]>
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" |
c93bb85b | 30 | #include "atom.h" |
3bc68535 | 31 | #include "rs690d.h" |
771fe6b9 | 32 | |
3bc68535 | 33 | static int rs690_mc_wait_for_idle(struct radeon_device *rdev) |
771fe6b9 JG |
34 | { |
35 | unsigned i; | |
36 | uint32_t tmp; | |
37 | ||
38 | for (i = 0; i < rdev->usec_timeout; i++) { | |
39 | /* read MC_STATUS */ | |
3bc68535 JG |
40 | tmp = RREG32_MC(R_000090_MC_SYSTEM_STATUS); |
41 | if (G_000090_MC_SYSTEM_IDLE(tmp)) | |
771fe6b9 | 42 | return 0; |
3bc68535 | 43 | udelay(1); |
771fe6b9 JG |
44 | } |
45 | return -1; | |
46 | } | |
47 | ||
3bc68535 | 48 | static void rs690_gpu_init(struct radeon_device *rdev) |
771fe6b9 JG |
49 | { |
50 | /* FIXME: HDP same place on rs690 ? */ | |
51 | r100_hdp_reset(rdev); | |
771fe6b9 JG |
52 | /* FIXME: is this correct ? */ |
53 | r420_pipes_init(rdev); | |
54 | if (rs690_mc_wait_for_idle(rdev)) { | |
55 | printk(KERN_WARNING "Failed to wait MC idle while " | |
56 | "programming pipes. Bad things might happen.\n"); | |
57 | } | |
58 | } | |
59 | ||
c93bb85b JG |
60 | void rs690_pm_info(struct radeon_device *rdev) |
61 | { | |
62 | int index = GetIndexIntoMasterTable(DATA, IntegratedSystemInfo); | |
63 | struct _ATOM_INTEGRATED_SYSTEM_INFO *info; | |
64 | struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 *info_v2; | |
65 | void *ptr; | |
66 | uint16_t data_offset; | |
67 | uint8_t frev, crev; | |
68 | fixed20_12 tmp; | |
69 | ||
70 | atom_parse_data_header(rdev->mode_info.atom_context, index, NULL, | |
71 | &frev, &crev, &data_offset); | |
72 | ptr = rdev->mode_info.atom_context->bios + data_offset; | |
73 | info = (struct _ATOM_INTEGRATED_SYSTEM_INFO *)ptr; | |
74 | info_v2 = (struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 *)ptr; | |
75 | /* Get various system informations from bios */ | |
76 | switch (crev) { | |
77 | case 1: | |
78 | tmp.full = rfixed_const(100); | |
79 | rdev->pm.igp_sideport_mclk.full = rfixed_const(info->ulBootUpMemoryClock); | |
80 | rdev->pm.igp_sideport_mclk.full = rfixed_div(rdev->pm.igp_sideport_mclk, tmp); | |
81 | rdev->pm.igp_system_mclk.full = rfixed_const(le16_to_cpu(info->usK8MemoryClock)); | |
82 | rdev->pm.igp_ht_link_clk.full = rfixed_const(le16_to_cpu(info->usFSBClock)); | |
83 | rdev->pm.igp_ht_link_width.full = rfixed_const(info->ucHTLinkWidth); | |
84 | break; | |
85 | case 2: | |
86 | tmp.full = rfixed_const(100); | |
87 | rdev->pm.igp_sideport_mclk.full = rfixed_const(info_v2->ulBootUpSidePortClock); | |
88 | rdev->pm.igp_sideport_mclk.full = rfixed_div(rdev->pm.igp_sideport_mclk, tmp); | |
89 | rdev->pm.igp_system_mclk.full = rfixed_const(info_v2->ulBootUpUMAClock); | |
90 | rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp); | |
91 | rdev->pm.igp_ht_link_clk.full = rfixed_const(info_v2->ulHTLinkFreq); | |
92 | rdev->pm.igp_ht_link_clk.full = rfixed_div(rdev->pm.igp_ht_link_clk, tmp); | |
93 | rdev->pm.igp_ht_link_width.full = rfixed_const(le16_to_cpu(info_v2->usMinHTLinkWidth)); | |
94 | break; | |
95 | default: | |
96 | tmp.full = rfixed_const(100); | |
97 | /* We assume the slower possible clock ie worst case */ | |
98 | /* DDR 333Mhz */ | |
99 | rdev->pm.igp_sideport_mclk.full = rfixed_const(333); | |
100 | /* FIXME: system clock ? */ | |
101 | rdev->pm.igp_system_mclk.full = rfixed_const(100); | |
102 | rdev->pm.igp_system_mclk.full = rfixed_div(rdev->pm.igp_system_mclk, tmp); | |
103 | rdev->pm.igp_ht_link_clk.full = rfixed_const(200); | |
104 | rdev->pm.igp_ht_link_width.full = rfixed_const(8); | |
105 | DRM_ERROR("No integrated system info for your GPU, using safe default\n"); | |
106 | break; | |
107 | } | |
108 | /* Compute various bandwidth */ | |
109 | /* k8_bandwidth = (memory_clk / 2) * 2 * 8 * 0.5 = memory_clk * 4 */ | |
110 | tmp.full = rfixed_const(4); | |
111 | rdev->pm.k8_bandwidth.full = rfixed_mul(rdev->pm.igp_system_mclk, tmp); | |
112 | /* ht_bandwidth = ht_clk * 2 * ht_width / 8 * 0.8 | |
113 | * = ht_clk * ht_width / 5 | |
114 | */ | |
115 | tmp.full = rfixed_const(5); | |
116 | rdev->pm.ht_bandwidth.full = rfixed_mul(rdev->pm.igp_ht_link_clk, | |
117 | rdev->pm.igp_ht_link_width); | |
118 | rdev->pm.ht_bandwidth.full = rfixed_div(rdev->pm.ht_bandwidth, tmp); | |
119 | if (tmp.full < rdev->pm.max_bandwidth.full) { | |
120 | /* HT link is a limiting factor */ | |
121 | rdev->pm.max_bandwidth.full = tmp.full; | |
122 | } | |
123 | /* sideport_bandwidth = (sideport_clk / 2) * 2 * 2 * 0.7 | |
124 | * = (sideport_clk * 14) / 10 | |
125 | */ | |
126 | tmp.full = rfixed_const(14); | |
127 | rdev->pm.sideport_bandwidth.full = rfixed_mul(rdev->pm.igp_sideport_mclk, tmp); | |
128 | tmp.full = rfixed_const(10); | |
129 | rdev->pm.sideport_bandwidth.full = rfixed_div(rdev->pm.sideport_bandwidth, tmp); | |
130 | } | |
131 | ||
771fe6b9 JG |
132 | void rs690_vram_info(struct radeon_device *rdev) |
133 | { | |
c93bb85b | 134 | fixed20_12 a; |
771fe6b9 JG |
135 | |
136 | rs400_gart_adjust_size(rdev); | |
0088dbdb | 137 | |
771fe6b9 | 138 | rdev->mc.vram_is_ddr = true; |
722f2943 AD |
139 | rdev->mc.vram_width = 128; |
140 | ||
7a50f01a DA |
141 | rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE); |
142 | rdev->mc.mc_vram_size = rdev->mc.real_vram_size; | |
771fe6b9 JG |
143 | |
144 | rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0); | |
145 | rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0); | |
0088dbdb AD |
146 | |
147 | if (rdev->mc.mc_vram_size > rdev->mc.aper_size) | |
148 | rdev->mc.mc_vram_size = rdev->mc.aper_size; | |
149 | ||
150 | if (rdev->mc.real_vram_size > rdev->mc.aper_size) | |
151 | rdev->mc.real_vram_size = rdev->mc.aper_size; | |
152 | ||
c93bb85b JG |
153 | rs690_pm_info(rdev); |
154 | /* FIXME: we should enforce default clock in case GPU is not in | |
155 | * default setup | |
156 | */ | |
157 | a.full = rfixed_const(100); | |
158 | rdev->pm.sclk.full = rfixed_const(rdev->clock.default_sclk); | |
159 | rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a); | |
160 | a.full = rfixed_const(16); | |
161 | /* core_bandwidth = sclk(Mhz) * 16 */ | |
162 | rdev->pm.core_bandwidth.full = rfixed_div(rdev->pm.sclk, a); | |
163 | } | |
164 | ||
22dd5013 AD |
165 | static int rs690_mc_init(struct radeon_device *rdev) |
166 | { | |
167 | int r; | |
168 | u32 tmp; | |
169 | ||
170 | /* Setup GPU memory space */ | |
171 | tmp = RREG32_MC(R_000100_MCCFG_FB_LOCATION); | |
172 | rdev->mc.vram_location = G_000100_MC_FB_START(tmp) << 16; | |
173 | rdev->mc.gtt_location = 0xFFFFFFFFUL; | |
174 | r = radeon_mc_setup(rdev); | |
175 | if (r) | |
176 | return r; | |
177 | return 0; | |
178 | } | |
179 | ||
c93bb85b JG |
180 | void rs690_line_buffer_adjust(struct radeon_device *rdev, |
181 | struct drm_display_mode *mode1, | |
182 | struct drm_display_mode *mode2) | |
183 | { | |
184 | u32 tmp; | |
185 | ||
186 | /* | |
187 | * Line Buffer Setup | |
188 | * There is a single line buffer shared by both display controllers. | |
3bc68535 | 189 | * R_006520_DC_LB_MEMORY_SPLIT controls how that line buffer is shared between |
c93bb85b JG |
190 | * the display controllers. The paritioning can either be done |
191 | * manually or via one of four preset allocations specified in bits 1:0: | |
192 | * 0 - line buffer is divided in half and shared between crtc | |
193 | * 1 - D1 gets 3/4 of the line buffer, D2 gets 1/4 | |
194 | * 2 - D1 gets the whole buffer | |
195 | * 3 - D1 gets 1/4 of the line buffer, D2 gets 3/4 | |
3bc68535 | 196 | * Setting bit 2 of R_006520_DC_LB_MEMORY_SPLIT controls switches to manual |
c93bb85b JG |
197 | * allocation mode. In manual allocation mode, D1 always starts at 0, |
198 | * D1 end/2 is specified in bits 14:4; D2 allocation follows D1. | |
199 | */ | |
3bc68535 JG |
200 | tmp = RREG32(R_006520_DC_LB_MEMORY_SPLIT) & C_006520_DC_LB_MEMORY_SPLIT; |
201 | tmp &= ~C_006520_DC_LB_MEMORY_SPLIT_MODE; | |
c93bb85b JG |
202 | /* auto */ |
203 | if (mode1 && mode2) { | |
204 | if (mode1->hdisplay > mode2->hdisplay) { | |
205 | if (mode1->hdisplay > 2560) | |
3bc68535 | 206 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q; |
c93bb85b | 207 | else |
3bc68535 | 208 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF; |
c93bb85b JG |
209 | } else if (mode2->hdisplay > mode1->hdisplay) { |
210 | if (mode2->hdisplay > 2560) | |
3bc68535 | 211 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q; |
c93bb85b | 212 | else |
3bc68535 | 213 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF; |
c93bb85b | 214 | } else |
3bc68535 | 215 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF; |
c93bb85b | 216 | } else if (mode1) { |
3bc68535 | 217 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_ONLY; |
c93bb85b | 218 | } else if (mode2) { |
3bc68535 | 219 | tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q; |
c93bb85b | 220 | } |
3bc68535 | 221 | WREG32(R_006520_DC_LB_MEMORY_SPLIT, tmp); |
771fe6b9 JG |
222 | } |
223 | ||
c93bb85b JG |
224 | struct rs690_watermark { |
225 | u32 lb_request_fifo_depth; | |
226 | fixed20_12 num_line_pair; | |
227 | fixed20_12 estimated_width; | |
228 | fixed20_12 worst_case_latency; | |
229 | fixed20_12 consumption_rate; | |
230 | fixed20_12 active_time; | |
231 | fixed20_12 dbpp; | |
232 | fixed20_12 priority_mark_max; | |
233 | fixed20_12 priority_mark; | |
234 | fixed20_12 sclk; | |
235 | }; | |
236 | ||
237 | void rs690_crtc_bandwidth_compute(struct radeon_device *rdev, | |
238 | struct radeon_crtc *crtc, | |
239 | struct rs690_watermark *wm) | |
240 | { | |
241 | struct drm_display_mode *mode = &crtc->base.mode; | |
242 | fixed20_12 a, b, c; | |
243 | fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width; | |
244 | fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency; | |
245 | /* FIXME: detect IGP with sideport memory, i don't think there is any | |
246 | * such product available | |
247 | */ | |
248 | bool sideport = false; | |
249 | ||
250 | if (!crtc->base.enabled) { | |
251 | /* FIXME: wouldn't it better to set priority mark to maximum */ | |
252 | wm->lb_request_fifo_depth = 4; | |
253 | return; | |
254 | } | |
255 | ||
256 | if (crtc->vsc.full > rfixed_const(2)) | |
257 | wm->num_line_pair.full = rfixed_const(2); | |
258 | else | |
259 | wm->num_line_pair.full = rfixed_const(1); | |
260 | ||
261 | b.full = rfixed_const(mode->crtc_hdisplay); | |
262 | c.full = rfixed_const(256); | |
263 | a.full = rfixed_mul(wm->num_line_pair, b); | |
264 | request_fifo_depth.full = rfixed_div(a, c); | |
265 | if (a.full < rfixed_const(4)) { | |
266 | wm->lb_request_fifo_depth = 4; | |
267 | } else { | |
268 | wm->lb_request_fifo_depth = rfixed_trunc(request_fifo_depth); | |
269 | } | |
270 | ||
271 | /* Determine consumption rate | |
272 | * pclk = pixel clock period(ns) = 1000 / (mode.clock / 1000) | |
273 | * vtaps = number of vertical taps, | |
274 | * vsc = vertical scaling ratio, defined as source/destination | |
275 | * hsc = horizontal scaling ration, defined as source/destination | |
276 | */ | |
277 | a.full = rfixed_const(mode->clock); | |
278 | b.full = rfixed_const(1000); | |
279 | a.full = rfixed_div(a, b); | |
280 | pclk.full = rfixed_div(b, a); | |
281 | if (crtc->rmx_type != RMX_OFF) { | |
282 | b.full = rfixed_const(2); | |
283 | if (crtc->vsc.full > b.full) | |
284 | b.full = crtc->vsc.full; | |
285 | b.full = rfixed_mul(b, crtc->hsc); | |
286 | c.full = rfixed_const(2); | |
287 | b.full = rfixed_div(b, c); | |
288 | consumption_time.full = rfixed_div(pclk, b); | |
289 | } else { | |
290 | consumption_time.full = pclk.full; | |
291 | } | |
292 | a.full = rfixed_const(1); | |
293 | wm->consumption_rate.full = rfixed_div(a, consumption_time); | |
294 | ||
295 | ||
296 | /* Determine line time | |
297 | * LineTime = total time for one line of displayhtotal | |
298 | * LineTime = total number of horizontal pixels | |
299 | * pclk = pixel clock period(ns) | |
300 | */ | |
301 | a.full = rfixed_const(crtc->base.mode.crtc_htotal); | |
302 | line_time.full = rfixed_mul(a, pclk); | |
303 | ||
304 | /* Determine active time | |
305 | * ActiveTime = time of active region of display within one line, | |
306 | * hactive = total number of horizontal active pixels | |
307 | * htotal = total number of horizontal pixels | |
308 | */ | |
309 | a.full = rfixed_const(crtc->base.mode.crtc_htotal); | |
310 | b.full = rfixed_const(crtc->base.mode.crtc_hdisplay); | |
311 | wm->active_time.full = rfixed_mul(line_time, b); | |
312 | wm->active_time.full = rfixed_div(wm->active_time, a); | |
313 | ||
314 | /* Maximun bandwidth is the minimun bandwidth of all component */ | |
315 | rdev->pm.max_bandwidth = rdev->pm.core_bandwidth; | |
316 | if (sideport) { | |
317 | if (rdev->pm.max_bandwidth.full > rdev->pm.sideport_bandwidth.full && | |
318 | rdev->pm.sideport_bandwidth.full) | |
319 | rdev->pm.max_bandwidth = rdev->pm.sideport_bandwidth; | |
320 | read_delay_latency.full = rfixed_const(370 * 800 * 1000); | |
321 | read_delay_latency.full = rfixed_div(read_delay_latency, | |
322 | rdev->pm.igp_sideport_mclk); | |
323 | } else { | |
324 | if (rdev->pm.max_bandwidth.full > rdev->pm.k8_bandwidth.full && | |
325 | rdev->pm.k8_bandwidth.full) | |
326 | rdev->pm.max_bandwidth = rdev->pm.k8_bandwidth; | |
327 | if (rdev->pm.max_bandwidth.full > rdev->pm.ht_bandwidth.full && | |
328 | rdev->pm.ht_bandwidth.full) | |
329 | rdev->pm.max_bandwidth = rdev->pm.ht_bandwidth; | |
330 | read_delay_latency.full = rfixed_const(5000); | |
331 | } | |
332 | ||
333 | /* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */ | |
334 | a.full = rfixed_const(16); | |
335 | rdev->pm.sclk.full = rfixed_mul(rdev->pm.max_bandwidth, a); | |
336 | a.full = rfixed_const(1000); | |
337 | rdev->pm.sclk.full = rfixed_div(a, rdev->pm.sclk); | |
338 | /* Determine chunk time | |
339 | * ChunkTime = the time it takes the DCP to send one chunk of data | |
340 | * to the LB which consists of pipeline delay and inter chunk gap | |
341 | * sclk = system clock(ns) | |
342 | */ | |
343 | a.full = rfixed_const(256 * 13); | |
344 | chunk_time.full = rfixed_mul(rdev->pm.sclk, a); | |
345 | a.full = rfixed_const(10); | |
346 | chunk_time.full = rfixed_div(chunk_time, a); | |
347 | ||
348 | /* Determine the worst case latency | |
349 | * NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines) | |
350 | * WorstCaseLatency = worst case time from urgent to when the MC starts | |
351 | * to return data | |
352 | * READ_DELAY_IDLE_MAX = constant of 1us | |
353 | * ChunkTime = time it takes the DCP to send one chunk of data to the LB | |
354 | * which consists of pipeline delay and inter chunk gap | |
355 | */ | |
356 | if (rfixed_trunc(wm->num_line_pair) > 1) { | |
357 | a.full = rfixed_const(3); | |
358 | wm->worst_case_latency.full = rfixed_mul(a, chunk_time); | |
359 | wm->worst_case_latency.full += read_delay_latency.full; | |
360 | } else { | |
361 | a.full = rfixed_const(2); | |
362 | wm->worst_case_latency.full = rfixed_mul(a, chunk_time); | |
363 | wm->worst_case_latency.full += read_delay_latency.full; | |
364 | } | |
365 | ||
366 | /* Determine the tolerable latency | |
367 | * TolerableLatency = Any given request has only 1 line time | |
368 | * for the data to be returned | |
369 | * LBRequestFifoDepth = Number of chunk requests the LB can | |
370 | * put into the request FIFO for a display | |
371 | * LineTime = total time for one line of display | |
372 | * ChunkTime = the time it takes the DCP to send one chunk | |
373 | * of data to the LB which consists of | |
374 | * pipeline delay and inter chunk gap | |
375 | */ | |
376 | if ((2+wm->lb_request_fifo_depth) >= rfixed_trunc(request_fifo_depth)) { | |
377 | tolerable_latency.full = line_time.full; | |
378 | } else { | |
379 | tolerable_latency.full = rfixed_const(wm->lb_request_fifo_depth - 2); | |
380 | tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full; | |
381 | tolerable_latency.full = rfixed_mul(tolerable_latency, chunk_time); | |
382 | tolerable_latency.full = line_time.full - tolerable_latency.full; | |
383 | } | |
384 | /* We assume worst case 32bits (4 bytes) */ | |
385 | wm->dbpp.full = rfixed_const(4 * 8); | |
386 | ||
387 | /* Determine the maximum priority mark | |
388 | * width = viewport width in pixels | |
389 | */ | |
390 | a.full = rfixed_const(16); | |
391 | wm->priority_mark_max.full = rfixed_const(crtc->base.mode.crtc_hdisplay); | |
392 | wm->priority_mark_max.full = rfixed_div(wm->priority_mark_max, a); | |
393 | ||
394 | /* Determine estimated width */ | |
395 | estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full; | |
396 | estimated_width.full = rfixed_div(estimated_width, consumption_time); | |
397 | if (rfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) { | |
398 | wm->priority_mark.full = rfixed_const(10); | |
399 | } else { | |
400 | a.full = rfixed_const(16); | |
401 | wm->priority_mark.full = rfixed_div(estimated_width, a); | |
402 | wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full; | |
403 | } | |
404 | } | |
405 | ||
406 | void rs690_bandwidth_update(struct radeon_device *rdev) | |
407 | { | |
408 | struct drm_display_mode *mode0 = NULL; | |
409 | struct drm_display_mode *mode1 = NULL; | |
410 | struct rs690_watermark wm0; | |
411 | struct rs690_watermark wm1; | |
412 | u32 tmp; | |
413 | fixed20_12 priority_mark02, priority_mark12, fill_rate; | |
414 | fixed20_12 a, b; | |
415 | ||
416 | if (rdev->mode_info.crtcs[0]->base.enabled) | |
417 | mode0 = &rdev->mode_info.crtcs[0]->base.mode; | |
418 | if (rdev->mode_info.crtcs[1]->base.enabled) | |
419 | mode1 = &rdev->mode_info.crtcs[1]->base.mode; | |
420 | /* | |
421 | * Set display0/1 priority up in the memory controller for | |
422 | * modes if the user specifies HIGH for displaypriority | |
423 | * option. | |
424 | */ | |
425 | if (rdev->disp_priority == 2) { | |
3bc68535 JG |
426 | tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER); |
427 | tmp &= C_000104_MC_DISP0R_INIT_LAT; | |
428 | tmp &= C_000104_MC_DISP1R_INIT_LAT; | |
c93bb85b | 429 | if (mode0) |
3bc68535 JG |
430 | tmp |= S_000104_MC_DISP0R_INIT_LAT(1); |
431 | if (mode1) | |
432 | tmp |= S_000104_MC_DISP1R_INIT_LAT(1); | |
433 | WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp); | |
c93bb85b JG |
434 | } |
435 | rs690_line_buffer_adjust(rdev, mode0, mode1); | |
436 | ||
437 | if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740)) | |
3bc68535 | 438 | WREG32(R_006C9C_DCP_CONTROL, 0); |
c93bb85b | 439 | if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880)) |
3bc68535 | 440 | WREG32(R_006C9C_DCP_CONTROL, 2); |
c93bb85b JG |
441 | |
442 | rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0); | |
443 | rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1); | |
444 | ||
445 | tmp = (wm0.lb_request_fifo_depth - 1); | |
446 | tmp |= (wm1.lb_request_fifo_depth - 1) << 16; | |
3bc68535 | 447 | WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp); |
c93bb85b JG |
448 | |
449 | if (mode0 && mode1) { | |
450 | if (rfixed_trunc(wm0.dbpp) > 64) | |
451 | a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair); | |
452 | else | |
453 | a.full = wm0.num_line_pair.full; | |
454 | if (rfixed_trunc(wm1.dbpp) > 64) | |
455 | b.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair); | |
456 | else | |
457 | b.full = wm1.num_line_pair.full; | |
458 | a.full += b.full; | |
459 | fill_rate.full = rfixed_div(wm0.sclk, a); | |
460 | if (wm0.consumption_rate.full > fill_rate.full) { | |
461 | b.full = wm0.consumption_rate.full - fill_rate.full; | |
462 | b.full = rfixed_mul(b, wm0.active_time); | |
463 | a.full = rfixed_mul(wm0.worst_case_latency, | |
464 | wm0.consumption_rate); | |
465 | a.full = a.full + b.full; | |
466 | b.full = rfixed_const(16 * 1000); | |
467 | priority_mark02.full = rfixed_div(a, b); | |
468 | } else { | |
469 | a.full = rfixed_mul(wm0.worst_case_latency, | |
470 | wm0.consumption_rate); | |
471 | b.full = rfixed_const(16 * 1000); | |
472 | priority_mark02.full = rfixed_div(a, b); | |
473 | } | |
474 | if (wm1.consumption_rate.full > fill_rate.full) { | |
475 | b.full = wm1.consumption_rate.full - fill_rate.full; | |
476 | b.full = rfixed_mul(b, wm1.active_time); | |
477 | a.full = rfixed_mul(wm1.worst_case_latency, | |
478 | wm1.consumption_rate); | |
479 | a.full = a.full + b.full; | |
480 | b.full = rfixed_const(16 * 1000); | |
481 | priority_mark12.full = rfixed_div(a, b); | |
482 | } else { | |
483 | a.full = rfixed_mul(wm1.worst_case_latency, | |
484 | wm1.consumption_rate); | |
485 | b.full = rfixed_const(16 * 1000); | |
486 | priority_mark12.full = rfixed_div(a, b); | |
487 | } | |
488 | if (wm0.priority_mark.full > priority_mark02.full) | |
489 | priority_mark02.full = wm0.priority_mark.full; | |
490 | if (rfixed_trunc(priority_mark02) < 0) | |
491 | priority_mark02.full = 0; | |
492 | if (wm0.priority_mark_max.full > priority_mark02.full) | |
493 | priority_mark02.full = wm0.priority_mark_max.full; | |
494 | if (wm1.priority_mark.full > priority_mark12.full) | |
495 | priority_mark12.full = wm1.priority_mark.full; | |
496 | if (rfixed_trunc(priority_mark12) < 0) | |
497 | priority_mark12.full = 0; | |
498 | if (wm1.priority_mark_max.full > priority_mark12.full) | |
499 | priority_mark12.full = wm1.priority_mark_max.full; | |
3bc68535 JG |
500 | WREG32(R_006548_D1MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark02)); |
501 | WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark02)); | |
502 | WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark12)); | |
503 | WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark12)); | |
c93bb85b JG |
504 | } else if (mode0) { |
505 | if (rfixed_trunc(wm0.dbpp) > 64) | |
506 | a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair); | |
507 | else | |
508 | a.full = wm0.num_line_pair.full; | |
509 | fill_rate.full = rfixed_div(wm0.sclk, a); | |
510 | if (wm0.consumption_rate.full > fill_rate.full) { | |
511 | b.full = wm0.consumption_rate.full - fill_rate.full; | |
512 | b.full = rfixed_mul(b, wm0.active_time); | |
513 | a.full = rfixed_mul(wm0.worst_case_latency, | |
514 | wm0.consumption_rate); | |
515 | a.full = a.full + b.full; | |
516 | b.full = rfixed_const(16 * 1000); | |
517 | priority_mark02.full = rfixed_div(a, b); | |
518 | } else { | |
519 | a.full = rfixed_mul(wm0.worst_case_latency, | |
520 | wm0.consumption_rate); | |
521 | b.full = rfixed_const(16 * 1000); | |
522 | priority_mark02.full = rfixed_div(a, b); | |
523 | } | |
524 | if (wm0.priority_mark.full > priority_mark02.full) | |
525 | priority_mark02.full = wm0.priority_mark.full; | |
526 | if (rfixed_trunc(priority_mark02) < 0) | |
527 | priority_mark02.full = 0; | |
528 | if (wm0.priority_mark_max.full > priority_mark02.full) | |
529 | priority_mark02.full = wm0.priority_mark_max.full; | |
3bc68535 JG |
530 | WREG32(R_006548_D1MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark02)); |
531 | WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark02)); | |
532 | WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, | |
533 | S_006D48_D2MODE_PRIORITY_A_OFF(1)); | |
534 | WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, | |
535 | S_006D4C_D2MODE_PRIORITY_B_OFF(1)); | |
c93bb85b JG |
536 | } else { |
537 | if (rfixed_trunc(wm1.dbpp) > 64) | |
538 | a.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair); | |
539 | else | |
540 | a.full = wm1.num_line_pair.full; | |
541 | fill_rate.full = rfixed_div(wm1.sclk, a); | |
542 | if (wm1.consumption_rate.full > fill_rate.full) { | |
543 | b.full = wm1.consumption_rate.full - fill_rate.full; | |
544 | b.full = rfixed_mul(b, wm1.active_time); | |
545 | a.full = rfixed_mul(wm1.worst_case_latency, | |
546 | wm1.consumption_rate); | |
547 | a.full = a.full + b.full; | |
548 | b.full = rfixed_const(16 * 1000); | |
549 | priority_mark12.full = rfixed_div(a, b); | |
550 | } else { | |
551 | a.full = rfixed_mul(wm1.worst_case_latency, | |
552 | wm1.consumption_rate); | |
553 | b.full = rfixed_const(16 * 1000); | |
554 | priority_mark12.full = rfixed_div(a, b); | |
555 | } | |
556 | if (wm1.priority_mark.full > priority_mark12.full) | |
557 | priority_mark12.full = wm1.priority_mark.full; | |
558 | if (rfixed_trunc(priority_mark12) < 0) | |
559 | priority_mark12.full = 0; | |
560 | if (wm1.priority_mark_max.full > priority_mark12.full) | |
561 | priority_mark12.full = wm1.priority_mark_max.full; | |
3bc68535 JG |
562 | WREG32(R_006548_D1MODE_PRIORITY_A_CNT, |
563 | S_006548_D1MODE_PRIORITY_A_OFF(1)); | |
564 | WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, | |
565 | S_00654C_D1MODE_PRIORITY_B_OFF(1)); | |
566 | WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark12)); | |
567 | WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark12)); | |
c93bb85b JG |
568 | } |
569 | } | |
771fe6b9 | 570 | |
771fe6b9 JG |
571 | uint32_t rs690_mc_rreg(struct radeon_device *rdev, uint32_t reg) |
572 | { | |
573 | uint32_t r; | |
574 | ||
3bc68535 JG |
575 | WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg)); |
576 | r = RREG32(R_00007C_MC_DATA); | |
577 | WREG32(R_000078_MC_INDEX, ~C_000078_MC_IND_ADDR); | |
771fe6b9 JG |
578 | return r; |
579 | } | |
580 | ||
581 | void rs690_mc_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v) | |
582 | { | |
3bc68535 JG |
583 | WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg) | |
584 | S_000078_MC_IND_WR_EN(1)); | |
585 | WREG32(R_00007C_MC_DATA, v); | |
586 | WREG32(R_000078_MC_INDEX, 0x7F); | |
587 | } | |
588 | ||
589 | void rs690_mc_program(struct radeon_device *rdev) | |
590 | { | |
591 | struct rv515_mc_save save; | |
592 | ||
593 | /* Stops all mc clients */ | |
594 | rv515_mc_stop(rdev, &save); | |
595 | ||
596 | /* Wait for mc idle */ | |
597 | if (rs690_mc_wait_for_idle(rdev)) | |
598 | dev_warn(rdev->dev, "Wait MC idle timeout before updating MC.\n"); | |
599 | /* Program MC, should be a 32bits limited address space */ | |
600 | WREG32_MC(R_000100_MCCFG_FB_LOCATION, | |
601 | S_000100_MC_FB_START(rdev->mc.vram_start >> 16) | | |
602 | S_000100_MC_FB_TOP(rdev->mc.vram_end >> 16)); | |
603 | WREG32(R_000134_HDP_FB_LOCATION, | |
604 | S_000134_HDP_FB_START(rdev->mc.vram_start >> 16)); | |
605 | ||
606 | rv515_mc_resume(rdev, &save); | |
607 | } | |
608 | ||
609 | static int rs690_startup(struct radeon_device *rdev) | |
610 | { | |
611 | int r; | |
612 | ||
613 | rs690_mc_program(rdev); | |
614 | /* Resume clock */ | |
615 | rv515_clock_startup(rdev); | |
616 | /* Initialize GPU configuration (# pipes, ...) */ | |
617 | rs690_gpu_init(rdev); | |
618 | /* Initialize GART (initialize after TTM so we can allocate | |
619 | * memory through TTM but finalize after TTM) */ | |
620 | r = rs400_gart_enable(rdev); | |
621 | if (r) | |
622 | return r; | |
623 | /* Enable IRQ */ | |
ac447df4 | 624 | rs600_irq_set(rdev); |
3bc68535 JG |
625 | /* 1M ring buffer */ |
626 | r = r100_cp_init(rdev, 1024 * 1024); | |
627 | if (r) { | |
628 | dev_err(rdev->dev, "failled initializing CP (%d).\n", r); | |
629 | return r; | |
630 | } | |
631 | r = r100_wb_init(rdev); | |
632 | if (r) | |
633 | dev_err(rdev->dev, "failled initializing WB (%d).\n", r); | |
634 | r = r100_ib_init(rdev); | |
635 | if (r) { | |
636 | dev_err(rdev->dev, "failled initializing IB (%d).\n", r); | |
637 | return r; | |
638 | } | |
639 | return 0; | |
640 | } | |
641 | ||
642 | int rs690_resume(struct radeon_device *rdev) | |
643 | { | |
644 | /* Make sur GART are not working */ | |
645 | rs400_gart_disable(rdev); | |
646 | /* Resume clock before doing reset */ | |
647 | rv515_clock_startup(rdev); | |
648 | /* Reset gpu before posting otherwise ATOM will enter infinite loop */ | |
649 | if (radeon_gpu_reset(rdev)) { | |
650 | dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n", | |
651 | RREG32(R_000E40_RBBM_STATUS), | |
652 | RREG32(R_0007C0_CP_STAT)); | |
653 | } | |
654 | /* post */ | |
655 | atom_asic_init(rdev->mode_info.atom_context); | |
656 | /* Resume clock after posting */ | |
657 | rv515_clock_startup(rdev); | |
658 | return rs690_startup(rdev); | |
659 | } | |
660 | ||
661 | int rs690_suspend(struct radeon_device *rdev) | |
662 | { | |
663 | r100_cp_disable(rdev); | |
664 | r100_wb_disable(rdev); | |
ac447df4 | 665 | rs600_irq_disable(rdev); |
3bc68535 JG |
666 | rs400_gart_disable(rdev); |
667 | return 0; | |
668 | } | |
669 | ||
670 | void rs690_fini(struct radeon_device *rdev) | |
671 | { | |
672 | rs690_suspend(rdev); | |
673 | r100_cp_fini(rdev); | |
674 | r100_wb_fini(rdev); | |
675 | r100_ib_fini(rdev); | |
676 | radeon_gem_fini(rdev); | |
677 | rs400_gart_fini(rdev); | |
678 | radeon_irq_kms_fini(rdev); | |
679 | radeon_fence_driver_fini(rdev); | |
4c788679 | 680 | radeon_bo_fini(rdev); |
3bc68535 JG |
681 | radeon_atombios_fini(rdev); |
682 | kfree(rdev->bios); | |
683 | rdev->bios = NULL; | |
684 | } | |
685 | ||
686 | int rs690_init(struct radeon_device *rdev) | |
687 | { | |
688 | int r; | |
689 | ||
3bc68535 JG |
690 | /* Disable VGA */ |
691 | rv515_vga_render_disable(rdev); | |
692 | /* Initialize scratch registers */ | |
693 | radeon_scratch_init(rdev); | |
694 | /* Initialize surface registers */ | |
695 | radeon_surface_init(rdev); | |
696 | /* TODO: disable VGA need to use VGA request */ | |
697 | /* BIOS*/ | |
698 | if (!radeon_get_bios(rdev)) { | |
699 | if (ASIC_IS_AVIVO(rdev)) | |
700 | return -EINVAL; | |
701 | } | |
702 | if (rdev->is_atom_bios) { | |
703 | r = radeon_atombios_init(rdev); | |
704 | if (r) | |
705 | return r; | |
706 | } else { | |
707 | dev_err(rdev->dev, "Expecting atombios for RV515 GPU\n"); | |
708 | return -EINVAL; | |
709 | } | |
710 | /* Reset gpu before posting otherwise ATOM will enter infinite loop */ | |
711 | if (radeon_gpu_reset(rdev)) { | |
712 | dev_warn(rdev->dev, | |
713 | "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n", | |
714 | RREG32(R_000E40_RBBM_STATUS), | |
715 | RREG32(R_0007C0_CP_STAT)); | |
716 | } | |
717 | /* check if cards are posted or not */ | |
72542d77 DA |
718 | if (radeon_boot_test_post_card(rdev) == false) |
719 | return -EINVAL; | |
720 | ||
3bc68535 JG |
721 | /* Initialize clocks */ |
722 | radeon_get_clock_info(rdev->ddev); | |
7433874e RM |
723 | /* Initialize power management */ |
724 | radeon_pm_init(rdev); | |
3bc68535 JG |
725 | /* Get vram informations */ |
726 | rs690_vram_info(rdev); | |
727 | /* Initialize memory controller (also test AGP) */ | |
22dd5013 | 728 | r = rs690_mc_init(rdev); |
3bc68535 JG |
729 | if (r) |
730 | return r; | |
731 | rv515_debugfs(rdev); | |
732 | /* Fence driver */ | |
733 | r = radeon_fence_driver_init(rdev); | |
734 | if (r) | |
735 | return r; | |
736 | r = radeon_irq_kms_init(rdev); | |
737 | if (r) | |
738 | return r; | |
739 | /* Memory manager */ | |
4c788679 | 740 | r = radeon_bo_init(rdev); |
3bc68535 JG |
741 | if (r) |
742 | return r; | |
743 | r = rs400_gart_init(rdev); | |
744 | if (r) | |
745 | return r; | |
746 | rs600_set_safe_registers(rdev); | |
747 | rdev->accel_working = true; | |
748 | r = rs690_startup(rdev); | |
749 | if (r) { | |
750 | /* Somethings want wront with the accel init stop accel */ | |
751 | dev_err(rdev->dev, "Disabling GPU acceleration\n"); | |
752 | rs690_suspend(rdev); | |
753 | r100_cp_fini(rdev); | |
754 | r100_wb_fini(rdev); | |
755 | r100_ib_fini(rdev); | |
756 | rs400_gart_fini(rdev); | |
757 | radeon_irq_kms_fini(rdev); | |
758 | rdev->accel_working = false; | |
759 | } | |
760 | return 0; | |
771fe6b9 | 761 | } |