rs690.c revision cafe6609d6dc0a6a278f9fdbb59ce4d761a35ddd
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 "atom.h"
31#include "rs690d.h"
32
33static int rs690_mc_wait_for_idle(struct radeon_device *rdev)
34{
35	unsigned i;
36	uint32_t tmp;
37
38	for (i = 0; i < rdev->usec_timeout; i++) {
39		/* read MC_STATUS */
40		tmp = RREG32_MC(R_000090_MC_SYSTEM_STATUS);
41		if (G_000090_MC_SYSTEM_IDLE(tmp))
42			return 0;
43		udelay(1);
44	}
45	return -1;
46}
47
48static void rs690_gpu_init(struct radeon_device *rdev)
49{
50	/* FIXME: HDP same place on rs690 ? */
51	r100_hdp_reset(rdev);
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
60void 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
132void rs690_vram_info(struct radeon_device *rdev)
133{
134	fixed20_12 a;
135
136	rs400_gart_adjust_size(rdev);
137
138	rdev->mc.vram_is_ddr = true;
139	rdev->mc.vram_width = 128;
140
141	rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
142	rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
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);
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
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
165static 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	rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
176	if (r)
177		return r;
178	return 0;
179}
180
181void rs690_line_buffer_adjust(struct radeon_device *rdev,
182			      struct drm_display_mode *mode1,
183			      struct drm_display_mode *mode2)
184{
185	u32 tmp;
186
187	/*
188	 * Line Buffer Setup
189	 * There is a single line buffer shared by both display controllers.
190	 * R_006520_DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
191	 * the display controllers.  The paritioning can either be done
192	 * manually or via one of four preset allocations specified in bits 1:0:
193	 *  0 - line buffer is divided in half and shared between crtc
194	 *  1 - D1 gets 3/4 of the line buffer, D2 gets 1/4
195	 *  2 - D1 gets the whole buffer
196	 *  3 - D1 gets 1/4 of the line buffer, D2 gets 3/4
197	 * Setting bit 2 of R_006520_DC_LB_MEMORY_SPLIT controls switches to manual
198	 * allocation mode. In manual allocation mode, D1 always starts at 0,
199	 * D1 end/2 is specified in bits 14:4; D2 allocation follows D1.
200	 */
201	tmp = RREG32(R_006520_DC_LB_MEMORY_SPLIT) & C_006520_DC_LB_MEMORY_SPLIT;
202	tmp &= ~C_006520_DC_LB_MEMORY_SPLIT_MODE;
203	/* auto */
204	if (mode1 && mode2) {
205		if (mode1->hdisplay > mode2->hdisplay) {
206			if (mode1->hdisplay > 2560)
207				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q;
208			else
209				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
210		} else if (mode2->hdisplay > mode1->hdisplay) {
211			if (mode2->hdisplay > 2560)
212				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
213			else
214				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
215		} else
216			tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
217	} else if (mode1) {
218		tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_ONLY;
219	} else if (mode2) {
220		tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
221	}
222	WREG32(R_006520_DC_LB_MEMORY_SPLIT, tmp);
223}
224
225struct rs690_watermark {
226	u32        lb_request_fifo_depth;
227	fixed20_12 num_line_pair;
228	fixed20_12 estimated_width;
229	fixed20_12 worst_case_latency;
230	fixed20_12 consumption_rate;
231	fixed20_12 active_time;
232	fixed20_12 dbpp;
233	fixed20_12 priority_mark_max;
234	fixed20_12 priority_mark;
235	fixed20_12 sclk;
236};
237
238void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
239				  struct radeon_crtc *crtc,
240				  struct rs690_watermark *wm)
241{
242	struct drm_display_mode *mode = &crtc->base.mode;
243	fixed20_12 a, b, c;
244	fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width;
245	fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency;
246	/* FIXME: detect IGP with sideport memory, i don't think there is any
247	 * such product available
248	 */
249	bool sideport = false;
250
251	if (!crtc->base.enabled) {
252		/* FIXME: wouldn't it better to set priority mark to maximum */
253		wm->lb_request_fifo_depth = 4;
254		return;
255	}
256
257	if (crtc->vsc.full > rfixed_const(2))
258		wm->num_line_pair.full = rfixed_const(2);
259	else
260		wm->num_line_pair.full = rfixed_const(1);
261
262	b.full = rfixed_const(mode->crtc_hdisplay);
263	c.full = rfixed_const(256);
264	a.full = rfixed_div(b, c);
265	request_fifo_depth.full = rfixed_mul(a, wm->num_line_pair);
266	request_fifo_depth.full = rfixed_ceil(request_fifo_depth);
267	if (a.full < rfixed_const(4)) {
268		wm->lb_request_fifo_depth = 4;
269	} else {
270		wm->lb_request_fifo_depth = rfixed_trunc(request_fifo_depth);
271	}
272
273	/* Determine consumption rate
274	 *  pclk = pixel clock period(ns) = 1000 / (mode.clock / 1000)
275	 *  vtaps = number of vertical taps,
276	 *  vsc = vertical scaling ratio, defined as source/destination
277	 *  hsc = horizontal scaling ration, defined as source/destination
278	 */
279	a.full = rfixed_const(mode->clock);
280	b.full = rfixed_const(1000);
281	a.full = rfixed_div(a, b);
282	pclk.full = rfixed_div(b, a);
283	if (crtc->rmx_type != RMX_OFF) {
284		b.full = rfixed_const(2);
285		if (crtc->vsc.full > b.full)
286			b.full = crtc->vsc.full;
287		b.full = rfixed_mul(b, crtc->hsc);
288		c.full = rfixed_const(2);
289		b.full = rfixed_div(b, c);
290		consumption_time.full = rfixed_div(pclk, b);
291	} else {
292		consumption_time.full = pclk.full;
293	}
294	a.full = rfixed_const(1);
295	wm->consumption_rate.full = rfixed_div(a, consumption_time);
296
297
298	/* Determine line time
299	 *  LineTime = total time for one line of displayhtotal
300	 *  LineTime = total number of horizontal pixels
301	 *  pclk = pixel clock period(ns)
302	 */
303	a.full = rfixed_const(crtc->base.mode.crtc_htotal);
304	line_time.full = rfixed_mul(a, pclk);
305
306	/* Determine active time
307	 *  ActiveTime = time of active region of display within one line,
308	 *  hactive = total number of horizontal active pixels
309	 *  htotal = total number of horizontal pixels
310	 */
311	a.full = rfixed_const(crtc->base.mode.crtc_htotal);
312	b.full = rfixed_const(crtc->base.mode.crtc_hdisplay);
313	wm->active_time.full = rfixed_mul(line_time, b);
314	wm->active_time.full = rfixed_div(wm->active_time, a);
315
316	/* Maximun bandwidth is the minimun bandwidth of all component */
317	rdev->pm.max_bandwidth = rdev->pm.core_bandwidth;
318	if (sideport) {
319		if (rdev->pm.max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
320			rdev->pm.sideport_bandwidth.full)
321			rdev->pm.max_bandwidth = rdev->pm.sideport_bandwidth;
322		read_delay_latency.full = rfixed_const(370 * 800 * 1000);
323		read_delay_latency.full = rfixed_div(read_delay_latency,
324			rdev->pm.igp_sideport_mclk);
325	} else {
326		if (rdev->pm.max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
327			rdev->pm.k8_bandwidth.full)
328			rdev->pm.max_bandwidth = rdev->pm.k8_bandwidth;
329		if (rdev->pm.max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
330			rdev->pm.ht_bandwidth.full)
331			rdev->pm.max_bandwidth = rdev->pm.ht_bandwidth;
332		read_delay_latency.full = rfixed_const(5000);
333	}
334
335	/* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */
336	a.full = rfixed_const(16);
337	rdev->pm.sclk.full = rfixed_mul(rdev->pm.max_bandwidth, a);
338	a.full = rfixed_const(1000);
339	rdev->pm.sclk.full = rfixed_div(a, rdev->pm.sclk);
340	/* Determine chunk time
341	 * ChunkTime = the time it takes the DCP to send one chunk of data
342	 * to the LB which consists of pipeline delay and inter chunk gap
343	 * sclk = system clock(ns)
344	 */
345	a.full = rfixed_const(256 * 13);
346	chunk_time.full = rfixed_mul(rdev->pm.sclk, a);
347	a.full = rfixed_const(10);
348	chunk_time.full = rfixed_div(chunk_time, a);
349
350	/* Determine the worst case latency
351	 * NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines)
352	 * WorstCaseLatency = worst case time from urgent to when the MC starts
353	 *                    to return data
354	 * READ_DELAY_IDLE_MAX = constant of 1us
355	 * ChunkTime = time it takes the DCP to send one chunk of data to the LB
356	 *             which consists of pipeline delay and inter chunk gap
357	 */
358	if (rfixed_trunc(wm->num_line_pair) > 1) {
359		a.full = rfixed_const(3);
360		wm->worst_case_latency.full = rfixed_mul(a, chunk_time);
361		wm->worst_case_latency.full += read_delay_latency.full;
362	} else {
363		a.full = rfixed_const(2);
364		wm->worst_case_latency.full = rfixed_mul(a, chunk_time);
365		wm->worst_case_latency.full += read_delay_latency.full;
366	}
367
368	/* Determine the tolerable latency
369	 * TolerableLatency = Any given request has only 1 line time
370	 *                    for the data to be returned
371	 * LBRequestFifoDepth = Number of chunk requests the LB can
372	 *                      put into the request FIFO for a display
373	 *  LineTime = total time for one line of display
374	 *  ChunkTime = the time it takes the DCP to send one chunk
375	 *              of data to the LB which consists of
376	 *  pipeline delay and inter chunk gap
377	 */
378	if ((2+wm->lb_request_fifo_depth) >= rfixed_trunc(request_fifo_depth)) {
379		tolerable_latency.full = line_time.full;
380	} else {
381		tolerable_latency.full = rfixed_const(wm->lb_request_fifo_depth - 2);
382		tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full;
383		tolerable_latency.full = rfixed_mul(tolerable_latency, chunk_time);
384		tolerable_latency.full = line_time.full - tolerable_latency.full;
385	}
386	/* We assume worst case 32bits (4 bytes) */
387	wm->dbpp.full = rfixed_const(4 * 8);
388
389	/* Determine the maximum priority mark
390	 *  width = viewport width in pixels
391	 */
392	a.full = rfixed_const(16);
393	wm->priority_mark_max.full = rfixed_const(crtc->base.mode.crtc_hdisplay);
394	wm->priority_mark_max.full = rfixed_div(wm->priority_mark_max, a);
395	wm->priority_mark_max.full = rfixed_ceil(wm->priority_mark_max);
396
397	/* Determine estimated width */
398	estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full;
399	estimated_width.full = rfixed_div(estimated_width, consumption_time);
400	if (rfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
401		wm->priority_mark.full = rfixed_const(10);
402	} else {
403		a.full = rfixed_const(16);
404		wm->priority_mark.full = rfixed_div(estimated_width, a);
405		wm->priority_mark.full = rfixed_ceil(wm->priority_mark);
406		wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full;
407	}
408}
409
410void rs690_bandwidth_update(struct radeon_device *rdev)
411{
412	struct drm_display_mode *mode0 = NULL;
413	struct drm_display_mode *mode1 = NULL;
414	struct rs690_watermark wm0;
415	struct rs690_watermark wm1;
416	u32 tmp;
417	fixed20_12 priority_mark02, priority_mark12, fill_rate;
418	fixed20_12 a, b;
419
420	if (rdev->mode_info.crtcs[0]->base.enabled)
421		mode0 = &rdev->mode_info.crtcs[0]->base.mode;
422	if (rdev->mode_info.crtcs[1]->base.enabled)
423		mode1 = &rdev->mode_info.crtcs[1]->base.mode;
424	/*
425	 * Set display0/1 priority up in the memory controller for
426	 * modes if the user specifies HIGH for displaypriority
427	 * option.
428	 */
429	if (rdev->disp_priority == 2) {
430		tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER);
431		tmp &= C_000104_MC_DISP0R_INIT_LAT;
432		tmp &= C_000104_MC_DISP1R_INIT_LAT;
433		if (mode0)
434			tmp |= S_000104_MC_DISP0R_INIT_LAT(1);
435		if (mode1)
436			tmp |= S_000104_MC_DISP1R_INIT_LAT(1);
437		WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp);
438	}
439	rs690_line_buffer_adjust(rdev, mode0, mode1);
440
441	if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))
442		WREG32(R_006C9C_DCP_CONTROL, 0);
443	if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880))
444		WREG32(R_006C9C_DCP_CONTROL, 2);
445
446	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0);
447	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1);
448
449	tmp = (wm0.lb_request_fifo_depth - 1);
450	tmp |= (wm1.lb_request_fifo_depth - 1) << 16;
451	WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
452
453	if (mode0 && mode1) {
454		if (rfixed_trunc(wm0.dbpp) > 64)
455			a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair);
456		else
457			a.full = wm0.num_line_pair.full;
458		if (rfixed_trunc(wm1.dbpp) > 64)
459			b.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair);
460		else
461			b.full = wm1.num_line_pair.full;
462		a.full += b.full;
463		fill_rate.full = rfixed_div(wm0.sclk, a);
464		if (wm0.consumption_rate.full > fill_rate.full) {
465			b.full = wm0.consumption_rate.full - fill_rate.full;
466			b.full = rfixed_mul(b, wm0.active_time);
467			a.full = rfixed_mul(wm0.worst_case_latency,
468						wm0.consumption_rate);
469			a.full = a.full + b.full;
470			b.full = rfixed_const(16 * 1000);
471			priority_mark02.full = rfixed_div(a, b);
472		} else {
473			a.full = rfixed_mul(wm0.worst_case_latency,
474						wm0.consumption_rate);
475			b.full = rfixed_const(16 * 1000);
476			priority_mark02.full = rfixed_div(a, b);
477		}
478		if (wm1.consumption_rate.full > fill_rate.full) {
479			b.full = wm1.consumption_rate.full - fill_rate.full;
480			b.full = rfixed_mul(b, wm1.active_time);
481			a.full = rfixed_mul(wm1.worst_case_latency,
482						wm1.consumption_rate);
483			a.full = a.full + b.full;
484			b.full = rfixed_const(16 * 1000);
485			priority_mark12.full = rfixed_div(a, b);
486		} else {
487			a.full = rfixed_mul(wm1.worst_case_latency,
488						wm1.consumption_rate);
489			b.full = rfixed_const(16 * 1000);
490			priority_mark12.full = rfixed_div(a, b);
491		}
492		if (wm0.priority_mark.full > priority_mark02.full)
493			priority_mark02.full = wm0.priority_mark.full;
494		if (rfixed_trunc(priority_mark02) < 0)
495			priority_mark02.full = 0;
496		if (wm0.priority_mark_max.full > priority_mark02.full)
497			priority_mark02.full = wm0.priority_mark_max.full;
498		if (wm1.priority_mark.full > priority_mark12.full)
499			priority_mark12.full = wm1.priority_mark.full;
500		if (rfixed_trunc(priority_mark12) < 0)
501			priority_mark12.full = 0;
502		if (wm1.priority_mark_max.full > priority_mark12.full)
503			priority_mark12.full = wm1.priority_mark_max.full;
504		WREG32(R_006548_D1MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark02));
505		WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark02));
506		WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark12));
507		WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark12));
508	} else if (mode0) {
509		if (rfixed_trunc(wm0.dbpp) > 64)
510			a.full = rfixed_mul(wm0.dbpp, wm0.num_line_pair);
511		else
512			a.full = wm0.num_line_pair.full;
513		fill_rate.full = rfixed_div(wm0.sclk, a);
514		if (wm0.consumption_rate.full > fill_rate.full) {
515			b.full = wm0.consumption_rate.full - fill_rate.full;
516			b.full = rfixed_mul(b, wm0.active_time);
517			a.full = rfixed_mul(wm0.worst_case_latency,
518						wm0.consumption_rate);
519			a.full = a.full + b.full;
520			b.full = rfixed_const(16 * 1000);
521			priority_mark02.full = rfixed_div(a, b);
522		} else {
523			a.full = rfixed_mul(wm0.worst_case_latency,
524						wm0.consumption_rate);
525			b.full = rfixed_const(16 * 1000);
526			priority_mark02.full = rfixed_div(a, b);
527		}
528		if (wm0.priority_mark.full > priority_mark02.full)
529			priority_mark02.full = wm0.priority_mark.full;
530		if (rfixed_trunc(priority_mark02) < 0)
531			priority_mark02.full = 0;
532		if (wm0.priority_mark_max.full > priority_mark02.full)
533			priority_mark02.full = wm0.priority_mark_max.full;
534		WREG32(R_006548_D1MODE_PRIORITY_A_CNT, rfixed_trunc(priority_mark02));
535		WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark02));
536		WREG32(R_006D48_D2MODE_PRIORITY_A_CNT,
537			S_006D48_D2MODE_PRIORITY_A_OFF(1));
538		WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT,
539			S_006D4C_D2MODE_PRIORITY_B_OFF(1));
540	} else {
541		if (rfixed_trunc(wm1.dbpp) > 64)
542			a.full = rfixed_mul(wm1.dbpp, wm1.num_line_pair);
543		else
544			a.full = wm1.num_line_pair.full;
545		fill_rate.full = rfixed_div(wm1.sclk, a);
546		if (wm1.consumption_rate.full > fill_rate.full) {
547			b.full = wm1.consumption_rate.full - fill_rate.full;
548			b.full = rfixed_mul(b, wm1.active_time);
549			a.full = rfixed_mul(wm1.worst_case_latency,
550						wm1.consumption_rate);
551			a.full = a.full + b.full;
552			b.full = rfixed_const(16 * 1000);
553			priority_mark12.full = rfixed_div(a, b);
554		} else {
555			a.full = rfixed_mul(wm1.worst_case_latency,
556						wm1.consumption_rate);
557			b.full = rfixed_const(16 * 1000);
558			priority_mark12.full = rfixed_div(a, b);
559		}
560		if (wm1.priority_mark.full > priority_mark12.full)
561			priority_mark12.full = wm1.priority_mark.full;
562		if (rfixed_trunc(priority_mark12) < 0)
563			priority_mark12.full = 0;
564		if (wm1.priority_mark_max.full > priority_mark12.full)
565			priority_mark12.full = wm1.priority_mark_max.full;
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, rfixed_trunc(priority_mark12));
571		WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, rfixed_trunc(priority_mark12));
572	}
573}
574
575uint32_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
585void 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
593void 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
613static 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
647int 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_gpu_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
668int 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
677void rs690_fini(struct radeon_device *rdev)
678{
679	rs690_suspend(rdev);
680	r100_cp_fini(rdev);
681	r100_wb_fini(rdev);
682	r100_ib_fini(rdev);
683	radeon_gem_fini(rdev);
684	rs400_gart_fini(rdev);
685	radeon_irq_kms_fini(rdev);
686	radeon_fence_driver_fini(rdev);
687	radeon_bo_fini(rdev);
688	radeon_atombios_fini(rdev);
689	kfree(rdev->bios);
690	rdev->bios = NULL;
691}
692
693int rs690_init(struct radeon_device *rdev)
694{
695	int r;
696
697	/* Disable VGA */
698	rv515_vga_render_disable(rdev);
699	/* Initialize scratch registers */
700	radeon_scratch_init(rdev);
701	/* Initialize surface registers */
702	radeon_surface_init(rdev);
703	/* TODO: disable VGA need to use VGA request */
704	/* BIOS*/
705	if (!radeon_get_bios(rdev)) {
706		if (ASIC_IS_AVIVO(rdev))
707			return -EINVAL;
708	}
709	if (rdev->is_atom_bios) {
710		r = radeon_atombios_init(rdev);
711		if (r)
712			return r;
713	} else {
714		dev_err(rdev->dev, "Expecting atombios for RV515 GPU\n");
715		return -EINVAL;
716	}
717	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
718	if (radeon_gpu_reset(rdev)) {
719		dev_warn(rdev->dev,
720			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
721			RREG32(R_000E40_RBBM_STATUS),
722			RREG32(R_0007C0_CP_STAT));
723	}
724	/* check if cards are posted or not */
725	if (radeon_boot_test_post_card(rdev) == false)
726		return -EINVAL;
727
728	/* Initialize clocks */
729	radeon_get_clock_info(rdev->ddev);
730	/* Initialize power management */
731	radeon_pm_init(rdev);
732	/* Get vram informations */
733	rs690_vram_info(rdev);
734	/* Initialize memory controller (also test AGP) */
735	r = rs690_mc_init(rdev);
736	if (r)
737		return r;
738	rv515_debugfs(rdev);
739	/* Fence driver */
740	r = radeon_fence_driver_init(rdev);
741	if (r)
742		return r;
743	r = radeon_irq_kms_init(rdev);
744	if (r)
745		return r;
746	/* Memory manager */
747	r = radeon_bo_init(rdev);
748	if (r)
749		return r;
750	r = rs400_gart_init(rdev);
751	if (r)
752		return r;
753	rs600_set_safe_registers(rdev);
754	rdev->accel_working = true;
755	r = rs690_startup(rdev);
756	if (r) {
757		/* Somethings want wront with the accel init stop accel */
758		dev_err(rdev->dev, "Disabling GPU acceleration\n");
759		rs690_suspend(rdev);
760		r100_cp_fini(rdev);
761		r100_wb_fini(rdev);
762		r100_ib_fini(rdev);
763		rs400_gart_fini(rdev);
764		radeon_irq_kms_fini(rdev);
765		rdev->accel_working = false;
766	}
767	return 0;
768}
769