xref: /linux/arch/arm64/kvm/sys_regs.c (revision b9527b38c66730061c245e353dab42ef7dda33c6)
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/cacheinfo.h>
15 #include <linux/debugfs.h>
16 #include <linux/kvm_host.h>
17 #include <linux/mm.h>
18 #include <linux/printk.h>
19 #include <linux/uaccess.h>
20 
21 #include <asm/arm_pmuv3.h>
22 #include <asm/cacheflush.h>
23 #include <asm/cputype.h>
24 #include <asm/debug-monitors.h>
25 #include <asm/esr.h>
26 #include <asm/kvm_arm.h>
27 #include <asm/kvm_emulate.h>
28 #include <asm/kvm_hyp.h>
29 #include <asm/kvm_mmu.h>
30 #include <asm/kvm_nested.h>
31 #include <asm/perf_event.h>
32 #include <asm/sysreg.h>
33 
34 #include <trace/events/kvm.h>
35 
36 #include "sys_regs.h"
37 #include "vgic/vgic.h"
38 
39 #include "trace.h"
40 
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 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
48 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
49 		      u64 val);
50 
51 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
52 			 const struct sys_reg_desc *r)
53 {
54 	kvm_inject_undefined(vcpu);
55 	return false;
56 }
57 
58 static bool bad_trap(struct kvm_vcpu *vcpu,
59 		     struct sys_reg_params *params,
60 		     const struct sys_reg_desc *r,
61 		     const char *msg)
62 {
63 	WARN_ONCE(1, "Unexpected %s\n", msg);
64 	print_sys_reg_instr(params);
65 	return undef_access(vcpu, params, r);
66 }
67 
68 static bool read_from_write_only(struct kvm_vcpu *vcpu,
69 				 struct sys_reg_params *params,
70 				 const struct sys_reg_desc *r)
71 {
72 	return bad_trap(vcpu, params, r,
73 			"sys_reg read to write-only register");
74 }
75 
76 static bool write_to_read_only(struct kvm_vcpu *vcpu,
77 			       struct sys_reg_params *params,
78 			       const struct sys_reg_desc *r)
79 {
80 	return bad_trap(vcpu, params, r,
81 			"sys_reg write to read-only register");
82 }
83 
84 #define PURE_EL2_SYSREG(el2)						\
85 	case el2: {							\
86 		*el1r = el2;						\
87 		return true;						\
88 	}
89 
90 #define MAPPED_EL2_SYSREG(el2, el1, fn)					\
91 	case el2: {							\
92 		*xlate = fn;						\
93 		*el1r = el1;						\
94 		return true;						\
95 	}
96 
97 static bool get_el2_to_el1_mapping(unsigned int reg,
98 				   unsigned int *el1r, u64 (**xlate)(u64))
99 {
100 	switch (reg) {
101 		PURE_EL2_SYSREG(  VPIDR_EL2	);
102 		PURE_EL2_SYSREG(  VMPIDR_EL2	);
103 		PURE_EL2_SYSREG(  ACTLR_EL2	);
104 		PURE_EL2_SYSREG(  HCR_EL2	);
105 		PURE_EL2_SYSREG(  MDCR_EL2	);
106 		PURE_EL2_SYSREG(  HSTR_EL2	);
107 		PURE_EL2_SYSREG(  HACR_EL2	);
108 		PURE_EL2_SYSREG(  VTTBR_EL2	);
109 		PURE_EL2_SYSREG(  VTCR_EL2	);
110 		PURE_EL2_SYSREG(  RVBAR_EL2	);
111 		PURE_EL2_SYSREG(  TPIDR_EL2	);
112 		PURE_EL2_SYSREG(  HPFAR_EL2	);
113 		PURE_EL2_SYSREG(  HCRX_EL2	);
114 		PURE_EL2_SYSREG(  HFGRTR_EL2	);
115 		PURE_EL2_SYSREG(  HFGWTR_EL2	);
116 		PURE_EL2_SYSREG(  HFGITR_EL2	);
117 		PURE_EL2_SYSREG(  HDFGRTR_EL2	);
118 		PURE_EL2_SYSREG(  HDFGWTR_EL2	);
119 		PURE_EL2_SYSREG(  HAFGRTR_EL2	);
120 		PURE_EL2_SYSREG(  CNTVOFF_EL2	);
121 		PURE_EL2_SYSREG(  CNTHCTL_EL2	);
122 		MAPPED_EL2_SYSREG(SCTLR_EL2,   SCTLR_EL1,
123 				  translate_sctlr_el2_to_sctlr_el1	     );
124 		MAPPED_EL2_SYSREG(CPTR_EL2,    CPACR_EL1,
125 				  translate_cptr_el2_to_cpacr_el1	     );
126 		MAPPED_EL2_SYSREG(TTBR0_EL2,   TTBR0_EL1,
127 				  translate_ttbr0_el2_to_ttbr0_el1	     );
128 		MAPPED_EL2_SYSREG(TTBR1_EL2,   TTBR1_EL1,   NULL	     );
129 		MAPPED_EL2_SYSREG(TCR_EL2,     TCR_EL1,
130 				  translate_tcr_el2_to_tcr_el1		     );
131 		MAPPED_EL2_SYSREG(VBAR_EL2,    VBAR_EL1,    NULL	     );
132 		MAPPED_EL2_SYSREG(AFSR0_EL2,   AFSR0_EL1,   NULL	     );
133 		MAPPED_EL2_SYSREG(AFSR1_EL2,   AFSR1_EL1,   NULL	     );
134 		MAPPED_EL2_SYSREG(ESR_EL2,     ESR_EL1,     NULL	     );
135 		MAPPED_EL2_SYSREG(FAR_EL2,     FAR_EL1,     NULL	     );
136 		MAPPED_EL2_SYSREG(MAIR_EL2,    MAIR_EL1,    NULL	     );
137 		MAPPED_EL2_SYSREG(AMAIR_EL2,   AMAIR_EL1,   NULL	     );
138 		MAPPED_EL2_SYSREG(ELR_EL2,     ELR_EL1,	    NULL	     );
139 		MAPPED_EL2_SYSREG(SPSR_EL2,    SPSR_EL1,    NULL	     );
140 		MAPPED_EL2_SYSREG(ZCR_EL2,     ZCR_EL1,     NULL	     );
141 		MAPPED_EL2_SYSREG(CONTEXTIDR_EL2, CONTEXTIDR_EL1, NULL	     );
142 	default:
143 		return false;
144 	}
145 }
146 
147 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
148 {
149 	u64 val = 0x8badf00d8badf00d;
150 	u64 (*xlate)(u64) = NULL;
151 	unsigned int el1r;
152 
153 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU))
154 		goto memory_read;
155 
156 	if (unlikely(get_el2_to_el1_mapping(reg, &el1r, &xlate))) {
157 		if (!is_hyp_ctxt(vcpu))
158 			goto memory_read;
159 
160 		/*
161 		 * CNTHCTL_EL2 requires some special treatment to
162 		 * account for the bits that can be set via CNTKCTL_EL1.
163 		 */
164 		switch (reg) {
165 		case CNTHCTL_EL2:
166 			if (vcpu_el2_e2h_is_set(vcpu)) {
167 				val = read_sysreg_el1(SYS_CNTKCTL);
168 				val &= CNTKCTL_VALID_BITS;
169 				val |= __vcpu_sys_reg(vcpu, reg) & ~CNTKCTL_VALID_BITS;
170 				return val;
171 			}
172 			break;
173 		}
174 
175 		/*
176 		 * If this register does not have an EL1 counterpart,
177 		 * then read the stored EL2 version.
178 		 */
179 		if (reg == el1r)
180 			goto memory_read;
181 
182 		/*
183 		 * If we have a non-VHE guest and that the sysreg
184 		 * requires translation to be used at EL1, use the
185 		 * in-memory copy instead.
186 		 */
187 		if (!vcpu_el2_e2h_is_set(vcpu) && xlate)
188 			goto memory_read;
189 
190 		/* Get the current version of the EL1 counterpart. */
191 		WARN_ON(!__vcpu_read_sys_reg_from_cpu(el1r, &val));
192 		return val;
193 	}
194 
195 	/* EL1 register can't be on the CPU if the guest is in vEL2. */
196 	if (unlikely(is_hyp_ctxt(vcpu)))
197 		goto memory_read;
198 
199 	if (__vcpu_read_sys_reg_from_cpu(reg, &val))
200 		return val;
201 
202 memory_read:
203 	return __vcpu_sys_reg(vcpu, reg);
204 }
205 
206 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
207 {
208 	u64 (*xlate)(u64) = NULL;
209 	unsigned int el1r;
210 
211 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU))
212 		goto memory_write;
213 
214 	if (unlikely(get_el2_to_el1_mapping(reg, &el1r, &xlate))) {
215 		if (!is_hyp_ctxt(vcpu))
216 			goto memory_write;
217 
218 		/*
219 		 * Always store a copy of the write to memory to avoid having
220 		 * to reverse-translate virtual EL2 system registers for a
221 		 * non-VHE guest hypervisor.
222 		 */
223 		__vcpu_sys_reg(vcpu, reg) = val;
224 
225 		switch (reg) {
226 		case CNTHCTL_EL2:
227 			/*
228 			 * If E2H=0, CNHTCTL_EL2 is a pure shadow register.
229 			 * Otherwise, some of the bits are backed by
230 			 * CNTKCTL_EL1, while the rest is kept in memory.
231 			 * Yes, this is fun stuff.
232 			 */
233 			if (vcpu_el2_e2h_is_set(vcpu))
234 				write_sysreg_el1(val, SYS_CNTKCTL);
235 			return;
236 		}
237 
238 		/* No EL1 counterpart? We're done here.? */
239 		if (reg == el1r)
240 			return;
241 
242 		if (!vcpu_el2_e2h_is_set(vcpu) && xlate)
243 			val = xlate(val);
244 
245 		/* Redirect this to the EL1 version of the register. */
246 		WARN_ON(!__vcpu_write_sys_reg_to_cpu(val, el1r));
247 		return;
248 	}
249 
250 	/* EL1 register can't be on the CPU if the guest is in vEL2. */
251 	if (unlikely(is_hyp_ctxt(vcpu)))
252 		goto memory_write;
253 
254 	if (__vcpu_write_sys_reg_to_cpu(val, reg))
255 		return;
256 
257 memory_write:
258 	 __vcpu_sys_reg(vcpu, reg) = val;
259 }
260 
261 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
262 #define CSSELR_MAX 14
263 
264 /*
265  * Returns the minimum line size for the selected cache, expressed as
266  * Log2(bytes).
267  */
268 static u8 get_min_cache_line_size(bool icache)
269 {
270 	u64 ctr = read_sanitised_ftr_reg(SYS_CTR_EL0);
271 	u8 field;
272 
273 	if (icache)
274 		field = SYS_FIELD_GET(CTR_EL0, IminLine, ctr);
275 	else
276 		field = SYS_FIELD_GET(CTR_EL0, DminLine, ctr);
277 
278 	/*
279 	 * Cache line size is represented as Log2(words) in CTR_EL0.
280 	 * Log2(bytes) can be derived with the following:
281 	 *
282 	 * Log2(words) + 2 = Log2(bytes / 4) + 2
283 	 * 		   = Log2(bytes) - 2 + 2
284 	 * 		   = Log2(bytes)
285 	 */
286 	return field + 2;
287 }
288 
289 /* Which cache CCSIDR represents depends on CSSELR value. */
290 static u32 get_ccsidr(struct kvm_vcpu *vcpu, u32 csselr)
291 {
292 	u8 line_size;
293 
294 	if (vcpu->arch.ccsidr)
295 		return vcpu->arch.ccsidr[csselr];
296 
297 	line_size = get_min_cache_line_size(csselr & CSSELR_EL1_InD);
298 
299 	/*
300 	 * Fabricate a CCSIDR value as the overriding value does not exist.
301 	 * The real CCSIDR value will not be used as it can vary by the
302 	 * physical CPU which the vcpu currently resides in.
303 	 *
304 	 * The line size is determined with get_min_cache_line_size(), which
305 	 * should be valid for all CPUs even if they have different cache
306 	 * configuration.
307 	 *
308 	 * The associativity bits are cleared, meaning the geometry of all data
309 	 * and unified caches (which are guaranteed to be PIPT and thus
310 	 * non-aliasing) are 1 set and 1 way.
311 	 * Guests should not be doing cache operations by set/way at all, and
312 	 * for this reason, we trap them and attempt to infer the intent, so
313 	 * that we can flush the entire guest's address space at the appropriate
314 	 * time. The exposed geometry minimizes the number of the traps.
315 	 * [If guests should attempt to infer aliasing properties from the
316 	 * geometry (which is not permitted by the architecture), they would
317 	 * only do so for virtually indexed caches.]
318 	 *
319 	 * We don't check if the cache level exists as it is allowed to return
320 	 * an UNKNOWN value if not.
321 	 */
322 	return SYS_FIELD_PREP(CCSIDR_EL1, LineSize, line_size - 4);
323 }
324 
325 static int set_ccsidr(struct kvm_vcpu *vcpu, u32 csselr, u32 val)
326 {
327 	u8 line_size = FIELD_GET(CCSIDR_EL1_LineSize, val) + 4;
328 	u32 *ccsidr = vcpu->arch.ccsidr;
329 	u32 i;
330 
331 	if ((val & CCSIDR_EL1_RES0) ||
332 	    line_size < get_min_cache_line_size(csselr & CSSELR_EL1_InD))
333 		return -EINVAL;
334 
335 	if (!ccsidr) {
336 		if (val == get_ccsidr(vcpu, csselr))
337 			return 0;
338 
339 		ccsidr = kmalloc_array(CSSELR_MAX, sizeof(u32), GFP_KERNEL_ACCOUNT);
340 		if (!ccsidr)
341 			return -ENOMEM;
342 
343 		for (i = 0; i < CSSELR_MAX; i++)
344 			ccsidr[i] = get_ccsidr(vcpu, i);
345 
346 		vcpu->arch.ccsidr = ccsidr;
347 	}
348 
349 	ccsidr[csselr] = val;
350 
351 	return 0;
352 }
353 
354 static bool access_rw(struct kvm_vcpu *vcpu,
355 		      struct sys_reg_params *p,
356 		      const struct sys_reg_desc *r)
357 {
358 	if (p->is_write)
359 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
360 	else
361 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
362 
363 	return true;
364 }
365 
366 /*
367  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
368  */
369 static bool access_dcsw(struct kvm_vcpu *vcpu,
370 			struct sys_reg_params *p,
371 			const struct sys_reg_desc *r)
372 {
373 	if (!p->is_write)
374 		return read_from_write_only(vcpu, p, r);
375 
376 	/*
377 	 * Only track S/W ops if we don't have FWB. It still indicates
378 	 * that the guest is a bit broken (S/W operations should only
379 	 * be done by firmware, knowing that there is only a single
380 	 * CPU left in the system, and certainly not from non-secure
381 	 * software).
382 	 */
383 	if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
384 		kvm_set_way_flush(vcpu);
385 
386 	return true;
387 }
388 
389 static bool access_dcgsw(struct kvm_vcpu *vcpu,
390 			 struct sys_reg_params *p,
391 			 const struct sys_reg_desc *r)
392 {
393 	if (!kvm_has_mte(vcpu->kvm))
394 		return undef_access(vcpu, p, r);
395 
396 	/* Treat MTE S/W ops as we treat the classic ones: with contempt */
397 	return access_dcsw(vcpu, p, r);
398 }
399 
400 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
401 {
402 	switch (r->aarch32_map) {
403 	case AA32_LO:
404 		*mask = GENMASK_ULL(31, 0);
405 		*shift = 0;
406 		break;
407 	case AA32_HI:
408 		*mask = GENMASK_ULL(63, 32);
409 		*shift = 32;
410 		break;
411 	default:
412 		*mask = GENMASK_ULL(63, 0);
413 		*shift = 0;
414 		break;
415 	}
416 }
417 
418 /*
419  * Generic accessor for VM registers. Only called as long as HCR_TVM
420  * is set. If the guest enables the MMU, we stop trapping the VM
421  * sys_regs and leave it in complete control of the caches.
422  */
423 static bool access_vm_reg(struct kvm_vcpu *vcpu,
424 			  struct sys_reg_params *p,
425 			  const struct sys_reg_desc *r)
426 {
427 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
428 	u64 val, mask, shift;
429 
430 	if (reg_to_encoding(r) == SYS_TCR2_EL1 &&
431 	    !kvm_has_feat(vcpu->kvm, ID_AA64MMFR3_EL1, TCRX, IMP))
432 		return undef_access(vcpu, p, r);
433 
434 	BUG_ON(!p->is_write);
435 
436 	get_access_mask(r, &mask, &shift);
437 
438 	if (~mask) {
439 		val = vcpu_read_sys_reg(vcpu, r->reg);
440 		val &= ~mask;
441 	} else {
442 		val = 0;
443 	}
444 
445 	val |= (p->regval & (mask >> shift)) << shift;
446 	vcpu_write_sys_reg(vcpu, val, r->reg);
447 
448 	kvm_toggle_cache(vcpu, was_enabled);
449 	return true;
450 }
451 
452 static bool access_actlr(struct kvm_vcpu *vcpu,
453 			 struct sys_reg_params *p,
454 			 const struct sys_reg_desc *r)
455 {
456 	u64 mask, shift;
457 
458 	if (p->is_write)
459 		return ignore_write(vcpu, p);
460 
461 	get_access_mask(r, &mask, &shift);
462 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
463 
464 	return true;
465 }
466 
467 /*
468  * Trap handler for the GICv3 SGI generation system register.
469  * Forward the request to the VGIC emulation.
470  * The cp15_64 code makes sure this automatically works
471  * for both AArch64 and AArch32 accesses.
472  */
473 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
474 			   struct sys_reg_params *p,
475 			   const struct sys_reg_desc *r)
476 {
477 	bool g1;
478 
479 	if (!kvm_has_gicv3(vcpu->kvm))
480 		return undef_access(vcpu, p, r);
481 
482 	if (!p->is_write)
483 		return read_from_write_only(vcpu, p, r);
484 
485 	/*
486 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
487 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
488 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
489 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
490 	 * group.
491 	 */
492 	if (p->Op0 == 0) {		/* AArch32 */
493 		switch (p->Op1) {
494 		default:		/* Keep GCC quiet */
495 		case 0:			/* ICC_SGI1R */
496 			g1 = true;
497 			break;
498 		case 1:			/* ICC_ASGI1R */
499 		case 2:			/* ICC_SGI0R */
500 			g1 = false;
501 			break;
502 		}
503 	} else {			/* AArch64 */
504 		switch (p->Op2) {
505 		default:		/* Keep GCC quiet */
506 		case 5:			/* ICC_SGI1R_EL1 */
507 			g1 = true;
508 			break;
509 		case 6:			/* ICC_ASGI1R_EL1 */
510 		case 7:			/* ICC_SGI0R_EL1 */
511 			g1 = false;
512 			break;
513 		}
514 	}
515 
516 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
517 
518 	return true;
519 }
520 
521 static bool access_gic_sre(struct kvm_vcpu *vcpu,
522 			   struct sys_reg_params *p,
523 			   const struct sys_reg_desc *r)
524 {
525 	if (!kvm_has_gicv3(vcpu->kvm))
526 		return undef_access(vcpu, p, r);
527 
528 	if (p->is_write)
529 		return ignore_write(vcpu, p);
530 
531 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
532 	return true;
533 }
534 
535 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
536 			struct sys_reg_params *p,
537 			const struct sys_reg_desc *r)
538 {
539 	if (p->is_write)
540 		return ignore_write(vcpu, p);
541 	else
542 		return read_zero(vcpu, p);
543 }
544 
545 /*
546  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
547  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
548  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
549  * treat it separately.
550  */
551 static bool trap_loregion(struct kvm_vcpu *vcpu,
552 			  struct sys_reg_params *p,
553 			  const struct sys_reg_desc *r)
554 {
555 	u32 sr = reg_to_encoding(r);
556 
557 	if (!kvm_has_feat(vcpu->kvm, ID_AA64MMFR1_EL1, LO, IMP))
558 		return undef_access(vcpu, p, r);
559 
560 	if (p->is_write && sr == SYS_LORID_EL1)
561 		return write_to_read_only(vcpu, p, r);
562 
563 	return trap_raz_wi(vcpu, p, r);
564 }
565 
566 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
567 			   struct sys_reg_params *p,
568 			   const struct sys_reg_desc *r)
569 {
570 	u64 oslsr;
571 
572 	if (!p->is_write)
573 		return read_from_write_only(vcpu, p, r);
574 
575 	/* Forward the OSLK bit to OSLSR */
576 	oslsr = __vcpu_sys_reg(vcpu, OSLSR_EL1) & ~OSLSR_EL1_OSLK;
577 	if (p->regval & OSLAR_EL1_OSLK)
578 		oslsr |= OSLSR_EL1_OSLK;
579 
580 	__vcpu_sys_reg(vcpu, OSLSR_EL1) = oslsr;
581 	return true;
582 }
583 
584 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
585 			   struct sys_reg_params *p,
586 			   const struct sys_reg_desc *r)
587 {
588 	if (p->is_write)
589 		return write_to_read_only(vcpu, p, r);
590 
591 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
592 	return true;
593 }
594 
595 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
596 			 u64 val)
597 {
598 	/*
599 	 * The only modifiable bit is the OSLK bit. Refuse the write if
600 	 * userspace attempts to change any other bit in the register.
601 	 */
602 	if ((val ^ rd->val) & ~OSLSR_EL1_OSLK)
603 		return -EINVAL;
604 
605 	__vcpu_sys_reg(vcpu, rd->reg) = val;
606 	return 0;
607 }
608 
609 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
610 				   struct sys_reg_params *p,
611 				   const struct sys_reg_desc *r)
612 {
613 	if (p->is_write) {
614 		return ignore_write(vcpu, p);
615 	} else {
616 		p->regval = read_sysreg(dbgauthstatus_el1);
617 		return true;
618 	}
619 }
620 
621 /*
622  * We want to avoid world-switching all the DBG registers all the
623  * time:
624  *
625  * - If we've touched any debug register, it is likely that we're
626  *   going to touch more of them. It then makes sense to disable the
627  *   traps and start doing the save/restore dance
628  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
629  *   then mandatory to save/restore the registers, as the guest
630  *   depends on them.
631  *
632  * For this, we use a DIRTY bit, indicating the guest has modified the
633  * debug registers, used as follow:
634  *
635  * On guest entry:
636  * - If the dirty bit is set (because we're coming back from trapping),
637  *   disable the traps, save host registers, restore guest registers.
638  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
639  *   set the dirty bit, disable the traps, save host registers,
640  *   restore guest registers.
641  * - Otherwise, enable the traps
642  *
643  * On guest exit:
644  * - If the dirty bit is set, save guest registers, restore host
645  *   registers and clear the dirty bit. This ensure that the host can
646  *   now use the debug registers.
647  */
648 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
649 			    struct sys_reg_params *p,
650 			    const struct sys_reg_desc *r)
651 {
652 	access_rw(vcpu, p, r);
653 	if (p->is_write)
654 		vcpu_set_flag(vcpu, DEBUG_DIRTY);
655 
656 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
657 
658 	return true;
659 }
660 
661 /*
662  * reg_to_dbg/dbg_to_reg
663  *
664  * A 32 bit write to a debug register leave top bits alone
665  * A 32 bit read from a debug register only returns the bottom bits
666  *
667  * All writes will set the DEBUG_DIRTY flag to ensure the hyp code
668  * switches between host and guest values in future.
669  */
670 static void reg_to_dbg(struct kvm_vcpu *vcpu,
671 		       struct sys_reg_params *p,
672 		       const struct sys_reg_desc *rd,
673 		       u64 *dbg_reg)
674 {
675 	u64 mask, shift, val;
676 
677 	get_access_mask(rd, &mask, &shift);
678 
679 	val = *dbg_reg;
680 	val &= ~mask;
681 	val |= (p->regval & (mask >> shift)) << shift;
682 	*dbg_reg = val;
683 
684 	vcpu_set_flag(vcpu, DEBUG_DIRTY);
685 }
686 
687 static void dbg_to_reg(struct kvm_vcpu *vcpu,
688 		       struct sys_reg_params *p,
689 		       const struct sys_reg_desc *rd,
690 		       u64 *dbg_reg)
691 {
692 	u64 mask, shift;
693 
694 	get_access_mask(rd, &mask, &shift);
695 	p->regval = (*dbg_reg & mask) >> shift;
696 }
697 
698 static bool trap_bvr(struct kvm_vcpu *vcpu,
699 		     struct sys_reg_params *p,
700 		     const struct sys_reg_desc *rd)
701 {
702 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
703 
704 	if (p->is_write)
705 		reg_to_dbg(vcpu, p, rd, dbg_reg);
706 	else
707 		dbg_to_reg(vcpu, p, rd, dbg_reg);
708 
709 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
710 
711 	return true;
712 }
713 
714 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
715 		   u64 val)
716 {
717 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = val;
718 	return 0;
719 }
720 
721 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
722 		   u64 *val)
723 {
724 	*val = vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
725 	return 0;
726 }
727 
728 static u64 reset_bvr(struct kvm_vcpu *vcpu,
729 		      const struct sys_reg_desc *rd)
730 {
731 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = rd->val;
732 	return rd->val;
733 }
734 
735 static bool trap_bcr(struct kvm_vcpu *vcpu,
736 		     struct sys_reg_params *p,
737 		     const struct sys_reg_desc *rd)
738 {
739 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
740 
741 	if (p->is_write)
742 		reg_to_dbg(vcpu, p, rd, dbg_reg);
743 	else
744 		dbg_to_reg(vcpu, p, rd, dbg_reg);
745 
746 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
747 
748 	return true;
749 }
750 
751 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
752 		   u64 val)
753 {
754 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = val;
755 	return 0;
756 }
757 
758 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
759 		   u64 *val)
760 {
761 	*val = vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
762 	return 0;
763 }
764 
765 static u64 reset_bcr(struct kvm_vcpu *vcpu,
766 		      const struct sys_reg_desc *rd)
767 {
768 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = rd->val;
769 	return rd->val;
770 }
771 
772 static bool trap_wvr(struct kvm_vcpu *vcpu,
773 		     struct sys_reg_params *p,
774 		     const struct sys_reg_desc *rd)
775 {
776 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
777 
778 	if (p->is_write)
779 		reg_to_dbg(vcpu, p, rd, dbg_reg);
780 	else
781 		dbg_to_reg(vcpu, p, rd, dbg_reg);
782 
783 	trace_trap_reg(__func__, rd->CRm, p->is_write,
784 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm]);
785 
786 	return true;
787 }
788 
789 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
790 		   u64 val)
791 {
792 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = val;
793 	return 0;
794 }
795 
796 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
797 		   u64 *val)
798 {
799 	*val = vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
800 	return 0;
801 }
802 
803 static u64 reset_wvr(struct kvm_vcpu *vcpu,
804 		      const struct sys_reg_desc *rd)
805 {
806 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = rd->val;
807 	return rd->val;
808 }
809 
810 static bool trap_wcr(struct kvm_vcpu *vcpu,
811 		     struct sys_reg_params *p,
812 		     const struct sys_reg_desc *rd)
813 {
814 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
815 
816 	if (p->is_write)
817 		reg_to_dbg(vcpu, p, rd, dbg_reg);
818 	else
819 		dbg_to_reg(vcpu, p, rd, dbg_reg);
820 
821 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
822 
823 	return true;
824 }
825 
826 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
827 		   u64 val)
828 {
829 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = val;
830 	return 0;
831 }
832 
833 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
834 		   u64 *val)
835 {
836 	*val = vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
837 	return 0;
838 }
839 
840 static u64 reset_wcr(struct kvm_vcpu *vcpu,
841 		      const struct sys_reg_desc *rd)
842 {
843 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = rd->val;
844 	return rd->val;
845 }
846 
847 static u64 reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
848 {
849 	u64 amair = read_sysreg(amair_el1);
850 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
851 	return amair;
852 }
853 
854 static u64 reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
855 {
856 	u64 actlr = read_sysreg(actlr_el1);
857 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
858 	return actlr;
859 }
860 
861 static u64 reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
862 {
863 	u64 mpidr;
864 
865 	/*
866 	 * Map the vcpu_id into the first three affinity level fields of
867 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
868 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
869 	 * of the GICv3 to be able to address each CPU directly when
870 	 * sending IPIs.
871 	 */
872 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
873 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
874 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
875 	mpidr |= (1ULL << 31);
876 	vcpu_write_sys_reg(vcpu, mpidr, MPIDR_EL1);
877 
878 	return mpidr;
879 }
880 
881 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
882 				   const struct sys_reg_desc *r)
883 {
884 	if (kvm_vcpu_has_pmu(vcpu))
885 		return 0;
886 
887 	return REG_HIDDEN;
888 }
889 
890 static u64 reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
891 {
892 	u64 mask = BIT(ARMV8_PMU_CYCLE_IDX);
893 	u8 n = vcpu->kvm->arch.pmcr_n;
894 
895 	if (n)
896 		mask |= GENMASK(n - 1, 0);
897 
898 	reset_unknown(vcpu, r);
899 	__vcpu_sys_reg(vcpu, r->reg) &= mask;
900 
901 	return __vcpu_sys_reg(vcpu, r->reg);
902 }
903 
904 static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
905 {
906 	reset_unknown(vcpu, r);
907 	__vcpu_sys_reg(vcpu, r->reg) &= GENMASK(31, 0);
908 
909 	return __vcpu_sys_reg(vcpu, r->reg);
910 }
911 
912 static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
913 {
914 	/* This thing will UNDEF, who cares about the reset value? */
915 	if (!kvm_vcpu_has_pmu(vcpu))
916 		return 0;
917 
918 	reset_unknown(vcpu, r);
919 	__vcpu_sys_reg(vcpu, r->reg) &= kvm_pmu_evtyper_mask(vcpu->kvm);
920 
921 	return __vcpu_sys_reg(vcpu, r->reg);
922 }
923 
924 static u64 reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
925 {
926 	reset_unknown(vcpu, r);
927 	__vcpu_sys_reg(vcpu, r->reg) &= PMSELR_EL0_SEL_MASK;
928 
929 	return __vcpu_sys_reg(vcpu, r->reg);
930 }
931 
932 static u64 reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
933 {
934 	u64 pmcr = 0;
935 
936 	if (!kvm_supports_32bit_el0())
937 		pmcr |= ARMV8_PMU_PMCR_LC;
938 
939 	/*
940 	 * The value of PMCR.N field is included when the
941 	 * vCPU register is read via kvm_vcpu_read_pmcr().
942 	 */
943 	__vcpu_sys_reg(vcpu, r->reg) = pmcr;
944 
945 	return __vcpu_sys_reg(vcpu, r->reg);
946 }
947 
948 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
949 {
950 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
951 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
952 
953 	if (!enabled)
954 		kvm_inject_undefined(vcpu);
955 
956 	return !enabled;
957 }
958 
959 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
960 {
961 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
962 }
963 
964 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
965 {
966 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
967 }
968 
969 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
970 {
971 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
972 }
973 
974 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
975 {
976 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
977 }
978 
979 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
980 			const struct sys_reg_desc *r)
981 {
982 	u64 val;
983 
984 	if (pmu_access_el0_disabled(vcpu))
985 		return false;
986 
987 	if (p->is_write) {
988 		/*
989 		 * Only update writeable bits of PMCR (continuing into
990 		 * kvm_pmu_handle_pmcr() as well)
991 		 */
992 		val = kvm_vcpu_read_pmcr(vcpu);
993 		val &= ~ARMV8_PMU_PMCR_MASK;
994 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
995 		if (!kvm_supports_32bit_el0())
996 			val |= ARMV8_PMU_PMCR_LC;
997 		kvm_pmu_handle_pmcr(vcpu, val);
998 	} else {
999 		/* PMCR.P & PMCR.C are RAZ */
1000 		val = kvm_vcpu_read_pmcr(vcpu)
1001 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
1002 		p->regval = val;
1003 	}
1004 
1005 	return true;
1006 }
1007 
1008 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1009 			  const struct sys_reg_desc *r)
1010 {
1011 	if (pmu_access_event_counter_el0_disabled(vcpu))
1012 		return false;
1013 
1014 	if (p->is_write)
1015 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
1016 	else
1017 		/* return PMSELR.SEL field */
1018 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
1019 			    & PMSELR_EL0_SEL_MASK;
1020 
1021 	return true;
1022 }
1023 
1024 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1025 			  const struct sys_reg_desc *r)
1026 {
1027 	u64 pmceid, mask, shift;
1028 
1029 	BUG_ON(p->is_write);
1030 
1031 	if (pmu_access_el0_disabled(vcpu))
1032 		return false;
1033 
1034 	get_access_mask(r, &mask, &shift);
1035 
1036 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
1037 	pmceid &= mask;
1038 	pmceid >>= shift;
1039 
1040 	p->regval = pmceid;
1041 
1042 	return true;
1043 }
1044 
1045 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
1046 {
1047 	u64 pmcr, val;
1048 
1049 	pmcr = kvm_vcpu_read_pmcr(vcpu);
1050 	val = FIELD_GET(ARMV8_PMU_PMCR_N, pmcr);
1051 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
1052 		kvm_inject_undefined(vcpu);
1053 		return false;
1054 	}
1055 
1056 	return true;
1057 }
1058 
1059 static int get_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1060 			  u64 *val)
1061 {
1062 	u64 idx;
1063 
1064 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1065 		/* PMCCNTR_EL0 */
1066 		idx = ARMV8_PMU_CYCLE_IDX;
1067 	else
1068 		/* PMEVCNTRn_EL0 */
1069 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1070 
1071 	*val = kvm_pmu_get_counter_value(vcpu, idx);
1072 	return 0;
1073 }
1074 
1075 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
1076 			      struct sys_reg_params *p,
1077 			      const struct sys_reg_desc *r)
1078 {
1079 	u64 idx = ~0UL;
1080 
1081 	if (r->CRn == 9 && r->CRm == 13) {
1082 		if (r->Op2 == 2) {
1083 			/* PMXEVCNTR_EL0 */
1084 			if (pmu_access_event_counter_el0_disabled(vcpu))
1085 				return false;
1086 
1087 			idx = SYS_FIELD_GET(PMSELR_EL0, SEL,
1088 					    __vcpu_sys_reg(vcpu, PMSELR_EL0));
1089 		} else if (r->Op2 == 0) {
1090 			/* PMCCNTR_EL0 */
1091 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
1092 				return false;
1093 
1094 			idx = ARMV8_PMU_CYCLE_IDX;
1095 		}
1096 	} else if (r->CRn == 0 && r->CRm == 9) {
1097 		/* PMCCNTR */
1098 		if (pmu_access_event_counter_el0_disabled(vcpu))
1099 			return false;
1100 
1101 		idx = ARMV8_PMU_CYCLE_IDX;
1102 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
1103 		/* PMEVCNTRn_EL0 */
1104 		if (pmu_access_event_counter_el0_disabled(vcpu))
1105 			return false;
1106 
1107 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1108 	}
1109 
1110 	/* Catch any decoding mistake */
1111 	WARN_ON(idx == ~0UL);
1112 
1113 	if (!pmu_counter_idx_valid(vcpu, idx))
1114 		return false;
1115 
1116 	if (p->is_write) {
1117 		if (pmu_access_el0_disabled(vcpu))
1118 			return false;
1119 
1120 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
1121 	} else {
1122 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
1123 	}
1124 
1125 	return true;
1126 }
1127 
1128 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1129 			       const struct sys_reg_desc *r)
1130 {
1131 	u64 idx, reg;
1132 
1133 	if (pmu_access_el0_disabled(vcpu))
1134 		return false;
1135 
1136 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
1137 		/* PMXEVTYPER_EL0 */
1138 		idx = SYS_FIELD_GET(PMSELR_EL0, SEL, __vcpu_sys_reg(vcpu, PMSELR_EL0));
1139 		reg = PMEVTYPER0_EL0 + idx;
1140 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
1141 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1142 		if (idx == ARMV8_PMU_CYCLE_IDX)
1143 			reg = PMCCFILTR_EL0;
1144 		else
1145 			/* PMEVTYPERn_EL0 */
1146 			reg = PMEVTYPER0_EL0 + idx;
1147 	} else {
1148 		BUG();
1149 	}
1150 
1151 	if (!pmu_counter_idx_valid(vcpu, idx))
1152 		return false;
1153 
1154 	if (p->is_write) {
1155 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
1156 		kvm_vcpu_pmu_restore_guest(vcpu);
1157 	} else {
1158 		p->regval = __vcpu_sys_reg(vcpu, reg);
1159 	}
1160 
1161 	return true;
1162 }
1163 
1164 static int set_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 val)
1165 {
1166 	bool set;
1167 
1168 	val &= kvm_pmu_valid_counter_mask(vcpu);
1169 
1170 	switch (r->reg) {
1171 	case PMOVSSET_EL0:
1172 		/* CRm[1] being set indicates a SET register, and CLR otherwise */
1173 		set = r->CRm & 2;
1174 		break;
1175 	default:
1176 		/* Op2[0] being set indicates a SET register, and CLR otherwise */
1177 		set = r->Op2 & 1;
1178 		break;
1179 	}
1180 
1181 	if (set)
1182 		__vcpu_sys_reg(vcpu, r->reg) |= val;
1183 	else
1184 		__vcpu_sys_reg(vcpu, r->reg) &= ~val;
1185 
1186 	return 0;
1187 }
1188 
1189 static int get_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 *val)
1190 {
1191 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
1192 
1193 	*val = __vcpu_sys_reg(vcpu, r->reg) & mask;
1194 	return 0;
1195 }
1196 
1197 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1198 			   const struct sys_reg_desc *r)
1199 {
1200 	u64 val, mask;
1201 
1202 	if (pmu_access_el0_disabled(vcpu))
1203 		return false;
1204 
1205 	mask = kvm_pmu_valid_counter_mask(vcpu);
1206 	if (p->is_write) {
1207 		val = p->regval & mask;
1208 		if (r->Op2 & 0x1) {
1209 			/* accessing PMCNTENSET_EL0 */
1210 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
1211 			kvm_pmu_enable_counter_mask(vcpu, val);
1212 			kvm_vcpu_pmu_restore_guest(vcpu);
1213 		} else {
1214 			/* accessing PMCNTENCLR_EL0 */
1215 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
1216 			kvm_pmu_disable_counter_mask(vcpu, val);
1217 		}
1218 	} else {
1219 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
1220 	}
1221 
1222 	return true;
1223 }
1224 
1225 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1226 			   const struct sys_reg_desc *r)
1227 {
1228 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
1229 
1230 	if (check_pmu_access_disabled(vcpu, 0))
1231 		return false;
1232 
1233 	if (p->is_write) {
1234 		u64 val = p->regval & mask;
1235 
1236 		if (r->Op2 & 0x1)
1237 			/* accessing PMINTENSET_EL1 */
1238 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
1239 		else
1240 			/* accessing PMINTENCLR_EL1 */
1241 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
1242 	} else {
1243 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
1244 	}
1245 
1246 	return true;
1247 }
1248 
1249 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1250 			 const struct sys_reg_desc *r)
1251 {
1252 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
1253 
1254 	if (pmu_access_el0_disabled(vcpu))
1255 		return false;
1256 
1257 	if (p->is_write) {
1258 		if (r->CRm & 0x2)
1259 			/* accessing PMOVSSET_EL0 */
1260 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
1261 		else
1262 			/* accessing PMOVSCLR_EL0 */
1263 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
1264 	} else {
1265 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
1266 	}
1267 
1268 	return true;
1269 }
1270 
1271 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1272 			   const struct sys_reg_desc *r)
1273 {
1274 	u64 mask;
1275 
1276 	if (!p->is_write)
1277 		return read_from_write_only(vcpu, p, r);
1278 
1279 	if (pmu_write_swinc_el0_disabled(vcpu))
1280 		return false;
1281 
1282 	mask = kvm_pmu_valid_counter_mask(vcpu);
1283 	kvm_pmu_software_increment(vcpu, p->regval & mask);
1284 	return true;
1285 }
1286 
1287 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1288 			     const struct sys_reg_desc *r)
1289 {
1290 	if (p->is_write) {
1291 		if (!vcpu_mode_priv(vcpu))
1292 			return undef_access(vcpu, p, r);
1293 
1294 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
1295 			       p->regval & ARMV8_PMU_USERENR_MASK;
1296 	} else {
1297 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1298 			    & ARMV8_PMU_USERENR_MASK;
1299 	}
1300 
1301 	return true;
1302 }
1303 
1304 static int get_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1305 		    u64 *val)
1306 {
1307 	*val = kvm_vcpu_read_pmcr(vcpu);
1308 	return 0;
1309 }
1310 
1311 static int set_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1312 		    u64 val)
1313 {
1314 	u8 new_n = FIELD_GET(ARMV8_PMU_PMCR_N, val);
1315 	struct kvm *kvm = vcpu->kvm;
1316 
1317 	mutex_lock(&kvm->arch.config_lock);
1318 
1319 	/*
1320 	 * The vCPU can't have more counters than the PMU hardware
1321 	 * implements. Ignore this error to maintain compatibility
1322 	 * with the existing KVM behavior.
1323 	 */
1324 	if (!kvm_vm_has_ran_once(kvm) &&
1325 	    new_n <= kvm_arm_pmu_get_max_counters(kvm))
1326 		kvm->arch.pmcr_n = new_n;
1327 
1328 	mutex_unlock(&kvm->arch.config_lock);
1329 
1330 	/*
1331 	 * Ignore writes to RES0 bits, read only bits that are cleared on
1332 	 * vCPU reset, and writable bits that KVM doesn't support yet.
1333 	 * (i.e. only PMCR.N and bits [7:0] are mutable from userspace)
1334 	 * The LP bit is RES0 when FEAT_PMUv3p5 is not supported on the vCPU.
1335 	 * But, we leave the bit as it is here, as the vCPU's PMUver might
1336 	 * be changed later (NOTE: the bit will be cleared on first vCPU run
1337 	 * if necessary).
1338 	 */
1339 	val &= ARMV8_PMU_PMCR_MASK;
1340 
1341 	/* The LC bit is RES1 when AArch32 is not supported */
1342 	if (!kvm_supports_32bit_el0())
1343 		val |= ARMV8_PMU_PMCR_LC;
1344 
1345 	__vcpu_sys_reg(vcpu, r->reg) = val;
1346 	return 0;
1347 }
1348 
1349 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1350 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1351 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1352 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1353 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1354 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1355 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1356 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1357 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1358 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1359 
1360 #define PMU_SYS_REG(name)						\
1361 	SYS_DESC(SYS_##name), .reset = reset_pmu_reg,			\
1362 	.visibility = pmu_visibility
1363 
1364 /* Macro to expand the PMEVCNTRn_EL0 register */
1365 #define PMU_PMEVCNTR_EL0(n)						\
1366 	{ PMU_SYS_REG(PMEVCNTRn_EL0(n)),				\
1367 	  .reset = reset_pmevcntr, .get_user = get_pmu_evcntr,		\
1368 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
1369 
1370 /* Macro to expand the PMEVTYPERn_EL0 register */
1371 #define PMU_PMEVTYPER_EL0(n)						\
1372 	{ PMU_SYS_REG(PMEVTYPERn_EL0(n)),				\
1373 	  .reset = reset_pmevtyper,					\
1374 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
1375 
1376 /* Macro to expand the AMU counter and type registers*/
1377 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1378 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1379 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1380 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1381 
1382 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1383 			const struct sys_reg_desc *rd)
1384 {
1385 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1386 }
1387 
1388 /*
1389  * If we land here on a PtrAuth access, that is because we didn't
1390  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1391  * way this happens is when the guest does not have PtrAuth support
1392  * enabled.
1393  */
1394 #define __PTRAUTH_KEY(k)						\
1395 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1396 	.visibility = ptrauth_visibility}
1397 
1398 #define PTRAUTH_KEY(k)							\
1399 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1400 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1401 
1402 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1403 			      struct sys_reg_params *p,
1404 			      const struct sys_reg_desc *r)
1405 {
1406 	enum kvm_arch_timers tmr;
1407 	enum kvm_arch_timer_regs treg;
1408 	u64 reg = reg_to_encoding(r);
1409 
1410 	switch (reg) {
1411 	case SYS_CNTP_TVAL_EL0:
1412 	case SYS_AARCH32_CNTP_TVAL:
1413 		tmr = TIMER_PTIMER;
1414 		treg = TIMER_REG_TVAL;
1415 		break;
1416 	case SYS_CNTP_CTL_EL0:
1417 	case SYS_AARCH32_CNTP_CTL:
1418 		tmr = TIMER_PTIMER;
1419 		treg = TIMER_REG_CTL;
1420 		break;
1421 	case SYS_CNTP_CVAL_EL0:
1422 	case SYS_AARCH32_CNTP_CVAL:
1423 		tmr = TIMER_PTIMER;
1424 		treg = TIMER_REG_CVAL;
1425 		break;
1426 	case SYS_CNTPCT_EL0:
1427 	case SYS_CNTPCTSS_EL0:
1428 	case SYS_AARCH32_CNTPCT:
1429 		tmr = TIMER_PTIMER;
1430 		treg = TIMER_REG_CNT;
1431 		break;
1432 	default:
1433 		print_sys_reg_msg(p, "%s", "Unhandled trapped timer register");
1434 		return undef_access(vcpu, p, r);
1435 	}
1436 
1437 	if (p->is_write)
1438 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1439 	else
1440 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1441 
1442 	return true;
1443 }
1444 
1445 static s64 kvm_arm64_ftr_safe_value(u32 id, const struct arm64_ftr_bits *ftrp,
1446 				    s64 new, s64 cur)
1447 {
1448 	struct arm64_ftr_bits kvm_ftr = *ftrp;
1449 
1450 	/* Some features have different safe value type in KVM than host features */
1451 	switch (id) {
1452 	case SYS_ID_AA64DFR0_EL1:
1453 		switch (kvm_ftr.shift) {
1454 		case ID_AA64DFR0_EL1_PMUVer_SHIFT:
1455 			kvm_ftr.type = FTR_LOWER_SAFE;
1456 			break;
1457 		case ID_AA64DFR0_EL1_DebugVer_SHIFT:
1458 			kvm_ftr.type = FTR_LOWER_SAFE;
1459 			break;
1460 		}
1461 		break;
1462 	case SYS_ID_DFR0_EL1:
1463 		if (kvm_ftr.shift == ID_DFR0_EL1_PerfMon_SHIFT)
1464 			kvm_ftr.type = FTR_LOWER_SAFE;
1465 		break;
1466 	}
1467 
1468 	return arm64_ftr_safe_value(&kvm_ftr, new, cur);
1469 }
1470 
1471 /*
1472  * arm64_check_features() - Check if a feature register value constitutes
1473  * a subset of features indicated by the idreg's KVM sanitised limit.
1474  *
1475  * This function will check if each feature field of @val is the "safe" value
1476  * against idreg's KVM sanitised limit return from reset() callback.
1477  * If a field value in @val is the same as the one in limit, it is always
1478  * considered the safe value regardless For register fields that are not in
1479  * writable, only the value in limit is considered the safe value.
1480  *
1481  * Return: 0 if all the fields are safe. Otherwise, return negative errno.
1482  */
1483 static int arm64_check_features(struct kvm_vcpu *vcpu,
1484 				const struct sys_reg_desc *rd,
1485 				u64 val)
1486 {
1487 	const struct arm64_ftr_reg *ftr_reg;
1488 	const struct arm64_ftr_bits *ftrp = NULL;
1489 	u32 id = reg_to_encoding(rd);
1490 	u64 writable_mask = rd->val;
1491 	u64 limit = rd->reset(vcpu, rd);
1492 	u64 mask = 0;
1493 
1494 	/*
1495 	 * Hidden and unallocated ID registers may not have a corresponding
1496 	 * struct arm64_ftr_reg. Of course, if the register is RAZ we know the
1497 	 * only safe value is 0.
1498 	 */
1499 	if (sysreg_visible_as_raz(vcpu, rd))
1500 		return val ? -E2BIG : 0;
1501 
1502 	ftr_reg = get_arm64_ftr_reg(id);
1503 	if (!ftr_reg)
1504 		return -EINVAL;
1505 
1506 	ftrp = ftr_reg->ftr_bits;
1507 
1508 	for (; ftrp && ftrp->width; ftrp++) {
1509 		s64 f_val, f_lim, safe_val;
1510 		u64 ftr_mask;
1511 
1512 		ftr_mask = arm64_ftr_mask(ftrp);
1513 		if ((ftr_mask & writable_mask) != ftr_mask)
1514 			continue;
1515 
1516 		f_val = arm64_ftr_value(ftrp, val);
1517 		f_lim = arm64_ftr_value(ftrp, limit);
1518 		mask |= ftr_mask;
1519 
1520 		if (f_val == f_lim)
1521 			safe_val = f_val;
1522 		else
1523 			safe_val = kvm_arm64_ftr_safe_value(id, ftrp, f_val, f_lim);
1524 
1525 		if (safe_val != f_val)
1526 			return -E2BIG;
1527 	}
1528 
1529 	/* For fields that are not writable, values in limit are the safe values. */
1530 	if ((val & ~mask) != (limit & ~mask))
1531 		return -E2BIG;
1532 
1533 	return 0;
1534 }
1535 
1536 static u8 pmuver_to_perfmon(u8 pmuver)
1537 {
1538 	switch (pmuver) {
1539 	case ID_AA64DFR0_EL1_PMUVer_IMP:
1540 		return ID_DFR0_EL1_PerfMon_PMUv3;
1541 	case ID_AA64DFR0_EL1_PMUVer_IMP_DEF:
1542 		return ID_DFR0_EL1_PerfMon_IMPDEF;
1543 	default:
1544 		/* Anything ARMv8.1+ and NI have the same value. For now. */
1545 		return pmuver;
1546 	}
1547 }
1548 
1549 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1550 static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu,
1551 				       const struct sys_reg_desc *r)
1552 {
1553 	u32 id = reg_to_encoding(r);
1554 	u64 val;
1555 
1556 	if (sysreg_visible_as_raz(vcpu, r))
1557 		return 0;
1558 
1559 	val = read_sanitised_ftr_reg(id);
1560 
1561 	switch (id) {
1562 	case SYS_ID_AA64PFR1_EL1:
1563 		if (!kvm_has_mte(vcpu->kvm))
1564 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTE);
1565 
1566 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_SME);
1567 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_RNDR_trap);
1568 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_NMI);
1569 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTE_frac);
1570 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_GCS);
1571 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_THE);
1572 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTEX);
1573 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_DF2);
1574 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_PFAR);
1575 		break;
1576 	case SYS_ID_AA64PFR2_EL1:
1577 		/* We only expose FPMR */
1578 		val &= ID_AA64PFR2_EL1_FPMR;
1579 		break;
1580 	case SYS_ID_AA64ISAR1_EL1:
1581 		if (!vcpu_has_ptrauth(vcpu))
1582 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_APA) |
1583 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_API) |
1584 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPA) |
1585 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPI));
1586 		break;
1587 	case SYS_ID_AA64ISAR2_EL1:
1588 		if (!vcpu_has_ptrauth(vcpu))
1589 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_APA3) |
1590 				 ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_GPA3));
1591 		if (!cpus_have_final_cap(ARM64_HAS_WFXT))
1592 			val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_WFxT);
1593 		break;
1594 	case SYS_ID_AA64MMFR2_EL1:
1595 		val &= ~ID_AA64MMFR2_EL1_CCIDX_MASK;
1596 		break;
1597 	case SYS_ID_AA64MMFR3_EL1:
1598 		val &= ID_AA64MMFR3_EL1_TCRX | ID_AA64MMFR3_EL1_S1POE |
1599 			ID_AA64MMFR3_EL1_S1PIE;
1600 		break;
1601 	case SYS_ID_MMFR4_EL1:
1602 		val &= ~ARM64_FEATURE_MASK(ID_MMFR4_EL1_CCIDX);
1603 		break;
1604 	}
1605 
1606 	return val;
1607 }
1608 
1609 static u64 kvm_read_sanitised_id_reg(struct kvm_vcpu *vcpu,
1610 				     const struct sys_reg_desc *r)
1611 {
1612 	return __kvm_read_sanitised_id_reg(vcpu, r);
1613 }
1614 
1615 static u64 read_id_reg(const struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1616 {
1617 	return kvm_read_vm_id_reg(vcpu->kvm, reg_to_encoding(r));
1618 }
1619 
1620 static bool is_feature_id_reg(u32 encoding)
1621 {
1622 	return (sys_reg_Op0(encoding) == 3 &&
1623 		(sys_reg_Op1(encoding) < 2 || sys_reg_Op1(encoding) == 3) &&
1624 		sys_reg_CRn(encoding) == 0 &&
1625 		sys_reg_CRm(encoding) <= 7);
1626 }
1627 
1628 /*
1629  * Return true if the register's (Op0, Op1, CRn, CRm, Op2) is
1630  * (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8, which is the range of ID
1631  * registers KVM maintains on a per-VM basis.
1632  */
1633 static inline bool is_vm_ftr_id_reg(u32 id)
1634 {
1635 	if (id == SYS_CTR_EL0)
1636 		return true;
1637 
1638 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1639 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1640 		sys_reg_CRm(id) < 8);
1641 }
1642 
1643 static inline bool is_vcpu_ftr_id_reg(u32 id)
1644 {
1645 	return is_feature_id_reg(id) && !is_vm_ftr_id_reg(id);
1646 }
1647 
1648 static inline bool is_aa32_id_reg(u32 id)
1649 {
1650 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1651 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1652 		sys_reg_CRm(id) <= 3);
1653 }
1654 
1655 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1656 				  const struct sys_reg_desc *r)
1657 {
1658 	u32 id = reg_to_encoding(r);
1659 
1660 	switch (id) {
1661 	case SYS_ID_AA64ZFR0_EL1:
1662 		if (!vcpu_has_sve(vcpu))
1663 			return REG_RAZ;
1664 		break;
1665 	}
1666 
1667 	return 0;
1668 }
1669 
1670 static unsigned int aa32_id_visibility(const struct kvm_vcpu *vcpu,
1671 				       const struct sys_reg_desc *r)
1672 {
1673 	/*
1674 	 * AArch32 ID registers are UNKNOWN if AArch32 isn't implemented at any
1675 	 * EL. Promote to RAZ/WI in order to guarantee consistency between
1676 	 * systems.
1677 	 */
1678 	if (!kvm_supports_32bit_el0())
1679 		return REG_RAZ | REG_USER_WI;
1680 
1681 	return id_visibility(vcpu, r);
1682 }
1683 
1684 static unsigned int raz_visibility(const struct kvm_vcpu *vcpu,
1685 				   const struct sys_reg_desc *r)
1686 {
1687 	return REG_RAZ;
1688 }
1689 
1690 /* cpufeature ID register access trap handlers */
1691 
1692 static bool access_id_reg(struct kvm_vcpu *vcpu,
1693 			  struct sys_reg_params *p,
1694 			  const struct sys_reg_desc *r)
1695 {
1696 	if (p->is_write)
1697 		return write_to_read_only(vcpu, p, r);
1698 
1699 	p->regval = read_id_reg(vcpu, r);
1700 
1701 	return true;
1702 }
1703 
1704 /* Visibility overrides for SVE-specific control registers */
1705 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1706 				   const struct sys_reg_desc *rd)
1707 {
1708 	if (vcpu_has_sve(vcpu))
1709 		return 0;
1710 
1711 	return REG_HIDDEN;
1712 }
1713 
1714 static unsigned int sme_visibility(const struct kvm_vcpu *vcpu,
1715 				   const struct sys_reg_desc *rd)
1716 {
1717 	if (kvm_has_feat(vcpu->kvm, ID_AA64PFR1_EL1, SME, IMP))
1718 		return 0;
1719 
1720 	return REG_HIDDEN;
1721 }
1722 
1723 static unsigned int fp8_visibility(const struct kvm_vcpu *vcpu,
1724 				   const struct sys_reg_desc *rd)
1725 {
1726 	if (kvm_has_fpmr(vcpu->kvm))
1727 		return 0;
1728 
1729 	return REG_HIDDEN;
1730 }
1731 
1732 static u64 read_sanitised_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1733 					  const struct sys_reg_desc *rd)
1734 {
1735 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1736 
1737 	if (!vcpu_has_sve(vcpu))
1738 		val &= ~ID_AA64PFR0_EL1_SVE_MASK;
1739 
1740 	/*
1741 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
1742 	 * Although this is a per-CPU feature, we make it global because
1743 	 * asymmetric systems are just a nuisance.
1744 	 *
1745 	 * Userspace can override this as long as it doesn't promise
1746 	 * the impossible.
1747 	 */
1748 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) {
1749 		val &= ~ID_AA64PFR0_EL1_CSV2_MASK;
1750 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV2, IMP);
1751 	}
1752 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) {
1753 		val &= ~ID_AA64PFR0_EL1_CSV3_MASK;
1754 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV3, IMP);
1755 	}
1756 
1757 	if (kvm_vgic_global_state.type == VGIC_V3) {
1758 		val &= ~ID_AA64PFR0_EL1_GIC_MASK;
1759 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, GIC, IMP);
1760 	}
1761 
1762 	val &= ~ID_AA64PFR0_EL1_AMU_MASK;
1763 
1764 	return val;
1765 }
1766 
1767 #define ID_REG_LIMIT_FIELD_ENUM(val, reg, field, limit)			       \
1768 ({									       \
1769 	u64 __f_val = FIELD_GET(reg##_##field##_MASK, val);		       \
1770 	(val) &= ~reg##_##field##_MASK;					       \
1771 	(val) |= FIELD_PREP(reg##_##field##_MASK,			       \
1772 			    min(__f_val,				       \
1773 				(u64)SYS_FIELD_VALUE(reg, field, limit)));     \
1774 	(val);								       \
1775 })
1776 
1777 static u64 read_sanitised_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
1778 					  const struct sys_reg_desc *rd)
1779 {
1780 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1781 
1782 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_AA64DFR0_EL1, DebugVer, V8P8);
1783 
1784 	/*
1785 	 * Only initialize the PMU version if the vCPU was configured with one.
1786 	 */
1787 	val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1788 	if (kvm_vcpu_has_pmu(vcpu))
1789 		val |= SYS_FIELD_PREP(ID_AA64DFR0_EL1, PMUVer,
1790 				      kvm_arm_pmu_get_pmuver_limit());
1791 
1792 	/* Hide SPE from guests */
1793 	val &= ~ID_AA64DFR0_EL1_PMSVer_MASK;
1794 
1795 	return val;
1796 }
1797 
1798 static int set_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
1799 			       const struct sys_reg_desc *rd,
1800 			       u64 val)
1801 {
1802 	u8 debugver = SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, val);
1803 	u8 pmuver = SYS_FIELD_GET(ID_AA64DFR0_EL1, PMUVer, val);
1804 
1805 	/*
1806 	 * Prior to commit 3d0dba5764b9 ("KVM: arm64: PMU: Move the
1807 	 * ID_AA64DFR0_EL1.PMUver limit to VM creation"), KVM erroneously
1808 	 * exposed an IMP_DEF PMU to userspace and the guest on systems w/
1809 	 * non-architectural PMUs. Of course, PMUv3 is the only game in town for
1810 	 * PMU virtualization, so the IMP_DEF value was rather user-hostile.
1811 	 *
1812 	 * At minimum, we're on the hook to allow values that were given to
1813 	 * userspace by KVM. Cover our tracks here and replace the IMP_DEF value
1814 	 * with a more sensible NI. The value of an ID register changing under
1815 	 * the nose of the guest is unfortunate, but is certainly no more
1816 	 * surprising than an ill-guided PMU driver poking at impdef system
1817 	 * registers that end in an UNDEF...
1818 	 */
1819 	if (pmuver == ID_AA64DFR0_EL1_PMUVer_IMP_DEF)
1820 		val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1821 
1822 	/*
1823 	 * ID_AA64DFR0_EL1.DebugVer is one of those awkward fields with a
1824 	 * nonzero minimum safe value.
1825 	 */
1826 	if (debugver < ID_AA64DFR0_EL1_DebugVer_IMP)
1827 		return -EINVAL;
1828 
1829 	return set_id_reg(vcpu, rd, val);
1830 }
1831 
1832 static u64 read_sanitised_id_dfr0_el1(struct kvm_vcpu *vcpu,
1833 				      const struct sys_reg_desc *rd)
1834 {
1835 	u8 perfmon = pmuver_to_perfmon(kvm_arm_pmu_get_pmuver_limit());
1836 	u64 val = read_sanitised_ftr_reg(SYS_ID_DFR0_EL1);
1837 
1838 	val &= ~ID_DFR0_EL1_PerfMon_MASK;
1839 	if (kvm_vcpu_has_pmu(vcpu))
1840 		val |= SYS_FIELD_PREP(ID_DFR0_EL1, PerfMon, perfmon);
1841 
1842 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_DFR0_EL1, CopDbg, Debugv8p8);
1843 
1844 	return val;
1845 }
1846 
1847 static int set_id_dfr0_el1(struct kvm_vcpu *vcpu,
1848 			   const struct sys_reg_desc *rd,
1849 			   u64 val)
1850 {
1851 	u8 perfmon = SYS_FIELD_GET(ID_DFR0_EL1, PerfMon, val);
1852 	u8 copdbg = SYS_FIELD_GET(ID_DFR0_EL1, CopDbg, val);
1853 
1854 	if (perfmon == ID_DFR0_EL1_PerfMon_IMPDEF) {
1855 		val &= ~ID_DFR0_EL1_PerfMon_MASK;
1856 		perfmon = 0;
1857 	}
1858 
1859 	/*
1860 	 * Allow DFR0_EL1.PerfMon to be set from userspace as long as
1861 	 * it doesn't promise more than what the HW gives us on the
1862 	 * AArch64 side (as everything is emulated with that), and
1863 	 * that this is a PMUv3.
1864 	 */
1865 	if (perfmon != 0 && perfmon < ID_DFR0_EL1_PerfMon_PMUv3)
1866 		return -EINVAL;
1867 
1868 	if (copdbg < ID_DFR0_EL1_CopDbg_Armv8)
1869 		return -EINVAL;
1870 
1871 	return set_id_reg(vcpu, rd, val);
1872 }
1873 
1874 /*
1875  * cpufeature ID register user accessors
1876  *
1877  * For now, these registers are immutable for userspace, so no values
1878  * are stored, and for set_id_reg() we don't allow the effective value
1879  * to be changed.
1880  */
1881 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1882 		      u64 *val)
1883 {
1884 	/*
1885 	 * Avoid locking if the VM has already started, as the ID registers are
1886 	 * guaranteed to be invariant at that point.
1887 	 */
1888 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
1889 		*val = read_id_reg(vcpu, rd);
1890 		return 0;
1891 	}
1892 
1893 	mutex_lock(&vcpu->kvm->arch.config_lock);
1894 	*val = read_id_reg(vcpu, rd);
1895 	mutex_unlock(&vcpu->kvm->arch.config_lock);
1896 
1897 	return 0;
1898 }
1899 
1900 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1901 		      u64 val)
1902 {
1903 	u32 id = reg_to_encoding(rd);
1904 	int ret;
1905 
1906 	mutex_lock(&vcpu->kvm->arch.config_lock);
1907 
1908 	/*
1909 	 * Once the VM has started the ID registers are immutable. Reject any
1910 	 * write that does not match the final register value.
1911 	 */
1912 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
1913 		if (val != read_id_reg(vcpu, rd))
1914 			ret = -EBUSY;
1915 		else
1916 			ret = 0;
1917 
1918 		mutex_unlock(&vcpu->kvm->arch.config_lock);
1919 		return ret;
1920 	}
1921 
1922 	ret = arm64_check_features(vcpu, rd, val);
1923 	if (!ret)
1924 		kvm_set_vm_id_reg(vcpu->kvm, id, val);
1925 
1926 	mutex_unlock(&vcpu->kvm->arch.config_lock);
1927 
1928 	/*
1929 	 * arm64_check_features() returns -E2BIG to indicate the register's
1930 	 * feature set is a superset of the maximally-allowed register value.
1931 	 * While it would be nice to precisely describe this to userspace, the
1932 	 * existing UAPI for KVM_SET_ONE_REG has it that invalid register
1933 	 * writes return -EINVAL.
1934 	 */
1935 	if (ret == -E2BIG)
1936 		ret = -EINVAL;
1937 	return ret;
1938 }
1939 
1940 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val)
1941 {
1942 	u64 *p = __vm_id_reg(&kvm->arch, reg);
1943 
1944 	lockdep_assert_held(&kvm->arch.config_lock);
1945 
1946 	if (KVM_BUG_ON(kvm_vm_has_ran_once(kvm) || !p, kvm))
1947 		return;
1948 
1949 	*p = val;
1950 }
1951 
1952 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1953 		       u64 *val)
1954 {
1955 	*val = 0;
1956 	return 0;
1957 }
1958 
1959 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1960 		      u64 val)
1961 {
1962 	return 0;
1963 }
1964 
1965 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1966 		       const struct sys_reg_desc *r)
1967 {
1968 	if (p->is_write)
1969 		return write_to_read_only(vcpu, p, r);
1970 
1971 	p->regval = kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0);
1972 	return true;
1973 }
1974 
1975 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1976 			 const struct sys_reg_desc *r)
1977 {
1978 	if (p->is_write)
1979 		return write_to_read_only(vcpu, p, r);
1980 
1981 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
1982 	return true;
1983 }
1984 
1985 /*
1986  * Fabricate a CLIDR_EL1 value instead of using the real value, which can vary
1987  * by the physical CPU which the vcpu currently resides in.
1988  */
1989 static u64 reset_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1990 {
1991 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
1992 	u64 clidr;
1993 	u8 loc;
1994 
1995 	if ((ctr_el0 & CTR_EL0_IDC)) {
1996 		/*
1997 		 * Data cache clean to the PoU is not required so LoUU and LoUIS
1998 		 * will not be set and a unified cache, which will be marked as
1999 		 * LoC, will be added.
2000 		 *
2001 		 * If not DIC, let the unified cache L2 so that an instruction
2002 		 * cache can be added as L1 later.
2003 		 */
2004 		loc = (ctr_el0 & CTR_EL0_DIC) ? 1 : 2;
2005 		clidr = CACHE_TYPE_UNIFIED << CLIDR_CTYPE_SHIFT(loc);
2006 	} else {
2007 		/*
2008 		 * Data cache clean to the PoU is required so let L1 have a data
2009 		 * cache and mark it as LoUU and LoUIS. As L1 has a data cache,
2010 		 * it can be marked as LoC too.
2011 		 */
2012 		loc = 1;
2013 		clidr = 1 << CLIDR_LOUU_SHIFT;
2014 		clidr |= 1 << CLIDR_LOUIS_SHIFT;
2015 		clidr |= CACHE_TYPE_DATA << CLIDR_CTYPE_SHIFT(1);
2016 	}
2017 
2018 	/*
2019 	 * Instruction cache invalidation to the PoU is required so let L1 have
2020 	 * an instruction cache. If L1 already has a data cache, it will be
2021 	 * CACHE_TYPE_SEPARATE.
2022 	 */
2023 	if (!(ctr_el0 & CTR_EL0_DIC))
2024 		clidr |= CACHE_TYPE_INST << CLIDR_CTYPE_SHIFT(1);
2025 
2026 	clidr |= loc << CLIDR_LOC_SHIFT;
2027 
2028 	/*
2029 	 * Add tag cache unified to data cache. Allocation tags and data are
2030 	 * unified in a cache line so that it looks valid even if there is only
2031 	 * one cache line.
2032 	 */
2033 	if (kvm_has_mte(vcpu->kvm))
2034 		clidr |= 2ULL << CLIDR_TTYPE_SHIFT(loc);
2035 
2036 	__vcpu_sys_reg(vcpu, r->reg) = clidr;
2037 
2038 	return __vcpu_sys_reg(vcpu, r->reg);
2039 }
2040 
2041 static int set_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2042 		      u64 val)
2043 {
2044 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2045 	u64 idc = !CLIDR_LOC(val) || (!CLIDR_LOUIS(val) && !CLIDR_LOUU(val));
2046 
2047 	if ((val & CLIDR_EL1_RES0) || (!(ctr_el0 & CTR_EL0_IDC) && idc))
2048 		return -EINVAL;
2049 
2050 	__vcpu_sys_reg(vcpu, rd->reg) = val;
2051 
2052 	return 0;
2053 }
2054 
2055 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2056 			  const struct sys_reg_desc *r)
2057 {
2058 	int reg = r->reg;
2059 
2060 	if (p->is_write)
2061 		vcpu_write_sys_reg(vcpu, p->regval, reg);
2062 	else
2063 		p->regval = vcpu_read_sys_reg(vcpu, reg);
2064 	return true;
2065 }
2066 
2067 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2068 			  const struct sys_reg_desc *r)
2069 {
2070 	u32 csselr;
2071 
2072 	if (p->is_write)
2073 		return write_to_read_only(vcpu, p, r);
2074 
2075 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
2076 	csselr &= CSSELR_EL1_Level | CSSELR_EL1_InD;
2077 	if (csselr < CSSELR_MAX)
2078 		p->regval = get_ccsidr(vcpu, csselr);
2079 
2080 	return true;
2081 }
2082 
2083 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
2084 				   const struct sys_reg_desc *rd)
2085 {
2086 	if (kvm_has_mte(vcpu->kvm))
2087 		return 0;
2088 
2089 	return REG_HIDDEN;
2090 }
2091 
2092 #define MTE_REG(name) {				\
2093 	SYS_DESC(SYS_##name),			\
2094 	.access = undef_access,			\
2095 	.reset = reset_unknown,			\
2096 	.reg = name,				\
2097 	.visibility = mte_visibility,		\
2098 }
2099 
2100 static unsigned int el2_visibility(const struct kvm_vcpu *vcpu,
2101 				   const struct sys_reg_desc *rd)
2102 {
2103 	if (vcpu_has_nv(vcpu))
2104 		return 0;
2105 
2106 	return REG_HIDDEN;
2107 }
2108 
2109 static bool bad_vncr_trap(struct kvm_vcpu *vcpu,
2110 			  struct sys_reg_params *p,
2111 			  const struct sys_reg_desc *r)
2112 {
2113 	/*
2114 	 * We really shouldn't be here, and this is likely the result
2115 	 * of a misconfigured trap, as this register should target the
2116 	 * VNCR page, and nothing else.
2117 	 */
2118 	return bad_trap(vcpu, p, r,
2119 			"trap of VNCR-backed register");
2120 }
2121 
2122 static bool bad_redir_trap(struct kvm_vcpu *vcpu,
2123 			   struct sys_reg_params *p,
2124 			   const struct sys_reg_desc *r)
2125 {
2126 	/*
2127 	 * We really shouldn't be here, and this is likely the result
2128 	 * of a misconfigured trap, as this register should target the
2129 	 * corresponding EL1, and nothing else.
2130 	 */
2131 	return bad_trap(vcpu, p, r,
2132 			"trap of EL2 register redirected to EL1");
2133 }
2134 
2135 #define EL2_REG(name, acc, rst, v) {		\
2136 	SYS_DESC(SYS_##name),			\
2137 	.access = acc,				\
2138 	.reset = rst,				\
2139 	.reg = name,				\
2140 	.visibility = el2_visibility,		\
2141 	.val = v,				\
2142 }
2143 
2144 #define EL2_REG_VNCR(name, rst, v)	EL2_REG(name, bad_vncr_trap, rst, v)
2145 #define EL2_REG_REDIR(name, rst, v)	EL2_REG(name, bad_redir_trap, rst, v)
2146 
2147 /*
2148  * Since reset() callback and field val are not used for idregs, they will be
2149  * used for specific purposes for idregs.
2150  * The reset() would return KVM sanitised register value. The value would be the
2151  * same as the host kernel sanitised value if there is no KVM sanitisation.
2152  * The val would be used as a mask indicating writable fields for the idreg.
2153  * Only bits with 1 are writable from userspace. This mask might not be
2154  * necessary in the future whenever all ID registers are enabled as writable
2155  * from userspace.
2156  */
2157 
2158 #define ID_DESC(name)				\
2159 	SYS_DESC(SYS_##name),			\
2160 	.access	= access_id_reg,		\
2161 	.get_user = get_id_reg			\
2162 
2163 /* sys_reg_desc initialiser for known cpufeature ID registers */
2164 #define ID_SANITISED(name) {			\
2165 	ID_DESC(name),				\
2166 	.set_user = set_id_reg,			\
2167 	.visibility = id_visibility,		\
2168 	.reset = kvm_read_sanitised_id_reg,	\
2169 	.val = 0,				\
2170 }
2171 
2172 /* sys_reg_desc initialiser for known cpufeature ID registers */
2173 #define AA32_ID_SANITISED(name) {		\
2174 	ID_DESC(name),				\
2175 	.set_user = set_id_reg,			\
2176 	.visibility = aa32_id_visibility,	\
2177 	.reset = kvm_read_sanitised_id_reg,	\
2178 	.val = 0,				\
2179 }
2180 
2181 /* sys_reg_desc initialiser for writable ID registers */
2182 #define ID_WRITABLE(name, mask) {		\
2183 	ID_DESC(name),				\
2184 	.set_user = set_id_reg,			\
2185 	.visibility = id_visibility,		\
2186 	.reset = kvm_read_sanitised_id_reg,	\
2187 	.val = mask,				\
2188 }
2189 
2190 /*
2191  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
2192  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
2193  * (1 <= crm < 8, 0 <= Op2 < 8).
2194  */
2195 #define ID_UNALLOCATED(crm, op2) {			\
2196 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
2197 	.access = access_id_reg,			\
2198 	.get_user = get_id_reg,				\
2199 	.set_user = set_id_reg,				\
2200 	.visibility = raz_visibility,			\
2201 	.reset = kvm_read_sanitised_id_reg,		\
2202 	.val = 0,					\
2203 }
2204 
2205 /*
2206  * sys_reg_desc initialiser for known ID registers that we hide from guests.
2207  * For now, these are exposed just like unallocated ID regs: they appear
2208  * RAZ for the guest.
2209  */
2210 #define ID_HIDDEN(name) {			\
2211 	ID_DESC(name),				\
2212 	.set_user = set_id_reg,			\
2213 	.visibility = raz_visibility,		\
2214 	.reset = kvm_read_sanitised_id_reg,	\
2215 	.val = 0,				\
2216 }
2217 
2218 static bool access_sp_el1(struct kvm_vcpu *vcpu,
2219 			  struct sys_reg_params *p,
2220 			  const struct sys_reg_desc *r)
2221 {
2222 	if (p->is_write)
2223 		__vcpu_sys_reg(vcpu, SP_EL1) = p->regval;
2224 	else
2225 		p->regval = __vcpu_sys_reg(vcpu, SP_EL1);
2226 
2227 	return true;
2228 }
2229 
2230 static bool access_elr(struct kvm_vcpu *vcpu,
2231 		       struct sys_reg_params *p,
2232 		       const struct sys_reg_desc *r)
2233 {
2234 	if (p->is_write)
2235 		vcpu_write_sys_reg(vcpu, p->regval, ELR_EL1);
2236 	else
2237 		p->regval = vcpu_read_sys_reg(vcpu, ELR_EL1);
2238 
2239 	return true;
2240 }
2241 
2242 static bool access_spsr(struct kvm_vcpu *vcpu,
2243 			struct sys_reg_params *p,
2244 			const struct sys_reg_desc *r)
2245 {
2246 	if (p->is_write)
2247 		__vcpu_sys_reg(vcpu, SPSR_EL1) = p->regval;
2248 	else
2249 		p->regval = __vcpu_sys_reg(vcpu, SPSR_EL1);
2250 
2251 	return true;
2252 }
2253 
2254 static bool access_cntkctl_el12(struct kvm_vcpu *vcpu,
2255 				struct sys_reg_params *p,
2256 				const struct sys_reg_desc *r)
2257 {
2258 	if (p->is_write)
2259 		__vcpu_sys_reg(vcpu, CNTKCTL_EL1) = p->regval;
2260 	else
2261 		p->regval = __vcpu_sys_reg(vcpu, CNTKCTL_EL1);
2262 
2263 	return true;
2264 }
2265 
2266 static u64 reset_hcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2267 {
2268 	u64 val = r->val;
2269 
2270 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
2271 		val |= HCR_E2H;
2272 
2273 	return __vcpu_sys_reg(vcpu, r->reg) = val;
2274 }
2275 
2276 static unsigned int sve_el2_visibility(const struct kvm_vcpu *vcpu,
2277 				       const struct sys_reg_desc *rd)
2278 {
2279 	unsigned int r;
2280 
2281 	r = el2_visibility(vcpu, rd);
2282 	if (r)
2283 		return r;
2284 
2285 	return sve_visibility(vcpu, rd);
2286 }
2287 
2288 static bool access_zcr_el2(struct kvm_vcpu *vcpu,
2289 			   struct sys_reg_params *p,
2290 			   const struct sys_reg_desc *r)
2291 {
2292 	unsigned int vq;
2293 
2294 	if (guest_hyp_sve_traps_enabled(vcpu)) {
2295 		kvm_inject_nested_sve_trap(vcpu);
2296 		return true;
2297 	}
2298 
2299 	if (!p->is_write) {
2300 		p->regval = vcpu_read_sys_reg(vcpu, ZCR_EL2);
2301 		return true;
2302 	}
2303 
2304 	vq = SYS_FIELD_GET(ZCR_ELx, LEN, p->regval) + 1;
2305 	vq = min(vq, vcpu_sve_max_vq(vcpu));
2306 	vcpu_write_sys_reg(vcpu, vq - 1, ZCR_EL2);
2307 	return true;
2308 }
2309 
2310 static unsigned int s1poe_visibility(const struct kvm_vcpu *vcpu,
2311 				     const struct sys_reg_desc *rd)
2312 {
2313 	if (kvm_has_feat(vcpu->kvm, ID_AA64MMFR3_EL1, S1POE, IMP))
2314 		return 0;
2315 
2316 	return REG_HIDDEN;
2317 }
2318 
2319 /*
2320  * Architected system registers.
2321  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
2322  *
2323  * Debug handling: We do trap most, if not all debug related system
2324  * registers. The implementation is good enough to ensure that a guest
2325  * can use these with minimal performance degradation. The drawback is
2326  * that we don't implement any of the external debug architecture.
2327  * This should be revisited if we ever encounter a more demanding
2328  * guest...
2329  */
2330 static const struct sys_reg_desc sys_reg_descs[] = {
2331 	DBG_BCR_BVR_WCR_WVR_EL1(0),
2332 	DBG_BCR_BVR_WCR_WVR_EL1(1),
2333 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
2334 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
2335 	DBG_BCR_BVR_WCR_WVR_EL1(2),
2336 	DBG_BCR_BVR_WCR_WVR_EL1(3),
2337 	DBG_BCR_BVR_WCR_WVR_EL1(4),
2338 	DBG_BCR_BVR_WCR_WVR_EL1(5),
2339 	DBG_BCR_BVR_WCR_WVR_EL1(6),
2340 	DBG_BCR_BVR_WCR_WVR_EL1(7),
2341 	DBG_BCR_BVR_WCR_WVR_EL1(8),
2342 	DBG_BCR_BVR_WCR_WVR_EL1(9),
2343 	DBG_BCR_BVR_WCR_WVR_EL1(10),
2344 	DBG_BCR_BVR_WCR_WVR_EL1(11),
2345 	DBG_BCR_BVR_WCR_WVR_EL1(12),
2346 	DBG_BCR_BVR_WCR_WVR_EL1(13),
2347 	DBG_BCR_BVR_WCR_WVR_EL1(14),
2348 	DBG_BCR_BVR_WCR_WVR_EL1(15),
2349 
2350 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
2351 	{ SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
2352 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
2353 		OSLSR_EL1_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
2354 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
2355 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
2356 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
2357 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
2358 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
2359 
2360 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
2361 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
2362 	// DBGDTR[TR]X_EL0 share the same encoding
2363 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
2364 
2365 	{ SYS_DESC(SYS_DBGVCR32_EL2), undef_access, reset_val, DBGVCR32_EL2, 0 },
2366 
2367 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
2368 
2369 	/*
2370 	 * ID regs: all ID_SANITISED() entries here must have corresponding
2371 	 * entries in arm64_ftr_regs[].
2372 	 */
2373 
2374 	/* AArch64 mappings of the AArch32 ID registers */
2375 	/* CRm=1 */
2376 	AA32_ID_SANITISED(ID_PFR0_EL1),
2377 	AA32_ID_SANITISED(ID_PFR1_EL1),
2378 	{ SYS_DESC(SYS_ID_DFR0_EL1),
2379 	  .access = access_id_reg,
2380 	  .get_user = get_id_reg,
2381 	  .set_user = set_id_dfr0_el1,
2382 	  .visibility = aa32_id_visibility,
2383 	  .reset = read_sanitised_id_dfr0_el1,
2384 	  .val = ID_DFR0_EL1_PerfMon_MASK |
2385 		 ID_DFR0_EL1_CopDbg_MASK, },
2386 	ID_HIDDEN(ID_AFR0_EL1),
2387 	AA32_ID_SANITISED(ID_MMFR0_EL1),
2388 	AA32_ID_SANITISED(ID_MMFR1_EL1),
2389 	AA32_ID_SANITISED(ID_MMFR2_EL1),
2390 	AA32_ID_SANITISED(ID_MMFR3_EL1),
2391 
2392 	/* CRm=2 */
2393 	AA32_ID_SANITISED(ID_ISAR0_EL1),
2394 	AA32_ID_SANITISED(ID_ISAR1_EL1),
2395 	AA32_ID_SANITISED(ID_ISAR2_EL1),
2396 	AA32_ID_SANITISED(ID_ISAR3_EL1),
2397 	AA32_ID_SANITISED(ID_ISAR4_EL1),
2398 	AA32_ID_SANITISED(ID_ISAR5_EL1),
2399 	AA32_ID_SANITISED(ID_MMFR4_EL1),
2400 	AA32_ID_SANITISED(ID_ISAR6_EL1),
2401 
2402 	/* CRm=3 */
2403 	AA32_ID_SANITISED(MVFR0_EL1),
2404 	AA32_ID_SANITISED(MVFR1_EL1),
2405 	AA32_ID_SANITISED(MVFR2_EL1),
2406 	ID_UNALLOCATED(3,3),
2407 	AA32_ID_SANITISED(ID_PFR2_EL1),
2408 	ID_HIDDEN(ID_DFR1_EL1),
2409 	AA32_ID_SANITISED(ID_MMFR5_EL1),
2410 	ID_UNALLOCATED(3,7),
2411 
2412 	/* AArch64 ID registers */
2413 	/* CRm=4 */
2414 	{ SYS_DESC(SYS_ID_AA64PFR0_EL1),
2415 	  .access = access_id_reg,
2416 	  .get_user = get_id_reg,
2417 	  .set_user = set_id_reg,
2418 	  .reset = read_sanitised_id_aa64pfr0_el1,
2419 	  .val = ~(ID_AA64PFR0_EL1_AMU |
2420 		   ID_AA64PFR0_EL1_MPAM |
2421 		   ID_AA64PFR0_EL1_SVE |
2422 		   ID_AA64PFR0_EL1_RAS |
2423 		   ID_AA64PFR0_EL1_AdvSIMD |
2424 		   ID_AA64PFR0_EL1_FP), },
2425 	ID_WRITABLE(ID_AA64PFR1_EL1, ~(ID_AA64PFR1_EL1_PFAR |
2426 				       ID_AA64PFR1_EL1_DF2 |
2427 				       ID_AA64PFR1_EL1_MTEX |
2428 				       ID_AA64PFR1_EL1_THE |
2429 				       ID_AA64PFR1_EL1_GCS |
2430 				       ID_AA64PFR1_EL1_MTE_frac |
2431 				       ID_AA64PFR1_EL1_NMI |
2432 				       ID_AA64PFR1_EL1_RNDR_trap |
2433 				       ID_AA64PFR1_EL1_SME |
2434 				       ID_AA64PFR1_EL1_RES0 |
2435 				       ID_AA64PFR1_EL1_MPAM_frac |
2436 				       ID_AA64PFR1_EL1_RAS_frac |
2437 				       ID_AA64PFR1_EL1_MTE)),
2438 	ID_WRITABLE(ID_AA64PFR2_EL1, ID_AA64PFR2_EL1_FPMR),
2439 	ID_UNALLOCATED(4,3),
2440 	ID_WRITABLE(ID_AA64ZFR0_EL1, ~ID_AA64ZFR0_EL1_RES0),
2441 	ID_HIDDEN(ID_AA64SMFR0_EL1),
2442 	ID_UNALLOCATED(4,6),
2443 	ID_WRITABLE(ID_AA64FPFR0_EL1, ~ID_AA64FPFR0_EL1_RES0),
2444 
2445 	/* CRm=5 */
2446 	{ SYS_DESC(SYS_ID_AA64DFR0_EL1),
2447 	  .access = access_id_reg,
2448 	  .get_user = get_id_reg,
2449 	  .set_user = set_id_aa64dfr0_el1,
2450 	  .reset = read_sanitised_id_aa64dfr0_el1,
2451 	/*
2452 	 * Prior to FEAT_Debugv8.9, the architecture defines context-aware
2453 	 * breakpoints (CTX_CMPs) as the highest numbered breakpoints (BRPs).
2454 	 * KVM does not trap + emulate the breakpoint registers, and as such
2455 	 * cannot support a layout that misaligns with the underlying hardware.
2456 	 * While it may be possible to describe a subset that aligns with
2457 	 * hardware, just prevent changes to BRPs and CTX_CMPs altogether for
2458 	 * simplicity.
2459 	 *
2460 	 * See DDI0487K.a, section D2.8.3 Breakpoint types and linking
2461 	 * of breakpoints for more details.
2462 	 */
2463 	  .val = ID_AA64DFR0_EL1_DoubleLock_MASK |
2464 		 ID_AA64DFR0_EL1_WRPs_MASK |
2465 		 ID_AA64DFR0_EL1_PMUVer_MASK |
2466 		 ID_AA64DFR0_EL1_DebugVer_MASK, },
2467 	ID_SANITISED(ID_AA64DFR1_EL1),
2468 	ID_UNALLOCATED(5,2),
2469 	ID_UNALLOCATED(5,3),
2470 	ID_HIDDEN(ID_AA64AFR0_EL1),
2471 	ID_HIDDEN(ID_AA64AFR1_EL1),
2472 	ID_UNALLOCATED(5,6),
2473 	ID_UNALLOCATED(5,7),
2474 
2475 	/* CRm=6 */
2476 	ID_WRITABLE(ID_AA64ISAR0_EL1, ~ID_AA64ISAR0_EL1_RES0),
2477 	ID_WRITABLE(ID_AA64ISAR1_EL1, ~(ID_AA64ISAR1_EL1_GPI |
2478 					ID_AA64ISAR1_EL1_GPA |
2479 					ID_AA64ISAR1_EL1_API |
2480 					ID_AA64ISAR1_EL1_APA)),
2481 	ID_WRITABLE(ID_AA64ISAR2_EL1, ~(ID_AA64ISAR2_EL1_RES0 |
2482 					ID_AA64ISAR2_EL1_APA3 |
2483 					ID_AA64ISAR2_EL1_GPA3)),
2484 	ID_UNALLOCATED(6,3),
2485 	ID_UNALLOCATED(6,4),
2486 	ID_UNALLOCATED(6,5),
2487 	ID_UNALLOCATED(6,6),
2488 	ID_UNALLOCATED(6,7),
2489 
2490 	/* CRm=7 */
2491 	ID_WRITABLE(ID_AA64MMFR0_EL1, ~(ID_AA64MMFR0_EL1_RES0 |
2492 					ID_AA64MMFR0_EL1_TGRAN4_2 |
2493 					ID_AA64MMFR0_EL1_TGRAN64_2 |
2494 					ID_AA64MMFR0_EL1_TGRAN16_2)),
2495 	ID_WRITABLE(ID_AA64MMFR1_EL1, ~(ID_AA64MMFR1_EL1_RES0 |
2496 					ID_AA64MMFR1_EL1_HCX |
2497 					ID_AA64MMFR1_EL1_TWED |
2498 					ID_AA64MMFR1_EL1_XNX |
2499 					ID_AA64MMFR1_EL1_VH |
2500 					ID_AA64MMFR1_EL1_VMIDBits)),
2501 	ID_WRITABLE(ID_AA64MMFR2_EL1, ~(ID_AA64MMFR2_EL1_RES0 |
2502 					ID_AA64MMFR2_EL1_EVT |
2503 					ID_AA64MMFR2_EL1_FWB |
2504 					ID_AA64MMFR2_EL1_IDS |
2505 					ID_AA64MMFR2_EL1_NV |
2506 					ID_AA64MMFR2_EL1_CCIDX)),
2507 	ID_WRITABLE(ID_AA64MMFR3_EL1, (ID_AA64MMFR3_EL1_TCRX	|
2508 				       ID_AA64MMFR3_EL1_S1PIE   |
2509 				       ID_AA64MMFR3_EL1_S1POE)),
2510 	ID_SANITISED(ID_AA64MMFR4_EL1),
2511 	ID_UNALLOCATED(7,5),
2512 	ID_UNALLOCATED(7,6),
2513 	ID_UNALLOCATED(7,7),
2514 
2515 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
2516 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
2517 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
2518 
2519 	MTE_REG(RGSR_EL1),
2520 	MTE_REG(GCR_EL1),
2521 
2522 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
2523 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
2524 	{ SYS_DESC(SYS_SMPRI_EL1), undef_access },
2525 	{ SYS_DESC(SYS_SMCR_EL1), undef_access },
2526 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
2527 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
2528 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
2529 	{ SYS_DESC(SYS_TCR2_EL1), access_vm_reg, reset_val, TCR2_EL1, 0 },
2530 
2531 	PTRAUTH_KEY(APIA),
2532 	PTRAUTH_KEY(APIB),
2533 	PTRAUTH_KEY(APDA),
2534 	PTRAUTH_KEY(APDB),
2535 	PTRAUTH_KEY(APGA),
2536 
2537 	{ SYS_DESC(SYS_SPSR_EL1), access_spsr},
2538 	{ SYS_DESC(SYS_ELR_EL1), access_elr},
2539 
2540 	{ SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
2541 
2542 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
2543 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
2544 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
2545 
2546 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
2547 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
2548 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
2549 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
2550 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
2551 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
2552 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
2553 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
2554 
2555 	MTE_REG(TFSR_EL1),
2556 	MTE_REG(TFSRE0_EL1),
2557 
2558 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
2559 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
2560 
2561 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
2562 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
2563 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
2564 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
2565 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
2566 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
2567 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
2568 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
2569 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
2570 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
2571 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
2572 	/* PMBIDR_EL1 is not trapped */
2573 
2574 	{ PMU_SYS_REG(PMINTENSET_EL1),
2575 	  .access = access_pminten, .reg = PMINTENSET_EL1,
2576 	  .get_user = get_pmreg, .set_user = set_pmreg },
2577 	{ PMU_SYS_REG(PMINTENCLR_EL1),
2578 	  .access = access_pminten, .reg = PMINTENSET_EL1,
2579 	  .get_user = get_pmreg, .set_user = set_pmreg },
2580 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
2581 
2582 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
2583 	{ SYS_DESC(SYS_PIRE0_EL1), NULL, reset_unknown, PIRE0_EL1 },
2584 	{ SYS_DESC(SYS_PIR_EL1), NULL, reset_unknown, PIR_EL1 },
2585 	{ SYS_DESC(SYS_POR_EL1), NULL, reset_unknown, POR_EL1,
2586 	  .visibility = s1poe_visibility },
2587 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
2588 
2589 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
2590 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
2591 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
2592 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
2593 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
2594 
2595 	{ SYS_DESC(SYS_VBAR_EL1), access_rw, reset_val, VBAR_EL1, 0 },
2596 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
2597 
2598 	{ SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
2599 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
2600 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
2601 	{ SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
2602 	{ SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
2603 	{ SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
2604 	{ SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
2605 	{ SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
2606 	{ SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
2607 	{ SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
2608 	{ SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
2609 	{ SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
2610 	{ SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
2611 	{ SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
2612 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
2613 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
2614 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
2615 	{ SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
2616 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
2617 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
2618 	{ SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
2619 	{ SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
2620 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
2621 	{ SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
2622 	{ SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
2623 
2624 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
2625 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
2626 
2627 	{ SYS_DESC(SYS_ACCDATA_EL1), undef_access },
2628 
2629 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
2630 
2631 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
2632 
2633 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
2634 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr, reset_clidr, CLIDR_EL1,
2635 	  .set_user = set_clidr, .val = ~CLIDR_EL1_RES0 },
2636 	{ SYS_DESC(SYS_CCSIDR2_EL1), undef_access },
2637 	{ SYS_DESC(SYS_SMIDR_EL1), undef_access },
2638 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
2639 	ID_WRITABLE(CTR_EL0, CTR_EL0_DIC_MASK |
2640 			     CTR_EL0_IDC_MASK |
2641 			     CTR_EL0_DminLine_MASK |
2642 			     CTR_EL0_IminLine_MASK),
2643 	{ SYS_DESC(SYS_SVCR), undef_access, reset_val, SVCR, 0, .visibility = sme_visibility  },
2644 	{ SYS_DESC(SYS_FPMR), undef_access, reset_val, FPMR, 0, .visibility = fp8_visibility },
2645 
2646 	{ PMU_SYS_REG(PMCR_EL0), .access = access_pmcr, .reset = reset_pmcr,
2647 	  .reg = PMCR_EL0, .get_user = get_pmcr, .set_user = set_pmcr },
2648 	{ PMU_SYS_REG(PMCNTENSET_EL0),
2649 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
2650 	  .get_user = get_pmreg, .set_user = set_pmreg },
2651 	{ PMU_SYS_REG(PMCNTENCLR_EL0),
2652 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
2653 	  .get_user = get_pmreg, .set_user = set_pmreg },
2654 	{ PMU_SYS_REG(PMOVSCLR_EL0),
2655 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
2656 	  .get_user = get_pmreg, .set_user = set_pmreg },
2657 	/*
2658 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
2659 	 * previously (and pointlessly) advertised in the past...
2660 	 */
2661 	{ PMU_SYS_REG(PMSWINC_EL0),
2662 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
2663 	  .access = access_pmswinc, .reset = NULL },
2664 	{ PMU_SYS_REG(PMSELR_EL0),
2665 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
2666 	{ PMU_SYS_REG(PMCEID0_EL0),
2667 	  .access = access_pmceid, .reset = NULL },
2668 	{ PMU_SYS_REG(PMCEID1_EL0),
2669 	  .access = access_pmceid, .reset = NULL },
2670 	{ PMU_SYS_REG(PMCCNTR_EL0),
2671 	  .access = access_pmu_evcntr, .reset = reset_unknown,
2672 	  .reg = PMCCNTR_EL0, .get_user = get_pmu_evcntr},
2673 	{ PMU_SYS_REG(PMXEVTYPER_EL0),
2674 	  .access = access_pmu_evtyper, .reset = NULL },
2675 	{ PMU_SYS_REG(PMXEVCNTR_EL0),
2676 	  .access = access_pmu_evcntr, .reset = NULL },
2677 	/*
2678 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
2679 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
2680 	 */
2681 	{ PMU_SYS_REG(PMUSERENR_EL0), .access = access_pmuserenr,
2682 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
2683 	{ PMU_SYS_REG(PMOVSSET_EL0),
2684 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
2685 	  .get_user = get_pmreg, .set_user = set_pmreg },
2686 
2687 	{ SYS_DESC(SYS_POR_EL0), NULL, reset_unknown, POR_EL0,
2688 	  .visibility = s1poe_visibility },
2689 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
2690 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
2691 	{ SYS_DESC(SYS_TPIDR2_EL0), undef_access },
2692 
2693 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
2694 
2695 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
2696 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
2697 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
2698 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
2699 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
2700 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
2701 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
2702 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
2703 	AMU_AMEVCNTR0_EL0(0),
2704 	AMU_AMEVCNTR0_EL0(1),
2705 	AMU_AMEVCNTR0_EL0(2),
2706 	AMU_AMEVCNTR0_EL0(3),
2707 	AMU_AMEVCNTR0_EL0(4),
2708 	AMU_AMEVCNTR0_EL0(5),
2709 	AMU_AMEVCNTR0_EL0(6),
2710 	AMU_AMEVCNTR0_EL0(7),
2711 	AMU_AMEVCNTR0_EL0(8),
2712 	AMU_AMEVCNTR0_EL0(9),
2713 	AMU_AMEVCNTR0_EL0(10),
2714 	AMU_AMEVCNTR0_EL0(11),
2715 	AMU_AMEVCNTR0_EL0(12),
2716 	AMU_AMEVCNTR0_EL0(13),
2717 	AMU_AMEVCNTR0_EL0(14),
2718 	AMU_AMEVCNTR0_EL0(15),
2719 	AMU_AMEVTYPER0_EL0(0),
2720 	AMU_AMEVTYPER0_EL0(1),
2721 	AMU_AMEVTYPER0_EL0(2),
2722 	AMU_AMEVTYPER0_EL0(3),
2723 	AMU_AMEVTYPER0_EL0(4),
2724 	AMU_AMEVTYPER0_EL0(5),
2725 	AMU_AMEVTYPER0_EL0(6),
2726 	AMU_AMEVTYPER0_EL0(7),
2727 	AMU_AMEVTYPER0_EL0(8),
2728 	AMU_AMEVTYPER0_EL0(9),
2729 	AMU_AMEVTYPER0_EL0(10),
2730 	AMU_AMEVTYPER0_EL0(11),
2731 	AMU_AMEVTYPER0_EL0(12),
2732 	AMU_AMEVTYPER0_EL0(13),
2733 	AMU_AMEVTYPER0_EL0(14),
2734 	AMU_AMEVTYPER0_EL0(15),
2735 	AMU_AMEVCNTR1_EL0(0),
2736 	AMU_AMEVCNTR1_EL0(1),
2737 	AMU_AMEVCNTR1_EL0(2),
2738 	AMU_AMEVCNTR1_EL0(3),
2739 	AMU_AMEVCNTR1_EL0(4),
2740 	AMU_AMEVCNTR1_EL0(5),
2741 	AMU_AMEVCNTR1_EL0(6),
2742 	AMU_AMEVCNTR1_EL0(7),
2743 	AMU_AMEVCNTR1_EL0(8),
2744 	AMU_AMEVCNTR1_EL0(9),
2745 	AMU_AMEVCNTR1_EL0(10),
2746 	AMU_AMEVCNTR1_EL0(11),
2747 	AMU_AMEVCNTR1_EL0(12),
2748 	AMU_AMEVCNTR1_EL0(13),
2749 	AMU_AMEVCNTR1_EL0(14),
2750 	AMU_AMEVCNTR1_EL0(15),
2751 	AMU_AMEVTYPER1_EL0(0),
2752 	AMU_AMEVTYPER1_EL0(1),
2753 	AMU_AMEVTYPER1_EL0(2),
2754 	AMU_AMEVTYPER1_EL0(3),
2755 	AMU_AMEVTYPER1_EL0(4),
2756 	AMU_AMEVTYPER1_EL0(5),
2757 	AMU_AMEVTYPER1_EL0(6),
2758 	AMU_AMEVTYPER1_EL0(7),
2759 	AMU_AMEVTYPER1_EL0(8),
2760 	AMU_AMEVTYPER1_EL0(9),
2761 	AMU_AMEVTYPER1_EL0(10),
2762 	AMU_AMEVTYPER1_EL0(11),
2763 	AMU_AMEVTYPER1_EL0(12),
2764 	AMU_AMEVTYPER1_EL0(13),
2765 	AMU_AMEVTYPER1_EL0(14),
2766 	AMU_AMEVTYPER1_EL0(15),
2767 
2768 	{ SYS_DESC(SYS_CNTPCT_EL0), access_arch_timer },
2769 	{ SYS_DESC(SYS_CNTPCTSS_EL0), access_arch_timer },
2770 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
2771 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
2772 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
2773 
2774 	/* PMEVCNTRn_EL0 */
2775 	PMU_PMEVCNTR_EL0(0),
2776 	PMU_PMEVCNTR_EL0(1),
2777 	PMU_PMEVCNTR_EL0(2),
2778 	PMU_PMEVCNTR_EL0(3),
2779 	PMU_PMEVCNTR_EL0(4),
2780 	PMU_PMEVCNTR_EL0(5),
2781 	PMU_PMEVCNTR_EL0(6),
2782 	PMU_PMEVCNTR_EL0(7),
2783 	PMU_PMEVCNTR_EL0(8),
2784 	PMU_PMEVCNTR_EL0(9),
2785 	PMU_PMEVCNTR_EL0(10),
2786 	PMU_PMEVCNTR_EL0(11),
2787 	PMU_PMEVCNTR_EL0(12),
2788 	PMU_PMEVCNTR_EL0(13),
2789 	PMU_PMEVCNTR_EL0(14),
2790 	PMU_PMEVCNTR_EL0(15),
2791 	PMU_PMEVCNTR_EL0(16),
2792 	PMU_PMEVCNTR_EL0(17),
2793 	PMU_PMEVCNTR_EL0(18),
2794 	PMU_PMEVCNTR_EL0(19),
2795 	PMU_PMEVCNTR_EL0(20),
2796 	PMU_PMEVCNTR_EL0(21),
2797 	PMU_PMEVCNTR_EL0(22),
2798 	PMU_PMEVCNTR_EL0(23),
2799 	PMU_PMEVCNTR_EL0(24),
2800 	PMU_PMEVCNTR_EL0(25),
2801 	PMU_PMEVCNTR_EL0(26),
2802 	PMU_PMEVCNTR_EL0(27),
2803 	PMU_PMEVCNTR_EL0(28),
2804 	PMU_PMEVCNTR_EL0(29),
2805 	PMU_PMEVCNTR_EL0(30),
2806 	/* PMEVTYPERn_EL0 */
2807 	PMU_PMEVTYPER_EL0(0),
2808 	PMU_PMEVTYPER_EL0(1),
2809 	PMU_PMEVTYPER_EL0(2),
2810 	PMU_PMEVTYPER_EL0(3),
2811 	PMU_PMEVTYPER_EL0(4),
2812 	PMU_PMEVTYPER_EL0(5),
2813 	PMU_PMEVTYPER_EL0(6),
2814 	PMU_PMEVTYPER_EL0(7),
2815 	PMU_PMEVTYPER_EL0(8),
2816 	PMU_PMEVTYPER_EL0(9),
2817 	PMU_PMEVTYPER_EL0(10),
2818 	PMU_PMEVTYPER_EL0(11),
2819 	PMU_PMEVTYPER_EL0(12),
2820 	PMU_PMEVTYPER_EL0(13),
2821 	PMU_PMEVTYPER_EL0(14),
2822 	PMU_PMEVTYPER_EL0(15),
2823 	PMU_PMEVTYPER_EL0(16),
2824 	PMU_PMEVTYPER_EL0(17),
2825 	PMU_PMEVTYPER_EL0(18),
2826 	PMU_PMEVTYPER_EL0(19),
2827 	PMU_PMEVTYPER_EL0(20),
2828 	PMU_PMEVTYPER_EL0(21),
2829 	PMU_PMEVTYPER_EL0(22),
2830 	PMU_PMEVTYPER_EL0(23),
2831 	PMU_PMEVTYPER_EL0(24),
2832 	PMU_PMEVTYPER_EL0(25),
2833 	PMU_PMEVTYPER_EL0(26),
2834 	PMU_PMEVTYPER_EL0(27),
2835 	PMU_PMEVTYPER_EL0(28),
2836 	PMU_PMEVTYPER_EL0(29),
2837 	PMU_PMEVTYPER_EL0(30),
2838 	/*
2839 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
2840 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
2841 	 */
2842 	{ PMU_SYS_REG(PMCCFILTR_EL0), .access = access_pmu_evtyper,
2843 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
2844 
2845 	EL2_REG_VNCR(VPIDR_EL2, reset_unknown, 0),
2846 	EL2_REG_VNCR(VMPIDR_EL2, reset_unknown, 0),
2847 	EL2_REG(SCTLR_EL2, access_rw, reset_val, SCTLR_EL2_RES1),
2848 	EL2_REG(ACTLR_EL2, access_rw, reset_val, 0),
2849 	EL2_REG_VNCR(HCR_EL2, reset_hcr, 0),
2850 	EL2_REG(MDCR_EL2, access_rw, reset_val, 0),
2851 	EL2_REG(CPTR_EL2, access_rw, reset_val, CPTR_NVHE_EL2_RES1),
2852 	EL2_REG_VNCR(HSTR_EL2, reset_val, 0),
2853 	EL2_REG_VNCR(HFGRTR_EL2, reset_val, 0),
2854 	EL2_REG_VNCR(HFGWTR_EL2, reset_val, 0),
2855 	EL2_REG_VNCR(HFGITR_EL2, reset_val, 0),
2856 	EL2_REG_VNCR(HACR_EL2, reset_val, 0),
2857 
2858 	{ SYS_DESC(SYS_ZCR_EL2), .access = access_zcr_el2, .reset = reset_val,
2859 	  .visibility = sve_el2_visibility, .reg = ZCR_EL2 },
2860 
2861 	EL2_REG_VNCR(HCRX_EL2, reset_val, 0),
2862 
2863 	EL2_REG(TTBR0_EL2, access_rw, reset_val, 0),
2864 	EL2_REG(TTBR1_EL2, access_rw, reset_val, 0),
2865 	EL2_REG(TCR_EL2, access_rw, reset_val, TCR_EL2_RES1),
2866 	EL2_REG_VNCR(VTTBR_EL2, reset_val, 0),
2867 	EL2_REG_VNCR(VTCR_EL2, reset_val, 0),
2868 
2869 	{ SYS_DESC(SYS_DACR32_EL2), undef_access, reset_unknown, DACR32_EL2 },
2870 	EL2_REG_VNCR(HDFGRTR_EL2, reset_val, 0),
2871 	EL2_REG_VNCR(HDFGWTR_EL2, reset_val, 0),
2872 	EL2_REG_VNCR(HAFGRTR_EL2, reset_val, 0),
2873 	EL2_REG_REDIR(SPSR_EL2, reset_val, 0),
2874 	EL2_REG_REDIR(ELR_EL2, reset_val, 0),
2875 	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},
2876 
2877 	/* AArch32 SPSR_* are RES0 if trapped from a NV guest */
2878 	{ SYS_DESC(SYS_SPSR_irq), .access = trap_raz_wi },
2879 	{ SYS_DESC(SYS_SPSR_abt), .access = trap_raz_wi },
2880 	{ SYS_DESC(SYS_SPSR_und), .access = trap_raz_wi },
2881 	{ SYS_DESC(SYS_SPSR_fiq), .access = trap_raz_wi },
2882 
2883 	{ SYS_DESC(SYS_IFSR32_EL2), undef_access, reset_unknown, IFSR32_EL2 },
2884 	EL2_REG(AFSR0_EL2, access_rw, reset_val, 0),
2885 	EL2_REG(AFSR1_EL2, access_rw, reset_val, 0),
2886 	EL2_REG_REDIR(ESR_EL2, reset_val, 0),
2887 	{ SYS_DESC(SYS_FPEXC32_EL2), undef_access, reset_val, FPEXC32_EL2, 0x700 },
2888 
2889 	EL2_REG_REDIR(FAR_EL2, reset_val, 0),
2890 	EL2_REG(HPFAR_EL2, access_rw, reset_val, 0),
2891 
2892 	EL2_REG(MAIR_EL2, access_rw, reset_val, 0),
2893 	EL2_REG(AMAIR_EL2, access_rw, reset_val, 0),
2894 
2895 	EL2_REG(VBAR_EL2, access_rw, reset_val, 0),
2896 	EL2_REG(RVBAR_EL2, access_rw, reset_val, 0),
2897 	{ SYS_DESC(SYS_RMR_EL2), undef_access },
2898 
2899 	EL2_REG_VNCR(ICH_HCR_EL2, reset_val, 0),
2900 
2901 	EL2_REG(CONTEXTIDR_EL2, access_rw, reset_val, 0),
2902 	EL2_REG(TPIDR_EL2, access_rw, reset_val, 0),
2903 
2904 	EL2_REG_VNCR(CNTVOFF_EL2, reset_val, 0),
2905 	EL2_REG(CNTHCTL_EL2, access_rw, reset_val, 0),
2906 
2907 	{ SYS_DESC(SYS_CNTKCTL_EL12), access_cntkctl_el12 },
2908 
2909 	EL2_REG(SP_EL2, NULL, reset_unknown, 0),
2910 };
2911 
2912 static bool handle_at_s1e01(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2913 			    const struct sys_reg_desc *r)
2914 {
2915 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2916 
2917 	__kvm_at_s1e01(vcpu, op, p->regval);
2918 
2919 	return true;
2920 }
2921 
2922 static bool handle_at_s1e2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2923 			   const struct sys_reg_desc *r)
2924 {
2925 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2926 
2927 	/* There is no FGT associated with AT S1E2A :-( */
2928 	if (op == OP_AT_S1E2A &&
2929 	    !kvm_has_feat(vcpu->kvm, ID_AA64ISAR2_EL1, ATS1A, IMP)) {
2930 		kvm_inject_undefined(vcpu);
2931 		return false;
2932 	}
2933 
2934 	__kvm_at_s1e2(vcpu, op, p->regval);
2935 
2936 	return true;
2937 }
2938 
2939 static bool handle_at_s12(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2940 			  const struct sys_reg_desc *r)
2941 {
2942 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2943 
2944 	__kvm_at_s12(vcpu, op, p->regval);
2945 
2946 	return true;
2947 }
2948 
2949 static bool kvm_supported_tlbi_s12_op(struct kvm_vcpu *vpcu, u32 instr)
2950 {
2951 	struct kvm *kvm = vpcu->kvm;
2952 	u8 CRm = sys_reg_CRm(instr);
2953 
2954 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
2955 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
2956 		return false;
2957 
2958 	if (CRm == TLBI_CRm_nROS &&
2959 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
2960 		return false;
2961 
2962 	return true;
2963 }
2964 
2965 static bool handle_alle1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2966 			   const struct sys_reg_desc *r)
2967 {
2968 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2969 
2970 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
2971 		return undef_access(vcpu, p, r);
2972 
2973 	write_lock(&vcpu->kvm->mmu_lock);
2974 
2975 	/*
2976 	 * Drop all shadow S2s, resulting in S1/S2 TLBIs for each of the
2977 	 * corresponding VMIDs.
2978 	 */
2979 	kvm_nested_s2_unmap(vcpu->kvm, true);
2980 
2981 	write_unlock(&vcpu->kvm->mmu_lock);
2982 
2983 	return true;
2984 }
2985 
2986 static bool kvm_supported_tlbi_ipas2_op(struct kvm_vcpu *vpcu, u32 instr)
2987 {
2988 	struct kvm *kvm = vpcu->kvm;
2989 	u8 CRm = sys_reg_CRm(instr);
2990 	u8 Op2 = sys_reg_Op2(instr);
2991 
2992 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
2993 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
2994 		return false;
2995 
2996 	if (CRm == TLBI_CRm_IPAIS && (Op2 == 2 || Op2 == 6) &&
2997 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
2998 		return false;
2999 
3000 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 0 || Op2 == 4) &&
3001 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3002 		return false;
3003 
3004 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 3 || Op2 == 7) &&
3005 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3006 		return false;
3007 
3008 	return true;
3009 }
3010 
3011 /* Only defined here as this is an internal "abstraction" */
3012 union tlbi_info {
3013 	struct {
3014 		u64	start;
3015 		u64	size;
3016 	} range;
3017 
3018 	struct {
3019 		u64	addr;
3020 	} ipa;
3021 
3022 	struct {
3023 		u64	addr;
3024 		u32	encoding;
3025 	} va;
3026 };
3027 
3028 static void s2_mmu_unmap_range(struct kvm_s2_mmu *mmu,
3029 			       const union tlbi_info *info)
3030 {
3031 	/*
3032 	 * The unmap operation is allowed to drop the MMU lock and block, which
3033 	 * means that @mmu could be used for a different context than the one
3034 	 * currently being invalidated.
3035 	 *
3036 	 * This behavior is still safe, as:
3037 	 *
3038 	 *  1) The vCPU(s) that recycled the MMU are responsible for invalidating
3039 	 *     the entire MMU before reusing it, which still honors the intent
3040 	 *     of a TLBI.
3041 	 *
3042 	 *  2) Until the guest TLBI instruction is 'retired' (i.e. increment PC
3043 	 *     and ERET to the guest), other vCPUs are allowed to use stale
3044 	 *     translations.
3045 	 *
3046 	 *  3) Accidentally unmapping an unrelated MMU context is nonfatal, and
3047 	 *     at worst may cause more aborts for shadow stage-2 fills.
3048 	 *
3049 	 * Dropping the MMU lock also implies that shadow stage-2 fills could
3050 	 * happen behind the back of the TLBI. This is still safe, though, as
3051 	 * the L1 needs to put its stage-2 in a consistent state before doing
3052 	 * the TLBI.
3053 	 */
3054 	kvm_stage2_unmap_range(mmu, info->range.start, info->range.size, true);
3055 }
3056 
3057 static bool handle_vmalls12e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3058 				const struct sys_reg_desc *r)
3059 {
3060 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3061 	u64 limit, vttbr;
3062 
3063 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3064 		return undef_access(vcpu, p, r);
3065 
3066 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3067 	limit = BIT_ULL(kvm_get_pa_bits(vcpu->kvm));
3068 
3069 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3070 				   &(union tlbi_info) {
3071 					   .range = {
3072 						   .start = 0,
3073 						   .size = limit,
3074 					   },
3075 				   },
3076 				   s2_mmu_unmap_range);
3077 
3078 	return true;
3079 }
3080 
3081 static bool handle_ripas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3082 			      const struct sys_reg_desc *r)
3083 {
3084 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3085 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3086 	u64 base, range, tg, num, scale;
3087 	int shift;
3088 
3089 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3090 		return undef_access(vcpu, p, r);
3091 
3092 	/*
3093 	 * Because the shadow S2 structure doesn't necessarily reflect that
3094 	 * of the guest's S2 (different base granule size, for example), we
3095 	 * decide to ignore TTL and only use the described range.
3096 	 */
3097 	tg	= FIELD_GET(GENMASK(47, 46), p->regval);
3098 	scale	= FIELD_GET(GENMASK(45, 44), p->regval);
3099 	num	= FIELD_GET(GENMASK(43, 39), p->regval);
3100 	base	= p->regval & GENMASK(36, 0);
3101 
3102 	switch(tg) {
3103 	case 1:
3104 		shift = 12;
3105 		break;
3106 	case 2:
3107 		shift = 14;
3108 		break;
3109 	case 3:
3110 	default:		/* IMPDEF: handle tg==0 as 64k */
3111 		shift = 16;
3112 		break;
3113 	}
3114 
3115 	base <<= shift;
3116 	range = __TLBI_RANGE_PAGES(num, scale) << shift;
3117 
3118 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3119 				   &(union tlbi_info) {
3120 					   .range = {
3121 						   .start = base,
3122 						   .size = range,
3123 					   },
3124 				   },
3125 				   s2_mmu_unmap_range);
3126 
3127 	return true;
3128 }
3129 
3130 static void s2_mmu_unmap_ipa(struct kvm_s2_mmu *mmu,
3131 			     const union tlbi_info *info)
3132 {
3133 	unsigned long max_size;
3134 	u64 base_addr;
3135 
3136 	/*
3137 	 * We drop a number of things from the supplied value:
3138 	 *
3139 	 * - NS bit: we're non-secure only.
3140 	 *
3141 	 * - IPA[51:48]: We don't support 52bit IPA just yet...
3142 	 *
3143 	 * And of course, adjust the IPA to be on an actual address.
3144 	 */
3145 	base_addr = (info->ipa.addr & GENMASK_ULL(35, 0)) << 12;
3146 	max_size = compute_tlb_inval_range(mmu, info->ipa.addr);
3147 	base_addr &= ~(max_size - 1);
3148 
3149 	/*
3150 	 * See comment in s2_mmu_unmap_range() for why this is allowed to
3151 	 * reschedule.
3152 	 */
3153 	kvm_stage2_unmap_range(mmu, base_addr, max_size, true);
3154 }
3155 
3156 static bool handle_ipas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3157 			     const struct sys_reg_desc *r)
3158 {
3159 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3160 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3161 
3162 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3163 		return undef_access(vcpu, p, r);
3164 
3165 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3166 				   &(union tlbi_info) {
3167 					   .ipa = {
3168 						   .addr = p->regval,
3169 					   },
3170 				   },
3171 				   s2_mmu_unmap_ipa);
3172 
3173 	return true;
3174 }
3175 
3176 static void s2_mmu_tlbi_s1e1(struct kvm_s2_mmu *mmu,
3177 			     const union tlbi_info *info)
3178 {
3179 	WARN_ON(__kvm_tlbi_s1e2(mmu, info->va.addr, info->va.encoding));
3180 }
3181 
3182 static bool handle_tlbi_el1(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3183 			    const struct sys_reg_desc *r)
3184 {
3185 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3186 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3187 
3188 	/*
3189 	 * If we're here, this is because we've trapped on a EL1 TLBI
3190 	 * instruction that affects the EL1 translation regime while
3191 	 * we're running in a context that doesn't allow us to let the
3192 	 * HW do its thing (aka vEL2):
3193 	 *
3194 	 * - HCR_EL2.E2H == 0 : a non-VHE guest
3195 	 * - HCR_EL2.{E2H,TGE} == { 1, 0 } : a VHE guest in guest mode
3196 	 *
3197 	 * We don't expect these helpers to ever be called when running
3198 	 * in a vEL1 context.
3199 	 */
3200 
3201 	WARN_ON(!vcpu_is_el2(vcpu));
3202 
3203 	if (!kvm_supported_tlbi_s1e1_op(vcpu, sys_encoding))
3204 		return undef_access(vcpu, p, r);
3205 
3206 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3207 				   &(union tlbi_info) {
3208 					   .va = {
3209 						   .addr = p->regval,
3210 						   .encoding = sys_encoding,
3211 					   },
3212 				   },
3213 				   s2_mmu_tlbi_s1e1);
3214 
3215 	return true;
3216 }
3217 
3218 #define SYS_INSN(insn, access_fn)					\
3219 	{								\
3220 		SYS_DESC(OP_##insn),					\
3221 		.access = (access_fn),					\
3222 	}
3223 
3224 static struct sys_reg_desc sys_insn_descs[] = {
3225 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
3226 	{ SYS_DESC(SYS_DC_IGSW), access_dcgsw },
3227 	{ SYS_DESC(SYS_DC_IGDSW), access_dcgsw },
3228 
3229 	SYS_INSN(AT_S1E1R, handle_at_s1e01),
3230 	SYS_INSN(AT_S1E1W, handle_at_s1e01),
3231 	SYS_INSN(AT_S1E0R, handle_at_s1e01),
3232 	SYS_INSN(AT_S1E0W, handle_at_s1e01),
3233 	SYS_INSN(AT_S1E1RP, handle_at_s1e01),
3234 	SYS_INSN(AT_S1E1WP, handle_at_s1e01),
3235 
3236 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
3237 	{ SYS_DESC(SYS_DC_CGSW), access_dcgsw },
3238 	{ SYS_DESC(SYS_DC_CGDSW), access_dcgsw },
3239 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
3240 	{ SYS_DESC(SYS_DC_CIGSW), access_dcgsw },
3241 	{ SYS_DESC(SYS_DC_CIGDSW), access_dcgsw },
3242 
3243 	SYS_INSN(TLBI_VMALLE1OS, handle_tlbi_el1),
3244 	SYS_INSN(TLBI_VAE1OS, handle_tlbi_el1),
3245 	SYS_INSN(TLBI_ASIDE1OS, handle_tlbi_el1),
3246 	SYS_INSN(TLBI_VAAE1OS, handle_tlbi_el1),
3247 	SYS_INSN(TLBI_VALE1OS, handle_tlbi_el1),
3248 	SYS_INSN(TLBI_VAALE1OS, handle_tlbi_el1),
3249 
3250 	SYS_INSN(TLBI_RVAE1IS, handle_tlbi_el1),
3251 	SYS_INSN(TLBI_RVAAE1IS, handle_tlbi_el1),
3252 	SYS_INSN(TLBI_RVALE1IS, handle_tlbi_el1),
3253 	SYS_INSN(TLBI_RVAALE1IS, handle_tlbi_el1),
3254 
3255 	SYS_INSN(TLBI_VMALLE1IS, handle_tlbi_el1),
3256 	SYS_INSN(TLBI_VAE1IS, handle_tlbi_el1),
3257 	SYS_INSN(TLBI_ASIDE1IS, handle_tlbi_el1),
3258 	SYS_INSN(TLBI_VAAE1IS, handle_tlbi_el1),
3259 	SYS_INSN(TLBI_VALE1IS, handle_tlbi_el1),
3260 	SYS_INSN(TLBI_VAALE1IS, handle_tlbi_el1),
3261 
3262 	SYS_INSN(TLBI_RVAE1OS, handle_tlbi_el1),
3263 	SYS_INSN(TLBI_RVAAE1OS, handle_tlbi_el1),
3264 	SYS_INSN(TLBI_RVALE1OS, handle_tlbi_el1),
3265 	SYS_INSN(TLBI_RVAALE1OS, handle_tlbi_el1),
3266 
3267 	SYS_INSN(TLBI_RVAE1, handle_tlbi_el1),
3268 	SYS_INSN(TLBI_RVAAE1, handle_tlbi_el1),
3269 	SYS_INSN(TLBI_RVALE1, handle_tlbi_el1),
3270 	SYS_INSN(TLBI_RVAALE1, handle_tlbi_el1),
3271 
3272 	SYS_INSN(TLBI_VMALLE1, handle_tlbi_el1),
3273 	SYS_INSN(TLBI_VAE1, handle_tlbi_el1),
3274 	SYS_INSN(TLBI_ASIDE1, handle_tlbi_el1),
3275 	SYS_INSN(TLBI_VAAE1, handle_tlbi_el1),
3276 	SYS_INSN(TLBI_VALE1, handle_tlbi_el1),
3277 	SYS_INSN(TLBI_VAALE1, handle_tlbi_el1),
3278 
3279 	SYS_INSN(TLBI_VMALLE1OSNXS, handle_tlbi_el1),
3280 	SYS_INSN(TLBI_VAE1OSNXS, handle_tlbi_el1),
3281 	SYS_INSN(TLBI_ASIDE1OSNXS, handle_tlbi_el1),
3282 	SYS_INSN(TLBI_VAAE1OSNXS, handle_tlbi_el1),
3283 	SYS_INSN(TLBI_VALE1OSNXS, handle_tlbi_el1),
3284 	SYS_INSN(TLBI_VAALE1OSNXS, handle_tlbi_el1),
3285 
3286 	SYS_INSN(TLBI_RVAE1ISNXS, handle_tlbi_el1),
3287 	SYS_INSN(TLBI_RVAAE1ISNXS, handle_tlbi_el1),
3288 	SYS_INSN(TLBI_RVALE1ISNXS, handle_tlbi_el1),
3289 	SYS_INSN(TLBI_RVAALE1ISNXS, handle_tlbi_el1),
3290 
3291 	SYS_INSN(TLBI_VMALLE1ISNXS, handle_tlbi_el1),
3292 	SYS_INSN(TLBI_VAE1ISNXS, handle_tlbi_el1),
3293 	SYS_INSN(TLBI_ASIDE1ISNXS, handle_tlbi_el1),
3294 	SYS_INSN(TLBI_VAAE1ISNXS, handle_tlbi_el1),
3295 	SYS_INSN(TLBI_VALE1ISNXS, handle_tlbi_el1),
3296 	SYS_INSN(TLBI_VAALE1ISNXS, handle_tlbi_el1),
3297 
3298 	SYS_INSN(TLBI_RVAE1OSNXS, handle_tlbi_el1),
3299 	SYS_INSN(TLBI_RVAAE1OSNXS, handle_tlbi_el1),
3300 	SYS_INSN(TLBI_RVALE1OSNXS, handle_tlbi_el1),
3301 	SYS_INSN(TLBI_RVAALE1OSNXS, handle_tlbi_el1),
3302 
3303 	SYS_INSN(TLBI_RVAE1NXS, handle_tlbi_el1),
3304 	SYS_INSN(TLBI_RVAAE1NXS, handle_tlbi_el1),
3305 	SYS_INSN(TLBI_RVALE1NXS, handle_tlbi_el1),
3306 	SYS_INSN(TLBI_RVAALE1NXS, handle_tlbi_el1),
3307 
3308 	SYS_INSN(TLBI_VMALLE1NXS, handle_tlbi_el1),
3309 	SYS_INSN(TLBI_VAE1NXS, handle_tlbi_el1),
3310 	SYS_INSN(TLBI_ASIDE1NXS, handle_tlbi_el1),
3311 	SYS_INSN(TLBI_VAAE1NXS, handle_tlbi_el1),
3312 	SYS_INSN(TLBI_VALE1NXS, handle_tlbi_el1),
3313 	SYS_INSN(TLBI_VAALE1NXS, handle_tlbi_el1),
3314 
3315 	SYS_INSN(AT_S1E2R, handle_at_s1e2),
3316 	SYS_INSN(AT_S1E2W, handle_at_s1e2),
3317 	SYS_INSN(AT_S12E1R, handle_at_s12),
3318 	SYS_INSN(AT_S12E1W, handle_at_s12),
3319 	SYS_INSN(AT_S12E0R, handle_at_s12),
3320 	SYS_INSN(AT_S12E0W, handle_at_s12),
3321 	SYS_INSN(AT_S1E2A, handle_at_s1e2),
3322 
3323 	SYS_INSN(TLBI_IPAS2E1IS, handle_ipas2e1is),
3324 	SYS_INSN(TLBI_RIPAS2E1IS, handle_ripas2e1is),
3325 	SYS_INSN(TLBI_IPAS2LE1IS, handle_ipas2e1is),
3326 	SYS_INSN(TLBI_RIPAS2LE1IS, handle_ripas2e1is),
3327 
3328 	SYS_INSN(TLBI_ALLE2OS, undef_access),
3329 	SYS_INSN(TLBI_VAE2OS, undef_access),
3330 	SYS_INSN(TLBI_ALLE1OS, handle_alle1is),
3331 	SYS_INSN(TLBI_VALE2OS, undef_access),
3332 	SYS_INSN(TLBI_VMALLS12E1OS, handle_vmalls12e1is),
3333 
3334 	SYS_INSN(TLBI_RVAE2IS, undef_access),
3335 	SYS_INSN(TLBI_RVALE2IS, undef_access),
3336 
3337 	SYS_INSN(TLBI_ALLE1IS, handle_alle1is),
3338 	SYS_INSN(TLBI_VMALLS12E1IS, handle_vmalls12e1is),
3339 	SYS_INSN(TLBI_IPAS2E1OS, handle_ipas2e1is),
3340 	SYS_INSN(TLBI_IPAS2E1, handle_ipas2e1is),
3341 	SYS_INSN(TLBI_RIPAS2E1, handle_ripas2e1is),
3342 	SYS_INSN(TLBI_RIPAS2E1OS, handle_ripas2e1is),
3343 	SYS_INSN(TLBI_IPAS2LE1OS, handle_ipas2e1is),
3344 	SYS_INSN(TLBI_IPAS2LE1, handle_ipas2e1is),
3345 	SYS_INSN(TLBI_RIPAS2LE1, handle_ripas2e1is),
3346 	SYS_INSN(TLBI_RIPAS2LE1OS, handle_ripas2e1is),
3347 	SYS_INSN(TLBI_RVAE2OS, undef_access),
3348 	SYS_INSN(TLBI_RVALE2OS, undef_access),
3349 	SYS_INSN(TLBI_RVAE2, undef_access),
3350 	SYS_INSN(TLBI_RVALE2, undef_access),
3351 	SYS_INSN(TLBI_ALLE1, handle_alle1is),
3352 	SYS_INSN(TLBI_VMALLS12E1, handle_vmalls12e1is),
3353 
3354 	SYS_INSN(TLBI_IPAS2E1ISNXS, handle_ipas2e1is),
3355 	SYS_INSN(TLBI_RIPAS2E1ISNXS, handle_ripas2e1is),
3356 	SYS_INSN(TLBI_IPAS2LE1ISNXS, handle_ipas2e1is),
3357 	SYS_INSN(TLBI_RIPAS2LE1ISNXS, handle_ripas2e1is),
3358 
3359 	SYS_INSN(TLBI_ALLE2OSNXS, undef_access),
3360 	SYS_INSN(TLBI_VAE2OSNXS, undef_access),
3361 	SYS_INSN(TLBI_ALLE1OSNXS, handle_alle1is),
3362 	SYS_INSN(TLBI_VALE2OSNXS, undef_access),
3363 	SYS_INSN(TLBI_VMALLS12E1OSNXS, handle_vmalls12e1is),
3364 
3365 	SYS_INSN(TLBI_RVAE2ISNXS, undef_access),
3366 	SYS_INSN(TLBI_RVALE2ISNXS, undef_access),
3367 	SYS_INSN(TLBI_ALLE2ISNXS, undef_access),
3368 	SYS_INSN(TLBI_VAE2ISNXS, undef_access),
3369 
3370 	SYS_INSN(TLBI_ALLE1ISNXS, handle_alle1is),
3371 	SYS_INSN(TLBI_VALE2ISNXS, undef_access),
3372 	SYS_INSN(TLBI_VMALLS12E1ISNXS, handle_vmalls12e1is),
3373 	SYS_INSN(TLBI_IPAS2E1OSNXS, handle_ipas2e1is),
3374 	SYS_INSN(TLBI_IPAS2E1NXS, handle_ipas2e1is),
3375 	SYS_INSN(TLBI_RIPAS2E1NXS, handle_ripas2e1is),
3376 	SYS_INSN(TLBI_RIPAS2E1OSNXS, handle_ripas2e1is),
3377 	SYS_INSN(TLBI_IPAS2LE1OSNXS, handle_ipas2e1is),
3378 	SYS_INSN(TLBI_IPAS2LE1NXS, handle_ipas2e1is),
3379 	SYS_INSN(TLBI_RIPAS2LE1NXS, handle_ripas2e1is),
3380 	SYS_INSN(TLBI_RIPAS2LE1OSNXS, handle_ripas2e1is),
3381 	SYS_INSN(TLBI_RVAE2OSNXS, undef_access),
3382 	SYS_INSN(TLBI_RVALE2OSNXS, undef_access),
3383 	SYS_INSN(TLBI_RVAE2NXS, undef_access),
3384 	SYS_INSN(TLBI_RVALE2NXS, undef_access),
3385 	SYS_INSN(TLBI_ALLE2NXS, undef_access),
3386 	SYS_INSN(TLBI_VAE2NXS, undef_access),
3387 	SYS_INSN(TLBI_ALLE1NXS, handle_alle1is),
3388 	SYS_INSN(TLBI_VALE2NXS, undef_access),
3389 	SYS_INSN(TLBI_VMALLS12E1NXS, handle_vmalls12e1is),
3390 };
3391 
3392 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
3393 			struct sys_reg_params *p,
3394 			const struct sys_reg_desc *r)
3395 {
3396 	if (p->is_write) {
3397 		return ignore_write(vcpu, p);
3398 	} else {
3399 		u64 dfr = kvm_read_vm_id_reg(vcpu->kvm, SYS_ID_AA64DFR0_EL1);
3400 		u32 el3 = kvm_has_feat(vcpu->kvm, ID_AA64PFR0_EL1, EL3, IMP);
3401 
3402 		p->regval = ((SYS_FIELD_GET(ID_AA64DFR0_EL1, WRPs, dfr) << 28) |
3403 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, BRPs, dfr) << 24) |
3404 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, CTX_CMPs, dfr) << 20) |
3405 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, dfr) << 16) |
3406 			     (1 << 15) | (el3 << 14) | (el3 << 12));
3407 		return true;
3408 	}
3409 }
3410 
3411 /*
3412  * AArch32 debug register mappings
3413  *
3414  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
3415  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
3416  *
3417  * None of the other registers share their location, so treat them as
3418  * if they were 64bit.
3419  */
3420 #define DBG_BCR_BVR_WCR_WVR(n)						      \
3421 	/* DBGBVRn */							      \
3422 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
3423 	/* DBGBCRn */							      \
3424 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
3425 	/* DBGWVRn */							      \
3426 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
3427 	/* DBGWCRn */							      \
3428 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
3429 
3430 #define DBGBXVR(n)							      \
3431 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
3432 
3433 /*
3434  * Trapped cp14 registers. We generally ignore most of the external
3435  * debug, on the principle that they don't really make sense to a
3436  * guest. Revisit this one day, would this principle change.
3437  */
3438 static const struct sys_reg_desc cp14_regs[] = {
3439 	/* DBGDIDR */
3440 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
3441 	/* DBGDTRRXext */
3442 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
3443 
3444 	DBG_BCR_BVR_WCR_WVR(0),
3445 	/* DBGDSCRint */
3446 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
3447 	DBG_BCR_BVR_WCR_WVR(1),
3448 	/* DBGDCCINT */
3449 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
3450 	/* DBGDSCRext */
3451 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
3452 	DBG_BCR_BVR_WCR_WVR(2),
3453 	/* DBGDTR[RT]Xint */
3454 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
3455 	/* DBGDTR[RT]Xext */
3456 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
3457 	DBG_BCR_BVR_WCR_WVR(3),
3458 	DBG_BCR_BVR_WCR_WVR(4),
3459 	DBG_BCR_BVR_WCR_WVR(5),
3460 	/* DBGWFAR */
3461 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
3462 	/* DBGOSECCR */
3463 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
3464 	DBG_BCR_BVR_WCR_WVR(6),
3465 	/* DBGVCR */
3466 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
3467 	DBG_BCR_BVR_WCR_WVR(7),
3468 	DBG_BCR_BVR_WCR_WVR(8),
3469 	DBG_BCR_BVR_WCR_WVR(9),
3470 	DBG_BCR_BVR_WCR_WVR(10),
3471 	DBG_BCR_BVR_WCR_WVR(11),
3472 	DBG_BCR_BVR_WCR_WVR(12),
3473 	DBG_BCR_BVR_WCR_WVR(13),
3474 	DBG_BCR_BVR_WCR_WVR(14),
3475 	DBG_BCR_BVR_WCR_WVR(15),
3476 
3477 	/* DBGDRAR (32bit) */
3478 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
3479 
3480 	DBGBXVR(0),
3481 	/* DBGOSLAR */
3482 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
3483 	DBGBXVR(1),
3484 	/* DBGOSLSR */
3485 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
3486 	DBGBXVR(2),
3487 	DBGBXVR(3),
3488 	/* DBGOSDLR */
3489 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
3490 	DBGBXVR(4),
3491 	/* DBGPRCR */
3492 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
3493 	DBGBXVR(5),
3494 	DBGBXVR(6),
3495 	DBGBXVR(7),
3496 	DBGBXVR(8),
3497 	DBGBXVR(9),
3498 	DBGBXVR(10),
3499 	DBGBXVR(11),
3500 	DBGBXVR(12),
3501 	DBGBXVR(13),
3502 	DBGBXVR(14),
3503 	DBGBXVR(15),
3504 
3505 	/* DBGDSAR (32bit) */
3506 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
3507 
3508 	/* DBGDEVID2 */
3509 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
3510 	/* DBGDEVID1 */
3511 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
3512 	/* DBGDEVID */
3513 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
3514 	/* DBGCLAIMSET */
3515 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
3516 	/* DBGCLAIMCLR */
3517 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
3518 	/* DBGAUTHSTATUS */
3519 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
3520 };
3521 
3522 /* Trapped cp14 64bit registers */
3523 static const struct sys_reg_desc cp14_64_regs[] = {
3524 	/* DBGDRAR (64bit) */
3525 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
3526 
3527 	/* DBGDSAR (64bit) */
3528 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
3529 };
3530 
3531 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)			\
3532 	AA32(_map),							\
3533 	Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),			\
3534 	.visibility = pmu_visibility
3535 
3536 /* Macro to expand the PMEVCNTRn register */
3537 #define PMU_PMEVCNTR(n)							\
3538 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
3539 	  (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
3540 	  .access = access_pmu_evcntr }
3541 
3542 /* Macro to expand the PMEVTYPERn register */
3543 #define PMU_PMEVTYPER(n)						\
3544 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
3545 	  (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
3546 	  .access = access_pmu_evtyper }
3547 /*
3548  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
3549  * depending on the way they are accessed (as a 32bit or a 64bit
3550  * register).
3551  */
3552 static const struct sys_reg_desc cp15_regs[] = {
3553 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
3554 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
3555 	/* ACTLR */
3556 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
3557 	/* ACTLR2 */
3558 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
3559 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
3560 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
3561 	/* TTBCR */
3562 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
3563 	/* TTBCR2 */
3564 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
3565 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
3566 	{ CP15_SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
3567 	/* DFSR */
3568 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
3569 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
3570 	/* ADFSR */
3571 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
3572 	/* AIFSR */
3573 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
3574 	/* DFAR */
3575 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
3576 	/* IFAR */
3577 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
3578 
3579 	/*
3580 	 * DC{C,I,CI}SW operations:
3581 	 */
3582 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
3583 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
3584 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
3585 
3586 	/* PMU */
3587 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
3588 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
3589 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
3590 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
3591 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
3592 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
3593 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
3594 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
3595 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
3596 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
3597 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
3598 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
3599 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
3600 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
3601 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
3602 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
3603 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
3604 	/* PMMIR */
3605 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
3606 
3607 	/* PRRR/MAIR0 */
3608 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
3609 	/* NMRR/MAIR1 */
3610 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
3611 	/* AMAIR0 */
3612 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
3613 	/* AMAIR1 */
3614 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
3615 
3616 	{ CP15_SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
3617 	{ CP15_SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
3618 	{ CP15_SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
3619 	{ CP15_SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
3620 	{ CP15_SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
3621 	{ CP15_SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
3622 	{ CP15_SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
3623 	{ CP15_SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
3624 	{ CP15_SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
3625 	{ CP15_SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
3626 	{ CP15_SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
3627 	{ CP15_SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
3628 	{ CP15_SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
3629 	{ CP15_SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
3630 	{ CP15_SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
3631 	{ CP15_SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
3632 	{ CP15_SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
3633 	{ CP15_SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
3634 	{ CP15_SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
3635 	{ CP15_SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
3636 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
3637 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
3638 
3639 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
3640 
3641 	/* Arch Tmers */
3642 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
3643 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
3644 
3645 	/* PMEVCNTRn */
3646 	PMU_PMEVCNTR(0),
3647 	PMU_PMEVCNTR(1),
3648 	PMU_PMEVCNTR(2),
3649 	PMU_PMEVCNTR(3),
3650 	PMU_PMEVCNTR(4),
3651 	PMU_PMEVCNTR(5),
3652 	PMU_PMEVCNTR(6),
3653 	PMU_PMEVCNTR(7),
3654 	PMU_PMEVCNTR(8),
3655 	PMU_PMEVCNTR(9),
3656 	PMU_PMEVCNTR(10),
3657 	PMU_PMEVCNTR(11),
3658 	PMU_PMEVCNTR(12),
3659 	PMU_PMEVCNTR(13),
3660 	PMU_PMEVCNTR(14),
3661 	PMU_PMEVCNTR(15),
3662 	PMU_PMEVCNTR(16),
3663 	PMU_PMEVCNTR(17),
3664 	PMU_PMEVCNTR(18),
3665 	PMU_PMEVCNTR(19),
3666 	PMU_PMEVCNTR(20),
3667 	PMU_PMEVCNTR(21),
3668 	PMU_PMEVCNTR(22),
3669 	PMU_PMEVCNTR(23),
3670 	PMU_PMEVCNTR(24),
3671 	PMU_PMEVCNTR(25),
3672 	PMU_PMEVCNTR(26),
3673 	PMU_PMEVCNTR(27),
3674 	PMU_PMEVCNTR(28),
3675 	PMU_PMEVCNTR(29),
3676 	PMU_PMEVCNTR(30),
3677 	/* PMEVTYPERn */
3678 	PMU_PMEVTYPER(0),
3679 	PMU_PMEVTYPER(1),
3680 	PMU_PMEVTYPER(2),
3681 	PMU_PMEVTYPER(3),
3682 	PMU_PMEVTYPER(4),
3683 	PMU_PMEVTYPER(5),
3684 	PMU_PMEVTYPER(6),
3685 	PMU_PMEVTYPER(7),
3686 	PMU_PMEVTYPER(8),
3687 	PMU_PMEVTYPER(9),
3688 	PMU_PMEVTYPER(10),
3689 	PMU_PMEVTYPER(11),
3690 	PMU_PMEVTYPER(12),
3691 	PMU_PMEVTYPER(13),
3692 	PMU_PMEVTYPER(14),
3693 	PMU_PMEVTYPER(15),
3694 	PMU_PMEVTYPER(16),
3695 	PMU_PMEVTYPER(17),
3696 	PMU_PMEVTYPER(18),
3697 	PMU_PMEVTYPER(19),
3698 	PMU_PMEVTYPER(20),
3699 	PMU_PMEVTYPER(21),
3700 	PMU_PMEVTYPER(22),
3701 	PMU_PMEVTYPER(23),
3702 	PMU_PMEVTYPER(24),
3703 	PMU_PMEVTYPER(25),
3704 	PMU_PMEVTYPER(26),
3705 	PMU_PMEVTYPER(27),
3706 	PMU_PMEVTYPER(28),
3707 	PMU_PMEVTYPER(29),
3708 	PMU_PMEVTYPER(30),
3709 	/* PMCCFILTR */
3710 	{ CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
3711 
3712 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
3713 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
3714 
3715 	/* CCSIDR2 */
3716 	{ Op1(1), CRn( 0), CRm( 0),  Op2(2), undef_access },
3717 
3718 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
3719 };
3720 
3721 static const struct sys_reg_desc cp15_64_regs[] = {
3722 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
3723 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
3724 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
3725 	{ SYS_DESC(SYS_AARCH32_CNTPCT),	      access_arch_timer },
3726 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
3727 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
3728 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
3729 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
3730 	{ SYS_DESC(SYS_AARCH32_CNTPCTSS),     access_arch_timer },
3731 };
3732 
3733 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
3734 			       bool is_32)
3735 {
3736 	unsigned int i;
3737 
3738 	for (i = 0; i < n; i++) {
3739 		if (!is_32 && table[i].reg && !table[i].reset) {
3740 			kvm_err("sys_reg table %pS entry %d (%s) lacks reset\n",
3741 				&table[i], i, table[i].name);
3742 			return false;
3743 		}
3744 
3745 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
3746 			kvm_err("sys_reg table %pS entry %d (%s -> %s) out of order\n",
3747 				&table[i], i, table[i - 1].name, table[i].name);
3748 			return false;
3749 		}
3750 	}
3751 
3752 	return true;
3753 }
3754 
3755 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
3756 {
3757 	kvm_inject_undefined(vcpu);
3758 	return 1;
3759 }
3760 
3761 static void perform_access(struct kvm_vcpu *vcpu,
3762 			   struct sys_reg_params *params,
3763 			   const struct sys_reg_desc *r)
3764 {
3765 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
3766 
3767 	/* Check for regs disabled by runtime config */
3768 	if (sysreg_hidden(vcpu, r)) {
3769 		kvm_inject_undefined(vcpu);
3770 		return;
3771 	}
3772 
3773 	/*
3774 	 * Not having an accessor means that we have configured a trap
3775 	 * that we don't know how to handle. This certainly qualifies
3776 	 * as a gross bug that should be fixed right away.
3777 	 */
3778 	BUG_ON(!r->access);
3779 
3780 	/* Skip instruction if instructed so */
3781 	if (likely(r->access(vcpu, params, r)))
3782 		kvm_incr_pc(vcpu);
3783 }
3784 
3785 /*
3786  * emulate_cp --  tries to match a sys_reg access in a handling table, and
3787  *                call the corresponding trap handler.
3788  *
3789  * @params: pointer to the descriptor of the access
3790  * @table: array of trap descriptors
3791  * @num: size of the trap descriptor array
3792  *
3793  * Return true if the access has been handled, false if not.
3794  */
3795 static bool emulate_cp(struct kvm_vcpu *vcpu,
3796 		       struct sys_reg_params *params,
3797 		       const struct sys_reg_desc *table,
3798 		       size_t num)
3799 {
3800 	const struct sys_reg_desc *r;
3801 
3802 	if (!table)
3803 		return false;	/* Not handled */
3804 
3805 	r = find_reg(params, table, num);
3806 
3807 	if (r) {
3808 		perform_access(vcpu, params, r);
3809 		return true;
3810 	}
3811 
3812 	/* Not handled */
3813 	return false;
3814 }
3815 
3816 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
3817 				struct sys_reg_params *params)
3818 {
3819 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
3820 	int cp = -1;
3821 
3822 	switch (esr_ec) {
3823 	case ESR_ELx_EC_CP15_32:
3824 	case ESR_ELx_EC_CP15_64:
3825 		cp = 15;
3826 		break;
3827 	case ESR_ELx_EC_CP14_MR:
3828 	case ESR_ELx_EC_CP14_64:
3829 		cp = 14;
3830 		break;
3831 	default:
3832 		WARN_ON(1);
3833 	}
3834 
3835 	print_sys_reg_msg(params,
3836 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
3837 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
3838 	kvm_inject_undefined(vcpu);
3839 }
3840 
3841 /**
3842  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
3843  * @vcpu: The VCPU pointer
3844  * @global: &struct sys_reg_desc
3845  * @nr_global: size of the @global array
3846  */
3847 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
3848 			    const struct sys_reg_desc *global,
3849 			    size_t nr_global)
3850 {
3851 	struct sys_reg_params params;
3852 	u64 esr = kvm_vcpu_get_esr(vcpu);
3853 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
3854 	int Rt2 = (esr >> 10) & 0x1f;
3855 
3856 	params.CRm = (esr >> 1) & 0xf;
3857 	params.is_write = ((esr & 1) == 0);
3858 
3859 	params.Op0 = 0;
3860 	params.Op1 = (esr >> 16) & 0xf;
3861 	params.Op2 = 0;
3862 	params.CRn = 0;
3863 
3864 	/*
3865 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
3866 	 * backends between AArch32 and AArch64, we get away with it.
3867 	 */
3868 	if (params.is_write) {
3869 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
3870 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
3871 	}
3872 
3873 	/*
3874 	 * If the table contains a handler, handle the
3875 	 * potential register operation in the case of a read and return
3876 	 * with success.
3877 	 */
3878 	if (emulate_cp(vcpu, &params, global, nr_global)) {
3879 		/* Split up the value between registers for the read side */
3880 		if (!params.is_write) {
3881 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
3882 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
3883 		}
3884 
3885 		return 1;
3886 	}
3887 
3888 	unhandled_cp_access(vcpu, &params);
3889 	return 1;
3890 }
3891 
3892 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
3893 
3894 /*
3895  * The CP10 ID registers are architecturally mapped to AArch64 feature
3896  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
3897  * from AArch32.
3898  */
3899 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
3900 {
3901 	u8 reg_id = (esr >> 10) & 0xf;
3902 	bool valid;
3903 
3904 	params->is_write = ((esr & 1) == 0);
3905 	params->Op0 = 3;
3906 	params->Op1 = 0;
3907 	params->CRn = 0;
3908 	params->CRm = 3;
3909 
3910 	/* CP10 ID registers are read-only */
3911 	valid = !params->is_write;
3912 
3913 	switch (reg_id) {
3914 	/* MVFR0 */
3915 	case 0b0111:
3916 		params->Op2 = 0;
3917 		break;
3918 	/* MVFR1 */
3919 	case 0b0110:
3920 		params->Op2 = 1;
3921 		break;
3922 	/* MVFR2 */
3923 	case 0b0101:
3924 		params->Op2 = 2;
3925 		break;
3926 	default:
3927 		valid = false;
3928 	}
3929 
3930 	if (valid)
3931 		return true;
3932 
3933 	kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
3934 		      params->is_write ? "write" : "read", reg_id);
3935 	return false;
3936 }
3937 
3938 /**
3939  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
3940  *			  VFP Register' from AArch32.
3941  * @vcpu: The vCPU pointer
3942  *
3943  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
3944  * Work out the correct AArch64 system register encoding and reroute to the
3945  * AArch64 system register emulation.
3946  */
3947 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
3948 {
3949 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
3950 	u64 esr = kvm_vcpu_get_esr(vcpu);
3951 	struct sys_reg_params params;
3952 
3953 	/* UNDEF on any unhandled register access */
3954 	if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
3955 		kvm_inject_undefined(vcpu);
3956 		return 1;
3957 	}
3958 
3959 	if (emulate_sys_reg(vcpu, &params))
3960 		vcpu_set_reg(vcpu, Rt, params.regval);
3961 
3962 	return 1;
3963 }
3964 
3965 /**
3966  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
3967  *			       CRn=0, which corresponds to the AArch32 feature
3968  *			       registers.
3969  * @vcpu: the vCPU pointer
3970  * @params: the system register access parameters.
3971  *
3972  * Our cp15 system register tables do not enumerate the AArch32 feature
3973  * registers. Conveniently, our AArch64 table does, and the AArch32 system
3974  * register encoding can be trivially remapped into the AArch64 for the feature
3975  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
3976  *
3977  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
3978  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
3979  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
3980  * treat undefined registers in this range as RAZ.
3981  */
3982 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
3983 				   struct sys_reg_params *params)
3984 {
3985 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
3986 
3987 	/* Treat impossible writes to RO registers as UNDEFINED */
3988 	if (params->is_write) {
3989 		unhandled_cp_access(vcpu, params);
3990 		return 1;
3991 	}
3992 
3993 	params->Op0 = 3;
3994 
3995 	/*
3996 	 * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
3997 	 * Avoid conflicting with future expansion of AArch64 feature registers
3998 	 * and simply treat them as RAZ here.
3999 	 */
4000 	if (params->CRm > 3)
4001 		params->regval = 0;
4002 	else if (!emulate_sys_reg(vcpu, params))
4003 		return 1;
4004 
4005 	vcpu_set_reg(vcpu, Rt, params->regval);
4006 	return 1;
4007 }
4008 
4009 /**
4010  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
4011  * @vcpu: The VCPU pointer
4012  * @params: &struct sys_reg_params
4013  * @global: &struct sys_reg_desc
4014  * @nr_global: size of the @global array
4015  */
4016 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
4017 			    struct sys_reg_params *params,
4018 			    const struct sys_reg_desc *global,
4019 			    size_t nr_global)
4020 {
4021 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
4022 
4023 	params->regval = vcpu_get_reg(vcpu, Rt);
4024 
4025 	if (emulate_cp(vcpu, params, global, nr_global)) {
4026 		if (!params->is_write)
4027 			vcpu_set_reg(vcpu, Rt, params->regval);
4028 		return 1;
4029 	}
4030 
4031 	unhandled_cp_access(vcpu, params);
4032 	return 1;
4033 }
4034 
4035 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
4036 {
4037 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
4038 }
4039 
4040 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
4041 {
4042 	struct sys_reg_params params;
4043 
4044 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4045 
4046 	/*
4047 	 * Certain AArch32 ID registers are handled by rerouting to the AArch64
4048 	 * system register table. Registers in the ID range where CRm=0 are
4049 	 * excluded from this scheme as they do not trivially map into AArch64
4050 	 * system register encodings.
4051 	 */
4052 	if (params.Op1 == 0 && params.CRn == 0 && params.CRm)
4053 		return kvm_emulate_cp15_id_reg(vcpu, &params);
4054 
4055 	return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
4056 }
4057 
4058 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
4059 {
4060 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
4061 }
4062 
4063 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
4064 {
4065 	struct sys_reg_params params;
4066 
4067 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4068 
4069 	return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
4070 }
4071 
4072 /**
4073  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
4074  * @vcpu: The VCPU pointer
4075  * @params: Decoded system register parameters
4076  *
4077  * Return: true if the system register access was successful, false otherwise.
4078  */
4079 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
4080 			    struct sys_reg_params *params)
4081 {
4082 	const struct sys_reg_desc *r;
4083 
4084 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4085 	if (likely(r)) {
4086 		perform_access(vcpu, params, r);
4087 		return true;
4088 	}
4089 
4090 	print_sys_reg_msg(params,
4091 			  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
4092 			  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
4093 	kvm_inject_undefined(vcpu);
4094 
4095 	return false;
4096 }
4097 
4098 static const struct sys_reg_desc *idregs_debug_find(struct kvm *kvm, u8 pos)
4099 {
4100 	unsigned long i, idreg_idx = 0;
4101 
4102 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4103 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4104 
4105 		if (!is_vm_ftr_id_reg(reg_to_encoding(r)))
4106 			continue;
4107 
4108 		if (idreg_idx == pos)
4109 			return r;
4110 
4111 		idreg_idx++;
4112 	}
4113 
4114 	return NULL;
4115 }
4116 
4117 static void *idregs_debug_start(struct seq_file *s, loff_t *pos)
4118 {
4119 	struct kvm *kvm = s->private;
4120 	u8 *iter;
4121 
4122 	mutex_lock(&kvm->arch.config_lock);
4123 
4124 	iter = &kvm->arch.idreg_debugfs_iter;
4125 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags) &&
4126 	    *iter == (u8)~0) {
4127 		*iter = *pos;
4128 		if (!idregs_debug_find(kvm, *iter))
4129 			iter = NULL;
4130 	} else {
4131 		iter = ERR_PTR(-EBUSY);
4132 	}
4133 
4134 	mutex_unlock(&kvm->arch.config_lock);
4135 
4136 	return iter;
4137 }
4138 
4139 static void *idregs_debug_next(struct seq_file *s, void *v, loff_t *pos)
4140 {
4141 	struct kvm *kvm = s->private;
4142 
4143 	(*pos)++;
4144 
4145 	if (idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter + 1)) {
4146 		kvm->arch.idreg_debugfs_iter++;
4147 
4148 		return &kvm->arch.idreg_debugfs_iter;
4149 	}
4150 
4151 	return NULL;
4152 }
4153 
4154 static void idregs_debug_stop(struct seq_file *s, void *v)
4155 {
4156 	struct kvm *kvm = s->private;
4157 
4158 	if (IS_ERR(v))
4159 		return;
4160 
4161 	mutex_lock(&kvm->arch.config_lock);
4162 
4163 	kvm->arch.idreg_debugfs_iter = ~0;
4164 
4165 	mutex_unlock(&kvm->arch.config_lock);
4166 }
4167 
4168 static int idregs_debug_show(struct seq_file *s, void *v)
4169 {
4170 	const struct sys_reg_desc *desc;
4171 	struct kvm *kvm = s->private;
4172 
4173 	desc = idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter);
4174 
4175 	if (!desc->name)
4176 		return 0;
4177 
4178 	seq_printf(s, "%20s:\t%016llx\n",
4179 		   desc->name, kvm_read_vm_id_reg(kvm, reg_to_encoding(desc)));
4180 
4181 	return 0;
4182 }
4183 
4184 static const struct seq_operations idregs_debug_sops = {
4185 	.start	= idregs_debug_start,
4186 	.next	= idregs_debug_next,
4187 	.stop	= idregs_debug_stop,
4188 	.show	= idregs_debug_show,
4189 };
4190 
4191 DEFINE_SEQ_ATTRIBUTE(idregs_debug);
4192 
4193 void kvm_sys_regs_create_debugfs(struct kvm *kvm)
4194 {
4195 	kvm->arch.idreg_debugfs_iter = ~0;
4196 
4197 	debugfs_create_file("idregs", 0444, kvm->debugfs_dentry, kvm,
4198 			    &idregs_debug_fops);
4199 }
4200 
4201 static void reset_vm_ftr_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *reg)
4202 {
4203 	u32 id = reg_to_encoding(reg);
4204 	struct kvm *kvm = vcpu->kvm;
4205 
4206 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags))
4207 		return;
4208 
4209 	kvm_set_vm_id_reg(kvm, id, reg->reset(vcpu, reg));
4210 }
4211 
4212 static void reset_vcpu_ftr_id_reg(struct kvm_vcpu *vcpu,
4213 				  const struct sys_reg_desc *reg)
4214 {
4215 	if (kvm_vcpu_initialized(vcpu))
4216 		return;
4217 
4218 	reg->reset(vcpu, reg);
4219 }
4220 
4221 /**
4222  * kvm_reset_sys_regs - sets system registers to reset value
4223  * @vcpu: The VCPU pointer
4224  *
4225  * This function finds the right table above and sets the registers on the
4226  * virtual CPU struct to their architecturally defined reset values.
4227  */
4228 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
4229 {
4230 	struct kvm *kvm = vcpu->kvm;
4231 	unsigned long i;
4232 
4233 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4234 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4235 
4236 		if (!r->reset)
4237 			continue;
4238 
4239 		if (is_vm_ftr_id_reg(reg_to_encoding(r)))
4240 			reset_vm_ftr_id_reg(vcpu, r);
4241 		else if (is_vcpu_ftr_id_reg(reg_to_encoding(r)))
4242 			reset_vcpu_ftr_id_reg(vcpu, r);
4243 		else
4244 			r->reset(vcpu, r);
4245 	}
4246 
4247 	set_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags);
4248 }
4249 
4250 /**
4251  * kvm_handle_sys_reg -- handles a system instruction or mrs/msr instruction
4252  *			 trap on a guest execution
4253  * @vcpu: The VCPU pointer
4254  */
4255 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
4256 {
4257 	const struct sys_reg_desc *desc = NULL;
4258 	struct sys_reg_params params;
4259 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
4260 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4261 	int sr_idx;
4262 
4263 	trace_kvm_handle_sys_reg(esr);
4264 
4265 	if (triage_sysreg_trap(vcpu, &sr_idx))
4266 		return 1;
4267 
4268 	params = esr_sys64_to_params(esr);
4269 	params.regval = vcpu_get_reg(vcpu, Rt);
4270 
4271 	/* System registers have Op0=={2,3}, as per DDI487 J.a C5.1.2 */
4272 	if (params.Op0 == 2 || params.Op0 == 3)
4273 		desc = &sys_reg_descs[sr_idx];
4274 	else
4275 		desc = &sys_insn_descs[sr_idx];
4276 
4277 	perform_access(vcpu, &params, desc);
4278 
4279 	/* Read from system register? */
4280 	if (!params.is_write &&
4281 	    (params.Op0 == 2 || params.Op0 == 3))
4282 		vcpu_set_reg(vcpu, Rt, params.regval);
4283 
4284 	return 1;
4285 }
4286 
4287 /******************************************************************************
4288  * Userspace API
4289  *****************************************************************************/
4290 
4291 static bool index_to_params(u64 id, struct sys_reg_params *params)
4292 {
4293 	switch (id & KVM_REG_SIZE_MASK) {
4294 	case KVM_REG_SIZE_U64:
4295 		/* Any unused index bits means it's not valid. */
4296 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
4297 			      | KVM_REG_ARM_COPROC_MASK
4298 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
4299 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
4300 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
4301 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
4302 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
4303 			return false;
4304 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
4305 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
4306 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
4307 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
4308 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
4309 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
4310 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
4311 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
4312 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
4313 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
4314 		return true;
4315 	default:
4316 		return false;
4317 	}
4318 }
4319 
4320 const struct sys_reg_desc *get_reg_by_id(u64 id,
4321 					 const struct sys_reg_desc table[],
4322 					 unsigned int num)
4323 {
4324 	struct sys_reg_params params;
4325 
4326 	if (!index_to_params(id, &params))
4327 		return NULL;
4328 
4329 	return find_reg(&params, table, num);
4330 }
4331 
4332 /* Decode an index value, and find the sys_reg_desc entry. */
4333 static const struct sys_reg_desc *
4334 id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
4335 		   const struct sys_reg_desc table[], unsigned int num)
4336 
4337 {
4338 	const struct sys_reg_desc *r;
4339 
4340 	/* We only do sys_reg for now. */
4341 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
4342 		return NULL;
4343 
4344 	r = get_reg_by_id(id, table, num);
4345 
4346 	/* Not saved in the sys_reg array and not otherwise accessible? */
4347 	if (r && (!(r->reg || r->get_user) || sysreg_hidden(vcpu, r)))
4348 		r = NULL;
4349 
4350 	return r;
4351 }
4352 
4353 /*
4354  * These are the invariant sys_reg registers: we let the guest see the
4355  * host versions of these, so they're part of the guest state.
4356  *
4357  * A future CPU may provide a mechanism to present different values to
4358  * the guest, or a future kvm may trap them.
4359  */
4360 
4361 #define FUNCTION_INVARIANT(reg)						\
4362 	static u64 reset_##reg(struct kvm_vcpu *v,			\
4363 			       const struct sys_reg_desc *r)		\
4364 	{								\
4365 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
4366 		return ((struct sys_reg_desc *)r)->val;			\
4367 	}
4368 
4369 FUNCTION_INVARIANT(midr_el1)
4370 FUNCTION_INVARIANT(revidr_el1)
4371 FUNCTION_INVARIANT(aidr_el1)
4372 
4373 /* ->val is filled in by kvm_sys_reg_table_init() */
4374 static struct sys_reg_desc invariant_sys_regs[] __ro_after_init = {
4375 	{ SYS_DESC(SYS_MIDR_EL1), NULL, reset_midr_el1 },
4376 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, reset_revidr_el1 },
4377 	{ SYS_DESC(SYS_AIDR_EL1), NULL, reset_aidr_el1 },
4378 };
4379 
4380 static int get_invariant_sys_reg(u64 id, u64 __user *uaddr)
4381 {
4382 	const struct sys_reg_desc *r;
4383 
4384 	r = get_reg_by_id(id, invariant_sys_regs,
4385 			  ARRAY_SIZE(invariant_sys_regs));
4386 	if (!r)
4387 		return -ENOENT;
4388 
4389 	return put_user(r->val, uaddr);
4390 }
4391 
4392 static int set_invariant_sys_reg(u64 id, u64 __user *uaddr)
4393 {
4394 	const struct sys_reg_desc *r;
4395 	u64 val;
4396 
4397 	r = get_reg_by_id(id, invariant_sys_regs,
4398 			  ARRAY_SIZE(invariant_sys_regs));
4399 	if (!r)
4400 		return -ENOENT;
4401 
4402 	if (get_user(val, uaddr))
4403 		return -EFAULT;
4404 
4405 	/* This is what we mean by invariant: you can't change it. */
4406 	if (r->val != val)
4407 		return -EINVAL;
4408 
4409 	return 0;
4410 }
4411 
4412 static int demux_c15_get(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
4413 {
4414 	u32 val;
4415 	u32 __user *uval = uaddr;
4416 
4417 	/* Fail if we have unknown bits set. */
4418 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
4419 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
4420 		return -ENOENT;
4421 
4422 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
4423 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
4424 		if (KVM_REG_SIZE(id) != 4)
4425 			return -ENOENT;
4426 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
4427 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
4428 		if (val >= CSSELR_MAX)
4429 			return -ENOENT;
4430 
4431 		return put_user(get_ccsidr(vcpu, val), uval);
4432 	default:
4433 		return -ENOENT;
4434 	}
4435 }
4436 
4437 static int demux_c15_set(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
4438 {
4439 	u32 val, newval;
4440 	u32 __user *uval = uaddr;
4441 
4442 	/* Fail if we have unknown bits set. */
4443 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
4444 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
4445 		return -ENOENT;
4446 
4447 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
4448 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
4449 		if (KVM_REG_SIZE(id) != 4)
4450 			return -ENOENT;
4451 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
4452 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
4453 		if (val >= CSSELR_MAX)
4454 			return -ENOENT;
4455 
4456 		if (get_user(newval, uval))
4457 			return -EFAULT;
4458 
4459 		return set_ccsidr(vcpu, val, newval);
4460 	default:
4461 		return -ENOENT;
4462 	}
4463 }
4464 
4465 int kvm_sys_reg_get_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
4466 			 const struct sys_reg_desc table[], unsigned int num)
4467 {
4468 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
4469 	const struct sys_reg_desc *r;
4470 	u64 val;
4471 	int ret;
4472 
4473 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
4474 	if (!r || sysreg_hidden(vcpu, r))
4475 		return -ENOENT;
4476 
4477 	if (r->get_user) {
4478 		ret = (r->get_user)(vcpu, r, &val);
4479 	} else {
4480 		val = __vcpu_sys_reg(vcpu, r->reg);
4481 		ret = 0;
4482 	}
4483 
4484 	if (!ret)
4485 		ret = put_user(val, uaddr);
4486 
4487 	return ret;
4488 }
4489 
4490 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
4491 {
4492 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
4493 	int err;
4494 
4495 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
4496 		return demux_c15_get(vcpu, reg->id, uaddr);
4497 
4498 	err = get_invariant_sys_reg(reg->id, uaddr);
4499 	if (err != -ENOENT)
4500 		return err;
4501 
4502 	return kvm_sys_reg_get_user(vcpu, reg,
4503 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4504 }
4505 
4506 int kvm_sys_reg_set_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
4507 			 const struct sys_reg_desc table[], unsigned int num)
4508 {
4509 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
4510 	const struct sys_reg_desc *r;
4511 	u64 val;
4512 	int ret;
4513 
4514 	if (get_user(val, uaddr))
4515 		return -EFAULT;
4516 
4517 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
4518 	if (!r || sysreg_hidden(vcpu, r))
4519 		return -ENOENT;
4520 
4521 	if (sysreg_user_write_ignore(vcpu, r))
4522 		return 0;
4523 
4524 	if (r->set_user) {
4525 		ret = (r->set_user)(vcpu, r, val);
4526 	} else {
4527 		__vcpu_sys_reg(vcpu, r->reg) = val;
4528 		ret = 0;
4529 	}
4530 
4531 	return ret;
4532 }
4533 
4534 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
4535 {
4536 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
4537 	int err;
4538 
4539 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
4540 		return demux_c15_set(vcpu, reg->id, uaddr);
4541 
4542 	err = set_invariant_sys_reg(reg->id, uaddr);
4543 	if (err != -ENOENT)
4544 		return err;
4545 
4546 	return kvm_sys_reg_set_user(vcpu, reg,
4547 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4548 }
4549 
4550 static unsigned int num_demux_regs(void)
4551 {
4552 	return CSSELR_MAX;
4553 }
4554 
4555 static int write_demux_regids(u64 __user *uindices)
4556 {
4557 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
4558 	unsigned int i;
4559 
4560 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
4561 	for (i = 0; i < CSSELR_MAX; i++) {
4562 		if (put_user(val | i, uindices))
4563 			return -EFAULT;
4564 		uindices++;
4565 	}
4566 	return 0;
4567 }
4568 
4569 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
4570 {
4571 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
4572 		KVM_REG_ARM64_SYSREG |
4573 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
4574 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
4575 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
4576 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
4577 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
4578 }
4579 
4580 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
4581 {
4582 	if (!*uind)
4583 		return true;
4584 
4585 	if (put_user(sys_reg_to_index(reg), *uind))
4586 		return false;
4587 
4588 	(*uind)++;
4589 	return true;
4590 }
4591 
4592 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
4593 			    const struct sys_reg_desc *rd,
4594 			    u64 __user **uind,
4595 			    unsigned int *total)
4596 {
4597 	/*
4598 	 * Ignore registers we trap but don't save,
4599 	 * and for which no custom user accessor is provided.
4600 	 */
4601 	if (!(rd->reg || rd->get_user))
4602 		return 0;
4603 
4604 	if (sysreg_hidden(vcpu, rd))
4605 		return 0;
4606 
4607 	if (!copy_reg_to_user(rd, uind))
4608 		return -EFAULT;
4609 
4610 	(*total)++;
4611 	return 0;
4612 }
4613 
4614 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
4615 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
4616 {
4617 	const struct sys_reg_desc *i2, *end2;
4618 	unsigned int total = 0;
4619 	int err;
4620 
4621 	i2 = sys_reg_descs;
4622 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
4623 
4624 	while (i2 != end2) {
4625 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
4626 		if (err)
4627 			return err;
4628 	}
4629 	return total;
4630 }
4631 
4632 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
4633 {
4634 	return ARRAY_SIZE(invariant_sys_regs)
4635 		+ num_demux_regs()
4636 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
4637 }
4638 
4639 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
4640 {
4641 	unsigned int i;
4642 	int err;
4643 
4644 	/* Then give them all the invariant registers' indices. */
4645 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
4646 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
4647 			return -EFAULT;
4648 		uindices++;
4649 	}
4650 
4651 	err = walk_sys_regs(vcpu, uindices);
4652 	if (err < 0)
4653 		return err;
4654 	uindices += err;
4655 
4656 	return write_demux_regids(uindices);
4657 }
4658 
4659 #define KVM_ARM_FEATURE_ID_RANGE_INDEX(r)			\
4660 	KVM_ARM_FEATURE_ID_RANGE_IDX(sys_reg_Op0(r),		\
4661 		sys_reg_Op1(r),					\
4662 		sys_reg_CRn(r),					\
4663 		sys_reg_CRm(r),					\
4664 		sys_reg_Op2(r))
4665 
4666 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, struct reg_mask_range *range)
4667 {
4668 	const void *zero_page = page_to_virt(ZERO_PAGE(0));
4669 	u64 __user *masks = (u64 __user *)range->addr;
4670 
4671 	/* Only feature id range is supported, reserved[13] must be zero. */
4672 	if (range->range ||
4673 	    memcmp(range->reserved, zero_page, sizeof(range->reserved)))
4674 		return -EINVAL;
4675 
4676 	/* Wipe the whole thing first */
4677 	if (clear_user(masks, KVM_ARM_FEATURE_ID_RANGE_SIZE * sizeof(__u64)))
4678 		return -EFAULT;
4679 
4680 	for (int i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4681 		const struct sys_reg_desc *reg = &sys_reg_descs[i];
4682 		u32 encoding = reg_to_encoding(reg);
4683 		u64 val;
4684 
4685 		if (!is_feature_id_reg(encoding) || !reg->set_user)
4686 			continue;
4687 
4688 		if (!reg->val ||
4689 		    (is_aa32_id_reg(encoding) && !kvm_supports_32bit_el0())) {
4690 			continue;
4691 		}
4692 		val = reg->val;
4693 
4694 		if (put_user(val, (masks + KVM_ARM_FEATURE_ID_RANGE_INDEX(encoding))))
4695 			return -EFAULT;
4696 	}
4697 
4698 	return 0;
4699 }
4700 
4701 static void vcpu_set_hcr(struct kvm_vcpu *vcpu)
4702 {
4703 	struct kvm *kvm = vcpu->kvm;
4704 
4705 	if (has_vhe() || has_hvhe())
4706 		vcpu->arch.hcr_el2 |= HCR_E2H;
4707 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN)) {
4708 		/* route synchronous external abort exceptions to EL2 */
4709 		vcpu->arch.hcr_el2 |= HCR_TEA;
4710 		/* trap error record accesses */
4711 		vcpu->arch.hcr_el2 |= HCR_TERR;
4712 	}
4713 
4714 	if (cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
4715 		vcpu->arch.hcr_el2 |= HCR_FWB;
4716 
4717 	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
4718 	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
4719 	    kvm_read_vm_id_reg(kvm, SYS_CTR_EL0) == read_sanitised_ftr_reg(SYS_CTR_EL0))
4720 		vcpu->arch.hcr_el2 |= HCR_TID4;
4721 	else
4722 		vcpu->arch.hcr_el2 |= HCR_TID2;
4723 
4724 	if (vcpu_el1_is_32bit(vcpu))
4725 		vcpu->arch.hcr_el2 &= ~HCR_RW;
4726 
4727 	if (kvm_has_mte(vcpu->kvm))
4728 		vcpu->arch.hcr_el2 |= HCR_ATA;
4729 
4730 	/*
4731 	 * In the absence of FGT, we cannot independently trap TLBI
4732 	 * Range instructions. This isn't great, but trapping all
4733 	 * TLBIs would be far worse. Live with it...
4734 	 */
4735 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
4736 		vcpu->arch.hcr_el2 |= HCR_TTLBOS;
4737 }
4738 
4739 void kvm_calculate_traps(struct kvm_vcpu *vcpu)
4740 {
4741 	struct kvm *kvm = vcpu->kvm;
4742 
4743 	mutex_lock(&kvm->arch.config_lock);
4744 	vcpu_set_hcr(vcpu);
4745 	vcpu_set_ich_hcr(vcpu);
4746 
4747 	if (cpus_have_final_cap(ARM64_HAS_HCX)) {
4748 		/*
4749 		 * In general, all HCRX_EL2 bits are gated by a feature.
4750 		 * The only reason we can set SMPME without checking any
4751 		 * feature is that its effects are not directly observable
4752 		 * from the guest.
4753 		 */
4754 		vcpu->arch.hcrx_el2 = HCRX_EL2_SMPME;
4755 
4756 		if (kvm_has_feat(kvm, ID_AA64ISAR2_EL1, MOPS, IMP))
4757 			vcpu->arch.hcrx_el2 |= (HCRX_EL2_MSCEn | HCRX_EL2_MCE2);
4758 
4759 		if (kvm_has_feat(kvm, ID_AA64MMFR3_EL1, TCRX, IMP))
4760 			vcpu->arch.hcrx_el2 |= HCRX_EL2_TCR2En;
4761 
4762 		if (kvm_has_fpmr(kvm))
4763 			vcpu->arch.hcrx_el2 |= HCRX_EL2_EnFPM;
4764 	}
4765 
4766 	if (test_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags))
4767 		goto out;
4768 
4769 	kvm->arch.fgu[HFGxTR_GROUP] = (HFGxTR_EL2_nAMAIR2_EL1		|
4770 				       HFGxTR_EL2_nMAIR2_EL1		|
4771 				       HFGxTR_EL2_nS2POR_EL1		|
4772 				       HFGxTR_EL2_nACCDATA_EL1		|
4773 				       HFGxTR_EL2_nSMPRI_EL1_MASK	|
4774 				       HFGxTR_EL2_nTPIDR2_EL0_MASK);
4775 
4776 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
4777 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_TLBIRVAALE1OS|
4778 						HFGITR_EL2_TLBIRVALE1OS	|
4779 						HFGITR_EL2_TLBIRVAAE1OS	|
4780 						HFGITR_EL2_TLBIRVAE1OS	|
4781 						HFGITR_EL2_TLBIVAALE1OS	|
4782 						HFGITR_EL2_TLBIVALE1OS	|
4783 						HFGITR_EL2_TLBIVAAE1OS	|
4784 						HFGITR_EL2_TLBIASIDE1OS	|
4785 						HFGITR_EL2_TLBIVAE1OS	|
4786 						HFGITR_EL2_TLBIVMALLE1OS);
4787 
4788 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
4789 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_TLBIRVAALE1	|
4790 						HFGITR_EL2_TLBIRVALE1	|
4791 						HFGITR_EL2_TLBIRVAAE1	|
4792 						HFGITR_EL2_TLBIRVAE1	|
4793 						HFGITR_EL2_TLBIRVAALE1IS|
4794 						HFGITR_EL2_TLBIRVALE1IS	|
4795 						HFGITR_EL2_TLBIRVAAE1IS	|
4796 						HFGITR_EL2_TLBIRVAE1IS	|
4797 						HFGITR_EL2_TLBIRVAALE1OS|
4798 						HFGITR_EL2_TLBIRVALE1OS	|
4799 						HFGITR_EL2_TLBIRVAAE1OS	|
4800 						HFGITR_EL2_TLBIRVAE1OS);
4801 
4802 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, ATS1A, IMP))
4803 		kvm->arch.fgu[HFGITR_GROUP] |= HFGITR_EL2_ATS1E1A;
4804 
4805 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, PAN, PAN2))
4806 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_ATS1E1RP |
4807 						HFGITR_EL2_ATS1E1WP);
4808 
4809 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1PIE, IMP))
4810 		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nPIRE0_EL1 |
4811 						HFGxTR_EL2_nPIR_EL1);
4812 
4813 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1POE, IMP))
4814 		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nPOR_EL1 |
4815 						HFGxTR_EL2_nPOR_EL0);
4816 
4817 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, IMP))
4818 		kvm->arch.fgu[HAFGRTR_GROUP] |= ~(HAFGRTR_EL2_RES0 |
4819 						  HAFGRTR_EL2_RES1);
4820 
4821 	set_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags);
4822 out:
4823 	mutex_unlock(&kvm->arch.config_lock);
4824 }
4825 
4826 /*
4827  * Perform last adjustments to the ID registers that are implied by the
4828  * configuration outside of the ID regs themselves, as well as any
4829  * initialisation that directly depend on these ID registers (such as
4830  * RES0/RES1 behaviours). This is not the place to configure traps though.
4831  *
4832  * Because this can be called once per CPU, changes must be idempotent.
4833  */
4834 int kvm_finalize_sys_regs(struct kvm_vcpu *vcpu)
4835 {
4836 	struct kvm *kvm = vcpu->kvm;
4837 
4838 	guard(mutex)(&kvm->arch.config_lock);
4839 
4840 	if (!(static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif) &&
4841 	      irqchip_in_kernel(kvm) &&
4842 	      kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3)) {
4843 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)] &= ~ID_AA64PFR0_EL1_GIC_MASK;
4844 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_PFR1_EL1)] &= ~ID_PFR1_EL1_GIC_MASK;
4845 	}
4846 
4847 	if (vcpu_has_nv(vcpu)) {
4848 		int ret = kvm_init_nv_sysregs(kvm);
4849 		if (ret)
4850 			return ret;
4851 	}
4852 
4853 	return 0;
4854 }
4855 
4856 int __init kvm_sys_reg_table_init(void)
4857 {
4858 	bool valid = true;
4859 	unsigned int i;
4860 	int ret = 0;
4861 
4862 	/* Make sure tables are unique and in order. */
4863 	valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false);
4864 	valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true);
4865 	valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true);
4866 	valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true);
4867 	valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true);
4868 	valid &= check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false);
4869 	valid &= check_sysreg_table(sys_insn_descs, ARRAY_SIZE(sys_insn_descs), false);
4870 
4871 	if (!valid)
4872 		return -EINVAL;
4873 
4874 	/* We abuse the reset function to overwrite the table itself. */
4875 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
4876 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
4877 
4878 	ret = populate_nv_trap_config();
4879 
4880 	for (i = 0; !ret && i < ARRAY_SIZE(sys_reg_descs); i++)
4881 		ret = populate_sysreg_config(sys_reg_descs + i, i);
4882 
4883 	for (i = 0; !ret && i < ARRAY_SIZE(sys_insn_descs); i++)
4884 		ret = populate_sysreg_config(sys_insn_descs + i, i);
4885 
4886 	return ret;
4887 }
4888