1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/init.h>
3 #include <linux/clocksource.h>
4 #include <linux/clockchips.h>
5 #include <linux/interrupt.h>
6 #include <linux/irq.h>
7 
8 #include <linux/clk.h>
9 #include <linux/delay.h>
10 #include <linux/err.h>
11 #include <linux/ioport.h>
12 #include <linux/io.h>
13 #include <linux/of_address.h>
14 #include <linux/of_irq.h>
15 #include <linux/sched_clock.h>
16 #include <linux/syscore_ops.h>
17 #include <soc/at91/atmel_tcb.h>
18 
19 
20 /*
21  * We're configured to use a specific TC block, one that's not hooked
22  * up to external hardware, to provide a time solution:
23  *
24  *   - Two channels combine to create a free-running 32 bit counter
25  *     with a base rate of 5+ MHz, packaged as a clocksource (with
26  *     resolution better than 200 nsec).
27  *   - Some chips support 32 bit counter. A single channel is used for
28  *     this 32 bit free-running counter. the second channel is not used.
29  *
30  *   - The third channel may be used to provide a 16-bit clockevent
31  *     source, used in either periodic or oneshot mode.  This runs
32  *     at 32 KiHZ, and can handle delays of up to two seconds.
33  *
34  * REVISIT behavior during system suspend states... we should disable
35  * all clocks and save the power.  Easily done for clockevent devices,
36  * but clocksources won't necessarily get the needed notifications.
37  * For deeper system sleep states, this will be mandatory...
38  */
39 
40 static void __iomem *tcaddr;
41 static struct
42 {
43 	u32 cmr;
44 	u32 imr;
45 	u32 rc;
46 	bool clken;
47 } tcb_cache[3];
48 static u32 bmr_cache;
49 
50 static u64 tc_get_cycles(struct clocksource *cs)
51 {
52 	unsigned long	flags;
53 	u32		lower, upper;
54 
55 	raw_local_irq_save(flags);
56 	do {
57 		upper = readl_relaxed(tcaddr + ATMEL_TC_REG(1, CV));
58 		lower = readl_relaxed(tcaddr + ATMEL_TC_REG(0, CV));
59 	} while (upper != readl_relaxed(tcaddr + ATMEL_TC_REG(1, CV)));
60 
61 	raw_local_irq_restore(flags);
62 	return (upper << 16) | lower;
63 }
64 
65 static u64 tc_get_cycles32(struct clocksource *cs)
66 {
67 	return readl_relaxed(tcaddr + ATMEL_TC_REG(0, CV));
68 }
69 
70 static void tc_clksrc_suspend(struct clocksource *cs)
71 {
72 	int i;
73 
74 	for (i = 0; i < ARRAY_SIZE(tcb_cache); i++) {
75 		tcb_cache[i].cmr = readl(tcaddr + ATMEL_TC_REG(i, CMR));
76 		tcb_cache[i].imr = readl(tcaddr + ATMEL_TC_REG(i, IMR));
77 		tcb_cache[i].rc = readl(tcaddr + ATMEL_TC_REG(i, RC));
78 		tcb_cache[i].clken = !!(readl(tcaddr + ATMEL_TC_REG(i, SR)) &
79 					ATMEL_TC_CLKSTA);
80 	}
81 
82 	bmr_cache = readl(tcaddr + ATMEL_TC_BMR);
83 }
84 
85 static void tc_clksrc_resume(struct clocksource *cs)
86 {
87 	int i;
88 
89 	for (i = 0; i < ARRAY_SIZE(tcb_cache); i++) {
90 		/* Restore registers for the channel, RA and RB are not used  */
91 		writel(tcb_cache[i].cmr, tcaddr + ATMEL_TC_REG(i, CMR));
92 		writel(tcb_cache[i].rc, tcaddr + ATMEL_TC_REG(i, RC));
93 		writel(0, tcaddr + ATMEL_TC_REG(i, RA));
94 		writel(0, tcaddr + ATMEL_TC_REG(i, RB));
95 		/* Disable all the interrupts */
96 		writel(0xff, tcaddr + ATMEL_TC_REG(i, IDR));
97 		/* Reenable interrupts that were enabled before suspending */
98 		writel(tcb_cache[i].imr, tcaddr + ATMEL_TC_REG(i, IER));
99 		/* Start the clock if it was used */
100 		if (tcb_cache[i].clken)
101 			writel(ATMEL_TC_CLKEN, tcaddr + ATMEL_TC_REG(i, CCR));
102 	}
103 
104 	/* Dual channel, chain channels */
105 	writel(bmr_cache, tcaddr + ATMEL_TC_BMR);
106 	/* Finally, trigger all the channels*/
107 	writel(ATMEL_TC_SYNC, tcaddr + ATMEL_TC_BCR);
108 }
109 
110 static struct clocksource clksrc = {
111 	.rating         = 200,
112 	.read           = tc_get_cycles,
113 	.mask           = CLOCKSOURCE_MASK(32),
114 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
115 	.suspend	= tc_clksrc_suspend,
116 	.resume		= tc_clksrc_resume,
117 };
118 
119 static u64 notrace tc_sched_clock_read(void)
120 {
121 	return tc_get_cycles(&clksrc);
122 }
123 
124 static u64 notrace tc_sched_clock_read32(void)
125 {
126 	return tc_get_cycles32(&clksrc);
127 }
128 
129 static struct delay_timer tc_delay_timer;
130 
131 static unsigned long tc_delay_timer_read(void)
132 {
133 	return tc_get_cycles(&clksrc);
134 }
135 
136 static unsigned long notrace tc_delay_timer_read32(void)
137 {
138 	return tc_get_cycles32(&clksrc);
139 }
140 
141 #ifdef CONFIG_GENERIC_CLOCKEVENTS
142 
143 struct tc_clkevt_device {
144 	struct clock_event_device	clkevt;
145 	struct clk			*clk;
146 	void __iomem			*regs;
147 };
148 
149 static struct tc_clkevt_device *to_tc_clkevt(struct clock_event_device *clkevt)
150 {
151 	return container_of(clkevt, struct tc_clkevt_device, clkevt);
152 }
153 
154 /* For now, we always use the 32K clock ... this optimizes for NO_HZ,
155  * because using one of the divided clocks would usually mean the
156  * tick rate can never be less than several dozen Hz (vs 0.5 Hz).
157  *
158  * A divided clock could be good for high resolution timers, since
159  * 30.5 usec resolution can seem "low".
160  */
161 static u32 timer_clock;
162 
163 static int tc_shutdown(struct clock_event_device *d)
164 {
165 	struct tc_clkevt_device *tcd = to_tc_clkevt(d);
166 	void __iomem		*regs = tcd->regs;
167 
168 	writel(0xff, regs + ATMEL_TC_REG(2, IDR));
169 	writel(ATMEL_TC_CLKDIS, regs + ATMEL_TC_REG(2, CCR));
170 	if (!clockevent_state_detached(d))
171 		clk_disable(tcd->clk);
172 
173 	return 0;
174 }
175 
176 static int tc_set_oneshot(struct clock_event_device *d)
177 {
178 	struct tc_clkevt_device *tcd = to_tc_clkevt(d);
179 	void __iomem		*regs = tcd->regs;
180 
181 	if (clockevent_state_oneshot(d) || clockevent_state_periodic(d))
182 		tc_shutdown(d);
183 
184 	clk_enable(tcd->clk);
185 
186 	/* slow clock, count up to RC, then irq and stop */
187 	writel(timer_clock | ATMEL_TC_CPCSTOP | ATMEL_TC_WAVE |
188 		     ATMEL_TC_WAVESEL_UP_AUTO, regs + ATMEL_TC_REG(2, CMR));
189 	writel(ATMEL_TC_CPCS, regs + ATMEL_TC_REG(2, IER));
190 
191 	/* set_next_event() configures and starts the timer */
192 	return 0;
193 }
194 
195 static int tc_set_periodic(struct clock_event_device *d)
196 {
197 	struct tc_clkevt_device *tcd = to_tc_clkevt(d);
198 	void __iomem		*regs = tcd->regs;
199 
200 	if (clockevent_state_oneshot(d) || clockevent_state_periodic(d))
201 		tc_shutdown(d);
202 
203 	/* By not making the gentime core emulate periodic mode on top
204 	 * of oneshot, we get lower overhead and improved accuracy.
205 	 */
206 	clk_enable(tcd->clk);
207 
208 	/* slow clock, count up to RC, then irq and restart */
209 	writel(timer_clock | ATMEL_TC_WAVE | ATMEL_TC_WAVESEL_UP_AUTO,
210 		     regs + ATMEL_TC_REG(2, CMR));
211 	writel((32768 + HZ / 2) / HZ, tcaddr + ATMEL_TC_REG(2, RC));
212 
213 	/* Enable clock and interrupts on RC compare */
214 	writel(ATMEL_TC_CPCS, regs + ATMEL_TC_REG(2, IER));
215 
216 	/* go go gadget! */
217 	writel(ATMEL_TC_CLKEN | ATMEL_TC_SWTRG, regs +
218 		     ATMEL_TC_REG(2, CCR));
219 	return 0;
220 }
221 
222 static int tc_next_event(unsigned long delta, struct clock_event_device *d)
223 {
224 	writel_relaxed(delta, tcaddr + ATMEL_TC_REG(2, RC));
225 
226 	/* go go gadget! */
227 	writel_relaxed(ATMEL_TC_CLKEN | ATMEL_TC_SWTRG,
228 			tcaddr + ATMEL_TC_REG(2, CCR));
229 	return 0;
230 }
231 
232 static struct tc_clkevt_device clkevt = {
233 	.clkevt	= {
234 		.features		= CLOCK_EVT_FEAT_PERIODIC |
235 					  CLOCK_EVT_FEAT_ONESHOT,
236 		/* Should be lower than at91rm9200's system timer */
237 		.rating			= 125,
238 		.set_next_event		= tc_next_event,
239 		.set_state_shutdown	= tc_shutdown,
240 		.set_state_periodic	= tc_set_periodic,
241 		.set_state_oneshot	= tc_set_oneshot,
242 	},
243 };
244 
245 static irqreturn_t ch2_irq(int irq, void *handle)
246 {
247 	struct tc_clkevt_device	*dev = handle;
248 	unsigned int		sr;
249 
250 	sr = readl_relaxed(dev->regs + ATMEL_TC_REG(2, SR));
251 	if (sr & ATMEL_TC_CPCS) {
252 		dev->clkevt.event_handler(&dev->clkevt);
253 		return IRQ_HANDLED;
254 	}
255 
256 	return IRQ_NONE;
257 }
258 
259 static int __init setup_clkevents(struct atmel_tc *tc, int clk32k_divisor_idx)
260 {
261 	int ret;
262 	struct clk *t2_clk = tc->clk[2];
263 	int irq = tc->irq[2];
264 
265 	ret = clk_prepare_enable(tc->slow_clk);
266 	if (ret)
267 		return ret;
268 
269 	/* try to enable t2 clk to avoid future errors in mode change */
270 	ret = clk_prepare_enable(t2_clk);
271 	if (ret) {
272 		clk_disable_unprepare(tc->slow_clk);
273 		return ret;
274 	}
275 
276 	clk_disable(t2_clk);
277 
278 	clkevt.regs = tc->regs;
279 	clkevt.clk = t2_clk;
280 
281 	timer_clock = clk32k_divisor_idx;
282 
283 	clkevt.clkevt.cpumask = cpumask_of(0);
284 
285 	ret = request_irq(irq, ch2_irq, IRQF_TIMER, "tc_clkevt", &clkevt);
286 	if (ret) {
287 		clk_unprepare(t2_clk);
288 		clk_disable_unprepare(tc->slow_clk);
289 		return ret;
290 	}
291 
292 	clockevents_config_and_register(&clkevt.clkevt, 32768, 1, 0xffff);
293 
294 	return ret;
295 }
296 
297 #else /* !CONFIG_GENERIC_CLOCKEVENTS */
298 
299 static int __init setup_clkevents(struct atmel_tc *tc, int clk32k_divisor_idx)
300 {
301 	/* NOTHING */
302 	return 0;
303 }
304 
305 #endif
306 
307 static void __init tcb_setup_dual_chan(struct atmel_tc *tc, int mck_divisor_idx)
308 {
309 	/* channel 0:  waveform mode, input mclk/8, clock TIOA0 on overflow */
310 	writel(mck_divisor_idx			/* likely divide-by-8 */
311 			| ATMEL_TC_WAVE
312 			| ATMEL_TC_WAVESEL_UP		/* free-run */
313 			| ATMEL_TC_ACPA_SET		/* TIOA0 rises at 0 */
314 			| ATMEL_TC_ACPC_CLEAR,		/* (duty cycle 50%) */
315 			tcaddr + ATMEL_TC_REG(0, CMR));
316 	writel(0x0000, tcaddr + ATMEL_TC_REG(0, RA));
317 	writel(0x8000, tcaddr + ATMEL_TC_REG(0, RC));
318 	writel(0xff, tcaddr + ATMEL_TC_REG(0, IDR));	/* no irqs */
319 	writel(ATMEL_TC_CLKEN, tcaddr + ATMEL_TC_REG(0, CCR));
320 
321 	/* channel 1:  waveform mode, input TIOA0 */
322 	writel(ATMEL_TC_XC1			/* input: TIOA0 */
323 			| ATMEL_TC_WAVE
324 			| ATMEL_TC_WAVESEL_UP,		/* free-run */
325 			tcaddr + ATMEL_TC_REG(1, CMR));
326 	writel(0xff, tcaddr + ATMEL_TC_REG(1, IDR));	/* no irqs */
327 	writel(ATMEL_TC_CLKEN, tcaddr + ATMEL_TC_REG(1, CCR));
328 
329 	/* chain channel 0 to channel 1*/
330 	writel(ATMEL_TC_TC1XC1S_TIOA0, tcaddr + ATMEL_TC_BMR);
331 	/* then reset all the timers */
332 	writel(ATMEL_TC_SYNC, tcaddr + ATMEL_TC_BCR);
333 }
334 
335 static void __init tcb_setup_single_chan(struct atmel_tc *tc, int mck_divisor_idx)
336 {
337 	/* channel 0:  waveform mode, input mclk/8 */
338 	writel(mck_divisor_idx			/* likely divide-by-8 */
339 			| ATMEL_TC_WAVE
340 			| ATMEL_TC_WAVESEL_UP,		/* free-run */
341 			tcaddr + ATMEL_TC_REG(0, CMR));
342 	writel(0xff, tcaddr + ATMEL_TC_REG(0, IDR));	/* no irqs */
343 	writel(ATMEL_TC_CLKEN, tcaddr + ATMEL_TC_REG(0, CCR));
344 
345 	/* then reset all the timers */
346 	writel(ATMEL_TC_SYNC, tcaddr + ATMEL_TC_BCR);
347 }
348 
349 static const u8 atmel_tcb_divisors[5] = { 2, 8, 32, 128, 0, };
350 
351 static const struct of_device_id atmel_tcb_of_match[] = {
352 	{ .compatible = "atmel,at91rm9200-tcb", .data = (void *)16, },
353 	{ .compatible = "atmel,at91sam9x5-tcb", .data = (void *)32, },
354 	{ /* sentinel */ }
355 };
356 
357 static int __init tcb_clksrc_init(struct device_node *node)
358 {
359 	struct atmel_tc tc;
360 	struct clk *t0_clk;
361 	const struct of_device_id *match;
362 	u64 (*tc_sched_clock)(void);
363 	u32 rate, divided_rate = 0;
364 	int best_divisor_idx = -1;
365 	int clk32k_divisor_idx = -1;
366 	int bits;
367 	int i;
368 	int ret;
369 
370 	/* Protect against multiple calls */
371 	if (tcaddr)
372 		return 0;
373 
374 	tc.regs = of_iomap(node->parent, 0);
375 	if (!tc.regs)
376 		return -ENXIO;
377 
378 	t0_clk = of_clk_get_by_name(node->parent, "t0_clk");
379 	if (IS_ERR(t0_clk))
380 		return PTR_ERR(t0_clk);
381 
382 	tc.slow_clk = of_clk_get_by_name(node->parent, "slow_clk");
383 	if (IS_ERR(tc.slow_clk))
384 		return PTR_ERR(tc.slow_clk);
385 
386 	tc.clk[0] = t0_clk;
387 	tc.clk[1] = of_clk_get_by_name(node->parent, "t1_clk");
388 	if (IS_ERR(tc.clk[1]))
389 		tc.clk[1] = t0_clk;
390 	tc.clk[2] = of_clk_get_by_name(node->parent, "t2_clk");
391 	if (IS_ERR(tc.clk[2]))
392 		tc.clk[2] = t0_clk;
393 
394 	tc.irq[2] = of_irq_get(node->parent, 2);
395 	if (tc.irq[2] <= 0) {
396 		tc.irq[2] = of_irq_get(node->parent, 0);
397 		if (tc.irq[2] <= 0)
398 			return -EINVAL;
399 	}
400 
401 	match = of_match_node(atmel_tcb_of_match, node->parent);
402 	bits = (uintptr_t)match->data;
403 
404 	for (i = 0; i < ARRAY_SIZE(tc.irq); i++)
405 		writel(ATMEL_TC_ALL_IRQ, tc.regs + ATMEL_TC_REG(i, IDR));
406 
407 	ret = clk_prepare_enable(t0_clk);
408 	if (ret) {
409 		pr_debug("can't enable T0 clk\n");
410 		return ret;
411 	}
412 
413 	/* How fast will we be counting?  Pick something over 5 MHz.  */
414 	rate = (u32) clk_get_rate(t0_clk);
415 	for (i = 0; i < ARRAY_SIZE(atmel_tcb_divisors); i++) {
416 		unsigned divisor = atmel_tcb_divisors[i];
417 		unsigned tmp;
418 
419 		/* remember 32 KiHz clock for later */
420 		if (!divisor) {
421 			clk32k_divisor_idx = i;
422 			continue;
423 		}
424 
425 		tmp = rate / divisor;
426 		pr_debug("TC: %u / %-3u [%d] --> %u\n", rate, divisor, i, tmp);
427 		if (best_divisor_idx > 0) {
428 			if (tmp < 5 * 1000 * 1000)
429 				continue;
430 		}
431 		divided_rate = tmp;
432 		best_divisor_idx = i;
433 	}
434 
435 	clksrc.name = kbasename(node->parent->full_name);
436 	clkevt.clkevt.name = kbasename(node->parent->full_name);
437 	pr_debug("%s at %d.%03d MHz\n", clksrc.name, divided_rate / 1000000,
438 			((divided_rate % 1000000) + 500) / 1000);
439 
440 	tcaddr = tc.regs;
441 
442 	if (bits == 32) {
443 		/* use apropriate function to read 32 bit counter */
444 		clksrc.read = tc_get_cycles32;
445 		/* setup ony channel 0 */
446 		tcb_setup_single_chan(&tc, best_divisor_idx);
447 		tc_sched_clock = tc_sched_clock_read32;
448 		tc_delay_timer.read_current_timer = tc_delay_timer_read32;
449 	} else {
450 		/* we have three clocks no matter what the
451 		 * underlying platform supports.
452 		 */
453 		ret = clk_prepare_enable(tc.clk[1]);
454 		if (ret) {
455 			pr_debug("can't enable T1 clk\n");
456 			goto err_disable_t0;
457 		}
458 		/* setup both channel 0 & 1 */
459 		tcb_setup_dual_chan(&tc, best_divisor_idx);
460 		tc_sched_clock = tc_sched_clock_read;
461 		tc_delay_timer.read_current_timer = tc_delay_timer_read;
462 	}
463 
464 	/* and away we go! */
465 	ret = clocksource_register_hz(&clksrc, divided_rate);
466 	if (ret)
467 		goto err_disable_t1;
468 
469 	/* channel 2:  periodic and oneshot timer support */
470 	ret = setup_clkevents(&tc, clk32k_divisor_idx);
471 	if (ret)
472 		goto err_unregister_clksrc;
473 
474 	sched_clock_register(tc_sched_clock, 32, divided_rate);
475 
476 	tc_delay_timer.freq = divided_rate;
477 	register_current_timer_delay(&tc_delay_timer);
478 
479 	return 0;
480 
481 err_unregister_clksrc:
482 	clocksource_unregister(&clksrc);
483 
484 err_disable_t1:
485 	if (bits != 32)
486 		clk_disable_unprepare(tc.clk[1]);
487 
488 err_disable_t0:
489 	clk_disable_unprepare(t0_clk);
490 
491 	tcaddr = NULL;
492 
493 	return ret;
494 }
495 TIMER_OF_DECLARE(atmel_tcb_clksrc, "atmel,tcb-timer", tcb_clksrc_init);
496