1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2014, Fuzhou Rockchip Electronics Co., Ltd
4 */
5
6 #include <linux/module.h>
7 #include <linux/platform_device.h>
8 #include <linux/clk.h>
9 #include <linux/mmc/host.h>
10 #include <linux/of_address.h>
11 #include <linux/mmc/slot-gpio.h>
12 #include <linux/pm_runtime.h>
13 #include <linux/slab.h>
14
15 #include "dw_mmc.h"
16 #include "dw_mmc-pltfm.h"
17
18 #define RK3288_CLKGEN_DIV 2
19 #define SDMMC_TIMING_CON0 0x130
20 #define SDMMC_TIMING_CON1 0x134
21 #define ROCKCHIP_MMC_DELAY_SEL BIT(10)
22 #define ROCKCHIP_MMC_DEGREE_MASK 0x3
23 #define ROCKCHIP_MMC_DEGREE_OFFSET 1
24 #define ROCKCHIP_MMC_DELAYNUM_OFFSET 2
25 #define ROCKCHIP_MMC_DELAYNUM_MASK (0xff << ROCKCHIP_MMC_DELAYNUM_OFFSET)
26 #define ROCKCHIP_MMC_DELAY_ELEMENT_PSEC 60
27 #define HIWORD_UPDATE(val, mask, shift) \
28 ((val) << (shift) | (mask) << ((shift) + 16))
29
30 static const unsigned int freqs[] = { 100000, 200000, 300000, 400000 };
31
32 struct dw_mci_rockchip_priv_data {
33 struct clk *drv_clk;
34 struct clk *sample_clk;
35 int default_sample_phase;
36 int num_phases;
37 bool internal_phase;
38 };
39
40 /*
41 * Each fine delay is between 44ps-77ps. Assume each fine delay is 60ps to
42 * simplify calculations. So 45degs could be anywhere between 33deg and 57.8deg.
43 */
rockchip_mmc_get_internal_phase(struct dw_mci * host,bool sample)44 static int rockchip_mmc_get_internal_phase(struct dw_mci *host, bool sample)
45 {
46 unsigned long rate = clk_get_rate(host->ciu_clk);
47 u32 raw_value;
48 u16 degrees;
49 u32 delay_num = 0;
50
51 /* Constant signal, no measurable phase shift */
52 if (!rate)
53 return 0;
54
55 if (sample)
56 raw_value = mci_readl(host, TIMING_CON1);
57 else
58 raw_value = mci_readl(host, TIMING_CON0);
59
60 raw_value >>= ROCKCHIP_MMC_DEGREE_OFFSET;
61 degrees = (raw_value & ROCKCHIP_MMC_DEGREE_MASK) * 90;
62
63 if (raw_value & ROCKCHIP_MMC_DELAY_SEL) {
64 /* degrees/delaynum * 1000000 */
65 unsigned long factor = (ROCKCHIP_MMC_DELAY_ELEMENT_PSEC / 10) *
66 36 * (rate / 10000);
67
68 delay_num = (raw_value & ROCKCHIP_MMC_DELAYNUM_MASK);
69 delay_num >>= ROCKCHIP_MMC_DELAYNUM_OFFSET;
70 degrees += DIV_ROUND_CLOSEST(delay_num * factor, 1000000);
71 }
72
73 return degrees % 360;
74 }
75
rockchip_mmc_get_phase(struct dw_mci * host,bool sample)76 static int rockchip_mmc_get_phase(struct dw_mci *host, bool sample)
77 {
78 struct dw_mci_rockchip_priv_data *priv = host->priv;
79 struct clk *clock = sample ? priv->sample_clk : priv->drv_clk;
80
81 if (priv->internal_phase)
82 return rockchip_mmc_get_internal_phase(host, sample);
83 else
84 return clk_get_phase(clock);
85 }
86
rockchip_mmc_set_internal_phase(struct dw_mci * host,bool sample,int degrees)87 static int rockchip_mmc_set_internal_phase(struct dw_mci *host, bool sample, int degrees)
88 {
89 unsigned long rate = clk_get_rate(host->ciu_clk);
90 u8 nineties, remainder;
91 u8 delay_num;
92 u32 raw_value;
93 u32 delay;
94
95 /*
96 * The below calculation is based on the output clock from
97 * MMC host to the card, which expects the phase clock inherits
98 * the clock rate from its parent, namely the output clock
99 * provider of MMC host. However, things may go wrong if
100 * (1) It is orphan.
101 * (2) It is assigned to the wrong parent.
102 *
103 * This check help debug the case (1), which seems to be the
104 * most likely problem we often face and which makes it difficult
105 * for people to debug unstable mmc tuning results.
106 */
107 if (!rate) {
108 dev_err(host->dev, "%s: invalid clk rate\n", __func__);
109 return -EINVAL;
110 }
111
112 nineties = degrees / 90;
113 remainder = (degrees % 90);
114
115 /*
116 * Due to the inexact nature of the "fine" delay, we might
117 * actually go non-monotonic. We don't go _too_ monotonic
118 * though, so we should be OK. Here are options of how we may
119 * work:
120 *
121 * Ideally we end up with:
122 * 1.0, 2.0, ..., 69.0, 70.0, ..., 89.0, 90.0
123 *
124 * On one extreme (if delay is actually 44ps):
125 * .73, 1.5, ..., 50.6, 51.3, ..., 65.3, 90.0
126 * The other (if delay is actually 77ps):
127 * 1.3, 2.6, ..., 88.6. 89.8, ..., 114.0, 90
128 *
129 * It's possible we might make a delay that is up to 25
130 * degrees off from what we think we're making. That's OK
131 * though because we should be REALLY far from any bad range.
132 */
133
134 /*
135 * Convert to delay; do a little extra work to make sure we
136 * don't overflow 32-bit / 64-bit numbers.
137 */
138 delay = 10000000; /* PSECS_PER_SEC / 10000 / 10 */
139 delay *= remainder;
140 delay = DIV_ROUND_CLOSEST(delay,
141 (rate / 1000) * 36 *
142 (ROCKCHIP_MMC_DELAY_ELEMENT_PSEC / 10));
143
144 delay_num = (u8) min_t(u32, delay, 255);
145
146 raw_value = delay_num ? ROCKCHIP_MMC_DELAY_SEL : 0;
147 raw_value |= delay_num << ROCKCHIP_MMC_DELAYNUM_OFFSET;
148 raw_value |= nineties;
149
150 if (sample)
151 mci_writel(host, TIMING_CON1, HIWORD_UPDATE(raw_value, 0x07ff, 1));
152 else
153 mci_writel(host, TIMING_CON0, HIWORD_UPDATE(raw_value, 0x07ff, 1));
154
155 dev_dbg(host->dev, "set %s_phase(%d) delay_nums=%u actual_degrees=%d\n",
156 sample ? "sample" : "drv", degrees, delay_num,
157 rockchip_mmc_get_phase(host, sample)
158 );
159
160 return 0;
161 }
162
rockchip_mmc_set_phase(struct dw_mci * host,bool sample,int degrees)163 static int rockchip_mmc_set_phase(struct dw_mci *host, bool sample, int degrees)
164 {
165 struct dw_mci_rockchip_priv_data *priv = host->priv;
166 struct clk *clock = sample ? priv->sample_clk : priv->drv_clk;
167
168 if (priv->internal_phase)
169 return rockchip_mmc_set_internal_phase(host, sample, degrees);
170 else
171 return clk_set_phase(clock, degrees);
172 }
173
dw_mci_rk3288_set_ios(struct dw_mci * host,struct mmc_ios * ios)174 static void dw_mci_rk3288_set_ios(struct dw_mci *host, struct mmc_ios *ios)
175 {
176 struct dw_mci_rockchip_priv_data *priv = host->priv;
177 int ret;
178 unsigned int cclkin;
179 u32 bus_hz;
180
181 if (ios->clock == 0)
182 return;
183
184 /*
185 * cclkin: source clock of mmc controller
186 * bus_hz: card interface clock generated by CLKGEN
187 * bus_hz = cclkin / RK3288_CLKGEN_DIV
188 * ios->clock = (div == 0) ? bus_hz : (bus_hz / (2 * div))
189 *
190 * Note: div can only be 0 or 1, but div must be set to 1 for eMMC
191 * DDR52 8-bit mode.
192 */
193 if (ios->bus_width == MMC_BUS_WIDTH_8 &&
194 ios->timing == MMC_TIMING_MMC_DDR52)
195 cclkin = 2 * ios->clock * RK3288_CLKGEN_DIV;
196 else
197 cclkin = ios->clock * RK3288_CLKGEN_DIV;
198
199 ret = clk_set_rate(host->ciu_clk, cclkin);
200 if (ret)
201 dev_warn(host->dev, "failed to set rate %uHz err: %d\n", cclkin, ret);
202
203 bus_hz = clk_get_rate(host->ciu_clk) / RK3288_CLKGEN_DIV;
204 if (bus_hz != host->bus_hz) {
205 host->bus_hz = bus_hz;
206 /* force dw_mci_setup_bus() */
207 host->current_speed = 0;
208 }
209
210 /* Make sure we use phases which we can enumerate with */
211 if (!IS_ERR(priv->sample_clk) && ios->timing <= MMC_TIMING_SD_HS)
212 rockchip_mmc_set_phase(host, true, priv->default_sample_phase);
213
214 /*
215 * Set the drive phase offset based on speed mode to achieve hold times.
216 *
217 * NOTE: this is _not_ a value that is dynamically tuned and is also
218 * _not_ a value that will vary from board to board. It is a value
219 * that could vary between different SoC models if they had massively
220 * different output clock delays inside their dw_mmc IP block (delay_o),
221 * but since it's OK to overshoot a little we don't need to do complex
222 * calculations and can pick values that will just work for everyone.
223 *
224 * When picking values we'll stick with picking 0/90/180/270 since
225 * those can be made very accurately on all known Rockchip SoCs.
226 *
227 * Note that these values match values from the DesignWare Databook
228 * tables for the most part except for SDR12 and "ID mode". For those
229 * two modes the databook calculations assume a clock in of 50MHz. As
230 * seen above, we always use a clock in rate that is exactly the
231 * card's input clock (times RK3288_CLKGEN_DIV, but that gets divided
232 * back out before the controller sees it).
233 *
234 * From measurement of a single device, it appears that delay_o is
235 * about .5 ns. Since we try to leave a bit of margin, it's expected
236 * that numbers here will be fine even with much larger delay_o
237 * (the 1.4 ns assumed by the DesignWare Databook would result in the
238 * same results, for instance).
239 */
240 if (!IS_ERR(priv->drv_clk)) {
241 int phase;
242
243 /*
244 * In almost all cases a 90 degree phase offset will provide
245 * sufficient hold times across all valid input clock rates
246 * assuming delay_o is not absurd for a given SoC. We'll use
247 * that as a default.
248 */
249 phase = 90;
250
251 switch (ios->timing) {
252 case MMC_TIMING_MMC_DDR52:
253 /*
254 * Since clock in rate with MMC_DDR52 is doubled when
255 * bus width is 8 we need to double the phase offset
256 * to get the same timings.
257 */
258 if (ios->bus_width == MMC_BUS_WIDTH_8)
259 phase = 180;
260 break;
261 case MMC_TIMING_UHS_SDR104:
262 case MMC_TIMING_MMC_HS200:
263 /*
264 * In the case of 150 MHz clock (typical max for
265 * Rockchip SoCs), 90 degree offset will add a delay
266 * of 1.67 ns. That will meet min hold time of .8 ns
267 * as long as clock output delay is < .87 ns. On
268 * SoCs measured this seems to be OK, but it doesn't
269 * hurt to give margin here, so we use 180.
270 */
271 phase = 180;
272 break;
273 }
274
275 rockchip_mmc_set_phase(host, false, phase);
276 }
277 }
278
279 #define TUNING_ITERATION_TO_PHASE(i, num_phases) \
280 (DIV_ROUND_UP((i) * 360, num_phases))
281
dw_mci_rk3288_execute_tuning(struct dw_mci_slot * slot,u32 opcode)282 static int dw_mci_rk3288_execute_tuning(struct dw_mci_slot *slot, u32 opcode)
283 {
284 struct dw_mci *host = slot->host;
285 struct dw_mci_rockchip_priv_data *priv = host->priv;
286 struct mmc_host *mmc = slot->mmc;
287 int ret = 0;
288 int i;
289 bool v, prev_v = 0, first_v;
290 struct range_t {
291 int start;
292 int end; /* inclusive */
293 };
294 struct range_t *ranges;
295 unsigned int range_count = 0;
296 int longest_range_len = -1;
297 int longest_range = -1;
298 int middle_phase;
299 int phase;
300
301 if (IS_ERR(priv->sample_clk)) {
302 dev_err(host->dev, "Tuning clock (sample_clk) not defined.\n");
303 return -EIO;
304 }
305
306 ranges = kmalloc_array(priv->num_phases / 2 + 1,
307 sizeof(*ranges), GFP_KERNEL);
308 if (!ranges)
309 return -ENOMEM;
310
311 /* Try each phase and extract good ranges */
312 for (i = 0; i < priv->num_phases; ) {
313 rockchip_mmc_set_phase(host, true,
314 TUNING_ITERATION_TO_PHASE(
315 i,
316 priv->num_phases));
317
318 v = !mmc_send_tuning(mmc, opcode, NULL);
319
320 if (i == 0)
321 first_v = v;
322
323 if ((!prev_v) && v) {
324 range_count++;
325 ranges[range_count-1].start = i;
326 }
327 if (v) {
328 ranges[range_count-1].end = i;
329 i++;
330 } else if (i == priv->num_phases - 1) {
331 /* No extra skipping rules if we're at the end */
332 i++;
333 } else {
334 /*
335 * No need to check too close to an invalid
336 * one since testing bad phases is slow. Skip
337 * 20 degrees.
338 */
339 i += DIV_ROUND_UP(20 * priv->num_phases, 360);
340
341 /* Always test the last one */
342 if (i >= priv->num_phases)
343 i = priv->num_phases - 1;
344 }
345
346 prev_v = v;
347 }
348
349 if (range_count == 0) {
350 dev_warn(host->dev, "All phases bad!");
351 ret = -EIO;
352 goto free;
353 }
354
355 /* wrap around case, merge the end points */
356 if ((range_count > 1) && first_v && v) {
357 ranges[0].start = ranges[range_count-1].start;
358 range_count--;
359 }
360
361 if (ranges[0].start == 0 && ranges[0].end == priv->num_phases - 1) {
362 rockchip_mmc_set_phase(host, true, priv->default_sample_phase);
363
364 dev_info(host->dev, "All phases work, using default phase %d.",
365 priv->default_sample_phase);
366 goto free;
367 }
368
369 /* Find the longest range */
370 for (i = 0; i < range_count; i++) {
371 int len = (ranges[i].end - ranges[i].start + 1);
372
373 if (len < 0)
374 len += priv->num_phases;
375
376 if (longest_range_len < len) {
377 longest_range_len = len;
378 longest_range = i;
379 }
380
381 dev_dbg(host->dev, "Good phase range %d-%d (%d len)\n",
382 TUNING_ITERATION_TO_PHASE(ranges[i].start,
383 priv->num_phases),
384 TUNING_ITERATION_TO_PHASE(ranges[i].end,
385 priv->num_phases),
386 len
387 );
388 }
389
390 dev_dbg(host->dev, "Best phase range %d-%d (%d len)\n",
391 TUNING_ITERATION_TO_PHASE(ranges[longest_range].start,
392 priv->num_phases),
393 TUNING_ITERATION_TO_PHASE(ranges[longest_range].end,
394 priv->num_phases),
395 longest_range_len
396 );
397
398 middle_phase = ranges[longest_range].start + longest_range_len / 2;
399 middle_phase %= priv->num_phases;
400 phase = TUNING_ITERATION_TO_PHASE(middle_phase, priv->num_phases);
401 dev_info(host->dev, "Successfully tuned phase to %d\n", phase);
402
403 rockchip_mmc_set_phase(host, true, phase);
404
405 free:
406 kfree(ranges);
407 return ret;
408 }
409
dw_mci_common_parse_dt(struct dw_mci * host)410 static int dw_mci_common_parse_dt(struct dw_mci *host)
411 {
412 struct device_node *np = host->dev->of_node;
413 struct dw_mci_rockchip_priv_data *priv;
414
415 priv = devm_kzalloc(host->dev, sizeof(*priv), GFP_KERNEL);
416 if (!priv)
417 return -ENOMEM;
418
419 if (of_property_read_u32(np, "rockchip,desired-num-phases",
420 &priv->num_phases))
421 priv->num_phases = 360;
422
423 if (of_property_read_u32(np, "rockchip,default-sample-phase",
424 &priv->default_sample_phase))
425 priv->default_sample_phase = 0;
426
427 host->priv = priv;
428
429 return 0;
430 }
431
dw_mci_rk3288_parse_dt(struct dw_mci * host)432 static int dw_mci_rk3288_parse_dt(struct dw_mci *host)
433 {
434 struct dw_mci_rockchip_priv_data *priv;
435 int err;
436
437 err = dw_mci_common_parse_dt(host);
438 if (err)
439 return err;
440
441 priv = host->priv;
442
443 priv->drv_clk = devm_clk_get(host->dev, "ciu-drive");
444 if (IS_ERR(priv->drv_clk))
445 dev_dbg(host->dev, "ciu-drive not available\n");
446
447 priv->sample_clk = devm_clk_get(host->dev, "ciu-sample");
448 if (IS_ERR(priv->sample_clk))
449 dev_dbg(host->dev, "ciu-sample not available\n");
450
451 priv->internal_phase = false;
452
453 return 0;
454 }
455
dw_mci_rk3576_parse_dt(struct dw_mci * host)456 static int dw_mci_rk3576_parse_dt(struct dw_mci *host)
457 {
458 struct dw_mci_rockchip_priv_data *priv;
459 int err = dw_mci_common_parse_dt(host);
460 if (err)
461 return err;
462
463 priv = host->priv;
464
465 priv->internal_phase = true;
466
467 return 0;
468 }
469
dw_mci_rockchip_init(struct dw_mci * host)470 static int dw_mci_rockchip_init(struct dw_mci *host)
471 {
472 int ret, i;
473
474 /* It is slot 8 on Rockchip SoCs */
475 host->sdio_id0 = 8;
476
477 if (of_device_is_compatible(host->dev->of_node, "rockchip,rk3288-dw-mshc")) {
478 host->bus_hz /= RK3288_CLKGEN_DIV;
479
480 /* clock driver will fail if the clock is less than the lowest source clock
481 * divided by the internal clock divider. Test for the lowest available
482 * clock and set the minimum freq to clock / clock divider.
483 */
484
485 for (i = 0; i < ARRAY_SIZE(freqs); i++) {
486 ret = clk_round_rate(host->ciu_clk, freqs[i] * RK3288_CLKGEN_DIV);
487 if (ret > 0) {
488 host->minimum_speed = ret / RK3288_CLKGEN_DIV;
489 break;
490 }
491 }
492 if (ret < 0)
493 dev_warn(host->dev, "no valid minimum freq: %d\n", ret);
494 }
495
496 return 0;
497 }
498
499 static const struct dw_mci_drv_data rk2928_drv_data = {
500 .init = dw_mci_rockchip_init,
501 };
502
503 static const struct dw_mci_drv_data rk3288_drv_data = {
504 .common_caps = MMC_CAP_CMD23,
505 .set_ios = dw_mci_rk3288_set_ios,
506 .execute_tuning = dw_mci_rk3288_execute_tuning,
507 .parse_dt = dw_mci_rk3288_parse_dt,
508 .init = dw_mci_rockchip_init,
509 };
510
511 static const struct dw_mci_drv_data rk3576_drv_data = {
512 .common_caps = MMC_CAP_CMD23,
513 .set_ios = dw_mci_rk3288_set_ios,
514 .execute_tuning = dw_mci_rk3288_execute_tuning,
515 .parse_dt = dw_mci_rk3576_parse_dt,
516 .init = dw_mci_rockchip_init,
517 };
518
519 static const struct of_device_id dw_mci_rockchip_match[] = {
520 { .compatible = "rockchip,rk2928-dw-mshc",
521 .data = &rk2928_drv_data },
522 { .compatible = "rockchip,rk3288-dw-mshc",
523 .data = &rk3288_drv_data },
524 { .compatible = "rockchip,rk3576-dw-mshc",
525 .data = &rk3576_drv_data },
526 {},
527 };
528 MODULE_DEVICE_TABLE(of, dw_mci_rockchip_match);
529
dw_mci_rockchip_probe(struct platform_device * pdev)530 static int dw_mci_rockchip_probe(struct platform_device *pdev)
531 {
532 const struct dw_mci_drv_data *drv_data;
533 const struct of_device_id *match;
534 int ret;
535
536 if (!pdev->dev.of_node)
537 return -ENODEV;
538
539 match = of_match_node(dw_mci_rockchip_match, pdev->dev.of_node);
540 drv_data = match->data;
541
542 pm_runtime_get_noresume(&pdev->dev);
543 pm_runtime_set_active(&pdev->dev);
544 pm_runtime_enable(&pdev->dev);
545 pm_runtime_set_autosuspend_delay(&pdev->dev, 50);
546 pm_runtime_use_autosuspend(&pdev->dev);
547
548 ret = dw_mci_pltfm_register(pdev, drv_data);
549 if (ret) {
550 pm_runtime_disable(&pdev->dev);
551 pm_runtime_set_suspended(&pdev->dev);
552 pm_runtime_put_noidle(&pdev->dev);
553 return ret;
554 }
555
556 pm_runtime_put_autosuspend(&pdev->dev);
557
558 return 0;
559 }
560
dw_mci_rockchip_remove(struct platform_device * pdev)561 static void dw_mci_rockchip_remove(struct platform_device *pdev)
562 {
563 pm_runtime_get_sync(&pdev->dev);
564 pm_runtime_disable(&pdev->dev);
565 pm_runtime_put_noidle(&pdev->dev);
566
567 dw_mci_pltfm_remove(pdev);
568 }
569
570 static const struct dev_pm_ops dw_mci_rockchip_dev_pm_ops = {
571 SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
572 pm_runtime_force_resume)
573 SET_RUNTIME_PM_OPS(dw_mci_runtime_suspend,
574 dw_mci_runtime_resume,
575 NULL)
576 };
577
578 static struct platform_driver dw_mci_rockchip_pltfm_driver = {
579 .probe = dw_mci_rockchip_probe,
580 .remove = dw_mci_rockchip_remove,
581 .driver = {
582 .name = "dwmmc_rockchip",
583 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
584 .of_match_table = dw_mci_rockchip_match,
585 .pm = &dw_mci_rockchip_dev_pm_ops,
586 },
587 };
588
589 module_platform_driver(dw_mci_rockchip_pltfm_driver);
590
591 MODULE_AUTHOR("Addy Ke <addy.ke@rock-chips.com>");
592 MODULE_DESCRIPTION("Rockchip Specific DW-MSHC Driver Extension");
593 MODULE_ALIAS("platform:dwmmc_rockchip");
594 MODULE_LICENSE("GPL v2");
595