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 <drm/drmP.h>
29#include "radeon.h"
30#include "radeon_asic.h"
31#include "radeon_audio.h"
32#include "atom.h"
33#include "rs690d.h"
34
35int rs690_mc_wait_for_idle(struct radeon_device *rdev)
36{
37	unsigned i;
38	uint32_t tmp;
39
40	for (i = 0; i < rdev->usec_timeout; i++) {
41		/* read MC_STATUS */
42		tmp = RREG32_MC(R_000090_MC_SYSTEM_STATUS);
43		if (G_000090_MC_SYSTEM_IDLE(tmp))
44			return 0;
45		udelay(1);
46	}
47	return -1;
48}
49
50static void rs690_gpu_init(struct radeon_device *rdev)
51{
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
60union igp_info {
61	struct _ATOM_INTEGRATED_SYSTEM_INFO info;
62	struct _ATOM_INTEGRATED_SYSTEM_INFO_V2 info_v2;
63};
64
65void rs690_pm_info(struct radeon_device *rdev)
66{
67	int index = GetIndexIntoMasterTable(DATA, IntegratedSystemInfo);
68	union igp_info *info;
69	uint16_t data_offset;
70	uint8_t frev, crev;
71	fixed20_12 tmp;
72
73	if (atom_parse_data_header(rdev->mode_info.atom_context, index, NULL,
74				   &frev, &crev, &data_offset)) {
75		info = (union igp_info *)(rdev->mode_info.atom_context->bios + data_offset);
76
77		/* Get various system informations from bios */
78		switch (crev) {
79		case 1:
80			tmp.full = dfixed_const(100);
81			rdev->pm.igp_sideport_mclk.full = dfixed_const(le32_to_cpu(info->info.ulBootUpMemoryClock));
82			rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
83			if (le16_to_cpu(info->info.usK8MemoryClock))
84				rdev->pm.igp_system_mclk.full = dfixed_const(le16_to_cpu(info->info.usK8MemoryClock));
85			else if (rdev->clock.default_mclk) {
86				rdev->pm.igp_system_mclk.full = dfixed_const(rdev->clock.default_mclk);
87				rdev->pm.igp_system_mclk.full = dfixed_div(rdev->pm.igp_system_mclk, tmp);
88			} else
89				rdev->pm.igp_system_mclk.full = dfixed_const(400);
90			rdev->pm.igp_ht_link_clk.full = dfixed_const(le16_to_cpu(info->info.usFSBClock));
91			rdev->pm.igp_ht_link_width.full = dfixed_const(info->info.ucHTLinkWidth);
92			break;
93		case 2:
94			tmp.full = dfixed_const(100);
95			rdev->pm.igp_sideport_mclk.full = dfixed_const(le32_to_cpu(info->info_v2.ulBootUpSidePortClock));
96			rdev->pm.igp_sideport_mclk.full = dfixed_div(rdev->pm.igp_sideport_mclk, tmp);
97			if (le32_to_cpu(info->info_v2.ulBootUpUMAClock))
98				rdev->pm.igp_system_mclk.full = dfixed_const(le32_to_cpu(info->info_v2.ulBootUpUMAClock));
99			else if (rdev->clock.default_mclk)
100				rdev->pm.igp_system_mclk.full = dfixed_const(rdev->clock.default_mclk);
101			else
102				rdev->pm.igp_system_mclk.full = dfixed_const(66700);
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(le32_to_cpu(info->info_v2.ulHTLinkFreq));
105			rdev->pm.igp_ht_link_clk.full = dfixed_div(rdev->pm.igp_ht_link_clk, tmp);
106			rdev->pm.igp_ht_link_width.full = dfixed_const(le16_to_cpu(info->info_v2.usMinHTLinkWidth));
107			break;
108		default:
109			/* We assume the slower possible clock ie worst case */
110			rdev->pm.igp_sideport_mclk.full = dfixed_const(200);
111			rdev->pm.igp_system_mclk.full = dfixed_const(200);
112			rdev->pm.igp_ht_link_clk.full = dfixed_const(1000);
113			rdev->pm.igp_ht_link_width.full = dfixed_const(8);
114			DRM_ERROR("No integrated system info for your GPU, using safe default\n");
115			break;
116		}
117	} else {
118		/* We assume the slower possible clock ie worst case */
119		rdev->pm.igp_sideport_mclk.full = dfixed_const(200);
120		rdev->pm.igp_system_mclk.full = dfixed_const(200);
121		rdev->pm.igp_ht_link_clk.full = dfixed_const(1000);
122		rdev->pm.igp_ht_link_width.full = dfixed_const(8);
123		DRM_ERROR("No integrated system info for your GPU, using safe default\n");
124	}
125	/* Compute various bandwidth */
126	/* k8_bandwidth = (memory_clk / 2) * 2 * 8 * 0.5 = memory_clk * 4  */
127	tmp.full = dfixed_const(4);
128	rdev->pm.k8_bandwidth.full = dfixed_mul(rdev->pm.igp_system_mclk, tmp);
129	/* ht_bandwidth = ht_clk * 2 * ht_width / 8 * 0.8
130	 *              = ht_clk * ht_width / 5
131	 */
132	tmp.full = dfixed_const(5);
133	rdev->pm.ht_bandwidth.full = dfixed_mul(rdev->pm.igp_ht_link_clk,
134						rdev->pm.igp_ht_link_width);
135	rdev->pm.ht_bandwidth.full = dfixed_div(rdev->pm.ht_bandwidth, tmp);
136	if (tmp.full < rdev->pm.max_bandwidth.full) {
137		/* HT link is a limiting factor */
138		rdev->pm.max_bandwidth.full = tmp.full;
139	}
140	/* sideport_bandwidth = (sideport_clk / 2) * 2 * 2 * 0.7
141	 *                    = (sideport_clk * 14) / 10
142	 */
143	tmp.full = dfixed_const(14);
144	rdev->pm.sideport_bandwidth.full = dfixed_mul(rdev->pm.igp_sideport_mclk, tmp);
145	tmp.full = dfixed_const(10);
146	rdev->pm.sideport_bandwidth.full = dfixed_div(rdev->pm.sideport_bandwidth, tmp);
147}
148
149static void rs690_mc_init(struct radeon_device *rdev)
150{
151	u64 base;
152	uint32_t h_addr, l_addr;
153	unsigned long long k8_addr;
154
155	rs400_gart_adjust_size(rdev);
156	rdev->mc.vram_is_ddr = true;
157	rdev->mc.vram_width = 128;
158	rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
159	rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
160	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
161	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
162	rdev->mc.visible_vram_size = rdev->mc.aper_size;
163	base = RREG32_MC(R_000100_MCCFG_FB_LOCATION);
164	base = G_000100_MC_FB_START(base) << 16;
165	rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
166	/* Some boards seem to be configured for 128MB of sideport memory,
167	 * but really only have 64MB.  Just skip the sideport and use
168	 * UMA memory.
169	 */
170	if (rdev->mc.igp_sideport_enabled &&
171	    (rdev->mc.real_vram_size == (384 * 1024 * 1024))) {
172		base += 128 * 1024 * 1024;
173		rdev->mc.real_vram_size -= 128 * 1024 * 1024;
174		rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
175	}
176
177	/* Use K8 direct mapping for fast fb access. */
178	rdev->fastfb_working = false;
179	h_addr = G_00005F_K8_ADDR_EXT(RREG32_MC(R_00005F_MC_MISC_UMA_CNTL));
180	l_addr = RREG32_MC(R_00001E_K8_FB_LOCATION);
181	k8_addr = ((unsigned long long)h_addr) << 32 | l_addr;
182#if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
183	if (k8_addr + rdev->mc.visible_vram_size < 0x100000000ULL)
184#endif
185	{
186		/* FastFB shall be used with UMA memory. Here it is simply disabled when sideport
187		 * memory is present.
188		 */
189		if (rdev->mc.igp_sideport_enabled == false && radeon_fastfb == 1) {
190			DRM_INFO("Direct mapping: aper base at 0x%llx, replaced by direct mapping base 0x%llx.\n",
191					(unsigned long long)rdev->mc.aper_base, k8_addr);
192			rdev->mc.aper_base = (resource_size_t)k8_addr;
193			rdev->fastfb_working = true;
194		}
195	}
196
197	rs690_pm_info(rdev);
198	radeon_vram_location(rdev, &rdev->mc, base);
199	rdev->mc.gtt_base_align = rdev->mc.gtt_size - 1;
200	radeon_gtt_location(rdev, &rdev->mc);
201	radeon_update_bandwidth_info(rdev);
202}
203
204void rs690_line_buffer_adjust(struct radeon_device *rdev,
205			      struct drm_display_mode *mode1,
206			      struct drm_display_mode *mode2)
207{
208	u32 tmp;
209
210	/* Guess line buffer size to be 8192 pixels */
211	u32 lb_size = 8192;
212
213	/*
214	 * Line Buffer Setup
215	 * There is a single line buffer shared by both display controllers.
216	 * R_006520_DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
217	 * the display controllers.  The paritioning can either be done
218	 * manually or via one of four preset allocations specified in bits 1:0:
219	 *  0 - line buffer is divided in half and shared between crtc
220	 *  1 - D1 gets 3/4 of the line buffer, D2 gets 1/4
221	 *  2 - D1 gets the whole buffer
222	 *  3 - D1 gets 1/4 of the line buffer, D2 gets 3/4
223	 * Setting bit 2 of R_006520_DC_LB_MEMORY_SPLIT controls switches to manual
224	 * allocation mode. In manual allocation mode, D1 always starts at 0,
225	 * D1 end/2 is specified in bits 14:4; D2 allocation follows D1.
226	 */
227	tmp = RREG32(R_006520_DC_LB_MEMORY_SPLIT) & C_006520_DC_LB_MEMORY_SPLIT;
228	tmp &= ~C_006520_DC_LB_MEMORY_SPLIT_MODE;
229	/* auto */
230	if (mode1 && mode2) {
231		if (mode1->hdisplay > mode2->hdisplay) {
232			if (mode1->hdisplay > 2560)
233				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_3Q_D2_1Q;
234			else
235				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
236		} else if (mode2->hdisplay > mode1->hdisplay) {
237			if (mode2->hdisplay > 2560)
238				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
239			else
240				tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
241		} else
242			tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1HALF_D2HALF;
243	} else if (mode1) {
244		tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_ONLY;
245	} else if (mode2) {
246		tmp |= V_006520_DC_LB_MEMORY_SPLIT_D1_1Q_D2_3Q;
247	}
248	WREG32(R_006520_DC_LB_MEMORY_SPLIT, tmp);
249
250	/* Save number of lines the linebuffer leads before the scanout */
251	if (mode1)
252		rdev->mode_info.crtcs[0]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode1->crtc_hdisplay);
253
254	if (mode2)
255		rdev->mode_info.crtcs[1]->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode2->crtc_hdisplay);
256}
257
258struct rs690_watermark {
259	u32        lb_request_fifo_depth;
260	fixed20_12 num_line_pair;
261	fixed20_12 estimated_width;
262	fixed20_12 worst_case_latency;
263	fixed20_12 consumption_rate;
264	fixed20_12 active_time;
265	fixed20_12 dbpp;
266	fixed20_12 priority_mark_max;
267	fixed20_12 priority_mark;
268	fixed20_12 sclk;
269};
270
271static void rs690_crtc_bandwidth_compute(struct radeon_device *rdev,
272					 struct radeon_crtc *crtc,
273					 struct rs690_watermark *wm,
274					 bool low)
275{
276	struct drm_display_mode *mode = &crtc->base.mode;
277	fixed20_12 a, b, c;
278	fixed20_12 pclk, request_fifo_depth, tolerable_latency, estimated_width;
279	fixed20_12 consumption_time, line_time, chunk_time, read_delay_latency;
280	fixed20_12 sclk, core_bandwidth, max_bandwidth;
281	u32 selected_sclk;
282
283	if (!crtc->base.enabled) {
284		/* FIXME: wouldn't it better to set priority mark to maximum */
285		wm->lb_request_fifo_depth = 4;
286		return;
287	}
288
289	if (((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880)) &&
290	    (rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled)
291		selected_sclk = radeon_dpm_get_sclk(rdev, low);
292	else
293		selected_sclk = rdev->pm.current_sclk;
294
295	/* sclk in Mhz */
296	a.full = dfixed_const(100);
297	sclk.full = dfixed_const(selected_sclk);
298	sclk.full = dfixed_div(sclk, a);
299
300	/* core_bandwidth = sclk(Mhz) * 16 */
301	a.full = dfixed_const(16);
302	core_bandwidth.full = dfixed_div(rdev->pm.sclk, a);
303
304	if (crtc->vsc.full > dfixed_const(2))
305		wm->num_line_pair.full = dfixed_const(2);
306	else
307		wm->num_line_pair.full = dfixed_const(1);
308
309	b.full = dfixed_const(mode->crtc_hdisplay);
310	c.full = dfixed_const(256);
311	a.full = dfixed_div(b, c);
312	request_fifo_depth.full = dfixed_mul(a, wm->num_line_pair);
313	request_fifo_depth.full = dfixed_ceil(request_fifo_depth);
314	if (a.full < dfixed_const(4)) {
315		wm->lb_request_fifo_depth = 4;
316	} else {
317		wm->lb_request_fifo_depth = dfixed_trunc(request_fifo_depth);
318	}
319
320	/* Determine consumption rate
321	 *  pclk = pixel clock period(ns) = 1000 / (mode.clock / 1000)
322	 *  vtaps = number of vertical taps,
323	 *  vsc = vertical scaling ratio, defined as source/destination
324	 *  hsc = horizontal scaling ration, defined as source/destination
325	 */
326	a.full = dfixed_const(mode->clock);
327	b.full = dfixed_const(1000);
328	a.full = dfixed_div(a, b);
329	pclk.full = dfixed_div(b, a);
330	if (crtc->rmx_type != RMX_OFF) {
331		b.full = dfixed_const(2);
332		if (crtc->vsc.full > b.full)
333			b.full = crtc->vsc.full;
334		b.full = dfixed_mul(b, crtc->hsc);
335		c.full = dfixed_const(2);
336		b.full = dfixed_div(b, c);
337		consumption_time.full = dfixed_div(pclk, b);
338	} else {
339		consumption_time.full = pclk.full;
340	}
341	a.full = dfixed_const(1);
342	wm->consumption_rate.full = dfixed_div(a, consumption_time);
343
344
345	/* Determine line time
346	 *  LineTime = total time for one line of displayhtotal
347	 *  LineTime = total number of horizontal pixels
348	 *  pclk = pixel clock period(ns)
349	 */
350	a.full = dfixed_const(crtc->base.mode.crtc_htotal);
351	line_time.full = dfixed_mul(a, pclk);
352
353	/* Determine active time
354	 *  ActiveTime = time of active region of display within one line,
355	 *  hactive = total number of horizontal active pixels
356	 *  htotal = total number of horizontal pixels
357	 */
358	a.full = dfixed_const(crtc->base.mode.crtc_htotal);
359	b.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
360	wm->active_time.full = dfixed_mul(line_time, b);
361	wm->active_time.full = dfixed_div(wm->active_time, a);
362
363	/* Maximun bandwidth is the minimun bandwidth of all component */
364	max_bandwidth = core_bandwidth;
365	if (rdev->mc.igp_sideport_enabled) {
366		if (max_bandwidth.full > rdev->pm.sideport_bandwidth.full &&
367			rdev->pm.sideport_bandwidth.full)
368			max_bandwidth = rdev->pm.sideport_bandwidth;
369		read_delay_latency.full = dfixed_const(370 * 800);
370		a.full = dfixed_const(1000);
371		b.full = dfixed_div(rdev->pm.igp_sideport_mclk, a);
372		read_delay_latency.full = dfixed_div(read_delay_latency, b);
373		read_delay_latency.full = dfixed_mul(read_delay_latency, a);
374	} else {
375		if (max_bandwidth.full > rdev->pm.k8_bandwidth.full &&
376			rdev->pm.k8_bandwidth.full)
377			max_bandwidth = rdev->pm.k8_bandwidth;
378		if (max_bandwidth.full > rdev->pm.ht_bandwidth.full &&
379			rdev->pm.ht_bandwidth.full)
380			max_bandwidth = rdev->pm.ht_bandwidth;
381		read_delay_latency.full = dfixed_const(5000);
382	}
383
384	/* sclk = system clocks(ns) = 1000 / max_bandwidth / 16 */
385	a.full = dfixed_const(16);
386	sclk.full = dfixed_mul(max_bandwidth, a);
387	a.full = dfixed_const(1000);
388	sclk.full = dfixed_div(a, sclk);
389	/* Determine chunk time
390	 * ChunkTime = the time it takes the DCP to send one chunk of data
391	 * to the LB which consists of pipeline delay and inter chunk gap
392	 * sclk = system clock(ns)
393	 */
394	a.full = dfixed_const(256 * 13);
395	chunk_time.full = dfixed_mul(sclk, a);
396	a.full = dfixed_const(10);
397	chunk_time.full = dfixed_div(chunk_time, a);
398
399	/* Determine the worst case latency
400	 * NumLinePair = Number of line pairs to request(1=2 lines, 2=4 lines)
401	 * WorstCaseLatency = worst case time from urgent to when the MC starts
402	 *                    to return data
403	 * READ_DELAY_IDLE_MAX = constant of 1us
404	 * ChunkTime = time it takes the DCP to send one chunk of data to the LB
405	 *             which consists of pipeline delay and inter chunk gap
406	 */
407	if (dfixed_trunc(wm->num_line_pair) > 1) {
408		a.full = dfixed_const(3);
409		wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
410		wm->worst_case_latency.full += read_delay_latency.full;
411	} else {
412		a.full = dfixed_const(2);
413		wm->worst_case_latency.full = dfixed_mul(a, chunk_time);
414		wm->worst_case_latency.full += read_delay_latency.full;
415	}
416
417	/* Determine the tolerable latency
418	 * TolerableLatency = Any given request has only 1 line time
419	 *                    for the data to be returned
420	 * LBRequestFifoDepth = Number of chunk requests the LB can
421	 *                      put into the request FIFO for a display
422	 *  LineTime = total time for one line of display
423	 *  ChunkTime = the time it takes the DCP to send one chunk
424	 *              of data to the LB which consists of
425	 *  pipeline delay and inter chunk gap
426	 */
427	if ((2+wm->lb_request_fifo_depth) >= dfixed_trunc(request_fifo_depth)) {
428		tolerable_latency.full = line_time.full;
429	} else {
430		tolerable_latency.full = dfixed_const(wm->lb_request_fifo_depth - 2);
431		tolerable_latency.full = request_fifo_depth.full - tolerable_latency.full;
432		tolerable_latency.full = dfixed_mul(tolerable_latency, chunk_time);
433		tolerable_latency.full = line_time.full - tolerable_latency.full;
434	}
435	/* We assume worst case 32bits (4 bytes) */
436	wm->dbpp.full = dfixed_const(4 * 8);
437
438	/* Determine the maximum priority mark
439	 *  width = viewport width in pixels
440	 */
441	a.full = dfixed_const(16);
442	wm->priority_mark_max.full = dfixed_const(crtc->base.mode.crtc_hdisplay);
443	wm->priority_mark_max.full = dfixed_div(wm->priority_mark_max, a);
444	wm->priority_mark_max.full = dfixed_ceil(wm->priority_mark_max);
445
446	/* Determine estimated width */
447	estimated_width.full = tolerable_latency.full - wm->worst_case_latency.full;
448	estimated_width.full = dfixed_div(estimated_width, consumption_time);
449	if (dfixed_trunc(estimated_width) > crtc->base.mode.crtc_hdisplay) {
450		wm->priority_mark.full = dfixed_const(10);
451	} else {
452		a.full = dfixed_const(16);
453		wm->priority_mark.full = dfixed_div(estimated_width, a);
454		wm->priority_mark.full = dfixed_ceil(wm->priority_mark);
455		wm->priority_mark.full = wm->priority_mark_max.full - wm->priority_mark.full;
456	}
457}
458
459static void rs690_compute_mode_priority(struct radeon_device *rdev,
460					struct rs690_watermark *wm0,
461					struct rs690_watermark *wm1,
462					struct drm_display_mode *mode0,
463					struct drm_display_mode *mode1,
464					u32 *d1mode_priority_a_cnt,
465					u32 *d2mode_priority_a_cnt)
466{
467	fixed20_12 priority_mark02, priority_mark12, fill_rate;
468	fixed20_12 a, b;
469
470	*d1mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
471	*d2mode_priority_a_cnt = S_006548_D1MODE_PRIORITY_A_OFF(1);
472
473	if (mode0 && mode1) {
474		if (dfixed_trunc(wm0->dbpp) > 64)
475			a.full = dfixed_mul(wm0->dbpp, wm0->num_line_pair);
476		else
477			a.full = wm0->num_line_pair.full;
478		if (dfixed_trunc(wm1->dbpp) > 64)
479			b.full = dfixed_mul(wm1->dbpp, wm1->num_line_pair);
480		else
481			b.full = wm1->num_line_pair.full;
482		a.full += b.full;
483		fill_rate.full = dfixed_div(wm0->sclk, a);
484		if (wm0->consumption_rate.full > fill_rate.full) {
485			b.full = wm0->consumption_rate.full - fill_rate.full;
486			b.full = dfixed_mul(b, wm0->active_time);
487			a.full = dfixed_mul(wm0->worst_case_latency,
488						wm0->consumption_rate);
489			a.full = a.full + b.full;
490			b.full = dfixed_const(16 * 1000);
491			priority_mark02.full = dfixed_div(a, b);
492		} else {
493			a.full = dfixed_mul(wm0->worst_case_latency,
494						wm0->consumption_rate);
495			b.full = dfixed_const(16 * 1000);
496			priority_mark02.full = dfixed_div(a, b);
497		}
498		if (wm1->consumption_rate.full > fill_rate.full) {
499			b.full = wm1->consumption_rate.full - fill_rate.full;
500			b.full = dfixed_mul(b, wm1->active_time);
501			a.full = dfixed_mul(wm1->worst_case_latency,
502						wm1->consumption_rate);
503			a.full = a.full + b.full;
504			b.full = dfixed_const(16 * 1000);
505			priority_mark12.full = dfixed_div(a, b);
506		} else {
507			a.full = dfixed_mul(wm1->worst_case_latency,
508						wm1->consumption_rate);
509			b.full = dfixed_const(16 * 1000);
510			priority_mark12.full = dfixed_div(a, b);
511		}
512		if (wm0->priority_mark.full > priority_mark02.full)
513			priority_mark02.full = wm0->priority_mark.full;
514		if (wm0->priority_mark_max.full > priority_mark02.full)
515			priority_mark02.full = wm0->priority_mark_max.full;
516		if (wm1->priority_mark.full > priority_mark12.full)
517			priority_mark12.full = wm1->priority_mark.full;
518		if (wm1->priority_mark_max.full > priority_mark12.full)
519			priority_mark12.full = wm1->priority_mark_max.full;
520		*d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
521		*d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
522		if (rdev->disp_priority == 2) {
523			*d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
524			*d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
525		}
526	} else if (mode0) {
527		if (dfixed_trunc(wm0->dbpp) > 64)
528			a.full = dfixed_mul(wm0->dbpp, wm0->num_line_pair);
529		else
530			a.full = wm0->num_line_pair.full;
531		fill_rate.full = dfixed_div(wm0->sclk, a);
532		if (wm0->consumption_rate.full > fill_rate.full) {
533			b.full = wm0->consumption_rate.full - fill_rate.full;
534			b.full = dfixed_mul(b, wm0->active_time);
535			a.full = dfixed_mul(wm0->worst_case_latency,
536						wm0->consumption_rate);
537			a.full = a.full + b.full;
538			b.full = dfixed_const(16 * 1000);
539			priority_mark02.full = dfixed_div(a, b);
540		} else {
541			a.full = dfixed_mul(wm0->worst_case_latency,
542						wm0->consumption_rate);
543			b.full = dfixed_const(16 * 1000);
544			priority_mark02.full = dfixed_div(a, b);
545		}
546		if (wm0->priority_mark.full > priority_mark02.full)
547			priority_mark02.full = wm0->priority_mark.full;
548		if (wm0->priority_mark_max.full > priority_mark02.full)
549			priority_mark02.full = wm0->priority_mark_max.full;
550		*d1mode_priority_a_cnt = dfixed_trunc(priority_mark02);
551		if (rdev->disp_priority == 2)
552			*d1mode_priority_a_cnt |= S_006548_D1MODE_PRIORITY_A_ALWAYS_ON(1);
553	} else if (mode1) {
554		if (dfixed_trunc(wm1->dbpp) > 64)
555			a.full = dfixed_mul(wm1->dbpp, wm1->num_line_pair);
556		else
557			a.full = wm1->num_line_pair.full;
558		fill_rate.full = dfixed_div(wm1->sclk, a);
559		if (wm1->consumption_rate.full > fill_rate.full) {
560			b.full = wm1->consumption_rate.full - fill_rate.full;
561			b.full = dfixed_mul(b, wm1->active_time);
562			a.full = dfixed_mul(wm1->worst_case_latency,
563						wm1->consumption_rate);
564			a.full = a.full + b.full;
565			b.full = dfixed_const(16 * 1000);
566			priority_mark12.full = dfixed_div(a, b);
567		} else {
568			a.full = dfixed_mul(wm1->worst_case_latency,
569						wm1->consumption_rate);
570			b.full = dfixed_const(16 * 1000);
571			priority_mark12.full = dfixed_div(a, b);
572		}
573		if (wm1->priority_mark.full > priority_mark12.full)
574			priority_mark12.full = wm1->priority_mark.full;
575		if (wm1->priority_mark_max.full > priority_mark12.full)
576			priority_mark12.full = wm1->priority_mark_max.full;
577		*d2mode_priority_a_cnt = dfixed_trunc(priority_mark12);
578		if (rdev->disp_priority == 2)
579			*d2mode_priority_a_cnt |= S_006D48_D2MODE_PRIORITY_A_ALWAYS_ON(1);
580	}
581}
582
583void rs690_bandwidth_update(struct radeon_device *rdev)
584{
585	struct drm_display_mode *mode0 = NULL;
586	struct drm_display_mode *mode1 = NULL;
587	struct rs690_watermark wm0_high, wm0_low;
588	struct rs690_watermark wm1_high, wm1_low;
589	u32 tmp;
590	u32 d1mode_priority_a_cnt, d1mode_priority_b_cnt;
591	u32 d2mode_priority_a_cnt, d2mode_priority_b_cnt;
592
593	if (!rdev->mode_info.mode_config_initialized)
594		return;
595
596	radeon_update_display_priority(rdev);
597
598	if (rdev->mode_info.crtcs[0]->base.enabled)
599		mode0 = &rdev->mode_info.crtcs[0]->base.mode;
600	if (rdev->mode_info.crtcs[1]->base.enabled)
601		mode1 = &rdev->mode_info.crtcs[1]->base.mode;
602	/*
603	 * Set display0/1 priority up in the memory controller for
604	 * modes if the user specifies HIGH for displaypriority
605	 * option.
606	 */
607	if ((rdev->disp_priority == 2) &&
608	    ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))) {
609		tmp = RREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER);
610		tmp &= C_000104_MC_DISP0R_INIT_LAT;
611		tmp &= C_000104_MC_DISP1R_INIT_LAT;
612		if (mode0)
613			tmp |= S_000104_MC_DISP0R_INIT_LAT(1);
614		if (mode1)
615			tmp |= S_000104_MC_DISP1R_INIT_LAT(1);
616		WREG32_MC(R_000104_MC_INIT_MISC_LAT_TIMER, tmp);
617	}
618	rs690_line_buffer_adjust(rdev, mode0, mode1);
619
620	if ((rdev->family == CHIP_RS690) || (rdev->family == CHIP_RS740))
621		WREG32(R_006C9C_DCP_CONTROL, 0);
622	if ((rdev->family == CHIP_RS780) || (rdev->family == CHIP_RS880))
623		WREG32(R_006C9C_DCP_CONTROL, 2);
624
625	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0_high, false);
626	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1_high, false);
627
628	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[0], &wm0_low, true);
629	rs690_crtc_bandwidth_compute(rdev, rdev->mode_info.crtcs[1], &wm1_low, true);
630
631	tmp = (wm0_high.lb_request_fifo_depth - 1);
632	tmp |= (wm1_high.lb_request_fifo_depth - 1) << 16;
633	WREG32(R_006D58_LB_MAX_REQ_OUTSTANDING, tmp);
634
635	rs690_compute_mode_priority(rdev,
636				    &wm0_high, &wm1_high,
637				    mode0, mode1,
638				    &d1mode_priority_a_cnt, &d2mode_priority_a_cnt);
639	rs690_compute_mode_priority(rdev,
640				    &wm0_low, &wm1_low,
641				    mode0, mode1,
642				    &d1mode_priority_b_cnt, &d2mode_priority_b_cnt);
643
644	WREG32(R_006548_D1MODE_PRIORITY_A_CNT, d1mode_priority_a_cnt);
645	WREG32(R_00654C_D1MODE_PRIORITY_B_CNT, d1mode_priority_b_cnt);
646	WREG32(R_006D48_D2MODE_PRIORITY_A_CNT, d2mode_priority_a_cnt);
647	WREG32(R_006D4C_D2MODE_PRIORITY_B_CNT, d2mode_priority_b_cnt);
648}
649
650uint32_t rs690_mc_rreg(struct radeon_device *rdev, uint32_t reg)
651{
652	unsigned long flags;
653	uint32_t r;
654
655	spin_lock_irqsave(&rdev->mc_idx_lock, flags);
656	WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg));
657	r = RREG32(R_00007C_MC_DATA);
658	WREG32(R_000078_MC_INDEX, ~C_000078_MC_IND_ADDR);
659	spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
660	return r;
661}
662
663void rs690_mc_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
664{
665	unsigned long flags;
666
667	spin_lock_irqsave(&rdev->mc_idx_lock, flags);
668	WREG32(R_000078_MC_INDEX, S_000078_MC_IND_ADDR(reg) |
669		S_000078_MC_IND_WR_EN(1));
670	WREG32(R_00007C_MC_DATA, v);
671	WREG32(R_000078_MC_INDEX, 0x7F);
672	spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
673}
674
675static void rs690_mc_program(struct radeon_device *rdev)
676{
677	struct rv515_mc_save save;
678
679	/* Stops all mc clients */
680	rv515_mc_stop(rdev, &save);
681
682	/* Wait for mc idle */
683	if (rs690_mc_wait_for_idle(rdev))
684		dev_warn(rdev->dev, "Wait MC idle timeout before updating MC.\n");
685	/* Program MC, should be a 32bits limited address space */
686	WREG32_MC(R_000100_MCCFG_FB_LOCATION,
687			S_000100_MC_FB_START(rdev->mc.vram_start >> 16) |
688			S_000100_MC_FB_TOP(rdev->mc.vram_end >> 16));
689	WREG32(R_000134_HDP_FB_LOCATION,
690		S_000134_HDP_FB_START(rdev->mc.vram_start >> 16));
691
692	rv515_mc_resume(rdev, &save);
693}
694
695static int rs690_startup(struct radeon_device *rdev)
696{
697	int r;
698
699	rs690_mc_program(rdev);
700	/* Resume clock */
701	rv515_clock_startup(rdev);
702	/* Initialize GPU configuration (# pipes, ...) */
703	rs690_gpu_init(rdev);
704	/* Initialize GART (initialize after TTM so we can allocate
705	 * memory through TTM but finalize after TTM) */
706	r = rs400_gart_enable(rdev);
707	if (r)
708		return r;
709
710	/* allocate wb buffer */
711	r = radeon_wb_init(rdev);
712	if (r)
713		return r;
714
715	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
716	if (r) {
717		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
718		return r;
719	}
720
721	/* Enable IRQ */
722	if (!rdev->irq.installed) {
723		r = radeon_irq_kms_init(rdev);
724		if (r)
725			return r;
726	}
727
728	rs600_irq_set(rdev);
729	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
730	/* 1M ring buffer */
731	r = r100_cp_init(rdev, 1024 * 1024);
732	if (r) {
733		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
734		return r;
735	}
736
737	r = radeon_ib_pool_init(rdev);
738	if (r) {
739		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
740		return r;
741	}
742
743	r = radeon_audio_init(rdev);
744	if (r) {
745		dev_err(rdev->dev, "failed initializing audio\n");
746		return r;
747	}
748
749	return 0;
750}
751
752int rs690_resume(struct radeon_device *rdev)
753{
754	int r;
755
756	/* Make sur GART are not working */
757	rs400_gart_disable(rdev);
758	/* Resume clock before doing reset */
759	rv515_clock_startup(rdev);
760	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
761	if (radeon_asic_reset(rdev)) {
762		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
763			RREG32(R_000E40_RBBM_STATUS),
764			RREG32(R_0007C0_CP_STAT));
765	}
766	/* post */
767	atom_asic_init(rdev->mode_info.atom_context);
768	/* Resume clock after posting */
769	rv515_clock_startup(rdev);
770	/* Initialize surface registers */
771	radeon_surface_init(rdev);
772
773	rdev->accel_working = true;
774	r = rs690_startup(rdev);
775	if (r) {
776		rdev->accel_working = false;
777	}
778	return r;
779}
780
781int rs690_suspend(struct radeon_device *rdev)
782{
783	radeon_pm_suspend(rdev);
784	radeon_audio_fini(rdev);
785	r100_cp_disable(rdev);
786	radeon_wb_disable(rdev);
787	rs600_irq_disable(rdev);
788	rs400_gart_disable(rdev);
789	return 0;
790}
791
792void rs690_fini(struct radeon_device *rdev)
793{
794	radeon_pm_fini(rdev);
795	radeon_audio_fini(rdev);
796	r100_cp_fini(rdev);
797	radeon_wb_fini(rdev);
798	radeon_ib_pool_fini(rdev);
799	radeon_gem_fini(rdev);
800	rs400_gart_fini(rdev);
801	radeon_irq_kms_fini(rdev);
802	radeon_fence_driver_fini(rdev);
803	radeon_bo_fini(rdev);
804	radeon_atombios_fini(rdev);
805	kfree(rdev->bios);
806	rdev->bios = NULL;
807}
808
809int rs690_init(struct radeon_device *rdev)
810{
811	int r;
812
813	/* Disable VGA */
814	rv515_vga_render_disable(rdev);
815	/* Initialize scratch registers */
816	radeon_scratch_init(rdev);
817	/* Initialize surface registers */
818	radeon_surface_init(rdev);
819	/* restore some register to sane defaults */
820	r100_restore_sanity(rdev);
821	/* TODO: disable VGA need to use VGA request */
822	/* BIOS*/
823	if (!radeon_get_bios(rdev)) {
824		if (ASIC_IS_AVIVO(rdev))
825			return -EINVAL;
826	}
827	if (rdev->is_atom_bios) {
828		r = radeon_atombios_init(rdev);
829		if (r)
830			return r;
831	} else {
832		dev_err(rdev->dev, "Expecting atombios for RV515 GPU\n");
833		return -EINVAL;
834	}
835	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
836	if (radeon_asic_reset(rdev)) {
837		dev_warn(rdev->dev,
838			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
839			RREG32(R_000E40_RBBM_STATUS),
840			RREG32(R_0007C0_CP_STAT));
841	}
842	/* check if cards are posted or not */
843	if (radeon_boot_test_post_card(rdev) == false)
844		return -EINVAL;
845
846	/* Initialize clocks */
847	radeon_get_clock_info(rdev->ddev);
848	/* initialize memory controller */
849	rs690_mc_init(rdev);
850	rv515_debugfs(rdev);
851	/* Fence driver */
852	r = radeon_fence_driver_init(rdev);
853	if (r)
854		return r;
855	/* Memory manager */
856	r = radeon_bo_init(rdev);
857	if (r)
858		return r;
859	r = rs400_gart_init(rdev);
860	if (r)
861		return r;
862	rs600_set_safe_registers(rdev);
863
864	/* Initialize power management */
865	radeon_pm_init(rdev);
866
867	rdev->accel_working = true;
868	r = rs690_startup(rdev);
869	if (r) {
870		/* Somethings want wront with the accel init stop accel */
871		dev_err(rdev->dev, "Disabling GPU acceleration\n");
872		r100_cp_fini(rdev);
873		radeon_wb_fini(rdev);
874		radeon_ib_pool_fini(rdev);
875		rs400_gart_fini(rdev);
876		radeon_irq_kms_fini(rdev);
877		rdev->accel_working = false;
878	}
879	return 0;
880}
881