Commit | Line | Data |
---|---|---|
771fe6b9 JG |
1 | /* |
2 | * Copyright 2008 Advanced Micro Devices, Inc. | |
3 | * Copyright 2008 Red Hat Inc. | |
4 | * Copyright 2009 Jerome Glisse. | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a | |
7 | * copy of this software and associated documentation files (the "Software"), | |
8 | * to deal in the Software without restriction, including without limitation | |
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
10 | * and/or sell copies of the Software, and to permit persons to whom the | |
11 | * Software is furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
22 | * OTHER DEALINGS IN THE SOFTWARE. | |
23 | * | |
24 | * Authors: Dave Airlie | |
25 | * Alex Deucher | |
26 | * Jerome Glisse | |
27 | */ | |
28 | #include "drmP.h" | |
771fe6b9 | 29 | #include "radeon.h" |
e6990375 | 30 | #include "radeon_asic.h" |
c93bb85b | 31 | #include "atom.h" |
3bc68535 | 32 | #include "rs690d.h" |
771fe6b9 | 33 | |
3bc68535 | 34 | static int rs690_mc_wait_for_idle(struct radeon_device *rdev) |
771fe6b9 JG |
35 | { |
36 | unsigned i; | |
37 | uint32_t tmp; | |
38 | ||
39 | for (i = 0; i < rdev->usec_timeout; i++) { | |
40 | /* read MC_STATUS */ | |
3bc68535 JG |
41 | tmp = RREG32_MC(R_000090_MC_SYSTEM_STATUS); |
42 | if (G_000090_MC_SYSTEM_IDLE(tmp)) | |
771fe6b9 | 43 | return 0; |
3bc68535 | 44 | udelay(1); |
771fe6b9 JG |
45 | } |
46 | return -1; | |
47 | } | |
48 | ||
3bc68535 | 49 | static void rs690_gpu_init(struct radeon_device *rdev) |
771fe6b9 JG |
50 | { |
51 | /* FIXME: HDP same place on rs690 ? */ | |
52 | r100_hdp_reset(rdev); | |
771fe6b9 JG |
53 | /* FIXME: is this correct ? */ |
54 | r420_pipes_init(rdev); | |
55 | if (rs690_mc_wait_for_idle(rdev)) { | |
56 | printk(KERN_WARNING "Failed to wait MC idle while " | |
57 | "programming pipes. Bad things might happen.\n"); | |
58 | } | |
59 | } | |
60 | ||
c93bb85b JG |
61 | void rs690_pm_info(struct radeon_device *rdev) |
62 | { | |
63 | int index = GetIndexIntoMasterTable(DATA, IntegratedSystemInfo); | |
64 | struct _ATOM_INTEGRATED_SYSTEM_INFO *info; | |
65 | struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 *info_v2; | |
66 | void *ptr; | |
67 | uint16_t data_offset; | |
68 | uint8_t frev, crev; | |
69 | fixed20_12 tmp; | |
70 | ||
71 | atom_parse_data_header(rdev->mode_info.atom_context, index, NULL, | |
72 | &frev, &crev, &data_offset); | |
73 | ptr = rdev->mode_info.atom_context->bios + data_offset; | |
74 | info = (struct _ATOM_INTEGRATED_SYSTEM_INFO *)ptr; | |
75 | info_v2 = (struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 *)ptr; | |
76 | /* Get various system informations from bios */ | |
77 | switch (crev) { | |
78 | case 1: | |
79 | tmp.full = rfixed_const(100); | |
80 | rdev->pm.igp_sideport_mclk.full = rfixed_const(info->ulBootUpMemoryClock); | |
81 | rdev->pm.igp_sideport_mclk.full = rfixed_div(rdev->pm.igp_sideport_mclk, tmp); | |
82 | rdev->pm.igp_system_mclk.full = rfixed_const(le16_to_cpu(info->usK8MemoryClock)); | |
83 | rdev->pm.igp_ht_link_clk.full = rfixed_const(le16_to_cpu(info->usFSBClock)); | |
84 | rdev->pm.igp_ht_link_width.full = rfixed_const(info->ucHTLinkWidth); | |
85 | break; | |
86 | case 2: | |
87 | tmp.full = rfixed_const(100); | |
88 | rdev->pm.igp_sideport_mclk.full = rfixed_const(info_v2->ulBootUpSidePortClock); | |
89 | rdev->pm.igp_sideport_mclk.full = rfixed_div(rdev->pm.igp_sideport_mclk, tmp); | |
90 | rdev->pm.igp_system_mclk.full = rfixed_const(info_v2->ulBootUpUMAClock); | |
91 | rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp); | |
92 | rdev->pm.igp_ht_link_clk.full = rfixed_const(info_v2->ulHTLinkFreq); | |
93 | rdev->pm.igp_ht_link_clk.full = rfixed_div(rdev->pm.igp_ht_link_clk, tmp); | |
94 | rdev->pm.igp_ht_link_width.full = rfixed_const(le16_to_cpu(info_v2->usMinHTLinkWidth)); | |
95 | break; | |
96 | default: | |
97 | tmp.full = rfixed_const(100); | |
98 | /* We assume the slower possible clock ie worst case */ | |
99 | /* DDR 333Mhz */ | |
100 | rdev->pm.igp_sideport_mclk.full = rfixed_const(333); | |
101 | /* FIXME: system clock ? */ | |
102 | rdev->pm.igp_system_mclk.full = rfixed_const(100); | |
103 | rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp); | |
104 | rdev->pm.igp_ht_link_clk.full = rfixed_const(200); | |
105 | rdev->pm.igp_ht_link_width.full = rfixed_const(8); | |
106 | DRM_ERROR("No integrated system info for your GPU, using safe default\n"); | |
107 | break; | |
108 | } | |
109 | /* Compute various bandwidth */ | |
110 | /* k8_bandwidth = (memory_clk / 2) * 2 * 8 * 0.5 = memory_clk * 4 */ | |
111 | tmp.full = rfixed_const(4); | |
112 | rdev->pm.k8_bandwidth.full = rfixed_mul(rdev->pm.igp_system_mclk, tmp); | |
113 | /* ht_bandwidth = ht_clk * 2 * ht_width / 8 * 0.8 | |
114 | * = ht_clk * ht_width / 5 | |
115 | */ | |
116 | tmp.full = rfixed_const(5); | |
117 | rdev->pm.ht_bandwidth.full = rfixed_mul(rdev->pm.igp_ht_link_clk, | |
118 | rdev->pm.igp_ht_link_width); | |
119 | rdev->pm.ht_bandwidth.full = rfixed_div(rdev->pm.ht_bandwidth, tmp); | |
120 | if (tmp.full < rdev->pm.max_bandwidth.full) { | |
121 | /* HT link is a limiting factor */ | |
122 | rdev->pm.max_bandwidth.full = tmp.full; | |
123 | } | |
124 | /* sideport_bandwidth = (sideport_clk / 2) * 2 * 2 * 0.7 | |
125 | * = (sideport_clk * 14) / 10 | |
126 | */ | |
127 | tmp.full = rfixed_const(14); | |
128 | rdev->pm.sideport_bandwidth.full = rfixed_mul(rdev->pm.igp_sideport_mclk, tmp); | |
129 | tmp.full = rfixed_const(10); | |
130 | rdev->pm.sideport_bandwidth.full = rfixed_div(rdev->pm.sideport_bandwidth, tmp); | |
131 | } | |
132 | ||
d594e46a | 133 | void rs690_mc_init(struct radeon_device *rdev) |
771fe6b9 | 134 | { |
d594e46a | 135 | u64 base; |
771fe6b9 JG |
136 | |
137 | rs400_gart_adjust_size(rdev); | |
771fe6b9 | 138 | rdev->mc.vram_is_ddr = true; |
722f2943 | 139 | rdev->mc.vram_width = 128; |
7a50f01a DA |
140 | rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE); |
141 | rdev->mc.mc_vram_size = rdev->mc.real_vram_size; | |
771fe6b9 JG |
142 | rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0); |
143 | rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0); | |
51e5fcd3 | 144 | rdev->mc.visible_vram_size = rdev->mc.aper_size; |
d594e46a JG |
145 | base = RREG32_MC(R_000100_MCCFG_FB_LOCATION); |
146 | base = G_000100_MC_FB_START(base) << 16; | |
c93bb85b | 147 | rs690_pm_info(rdev); |
06b6476d | 148 | rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev); |
d594e46a JG |
149 | radeon_vram_location(rdev, &rdev->mc, base); |
150 | radeon_gtt_location(rdev, &rdev->mc); | |
f47299c5 | 151 | radeon_update_bandwidth_info(rdev); |
22dd5013 AD |
152 | } |
153 | ||
c93bb85b JG |
154 | void rs690_line_buffer_adjust(struct radeon_device *rdev, |
155 | struct drm_display_mode *mode1, | |
156 | struct drm_display_mode *mode2) | |
157 | { | |
158 | u32 tmp; | |
159 | ||
160 | /* | |
161 | * Line Buffer Setup | |
162 | * There is a single line buffer shared by both display controllers. | |
3bc68535 | 163 | * R_006520_DC_LB_MEMORY_SPLIT controls how that line buffer is shared between |
c93bb85b JG |
164 | * the display controllers. The paritioning can either be done |
165 | * manually or via one of four preset allocations specified in bits 1:0: | |
166 | * 0 - line buffer is divided in half and shared between crtc | |
167 | * 1 - D1 gets 3/4 of the line buffer, D2 gets 1/4 | |
168 | * 2 - D1 gets the whole buffer | |
169 | * 3 - D1 gets 1/4 of the line buffer, D2 gets 3/4 | |
3bc68535 | 170 | * Setting bit 2 of R_006520_DC_LB_MEMORY_SPLIT controls switches to manual |
c93bb85b JG |
171 | * allocation mode. In manual allocation mode, D1 always starts at 0, |
172 | * D1 end/2 is specified in bits 14:4; D2 allocation follows D1. | |
173 | */ | |
3bc68535 JG |
174 | tmp = RREG32(R_006520_DC_LB_MEMORY_SPLIT) & C_006520_DC_LB_MEMORY_SPLIT; |
175 | tmp &= ~C_006520_DC_LB_MEMORY_SPLIT_MODE; | |
c93bb85b JG |
176 | /* auto */ |
177 | if (mode1 && mode2) { | |
178 | if (mode1->hdisplay > mode2->hdisplay) { | |
179 | if (mode1->hdisplay > 2560) | |
3bc68535 | 180 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q; |
c93bb85b | 181 | else |
3bc68535 | 182 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF; |
c93bb85b JG |
183 | } else if (mode2->hdisplay > mode1->hdisplay) { |
184 | if (mode2->hdisplay > 2560) | |
3bc68535 | 185 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q; |
c93bb85b | 186 | else |
3bc68535 | 187 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF; |
c93bb85b | 188 | } else |
3bc68535 | 189 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF; |
c93bb85b | 190 | } else if (mode1) { |
3bc68535 | 191 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_ONLY; |
c93bb85b | 192 | } else if (mode2) { |
3bc68535 | 193 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q; |
c93bb85b | 194 | } |
3bc68535 | 195 | WREG32(R_006520_DC_LB_MEMORY_SPLIT, tmp); |
771fe6b9 JG |
196 | } |
197 | ||
c93bb85b JG |
198 | struct rs690_watermark { |
199 | u32 lb_request_fifo_depth; | |
200 | fixed20_12 num_line_pair; | |
201 | fixed20_12 estimated_width; | |
202 | fixed20_12 worst_case_latency; | |
203 | fixed20_12 consumption_rate; | |
204 | fixed20_12 active_time; | |
205 | fixed20_12 dbpp; | |
206 | fixed20_12 priority_mark_max; | |
207 | fixed20_12 priority_mark; | |
208 | fixed20_12 sclk; | |
209 | }; | |
210 | ||
211 | void rs690_crtc_bandwidth_compute(struct radeon_device *rdev, | |
212 | struct radeon_crtc *crtc, | |
213 | struct rs690_watermark *wm) | |
214 | { | |
215 | struct drm_display_mode *mode = &crtc->base.mode; | |
216 | fixed20_12 a, b, c; | |
217 | fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width; | |
218 | fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency; | |
219 | /* FIXME: detect IGP with sideport memory, i don't think there is any | |
220 | * such product available | |
221 | */ | |
222 | bool sideport = false; | |
223 | ||
224 | if (!crtc->base.enabled) { | |
225 | /* FIXME: wouldn't it better to set priority mark to maximum */ | |
226 | wm->lb_request_fifo_depth = 4; | |
227 | return; | |
228 | } | |
229 | ||
230 | if (crtc->vsc.full > rfixed_const(2)) | |
231 | wm->num_line_pair.full = rfixed_const(2); | |
232 | else | |
233 | wm->num_line_pair.full = rfixed_const(1); | |
234 | ||
235 | b.full = rfixed_const(mode->crtc_hdisplay); | |
236 | c.full = rfixed_const(256); | |
69b3b5e5 AD |
237 | a.full = rfixed_div(b, c); |
238 | request_fifo_depth.full = rfixed_mul(a, wm->num_line_pair); | |
239 | request_fifo_depth.full = rfixed_ceil(request_fifo_depth); | |
c93bb85b JG |
240 | if (a.full < rfixed_const(4)) { |
241 | wm->lb_request_fifo_depth = 4; | |
242 | } else { | |
243 | wm->lb_request_fifo_depth = rfixed_trunc(request_fifo_depth); | |
244 | } | |
245 | ||
246 | /* Determine consumption rate | |
247 | * pclk = pixel clock period(ns) = 1000 / (mode.clock / 1000) | |
248 | * vtaps = number of vertical taps, | |
249 | * vsc = vertical scaling ratio, defined as source/destination | |
250 | * hsc = horizontal scaling ration, defined as source/destination | |
251 | */ | |
252 | a.full = rfixed_const(mode->clock); | |
253 | b.full = rfixed_const(1000); | |
254 | a.full = rfixed_div(a, b); | |
255 | pclk.full = rfixed_div(b, a); | |
256 | if (crtc->rmx_type != RMX_OFF) { | |
257 | b.full = rfixed_const(2); | |
258 | if (crtc->vsc.full > b.full) | |
259 | b.full = crtc->vsc.full; | |
260 | b.full = rfixed_mul(b, crtc->hsc); | |
261 | c.full = rfixed_const(2); | |
262 | b.full = rfixed_div(b, c); | |
263 | consumption_time.full = rfixed_div(pclk, b); | |
264 | } else { | |
265 | consumption_time.full = pclk.full; | |
266 | } | |
267 | a.full = rfixed_const(1); | |
268 | wm->consumption_rate.full = rfixed_div(a, consumption_time); | |
269 | ||
270 | ||
271 | /* Determine line time | |
272 | * LineTime = total time for one line of displayhtotal | |
273 | * LineTime = total number of horizontal pixels | |
274 | * pclk = pixel clock period(ns) | |
275 | */ | |
276 | a.full = rfixed_const(crtc->base.mode.crtc_htotal); | |
277 | line_time.full = rfixed_mul(a, pclk); | |
278 | ||
279 | /* Determine active time | |
280 | * ActiveTime = time of active region of display within one line, | |
281 | * hactive = total number of horizontal active pixels | |
282 | * htotal = total number of horizontal pixels | |
283 | */ | |
284 | a.full = rfixed_const(crtc->base.mode.crtc_htotal); | |
285 | b.full = rfixed_const(crtc->base.mode.crtc_hdisplay); | |
286 | wm->active_time.full = rfixed_mul(line_time, b); | |
287 | wm->active_time.full = rfixed_div(wm->active_time, a); | |
288 | ||
289 | /* Maximun bandwidth is the minimun bandwidth of all component */ | |
290 | rdev->pm.max_bandwidth = rdev->pm.core_bandwidth; | |
291 | if (sideport) { | |
292 | if (rdev->pm.max_bandwidth.full > rdev->pm.sideport_bandwidth.full && | |
293 | rdev->pm.sideport_bandwidth.full) | |
294 | rdev->pm.max_bandwidth = rdev->pm.sideport_bandwidth; | |
295 | read_delay_latency.full = rfixed_const(370 * 800 * 1000); | |
296 | read_delay_latency.full = rfixed_div(read_delay_latency, | |
297 | rdev->pm.igp_sideport_mclk); | |
298 | } else { | |
299 | if (rdev->pm.max_bandwidth.full > rdev->pm.k8_bandwidth.full && | |
300 | rdev->pm.k8_bandwidth.full) | |
301 | rdev->pm.max_bandwidth = rdev->pm.k8_bandwidth; | |
302 | if (rdev->pm.max_bandwidth.full > rdev->pm.ht_bandwidth.full && | |
303 | rdev->pm.ht_bandwidth.full) | |
304 | rdev->pm.max_bandwidth = rdev->pm.ht_bandwidth; | |
305 | read_delay_latency.full = rfixed_const(5000); | |
306 | } | |
307 | ||
308 | /* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */ | |
309 | a.full = rfixed_const(16); | |
310 | rdev->pm.sclk.full = rfixed_mul(rdev->pm.max_bandwidth, a); | |
311 | a.full = rfixed_const(1000); | |
312 | rdev->pm.sclk.full = rfixed_div(a, rdev->pm.sclk); | |
313 | /* Determine chunk time | |
314 | * ChunkTime = the time it takes the DCP to send one chunk of data | |
315 | * to the LB which consists of pipeline delay and inter chunk gap | |
316 | * sclk = system clock(ns) | |
317 | */ | |
318 | a.full = rfixed_const(256 * 13); | |
319 | chunk_time.full = rfixed_mul(rdev->pm.sclk, a); | |
320 | a.full = rfixed_const(10); | |
321 | chunk_time.full = rfixed_div(chunk_time, a); | |
322 | ||
323 | /* Determine the worst case latency | |
324 | * NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines) | |
325 | * WorstCaseLatency = worst case time from urgent to when the MC starts | |
326 | * to return data | |
327 | * READ_DELAY_IDLE_MAX = constant of 1us | |
328 | * ChunkTime = time it takes the DCP to send one chunk of data to the LB | |
329 | * which consists of pipeline delay and inter chunk gap | |
330 | */ | |
331 | if (rfixed_trunc(wm->num_line_pair) > 1) { | |
332 | a.full = rfixed_const(3); | |
333 | wm->worst_case_latency.full = rfixed_mul(a, chunk_time); | |
334 | wm->worst_case_latency.full += read_delay_latency.full; | |
335 | } else { | |
336 | a.full = rfixed_const(2); | |
337 | wm->worst_case_latency.full = rfixed_mul(a, chunk_time); | |
338 | wm->worst_case_latency.full += read_delay_latency.full; | |
339 | } | |
340 | ||
341 | /* Determine the tolerable latency | |
342 | * TolerableLatency = Any given request has only 1 line time | |
343 | * for the data to be returned | |
344 | * LBRequestFifoDepth = Number of chunk requests the LB can | |
345 | * put into the request FIFO for a display | |
346 | * LineTime = total time for one line of display | |
347 | * ChunkTime = the time it takes the DCP to send one chunk | |
348 | * of data to the LB which consists of | |
349 | * pipeline delay and inter chunk gap | |
350 | */ | |
351 | if ((2+wm->lb_request_fifo_depth) >= rfixed_trunc(request_fifo_depth)) { | |
352 | tolerable_latency.full = line_time.full; | |
353 | } else { | |
354 | tolerable_latency.full = rfixed_const(wm->lb_request_fifo_depth - 2); | |
355 | tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full; | |
356 | tolerable_latency.full = rfixed_mul(tolerable_latency, chunk_time); | |
357 | tolerable_latency.full = line_time.full - tolerable_latency.full; | |
358 | } | |
359 | /* We assume worst case 32bits (4 bytes) */ | |
360 | wm->dbpp.full = rfixed_const(4 * 8); | |
361 | ||
362 | /* Determine the maximum priority mark | |
363 | * width = viewport width in pixels | |
364 | */ | |
365 | a.full = rfixed_const(16); | |
366 | wm->priority_mark_max.full = rfixed_const(crtc->base.mode.crtc_hdisplay); | |
367 | wm->priority_mark_max.full = rfixed_div(wm->priority_mark_max, a); | |
69b3b5e5 | 368 | wm->priority_mark_max.full = rfixed_ceil(wm->priority_mark_max); |
c93bb85b JG |
369 | |
370 | /* Determine estimated width */ | |
371 | estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full; | |
372 | estimated_width.full = rfixed_div(estimated_width, consumption_time); | |
373 | if (rfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) { | |
374 | wm->priority_mark.full = rfixed_const(10); | |
375 | } else { | |
376 | a.full = rfixed_const(16); | |
377 | wm->priority_mark.full = rfixed_div(estimated_width, a); | |
69b3b5e5 | 378 | wm->priority_mark.full = rfixed_ceil(wm->priority_mark); |
c93bb85b JG |
379 | wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full; |
380 | } | |
381 | } | |
382 | ||
383 | void rs690_bandwidth_update(struct radeon_device *rdev) | |
384 | { | |
385 | struct drm_display_mode *mode0 = NULL; | |
386 | struct drm_display_mode *mode1 = NULL; | |
387 | struct rs690_watermark wm0; | |
388 | struct rs690_watermark wm1; | |
389 | u32 tmp; | |
390 | fixed20_12 priority_mark02, priority_mark12, fill_rate; | |
391 | fixed20_12 a, b; | |
392 | ||
393 | if (rdev->mode_info.crtcs[0]->base.enabled) | |
394 | mode0 = &rdev->mode_info.crtcs[0]->base.mode; | |
395 | if (rdev->mode_info.crtcs[1]->base.enabled) | |
396 | mode1 = &rdev->mode_info.crtcs[1]->base.mode; | |
397 | /* | |
398 | * Set display0/1 priority up in the memory controller for | |
399 | * modes if the user specifies HIGH for displaypriority | |
400 | * option. | |
401 | */ | |
402 | if (rdev->disp_priority == 2) { | |
3bc68535 JG |
403 | tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER); |
404 | tmp &= C_000104_MC_DISP0R_INIT_LAT; | |
405 | tmp &= C_000104_MC_DISP1R_INIT_LAT; | |
c93bb85b | 406 | if (mode0) |
3bc68535 JG |
407 | tmp |= S_000104_MC_DISP0R_INIT_LAT(1); |
408 | if (mode1) | |
409 | tmp |= S_000104_MC_DISP1R_INIT_LAT(1); | |
410 | WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp); | |
c93bb85b JG |
411 | } |
412 | rs690_line_buffer_adjust(rdev, mode0, mode1); | |
413 | ||
414 | if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740)) | |
3bc68535 | 415 | WREG32(R_006C9C_DCP_CONTROL, 0); |
c93bb85b | 416 | if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880)) |
3bc68535 | 417 | WREG32(R_006C9C_DCP_CONTROL, 2); |
c93bb85b JG |
418 | |
419 | rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0); | |
420 | rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1); | |
421 | ||
422 | tmp = (wm0.lb_request_fifo_depth - 1); | |
423 | tmp |= (wm1.lb_request_fifo_depth - 1) << 16; | |
3bc68535 | 424 | WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp); |
c93bb85b JG |
425 | |
426 | if (mode0 && mode1) { | |
427 | if (rfixed_trunc(wm0.dbpp) > 64) | |
428 | a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair); | |
429 | else | |
430 | a.full = wm0.num_line_pair.full; | |
431 | if (rfixed_trunc(wm1.dbpp) > 64) | |
432 | b.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair); | |
433 | else | |
434 | b.full = wm1.num_line_pair.full; | |
435 | a.full += b.full; | |
436 | fill_rate.full = rfixed_div(wm0.sclk, a); | |
437 | if (wm0.consumption_rate.full > fill_rate.full) { | |
438 | b.full = wm0.consumption_rate.full - fill_rate.full; | |
439 | b.full = rfixed_mul(b, wm0.active_time); | |
440 | a.full = rfixed_mul(wm0.worst_case_latency, | |
441 | wm0.consumption_rate); | |
442 | a.full = a.full + b.full; | |
443 | b.full = rfixed_const(16 * 1000); | |
444 | priority_mark02.full = rfixed_div(a, b); | |
445 | } else { | |
446 | a.full = rfixed_mul(wm0.worst_case_latency, | |
447 | wm0.consumption_rate); | |
448 | b.full = rfixed_const(16 * 1000); | |
449 | priority_mark02.full = rfixed_div(a, b); | |
450 | } | |
451 | if (wm1.consumption_rate.full > fill_rate.full) { | |
452 | b.full = wm1.consumption_rate.full - fill_rate.full; | |
453 | b.full = rfixed_mul(b, wm1.active_time); | |
454 | a.full = rfixed_mul(wm1.worst_case_latency, | |
455 | wm1.consumption_rate); | |
456 | a.full = a.full + b.full; | |
457 | b.full = rfixed_const(16 * 1000); | |
458 | priority_mark12.full = rfixed_div(a, b); | |
459 | } else { | |
460 | a.full = rfixed_mul(wm1.worst_case_latency, | |
461 | wm1.consumption_rate); | |
462 | b.full = rfixed_const(16 * 1000); | |
463 | priority_mark12.full = rfixed_div(a, b); | |
464 | } | |
465 | if (wm0.priority_mark.full > priority_mark02.full) | |
466 | priority_mark02.full = wm0.priority_mark.full; | |
467 | if (rfixed_trunc(priority_mark02) < 0) | |
468 | priority_mark02.full = 0; | |
469 | if (wm0.priority_mark_max.full > priority_mark02.full) | |
470 | priority_mark02.full = wm0.priority_mark_max.full; | |
471 | if (wm1.priority_mark.full > priority_mark12.full) | |
472 | priority_mark12.full = wm1.priority_mark.full; | |
473 | if (rfixed_trunc(priority_mark12) < 0) | |
474 | priority_mark12.full = 0; | |
475 | if (wm1.priority_mark_max.full > priority_mark12.full) | |
476 | priority_mark12.full = wm1.priority_mark_max.full; | |
3bc68535 JG |
477 | WREG32(R_006548_D1MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark02)); |
478 | WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark02)); | |
479 | WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark12)); | |
480 | WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark12)); | |
c93bb85b JG |
481 | } else if (mode0) { |
482 | if (rfixed_trunc(wm0.dbpp) > 64) | |
483 | a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair); | |
484 | else | |
485 | a.full = wm0.num_line_pair.full; | |
486 | fill_rate.full = rfixed_div(wm0.sclk, a); | |
487 | if (wm0.consumption_rate.full > fill_rate.full) { | |
488 | b.full = wm0.consumption_rate.full - fill_rate.full; | |
489 | b.full = rfixed_mul(b, wm0.active_time); | |
490 | a.full = rfixed_mul(wm0.worst_case_latency, | |
491 | wm0.consumption_rate); | |
492 | a.full = a.full + b.full; | |
493 | b.full = rfixed_const(16 * 1000); | |
494 | priority_mark02.full = rfixed_div(a, b); | |
495 | } else { | |
496 | a.full = rfixed_mul(wm0.worst_case_latency, | |
497 | wm0.consumption_rate); | |
498 | b.full = rfixed_const(16 * 1000); | |
499 | priority_mark02.full = rfixed_div(a, b); | |
500 | } | |
501 | if (wm0.priority_mark.full > priority_mark02.full) | |
502 | priority_mark02.full = wm0.priority_mark.full; | |
503 | if (rfixed_trunc(priority_mark02) < 0) | |
504 | priority_mark02.full = 0; | |
505 | if (wm0.priority_mark_max.full > priority_mark02.full) | |
506 | priority_mark02.full = wm0.priority_mark_max.full; | |
3bc68535 JG |
507 | WREG32(R_006548_D1MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark02)); |
508 | WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark02)); | |
509 | WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, | |
510 | S_006D48_D2MODE_PRIORITY_A_OFF(1)); | |
511 | WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, | |
512 | S_006D4C_D2MODE_PRIORITY_B_OFF(1)); | |
c93bb85b JG |
513 | } else { |
514 | if (rfixed_trunc(wm1.dbpp) > 64) | |
515 | a.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair); | |
516 | else | |
517 | a.full = wm1.num_line_pair.full; | |
518 | fill_rate.full = rfixed_div(wm1.sclk, a); | |
519 | if (wm1.consumption_rate.full > fill_rate.full) { | |
520 | b.full = wm1.consumption_rate.full - fill_rate.full; | |
521 | b.full = rfixed_mul(b, wm1.active_time); | |
522 | a.full = rfixed_mul(wm1.worst_case_latency, | |
523 | wm1.consumption_rate); | |
524 | a.full = a.full + b.full; | |
525 | b.full = rfixed_const(16 * 1000); | |
526 | priority_mark12.full = rfixed_div(a, b); | |
527 | } else { | |
528 | a.full = rfixed_mul(wm1.worst_case_latency, | |
529 | wm1.consumption_rate); | |
530 | b.full = rfixed_const(16 * 1000); | |
531 | priority_mark12.full = rfixed_div(a, b); | |
532 | } | |
533 | if (wm1.priority_mark.full > priority_mark12.full) | |
534 | priority_mark12.full = wm1.priority_mark.full; | |
535 | if (rfixed_trunc(priority_mark12) < 0) | |
536 | priority_mark12.full = 0; | |
537 | if (wm1.priority_mark_max.full > priority_mark12.full) | |
538 | priority_mark12.full = wm1.priority_mark_max.full; | |
3bc68535 JG |
539 | WREG32(R_006548_D1MODE_PRIORITY_A_CNT, |
540 | S_006548_D1MODE_PRIORITY_A_OFF(1)); | |
541 | WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, | |
542 | S_00654C_D1MODE_PRIORITY_B_OFF(1)); | |
543 | WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark12)); | |
544 | WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark12)); | |
c93bb85b JG |
545 | } |
546 | } | |
771fe6b9 | 547 | |
771fe6b9 JG |
548 | uint32_t rs690_mc_rreg(struct radeon_device *rdev, uint32_t reg) |
549 | { | |
550 | uint32_t r; | |
551 | ||
3bc68535 JG |
552 | WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg)); |
553 | r = RREG32(R_00007C_MC_DATA); | |
554 | WREG32(R_000078_MC_INDEX, ~C_000078_MC_IND_ADDR); | |
771fe6b9 JG |
555 | return r; |
556 | } | |
557 | ||
558 | void rs690_mc_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v) | |
559 | { | |
3bc68535 JG |
560 | WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg) | |
561 | S_000078_MC_IND_WR_EN(1)); | |
562 | WREG32(R_00007C_MC_DATA, v); | |
563 | WREG32(R_000078_MC_INDEX, 0x7F); | |
564 | } | |
565 | ||
566 | void rs690_mc_program(struct radeon_device *rdev) | |
567 | { | |
568 | struct rv515_mc_save save; | |
569 | ||
570 | /* Stops all mc clients */ | |
571 | rv515_mc_stop(rdev, &save); | |
572 | ||
573 | /* Wait for mc idle */ | |
574 | if (rs690_mc_wait_for_idle(rdev)) | |
575 | dev_warn(rdev->dev, "Wait MC idle timeout before updating MC.\n"); | |
576 | /* Program MC, should be a 32bits limited address space */ | |
577 | WREG32_MC(R_000100_MCCFG_FB_LOCATION, | |
578 | S_000100_MC_FB_START(rdev->mc.vram_start >> 16) | | |
579 | S_000100_MC_FB_TOP(rdev->mc.vram_end >> 16)); | |
580 | WREG32(R_000134_HDP_FB_LOCATION, | |
581 | S_000134_HDP_FB_START(rdev->mc.vram_start >> 16)); | |
582 | ||
583 | rv515_mc_resume(rdev, &save); | |
584 | } | |
585 | ||
586 | static int rs690_startup(struct radeon_device *rdev) | |
587 | { | |
588 | int r; | |
589 | ||
590 | rs690_mc_program(rdev); | |
591 | /* Resume clock */ | |
592 | rv515_clock_startup(rdev); | |
593 | /* Initialize GPU configuration (# pipes, ...) */ | |
594 | rs690_gpu_init(rdev); | |
595 | /* Initialize GART (initialize after TTM so we can allocate | |
596 | * memory through TTM but finalize after TTM) */ | |
597 | r = rs400_gart_enable(rdev); | |
598 | if (r) | |
599 | return r; | |
600 | /* Enable IRQ */ | |
ac447df4 | 601 | rs600_irq_set(rdev); |
cafe6609 | 602 | rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL); |
3bc68535 JG |
603 | /* 1M ring buffer */ |
604 | r = r100_cp_init(rdev, 1024 * 1024); | |
605 | if (r) { | |
606 | dev_err(rdev->dev, "failled initializing CP (%d).\n", r); | |
607 | return r; | |
608 | } | |
609 | r = r100_wb_init(rdev); | |
610 | if (r) | |
611 | dev_err(rdev->dev, "failled initializing WB (%d).\n", r); | |
612 | r = r100_ib_init(rdev); | |
613 | if (r) { | |
614 | dev_err(rdev->dev, "failled initializing IB (%d).\n", r); | |
615 | return r; | |
616 | } | |
617 | return 0; | |
618 | } | |
619 | ||
620 | int rs690_resume(struct radeon_device *rdev) | |
621 | { | |
622 | /* Make sur GART are not working */ | |
623 | rs400_gart_disable(rdev); | |
624 | /* Resume clock before doing reset */ | |
625 | rv515_clock_startup(rdev); | |
626 | /* Reset gpu before posting otherwise ATOM will enter infinite loop */ | |
627 | if (radeon_gpu_reset(rdev)) { | |
628 | dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n", | |
629 | RREG32(R_000E40_RBBM_STATUS), | |
630 | RREG32(R_0007C0_CP_STAT)); | |
631 | } | |
632 | /* post */ | |
633 | atom_asic_init(rdev->mode_info.atom_context); | |
634 | /* Resume clock after posting */ | |
635 | rv515_clock_startup(rdev); | |
550e2d92 DA |
636 | /* Initialize surface registers */ |
637 | radeon_surface_init(rdev); | |
3bc68535 JG |
638 | return rs690_startup(rdev); |
639 | } | |
640 | ||
641 | int rs690_suspend(struct radeon_device *rdev) | |
642 | { | |
643 | r100_cp_disable(rdev); | |
644 | r100_wb_disable(rdev); | |
ac447df4 | 645 | rs600_irq_disable(rdev); |
3bc68535 JG |
646 | rs400_gart_disable(rdev); |
647 | return 0; | |
648 | } | |
649 | ||
650 | void rs690_fini(struct radeon_device *rdev) | |
651 | { | |
29fb52ca | 652 | radeon_pm_fini(rdev); |
3bc68535 JG |
653 | r100_cp_fini(rdev); |
654 | r100_wb_fini(rdev); | |
655 | r100_ib_fini(rdev); | |
656 | radeon_gem_fini(rdev); | |
657 | rs400_gart_fini(rdev); | |
658 | radeon_irq_kms_fini(rdev); | |
659 | radeon_fence_driver_fini(rdev); | |
4c788679 | 660 | radeon_bo_fini(rdev); |
3bc68535 JG |
661 | radeon_atombios_fini(rdev); |
662 | kfree(rdev->bios); | |
663 | rdev->bios = NULL; | |
664 | } | |
665 | ||
666 | int rs690_init(struct radeon_device *rdev) | |
667 | { | |
668 | int r; | |
669 | ||
3bc68535 JG |
670 | /* Disable VGA */ |
671 | rv515_vga_render_disable(rdev); | |
672 | /* Initialize scratch registers */ | |
673 | radeon_scratch_init(rdev); | |
674 | /* Initialize surface registers */ | |
675 | radeon_surface_init(rdev); | |
676 | /* TODO: disable VGA need to use VGA request */ | |
677 | /* BIOS*/ | |
678 | if (!radeon_get_bios(rdev)) { | |
679 | if (ASIC_IS_AVIVO(rdev)) | |
680 | return -EINVAL; | |
681 | } | |
682 | if (rdev->is_atom_bios) { | |
683 | r = radeon_atombios_init(rdev); | |
684 | if (r) | |
685 | return r; | |
686 | } else { | |
687 | dev_err(rdev->dev, "Expecting atombios for RV515 GPU\n"); | |
688 | return -EINVAL; | |
689 | } | |
690 | /* Reset gpu before posting otherwise ATOM will enter infinite loop */ | |
691 | if (radeon_gpu_reset(rdev)) { | |
692 | dev_warn(rdev->dev, | |
693 | "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n", | |
694 | RREG32(R_000E40_RBBM_STATUS), | |
695 | RREG32(R_0007C0_CP_STAT)); | |
696 | } | |
697 | /* check if cards are posted or not */ | |
72542d77 DA |
698 | if (radeon_boot_test_post_card(rdev) == false) |
699 | return -EINVAL; | |
700 | ||
3bc68535 JG |
701 | /* Initialize clocks */ |
702 | radeon_get_clock_info(rdev->ddev); | |
7433874e RM |
703 | /* Initialize power management */ |
704 | radeon_pm_init(rdev); | |
d594e46a JG |
705 | /* initialize memory controller */ |
706 | rs690_mc_init(rdev); | |
3bc68535 JG |
707 | rv515_debugfs(rdev); |
708 | /* Fence driver */ | |
709 | r = radeon_fence_driver_init(rdev); | |
710 | if (r) | |
711 | return r; | |
712 | r = radeon_irq_kms_init(rdev); | |
713 | if (r) | |
714 | return r; | |
715 | /* Memory manager */ | |
4c788679 | 716 | r = radeon_bo_init(rdev); |
3bc68535 JG |
717 | if (r) |
718 | return r; | |
719 | r = rs400_gart_init(rdev); | |
720 | if (r) | |
721 | return r; | |
722 | rs600_set_safe_registers(rdev); | |
723 | rdev->accel_working = true; | |
724 | r = rs690_startup(rdev); | |
725 | if (r) { | |
726 | /* Somethings want wront with the accel init stop accel */ | |
727 | dev_err(rdev->dev, "Disabling GPU acceleration\n"); | |
3bc68535 JG |
728 | r100_cp_fini(rdev); |
729 | r100_wb_fini(rdev); | |
730 | r100_ib_fini(rdev); | |
731 | rs400_gart_fini(rdev); | |
732 | radeon_irq_kms_fini(rdev); | |
733 | rdev->accel_working = false; | |
734 | } | |
735 | return 0; | |
771fe6b9 | 736 | } |