: radeon_device pointer | |
5229 | * | |
5230 | * Tear down the asic specific driver variables and program the hw | |
5231 | * to an idle state (CIK). | |
5232 | * Called at driver unload. | |
5233 | */ | |
5234 | void cik_fini(struct radeon_device *rdev) | |
5235 | { | |
5236 | cik_cp_fini(rdev); | |
5237 | cik_sdma_fini(rdev); | |
5238 | cik_irq_fini(rdev); | |
5239 | si_rlc_fini(rdev); | |
5240 | radeon_wb_fini(rdev); | |
5241 | radeon_vm_manager_fini(rdev); | |
5242 | radeon_ib_pool_fini(rdev); | |
5243 | radeon_irq_kms_fini(rdev); | |
87167bb1 | 5244 | radeon_uvd_fini(rdev); |
7bf94a2c AD |
5245 | cik_pcie_gart_fini(rdev); |
5246 | r600_vram_scratch_fini(rdev); | |
5247 | radeon_gem_fini(rdev); | |
5248 | radeon_fence_driver_fini(rdev); | |
5249 | radeon_bo_fini(rdev); | |
5250 | radeon_atombios_fini(rdev); | |
5251 | kfree(rdev->bios); | |
5252 | rdev->bios = NULL; | |
5253 | } | |
cd84a27d AD |
5254 | |
5255 | /* display watermark setup */ | |
5256 | /** | |
5257 | * dce8_line_buffer_adjust - Set up the line buffer | |
5258 | * | |
5259 | * @rdev: radeon_device pointer | |
5260 | * @radeon_crtc: the selected display controller | |
5261 | * @mode: the current display mode on the selected display | |
5262 | * controller | |
5263 | * | |
5264 | * Setup up the line buffer allocation for | |
5265 | * the selected display controller (CIK). | |
5266 | * Returns the line buffer size in pixels. | |
5267 | */ | |
5268 | static u32 dce8_line_buffer_adjust(struct radeon_device *rdev, | |
5269 | struct radeon_crtc *radeon_crtc, | |
5270 | struct drm_display_mode *mode) | |
5271 | { | |
5272 | u32 tmp; | |
5273 | ||
5274 | /* | |
5275 | * Line Buffer Setup | |
5276 | * There are 6 line buffers, one for each display controllers. | |
5277 | * There are 3 partitions per LB. Select the number of partitions | |
5278 | * to enable based on the display width. For display widths larger | |
5279 | * than 4096, you need use to use 2 display controllers and combine | |
5280 | * them using the stereo blender. | |
5281 | */ | |
5282 | if (radeon_crtc->base.enabled && mode) { | |
5283 | if (mode->crtc_hdisplay < 1920) | |
5284 | tmp = 1; | |
5285 | else if (mode->crtc_hdisplay < 2560) | |
5286 | tmp = 2; | |
5287 | else if (mode->crtc_hdisplay < 4096) | |
5288 | tmp = 0; | |
5289 | else { | |
5290 | DRM_DEBUG_KMS("Mode too big for LB!\n"); | |
5291 | tmp = 0; | |
5292 | } | |
5293 | } else | |
5294 | tmp = 1; | |
5295 | ||
5296 | WREG32(LB_MEMORY_CTRL + radeon_crtc->crtc_offset, | |
5297 | LB_MEMORY_CONFIG(tmp) | LB_MEMORY_SIZE(0x6B0)); | |
5298 | ||
5299 | if (radeon_crtc->base.enabled && mode) { | |
5300 | switch (tmp) { | |
5301 | case 0: | |
5302 | default: | |
5303 | return 4096 * 2; | |
5304 | case 1: | |
5305 | return 1920 * 2; | |
5306 | case 2: | |
5307 | return 2560 * 2; | |
5308 | } | |
5309 | } | |
5310 | ||
5311 | /* controller not enabled, so no lb used */ | |
5312 | return 0; | |
5313 | } | |
5314 | ||
5315 | /** | |
5316 | * cik_get_number_of_dram_channels - get the number of dram channels | |
5317 | * | |
5318 | * @rdev: radeon_device pointer | |
5319 | * | |
5320 | * Look up the number of video ram channels (CIK). | |
5321 | * Used for display watermark bandwidth calculations | |
5322 | * Returns the number of dram channels | |
5323 | */ | |
5324 | static u32 cik_get_number_of_dram_channels(struct radeon_device *rdev) | |
5325 | { | |
5326 | u32 tmp = RREG32(MC_SHARED_CHMAP); | |
5327 | ||
5328 | switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) { | |
5329 | case 0: | |
5330 | default: | |
5331 | return 1; | |
5332 | case 1: | |
5333 | return 2; | |
5334 | case 2: | |
5335 | return 4; | |
5336 | case 3: | |
5337 | return 8; | |
5338 | case 4: | |
5339 | return 3; | |
5340 | case 5: | |
5341 | return 6; | |
5342 | case 6: | |
5343 | return 10; | |
5344 | case 7: | |
5345 | return 12; | |
5346 | case 8: | |
5347 | return 16; | |
5348 | } | |
5349 | } | |
5350 | ||
5351 | struct dce8_wm_params { | |
5352 | u32 dram_channels; /* number of dram channels */ | |
5353 | u32 yclk; /* bandwidth per dram data pin in kHz */ | |
5354 | u32 sclk; /* engine clock in kHz */ | |
5355 | u32 disp_clk; /* display clock in kHz */ | |
5356 | u32 src_width; /* viewport width */ | |
5357 | u32 active_time; /* active display time in ns */ | |
5358 | u32 blank_time; /* blank time in ns */ | |
5359 | bool interlaced; /* mode is interlaced */ | |
5360 | fixed20_12 vsc; /* vertical scale ratio */ | |
5361 | u32 num_heads; /* number of active crtcs */ | |
5362 | u32 bytes_per_pixel; /* bytes per pixel display + overlay */ | |
5363 | u32 lb_size; /* line buffer allocated to pipe */ | |
5364 | u32 vtaps; /* vertical scaler taps */ | |
5365 | }; | |
5366 | ||
5367 | /** | |
5368 | * dce8_dram_bandwidth - get the dram bandwidth | |
5369 | * | |
5370 | * @wm: watermark calculation data | |
5371 | * | |
5372 | * Calculate the raw dram bandwidth (CIK). | |
5373 | * Used for display watermark bandwidth calculations | |
5374 | * Returns the dram bandwidth in MBytes/s | |
5375 | */ | |
5376 | static u32 dce8_dram_bandwidth(struct dce8_wm_params *wm) | |
5377 | { | |
5378 | /* Calculate raw DRAM Bandwidth */ | |
5379 | fixed20_12 dram_efficiency; /* 0.7 */ | |
5380 | fixed20_12 yclk, dram_channels, bandwidth; | |
5381 | fixed20_12 a; | |
5382 | ||
5383 | a.full = dfixed_const(1000); | |
5384 | yclk.full = dfixed_const(wm->yclk); | |
5385 | yclk.full = dfixed_div(yclk, a); | |
5386 | dram_channels.full = dfixed_const(wm->dram_channels * 4); | |
5387 | a.full = dfixed_const(10); | |
5388 | dram_efficiency.full = dfixed_const(7); | |
5389 | dram_efficiency.full = dfixed_div(dram_efficiency, a); | |
5390 | bandwidth.full = dfixed_mul(dram_channels, yclk); | |
5391 | bandwidth.full = dfixed_mul(bandwidth, dram_efficiency); | |
5392 | ||
5393 | return dfixed_trunc(bandwidth); | |
5394 | } | |
5395 | ||
5396 | /** | |
5397 | * dce8_dram_bandwidth_for_display - get the dram bandwidth for display | |
5398 | * | |
5399 | * @wm: watermark calculation data | |
5400 | * | |
5401 | * Calculate the dram bandwidth used for display (CIK). | |
5402 | * Used for display watermark bandwidth calculations | |
5403 | * Returns the dram bandwidth for display in MBytes/s | |
5404 | */ | |
5405 | static u32 dce8_dram_bandwidth_for_display(struct dce8_wm_params *wm) | |
5406 | { | |
5407 | /* Calculate DRAM Bandwidth and the part allocated to display. */ | |
5408 | fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */ | |
5409 | fixed20_12 yclk, dram_channels, bandwidth; | |
5410 | fixed20_12 a; | |
5411 | ||
5412 | a.full = dfixed_const(1000); | |
5413 | yclk.full = dfixed_const(wm->yclk); | |
5414 | yclk.full = dfixed_div(yclk, a); | |
5415 | dram_channels.full = dfixed_const(wm->dram_channels * 4); | |
5416 | a.full = dfixed_const(10); | |
5417 | disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */ | |
5418 | disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a); | |
5419 | bandwidth.full = dfixed_mul(dram_channels, yclk); | |
5420 | bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation); | |
5421 | ||
5422 | return dfixed_trunc(bandwidth); | |
5423 | } | |
5424 | ||
5425 | /** | |
5426 | * dce8_data_return_bandwidth - get the data return bandwidth | |
5427 | * | |
5428 | * @wm: watermark calculation data | |
5429 | * | |
5430 | * Calculate the data return bandwidth used for display (CIK). | |
5431 | * Used for display watermark bandwidth calculations | |
5432 | * Returns the data return bandwidth in MBytes/s | |
5433 | */ | |
5434 | static u32 dce8_data_return_bandwidth(struct dce8_wm_params *wm) | |
5435 | { | |
5436 | /* Calculate the display Data return Bandwidth */ | |
5437 | fixed20_12 return_efficiency; /* 0.8 */ | |
5438 | fixed20_12 sclk, bandwidth; | |
5439 | fixed20_12 a; | |
5440 | ||
5441 | a.full = dfixed_const(1000); | |
5442 | sclk.full = dfixed_const(wm->sclk); | |
5443 | sclk.full = dfixed_div(sclk, a); | |
5444 | a.full = dfixed_const(10); | |
5445 | return_efficiency.full = dfixed_const(8); | |
5446 | return_efficiency.full = dfixed_div(return_efficiency, a); | |
5447 | a.full = dfixed_const(32); | |
5448 | bandwidth.full = dfixed_mul(a, sclk); | |
5449 | bandwidth.full = dfixed_mul(bandwidth, return_efficiency); | |
5450 | ||
5451 | return dfixed_trunc(bandwidth); | |
5452 | } | |
5453 | ||
5454 | /** | |
5455 | * dce8_dmif_request_bandwidth - get the dmif bandwidth | |
5456 | * | |
5457 | * @wm: watermark calculation data | |
5458 | * | |
5459 | * Calculate the dmif bandwidth used for display (CIK). | |
5460 | * Used for display watermark bandwidth calculations | |
5461 | * Returns the dmif bandwidth in MBytes/s | |
5462 | */ | |
5463 | static u32 dce8_dmif_request_bandwidth(struct dce8_wm_params *wm) | |
5464 | { | |
5465 | /* Calculate the DMIF Request Bandwidth */ | |
5466 | fixed20_12 disp_clk_request_efficiency; /* 0.8 */ | |
5467 | fixed20_12 disp_clk, bandwidth; | |
5468 | fixed20_12 a, b; | |
5469 | ||
5470 | a.full = dfixed_const(1000); | |
5471 | disp_clk.full = dfixed_const(wm->disp_clk); | |
5472 | disp_clk.full = dfixed_div(disp_clk, a); | |
5473 | a.full = dfixed_const(32); | |
5474 | b.full = dfixed_mul(a, disp_clk); | |
5475 | ||
5476 | a.full = dfixed_const(10); | |
5477 | disp_clk_request_efficiency.full = dfixed_const(8); | |
5478 | disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a); | |
5479 | ||
5480 | bandwidth.full = dfixed_mul(b, disp_clk_request_efficiency); | |
5481 | ||
5482 | return dfixed_trunc(bandwidth); | |
5483 | } | |
5484 | ||
5485 | /** | |
5486 | * dce8_available_bandwidth - get the min available bandwidth | |
5487 | * | |
5488 | * @wm: watermark calculation data | |
5489 | * | |
5490 | * Calculate the min available bandwidth used for display (CIK). | |
5491 | * Used for display watermark bandwidth calculations | |
5492 | * Returns the min available bandwidth in MBytes/s | |
5493 | */ | |
5494 | static u32 dce8_available_bandwidth(struct dce8_wm_params *wm) | |
5495 | { | |
5496 | /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */ | |
5497 | u32 dram_bandwidth = dce8_dram_bandwidth(wm); | |
5498 | u32 data_return_bandwidth = dce8_data_return_bandwidth(wm); | |
5499 | u32 dmif_req_bandwidth = dce8_dmif_request_bandwidth(wm); | |
5500 | ||
5501 | return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth)); | |
5502 | } | |
5503 | ||
5504 | /** | |
5505 | * dce8_average_bandwidth - get the average available bandwidth | |
5506 | * | |
5507 | * @wm: watermark calculation data | |
5508 | * | |
5509 | * Calculate the average available bandwidth used for display (CIK). | |
5510 | * Used for display watermark bandwidth calculations | |
5511 | * Returns the average available bandwidth in MBytes/s | |
5512 | */ | |
5513 | static u32 dce8_average_bandwidth(struct dce8_wm_params *wm) | |
5514 | { | |
5515 | /* Calculate the display mode Average Bandwidth | |
5516 | * DisplayMode should contain the source and destination dimensions, | |
5517 | * timing, etc. | |
5518 | */ | |
5519 | fixed20_12 bpp; | |
5520 | fixed20_12 line_time; | |
5521 | fixed20_12 src_width; | |
5522 | fixed20_12 bandwidth; | |
5523 | fixed20_12 a; | |
5524 | ||
5525 | a.full = dfixed_const(1000); | |
5526 | line_time.full = dfixed_const(wm->active_time + wm->blank_time); | |
5527 | line_time.full = dfixed_div(line_time, a); | |
5528 | bpp.full = dfixed_const(wm->bytes_per_pixel); | |
5529 | src_width.full = dfixed_const(wm->src_width); | |
5530 | bandwidth.full = dfixed_mul(src_width, bpp); | |
5531 | bandwidth.full = dfixed_mul(bandwidth, wm->vsc); | |
5532 | bandwidth.full = dfixed_div(bandwidth, line_time); | |
5533 | ||
5534 | return dfixed_trunc(bandwidth); | |
5535 | } | |
5536 | ||
5537 | /** | |
5538 | * dce8_latency_watermark - get the latency watermark | |
5539 | * | |
5540 | * @wm: watermark calculation data | |
5541 | * | |
5542 | * Calculate the latency watermark (CIK). | |
5543 | * Used for display watermark bandwidth calculations | |
5544 | * Returns the latency watermark in ns | |
5545 | */ | |
5546 | static u32 dce8_latency_watermark(struct dce8_wm_params *wm) | |
5547 | { | |
5548 | /* First calculate the latency in ns */ | |
5549 | u32 mc_latency = 2000; /* 2000 ns. */ | |
5550 | u32 available_bandwidth = dce8_available_bandwidth(wm); | |
5551 | u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth; | |
5552 | u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth; | |
5553 | u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */ | |
5554 | u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) + | |
5555 | (wm->num_heads * cursor_line_pair_return_time); | |
5556 | u32 latency = mc_latency + other_heads_data_return_time + dc_latency; | |
5557 | u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time; | |
5558 | u32 tmp, dmif_size = 12288; | |
5559 | fixed20_12 a, b, c; | |
5560 | ||
5561 | if (wm->num_heads == 0) | |
5562 | return 0; | |
5563 | ||
5564 | a.full = dfixed_const(2); | |
5565 | b.full = dfixed_const(1); | |
5566 | if ((wm->vsc.full > a.full) || | |
5567 | ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) || | |
5568 | (wm->vtaps >= 5) || | |
5569 | ((wm->vsc.full >= a.full) && wm->interlaced)) | |
5570 | max_src_lines_per_dst_line = 4; | |
5571 | else | |
5572 | max_src_lines_per_dst_line = 2; | |
5573 | ||
5574 | a.full = dfixed_const(available_bandwidth); | |
5575 | b.full = dfixed_const(wm->num_heads); | |
5576 | a.full = dfixed_div(a, b); | |
5577 | ||
5578 | b.full = dfixed_const(mc_latency + 512); | |
5579 | c.full = dfixed_const(wm->disp_clk); | |
5580 | b.full = dfixed_div(b, c); | |
5581 | ||
5582 | c.full = dfixed_const(dmif_size); | |
5583 | b.full = dfixed_div(c, b); | |
5584 | ||
5585 | tmp = min(dfixed_trunc(a), dfixed_trunc(b)); | |
5586 | ||
5587 | b.full = dfixed_const(1000); | |
5588 | c.full = dfixed_const(wm->disp_clk); | |
5589 | b.full = dfixed_div(c, b); | |
5590 | c.full = dfixed_const(wm->bytes_per_pixel); | |
5591 | b.full = dfixed_mul(b, c); | |
5592 | ||
5593 | lb_fill_bw = min(tmp, dfixed_trunc(b)); | |
5594 | ||
5595 | a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel); | |
5596 | b.full = dfixed_const(1000); | |
5597 | c.full = dfixed_const(lb_fill_bw); | |
5598 | b.full = dfixed_div(c, b); | |
5599 | a.full = dfixed_div(a, b); | |
5600 | line_fill_time = dfixed_trunc(a); | |
5601 | ||
5602 | if (line_fill_time < wm->active_time) | |
5603 | return latency; | |
5604 | else | |
5605 | return latency + (line_fill_time - wm->active_time); | |
5606 | ||
5607 | } | |
5608 | ||
5609 | /** | |
5610 | * dce8_average_bandwidth_vs_dram_bandwidth_for_display - check | |
5611 | * average and available dram bandwidth | |
5612 | * | |
5613 | * @wm: watermark calculation data | |
5614 | * | |
5615 | * Check if the display average bandwidth fits in the display | |
5616 | * dram bandwidth (CIK). | |
5617 | * Used for display watermark bandwidth calculations | |
5618 | * Returns true if the display fits, false if not. | |
5619 | */ | |
5620 | static bool dce8_average_bandwidth_vs_dram_bandwidth_for_display(struct dce8_wm_params *wm) | |
5621 | { | |
5622 | if (dce8_average_bandwidth(wm) <= | |
5623 | (dce8_dram_bandwidth_for_display(wm) / wm->num_heads)) | |
5624 | return true; | |
5625 | else | |
5626 | return false; | |
5627 | } | |
5628 | ||
5629 | /** | |
5630 | * dce8_average_bandwidth_vs_available_bandwidth - check | |
5631 | * average and available bandwidth | |
5632 | * | |
5633 | * @wm: watermark calculation data | |
5634 | * | |
5635 | * Check if the display average bandwidth fits in the display | |
5636 | * available bandwidth (CIK). | |
5637 | * Used for display watermark bandwidth calculations | |
5638 | * Returns true if the display fits, false if not. | |
5639 | */ | |
5640 | static bool dce8_average_bandwidth_vs_available_bandwidth(struct dce8_wm_params *wm) | |
5641 | { | |
5642 | if (dce8_average_bandwidth(wm) <= | |
5643 | (dce8_available_bandwidth(wm) / wm->num_heads)) | |
5644 | return true; | |
5645 | else | |
5646 | return false; | |
5647 | } | |
5648 | ||
5649 | /** | |
5650 | * dce8_check_latency_hiding - check latency hiding | |
5651 | * | |
5652 | * @wm: watermark calculation data | |
5653 | * | |
5654 | * Check latency hiding (CIK). | |
5655 | * Used for display watermark bandwidth calculations | |
5656 | * Returns true if the display fits, false if not. | |
5657 | */ | |
5658 | static bool dce8_check_latency_hiding(struct dce8_wm_params *wm) | |
5659 | { | |
5660 | u32 lb_partitions = wm->lb_size / wm->src_width; | |
5661 | u32 line_time = wm->active_time + wm->blank_time; | |
5662 | u32 latency_tolerant_lines; | |
5663 | u32 latency_hiding; | |
5664 | fixed20_12 a; | |
5665 | ||
5666 | a.full = dfixed_const(1); | |
5667 | if (wm->vsc.full > a.full) | |
5668 | latency_tolerant_lines = 1; | |
5669 | else { | |
5670 | if (lb_partitions <= (wm->vtaps + 1)) | |
5671 | latency_tolerant_lines = 1; | |
5672 | else | |
5673 | latency_tolerant_lines = 2; | |
5674 | } | |
5675 | ||
5676 | latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time); | |
5677 | ||
5678 | if (dce8_latency_watermark(wm) <= latency_hiding) | |
5679 | return true; | |
5680 | else | |
5681 | return false; | |
5682 | } | |
5683 | ||
5684 | /** | |
5685 | * dce8_program_watermarks - program display watermarks | |
5686 | * | |
5687 | * @rdev: radeon_device pointer | |
5688 | * @radeon_crtc: the selected display controller | |
5689 | * @lb_size: line buffer size | |
5690 | * @num_heads: number of display controllers in use | |
5691 | * | |
5692 | * Calculate and program the display watermarks for the | |
5693 | * selected display controller (CIK). | |
5694 | */ | |
5695 | static void dce8_program_watermarks(struct radeon_device *rdev, | |
5696 | struct radeon_crtc *radeon_crtc, | |
5697 | u32 lb_size, u32 num_heads) | |
5698 | { | |
5699 | struct drm_display_mode *mode = &radeon_crtc->base.mode; | |
5700 | struct dce8_wm_params wm; | |
5701 | u32 pixel_period; | |
5702 | u32 line_time = 0; | |
5703 | u32 latency_watermark_a = 0, latency_watermark_b = 0; | |
5704 | u32 tmp, wm_mask; | |
5705 | ||
5706 | if (radeon_crtc->base.enabled && num_heads && mode) { | |
5707 | pixel_period = 1000000 / (u32)mode->clock; | |
5708 | line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535); | |
5709 | ||
5710 | wm.yclk = rdev->pm.current_mclk * 10; | |
5711 | wm.sclk = rdev->pm.current_sclk * 10; | |
5712 | wm.disp_clk = mode->clock; | |
5713 | wm.src_width = mode->crtc_hdisplay; | |
5714 | wm.active_time = mode->crtc_hdisplay * pixel_period; | |
5715 | wm.blank_time = line_time - wm.active_time; | |
5716 | wm.interlaced = false; | |
5717 | if (mode->flags & DRM_MODE_FLAG_INTERLACE) | |
5718 | wm.interlaced = true; | |
5719 | wm.vsc = radeon_crtc->vsc; | |
5720 | wm.vtaps = 1; | |
5721 | if (radeon_crtc->rmx_type != RMX_OFF) | |
5722 | wm.vtaps = 2; | |
5723 | wm.bytes_per_pixel = 4; /* XXX: get this from fb config */ | |
5724 | wm.lb_size = lb_size; | |
5725 | wm.dram_channels = cik_get_number_of_dram_channels(rdev); | |
5726 | wm.num_heads = num_heads; | |
5727 | ||
5728 | /* set for high clocks */ | |
5729 | latency_watermark_a = min(dce8_latency_watermark(&wm), (u32)65535); | |
5730 | /* set for low clocks */ | |
5731 | /* wm.yclk = low clk; wm.sclk = low clk */ | |
5732 | latency_watermark_b = min(dce8_latency_watermark(&wm), (u32)65535); | |
5733 | ||
5734 | /* possibly force display priority to high */ | |
5735 | /* should really do this at mode validation time... */ | |
5736 | if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm) || | |
5737 | !dce8_average_bandwidth_vs_available_bandwidth(&wm) || | |
5738 | !dce8_check_latency_hiding(&wm) || | |
5739 | (rdev->disp_priority == 2)) { | |
5740 | DRM_DEBUG_KMS("force priority to high\n"); | |
5741 | } | |
5742 | } | |
5743 | ||
5744 | /* select wm A */ | |
5745 | wm_mask = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset); | |
5746 | tmp = wm_mask; | |
5747 | tmp &= ~LATENCY_WATERMARK_MASK(3); | |
5748 | tmp |= LATENCY_WATERMARK_MASK(1); | |
5749 | WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp); | |
5750 | WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset, | |
5751 | (LATENCY_LOW_WATERMARK(latency_watermark_a) | | |
5752 | LATENCY_HIGH_WATERMARK(line_time))); | |
5753 | /* select wm B */ | |
5754 | tmp = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset); | |
5755 | tmp &= ~LATENCY_WATERMARK_MASK(3); | |
5756 | tmp |= LATENCY_WATERMARK_MASK(2); | |
5757 | WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp); | |
5758 | WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset, | |
5759 | (LATENCY_LOW_WATERMARK(latency_watermark_b) | | |
5760 | LATENCY_HIGH_WATERMARK(line_time))); | |
5761 | /* restore original selection */ | |
5762 | WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, wm_mask); | |
5763 | } | |
5764 | ||
5765 | /** | |
5766 | * dce8_bandwidth_update - program display watermarks | |
5767 | * | |
5768 | * @rdev: radeon_device pointer | |
5769 | * | |
5770 | * Calculate and program the display watermarks and line | |
5771 | * buffer allocation (CIK). | |
5772 | */ | |
5773 | void dce8_bandwidth_update(struct radeon_device *rdev) | |
5774 | { | |
5775 | struct drm_display_mode *mode = NULL; | |
5776 | u32 num_heads = 0, lb_size; | |
5777 | int i; | |
5778 | ||
5779 | radeon_update_display_priority(rdev); | |
5780 | ||
5781 | for (i = 0; i < rdev->num_crtc; i++) { | |
5782 | if (rdev->mode_info.crtcs[i]->base.enabled) | |
5783 | num_heads++; | |
5784 | } | |
5785 | for (i = 0; i < rdev->num_crtc; i++) { | |
5786 | mode = &rdev->mode_info.crtcs[i]->base.mode; | |
5787 | lb_size = dce8_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode); | |
5788 | dce8_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads); | |
5789 | } | |
5790 | } | |
44fa346f AD |
5791 | |
5792 | /** | |
5793 | * cik_get_gpu_clock_counter - return GPU clock counter snapshot | |
5794 | * | |
5795 | * @rdev: radeon_device pointer | |
5796 | * | |
5797 | * Fetches a GPU clock counter snapshot (SI). | |
5798 | * Returns the 64 bit clock counter snapshot. | |
5799 | */ | |
5800 | uint64_t cik_get_gpu_clock_counter(struct radeon_device *rdev) | |
5801 | { | |
5802 | uint64_t clock; | |
5803 | ||
5804 | mutex_lock(&rdev->gpu_clock_mutex); | |
5805 | WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1); | |
5806 | clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) | | |
5807 | ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL); | |
5808 | mutex_unlock(&rdev->gpu_clock_mutex); | |
5809 | return clock; | |
5810 | } | |
5811 | ||
87167bb1 CK |
5812 | static int cik_set_uvd_clock(struct radeon_device *rdev, u32 clock, |
5813 | u32 cntl_reg, u32 status_reg) | |
5814 | { | |
5815 | int r, i; | |
5816 | struct atom_clock_dividers dividers; | |
5817 | uint32_t tmp; | |
5818 | ||
5819 | r = radeon_atom_get_clock_dividers(rdev, COMPUTE_GPUCLK_INPUT_FLAG_DEFAULT_GPUCLK, | |
5820 | clock, false, ÷rs); | |
5821 | if (r) | |
5822 | return r; | |
5823 | ||
5824 | tmp = RREG32_SMC(cntl_reg); | |
5825 | tmp &= ~(DCLK_DIR_CNTL_EN|DCLK_DIVIDER_MASK); | |
5826 | tmp |= dividers.post_divider; | |
5827 | WREG32_SMC(cntl_reg, tmp); | |
5828 | ||
5829 | for (i = 0; i < 100; i++) { | |
5830 | if (RREG32_SMC(status_reg) & DCLK_STATUS) | |
5831 | break; | |
5832 | mdelay(10); | |
5833 | } | |
5834 | if (i == 100) | |
5835 | return -ETIMEDOUT; | |
5836 | ||
5837 | return 0; | |
5838 | } | |
5839 | ||
5840 | int cik_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk) | |
5841 | { | |
5842 | int r = 0; | |
5843 | ||
5844 | r = cik_set_uvd_clock(rdev, vclk, CG_VCLK_CNTL, CG_VCLK_STATUS); | |
5845 | if (r) | |
5846 | return r; | |
5847 | ||
5848 | r = cik_set_uvd_clock(rdev, dclk, CG_DCLK_CNTL, CG_DCLK_STATUS); | |
5849 | return r; | |
5850 | } | |
5851 | ||
5852 | int cik_uvd_resume(struct radeon_device *rdev) | |
5853 | { | |
5854 | uint64_t addr; | |
5855 | uint32_t size; | |
5856 | int r; | |
5857 | ||
5858 | r = radeon_uvd_resume(rdev); | |
5859 | if (r) | |
5860 | return r; | |
5861 | ||
5862 | /* programm the VCPU memory controller bits 0-27 */ | |
5863 | addr = rdev->uvd.gpu_addr >> 3; | |
5864 | size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 4) >> 3; | |
5865 | WREG32(UVD_VCPU_CACHE_OFFSET0, addr); | |
5866 | WREG32(UVD_VCPU_CACHE_SIZE0, size); | |
5867 | ||
5868 | addr += size; | |
5869 | size = RADEON_UVD_STACK_SIZE >> 3; | |
5870 | WREG32(UVD_VCPU_CACHE_OFFSET1, addr); | |
5871 | WREG32(UVD_VCPU_CACHE_SIZE1, size); | |
5872 | ||
5873 | addr += size; | |
5874 | size = RADEON_UVD_HEAP_SIZE >> 3; | |
5875 | WREG32(UVD_VCPU_CACHE_OFFSET2, addr); | |
5876 | WREG32(UVD_VCPU_CACHE_SIZE2, size); | |
5877 | ||
5878 | /* bits 28-31 */ | |
5879 | addr = (rdev->uvd.gpu_addr >> 28) & 0xF; | |
5880 | WREG32(UVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0)); | |
5881 | ||
5882 | /* bits 32-39 */ | |
5883 | addr = (rdev->uvd.gpu_addr >> 32) & 0xFF; | |
5884 | WREG32(UVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31)); | |
5885 | ||
5886 | return 0; | |
5887 | } |