1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright The Asahi Linux Contributors
4  *
5  * Based on irq-lpc32xx:
6  *   Copyright 2015-2016 Vladimir Zapolskiy <vz@mleia.com>
7  * Based on irq-bcm2836:
8  *   Copyright 2015 Broadcom
9  */
10 
11 /*
12  * AIC is a fairly simple interrupt controller with the following features:
13  *
14  * - 896 level-triggered hardware IRQs
15  *   - Single mask bit per IRQ
16  *   - Per-IRQ affinity setting
17  *   - Automatic masking on event delivery (auto-ack)
18  *   - Software triggering (ORed with hw line)
19  * - 2 per-CPU IPIs (meant as "self" and "other", but they are
20  *   interchangeable if not symmetric)
21  * - Automatic prioritization (single event/ack register per CPU, lower IRQs =
22  *   higher priority)
23  * - Automatic masking on ack
24  * - Default "this CPU" register view and explicit per-CPU views
25  *
26  * In addition, this driver also handles FIQs, as these are routed to the same
27  * IRQ vector. These are used for Fast IPIs, the ARMv8 timer IRQs, and
28  * performance counters (TODO).
29  *
30  * Implementation notes:
31  *
32  * - This driver creates two IRQ domains, one for HW IRQs and internal FIQs,
33  *   and one for IPIs.
34  * - Since Linux needs more than 2 IPIs, we implement a software IRQ controller
35  *   and funnel all IPIs into one per-CPU IPI (the second "self" IPI is unused).
36  * - FIQ hwirq numbers are assigned after true hwirqs, and are per-cpu.
37  * - DT bindings use 3-cell form (like GIC):
38  *   - <0 nr flags> - hwirq #nr
39  *   - <1 nr flags> - FIQ #nr
40  *     - nr=0  Physical HV timer
41  *     - nr=1  Virtual HV timer
42  *     - nr=2  Physical guest timer
43  *     - nr=3  Virtual guest timer
44  */
45 
46 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
47 
48 #include <linux/bits.h>
49 #include <linux/bitfield.h>
50 #include <linux/cpuhotplug.h>
51 #include <linux/io.h>
52 #include <linux/irqchip.h>
53 #include <linux/irqchip/arm-vgic-info.h>
54 #include <linux/irqdomain.h>
55 #include <linux/jump_label.h>
56 #include <linux/limits.h>
57 #include <linux/of_address.h>
58 #include <linux/slab.h>
59 #include <asm/apple_m1_pmu.h>
60 #include <asm/cputype.h>
61 #include <asm/exception.h>
62 #include <asm/sysreg.h>
63 #include <asm/virt.h>
64 
65 #include <dt-bindings/interrupt-controller/apple-aic.h>
66 
67 /*
68  * AIC v1 registers (MMIO)
69  */
70 
71 #define AIC_INFO		0x0004
72 #define AIC_INFO_NR_IRQ		GENMASK(15, 0)
73 
74 #define AIC_CONFIG		0x0010
75 
76 #define AIC_WHOAMI		0x2000
77 #define AIC_EVENT		0x2004
78 #define AIC_EVENT_DIE		GENMASK(31, 24)
79 #define AIC_EVENT_TYPE		GENMASK(23, 16)
80 #define AIC_EVENT_NUM		GENMASK(15, 0)
81 
82 #define AIC_EVENT_TYPE_FIQ	0 /* Software use */
83 #define AIC_EVENT_TYPE_IRQ	1
84 #define AIC_EVENT_TYPE_IPI	4
85 #define AIC_EVENT_IPI_OTHER	1
86 #define AIC_EVENT_IPI_SELF	2
87 
88 #define AIC_IPI_SEND		0x2008
89 #define AIC_IPI_ACK		0x200c
90 #define AIC_IPI_MASK_SET	0x2024
91 #define AIC_IPI_MASK_CLR	0x2028
92 
93 #define AIC_IPI_SEND_CPU(cpu)	BIT(cpu)
94 
95 #define AIC_IPI_OTHER		BIT(0)
96 #define AIC_IPI_SELF		BIT(31)
97 
98 #define AIC_TARGET_CPU		0x3000
99 
100 #define AIC_CPU_IPI_SET(cpu)	(0x5008 + ((cpu) << 7))
101 #define AIC_CPU_IPI_CLR(cpu)	(0x500c + ((cpu) << 7))
102 #define AIC_CPU_IPI_MASK_SET(cpu) (0x5024 + ((cpu) << 7))
103 #define AIC_CPU_IPI_MASK_CLR(cpu) (0x5028 + ((cpu) << 7))
104 
105 #define AIC_MAX_IRQ		0x400
106 
107 /*
108  * AIC v2 registers (MMIO)
109  */
110 
111 #define AIC2_VERSION		0x0000
112 #define AIC2_VERSION_VER	GENMASK(7, 0)
113 
114 #define AIC2_INFO1		0x0004
115 #define AIC2_INFO1_NR_IRQ	GENMASK(15, 0)
116 #define AIC2_INFO1_LAST_DIE	GENMASK(27, 24)
117 
118 #define AIC2_INFO2		0x0008
119 
120 #define AIC2_INFO3		0x000c
121 #define AIC2_INFO3_MAX_IRQ	GENMASK(15, 0)
122 #define AIC2_INFO3_MAX_DIE	GENMASK(27, 24)
123 
124 #define AIC2_RESET		0x0010
125 #define AIC2_RESET_RESET	BIT(0)
126 
127 #define AIC2_CONFIG		0x0014
128 #define AIC2_CONFIG_ENABLE	BIT(0)
129 #define AIC2_CONFIG_PREFER_PCPU	BIT(28)
130 
131 #define AIC2_TIMEOUT		0x0028
132 #define AIC2_CLUSTER_PRIO	0x0030
133 #define AIC2_DELAY_GROUPS	0x0100
134 
135 #define AIC2_IRQ_CFG		0x2000
136 
137 /*
138  * AIC2 registers are laid out like this, starting at AIC2_IRQ_CFG:
139  *
140  * Repeat for each die:
141  *   IRQ_CFG: u32 * MAX_IRQS
142  *   SW_SET: u32 * (MAX_IRQS / 32)
143  *   SW_CLR: u32 * (MAX_IRQS / 32)
144  *   MASK_SET: u32 * (MAX_IRQS / 32)
145  *   MASK_CLR: u32 * (MAX_IRQS / 32)
146  *   HW_STATE: u32 * (MAX_IRQS / 32)
147  *
148  * This is followed by a set of event registers, each 16K page aligned.
149  * The first one is the AP event register we will use. Unfortunately,
150  * the actual implemented die count is not specified anywhere in the
151  * capability registers, so we have to explicitly specify the event
152  * register as a second reg entry in the device tree to remain
153  * forward-compatible.
154  */
155 
156 #define AIC2_IRQ_CFG_TARGET	GENMASK(3, 0)
157 #define AIC2_IRQ_CFG_DELAY_IDX	GENMASK(7, 5)
158 
159 #define MASK_REG(x)		(4 * ((x) >> 5))
160 #define MASK_BIT(x)		BIT((x) & GENMASK(4, 0))
161 
162 /*
163  * IMP-DEF sysregs that control FIQ sources
164  */
165 
166 /* IPI request registers */
167 #define SYS_IMP_APL_IPI_RR_LOCAL_EL1	sys_reg(3, 5, 15, 0, 0)
168 #define SYS_IMP_APL_IPI_RR_GLOBAL_EL1	sys_reg(3, 5, 15, 0, 1)
169 #define IPI_RR_CPU			GENMASK(7, 0)
170 /* Cluster only used for the GLOBAL register */
171 #define IPI_RR_CLUSTER			GENMASK(23, 16)
172 #define IPI_RR_TYPE			GENMASK(29, 28)
173 #define IPI_RR_IMMEDIATE		0
174 #define IPI_RR_RETRACT			1
175 #define IPI_RR_DEFERRED			2
176 #define IPI_RR_NOWAKE			3
177 
178 /* IPI status register */
179 #define SYS_IMP_APL_IPI_SR_EL1		sys_reg(3, 5, 15, 1, 1)
180 #define IPI_SR_PENDING			BIT(0)
181 
182 /* Guest timer FIQ enable register */
183 #define SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2	sys_reg(3, 5, 15, 1, 3)
184 #define VM_TMR_FIQ_ENABLE_V		BIT(0)
185 #define VM_TMR_FIQ_ENABLE_P		BIT(1)
186 
187 /* Deferred IPI countdown register */
188 #define SYS_IMP_APL_IPI_CR_EL1		sys_reg(3, 5, 15, 3, 1)
189 
190 /* Uncore PMC control register */
191 #define SYS_IMP_APL_UPMCR0_EL1		sys_reg(3, 7, 15, 0, 4)
192 #define UPMCR0_IMODE			GENMASK(18, 16)
193 #define UPMCR0_IMODE_OFF		0
194 #define UPMCR0_IMODE_AIC		2
195 #define UPMCR0_IMODE_HALT		3
196 #define UPMCR0_IMODE_FIQ		4
197 
198 /* Uncore PMC status register */
199 #define SYS_IMP_APL_UPMSR_EL1		sys_reg(3, 7, 15, 6, 4)
200 #define UPMSR_IACT			BIT(0)
201 
202 /* MPIDR fields */
203 #define MPIDR_CPU(x)			MPIDR_AFFINITY_LEVEL(x, 0)
204 #define MPIDR_CLUSTER(x)		MPIDR_AFFINITY_LEVEL(x, 1)
205 
206 #define AIC_IRQ_HWIRQ(die, irq)	(FIELD_PREP(AIC_EVENT_DIE, die) | \
207 				 FIELD_PREP(AIC_EVENT_TYPE, AIC_EVENT_TYPE_IRQ) | \
208 				 FIELD_PREP(AIC_EVENT_NUM, irq))
209 #define AIC_FIQ_HWIRQ(x)	(FIELD_PREP(AIC_EVENT_TYPE, AIC_EVENT_TYPE_FIQ) | \
210 				 FIELD_PREP(AIC_EVENT_NUM, x))
211 #define AIC_HWIRQ_IRQ(x)	FIELD_GET(AIC_EVENT_NUM, x)
212 #define AIC_HWIRQ_DIE(x)	FIELD_GET(AIC_EVENT_DIE, x)
213 #define AIC_NR_FIQ		6
214 #define AIC_NR_SWIPI		32
215 
216 /*
217  * FIQ hwirq index definitions: FIQ sources use the DT binding defines
218  * directly, except that timers are special. At the irqchip level, the
219  * two timer types are represented by their access method: _EL0 registers
220  * or _EL02 registers. In the DT binding, the timers are represented
221  * by their purpose (HV or guest). This mapping is for when the kernel is
222  * running at EL2 (with VHE). When the kernel is running at EL1, the
223  * mapping differs and aic_irq_domain_translate() performs the remapping.
224  */
225 
226 #define AIC_TMR_EL0_PHYS	AIC_TMR_HV_PHYS
227 #define AIC_TMR_EL0_VIRT	AIC_TMR_HV_VIRT
228 #define AIC_TMR_EL02_PHYS	AIC_TMR_GUEST_PHYS
229 #define AIC_TMR_EL02_VIRT	AIC_TMR_GUEST_VIRT
230 
231 static DEFINE_STATIC_KEY_TRUE(use_fast_ipi);
232 
233 struct aic_info {
234 	int version;
235 
236 	/* Register offsets */
237 	u32 event;
238 	u32 target_cpu;
239 	u32 irq_cfg;
240 	u32 sw_set;
241 	u32 sw_clr;
242 	u32 mask_set;
243 	u32 mask_clr;
244 
245 	u32 die_stride;
246 
247 	/* Features */
248 	bool fast_ipi;
249 };
250 
251 static const struct aic_info aic1_info __initconst = {
252 	.version	= 1,
253 
254 	.event		= AIC_EVENT,
255 	.target_cpu	= AIC_TARGET_CPU,
256 };
257 
258 static const struct aic_info aic1_fipi_info __initconst = {
259 	.version	= 1,
260 
261 	.event		= AIC_EVENT,
262 	.target_cpu	= AIC_TARGET_CPU,
263 
264 	.fast_ipi	= true,
265 };
266 
267 static const struct aic_info aic2_info __initconst = {
268 	.version	= 2,
269 
270 	.irq_cfg	= AIC2_IRQ_CFG,
271 
272 	.fast_ipi	= true,
273 };
274 
275 static const struct of_device_id aic_info_match[] = {
276 	{
277 		.compatible = "apple,t8103-aic",
278 		.data = &aic1_fipi_info,
279 	},
280 	{
281 		.compatible = "apple,aic",
282 		.data = &aic1_info,
283 	},
284 	{
285 		.compatible = "apple,aic2",
286 		.data = &aic2_info,
287 	},
288 	{}
289 };
290 
291 struct aic_irq_chip {
292 	void __iomem *base;
293 	void __iomem *event;
294 	struct irq_domain *hw_domain;
295 	struct {
296 		cpumask_t aff;
297 	} *fiq_aff[AIC_NR_FIQ];
298 
299 	int nr_irq;
300 	int max_irq;
301 	int nr_die;
302 	int max_die;
303 
304 	struct aic_info info;
305 };
306 
307 static DEFINE_PER_CPU(uint32_t, aic_fiq_unmasked);
308 
309 static struct aic_irq_chip *aic_irqc;
310 
311 static void aic_handle_ipi(struct pt_regs *regs);
312 
313 static u32 aic_ic_read(struct aic_irq_chip *ic, u32 reg)
314 {
315 	return readl_relaxed(ic->base + reg);
316 }
317 
318 static void aic_ic_write(struct aic_irq_chip *ic, u32 reg, u32 val)
319 {
320 	writel_relaxed(val, ic->base + reg);
321 }
322 
323 /*
324  * IRQ irqchip
325  */
326 
327 static void aic_irq_mask(struct irq_data *d)
328 {
329 	irq_hw_number_t hwirq = irqd_to_hwirq(d);
330 	struct aic_irq_chip *ic = irq_data_get_irq_chip_data(d);
331 
332 	u32 off = AIC_HWIRQ_DIE(hwirq) * ic->info.die_stride;
333 	u32 irq = AIC_HWIRQ_IRQ(hwirq);
334 
335 	aic_ic_write(ic, ic->info.mask_set + off + MASK_REG(irq), MASK_BIT(irq));
336 }
337 
338 static void aic_irq_unmask(struct irq_data *d)
339 {
340 	irq_hw_number_t hwirq = irqd_to_hwirq(d);
341 	struct aic_irq_chip *ic = irq_data_get_irq_chip_data(d);
342 
343 	u32 off = AIC_HWIRQ_DIE(hwirq) * ic->info.die_stride;
344 	u32 irq = AIC_HWIRQ_IRQ(hwirq);
345 
346 	aic_ic_write(ic, ic->info.mask_clr + off + MASK_REG(irq), MASK_BIT(irq));
347 }
348 
349 static void aic_irq_eoi(struct irq_data *d)
350 {
351 	/*
352 	 * Reading the interrupt reason automatically acknowledges and masks
353 	 * the IRQ, so we just unmask it here if needed.
354 	 */
355 	if (!irqd_irq_masked(d))
356 		aic_irq_unmask(d);
357 }
358 
359 static void __exception_irq_entry aic_handle_irq(struct pt_regs *regs)
360 {
361 	struct aic_irq_chip *ic = aic_irqc;
362 	u32 event, type, irq;
363 
364 	do {
365 		/*
366 		 * We cannot use a relaxed read here, as reads from DMA buffers
367 		 * need to be ordered after the IRQ fires.
368 		 */
369 		event = readl(ic->event + ic->info.event);
370 		type = FIELD_GET(AIC_EVENT_TYPE, event);
371 		irq = FIELD_GET(AIC_EVENT_NUM, event);
372 
373 		if (type == AIC_EVENT_TYPE_IRQ)
374 			generic_handle_domain_irq(aic_irqc->hw_domain, event);
375 		else if (type == AIC_EVENT_TYPE_IPI && irq == 1)
376 			aic_handle_ipi(regs);
377 		else if (event != 0)
378 			pr_err_ratelimited("Unknown IRQ event %d, %d\n", type, irq);
379 	} while (event);
380 
381 	/*
382 	 * vGIC maintenance interrupts end up here too, so we need to check
383 	 * for them separately. This should never trigger if KVM is working
384 	 * properly, because it will have already taken care of clearing it
385 	 * on guest exit before this handler runs.
386 	 */
387 	if (is_kernel_in_hyp_mode() && (read_sysreg_s(SYS_ICH_HCR_EL2) & ICH_HCR_EN) &&
388 		read_sysreg_s(SYS_ICH_MISR_EL2) != 0) {
389 		pr_err_ratelimited("vGIC IRQ fired and not handled by KVM, disabling.\n");
390 		sysreg_clear_set_s(SYS_ICH_HCR_EL2, ICH_HCR_EN, 0);
391 	}
392 }
393 
394 static int aic_irq_set_affinity(struct irq_data *d,
395 				const struct cpumask *mask_val, bool force)
396 {
397 	irq_hw_number_t hwirq = irqd_to_hwirq(d);
398 	struct aic_irq_chip *ic = irq_data_get_irq_chip_data(d);
399 	int cpu;
400 
401 	BUG_ON(!ic->info.target_cpu);
402 
403 	if (force)
404 		cpu = cpumask_first(mask_val);
405 	else
406 		cpu = cpumask_any_and(mask_val, cpu_online_mask);
407 
408 	aic_ic_write(ic, ic->info.target_cpu + AIC_HWIRQ_IRQ(hwirq) * 4, BIT(cpu));
409 	irq_data_update_effective_affinity(d, cpumask_of(cpu));
410 
411 	return IRQ_SET_MASK_OK;
412 }
413 
414 static int aic_irq_set_type(struct irq_data *d, unsigned int type)
415 {
416 	/*
417 	 * Some IRQs (e.g. MSIs) implicitly have edge semantics, and we don't
418 	 * have a way to find out the type of any given IRQ, so just allow both.
419 	 */
420 	return (type == IRQ_TYPE_LEVEL_HIGH || type == IRQ_TYPE_EDGE_RISING) ? 0 : -EINVAL;
421 }
422 
423 static struct irq_chip aic_chip = {
424 	.name = "AIC",
425 	.irq_mask = aic_irq_mask,
426 	.irq_unmask = aic_irq_unmask,
427 	.irq_eoi = aic_irq_eoi,
428 	.irq_set_affinity = aic_irq_set_affinity,
429 	.irq_set_type = aic_irq_set_type,
430 };
431 
432 static struct irq_chip aic2_chip = {
433 	.name = "AIC2",
434 	.irq_mask = aic_irq_mask,
435 	.irq_unmask = aic_irq_unmask,
436 	.irq_eoi = aic_irq_eoi,
437 	.irq_set_type = aic_irq_set_type,
438 };
439 
440 /*
441  * FIQ irqchip
442  */
443 
444 static unsigned long aic_fiq_get_idx(struct irq_data *d)
445 {
446 	return AIC_HWIRQ_IRQ(irqd_to_hwirq(d));
447 }
448 
449 static void aic_fiq_set_mask(struct irq_data *d)
450 {
451 	/* Only the guest timers have real mask bits, unfortunately. */
452 	switch (aic_fiq_get_idx(d)) {
453 	case AIC_TMR_EL02_PHYS:
454 		sysreg_clear_set_s(SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2, VM_TMR_FIQ_ENABLE_P, 0);
455 		isb();
456 		break;
457 	case AIC_TMR_EL02_VIRT:
458 		sysreg_clear_set_s(SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2, VM_TMR_FIQ_ENABLE_V, 0);
459 		isb();
460 		break;
461 	default:
462 		break;
463 	}
464 }
465 
466 static void aic_fiq_clear_mask(struct irq_data *d)
467 {
468 	switch (aic_fiq_get_idx(d)) {
469 	case AIC_TMR_EL02_PHYS:
470 		sysreg_clear_set_s(SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2, 0, VM_TMR_FIQ_ENABLE_P);
471 		isb();
472 		break;
473 	case AIC_TMR_EL02_VIRT:
474 		sysreg_clear_set_s(SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2, 0, VM_TMR_FIQ_ENABLE_V);
475 		isb();
476 		break;
477 	default:
478 		break;
479 	}
480 }
481 
482 static void aic_fiq_mask(struct irq_data *d)
483 {
484 	aic_fiq_set_mask(d);
485 	__this_cpu_and(aic_fiq_unmasked, ~BIT(aic_fiq_get_idx(d)));
486 }
487 
488 static void aic_fiq_unmask(struct irq_data *d)
489 {
490 	aic_fiq_clear_mask(d);
491 	__this_cpu_or(aic_fiq_unmasked, BIT(aic_fiq_get_idx(d)));
492 }
493 
494 static void aic_fiq_eoi(struct irq_data *d)
495 {
496 	/* We mask to ack (where we can), so we need to unmask at EOI. */
497 	if (__this_cpu_read(aic_fiq_unmasked) & BIT(aic_fiq_get_idx(d)))
498 		aic_fiq_clear_mask(d);
499 }
500 
501 #define TIMER_FIRING(x)                                                        \
502 	(((x) & (ARCH_TIMER_CTRL_ENABLE | ARCH_TIMER_CTRL_IT_MASK |            \
503 		 ARCH_TIMER_CTRL_IT_STAT)) ==                                  \
504 	 (ARCH_TIMER_CTRL_ENABLE | ARCH_TIMER_CTRL_IT_STAT))
505 
506 static void __exception_irq_entry aic_handle_fiq(struct pt_regs *regs)
507 {
508 	/*
509 	 * It would be really nice if we had a system register that lets us get
510 	 * the FIQ source state without having to peek down into sources...
511 	 * but such a register does not seem to exist.
512 	 *
513 	 * So, we have these potential sources to test for:
514 	 *  - Fast IPIs (not yet used)
515 	 *  - The 4 timers (CNTP, CNTV for each of HV and guest)
516 	 *  - Per-core PMCs (not yet supported)
517 	 *  - Per-cluster uncore PMCs (not yet supported)
518 	 *
519 	 * Since not dealing with any of these results in a FIQ storm,
520 	 * we check for everything here, even things we don't support yet.
521 	 */
522 
523 	if (read_sysreg_s(SYS_IMP_APL_IPI_SR_EL1) & IPI_SR_PENDING) {
524 		if (static_branch_likely(&use_fast_ipi)) {
525 			aic_handle_ipi(regs);
526 		} else {
527 			pr_err_ratelimited("Fast IPI fired. Acking.\n");
528 			write_sysreg_s(IPI_SR_PENDING, SYS_IMP_APL_IPI_SR_EL1);
529 		}
530 	}
531 
532 	if (TIMER_FIRING(read_sysreg(cntp_ctl_el0)))
533 		generic_handle_domain_irq(aic_irqc->hw_domain,
534 					  AIC_FIQ_HWIRQ(AIC_TMR_EL0_PHYS));
535 
536 	if (TIMER_FIRING(read_sysreg(cntv_ctl_el0)))
537 		generic_handle_domain_irq(aic_irqc->hw_domain,
538 					  AIC_FIQ_HWIRQ(AIC_TMR_EL0_VIRT));
539 
540 	if (is_kernel_in_hyp_mode()) {
541 		uint64_t enabled = read_sysreg_s(SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2);
542 
543 		if ((enabled & VM_TMR_FIQ_ENABLE_P) &&
544 		    TIMER_FIRING(read_sysreg_s(SYS_CNTP_CTL_EL02)))
545 			generic_handle_domain_irq(aic_irqc->hw_domain,
546 						  AIC_FIQ_HWIRQ(AIC_TMR_EL02_PHYS));
547 
548 		if ((enabled & VM_TMR_FIQ_ENABLE_V) &&
549 		    TIMER_FIRING(read_sysreg_s(SYS_CNTV_CTL_EL02)))
550 			generic_handle_domain_irq(aic_irqc->hw_domain,
551 						  AIC_FIQ_HWIRQ(AIC_TMR_EL02_VIRT));
552 	}
553 
554 	if (read_sysreg_s(SYS_IMP_APL_PMCR0_EL1) & PMCR0_IACT) {
555 		int irq;
556 		if (cpumask_test_cpu(smp_processor_id(),
557 				     &aic_irqc->fiq_aff[AIC_CPU_PMU_P]->aff))
558 			irq = AIC_CPU_PMU_P;
559 		else
560 			irq = AIC_CPU_PMU_E;
561 		generic_handle_domain_irq(aic_irqc->hw_domain,
562 					  AIC_FIQ_HWIRQ(irq));
563 	}
564 
565 	if (FIELD_GET(UPMCR0_IMODE, read_sysreg_s(SYS_IMP_APL_UPMCR0_EL1)) == UPMCR0_IMODE_FIQ &&
566 			(read_sysreg_s(SYS_IMP_APL_UPMSR_EL1) & UPMSR_IACT)) {
567 		/* Same story with uncore PMCs */
568 		pr_err_ratelimited("Uncore PMC FIQ fired. Masking.\n");
569 		sysreg_clear_set_s(SYS_IMP_APL_UPMCR0_EL1, UPMCR0_IMODE,
570 				   FIELD_PREP(UPMCR0_IMODE, UPMCR0_IMODE_OFF));
571 	}
572 }
573 
574 static int aic_fiq_set_type(struct irq_data *d, unsigned int type)
575 {
576 	return (type == IRQ_TYPE_LEVEL_HIGH) ? 0 : -EINVAL;
577 }
578 
579 static struct irq_chip fiq_chip = {
580 	.name = "AIC-FIQ",
581 	.irq_mask = aic_fiq_mask,
582 	.irq_unmask = aic_fiq_unmask,
583 	.irq_ack = aic_fiq_set_mask,
584 	.irq_eoi = aic_fiq_eoi,
585 	.irq_set_type = aic_fiq_set_type,
586 };
587 
588 /*
589  * Main IRQ domain
590  */
591 
592 static int aic_irq_domain_map(struct irq_domain *id, unsigned int irq,
593 			      irq_hw_number_t hw)
594 {
595 	struct aic_irq_chip *ic = id->host_data;
596 	u32 type = FIELD_GET(AIC_EVENT_TYPE, hw);
597 	struct irq_chip *chip = &aic_chip;
598 
599 	if (ic->info.version == 2)
600 		chip = &aic2_chip;
601 
602 	if (type == AIC_EVENT_TYPE_IRQ) {
603 		irq_domain_set_info(id, irq, hw, chip, id->host_data,
604 				    handle_fasteoi_irq, NULL, NULL);
605 		irqd_set_single_target(irq_desc_get_irq_data(irq_to_desc(irq)));
606 	} else {
607 		int fiq = FIELD_GET(AIC_EVENT_NUM, hw);
608 
609 		switch (fiq) {
610 		case AIC_CPU_PMU_P:
611 		case AIC_CPU_PMU_E:
612 			irq_set_percpu_devid_partition(irq, &ic->fiq_aff[fiq]->aff);
613 			break;
614 		default:
615 			irq_set_percpu_devid(irq);
616 			break;
617 		}
618 
619 		irq_domain_set_info(id, irq, hw, &fiq_chip, id->host_data,
620 				    handle_percpu_devid_irq, NULL, NULL);
621 	}
622 
623 	return 0;
624 }
625 
626 static int aic_irq_domain_translate(struct irq_domain *id,
627 				    struct irq_fwspec *fwspec,
628 				    unsigned long *hwirq,
629 				    unsigned int *type)
630 {
631 	struct aic_irq_chip *ic = id->host_data;
632 	u32 *args;
633 	u32 die = 0;
634 
635 	if (fwspec->param_count < 3 || fwspec->param_count > 4 ||
636 	    !is_of_node(fwspec->fwnode))
637 		return -EINVAL;
638 
639 	args = &fwspec->param[1];
640 
641 	if (fwspec->param_count == 4) {
642 		die = args[0];
643 		args++;
644 	}
645 
646 	switch (fwspec->param[0]) {
647 	case AIC_IRQ:
648 		if (die >= ic->nr_die)
649 			return -EINVAL;
650 		if (args[0] >= ic->nr_irq)
651 			return -EINVAL;
652 		*hwirq = AIC_IRQ_HWIRQ(die, args[0]);
653 		break;
654 	case AIC_FIQ:
655 		if (die != 0)
656 			return -EINVAL;
657 		if (args[0] >= AIC_NR_FIQ)
658 			return -EINVAL;
659 		*hwirq = AIC_FIQ_HWIRQ(args[0]);
660 
661 		/*
662 		 * In EL1 the non-redirected registers are the guest's,
663 		 * not EL2's, so remap the hwirqs to match.
664 		 */
665 		if (!is_kernel_in_hyp_mode()) {
666 			switch (args[0]) {
667 			case AIC_TMR_GUEST_PHYS:
668 				*hwirq = AIC_FIQ_HWIRQ(AIC_TMR_EL0_PHYS);
669 				break;
670 			case AIC_TMR_GUEST_VIRT:
671 				*hwirq = AIC_FIQ_HWIRQ(AIC_TMR_EL0_VIRT);
672 				break;
673 			case AIC_TMR_HV_PHYS:
674 			case AIC_TMR_HV_VIRT:
675 				return -ENOENT;
676 			default:
677 				break;
678 			}
679 		}
680 		break;
681 	default:
682 		return -EINVAL;
683 	}
684 
685 	*type = args[1] & IRQ_TYPE_SENSE_MASK;
686 
687 	return 0;
688 }
689 
690 static int aic_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
691 				unsigned int nr_irqs, void *arg)
692 {
693 	unsigned int type = IRQ_TYPE_NONE;
694 	struct irq_fwspec *fwspec = arg;
695 	irq_hw_number_t hwirq;
696 	int i, ret;
697 
698 	ret = aic_irq_domain_translate(domain, fwspec, &hwirq, &type);
699 	if (ret)
700 		return ret;
701 
702 	for (i = 0; i < nr_irqs; i++) {
703 		ret = aic_irq_domain_map(domain, virq + i, hwirq + i);
704 		if (ret)
705 			return ret;
706 	}
707 
708 	return 0;
709 }
710 
711 static void aic_irq_domain_free(struct irq_domain *domain, unsigned int virq,
712 				unsigned int nr_irqs)
713 {
714 	int i;
715 
716 	for (i = 0; i < nr_irqs; i++) {
717 		struct irq_data *d = irq_domain_get_irq_data(domain, virq + i);
718 
719 		irq_set_handler(virq + i, NULL);
720 		irq_domain_reset_irq_data(d);
721 	}
722 }
723 
724 static const struct irq_domain_ops aic_irq_domain_ops = {
725 	.translate	= aic_irq_domain_translate,
726 	.alloc		= aic_irq_domain_alloc,
727 	.free		= aic_irq_domain_free,
728 };
729 
730 /*
731  * IPI irqchip
732  */
733 
734 static void aic_ipi_send_fast(int cpu)
735 {
736 	u64 mpidr = cpu_logical_map(cpu);
737 	u64 my_mpidr = read_cpuid_mpidr();
738 	u64 cluster = MPIDR_CLUSTER(mpidr);
739 	u64 idx = MPIDR_CPU(mpidr);
740 
741 	if (MPIDR_CLUSTER(my_mpidr) == cluster)
742 		write_sysreg_s(FIELD_PREP(IPI_RR_CPU, idx),
743 			       SYS_IMP_APL_IPI_RR_LOCAL_EL1);
744 	else
745 		write_sysreg_s(FIELD_PREP(IPI_RR_CPU, idx) | FIELD_PREP(IPI_RR_CLUSTER, cluster),
746 			       SYS_IMP_APL_IPI_RR_GLOBAL_EL1);
747 	isb();
748 }
749 
750 static void aic_handle_ipi(struct pt_regs *regs)
751 {
752 	/*
753 	 * Ack the IPI. We need to order this after the AIC event read, but
754 	 * that is enforced by normal MMIO ordering guarantees.
755 	 *
756 	 * For the Fast IPI case, this needs to be ordered before the vIPI
757 	 * handling below, so we need to isb();
758 	 */
759 	if (static_branch_likely(&use_fast_ipi)) {
760 		write_sysreg_s(IPI_SR_PENDING, SYS_IMP_APL_IPI_SR_EL1);
761 		isb();
762 	} else {
763 		aic_ic_write(aic_irqc, AIC_IPI_ACK, AIC_IPI_OTHER);
764 	}
765 
766 	ipi_mux_process();
767 
768 	/*
769 	 * No ordering needed here; at worst this just changes the timing of
770 	 * when the next IPI will be delivered.
771 	 */
772 	if (!static_branch_likely(&use_fast_ipi))
773 		aic_ic_write(aic_irqc, AIC_IPI_MASK_CLR, AIC_IPI_OTHER);
774 }
775 
776 static void aic_ipi_send_single(unsigned int cpu)
777 {
778 	if (static_branch_likely(&use_fast_ipi))
779 		aic_ipi_send_fast(cpu);
780 	else
781 		aic_ic_write(aic_irqc, AIC_IPI_SEND, AIC_IPI_SEND_CPU(cpu));
782 }
783 
784 static int __init aic_init_smp(struct aic_irq_chip *irqc, struct device_node *node)
785 {
786 	int base_ipi;
787 
788 	base_ipi = ipi_mux_create(AIC_NR_SWIPI, aic_ipi_send_single);
789 	if (WARN_ON(base_ipi <= 0))
790 		return -ENODEV;
791 
792 	set_smp_ipi_range(base_ipi, AIC_NR_SWIPI);
793 
794 	return 0;
795 }
796 
797 static int aic_init_cpu(unsigned int cpu)
798 {
799 	/* Mask all hard-wired per-CPU IRQ/FIQ sources */
800 
801 	/* Pending Fast IPI FIQs */
802 	write_sysreg_s(IPI_SR_PENDING, SYS_IMP_APL_IPI_SR_EL1);
803 
804 	/* Timer FIQs */
805 	sysreg_clear_set(cntp_ctl_el0, 0, ARCH_TIMER_CTRL_IT_MASK);
806 	sysreg_clear_set(cntv_ctl_el0, 0, ARCH_TIMER_CTRL_IT_MASK);
807 
808 	/* EL2-only (VHE mode) IRQ sources */
809 	if (is_kernel_in_hyp_mode()) {
810 		/* Guest timers */
811 		sysreg_clear_set_s(SYS_IMP_APL_VM_TMR_FIQ_ENA_EL2,
812 				   VM_TMR_FIQ_ENABLE_V | VM_TMR_FIQ_ENABLE_P, 0);
813 
814 		/* vGIC maintenance IRQ */
815 		sysreg_clear_set_s(SYS_ICH_HCR_EL2, ICH_HCR_EN, 0);
816 	}
817 
818 	/* PMC FIQ */
819 	sysreg_clear_set_s(SYS_IMP_APL_PMCR0_EL1, PMCR0_IMODE | PMCR0_IACT,
820 			   FIELD_PREP(PMCR0_IMODE, PMCR0_IMODE_OFF));
821 
822 	/* Uncore PMC FIQ */
823 	sysreg_clear_set_s(SYS_IMP_APL_UPMCR0_EL1, UPMCR0_IMODE,
824 			   FIELD_PREP(UPMCR0_IMODE, UPMCR0_IMODE_OFF));
825 
826 	/* Commit all of the above */
827 	isb();
828 
829 	if (aic_irqc->info.version == 1) {
830 		/*
831 		 * Make sure the kernel's idea of logical CPU order is the same as AIC's
832 		 * If we ever end up with a mismatch here, we will have to introduce
833 		 * a mapping table similar to what other irqchip drivers do.
834 		 */
835 		WARN_ON(aic_ic_read(aic_irqc, AIC_WHOAMI) != smp_processor_id());
836 
837 		/*
838 		 * Always keep IPIs unmasked at the hardware level (except auto-masking
839 		 * by AIC during processing). We manage masks at the vIPI level.
840 		 * These registers only exist on AICv1, AICv2 always uses fast IPIs.
841 		 */
842 		aic_ic_write(aic_irqc, AIC_IPI_ACK, AIC_IPI_SELF | AIC_IPI_OTHER);
843 		if (static_branch_likely(&use_fast_ipi)) {
844 			aic_ic_write(aic_irqc, AIC_IPI_MASK_SET, AIC_IPI_SELF | AIC_IPI_OTHER);
845 		} else {
846 			aic_ic_write(aic_irqc, AIC_IPI_MASK_SET, AIC_IPI_SELF);
847 			aic_ic_write(aic_irqc, AIC_IPI_MASK_CLR, AIC_IPI_OTHER);
848 		}
849 	}
850 
851 	/* Initialize the local mask state */
852 	__this_cpu_write(aic_fiq_unmasked, 0);
853 
854 	return 0;
855 }
856 
857 static struct gic_kvm_info vgic_info __initdata = {
858 	.type			= GIC_V3,
859 	.no_maint_irq_mask	= true,
860 	.no_hw_deactivation	= true,
861 };
862 
863 static void build_fiq_affinity(struct aic_irq_chip *ic, struct device_node *aff)
864 {
865 	int i, n;
866 	u32 fiq;
867 
868 	if (of_property_read_u32(aff, "apple,fiq-index", &fiq) ||
869 	    WARN_ON(fiq >= AIC_NR_FIQ) || ic->fiq_aff[fiq])
870 		return;
871 
872 	n = of_property_count_elems_of_size(aff, "cpus", sizeof(u32));
873 	if (WARN_ON(n < 0))
874 		return;
875 
876 	ic->fiq_aff[fiq] = kzalloc(sizeof(*ic->fiq_aff[fiq]), GFP_KERNEL);
877 	if (!ic->fiq_aff[fiq])
878 		return;
879 
880 	for (i = 0; i < n; i++) {
881 		struct device_node *cpu_node;
882 		u32 cpu_phandle;
883 		int cpu;
884 
885 		if (of_property_read_u32_index(aff, "cpus", i, &cpu_phandle))
886 			continue;
887 
888 		cpu_node = of_find_node_by_phandle(cpu_phandle);
889 		if (WARN_ON(!cpu_node))
890 			continue;
891 
892 		cpu = of_cpu_node_to_id(cpu_node);
893 		of_node_put(cpu_node);
894 		if (WARN_ON(cpu < 0))
895 			continue;
896 
897 		cpumask_set_cpu(cpu, &ic->fiq_aff[fiq]->aff);
898 	}
899 }
900 
901 static int __init aic_of_ic_init(struct device_node *node, struct device_node *parent)
902 {
903 	int i, die;
904 	u32 off, start_off;
905 	void __iomem *regs;
906 	struct aic_irq_chip *irqc;
907 	struct device_node *affs;
908 	const struct of_device_id *match;
909 
910 	regs = of_iomap(node, 0);
911 	if (WARN_ON(!regs))
912 		return -EIO;
913 
914 	irqc = kzalloc(sizeof(*irqc), GFP_KERNEL);
915 	if (!irqc) {
916 		iounmap(regs);
917 		return -ENOMEM;
918 	}
919 
920 	irqc->base = regs;
921 
922 	match = of_match_node(aic_info_match, node);
923 	if (!match)
924 		goto err_unmap;
925 
926 	irqc->info = *(struct aic_info *)match->data;
927 
928 	aic_irqc = irqc;
929 
930 	switch (irqc->info.version) {
931 	case 1: {
932 		u32 info;
933 
934 		info = aic_ic_read(irqc, AIC_INFO);
935 		irqc->nr_irq = FIELD_GET(AIC_INFO_NR_IRQ, info);
936 		irqc->max_irq = AIC_MAX_IRQ;
937 		irqc->nr_die = irqc->max_die = 1;
938 
939 		off = start_off = irqc->info.target_cpu;
940 		off += sizeof(u32) * irqc->max_irq; /* TARGET_CPU */
941 
942 		irqc->event = irqc->base;
943 
944 		break;
945 	}
946 	case 2: {
947 		u32 info1, info3;
948 
949 		info1 = aic_ic_read(irqc, AIC2_INFO1);
950 		info3 = aic_ic_read(irqc, AIC2_INFO3);
951 
952 		irqc->nr_irq = FIELD_GET(AIC2_INFO1_NR_IRQ, info1);
953 		irqc->max_irq = FIELD_GET(AIC2_INFO3_MAX_IRQ, info3);
954 		irqc->nr_die = FIELD_GET(AIC2_INFO1_LAST_DIE, info1) + 1;
955 		irqc->max_die = FIELD_GET(AIC2_INFO3_MAX_DIE, info3);
956 
957 		off = start_off = irqc->info.irq_cfg;
958 		off += sizeof(u32) * irqc->max_irq; /* IRQ_CFG */
959 
960 		irqc->event = of_iomap(node, 1);
961 		if (WARN_ON(!irqc->event))
962 			goto err_unmap;
963 
964 		break;
965 	}
966 	}
967 
968 	irqc->info.sw_set = off;
969 	off += sizeof(u32) * (irqc->max_irq >> 5); /* SW_SET */
970 	irqc->info.sw_clr = off;
971 	off += sizeof(u32) * (irqc->max_irq >> 5); /* SW_CLR */
972 	irqc->info.mask_set = off;
973 	off += sizeof(u32) * (irqc->max_irq >> 5); /* MASK_SET */
974 	irqc->info.mask_clr = off;
975 	off += sizeof(u32) * (irqc->max_irq >> 5); /* MASK_CLR */
976 	off += sizeof(u32) * (irqc->max_irq >> 5); /* HW_STATE */
977 
978 	if (irqc->info.fast_ipi)
979 		static_branch_enable(&use_fast_ipi);
980 	else
981 		static_branch_disable(&use_fast_ipi);
982 
983 	irqc->info.die_stride = off - start_off;
984 
985 	irqc->hw_domain = irq_domain_create_tree(of_node_to_fwnode(node),
986 						 &aic_irq_domain_ops, irqc);
987 	if (WARN_ON(!irqc->hw_domain))
988 		goto err_unmap;
989 
990 	irq_domain_update_bus_token(irqc->hw_domain, DOMAIN_BUS_WIRED);
991 
992 	if (aic_init_smp(irqc, node))
993 		goto err_remove_domain;
994 
995 	affs = of_get_child_by_name(node, "affinities");
996 	if (affs) {
997 		struct device_node *chld;
998 
999 		for_each_child_of_node(affs, chld)
1000 			build_fiq_affinity(irqc, chld);
1001 	}
1002 	of_node_put(affs);
1003 
1004 	set_handle_irq(aic_handle_irq);
1005 	set_handle_fiq(aic_handle_fiq);
1006 
1007 	off = 0;
1008 	for (die = 0; die < irqc->nr_die; die++) {
1009 		for (i = 0; i < BITS_TO_U32(irqc->nr_irq); i++)
1010 			aic_ic_write(irqc, irqc->info.mask_set + off + i * 4, U32_MAX);
1011 		for (i = 0; i < BITS_TO_U32(irqc->nr_irq); i++)
1012 			aic_ic_write(irqc, irqc->info.sw_clr + off + i * 4, U32_MAX);
1013 		if (irqc->info.target_cpu)
1014 			for (i = 0; i < irqc->nr_irq; i++)
1015 				aic_ic_write(irqc, irqc->info.target_cpu + off + i * 4, 1);
1016 		off += irqc->info.die_stride;
1017 	}
1018 
1019 	if (irqc->info.version == 2) {
1020 		u32 config = aic_ic_read(irqc, AIC2_CONFIG);
1021 
1022 		config |= AIC2_CONFIG_ENABLE;
1023 		aic_ic_write(irqc, AIC2_CONFIG, config);
1024 	}
1025 
1026 	if (!is_kernel_in_hyp_mode())
1027 		pr_info("Kernel running in EL1, mapping interrupts");
1028 
1029 	if (static_branch_likely(&use_fast_ipi))
1030 		pr_info("Using Fast IPIs");
1031 
1032 	cpuhp_setup_state(CPUHP_AP_IRQ_APPLE_AIC_STARTING,
1033 			  "irqchip/apple-aic/ipi:starting",
1034 			  aic_init_cpu, NULL);
1035 
1036 	vgic_set_kvm_info(&vgic_info);
1037 
1038 	pr_info("Initialized with %d/%d IRQs * %d/%d die(s), %d FIQs, %d vIPIs",
1039 		irqc->nr_irq, irqc->max_irq, irqc->nr_die, irqc->max_die, AIC_NR_FIQ, AIC_NR_SWIPI);
1040 
1041 	return 0;
1042 
1043 err_remove_domain:
1044 	irq_domain_remove(irqc->hw_domain);
1045 err_unmap:
1046 	if (irqc->event && irqc->event != irqc->base)
1047 		iounmap(irqc->event);
1048 	iounmap(irqc->base);
1049 	kfree(irqc);
1050 	return -ENODEV;
1051 }
1052 
1053 IRQCHIP_DECLARE(apple_aic, "apple,aic", aic_of_ic_init);
1054 IRQCHIP_DECLARE(apple_aic2, "apple,aic2", aic_of_ic_init);
1055