xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 276e552e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/coproc.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
9  *          Christoffer Dall <c.dall@virtualopensystems.com>
10  */
11 
12 #include <linux/bitfield.h>
13 #include <linux/bsearch.h>
14 #include <linux/kvm_host.h>
15 #include <linux/mm.h>
16 #include <linux/printk.h>
17 #include <linux/uaccess.h>
18 
19 #include <asm/cacheflush.h>
20 #include <asm/cputype.h>
21 #include <asm/debug-monitors.h>
22 #include <asm/esr.h>
23 #include <asm/kvm_arm.h>
24 #include <asm/kvm_emulate.h>
25 #include <asm/kvm_hyp.h>
26 #include <asm/kvm_mmu.h>
27 #include <asm/perf_event.h>
28 #include <asm/sysreg.h>
29 
30 #include <trace/events/kvm.h>
31 
32 #include "sys_regs.h"
33 
34 #include "trace.h"
35 
36 /*
37  * All of this file is extremely similar to the ARM coproc.c, but the
38  * types are different. My gut feeling is that it should be pretty
39  * easy to merge, but that would be an ABI breakage -- again. VFP
40  * would also need to be abstracted.
41  *
42  * For AArch32, we only take care of what is being trapped. Anything
43  * that has to do with init and userspace access has to go via the
44  * 64bit interface.
45  */
46 
47 #define reg_to_encoding(x)						\
48 	sys_reg((u32)(x)->Op0, (u32)(x)->Op1,				\
49 		(u32)(x)->CRn, (u32)(x)->CRm, (u32)(x)->Op2)
50 
51 static bool read_from_write_only(struct kvm_vcpu *vcpu,
52 				 struct sys_reg_params *params,
53 				 const struct sys_reg_desc *r)
54 {
55 	WARN_ONCE(1, "Unexpected sys_reg read to write-only register\n");
56 	print_sys_reg_instr(params);
57 	kvm_inject_undefined(vcpu);
58 	return false;
59 }
60 
61 static bool write_to_read_only(struct kvm_vcpu *vcpu,
62 			       struct sys_reg_params *params,
63 			       const struct sys_reg_desc *r)
64 {
65 	WARN_ONCE(1, "Unexpected sys_reg write to read-only register\n");
66 	print_sys_reg_instr(params);
67 	kvm_inject_undefined(vcpu);
68 	return false;
69 }
70 
71 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
72 {
73 	u64 val = 0x8badf00d8badf00d;
74 
75 	if (vcpu->arch.sysregs_loaded_on_cpu &&
76 	    __vcpu_read_sys_reg_from_cpu(reg, &val))
77 		return val;
78 
79 	return __vcpu_sys_reg(vcpu, reg);
80 }
81 
82 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
83 {
84 	if (vcpu->arch.sysregs_loaded_on_cpu &&
85 	    __vcpu_write_sys_reg_to_cpu(val, reg))
86 		return;
87 
88 	 __vcpu_sys_reg(vcpu, reg) = val;
89 }
90 
91 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
92 static u32 cache_levels;
93 
94 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
95 #define CSSELR_MAX 14
96 
97 /* Which cache CCSIDR represents depends on CSSELR value. */
98 static u32 get_ccsidr(u32 csselr)
99 {
100 	u32 ccsidr;
101 
102 	/* Make sure noone else changes CSSELR during this! */
103 	local_irq_disable();
104 	write_sysreg(csselr, csselr_el1);
105 	isb();
106 	ccsidr = read_sysreg(ccsidr_el1);
107 	local_irq_enable();
108 
109 	return ccsidr;
110 }
111 
112 /*
113  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
114  */
115 static bool access_dcsw(struct kvm_vcpu *vcpu,
116 			struct sys_reg_params *p,
117 			const struct sys_reg_desc *r)
118 {
119 	if (!p->is_write)
120 		return read_from_write_only(vcpu, p, r);
121 
122 	/*
123 	 * Only track S/W ops if we don't have FWB. It still indicates
124 	 * that the guest is a bit broken (S/W operations should only
125 	 * be done by firmware, knowing that there is only a single
126 	 * CPU left in the system, and certainly not from non-secure
127 	 * software).
128 	 */
129 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
130 		kvm_set_way_flush(vcpu);
131 
132 	return true;
133 }
134 
135 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
136 {
137 	switch (r->aarch32_map) {
138 	case AA32_LO:
139 		*mask = GENMASK_ULL(31, 0);
140 		*shift = 0;
141 		break;
142 	case AA32_HI:
143 		*mask = GENMASK_ULL(63, 32);
144 		*shift = 32;
145 		break;
146 	default:
147 		*mask = GENMASK_ULL(63, 0);
148 		*shift = 0;
149 		break;
150 	}
151 }
152 
153 /*
154  * Generic accessor for VM registers. Only called as long as HCR_TVM
155  * is set. If the guest enables the MMU, we stop trapping the VM
156  * sys_regs and leave it in complete control of the caches.
157  */
158 static bool access_vm_reg(struct kvm_vcpu *vcpu,
159 			  struct sys_reg_params *p,
160 			  const struct sys_reg_desc *r)
161 {
162 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
163 	u64 val, mask, shift;
164 
165 	BUG_ON(!p->is_write);
166 
167 	get_access_mask(r, &mask, &shift);
168 
169 	if (~mask) {
170 		val = vcpu_read_sys_reg(vcpu, r->reg);
171 		val &= ~mask;
172 	} else {
173 		val = 0;
174 	}
175 
176 	val |= (p->regval & (mask >> shift)) << shift;
177 	vcpu_write_sys_reg(vcpu, val, r->reg);
178 
179 	kvm_toggle_cache(vcpu, was_enabled);
180 	return true;
181 }
182 
183 static bool access_actlr(struct kvm_vcpu *vcpu,
184 			 struct sys_reg_params *p,
185 			 const struct sys_reg_desc *r)
186 {
187 	u64 mask, shift;
188 
189 	if (p->is_write)
190 		return ignore_write(vcpu, p);
191 
192 	get_access_mask(r, &mask, &shift);
193 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
194 
195 	return true;
196 }
197 
198 /*
199  * Trap handler for the GICv3 SGI generation system register.
200  * Forward the request to the VGIC emulation.
201  * The cp15_64 code makes sure this automatically works
202  * for both AArch64 and AArch32 accesses.
203  */
204 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
205 			   struct sys_reg_params *p,
206 			   const struct sys_reg_desc *r)
207 {
208 	bool g1;
209 
210 	if (!p->is_write)
211 		return read_from_write_only(vcpu, p, r);
212 
213 	/*
214 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
215 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
216 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
217 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
218 	 * group.
219 	 */
220 	if (p->Op0 == 0) {		/* AArch32 */
221 		switch (p->Op1) {
222 		default:		/* Keep GCC quiet */
223 		case 0:			/* ICC_SGI1R */
224 			g1 = true;
225 			break;
226 		case 1:			/* ICC_ASGI1R */
227 		case 2:			/* ICC_SGI0R */
228 			g1 = false;
229 			break;
230 		}
231 	} else {			/* AArch64 */
232 		switch (p->Op2) {
233 		default:		/* Keep GCC quiet */
234 		case 5:			/* ICC_SGI1R_EL1 */
235 			g1 = true;
236 			break;
237 		case 6:			/* ICC_ASGI1R_EL1 */
238 		case 7:			/* ICC_SGI0R_EL1 */
239 			g1 = false;
240 			break;
241 		}
242 	}
243 
244 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
245 
246 	return true;
247 }
248 
249 static bool access_gic_sre(struct kvm_vcpu *vcpu,
250 			   struct sys_reg_params *p,
251 			   const struct sys_reg_desc *r)
252 {
253 	if (p->is_write)
254 		return ignore_write(vcpu, p);
255 
256 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
257 	return true;
258 }
259 
260 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
261 			struct sys_reg_params *p,
262 			const struct sys_reg_desc *r)
263 {
264 	if (p->is_write)
265 		return ignore_write(vcpu, p);
266 	else
267 		return read_zero(vcpu, p);
268 }
269 
270 /*
271  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
272  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
273  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
274  * treat it separately.
275  */
276 static bool trap_loregion(struct kvm_vcpu *vcpu,
277 			  struct sys_reg_params *p,
278 			  const struct sys_reg_desc *r)
279 {
280 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
281 	u32 sr = reg_to_encoding(r);
282 
283 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
284 		kvm_inject_undefined(vcpu);
285 		return false;
286 	}
287 
288 	if (p->is_write && sr == SYS_LORID_EL1)
289 		return write_to_read_only(vcpu, p, r);
290 
291 	return trap_raz_wi(vcpu, p, r);
292 }
293 
294 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
295 			   struct sys_reg_params *p,
296 			   const struct sys_reg_desc *r)
297 {
298 	if (p->is_write) {
299 		return ignore_write(vcpu, p);
300 	} else {
301 		p->regval = (1 << 3);
302 		return true;
303 	}
304 }
305 
306 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
307 				   struct sys_reg_params *p,
308 				   const struct sys_reg_desc *r)
309 {
310 	if (p->is_write) {
311 		return ignore_write(vcpu, p);
312 	} else {
313 		p->regval = read_sysreg(dbgauthstatus_el1);
314 		return true;
315 	}
316 }
317 
318 /*
319  * We want to avoid world-switching all the DBG registers all the
320  * time:
321  *
322  * - If we've touched any debug register, it is likely that we're
323  *   going to touch more of them. It then makes sense to disable the
324  *   traps and start doing the save/restore dance
325  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
326  *   then mandatory to save/restore the registers, as the guest
327  *   depends on them.
328  *
329  * For this, we use a DIRTY bit, indicating the guest has modified the
330  * debug registers, used as follow:
331  *
332  * On guest entry:
333  * - If the dirty bit is set (because we're coming back from trapping),
334  *   disable the traps, save host registers, restore guest registers.
335  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
336  *   set the dirty bit, disable the traps, save host registers,
337  *   restore guest registers.
338  * - Otherwise, enable the traps
339  *
340  * On guest exit:
341  * - If the dirty bit is set, save guest registers, restore host
342  *   registers and clear the dirty bit. This ensure that the host can
343  *   now use the debug registers.
344  */
345 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
346 			    struct sys_reg_params *p,
347 			    const struct sys_reg_desc *r)
348 {
349 	if (p->is_write) {
350 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
351 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
352 	} else {
353 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
354 	}
355 
356 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
357 
358 	return true;
359 }
360 
361 /*
362  * reg_to_dbg/dbg_to_reg
363  *
364  * A 32 bit write to a debug register leave top bits alone
365  * A 32 bit read from a debug register only returns the bottom bits
366  *
367  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
368  * hyp.S code switches between host and guest values in future.
369  */
370 static void reg_to_dbg(struct kvm_vcpu *vcpu,
371 		       struct sys_reg_params *p,
372 		       const struct sys_reg_desc *rd,
373 		       u64 *dbg_reg)
374 {
375 	u64 mask, shift, val;
376 
377 	get_access_mask(rd, &mask, &shift);
378 
379 	val = *dbg_reg;
380 	val &= ~mask;
381 	val |= (p->regval & (mask >> shift)) << shift;
382 	*dbg_reg = val;
383 
384 	vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
385 }
386 
387 static void dbg_to_reg(struct kvm_vcpu *vcpu,
388 		       struct sys_reg_params *p,
389 		       const struct sys_reg_desc *rd,
390 		       u64 *dbg_reg)
391 {
392 	u64 mask, shift;
393 
394 	get_access_mask(rd, &mask, &shift);
395 	p->regval = (*dbg_reg & mask) >> shift;
396 }
397 
398 static bool trap_bvr(struct kvm_vcpu *vcpu,
399 		     struct sys_reg_params *p,
400 		     const struct sys_reg_desc *rd)
401 {
402 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
403 
404 	if (p->is_write)
405 		reg_to_dbg(vcpu, p, rd, dbg_reg);
406 	else
407 		dbg_to_reg(vcpu, p, rd, dbg_reg);
408 
409 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
410 
411 	return true;
412 }
413 
414 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
415 		const struct kvm_one_reg *reg, void __user *uaddr)
416 {
417 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
418 
419 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
420 		return -EFAULT;
421 	return 0;
422 }
423 
424 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
425 	const struct kvm_one_reg *reg, void __user *uaddr)
426 {
427 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
428 
429 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
430 		return -EFAULT;
431 	return 0;
432 }
433 
434 static void reset_bvr(struct kvm_vcpu *vcpu,
435 		      const struct sys_reg_desc *rd)
436 {
437 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
438 }
439 
440 static bool trap_bcr(struct kvm_vcpu *vcpu,
441 		     struct sys_reg_params *p,
442 		     const struct sys_reg_desc *rd)
443 {
444 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
445 
446 	if (p->is_write)
447 		reg_to_dbg(vcpu, p, rd, dbg_reg);
448 	else
449 		dbg_to_reg(vcpu, p, rd, dbg_reg);
450 
451 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
452 
453 	return true;
454 }
455 
456 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
457 		const struct kvm_one_reg *reg, void __user *uaddr)
458 {
459 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
460 
461 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
462 		return -EFAULT;
463 
464 	return 0;
465 }
466 
467 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
468 	const struct kvm_one_reg *reg, void __user *uaddr)
469 {
470 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
471 
472 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
473 		return -EFAULT;
474 	return 0;
475 }
476 
477 static void reset_bcr(struct kvm_vcpu *vcpu,
478 		      const struct sys_reg_desc *rd)
479 {
480 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
481 }
482 
483 static bool trap_wvr(struct kvm_vcpu *vcpu,
484 		     struct sys_reg_params *p,
485 		     const struct sys_reg_desc *rd)
486 {
487 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
488 
489 	if (p->is_write)
490 		reg_to_dbg(vcpu, p, rd, dbg_reg);
491 	else
492 		dbg_to_reg(vcpu, p, rd, dbg_reg);
493 
494 	trace_trap_reg(__func__, rd->reg, p->is_write,
495 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
496 
497 	return true;
498 }
499 
500 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
501 		const struct kvm_one_reg *reg, void __user *uaddr)
502 {
503 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
504 
505 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
506 		return -EFAULT;
507 	return 0;
508 }
509 
510 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
511 	const struct kvm_one_reg *reg, void __user *uaddr)
512 {
513 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
514 
515 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
516 		return -EFAULT;
517 	return 0;
518 }
519 
520 static void reset_wvr(struct kvm_vcpu *vcpu,
521 		      const struct sys_reg_desc *rd)
522 {
523 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
524 }
525 
526 static bool trap_wcr(struct kvm_vcpu *vcpu,
527 		     struct sys_reg_params *p,
528 		     const struct sys_reg_desc *rd)
529 {
530 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
531 
532 	if (p->is_write)
533 		reg_to_dbg(vcpu, p, rd, dbg_reg);
534 	else
535 		dbg_to_reg(vcpu, p, rd, dbg_reg);
536 
537 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
538 
539 	return true;
540 }
541 
542 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
543 		const struct kvm_one_reg *reg, void __user *uaddr)
544 {
545 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
546 
547 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
548 		return -EFAULT;
549 	return 0;
550 }
551 
552 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
553 	const struct kvm_one_reg *reg, void __user *uaddr)
554 {
555 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
556 
557 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
558 		return -EFAULT;
559 	return 0;
560 }
561 
562 static void reset_wcr(struct kvm_vcpu *vcpu,
563 		      const struct sys_reg_desc *rd)
564 {
565 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
566 }
567 
568 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
569 {
570 	u64 amair = read_sysreg(amair_el1);
571 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
572 }
573 
574 static void reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
575 {
576 	u64 actlr = read_sysreg(actlr_el1);
577 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
578 }
579 
580 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
581 {
582 	u64 mpidr;
583 
584 	/*
585 	 * Map the vcpu_id into the first three affinity level fields of
586 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
587 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
588 	 * of the GICv3 to be able to address each CPU directly when
589 	 * sending IPIs.
590 	 */
591 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
592 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
593 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
594 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
595 }
596 
597 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
598 				   const struct sys_reg_desc *r)
599 {
600 	if (kvm_vcpu_has_pmu(vcpu))
601 		return 0;
602 
603 	return REG_HIDDEN;
604 }
605 
606 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
607 {
608 	u64 pmcr, val;
609 
610 	/* No PMU available, PMCR_EL0 may UNDEF... */
611 	if (!kvm_arm_support_pmu_v3())
612 		return;
613 
614 	pmcr = read_sysreg(pmcr_el0);
615 	/*
616 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
617 	 * except PMCR.E resetting to zero.
618 	 */
619 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
620 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
621 	if (!system_supports_32bit_el0())
622 		val |= ARMV8_PMU_PMCR_LC;
623 	__vcpu_sys_reg(vcpu, r->reg) = val;
624 }
625 
626 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
627 {
628 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
629 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
630 
631 	if (!enabled)
632 		kvm_inject_undefined(vcpu);
633 
634 	return !enabled;
635 }
636 
637 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
638 {
639 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
640 }
641 
642 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
643 {
644 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
645 }
646 
647 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
648 {
649 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
650 }
651 
652 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
653 {
654 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
655 }
656 
657 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
658 			const struct sys_reg_desc *r)
659 {
660 	u64 val;
661 
662 	if (pmu_access_el0_disabled(vcpu))
663 		return false;
664 
665 	if (p->is_write) {
666 		/* Only update writeable bits of PMCR */
667 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
668 		val &= ~ARMV8_PMU_PMCR_MASK;
669 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
670 		if (!system_supports_32bit_el0())
671 			val |= ARMV8_PMU_PMCR_LC;
672 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
673 		kvm_pmu_handle_pmcr(vcpu, val);
674 		kvm_vcpu_pmu_restore_guest(vcpu);
675 	} else {
676 		/* PMCR.P & PMCR.C are RAZ */
677 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
678 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
679 		p->regval = val;
680 	}
681 
682 	return true;
683 }
684 
685 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
686 			  const struct sys_reg_desc *r)
687 {
688 	if (pmu_access_event_counter_el0_disabled(vcpu))
689 		return false;
690 
691 	if (p->is_write)
692 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
693 	else
694 		/* return PMSELR.SEL field */
695 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
696 			    & ARMV8_PMU_COUNTER_MASK;
697 
698 	return true;
699 }
700 
701 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
702 			  const struct sys_reg_desc *r)
703 {
704 	u64 pmceid, mask, shift;
705 
706 	BUG_ON(p->is_write);
707 
708 	if (pmu_access_el0_disabled(vcpu))
709 		return false;
710 
711 	get_access_mask(r, &mask, &shift);
712 
713 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
714 	pmceid &= mask;
715 	pmceid >>= shift;
716 
717 	p->regval = pmceid;
718 
719 	return true;
720 }
721 
722 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
723 {
724 	u64 pmcr, val;
725 
726 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
727 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
728 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
729 		kvm_inject_undefined(vcpu);
730 		return false;
731 	}
732 
733 	return true;
734 }
735 
736 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
737 			      struct sys_reg_params *p,
738 			      const struct sys_reg_desc *r)
739 {
740 	u64 idx = ~0UL;
741 
742 	if (r->CRn == 9 && r->CRm == 13) {
743 		if (r->Op2 == 2) {
744 			/* PMXEVCNTR_EL0 */
745 			if (pmu_access_event_counter_el0_disabled(vcpu))
746 				return false;
747 
748 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
749 			      & ARMV8_PMU_COUNTER_MASK;
750 		} else if (r->Op2 == 0) {
751 			/* PMCCNTR_EL0 */
752 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
753 				return false;
754 
755 			idx = ARMV8_PMU_CYCLE_IDX;
756 		}
757 	} else if (r->CRn == 0 && r->CRm == 9) {
758 		/* PMCCNTR */
759 		if (pmu_access_event_counter_el0_disabled(vcpu))
760 			return false;
761 
762 		idx = ARMV8_PMU_CYCLE_IDX;
763 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
764 		/* PMEVCNTRn_EL0 */
765 		if (pmu_access_event_counter_el0_disabled(vcpu))
766 			return false;
767 
768 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
769 	}
770 
771 	/* Catch any decoding mistake */
772 	WARN_ON(idx == ~0UL);
773 
774 	if (!pmu_counter_idx_valid(vcpu, idx))
775 		return false;
776 
777 	if (p->is_write) {
778 		if (pmu_access_el0_disabled(vcpu))
779 			return false;
780 
781 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
782 	} else {
783 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
784 	}
785 
786 	return true;
787 }
788 
789 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
790 			       const struct sys_reg_desc *r)
791 {
792 	u64 idx, reg;
793 
794 	if (pmu_access_el0_disabled(vcpu))
795 		return false;
796 
797 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
798 		/* PMXEVTYPER_EL0 */
799 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
800 		reg = PMEVTYPER0_EL0 + idx;
801 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
802 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
803 		if (idx == ARMV8_PMU_CYCLE_IDX)
804 			reg = PMCCFILTR_EL0;
805 		else
806 			/* PMEVTYPERn_EL0 */
807 			reg = PMEVTYPER0_EL0 + idx;
808 	} else {
809 		BUG();
810 	}
811 
812 	if (!pmu_counter_idx_valid(vcpu, idx))
813 		return false;
814 
815 	if (p->is_write) {
816 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
817 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
818 		kvm_vcpu_pmu_restore_guest(vcpu);
819 	} else {
820 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
821 	}
822 
823 	return true;
824 }
825 
826 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
827 			   const struct sys_reg_desc *r)
828 {
829 	u64 val, mask;
830 
831 	if (pmu_access_el0_disabled(vcpu))
832 		return false;
833 
834 	mask = kvm_pmu_valid_counter_mask(vcpu);
835 	if (p->is_write) {
836 		val = p->regval & mask;
837 		if (r->Op2 & 0x1) {
838 			/* accessing PMCNTENSET_EL0 */
839 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
840 			kvm_pmu_enable_counter_mask(vcpu, val);
841 			kvm_vcpu_pmu_restore_guest(vcpu);
842 		} else {
843 			/* accessing PMCNTENCLR_EL0 */
844 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
845 			kvm_pmu_disable_counter_mask(vcpu, val);
846 		}
847 	} else {
848 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
849 	}
850 
851 	return true;
852 }
853 
854 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
855 			   const struct sys_reg_desc *r)
856 {
857 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
858 
859 	if (check_pmu_access_disabled(vcpu, 0))
860 		return false;
861 
862 	if (p->is_write) {
863 		u64 val = p->regval & mask;
864 
865 		if (r->Op2 & 0x1)
866 			/* accessing PMINTENSET_EL1 */
867 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
868 		else
869 			/* accessing PMINTENCLR_EL1 */
870 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
871 	} else {
872 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
873 	}
874 
875 	return true;
876 }
877 
878 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
879 			 const struct sys_reg_desc *r)
880 {
881 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
882 
883 	if (pmu_access_el0_disabled(vcpu))
884 		return false;
885 
886 	if (p->is_write) {
887 		if (r->CRm & 0x2)
888 			/* accessing PMOVSSET_EL0 */
889 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
890 		else
891 			/* accessing PMOVSCLR_EL0 */
892 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
893 	} else {
894 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
895 	}
896 
897 	return true;
898 }
899 
900 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
901 			   const struct sys_reg_desc *r)
902 {
903 	u64 mask;
904 
905 	if (!p->is_write)
906 		return read_from_write_only(vcpu, p, r);
907 
908 	if (pmu_write_swinc_el0_disabled(vcpu))
909 		return false;
910 
911 	mask = kvm_pmu_valid_counter_mask(vcpu);
912 	kvm_pmu_software_increment(vcpu, p->regval & mask);
913 	return true;
914 }
915 
916 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
917 			     const struct sys_reg_desc *r)
918 {
919 	if (p->is_write) {
920 		if (!vcpu_mode_priv(vcpu)) {
921 			kvm_inject_undefined(vcpu);
922 			return false;
923 		}
924 
925 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
926 			       p->regval & ARMV8_PMU_USERENR_MASK;
927 	} else {
928 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
929 			    & ARMV8_PMU_USERENR_MASK;
930 	}
931 
932 	return true;
933 }
934 
935 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
936 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
937 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
938 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
939 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
940 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
941 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
942 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
943 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
944 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
945 
946 #define PMU_SYS_REG(r)						\
947 	SYS_DESC(r), .reset = reset_unknown, .visibility = pmu_visibility
948 
949 /* Macro to expand the PMEVCNTRn_EL0 register */
950 #define PMU_PMEVCNTR_EL0(n)						\
951 	{ PMU_SYS_REG(SYS_PMEVCNTRn_EL0(n)),				\
952 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
953 
954 /* Macro to expand the PMEVTYPERn_EL0 register */
955 #define PMU_PMEVTYPER_EL0(n)						\
956 	{ PMU_SYS_REG(SYS_PMEVTYPERn_EL0(n)),				\
957 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
958 
959 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
960 			 const struct sys_reg_desc *r)
961 {
962 	kvm_inject_undefined(vcpu);
963 
964 	return false;
965 }
966 
967 /* Macro to expand the AMU counter and type registers*/
968 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
969 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
970 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
971 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
972 
973 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
974 			const struct sys_reg_desc *rd)
975 {
976 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
977 }
978 
979 /*
980  * If we land here on a PtrAuth access, that is because we didn't
981  * fixup the access on exit by allowing the PtrAuth sysregs. The only
982  * way this happens is when the guest does not have PtrAuth support
983  * enabled.
984  */
985 #define __PTRAUTH_KEY(k)						\
986 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
987 	.visibility = ptrauth_visibility}
988 
989 #define PTRAUTH_KEY(k)							\
990 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
991 	__PTRAUTH_KEY(k ## KEYHI_EL1)
992 
993 static bool access_arch_timer(struct kvm_vcpu *vcpu,
994 			      struct sys_reg_params *p,
995 			      const struct sys_reg_desc *r)
996 {
997 	enum kvm_arch_timers tmr;
998 	enum kvm_arch_timer_regs treg;
999 	u64 reg = reg_to_encoding(r);
1000 
1001 	switch (reg) {
1002 	case SYS_CNTP_TVAL_EL0:
1003 	case SYS_AARCH32_CNTP_TVAL:
1004 		tmr = TIMER_PTIMER;
1005 		treg = TIMER_REG_TVAL;
1006 		break;
1007 	case SYS_CNTP_CTL_EL0:
1008 	case SYS_AARCH32_CNTP_CTL:
1009 		tmr = TIMER_PTIMER;
1010 		treg = TIMER_REG_CTL;
1011 		break;
1012 	case SYS_CNTP_CVAL_EL0:
1013 	case SYS_AARCH32_CNTP_CVAL:
1014 		tmr = TIMER_PTIMER;
1015 		treg = TIMER_REG_CVAL;
1016 		break;
1017 	default:
1018 		BUG();
1019 	}
1020 
1021 	if (p->is_write)
1022 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1023 	else
1024 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1025 
1026 	return true;
1027 }
1028 
1029 #define FEATURE(x)	(GENMASK_ULL(x##_SHIFT + 3, x##_SHIFT))
1030 
1031 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1032 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1033 		struct sys_reg_desc const *r, bool raz)
1034 {
1035 	u32 id = reg_to_encoding(r);
1036 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1037 
1038 	switch (id) {
1039 	case SYS_ID_AA64PFR0_EL1:
1040 		if (!vcpu_has_sve(vcpu))
1041 			val &= ~FEATURE(ID_AA64PFR0_SVE);
1042 		val &= ~FEATURE(ID_AA64PFR0_AMU);
1043 		val &= ~FEATURE(ID_AA64PFR0_CSV2);
1044 		val |= FIELD_PREP(FEATURE(ID_AA64PFR0_CSV2), (u64)vcpu->kvm->arch.pfr0_csv2);
1045 		val &= ~FEATURE(ID_AA64PFR0_CSV3);
1046 		val |= FIELD_PREP(FEATURE(ID_AA64PFR0_CSV3), (u64)vcpu->kvm->arch.pfr0_csv3);
1047 		break;
1048 	case SYS_ID_AA64PFR1_EL1:
1049 		val &= ~FEATURE(ID_AA64PFR1_MTE);
1050 		break;
1051 	case SYS_ID_AA64ISAR1_EL1:
1052 		if (!vcpu_has_ptrauth(vcpu))
1053 			val &= ~(FEATURE(ID_AA64ISAR1_APA) |
1054 				 FEATURE(ID_AA64ISAR1_API) |
1055 				 FEATURE(ID_AA64ISAR1_GPA) |
1056 				 FEATURE(ID_AA64ISAR1_GPI));
1057 		break;
1058 	case SYS_ID_AA64DFR0_EL1:
1059 		/* Limit debug to ARMv8.0 */
1060 		val &= ~FEATURE(ID_AA64DFR0_DEBUGVER);
1061 		val |= FIELD_PREP(FEATURE(ID_AA64DFR0_DEBUGVER), 6);
1062 		/* Limit guests to PMUv3 for ARMv8.4 */
1063 		val = cpuid_feature_cap_perfmon_field(val,
1064 						      ID_AA64DFR0_PMUVER_SHIFT,
1065 						      kvm_vcpu_has_pmu(vcpu) ? ID_AA64DFR0_PMUVER_8_4 : 0);
1066 		/* Hide SPE from guests */
1067 		val &= ~FEATURE(ID_AA64DFR0_PMSVER);
1068 		break;
1069 	case SYS_ID_DFR0_EL1:
1070 		/* Limit guests to PMUv3 for ARMv8.4 */
1071 		val = cpuid_feature_cap_perfmon_field(val,
1072 						      ID_DFR0_PERFMON_SHIFT,
1073 						      kvm_vcpu_has_pmu(vcpu) ? ID_DFR0_PERFMON_8_4 : 0);
1074 		break;
1075 	}
1076 
1077 	return val;
1078 }
1079 
1080 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1081 				  const struct sys_reg_desc *r)
1082 {
1083 	u32 id = reg_to_encoding(r);
1084 
1085 	switch (id) {
1086 	case SYS_ID_AA64ZFR0_EL1:
1087 		if (!vcpu_has_sve(vcpu))
1088 			return REG_RAZ;
1089 		break;
1090 	}
1091 
1092 	return 0;
1093 }
1094 
1095 /* cpufeature ID register access trap handlers */
1096 
1097 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1098 			    struct sys_reg_params *p,
1099 			    const struct sys_reg_desc *r,
1100 			    bool raz)
1101 {
1102 	if (p->is_write)
1103 		return write_to_read_only(vcpu, p, r);
1104 
1105 	p->regval = read_id_reg(vcpu, r, raz);
1106 	return true;
1107 }
1108 
1109 static bool access_id_reg(struct kvm_vcpu *vcpu,
1110 			  struct sys_reg_params *p,
1111 			  const struct sys_reg_desc *r)
1112 {
1113 	bool raz = sysreg_visible_as_raz(vcpu, r);
1114 
1115 	return __access_id_reg(vcpu, p, r, raz);
1116 }
1117 
1118 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1119 			      struct sys_reg_params *p,
1120 			      const struct sys_reg_desc *r)
1121 {
1122 	return __access_id_reg(vcpu, p, r, true);
1123 }
1124 
1125 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1126 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1127 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1128 
1129 /* Visibility overrides for SVE-specific control registers */
1130 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1131 				   const struct sys_reg_desc *rd)
1132 {
1133 	if (vcpu_has_sve(vcpu))
1134 		return 0;
1135 
1136 	return REG_HIDDEN;
1137 }
1138 
1139 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1140 			       const struct sys_reg_desc *rd,
1141 			       const struct kvm_one_reg *reg, void __user *uaddr)
1142 {
1143 	const u64 id = sys_reg_to_index(rd);
1144 	u8 csv2, csv3;
1145 	int err;
1146 	u64 val;
1147 
1148 	err = reg_from_user(&val, uaddr, id);
1149 	if (err)
1150 		return err;
1151 
1152 	/*
1153 	 * Allow AA64PFR0_EL1.CSV2 to be set from userspace as long as
1154 	 * it doesn't promise more than what is actually provided (the
1155 	 * guest could otherwise be covered in ectoplasmic residue).
1156 	 */
1157 	csv2 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV2_SHIFT);
1158 	if (csv2 > 1 ||
1159 	    (csv2 && arm64_get_spectre_v2_state() != SPECTRE_UNAFFECTED))
1160 		return -EINVAL;
1161 
1162 	/* Same thing for CSV3 */
1163 	csv3 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV3_SHIFT);
1164 	if (csv3 > 1 ||
1165 	    (csv3 && arm64_get_meltdown_state() != SPECTRE_UNAFFECTED))
1166 		return -EINVAL;
1167 
1168 	/* We can only differ with CSV[23], and anything else is an error */
1169 	val ^= read_id_reg(vcpu, rd, false);
1170 	val &= ~((0xFUL << ID_AA64PFR0_CSV2_SHIFT) |
1171 		 (0xFUL << ID_AA64PFR0_CSV3_SHIFT));
1172 	if (val)
1173 		return -EINVAL;
1174 
1175 	vcpu->kvm->arch.pfr0_csv2 = csv2;
1176 	vcpu->kvm->arch.pfr0_csv3 = csv3 ;
1177 
1178 	return 0;
1179 }
1180 
1181 /*
1182  * cpufeature ID register user accessors
1183  *
1184  * For now, these registers are immutable for userspace, so no values
1185  * are stored, and for set_id_reg() we don't allow the effective value
1186  * to be changed.
1187  */
1188 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1189 			const struct sys_reg_desc *rd, void __user *uaddr,
1190 			bool raz)
1191 {
1192 	const u64 id = sys_reg_to_index(rd);
1193 	const u64 val = read_id_reg(vcpu, rd, raz);
1194 
1195 	return reg_to_user(uaddr, &val, id);
1196 }
1197 
1198 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1199 			const struct sys_reg_desc *rd, void __user *uaddr,
1200 			bool raz)
1201 {
1202 	const u64 id = sys_reg_to_index(rd);
1203 	int err;
1204 	u64 val;
1205 
1206 	err = reg_from_user(&val, uaddr, id);
1207 	if (err)
1208 		return err;
1209 
1210 	/* This is what we mean by invariant: you can't change it. */
1211 	if (val != read_id_reg(vcpu, rd, raz))
1212 		return -EINVAL;
1213 
1214 	return 0;
1215 }
1216 
1217 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1218 		      const struct kvm_one_reg *reg, void __user *uaddr)
1219 {
1220 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1221 
1222 	return __get_id_reg(vcpu, rd, uaddr, raz);
1223 }
1224 
1225 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1226 		      const struct kvm_one_reg *reg, void __user *uaddr)
1227 {
1228 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1229 
1230 	return __set_id_reg(vcpu, rd, uaddr, raz);
1231 }
1232 
1233 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1234 			  const struct kvm_one_reg *reg, void __user *uaddr)
1235 {
1236 	return __get_id_reg(vcpu, rd, uaddr, true);
1237 }
1238 
1239 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1240 			  const struct kvm_one_reg *reg, void __user *uaddr)
1241 {
1242 	return __set_id_reg(vcpu, rd, uaddr, true);
1243 }
1244 
1245 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1246 		       const struct sys_reg_desc *r)
1247 {
1248 	if (p->is_write)
1249 		return write_to_read_only(vcpu, p, r);
1250 
1251 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1252 	return true;
1253 }
1254 
1255 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1256 			 const struct sys_reg_desc *r)
1257 {
1258 	if (p->is_write)
1259 		return write_to_read_only(vcpu, p, r);
1260 
1261 	p->regval = read_sysreg(clidr_el1);
1262 	return true;
1263 }
1264 
1265 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1266 			  const struct sys_reg_desc *r)
1267 {
1268 	int reg = r->reg;
1269 
1270 	if (p->is_write)
1271 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1272 	else
1273 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1274 	return true;
1275 }
1276 
1277 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1278 			  const struct sys_reg_desc *r)
1279 {
1280 	u32 csselr;
1281 
1282 	if (p->is_write)
1283 		return write_to_read_only(vcpu, p, r);
1284 
1285 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1286 	p->regval = get_ccsidr(csselr);
1287 
1288 	/*
1289 	 * Guests should not be doing cache operations by set/way at all, and
1290 	 * for this reason, we trap them and attempt to infer the intent, so
1291 	 * that we can flush the entire guest's address space at the appropriate
1292 	 * time.
1293 	 * To prevent this trapping from causing performance problems, let's
1294 	 * expose the geometry of all data and unified caches (which are
1295 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1296 	 * [If guests should attempt to infer aliasing properties from the
1297 	 * geometry (which is not permitted by the architecture), they would
1298 	 * only do so for virtually indexed caches.]
1299 	 */
1300 	if (!(csselr & 1)) // data or unified cache
1301 		p->regval &= ~GENMASK(27, 3);
1302 	return true;
1303 }
1304 
1305 /* sys_reg_desc initialiser for known cpufeature ID registers */
1306 #define ID_SANITISED(name) {			\
1307 	SYS_DESC(SYS_##name),			\
1308 	.access	= access_id_reg,		\
1309 	.get_user = get_id_reg,			\
1310 	.set_user = set_id_reg,			\
1311 	.visibility = id_visibility,		\
1312 }
1313 
1314 /*
1315  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1316  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1317  * (1 <= crm < 8, 0 <= Op2 < 8).
1318  */
1319 #define ID_UNALLOCATED(crm, op2) {			\
1320 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1321 	.access = access_raz_id_reg,			\
1322 	.get_user = get_raz_id_reg,			\
1323 	.set_user = set_raz_id_reg,			\
1324 }
1325 
1326 /*
1327  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1328  * For now, these are exposed just like unallocated ID regs: they appear
1329  * RAZ for the guest.
1330  */
1331 #define ID_HIDDEN(name) {			\
1332 	SYS_DESC(SYS_##name),			\
1333 	.access = access_raz_id_reg,		\
1334 	.get_user = get_raz_id_reg,		\
1335 	.set_user = set_raz_id_reg,		\
1336 }
1337 
1338 /*
1339  * Architected system registers.
1340  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1341  *
1342  * Debug handling: We do trap most, if not all debug related system
1343  * registers. The implementation is good enough to ensure that a guest
1344  * can use these with minimal performance degradation. The drawback is
1345  * that we don't implement any of the external debug, none of the
1346  * OSlock protocol. This should be revisited if we ever encounter a
1347  * more demanding guest...
1348  */
1349 static const struct sys_reg_desc sys_reg_descs[] = {
1350 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1351 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1352 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1353 
1354 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1355 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1356 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1357 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1358 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1359 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1360 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1361 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1362 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1363 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1364 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1365 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1366 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1367 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1368 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1369 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1370 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1371 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1372 
1373 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1374 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1375 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1376 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1377 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1378 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1379 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1380 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1381 
1382 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1383 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1384 	// DBGDTR[TR]X_EL0 share the same encoding
1385 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1386 
1387 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1388 
1389 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1390 
1391 	/*
1392 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1393 	 * entries in arm64_ftr_regs[].
1394 	 */
1395 
1396 	/* AArch64 mappings of the AArch32 ID registers */
1397 	/* CRm=1 */
1398 	ID_SANITISED(ID_PFR0_EL1),
1399 	ID_SANITISED(ID_PFR1_EL1),
1400 	ID_SANITISED(ID_DFR0_EL1),
1401 	ID_HIDDEN(ID_AFR0_EL1),
1402 	ID_SANITISED(ID_MMFR0_EL1),
1403 	ID_SANITISED(ID_MMFR1_EL1),
1404 	ID_SANITISED(ID_MMFR2_EL1),
1405 	ID_SANITISED(ID_MMFR3_EL1),
1406 
1407 	/* CRm=2 */
1408 	ID_SANITISED(ID_ISAR0_EL1),
1409 	ID_SANITISED(ID_ISAR1_EL1),
1410 	ID_SANITISED(ID_ISAR2_EL1),
1411 	ID_SANITISED(ID_ISAR3_EL1),
1412 	ID_SANITISED(ID_ISAR4_EL1),
1413 	ID_SANITISED(ID_ISAR5_EL1),
1414 	ID_SANITISED(ID_MMFR4_EL1),
1415 	ID_SANITISED(ID_ISAR6_EL1),
1416 
1417 	/* CRm=3 */
1418 	ID_SANITISED(MVFR0_EL1),
1419 	ID_SANITISED(MVFR1_EL1),
1420 	ID_SANITISED(MVFR2_EL1),
1421 	ID_UNALLOCATED(3,3),
1422 	ID_SANITISED(ID_PFR2_EL1),
1423 	ID_HIDDEN(ID_DFR1_EL1),
1424 	ID_SANITISED(ID_MMFR5_EL1),
1425 	ID_UNALLOCATED(3,7),
1426 
1427 	/* AArch64 ID registers */
1428 	/* CRm=4 */
1429 	{ SYS_DESC(SYS_ID_AA64PFR0_EL1), .access = access_id_reg,
1430 	  .get_user = get_id_reg, .set_user = set_id_aa64pfr0_el1, },
1431 	ID_SANITISED(ID_AA64PFR1_EL1),
1432 	ID_UNALLOCATED(4,2),
1433 	ID_UNALLOCATED(4,3),
1434 	ID_SANITISED(ID_AA64ZFR0_EL1),
1435 	ID_UNALLOCATED(4,5),
1436 	ID_UNALLOCATED(4,6),
1437 	ID_UNALLOCATED(4,7),
1438 
1439 	/* CRm=5 */
1440 	ID_SANITISED(ID_AA64DFR0_EL1),
1441 	ID_SANITISED(ID_AA64DFR1_EL1),
1442 	ID_UNALLOCATED(5,2),
1443 	ID_UNALLOCATED(5,3),
1444 	ID_HIDDEN(ID_AA64AFR0_EL1),
1445 	ID_HIDDEN(ID_AA64AFR1_EL1),
1446 	ID_UNALLOCATED(5,6),
1447 	ID_UNALLOCATED(5,7),
1448 
1449 	/* CRm=6 */
1450 	ID_SANITISED(ID_AA64ISAR0_EL1),
1451 	ID_SANITISED(ID_AA64ISAR1_EL1),
1452 	ID_UNALLOCATED(6,2),
1453 	ID_UNALLOCATED(6,3),
1454 	ID_UNALLOCATED(6,4),
1455 	ID_UNALLOCATED(6,5),
1456 	ID_UNALLOCATED(6,6),
1457 	ID_UNALLOCATED(6,7),
1458 
1459 	/* CRm=7 */
1460 	ID_SANITISED(ID_AA64MMFR0_EL1),
1461 	ID_SANITISED(ID_AA64MMFR1_EL1),
1462 	ID_SANITISED(ID_AA64MMFR2_EL1),
1463 	ID_UNALLOCATED(7,3),
1464 	ID_UNALLOCATED(7,4),
1465 	ID_UNALLOCATED(7,5),
1466 	ID_UNALLOCATED(7,6),
1467 	ID_UNALLOCATED(7,7),
1468 
1469 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1470 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1471 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1472 
1473 	{ SYS_DESC(SYS_RGSR_EL1), undef_access },
1474 	{ SYS_DESC(SYS_GCR_EL1), undef_access },
1475 
1476 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1477 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
1478 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1479 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1480 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1481 
1482 	PTRAUTH_KEY(APIA),
1483 	PTRAUTH_KEY(APIB),
1484 	PTRAUTH_KEY(APDA),
1485 	PTRAUTH_KEY(APDB),
1486 	PTRAUTH_KEY(APGA),
1487 
1488 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1489 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1490 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1491 
1492 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1493 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1494 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1495 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1496 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1497 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1498 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1499 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1500 
1501 	{ SYS_DESC(SYS_TFSR_EL1), undef_access },
1502 	{ SYS_DESC(SYS_TFSRE0_EL1), undef_access },
1503 
1504 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1505 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1506 
1507 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
1508 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
1509 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
1510 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
1511 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
1512 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
1513 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
1514 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
1515 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
1516 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
1517 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
1518 	/* PMBIDR_EL1 is not trapped */
1519 
1520 	{ PMU_SYS_REG(SYS_PMINTENSET_EL1),
1521 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1522 	{ PMU_SYS_REG(SYS_PMINTENCLR_EL1),
1523 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1524 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
1525 
1526 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1527 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1528 
1529 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1530 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1531 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1532 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1533 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1534 
1535 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1536 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1537 
1538 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1539 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1540 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1541 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1542 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1543 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1544 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1545 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1546 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1547 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1548 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1549 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1550 
1551 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1552 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1553 
1554 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
1555 
1556 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1557 
1558 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1559 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1560 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1561 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1562 
1563 	{ PMU_SYS_REG(SYS_PMCR_EL0), .access = access_pmcr,
1564 	  .reset = reset_pmcr, .reg = PMCR_EL0 },
1565 	{ PMU_SYS_REG(SYS_PMCNTENSET_EL0),
1566 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1567 	{ PMU_SYS_REG(SYS_PMCNTENCLR_EL0),
1568 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1569 	{ PMU_SYS_REG(SYS_PMOVSCLR_EL0),
1570 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1571 	{ PMU_SYS_REG(SYS_PMSWINC_EL0),
1572 	  .access = access_pmswinc, .reg = PMSWINC_EL0 },
1573 	{ PMU_SYS_REG(SYS_PMSELR_EL0),
1574 	  .access = access_pmselr, .reg = PMSELR_EL0 },
1575 	{ PMU_SYS_REG(SYS_PMCEID0_EL0),
1576 	  .access = access_pmceid, .reset = NULL },
1577 	{ PMU_SYS_REG(SYS_PMCEID1_EL0),
1578 	  .access = access_pmceid, .reset = NULL },
1579 	{ PMU_SYS_REG(SYS_PMCCNTR_EL0),
1580 	  .access = access_pmu_evcntr, .reg = PMCCNTR_EL0 },
1581 	{ PMU_SYS_REG(SYS_PMXEVTYPER_EL0),
1582 	  .access = access_pmu_evtyper, .reset = NULL },
1583 	{ PMU_SYS_REG(SYS_PMXEVCNTR_EL0),
1584 	  .access = access_pmu_evcntr, .reset = NULL },
1585 	/*
1586 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1587 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1588 	 */
1589 	{ PMU_SYS_REG(SYS_PMUSERENR_EL0), .access = access_pmuserenr,
1590 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
1591 	{ PMU_SYS_REG(SYS_PMOVSSET_EL0),
1592 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1593 
1594 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1595 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1596 
1597 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
1598 
1599 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
1600 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
1601 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
1602 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
1603 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
1604 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
1605 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
1606 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
1607 	AMU_AMEVCNTR0_EL0(0),
1608 	AMU_AMEVCNTR0_EL0(1),
1609 	AMU_AMEVCNTR0_EL0(2),
1610 	AMU_AMEVCNTR0_EL0(3),
1611 	AMU_AMEVCNTR0_EL0(4),
1612 	AMU_AMEVCNTR0_EL0(5),
1613 	AMU_AMEVCNTR0_EL0(6),
1614 	AMU_AMEVCNTR0_EL0(7),
1615 	AMU_AMEVCNTR0_EL0(8),
1616 	AMU_AMEVCNTR0_EL0(9),
1617 	AMU_AMEVCNTR0_EL0(10),
1618 	AMU_AMEVCNTR0_EL0(11),
1619 	AMU_AMEVCNTR0_EL0(12),
1620 	AMU_AMEVCNTR0_EL0(13),
1621 	AMU_AMEVCNTR0_EL0(14),
1622 	AMU_AMEVCNTR0_EL0(15),
1623 	AMU_AMEVTYPER0_EL0(0),
1624 	AMU_AMEVTYPER0_EL0(1),
1625 	AMU_AMEVTYPER0_EL0(2),
1626 	AMU_AMEVTYPER0_EL0(3),
1627 	AMU_AMEVTYPER0_EL0(4),
1628 	AMU_AMEVTYPER0_EL0(5),
1629 	AMU_AMEVTYPER0_EL0(6),
1630 	AMU_AMEVTYPER0_EL0(7),
1631 	AMU_AMEVTYPER0_EL0(8),
1632 	AMU_AMEVTYPER0_EL0(9),
1633 	AMU_AMEVTYPER0_EL0(10),
1634 	AMU_AMEVTYPER0_EL0(11),
1635 	AMU_AMEVTYPER0_EL0(12),
1636 	AMU_AMEVTYPER0_EL0(13),
1637 	AMU_AMEVTYPER0_EL0(14),
1638 	AMU_AMEVTYPER0_EL0(15),
1639 	AMU_AMEVCNTR1_EL0(0),
1640 	AMU_AMEVCNTR1_EL0(1),
1641 	AMU_AMEVCNTR1_EL0(2),
1642 	AMU_AMEVCNTR1_EL0(3),
1643 	AMU_AMEVCNTR1_EL0(4),
1644 	AMU_AMEVCNTR1_EL0(5),
1645 	AMU_AMEVCNTR1_EL0(6),
1646 	AMU_AMEVCNTR1_EL0(7),
1647 	AMU_AMEVCNTR1_EL0(8),
1648 	AMU_AMEVCNTR1_EL0(9),
1649 	AMU_AMEVCNTR1_EL0(10),
1650 	AMU_AMEVCNTR1_EL0(11),
1651 	AMU_AMEVCNTR1_EL0(12),
1652 	AMU_AMEVCNTR1_EL0(13),
1653 	AMU_AMEVCNTR1_EL0(14),
1654 	AMU_AMEVCNTR1_EL0(15),
1655 	AMU_AMEVTYPER1_EL0(0),
1656 	AMU_AMEVTYPER1_EL0(1),
1657 	AMU_AMEVTYPER1_EL0(2),
1658 	AMU_AMEVTYPER1_EL0(3),
1659 	AMU_AMEVTYPER1_EL0(4),
1660 	AMU_AMEVTYPER1_EL0(5),
1661 	AMU_AMEVTYPER1_EL0(6),
1662 	AMU_AMEVTYPER1_EL0(7),
1663 	AMU_AMEVTYPER1_EL0(8),
1664 	AMU_AMEVTYPER1_EL0(9),
1665 	AMU_AMEVTYPER1_EL0(10),
1666 	AMU_AMEVTYPER1_EL0(11),
1667 	AMU_AMEVTYPER1_EL0(12),
1668 	AMU_AMEVTYPER1_EL0(13),
1669 	AMU_AMEVTYPER1_EL0(14),
1670 	AMU_AMEVTYPER1_EL0(15),
1671 
1672 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1673 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1674 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1675 
1676 	/* PMEVCNTRn_EL0 */
1677 	PMU_PMEVCNTR_EL0(0),
1678 	PMU_PMEVCNTR_EL0(1),
1679 	PMU_PMEVCNTR_EL0(2),
1680 	PMU_PMEVCNTR_EL0(3),
1681 	PMU_PMEVCNTR_EL0(4),
1682 	PMU_PMEVCNTR_EL0(5),
1683 	PMU_PMEVCNTR_EL0(6),
1684 	PMU_PMEVCNTR_EL0(7),
1685 	PMU_PMEVCNTR_EL0(8),
1686 	PMU_PMEVCNTR_EL0(9),
1687 	PMU_PMEVCNTR_EL0(10),
1688 	PMU_PMEVCNTR_EL0(11),
1689 	PMU_PMEVCNTR_EL0(12),
1690 	PMU_PMEVCNTR_EL0(13),
1691 	PMU_PMEVCNTR_EL0(14),
1692 	PMU_PMEVCNTR_EL0(15),
1693 	PMU_PMEVCNTR_EL0(16),
1694 	PMU_PMEVCNTR_EL0(17),
1695 	PMU_PMEVCNTR_EL0(18),
1696 	PMU_PMEVCNTR_EL0(19),
1697 	PMU_PMEVCNTR_EL0(20),
1698 	PMU_PMEVCNTR_EL0(21),
1699 	PMU_PMEVCNTR_EL0(22),
1700 	PMU_PMEVCNTR_EL0(23),
1701 	PMU_PMEVCNTR_EL0(24),
1702 	PMU_PMEVCNTR_EL0(25),
1703 	PMU_PMEVCNTR_EL0(26),
1704 	PMU_PMEVCNTR_EL0(27),
1705 	PMU_PMEVCNTR_EL0(28),
1706 	PMU_PMEVCNTR_EL0(29),
1707 	PMU_PMEVCNTR_EL0(30),
1708 	/* PMEVTYPERn_EL0 */
1709 	PMU_PMEVTYPER_EL0(0),
1710 	PMU_PMEVTYPER_EL0(1),
1711 	PMU_PMEVTYPER_EL0(2),
1712 	PMU_PMEVTYPER_EL0(3),
1713 	PMU_PMEVTYPER_EL0(4),
1714 	PMU_PMEVTYPER_EL0(5),
1715 	PMU_PMEVTYPER_EL0(6),
1716 	PMU_PMEVTYPER_EL0(7),
1717 	PMU_PMEVTYPER_EL0(8),
1718 	PMU_PMEVTYPER_EL0(9),
1719 	PMU_PMEVTYPER_EL0(10),
1720 	PMU_PMEVTYPER_EL0(11),
1721 	PMU_PMEVTYPER_EL0(12),
1722 	PMU_PMEVTYPER_EL0(13),
1723 	PMU_PMEVTYPER_EL0(14),
1724 	PMU_PMEVTYPER_EL0(15),
1725 	PMU_PMEVTYPER_EL0(16),
1726 	PMU_PMEVTYPER_EL0(17),
1727 	PMU_PMEVTYPER_EL0(18),
1728 	PMU_PMEVTYPER_EL0(19),
1729 	PMU_PMEVTYPER_EL0(20),
1730 	PMU_PMEVTYPER_EL0(21),
1731 	PMU_PMEVTYPER_EL0(22),
1732 	PMU_PMEVTYPER_EL0(23),
1733 	PMU_PMEVTYPER_EL0(24),
1734 	PMU_PMEVTYPER_EL0(25),
1735 	PMU_PMEVTYPER_EL0(26),
1736 	PMU_PMEVTYPER_EL0(27),
1737 	PMU_PMEVTYPER_EL0(28),
1738 	PMU_PMEVTYPER_EL0(29),
1739 	PMU_PMEVTYPER_EL0(30),
1740 	/*
1741 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1742 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1743 	 */
1744 	{ PMU_SYS_REG(SYS_PMCCFILTR_EL0), .access = access_pmu_evtyper,
1745 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
1746 
1747 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1748 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1749 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1750 };
1751 
1752 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
1753 			struct sys_reg_params *p,
1754 			const struct sys_reg_desc *r)
1755 {
1756 	if (p->is_write) {
1757 		return ignore_write(vcpu, p);
1758 	} else {
1759 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1760 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1761 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1762 
1763 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1764 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1765 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1766 			     | (6 << 16) | (1 << 15) | (el3 << 14) | (el3 << 12));
1767 		return true;
1768 	}
1769 }
1770 
1771 /*
1772  * AArch32 debug register mappings
1773  *
1774  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1775  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1776  *
1777  * None of the other registers share their location, so treat them as
1778  * if they were 64bit.
1779  */
1780 #define DBG_BCR_BVR_WCR_WVR(n)						      \
1781 	/* DBGBVRn */							      \
1782 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
1783 	/* DBGBCRn */							      \
1784 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
1785 	/* DBGWVRn */							      \
1786 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
1787 	/* DBGWCRn */							      \
1788 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1789 
1790 #define DBGBXVR(n)							      \
1791 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
1792 
1793 /*
1794  * Trapped cp14 registers. We generally ignore most of the external
1795  * debug, on the principle that they don't really make sense to a
1796  * guest. Revisit this one day, would this principle change.
1797  */
1798 static const struct sys_reg_desc cp14_regs[] = {
1799 	/* DBGDIDR */
1800 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
1801 	/* DBGDTRRXext */
1802 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1803 
1804 	DBG_BCR_BVR_WCR_WVR(0),
1805 	/* DBGDSCRint */
1806 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1807 	DBG_BCR_BVR_WCR_WVR(1),
1808 	/* DBGDCCINT */
1809 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
1810 	/* DBGDSCRext */
1811 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
1812 	DBG_BCR_BVR_WCR_WVR(2),
1813 	/* DBGDTR[RT]Xint */
1814 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1815 	/* DBGDTR[RT]Xext */
1816 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1817 	DBG_BCR_BVR_WCR_WVR(3),
1818 	DBG_BCR_BVR_WCR_WVR(4),
1819 	DBG_BCR_BVR_WCR_WVR(5),
1820 	/* DBGWFAR */
1821 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1822 	/* DBGOSECCR */
1823 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1824 	DBG_BCR_BVR_WCR_WVR(6),
1825 	/* DBGVCR */
1826 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
1827 	DBG_BCR_BVR_WCR_WVR(7),
1828 	DBG_BCR_BVR_WCR_WVR(8),
1829 	DBG_BCR_BVR_WCR_WVR(9),
1830 	DBG_BCR_BVR_WCR_WVR(10),
1831 	DBG_BCR_BVR_WCR_WVR(11),
1832 	DBG_BCR_BVR_WCR_WVR(12),
1833 	DBG_BCR_BVR_WCR_WVR(13),
1834 	DBG_BCR_BVR_WCR_WVR(14),
1835 	DBG_BCR_BVR_WCR_WVR(15),
1836 
1837 	/* DBGDRAR (32bit) */
1838 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1839 
1840 	DBGBXVR(0),
1841 	/* DBGOSLAR */
1842 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1843 	DBGBXVR(1),
1844 	/* DBGOSLSR */
1845 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1846 	DBGBXVR(2),
1847 	DBGBXVR(3),
1848 	/* DBGOSDLR */
1849 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1850 	DBGBXVR(4),
1851 	/* DBGPRCR */
1852 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1853 	DBGBXVR(5),
1854 	DBGBXVR(6),
1855 	DBGBXVR(7),
1856 	DBGBXVR(8),
1857 	DBGBXVR(9),
1858 	DBGBXVR(10),
1859 	DBGBXVR(11),
1860 	DBGBXVR(12),
1861 	DBGBXVR(13),
1862 	DBGBXVR(14),
1863 	DBGBXVR(15),
1864 
1865 	/* DBGDSAR (32bit) */
1866 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1867 
1868 	/* DBGDEVID2 */
1869 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1870 	/* DBGDEVID1 */
1871 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1872 	/* DBGDEVID */
1873 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1874 	/* DBGCLAIMSET */
1875 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1876 	/* DBGCLAIMCLR */
1877 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1878 	/* DBGAUTHSTATUS */
1879 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1880 };
1881 
1882 /* Trapped cp14 64bit registers */
1883 static const struct sys_reg_desc cp14_64_regs[] = {
1884 	/* DBGDRAR (64bit) */
1885 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1886 
1887 	/* DBGDSAR (64bit) */
1888 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1889 };
1890 
1891 /* Macro to expand the PMEVCNTRn register */
1892 #define PMU_PMEVCNTR(n)							\
1893 	/* PMEVCNTRn */							\
1894 	{ Op1(0), CRn(0b1110),						\
1895 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1896 	  access_pmu_evcntr }
1897 
1898 /* Macro to expand the PMEVTYPERn register */
1899 #define PMU_PMEVTYPER(n)						\
1900 	/* PMEVTYPERn */						\
1901 	{ Op1(0), CRn(0b1110),						\
1902 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1903 	  access_pmu_evtyper }
1904 
1905 /*
1906  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1907  * depending on the way they are accessed (as a 32bit or a 64bit
1908  * register).
1909  */
1910 static const struct sys_reg_desc cp15_regs[] = {
1911 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1912 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
1913 	/* ACTLR */
1914 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
1915 	/* ACTLR2 */
1916 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
1917 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
1918 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
1919 	/* TTBCR */
1920 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
1921 	/* TTBCR2 */
1922 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
1923 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
1924 	/* DFSR */
1925 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
1926 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
1927 	/* ADFSR */
1928 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
1929 	/* AIFSR */
1930 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
1931 	/* DFAR */
1932 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
1933 	/* IFAR */
1934 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
1935 
1936 	/*
1937 	 * DC{C,I,CI}SW operations:
1938 	 */
1939 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1940 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1941 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1942 
1943 	/* PMU */
1944 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1945 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1946 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1947 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1948 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1949 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1950 	{ AA32(LO), Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1951 	{ AA32(LO), Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1952 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1953 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1954 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1955 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1956 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1957 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1958 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1959 	{ AA32(HI), Op1( 0), CRn( 9), CRm(14), Op2( 4), access_pmceid },
1960 	{ AA32(HI), Op1( 0), CRn( 9), CRm(14), Op2( 5), access_pmceid },
1961 	/* PMMIR */
1962 	{ Op1( 0), CRn( 9), CRm(14), Op2( 6), trap_raz_wi },
1963 
1964 	/* PRRR/MAIR0 */
1965 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
1966 	/* NMRR/MAIR1 */
1967 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
1968 	/* AMAIR0 */
1969 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
1970 	/* AMAIR1 */
1971 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
1972 
1973 	/* ICC_SRE */
1974 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
1975 
1976 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
1977 
1978 	/* Arch Tmers */
1979 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
1980 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
1981 
1982 	/* PMEVCNTRn */
1983 	PMU_PMEVCNTR(0),
1984 	PMU_PMEVCNTR(1),
1985 	PMU_PMEVCNTR(2),
1986 	PMU_PMEVCNTR(3),
1987 	PMU_PMEVCNTR(4),
1988 	PMU_PMEVCNTR(5),
1989 	PMU_PMEVCNTR(6),
1990 	PMU_PMEVCNTR(7),
1991 	PMU_PMEVCNTR(8),
1992 	PMU_PMEVCNTR(9),
1993 	PMU_PMEVCNTR(10),
1994 	PMU_PMEVCNTR(11),
1995 	PMU_PMEVCNTR(12),
1996 	PMU_PMEVCNTR(13),
1997 	PMU_PMEVCNTR(14),
1998 	PMU_PMEVCNTR(15),
1999 	PMU_PMEVCNTR(16),
2000 	PMU_PMEVCNTR(17),
2001 	PMU_PMEVCNTR(18),
2002 	PMU_PMEVCNTR(19),
2003 	PMU_PMEVCNTR(20),
2004 	PMU_PMEVCNTR(21),
2005 	PMU_PMEVCNTR(22),
2006 	PMU_PMEVCNTR(23),
2007 	PMU_PMEVCNTR(24),
2008 	PMU_PMEVCNTR(25),
2009 	PMU_PMEVCNTR(26),
2010 	PMU_PMEVCNTR(27),
2011 	PMU_PMEVCNTR(28),
2012 	PMU_PMEVCNTR(29),
2013 	PMU_PMEVCNTR(30),
2014 	/* PMEVTYPERn */
2015 	PMU_PMEVTYPER(0),
2016 	PMU_PMEVTYPER(1),
2017 	PMU_PMEVTYPER(2),
2018 	PMU_PMEVTYPER(3),
2019 	PMU_PMEVTYPER(4),
2020 	PMU_PMEVTYPER(5),
2021 	PMU_PMEVTYPER(6),
2022 	PMU_PMEVTYPER(7),
2023 	PMU_PMEVTYPER(8),
2024 	PMU_PMEVTYPER(9),
2025 	PMU_PMEVTYPER(10),
2026 	PMU_PMEVTYPER(11),
2027 	PMU_PMEVTYPER(12),
2028 	PMU_PMEVTYPER(13),
2029 	PMU_PMEVTYPER(14),
2030 	PMU_PMEVTYPER(15),
2031 	PMU_PMEVTYPER(16),
2032 	PMU_PMEVTYPER(17),
2033 	PMU_PMEVTYPER(18),
2034 	PMU_PMEVTYPER(19),
2035 	PMU_PMEVTYPER(20),
2036 	PMU_PMEVTYPER(21),
2037 	PMU_PMEVTYPER(22),
2038 	PMU_PMEVTYPER(23),
2039 	PMU_PMEVTYPER(24),
2040 	PMU_PMEVTYPER(25),
2041 	PMU_PMEVTYPER(26),
2042 	PMU_PMEVTYPER(27),
2043 	PMU_PMEVTYPER(28),
2044 	PMU_PMEVTYPER(29),
2045 	PMU_PMEVTYPER(30),
2046 	/* PMCCFILTR */
2047 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2048 
2049 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2050 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2051 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
2052 };
2053 
2054 static const struct sys_reg_desc cp15_64_regs[] = {
2055 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
2056 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2057 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2058 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
2059 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2060 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2061 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2062 };
2063 
2064 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2065 			      bool is_32)
2066 {
2067 	unsigned int i;
2068 
2069 	for (i = 0; i < n; i++) {
2070 		if (!is_32 && table[i].reg && !table[i].reset) {
2071 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2072 				table, i);
2073 			return 1;
2074 		}
2075 
2076 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2077 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2078 			return 1;
2079 		}
2080 	}
2081 
2082 	return 0;
2083 }
2084 
2085 static int match_sys_reg(const void *key, const void *elt)
2086 {
2087 	const unsigned long pval = (unsigned long)key;
2088 	const struct sys_reg_desc *r = elt;
2089 
2090 	return pval - reg_to_encoding(r);
2091 }
2092 
2093 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
2094 					 const struct sys_reg_desc table[],
2095 					 unsigned int num)
2096 {
2097 	unsigned long pval = reg_to_encoding(params);
2098 
2099 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
2100 }
2101 
2102 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2103 {
2104 	kvm_inject_undefined(vcpu);
2105 	return 1;
2106 }
2107 
2108 static void perform_access(struct kvm_vcpu *vcpu,
2109 			   struct sys_reg_params *params,
2110 			   const struct sys_reg_desc *r)
2111 {
2112 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2113 
2114 	/* Check for regs disabled by runtime config */
2115 	if (sysreg_hidden(vcpu, r)) {
2116 		kvm_inject_undefined(vcpu);
2117 		return;
2118 	}
2119 
2120 	/*
2121 	 * Not having an accessor means that we have configured a trap
2122 	 * that we don't know how to handle. This certainly qualifies
2123 	 * as a gross bug that should be fixed right away.
2124 	 */
2125 	BUG_ON(!r->access);
2126 
2127 	/* Skip instruction if instructed so */
2128 	if (likely(r->access(vcpu, params, r)))
2129 		kvm_incr_pc(vcpu);
2130 }
2131 
2132 /*
2133  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2134  *                call the corresponding trap handler.
2135  *
2136  * @params: pointer to the descriptor of the access
2137  * @table: array of trap descriptors
2138  * @num: size of the trap descriptor array
2139  *
2140  * Return 0 if the access has been handled, and -1 if not.
2141  */
2142 static int emulate_cp(struct kvm_vcpu *vcpu,
2143 		      struct sys_reg_params *params,
2144 		      const struct sys_reg_desc *table,
2145 		      size_t num)
2146 {
2147 	const struct sys_reg_desc *r;
2148 
2149 	if (!table)
2150 		return -1;	/* Not handled */
2151 
2152 	r = find_reg(params, table, num);
2153 
2154 	if (r) {
2155 		perform_access(vcpu, params, r);
2156 		return 0;
2157 	}
2158 
2159 	/* Not handled */
2160 	return -1;
2161 }
2162 
2163 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2164 				struct sys_reg_params *params)
2165 {
2166 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2167 	int cp = -1;
2168 
2169 	switch (esr_ec) {
2170 	case ESR_ELx_EC_CP15_32:
2171 	case ESR_ELx_EC_CP15_64:
2172 		cp = 15;
2173 		break;
2174 	case ESR_ELx_EC_CP14_MR:
2175 	case ESR_ELx_EC_CP14_64:
2176 		cp = 14;
2177 		break;
2178 	default:
2179 		WARN_ON(1);
2180 	}
2181 
2182 	print_sys_reg_msg(params,
2183 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2184 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2185 	kvm_inject_undefined(vcpu);
2186 }
2187 
2188 /**
2189  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2190  * @vcpu: The VCPU pointer
2191  * @run:  The kvm_run struct
2192  */
2193 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2194 			    const struct sys_reg_desc *global,
2195 			    size_t nr_global)
2196 {
2197 	struct sys_reg_params params;
2198 	u32 esr = kvm_vcpu_get_esr(vcpu);
2199 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2200 	int Rt2 = (esr >> 10) & 0x1f;
2201 
2202 	params.CRm = (esr >> 1) & 0xf;
2203 	params.is_write = ((esr & 1) == 0);
2204 
2205 	params.Op0 = 0;
2206 	params.Op1 = (esr >> 16) & 0xf;
2207 	params.Op2 = 0;
2208 	params.CRn = 0;
2209 
2210 	/*
2211 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2212 	 * backends between AArch32 and AArch64, we get away with it.
2213 	 */
2214 	if (params.is_write) {
2215 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2216 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2217 	}
2218 
2219 	/*
2220 	 * If the table contains a handler, handle the
2221 	 * potential register operation in the case of a read and return
2222 	 * with success.
2223 	 */
2224 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2225 		/* Split up the value between registers for the read side */
2226 		if (!params.is_write) {
2227 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2228 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2229 		}
2230 
2231 		return 1;
2232 	}
2233 
2234 	unhandled_cp_access(vcpu, &params);
2235 	return 1;
2236 }
2237 
2238 /**
2239  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2240  * @vcpu: The VCPU pointer
2241  * @run:  The kvm_run struct
2242  */
2243 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2244 			    const struct sys_reg_desc *global,
2245 			    size_t nr_global)
2246 {
2247 	struct sys_reg_params params;
2248 	u32 esr = kvm_vcpu_get_esr(vcpu);
2249 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2250 
2251 	params.CRm = (esr >> 1) & 0xf;
2252 	params.regval = vcpu_get_reg(vcpu, Rt);
2253 	params.is_write = ((esr & 1) == 0);
2254 	params.CRn = (esr >> 10) & 0xf;
2255 	params.Op0 = 0;
2256 	params.Op1 = (esr >> 14) & 0x7;
2257 	params.Op2 = (esr >> 17) & 0x7;
2258 
2259 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2260 		if (!params.is_write)
2261 			vcpu_set_reg(vcpu, Rt, params.regval);
2262 		return 1;
2263 	}
2264 
2265 	unhandled_cp_access(vcpu, &params);
2266 	return 1;
2267 }
2268 
2269 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2270 {
2271 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2272 }
2273 
2274 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2275 {
2276 	return kvm_handle_cp_32(vcpu, cp15_regs, ARRAY_SIZE(cp15_regs));
2277 }
2278 
2279 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2280 {
2281 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2282 }
2283 
2284 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2285 {
2286 	return kvm_handle_cp_32(vcpu, cp14_regs, ARRAY_SIZE(cp14_regs));
2287 }
2288 
2289 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2290 {
2291 	// See ARM DDI 0487E.a, section D12.3.2
2292 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2293 }
2294 
2295 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2296 			   struct sys_reg_params *params)
2297 {
2298 	const struct sys_reg_desc *r;
2299 
2300 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2301 
2302 	if (likely(r)) {
2303 		perform_access(vcpu, params, r);
2304 	} else if (is_imp_def_sys_reg(params)) {
2305 		kvm_inject_undefined(vcpu);
2306 	} else {
2307 		print_sys_reg_msg(params,
2308 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2309 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2310 		kvm_inject_undefined(vcpu);
2311 	}
2312 	return 1;
2313 }
2314 
2315 /**
2316  * kvm_reset_sys_regs - sets system registers to reset value
2317  * @vcpu: The VCPU pointer
2318  *
2319  * This function finds the right table above and sets the registers on the
2320  * virtual CPU struct to their architecturally defined reset values.
2321  */
2322 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2323 {
2324 	unsigned long i;
2325 
2326 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2327 		if (sys_reg_descs[i].reset)
2328 			sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2329 }
2330 
2331 /**
2332  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2333  * @vcpu: The VCPU pointer
2334  */
2335 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2336 {
2337 	struct sys_reg_params params;
2338 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
2339 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2340 	int ret;
2341 
2342 	trace_kvm_handle_sys_reg(esr);
2343 
2344 	params.Op0 = (esr >> 20) & 3;
2345 	params.Op1 = (esr >> 14) & 0x7;
2346 	params.CRn = (esr >> 10) & 0xf;
2347 	params.CRm = (esr >> 1) & 0xf;
2348 	params.Op2 = (esr >> 17) & 0x7;
2349 	params.regval = vcpu_get_reg(vcpu, Rt);
2350 	params.is_write = !(esr & 1);
2351 
2352 	ret = emulate_sys_reg(vcpu, &params);
2353 
2354 	if (!params.is_write)
2355 		vcpu_set_reg(vcpu, Rt, params.regval);
2356 	return ret;
2357 }
2358 
2359 /******************************************************************************
2360  * Userspace API
2361  *****************************************************************************/
2362 
2363 static bool index_to_params(u64 id, struct sys_reg_params *params)
2364 {
2365 	switch (id & KVM_REG_SIZE_MASK) {
2366 	case KVM_REG_SIZE_U64:
2367 		/* Any unused index bits means it's not valid. */
2368 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2369 			      | KVM_REG_ARM_COPROC_MASK
2370 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2371 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2372 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2373 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2374 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2375 			return false;
2376 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2377 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2378 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2379 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2380 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2381 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2382 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2383 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2384 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2385 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2386 		return true;
2387 	default:
2388 		return false;
2389 	}
2390 }
2391 
2392 const struct sys_reg_desc *find_reg_by_id(u64 id,
2393 					  struct sys_reg_params *params,
2394 					  const struct sys_reg_desc table[],
2395 					  unsigned int num)
2396 {
2397 	if (!index_to_params(id, params))
2398 		return NULL;
2399 
2400 	return find_reg(params, table, num);
2401 }
2402 
2403 /* Decode an index value, and find the sys_reg_desc entry. */
2404 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2405 						    u64 id)
2406 {
2407 	const struct sys_reg_desc *r;
2408 	struct sys_reg_params params;
2409 
2410 	/* We only do sys_reg for now. */
2411 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2412 		return NULL;
2413 
2414 	if (!index_to_params(id, &params))
2415 		return NULL;
2416 
2417 	r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2418 
2419 	/* Not saved in the sys_reg array and not otherwise accessible? */
2420 	if (r && !(r->reg || r->get_user))
2421 		r = NULL;
2422 
2423 	return r;
2424 }
2425 
2426 /*
2427  * These are the invariant sys_reg registers: we let the guest see the
2428  * host versions of these, so they're part of the guest state.
2429  *
2430  * A future CPU may provide a mechanism to present different values to
2431  * the guest, or a future kvm may trap them.
2432  */
2433 
2434 #define FUNCTION_INVARIANT(reg)						\
2435 	static void get_##reg(struct kvm_vcpu *v,			\
2436 			      const struct sys_reg_desc *r)		\
2437 	{								\
2438 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2439 	}
2440 
2441 FUNCTION_INVARIANT(midr_el1)
2442 FUNCTION_INVARIANT(revidr_el1)
2443 FUNCTION_INVARIANT(clidr_el1)
2444 FUNCTION_INVARIANT(aidr_el1)
2445 
2446 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2447 {
2448 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2449 }
2450 
2451 /* ->val is filled in by kvm_sys_reg_table_init() */
2452 static struct sys_reg_desc invariant_sys_regs[] = {
2453 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2454 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2455 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2456 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2457 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2458 };
2459 
2460 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2461 {
2462 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2463 		return -EFAULT;
2464 	return 0;
2465 }
2466 
2467 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2468 {
2469 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2470 		return -EFAULT;
2471 	return 0;
2472 }
2473 
2474 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2475 {
2476 	struct sys_reg_params params;
2477 	const struct sys_reg_desc *r;
2478 
2479 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2480 			   ARRAY_SIZE(invariant_sys_regs));
2481 	if (!r)
2482 		return -ENOENT;
2483 
2484 	return reg_to_user(uaddr, &r->val, id);
2485 }
2486 
2487 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2488 {
2489 	struct sys_reg_params params;
2490 	const struct sys_reg_desc *r;
2491 	int err;
2492 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2493 
2494 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2495 			   ARRAY_SIZE(invariant_sys_regs));
2496 	if (!r)
2497 		return -ENOENT;
2498 
2499 	err = reg_from_user(&val, uaddr, id);
2500 	if (err)
2501 		return err;
2502 
2503 	/* This is what we mean by invariant: you can't change it. */
2504 	if (r->val != val)
2505 		return -EINVAL;
2506 
2507 	return 0;
2508 }
2509 
2510 static bool is_valid_cache(u32 val)
2511 {
2512 	u32 level, ctype;
2513 
2514 	if (val >= CSSELR_MAX)
2515 		return false;
2516 
2517 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2518 	level = (val >> 1);
2519 	ctype = (cache_levels >> (level * 3)) & 7;
2520 
2521 	switch (ctype) {
2522 	case 0: /* No cache */
2523 		return false;
2524 	case 1: /* Instruction cache only */
2525 		return (val & 1);
2526 	case 2: /* Data cache only */
2527 	case 4: /* Unified cache */
2528 		return !(val & 1);
2529 	case 3: /* Separate instruction and data caches */
2530 		return true;
2531 	default: /* Reserved: we can't know instruction or data. */
2532 		return false;
2533 	}
2534 }
2535 
2536 static int demux_c15_get(u64 id, void __user *uaddr)
2537 {
2538 	u32 val;
2539 	u32 __user *uval = uaddr;
2540 
2541 	/* Fail if we have unknown bits set. */
2542 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2543 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2544 		return -ENOENT;
2545 
2546 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2547 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2548 		if (KVM_REG_SIZE(id) != 4)
2549 			return -ENOENT;
2550 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2551 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2552 		if (!is_valid_cache(val))
2553 			return -ENOENT;
2554 
2555 		return put_user(get_ccsidr(val), uval);
2556 	default:
2557 		return -ENOENT;
2558 	}
2559 }
2560 
2561 static int demux_c15_set(u64 id, void __user *uaddr)
2562 {
2563 	u32 val, newval;
2564 	u32 __user *uval = uaddr;
2565 
2566 	/* Fail if we have unknown bits set. */
2567 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2568 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2569 		return -ENOENT;
2570 
2571 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2572 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2573 		if (KVM_REG_SIZE(id) != 4)
2574 			return -ENOENT;
2575 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2576 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2577 		if (!is_valid_cache(val))
2578 			return -ENOENT;
2579 
2580 		if (get_user(newval, uval))
2581 			return -EFAULT;
2582 
2583 		/* This is also invariant: you can't change it. */
2584 		if (newval != get_ccsidr(val))
2585 			return -EINVAL;
2586 		return 0;
2587 	default:
2588 		return -ENOENT;
2589 	}
2590 }
2591 
2592 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2593 {
2594 	const struct sys_reg_desc *r;
2595 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2596 
2597 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2598 		return demux_c15_get(reg->id, uaddr);
2599 
2600 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2601 		return -ENOENT;
2602 
2603 	r = index_to_sys_reg_desc(vcpu, reg->id);
2604 	if (!r)
2605 		return get_invariant_sys_reg(reg->id, uaddr);
2606 
2607 	/* Check for regs disabled by runtime config */
2608 	if (sysreg_hidden(vcpu, r))
2609 		return -ENOENT;
2610 
2611 	if (r->get_user)
2612 		return (r->get_user)(vcpu, r, reg, uaddr);
2613 
2614 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2615 }
2616 
2617 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2618 {
2619 	const struct sys_reg_desc *r;
2620 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2621 
2622 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2623 		return demux_c15_set(reg->id, uaddr);
2624 
2625 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2626 		return -ENOENT;
2627 
2628 	r = index_to_sys_reg_desc(vcpu, reg->id);
2629 	if (!r)
2630 		return set_invariant_sys_reg(reg->id, uaddr);
2631 
2632 	/* Check for regs disabled by runtime config */
2633 	if (sysreg_hidden(vcpu, r))
2634 		return -ENOENT;
2635 
2636 	if (r->set_user)
2637 		return (r->set_user)(vcpu, r, reg, uaddr);
2638 
2639 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2640 }
2641 
2642 static unsigned int num_demux_regs(void)
2643 {
2644 	unsigned int i, count = 0;
2645 
2646 	for (i = 0; i < CSSELR_MAX; i++)
2647 		if (is_valid_cache(i))
2648 			count++;
2649 
2650 	return count;
2651 }
2652 
2653 static int write_demux_regids(u64 __user *uindices)
2654 {
2655 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2656 	unsigned int i;
2657 
2658 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2659 	for (i = 0; i < CSSELR_MAX; i++) {
2660 		if (!is_valid_cache(i))
2661 			continue;
2662 		if (put_user(val | i, uindices))
2663 			return -EFAULT;
2664 		uindices++;
2665 	}
2666 	return 0;
2667 }
2668 
2669 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2670 {
2671 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2672 		KVM_REG_ARM64_SYSREG |
2673 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2674 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2675 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2676 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2677 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2678 }
2679 
2680 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2681 {
2682 	if (!*uind)
2683 		return true;
2684 
2685 	if (put_user(sys_reg_to_index(reg), *uind))
2686 		return false;
2687 
2688 	(*uind)++;
2689 	return true;
2690 }
2691 
2692 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2693 			    const struct sys_reg_desc *rd,
2694 			    u64 __user **uind,
2695 			    unsigned int *total)
2696 {
2697 	/*
2698 	 * Ignore registers we trap but don't save,
2699 	 * and for which no custom user accessor is provided.
2700 	 */
2701 	if (!(rd->reg || rd->get_user))
2702 		return 0;
2703 
2704 	if (sysreg_hidden(vcpu, rd))
2705 		return 0;
2706 
2707 	if (!copy_reg_to_user(rd, uind))
2708 		return -EFAULT;
2709 
2710 	(*total)++;
2711 	return 0;
2712 }
2713 
2714 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2715 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2716 {
2717 	const struct sys_reg_desc *i2, *end2;
2718 	unsigned int total = 0;
2719 	int err;
2720 
2721 	i2 = sys_reg_descs;
2722 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2723 
2724 	while (i2 != end2) {
2725 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2726 		if (err)
2727 			return err;
2728 	}
2729 	return total;
2730 }
2731 
2732 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2733 {
2734 	return ARRAY_SIZE(invariant_sys_regs)
2735 		+ num_demux_regs()
2736 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2737 }
2738 
2739 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2740 {
2741 	unsigned int i;
2742 	int err;
2743 
2744 	/* Then give them all the invariant registers' indices. */
2745 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2746 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2747 			return -EFAULT;
2748 		uindices++;
2749 	}
2750 
2751 	err = walk_sys_regs(vcpu, uindices);
2752 	if (err < 0)
2753 		return err;
2754 	uindices += err;
2755 
2756 	return write_demux_regids(uindices);
2757 }
2758 
2759 void kvm_sys_reg_table_init(void)
2760 {
2761 	unsigned int i;
2762 	struct sys_reg_desc clidr;
2763 
2764 	/* Make sure tables are unique and in order. */
2765 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2766 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2767 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2768 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2769 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2770 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2771 
2772 	/* We abuse the reset function to overwrite the table itself. */
2773 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2774 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2775 
2776 	/*
2777 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2778 	 *
2779 	 *   If software reads the Cache Type fields from Ctype1
2780 	 *   upwards, once it has seen a value of 0b000, no caches
2781 	 *   exist at further-out levels of the hierarchy. So, for
2782 	 *   example, if Ctype3 is the first Cache Type field with a
2783 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2784 	 *   ignored.
2785 	 */
2786 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2787 	cache_levels = clidr.val;
2788 	for (i = 0; i < 7; i++)
2789 		if (((cache_levels >> (i*3)) & 7) == 0)
2790 			break;
2791 	/* Clear all higher bits. */
2792 	cache_levels &= (1 << (i*3))-1;
2793 }
2794