1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/drivers/clocksource/arm_arch_timer.c
4  *
5  *  Copyright (C) 2011 ARM Ltd.
6  *  All Rights Reserved
7  */
8 
9 #define pr_fmt(fmt) 	"arch_timer: " fmt
10 
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/device.h>
14 #include <linux/smp.h>
15 #include <linux/cpu.h>
16 #include <linux/cpu_pm.h>
17 #include <linux/clockchips.h>
18 #include <linux/clocksource.h>
19 #include <linux/interrupt.h>
20 #include <linux/of_irq.h>
21 #include <linux/of_address.h>
22 #include <linux/io.h>
23 #include <linux/slab.h>
24 #include <linux/sched/clock.h>
25 #include <linux/sched_clock.h>
26 #include <linux/acpi.h>
27 
28 #include <asm/arch_timer.h>
29 #include <asm/virt.h>
30 
31 #include <clocksource/arm_arch_timer.h>
32 
33 #define CNTTIDR		0x08
34 #define CNTTIDR_VIRT(n)	(BIT(1) << ((n) * 4))
35 
36 #define CNTACR(n)	(0x40 + ((n) * 4))
37 #define CNTACR_RPCT	BIT(0)
38 #define CNTACR_RVCT	BIT(1)
39 #define CNTACR_RFRQ	BIT(2)
40 #define CNTACR_RVOFF	BIT(3)
41 #define CNTACR_RWVT	BIT(4)
42 #define CNTACR_RWPT	BIT(5)
43 
44 #define CNTVCT_LO	0x08
45 #define CNTVCT_HI	0x0c
46 #define CNTFRQ		0x10
47 #define CNTP_TVAL	0x28
48 #define CNTP_CTL	0x2c
49 #define CNTV_TVAL	0x38
50 #define CNTV_CTL	0x3c
51 
52 static unsigned arch_timers_present __initdata;
53 
54 static void __iomem *arch_counter_base;
55 
56 struct arch_timer {
57 	void __iomem *base;
58 	struct clock_event_device evt;
59 };
60 
61 #define to_arch_timer(e) container_of(e, struct arch_timer, evt)
62 
63 static u32 arch_timer_rate;
64 static int arch_timer_ppi[ARCH_TIMER_MAX_TIMER_PPI];
65 
66 static struct clock_event_device __percpu *arch_timer_evt;
67 
68 static enum arch_timer_ppi_nr arch_timer_uses_ppi = ARCH_TIMER_VIRT_PPI;
69 static bool arch_timer_c3stop;
70 static bool arch_timer_mem_use_virtual;
71 static bool arch_counter_suspend_stop;
72 #ifdef CONFIG_GENERIC_GETTIMEOFDAY
73 static enum vdso_clock_mode vdso_default = VDSO_CLOCKMODE_ARCHTIMER;
74 #else
75 static enum vdso_clock_mode vdso_default = VDSO_CLOCKMODE_NONE;
76 #endif /* CONFIG_GENERIC_GETTIMEOFDAY */
77 
78 static cpumask_t evtstrm_available = CPU_MASK_NONE;
79 static bool evtstrm_enable = IS_ENABLED(CONFIG_ARM_ARCH_TIMER_EVTSTREAM);
80 
81 static int __init early_evtstrm_cfg(char *buf)
82 {
83 	return strtobool(buf, &evtstrm_enable);
84 }
85 early_param("clocksource.arm_arch_timer.evtstrm", early_evtstrm_cfg);
86 
87 /*
88  * Architected system timer support.
89  */
90 
91 static __always_inline
92 void arch_timer_reg_write(int access, enum arch_timer_reg reg, u32 val,
93 			  struct clock_event_device *clk)
94 {
95 	if (access == ARCH_TIMER_MEM_PHYS_ACCESS) {
96 		struct arch_timer *timer = to_arch_timer(clk);
97 		switch (reg) {
98 		case ARCH_TIMER_REG_CTRL:
99 			writel_relaxed(val, timer->base + CNTP_CTL);
100 			break;
101 		case ARCH_TIMER_REG_TVAL:
102 			writel_relaxed(val, timer->base + CNTP_TVAL);
103 			break;
104 		}
105 	} else if (access == ARCH_TIMER_MEM_VIRT_ACCESS) {
106 		struct arch_timer *timer = to_arch_timer(clk);
107 		switch (reg) {
108 		case ARCH_TIMER_REG_CTRL:
109 			writel_relaxed(val, timer->base + CNTV_CTL);
110 			break;
111 		case ARCH_TIMER_REG_TVAL:
112 			writel_relaxed(val, timer->base + CNTV_TVAL);
113 			break;
114 		}
115 	} else {
116 		arch_timer_reg_write_cp15(access, reg, val);
117 	}
118 }
119 
120 static __always_inline
121 u32 arch_timer_reg_read(int access, enum arch_timer_reg reg,
122 			struct clock_event_device *clk)
123 {
124 	u32 val;
125 
126 	if (access == ARCH_TIMER_MEM_PHYS_ACCESS) {
127 		struct arch_timer *timer = to_arch_timer(clk);
128 		switch (reg) {
129 		case ARCH_TIMER_REG_CTRL:
130 			val = readl_relaxed(timer->base + CNTP_CTL);
131 			break;
132 		case ARCH_TIMER_REG_TVAL:
133 			val = readl_relaxed(timer->base + CNTP_TVAL);
134 			break;
135 		}
136 	} else if (access == ARCH_TIMER_MEM_VIRT_ACCESS) {
137 		struct arch_timer *timer = to_arch_timer(clk);
138 		switch (reg) {
139 		case ARCH_TIMER_REG_CTRL:
140 			val = readl_relaxed(timer->base + CNTV_CTL);
141 			break;
142 		case ARCH_TIMER_REG_TVAL:
143 			val = readl_relaxed(timer->base + CNTV_TVAL);
144 			break;
145 		}
146 	} else {
147 		val = arch_timer_reg_read_cp15(access, reg);
148 	}
149 
150 	return val;
151 }
152 
153 static notrace u64 arch_counter_get_cntpct_stable(void)
154 {
155 	return __arch_counter_get_cntpct_stable();
156 }
157 
158 static notrace u64 arch_counter_get_cntpct(void)
159 {
160 	return __arch_counter_get_cntpct();
161 }
162 
163 static notrace u64 arch_counter_get_cntvct_stable(void)
164 {
165 	return __arch_counter_get_cntvct_stable();
166 }
167 
168 static notrace u64 arch_counter_get_cntvct(void)
169 {
170 	return __arch_counter_get_cntvct();
171 }
172 
173 /*
174  * Default to cp15 based access because arm64 uses this function for
175  * sched_clock() before DT is probed and the cp15 method is guaranteed
176  * to exist on arm64. arm doesn't use this before DT is probed so even
177  * if we don't have the cp15 accessors we won't have a problem.
178  */
179 u64 (*arch_timer_read_counter)(void) = arch_counter_get_cntvct;
180 EXPORT_SYMBOL_GPL(arch_timer_read_counter);
181 
182 static u64 arch_counter_read(struct clocksource *cs)
183 {
184 	return arch_timer_read_counter();
185 }
186 
187 static u64 arch_counter_read_cc(const struct cyclecounter *cc)
188 {
189 	return arch_timer_read_counter();
190 }
191 
192 static struct clocksource clocksource_counter = {
193 	.name	= "arch_sys_counter",
194 	.rating	= 400,
195 	.read	= arch_counter_read,
196 	.mask	= CLOCKSOURCE_MASK(56),
197 	.flags	= CLOCK_SOURCE_IS_CONTINUOUS,
198 };
199 
200 static struct cyclecounter cyclecounter __ro_after_init = {
201 	.read	= arch_counter_read_cc,
202 	.mask	= CLOCKSOURCE_MASK(56),
203 };
204 
205 struct ate_acpi_oem_info {
206 	char oem_id[ACPI_OEM_ID_SIZE + 1];
207 	char oem_table_id[ACPI_OEM_TABLE_ID_SIZE + 1];
208 	u32 oem_revision;
209 };
210 
211 #ifdef CONFIG_FSL_ERRATUM_A008585
212 /*
213  * The number of retries is an arbitrary value well beyond the highest number
214  * of iterations the loop has been observed to take.
215  */
216 #define __fsl_a008585_read_reg(reg) ({			\
217 	u64 _old, _new;					\
218 	int _retries = 200;				\
219 							\
220 	do {						\
221 		_old = read_sysreg(reg);		\
222 		_new = read_sysreg(reg);		\
223 		_retries--;				\
224 	} while (unlikely(_old != _new) && _retries);	\
225 							\
226 	WARN_ON_ONCE(!_retries);			\
227 	_new;						\
228 })
229 
230 static u32 notrace fsl_a008585_read_cntp_tval_el0(void)
231 {
232 	return __fsl_a008585_read_reg(cntp_tval_el0);
233 }
234 
235 static u32 notrace fsl_a008585_read_cntv_tval_el0(void)
236 {
237 	return __fsl_a008585_read_reg(cntv_tval_el0);
238 }
239 
240 static u64 notrace fsl_a008585_read_cntpct_el0(void)
241 {
242 	return __fsl_a008585_read_reg(cntpct_el0);
243 }
244 
245 static u64 notrace fsl_a008585_read_cntvct_el0(void)
246 {
247 	return __fsl_a008585_read_reg(cntvct_el0);
248 }
249 #endif
250 
251 #ifdef CONFIG_HISILICON_ERRATUM_161010101
252 /*
253  * Verify whether the value of the second read is larger than the first by
254  * less than 32 is the only way to confirm the value is correct, so clear the
255  * lower 5 bits to check whether the difference is greater than 32 or not.
256  * Theoretically the erratum should not occur more than twice in succession
257  * when reading the system counter, but it is possible that some interrupts
258  * may lead to more than twice read errors, triggering the warning, so setting
259  * the number of retries far beyond the number of iterations the loop has been
260  * observed to take.
261  */
262 #define __hisi_161010101_read_reg(reg) ({				\
263 	u64 _old, _new;						\
264 	int _retries = 50;					\
265 								\
266 	do {							\
267 		_old = read_sysreg(reg);			\
268 		_new = read_sysreg(reg);			\
269 		_retries--;					\
270 	} while (unlikely((_new - _old) >> 5) && _retries);	\
271 								\
272 	WARN_ON_ONCE(!_retries);				\
273 	_new;							\
274 })
275 
276 static u32 notrace hisi_161010101_read_cntp_tval_el0(void)
277 {
278 	return __hisi_161010101_read_reg(cntp_tval_el0);
279 }
280 
281 static u32 notrace hisi_161010101_read_cntv_tval_el0(void)
282 {
283 	return __hisi_161010101_read_reg(cntv_tval_el0);
284 }
285 
286 static u64 notrace hisi_161010101_read_cntpct_el0(void)
287 {
288 	return __hisi_161010101_read_reg(cntpct_el0);
289 }
290 
291 static u64 notrace hisi_161010101_read_cntvct_el0(void)
292 {
293 	return __hisi_161010101_read_reg(cntvct_el0);
294 }
295 
296 static struct ate_acpi_oem_info hisi_161010101_oem_info[] = {
297 	/*
298 	 * Note that trailing spaces are required to properly match
299 	 * the OEM table information.
300 	 */
301 	{
302 		.oem_id		= "HISI  ",
303 		.oem_table_id	= "HIP05   ",
304 		.oem_revision	= 0,
305 	},
306 	{
307 		.oem_id		= "HISI  ",
308 		.oem_table_id	= "HIP06   ",
309 		.oem_revision	= 0,
310 	},
311 	{
312 		.oem_id		= "HISI  ",
313 		.oem_table_id	= "HIP07   ",
314 		.oem_revision	= 0,
315 	},
316 	{ /* Sentinel indicating the end of the OEM array */ },
317 };
318 #endif
319 
320 #ifdef CONFIG_ARM64_ERRATUM_858921
321 static u64 notrace arm64_858921_read_cntpct_el0(void)
322 {
323 	u64 old, new;
324 
325 	old = read_sysreg(cntpct_el0);
326 	new = read_sysreg(cntpct_el0);
327 	return (((old ^ new) >> 32) & 1) ? old : new;
328 }
329 
330 static u64 notrace arm64_858921_read_cntvct_el0(void)
331 {
332 	u64 old, new;
333 
334 	old = read_sysreg(cntvct_el0);
335 	new = read_sysreg(cntvct_el0);
336 	return (((old ^ new) >> 32) & 1) ? old : new;
337 }
338 #endif
339 
340 #ifdef CONFIG_SUN50I_ERRATUM_UNKNOWN1
341 /*
342  * The low bits of the counter registers are indeterminate while bit 10 or
343  * greater is rolling over. Since the counter value can jump both backward
344  * (7ff -> 000 -> 800) and forward (7ff -> fff -> 800), ignore register values
345  * with all ones or all zeros in the low bits. Bound the loop by the maximum
346  * number of CPU cycles in 3 consecutive 24 MHz counter periods.
347  */
348 #define __sun50i_a64_read_reg(reg) ({					\
349 	u64 _val;							\
350 	int _retries = 150;						\
351 									\
352 	do {								\
353 		_val = read_sysreg(reg);				\
354 		_retries--;						\
355 	} while (((_val + 1) & GENMASK(9, 0)) <= 1 && _retries);	\
356 									\
357 	WARN_ON_ONCE(!_retries);					\
358 	_val;								\
359 })
360 
361 static u64 notrace sun50i_a64_read_cntpct_el0(void)
362 {
363 	return __sun50i_a64_read_reg(cntpct_el0);
364 }
365 
366 static u64 notrace sun50i_a64_read_cntvct_el0(void)
367 {
368 	return __sun50i_a64_read_reg(cntvct_el0);
369 }
370 
371 static u32 notrace sun50i_a64_read_cntp_tval_el0(void)
372 {
373 	return read_sysreg(cntp_cval_el0) - sun50i_a64_read_cntpct_el0();
374 }
375 
376 static u32 notrace sun50i_a64_read_cntv_tval_el0(void)
377 {
378 	return read_sysreg(cntv_cval_el0) - sun50i_a64_read_cntvct_el0();
379 }
380 #endif
381 
382 #ifdef CONFIG_ARM_ARCH_TIMER_OOL_WORKAROUND
383 DEFINE_PER_CPU(const struct arch_timer_erratum_workaround *, timer_unstable_counter_workaround);
384 EXPORT_SYMBOL_GPL(timer_unstable_counter_workaround);
385 
386 static atomic_t timer_unstable_counter_workaround_in_use = ATOMIC_INIT(0);
387 
388 static void erratum_set_next_event_tval_generic(const int access, unsigned long evt,
389 						struct clock_event_device *clk)
390 {
391 	unsigned long ctrl;
392 	u64 cval;
393 
394 	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
395 	ctrl |= ARCH_TIMER_CTRL_ENABLE;
396 	ctrl &= ~ARCH_TIMER_CTRL_IT_MASK;
397 
398 	if (access == ARCH_TIMER_PHYS_ACCESS) {
399 		cval = evt + arch_counter_get_cntpct();
400 		write_sysreg(cval, cntp_cval_el0);
401 	} else {
402 		cval = evt + arch_counter_get_cntvct();
403 		write_sysreg(cval, cntv_cval_el0);
404 	}
405 
406 	arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
407 }
408 
409 static __maybe_unused int erratum_set_next_event_tval_virt(unsigned long evt,
410 					    struct clock_event_device *clk)
411 {
412 	erratum_set_next_event_tval_generic(ARCH_TIMER_VIRT_ACCESS, evt, clk);
413 	return 0;
414 }
415 
416 static __maybe_unused int erratum_set_next_event_tval_phys(unsigned long evt,
417 					    struct clock_event_device *clk)
418 {
419 	erratum_set_next_event_tval_generic(ARCH_TIMER_PHYS_ACCESS, evt, clk);
420 	return 0;
421 }
422 
423 static const struct arch_timer_erratum_workaround ool_workarounds[] = {
424 #ifdef CONFIG_FSL_ERRATUM_A008585
425 	{
426 		.match_type = ate_match_dt,
427 		.id = "fsl,erratum-a008585",
428 		.desc = "Freescale erratum a005858",
429 		.read_cntp_tval_el0 = fsl_a008585_read_cntp_tval_el0,
430 		.read_cntv_tval_el0 = fsl_a008585_read_cntv_tval_el0,
431 		.read_cntpct_el0 = fsl_a008585_read_cntpct_el0,
432 		.read_cntvct_el0 = fsl_a008585_read_cntvct_el0,
433 		.set_next_event_phys = erratum_set_next_event_tval_phys,
434 		.set_next_event_virt = erratum_set_next_event_tval_virt,
435 	},
436 #endif
437 #ifdef CONFIG_HISILICON_ERRATUM_161010101
438 	{
439 		.match_type = ate_match_dt,
440 		.id = "hisilicon,erratum-161010101",
441 		.desc = "HiSilicon erratum 161010101",
442 		.read_cntp_tval_el0 = hisi_161010101_read_cntp_tval_el0,
443 		.read_cntv_tval_el0 = hisi_161010101_read_cntv_tval_el0,
444 		.read_cntpct_el0 = hisi_161010101_read_cntpct_el0,
445 		.read_cntvct_el0 = hisi_161010101_read_cntvct_el0,
446 		.set_next_event_phys = erratum_set_next_event_tval_phys,
447 		.set_next_event_virt = erratum_set_next_event_tval_virt,
448 	},
449 	{
450 		.match_type = ate_match_acpi_oem_info,
451 		.id = hisi_161010101_oem_info,
452 		.desc = "HiSilicon erratum 161010101",
453 		.read_cntp_tval_el0 = hisi_161010101_read_cntp_tval_el0,
454 		.read_cntv_tval_el0 = hisi_161010101_read_cntv_tval_el0,
455 		.read_cntpct_el0 = hisi_161010101_read_cntpct_el0,
456 		.read_cntvct_el0 = hisi_161010101_read_cntvct_el0,
457 		.set_next_event_phys = erratum_set_next_event_tval_phys,
458 		.set_next_event_virt = erratum_set_next_event_tval_virt,
459 	},
460 #endif
461 #ifdef CONFIG_ARM64_ERRATUM_858921
462 	{
463 		.match_type = ate_match_local_cap_id,
464 		.id = (void *)ARM64_WORKAROUND_858921,
465 		.desc = "ARM erratum 858921",
466 		.read_cntpct_el0 = arm64_858921_read_cntpct_el0,
467 		.read_cntvct_el0 = arm64_858921_read_cntvct_el0,
468 	},
469 #endif
470 #ifdef CONFIG_SUN50I_ERRATUM_UNKNOWN1
471 	{
472 		.match_type = ate_match_dt,
473 		.id = "allwinner,erratum-unknown1",
474 		.desc = "Allwinner erratum UNKNOWN1",
475 		.read_cntp_tval_el0 = sun50i_a64_read_cntp_tval_el0,
476 		.read_cntv_tval_el0 = sun50i_a64_read_cntv_tval_el0,
477 		.read_cntpct_el0 = sun50i_a64_read_cntpct_el0,
478 		.read_cntvct_el0 = sun50i_a64_read_cntvct_el0,
479 		.set_next_event_phys = erratum_set_next_event_tval_phys,
480 		.set_next_event_virt = erratum_set_next_event_tval_virt,
481 	},
482 #endif
483 };
484 
485 typedef bool (*ate_match_fn_t)(const struct arch_timer_erratum_workaround *,
486 			       const void *);
487 
488 static
489 bool arch_timer_check_dt_erratum(const struct arch_timer_erratum_workaround *wa,
490 				 const void *arg)
491 {
492 	const struct device_node *np = arg;
493 
494 	return of_property_read_bool(np, wa->id);
495 }
496 
497 static
498 bool arch_timer_check_local_cap_erratum(const struct arch_timer_erratum_workaround *wa,
499 					const void *arg)
500 {
501 	return this_cpu_has_cap((uintptr_t)wa->id);
502 }
503 
504 
505 static
506 bool arch_timer_check_acpi_oem_erratum(const struct arch_timer_erratum_workaround *wa,
507 				       const void *arg)
508 {
509 	static const struct ate_acpi_oem_info empty_oem_info = {};
510 	const struct ate_acpi_oem_info *info = wa->id;
511 	const struct acpi_table_header *table = arg;
512 
513 	/* Iterate over the ACPI OEM info array, looking for a match */
514 	while (memcmp(info, &empty_oem_info, sizeof(*info))) {
515 		if (!memcmp(info->oem_id, table->oem_id, ACPI_OEM_ID_SIZE) &&
516 		    !memcmp(info->oem_table_id, table->oem_table_id, ACPI_OEM_TABLE_ID_SIZE) &&
517 		    info->oem_revision == table->oem_revision)
518 			return true;
519 
520 		info++;
521 	}
522 
523 	return false;
524 }
525 
526 static const struct arch_timer_erratum_workaround *
527 arch_timer_iterate_errata(enum arch_timer_erratum_match_type type,
528 			  ate_match_fn_t match_fn,
529 			  void *arg)
530 {
531 	int i;
532 
533 	for (i = 0; i < ARRAY_SIZE(ool_workarounds); i++) {
534 		if (ool_workarounds[i].match_type != type)
535 			continue;
536 
537 		if (match_fn(&ool_workarounds[i], arg))
538 			return &ool_workarounds[i];
539 	}
540 
541 	return NULL;
542 }
543 
544 static
545 void arch_timer_enable_workaround(const struct arch_timer_erratum_workaround *wa,
546 				  bool local)
547 {
548 	int i;
549 
550 	if (local) {
551 		__this_cpu_write(timer_unstable_counter_workaround, wa);
552 	} else {
553 		for_each_possible_cpu(i)
554 			per_cpu(timer_unstable_counter_workaround, i) = wa;
555 	}
556 
557 	if (wa->read_cntvct_el0 || wa->read_cntpct_el0)
558 		atomic_set(&timer_unstable_counter_workaround_in_use, 1);
559 
560 	/*
561 	 * Don't use the vdso fastpath if errata require using the
562 	 * out-of-line counter accessor. We may change our mind pretty
563 	 * late in the game (with a per-CPU erratum, for example), so
564 	 * change both the default value and the vdso itself.
565 	 */
566 	if (wa->read_cntvct_el0) {
567 		clocksource_counter.vdso_clock_mode = VDSO_CLOCKMODE_NONE;
568 		vdso_default = VDSO_CLOCKMODE_NONE;
569 	}
570 }
571 
572 static void arch_timer_check_ool_workaround(enum arch_timer_erratum_match_type type,
573 					    void *arg)
574 {
575 	const struct arch_timer_erratum_workaround *wa, *__wa;
576 	ate_match_fn_t match_fn = NULL;
577 	bool local = false;
578 
579 	switch (type) {
580 	case ate_match_dt:
581 		match_fn = arch_timer_check_dt_erratum;
582 		break;
583 	case ate_match_local_cap_id:
584 		match_fn = arch_timer_check_local_cap_erratum;
585 		local = true;
586 		break;
587 	case ate_match_acpi_oem_info:
588 		match_fn = arch_timer_check_acpi_oem_erratum;
589 		break;
590 	default:
591 		WARN_ON(1);
592 		return;
593 	}
594 
595 	wa = arch_timer_iterate_errata(type, match_fn, arg);
596 	if (!wa)
597 		return;
598 
599 	__wa = __this_cpu_read(timer_unstable_counter_workaround);
600 	if (__wa && wa != __wa)
601 		pr_warn("Can't enable workaround for %s (clashes with %s\n)",
602 			wa->desc, __wa->desc);
603 
604 	if (__wa)
605 		return;
606 
607 	arch_timer_enable_workaround(wa, local);
608 	pr_info("Enabling %s workaround for %s\n",
609 		local ? "local" : "global", wa->desc);
610 }
611 
612 static bool arch_timer_this_cpu_has_cntvct_wa(void)
613 {
614 	return has_erratum_handler(read_cntvct_el0);
615 }
616 
617 static bool arch_timer_counter_has_wa(void)
618 {
619 	return atomic_read(&timer_unstable_counter_workaround_in_use);
620 }
621 #else
622 #define arch_timer_check_ool_workaround(t,a)		do { } while(0)
623 #define arch_timer_this_cpu_has_cntvct_wa()		({false;})
624 #define arch_timer_counter_has_wa()			({false;})
625 #endif /* CONFIG_ARM_ARCH_TIMER_OOL_WORKAROUND */
626 
627 static __always_inline irqreturn_t timer_handler(const int access,
628 					struct clock_event_device *evt)
629 {
630 	unsigned long ctrl;
631 
632 	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, evt);
633 	if (ctrl & ARCH_TIMER_CTRL_IT_STAT) {
634 		ctrl |= ARCH_TIMER_CTRL_IT_MASK;
635 		arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, evt);
636 		evt->event_handler(evt);
637 		return IRQ_HANDLED;
638 	}
639 
640 	return IRQ_NONE;
641 }
642 
643 static irqreturn_t arch_timer_handler_virt(int irq, void *dev_id)
644 {
645 	struct clock_event_device *evt = dev_id;
646 
647 	return timer_handler(ARCH_TIMER_VIRT_ACCESS, evt);
648 }
649 
650 static irqreturn_t arch_timer_handler_phys(int irq, void *dev_id)
651 {
652 	struct clock_event_device *evt = dev_id;
653 
654 	return timer_handler(ARCH_TIMER_PHYS_ACCESS, evt);
655 }
656 
657 static irqreturn_t arch_timer_handler_phys_mem(int irq, void *dev_id)
658 {
659 	struct clock_event_device *evt = dev_id;
660 
661 	return timer_handler(ARCH_TIMER_MEM_PHYS_ACCESS, evt);
662 }
663 
664 static irqreturn_t arch_timer_handler_virt_mem(int irq, void *dev_id)
665 {
666 	struct clock_event_device *evt = dev_id;
667 
668 	return timer_handler(ARCH_TIMER_MEM_VIRT_ACCESS, evt);
669 }
670 
671 static __always_inline int timer_shutdown(const int access,
672 					  struct clock_event_device *clk)
673 {
674 	unsigned long ctrl;
675 
676 	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
677 	ctrl &= ~ARCH_TIMER_CTRL_ENABLE;
678 	arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
679 
680 	return 0;
681 }
682 
683 static int arch_timer_shutdown_virt(struct clock_event_device *clk)
684 {
685 	return timer_shutdown(ARCH_TIMER_VIRT_ACCESS, clk);
686 }
687 
688 static int arch_timer_shutdown_phys(struct clock_event_device *clk)
689 {
690 	return timer_shutdown(ARCH_TIMER_PHYS_ACCESS, clk);
691 }
692 
693 static int arch_timer_shutdown_virt_mem(struct clock_event_device *clk)
694 {
695 	return timer_shutdown(ARCH_TIMER_MEM_VIRT_ACCESS, clk);
696 }
697 
698 static int arch_timer_shutdown_phys_mem(struct clock_event_device *clk)
699 {
700 	return timer_shutdown(ARCH_TIMER_MEM_PHYS_ACCESS, clk);
701 }
702 
703 static __always_inline void set_next_event(const int access, unsigned long evt,
704 					   struct clock_event_device *clk)
705 {
706 	unsigned long ctrl;
707 	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
708 	ctrl |= ARCH_TIMER_CTRL_ENABLE;
709 	ctrl &= ~ARCH_TIMER_CTRL_IT_MASK;
710 	arch_timer_reg_write(access, ARCH_TIMER_REG_TVAL, evt, clk);
711 	arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
712 }
713 
714 static int arch_timer_set_next_event_virt(unsigned long evt,
715 					  struct clock_event_device *clk)
716 {
717 	set_next_event(ARCH_TIMER_VIRT_ACCESS, evt, clk);
718 	return 0;
719 }
720 
721 static int arch_timer_set_next_event_phys(unsigned long evt,
722 					  struct clock_event_device *clk)
723 {
724 	set_next_event(ARCH_TIMER_PHYS_ACCESS, evt, clk);
725 	return 0;
726 }
727 
728 static int arch_timer_set_next_event_virt_mem(unsigned long evt,
729 					      struct clock_event_device *clk)
730 {
731 	set_next_event(ARCH_TIMER_MEM_VIRT_ACCESS, evt, clk);
732 	return 0;
733 }
734 
735 static int arch_timer_set_next_event_phys_mem(unsigned long evt,
736 					      struct clock_event_device *clk)
737 {
738 	set_next_event(ARCH_TIMER_MEM_PHYS_ACCESS, evt, clk);
739 	return 0;
740 }
741 
742 static void __arch_timer_setup(unsigned type,
743 			       struct clock_event_device *clk)
744 {
745 	clk->features = CLOCK_EVT_FEAT_ONESHOT;
746 
747 	if (type == ARCH_TIMER_TYPE_CP15) {
748 		typeof(clk->set_next_event) sne;
749 
750 		arch_timer_check_ool_workaround(ate_match_local_cap_id, NULL);
751 
752 		if (arch_timer_c3stop)
753 			clk->features |= CLOCK_EVT_FEAT_C3STOP;
754 		clk->name = "arch_sys_timer";
755 		clk->rating = 450;
756 		clk->cpumask = cpumask_of(smp_processor_id());
757 		clk->irq = arch_timer_ppi[arch_timer_uses_ppi];
758 		switch (arch_timer_uses_ppi) {
759 		case ARCH_TIMER_VIRT_PPI:
760 			clk->set_state_shutdown = arch_timer_shutdown_virt;
761 			clk->set_state_oneshot_stopped = arch_timer_shutdown_virt;
762 			sne = erratum_handler(set_next_event_virt);
763 			break;
764 		case ARCH_TIMER_PHYS_SECURE_PPI:
765 		case ARCH_TIMER_PHYS_NONSECURE_PPI:
766 		case ARCH_TIMER_HYP_PPI:
767 			clk->set_state_shutdown = arch_timer_shutdown_phys;
768 			clk->set_state_oneshot_stopped = arch_timer_shutdown_phys;
769 			sne = erratum_handler(set_next_event_phys);
770 			break;
771 		default:
772 			BUG();
773 		}
774 
775 		clk->set_next_event = sne;
776 	} else {
777 		clk->features |= CLOCK_EVT_FEAT_DYNIRQ;
778 		clk->name = "arch_mem_timer";
779 		clk->rating = 400;
780 		clk->cpumask = cpu_possible_mask;
781 		if (arch_timer_mem_use_virtual) {
782 			clk->set_state_shutdown = arch_timer_shutdown_virt_mem;
783 			clk->set_state_oneshot_stopped = arch_timer_shutdown_virt_mem;
784 			clk->set_next_event =
785 				arch_timer_set_next_event_virt_mem;
786 		} else {
787 			clk->set_state_shutdown = arch_timer_shutdown_phys_mem;
788 			clk->set_state_oneshot_stopped = arch_timer_shutdown_phys_mem;
789 			clk->set_next_event =
790 				arch_timer_set_next_event_phys_mem;
791 		}
792 	}
793 
794 	clk->set_state_shutdown(clk);
795 
796 	clockevents_config_and_register(clk, arch_timer_rate, 0xf, 0x7fffffff);
797 }
798 
799 static void arch_timer_evtstrm_enable(int divider)
800 {
801 	u32 cntkctl = arch_timer_get_cntkctl();
802 
803 	cntkctl &= ~ARCH_TIMER_EVT_TRIGGER_MASK;
804 	/* Set the divider and enable virtual event stream */
805 	cntkctl |= (divider << ARCH_TIMER_EVT_TRIGGER_SHIFT)
806 			| ARCH_TIMER_VIRT_EVT_EN;
807 	arch_timer_set_cntkctl(cntkctl);
808 	arch_timer_set_evtstrm_feature();
809 	cpumask_set_cpu(smp_processor_id(), &evtstrm_available);
810 }
811 
812 static void arch_timer_configure_evtstream(void)
813 {
814 	int evt_stream_div, pos;
815 
816 	/* Find the closest power of two to the divisor */
817 	evt_stream_div = arch_timer_rate / ARCH_TIMER_EVT_STREAM_FREQ;
818 	pos = fls(evt_stream_div);
819 	if (pos > 1 && !(evt_stream_div & (1 << (pos - 2))))
820 		pos--;
821 	/* enable event stream */
822 	arch_timer_evtstrm_enable(min(pos, 15));
823 }
824 
825 static void arch_counter_set_user_access(void)
826 {
827 	u32 cntkctl = arch_timer_get_cntkctl();
828 
829 	/* Disable user access to the timers and both counters */
830 	/* Also disable virtual event stream */
831 	cntkctl &= ~(ARCH_TIMER_USR_PT_ACCESS_EN
832 			| ARCH_TIMER_USR_VT_ACCESS_EN
833 		        | ARCH_TIMER_USR_VCT_ACCESS_EN
834 			| ARCH_TIMER_VIRT_EVT_EN
835 			| ARCH_TIMER_USR_PCT_ACCESS_EN);
836 
837 	/*
838 	 * Enable user access to the virtual counter if it doesn't
839 	 * need to be workaround. The vdso may have been already
840 	 * disabled though.
841 	 */
842 	if (arch_timer_this_cpu_has_cntvct_wa())
843 		pr_info("CPU%d: Trapping CNTVCT access\n", smp_processor_id());
844 	else
845 		cntkctl |= ARCH_TIMER_USR_VCT_ACCESS_EN;
846 
847 	arch_timer_set_cntkctl(cntkctl);
848 }
849 
850 static bool arch_timer_has_nonsecure_ppi(void)
851 {
852 	return (arch_timer_uses_ppi == ARCH_TIMER_PHYS_SECURE_PPI &&
853 		arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
854 }
855 
856 static u32 check_ppi_trigger(int irq)
857 {
858 	u32 flags = irq_get_trigger_type(irq);
859 
860 	if (flags != IRQF_TRIGGER_HIGH && flags != IRQF_TRIGGER_LOW) {
861 		pr_warn("WARNING: Invalid trigger for IRQ%d, assuming level low\n", irq);
862 		pr_warn("WARNING: Please fix your firmware\n");
863 		flags = IRQF_TRIGGER_LOW;
864 	}
865 
866 	return flags;
867 }
868 
869 static int arch_timer_starting_cpu(unsigned int cpu)
870 {
871 	struct clock_event_device *clk = this_cpu_ptr(arch_timer_evt);
872 	u32 flags;
873 
874 	__arch_timer_setup(ARCH_TIMER_TYPE_CP15, clk);
875 
876 	flags = check_ppi_trigger(arch_timer_ppi[arch_timer_uses_ppi]);
877 	enable_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], flags);
878 
879 	if (arch_timer_has_nonsecure_ppi()) {
880 		flags = check_ppi_trigger(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
881 		enable_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI],
882 				  flags);
883 	}
884 
885 	arch_counter_set_user_access();
886 	if (evtstrm_enable)
887 		arch_timer_configure_evtstream();
888 
889 	return 0;
890 }
891 
892 static int validate_timer_rate(void)
893 {
894 	if (!arch_timer_rate)
895 		return -EINVAL;
896 
897 	/* Arch timer frequency < 1MHz can cause trouble */
898 	WARN_ON(arch_timer_rate < 1000000);
899 
900 	return 0;
901 }
902 
903 /*
904  * For historical reasons, when probing with DT we use whichever (non-zero)
905  * rate was probed first, and don't verify that others match. If the first node
906  * probed has a clock-frequency property, this overrides the HW register.
907  */
908 static void arch_timer_of_configure_rate(u32 rate, struct device_node *np)
909 {
910 	/* Who has more than one independent system counter? */
911 	if (arch_timer_rate)
912 		return;
913 
914 	if (of_property_read_u32(np, "clock-frequency", &arch_timer_rate))
915 		arch_timer_rate = rate;
916 
917 	/* Check the timer frequency. */
918 	if (validate_timer_rate())
919 		pr_warn("frequency not available\n");
920 }
921 
922 static void arch_timer_banner(unsigned type)
923 {
924 	pr_info("%s%s%s timer(s) running at %lu.%02luMHz (%s%s%s).\n",
925 		type & ARCH_TIMER_TYPE_CP15 ? "cp15" : "",
926 		type == (ARCH_TIMER_TYPE_CP15 | ARCH_TIMER_TYPE_MEM) ?
927 			" and " : "",
928 		type & ARCH_TIMER_TYPE_MEM ? "mmio" : "",
929 		(unsigned long)arch_timer_rate / 1000000,
930 		(unsigned long)(arch_timer_rate / 10000) % 100,
931 		type & ARCH_TIMER_TYPE_CP15 ?
932 			(arch_timer_uses_ppi == ARCH_TIMER_VIRT_PPI) ? "virt" : "phys" :
933 			"",
934 		type == (ARCH_TIMER_TYPE_CP15 | ARCH_TIMER_TYPE_MEM) ? "/" : "",
935 		type & ARCH_TIMER_TYPE_MEM ?
936 			arch_timer_mem_use_virtual ? "virt" : "phys" :
937 			"");
938 }
939 
940 u32 arch_timer_get_rate(void)
941 {
942 	return arch_timer_rate;
943 }
944 
945 bool arch_timer_evtstrm_available(void)
946 {
947 	/*
948 	 * We might get called from a preemptible context. This is fine
949 	 * because availability of the event stream should be always the same
950 	 * for a preemptible context and context where we might resume a task.
951 	 */
952 	return cpumask_test_cpu(raw_smp_processor_id(), &evtstrm_available);
953 }
954 
955 static u64 arch_counter_get_cntvct_mem(void)
956 {
957 	u32 vct_lo, vct_hi, tmp_hi;
958 
959 	do {
960 		vct_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
961 		vct_lo = readl_relaxed(arch_counter_base + CNTVCT_LO);
962 		tmp_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
963 	} while (vct_hi != tmp_hi);
964 
965 	return ((u64) vct_hi << 32) | vct_lo;
966 }
967 
968 static struct arch_timer_kvm_info arch_timer_kvm_info;
969 
970 struct arch_timer_kvm_info *arch_timer_get_kvm_info(void)
971 {
972 	return &arch_timer_kvm_info;
973 }
974 
975 static void __init arch_counter_register(unsigned type)
976 {
977 	u64 start_count;
978 
979 	/* Register the CP15 based counter if we have one */
980 	if (type & ARCH_TIMER_TYPE_CP15) {
981 		u64 (*rd)(void);
982 
983 		if ((IS_ENABLED(CONFIG_ARM64) && !is_hyp_mode_available()) ||
984 		    arch_timer_uses_ppi == ARCH_TIMER_VIRT_PPI) {
985 			if (arch_timer_counter_has_wa())
986 				rd = arch_counter_get_cntvct_stable;
987 			else
988 				rd = arch_counter_get_cntvct;
989 		} else {
990 			if (arch_timer_counter_has_wa())
991 				rd = arch_counter_get_cntpct_stable;
992 			else
993 				rd = arch_counter_get_cntpct;
994 		}
995 
996 		arch_timer_read_counter = rd;
997 		clocksource_counter.vdso_clock_mode = vdso_default;
998 	} else {
999 		arch_timer_read_counter = arch_counter_get_cntvct_mem;
1000 	}
1001 
1002 	if (!arch_counter_suspend_stop)
1003 		clocksource_counter.flags |= CLOCK_SOURCE_SUSPEND_NONSTOP;
1004 	start_count = arch_timer_read_counter();
1005 	clocksource_register_hz(&clocksource_counter, arch_timer_rate);
1006 	cyclecounter.mult = clocksource_counter.mult;
1007 	cyclecounter.shift = clocksource_counter.shift;
1008 	timecounter_init(&arch_timer_kvm_info.timecounter,
1009 			 &cyclecounter, start_count);
1010 
1011 	/* 56 bits minimum, so we assume worst case rollover */
1012 	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
1013 }
1014 
1015 static void arch_timer_stop(struct clock_event_device *clk)
1016 {
1017 	pr_debug("disable IRQ%d cpu #%d\n", clk->irq, smp_processor_id());
1018 
1019 	disable_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi]);
1020 	if (arch_timer_has_nonsecure_ppi())
1021 		disable_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
1022 
1023 	clk->set_state_shutdown(clk);
1024 }
1025 
1026 static int arch_timer_dying_cpu(unsigned int cpu)
1027 {
1028 	struct clock_event_device *clk = this_cpu_ptr(arch_timer_evt);
1029 
1030 	cpumask_clear_cpu(smp_processor_id(), &evtstrm_available);
1031 
1032 	arch_timer_stop(clk);
1033 	return 0;
1034 }
1035 
1036 #ifdef CONFIG_CPU_PM
1037 static DEFINE_PER_CPU(unsigned long, saved_cntkctl);
1038 static int arch_timer_cpu_pm_notify(struct notifier_block *self,
1039 				    unsigned long action, void *hcpu)
1040 {
1041 	if (action == CPU_PM_ENTER) {
1042 		__this_cpu_write(saved_cntkctl, arch_timer_get_cntkctl());
1043 
1044 		cpumask_clear_cpu(smp_processor_id(), &evtstrm_available);
1045 	} else if (action == CPU_PM_ENTER_FAILED || action == CPU_PM_EXIT) {
1046 		arch_timer_set_cntkctl(__this_cpu_read(saved_cntkctl));
1047 
1048 		if (arch_timer_have_evtstrm_feature())
1049 			cpumask_set_cpu(smp_processor_id(), &evtstrm_available);
1050 	}
1051 	return NOTIFY_OK;
1052 }
1053 
1054 static struct notifier_block arch_timer_cpu_pm_notifier = {
1055 	.notifier_call = arch_timer_cpu_pm_notify,
1056 };
1057 
1058 static int __init arch_timer_cpu_pm_init(void)
1059 {
1060 	return cpu_pm_register_notifier(&arch_timer_cpu_pm_notifier);
1061 }
1062 
1063 static void __init arch_timer_cpu_pm_deinit(void)
1064 {
1065 	WARN_ON(cpu_pm_unregister_notifier(&arch_timer_cpu_pm_notifier));
1066 }
1067 
1068 #else
1069 static int __init arch_timer_cpu_pm_init(void)
1070 {
1071 	return 0;
1072 }
1073 
1074 static void __init arch_timer_cpu_pm_deinit(void)
1075 {
1076 }
1077 #endif
1078 
1079 static int __init arch_timer_register(void)
1080 {
1081 	int err;
1082 	int ppi;
1083 
1084 	arch_timer_evt = alloc_percpu(struct clock_event_device);
1085 	if (!arch_timer_evt) {
1086 		err = -ENOMEM;
1087 		goto out;
1088 	}
1089 
1090 	ppi = arch_timer_ppi[arch_timer_uses_ppi];
1091 	switch (arch_timer_uses_ppi) {
1092 	case ARCH_TIMER_VIRT_PPI:
1093 		err = request_percpu_irq(ppi, arch_timer_handler_virt,
1094 					 "arch_timer", arch_timer_evt);
1095 		break;
1096 	case ARCH_TIMER_PHYS_SECURE_PPI:
1097 	case ARCH_TIMER_PHYS_NONSECURE_PPI:
1098 		err = request_percpu_irq(ppi, arch_timer_handler_phys,
1099 					 "arch_timer", arch_timer_evt);
1100 		if (!err && arch_timer_has_nonsecure_ppi()) {
1101 			ppi = arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI];
1102 			err = request_percpu_irq(ppi, arch_timer_handler_phys,
1103 						 "arch_timer", arch_timer_evt);
1104 			if (err)
1105 				free_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_SECURE_PPI],
1106 						arch_timer_evt);
1107 		}
1108 		break;
1109 	case ARCH_TIMER_HYP_PPI:
1110 		err = request_percpu_irq(ppi, arch_timer_handler_phys,
1111 					 "arch_timer", arch_timer_evt);
1112 		break;
1113 	default:
1114 		BUG();
1115 	}
1116 
1117 	if (err) {
1118 		pr_err("can't register interrupt %d (%d)\n", ppi, err);
1119 		goto out_free;
1120 	}
1121 
1122 	err = arch_timer_cpu_pm_init();
1123 	if (err)
1124 		goto out_unreg_notify;
1125 
1126 	/* Register and immediately configure the timer on the boot CPU */
1127 	err = cpuhp_setup_state(CPUHP_AP_ARM_ARCH_TIMER_STARTING,
1128 				"clockevents/arm/arch_timer:starting",
1129 				arch_timer_starting_cpu, arch_timer_dying_cpu);
1130 	if (err)
1131 		goto out_unreg_cpupm;
1132 	return 0;
1133 
1134 out_unreg_cpupm:
1135 	arch_timer_cpu_pm_deinit();
1136 
1137 out_unreg_notify:
1138 	free_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], arch_timer_evt);
1139 	if (arch_timer_has_nonsecure_ppi())
1140 		free_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI],
1141 				arch_timer_evt);
1142 
1143 out_free:
1144 	free_percpu(arch_timer_evt);
1145 out:
1146 	return err;
1147 }
1148 
1149 static int __init arch_timer_mem_register(void __iomem *base, unsigned int irq)
1150 {
1151 	int ret;
1152 	irq_handler_t func;
1153 	struct arch_timer *t;
1154 
1155 	t = kzalloc(sizeof(*t), GFP_KERNEL);
1156 	if (!t)
1157 		return -ENOMEM;
1158 
1159 	t->base = base;
1160 	t->evt.irq = irq;
1161 	__arch_timer_setup(ARCH_TIMER_TYPE_MEM, &t->evt);
1162 
1163 	if (arch_timer_mem_use_virtual)
1164 		func = arch_timer_handler_virt_mem;
1165 	else
1166 		func = arch_timer_handler_phys_mem;
1167 
1168 	ret = request_irq(irq, func, IRQF_TIMER, "arch_mem_timer", &t->evt);
1169 	if (ret) {
1170 		pr_err("Failed to request mem timer irq\n");
1171 		kfree(t);
1172 	}
1173 
1174 	return ret;
1175 }
1176 
1177 static const struct of_device_id arch_timer_of_match[] __initconst = {
1178 	{ .compatible   = "arm,armv7-timer",    },
1179 	{ .compatible   = "arm,armv8-timer",    },
1180 	{},
1181 };
1182 
1183 static const struct of_device_id arch_timer_mem_of_match[] __initconst = {
1184 	{ .compatible   = "arm,armv7-timer-mem", },
1185 	{},
1186 };
1187 
1188 static bool __init arch_timer_needs_of_probing(void)
1189 {
1190 	struct device_node *dn;
1191 	bool needs_probing = false;
1192 	unsigned int mask = ARCH_TIMER_TYPE_CP15 | ARCH_TIMER_TYPE_MEM;
1193 
1194 	/* We have two timers, and both device-tree nodes are probed. */
1195 	if ((arch_timers_present & mask) == mask)
1196 		return false;
1197 
1198 	/*
1199 	 * Only one type of timer is probed,
1200 	 * check if we have another type of timer node in device-tree.
1201 	 */
1202 	if (arch_timers_present & ARCH_TIMER_TYPE_CP15)
1203 		dn = of_find_matching_node(NULL, arch_timer_mem_of_match);
1204 	else
1205 		dn = of_find_matching_node(NULL, arch_timer_of_match);
1206 
1207 	if (dn && of_device_is_available(dn))
1208 		needs_probing = true;
1209 
1210 	of_node_put(dn);
1211 
1212 	return needs_probing;
1213 }
1214 
1215 static int __init arch_timer_common_init(void)
1216 {
1217 	arch_timer_banner(arch_timers_present);
1218 	arch_counter_register(arch_timers_present);
1219 	return arch_timer_arch_init();
1220 }
1221 
1222 /**
1223  * arch_timer_select_ppi() - Select suitable PPI for the current system.
1224  *
1225  * If HYP mode is available, we know that the physical timer
1226  * has been configured to be accessible from PL1. Use it, so
1227  * that a guest can use the virtual timer instead.
1228  *
1229  * On ARMv8.1 with VH extensions, the kernel runs in HYP. VHE
1230  * accesses to CNTP_*_EL1 registers are silently redirected to
1231  * their CNTHP_*_EL2 counterparts, and use a different PPI
1232  * number.
1233  *
1234  * If no interrupt provided for virtual timer, we'll have to
1235  * stick to the physical timer. It'd better be accessible...
1236  * For arm64 we never use the secure interrupt.
1237  *
1238  * Return: a suitable PPI type for the current system.
1239  */
1240 static enum arch_timer_ppi_nr __init arch_timer_select_ppi(void)
1241 {
1242 	if (is_kernel_in_hyp_mode())
1243 		return ARCH_TIMER_HYP_PPI;
1244 
1245 	if (!is_hyp_mode_available() && arch_timer_ppi[ARCH_TIMER_VIRT_PPI])
1246 		return ARCH_TIMER_VIRT_PPI;
1247 
1248 	if (IS_ENABLED(CONFIG_ARM64))
1249 		return ARCH_TIMER_PHYS_NONSECURE_PPI;
1250 
1251 	return ARCH_TIMER_PHYS_SECURE_PPI;
1252 }
1253 
1254 static void __init arch_timer_populate_kvm_info(void)
1255 {
1256 	arch_timer_kvm_info.virtual_irq = arch_timer_ppi[ARCH_TIMER_VIRT_PPI];
1257 	if (is_kernel_in_hyp_mode())
1258 		arch_timer_kvm_info.physical_irq = arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI];
1259 }
1260 
1261 static int __init arch_timer_of_init(struct device_node *np)
1262 {
1263 	int i, ret;
1264 	u32 rate;
1265 
1266 	if (arch_timers_present & ARCH_TIMER_TYPE_CP15) {
1267 		pr_warn("multiple nodes in dt, skipping\n");
1268 		return 0;
1269 	}
1270 
1271 	arch_timers_present |= ARCH_TIMER_TYPE_CP15;
1272 	for (i = ARCH_TIMER_PHYS_SECURE_PPI; i < ARCH_TIMER_MAX_TIMER_PPI; i++)
1273 		arch_timer_ppi[i] = irq_of_parse_and_map(np, i);
1274 
1275 	arch_timer_populate_kvm_info();
1276 
1277 	rate = arch_timer_get_cntfrq();
1278 	arch_timer_of_configure_rate(rate, np);
1279 
1280 	arch_timer_c3stop = !of_property_read_bool(np, "always-on");
1281 
1282 	/* Check for globally applicable workarounds */
1283 	arch_timer_check_ool_workaround(ate_match_dt, np);
1284 
1285 	/*
1286 	 * If we cannot rely on firmware initializing the timer registers then
1287 	 * we should use the physical timers instead.
1288 	 */
1289 	if (IS_ENABLED(CONFIG_ARM) &&
1290 	    of_property_read_bool(np, "arm,cpu-registers-not-fw-configured"))
1291 		arch_timer_uses_ppi = ARCH_TIMER_PHYS_SECURE_PPI;
1292 	else
1293 		arch_timer_uses_ppi = arch_timer_select_ppi();
1294 
1295 	if (!arch_timer_ppi[arch_timer_uses_ppi]) {
1296 		pr_err("No interrupt available, giving up\n");
1297 		return -EINVAL;
1298 	}
1299 
1300 	/* On some systems, the counter stops ticking when in suspend. */
1301 	arch_counter_suspend_stop = of_property_read_bool(np,
1302 							 "arm,no-tick-in-suspend");
1303 
1304 	ret = arch_timer_register();
1305 	if (ret)
1306 		return ret;
1307 
1308 	if (arch_timer_needs_of_probing())
1309 		return 0;
1310 
1311 	return arch_timer_common_init();
1312 }
1313 TIMER_OF_DECLARE(armv7_arch_timer, "arm,armv7-timer", arch_timer_of_init);
1314 TIMER_OF_DECLARE(armv8_arch_timer, "arm,armv8-timer", arch_timer_of_init);
1315 
1316 static u32 __init
1317 arch_timer_mem_frame_get_cntfrq(struct arch_timer_mem_frame *frame)
1318 {
1319 	void __iomem *base;
1320 	u32 rate;
1321 
1322 	base = ioremap(frame->cntbase, frame->size);
1323 	if (!base) {
1324 		pr_err("Unable to map frame @ %pa\n", &frame->cntbase);
1325 		return 0;
1326 	}
1327 
1328 	rate = readl_relaxed(base + CNTFRQ);
1329 
1330 	iounmap(base);
1331 
1332 	return rate;
1333 }
1334 
1335 static struct arch_timer_mem_frame * __init
1336 arch_timer_mem_find_best_frame(struct arch_timer_mem *timer_mem)
1337 {
1338 	struct arch_timer_mem_frame *frame, *best_frame = NULL;
1339 	void __iomem *cntctlbase;
1340 	u32 cnttidr;
1341 	int i;
1342 
1343 	cntctlbase = ioremap(timer_mem->cntctlbase, timer_mem->size);
1344 	if (!cntctlbase) {
1345 		pr_err("Can't map CNTCTLBase @ %pa\n",
1346 			&timer_mem->cntctlbase);
1347 		return NULL;
1348 	}
1349 
1350 	cnttidr = readl_relaxed(cntctlbase + CNTTIDR);
1351 
1352 	/*
1353 	 * Try to find a virtual capable frame. Otherwise fall back to a
1354 	 * physical capable frame.
1355 	 */
1356 	for (i = 0; i < ARCH_TIMER_MEM_MAX_FRAMES; i++) {
1357 		u32 cntacr = CNTACR_RFRQ | CNTACR_RWPT | CNTACR_RPCT |
1358 			     CNTACR_RWVT | CNTACR_RVOFF | CNTACR_RVCT;
1359 
1360 		frame = &timer_mem->frame[i];
1361 		if (!frame->valid)
1362 			continue;
1363 
1364 		/* Try enabling everything, and see what sticks */
1365 		writel_relaxed(cntacr, cntctlbase + CNTACR(i));
1366 		cntacr = readl_relaxed(cntctlbase + CNTACR(i));
1367 
1368 		if ((cnttidr & CNTTIDR_VIRT(i)) &&
1369 		    !(~cntacr & (CNTACR_RWVT | CNTACR_RVCT))) {
1370 			best_frame = frame;
1371 			arch_timer_mem_use_virtual = true;
1372 			break;
1373 		}
1374 
1375 		if (~cntacr & (CNTACR_RWPT | CNTACR_RPCT))
1376 			continue;
1377 
1378 		best_frame = frame;
1379 	}
1380 
1381 	iounmap(cntctlbase);
1382 
1383 	return best_frame;
1384 }
1385 
1386 static int __init
1387 arch_timer_mem_frame_register(struct arch_timer_mem_frame *frame)
1388 {
1389 	void __iomem *base;
1390 	int ret, irq = 0;
1391 
1392 	if (arch_timer_mem_use_virtual)
1393 		irq = frame->virt_irq;
1394 	else
1395 		irq = frame->phys_irq;
1396 
1397 	if (!irq) {
1398 		pr_err("Frame missing %s irq.\n",
1399 		       arch_timer_mem_use_virtual ? "virt" : "phys");
1400 		return -EINVAL;
1401 	}
1402 
1403 	if (!request_mem_region(frame->cntbase, frame->size,
1404 				"arch_mem_timer"))
1405 		return -EBUSY;
1406 
1407 	base = ioremap(frame->cntbase, frame->size);
1408 	if (!base) {
1409 		pr_err("Can't map frame's registers\n");
1410 		return -ENXIO;
1411 	}
1412 
1413 	ret = arch_timer_mem_register(base, irq);
1414 	if (ret) {
1415 		iounmap(base);
1416 		return ret;
1417 	}
1418 
1419 	arch_counter_base = base;
1420 	arch_timers_present |= ARCH_TIMER_TYPE_MEM;
1421 
1422 	return 0;
1423 }
1424 
1425 static int __init arch_timer_mem_of_init(struct device_node *np)
1426 {
1427 	struct arch_timer_mem *timer_mem;
1428 	struct arch_timer_mem_frame *frame;
1429 	struct device_node *frame_node;
1430 	struct resource res;
1431 	int ret = -EINVAL;
1432 	u32 rate;
1433 
1434 	timer_mem = kzalloc(sizeof(*timer_mem), GFP_KERNEL);
1435 	if (!timer_mem)
1436 		return -ENOMEM;
1437 
1438 	if (of_address_to_resource(np, 0, &res))
1439 		goto out;
1440 	timer_mem->cntctlbase = res.start;
1441 	timer_mem->size = resource_size(&res);
1442 
1443 	for_each_available_child_of_node(np, frame_node) {
1444 		u32 n;
1445 		struct arch_timer_mem_frame *frame;
1446 
1447 		if (of_property_read_u32(frame_node, "frame-number", &n)) {
1448 			pr_err(FW_BUG "Missing frame-number.\n");
1449 			of_node_put(frame_node);
1450 			goto out;
1451 		}
1452 		if (n >= ARCH_TIMER_MEM_MAX_FRAMES) {
1453 			pr_err(FW_BUG "Wrong frame-number, only 0-%u are permitted.\n",
1454 			       ARCH_TIMER_MEM_MAX_FRAMES - 1);
1455 			of_node_put(frame_node);
1456 			goto out;
1457 		}
1458 		frame = &timer_mem->frame[n];
1459 
1460 		if (frame->valid) {
1461 			pr_err(FW_BUG "Duplicated frame-number.\n");
1462 			of_node_put(frame_node);
1463 			goto out;
1464 		}
1465 
1466 		if (of_address_to_resource(frame_node, 0, &res)) {
1467 			of_node_put(frame_node);
1468 			goto out;
1469 		}
1470 		frame->cntbase = res.start;
1471 		frame->size = resource_size(&res);
1472 
1473 		frame->virt_irq = irq_of_parse_and_map(frame_node,
1474 						       ARCH_TIMER_VIRT_SPI);
1475 		frame->phys_irq = irq_of_parse_and_map(frame_node,
1476 						       ARCH_TIMER_PHYS_SPI);
1477 
1478 		frame->valid = true;
1479 	}
1480 
1481 	frame = arch_timer_mem_find_best_frame(timer_mem);
1482 	if (!frame) {
1483 		pr_err("Unable to find a suitable frame in timer @ %pa\n",
1484 			&timer_mem->cntctlbase);
1485 		ret = -EINVAL;
1486 		goto out;
1487 	}
1488 
1489 	rate = arch_timer_mem_frame_get_cntfrq(frame);
1490 	arch_timer_of_configure_rate(rate, np);
1491 
1492 	ret = arch_timer_mem_frame_register(frame);
1493 	if (!ret && !arch_timer_needs_of_probing())
1494 		ret = arch_timer_common_init();
1495 out:
1496 	kfree(timer_mem);
1497 	return ret;
1498 }
1499 TIMER_OF_DECLARE(armv7_arch_timer_mem, "arm,armv7-timer-mem",
1500 		       arch_timer_mem_of_init);
1501 
1502 #ifdef CONFIG_ACPI_GTDT
1503 static int __init
1504 arch_timer_mem_verify_cntfrq(struct arch_timer_mem *timer_mem)
1505 {
1506 	struct arch_timer_mem_frame *frame;
1507 	u32 rate;
1508 	int i;
1509 
1510 	for (i = 0; i < ARCH_TIMER_MEM_MAX_FRAMES; i++) {
1511 		frame = &timer_mem->frame[i];
1512 
1513 		if (!frame->valid)
1514 			continue;
1515 
1516 		rate = arch_timer_mem_frame_get_cntfrq(frame);
1517 		if (rate == arch_timer_rate)
1518 			continue;
1519 
1520 		pr_err(FW_BUG "CNTFRQ mismatch: frame @ %pa: (0x%08lx), CPU: (0x%08lx)\n",
1521 			&frame->cntbase,
1522 			(unsigned long)rate, (unsigned long)arch_timer_rate);
1523 
1524 		return -EINVAL;
1525 	}
1526 
1527 	return 0;
1528 }
1529 
1530 static int __init arch_timer_mem_acpi_init(int platform_timer_count)
1531 {
1532 	struct arch_timer_mem *timers, *timer;
1533 	struct arch_timer_mem_frame *frame, *best_frame = NULL;
1534 	int timer_count, i, ret = 0;
1535 
1536 	timers = kcalloc(platform_timer_count, sizeof(*timers),
1537 			    GFP_KERNEL);
1538 	if (!timers)
1539 		return -ENOMEM;
1540 
1541 	ret = acpi_arch_timer_mem_init(timers, &timer_count);
1542 	if (ret || !timer_count)
1543 		goto out;
1544 
1545 	/*
1546 	 * While unlikely, it's theoretically possible that none of the frames
1547 	 * in a timer expose the combination of feature we want.
1548 	 */
1549 	for (i = 0; i < timer_count; i++) {
1550 		timer = &timers[i];
1551 
1552 		frame = arch_timer_mem_find_best_frame(timer);
1553 		if (!best_frame)
1554 			best_frame = frame;
1555 
1556 		ret = arch_timer_mem_verify_cntfrq(timer);
1557 		if (ret) {
1558 			pr_err("Disabling MMIO timers due to CNTFRQ mismatch\n");
1559 			goto out;
1560 		}
1561 
1562 		if (!best_frame) /* implies !frame */
1563 			/*
1564 			 * Only complain about missing suitable frames if we
1565 			 * haven't already found one in a previous iteration.
1566 			 */
1567 			pr_err("Unable to find a suitable frame in timer @ %pa\n",
1568 				&timer->cntctlbase);
1569 	}
1570 
1571 	if (best_frame)
1572 		ret = arch_timer_mem_frame_register(best_frame);
1573 out:
1574 	kfree(timers);
1575 	return ret;
1576 }
1577 
1578 /* Initialize per-processor generic timer and memory-mapped timer(if present) */
1579 static int __init arch_timer_acpi_init(struct acpi_table_header *table)
1580 {
1581 	int ret, platform_timer_count;
1582 
1583 	if (arch_timers_present & ARCH_TIMER_TYPE_CP15) {
1584 		pr_warn("already initialized, skipping\n");
1585 		return -EINVAL;
1586 	}
1587 
1588 	arch_timers_present |= ARCH_TIMER_TYPE_CP15;
1589 
1590 	ret = acpi_gtdt_init(table, &platform_timer_count);
1591 	if (ret)
1592 		return ret;
1593 
1594 	arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI] =
1595 		acpi_gtdt_map_ppi(ARCH_TIMER_PHYS_NONSECURE_PPI);
1596 
1597 	arch_timer_ppi[ARCH_TIMER_VIRT_PPI] =
1598 		acpi_gtdt_map_ppi(ARCH_TIMER_VIRT_PPI);
1599 
1600 	arch_timer_ppi[ARCH_TIMER_HYP_PPI] =
1601 		acpi_gtdt_map_ppi(ARCH_TIMER_HYP_PPI);
1602 
1603 	arch_timer_populate_kvm_info();
1604 
1605 	/*
1606 	 * When probing via ACPI, we have no mechanism to override the sysreg
1607 	 * CNTFRQ value. This *must* be correct.
1608 	 */
1609 	arch_timer_rate = arch_timer_get_cntfrq();
1610 	ret = validate_timer_rate();
1611 	if (ret) {
1612 		pr_err(FW_BUG "frequency not available.\n");
1613 		return ret;
1614 	}
1615 
1616 	arch_timer_uses_ppi = arch_timer_select_ppi();
1617 	if (!arch_timer_ppi[arch_timer_uses_ppi]) {
1618 		pr_err("No interrupt available, giving up\n");
1619 		return -EINVAL;
1620 	}
1621 
1622 	/* Always-on capability */
1623 	arch_timer_c3stop = acpi_gtdt_c3stop(arch_timer_uses_ppi);
1624 
1625 	/* Check for globally applicable workarounds */
1626 	arch_timer_check_ool_workaround(ate_match_acpi_oem_info, table);
1627 
1628 	ret = arch_timer_register();
1629 	if (ret)
1630 		return ret;
1631 
1632 	if (platform_timer_count &&
1633 	    arch_timer_mem_acpi_init(platform_timer_count))
1634 		pr_err("Failed to initialize memory-mapped timer.\n");
1635 
1636 	return arch_timer_common_init();
1637 }
1638 TIMER_ACPI_DECLARE(arch_timer, ACPI_SIG_GTDT, arch_timer_acpi_init);
1639 #endif
1640