xref: /linux/arch/arm64/kvm/sys_regs.c (revision b3ad940a088761fd183dccd65c6ee20b360e8c4b)
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(TCR2_EL2,    TCR2_EL1,    NULL	     );
138 		MAPPED_EL2_SYSREG(PIR_EL2,     PIR_EL1,     NULL	     );
139 		MAPPED_EL2_SYSREG(PIRE0_EL2,   PIRE0_EL1,   NULL	     );
140 		MAPPED_EL2_SYSREG(AMAIR_EL2,   AMAIR_EL1,   NULL	     );
141 		MAPPED_EL2_SYSREG(ELR_EL2,     ELR_EL1,	    NULL	     );
142 		MAPPED_EL2_SYSREG(SPSR_EL2,    SPSR_EL1,    NULL	     );
143 		MAPPED_EL2_SYSREG(ZCR_EL2,     ZCR_EL1,     NULL	     );
144 		MAPPED_EL2_SYSREG(CONTEXTIDR_EL2, CONTEXTIDR_EL1, NULL	     );
145 	default:
146 		return false;
147 	}
148 }
149 
150 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
151 {
152 	u64 val = 0x8badf00d8badf00d;
153 	u64 (*xlate)(u64) = NULL;
154 	unsigned int el1r;
155 
156 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU))
157 		goto memory_read;
158 
159 	if (unlikely(get_el2_to_el1_mapping(reg, &el1r, &xlate))) {
160 		if (!is_hyp_ctxt(vcpu))
161 			goto memory_read;
162 
163 		/*
164 		 * CNTHCTL_EL2 requires some special treatment to
165 		 * account for the bits that can be set via CNTKCTL_EL1.
166 		 */
167 		switch (reg) {
168 		case CNTHCTL_EL2:
169 			if (vcpu_el2_e2h_is_set(vcpu)) {
170 				val = read_sysreg_el1(SYS_CNTKCTL);
171 				val &= CNTKCTL_VALID_BITS;
172 				val |= __vcpu_sys_reg(vcpu, reg) & ~CNTKCTL_VALID_BITS;
173 				return val;
174 			}
175 			break;
176 		}
177 
178 		/*
179 		 * If this register does not have an EL1 counterpart,
180 		 * then read the stored EL2 version.
181 		 */
182 		if (reg == el1r)
183 			goto memory_read;
184 
185 		/*
186 		 * If we have a non-VHE guest and that the sysreg
187 		 * requires translation to be used at EL1, use the
188 		 * in-memory copy instead.
189 		 */
190 		if (!vcpu_el2_e2h_is_set(vcpu) && xlate)
191 			goto memory_read;
192 
193 		/* Get the current version of the EL1 counterpart. */
194 		WARN_ON(!__vcpu_read_sys_reg_from_cpu(el1r, &val));
195 		if (reg >= __SANITISED_REG_START__)
196 			val = kvm_vcpu_apply_reg_masks(vcpu, reg, val);
197 
198 		return val;
199 	}
200 
201 	/* EL1 register can't be on the CPU if the guest is in vEL2. */
202 	if (unlikely(is_hyp_ctxt(vcpu)))
203 		goto memory_read;
204 
205 	if (__vcpu_read_sys_reg_from_cpu(reg, &val))
206 		return val;
207 
208 memory_read:
209 	return __vcpu_sys_reg(vcpu, reg);
210 }
211 
212 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
213 {
214 	u64 (*xlate)(u64) = NULL;
215 	unsigned int el1r;
216 
217 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU))
218 		goto memory_write;
219 
220 	if (unlikely(get_el2_to_el1_mapping(reg, &el1r, &xlate))) {
221 		if (!is_hyp_ctxt(vcpu))
222 			goto memory_write;
223 
224 		/*
225 		 * Always store a copy of the write to memory to avoid having
226 		 * to reverse-translate virtual EL2 system registers for a
227 		 * non-VHE guest hypervisor.
228 		 */
229 		__vcpu_sys_reg(vcpu, reg) = val;
230 
231 		switch (reg) {
232 		case CNTHCTL_EL2:
233 			/*
234 			 * If E2H=0, CNHTCTL_EL2 is a pure shadow register.
235 			 * Otherwise, some of the bits are backed by
236 			 * CNTKCTL_EL1, while the rest is kept in memory.
237 			 * Yes, this is fun stuff.
238 			 */
239 			if (vcpu_el2_e2h_is_set(vcpu))
240 				write_sysreg_el1(val, SYS_CNTKCTL);
241 			return;
242 		}
243 
244 		/* No EL1 counterpart? We're done here.? */
245 		if (reg == el1r)
246 			return;
247 
248 		if (!vcpu_el2_e2h_is_set(vcpu) && xlate)
249 			val = xlate(val);
250 
251 		/* Redirect this to the EL1 version of the register. */
252 		WARN_ON(!__vcpu_write_sys_reg_to_cpu(val, el1r));
253 		return;
254 	}
255 
256 	/* EL1 register can't be on the CPU if the guest is in vEL2. */
257 	if (unlikely(is_hyp_ctxt(vcpu)))
258 		goto memory_write;
259 
260 	if (__vcpu_write_sys_reg_to_cpu(val, reg))
261 		return;
262 
263 memory_write:
264 	 __vcpu_sys_reg(vcpu, reg) = val;
265 }
266 
267 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
268 #define CSSELR_MAX 14
269 
270 /*
271  * Returns the minimum line size for the selected cache, expressed as
272  * Log2(bytes).
273  */
274 static u8 get_min_cache_line_size(bool icache)
275 {
276 	u64 ctr = read_sanitised_ftr_reg(SYS_CTR_EL0);
277 	u8 field;
278 
279 	if (icache)
280 		field = SYS_FIELD_GET(CTR_EL0, IminLine, ctr);
281 	else
282 		field = SYS_FIELD_GET(CTR_EL0, DminLine, ctr);
283 
284 	/*
285 	 * Cache line size is represented as Log2(words) in CTR_EL0.
286 	 * Log2(bytes) can be derived with the following:
287 	 *
288 	 * Log2(words) + 2 = Log2(bytes / 4) + 2
289 	 * 		   = Log2(bytes) - 2 + 2
290 	 * 		   = Log2(bytes)
291 	 */
292 	return field + 2;
293 }
294 
295 /* Which cache CCSIDR represents depends on CSSELR value. */
296 static u32 get_ccsidr(struct kvm_vcpu *vcpu, u32 csselr)
297 {
298 	u8 line_size;
299 
300 	if (vcpu->arch.ccsidr)
301 		return vcpu->arch.ccsidr[csselr];
302 
303 	line_size = get_min_cache_line_size(csselr & CSSELR_EL1_InD);
304 
305 	/*
306 	 * Fabricate a CCSIDR value as the overriding value does not exist.
307 	 * The real CCSIDR value will not be used as it can vary by the
308 	 * physical CPU which the vcpu currently resides in.
309 	 *
310 	 * The line size is determined with get_min_cache_line_size(), which
311 	 * should be valid for all CPUs even if they have different cache
312 	 * configuration.
313 	 *
314 	 * The associativity bits are cleared, meaning the geometry of all data
315 	 * and unified caches (which are guaranteed to be PIPT and thus
316 	 * non-aliasing) are 1 set and 1 way.
317 	 * Guests should not be doing cache operations by set/way at all, and
318 	 * for this reason, we trap them and attempt to infer the intent, so
319 	 * that we can flush the entire guest's address space at the appropriate
320 	 * time. The exposed geometry minimizes the number of the traps.
321 	 * [If guests should attempt to infer aliasing properties from the
322 	 * geometry (which is not permitted by the architecture), they would
323 	 * only do so for virtually indexed caches.]
324 	 *
325 	 * We don't check if the cache level exists as it is allowed to return
326 	 * an UNKNOWN value if not.
327 	 */
328 	return SYS_FIELD_PREP(CCSIDR_EL1, LineSize, line_size - 4);
329 }
330 
331 static int set_ccsidr(struct kvm_vcpu *vcpu, u32 csselr, u32 val)
332 {
333 	u8 line_size = FIELD_GET(CCSIDR_EL1_LineSize, val) + 4;
334 	u32 *ccsidr = vcpu->arch.ccsidr;
335 	u32 i;
336 
337 	if ((val & CCSIDR_EL1_RES0) ||
338 	    line_size < get_min_cache_line_size(csselr & CSSELR_EL1_InD))
339 		return -EINVAL;
340 
341 	if (!ccsidr) {
342 		if (val == get_ccsidr(vcpu, csselr))
343 			return 0;
344 
345 		ccsidr = kmalloc_array(CSSELR_MAX, sizeof(u32), GFP_KERNEL_ACCOUNT);
346 		if (!ccsidr)
347 			return -ENOMEM;
348 
349 		for (i = 0; i < CSSELR_MAX; i++)
350 			ccsidr[i] = get_ccsidr(vcpu, i);
351 
352 		vcpu->arch.ccsidr = ccsidr;
353 	}
354 
355 	ccsidr[csselr] = val;
356 
357 	return 0;
358 }
359 
360 static bool access_rw(struct kvm_vcpu *vcpu,
361 		      struct sys_reg_params *p,
362 		      const struct sys_reg_desc *r)
363 {
364 	if (p->is_write)
365 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
366 	else
367 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
368 
369 	return true;
370 }
371 
372 /*
373  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
374  */
375 static bool access_dcsw(struct kvm_vcpu *vcpu,
376 			struct sys_reg_params *p,
377 			const struct sys_reg_desc *r)
378 {
379 	if (!p->is_write)
380 		return read_from_write_only(vcpu, p, r);
381 
382 	/*
383 	 * Only track S/W ops if we don't have FWB. It still indicates
384 	 * that the guest is a bit broken (S/W operations should only
385 	 * be done by firmware, knowing that there is only a single
386 	 * CPU left in the system, and certainly not from non-secure
387 	 * software).
388 	 */
389 	if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
390 		kvm_set_way_flush(vcpu);
391 
392 	return true;
393 }
394 
395 static bool access_dcgsw(struct kvm_vcpu *vcpu,
396 			 struct sys_reg_params *p,
397 			 const struct sys_reg_desc *r)
398 {
399 	if (!kvm_has_mte(vcpu->kvm))
400 		return undef_access(vcpu, p, r);
401 
402 	/* Treat MTE S/W ops as we treat the classic ones: with contempt */
403 	return access_dcsw(vcpu, p, r);
404 }
405 
406 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
407 {
408 	switch (r->aarch32_map) {
409 	case AA32_LO:
410 		*mask = GENMASK_ULL(31, 0);
411 		*shift = 0;
412 		break;
413 	case AA32_HI:
414 		*mask = GENMASK_ULL(63, 32);
415 		*shift = 32;
416 		break;
417 	default:
418 		*mask = GENMASK_ULL(63, 0);
419 		*shift = 0;
420 		break;
421 	}
422 }
423 
424 /*
425  * Generic accessor for VM registers. Only called as long as HCR_TVM
426  * is set. If the guest enables the MMU, we stop trapping the VM
427  * sys_regs and leave it in complete control of the caches.
428  */
429 static bool access_vm_reg(struct kvm_vcpu *vcpu,
430 			  struct sys_reg_params *p,
431 			  const struct sys_reg_desc *r)
432 {
433 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
434 	u64 val, mask, shift;
435 
436 	if (reg_to_encoding(r) == SYS_TCR2_EL1 &&
437 	    !kvm_has_feat(vcpu->kvm, ID_AA64MMFR3_EL1, TCRX, IMP))
438 		return undef_access(vcpu, p, r);
439 
440 	BUG_ON(!p->is_write);
441 
442 	get_access_mask(r, &mask, &shift);
443 
444 	if (~mask) {
445 		val = vcpu_read_sys_reg(vcpu, r->reg);
446 		val &= ~mask;
447 	} else {
448 		val = 0;
449 	}
450 
451 	val |= (p->regval & (mask >> shift)) << shift;
452 	vcpu_write_sys_reg(vcpu, val, r->reg);
453 
454 	kvm_toggle_cache(vcpu, was_enabled);
455 	return true;
456 }
457 
458 static bool access_tcr2_el2(struct kvm_vcpu *vcpu,
459 			    struct sys_reg_params *p,
460 			    const struct sys_reg_desc *r)
461 {
462 	if (!kvm_has_feat(vcpu->kvm, ID_AA64MMFR3_EL1, TCRX, IMP)) {
463 		kvm_inject_undefined(vcpu);
464 		return false;
465 	}
466 
467 	return access_rw(vcpu, p, r);
468 }
469 
470 static bool access_actlr(struct kvm_vcpu *vcpu,
471 			 struct sys_reg_params *p,
472 			 const struct sys_reg_desc *r)
473 {
474 	u64 mask, shift;
475 
476 	if (p->is_write)
477 		return ignore_write(vcpu, p);
478 
479 	get_access_mask(r, &mask, &shift);
480 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
481 
482 	return true;
483 }
484 
485 /*
486  * Trap handler for the GICv3 SGI generation system register.
487  * Forward the request to the VGIC emulation.
488  * The cp15_64 code makes sure this automatically works
489  * for both AArch64 and AArch32 accesses.
490  */
491 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
492 			   struct sys_reg_params *p,
493 			   const struct sys_reg_desc *r)
494 {
495 	bool g1;
496 
497 	if (!kvm_has_gicv3(vcpu->kvm))
498 		return undef_access(vcpu, p, r);
499 
500 	if (!p->is_write)
501 		return read_from_write_only(vcpu, p, r);
502 
503 	/*
504 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
505 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
506 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
507 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
508 	 * group.
509 	 */
510 	if (p->Op0 == 0) {		/* AArch32 */
511 		switch (p->Op1) {
512 		default:		/* Keep GCC quiet */
513 		case 0:			/* ICC_SGI1R */
514 			g1 = true;
515 			break;
516 		case 1:			/* ICC_ASGI1R */
517 		case 2:			/* ICC_SGI0R */
518 			g1 = false;
519 			break;
520 		}
521 	} else {			/* AArch64 */
522 		switch (p->Op2) {
523 		default:		/* Keep GCC quiet */
524 		case 5:			/* ICC_SGI1R_EL1 */
525 			g1 = true;
526 			break;
527 		case 6:			/* ICC_ASGI1R_EL1 */
528 		case 7:			/* ICC_SGI0R_EL1 */
529 			g1 = false;
530 			break;
531 		}
532 	}
533 
534 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
535 
536 	return true;
537 }
538 
539 static bool access_gic_sre(struct kvm_vcpu *vcpu,
540 			   struct sys_reg_params *p,
541 			   const struct sys_reg_desc *r)
542 {
543 	if (!kvm_has_gicv3(vcpu->kvm))
544 		return undef_access(vcpu, p, r);
545 
546 	if (p->is_write)
547 		return ignore_write(vcpu, p);
548 
549 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
550 	return true;
551 }
552 
553 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
554 			struct sys_reg_params *p,
555 			const struct sys_reg_desc *r)
556 {
557 	if (p->is_write)
558 		return ignore_write(vcpu, p);
559 	else
560 		return read_zero(vcpu, p);
561 }
562 
563 /*
564  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
565  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
566  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
567  * treat it separately.
568  */
569 static bool trap_loregion(struct kvm_vcpu *vcpu,
570 			  struct sys_reg_params *p,
571 			  const struct sys_reg_desc *r)
572 {
573 	u32 sr = reg_to_encoding(r);
574 
575 	if (!kvm_has_feat(vcpu->kvm, ID_AA64MMFR1_EL1, LO, IMP))
576 		return undef_access(vcpu, p, r);
577 
578 	if (p->is_write && sr == SYS_LORID_EL1)
579 		return write_to_read_only(vcpu, p, r);
580 
581 	return trap_raz_wi(vcpu, p, r);
582 }
583 
584 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
585 			   struct sys_reg_params *p,
586 			   const struct sys_reg_desc *r)
587 {
588 	u64 oslsr;
589 
590 	if (!p->is_write)
591 		return read_from_write_only(vcpu, p, r);
592 
593 	/* Forward the OSLK bit to OSLSR */
594 	oslsr = __vcpu_sys_reg(vcpu, OSLSR_EL1) & ~OSLSR_EL1_OSLK;
595 	if (p->regval & OSLAR_EL1_OSLK)
596 		oslsr |= OSLSR_EL1_OSLK;
597 
598 	__vcpu_sys_reg(vcpu, OSLSR_EL1) = oslsr;
599 	return true;
600 }
601 
602 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
603 			   struct sys_reg_params *p,
604 			   const struct sys_reg_desc *r)
605 {
606 	if (p->is_write)
607 		return write_to_read_only(vcpu, p, r);
608 
609 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
610 	return true;
611 }
612 
613 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
614 			 u64 val)
615 {
616 	/*
617 	 * The only modifiable bit is the OSLK bit. Refuse the write if
618 	 * userspace attempts to change any other bit in the register.
619 	 */
620 	if ((val ^ rd->val) & ~OSLSR_EL1_OSLK)
621 		return -EINVAL;
622 
623 	__vcpu_sys_reg(vcpu, rd->reg) = val;
624 	return 0;
625 }
626 
627 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
628 				   struct sys_reg_params *p,
629 				   const struct sys_reg_desc *r)
630 {
631 	if (p->is_write) {
632 		return ignore_write(vcpu, p);
633 	} else {
634 		p->regval = read_sysreg(dbgauthstatus_el1);
635 		return true;
636 	}
637 }
638 
639 /*
640  * We want to avoid world-switching all the DBG registers all the
641  * time:
642  *
643  * - If we've touched any debug register, it is likely that we're
644  *   going to touch more of them. It then makes sense to disable the
645  *   traps and start doing the save/restore dance
646  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
647  *   then mandatory to save/restore the registers, as the guest
648  *   depends on them.
649  *
650  * For this, we use a DIRTY bit, indicating the guest has modified the
651  * debug registers, used as follow:
652  *
653  * On guest entry:
654  * - If the dirty bit is set (because we're coming back from trapping),
655  *   disable the traps, save host registers, restore guest registers.
656  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
657  *   set the dirty bit, disable the traps, save host registers,
658  *   restore guest registers.
659  * - Otherwise, enable the traps
660  *
661  * On guest exit:
662  * - If the dirty bit is set, save guest registers, restore host
663  *   registers and clear the dirty bit. This ensure that the host can
664  *   now use the debug registers.
665  */
666 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
667 			    struct sys_reg_params *p,
668 			    const struct sys_reg_desc *r)
669 {
670 	access_rw(vcpu, p, r);
671 	if (p->is_write)
672 		vcpu_set_flag(vcpu, DEBUG_DIRTY);
673 
674 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
675 
676 	return true;
677 }
678 
679 /*
680  * reg_to_dbg/dbg_to_reg
681  *
682  * A 32 bit write to a debug register leave top bits alone
683  * A 32 bit read from a debug register only returns the bottom bits
684  *
685  * All writes will set the DEBUG_DIRTY flag to ensure the hyp code
686  * switches between host and guest values in future.
687  */
688 static void reg_to_dbg(struct kvm_vcpu *vcpu,
689 		       struct sys_reg_params *p,
690 		       const struct sys_reg_desc *rd,
691 		       u64 *dbg_reg)
692 {
693 	u64 mask, shift, val;
694 
695 	get_access_mask(rd, &mask, &shift);
696 
697 	val = *dbg_reg;
698 	val &= ~mask;
699 	val |= (p->regval & (mask >> shift)) << shift;
700 	*dbg_reg = val;
701 
702 	vcpu_set_flag(vcpu, DEBUG_DIRTY);
703 }
704 
705 static void dbg_to_reg(struct kvm_vcpu *vcpu,
706 		       struct sys_reg_params *p,
707 		       const struct sys_reg_desc *rd,
708 		       u64 *dbg_reg)
709 {
710 	u64 mask, shift;
711 
712 	get_access_mask(rd, &mask, &shift);
713 	p->regval = (*dbg_reg & mask) >> shift;
714 }
715 
716 static bool trap_bvr(struct kvm_vcpu *vcpu,
717 		     struct sys_reg_params *p,
718 		     const struct sys_reg_desc *rd)
719 {
720 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
721 
722 	if (p->is_write)
723 		reg_to_dbg(vcpu, p, rd, dbg_reg);
724 	else
725 		dbg_to_reg(vcpu, p, rd, dbg_reg);
726 
727 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
728 
729 	return true;
730 }
731 
732 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
733 		   u64 val)
734 {
735 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = val;
736 	return 0;
737 }
738 
739 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
740 		   u64 *val)
741 {
742 	*val = vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
743 	return 0;
744 }
745 
746 static u64 reset_bvr(struct kvm_vcpu *vcpu,
747 		      const struct sys_reg_desc *rd)
748 {
749 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = rd->val;
750 	return rd->val;
751 }
752 
753 static bool trap_bcr(struct kvm_vcpu *vcpu,
754 		     struct sys_reg_params *p,
755 		     const struct sys_reg_desc *rd)
756 {
757 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
758 
759 	if (p->is_write)
760 		reg_to_dbg(vcpu, p, rd, dbg_reg);
761 	else
762 		dbg_to_reg(vcpu, p, rd, dbg_reg);
763 
764 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
765 
766 	return true;
767 }
768 
769 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
770 		   u64 val)
771 {
772 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = val;
773 	return 0;
774 }
775 
776 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
777 		   u64 *val)
778 {
779 	*val = vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
780 	return 0;
781 }
782 
783 static u64 reset_bcr(struct kvm_vcpu *vcpu,
784 		      const struct sys_reg_desc *rd)
785 {
786 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = rd->val;
787 	return rd->val;
788 }
789 
790 static bool trap_wvr(struct kvm_vcpu *vcpu,
791 		     struct sys_reg_params *p,
792 		     const struct sys_reg_desc *rd)
793 {
794 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
795 
796 	if (p->is_write)
797 		reg_to_dbg(vcpu, p, rd, dbg_reg);
798 	else
799 		dbg_to_reg(vcpu, p, rd, dbg_reg);
800 
801 	trace_trap_reg(__func__, rd->CRm, p->is_write,
802 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm]);
803 
804 	return true;
805 }
806 
807 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
808 		   u64 val)
809 {
810 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = val;
811 	return 0;
812 }
813 
814 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
815 		   u64 *val)
816 {
817 	*val = vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
818 	return 0;
819 }
820 
821 static u64 reset_wvr(struct kvm_vcpu *vcpu,
822 		      const struct sys_reg_desc *rd)
823 {
824 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = rd->val;
825 	return rd->val;
826 }
827 
828 static bool trap_wcr(struct kvm_vcpu *vcpu,
829 		     struct sys_reg_params *p,
830 		     const struct sys_reg_desc *rd)
831 {
832 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
833 
834 	if (p->is_write)
835 		reg_to_dbg(vcpu, p, rd, dbg_reg);
836 	else
837 		dbg_to_reg(vcpu, p, rd, dbg_reg);
838 
839 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
840 
841 	return true;
842 }
843 
844 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
845 		   u64 val)
846 {
847 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = val;
848 	return 0;
849 }
850 
851 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
852 		   u64 *val)
853 {
854 	*val = vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
855 	return 0;
856 }
857 
858 static u64 reset_wcr(struct kvm_vcpu *vcpu,
859 		      const struct sys_reg_desc *rd)
860 {
861 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = rd->val;
862 	return rd->val;
863 }
864 
865 static u64 reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
866 {
867 	u64 amair = read_sysreg(amair_el1);
868 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
869 	return amair;
870 }
871 
872 static u64 reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
873 {
874 	u64 actlr = read_sysreg(actlr_el1);
875 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
876 	return actlr;
877 }
878 
879 static u64 reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
880 {
881 	u64 mpidr;
882 
883 	/*
884 	 * Map the vcpu_id into the first three affinity level fields of
885 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
886 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
887 	 * of the GICv3 to be able to address each CPU directly when
888 	 * sending IPIs.
889 	 */
890 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
891 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
892 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
893 	mpidr |= (1ULL << 31);
894 	vcpu_write_sys_reg(vcpu, mpidr, MPIDR_EL1);
895 
896 	return mpidr;
897 }
898 
899 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
900 				   const struct sys_reg_desc *r)
901 {
902 	if (kvm_vcpu_has_pmu(vcpu))
903 		return 0;
904 
905 	return REG_HIDDEN;
906 }
907 
908 static u64 reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
909 {
910 	u64 mask = BIT(ARMV8_PMU_CYCLE_IDX);
911 	u8 n = vcpu->kvm->arch.pmcr_n;
912 
913 	if (n)
914 		mask |= GENMASK(n - 1, 0);
915 
916 	reset_unknown(vcpu, r);
917 	__vcpu_sys_reg(vcpu, r->reg) &= mask;
918 
919 	return __vcpu_sys_reg(vcpu, r->reg);
920 }
921 
922 static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
923 {
924 	reset_unknown(vcpu, r);
925 	__vcpu_sys_reg(vcpu, r->reg) &= GENMASK(31, 0);
926 
927 	return __vcpu_sys_reg(vcpu, r->reg);
928 }
929 
930 static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
931 {
932 	/* This thing will UNDEF, who cares about the reset value? */
933 	if (!kvm_vcpu_has_pmu(vcpu))
934 		return 0;
935 
936 	reset_unknown(vcpu, r);
937 	__vcpu_sys_reg(vcpu, r->reg) &= kvm_pmu_evtyper_mask(vcpu->kvm);
938 
939 	return __vcpu_sys_reg(vcpu, r->reg);
940 }
941 
942 static u64 reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
943 {
944 	reset_unknown(vcpu, r);
945 	__vcpu_sys_reg(vcpu, r->reg) &= PMSELR_EL0_SEL_MASK;
946 
947 	return __vcpu_sys_reg(vcpu, r->reg);
948 }
949 
950 static u64 reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
951 {
952 	u64 pmcr = 0;
953 
954 	if (!kvm_supports_32bit_el0())
955 		pmcr |= ARMV8_PMU_PMCR_LC;
956 
957 	/*
958 	 * The value of PMCR.N field is included when the
959 	 * vCPU register is read via kvm_vcpu_read_pmcr().
960 	 */
961 	__vcpu_sys_reg(vcpu, r->reg) = pmcr;
962 
963 	return __vcpu_sys_reg(vcpu, r->reg);
964 }
965 
966 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
967 {
968 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
969 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
970 
971 	if (!enabled)
972 		kvm_inject_undefined(vcpu);
973 
974 	return !enabled;
975 }
976 
977 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
978 {
979 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
980 }
981 
982 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
983 {
984 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
985 }
986 
987 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
988 {
989 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
990 }
991 
992 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
993 {
994 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
995 }
996 
997 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
998 			const struct sys_reg_desc *r)
999 {
1000 	u64 val;
1001 
1002 	if (pmu_access_el0_disabled(vcpu))
1003 		return false;
1004 
1005 	if (p->is_write) {
1006 		/*
1007 		 * Only update writeable bits of PMCR (continuing into
1008 		 * kvm_pmu_handle_pmcr() as well)
1009 		 */
1010 		val = kvm_vcpu_read_pmcr(vcpu);
1011 		val &= ~ARMV8_PMU_PMCR_MASK;
1012 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
1013 		if (!kvm_supports_32bit_el0())
1014 			val |= ARMV8_PMU_PMCR_LC;
1015 		kvm_pmu_handle_pmcr(vcpu, val);
1016 	} else {
1017 		/* PMCR.P & PMCR.C are RAZ */
1018 		val = kvm_vcpu_read_pmcr(vcpu)
1019 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
1020 		p->regval = val;
1021 	}
1022 
1023 	return true;
1024 }
1025 
1026 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1027 			  const struct sys_reg_desc *r)
1028 {
1029 	if (pmu_access_event_counter_el0_disabled(vcpu))
1030 		return false;
1031 
1032 	if (p->is_write)
1033 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
1034 	else
1035 		/* return PMSELR.SEL field */
1036 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
1037 			    & PMSELR_EL0_SEL_MASK;
1038 
1039 	return true;
1040 }
1041 
1042 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1043 			  const struct sys_reg_desc *r)
1044 {
1045 	u64 pmceid, mask, shift;
1046 
1047 	BUG_ON(p->is_write);
1048 
1049 	if (pmu_access_el0_disabled(vcpu))
1050 		return false;
1051 
1052 	get_access_mask(r, &mask, &shift);
1053 
1054 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
1055 	pmceid &= mask;
1056 	pmceid >>= shift;
1057 
1058 	p->regval = pmceid;
1059 
1060 	return true;
1061 }
1062 
1063 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
1064 {
1065 	u64 pmcr, val;
1066 
1067 	pmcr = kvm_vcpu_read_pmcr(vcpu);
1068 	val = FIELD_GET(ARMV8_PMU_PMCR_N, pmcr);
1069 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
1070 		kvm_inject_undefined(vcpu);
1071 		return false;
1072 	}
1073 
1074 	return true;
1075 }
1076 
1077 static int get_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1078 			  u64 *val)
1079 {
1080 	u64 idx;
1081 
1082 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1083 		/* PMCCNTR_EL0 */
1084 		idx = ARMV8_PMU_CYCLE_IDX;
1085 	else
1086 		/* PMEVCNTRn_EL0 */
1087 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1088 
1089 	*val = kvm_pmu_get_counter_value(vcpu, idx);
1090 	return 0;
1091 }
1092 
1093 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
1094 			      struct sys_reg_params *p,
1095 			      const struct sys_reg_desc *r)
1096 {
1097 	u64 idx = ~0UL;
1098 
1099 	if (r->CRn == 9 && r->CRm == 13) {
1100 		if (r->Op2 == 2) {
1101 			/* PMXEVCNTR_EL0 */
1102 			if (pmu_access_event_counter_el0_disabled(vcpu))
1103 				return false;
1104 
1105 			idx = SYS_FIELD_GET(PMSELR_EL0, SEL,
1106 					    __vcpu_sys_reg(vcpu, PMSELR_EL0));
1107 		} else if (r->Op2 == 0) {
1108 			/* PMCCNTR_EL0 */
1109 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
1110 				return false;
1111 
1112 			idx = ARMV8_PMU_CYCLE_IDX;
1113 		}
1114 	} else if (r->CRn == 0 && r->CRm == 9) {
1115 		/* PMCCNTR */
1116 		if (pmu_access_event_counter_el0_disabled(vcpu))
1117 			return false;
1118 
1119 		idx = ARMV8_PMU_CYCLE_IDX;
1120 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
1121 		/* PMEVCNTRn_EL0 */
1122 		if (pmu_access_event_counter_el0_disabled(vcpu))
1123 			return false;
1124 
1125 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1126 	}
1127 
1128 	/* Catch any decoding mistake */
1129 	WARN_ON(idx == ~0UL);
1130 
1131 	if (!pmu_counter_idx_valid(vcpu, idx))
1132 		return false;
1133 
1134 	if (p->is_write) {
1135 		if (pmu_access_el0_disabled(vcpu))
1136 			return false;
1137 
1138 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
1139 	} else {
1140 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
1141 	}
1142 
1143 	return true;
1144 }
1145 
1146 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1147 			       const struct sys_reg_desc *r)
1148 {
1149 	u64 idx, reg;
1150 
1151 	if (pmu_access_el0_disabled(vcpu))
1152 		return false;
1153 
1154 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
1155 		/* PMXEVTYPER_EL0 */
1156 		idx = SYS_FIELD_GET(PMSELR_EL0, SEL, __vcpu_sys_reg(vcpu, PMSELR_EL0));
1157 		reg = PMEVTYPER0_EL0 + idx;
1158 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
1159 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1160 		if (idx == ARMV8_PMU_CYCLE_IDX)
1161 			reg = PMCCFILTR_EL0;
1162 		else
1163 			/* PMEVTYPERn_EL0 */
1164 			reg = PMEVTYPER0_EL0 + idx;
1165 	} else {
1166 		BUG();
1167 	}
1168 
1169 	if (!pmu_counter_idx_valid(vcpu, idx))
1170 		return false;
1171 
1172 	if (p->is_write) {
1173 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
1174 		kvm_vcpu_pmu_restore_guest(vcpu);
1175 	} else {
1176 		p->regval = __vcpu_sys_reg(vcpu, reg);
1177 	}
1178 
1179 	return true;
1180 }
1181 
1182 static int set_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 val)
1183 {
1184 	bool set;
1185 
1186 	val &= kvm_pmu_valid_counter_mask(vcpu);
1187 
1188 	switch (r->reg) {
1189 	case PMOVSSET_EL0:
1190 		/* CRm[1] being set indicates a SET register, and CLR otherwise */
1191 		set = r->CRm & 2;
1192 		break;
1193 	default:
1194 		/* Op2[0] being set indicates a SET register, and CLR otherwise */
1195 		set = r->Op2 & 1;
1196 		break;
1197 	}
1198 
1199 	if (set)
1200 		__vcpu_sys_reg(vcpu, r->reg) |= val;
1201 	else
1202 		__vcpu_sys_reg(vcpu, r->reg) &= ~val;
1203 
1204 	return 0;
1205 }
1206 
1207 static int get_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 *val)
1208 {
1209 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
1210 
1211 	*val = __vcpu_sys_reg(vcpu, r->reg) & mask;
1212 	return 0;
1213 }
1214 
1215 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1216 			   const struct sys_reg_desc *r)
1217 {
1218 	u64 val, mask;
1219 
1220 	if (pmu_access_el0_disabled(vcpu))
1221 		return false;
1222 
1223 	mask = kvm_pmu_valid_counter_mask(vcpu);
1224 	if (p->is_write) {
1225 		val = p->regval & mask;
1226 		if (r->Op2 & 0x1) {
1227 			/* accessing PMCNTENSET_EL0 */
1228 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
1229 			kvm_pmu_enable_counter_mask(vcpu, val);
1230 			kvm_vcpu_pmu_restore_guest(vcpu);
1231 		} else {
1232 			/* accessing PMCNTENCLR_EL0 */
1233 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
1234 			kvm_pmu_disable_counter_mask(vcpu, val);
1235 		}
1236 	} else {
1237 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
1238 	}
1239 
1240 	return true;
1241 }
1242 
1243 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1244 			   const struct sys_reg_desc *r)
1245 {
1246 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
1247 
1248 	if (check_pmu_access_disabled(vcpu, 0))
1249 		return false;
1250 
1251 	if (p->is_write) {
1252 		u64 val = p->regval & mask;
1253 
1254 		if (r->Op2 & 0x1)
1255 			/* accessing PMINTENSET_EL1 */
1256 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
1257 		else
1258 			/* accessing PMINTENCLR_EL1 */
1259 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
1260 	} else {
1261 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
1262 	}
1263 
1264 	return true;
1265 }
1266 
1267 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1268 			 const struct sys_reg_desc *r)
1269 {
1270 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
1271 
1272 	if (pmu_access_el0_disabled(vcpu))
1273 		return false;
1274 
1275 	if (p->is_write) {
1276 		if (r->CRm & 0x2)
1277 			/* accessing PMOVSSET_EL0 */
1278 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
1279 		else
1280 			/* accessing PMOVSCLR_EL0 */
1281 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
1282 	} else {
1283 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
1284 	}
1285 
1286 	return true;
1287 }
1288 
1289 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1290 			   const struct sys_reg_desc *r)
1291 {
1292 	u64 mask;
1293 
1294 	if (!p->is_write)
1295 		return read_from_write_only(vcpu, p, r);
1296 
1297 	if (pmu_write_swinc_el0_disabled(vcpu))
1298 		return false;
1299 
1300 	mask = kvm_pmu_valid_counter_mask(vcpu);
1301 	kvm_pmu_software_increment(vcpu, p->regval & mask);
1302 	return true;
1303 }
1304 
1305 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1306 			     const struct sys_reg_desc *r)
1307 {
1308 	if (p->is_write) {
1309 		if (!vcpu_mode_priv(vcpu))
1310 			return undef_access(vcpu, p, r);
1311 
1312 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
1313 			       p->regval & ARMV8_PMU_USERENR_MASK;
1314 	} else {
1315 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1316 			    & ARMV8_PMU_USERENR_MASK;
1317 	}
1318 
1319 	return true;
1320 }
1321 
1322 static int get_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1323 		    u64 *val)
1324 {
1325 	*val = kvm_vcpu_read_pmcr(vcpu);
1326 	return 0;
1327 }
1328 
1329 static int set_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1330 		    u64 val)
1331 {
1332 	u8 new_n = FIELD_GET(ARMV8_PMU_PMCR_N, val);
1333 	struct kvm *kvm = vcpu->kvm;
1334 
1335 	mutex_lock(&kvm->arch.config_lock);
1336 
1337 	/*
1338 	 * The vCPU can't have more counters than the PMU hardware
1339 	 * implements. Ignore this error to maintain compatibility
1340 	 * with the existing KVM behavior.
1341 	 */
1342 	if (!kvm_vm_has_ran_once(kvm) &&
1343 	    new_n <= kvm_arm_pmu_get_max_counters(kvm))
1344 		kvm->arch.pmcr_n = new_n;
1345 
1346 	mutex_unlock(&kvm->arch.config_lock);
1347 
1348 	/*
1349 	 * Ignore writes to RES0 bits, read only bits that are cleared on
1350 	 * vCPU reset, and writable bits that KVM doesn't support yet.
1351 	 * (i.e. only PMCR.N and bits [7:0] are mutable from userspace)
1352 	 * The LP bit is RES0 when FEAT_PMUv3p5 is not supported on the vCPU.
1353 	 * But, we leave the bit as it is here, as the vCPU's PMUver might
1354 	 * be changed later (NOTE: the bit will be cleared on first vCPU run
1355 	 * if necessary).
1356 	 */
1357 	val &= ARMV8_PMU_PMCR_MASK;
1358 
1359 	/* The LC bit is RES1 when AArch32 is not supported */
1360 	if (!kvm_supports_32bit_el0())
1361 		val |= ARMV8_PMU_PMCR_LC;
1362 
1363 	__vcpu_sys_reg(vcpu, r->reg) = val;
1364 	return 0;
1365 }
1366 
1367 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1368 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1369 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1370 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1371 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1372 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1373 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1374 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1375 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1376 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1377 
1378 #define PMU_SYS_REG(name)						\
1379 	SYS_DESC(SYS_##name), .reset = reset_pmu_reg,			\
1380 	.visibility = pmu_visibility
1381 
1382 /* Macro to expand the PMEVCNTRn_EL0 register */
1383 #define PMU_PMEVCNTR_EL0(n)						\
1384 	{ PMU_SYS_REG(PMEVCNTRn_EL0(n)),				\
1385 	  .reset = reset_pmevcntr, .get_user = get_pmu_evcntr,		\
1386 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
1387 
1388 /* Macro to expand the PMEVTYPERn_EL0 register */
1389 #define PMU_PMEVTYPER_EL0(n)						\
1390 	{ PMU_SYS_REG(PMEVTYPERn_EL0(n)),				\
1391 	  .reset = reset_pmevtyper,					\
1392 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
1393 
1394 /* Macro to expand the AMU counter and type registers*/
1395 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1396 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1397 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1398 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1399 
1400 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1401 			const struct sys_reg_desc *rd)
1402 {
1403 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1404 }
1405 
1406 /*
1407  * If we land here on a PtrAuth access, that is because we didn't
1408  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1409  * way this happens is when the guest does not have PtrAuth support
1410  * enabled.
1411  */
1412 #define __PTRAUTH_KEY(k)						\
1413 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1414 	.visibility = ptrauth_visibility}
1415 
1416 #define PTRAUTH_KEY(k)							\
1417 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1418 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1419 
1420 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1421 			      struct sys_reg_params *p,
1422 			      const struct sys_reg_desc *r)
1423 {
1424 	enum kvm_arch_timers tmr;
1425 	enum kvm_arch_timer_regs treg;
1426 	u64 reg = reg_to_encoding(r);
1427 
1428 	switch (reg) {
1429 	case SYS_CNTP_TVAL_EL0:
1430 	case SYS_AARCH32_CNTP_TVAL:
1431 		tmr = TIMER_PTIMER;
1432 		treg = TIMER_REG_TVAL;
1433 		break;
1434 	case SYS_CNTP_CTL_EL0:
1435 	case SYS_AARCH32_CNTP_CTL:
1436 		tmr = TIMER_PTIMER;
1437 		treg = TIMER_REG_CTL;
1438 		break;
1439 	case SYS_CNTP_CVAL_EL0:
1440 	case SYS_AARCH32_CNTP_CVAL:
1441 		tmr = TIMER_PTIMER;
1442 		treg = TIMER_REG_CVAL;
1443 		break;
1444 	case SYS_CNTPCT_EL0:
1445 	case SYS_CNTPCTSS_EL0:
1446 	case SYS_AARCH32_CNTPCT:
1447 		tmr = TIMER_PTIMER;
1448 		treg = TIMER_REG_CNT;
1449 		break;
1450 	default:
1451 		print_sys_reg_msg(p, "%s", "Unhandled trapped timer register");
1452 		return undef_access(vcpu, p, r);
1453 	}
1454 
1455 	if (p->is_write)
1456 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1457 	else
1458 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1459 
1460 	return true;
1461 }
1462 
1463 static s64 kvm_arm64_ftr_safe_value(u32 id, const struct arm64_ftr_bits *ftrp,
1464 				    s64 new, s64 cur)
1465 {
1466 	struct arm64_ftr_bits kvm_ftr = *ftrp;
1467 
1468 	/* Some features have different safe value type in KVM than host features */
1469 	switch (id) {
1470 	case SYS_ID_AA64DFR0_EL1:
1471 		switch (kvm_ftr.shift) {
1472 		case ID_AA64DFR0_EL1_PMUVer_SHIFT:
1473 			kvm_ftr.type = FTR_LOWER_SAFE;
1474 			break;
1475 		case ID_AA64DFR0_EL1_DebugVer_SHIFT:
1476 			kvm_ftr.type = FTR_LOWER_SAFE;
1477 			break;
1478 		}
1479 		break;
1480 	case SYS_ID_DFR0_EL1:
1481 		if (kvm_ftr.shift == ID_DFR0_EL1_PerfMon_SHIFT)
1482 			kvm_ftr.type = FTR_LOWER_SAFE;
1483 		break;
1484 	}
1485 
1486 	return arm64_ftr_safe_value(&kvm_ftr, new, cur);
1487 }
1488 
1489 /*
1490  * arm64_check_features() - Check if a feature register value constitutes
1491  * a subset of features indicated by the idreg's KVM sanitised limit.
1492  *
1493  * This function will check if each feature field of @val is the "safe" value
1494  * against idreg's KVM sanitised limit return from reset() callback.
1495  * If a field value in @val is the same as the one in limit, it is always
1496  * considered the safe value regardless For register fields that are not in
1497  * writable, only the value in limit is considered the safe value.
1498  *
1499  * Return: 0 if all the fields are safe. Otherwise, return negative errno.
1500  */
1501 static int arm64_check_features(struct kvm_vcpu *vcpu,
1502 				const struct sys_reg_desc *rd,
1503 				u64 val)
1504 {
1505 	const struct arm64_ftr_reg *ftr_reg;
1506 	const struct arm64_ftr_bits *ftrp = NULL;
1507 	u32 id = reg_to_encoding(rd);
1508 	u64 writable_mask = rd->val;
1509 	u64 limit = rd->reset(vcpu, rd);
1510 	u64 mask = 0;
1511 
1512 	/*
1513 	 * Hidden and unallocated ID registers may not have a corresponding
1514 	 * struct arm64_ftr_reg. Of course, if the register is RAZ we know the
1515 	 * only safe value is 0.
1516 	 */
1517 	if (sysreg_visible_as_raz(vcpu, rd))
1518 		return val ? -E2BIG : 0;
1519 
1520 	ftr_reg = get_arm64_ftr_reg(id);
1521 	if (!ftr_reg)
1522 		return -EINVAL;
1523 
1524 	ftrp = ftr_reg->ftr_bits;
1525 
1526 	for (; ftrp && ftrp->width; ftrp++) {
1527 		s64 f_val, f_lim, safe_val;
1528 		u64 ftr_mask;
1529 
1530 		ftr_mask = arm64_ftr_mask(ftrp);
1531 		if ((ftr_mask & writable_mask) != ftr_mask)
1532 			continue;
1533 
1534 		f_val = arm64_ftr_value(ftrp, val);
1535 		f_lim = arm64_ftr_value(ftrp, limit);
1536 		mask |= ftr_mask;
1537 
1538 		if (f_val == f_lim)
1539 			safe_val = f_val;
1540 		else
1541 			safe_val = kvm_arm64_ftr_safe_value(id, ftrp, f_val, f_lim);
1542 
1543 		if (safe_val != f_val)
1544 			return -E2BIG;
1545 	}
1546 
1547 	/* For fields that are not writable, values in limit are the safe values. */
1548 	if ((val & ~mask) != (limit & ~mask))
1549 		return -E2BIG;
1550 
1551 	return 0;
1552 }
1553 
1554 static u8 pmuver_to_perfmon(u8 pmuver)
1555 {
1556 	switch (pmuver) {
1557 	case ID_AA64DFR0_EL1_PMUVer_IMP:
1558 		return ID_DFR0_EL1_PerfMon_PMUv3;
1559 	case ID_AA64DFR0_EL1_PMUVer_IMP_DEF:
1560 		return ID_DFR0_EL1_PerfMon_IMPDEF;
1561 	default:
1562 		/* Anything ARMv8.1+ and NI have the same value. For now. */
1563 		return pmuver;
1564 	}
1565 }
1566 
1567 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1568 static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu,
1569 				       const struct sys_reg_desc *r)
1570 {
1571 	u32 id = reg_to_encoding(r);
1572 	u64 val;
1573 
1574 	if (sysreg_visible_as_raz(vcpu, r))
1575 		return 0;
1576 
1577 	val = read_sanitised_ftr_reg(id);
1578 
1579 	switch (id) {
1580 	case SYS_ID_AA64PFR1_EL1:
1581 		if (!kvm_has_mte(vcpu->kvm))
1582 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTE);
1583 
1584 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_SME);
1585 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_RNDR_trap);
1586 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_NMI);
1587 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTE_frac);
1588 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_GCS);
1589 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_THE);
1590 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_MTEX);
1591 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_DF2);
1592 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_EL1_PFAR);
1593 		break;
1594 	case SYS_ID_AA64PFR2_EL1:
1595 		/* We only expose FPMR */
1596 		val &= ID_AA64PFR2_EL1_FPMR;
1597 		break;
1598 	case SYS_ID_AA64ISAR1_EL1:
1599 		if (!vcpu_has_ptrauth(vcpu))
1600 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_APA) |
1601 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_API) |
1602 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPA) |
1603 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPI));
1604 		break;
1605 	case SYS_ID_AA64ISAR2_EL1:
1606 		if (!vcpu_has_ptrauth(vcpu))
1607 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_APA3) |
1608 				 ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_GPA3));
1609 		if (!cpus_have_final_cap(ARM64_HAS_WFXT))
1610 			val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_WFxT);
1611 		break;
1612 	case SYS_ID_AA64MMFR2_EL1:
1613 		val &= ~ID_AA64MMFR2_EL1_CCIDX_MASK;
1614 		break;
1615 	case SYS_ID_AA64MMFR3_EL1:
1616 		val &= ID_AA64MMFR3_EL1_TCRX | ID_AA64MMFR3_EL1_S1POE |
1617 			ID_AA64MMFR3_EL1_S1PIE;
1618 		break;
1619 	case SYS_ID_MMFR4_EL1:
1620 		val &= ~ARM64_FEATURE_MASK(ID_MMFR4_EL1_CCIDX);
1621 		break;
1622 	}
1623 
1624 	return val;
1625 }
1626 
1627 static u64 kvm_read_sanitised_id_reg(struct kvm_vcpu *vcpu,
1628 				     const struct sys_reg_desc *r)
1629 {
1630 	return __kvm_read_sanitised_id_reg(vcpu, r);
1631 }
1632 
1633 static u64 read_id_reg(const struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1634 {
1635 	return kvm_read_vm_id_reg(vcpu->kvm, reg_to_encoding(r));
1636 }
1637 
1638 static bool is_feature_id_reg(u32 encoding)
1639 {
1640 	return (sys_reg_Op0(encoding) == 3 &&
1641 		(sys_reg_Op1(encoding) < 2 || sys_reg_Op1(encoding) == 3) &&
1642 		sys_reg_CRn(encoding) == 0 &&
1643 		sys_reg_CRm(encoding) <= 7);
1644 }
1645 
1646 /*
1647  * Return true if the register's (Op0, Op1, CRn, CRm, Op2) is
1648  * (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8, which is the range of ID
1649  * registers KVM maintains on a per-VM basis.
1650  */
1651 static inline bool is_vm_ftr_id_reg(u32 id)
1652 {
1653 	if (id == SYS_CTR_EL0)
1654 		return true;
1655 
1656 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1657 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1658 		sys_reg_CRm(id) < 8);
1659 }
1660 
1661 static inline bool is_vcpu_ftr_id_reg(u32 id)
1662 {
1663 	return is_feature_id_reg(id) && !is_vm_ftr_id_reg(id);
1664 }
1665 
1666 static inline bool is_aa32_id_reg(u32 id)
1667 {
1668 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1669 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1670 		sys_reg_CRm(id) <= 3);
1671 }
1672 
1673 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1674 				  const struct sys_reg_desc *r)
1675 {
1676 	u32 id = reg_to_encoding(r);
1677 
1678 	switch (id) {
1679 	case SYS_ID_AA64ZFR0_EL1:
1680 		if (!vcpu_has_sve(vcpu))
1681 			return REG_RAZ;
1682 		break;
1683 	}
1684 
1685 	return 0;
1686 }
1687 
1688 static unsigned int aa32_id_visibility(const struct kvm_vcpu *vcpu,
1689 				       const struct sys_reg_desc *r)
1690 {
1691 	/*
1692 	 * AArch32 ID registers are UNKNOWN if AArch32 isn't implemented at any
1693 	 * EL. Promote to RAZ/WI in order to guarantee consistency between
1694 	 * systems.
1695 	 */
1696 	if (!kvm_supports_32bit_el0())
1697 		return REG_RAZ | REG_USER_WI;
1698 
1699 	return id_visibility(vcpu, r);
1700 }
1701 
1702 static unsigned int raz_visibility(const struct kvm_vcpu *vcpu,
1703 				   const struct sys_reg_desc *r)
1704 {
1705 	return REG_RAZ;
1706 }
1707 
1708 /* cpufeature ID register access trap handlers */
1709 
1710 static bool access_id_reg(struct kvm_vcpu *vcpu,
1711 			  struct sys_reg_params *p,
1712 			  const struct sys_reg_desc *r)
1713 {
1714 	if (p->is_write)
1715 		return write_to_read_only(vcpu, p, r);
1716 
1717 	p->regval = read_id_reg(vcpu, r);
1718 
1719 	return true;
1720 }
1721 
1722 /* Visibility overrides for SVE-specific control registers */
1723 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1724 				   const struct sys_reg_desc *rd)
1725 {
1726 	if (vcpu_has_sve(vcpu))
1727 		return 0;
1728 
1729 	return REG_HIDDEN;
1730 }
1731 
1732 static unsigned int sme_visibility(const struct kvm_vcpu *vcpu,
1733 				   const struct sys_reg_desc *rd)
1734 {
1735 	if (kvm_has_feat(vcpu->kvm, ID_AA64PFR1_EL1, SME, IMP))
1736 		return 0;
1737 
1738 	return REG_HIDDEN;
1739 }
1740 
1741 static unsigned int fp8_visibility(const struct kvm_vcpu *vcpu,
1742 				   const struct sys_reg_desc *rd)
1743 {
1744 	if (kvm_has_fpmr(vcpu->kvm))
1745 		return 0;
1746 
1747 	return REG_HIDDEN;
1748 }
1749 
1750 static u64 read_sanitised_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1751 					  const struct sys_reg_desc *rd)
1752 {
1753 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1754 
1755 	if (!vcpu_has_sve(vcpu))
1756 		val &= ~ID_AA64PFR0_EL1_SVE_MASK;
1757 
1758 	/*
1759 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
1760 	 * Although this is a per-CPU feature, we make it global because
1761 	 * asymmetric systems are just a nuisance.
1762 	 *
1763 	 * Userspace can override this as long as it doesn't promise
1764 	 * the impossible.
1765 	 */
1766 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) {
1767 		val &= ~ID_AA64PFR0_EL1_CSV2_MASK;
1768 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV2, IMP);
1769 	}
1770 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) {
1771 		val &= ~ID_AA64PFR0_EL1_CSV3_MASK;
1772 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV3, IMP);
1773 	}
1774 
1775 	if (kvm_vgic_global_state.type == VGIC_V3) {
1776 		val &= ~ID_AA64PFR0_EL1_GIC_MASK;
1777 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, GIC, IMP);
1778 	}
1779 
1780 	val &= ~ID_AA64PFR0_EL1_AMU_MASK;
1781 
1782 	return val;
1783 }
1784 
1785 #define ID_REG_LIMIT_FIELD_ENUM(val, reg, field, limit)			       \
1786 ({									       \
1787 	u64 __f_val = FIELD_GET(reg##_##field##_MASK, val);		       \
1788 	(val) &= ~reg##_##field##_MASK;					       \
1789 	(val) |= FIELD_PREP(reg##_##field##_MASK,			       \
1790 			    min(__f_val,				       \
1791 				(u64)SYS_FIELD_VALUE(reg, field, limit)));     \
1792 	(val);								       \
1793 })
1794 
1795 static u64 read_sanitised_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
1796 					  const struct sys_reg_desc *rd)
1797 {
1798 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1799 
1800 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_AA64DFR0_EL1, DebugVer, V8P8);
1801 
1802 	/*
1803 	 * Only initialize the PMU version if the vCPU was configured with one.
1804 	 */
1805 	val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1806 	if (kvm_vcpu_has_pmu(vcpu))
1807 		val |= SYS_FIELD_PREP(ID_AA64DFR0_EL1, PMUVer,
1808 				      kvm_arm_pmu_get_pmuver_limit());
1809 
1810 	/* Hide SPE from guests */
1811 	val &= ~ID_AA64DFR0_EL1_PMSVer_MASK;
1812 
1813 	return val;
1814 }
1815 
1816 static int set_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
1817 			       const struct sys_reg_desc *rd,
1818 			       u64 val)
1819 {
1820 	u8 debugver = SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, val);
1821 	u8 pmuver = SYS_FIELD_GET(ID_AA64DFR0_EL1, PMUVer, val);
1822 
1823 	/*
1824 	 * Prior to commit 3d0dba5764b9 ("KVM: arm64: PMU: Move the
1825 	 * ID_AA64DFR0_EL1.PMUver limit to VM creation"), KVM erroneously
1826 	 * exposed an IMP_DEF PMU to userspace and the guest on systems w/
1827 	 * non-architectural PMUs. Of course, PMUv3 is the only game in town for
1828 	 * PMU virtualization, so the IMP_DEF value was rather user-hostile.
1829 	 *
1830 	 * At minimum, we're on the hook to allow values that were given to
1831 	 * userspace by KVM. Cover our tracks here and replace the IMP_DEF value
1832 	 * with a more sensible NI. The value of an ID register changing under
1833 	 * the nose of the guest is unfortunate, but is certainly no more
1834 	 * surprising than an ill-guided PMU driver poking at impdef system
1835 	 * registers that end in an UNDEF...
1836 	 */
1837 	if (pmuver == ID_AA64DFR0_EL1_PMUVer_IMP_DEF)
1838 		val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1839 
1840 	/*
1841 	 * ID_AA64DFR0_EL1.DebugVer is one of those awkward fields with a
1842 	 * nonzero minimum safe value.
1843 	 */
1844 	if (debugver < ID_AA64DFR0_EL1_DebugVer_IMP)
1845 		return -EINVAL;
1846 
1847 	return set_id_reg(vcpu, rd, val);
1848 }
1849 
1850 static u64 read_sanitised_id_dfr0_el1(struct kvm_vcpu *vcpu,
1851 				      const struct sys_reg_desc *rd)
1852 {
1853 	u8 perfmon = pmuver_to_perfmon(kvm_arm_pmu_get_pmuver_limit());
1854 	u64 val = read_sanitised_ftr_reg(SYS_ID_DFR0_EL1);
1855 
1856 	val &= ~ID_DFR0_EL1_PerfMon_MASK;
1857 	if (kvm_vcpu_has_pmu(vcpu))
1858 		val |= SYS_FIELD_PREP(ID_DFR0_EL1, PerfMon, perfmon);
1859 
1860 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_DFR0_EL1, CopDbg, Debugv8p8);
1861 
1862 	return val;
1863 }
1864 
1865 static int set_id_dfr0_el1(struct kvm_vcpu *vcpu,
1866 			   const struct sys_reg_desc *rd,
1867 			   u64 val)
1868 {
1869 	u8 perfmon = SYS_FIELD_GET(ID_DFR0_EL1, PerfMon, val);
1870 	u8 copdbg = SYS_FIELD_GET(ID_DFR0_EL1, CopDbg, val);
1871 
1872 	if (perfmon == ID_DFR0_EL1_PerfMon_IMPDEF) {
1873 		val &= ~ID_DFR0_EL1_PerfMon_MASK;
1874 		perfmon = 0;
1875 	}
1876 
1877 	/*
1878 	 * Allow DFR0_EL1.PerfMon to be set from userspace as long as
1879 	 * it doesn't promise more than what the HW gives us on the
1880 	 * AArch64 side (as everything is emulated with that), and
1881 	 * that this is a PMUv3.
1882 	 */
1883 	if (perfmon != 0 && perfmon < ID_DFR0_EL1_PerfMon_PMUv3)
1884 		return -EINVAL;
1885 
1886 	if (copdbg < ID_DFR0_EL1_CopDbg_Armv8)
1887 		return -EINVAL;
1888 
1889 	return set_id_reg(vcpu, rd, val);
1890 }
1891 
1892 /*
1893  * cpufeature ID register user accessors
1894  *
1895  * For now, these registers are immutable for userspace, so no values
1896  * are stored, and for set_id_reg() we don't allow the effective value
1897  * to be changed.
1898  */
1899 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1900 		      u64 *val)
1901 {
1902 	/*
1903 	 * Avoid locking if the VM has already started, as the ID registers are
1904 	 * guaranteed to be invariant at that point.
1905 	 */
1906 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
1907 		*val = read_id_reg(vcpu, rd);
1908 		return 0;
1909 	}
1910 
1911 	mutex_lock(&vcpu->kvm->arch.config_lock);
1912 	*val = read_id_reg(vcpu, rd);
1913 	mutex_unlock(&vcpu->kvm->arch.config_lock);
1914 
1915 	return 0;
1916 }
1917 
1918 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1919 		      u64 val)
1920 {
1921 	u32 id = reg_to_encoding(rd);
1922 	int ret;
1923 
1924 	mutex_lock(&vcpu->kvm->arch.config_lock);
1925 
1926 	/*
1927 	 * Once the VM has started the ID registers are immutable. Reject any
1928 	 * write that does not match the final register value.
1929 	 */
1930 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
1931 		if (val != read_id_reg(vcpu, rd))
1932 			ret = -EBUSY;
1933 		else
1934 			ret = 0;
1935 
1936 		mutex_unlock(&vcpu->kvm->arch.config_lock);
1937 		return ret;
1938 	}
1939 
1940 	ret = arm64_check_features(vcpu, rd, val);
1941 	if (!ret)
1942 		kvm_set_vm_id_reg(vcpu->kvm, id, val);
1943 
1944 	mutex_unlock(&vcpu->kvm->arch.config_lock);
1945 
1946 	/*
1947 	 * arm64_check_features() returns -E2BIG to indicate the register's
1948 	 * feature set is a superset of the maximally-allowed register value.
1949 	 * While it would be nice to precisely describe this to userspace, the
1950 	 * existing UAPI for KVM_SET_ONE_REG has it that invalid register
1951 	 * writes return -EINVAL.
1952 	 */
1953 	if (ret == -E2BIG)
1954 		ret = -EINVAL;
1955 	return ret;
1956 }
1957 
1958 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val)
1959 {
1960 	u64 *p = __vm_id_reg(&kvm->arch, reg);
1961 
1962 	lockdep_assert_held(&kvm->arch.config_lock);
1963 
1964 	if (KVM_BUG_ON(kvm_vm_has_ran_once(kvm) || !p, kvm))
1965 		return;
1966 
1967 	*p = val;
1968 }
1969 
1970 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1971 		       u64 *val)
1972 {
1973 	*val = 0;
1974 	return 0;
1975 }
1976 
1977 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1978 		      u64 val)
1979 {
1980 	return 0;
1981 }
1982 
1983 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1984 		       const struct sys_reg_desc *r)
1985 {
1986 	if (p->is_write)
1987 		return write_to_read_only(vcpu, p, r);
1988 
1989 	p->regval = kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0);
1990 	return true;
1991 }
1992 
1993 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1994 			 const struct sys_reg_desc *r)
1995 {
1996 	if (p->is_write)
1997 		return write_to_read_only(vcpu, p, r);
1998 
1999 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
2000 	return true;
2001 }
2002 
2003 /*
2004  * Fabricate a CLIDR_EL1 value instead of using the real value, which can vary
2005  * by the physical CPU which the vcpu currently resides in.
2006  */
2007 static u64 reset_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2008 {
2009 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2010 	u64 clidr;
2011 	u8 loc;
2012 
2013 	if ((ctr_el0 & CTR_EL0_IDC)) {
2014 		/*
2015 		 * Data cache clean to the PoU is not required so LoUU and LoUIS
2016 		 * will not be set and a unified cache, which will be marked as
2017 		 * LoC, will be added.
2018 		 *
2019 		 * If not DIC, let the unified cache L2 so that an instruction
2020 		 * cache can be added as L1 later.
2021 		 */
2022 		loc = (ctr_el0 & CTR_EL0_DIC) ? 1 : 2;
2023 		clidr = CACHE_TYPE_UNIFIED << CLIDR_CTYPE_SHIFT(loc);
2024 	} else {
2025 		/*
2026 		 * Data cache clean to the PoU is required so let L1 have a data
2027 		 * cache and mark it as LoUU and LoUIS. As L1 has a data cache,
2028 		 * it can be marked as LoC too.
2029 		 */
2030 		loc = 1;
2031 		clidr = 1 << CLIDR_LOUU_SHIFT;
2032 		clidr |= 1 << CLIDR_LOUIS_SHIFT;
2033 		clidr |= CACHE_TYPE_DATA << CLIDR_CTYPE_SHIFT(1);
2034 	}
2035 
2036 	/*
2037 	 * Instruction cache invalidation to the PoU is required so let L1 have
2038 	 * an instruction cache. If L1 already has a data cache, it will be
2039 	 * CACHE_TYPE_SEPARATE.
2040 	 */
2041 	if (!(ctr_el0 & CTR_EL0_DIC))
2042 		clidr |= CACHE_TYPE_INST << CLIDR_CTYPE_SHIFT(1);
2043 
2044 	clidr |= loc << CLIDR_LOC_SHIFT;
2045 
2046 	/*
2047 	 * Add tag cache unified to data cache. Allocation tags and data are
2048 	 * unified in a cache line so that it looks valid even if there is only
2049 	 * one cache line.
2050 	 */
2051 	if (kvm_has_mte(vcpu->kvm))
2052 		clidr |= 2ULL << CLIDR_TTYPE_SHIFT(loc);
2053 
2054 	__vcpu_sys_reg(vcpu, r->reg) = clidr;
2055 
2056 	return __vcpu_sys_reg(vcpu, r->reg);
2057 }
2058 
2059 static int set_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2060 		      u64 val)
2061 {
2062 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2063 	u64 idc = !CLIDR_LOC(val) || (!CLIDR_LOUIS(val) && !CLIDR_LOUU(val));
2064 
2065 	if ((val & CLIDR_EL1_RES0) || (!(ctr_el0 & CTR_EL0_IDC) && idc))
2066 		return -EINVAL;
2067 
2068 	__vcpu_sys_reg(vcpu, rd->reg) = val;
2069 
2070 	return 0;
2071 }
2072 
2073 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2074 			  const struct sys_reg_desc *r)
2075 {
2076 	int reg = r->reg;
2077 
2078 	if (p->is_write)
2079 		vcpu_write_sys_reg(vcpu, p->regval, reg);
2080 	else
2081 		p->regval = vcpu_read_sys_reg(vcpu, reg);
2082 	return true;
2083 }
2084 
2085 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2086 			  const struct sys_reg_desc *r)
2087 {
2088 	u32 csselr;
2089 
2090 	if (p->is_write)
2091 		return write_to_read_only(vcpu, p, r);
2092 
2093 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
2094 	csselr &= CSSELR_EL1_Level | CSSELR_EL1_InD;
2095 	if (csselr < CSSELR_MAX)
2096 		p->regval = get_ccsidr(vcpu, csselr);
2097 
2098 	return true;
2099 }
2100 
2101 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
2102 				   const struct sys_reg_desc *rd)
2103 {
2104 	if (kvm_has_mte(vcpu->kvm))
2105 		return 0;
2106 
2107 	return REG_HIDDEN;
2108 }
2109 
2110 #define MTE_REG(name) {				\
2111 	SYS_DESC(SYS_##name),			\
2112 	.access = undef_access,			\
2113 	.reset = reset_unknown,			\
2114 	.reg = name,				\
2115 	.visibility = mte_visibility,		\
2116 }
2117 
2118 static unsigned int el2_visibility(const struct kvm_vcpu *vcpu,
2119 				   const struct sys_reg_desc *rd)
2120 {
2121 	if (vcpu_has_nv(vcpu))
2122 		return 0;
2123 
2124 	return REG_HIDDEN;
2125 }
2126 
2127 static bool bad_vncr_trap(struct kvm_vcpu *vcpu,
2128 			  struct sys_reg_params *p,
2129 			  const struct sys_reg_desc *r)
2130 {
2131 	/*
2132 	 * We really shouldn't be here, and this is likely the result
2133 	 * of a misconfigured trap, as this register should target the
2134 	 * VNCR page, and nothing else.
2135 	 */
2136 	return bad_trap(vcpu, p, r,
2137 			"trap of VNCR-backed register");
2138 }
2139 
2140 static bool bad_redir_trap(struct kvm_vcpu *vcpu,
2141 			   struct sys_reg_params *p,
2142 			   const struct sys_reg_desc *r)
2143 {
2144 	/*
2145 	 * We really shouldn't be here, and this is likely the result
2146 	 * of a misconfigured trap, as this register should target the
2147 	 * corresponding EL1, and nothing else.
2148 	 */
2149 	return bad_trap(vcpu, p, r,
2150 			"trap of EL2 register redirected to EL1");
2151 }
2152 
2153 #define EL2_REG(name, acc, rst, v) {		\
2154 	SYS_DESC(SYS_##name),			\
2155 	.access = acc,				\
2156 	.reset = rst,				\
2157 	.reg = name,				\
2158 	.visibility = el2_visibility,		\
2159 	.val = v,				\
2160 }
2161 
2162 #define EL2_REG_VNCR(name, rst, v)	EL2_REG(name, bad_vncr_trap, rst, v)
2163 #define EL2_REG_REDIR(name, rst, v)	EL2_REG(name, bad_redir_trap, rst, v)
2164 
2165 /*
2166  * Since reset() callback and field val are not used for idregs, they will be
2167  * used for specific purposes for idregs.
2168  * The reset() would return KVM sanitised register value. The value would be the
2169  * same as the host kernel sanitised value if there is no KVM sanitisation.
2170  * The val would be used as a mask indicating writable fields for the idreg.
2171  * Only bits with 1 are writable from userspace. This mask might not be
2172  * necessary in the future whenever all ID registers are enabled as writable
2173  * from userspace.
2174  */
2175 
2176 #define ID_DESC(name)				\
2177 	SYS_DESC(SYS_##name),			\
2178 	.access	= access_id_reg,		\
2179 	.get_user = get_id_reg			\
2180 
2181 /* sys_reg_desc initialiser for known cpufeature ID registers */
2182 #define ID_SANITISED(name) {			\
2183 	ID_DESC(name),				\
2184 	.set_user = set_id_reg,			\
2185 	.visibility = id_visibility,		\
2186 	.reset = kvm_read_sanitised_id_reg,	\
2187 	.val = 0,				\
2188 }
2189 
2190 /* sys_reg_desc initialiser for known cpufeature ID registers */
2191 #define AA32_ID_SANITISED(name) {		\
2192 	ID_DESC(name),				\
2193 	.set_user = set_id_reg,			\
2194 	.visibility = aa32_id_visibility,	\
2195 	.reset = kvm_read_sanitised_id_reg,	\
2196 	.val = 0,				\
2197 }
2198 
2199 /* sys_reg_desc initialiser for writable ID registers */
2200 #define ID_WRITABLE(name, mask) {		\
2201 	ID_DESC(name),				\
2202 	.set_user = set_id_reg,			\
2203 	.visibility = id_visibility,		\
2204 	.reset = kvm_read_sanitised_id_reg,	\
2205 	.val = mask,				\
2206 }
2207 
2208 /*
2209  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
2210  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
2211  * (1 <= crm < 8, 0 <= Op2 < 8).
2212  */
2213 #define ID_UNALLOCATED(crm, op2) {			\
2214 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
2215 	.access = access_id_reg,			\
2216 	.get_user = get_id_reg,				\
2217 	.set_user = set_id_reg,				\
2218 	.visibility = raz_visibility,			\
2219 	.reset = kvm_read_sanitised_id_reg,		\
2220 	.val = 0,					\
2221 }
2222 
2223 /*
2224  * sys_reg_desc initialiser for known ID registers that we hide from guests.
2225  * For now, these are exposed just like unallocated ID regs: they appear
2226  * RAZ for the guest.
2227  */
2228 #define ID_HIDDEN(name) {			\
2229 	ID_DESC(name),				\
2230 	.set_user = set_id_reg,			\
2231 	.visibility = raz_visibility,		\
2232 	.reset = kvm_read_sanitised_id_reg,	\
2233 	.val = 0,				\
2234 }
2235 
2236 static bool access_sp_el1(struct kvm_vcpu *vcpu,
2237 			  struct sys_reg_params *p,
2238 			  const struct sys_reg_desc *r)
2239 {
2240 	if (p->is_write)
2241 		__vcpu_sys_reg(vcpu, SP_EL1) = p->regval;
2242 	else
2243 		p->regval = __vcpu_sys_reg(vcpu, SP_EL1);
2244 
2245 	return true;
2246 }
2247 
2248 static bool access_elr(struct kvm_vcpu *vcpu,
2249 		       struct sys_reg_params *p,
2250 		       const struct sys_reg_desc *r)
2251 {
2252 	if (p->is_write)
2253 		vcpu_write_sys_reg(vcpu, p->regval, ELR_EL1);
2254 	else
2255 		p->regval = vcpu_read_sys_reg(vcpu, ELR_EL1);
2256 
2257 	return true;
2258 }
2259 
2260 static bool access_spsr(struct kvm_vcpu *vcpu,
2261 			struct sys_reg_params *p,
2262 			const struct sys_reg_desc *r)
2263 {
2264 	if (p->is_write)
2265 		__vcpu_sys_reg(vcpu, SPSR_EL1) = p->regval;
2266 	else
2267 		p->regval = __vcpu_sys_reg(vcpu, SPSR_EL1);
2268 
2269 	return true;
2270 }
2271 
2272 static bool access_cntkctl_el12(struct kvm_vcpu *vcpu,
2273 				struct sys_reg_params *p,
2274 				const struct sys_reg_desc *r)
2275 {
2276 	if (p->is_write)
2277 		__vcpu_sys_reg(vcpu, CNTKCTL_EL1) = p->regval;
2278 	else
2279 		p->regval = __vcpu_sys_reg(vcpu, CNTKCTL_EL1);
2280 
2281 	return true;
2282 }
2283 
2284 static u64 reset_hcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2285 {
2286 	u64 val = r->val;
2287 
2288 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
2289 		val |= HCR_E2H;
2290 
2291 	return __vcpu_sys_reg(vcpu, r->reg) = val;
2292 }
2293 
2294 static unsigned int sve_el2_visibility(const struct kvm_vcpu *vcpu,
2295 				       const struct sys_reg_desc *rd)
2296 {
2297 	unsigned int r;
2298 
2299 	r = el2_visibility(vcpu, rd);
2300 	if (r)
2301 		return r;
2302 
2303 	return sve_visibility(vcpu, rd);
2304 }
2305 
2306 static bool access_zcr_el2(struct kvm_vcpu *vcpu,
2307 			   struct sys_reg_params *p,
2308 			   const struct sys_reg_desc *r)
2309 {
2310 	unsigned int vq;
2311 
2312 	if (guest_hyp_sve_traps_enabled(vcpu)) {
2313 		kvm_inject_nested_sve_trap(vcpu);
2314 		return true;
2315 	}
2316 
2317 	if (!p->is_write) {
2318 		p->regval = vcpu_read_sys_reg(vcpu, ZCR_EL2);
2319 		return true;
2320 	}
2321 
2322 	vq = SYS_FIELD_GET(ZCR_ELx, LEN, p->regval) + 1;
2323 	vq = min(vq, vcpu_sve_max_vq(vcpu));
2324 	vcpu_write_sys_reg(vcpu, vq - 1, ZCR_EL2);
2325 	return true;
2326 }
2327 
2328 static unsigned int s1poe_visibility(const struct kvm_vcpu *vcpu,
2329 				     const struct sys_reg_desc *rd)
2330 {
2331 	if (kvm_has_feat(vcpu->kvm, ID_AA64MMFR3_EL1, S1POE, IMP))
2332 		return 0;
2333 
2334 	return REG_HIDDEN;
2335 }
2336 
2337 /*
2338  * Architected system registers.
2339  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
2340  *
2341  * Debug handling: We do trap most, if not all debug related system
2342  * registers. The implementation is good enough to ensure that a guest
2343  * can use these with minimal performance degradation. The drawback is
2344  * that we don't implement any of the external debug architecture.
2345  * This should be revisited if we ever encounter a more demanding
2346  * guest...
2347  */
2348 static const struct sys_reg_desc sys_reg_descs[] = {
2349 	DBG_BCR_BVR_WCR_WVR_EL1(0),
2350 	DBG_BCR_BVR_WCR_WVR_EL1(1),
2351 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
2352 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
2353 	DBG_BCR_BVR_WCR_WVR_EL1(2),
2354 	DBG_BCR_BVR_WCR_WVR_EL1(3),
2355 	DBG_BCR_BVR_WCR_WVR_EL1(4),
2356 	DBG_BCR_BVR_WCR_WVR_EL1(5),
2357 	DBG_BCR_BVR_WCR_WVR_EL1(6),
2358 	DBG_BCR_BVR_WCR_WVR_EL1(7),
2359 	DBG_BCR_BVR_WCR_WVR_EL1(8),
2360 	DBG_BCR_BVR_WCR_WVR_EL1(9),
2361 	DBG_BCR_BVR_WCR_WVR_EL1(10),
2362 	DBG_BCR_BVR_WCR_WVR_EL1(11),
2363 	DBG_BCR_BVR_WCR_WVR_EL1(12),
2364 	DBG_BCR_BVR_WCR_WVR_EL1(13),
2365 	DBG_BCR_BVR_WCR_WVR_EL1(14),
2366 	DBG_BCR_BVR_WCR_WVR_EL1(15),
2367 
2368 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
2369 	{ SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
2370 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
2371 		OSLSR_EL1_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
2372 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
2373 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
2374 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
2375 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
2376 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
2377 
2378 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
2379 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
2380 	// DBGDTR[TR]X_EL0 share the same encoding
2381 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
2382 
2383 	{ SYS_DESC(SYS_DBGVCR32_EL2), undef_access, reset_val, DBGVCR32_EL2, 0 },
2384 
2385 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
2386 
2387 	/*
2388 	 * ID regs: all ID_SANITISED() entries here must have corresponding
2389 	 * entries in arm64_ftr_regs[].
2390 	 */
2391 
2392 	/* AArch64 mappings of the AArch32 ID registers */
2393 	/* CRm=1 */
2394 	AA32_ID_SANITISED(ID_PFR0_EL1),
2395 	AA32_ID_SANITISED(ID_PFR1_EL1),
2396 	{ SYS_DESC(SYS_ID_DFR0_EL1),
2397 	  .access = access_id_reg,
2398 	  .get_user = get_id_reg,
2399 	  .set_user = set_id_dfr0_el1,
2400 	  .visibility = aa32_id_visibility,
2401 	  .reset = read_sanitised_id_dfr0_el1,
2402 	  .val = ID_DFR0_EL1_PerfMon_MASK |
2403 		 ID_DFR0_EL1_CopDbg_MASK, },
2404 	ID_HIDDEN(ID_AFR0_EL1),
2405 	AA32_ID_SANITISED(ID_MMFR0_EL1),
2406 	AA32_ID_SANITISED(ID_MMFR1_EL1),
2407 	AA32_ID_SANITISED(ID_MMFR2_EL1),
2408 	AA32_ID_SANITISED(ID_MMFR3_EL1),
2409 
2410 	/* CRm=2 */
2411 	AA32_ID_SANITISED(ID_ISAR0_EL1),
2412 	AA32_ID_SANITISED(ID_ISAR1_EL1),
2413 	AA32_ID_SANITISED(ID_ISAR2_EL1),
2414 	AA32_ID_SANITISED(ID_ISAR3_EL1),
2415 	AA32_ID_SANITISED(ID_ISAR4_EL1),
2416 	AA32_ID_SANITISED(ID_ISAR5_EL1),
2417 	AA32_ID_SANITISED(ID_MMFR4_EL1),
2418 	AA32_ID_SANITISED(ID_ISAR6_EL1),
2419 
2420 	/* CRm=3 */
2421 	AA32_ID_SANITISED(MVFR0_EL1),
2422 	AA32_ID_SANITISED(MVFR1_EL1),
2423 	AA32_ID_SANITISED(MVFR2_EL1),
2424 	ID_UNALLOCATED(3,3),
2425 	AA32_ID_SANITISED(ID_PFR2_EL1),
2426 	ID_HIDDEN(ID_DFR1_EL1),
2427 	AA32_ID_SANITISED(ID_MMFR5_EL1),
2428 	ID_UNALLOCATED(3,7),
2429 
2430 	/* AArch64 ID registers */
2431 	/* CRm=4 */
2432 	{ SYS_DESC(SYS_ID_AA64PFR0_EL1),
2433 	  .access = access_id_reg,
2434 	  .get_user = get_id_reg,
2435 	  .set_user = set_id_reg,
2436 	  .reset = read_sanitised_id_aa64pfr0_el1,
2437 	  .val = ~(ID_AA64PFR0_EL1_AMU |
2438 		   ID_AA64PFR0_EL1_MPAM |
2439 		   ID_AA64PFR0_EL1_SVE |
2440 		   ID_AA64PFR0_EL1_RAS |
2441 		   ID_AA64PFR0_EL1_AdvSIMD |
2442 		   ID_AA64PFR0_EL1_FP), },
2443 	ID_WRITABLE(ID_AA64PFR1_EL1, ~(ID_AA64PFR1_EL1_PFAR |
2444 				       ID_AA64PFR1_EL1_DF2 |
2445 				       ID_AA64PFR1_EL1_MTEX |
2446 				       ID_AA64PFR1_EL1_THE |
2447 				       ID_AA64PFR1_EL1_GCS |
2448 				       ID_AA64PFR1_EL1_MTE_frac |
2449 				       ID_AA64PFR1_EL1_NMI |
2450 				       ID_AA64PFR1_EL1_RNDR_trap |
2451 				       ID_AA64PFR1_EL1_SME |
2452 				       ID_AA64PFR1_EL1_RES0 |
2453 				       ID_AA64PFR1_EL1_MPAM_frac |
2454 				       ID_AA64PFR1_EL1_RAS_frac |
2455 				       ID_AA64PFR1_EL1_MTE)),
2456 	ID_WRITABLE(ID_AA64PFR2_EL1, ID_AA64PFR2_EL1_FPMR),
2457 	ID_UNALLOCATED(4,3),
2458 	ID_WRITABLE(ID_AA64ZFR0_EL1, ~ID_AA64ZFR0_EL1_RES0),
2459 	ID_HIDDEN(ID_AA64SMFR0_EL1),
2460 	ID_UNALLOCATED(4,6),
2461 	ID_WRITABLE(ID_AA64FPFR0_EL1, ~ID_AA64FPFR0_EL1_RES0),
2462 
2463 	/* CRm=5 */
2464 	{ SYS_DESC(SYS_ID_AA64DFR0_EL1),
2465 	  .access = access_id_reg,
2466 	  .get_user = get_id_reg,
2467 	  .set_user = set_id_aa64dfr0_el1,
2468 	  .reset = read_sanitised_id_aa64dfr0_el1,
2469 	/*
2470 	 * Prior to FEAT_Debugv8.9, the architecture defines context-aware
2471 	 * breakpoints (CTX_CMPs) as the highest numbered breakpoints (BRPs).
2472 	 * KVM does not trap + emulate the breakpoint registers, and as such
2473 	 * cannot support a layout that misaligns with the underlying hardware.
2474 	 * While it may be possible to describe a subset that aligns with
2475 	 * hardware, just prevent changes to BRPs and CTX_CMPs altogether for
2476 	 * simplicity.
2477 	 *
2478 	 * See DDI0487K.a, section D2.8.3 Breakpoint types and linking
2479 	 * of breakpoints for more details.
2480 	 */
2481 	  .val = ID_AA64DFR0_EL1_DoubleLock_MASK |
2482 		 ID_AA64DFR0_EL1_WRPs_MASK |
2483 		 ID_AA64DFR0_EL1_PMUVer_MASK |
2484 		 ID_AA64DFR0_EL1_DebugVer_MASK, },
2485 	ID_SANITISED(ID_AA64DFR1_EL1),
2486 	ID_UNALLOCATED(5,2),
2487 	ID_UNALLOCATED(5,3),
2488 	ID_HIDDEN(ID_AA64AFR0_EL1),
2489 	ID_HIDDEN(ID_AA64AFR1_EL1),
2490 	ID_UNALLOCATED(5,6),
2491 	ID_UNALLOCATED(5,7),
2492 
2493 	/* CRm=6 */
2494 	ID_WRITABLE(ID_AA64ISAR0_EL1, ~ID_AA64ISAR0_EL1_RES0),
2495 	ID_WRITABLE(ID_AA64ISAR1_EL1, ~(ID_AA64ISAR1_EL1_GPI |
2496 					ID_AA64ISAR1_EL1_GPA |
2497 					ID_AA64ISAR1_EL1_API |
2498 					ID_AA64ISAR1_EL1_APA)),
2499 	ID_WRITABLE(ID_AA64ISAR2_EL1, ~(ID_AA64ISAR2_EL1_RES0 |
2500 					ID_AA64ISAR2_EL1_APA3 |
2501 					ID_AA64ISAR2_EL1_GPA3)),
2502 	ID_UNALLOCATED(6,3),
2503 	ID_UNALLOCATED(6,4),
2504 	ID_UNALLOCATED(6,5),
2505 	ID_UNALLOCATED(6,6),
2506 	ID_UNALLOCATED(6,7),
2507 
2508 	/* CRm=7 */
2509 	ID_WRITABLE(ID_AA64MMFR0_EL1, ~(ID_AA64MMFR0_EL1_RES0 |
2510 					ID_AA64MMFR0_EL1_TGRAN4_2 |
2511 					ID_AA64MMFR0_EL1_TGRAN64_2 |
2512 					ID_AA64MMFR0_EL1_TGRAN16_2)),
2513 	ID_WRITABLE(ID_AA64MMFR1_EL1, ~(ID_AA64MMFR1_EL1_RES0 |
2514 					ID_AA64MMFR1_EL1_HCX |
2515 					ID_AA64MMFR1_EL1_TWED |
2516 					ID_AA64MMFR1_EL1_XNX |
2517 					ID_AA64MMFR1_EL1_VH |
2518 					ID_AA64MMFR1_EL1_VMIDBits)),
2519 	ID_WRITABLE(ID_AA64MMFR2_EL1, ~(ID_AA64MMFR2_EL1_RES0 |
2520 					ID_AA64MMFR2_EL1_EVT |
2521 					ID_AA64MMFR2_EL1_FWB |
2522 					ID_AA64MMFR2_EL1_IDS |
2523 					ID_AA64MMFR2_EL1_NV |
2524 					ID_AA64MMFR2_EL1_CCIDX)),
2525 	ID_WRITABLE(ID_AA64MMFR3_EL1, (ID_AA64MMFR3_EL1_TCRX	|
2526 				       ID_AA64MMFR3_EL1_S1PIE   |
2527 				       ID_AA64MMFR3_EL1_S1POE)),
2528 	ID_SANITISED(ID_AA64MMFR4_EL1),
2529 	ID_UNALLOCATED(7,5),
2530 	ID_UNALLOCATED(7,6),
2531 	ID_UNALLOCATED(7,7),
2532 
2533 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
2534 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
2535 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
2536 
2537 	MTE_REG(RGSR_EL1),
2538 	MTE_REG(GCR_EL1),
2539 
2540 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
2541 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
2542 	{ SYS_DESC(SYS_SMPRI_EL1), undef_access },
2543 	{ SYS_DESC(SYS_SMCR_EL1), undef_access },
2544 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
2545 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
2546 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
2547 	{ SYS_DESC(SYS_TCR2_EL1), access_vm_reg, reset_val, TCR2_EL1, 0 },
2548 
2549 	PTRAUTH_KEY(APIA),
2550 	PTRAUTH_KEY(APIB),
2551 	PTRAUTH_KEY(APDA),
2552 	PTRAUTH_KEY(APDB),
2553 	PTRAUTH_KEY(APGA),
2554 
2555 	{ SYS_DESC(SYS_SPSR_EL1), access_spsr},
2556 	{ SYS_DESC(SYS_ELR_EL1), access_elr},
2557 
2558 	{ SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
2559 
2560 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
2561 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
2562 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
2563 
2564 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
2565 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
2566 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
2567 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
2568 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
2569 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
2570 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
2571 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
2572 
2573 	MTE_REG(TFSR_EL1),
2574 	MTE_REG(TFSRE0_EL1),
2575 
2576 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
2577 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
2578 
2579 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
2580 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
2581 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
2582 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
2583 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
2584 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
2585 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
2586 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
2587 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
2588 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
2589 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
2590 	/* PMBIDR_EL1 is not trapped */
2591 
2592 	{ PMU_SYS_REG(PMINTENSET_EL1),
2593 	  .access = access_pminten, .reg = PMINTENSET_EL1,
2594 	  .get_user = get_pmreg, .set_user = set_pmreg },
2595 	{ PMU_SYS_REG(PMINTENCLR_EL1),
2596 	  .access = access_pminten, .reg = PMINTENSET_EL1,
2597 	  .get_user = get_pmreg, .set_user = set_pmreg },
2598 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
2599 
2600 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
2601 	{ SYS_DESC(SYS_PIRE0_EL1), NULL, reset_unknown, PIRE0_EL1 },
2602 	{ SYS_DESC(SYS_PIR_EL1), NULL, reset_unknown, PIR_EL1 },
2603 	{ SYS_DESC(SYS_POR_EL1), NULL, reset_unknown, POR_EL1,
2604 	  .visibility = s1poe_visibility },
2605 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
2606 
2607 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
2608 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
2609 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
2610 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
2611 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
2612 
2613 	{ SYS_DESC(SYS_VBAR_EL1), access_rw, reset_val, VBAR_EL1, 0 },
2614 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
2615 
2616 	{ SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
2617 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
2618 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
2619 	{ SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
2620 	{ SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
2621 	{ SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
2622 	{ SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
2623 	{ SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
2624 	{ SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
2625 	{ SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
2626 	{ SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
2627 	{ SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
2628 	{ SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
2629 	{ SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
2630 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
2631 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
2632 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
2633 	{ SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
2634 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
2635 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
2636 	{ SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
2637 	{ SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
2638 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
2639 	{ SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
2640 	{ SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
2641 
2642 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
2643 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
2644 
2645 	{ SYS_DESC(SYS_ACCDATA_EL1), undef_access },
2646 
2647 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
2648 
2649 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
2650 
2651 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
2652 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr, reset_clidr, CLIDR_EL1,
2653 	  .set_user = set_clidr, .val = ~CLIDR_EL1_RES0 },
2654 	{ SYS_DESC(SYS_CCSIDR2_EL1), undef_access },
2655 	{ SYS_DESC(SYS_SMIDR_EL1), undef_access },
2656 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
2657 	ID_WRITABLE(CTR_EL0, CTR_EL0_DIC_MASK |
2658 			     CTR_EL0_IDC_MASK |
2659 			     CTR_EL0_DminLine_MASK |
2660 			     CTR_EL0_IminLine_MASK),
2661 	{ SYS_DESC(SYS_SVCR), undef_access, reset_val, SVCR, 0, .visibility = sme_visibility  },
2662 	{ SYS_DESC(SYS_FPMR), undef_access, reset_val, FPMR, 0, .visibility = fp8_visibility },
2663 
2664 	{ PMU_SYS_REG(PMCR_EL0), .access = access_pmcr, .reset = reset_pmcr,
2665 	  .reg = PMCR_EL0, .get_user = get_pmcr, .set_user = set_pmcr },
2666 	{ PMU_SYS_REG(PMCNTENSET_EL0),
2667 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
2668 	  .get_user = get_pmreg, .set_user = set_pmreg },
2669 	{ PMU_SYS_REG(PMCNTENCLR_EL0),
2670 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
2671 	  .get_user = get_pmreg, .set_user = set_pmreg },
2672 	{ PMU_SYS_REG(PMOVSCLR_EL0),
2673 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
2674 	  .get_user = get_pmreg, .set_user = set_pmreg },
2675 	/*
2676 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
2677 	 * previously (and pointlessly) advertised in the past...
2678 	 */
2679 	{ PMU_SYS_REG(PMSWINC_EL0),
2680 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
2681 	  .access = access_pmswinc, .reset = NULL },
2682 	{ PMU_SYS_REG(PMSELR_EL0),
2683 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
2684 	{ PMU_SYS_REG(PMCEID0_EL0),
2685 	  .access = access_pmceid, .reset = NULL },
2686 	{ PMU_SYS_REG(PMCEID1_EL0),
2687 	  .access = access_pmceid, .reset = NULL },
2688 	{ PMU_SYS_REG(PMCCNTR_EL0),
2689 	  .access = access_pmu_evcntr, .reset = reset_unknown,
2690 	  .reg = PMCCNTR_EL0, .get_user = get_pmu_evcntr},
2691 	{ PMU_SYS_REG(PMXEVTYPER_EL0),
2692 	  .access = access_pmu_evtyper, .reset = NULL },
2693 	{ PMU_SYS_REG(PMXEVCNTR_EL0),
2694 	  .access = access_pmu_evcntr, .reset = NULL },
2695 	/*
2696 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
2697 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
2698 	 */
2699 	{ PMU_SYS_REG(PMUSERENR_EL0), .access = access_pmuserenr,
2700 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
2701 	{ PMU_SYS_REG(PMOVSSET_EL0),
2702 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
2703 	  .get_user = get_pmreg, .set_user = set_pmreg },
2704 
2705 	{ SYS_DESC(SYS_POR_EL0), NULL, reset_unknown, POR_EL0,
2706 	  .visibility = s1poe_visibility },
2707 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
2708 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
2709 	{ SYS_DESC(SYS_TPIDR2_EL0), undef_access },
2710 
2711 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
2712 
2713 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
2714 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
2715 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
2716 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
2717 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
2718 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
2719 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
2720 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
2721 	AMU_AMEVCNTR0_EL0(0),
2722 	AMU_AMEVCNTR0_EL0(1),
2723 	AMU_AMEVCNTR0_EL0(2),
2724 	AMU_AMEVCNTR0_EL0(3),
2725 	AMU_AMEVCNTR0_EL0(4),
2726 	AMU_AMEVCNTR0_EL0(5),
2727 	AMU_AMEVCNTR0_EL0(6),
2728 	AMU_AMEVCNTR0_EL0(7),
2729 	AMU_AMEVCNTR0_EL0(8),
2730 	AMU_AMEVCNTR0_EL0(9),
2731 	AMU_AMEVCNTR0_EL0(10),
2732 	AMU_AMEVCNTR0_EL0(11),
2733 	AMU_AMEVCNTR0_EL0(12),
2734 	AMU_AMEVCNTR0_EL0(13),
2735 	AMU_AMEVCNTR0_EL0(14),
2736 	AMU_AMEVCNTR0_EL0(15),
2737 	AMU_AMEVTYPER0_EL0(0),
2738 	AMU_AMEVTYPER0_EL0(1),
2739 	AMU_AMEVTYPER0_EL0(2),
2740 	AMU_AMEVTYPER0_EL0(3),
2741 	AMU_AMEVTYPER0_EL0(4),
2742 	AMU_AMEVTYPER0_EL0(5),
2743 	AMU_AMEVTYPER0_EL0(6),
2744 	AMU_AMEVTYPER0_EL0(7),
2745 	AMU_AMEVTYPER0_EL0(8),
2746 	AMU_AMEVTYPER0_EL0(9),
2747 	AMU_AMEVTYPER0_EL0(10),
2748 	AMU_AMEVTYPER0_EL0(11),
2749 	AMU_AMEVTYPER0_EL0(12),
2750 	AMU_AMEVTYPER0_EL0(13),
2751 	AMU_AMEVTYPER0_EL0(14),
2752 	AMU_AMEVTYPER0_EL0(15),
2753 	AMU_AMEVCNTR1_EL0(0),
2754 	AMU_AMEVCNTR1_EL0(1),
2755 	AMU_AMEVCNTR1_EL0(2),
2756 	AMU_AMEVCNTR1_EL0(3),
2757 	AMU_AMEVCNTR1_EL0(4),
2758 	AMU_AMEVCNTR1_EL0(5),
2759 	AMU_AMEVCNTR1_EL0(6),
2760 	AMU_AMEVCNTR1_EL0(7),
2761 	AMU_AMEVCNTR1_EL0(8),
2762 	AMU_AMEVCNTR1_EL0(9),
2763 	AMU_AMEVCNTR1_EL0(10),
2764 	AMU_AMEVCNTR1_EL0(11),
2765 	AMU_AMEVCNTR1_EL0(12),
2766 	AMU_AMEVCNTR1_EL0(13),
2767 	AMU_AMEVCNTR1_EL0(14),
2768 	AMU_AMEVCNTR1_EL0(15),
2769 	AMU_AMEVTYPER1_EL0(0),
2770 	AMU_AMEVTYPER1_EL0(1),
2771 	AMU_AMEVTYPER1_EL0(2),
2772 	AMU_AMEVTYPER1_EL0(3),
2773 	AMU_AMEVTYPER1_EL0(4),
2774 	AMU_AMEVTYPER1_EL0(5),
2775 	AMU_AMEVTYPER1_EL0(6),
2776 	AMU_AMEVTYPER1_EL0(7),
2777 	AMU_AMEVTYPER1_EL0(8),
2778 	AMU_AMEVTYPER1_EL0(9),
2779 	AMU_AMEVTYPER1_EL0(10),
2780 	AMU_AMEVTYPER1_EL0(11),
2781 	AMU_AMEVTYPER1_EL0(12),
2782 	AMU_AMEVTYPER1_EL0(13),
2783 	AMU_AMEVTYPER1_EL0(14),
2784 	AMU_AMEVTYPER1_EL0(15),
2785 
2786 	{ SYS_DESC(SYS_CNTPCT_EL0), access_arch_timer },
2787 	{ SYS_DESC(SYS_CNTPCTSS_EL0), access_arch_timer },
2788 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
2789 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
2790 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
2791 
2792 	/* PMEVCNTRn_EL0 */
2793 	PMU_PMEVCNTR_EL0(0),
2794 	PMU_PMEVCNTR_EL0(1),
2795 	PMU_PMEVCNTR_EL0(2),
2796 	PMU_PMEVCNTR_EL0(3),
2797 	PMU_PMEVCNTR_EL0(4),
2798 	PMU_PMEVCNTR_EL0(5),
2799 	PMU_PMEVCNTR_EL0(6),
2800 	PMU_PMEVCNTR_EL0(7),
2801 	PMU_PMEVCNTR_EL0(8),
2802 	PMU_PMEVCNTR_EL0(9),
2803 	PMU_PMEVCNTR_EL0(10),
2804 	PMU_PMEVCNTR_EL0(11),
2805 	PMU_PMEVCNTR_EL0(12),
2806 	PMU_PMEVCNTR_EL0(13),
2807 	PMU_PMEVCNTR_EL0(14),
2808 	PMU_PMEVCNTR_EL0(15),
2809 	PMU_PMEVCNTR_EL0(16),
2810 	PMU_PMEVCNTR_EL0(17),
2811 	PMU_PMEVCNTR_EL0(18),
2812 	PMU_PMEVCNTR_EL0(19),
2813 	PMU_PMEVCNTR_EL0(20),
2814 	PMU_PMEVCNTR_EL0(21),
2815 	PMU_PMEVCNTR_EL0(22),
2816 	PMU_PMEVCNTR_EL0(23),
2817 	PMU_PMEVCNTR_EL0(24),
2818 	PMU_PMEVCNTR_EL0(25),
2819 	PMU_PMEVCNTR_EL0(26),
2820 	PMU_PMEVCNTR_EL0(27),
2821 	PMU_PMEVCNTR_EL0(28),
2822 	PMU_PMEVCNTR_EL0(29),
2823 	PMU_PMEVCNTR_EL0(30),
2824 	/* PMEVTYPERn_EL0 */
2825 	PMU_PMEVTYPER_EL0(0),
2826 	PMU_PMEVTYPER_EL0(1),
2827 	PMU_PMEVTYPER_EL0(2),
2828 	PMU_PMEVTYPER_EL0(3),
2829 	PMU_PMEVTYPER_EL0(4),
2830 	PMU_PMEVTYPER_EL0(5),
2831 	PMU_PMEVTYPER_EL0(6),
2832 	PMU_PMEVTYPER_EL0(7),
2833 	PMU_PMEVTYPER_EL0(8),
2834 	PMU_PMEVTYPER_EL0(9),
2835 	PMU_PMEVTYPER_EL0(10),
2836 	PMU_PMEVTYPER_EL0(11),
2837 	PMU_PMEVTYPER_EL0(12),
2838 	PMU_PMEVTYPER_EL0(13),
2839 	PMU_PMEVTYPER_EL0(14),
2840 	PMU_PMEVTYPER_EL0(15),
2841 	PMU_PMEVTYPER_EL0(16),
2842 	PMU_PMEVTYPER_EL0(17),
2843 	PMU_PMEVTYPER_EL0(18),
2844 	PMU_PMEVTYPER_EL0(19),
2845 	PMU_PMEVTYPER_EL0(20),
2846 	PMU_PMEVTYPER_EL0(21),
2847 	PMU_PMEVTYPER_EL0(22),
2848 	PMU_PMEVTYPER_EL0(23),
2849 	PMU_PMEVTYPER_EL0(24),
2850 	PMU_PMEVTYPER_EL0(25),
2851 	PMU_PMEVTYPER_EL0(26),
2852 	PMU_PMEVTYPER_EL0(27),
2853 	PMU_PMEVTYPER_EL0(28),
2854 	PMU_PMEVTYPER_EL0(29),
2855 	PMU_PMEVTYPER_EL0(30),
2856 	/*
2857 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
2858 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
2859 	 */
2860 	{ PMU_SYS_REG(PMCCFILTR_EL0), .access = access_pmu_evtyper,
2861 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
2862 
2863 	EL2_REG_VNCR(VPIDR_EL2, reset_unknown, 0),
2864 	EL2_REG_VNCR(VMPIDR_EL2, reset_unknown, 0),
2865 	EL2_REG(SCTLR_EL2, access_rw, reset_val, SCTLR_EL2_RES1),
2866 	EL2_REG(ACTLR_EL2, access_rw, reset_val, 0),
2867 	EL2_REG_VNCR(HCR_EL2, reset_hcr, 0),
2868 	EL2_REG(MDCR_EL2, access_rw, reset_val, 0),
2869 	EL2_REG(CPTR_EL2, access_rw, reset_val, CPTR_NVHE_EL2_RES1),
2870 	EL2_REG_VNCR(HSTR_EL2, reset_val, 0),
2871 	EL2_REG_VNCR(HFGRTR_EL2, reset_val, 0),
2872 	EL2_REG_VNCR(HFGWTR_EL2, reset_val, 0),
2873 	EL2_REG_VNCR(HFGITR_EL2, reset_val, 0),
2874 	EL2_REG_VNCR(HACR_EL2, reset_val, 0),
2875 
2876 	{ SYS_DESC(SYS_ZCR_EL2), .access = access_zcr_el2, .reset = reset_val,
2877 	  .visibility = sve_el2_visibility, .reg = ZCR_EL2 },
2878 
2879 	EL2_REG_VNCR(HCRX_EL2, reset_val, 0),
2880 
2881 	EL2_REG(TTBR0_EL2, access_rw, reset_val, 0),
2882 	EL2_REG(TTBR1_EL2, access_rw, reset_val, 0),
2883 	EL2_REG(TCR_EL2, access_rw, reset_val, TCR_EL2_RES1),
2884 	EL2_REG(TCR2_EL2, access_tcr2_el2, reset_val, TCR2_EL2_RES1),
2885 	EL2_REG_VNCR(VTTBR_EL2, reset_val, 0),
2886 	EL2_REG_VNCR(VTCR_EL2, reset_val, 0),
2887 
2888 	{ SYS_DESC(SYS_DACR32_EL2), undef_access, reset_unknown, DACR32_EL2 },
2889 	EL2_REG_VNCR(HDFGRTR_EL2, reset_val, 0),
2890 	EL2_REG_VNCR(HDFGWTR_EL2, reset_val, 0),
2891 	EL2_REG_VNCR(HAFGRTR_EL2, reset_val, 0),
2892 	EL2_REG_REDIR(SPSR_EL2, reset_val, 0),
2893 	EL2_REG_REDIR(ELR_EL2, reset_val, 0),
2894 	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},
2895 
2896 	/* AArch32 SPSR_* are RES0 if trapped from a NV guest */
2897 	{ SYS_DESC(SYS_SPSR_irq), .access = trap_raz_wi },
2898 	{ SYS_DESC(SYS_SPSR_abt), .access = trap_raz_wi },
2899 	{ SYS_DESC(SYS_SPSR_und), .access = trap_raz_wi },
2900 	{ SYS_DESC(SYS_SPSR_fiq), .access = trap_raz_wi },
2901 
2902 	{ SYS_DESC(SYS_IFSR32_EL2), undef_access, reset_unknown, IFSR32_EL2 },
2903 	EL2_REG(AFSR0_EL2, access_rw, reset_val, 0),
2904 	EL2_REG(AFSR1_EL2, access_rw, reset_val, 0),
2905 	EL2_REG_REDIR(ESR_EL2, reset_val, 0),
2906 	{ SYS_DESC(SYS_FPEXC32_EL2), undef_access, reset_val, FPEXC32_EL2, 0x700 },
2907 
2908 	EL2_REG_REDIR(FAR_EL2, reset_val, 0),
2909 	EL2_REG(HPFAR_EL2, access_rw, reset_val, 0),
2910 
2911 	EL2_REG(MAIR_EL2, access_rw, reset_val, 0),
2912 	EL2_REG(AMAIR_EL2, access_rw, reset_val, 0),
2913 
2914 	EL2_REG(VBAR_EL2, access_rw, reset_val, 0),
2915 	EL2_REG(RVBAR_EL2, access_rw, reset_val, 0),
2916 	{ SYS_DESC(SYS_RMR_EL2), undef_access },
2917 
2918 	EL2_REG_VNCR(ICH_HCR_EL2, reset_val, 0),
2919 
2920 	EL2_REG(CONTEXTIDR_EL2, access_rw, reset_val, 0),
2921 	EL2_REG(TPIDR_EL2, access_rw, reset_val, 0),
2922 
2923 	EL2_REG_VNCR(CNTVOFF_EL2, reset_val, 0),
2924 	EL2_REG(CNTHCTL_EL2, access_rw, reset_val, 0),
2925 
2926 	{ SYS_DESC(SYS_CNTKCTL_EL12), access_cntkctl_el12 },
2927 
2928 	EL2_REG(SP_EL2, NULL, reset_unknown, 0),
2929 };
2930 
2931 static bool handle_at_s1e01(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2932 			    const struct sys_reg_desc *r)
2933 {
2934 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2935 
2936 	__kvm_at_s1e01(vcpu, op, p->regval);
2937 
2938 	return true;
2939 }
2940 
2941 static bool handle_at_s1e2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2942 			   const struct sys_reg_desc *r)
2943 {
2944 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2945 
2946 	/* There is no FGT associated with AT S1E2A :-( */
2947 	if (op == OP_AT_S1E2A &&
2948 	    !kvm_has_feat(vcpu->kvm, ID_AA64ISAR2_EL1, ATS1A, IMP)) {
2949 		kvm_inject_undefined(vcpu);
2950 		return false;
2951 	}
2952 
2953 	__kvm_at_s1e2(vcpu, op, p->regval);
2954 
2955 	return true;
2956 }
2957 
2958 static bool handle_at_s12(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2959 			  const struct sys_reg_desc *r)
2960 {
2961 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2962 
2963 	__kvm_at_s12(vcpu, op, p->regval);
2964 
2965 	return true;
2966 }
2967 
2968 static bool kvm_supported_tlbi_s12_op(struct kvm_vcpu *vpcu, u32 instr)
2969 {
2970 	struct kvm *kvm = vpcu->kvm;
2971 	u8 CRm = sys_reg_CRm(instr);
2972 
2973 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
2974 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
2975 		return false;
2976 
2977 	if (CRm == TLBI_CRm_nROS &&
2978 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
2979 		return false;
2980 
2981 	return true;
2982 }
2983 
2984 static bool handle_alle1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2985 			   const struct sys_reg_desc *r)
2986 {
2987 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
2988 
2989 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
2990 		return undef_access(vcpu, p, r);
2991 
2992 	write_lock(&vcpu->kvm->mmu_lock);
2993 
2994 	/*
2995 	 * Drop all shadow S2s, resulting in S1/S2 TLBIs for each of the
2996 	 * corresponding VMIDs.
2997 	 */
2998 	kvm_nested_s2_unmap(vcpu->kvm, true);
2999 
3000 	write_unlock(&vcpu->kvm->mmu_lock);
3001 
3002 	return true;
3003 }
3004 
3005 static bool kvm_supported_tlbi_ipas2_op(struct kvm_vcpu *vpcu, u32 instr)
3006 {
3007 	struct kvm *kvm = vpcu->kvm;
3008 	u8 CRm = sys_reg_CRm(instr);
3009 	u8 Op2 = sys_reg_Op2(instr);
3010 
3011 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3012 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3013 		return false;
3014 
3015 	if (CRm == TLBI_CRm_IPAIS && (Op2 == 2 || Op2 == 6) &&
3016 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3017 		return false;
3018 
3019 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 0 || Op2 == 4) &&
3020 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3021 		return false;
3022 
3023 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 3 || Op2 == 7) &&
3024 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3025 		return false;
3026 
3027 	return true;
3028 }
3029 
3030 /* Only defined here as this is an internal "abstraction" */
3031 union tlbi_info {
3032 	struct {
3033 		u64	start;
3034 		u64	size;
3035 	} range;
3036 
3037 	struct {
3038 		u64	addr;
3039 	} ipa;
3040 
3041 	struct {
3042 		u64	addr;
3043 		u32	encoding;
3044 	} va;
3045 };
3046 
3047 static void s2_mmu_unmap_range(struct kvm_s2_mmu *mmu,
3048 			       const union tlbi_info *info)
3049 {
3050 	/*
3051 	 * The unmap operation is allowed to drop the MMU lock and block, which
3052 	 * means that @mmu could be used for a different context than the one
3053 	 * currently being invalidated.
3054 	 *
3055 	 * This behavior is still safe, as:
3056 	 *
3057 	 *  1) The vCPU(s) that recycled the MMU are responsible for invalidating
3058 	 *     the entire MMU before reusing it, which still honors the intent
3059 	 *     of a TLBI.
3060 	 *
3061 	 *  2) Until the guest TLBI instruction is 'retired' (i.e. increment PC
3062 	 *     and ERET to the guest), other vCPUs are allowed to use stale
3063 	 *     translations.
3064 	 *
3065 	 *  3) Accidentally unmapping an unrelated MMU context is nonfatal, and
3066 	 *     at worst may cause more aborts for shadow stage-2 fills.
3067 	 *
3068 	 * Dropping the MMU lock also implies that shadow stage-2 fills could
3069 	 * happen behind the back of the TLBI. This is still safe, though, as
3070 	 * the L1 needs to put its stage-2 in a consistent state before doing
3071 	 * the TLBI.
3072 	 */
3073 	kvm_stage2_unmap_range(mmu, info->range.start, info->range.size, true);
3074 }
3075 
3076 static bool handle_vmalls12e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3077 				const struct sys_reg_desc *r)
3078 {
3079 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3080 	u64 limit, vttbr;
3081 
3082 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3083 		return undef_access(vcpu, p, r);
3084 
3085 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3086 	limit = BIT_ULL(kvm_get_pa_bits(vcpu->kvm));
3087 
3088 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3089 				   &(union tlbi_info) {
3090 					   .range = {
3091 						   .start = 0,
3092 						   .size = limit,
3093 					   },
3094 				   },
3095 				   s2_mmu_unmap_range);
3096 
3097 	return true;
3098 }
3099 
3100 static bool handle_ripas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3101 			      const struct sys_reg_desc *r)
3102 {
3103 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3104 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3105 	u64 base, range, tg, num, scale;
3106 	int shift;
3107 
3108 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3109 		return undef_access(vcpu, p, r);
3110 
3111 	/*
3112 	 * Because the shadow S2 structure doesn't necessarily reflect that
3113 	 * of the guest's S2 (different base granule size, for example), we
3114 	 * decide to ignore TTL and only use the described range.
3115 	 */
3116 	tg	= FIELD_GET(GENMASK(47, 46), p->regval);
3117 	scale	= FIELD_GET(GENMASK(45, 44), p->regval);
3118 	num	= FIELD_GET(GENMASK(43, 39), p->regval);
3119 	base	= p->regval & GENMASK(36, 0);
3120 
3121 	switch(tg) {
3122 	case 1:
3123 		shift = 12;
3124 		break;
3125 	case 2:
3126 		shift = 14;
3127 		break;
3128 	case 3:
3129 	default:		/* IMPDEF: handle tg==0 as 64k */
3130 		shift = 16;
3131 		break;
3132 	}
3133 
3134 	base <<= shift;
3135 	range = __TLBI_RANGE_PAGES(num, scale) << shift;
3136 
3137 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3138 				   &(union tlbi_info) {
3139 					   .range = {
3140 						   .start = base,
3141 						   .size = range,
3142 					   },
3143 				   },
3144 				   s2_mmu_unmap_range);
3145 
3146 	return true;
3147 }
3148 
3149 static void s2_mmu_unmap_ipa(struct kvm_s2_mmu *mmu,
3150 			     const union tlbi_info *info)
3151 {
3152 	unsigned long max_size;
3153 	u64 base_addr;
3154 
3155 	/*
3156 	 * We drop a number of things from the supplied value:
3157 	 *
3158 	 * - NS bit: we're non-secure only.
3159 	 *
3160 	 * - IPA[51:48]: We don't support 52bit IPA just yet...
3161 	 *
3162 	 * And of course, adjust the IPA to be on an actual address.
3163 	 */
3164 	base_addr = (info->ipa.addr & GENMASK_ULL(35, 0)) << 12;
3165 	max_size = compute_tlb_inval_range(mmu, info->ipa.addr);
3166 	base_addr &= ~(max_size - 1);
3167 
3168 	/*
3169 	 * See comment in s2_mmu_unmap_range() for why this is allowed to
3170 	 * reschedule.
3171 	 */
3172 	kvm_stage2_unmap_range(mmu, base_addr, max_size, true);
3173 }
3174 
3175 static bool handle_ipas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3176 			     const struct sys_reg_desc *r)
3177 {
3178 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3179 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3180 
3181 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3182 		return undef_access(vcpu, p, r);
3183 
3184 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3185 				   &(union tlbi_info) {
3186 					   .ipa = {
3187 						   .addr = p->regval,
3188 					   },
3189 				   },
3190 				   s2_mmu_unmap_ipa);
3191 
3192 	return true;
3193 }
3194 
3195 static void s2_mmu_tlbi_s1e1(struct kvm_s2_mmu *mmu,
3196 			     const union tlbi_info *info)
3197 {
3198 	WARN_ON(__kvm_tlbi_s1e2(mmu, info->va.addr, info->va.encoding));
3199 }
3200 
3201 static bool handle_tlbi_el1(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3202 			    const struct sys_reg_desc *r)
3203 {
3204 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3205 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3206 
3207 	/*
3208 	 * If we're here, this is because we've trapped on a EL1 TLBI
3209 	 * instruction that affects the EL1 translation regime while
3210 	 * we're running in a context that doesn't allow us to let the
3211 	 * HW do its thing (aka vEL2):
3212 	 *
3213 	 * - HCR_EL2.E2H == 0 : a non-VHE guest
3214 	 * - HCR_EL2.{E2H,TGE} == { 1, 0 } : a VHE guest in guest mode
3215 	 *
3216 	 * We don't expect these helpers to ever be called when running
3217 	 * in a vEL1 context.
3218 	 */
3219 
3220 	WARN_ON(!vcpu_is_el2(vcpu));
3221 
3222 	if (!kvm_supported_tlbi_s1e1_op(vcpu, sys_encoding))
3223 		return undef_access(vcpu, p, r);
3224 
3225 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3226 				   &(union tlbi_info) {
3227 					   .va = {
3228 						   .addr = p->regval,
3229 						   .encoding = sys_encoding,
3230 					   },
3231 				   },
3232 				   s2_mmu_tlbi_s1e1);
3233 
3234 	return true;
3235 }
3236 
3237 #define SYS_INSN(insn, access_fn)					\
3238 	{								\
3239 		SYS_DESC(OP_##insn),					\
3240 		.access = (access_fn),					\
3241 	}
3242 
3243 static struct sys_reg_desc sys_insn_descs[] = {
3244 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
3245 	{ SYS_DESC(SYS_DC_IGSW), access_dcgsw },
3246 	{ SYS_DESC(SYS_DC_IGDSW), access_dcgsw },
3247 
3248 	SYS_INSN(AT_S1E1R, handle_at_s1e01),
3249 	SYS_INSN(AT_S1E1W, handle_at_s1e01),
3250 	SYS_INSN(AT_S1E0R, handle_at_s1e01),
3251 	SYS_INSN(AT_S1E0W, handle_at_s1e01),
3252 	SYS_INSN(AT_S1E1RP, handle_at_s1e01),
3253 	SYS_INSN(AT_S1E1WP, handle_at_s1e01),
3254 
3255 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
3256 	{ SYS_DESC(SYS_DC_CGSW), access_dcgsw },
3257 	{ SYS_DESC(SYS_DC_CGDSW), access_dcgsw },
3258 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
3259 	{ SYS_DESC(SYS_DC_CIGSW), access_dcgsw },
3260 	{ SYS_DESC(SYS_DC_CIGDSW), access_dcgsw },
3261 
3262 	SYS_INSN(TLBI_VMALLE1OS, handle_tlbi_el1),
3263 	SYS_INSN(TLBI_VAE1OS, handle_tlbi_el1),
3264 	SYS_INSN(TLBI_ASIDE1OS, handle_tlbi_el1),
3265 	SYS_INSN(TLBI_VAAE1OS, handle_tlbi_el1),
3266 	SYS_INSN(TLBI_VALE1OS, handle_tlbi_el1),
3267 	SYS_INSN(TLBI_VAALE1OS, handle_tlbi_el1),
3268 
3269 	SYS_INSN(TLBI_RVAE1IS, handle_tlbi_el1),
3270 	SYS_INSN(TLBI_RVAAE1IS, handle_tlbi_el1),
3271 	SYS_INSN(TLBI_RVALE1IS, handle_tlbi_el1),
3272 	SYS_INSN(TLBI_RVAALE1IS, handle_tlbi_el1),
3273 
3274 	SYS_INSN(TLBI_VMALLE1IS, handle_tlbi_el1),
3275 	SYS_INSN(TLBI_VAE1IS, handle_tlbi_el1),
3276 	SYS_INSN(TLBI_ASIDE1IS, handle_tlbi_el1),
3277 	SYS_INSN(TLBI_VAAE1IS, handle_tlbi_el1),
3278 	SYS_INSN(TLBI_VALE1IS, handle_tlbi_el1),
3279 	SYS_INSN(TLBI_VAALE1IS, handle_tlbi_el1),
3280 
3281 	SYS_INSN(TLBI_RVAE1OS, handle_tlbi_el1),
3282 	SYS_INSN(TLBI_RVAAE1OS, handle_tlbi_el1),
3283 	SYS_INSN(TLBI_RVALE1OS, handle_tlbi_el1),
3284 	SYS_INSN(TLBI_RVAALE1OS, handle_tlbi_el1),
3285 
3286 	SYS_INSN(TLBI_RVAE1, handle_tlbi_el1),
3287 	SYS_INSN(TLBI_RVAAE1, handle_tlbi_el1),
3288 	SYS_INSN(TLBI_RVALE1, handle_tlbi_el1),
3289 	SYS_INSN(TLBI_RVAALE1, handle_tlbi_el1),
3290 
3291 	SYS_INSN(TLBI_VMALLE1, handle_tlbi_el1),
3292 	SYS_INSN(TLBI_VAE1, handle_tlbi_el1),
3293 	SYS_INSN(TLBI_ASIDE1, handle_tlbi_el1),
3294 	SYS_INSN(TLBI_VAAE1, handle_tlbi_el1),
3295 	SYS_INSN(TLBI_VALE1, handle_tlbi_el1),
3296 	SYS_INSN(TLBI_VAALE1, handle_tlbi_el1),
3297 
3298 	SYS_INSN(TLBI_VMALLE1OSNXS, handle_tlbi_el1),
3299 	SYS_INSN(TLBI_VAE1OSNXS, handle_tlbi_el1),
3300 	SYS_INSN(TLBI_ASIDE1OSNXS, handle_tlbi_el1),
3301 	SYS_INSN(TLBI_VAAE1OSNXS, handle_tlbi_el1),
3302 	SYS_INSN(TLBI_VALE1OSNXS, handle_tlbi_el1),
3303 	SYS_INSN(TLBI_VAALE1OSNXS, handle_tlbi_el1),
3304 
3305 	SYS_INSN(TLBI_RVAE1ISNXS, handle_tlbi_el1),
3306 	SYS_INSN(TLBI_RVAAE1ISNXS, handle_tlbi_el1),
3307 	SYS_INSN(TLBI_RVALE1ISNXS, handle_tlbi_el1),
3308 	SYS_INSN(TLBI_RVAALE1ISNXS, handle_tlbi_el1),
3309 
3310 	SYS_INSN(TLBI_VMALLE1ISNXS, handle_tlbi_el1),
3311 	SYS_INSN(TLBI_VAE1ISNXS, handle_tlbi_el1),
3312 	SYS_INSN(TLBI_ASIDE1ISNXS, handle_tlbi_el1),
3313 	SYS_INSN(TLBI_VAAE1ISNXS, handle_tlbi_el1),
3314 	SYS_INSN(TLBI_VALE1ISNXS, handle_tlbi_el1),
3315 	SYS_INSN(TLBI_VAALE1ISNXS, handle_tlbi_el1),
3316 
3317 	SYS_INSN(TLBI_RVAE1OSNXS, handle_tlbi_el1),
3318 	SYS_INSN(TLBI_RVAAE1OSNXS, handle_tlbi_el1),
3319 	SYS_INSN(TLBI_RVALE1OSNXS, handle_tlbi_el1),
3320 	SYS_INSN(TLBI_RVAALE1OSNXS, handle_tlbi_el1),
3321 
3322 	SYS_INSN(TLBI_RVAE1NXS, handle_tlbi_el1),
3323 	SYS_INSN(TLBI_RVAAE1NXS, handle_tlbi_el1),
3324 	SYS_INSN(TLBI_RVALE1NXS, handle_tlbi_el1),
3325 	SYS_INSN(TLBI_RVAALE1NXS, handle_tlbi_el1),
3326 
3327 	SYS_INSN(TLBI_VMALLE1NXS, handle_tlbi_el1),
3328 	SYS_INSN(TLBI_VAE1NXS, handle_tlbi_el1),
3329 	SYS_INSN(TLBI_ASIDE1NXS, handle_tlbi_el1),
3330 	SYS_INSN(TLBI_VAAE1NXS, handle_tlbi_el1),
3331 	SYS_INSN(TLBI_VALE1NXS, handle_tlbi_el1),
3332 	SYS_INSN(TLBI_VAALE1NXS, handle_tlbi_el1),
3333 
3334 	SYS_INSN(AT_S1E2R, handle_at_s1e2),
3335 	SYS_INSN(AT_S1E2W, handle_at_s1e2),
3336 	SYS_INSN(AT_S12E1R, handle_at_s12),
3337 	SYS_INSN(AT_S12E1W, handle_at_s12),
3338 	SYS_INSN(AT_S12E0R, handle_at_s12),
3339 	SYS_INSN(AT_S12E0W, handle_at_s12),
3340 	SYS_INSN(AT_S1E2A, handle_at_s1e2),
3341 
3342 	SYS_INSN(TLBI_IPAS2E1IS, handle_ipas2e1is),
3343 	SYS_INSN(TLBI_RIPAS2E1IS, handle_ripas2e1is),
3344 	SYS_INSN(TLBI_IPAS2LE1IS, handle_ipas2e1is),
3345 	SYS_INSN(TLBI_RIPAS2LE1IS, handle_ripas2e1is),
3346 
3347 	SYS_INSN(TLBI_ALLE2OS, undef_access),
3348 	SYS_INSN(TLBI_VAE2OS, undef_access),
3349 	SYS_INSN(TLBI_ALLE1OS, handle_alle1is),
3350 	SYS_INSN(TLBI_VALE2OS, undef_access),
3351 	SYS_INSN(TLBI_VMALLS12E1OS, handle_vmalls12e1is),
3352 
3353 	SYS_INSN(TLBI_RVAE2IS, undef_access),
3354 	SYS_INSN(TLBI_RVALE2IS, undef_access),
3355 
3356 	SYS_INSN(TLBI_ALLE1IS, handle_alle1is),
3357 	SYS_INSN(TLBI_VMALLS12E1IS, handle_vmalls12e1is),
3358 	SYS_INSN(TLBI_IPAS2E1OS, handle_ipas2e1is),
3359 	SYS_INSN(TLBI_IPAS2E1, handle_ipas2e1is),
3360 	SYS_INSN(TLBI_RIPAS2E1, handle_ripas2e1is),
3361 	SYS_INSN(TLBI_RIPAS2E1OS, handle_ripas2e1is),
3362 	SYS_INSN(TLBI_IPAS2LE1OS, handle_ipas2e1is),
3363 	SYS_INSN(TLBI_IPAS2LE1, handle_ipas2e1is),
3364 	SYS_INSN(TLBI_RIPAS2LE1, handle_ripas2e1is),
3365 	SYS_INSN(TLBI_RIPAS2LE1OS, handle_ripas2e1is),
3366 	SYS_INSN(TLBI_RVAE2OS, undef_access),
3367 	SYS_INSN(TLBI_RVALE2OS, undef_access),
3368 	SYS_INSN(TLBI_RVAE2, undef_access),
3369 	SYS_INSN(TLBI_RVALE2, undef_access),
3370 	SYS_INSN(TLBI_ALLE1, handle_alle1is),
3371 	SYS_INSN(TLBI_VMALLS12E1, handle_vmalls12e1is),
3372 
3373 	SYS_INSN(TLBI_IPAS2E1ISNXS, handle_ipas2e1is),
3374 	SYS_INSN(TLBI_RIPAS2E1ISNXS, handle_ripas2e1is),
3375 	SYS_INSN(TLBI_IPAS2LE1ISNXS, handle_ipas2e1is),
3376 	SYS_INSN(TLBI_RIPAS2LE1ISNXS, handle_ripas2e1is),
3377 
3378 	SYS_INSN(TLBI_ALLE2OSNXS, undef_access),
3379 	SYS_INSN(TLBI_VAE2OSNXS, undef_access),
3380 	SYS_INSN(TLBI_ALLE1OSNXS, handle_alle1is),
3381 	SYS_INSN(TLBI_VALE2OSNXS, undef_access),
3382 	SYS_INSN(TLBI_VMALLS12E1OSNXS, handle_vmalls12e1is),
3383 
3384 	SYS_INSN(TLBI_RVAE2ISNXS, undef_access),
3385 	SYS_INSN(TLBI_RVALE2ISNXS, undef_access),
3386 	SYS_INSN(TLBI_ALLE2ISNXS, undef_access),
3387 	SYS_INSN(TLBI_VAE2ISNXS, undef_access),
3388 
3389 	SYS_INSN(TLBI_ALLE1ISNXS, handle_alle1is),
3390 	SYS_INSN(TLBI_VALE2ISNXS, undef_access),
3391 	SYS_INSN(TLBI_VMALLS12E1ISNXS, handle_vmalls12e1is),
3392 	SYS_INSN(TLBI_IPAS2E1OSNXS, handle_ipas2e1is),
3393 	SYS_INSN(TLBI_IPAS2E1NXS, handle_ipas2e1is),
3394 	SYS_INSN(TLBI_RIPAS2E1NXS, handle_ripas2e1is),
3395 	SYS_INSN(TLBI_RIPAS2E1OSNXS, handle_ripas2e1is),
3396 	SYS_INSN(TLBI_IPAS2LE1OSNXS, handle_ipas2e1is),
3397 	SYS_INSN(TLBI_IPAS2LE1NXS, handle_ipas2e1is),
3398 	SYS_INSN(TLBI_RIPAS2LE1NXS, handle_ripas2e1is),
3399 	SYS_INSN(TLBI_RIPAS2LE1OSNXS, handle_ripas2e1is),
3400 	SYS_INSN(TLBI_RVAE2OSNXS, undef_access),
3401 	SYS_INSN(TLBI_RVALE2OSNXS, undef_access),
3402 	SYS_INSN(TLBI_RVAE2NXS, undef_access),
3403 	SYS_INSN(TLBI_RVALE2NXS, undef_access),
3404 	SYS_INSN(TLBI_ALLE2NXS, undef_access),
3405 	SYS_INSN(TLBI_VAE2NXS, undef_access),
3406 	SYS_INSN(TLBI_ALLE1NXS, handle_alle1is),
3407 	SYS_INSN(TLBI_VALE2NXS, undef_access),
3408 	SYS_INSN(TLBI_VMALLS12E1NXS, handle_vmalls12e1is),
3409 };
3410 
3411 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
3412 			struct sys_reg_params *p,
3413 			const struct sys_reg_desc *r)
3414 {
3415 	if (p->is_write) {
3416 		return ignore_write(vcpu, p);
3417 	} else {
3418 		u64 dfr = kvm_read_vm_id_reg(vcpu->kvm, SYS_ID_AA64DFR0_EL1);
3419 		u32 el3 = kvm_has_feat(vcpu->kvm, ID_AA64PFR0_EL1, EL3, IMP);
3420 
3421 		p->regval = ((SYS_FIELD_GET(ID_AA64DFR0_EL1, WRPs, dfr) << 28) |
3422 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, BRPs, dfr) << 24) |
3423 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, CTX_CMPs, dfr) << 20) |
3424 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, dfr) << 16) |
3425 			     (1 << 15) | (el3 << 14) | (el3 << 12));
3426 		return true;
3427 	}
3428 }
3429 
3430 /*
3431  * AArch32 debug register mappings
3432  *
3433  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
3434  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
3435  *
3436  * None of the other registers share their location, so treat them as
3437  * if they were 64bit.
3438  */
3439 #define DBG_BCR_BVR_WCR_WVR(n)						      \
3440 	/* DBGBVRn */							      \
3441 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
3442 	/* DBGBCRn */							      \
3443 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
3444 	/* DBGWVRn */							      \
3445 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
3446 	/* DBGWCRn */							      \
3447 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
3448 
3449 #define DBGBXVR(n)							      \
3450 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
3451 
3452 /*
3453  * Trapped cp14 registers. We generally ignore most of the external
3454  * debug, on the principle that they don't really make sense to a
3455  * guest. Revisit this one day, would this principle change.
3456  */
3457 static const struct sys_reg_desc cp14_regs[] = {
3458 	/* DBGDIDR */
3459 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
3460 	/* DBGDTRRXext */
3461 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
3462 
3463 	DBG_BCR_BVR_WCR_WVR(0),
3464 	/* DBGDSCRint */
3465 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
3466 	DBG_BCR_BVR_WCR_WVR(1),
3467 	/* DBGDCCINT */
3468 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
3469 	/* DBGDSCRext */
3470 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
3471 	DBG_BCR_BVR_WCR_WVR(2),
3472 	/* DBGDTR[RT]Xint */
3473 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
3474 	/* DBGDTR[RT]Xext */
3475 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
3476 	DBG_BCR_BVR_WCR_WVR(3),
3477 	DBG_BCR_BVR_WCR_WVR(4),
3478 	DBG_BCR_BVR_WCR_WVR(5),
3479 	/* DBGWFAR */
3480 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
3481 	/* DBGOSECCR */
3482 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
3483 	DBG_BCR_BVR_WCR_WVR(6),
3484 	/* DBGVCR */
3485 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
3486 	DBG_BCR_BVR_WCR_WVR(7),
3487 	DBG_BCR_BVR_WCR_WVR(8),
3488 	DBG_BCR_BVR_WCR_WVR(9),
3489 	DBG_BCR_BVR_WCR_WVR(10),
3490 	DBG_BCR_BVR_WCR_WVR(11),
3491 	DBG_BCR_BVR_WCR_WVR(12),
3492 	DBG_BCR_BVR_WCR_WVR(13),
3493 	DBG_BCR_BVR_WCR_WVR(14),
3494 	DBG_BCR_BVR_WCR_WVR(15),
3495 
3496 	/* DBGDRAR (32bit) */
3497 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
3498 
3499 	DBGBXVR(0),
3500 	/* DBGOSLAR */
3501 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
3502 	DBGBXVR(1),
3503 	/* DBGOSLSR */
3504 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
3505 	DBGBXVR(2),
3506 	DBGBXVR(3),
3507 	/* DBGOSDLR */
3508 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
3509 	DBGBXVR(4),
3510 	/* DBGPRCR */
3511 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
3512 	DBGBXVR(5),
3513 	DBGBXVR(6),
3514 	DBGBXVR(7),
3515 	DBGBXVR(8),
3516 	DBGBXVR(9),
3517 	DBGBXVR(10),
3518 	DBGBXVR(11),
3519 	DBGBXVR(12),
3520 	DBGBXVR(13),
3521 	DBGBXVR(14),
3522 	DBGBXVR(15),
3523 
3524 	/* DBGDSAR (32bit) */
3525 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
3526 
3527 	/* DBGDEVID2 */
3528 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
3529 	/* DBGDEVID1 */
3530 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
3531 	/* DBGDEVID */
3532 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
3533 	/* DBGCLAIMSET */
3534 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
3535 	/* DBGCLAIMCLR */
3536 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
3537 	/* DBGAUTHSTATUS */
3538 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
3539 };
3540 
3541 /* Trapped cp14 64bit registers */
3542 static const struct sys_reg_desc cp14_64_regs[] = {
3543 	/* DBGDRAR (64bit) */
3544 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
3545 
3546 	/* DBGDSAR (64bit) */
3547 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
3548 };
3549 
3550 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)			\
3551 	AA32(_map),							\
3552 	Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),			\
3553 	.visibility = pmu_visibility
3554 
3555 /* Macro to expand the PMEVCNTRn register */
3556 #define PMU_PMEVCNTR(n)							\
3557 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
3558 	  (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
3559 	  .access = access_pmu_evcntr }
3560 
3561 /* Macro to expand the PMEVTYPERn register */
3562 #define PMU_PMEVTYPER(n)						\
3563 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
3564 	  (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
3565 	  .access = access_pmu_evtyper }
3566 /*
3567  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
3568  * depending on the way they are accessed (as a 32bit or a 64bit
3569  * register).
3570  */
3571 static const struct sys_reg_desc cp15_regs[] = {
3572 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
3573 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
3574 	/* ACTLR */
3575 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
3576 	/* ACTLR2 */
3577 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
3578 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
3579 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
3580 	/* TTBCR */
3581 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
3582 	/* TTBCR2 */
3583 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
3584 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
3585 	{ CP15_SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
3586 	/* DFSR */
3587 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
3588 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
3589 	/* ADFSR */
3590 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
3591 	/* AIFSR */
3592 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
3593 	/* DFAR */
3594 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
3595 	/* IFAR */
3596 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
3597 
3598 	/*
3599 	 * DC{C,I,CI}SW operations:
3600 	 */
3601 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
3602 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
3603 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
3604 
3605 	/* PMU */
3606 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
3607 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
3608 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
3609 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
3610 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
3611 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
3612 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
3613 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
3614 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
3615 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
3616 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
3617 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
3618 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
3619 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
3620 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
3621 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
3622 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
3623 	/* PMMIR */
3624 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
3625 
3626 	/* PRRR/MAIR0 */
3627 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
3628 	/* NMRR/MAIR1 */
3629 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
3630 	/* AMAIR0 */
3631 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
3632 	/* AMAIR1 */
3633 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
3634 
3635 	{ CP15_SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
3636 	{ CP15_SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
3637 	{ CP15_SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
3638 	{ CP15_SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
3639 	{ CP15_SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
3640 	{ CP15_SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
3641 	{ CP15_SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
3642 	{ CP15_SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
3643 	{ CP15_SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
3644 	{ CP15_SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
3645 	{ CP15_SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
3646 	{ CP15_SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
3647 	{ CP15_SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
3648 	{ CP15_SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
3649 	{ CP15_SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
3650 	{ CP15_SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
3651 	{ CP15_SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
3652 	{ CP15_SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
3653 	{ CP15_SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
3654 	{ CP15_SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
3655 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
3656 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
3657 
3658 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
3659 
3660 	/* Arch Tmers */
3661 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
3662 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
3663 
3664 	/* PMEVCNTRn */
3665 	PMU_PMEVCNTR(0),
3666 	PMU_PMEVCNTR(1),
3667 	PMU_PMEVCNTR(2),
3668 	PMU_PMEVCNTR(3),
3669 	PMU_PMEVCNTR(4),
3670 	PMU_PMEVCNTR(5),
3671 	PMU_PMEVCNTR(6),
3672 	PMU_PMEVCNTR(7),
3673 	PMU_PMEVCNTR(8),
3674 	PMU_PMEVCNTR(9),
3675 	PMU_PMEVCNTR(10),
3676 	PMU_PMEVCNTR(11),
3677 	PMU_PMEVCNTR(12),
3678 	PMU_PMEVCNTR(13),
3679 	PMU_PMEVCNTR(14),
3680 	PMU_PMEVCNTR(15),
3681 	PMU_PMEVCNTR(16),
3682 	PMU_PMEVCNTR(17),
3683 	PMU_PMEVCNTR(18),
3684 	PMU_PMEVCNTR(19),
3685 	PMU_PMEVCNTR(20),
3686 	PMU_PMEVCNTR(21),
3687 	PMU_PMEVCNTR(22),
3688 	PMU_PMEVCNTR(23),
3689 	PMU_PMEVCNTR(24),
3690 	PMU_PMEVCNTR(25),
3691 	PMU_PMEVCNTR(26),
3692 	PMU_PMEVCNTR(27),
3693 	PMU_PMEVCNTR(28),
3694 	PMU_PMEVCNTR(29),
3695 	PMU_PMEVCNTR(30),
3696 	/* PMEVTYPERn */
3697 	PMU_PMEVTYPER(0),
3698 	PMU_PMEVTYPER(1),
3699 	PMU_PMEVTYPER(2),
3700 	PMU_PMEVTYPER(3),
3701 	PMU_PMEVTYPER(4),
3702 	PMU_PMEVTYPER(5),
3703 	PMU_PMEVTYPER(6),
3704 	PMU_PMEVTYPER(7),
3705 	PMU_PMEVTYPER(8),
3706 	PMU_PMEVTYPER(9),
3707 	PMU_PMEVTYPER(10),
3708 	PMU_PMEVTYPER(11),
3709 	PMU_PMEVTYPER(12),
3710 	PMU_PMEVTYPER(13),
3711 	PMU_PMEVTYPER(14),
3712 	PMU_PMEVTYPER(15),
3713 	PMU_PMEVTYPER(16),
3714 	PMU_PMEVTYPER(17),
3715 	PMU_PMEVTYPER(18),
3716 	PMU_PMEVTYPER(19),
3717 	PMU_PMEVTYPER(20),
3718 	PMU_PMEVTYPER(21),
3719 	PMU_PMEVTYPER(22),
3720 	PMU_PMEVTYPER(23),
3721 	PMU_PMEVTYPER(24),
3722 	PMU_PMEVTYPER(25),
3723 	PMU_PMEVTYPER(26),
3724 	PMU_PMEVTYPER(27),
3725 	PMU_PMEVTYPER(28),
3726 	PMU_PMEVTYPER(29),
3727 	PMU_PMEVTYPER(30),
3728 	/* PMCCFILTR */
3729 	{ CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
3730 
3731 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
3732 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
3733 
3734 	/* CCSIDR2 */
3735 	{ Op1(1), CRn( 0), CRm( 0),  Op2(2), undef_access },
3736 
3737 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
3738 };
3739 
3740 static const struct sys_reg_desc cp15_64_regs[] = {
3741 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
3742 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
3743 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
3744 	{ SYS_DESC(SYS_AARCH32_CNTPCT),	      access_arch_timer },
3745 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
3746 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
3747 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
3748 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
3749 	{ SYS_DESC(SYS_AARCH32_CNTPCTSS),     access_arch_timer },
3750 };
3751 
3752 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
3753 			       bool is_32)
3754 {
3755 	unsigned int i;
3756 
3757 	for (i = 0; i < n; i++) {
3758 		if (!is_32 && table[i].reg && !table[i].reset) {
3759 			kvm_err("sys_reg table %pS entry %d (%s) lacks reset\n",
3760 				&table[i], i, table[i].name);
3761 			return false;
3762 		}
3763 
3764 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
3765 			kvm_err("sys_reg table %pS entry %d (%s -> %s) out of order\n",
3766 				&table[i], i, table[i - 1].name, table[i].name);
3767 			return false;
3768 		}
3769 	}
3770 
3771 	return true;
3772 }
3773 
3774 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
3775 {
3776 	kvm_inject_undefined(vcpu);
3777 	return 1;
3778 }
3779 
3780 static void perform_access(struct kvm_vcpu *vcpu,
3781 			   struct sys_reg_params *params,
3782 			   const struct sys_reg_desc *r)
3783 {
3784 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
3785 
3786 	/* Check for regs disabled by runtime config */
3787 	if (sysreg_hidden(vcpu, r)) {
3788 		kvm_inject_undefined(vcpu);
3789 		return;
3790 	}
3791 
3792 	/*
3793 	 * Not having an accessor means that we have configured a trap
3794 	 * that we don't know how to handle. This certainly qualifies
3795 	 * as a gross bug that should be fixed right away.
3796 	 */
3797 	BUG_ON(!r->access);
3798 
3799 	/* Skip instruction if instructed so */
3800 	if (likely(r->access(vcpu, params, r)))
3801 		kvm_incr_pc(vcpu);
3802 }
3803 
3804 /*
3805  * emulate_cp --  tries to match a sys_reg access in a handling table, and
3806  *                call the corresponding trap handler.
3807  *
3808  * @params: pointer to the descriptor of the access
3809  * @table: array of trap descriptors
3810  * @num: size of the trap descriptor array
3811  *
3812  * Return true if the access has been handled, false if not.
3813  */
3814 static bool emulate_cp(struct kvm_vcpu *vcpu,
3815 		       struct sys_reg_params *params,
3816 		       const struct sys_reg_desc *table,
3817 		       size_t num)
3818 {
3819 	const struct sys_reg_desc *r;
3820 
3821 	if (!table)
3822 		return false;	/* Not handled */
3823 
3824 	r = find_reg(params, table, num);
3825 
3826 	if (r) {
3827 		perform_access(vcpu, params, r);
3828 		return true;
3829 	}
3830 
3831 	/* Not handled */
3832 	return false;
3833 }
3834 
3835 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
3836 				struct sys_reg_params *params)
3837 {
3838 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
3839 	int cp = -1;
3840 
3841 	switch (esr_ec) {
3842 	case ESR_ELx_EC_CP15_32:
3843 	case ESR_ELx_EC_CP15_64:
3844 		cp = 15;
3845 		break;
3846 	case ESR_ELx_EC_CP14_MR:
3847 	case ESR_ELx_EC_CP14_64:
3848 		cp = 14;
3849 		break;
3850 	default:
3851 		WARN_ON(1);
3852 	}
3853 
3854 	print_sys_reg_msg(params,
3855 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
3856 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
3857 	kvm_inject_undefined(vcpu);
3858 }
3859 
3860 /**
3861  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
3862  * @vcpu: The VCPU pointer
3863  * @global: &struct sys_reg_desc
3864  * @nr_global: size of the @global array
3865  */
3866 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
3867 			    const struct sys_reg_desc *global,
3868 			    size_t nr_global)
3869 {
3870 	struct sys_reg_params params;
3871 	u64 esr = kvm_vcpu_get_esr(vcpu);
3872 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
3873 	int Rt2 = (esr >> 10) & 0x1f;
3874 
3875 	params.CRm = (esr >> 1) & 0xf;
3876 	params.is_write = ((esr & 1) == 0);
3877 
3878 	params.Op0 = 0;
3879 	params.Op1 = (esr >> 16) & 0xf;
3880 	params.Op2 = 0;
3881 	params.CRn = 0;
3882 
3883 	/*
3884 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
3885 	 * backends between AArch32 and AArch64, we get away with it.
3886 	 */
3887 	if (params.is_write) {
3888 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
3889 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
3890 	}
3891 
3892 	/*
3893 	 * If the table contains a handler, handle the
3894 	 * potential register operation in the case of a read and return
3895 	 * with success.
3896 	 */
3897 	if (emulate_cp(vcpu, &params, global, nr_global)) {
3898 		/* Split up the value between registers for the read side */
3899 		if (!params.is_write) {
3900 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
3901 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
3902 		}
3903 
3904 		return 1;
3905 	}
3906 
3907 	unhandled_cp_access(vcpu, &params);
3908 	return 1;
3909 }
3910 
3911 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
3912 
3913 /*
3914  * The CP10 ID registers are architecturally mapped to AArch64 feature
3915  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
3916  * from AArch32.
3917  */
3918 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
3919 {
3920 	u8 reg_id = (esr >> 10) & 0xf;
3921 	bool valid;
3922 
3923 	params->is_write = ((esr & 1) == 0);
3924 	params->Op0 = 3;
3925 	params->Op1 = 0;
3926 	params->CRn = 0;
3927 	params->CRm = 3;
3928 
3929 	/* CP10 ID registers are read-only */
3930 	valid = !params->is_write;
3931 
3932 	switch (reg_id) {
3933 	/* MVFR0 */
3934 	case 0b0111:
3935 		params->Op2 = 0;
3936 		break;
3937 	/* MVFR1 */
3938 	case 0b0110:
3939 		params->Op2 = 1;
3940 		break;
3941 	/* MVFR2 */
3942 	case 0b0101:
3943 		params->Op2 = 2;
3944 		break;
3945 	default:
3946 		valid = false;
3947 	}
3948 
3949 	if (valid)
3950 		return true;
3951 
3952 	kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
3953 		      params->is_write ? "write" : "read", reg_id);
3954 	return false;
3955 }
3956 
3957 /**
3958  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
3959  *			  VFP Register' from AArch32.
3960  * @vcpu: The vCPU pointer
3961  *
3962  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
3963  * Work out the correct AArch64 system register encoding and reroute to the
3964  * AArch64 system register emulation.
3965  */
3966 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
3967 {
3968 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
3969 	u64 esr = kvm_vcpu_get_esr(vcpu);
3970 	struct sys_reg_params params;
3971 
3972 	/* UNDEF on any unhandled register access */
3973 	if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
3974 		kvm_inject_undefined(vcpu);
3975 		return 1;
3976 	}
3977 
3978 	if (emulate_sys_reg(vcpu, &params))
3979 		vcpu_set_reg(vcpu, Rt, params.regval);
3980 
3981 	return 1;
3982 }
3983 
3984 /**
3985  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
3986  *			       CRn=0, which corresponds to the AArch32 feature
3987  *			       registers.
3988  * @vcpu: the vCPU pointer
3989  * @params: the system register access parameters.
3990  *
3991  * Our cp15 system register tables do not enumerate the AArch32 feature
3992  * registers. Conveniently, our AArch64 table does, and the AArch32 system
3993  * register encoding can be trivially remapped into the AArch64 for the feature
3994  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
3995  *
3996  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
3997  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
3998  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
3999  * treat undefined registers in this range as RAZ.
4000  */
4001 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
4002 				   struct sys_reg_params *params)
4003 {
4004 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4005 
4006 	/* Treat impossible writes to RO registers as UNDEFINED */
4007 	if (params->is_write) {
4008 		unhandled_cp_access(vcpu, params);
4009 		return 1;
4010 	}
4011 
4012 	params->Op0 = 3;
4013 
4014 	/*
4015 	 * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
4016 	 * Avoid conflicting with future expansion of AArch64 feature registers
4017 	 * and simply treat them as RAZ here.
4018 	 */
4019 	if (params->CRm > 3)
4020 		params->regval = 0;
4021 	else if (!emulate_sys_reg(vcpu, params))
4022 		return 1;
4023 
4024 	vcpu_set_reg(vcpu, Rt, params->regval);
4025 	return 1;
4026 }
4027 
4028 /**
4029  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
4030  * @vcpu: The VCPU pointer
4031  * @params: &struct sys_reg_params
4032  * @global: &struct sys_reg_desc
4033  * @nr_global: size of the @global array
4034  */
4035 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
4036 			    struct sys_reg_params *params,
4037 			    const struct sys_reg_desc *global,
4038 			    size_t nr_global)
4039 {
4040 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
4041 
4042 	params->regval = vcpu_get_reg(vcpu, Rt);
4043 
4044 	if (emulate_cp(vcpu, params, global, nr_global)) {
4045 		if (!params->is_write)
4046 			vcpu_set_reg(vcpu, Rt, params->regval);
4047 		return 1;
4048 	}
4049 
4050 	unhandled_cp_access(vcpu, params);
4051 	return 1;
4052 }
4053 
4054 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
4055 {
4056 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
4057 }
4058 
4059 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
4060 {
4061 	struct sys_reg_params params;
4062 
4063 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4064 
4065 	/*
4066 	 * Certain AArch32 ID registers are handled by rerouting to the AArch64
4067 	 * system register table. Registers in the ID range where CRm=0 are
4068 	 * excluded from this scheme as they do not trivially map into AArch64
4069 	 * system register encodings.
4070 	 */
4071 	if (params.Op1 == 0 && params.CRn == 0 && params.CRm)
4072 		return kvm_emulate_cp15_id_reg(vcpu, &params);
4073 
4074 	return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
4075 }
4076 
4077 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
4078 {
4079 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
4080 }
4081 
4082 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
4083 {
4084 	struct sys_reg_params params;
4085 
4086 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4087 
4088 	return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
4089 }
4090 
4091 /**
4092  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
4093  * @vcpu: The VCPU pointer
4094  * @params: Decoded system register parameters
4095  *
4096  * Return: true if the system register access was successful, false otherwise.
4097  */
4098 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
4099 			    struct sys_reg_params *params)
4100 {
4101 	const struct sys_reg_desc *r;
4102 
4103 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4104 	if (likely(r)) {
4105 		perform_access(vcpu, params, r);
4106 		return true;
4107 	}
4108 
4109 	print_sys_reg_msg(params,
4110 			  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
4111 			  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
4112 	kvm_inject_undefined(vcpu);
4113 
4114 	return false;
4115 }
4116 
4117 static const struct sys_reg_desc *idregs_debug_find(struct kvm *kvm, u8 pos)
4118 {
4119 	unsigned long i, idreg_idx = 0;
4120 
4121 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4122 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4123 
4124 		if (!is_vm_ftr_id_reg(reg_to_encoding(r)))
4125 			continue;
4126 
4127 		if (idreg_idx == pos)
4128 			return r;
4129 
4130 		idreg_idx++;
4131 	}
4132 
4133 	return NULL;
4134 }
4135 
4136 static void *idregs_debug_start(struct seq_file *s, loff_t *pos)
4137 {
4138 	struct kvm *kvm = s->private;
4139 	u8 *iter;
4140 
4141 	mutex_lock(&kvm->arch.config_lock);
4142 
4143 	iter = &kvm->arch.idreg_debugfs_iter;
4144 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags) &&
4145 	    *iter == (u8)~0) {
4146 		*iter = *pos;
4147 		if (!idregs_debug_find(kvm, *iter))
4148 			iter = NULL;
4149 	} else {
4150 		iter = ERR_PTR(-EBUSY);
4151 	}
4152 
4153 	mutex_unlock(&kvm->arch.config_lock);
4154 
4155 	return iter;
4156 }
4157 
4158 static void *idregs_debug_next(struct seq_file *s, void *v, loff_t *pos)
4159 {
4160 	struct kvm *kvm = s->private;
4161 
4162 	(*pos)++;
4163 
4164 	if (idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter + 1)) {
4165 		kvm->arch.idreg_debugfs_iter++;
4166 
4167 		return &kvm->arch.idreg_debugfs_iter;
4168 	}
4169 
4170 	return NULL;
4171 }
4172 
4173 static void idregs_debug_stop(struct seq_file *s, void *v)
4174 {
4175 	struct kvm *kvm = s->private;
4176 
4177 	if (IS_ERR(v))
4178 		return;
4179 
4180 	mutex_lock(&kvm->arch.config_lock);
4181 
4182 	kvm->arch.idreg_debugfs_iter = ~0;
4183 
4184 	mutex_unlock(&kvm->arch.config_lock);
4185 }
4186 
4187 static int idregs_debug_show(struct seq_file *s, void *v)
4188 {
4189 	const struct sys_reg_desc *desc;
4190 	struct kvm *kvm = s->private;
4191 
4192 	desc = idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter);
4193 
4194 	if (!desc->name)
4195 		return 0;
4196 
4197 	seq_printf(s, "%20s:\t%016llx\n",
4198 		   desc->name, kvm_read_vm_id_reg(kvm, reg_to_encoding(desc)));
4199 
4200 	return 0;
4201 }
4202 
4203 static const struct seq_operations idregs_debug_sops = {
4204 	.start	= idregs_debug_start,
4205 	.next	= idregs_debug_next,
4206 	.stop	= idregs_debug_stop,
4207 	.show	= idregs_debug_show,
4208 };
4209 
4210 DEFINE_SEQ_ATTRIBUTE(idregs_debug);
4211 
4212 void kvm_sys_regs_create_debugfs(struct kvm *kvm)
4213 {
4214 	kvm->arch.idreg_debugfs_iter = ~0;
4215 
4216 	debugfs_create_file("idregs", 0444, kvm->debugfs_dentry, kvm,
4217 			    &idregs_debug_fops);
4218 }
4219 
4220 static void reset_vm_ftr_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *reg)
4221 {
4222 	u32 id = reg_to_encoding(reg);
4223 	struct kvm *kvm = vcpu->kvm;
4224 
4225 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags))
4226 		return;
4227 
4228 	kvm_set_vm_id_reg(kvm, id, reg->reset(vcpu, reg));
4229 }
4230 
4231 static void reset_vcpu_ftr_id_reg(struct kvm_vcpu *vcpu,
4232 				  const struct sys_reg_desc *reg)
4233 {
4234 	if (kvm_vcpu_initialized(vcpu))
4235 		return;
4236 
4237 	reg->reset(vcpu, reg);
4238 }
4239 
4240 /**
4241  * kvm_reset_sys_regs - sets system registers to reset value
4242  * @vcpu: The VCPU pointer
4243  *
4244  * This function finds the right table above and sets the registers on the
4245  * virtual CPU struct to their architecturally defined reset values.
4246  */
4247 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
4248 {
4249 	struct kvm *kvm = vcpu->kvm;
4250 	unsigned long i;
4251 
4252 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4253 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4254 
4255 		if (!r->reset)
4256 			continue;
4257 
4258 		if (is_vm_ftr_id_reg(reg_to_encoding(r)))
4259 			reset_vm_ftr_id_reg(vcpu, r);
4260 		else if (is_vcpu_ftr_id_reg(reg_to_encoding(r)))
4261 			reset_vcpu_ftr_id_reg(vcpu, r);
4262 		else
4263 			r->reset(vcpu, r);
4264 	}
4265 
4266 	set_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags);
4267 }
4268 
4269 /**
4270  * kvm_handle_sys_reg -- handles a system instruction or mrs/msr instruction
4271  *			 trap on a guest execution
4272  * @vcpu: The VCPU pointer
4273  */
4274 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
4275 {
4276 	const struct sys_reg_desc *desc = NULL;
4277 	struct sys_reg_params params;
4278 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
4279 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4280 	int sr_idx;
4281 
4282 	trace_kvm_handle_sys_reg(esr);
4283 
4284 	if (triage_sysreg_trap(vcpu, &sr_idx))
4285 		return 1;
4286 
4287 	params = esr_sys64_to_params(esr);
4288 	params.regval = vcpu_get_reg(vcpu, Rt);
4289 
4290 	/* System registers have Op0=={2,3}, as per DDI487 J.a C5.1.2 */
4291 	if (params.Op0 == 2 || params.Op0 == 3)
4292 		desc = &sys_reg_descs[sr_idx];
4293 	else
4294 		desc = &sys_insn_descs[sr_idx];
4295 
4296 	perform_access(vcpu, &params, desc);
4297 
4298 	/* Read from system register? */
4299 	if (!params.is_write &&
4300 	    (params.Op0 == 2 || params.Op0 == 3))
4301 		vcpu_set_reg(vcpu, Rt, params.regval);
4302 
4303 	return 1;
4304 }
4305 
4306 /******************************************************************************
4307  * Userspace API
4308  *****************************************************************************/
4309 
4310 static bool index_to_params(u64 id, struct sys_reg_params *params)
4311 {
4312 	switch (id & KVM_REG_SIZE_MASK) {
4313 	case KVM_REG_SIZE_U64:
4314 		/* Any unused index bits means it's not valid. */
4315 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
4316 			      | KVM_REG_ARM_COPROC_MASK
4317 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
4318 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
4319 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
4320 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
4321 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
4322 			return false;
4323 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
4324 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
4325 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
4326 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
4327 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
4328 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
4329 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
4330 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
4331 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
4332 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
4333 		return true;
4334 	default:
4335 		return false;
4336 	}
4337 }
4338 
4339 const struct sys_reg_desc *get_reg_by_id(u64 id,
4340 					 const struct sys_reg_desc table[],
4341 					 unsigned int num)
4342 {
4343 	struct sys_reg_params params;
4344 
4345 	if (!index_to_params(id, &params))
4346 		return NULL;
4347 
4348 	return find_reg(&params, table, num);
4349 }
4350 
4351 /* Decode an index value, and find the sys_reg_desc entry. */
4352 static const struct sys_reg_desc *
4353 id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
4354 		   const struct sys_reg_desc table[], unsigned int num)
4355 
4356 {
4357 	const struct sys_reg_desc *r;
4358 
4359 	/* We only do sys_reg for now. */
4360 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
4361 		return NULL;
4362 
4363 	r = get_reg_by_id(id, table, num);
4364 
4365 	/* Not saved in the sys_reg array and not otherwise accessible? */
4366 	if (r && (!(r->reg || r->get_user) || sysreg_hidden(vcpu, r)))
4367 		r = NULL;
4368 
4369 	return r;
4370 }
4371 
4372 /*
4373  * These are the invariant sys_reg registers: we let the guest see the
4374  * host versions of these, so they're part of the guest state.
4375  *
4376  * A future CPU may provide a mechanism to present different values to
4377  * the guest, or a future kvm may trap them.
4378  */
4379 
4380 #define FUNCTION_INVARIANT(reg)						\
4381 	static u64 reset_##reg(struct kvm_vcpu *v,			\
4382 			       const struct sys_reg_desc *r)		\
4383 	{								\
4384 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
4385 		return ((struct sys_reg_desc *)r)->val;			\
4386 	}
4387 
4388 FUNCTION_INVARIANT(midr_el1)
4389 FUNCTION_INVARIANT(revidr_el1)
4390 FUNCTION_INVARIANT(aidr_el1)
4391 
4392 /* ->val is filled in by kvm_sys_reg_table_init() */
4393 static struct sys_reg_desc invariant_sys_regs[] __ro_after_init = {
4394 	{ SYS_DESC(SYS_MIDR_EL1), NULL, reset_midr_el1 },
4395 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, reset_revidr_el1 },
4396 	{ SYS_DESC(SYS_AIDR_EL1), NULL, reset_aidr_el1 },
4397 };
4398 
4399 static int get_invariant_sys_reg(u64 id, u64 __user *uaddr)
4400 {
4401 	const struct sys_reg_desc *r;
4402 
4403 	r = get_reg_by_id(id, invariant_sys_regs,
4404 			  ARRAY_SIZE(invariant_sys_regs));
4405 	if (!r)
4406 		return -ENOENT;
4407 
4408 	return put_user(r->val, uaddr);
4409 }
4410 
4411 static int set_invariant_sys_reg(u64 id, u64 __user *uaddr)
4412 {
4413 	const struct sys_reg_desc *r;
4414 	u64 val;
4415 
4416 	r = get_reg_by_id(id, invariant_sys_regs,
4417 			  ARRAY_SIZE(invariant_sys_regs));
4418 	if (!r)
4419 		return -ENOENT;
4420 
4421 	if (get_user(val, uaddr))
4422 		return -EFAULT;
4423 
4424 	/* This is what we mean by invariant: you can't change it. */
4425 	if (r->val != val)
4426 		return -EINVAL;
4427 
4428 	return 0;
4429 }
4430 
4431 static int demux_c15_get(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
4432 {
4433 	u32 val;
4434 	u32 __user *uval = uaddr;
4435 
4436 	/* Fail if we have unknown bits set. */
4437 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
4438 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
4439 		return -ENOENT;
4440 
4441 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
4442 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
4443 		if (KVM_REG_SIZE(id) != 4)
4444 			return -ENOENT;
4445 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
4446 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
4447 		if (val >= CSSELR_MAX)
4448 			return -ENOENT;
4449 
4450 		return put_user(get_ccsidr(vcpu, val), uval);
4451 	default:
4452 		return -ENOENT;
4453 	}
4454 }
4455 
4456 static int demux_c15_set(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
4457 {
4458 	u32 val, newval;
4459 	u32 __user *uval = uaddr;
4460 
4461 	/* Fail if we have unknown bits set. */
4462 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
4463 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
4464 		return -ENOENT;
4465 
4466 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
4467 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
4468 		if (KVM_REG_SIZE(id) != 4)
4469 			return -ENOENT;
4470 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
4471 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
4472 		if (val >= CSSELR_MAX)
4473 			return -ENOENT;
4474 
4475 		if (get_user(newval, uval))
4476 			return -EFAULT;
4477 
4478 		return set_ccsidr(vcpu, val, newval);
4479 	default:
4480 		return -ENOENT;
4481 	}
4482 }
4483 
4484 int kvm_sys_reg_get_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
4485 			 const struct sys_reg_desc table[], unsigned int num)
4486 {
4487 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
4488 	const struct sys_reg_desc *r;
4489 	u64 val;
4490 	int ret;
4491 
4492 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
4493 	if (!r || sysreg_hidden(vcpu, r))
4494 		return -ENOENT;
4495 
4496 	if (r->get_user) {
4497 		ret = (r->get_user)(vcpu, r, &val);
4498 	} else {
4499 		val = __vcpu_sys_reg(vcpu, r->reg);
4500 		ret = 0;
4501 	}
4502 
4503 	if (!ret)
4504 		ret = put_user(val, uaddr);
4505 
4506 	return ret;
4507 }
4508 
4509 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
4510 {
4511 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
4512 	int err;
4513 
4514 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
4515 		return demux_c15_get(vcpu, reg->id, uaddr);
4516 
4517 	err = get_invariant_sys_reg(reg->id, uaddr);
4518 	if (err != -ENOENT)
4519 		return err;
4520 
4521 	return kvm_sys_reg_get_user(vcpu, reg,
4522 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4523 }
4524 
4525 int kvm_sys_reg_set_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
4526 			 const struct sys_reg_desc table[], unsigned int num)
4527 {
4528 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
4529 	const struct sys_reg_desc *r;
4530 	u64 val;
4531 	int ret;
4532 
4533 	if (get_user(val, uaddr))
4534 		return -EFAULT;
4535 
4536 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
4537 	if (!r || sysreg_hidden(vcpu, r))
4538 		return -ENOENT;
4539 
4540 	if (sysreg_user_write_ignore(vcpu, r))
4541 		return 0;
4542 
4543 	if (r->set_user) {
4544 		ret = (r->set_user)(vcpu, r, val);
4545 	} else {
4546 		__vcpu_sys_reg(vcpu, r->reg) = val;
4547 		ret = 0;
4548 	}
4549 
4550 	return ret;
4551 }
4552 
4553 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
4554 {
4555 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
4556 	int err;
4557 
4558 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
4559 		return demux_c15_set(vcpu, reg->id, uaddr);
4560 
4561 	err = set_invariant_sys_reg(reg->id, uaddr);
4562 	if (err != -ENOENT)
4563 		return err;
4564 
4565 	return kvm_sys_reg_set_user(vcpu, reg,
4566 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4567 }
4568 
4569 static unsigned int num_demux_regs(void)
4570 {
4571 	return CSSELR_MAX;
4572 }
4573 
4574 static int write_demux_regids(u64 __user *uindices)
4575 {
4576 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
4577 	unsigned int i;
4578 
4579 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
4580 	for (i = 0; i < CSSELR_MAX; i++) {
4581 		if (put_user(val | i, uindices))
4582 			return -EFAULT;
4583 		uindices++;
4584 	}
4585 	return 0;
4586 }
4587 
4588 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
4589 {
4590 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
4591 		KVM_REG_ARM64_SYSREG |
4592 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
4593 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
4594 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
4595 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
4596 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
4597 }
4598 
4599 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
4600 {
4601 	if (!*uind)
4602 		return true;
4603 
4604 	if (put_user(sys_reg_to_index(reg), *uind))
4605 		return false;
4606 
4607 	(*uind)++;
4608 	return true;
4609 }
4610 
4611 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
4612 			    const struct sys_reg_desc *rd,
4613 			    u64 __user **uind,
4614 			    unsigned int *total)
4615 {
4616 	/*
4617 	 * Ignore registers we trap but don't save,
4618 	 * and for which no custom user accessor is provided.
4619 	 */
4620 	if (!(rd->reg || rd->get_user))
4621 		return 0;
4622 
4623 	if (sysreg_hidden(vcpu, rd))
4624 		return 0;
4625 
4626 	if (!copy_reg_to_user(rd, uind))
4627 		return -EFAULT;
4628 
4629 	(*total)++;
4630 	return 0;
4631 }
4632 
4633 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
4634 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
4635 {
4636 	const struct sys_reg_desc *i2, *end2;
4637 	unsigned int total = 0;
4638 	int err;
4639 
4640 	i2 = sys_reg_descs;
4641 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
4642 
4643 	while (i2 != end2) {
4644 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
4645 		if (err)
4646 			return err;
4647 	}
4648 	return total;
4649 }
4650 
4651 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
4652 {
4653 	return ARRAY_SIZE(invariant_sys_regs)
4654 		+ num_demux_regs()
4655 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
4656 }
4657 
4658 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
4659 {
4660 	unsigned int i;
4661 	int err;
4662 
4663 	/* Then give them all the invariant registers' indices. */
4664 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
4665 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
4666 			return -EFAULT;
4667 		uindices++;
4668 	}
4669 
4670 	err = walk_sys_regs(vcpu, uindices);
4671 	if (err < 0)
4672 		return err;
4673 	uindices += err;
4674 
4675 	return write_demux_regids(uindices);
4676 }
4677 
4678 #define KVM_ARM_FEATURE_ID_RANGE_INDEX(r)			\
4679 	KVM_ARM_FEATURE_ID_RANGE_IDX(sys_reg_Op0(r),		\
4680 		sys_reg_Op1(r),					\
4681 		sys_reg_CRn(r),					\
4682 		sys_reg_CRm(r),					\
4683 		sys_reg_Op2(r))
4684 
4685 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, struct reg_mask_range *range)
4686 {
4687 	const void *zero_page = page_to_virt(ZERO_PAGE(0));
4688 	u64 __user *masks = (u64 __user *)range->addr;
4689 
4690 	/* Only feature id range is supported, reserved[13] must be zero. */
4691 	if (range->range ||
4692 	    memcmp(range->reserved, zero_page, sizeof(range->reserved)))
4693 		return -EINVAL;
4694 
4695 	/* Wipe the whole thing first */
4696 	if (clear_user(masks, KVM_ARM_FEATURE_ID_RANGE_SIZE * sizeof(__u64)))
4697 		return -EFAULT;
4698 
4699 	for (int i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4700 		const struct sys_reg_desc *reg = &sys_reg_descs[i];
4701 		u32 encoding = reg_to_encoding(reg);
4702 		u64 val;
4703 
4704 		if (!is_feature_id_reg(encoding) || !reg->set_user)
4705 			continue;
4706 
4707 		if (!reg->val ||
4708 		    (is_aa32_id_reg(encoding) && !kvm_supports_32bit_el0())) {
4709 			continue;
4710 		}
4711 		val = reg->val;
4712 
4713 		if (put_user(val, (masks + KVM_ARM_FEATURE_ID_RANGE_INDEX(encoding))))
4714 			return -EFAULT;
4715 	}
4716 
4717 	return 0;
4718 }
4719 
4720 static void vcpu_set_hcr(struct kvm_vcpu *vcpu)
4721 {
4722 	struct kvm *kvm = vcpu->kvm;
4723 
4724 	if (has_vhe() || has_hvhe())
4725 		vcpu->arch.hcr_el2 |= HCR_E2H;
4726 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN)) {
4727 		/* route synchronous external abort exceptions to EL2 */
4728 		vcpu->arch.hcr_el2 |= HCR_TEA;
4729 		/* trap error record accesses */
4730 		vcpu->arch.hcr_el2 |= HCR_TERR;
4731 	}
4732 
4733 	if (cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
4734 		vcpu->arch.hcr_el2 |= HCR_FWB;
4735 
4736 	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
4737 	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
4738 	    kvm_read_vm_id_reg(kvm, SYS_CTR_EL0) == read_sanitised_ftr_reg(SYS_CTR_EL0))
4739 		vcpu->arch.hcr_el2 |= HCR_TID4;
4740 	else
4741 		vcpu->arch.hcr_el2 |= HCR_TID2;
4742 
4743 	if (vcpu_el1_is_32bit(vcpu))
4744 		vcpu->arch.hcr_el2 &= ~HCR_RW;
4745 
4746 	if (kvm_has_mte(vcpu->kvm))
4747 		vcpu->arch.hcr_el2 |= HCR_ATA;
4748 
4749 	/*
4750 	 * In the absence of FGT, we cannot independently trap TLBI
4751 	 * Range instructions. This isn't great, but trapping all
4752 	 * TLBIs would be far worse. Live with it...
4753 	 */
4754 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
4755 		vcpu->arch.hcr_el2 |= HCR_TTLBOS;
4756 }
4757 
4758 void kvm_calculate_traps(struct kvm_vcpu *vcpu)
4759 {
4760 	struct kvm *kvm = vcpu->kvm;
4761 
4762 	mutex_lock(&kvm->arch.config_lock);
4763 	vcpu_set_hcr(vcpu);
4764 	vcpu_set_ich_hcr(vcpu);
4765 
4766 	if (cpus_have_final_cap(ARM64_HAS_HCX)) {
4767 		/*
4768 		 * In general, all HCRX_EL2 bits are gated by a feature.
4769 		 * The only reason we can set SMPME without checking any
4770 		 * feature is that its effects are not directly observable
4771 		 * from the guest.
4772 		 */
4773 		vcpu->arch.hcrx_el2 = HCRX_EL2_SMPME;
4774 
4775 		if (kvm_has_feat(kvm, ID_AA64ISAR2_EL1, MOPS, IMP))
4776 			vcpu->arch.hcrx_el2 |= (HCRX_EL2_MSCEn | HCRX_EL2_MCE2);
4777 
4778 		if (kvm_has_feat(kvm, ID_AA64MMFR3_EL1, TCRX, IMP))
4779 			vcpu->arch.hcrx_el2 |= HCRX_EL2_TCR2En;
4780 
4781 		if (kvm_has_fpmr(kvm))
4782 			vcpu->arch.hcrx_el2 |= HCRX_EL2_EnFPM;
4783 	}
4784 
4785 	if (test_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags))
4786 		goto out;
4787 
4788 	kvm->arch.fgu[HFGxTR_GROUP] = (HFGxTR_EL2_nAMAIR2_EL1		|
4789 				       HFGxTR_EL2_nMAIR2_EL1		|
4790 				       HFGxTR_EL2_nS2POR_EL1		|
4791 				       HFGxTR_EL2_nACCDATA_EL1		|
4792 				       HFGxTR_EL2_nSMPRI_EL1_MASK	|
4793 				       HFGxTR_EL2_nTPIDR2_EL0_MASK);
4794 
4795 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
4796 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_TLBIRVAALE1OS|
4797 						HFGITR_EL2_TLBIRVALE1OS	|
4798 						HFGITR_EL2_TLBIRVAAE1OS	|
4799 						HFGITR_EL2_TLBIRVAE1OS	|
4800 						HFGITR_EL2_TLBIVAALE1OS	|
4801 						HFGITR_EL2_TLBIVALE1OS	|
4802 						HFGITR_EL2_TLBIVAAE1OS	|
4803 						HFGITR_EL2_TLBIASIDE1OS	|
4804 						HFGITR_EL2_TLBIVAE1OS	|
4805 						HFGITR_EL2_TLBIVMALLE1OS);
4806 
4807 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
4808 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_TLBIRVAALE1	|
4809 						HFGITR_EL2_TLBIRVALE1	|
4810 						HFGITR_EL2_TLBIRVAAE1	|
4811 						HFGITR_EL2_TLBIRVAE1	|
4812 						HFGITR_EL2_TLBIRVAALE1IS|
4813 						HFGITR_EL2_TLBIRVALE1IS	|
4814 						HFGITR_EL2_TLBIRVAAE1IS	|
4815 						HFGITR_EL2_TLBIRVAE1IS	|
4816 						HFGITR_EL2_TLBIRVAALE1OS|
4817 						HFGITR_EL2_TLBIRVALE1OS	|
4818 						HFGITR_EL2_TLBIRVAAE1OS	|
4819 						HFGITR_EL2_TLBIRVAE1OS);
4820 
4821 	if (!kvm_has_feat(kvm, ID_AA64ISAR2_EL1, ATS1A, IMP))
4822 		kvm->arch.fgu[HFGITR_GROUP] |= HFGITR_EL2_ATS1E1A;
4823 
4824 	if (!kvm_has_feat(kvm, ID_AA64MMFR1_EL1, PAN, PAN2))
4825 		kvm->arch.fgu[HFGITR_GROUP] |= (HFGITR_EL2_ATS1E1RP |
4826 						HFGITR_EL2_ATS1E1WP);
4827 
4828 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1PIE, IMP))
4829 		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nPIRE0_EL1 |
4830 						HFGxTR_EL2_nPIR_EL1);
4831 
4832 	if (!kvm_has_feat(kvm, ID_AA64MMFR3_EL1, S1POE, IMP))
4833 		kvm->arch.fgu[HFGxTR_GROUP] |= (HFGxTR_EL2_nPOR_EL1 |
4834 						HFGxTR_EL2_nPOR_EL0);
4835 
4836 	if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, AMU, IMP))
4837 		kvm->arch.fgu[HAFGRTR_GROUP] |= ~(HAFGRTR_EL2_RES0 |
4838 						  HAFGRTR_EL2_RES1);
4839 
4840 	set_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags);
4841 out:
4842 	mutex_unlock(&kvm->arch.config_lock);
4843 }
4844 
4845 /*
4846  * Perform last adjustments to the ID registers that are implied by the
4847  * configuration outside of the ID regs themselves, as well as any
4848  * initialisation that directly depend on these ID registers (such as
4849  * RES0/RES1 behaviours). This is not the place to configure traps though.
4850  *
4851  * Because this can be called once per CPU, changes must be idempotent.
4852  */
4853 int kvm_finalize_sys_regs(struct kvm_vcpu *vcpu)
4854 {
4855 	struct kvm *kvm = vcpu->kvm;
4856 
4857 	guard(mutex)(&kvm->arch.config_lock);
4858 
4859 	if (!(static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif) &&
4860 	      irqchip_in_kernel(kvm) &&
4861 	      kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3)) {
4862 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)] &= ~ID_AA64PFR0_EL1_GIC_MASK;
4863 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_PFR1_EL1)] &= ~ID_PFR1_EL1_GIC_MASK;
4864 	}
4865 
4866 	if (vcpu_has_nv(vcpu)) {
4867 		int ret = kvm_init_nv_sysregs(kvm);
4868 		if (ret)
4869 			return ret;
4870 	}
4871 
4872 	return 0;
4873 }
4874 
4875 int __init kvm_sys_reg_table_init(void)
4876 {
4877 	bool valid = true;
4878 	unsigned int i;
4879 	int ret = 0;
4880 
4881 	/* Make sure tables are unique and in order. */
4882 	valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false);
4883 	valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true);
4884 	valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true);
4885 	valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true);
4886 	valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true);
4887 	valid &= check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false);
4888 	valid &= check_sysreg_table(sys_insn_descs, ARRAY_SIZE(sys_insn_descs), false);
4889 
4890 	if (!valid)
4891 		return -EINVAL;
4892 
4893 	/* We abuse the reset function to overwrite the table itself. */
4894 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
4895 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
4896 
4897 	ret = populate_nv_trap_config();
4898 
4899 	for (i = 0; !ret && i < ARRAY_SIZE(sys_reg_descs); i++)
4900 		ret = populate_sysreg_config(sys_reg_descs + i, i);
4901 
4902 	for (i = 0; !ret && i < ARRAY_SIZE(sys_insn_descs); i++)
4903 		ret = populate_sysreg_config(sys_insn_descs + i, i);
4904 
4905 	return ret;
4906 }
4907