1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2016-2018, The Linux Foundation. All rights reserved.
3 */
4
5#define pr_fmt(fmt)	"[drm:%s:%d] " fmt, __func__, __LINE__
6
7#include <linux/debugfs.h>
8#include <linux/errno.h>
9#include <linux/mutex.h>
10#include <linux/pm_opp.h>
11#include <linux/sort.h>
12#include <linux/clk.h>
13#include <linux/bitmap.h>
14
15#include "dpu_kms.h"
16#include "dpu_trace.h"
17#include "dpu_crtc.h"
18#include "dpu_core_perf.h"
19
20/**
21 * enum dpu_perf_mode - performance tuning mode
22 * @DPU_PERF_MODE_NORMAL: performance controlled by user mode client
23 * @DPU_PERF_MODE_MINIMUM: performance bounded by minimum setting
24 * @DPU_PERF_MODE_FIXED: performance bounded by fixed setting
25 * @DPU_PERF_MODE_MAX: maximum value, used for error checking
26 */
27enum dpu_perf_mode {
28	DPU_PERF_MODE_NORMAL,
29	DPU_PERF_MODE_MINIMUM,
30	DPU_PERF_MODE_FIXED,
31	DPU_PERF_MODE_MAX
32};
33
34/**
35 * _dpu_core_perf_calc_bw() - to calculate BW per crtc
36 * @perf_cfg: performance configuration
37 * @crtc: pointer to a crtc
38 * Return: returns aggregated BW for all planes in crtc.
39 */
40static u64 _dpu_core_perf_calc_bw(const struct dpu_perf_cfg *perf_cfg,
41		struct drm_crtc *crtc)
42{
43	struct drm_plane *plane;
44	struct dpu_plane_state *pstate;
45	u64 crtc_plane_bw = 0;
46	u32 bw_factor;
47
48	drm_atomic_crtc_for_each_plane(plane, crtc) {
49		pstate = to_dpu_plane_state(plane->state);
50		if (!pstate)
51			continue;
52
53		crtc_plane_bw += pstate->plane_fetch_bw;
54	}
55
56	bw_factor = perf_cfg->bw_inefficiency_factor;
57	if (bw_factor) {
58		crtc_plane_bw *= bw_factor;
59		do_div(crtc_plane_bw, 100);
60	}
61
62	return crtc_plane_bw;
63}
64
65/**
66 * _dpu_core_perf_calc_clk() - to calculate clock per crtc
67 * @perf_cfg: performance configuration
68 * @crtc: pointer to a crtc
69 * @state: pointer to a crtc state
70 * Return: returns max clk for all planes in crtc.
71 */
72static u64 _dpu_core_perf_calc_clk(const struct dpu_perf_cfg *perf_cfg,
73		struct drm_crtc *crtc, struct drm_crtc_state *state)
74{
75	struct drm_plane *plane;
76	struct dpu_plane_state *pstate;
77	struct drm_display_mode *mode;
78	u64 crtc_clk;
79	u32 clk_factor;
80
81	mode = &state->adjusted_mode;
82
83	crtc_clk = mode->vtotal * mode->hdisplay * drm_mode_vrefresh(mode);
84
85	drm_atomic_crtc_for_each_plane(plane, crtc) {
86		pstate = to_dpu_plane_state(plane->state);
87		if (!pstate)
88			continue;
89
90		crtc_clk = max(pstate->plane_clk, crtc_clk);
91	}
92
93	clk_factor = perf_cfg->clk_inefficiency_factor;
94	if (clk_factor) {
95		crtc_clk *= clk_factor;
96		do_div(crtc_clk, 100);
97	}
98
99	return crtc_clk;
100}
101
102static struct dpu_kms *_dpu_crtc_get_kms(struct drm_crtc *crtc)
103{
104	struct msm_drm_private *priv;
105	priv = crtc->dev->dev_private;
106	return to_dpu_kms(priv->kms);
107}
108
109static void _dpu_core_perf_calc_crtc(const struct dpu_core_perf *core_perf,
110				     struct drm_crtc *crtc,
111				     struct drm_crtc_state *state,
112				     struct dpu_core_perf_params *perf)
113{
114	const struct dpu_perf_cfg *perf_cfg = core_perf->perf_cfg;
115
116	if (!perf_cfg || !crtc || !state || !perf) {
117		DPU_ERROR("invalid parameters\n");
118		return;
119	}
120
121	memset(perf, 0, sizeof(struct dpu_core_perf_params));
122
123	if (core_perf->perf_tune.mode == DPU_PERF_MODE_MINIMUM) {
124		perf->bw_ctl = 0;
125		perf->max_per_pipe_ib = 0;
126		perf->core_clk_rate = 0;
127	} else if (core_perf->perf_tune.mode == DPU_PERF_MODE_FIXED) {
128		perf->bw_ctl = core_perf->fix_core_ab_vote;
129		perf->max_per_pipe_ib = core_perf->fix_core_ib_vote;
130		perf->core_clk_rate = core_perf->fix_core_clk_rate;
131	} else {
132		perf->bw_ctl = _dpu_core_perf_calc_bw(perf_cfg, crtc);
133		perf->max_per_pipe_ib = perf_cfg->min_dram_ib;
134		perf->core_clk_rate = _dpu_core_perf_calc_clk(perf_cfg, crtc, state);
135	}
136
137	DRM_DEBUG_ATOMIC(
138		"crtc=%d clk_rate=%llu core_ib=%llu core_ab=%llu\n",
139			crtc->base.id, perf->core_clk_rate,
140			perf->max_per_pipe_ib, perf->bw_ctl);
141}
142
143int dpu_core_perf_crtc_check(struct drm_crtc *crtc,
144		struct drm_crtc_state *state)
145{
146	u32 bw, threshold;
147	u64 bw_sum_of_intfs = 0;
148	enum dpu_crtc_client_type curr_client_type;
149	struct dpu_crtc_state *dpu_cstate;
150	struct drm_crtc *tmp_crtc;
151	struct dpu_kms *kms;
152
153	if (!crtc || !state) {
154		DPU_ERROR("invalid crtc\n");
155		return -EINVAL;
156	}
157
158	kms = _dpu_crtc_get_kms(crtc);
159
160	/* we only need bandwidth check on real-time clients (interfaces) */
161	if (dpu_crtc_get_client_type(crtc) == NRT_CLIENT)
162		return 0;
163
164	dpu_cstate = to_dpu_crtc_state(state);
165
166	/* obtain new values */
167	_dpu_core_perf_calc_crtc(&kms->perf, crtc, state, &dpu_cstate->new_perf);
168
169	bw_sum_of_intfs = dpu_cstate->new_perf.bw_ctl;
170	curr_client_type = dpu_crtc_get_client_type(crtc);
171
172	drm_for_each_crtc(tmp_crtc, crtc->dev) {
173		if (tmp_crtc->enabled &&
174		    dpu_crtc_get_client_type(tmp_crtc) == curr_client_type &&
175		    tmp_crtc != crtc) {
176			struct dpu_crtc_state *tmp_cstate =
177				to_dpu_crtc_state(tmp_crtc->state);
178
179			DRM_DEBUG_ATOMIC("crtc:%d bw:%llu ctrl:%d\n",
180					 tmp_crtc->base.id, tmp_cstate->new_perf.bw_ctl,
181					 tmp_cstate->bw_control);
182
183			bw_sum_of_intfs += tmp_cstate->new_perf.bw_ctl;
184		}
185
186		/* convert bandwidth to kb */
187		bw = DIV_ROUND_UP_ULL(bw_sum_of_intfs, 1000);
188		DRM_DEBUG_ATOMIC("calculated bandwidth=%uk\n", bw);
189
190		threshold = kms->perf.perf_cfg->max_bw_high;
191
192		DRM_DEBUG_ATOMIC("final threshold bw limit = %d\n", threshold);
193
194		if (!threshold) {
195			DPU_ERROR("no bandwidth limits specified\n");
196			return -E2BIG;
197		} else if (bw > threshold) {
198			DPU_ERROR("exceeds bandwidth: %ukb > %ukb\n", bw,
199					threshold);
200			return -E2BIG;
201		}
202	}
203
204	return 0;
205}
206
207static int _dpu_core_perf_crtc_update_bus(struct dpu_kms *kms,
208		struct drm_crtc *crtc)
209{
210	struct dpu_core_perf_params perf = { 0 };
211	enum dpu_crtc_client_type curr_client_type
212					= dpu_crtc_get_client_type(crtc);
213	struct drm_crtc *tmp_crtc;
214	struct dpu_crtc_state *dpu_cstate;
215	int i, ret = 0;
216	u64 avg_bw;
217
218	if (!kms->num_paths)
219		return 0;
220
221	drm_for_each_crtc(tmp_crtc, crtc->dev) {
222		if (tmp_crtc->enabled &&
223			curr_client_type ==
224				dpu_crtc_get_client_type(tmp_crtc)) {
225			dpu_cstate = to_dpu_crtc_state(tmp_crtc->state);
226
227			perf.max_per_pipe_ib = max(perf.max_per_pipe_ib,
228					dpu_cstate->new_perf.max_per_pipe_ib);
229
230			perf.bw_ctl += dpu_cstate->new_perf.bw_ctl;
231
232			DRM_DEBUG_ATOMIC("crtc=%d bw=%llu paths:%d\n",
233				  tmp_crtc->base.id,
234				  dpu_cstate->new_perf.bw_ctl, kms->num_paths);
235		}
236	}
237
238	avg_bw = perf.bw_ctl;
239	do_div(avg_bw, (kms->num_paths * 1000)); /*Bps_to_icc*/
240
241	for (i = 0; i < kms->num_paths; i++)
242		icc_set_bw(kms->path[i], avg_bw, perf.max_per_pipe_ib);
243
244	return ret;
245}
246
247/**
248 * dpu_core_perf_crtc_release_bw() - request zero bandwidth
249 * @crtc: pointer to a crtc
250 *
251 * Function checks a state variable for the crtc, if all pending commit
252 * requests are done, meaning no more bandwidth is needed, release
253 * bandwidth request.
254 */
255void dpu_core_perf_crtc_release_bw(struct drm_crtc *crtc)
256{
257	struct dpu_crtc *dpu_crtc;
258	struct dpu_kms *kms;
259
260	if (!crtc) {
261		DPU_ERROR("invalid crtc\n");
262		return;
263	}
264
265	kms = _dpu_crtc_get_kms(crtc);
266	dpu_crtc = to_dpu_crtc(crtc);
267
268	if (atomic_dec_return(&kms->bandwidth_ref) > 0)
269		return;
270
271	/* Release the bandwidth */
272	if (kms->perf.enable_bw_release) {
273		trace_dpu_cmd_release_bw(crtc->base.id);
274		DRM_DEBUG_ATOMIC("Release BW crtc=%d\n", crtc->base.id);
275		dpu_crtc->cur_perf.bw_ctl = 0;
276		_dpu_core_perf_crtc_update_bus(kms, crtc);
277	}
278}
279
280static u64 _dpu_core_perf_get_core_clk_rate(struct dpu_kms *kms)
281{
282	u64 clk_rate;
283	struct drm_crtc *crtc;
284	struct dpu_crtc_state *dpu_cstate;
285
286	if (kms->perf.perf_tune.mode == DPU_PERF_MODE_FIXED)
287		return kms->perf.fix_core_clk_rate;
288
289	if (kms->perf.perf_tune.mode == DPU_PERF_MODE_MINIMUM)
290		return kms->perf.max_core_clk_rate;
291
292	clk_rate = 0;
293	drm_for_each_crtc(crtc, kms->dev) {
294		if (crtc->enabled) {
295			dpu_cstate = to_dpu_crtc_state(crtc->state);
296			clk_rate = max(dpu_cstate->new_perf.core_clk_rate,
297							clk_rate);
298		}
299	}
300
301	return clk_rate;
302}
303
304int dpu_core_perf_crtc_update(struct drm_crtc *crtc,
305			      int params_changed)
306{
307	struct dpu_core_perf_params *new, *old;
308	bool update_bus = false, update_clk = false;
309	u64 clk_rate = 0;
310	struct dpu_crtc *dpu_crtc;
311	struct dpu_crtc_state *dpu_cstate;
312	struct dpu_kms *kms;
313	int ret;
314
315	if (!crtc) {
316		DPU_ERROR("invalid crtc\n");
317		return -EINVAL;
318	}
319
320	kms = _dpu_crtc_get_kms(crtc);
321
322	dpu_crtc = to_dpu_crtc(crtc);
323	dpu_cstate = to_dpu_crtc_state(crtc->state);
324
325	DRM_DEBUG_ATOMIC("crtc:%d enabled:%d core_clk:%llu\n",
326			crtc->base.id, crtc->enabled, kms->perf.core_clk_rate);
327
328	old = &dpu_crtc->cur_perf;
329	new = &dpu_cstate->new_perf;
330
331	if (crtc->enabled) {
332		/*
333		 * cases for bus bandwidth update.
334		 * 1. new bandwidth vote - "ab or ib vote" is higher
335		 *    than current vote for update request.
336		 * 2. new bandwidth vote - "ab or ib vote" is lower
337		 *    than current vote at end of commit or stop.
338		 */
339		if ((params_changed && ((new->bw_ctl > old->bw_ctl) ||
340			(new->max_per_pipe_ib > old->max_per_pipe_ib)))	||
341			(!params_changed && ((new->bw_ctl < old->bw_ctl) ||
342			(new->max_per_pipe_ib < old->max_per_pipe_ib)))) {
343			DRM_DEBUG_ATOMIC("crtc=%d p=%d new_bw=%llu,old_bw=%llu\n",
344				crtc->base.id, params_changed,
345				new->bw_ctl, old->bw_ctl);
346			old->bw_ctl = new->bw_ctl;
347			old->max_per_pipe_ib = new->max_per_pipe_ib;
348			update_bus = true;
349		}
350
351		if ((params_changed && new->core_clk_rate > old->core_clk_rate) ||
352		    (!params_changed && new->core_clk_rate < old->core_clk_rate)) {
353			old->core_clk_rate = new->core_clk_rate;
354			update_clk = true;
355		}
356	} else {
357		DRM_DEBUG_ATOMIC("crtc=%d disable\n", crtc->base.id);
358		memset(old, 0, sizeof(*old));
359		update_bus = true;
360		update_clk = true;
361	}
362
363	trace_dpu_perf_crtc_update(crtc->base.id, new->bw_ctl,
364		new->core_clk_rate, !crtc->enabled, update_bus, update_clk);
365
366	if (update_bus) {
367		ret = _dpu_core_perf_crtc_update_bus(kms, crtc);
368		if (ret) {
369			DPU_ERROR("crtc-%d: failed to update bus bw vote\n",
370				  crtc->base.id);
371			return ret;
372		}
373	}
374
375	/*
376	 * Update the clock after bandwidth vote to ensure
377	 * bandwidth is available before clock rate is increased.
378	 */
379	if (update_clk) {
380		clk_rate = _dpu_core_perf_get_core_clk_rate(kms);
381
382		DRM_DEBUG_ATOMIC("clk:%llu\n", clk_rate);
383
384		trace_dpu_core_perf_update_clk(kms->dev, !crtc->enabled, clk_rate);
385
386		clk_rate = min(clk_rate, kms->perf.max_core_clk_rate);
387		ret = dev_pm_opp_set_rate(&kms->pdev->dev, clk_rate);
388		if (ret) {
389			DPU_ERROR("failed to set core clock rate %llu\n", clk_rate);
390			return ret;
391		}
392
393		kms->perf.core_clk_rate = clk_rate;
394		DRM_DEBUG_ATOMIC("update clk rate = %lld HZ\n", clk_rate);
395	}
396	return 0;
397}
398
399#ifdef CONFIG_DEBUG_FS
400
401static ssize_t _dpu_core_perf_mode_write(struct file *file,
402		    const char __user *user_buf, size_t count, loff_t *ppos)
403{
404	struct dpu_core_perf *perf = file->private_data;
405	u32 perf_mode = 0;
406	int ret;
407
408	ret = kstrtouint_from_user(user_buf, count, 0, &perf_mode);
409	if (ret)
410		return ret;
411
412	if (perf_mode >= DPU_PERF_MODE_MAX)
413		return -EINVAL;
414
415	if (perf_mode == DPU_PERF_MODE_FIXED) {
416		DRM_INFO("fix performance mode\n");
417	} else if (perf_mode == DPU_PERF_MODE_MINIMUM) {
418		/* run the driver with max clk and BW vote */
419		DRM_INFO("minimum performance mode\n");
420	} else if (perf_mode == DPU_PERF_MODE_NORMAL) {
421		/* reset the perf tune params to 0 */
422		DRM_INFO("normal performance mode\n");
423	}
424	perf->perf_tune.mode = perf_mode;
425
426	return count;
427}
428
429static ssize_t _dpu_core_perf_mode_read(struct file *file,
430			char __user *buff, size_t count, loff_t *ppos)
431{
432	struct dpu_core_perf *perf = file->private_data;
433	int len;
434	char buf[128];
435
436	len = scnprintf(buf, sizeof(buf),
437			"mode %d\n",
438			perf->perf_tune.mode);
439
440	return simple_read_from_buffer(buff, count, ppos, buf, len);
441}
442
443static const struct file_operations dpu_core_perf_mode_fops = {
444	.open = simple_open,
445	.read = _dpu_core_perf_mode_read,
446	.write = _dpu_core_perf_mode_write,
447};
448
449int dpu_core_perf_debugfs_init(struct dpu_kms *dpu_kms, struct dentry *parent)
450{
451	struct dpu_core_perf *perf = &dpu_kms->perf;
452	struct dentry *entry;
453
454	entry = debugfs_create_dir("core_perf", parent);
455
456	debugfs_create_u64("max_core_clk_rate", 0600, entry,
457			&perf->max_core_clk_rate);
458	debugfs_create_u64("core_clk_rate", 0600, entry,
459			&perf->core_clk_rate);
460	debugfs_create_u32("enable_bw_release", 0600, entry,
461			(u32 *)&perf->enable_bw_release);
462	debugfs_create_u32("threshold_low", 0400, entry,
463			(u32 *)&perf->perf_cfg->max_bw_low);
464	debugfs_create_u32("threshold_high", 0400, entry,
465			(u32 *)&perf->perf_cfg->max_bw_high);
466	debugfs_create_u32("min_core_ib", 0400, entry,
467			(u32 *)&perf->perf_cfg->min_core_ib);
468	debugfs_create_u32("min_llcc_ib", 0400, entry,
469			(u32 *)&perf->perf_cfg->min_llcc_ib);
470	debugfs_create_u32("min_dram_ib", 0400, entry,
471			(u32 *)&perf->perf_cfg->min_dram_ib);
472	debugfs_create_file("perf_mode", 0600, entry,
473			(u32 *)perf, &dpu_core_perf_mode_fops);
474	debugfs_create_u64("fix_core_clk_rate", 0600, entry,
475			&perf->fix_core_clk_rate);
476	debugfs_create_u64("fix_core_ib_vote", 0600, entry,
477			&perf->fix_core_ib_vote);
478	debugfs_create_u64("fix_core_ab_vote", 0600, entry,
479			&perf->fix_core_ab_vote);
480
481	return 0;
482}
483#endif
484
485int dpu_core_perf_init(struct dpu_core_perf *perf,
486		const struct dpu_perf_cfg *perf_cfg,
487		unsigned long max_core_clk_rate)
488{
489	perf->perf_cfg = perf_cfg;
490	perf->max_core_clk_rate = max_core_clk_rate;
491
492	return 0;
493}
494