xref: /linux/drivers/gpu/drm/radeon/rs690.c (revision 5d4a2e29fba5b2bef95b96a46b338ec4d76fa4fd)
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"
29 #include "radeon.h"
30 #include "radeon_asic.h"
31 #include "atom.h"
32 #include "rs690d.h"
33 
34 static int rs690_mc_wait_for_idle(struct radeon_device *rdev)
35 {
36 	unsigned i;
37 	uint32_t tmp;
38 
39 	for (i = 0; i < rdev->usec_timeout; i++) {
40 		/* read MC_STATUS */
41 		tmp = RREG32_MC(R_000090_MC_SYSTEM_STATUS);
42 		if (G_000090_MC_SYSTEM_IDLE(tmp))
43 			return 0;
44 		udelay(1);
45 	}
46 	return -1;
47 }
48 
49 static void rs690_gpu_init(struct radeon_device *rdev)
50 {
51 	/* FIXME: is this correct ? */
52 	r420_pipes_init(rdev);
53 	if (rs690_mc_wait_for_idle(rdev)) {
54 		printk(KERN_WARNING "Failed to wait MC idle while "
55 		       "programming pipes. Bad things might happen.\n");
56 	}
57 }
58 
59 union igp_info {
60 	struct _ATOM_INTEGRATED_SYSTEM_INFO info;
61 	struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 info_v2;
62 };
63 
64 void rs690_pm_info(struct radeon_device *rdev)
65 {
66 	int index = GetIndexIntoMasterTable(DATA, IntegratedSystemInfo);
67 	union igp_info *info;
68 	uint16_t data_offset;
69 	uint8_t frev, crev;
70 	fixed20_12 tmp;
71 
72 	if (atom_parse_data_header(rdev->mode_info.atom_context, index, NULL,
73 				   &frev, &crev, &data_offset)) {
74 		info = (union igp_info *)(rdev->mode_info.atom_context->bios + data_offset);
75 
76 		/* Get various system informations from bios */
77 		switch (crev) {
78 		case 1:
79 			tmp.full = dfixed_const(100);
80 			rdev->pm.igp_sideport_mclk.full = dfixed_const(info->info.ulBootUpMemoryClock);
81 			rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
82 			rdev->pm.igp_system_mclk.full = dfixed_const(le16_to_cpu(info->info.usK8MemoryClock));
83 			rdev->pm.igp_ht_link_clk.full = dfixed_const(le16_to_cpu(info->info.usFSBClock));
84 			rdev->pm.igp_ht_link_width.full = dfixed_const(info->info.ucHTLinkWidth);
85 			break;
86 		case 2:
87 			tmp.full = dfixed_const(100);
88 			rdev->pm.igp_sideport_mclk.full = dfixed_const(info->info_v2.ulBootUpSidePortClock);
89 			rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
90 			rdev->pm.igp_system_mclk.full = dfixed_const(info->info_v2.ulBootUpUMAClock);
91 			rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
92 			rdev->pm.igp_ht_link_clk.full = dfixed_const(info->info_v2.ulHTLinkFreq);
93 			rdev->pm.igp_ht_link_clk.full = dfixed_div(rdev->pm.igp_ht_link_clk, tmp);
94 			rdev->pm.igp_ht_link_width.full = dfixed_const(le16_to_cpu(info->info_v2.usMinHTLinkWidth));
95 			break;
96 		default:
97 			tmp.full = dfixed_const(100);
98 			/* We assume the slower possible clock ie worst case */
99 			/* DDR 333Mhz */
100 			rdev->pm.igp_sideport_mclk.full = dfixed_const(333);
101 			/* FIXME: system clock ? */
102 			rdev->pm.igp_system_mclk.full = dfixed_const(100);
103 			rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
104 			rdev->pm.igp_ht_link_clk.full = dfixed_const(200);
105 			rdev->pm.igp_ht_link_width.full = dfixed_const(8);
106 			DRM_ERROR("No integrated system info for your GPU, using safe default\n");
107 			break;
108 		}
109 	} else {
110 		tmp.full = dfixed_const(100);
111 		/* We assume the slower possible clock ie worst case */
112 		/* DDR 333Mhz */
113 		rdev->pm.igp_sideport_mclk.full = dfixed_const(333);
114 		/* FIXME: system clock ? */
115 		rdev->pm.igp_system_mclk.full = dfixed_const(100);
116 		rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
117 		rdev->pm.igp_ht_link_clk.full = dfixed_const(200);
118 		rdev->pm.igp_ht_link_width.full = dfixed_const(8);
119 		DRM_ERROR("No integrated system info for your GPU, using safe default\n");
120 	}
121 	/* Compute various bandwidth */
122 	/* k8_bandwidth = (memory_clk / 2) * 2 * 8 * 0.5 = memory_clk * 4  */
123 	tmp.full = dfixed_const(4);
124 	rdev->pm.k8_bandwidth.full = dfixed_mul(rdev->pm.igp_system_mclk, tmp);
125 	/* ht_bandwidth = ht_clk * 2 * ht_width / 8 * 0.8
126 	 *              = ht_clk * ht_width / 5
127 	 */
128 	tmp.full = dfixed_const(5);
129 	rdev->pm.ht_bandwidth.full = dfixed_mul(rdev->pm.igp_ht_link_clk,
130 						rdev->pm.igp_ht_link_width);
131 	rdev->pm.ht_bandwidth.full = dfixed_div(rdev->pm.ht_bandwidth, tmp);
132 	if (tmp.full < rdev->pm.max_bandwidth.full) {
133 		/* HT link is a limiting factor */
134 		rdev->pm.max_bandwidth.full = tmp.full;
135 	}
136 	/* sideport_bandwidth = (sideport_clk / 2) * 2 * 2 * 0.7
137 	 *                    = (sideport_clk * 14) / 10
138 	 */
139 	tmp.full = dfixed_const(14);
140 	rdev->pm.sideport_bandwidth.full = dfixed_mul(rdev->pm.igp_sideport_mclk, tmp);
141 	tmp.full = dfixed_const(10);
142 	rdev->pm.sideport_bandwidth.full = dfixed_div(rdev->pm.sideport_bandwidth, tmp);
143 }
144 
145 void rs690_mc_init(struct radeon_device *rdev)
146 {
147 	u64 base;
148 
149 	rs400_gart_adjust_size(rdev);
150 	rdev->mc.vram_is_ddr = true;
151 	rdev->mc.vram_width = 128;
152 	rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
153 	rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
154 	rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
155 	rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
156 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
157 	base = RREG32_MC(R_000100_MCCFG_FB_LOCATION);
158 	base = G_000100_MC_FB_START(base) << 16;
159 	rs690_pm_info(rdev);
160 	rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
161 	radeon_vram_location(rdev, &rdev->mc, base);
162 	radeon_gtt_location(rdev, &rdev->mc);
163 	radeon_update_bandwidth_info(rdev);
164 }
165 
166 void rs690_line_buffer_adjust(struct radeon_device *rdev,
167 			      struct drm_display_mode *mode1,
168 			      struct drm_display_mode *mode2)
169 {
170 	u32 tmp;
171 
172 	/*
173 	 * Line Buffer Setup
174 	 * There is a single line buffer shared by both display controllers.
175 	 * R_006520_DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
176 	 * the display controllers.  The paritioning can either be done
177 	 * manually or via one of four preset allocations specified in bits 1:0:
178 	 *  0 - line buffer is divided in half and shared between crtc
179 	 *  1 - D1 gets 3/4 of the line buffer, D2 gets 1/4
180 	 *  2 - D1 gets the whole buffer
181 	 *  3 - D1 gets 1/4 of the line buffer, D2 gets 3/4
182 	 * Setting bit 2 of R_006520_DC_LB_MEMORY_SPLIT controls switches to manual
183 	 * allocation mode. In manual allocation mode, D1 always starts at 0,
184 	 * D1 end/2 is specified in bits 14:4; D2 allocation follows D1.
185 	 */
186 	tmp = RREG32(R_006520_DC_LB_MEMORY_SPLIT) & C_006520_DC_LB_MEMORY_SPLIT;
187 	tmp &= ~C_006520_DC_LB_MEMORY_SPLIT_MODE;
188 	/* auto */
189 	if (mode1 && mode2) {
190 		if (mode1->hdisplay > mode2->hdisplay) {
191 			if (mode1->hdisplay > 2560)
192 				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q;
193 			else
194 				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
195 		} else if (mode2->hdisplay > mode1->hdisplay) {
196 			if (mode2->hdisplay > 2560)
197 				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
198 			else
199 				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
200 		} else
201 			tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
202 	} else if (mode1) {
203 		tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_ONLY;
204 	} else if (mode2) {
205 		tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
206 	}
207 	WREG32(R_006520_DC_LB_MEMORY_SPLIT, tmp);
208 }
209 
210 struct rs690_watermark {
211 	u32        lb_request_fifo_depth;
212 	fixed20_12 num_line_pair;
213 	fixed20_12 estimated_width;
214 	fixed20_12 worst_case_latency;
215 	fixed20_12 consumption_rate;
216 	fixed20_12 active_time;
217 	fixed20_12 dbpp;
218 	fixed20_12 priority_mark_max;
219 	fixed20_12 priority_mark;
220 	fixed20_12 sclk;
221 };
222 
223 void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
224 				  struct radeon_crtc *crtc,
225 				  struct rs690_watermark *wm)
226 {
227 	struct drm_display_mode *mode = &crtc->base.mode;
228 	fixed20_12 a, b, c;
229 	fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width;
230 	fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency;
231 	/* FIXME: detect IGP with sideport memory, i don't think there is any
232 	 * such product available
233 	 */
234 	bool sideport = false;
235 
236 	if (!crtc->base.enabled) {
237 		/* FIXME: wouldn't it better to set priority mark to maximum */
238 		wm->lb_request_fifo_depth = 4;
239 		return;
240 	}
241 
242 	if (crtc->vsc.full > dfixed_const(2))
243 		wm->num_line_pair.full = dfixed_const(2);
244 	else
245 		wm->num_line_pair.full = dfixed_const(1);
246 
247 	b.full = dfixed_const(mode->crtc_hdisplay);
248 	c.full = dfixed_const(256);
249 	a.full = dfixed_div(b, c);
250 	request_fifo_depth.full = dfixed_mul(a, wm->num_line_pair);
251 	request_fifo_depth.full = dfixed_ceil(request_fifo_depth);
252 	if (a.full < dfixed_const(4)) {
253 		wm->lb_request_fifo_depth = 4;
254 	} else {
255 		wm->lb_request_fifo_depth = dfixed_trunc(request_fifo_depth);
256 	}
257 
258 	/* Determine consumption rate
259 	 *  pclk = pixel clock period(ns) = 1000 / (mode.clock / 1000)
260 	 *  vtaps = number of vertical taps,
261 	 *  vsc = vertical scaling ratio, defined as source/destination
262 	 *  hsc = horizontal scaling ration, defined as source/destination
263 	 */
264 	a.full = dfixed_const(mode->clock);
265 	b.full = dfixed_const(1000);
266 	a.full = dfixed_div(a, b);
267 	pclk.full = dfixed_div(b, a);
268 	if (crtc->rmx_type != RMX_OFF) {
269 		b.full = dfixed_const(2);
270 		if (crtc->vsc.full > b.full)
271 			b.full = crtc->vsc.full;
272 		b.full = dfixed_mul(b, crtc->hsc);
273 		c.full = dfixed_const(2);
274 		b.full = dfixed_div(b, c);
275 		consumption_time.full = dfixed_div(pclk, b);
276 	} else {
277 		consumption_time.full = pclk.full;
278 	}
279 	a.full = dfixed_const(1);
280 	wm->consumption_rate.full = dfixed_div(a, consumption_time);
281 
282 
283 	/* Determine line time
284 	 *  LineTime = total time for one line of displayhtotal
285 	 *  LineTime = total number of horizontal pixels
286 	 *  pclk = pixel clock period(ns)
287 	 */
288 	a.full = dfixed_const(crtc->base.mode.crtc_htotal);
289 	line_time.full = dfixed_mul(a, pclk);
290 
291 	/* Determine active time
292 	 *  ActiveTime = time of active region of display within one line,
293 	 *  hactive = total number of horizontal active pixels
294 	 *  htotal = total number of horizontal pixels
295 	 */
296 	a.full = dfixed_const(crtc->base.mode.crtc_htotal);
297 	b.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
298 	wm->active_time.full = dfixed_mul(line_time, b);
299 	wm->active_time.full = dfixed_div(wm->active_time, a);
300 
301 	/* Maximun bandwidth is the minimun bandwidth of all component */
302 	rdev->pm.max_bandwidth = rdev->pm.core_bandwidth;
303 	if (sideport) {
304 		if (rdev->pm.max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
305 			rdev->pm.sideport_bandwidth.full)
306 			rdev->pm.max_bandwidth = rdev->pm.sideport_bandwidth;
307 		read_delay_latency.full = dfixed_const(370 * 800 * 1000);
308 		read_delay_latency.full = dfixed_div(read_delay_latency,
309 			rdev->pm.igp_sideport_mclk);
310 	} else {
311 		if (rdev->pm.max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
312 			rdev->pm.k8_bandwidth.full)
313 			rdev->pm.max_bandwidth = rdev->pm.k8_bandwidth;
314 		if (rdev->pm.max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
315 			rdev->pm.ht_bandwidth.full)
316 			rdev->pm.max_bandwidth = rdev->pm.ht_bandwidth;
317 		read_delay_latency.full = dfixed_const(5000);
318 	}
319 
320 	/* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */
321 	a.full = dfixed_const(16);
322 	rdev->pm.sclk.full = dfixed_mul(rdev->pm.max_bandwidth, a);
323 	a.full = dfixed_const(1000);
324 	rdev->pm.sclk.full = dfixed_div(a, rdev->pm.sclk);
325 	/* Determine chunk time
326 	 * ChunkTime = the time it takes the DCP to send one chunk of data
327 	 * to the LB which consists of pipeline delay and inter chunk gap
328 	 * sclk = system clock(ns)
329 	 */
330 	a.full = dfixed_const(256 * 13);
331 	chunk_time.full = dfixed_mul(rdev->pm.sclk, a);
332 	a.full = dfixed_const(10);
333 	chunk_time.full = dfixed_div(chunk_time, a);
334 
335 	/* Determine the worst case latency
336 	 * NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines)
337 	 * WorstCaseLatency = worst case time from urgent to when the MC starts
338 	 *                    to return data
339 	 * READ_DELAY_IDLE_MAX = constant of 1us
340 	 * ChunkTime = time it takes the DCP to send one chunk of data to the LB
341 	 *             which consists of pipeline delay and inter chunk gap
342 	 */
343 	if (dfixed_trunc(wm->num_line_pair) > 1) {
344 		a.full = dfixed_const(3);
345 		wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
346 		wm->worst_case_latency.full += read_delay_latency.full;
347 	} else {
348 		a.full = dfixed_const(2);
349 		wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
350 		wm->worst_case_latency.full += read_delay_latency.full;
351 	}
352 
353 	/* Determine the tolerable latency
354 	 * TolerableLatency = Any given request has only 1 line time
355 	 *                    for the data to be returned
356 	 * LBRequestFifoDepth = Number of chunk requests the LB can
357 	 *                      put into the request FIFO for a display
358 	 *  LineTime = total time for one line of display
359 	 *  ChunkTime = the time it takes the DCP to send one chunk
360 	 *              of data to the LB which consists of
361 	 *  pipeline delay and inter chunk gap
362 	 */
363 	if ((2+wm->lb_request_fifo_depth) >= dfixed_trunc(request_fifo_depth)) {
364 		tolerable_latency.full = line_time.full;
365 	} else {
366 		tolerable_latency.full = dfixed_const(wm->lb_request_fifo_depth - 2);
367 		tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full;
368 		tolerable_latency.full = dfixed_mul(tolerable_latency, chunk_time);
369 		tolerable_latency.full = line_time.full - tolerable_latency.full;
370 	}
371 	/* We assume worst case 32bits (4 bytes) */
372 	wm->dbpp.full = dfixed_const(4 * 8);
373 
374 	/* Determine the maximum priority mark
375 	 *  width = viewport width in pixels
376 	 */
377 	a.full = dfixed_const(16);
378 	wm->priority_mark_max.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
379 	wm->priority_mark_max.full = dfixed_div(wm->priority_mark_max, a);
380 	wm->priority_mark_max.full = dfixed_ceil(wm->priority_mark_max);
381 
382 	/* Determine estimated width */
383 	estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full;
384 	estimated_width.full = dfixed_div(estimated_width, consumption_time);
385 	if (dfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
386 		wm->priority_mark.full = dfixed_const(10);
387 	} else {
388 		a.full = dfixed_const(16);
389 		wm->priority_mark.full = dfixed_div(estimated_width, a);
390 		wm->priority_mark.full = dfixed_ceil(wm->priority_mark);
391 		wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full;
392 	}
393 }
394 
395 void rs690_bandwidth_update(struct radeon_device *rdev)
396 {
397 	struct drm_display_mode *mode0 = NULL;
398 	struct drm_display_mode *mode1 = NULL;
399 	struct rs690_watermark wm0;
400 	struct rs690_watermark wm1;
401 	u32 tmp, d1mode_priority_a_cnt, d2mode_priority_a_cnt;
402 	fixed20_12 priority_mark02, priority_mark12, fill_rate;
403 	fixed20_12 a, b;
404 
405 	radeon_update_display_priority(rdev);
406 
407 	if (rdev->mode_info.crtcs[0]->base.enabled)
408 		mode0 = &rdev->mode_info.crtcs[0]->base.mode;
409 	if (rdev->mode_info.crtcs[1]->base.enabled)
410 		mode1 = &rdev->mode_info.crtcs[1]->base.mode;
411 	/*
412 	 * Set display0/1 priority up in the memory controller for
413 	 * modes if the user specifies HIGH for displaypriority
414 	 * option.
415 	 */
416 	if ((rdev->disp_priority == 2) &&
417 	    ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))) {
418 		tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER);
419 		tmp &= C_000104_MC_DISP0R_INIT_LAT;
420 		tmp &= C_000104_MC_DISP1R_INIT_LAT;
421 		if (mode0)
422 			tmp |= S_000104_MC_DISP0R_INIT_LAT(1);
423 		if (mode1)
424 			tmp |= S_000104_MC_DISP1R_INIT_LAT(1);
425 		WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp);
426 	}
427 	rs690_line_buffer_adjust(rdev, mode0, mode1);
428 
429 	if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))
430 		WREG32(R_006C9C_DCP_CONTROL, 0);
431 	if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880))
432 		WREG32(R_006C9C_DCP_CONTROL, 2);
433 
434 	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0);
435 	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1);
436 
437 	tmp = (wm0.lb_request_fifo_depth - 1);
438 	tmp |= (wm1.lb_request_fifo_depth - 1) << 16;
439 	WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
440 
441 	if (mode0 && mode1) {
442 		if (dfixed_trunc(wm0.dbpp) > 64)
443 			a.full = dfixed_mul(wm0.dbpp, wm0.num_line_pair);
444 		else
445 			a.full = wm0.num_line_pair.full;
446 		if (dfixed_trunc(wm1.dbpp) > 64)
447 			b.full = dfixed_mul(wm1.dbpp, wm1.num_line_pair);
448 		else
449 			b.full = wm1.num_line_pair.full;
450 		a.full += b.full;
451 		fill_rate.full = dfixed_div(wm0.sclk, a);
452 		if (wm0.consumption_rate.full > fill_rate.full) {
453 			b.full = wm0.consumption_rate.full - fill_rate.full;
454 			b.full = dfixed_mul(b, wm0.active_time);
455 			a.full = dfixed_mul(wm0.worst_case_latency,
456 						wm0.consumption_rate);
457 			a.full = a.full + b.full;
458 			b.full = dfixed_const(16 * 1000);
459 			priority_mark02.full = dfixed_div(a, b);
460 		} else {
461 			a.full = dfixed_mul(wm0.worst_case_latency,
462 						wm0.consumption_rate);
463 			b.full = dfixed_const(16 * 1000);
464 			priority_mark02.full = dfixed_div(a, b);
465 		}
466 		if (wm1.consumption_rate.full > fill_rate.full) {
467 			b.full = wm1.consumption_rate.full - fill_rate.full;
468 			b.full = dfixed_mul(b, wm1.active_time);
469 			a.full = dfixed_mul(wm1.worst_case_latency,
470 						wm1.consumption_rate);
471 			a.full = a.full + b.full;
472 			b.full = dfixed_const(16 * 1000);
473 			priority_mark12.full = dfixed_div(a, b);
474 		} else {
475 			a.full = dfixed_mul(wm1.worst_case_latency,
476 						wm1.consumption_rate);
477 			b.full = dfixed_const(16 * 1000);
478 			priority_mark12.full = dfixed_div(a, b);
479 		}
480 		if (wm0.priority_mark.full > priority_mark02.full)
481 			priority_mark02.full = wm0.priority_mark.full;
482 		if (dfixed_trunc(priority_mark02) < 0)
483 			priority_mark02.full = 0;
484 		if (wm0.priority_mark_max.full > priority_mark02.full)
485 			priority_mark02.full = wm0.priority_mark_max.full;
486 		if (wm1.priority_mark.full > priority_mark12.full)
487 			priority_mark12.full = wm1.priority_mark.full;
488 		if (dfixed_trunc(priority_mark12) < 0)
489 			priority_mark12.full = 0;
490 		if (wm1.priority_mark_max.full > priority_mark12.full)
491 			priority_mark12.full = wm1.priority_mark_max.full;
492 		d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
493 		d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
494 		if (rdev->disp_priority == 2) {
495 			d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
496 			d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
497 		}
498 		WREG32(R_006548_D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
499 		WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, d1mode_priority_a_cnt);
500 		WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
501 		WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_a_cnt);
502 	} else if (mode0) {
503 		if (dfixed_trunc(wm0.dbpp) > 64)
504 			a.full = dfixed_mul(wm0.dbpp, wm0.num_line_pair);
505 		else
506 			a.full = wm0.num_line_pair.full;
507 		fill_rate.full = dfixed_div(wm0.sclk, a);
508 		if (wm0.consumption_rate.full > fill_rate.full) {
509 			b.full = wm0.consumption_rate.full - fill_rate.full;
510 			b.full = dfixed_mul(b, wm0.active_time);
511 			a.full = dfixed_mul(wm0.worst_case_latency,
512 						wm0.consumption_rate);
513 			a.full = a.full + b.full;
514 			b.full = dfixed_const(16 * 1000);
515 			priority_mark02.full = dfixed_div(a, b);
516 		} else {
517 			a.full = dfixed_mul(wm0.worst_case_latency,
518 						wm0.consumption_rate);
519 			b.full = dfixed_const(16 * 1000);
520 			priority_mark02.full = dfixed_div(a, b);
521 		}
522 		if (wm0.priority_mark.full > priority_mark02.full)
523 			priority_mark02.full = wm0.priority_mark.full;
524 		if (dfixed_trunc(priority_mark02) < 0)
525 			priority_mark02.full = 0;
526 		if (wm0.priority_mark_max.full > priority_mark02.full)
527 			priority_mark02.full = wm0.priority_mark_max.full;
528 		d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
529 		if (rdev->disp_priority == 2)
530 			d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
531 		WREG32(R_006548_D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
532 		WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, d1mode_priority_a_cnt);
533 		WREG32(R_006D48_D2MODE_PRIORITY_A_CNT,
534 			S_006D48_D2MODE_PRIORITY_A_OFF(1));
535 		WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT,
536 			S_006D4C_D2MODE_PRIORITY_B_OFF(1));
537 	} else {
538 		if (dfixed_trunc(wm1.dbpp) > 64)
539 			a.full = dfixed_mul(wm1.dbpp, wm1.num_line_pair);
540 		else
541 			a.full = wm1.num_line_pair.full;
542 		fill_rate.full = dfixed_div(wm1.sclk, a);
543 		if (wm1.consumption_rate.full > fill_rate.full) {
544 			b.full = wm1.consumption_rate.full - fill_rate.full;
545 			b.full = dfixed_mul(b, wm1.active_time);
546 			a.full = dfixed_mul(wm1.worst_case_latency,
547 						wm1.consumption_rate);
548 			a.full = a.full + b.full;
549 			b.full = dfixed_const(16 * 1000);
550 			priority_mark12.full = dfixed_div(a, b);
551 		} else {
552 			a.full = dfixed_mul(wm1.worst_case_latency,
553 						wm1.consumption_rate);
554 			b.full = dfixed_const(16 * 1000);
555 			priority_mark12.full = dfixed_div(a, b);
556 		}
557 		if (wm1.priority_mark.full > priority_mark12.full)
558 			priority_mark12.full = wm1.priority_mark.full;
559 		if (dfixed_trunc(priority_mark12) < 0)
560 			priority_mark12.full = 0;
561 		if (wm1.priority_mark_max.full > priority_mark12.full)
562 			priority_mark12.full = wm1.priority_mark_max.full;
563 		d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
564 		if (rdev->disp_priority == 2)
565 			d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
566 		WREG32(R_006548_D1MODE_PRIORITY_A_CNT,
567 			S_006548_D1MODE_PRIORITY_A_OFF(1));
568 		WREG32(R_00654C_D1MODE_PRIORITY_B_CNT,
569 			S_00654C_D1MODE_PRIORITY_B_OFF(1));
570 		WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
571 		WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_a_cnt);
572 	}
573 }
574 
575 uint32_t rs690_mc_rreg(struct radeon_device *rdev, uint32_t reg)
576 {
577 	uint32_t r;
578 
579 	WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg));
580 	r = RREG32(R_00007C_MC_DATA);
581 	WREG32(R_000078_MC_INDEX, ~C_000078_MC_IND_ADDR);
582 	return r;
583 }
584 
585 void rs690_mc_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
586 {
587 	WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg) |
588 		S_000078_MC_IND_WR_EN(1));
589 	WREG32(R_00007C_MC_DATA, v);
590 	WREG32(R_000078_MC_INDEX, 0x7F);
591 }
592 
593 void rs690_mc_program(struct radeon_device *rdev)
594 {
595 	struct rv515_mc_save save;
596 
597 	/* Stops all mc clients */
598 	rv515_mc_stop(rdev, &save);
599 
600 	/* Wait for mc idle */
601 	if (rs690_mc_wait_for_idle(rdev))
602 		dev_warn(rdev->dev, "Wait MC idle timeout before updating MC.\n");
603 	/* Program MC, should be a 32bits limited address space */
604 	WREG32_MC(R_000100_MCCFG_FB_LOCATION,
605 			S_000100_MC_FB_START(rdev->mc.vram_start >> 16) |
606 			S_000100_MC_FB_TOP(rdev->mc.vram_end >> 16));
607 	WREG32(R_000134_HDP_FB_LOCATION,
608 		S_000134_HDP_FB_START(rdev->mc.vram_start >> 16));
609 
610 	rv515_mc_resume(rdev, &save);
611 }
612 
613 static int rs690_startup(struct radeon_device *rdev)
614 {
615 	int r;
616 
617 	rs690_mc_program(rdev);
618 	/* Resume clock */
619 	rv515_clock_startup(rdev);
620 	/* Initialize GPU configuration (# pipes, ...) */
621 	rs690_gpu_init(rdev);
622 	/* Initialize GART (initialize after TTM so we can allocate
623 	 * memory through TTM but finalize after TTM) */
624 	r = rs400_gart_enable(rdev);
625 	if (r)
626 		return r;
627 	/* Enable IRQ */
628 	rs600_irq_set(rdev);
629 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
630 	/* 1M ring buffer */
631 	r = r100_cp_init(rdev, 1024 * 1024);
632 	if (r) {
633 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
634 		return r;
635 	}
636 	r = r100_wb_init(rdev);
637 	if (r)
638 		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
639 	r = r100_ib_init(rdev);
640 	if (r) {
641 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
642 		return r;
643 	}
644 	return 0;
645 }
646 
647 int rs690_resume(struct radeon_device *rdev)
648 {
649 	/* Make sur GART are not working */
650 	rs400_gart_disable(rdev);
651 	/* Resume clock before doing reset */
652 	rv515_clock_startup(rdev);
653 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
654 	if (radeon_asic_reset(rdev)) {
655 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
656 			RREG32(R_000E40_RBBM_STATUS),
657 			RREG32(R_0007C0_CP_STAT));
658 	}
659 	/* post */
660 	atom_asic_init(rdev->mode_info.atom_context);
661 	/* Resume clock after posting */
662 	rv515_clock_startup(rdev);
663 	/* Initialize surface registers */
664 	radeon_surface_init(rdev);
665 	return rs690_startup(rdev);
666 }
667 
668 int rs690_suspend(struct radeon_device *rdev)
669 {
670 	r100_cp_disable(rdev);
671 	r100_wb_disable(rdev);
672 	rs600_irq_disable(rdev);
673 	rs400_gart_disable(rdev);
674 	return 0;
675 }
676 
677 void rs690_fini(struct radeon_device *rdev)
678 {
679 	r100_cp_fini(rdev);
680 	r100_wb_fini(rdev);
681 	r100_ib_fini(rdev);
682 	radeon_gem_fini(rdev);
683 	rs400_gart_fini(rdev);
684 	radeon_irq_kms_fini(rdev);
685 	radeon_fence_driver_fini(rdev);
686 	radeon_bo_fini(rdev);
687 	radeon_atombios_fini(rdev);
688 	kfree(rdev->bios);
689 	rdev->bios = NULL;
690 }
691 
692 int rs690_init(struct radeon_device *rdev)
693 {
694 	int r;
695 
696 	/* Disable VGA */
697 	rv515_vga_render_disable(rdev);
698 	/* Initialize scratch registers */
699 	radeon_scratch_init(rdev);
700 	/* Initialize surface registers */
701 	radeon_surface_init(rdev);
702 	/* TODO: disable VGA need to use VGA request */
703 	/* BIOS*/
704 	if (!radeon_get_bios(rdev)) {
705 		if (ASIC_IS_AVIVO(rdev))
706 			return -EINVAL;
707 	}
708 	if (rdev->is_atom_bios) {
709 		r = radeon_atombios_init(rdev);
710 		if (r)
711 			return r;
712 	} else {
713 		dev_err(rdev->dev, "Expecting atombios for RV515 GPU\n");
714 		return -EINVAL;
715 	}
716 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
717 	if (radeon_asic_reset(rdev)) {
718 		dev_warn(rdev->dev,
719 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
720 			RREG32(R_000E40_RBBM_STATUS),
721 			RREG32(R_0007C0_CP_STAT));
722 	}
723 	/* check if cards are posted or not */
724 	if (radeon_boot_test_post_card(rdev) == false)
725 		return -EINVAL;
726 
727 	/* Initialize clocks */
728 	radeon_get_clock_info(rdev->ddev);
729 	/* initialize memory controller */
730 	rs690_mc_init(rdev);
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 */
740 	r = radeon_bo_init(rdev);
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 		r100_cp_fini(rdev);
753 		r100_wb_fini(rdev);
754 		r100_ib_fini(rdev);
755 		rs400_gart_fini(rdev);
756 		radeon_irq_kms_fini(rdev);
757 		rdev->accel_working = false;
758 	}
759 	return 0;
760 }
761