xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 0e407915)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/coproc.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
9  *          Christoffer Dall <c.dall@virtualopensystems.com>
10  */
11 
12 #include <linux/bitfield.h>
13 #include <linux/bsearch.h>
14 #include <linux/kvm_host.h>
15 #include <linux/mm.h>
16 #include <linux/printk.h>
17 #include <linux/uaccess.h>
18 
19 #include <asm/cacheflush.h>
20 #include <asm/cputype.h>
21 #include <asm/debug-monitors.h>
22 #include <asm/esr.h>
23 #include <asm/kvm_arm.h>
24 #include <asm/kvm_emulate.h>
25 #include <asm/kvm_hyp.h>
26 #include <asm/kvm_mmu.h>
27 #include <asm/perf_event.h>
28 #include <asm/sysreg.h>
29 
30 #include <trace/events/kvm.h>
31 
32 #include "sys_regs.h"
33 
34 #include "trace.h"
35 
36 /*
37  * For AArch32, we only take care of what is being trapped. Anything
38  * that has to do with init and userspace access has to go via the
39  * 64bit interface.
40  */
41 
42 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
43 
44 static bool read_from_write_only(struct kvm_vcpu *vcpu,
45 				 struct sys_reg_params *params,
46 				 const struct sys_reg_desc *r)
47 {
48 	WARN_ONCE(1, "Unexpected sys_reg read to write-only register\n");
49 	print_sys_reg_instr(params);
50 	kvm_inject_undefined(vcpu);
51 	return false;
52 }
53 
54 static bool write_to_read_only(struct kvm_vcpu *vcpu,
55 			       struct sys_reg_params *params,
56 			       const struct sys_reg_desc *r)
57 {
58 	WARN_ONCE(1, "Unexpected sys_reg write to read-only register\n");
59 	print_sys_reg_instr(params);
60 	kvm_inject_undefined(vcpu);
61 	return false;
62 }
63 
64 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
65 {
66 	u64 val = 0x8badf00d8badf00d;
67 
68 	if (vcpu_get_flag(vcpu, SYSREGS_ON_CPU) &&
69 	    __vcpu_read_sys_reg_from_cpu(reg, &val))
70 		return val;
71 
72 	return __vcpu_sys_reg(vcpu, reg);
73 }
74 
75 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
76 {
77 	if (vcpu_get_flag(vcpu, SYSREGS_ON_CPU) &&
78 	    __vcpu_write_sys_reg_to_cpu(val, reg))
79 		return;
80 
81 	 __vcpu_sys_reg(vcpu, reg) = val;
82 }
83 
84 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
85 static u32 cache_levels;
86 
87 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
88 #define CSSELR_MAX 14
89 
90 /* Which cache CCSIDR represents depends on CSSELR value. */
91 static u32 get_ccsidr(u32 csselr)
92 {
93 	u32 ccsidr;
94 
95 	/* Make sure noone else changes CSSELR during this! */
96 	local_irq_disable();
97 	write_sysreg(csselr, csselr_el1);
98 	isb();
99 	ccsidr = read_sysreg(ccsidr_el1);
100 	local_irq_enable();
101 
102 	return ccsidr;
103 }
104 
105 /*
106  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
107  */
108 static bool access_dcsw(struct kvm_vcpu *vcpu,
109 			struct sys_reg_params *p,
110 			const struct sys_reg_desc *r)
111 {
112 	if (!p->is_write)
113 		return read_from_write_only(vcpu, p, r);
114 
115 	/*
116 	 * Only track S/W ops if we don't have FWB. It still indicates
117 	 * that the guest is a bit broken (S/W operations should only
118 	 * be done by firmware, knowing that there is only a single
119 	 * CPU left in the system, and certainly not from non-secure
120 	 * software).
121 	 */
122 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
123 		kvm_set_way_flush(vcpu);
124 
125 	return true;
126 }
127 
128 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
129 {
130 	switch (r->aarch32_map) {
131 	case AA32_LO:
132 		*mask = GENMASK_ULL(31, 0);
133 		*shift = 0;
134 		break;
135 	case AA32_HI:
136 		*mask = GENMASK_ULL(63, 32);
137 		*shift = 32;
138 		break;
139 	default:
140 		*mask = GENMASK_ULL(63, 0);
141 		*shift = 0;
142 		break;
143 	}
144 }
145 
146 /*
147  * Generic accessor for VM registers. Only called as long as HCR_TVM
148  * is set. If the guest enables the MMU, we stop trapping the VM
149  * sys_regs and leave it in complete control of the caches.
150  */
151 static bool access_vm_reg(struct kvm_vcpu *vcpu,
152 			  struct sys_reg_params *p,
153 			  const struct sys_reg_desc *r)
154 {
155 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
156 	u64 val, mask, shift;
157 
158 	BUG_ON(!p->is_write);
159 
160 	get_access_mask(r, &mask, &shift);
161 
162 	if (~mask) {
163 		val = vcpu_read_sys_reg(vcpu, r->reg);
164 		val &= ~mask;
165 	} else {
166 		val = 0;
167 	}
168 
169 	val |= (p->regval & (mask >> shift)) << shift;
170 	vcpu_write_sys_reg(vcpu, val, r->reg);
171 
172 	kvm_toggle_cache(vcpu, was_enabled);
173 	return true;
174 }
175 
176 static bool access_actlr(struct kvm_vcpu *vcpu,
177 			 struct sys_reg_params *p,
178 			 const struct sys_reg_desc *r)
179 {
180 	u64 mask, shift;
181 
182 	if (p->is_write)
183 		return ignore_write(vcpu, p);
184 
185 	get_access_mask(r, &mask, &shift);
186 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
187 
188 	return true;
189 }
190 
191 /*
192  * Trap handler for the GICv3 SGI generation system register.
193  * Forward the request to the VGIC emulation.
194  * The cp15_64 code makes sure this automatically works
195  * for both AArch64 and AArch32 accesses.
196  */
197 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
198 			   struct sys_reg_params *p,
199 			   const struct sys_reg_desc *r)
200 {
201 	bool g1;
202 
203 	if (!p->is_write)
204 		return read_from_write_only(vcpu, p, r);
205 
206 	/*
207 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
208 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
209 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
210 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
211 	 * group.
212 	 */
213 	if (p->Op0 == 0) {		/* AArch32 */
214 		switch (p->Op1) {
215 		default:		/* Keep GCC quiet */
216 		case 0:			/* ICC_SGI1R */
217 			g1 = true;
218 			break;
219 		case 1:			/* ICC_ASGI1R */
220 		case 2:			/* ICC_SGI0R */
221 			g1 = false;
222 			break;
223 		}
224 	} else {			/* AArch64 */
225 		switch (p->Op2) {
226 		default:		/* Keep GCC quiet */
227 		case 5:			/* ICC_SGI1R_EL1 */
228 			g1 = true;
229 			break;
230 		case 6:			/* ICC_ASGI1R_EL1 */
231 		case 7:			/* ICC_SGI0R_EL1 */
232 			g1 = false;
233 			break;
234 		}
235 	}
236 
237 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
238 
239 	return true;
240 }
241 
242 static bool access_gic_sre(struct kvm_vcpu *vcpu,
243 			   struct sys_reg_params *p,
244 			   const struct sys_reg_desc *r)
245 {
246 	if (p->is_write)
247 		return ignore_write(vcpu, p);
248 
249 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
250 	return true;
251 }
252 
253 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
254 			struct sys_reg_params *p,
255 			const struct sys_reg_desc *r)
256 {
257 	if (p->is_write)
258 		return ignore_write(vcpu, p);
259 	else
260 		return read_zero(vcpu, p);
261 }
262 
263 /*
264  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
265  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
266  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
267  * treat it separately.
268  */
269 static bool trap_loregion(struct kvm_vcpu *vcpu,
270 			  struct sys_reg_params *p,
271 			  const struct sys_reg_desc *r)
272 {
273 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
274 	u32 sr = reg_to_encoding(r);
275 
276 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
277 		kvm_inject_undefined(vcpu);
278 		return false;
279 	}
280 
281 	if (p->is_write && sr == SYS_LORID_EL1)
282 		return write_to_read_only(vcpu, p, r);
283 
284 	return trap_raz_wi(vcpu, p, r);
285 }
286 
287 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
288 			   struct sys_reg_params *p,
289 			   const struct sys_reg_desc *r)
290 {
291 	u64 oslsr;
292 
293 	if (!p->is_write)
294 		return read_from_write_only(vcpu, p, r);
295 
296 	/* Forward the OSLK bit to OSLSR */
297 	oslsr = __vcpu_sys_reg(vcpu, OSLSR_EL1) & ~SYS_OSLSR_OSLK;
298 	if (p->regval & SYS_OSLAR_OSLK)
299 		oslsr |= SYS_OSLSR_OSLK;
300 
301 	__vcpu_sys_reg(vcpu, OSLSR_EL1) = oslsr;
302 	return true;
303 }
304 
305 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
306 			   struct sys_reg_params *p,
307 			   const struct sys_reg_desc *r)
308 {
309 	if (p->is_write)
310 		return write_to_read_only(vcpu, p, r);
311 
312 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
313 	return true;
314 }
315 
316 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
317 			 u64 val)
318 {
319 	/*
320 	 * The only modifiable bit is the OSLK bit. Refuse the write if
321 	 * userspace attempts to change any other bit in the register.
322 	 */
323 	if ((val ^ rd->val) & ~SYS_OSLSR_OSLK)
324 		return -EINVAL;
325 
326 	__vcpu_sys_reg(vcpu, rd->reg) = val;
327 	return 0;
328 }
329 
330 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
331 				   struct sys_reg_params *p,
332 				   const struct sys_reg_desc *r)
333 {
334 	if (p->is_write) {
335 		return ignore_write(vcpu, p);
336 	} else {
337 		p->regval = read_sysreg(dbgauthstatus_el1);
338 		return true;
339 	}
340 }
341 
342 /*
343  * We want to avoid world-switching all the DBG registers all the
344  * time:
345  *
346  * - If we've touched any debug register, it is likely that we're
347  *   going to touch more of them. It then makes sense to disable the
348  *   traps and start doing the save/restore dance
349  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
350  *   then mandatory to save/restore the registers, as the guest
351  *   depends on them.
352  *
353  * For this, we use a DIRTY bit, indicating the guest has modified the
354  * debug registers, used as follow:
355  *
356  * On guest entry:
357  * - If the dirty bit is set (because we're coming back from trapping),
358  *   disable the traps, save host registers, restore guest registers.
359  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
360  *   set the dirty bit, disable the traps, save host registers,
361  *   restore guest registers.
362  * - Otherwise, enable the traps
363  *
364  * On guest exit:
365  * - If the dirty bit is set, save guest registers, restore host
366  *   registers and clear the dirty bit. This ensure that the host can
367  *   now use the debug registers.
368  */
369 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
370 			    struct sys_reg_params *p,
371 			    const struct sys_reg_desc *r)
372 {
373 	if (p->is_write) {
374 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
375 		vcpu_set_flag(vcpu, DEBUG_DIRTY);
376 	} else {
377 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
378 	}
379 
380 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
381 
382 	return true;
383 }
384 
385 /*
386  * reg_to_dbg/dbg_to_reg
387  *
388  * A 32 bit write to a debug register leave top bits alone
389  * A 32 bit read from a debug register only returns the bottom bits
390  *
391  * All writes will set the DEBUG_DIRTY flag to ensure the hyp code
392  * switches between host and guest values in future.
393  */
394 static void reg_to_dbg(struct kvm_vcpu *vcpu,
395 		       struct sys_reg_params *p,
396 		       const struct sys_reg_desc *rd,
397 		       u64 *dbg_reg)
398 {
399 	u64 mask, shift, val;
400 
401 	get_access_mask(rd, &mask, &shift);
402 
403 	val = *dbg_reg;
404 	val &= ~mask;
405 	val |= (p->regval & (mask >> shift)) << shift;
406 	*dbg_reg = val;
407 
408 	vcpu_set_flag(vcpu, DEBUG_DIRTY);
409 }
410 
411 static void dbg_to_reg(struct kvm_vcpu *vcpu,
412 		       struct sys_reg_params *p,
413 		       const struct sys_reg_desc *rd,
414 		       u64 *dbg_reg)
415 {
416 	u64 mask, shift;
417 
418 	get_access_mask(rd, &mask, &shift);
419 	p->regval = (*dbg_reg & mask) >> shift;
420 }
421 
422 static bool trap_bvr(struct kvm_vcpu *vcpu,
423 		     struct sys_reg_params *p,
424 		     const struct sys_reg_desc *rd)
425 {
426 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
427 
428 	if (p->is_write)
429 		reg_to_dbg(vcpu, p, rd, dbg_reg);
430 	else
431 		dbg_to_reg(vcpu, p, rd, dbg_reg);
432 
433 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
434 
435 	return true;
436 }
437 
438 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
439 		   u64 val)
440 {
441 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = val;
442 	return 0;
443 }
444 
445 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
446 		   u64 *val)
447 {
448 	*val = vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
449 	return 0;
450 }
451 
452 static void reset_bvr(struct kvm_vcpu *vcpu,
453 		      const struct sys_reg_desc *rd)
454 {
455 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = rd->val;
456 }
457 
458 static bool trap_bcr(struct kvm_vcpu *vcpu,
459 		     struct sys_reg_params *p,
460 		     const struct sys_reg_desc *rd)
461 {
462 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
463 
464 	if (p->is_write)
465 		reg_to_dbg(vcpu, p, rd, dbg_reg);
466 	else
467 		dbg_to_reg(vcpu, p, rd, dbg_reg);
468 
469 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
470 
471 	return true;
472 }
473 
474 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
475 		   u64 val)
476 {
477 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = val;
478 	return 0;
479 }
480 
481 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
482 		   u64 *val)
483 {
484 	*val = vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
485 	return 0;
486 }
487 
488 static void reset_bcr(struct kvm_vcpu *vcpu,
489 		      const struct sys_reg_desc *rd)
490 {
491 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = rd->val;
492 }
493 
494 static bool trap_wvr(struct kvm_vcpu *vcpu,
495 		     struct sys_reg_params *p,
496 		     const struct sys_reg_desc *rd)
497 {
498 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
499 
500 	if (p->is_write)
501 		reg_to_dbg(vcpu, p, rd, dbg_reg);
502 	else
503 		dbg_to_reg(vcpu, p, rd, dbg_reg);
504 
505 	trace_trap_reg(__func__, rd->CRm, p->is_write,
506 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm]);
507 
508 	return true;
509 }
510 
511 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
512 		   u64 val)
513 {
514 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = val;
515 	return 0;
516 }
517 
518 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
519 		   u64 *val)
520 {
521 	*val = vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
522 	return 0;
523 }
524 
525 static void reset_wvr(struct kvm_vcpu *vcpu,
526 		      const struct sys_reg_desc *rd)
527 {
528 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = rd->val;
529 }
530 
531 static bool trap_wcr(struct kvm_vcpu *vcpu,
532 		     struct sys_reg_params *p,
533 		     const struct sys_reg_desc *rd)
534 {
535 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
536 
537 	if (p->is_write)
538 		reg_to_dbg(vcpu, p, rd, dbg_reg);
539 	else
540 		dbg_to_reg(vcpu, p, rd, dbg_reg);
541 
542 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
543 
544 	return true;
545 }
546 
547 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
548 		   u64 val)
549 {
550 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = val;
551 	return 0;
552 }
553 
554 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
555 		   u64 *val)
556 {
557 	*val = vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
558 	return 0;
559 }
560 
561 static void reset_wcr(struct kvm_vcpu *vcpu,
562 		      const struct sys_reg_desc *rd)
563 {
564 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = rd->val;
565 }
566 
567 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
568 {
569 	u64 amair = read_sysreg(amair_el1);
570 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
571 }
572 
573 static void reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
574 {
575 	u64 actlr = read_sysreg(actlr_el1);
576 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
577 }
578 
579 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
580 {
581 	u64 mpidr;
582 
583 	/*
584 	 * Map the vcpu_id into the first three affinity level fields of
585 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
586 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
587 	 * of the GICv3 to be able to address each CPU directly when
588 	 * sending IPIs.
589 	 */
590 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
591 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
592 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
593 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
594 }
595 
596 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
597 				   const struct sys_reg_desc *r)
598 {
599 	if (kvm_vcpu_has_pmu(vcpu))
600 		return 0;
601 
602 	return REG_HIDDEN;
603 }
604 
605 static void reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
606 {
607 	u64 n, mask = BIT(ARMV8_PMU_CYCLE_IDX);
608 
609 	/* No PMU available, any PMU reg may UNDEF... */
610 	if (!kvm_arm_support_pmu_v3())
611 		return;
612 
613 	n = read_sysreg(pmcr_el0) >> ARMV8_PMU_PMCR_N_SHIFT;
614 	n &= ARMV8_PMU_PMCR_N_MASK;
615 	if (n)
616 		mask |= GENMASK(n - 1, 0);
617 
618 	reset_unknown(vcpu, r);
619 	__vcpu_sys_reg(vcpu, r->reg) &= mask;
620 }
621 
622 static void reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
623 {
624 	reset_unknown(vcpu, r);
625 	__vcpu_sys_reg(vcpu, r->reg) &= GENMASK(31, 0);
626 }
627 
628 static void reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
629 {
630 	reset_unknown(vcpu, r);
631 	__vcpu_sys_reg(vcpu, r->reg) &= ARMV8_PMU_EVTYPE_MASK;
632 }
633 
634 static void reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
635 {
636 	reset_unknown(vcpu, r);
637 	__vcpu_sys_reg(vcpu, r->reg) &= ARMV8_PMU_COUNTER_MASK;
638 }
639 
640 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
641 {
642 	u64 pmcr, val;
643 
644 	/* No PMU available, PMCR_EL0 may UNDEF... */
645 	if (!kvm_arm_support_pmu_v3())
646 		return;
647 
648 	pmcr = read_sysreg(pmcr_el0);
649 	/*
650 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
651 	 * except PMCR.E resetting to zero.
652 	 */
653 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
654 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
655 	if (!kvm_supports_32bit_el0())
656 		val |= ARMV8_PMU_PMCR_LC;
657 	__vcpu_sys_reg(vcpu, r->reg) = val;
658 }
659 
660 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
661 {
662 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
663 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
664 
665 	if (!enabled)
666 		kvm_inject_undefined(vcpu);
667 
668 	return !enabled;
669 }
670 
671 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
672 {
673 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
674 }
675 
676 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
677 {
678 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
679 }
680 
681 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
682 {
683 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
684 }
685 
686 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
687 {
688 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
689 }
690 
691 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
692 			const struct sys_reg_desc *r)
693 {
694 	u64 val;
695 
696 	if (pmu_access_el0_disabled(vcpu))
697 		return false;
698 
699 	if (p->is_write) {
700 		/* Only update writeable bits of PMCR */
701 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
702 		val &= ~ARMV8_PMU_PMCR_MASK;
703 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
704 		if (!kvm_supports_32bit_el0())
705 			val |= ARMV8_PMU_PMCR_LC;
706 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
707 		kvm_pmu_handle_pmcr(vcpu, val);
708 		kvm_vcpu_pmu_restore_guest(vcpu);
709 	} else {
710 		/* PMCR.P & PMCR.C are RAZ */
711 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
712 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
713 		p->regval = val;
714 	}
715 
716 	return true;
717 }
718 
719 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
720 			  const struct sys_reg_desc *r)
721 {
722 	if (pmu_access_event_counter_el0_disabled(vcpu))
723 		return false;
724 
725 	if (p->is_write)
726 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
727 	else
728 		/* return PMSELR.SEL field */
729 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
730 			    & ARMV8_PMU_COUNTER_MASK;
731 
732 	return true;
733 }
734 
735 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
736 			  const struct sys_reg_desc *r)
737 {
738 	u64 pmceid, mask, shift;
739 
740 	BUG_ON(p->is_write);
741 
742 	if (pmu_access_el0_disabled(vcpu))
743 		return false;
744 
745 	get_access_mask(r, &mask, &shift);
746 
747 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
748 	pmceid &= mask;
749 	pmceid >>= shift;
750 
751 	p->regval = pmceid;
752 
753 	return true;
754 }
755 
756 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
757 {
758 	u64 pmcr, val;
759 
760 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
761 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
762 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
763 		kvm_inject_undefined(vcpu);
764 		return false;
765 	}
766 
767 	return true;
768 }
769 
770 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
771 			      struct sys_reg_params *p,
772 			      const struct sys_reg_desc *r)
773 {
774 	u64 idx = ~0UL;
775 
776 	if (r->CRn == 9 && r->CRm == 13) {
777 		if (r->Op2 == 2) {
778 			/* PMXEVCNTR_EL0 */
779 			if (pmu_access_event_counter_el0_disabled(vcpu))
780 				return false;
781 
782 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
783 			      & ARMV8_PMU_COUNTER_MASK;
784 		} else if (r->Op2 == 0) {
785 			/* PMCCNTR_EL0 */
786 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
787 				return false;
788 
789 			idx = ARMV8_PMU_CYCLE_IDX;
790 		}
791 	} else if (r->CRn == 0 && r->CRm == 9) {
792 		/* PMCCNTR */
793 		if (pmu_access_event_counter_el0_disabled(vcpu))
794 			return false;
795 
796 		idx = ARMV8_PMU_CYCLE_IDX;
797 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
798 		/* PMEVCNTRn_EL0 */
799 		if (pmu_access_event_counter_el0_disabled(vcpu))
800 			return false;
801 
802 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
803 	}
804 
805 	/* Catch any decoding mistake */
806 	WARN_ON(idx == ~0UL);
807 
808 	if (!pmu_counter_idx_valid(vcpu, idx))
809 		return false;
810 
811 	if (p->is_write) {
812 		if (pmu_access_el0_disabled(vcpu))
813 			return false;
814 
815 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
816 	} else {
817 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
818 	}
819 
820 	return true;
821 }
822 
823 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
824 			       const struct sys_reg_desc *r)
825 {
826 	u64 idx, reg;
827 
828 	if (pmu_access_el0_disabled(vcpu))
829 		return false;
830 
831 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
832 		/* PMXEVTYPER_EL0 */
833 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
834 		reg = PMEVTYPER0_EL0 + idx;
835 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
836 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
837 		if (idx == ARMV8_PMU_CYCLE_IDX)
838 			reg = PMCCFILTR_EL0;
839 		else
840 			/* PMEVTYPERn_EL0 */
841 			reg = PMEVTYPER0_EL0 + idx;
842 	} else {
843 		BUG();
844 	}
845 
846 	if (!pmu_counter_idx_valid(vcpu, idx))
847 		return false;
848 
849 	if (p->is_write) {
850 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
851 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
852 		kvm_vcpu_pmu_restore_guest(vcpu);
853 	} else {
854 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
855 	}
856 
857 	return true;
858 }
859 
860 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
861 			   const struct sys_reg_desc *r)
862 {
863 	u64 val, mask;
864 
865 	if (pmu_access_el0_disabled(vcpu))
866 		return false;
867 
868 	mask = kvm_pmu_valid_counter_mask(vcpu);
869 	if (p->is_write) {
870 		val = p->regval & mask;
871 		if (r->Op2 & 0x1) {
872 			/* accessing PMCNTENSET_EL0 */
873 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
874 			kvm_pmu_enable_counter_mask(vcpu, val);
875 			kvm_vcpu_pmu_restore_guest(vcpu);
876 		} else {
877 			/* accessing PMCNTENCLR_EL0 */
878 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
879 			kvm_pmu_disable_counter_mask(vcpu, val);
880 		}
881 	} else {
882 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
883 	}
884 
885 	return true;
886 }
887 
888 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
889 			   const struct sys_reg_desc *r)
890 {
891 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
892 
893 	if (check_pmu_access_disabled(vcpu, 0))
894 		return false;
895 
896 	if (p->is_write) {
897 		u64 val = p->regval & mask;
898 
899 		if (r->Op2 & 0x1)
900 			/* accessing PMINTENSET_EL1 */
901 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
902 		else
903 			/* accessing PMINTENCLR_EL1 */
904 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
905 	} else {
906 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
907 	}
908 
909 	return true;
910 }
911 
912 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
913 			 const struct sys_reg_desc *r)
914 {
915 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
916 
917 	if (pmu_access_el0_disabled(vcpu))
918 		return false;
919 
920 	if (p->is_write) {
921 		if (r->CRm & 0x2)
922 			/* accessing PMOVSSET_EL0 */
923 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
924 		else
925 			/* accessing PMOVSCLR_EL0 */
926 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
927 	} else {
928 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
929 	}
930 
931 	return true;
932 }
933 
934 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
935 			   const struct sys_reg_desc *r)
936 {
937 	u64 mask;
938 
939 	if (!p->is_write)
940 		return read_from_write_only(vcpu, p, r);
941 
942 	if (pmu_write_swinc_el0_disabled(vcpu))
943 		return false;
944 
945 	mask = kvm_pmu_valid_counter_mask(vcpu);
946 	kvm_pmu_software_increment(vcpu, p->regval & mask);
947 	return true;
948 }
949 
950 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
951 			     const struct sys_reg_desc *r)
952 {
953 	if (p->is_write) {
954 		if (!vcpu_mode_priv(vcpu)) {
955 			kvm_inject_undefined(vcpu);
956 			return false;
957 		}
958 
959 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
960 			       p->regval & ARMV8_PMU_USERENR_MASK;
961 	} else {
962 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
963 			    & ARMV8_PMU_USERENR_MASK;
964 	}
965 
966 	return true;
967 }
968 
969 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
970 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
971 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
972 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
973 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
974 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
975 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
976 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
977 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
978 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
979 
980 #define PMU_SYS_REG(r)						\
981 	SYS_DESC(r), .reset = reset_pmu_reg, .visibility = pmu_visibility
982 
983 /* Macro to expand the PMEVCNTRn_EL0 register */
984 #define PMU_PMEVCNTR_EL0(n)						\
985 	{ PMU_SYS_REG(SYS_PMEVCNTRn_EL0(n)),				\
986 	  .reset = reset_pmevcntr,					\
987 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
988 
989 /* Macro to expand the PMEVTYPERn_EL0 register */
990 #define PMU_PMEVTYPER_EL0(n)						\
991 	{ PMU_SYS_REG(SYS_PMEVTYPERn_EL0(n)),				\
992 	  .reset = reset_pmevtyper,					\
993 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
994 
995 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
996 			 const struct sys_reg_desc *r)
997 {
998 	kvm_inject_undefined(vcpu);
999 
1000 	return false;
1001 }
1002 
1003 /* Macro to expand the AMU counter and type registers*/
1004 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1005 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1006 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1007 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1008 
1009 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1010 			const struct sys_reg_desc *rd)
1011 {
1012 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1013 }
1014 
1015 /*
1016  * If we land here on a PtrAuth access, that is because we didn't
1017  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1018  * way this happens is when the guest does not have PtrAuth support
1019  * enabled.
1020  */
1021 #define __PTRAUTH_KEY(k)						\
1022 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1023 	.visibility = ptrauth_visibility}
1024 
1025 #define PTRAUTH_KEY(k)							\
1026 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1027 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1028 
1029 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1030 			      struct sys_reg_params *p,
1031 			      const struct sys_reg_desc *r)
1032 {
1033 	enum kvm_arch_timers tmr;
1034 	enum kvm_arch_timer_regs treg;
1035 	u64 reg = reg_to_encoding(r);
1036 
1037 	switch (reg) {
1038 	case SYS_CNTP_TVAL_EL0:
1039 	case SYS_AARCH32_CNTP_TVAL:
1040 		tmr = TIMER_PTIMER;
1041 		treg = TIMER_REG_TVAL;
1042 		break;
1043 	case SYS_CNTP_CTL_EL0:
1044 	case SYS_AARCH32_CNTP_CTL:
1045 		tmr = TIMER_PTIMER;
1046 		treg = TIMER_REG_CTL;
1047 		break;
1048 	case SYS_CNTP_CVAL_EL0:
1049 	case SYS_AARCH32_CNTP_CVAL:
1050 		tmr = TIMER_PTIMER;
1051 		treg = TIMER_REG_CVAL;
1052 		break;
1053 	default:
1054 		BUG();
1055 	}
1056 
1057 	if (p->is_write)
1058 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1059 	else
1060 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1061 
1062 	return true;
1063 }
1064 
1065 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1066 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1067 		struct sys_reg_desc const *r, bool raz)
1068 {
1069 	u32 id = reg_to_encoding(r);
1070 	u64 val;
1071 
1072 	if (raz)
1073 		return 0;
1074 
1075 	val = read_sanitised_ftr_reg(id);
1076 
1077 	switch (id) {
1078 	case SYS_ID_AA64PFR0_EL1:
1079 		if (!vcpu_has_sve(vcpu))
1080 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_SVE);
1081 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_AMU);
1082 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2);
1083 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2), (u64)vcpu->kvm->arch.pfr0_csv2);
1084 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3);
1085 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3), (u64)vcpu->kvm->arch.pfr0_csv3);
1086 		if (kvm_vgic_global_state.type == VGIC_V3) {
1087 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_GIC);
1088 			val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_GIC), 1);
1089 		}
1090 		break;
1091 	case SYS_ID_AA64PFR1_EL1:
1092 		if (!kvm_has_mte(vcpu->kvm))
1093 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_MTE);
1094 
1095 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_SME);
1096 		break;
1097 	case SYS_ID_AA64ISAR1_EL1:
1098 		if (!vcpu_has_ptrauth(vcpu))
1099 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_APA) |
1100 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_API) |
1101 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPA) |
1102 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_EL1_GPI));
1103 		break;
1104 	case SYS_ID_AA64ISAR2_EL1:
1105 		if (!vcpu_has_ptrauth(vcpu))
1106 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_APA3) |
1107 				 ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_GPA3));
1108 		if (!cpus_have_final_cap(ARM64_HAS_WFXT))
1109 			val &= ~ARM64_FEATURE_MASK(ID_AA64ISAR2_EL1_WFxT);
1110 		break;
1111 	case SYS_ID_AA64DFR0_EL1:
1112 		/* Limit debug to ARMv8.0 */
1113 		val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER);
1114 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER), 6);
1115 		/* Limit guests to PMUv3 for ARMv8.4 */
1116 		val = cpuid_feature_cap_perfmon_field(val,
1117 						      ID_AA64DFR0_PMUVER_SHIFT,
1118 						      kvm_vcpu_has_pmu(vcpu) ? ID_AA64DFR0_PMUVER_8_4 : 0);
1119 		/* Hide SPE from guests */
1120 		val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_PMSVER);
1121 		break;
1122 	case SYS_ID_DFR0_EL1:
1123 		/* Limit guests to PMUv3 for ARMv8.4 */
1124 		val = cpuid_feature_cap_perfmon_field(val,
1125 						      ID_DFR0_PERFMON_SHIFT,
1126 						      kvm_vcpu_has_pmu(vcpu) ? ID_DFR0_PERFMON_8_4 : 0);
1127 		break;
1128 	}
1129 
1130 	return val;
1131 }
1132 
1133 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1134 				  const struct sys_reg_desc *r)
1135 {
1136 	u32 id = reg_to_encoding(r);
1137 
1138 	switch (id) {
1139 	case SYS_ID_AA64ZFR0_EL1:
1140 		if (!vcpu_has_sve(vcpu))
1141 			return REG_RAZ;
1142 		break;
1143 	}
1144 
1145 	return 0;
1146 }
1147 
1148 /* cpufeature ID register access trap handlers */
1149 
1150 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1151 			    struct sys_reg_params *p,
1152 			    const struct sys_reg_desc *r,
1153 			    bool raz)
1154 {
1155 	if (p->is_write)
1156 		return write_to_read_only(vcpu, p, r);
1157 
1158 	p->regval = read_id_reg(vcpu, r, raz);
1159 	return true;
1160 }
1161 
1162 static bool access_id_reg(struct kvm_vcpu *vcpu,
1163 			  struct sys_reg_params *p,
1164 			  const struct sys_reg_desc *r)
1165 {
1166 	bool raz = sysreg_visible_as_raz(vcpu, r);
1167 
1168 	return __access_id_reg(vcpu, p, r, raz);
1169 }
1170 
1171 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1172 			      struct sys_reg_params *p,
1173 			      const struct sys_reg_desc *r)
1174 {
1175 	return __access_id_reg(vcpu, p, r, true);
1176 }
1177 
1178 /* Visibility overrides for SVE-specific control registers */
1179 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1180 				   const struct sys_reg_desc *rd)
1181 {
1182 	if (vcpu_has_sve(vcpu))
1183 		return 0;
1184 
1185 	return REG_HIDDEN;
1186 }
1187 
1188 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1189 			       const struct sys_reg_desc *rd,
1190 			       u64 val)
1191 {
1192 	u8 csv2, csv3;
1193 
1194 	/*
1195 	 * Allow AA64PFR0_EL1.CSV2 to be set from userspace as long as
1196 	 * it doesn't promise more than what is actually provided (the
1197 	 * guest could otherwise be covered in ectoplasmic residue).
1198 	 */
1199 	csv2 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV2_SHIFT);
1200 	if (csv2 > 1 ||
1201 	    (csv2 && arm64_get_spectre_v2_state() != SPECTRE_UNAFFECTED))
1202 		return -EINVAL;
1203 
1204 	/* Same thing for CSV3 */
1205 	csv3 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV3_SHIFT);
1206 	if (csv3 > 1 ||
1207 	    (csv3 && arm64_get_meltdown_state() != SPECTRE_UNAFFECTED))
1208 		return -EINVAL;
1209 
1210 	/* We can only differ with CSV[23], and anything else is an error */
1211 	val ^= read_id_reg(vcpu, rd, false);
1212 	val &= ~((0xFUL << ID_AA64PFR0_CSV2_SHIFT) |
1213 		 (0xFUL << ID_AA64PFR0_CSV3_SHIFT));
1214 	if (val)
1215 		return -EINVAL;
1216 
1217 	vcpu->kvm->arch.pfr0_csv2 = csv2;
1218 	vcpu->kvm->arch.pfr0_csv3 = csv3;
1219 
1220 	return 0;
1221 }
1222 
1223 /*
1224  * cpufeature ID register user accessors
1225  *
1226  * For now, these registers are immutable for userspace, so no values
1227  * are stored, and for set_id_reg() we don't allow the effective value
1228  * to be changed.
1229  */
1230 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1231 			const struct sys_reg_desc *rd, u64 *val,
1232 			bool raz)
1233 {
1234 	*val = read_id_reg(vcpu, rd, raz);
1235 	return 0;
1236 }
1237 
1238 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1239 			const struct sys_reg_desc *rd, u64 val,
1240 			bool raz)
1241 {
1242 	/* This is what we mean by invariant: you can't change it. */
1243 	if (val != read_id_reg(vcpu, rd, raz))
1244 		return -EINVAL;
1245 
1246 	return 0;
1247 }
1248 
1249 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1250 		      u64 *val)
1251 {
1252 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1253 
1254 	return __get_id_reg(vcpu, rd, val, raz);
1255 }
1256 
1257 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1258 		      u64 val)
1259 {
1260 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1261 
1262 	return __set_id_reg(vcpu, rd, val, raz);
1263 }
1264 
1265 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1266 			  u64 val)
1267 {
1268 	return __set_id_reg(vcpu, rd, val, true);
1269 }
1270 
1271 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1272 		       u64 *val)
1273 {
1274 	*val = 0;
1275 	return 0;
1276 }
1277 
1278 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1279 		      u64 val)
1280 {
1281 	return 0;
1282 }
1283 
1284 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1285 		       const struct sys_reg_desc *r)
1286 {
1287 	if (p->is_write)
1288 		return write_to_read_only(vcpu, p, r);
1289 
1290 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1291 	return true;
1292 }
1293 
1294 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1295 			 const struct sys_reg_desc *r)
1296 {
1297 	if (p->is_write)
1298 		return write_to_read_only(vcpu, p, r);
1299 
1300 	p->regval = read_sysreg(clidr_el1);
1301 	return true;
1302 }
1303 
1304 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1305 			  const struct sys_reg_desc *r)
1306 {
1307 	int reg = r->reg;
1308 
1309 	if (p->is_write)
1310 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1311 	else
1312 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1313 	return true;
1314 }
1315 
1316 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1317 			  const struct sys_reg_desc *r)
1318 {
1319 	u32 csselr;
1320 
1321 	if (p->is_write)
1322 		return write_to_read_only(vcpu, p, r);
1323 
1324 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1325 	p->regval = get_ccsidr(csselr);
1326 
1327 	/*
1328 	 * Guests should not be doing cache operations by set/way at all, and
1329 	 * for this reason, we trap them and attempt to infer the intent, so
1330 	 * that we can flush the entire guest's address space at the appropriate
1331 	 * time.
1332 	 * To prevent this trapping from causing performance problems, let's
1333 	 * expose the geometry of all data and unified caches (which are
1334 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1335 	 * [If guests should attempt to infer aliasing properties from the
1336 	 * geometry (which is not permitted by the architecture), they would
1337 	 * only do so for virtually indexed caches.]
1338 	 */
1339 	if (!(csselr & 1)) // data or unified cache
1340 		p->regval &= ~GENMASK(27, 3);
1341 	return true;
1342 }
1343 
1344 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
1345 				   const struct sys_reg_desc *rd)
1346 {
1347 	if (kvm_has_mte(vcpu->kvm))
1348 		return 0;
1349 
1350 	return REG_HIDDEN;
1351 }
1352 
1353 #define MTE_REG(name) {				\
1354 	SYS_DESC(SYS_##name),			\
1355 	.access = undef_access,			\
1356 	.reset = reset_unknown,			\
1357 	.reg = name,				\
1358 	.visibility = mte_visibility,		\
1359 }
1360 
1361 /* sys_reg_desc initialiser for known cpufeature ID registers */
1362 #define ID_SANITISED(name) {			\
1363 	SYS_DESC(SYS_##name),			\
1364 	.access	= access_id_reg,		\
1365 	.get_user = get_id_reg,			\
1366 	.set_user = set_id_reg,			\
1367 	.visibility = id_visibility,		\
1368 }
1369 
1370 /*
1371  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1372  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1373  * (1 <= crm < 8, 0 <= Op2 < 8).
1374  */
1375 #define ID_UNALLOCATED(crm, op2) {			\
1376 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1377 	.access = access_raz_id_reg,			\
1378 	.get_user = get_raz_reg,			\
1379 	.set_user = set_raz_id_reg,			\
1380 }
1381 
1382 /*
1383  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1384  * For now, these are exposed just like unallocated ID regs: they appear
1385  * RAZ for the guest.
1386  */
1387 #define ID_HIDDEN(name) {			\
1388 	SYS_DESC(SYS_##name),			\
1389 	.access = access_raz_id_reg,		\
1390 	.get_user = get_raz_reg,		\
1391 	.set_user = set_raz_id_reg,		\
1392 }
1393 
1394 /*
1395  * Architected system registers.
1396  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1397  *
1398  * Debug handling: We do trap most, if not all debug related system
1399  * registers. The implementation is good enough to ensure that a guest
1400  * can use these with minimal performance degradation. The drawback is
1401  * that we don't implement any of the external debug architecture.
1402  * This should be revisited if we ever encounter a more demanding
1403  * guest...
1404  */
1405 static const struct sys_reg_desc sys_reg_descs[] = {
1406 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1407 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1408 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1409 
1410 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1411 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1412 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1413 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1414 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1415 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1416 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1417 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1418 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1419 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1420 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1421 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1422 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1423 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1424 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1425 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1426 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1427 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1428 
1429 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1430 	{ SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
1431 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
1432 		SYS_OSLSR_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
1433 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1434 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1435 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1436 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1437 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1438 
1439 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1440 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1441 	// DBGDTR[TR]X_EL0 share the same encoding
1442 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1443 
1444 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1445 
1446 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1447 
1448 	/*
1449 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1450 	 * entries in arm64_ftr_regs[].
1451 	 */
1452 
1453 	/* AArch64 mappings of the AArch32 ID registers */
1454 	/* CRm=1 */
1455 	ID_SANITISED(ID_PFR0_EL1),
1456 	ID_SANITISED(ID_PFR1_EL1),
1457 	ID_SANITISED(ID_DFR0_EL1),
1458 	ID_HIDDEN(ID_AFR0_EL1),
1459 	ID_SANITISED(ID_MMFR0_EL1),
1460 	ID_SANITISED(ID_MMFR1_EL1),
1461 	ID_SANITISED(ID_MMFR2_EL1),
1462 	ID_SANITISED(ID_MMFR3_EL1),
1463 
1464 	/* CRm=2 */
1465 	ID_SANITISED(ID_ISAR0_EL1),
1466 	ID_SANITISED(ID_ISAR1_EL1),
1467 	ID_SANITISED(ID_ISAR2_EL1),
1468 	ID_SANITISED(ID_ISAR3_EL1),
1469 	ID_SANITISED(ID_ISAR4_EL1),
1470 	ID_SANITISED(ID_ISAR5_EL1),
1471 	ID_SANITISED(ID_MMFR4_EL1),
1472 	ID_SANITISED(ID_ISAR6_EL1),
1473 
1474 	/* CRm=3 */
1475 	ID_SANITISED(MVFR0_EL1),
1476 	ID_SANITISED(MVFR1_EL1),
1477 	ID_SANITISED(MVFR2_EL1),
1478 	ID_UNALLOCATED(3,3),
1479 	ID_SANITISED(ID_PFR2_EL1),
1480 	ID_HIDDEN(ID_DFR1_EL1),
1481 	ID_SANITISED(ID_MMFR5_EL1),
1482 	ID_UNALLOCATED(3,7),
1483 
1484 	/* AArch64 ID registers */
1485 	/* CRm=4 */
1486 	{ SYS_DESC(SYS_ID_AA64PFR0_EL1), .access = access_id_reg,
1487 	  .get_user = get_id_reg, .set_user = set_id_aa64pfr0_el1, },
1488 	ID_SANITISED(ID_AA64PFR1_EL1),
1489 	ID_UNALLOCATED(4,2),
1490 	ID_UNALLOCATED(4,3),
1491 	ID_SANITISED(ID_AA64ZFR0_EL1),
1492 	ID_HIDDEN(ID_AA64SMFR0_EL1),
1493 	ID_UNALLOCATED(4,6),
1494 	ID_UNALLOCATED(4,7),
1495 
1496 	/* CRm=5 */
1497 	ID_SANITISED(ID_AA64DFR0_EL1),
1498 	ID_SANITISED(ID_AA64DFR1_EL1),
1499 	ID_UNALLOCATED(5,2),
1500 	ID_UNALLOCATED(5,3),
1501 	ID_HIDDEN(ID_AA64AFR0_EL1),
1502 	ID_HIDDEN(ID_AA64AFR1_EL1),
1503 	ID_UNALLOCATED(5,6),
1504 	ID_UNALLOCATED(5,7),
1505 
1506 	/* CRm=6 */
1507 	ID_SANITISED(ID_AA64ISAR0_EL1),
1508 	ID_SANITISED(ID_AA64ISAR1_EL1),
1509 	ID_SANITISED(ID_AA64ISAR2_EL1),
1510 	ID_UNALLOCATED(6,3),
1511 	ID_UNALLOCATED(6,4),
1512 	ID_UNALLOCATED(6,5),
1513 	ID_UNALLOCATED(6,6),
1514 	ID_UNALLOCATED(6,7),
1515 
1516 	/* CRm=7 */
1517 	ID_SANITISED(ID_AA64MMFR0_EL1),
1518 	ID_SANITISED(ID_AA64MMFR1_EL1),
1519 	ID_SANITISED(ID_AA64MMFR2_EL1),
1520 	ID_UNALLOCATED(7,3),
1521 	ID_UNALLOCATED(7,4),
1522 	ID_UNALLOCATED(7,5),
1523 	ID_UNALLOCATED(7,6),
1524 	ID_UNALLOCATED(7,7),
1525 
1526 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1527 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1528 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1529 
1530 	MTE_REG(RGSR_EL1),
1531 	MTE_REG(GCR_EL1),
1532 
1533 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1534 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
1535 	{ SYS_DESC(SYS_SMPRI_EL1), undef_access },
1536 	{ SYS_DESC(SYS_SMCR_EL1), undef_access },
1537 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1538 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1539 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1540 
1541 	PTRAUTH_KEY(APIA),
1542 	PTRAUTH_KEY(APIB),
1543 	PTRAUTH_KEY(APDA),
1544 	PTRAUTH_KEY(APDB),
1545 	PTRAUTH_KEY(APGA),
1546 
1547 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1548 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1549 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1550 
1551 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1552 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1553 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1554 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1555 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1556 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1557 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1558 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1559 
1560 	MTE_REG(TFSR_EL1),
1561 	MTE_REG(TFSRE0_EL1),
1562 
1563 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1564 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1565 
1566 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
1567 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
1568 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
1569 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
1570 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
1571 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
1572 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
1573 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
1574 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
1575 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
1576 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
1577 	/* PMBIDR_EL1 is not trapped */
1578 
1579 	{ PMU_SYS_REG(SYS_PMINTENSET_EL1),
1580 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1581 	{ PMU_SYS_REG(SYS_PMINTENCLR_EL1),
1582 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1583 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
1584 
1585 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1586 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1587 
1588 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1589 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1590 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1591 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1592 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1593 
1594 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1595 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1596 
1597 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1598 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1599 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1600 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1601 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1602 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1603 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1604 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1605 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1606 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1607 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1608 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1609 
1610 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1611 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1612 
1613 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
1614 
1615 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1616 
1617 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1618 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1619 	{ SYS_DESC(SYS_SMIDR_EL1), undef_access },
1620 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1621 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1622 	{ SYS_DESC(SYS_SVCR), undef_access },
1623 
1624 	{ PMU_SYS_REG(SYS_PMCR_EL0), .access = access_pmcr,
1625 	  .reset = reset_pmcr, .reg = PMCR_EL0 },
1626 	{ PMU_SYS_REG(SYS_PMCNTENSET_EL0),
1627 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1628 	{ PMU_SYS_REG(SYS_PMCNTENCLR_EL0),
1629 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1630 	{ PMU_SYS_REG(SYS_PMOVSCLR_EL0),
1631 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1632 	/*
1633 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
1634 	 * previously (and pointlessly) advertised in the past...
1635 	 */
1636 	{ PMU_SYS_REG(SYS_PMSWINC_EL0),
1637 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
1638 	  .access = access_pmswinc, .reset = NULL },
1639 	{ PMU_SYS_REG(SYS_PMSELR_EL0),
1640 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
1641 	{ PMU_SYS_REG(SYS_PMCEID0_EL0),
1642 	  .access = access_pmceid, .reset = NULL },
1643 	{ PMU_SYS_REG(SYS_PMCEID1_EL0),
1644 	  .access = access_pmceid, .reset = NULL },
1645 	{ PMU_SYS_REG(SYS_PMCCNTR_EL0),
1646 	  .access = access_pmu_evcntr, .reset = reset_unknown, .reg = PMCCNTR_EL0 },
1647 	{ PMU_SYS_REG(SYS_PMXEVTYPER_EL0),
1648 	  .access = access_pmu_evtyper, .reset = NULL },
1649 	{ PMU_SYS_REG(SYS_PMXEVCNTR_EL0),
1650 	  .access = access_pmu_evcntr, .reset = NULL },
1651 	/*
1652 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1653 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1654 	 */
1655 	{ PMU_SYS_REG(SYS_PMUSERENR_EL0), .access = access_pmuserenr,
1656 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
1657 	{ PMU_SYS_REG(SYS_PMOVSSET_EL0),
1658 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1659 
1660 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1661 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1662 	{ SYS_DESC(SYS_TPIDR2_EL0), undef_access },
1663 
1664 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
1665 
1666 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
1667 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
1668 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
1669 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
1670 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
1671 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
1672 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
1673 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
1674 	AMU_AMEVCNTR0_EL0(0),
1675 	AMU_AMEVCNTR0_EL0(1),
1676 	AMU_AMEVCNTR0_EL0(2),
1677 	AMU_AMEVCNTR0_EL0(3),
1678 	AMU_AMEVCNTR0_EL0(4),
1679 	AMU_AMEVCNTR0_EL0(5),
1680 	AMU_AMEVCNTR0_EL0(6),
1681 	AMU_AMEVCNTR0_EL0(7),
1682 	AMU_AMEVCNTR0_EL0(8),
1683 	AMU_AMEVCNTR0_EL0(9),
1684 	AMU_AMEVCNTR0_EL0(10),
1685 	AMU_AMEVCNTR0_EL0(11),
1686 	AMU_AMEVCNTR0_EL0(12),
1687 	AMU_AMEVCNTR0_EL0(13),
1688 	AMU_AMEVCNTR0_EL0(14),
1689 	AMU_AMEVCNTR0_EL0(15),
1690 	AMU_AMEVTYPER0_EL0(0),
1691 	AMU_AMEVTYPER0_EL0(1),
1692 	AMU_AMEVTYPER0_EL0(2),
1693 	AMU_AMEVTYPER0_EL0(3),
1694 	AMU_AMEVTYPER0_EL0(4),
1695 	AMU_AMEVTYPER0_EL0(5),
1696 	AMU_AMEVTYPER0_EL0(6),
1697 	AMU_AMEVTYPER0_EL0(7),
1698 	AMU_AMEVTYPER0_EL0(8),
1699 	AMU_AMEVTYPER0_EL0(9),
1700 	AMU_AMEVTYPER0_EL0(10),
1701 	AMU_AMEVTYPER0_EL0(11),
1702 	AMU_AMEVTYPER0_EL0(12),
1703 	AMU_AMEVTYPER0_EL0(13),
1704 	AMU_AMEVTYPER0_EL0(14),
1705 	AMU_AMEVTYPER0_EL0(15),
1706 	AMU_AMEVCNTR1_EL0(0),
1707 	AMU_AMEVCNTR1_EL0(1),
1708 	AMU_AMEVCNTR1_EL0(2),
1709 	AMU_AMEVCNTR1_EL0(3),
1710 	AMU_AMEVCNTR1_EL0(4),
1711 	AMU_AMEVCNTR1_EL0(5),
1712 	AMU_AMEVCNTR1_EL0(6),
1713 	AMU_AMEVCNTR1_EL0(7),
1714 	AMU_AMEVCNTR1_EL0(8),
1715 	AMU_AMEVCNTR1_EL0(9),
1716 	AMU_AMEVCNTR1_EL0(10),
1717 	AMU_AMEVCNTR1_EL0(11),
1718 	AMU_AMEVCNTR1_EL0(12),
1719 	AMU_AMEVCNTR1_EL0(13),
1720 	AMU_AMEVCNTR1_EL0(14),
1721 	AMU_AMEVCNTR1_EL0(15),
1722 	AMU_AMEVTYPER1_EL0(0),
1723 	AMU_AMEVTYPER1_EL0(1),
1724 	AMU_AMEVTYPER1_EL0(2),
1725 	AMU_AMEVTYPER1_EL0(3),
1726 	AMU_AMEVTYPER1_EL0(4),
1727 	AMU_AMEVTYPER1_EL0(5),
1728 	AMU_AMEVTYPER1_EL0(6),
1729 	AMU_AMEVTYPER1_EL0(7),
1730 	AMU_AMEVTYPER1_EL0(8),
1731 	AMU_AMEVTYPER1_EL0(9),
1732 	AMU_AMEVTYPER1_EL0(10),
1733 	AMU_AMEVTYPER1_EL0(11),
1734 	AMU_AMEVTYPER1_EL0(12),
1735 	AMU_AMEVTYPER1_EL0(13),
1736 	AMU_AMEVTYPER1_EL0(14),
1737 	AMU_AMEVTYPER1_EL0(15),
1738 
1739 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1740 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1741 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1742 
1743 	/* PMEVCNTRn_EL0 */
1744 	PMU_PMEVCNTR_EL0(0),
1745 	PMU_PMEVCNTR_EL0(1),
1746 	PMU_PMEVCNTR_EL0(2),
1747 	PMU_PMEVCNTR_EL0(3),
1748 	PMU_PMEVCNTR_EL0(4),
1749 	PMU_PMEVCNTR_EL0(5),
1750 	PMU_PMEVCNTR_EL0(6),
1751 	PMU_PMEVCNTR_EL0(7),
1752 	PMU_PMEVCNTR_EL0(8),
1753 	PMU_PMEVCNTR_EL0(9),
1754 	PMU_PMEVCNTR_EL0(10),
1755 	PMU_PMEVCNTR_EL0(11),
1756 	PMU_PMEVCNTR_EL0(12),
1757 	PMU_PMEVCNTR_EL0(13),
1758 	PMU_PMEVCNTR_EL0(14),
1759 	PMU_PMEVCNTR_EL0(15),
1760 	PMU_PMEVCNTR_EL0(16),
1761 	PMU_PMEVCNTR_EL0(17),
1762 	PMU_PMEVCNTR_EL0(18),
1763 	PMU_PMEVCNTR_EL0(19),
1764 	PMU_PMEVCNTR_EL0(20),
1765 	PMU_PMEVCNTR_EL0(21),
1766 	PMU_PMEVCNTR_EL0(22),
1767 	PMU_PMEVCNTR_EL0(23),
1768 	PMU_PMEVCNTR_EL0(24),
1769 	PMU_PMEVCNTR_EL0(25),
1770 	PMU_PMEVCNTR_EL0(26),
1771 	PMU_PMEVCNTR_EL0(27),
1772 	PMU_PMEVCNTR_EL0(28),
1773 	PMU_PMEVCNTR_EL0(29),
1774 	PMU_PMEVCNTR_EL0(30),
1775 	/* PMEVTYPERn_EL0 */
1776 	PMU_PMEVTYPER_EL0(0),
1777 	PMU_PMEVTYPER_EL0(1),
1778 	PMU_PMEVTYPER_EL0(2),
1779 	PMU_PMEVTYPER_EL0(3),
1780 	PMU_PMEVTYPER_EL0(4),
1781 	PMU_PMEVTYPER_EL0(5),
1782 	PMU_PMEVTYPER_EL0(6),
1783 	PMU_PMEVTYPER_EL0(7),
1784 	PMU_PMEVTYPER_EL0(8),
1785 	PMU_PMEVTYPER_EL0(9),
1786 	PMU_PMEVTYPER_EL0(10),
1787 	PMU_PMEVTYPER_EL0(11),
1788 	PMU_PMEVTYPER_EL0(12),
1789 	PMU_PMEVTYPER_EL0(13),
1790 	PMU_PMEVTYPER_EL0(14),
1791 	PMU_PMEVTYPER_EL0(15),
1792 	PMU_PMEVTYPER_EL0(16),
1793 	PMU_PMEVTYPER_EL0(17),
1794 	PMU_PMEVTYPER_EL0(18),
1795 	PMU_PMEVTYPER_EL0(19),
1796 	PMU_PMEVTYPER_EL0(20),
1797 	PMU_PMEVTYPER_EL0(21),
1798 	PMU_PMEVTYPER_EL0(22),
1799 	PMU_PMEVTYPER_EL0(23),
1800 	PMU_PMEVTYPER_EL0(24),
1801 	PMU_PMEVTYPER_EL0(25),
1802 	PMU_PMEVTYPER_EL0(26),
1803 	PMU_PMEVTYPER_EL0(27),
1804 	PMU_PMEVTYPER_EL0(28),
1805 	PMU_PMEVTYPER_EL0(29),
1806 	PMU_PMEVTYPER_EL0(30),
1807 	/*
1808 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1809 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1810 	 */
1811 	{ PMU_SYS_REG(SYS_PMCCFILTR_EL0), .access = access_pmu_evtyper,
1812 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
1813 
1814 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1815 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1816 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1817 };
1818 
1819 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
1820 			struct sys_reg_params *p,
1821 			const struct sys_reg_desc *r)
1822 {
1823 	if (p->is_write) {
1824 		return ignore_write(vcpu, p);
1825 	} else {
1826 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1827 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1828 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1829 
1830 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1831 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1832 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1833 			     | (6 << 16) | (1 << 15) | (el3 << 14) | (el3 << 12));
1834 		return true;
1835 	}
1836 }
1837 
1838 /*
1839  * AArch32 debug register mappings
1840  *
1841  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1842  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1843  *
1844  * None of the other registers share their location, so treat them as
1845  * if they were 64bit.
1846  */
1847 #define DBG_BCR_BVR_WCR_WVR(n)						      \
1848 	/* DBGBVRn */							      \
1849 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
1850 	/* DBGBCRn */							      \
1851 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
1852 	/* DBGWVRn */							      \
1853 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
1854 	/* DBGWCRn */							      \
1855 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1856 
1857 #define DBGBXVR(n)							      \
1858 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
1859 
1860 /*
1861  * Trapped cp14 registers. We generally ignore most of the external
1862  * debug, on the principle that they don't really make sense to a
1863  * guest. Revisit this one day, would this principle change.
1864  */
1865 static const struct sys_reg_desc cp14_regs[] = {
1866 	/* DBGDIDR */
1867 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
1868 	/* DBGDTRRXext */
1869 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1870 
1871 	DBG_BCR_BVR_WCR_WVR(0),
1872 	/* DBGDSCRint */
1873 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1874 	DBG_BCR_BVR_WCR_WVR(1),
1875 	/* DBGDCCINT */
1876 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
1877 	/* DBGDSCRext */
1878 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
1879 	DBG_BCR_BVR_WCR_WVR(2),
1880 	/* DBGDTR[RT]Xint */
1881 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1882 	/* DBGDTR[RT]Xext */
1883 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1884 	DBG_BCR_BVR_WCR_WVR(3),
1885 	DBG_BCR_BVR_WCR_WVR(4),
1886 	DBG_BCR_BVR_WCR_WVR(5),
1887 	/* DBGWFAR */
1888 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1889 	/* DBGOSECCR */
1890 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1891 	DBG_BCR_BVR_WCR_WVR(6),
1892 	/* DBGVCR */
1893 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
1894 	DBG_BCR_BVR_WCR_WVR(7),
1895 	DBG_BCR_BVR_WCR_WVR(8),
1896 	DBG_BCR_BVR_WCR_WVR(9),
1897 	DBG_BCR_BVR_WCR_WVR(10),
1898 	DBG_BCR_BVR_WCR_WVR(11),
1899 	DBG_BCR_BVR_WCR_WVR(12),
1900 	DBG_BCR_BVR_WCR_WVR(13),
1901 	DBG_BCR_BVR_WCR_WVR(14),
1902 	DBG_BCR_BVR_WCR_WVR(15),
1903 
1904 	/* DBGDRAR (32bit) */
1905 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1906 
1907 	DBGBXVR(0),
1908 	/* DBGOSLAR */
1909 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
1910 	DBGBXVR(1),
1911 	/* DBGOSLSR */
1912 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
1913 	DBGBXVR(2),
1914 	DBGBXVR(3),
1915 	/* DBGOSDLR */
1916 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1917 	DBGBXVR(4),
1918 	/* DBGPRCR */
1919 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1920 	DBGBXVR(5),
1921 	DBGBXVR(6),
1922 	DBGBXVR(7),
1923 	DBGBXVR(8),
1924 	DBGBXVR(9),
1925 	DBGBXVR(10),
1926 	DBGBXVR(11),
1927 	DBGBXVR(12),
1928 	DBGBXVR(13),
1929 	DBGBXVR(14),
1930 	DBGBXVR(15),
1931 
1932 	/* DBGDSAR (32bit) */
1933 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1934 
1935 	/* DBGDEVID2 */
1936 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1937 	/* DBGDEVID1 */
1938 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1939 	/* DBGDEVID */
1940 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1941 	/* DBGCLAIMSET */
1942 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1943 	/* DBGCLAIMCLR */
1944 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1945 	/* DBGAUTHSTATUS */
1946 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1947 };
1948 
1949 /* Trapped cp14 64bit registers */
1950 static const struct sys_reg_desc cp14_64_regs[] = {
1951 	/* DBGDRAR (64bit) */
1952 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1953 
1954 	/* DBGDSAR (64bit) */
1955 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1956 };
1957 
1958 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)			\
1959 	AA32(_map),							\
1960 	Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),			\
1961 	.visibility = pmu_visibility
1962 
1963 /* Macro to expand the PMEVCNTRn register */
1964 #define PMU_PMEVCNTR(n)							\
1965 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
1966 	  (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
1967 	  .access = access_pmu_evcntr }
1968 
1969 /* Macro to expand the PMEVTYPERn register */
1970 #define PMU_PMEVTYPER(n)						\
1971 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
1972 	  (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
1973 	  .access = access_pmu_evtyper }
1974 /*
1975  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1976  * depending on the way they are accessed (as a 32bit or a 64bit
1977  * register).
1978  */
1979 static const struct sys_reg_desc cp15_regs[] = {
1980 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1981 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
1982 	/* ACTLR */
1983 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
1984 	/* ACTLR2 */
1985 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
1986 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
1987 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
1988 	/* TTBCR */
1989 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
1990 	/* TTBCR2 */
1991 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
1992 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
1993 	/* DFSR */
1994 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
1995 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
1996 	/* ADFSR */
1997 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
1998 	/* AIFSR */
1999 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
2000 	/* DFAR */
2001 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
2002 	/* IFAR */
2003 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
2004 
2005 	/*
2006 	 * DC{C,I,CI}SW operations:
2007 	 */
2008 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
2009 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
2010 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
2011 
2012 	/* PMU */
2013 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
2014 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
2015 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
2016 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
2017 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
2018 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
2019 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
2020 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
2021 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
2022 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
2023 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
2024 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
2025 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
2026 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
2027 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
2028 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
2029 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
2030 	/* PMMIR */
2031 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
2032 
2033 	/* PRRR/MAIR0 */
2034 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
2035 	/* NMRR/MAIR1 */
2036 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
2037 	/* AMAIR0 */
2038 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
2039 	/* AMAIR1 */
2040 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
2041 
2042 	/* ICC_SRE */
2043 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2044 
2045 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
2046 
2047 	/* Arch Tmers */
2048 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2049 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2050 
2051 	/* PMEVCNTRn */
2052 	PMU_PMEVCNTR(0),
2053 	PMU_PMEVCNTR(1),
2054 	PMU_PMEVCNTR(2),
2055 	PMU_PMEVCNTR(3),
2056 	PMU_PMEVCNTR(4),
2057 	PMU_PMEVCNTR(5),
2058 	PMU_PMEVCNTR(6),
2059 	PMU_PMEVCNTR(7),
2060 	PMU_PMEVCNTR(8),
2061 	PMU_PMEVCNTR(9),
2062 	PMU_PMEVCNTR(10),
2063 	PMU_PMEVCNTR(11),
2064 	PMU_PMEVCNTR(12),
2065 	PMU_PMEVCNTR(13),
2066 	PMU_PMEVCNTR(14),
2067 	PMU_PMEVCNTR(15),
2068 	PMU_PMEVCNTR(16),
2069 	PMU_PMEVCNTR(17),
2070 	PMU_PMEVCNTR(18),
2071 	PMU_PMEVCNTR(19),
2072 	PMU_PMEVCNTR(20),
2073 	PMU_PMEVCNTR(21),
2074 	PMU_PMEVCNTR(22),
2075 	PMU_PMEVCNTR(23),
2076 	PMU_PMEVCNTR(24),
2077 	PMU_PMEVCNTR(25),
2078 	PMU_PMEVCNTR(26),
2079 	PMU_PMEVCNTR(27),
2080 	PMU_PMEVCNTR(28),
2081 	PMU_PMEVCNTR(29),
2082 	PMU_PMEVCNTR(30),
2083 	/* PMEVTYPERn */
2084 	PMU_PMEVTYPER(0),
2085 	PMU_PMEVTYPER(1),
2086 	PMU_PMEVTYPER(2),
2087 	PMU_PMEVTYPER(3),
2088 	PMU_PMEVTYPER(4),
2089 	PMU_PMEVTYPER(5),
2090 	PMU_PMEVTYPER(6),
2091 	PMU_PMEVTYPER(7),
2092 	PMU_PMEVTYPER(8),
2093 	PMU_PMEVTYPER(9),
2094 	PMU_PMEVTYPER(10),
2095 	PMU_PMEVTYPER(11),
2096 	PMU_PMEVTYPER(12),
2097 	PMU_PMEVTYPER(13),
2098 	PMU_PMEVTYPER(14),
2099 	PMU_PMEVTYPER(15),
2100 	PMU_PMEVTYPER(16),
2101 	PMU_PMEVTYPER(17),
2102 	PMU_PMEVTYPER(18),
2103 	PMU_PMEVTYPER(19),
2104 	PMU_PMEVTYPER(20),
2105 	PMU_PMEVTYPER(21),
2106 	PMU_PMEVTYPER(22),
2107 	PMU_PMEVTYPER(23),
2108 	PMU_PMEVTYPER(24),
2109 	PMU_PMEVTYPER(25),
2110 	PMU_PMEVTYPER(26),
2111 	PMU_PMEVTYPER(27),
2112 	PMU_PMEVTYPER(28),
2113 	PMU_PMEVTYPER(29),
2114 	PMU_PMEVTYPER(30),
2115 	/* PMCCFILTR */
2116 	{ CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
2117 
2118 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2119 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2120 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
2121 };
2122 
2123 static const struct sys_reg_desc cp15_64_regs[] = {
2124 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
2125 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
2126 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2127 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
2128 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2129 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2130 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2131 };
2132 
2133 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2134 			       bool is_32)
2135 {
2136 	unsigned int i;
2137 
2138 	for (i = 0; i < n; i++) {
2139 		if (!is_32 && table[i].reg && !table[i].reset) {
2140 			kvm_err("sys_reg table %pS entry %d lacks reset\n", &table[i], i);
2141 			return false;
2142 		}
2143 
2144 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2145 			kvm_err("sys_reg table %pS entry %d out of order\n", &table[i - 1], i - 1);
2146 			return false;
2147 		}
2148 	}
2149 
2150 	return true;
2151 }
2152 
2153 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2154 {
2155 	kvm_inject_undefined(vcpu);
2156 	return 1;
2157 }
2158 
2159 static void perform_access(struct kvm_vcpu *vcpu,
2160 			   struct sys_reg_params *params,
2161 			   const struct sys_reg_desc *r)
2162 {
2163 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2164 
2165 	/* Check for regs disabled by runtime config */
2166 	if (sysreg_hidden(vcpu, r)) {
2167 		kvm_inject_undefined(vcpu);
2168 		return;
2169 	}
2170 
2171 	/*
2172 	 * Not having an accessor means that we have configured a trap
2173 	 * that we don't know how to handle. This certainly qualifies
2174 	 * as a gross bug that should be fixed right away.
2175 	 */
2176 	BUG_ON(!r->access);
2177 
2178 	/* Skip instruction if instructed so */
2179 	if (likely(r->access(vcpu, params, r)))
2180 		kvm_incr_pc(vcpu);
2181 }
2182 
2183 /*
2184  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2185  *                call the corresponding trap handler.
2186  *
2187  * @params: pointer to the descriptor of the access
2188  * @table: array of trap descriptors
2189  * @num: size of the trap descriptor array
2190  *
2191  * Return true if the access has been handled, false if not.
2192  */
2193 static bool emulate_cp(struct kvm_vcpu *vcpu,
2194 		       struct sys_reg_params *params,
2195 		       const struct sys_reg_desc *table,
2196 		       size_t num)
2197 {
2198 	const struct sys_reg_desc *r;
2199 
2200 	if (!table)
2201 		return false;	/* Not handled */
2202 
2203 	r = find_reg(params, table, num);
2204 
2205 	if (r) {
2206 		perform_access(vcpu, params, r);
2207 		return true;
2208 	}
2209 
2210 	/* Not handled */
2211 	return false;
2212 }
2213 
2214 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2215 				struct sys_reg_params *params)
2216 {
2217 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2218 	int cp = -1;
2219 
2220 	switch (esr_ec) {
2221 	case ESR_ELx_EC_CP15_32:
2222 	case ESR_ELx_EC_CP15_64:
2223 		cp = 15;
2224 		break;
2225 	case ESR_ELx_EC_CP14_MR:
2226 	case ESR_ELx_EC_CP14_64:
2227 		cp = 14;
2228 		break;
2229 	default:
2230 		WARN_ON(1);
2231 	}
2232 
2233 	print_sys_reg_msg(params,
2234 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2235 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2236 	kvm_inject_undefined(vcpu);
2237 }
2238 
2239 /**
2240  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2241  * @vcpu: The VCPU pointer
2242  * @run:  The kvm_run struct
2243  */
2244 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2245 			    const struct sys_reg_desc *global,
2246 			    size_t nr_global)
2247 {
2248 	struct sys_reg_params params;
2249 	u64 esr = kvm_vcpu_get_esr(vcpu);
2250 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2251 	int Rt2 = (esr >> 10) & 0x1f;
2252 
2253 	params.CRm = (esr >> 1) & 0xf;
2254 	params.is_write = ((esr & 1) == 0);
2255 
2256 	params.Op0 = 0;
2257 	params.Op1 = (esr >> 16) & 0xf;
2258 	params.Op2 = 0;
2259 	params.CRn = 0;
2260 
2261 	/*
2262 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2263 	 * backends between AArch32 and AArch64, we get away with it.
2264 	 */
2265 	if (params.is_write) {
2266 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2267 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2268 	}
2269 
2270 	/*
2271 	 * If the table contains a handler, handle the
2272 	 * potential register operation in the case of a read and return
2273 	 * with success.
2274 	 */
2275 	if (emulate_cp(vcpu, &params, global, nr_global)) {
2276 		/* Split up the value between registers for the read side */
2277 		if (!params.is_write) {
2278 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2279 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2280 		}
2281 
2282 		return 1;
2283 	}
2284 
2285 	unhandled_cp_access(vcpu, &params);
2286 	return 1;
2287 }
2288 
2289 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
2290 
2291 /*
2292  * The CP10 ID registers are architecturally mapped to AArch64 feature
2293  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
2294  * from AArch32.
2295  */
2296 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
2297 {
2298 	u8 reg_id = (esr >> 10) & 0xf;
2299 	bool valid;
2300 
2301 	params->is_write = ((esr & 1) == 0);
2302 	params->Op0 = 3;
2303 	params->Op1 = 0;
2304 	params->CRn = 0;
2305 	params->CRm = 3;
2306 
2307 	/* CP10 ID registers are read-only */
2308 	valid = !params->is_write;
2309 
2310 	switch (reg_id) {
2311 	/* MVFR0 */
2312 	case 0b0111:
2313 		params->Op2 = 0;
2314 		break;
2315 	/* MVFR1 */
2316 	case 0b0110:
2317 		params->Op2 = 1;
2318 		break;
2319 	/* MVFR2 */
2320 	case 0b0101:
2321 		params->Op2 = 2;
2322 		break;
2323 	default:
2324 		valid = false;
2325 	}
2326 
2327 	if (valid)
2328 		return true;
2329 
2330 	kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
2331 		      params->is_write ? "write" : "read", reg_id);
2332 	return false;
2333 }
2334 
2335 /**
2336  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
2337  *			  VFP Register' from AArch32.
2338  * @vcpu: The vCPU pointer
2339  *
2340  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
2341  * Work out the correct AArch64 system register encoding and reroute to the
2342  * AArch64 system register emulation.
2343  */
2344 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
2345 {
2346 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2347 	u64 esr = kvm_vcpu_get_esr(vcpu);
2348 	struct sys_reg_params params;
2349 
2350 	/* UNDEF on any unhandled register access */
2351 	if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
2352 		kvm_inject_undefined(vcpu);
2353 		return 1;
2354 	}
2355 
2356 	if (emulate_sys_reg(vcpu, &params))
2357 		vcpu_set_reg(vcpu, Rt, params.regval);
2358 
2359 	return 1;
2360 }
2361 
2362 /**
2363  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
2364  *			       CRn=0, which corresponds to the AArch32 feature
2365  *			       registers.
2366  * @vcpu: the vCPU pointer
2367  * @params: the system register access parameters.
2368  *
2369  * Our cp15 system register tables do not enumerate the AArch32 feature
2370  * registers. Conveniently, our AArch64 table does, and the AArch32 system
2371  * register encoding can be trivially remapped into the AArch64 for the feature
2372  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
2373  *
2374  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
2375  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
2376  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
2377  * treat undefined registers in this range as RAZ.
2378  */
2379 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
2380 				   struct sys_reg_params *params)
2381 {
2382 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2383 
2384 	/* Treat impossible writes to RO registers as UNDEFINED */
2385 	if (params->is_write) {
2386 		unhandled_cp_access(vcpu, params);
2387 		return 1;
2388 	}
2389 
2390 	params->Op0 = 3;
2391 
2392 	/*
2393 	 * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
2394 	 * Avoid conflicting with future expansion of AArch64 feature registers
2395 	 * and simply treat them as RAZ here.
2396 	 */
2397 	if (params->CRm > 3)
2398 		params->regval = 0;
2399 	else if (!emulate_sys_reg(vcpu, params))
2400 		return 1;
2401 
2402 	vcpu_set_reg(vcpu, Rt, params->regval);
2403 	return 1;
2404 }
2405 
2406 /**
2407  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2408  * @vcpu: The VCPU pointer
2409  * @run:  The kvm_run struct
2410  */
2411 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2412 			    struct sys_reg_params *params,
2413 			    const struct sys_reg_desc *global,
2414 			    size_t nr_global)
2415 {
2416 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2417 
2418 	params->regval = vcpu_get_reg(vcpu, Rt);
2419 
2420 	if (emulate_cp(vcpu, params, global, nr_global)) {
2421 		if (!params->is_write)
2422 			vcpu_set_reg(vcpu, Rt, params->regval);
2423 		return 1;
2424 	}
2425 
2426 	unhandled_cp_access(vcpu, params);
2427 	return 1;
2428 }
2429 
2430 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2431 {
2432 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2433 }
2434 
2435 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2436 {
2437 	struct sys_reg_params params;
2438 
2439 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
2440 
2441 	/*
2442 	 * Certain AArch32 ID registers are handled by rerouting to the AArch64
2443 	 * system register table. Registers in the ID range where CRm=0 are
2444 	 * excluded from this scheme as they do not trivially map into AArch64
2445 	 * system register encodings.
2446 	 */
2447 	if (params.Op1 == 0 && params.CRn == 0 && params.CRm)
2448 		return kvm_emulate_cp15_id_reg(vcpu, &params);
2449 
2450 	return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
2451 }
2452 
2453 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2454 {
2455 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2456 }
2457 
2458 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2459 {
2460 	struct sys_reg_params params;
2461 
2462 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
2463 
2464 	return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
2465 }
2466 
2467 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2468 {
2469 	// See ARM DDI 0487E.a, section D12.3.2
2470 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2471 }
2472 
2473 /**
2474  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
2475  * @vcpu: The VCPU pointer
2476  * @params: Decoded system register parameters
2477  *
2478  * Return: true if the system register access was successful, false otherwise.
2479  */
2480 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
2481 			   struct sys_reg_params *params)
2482 {
2483 	const struct sys_reg_desc *r;
2484 
2485 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2486 
2487 	if (likely(r)) {
2488 		perform_access(vcpu, params, r);
2489 		return true;
2490 	}
2491 
2492 	if (is_imp_def_sys_reg(params)) {
2493 		kvm_inject_undefined(vcpu);
2494 	} else {
2495 		print_sys_reg_msg(params,
2496 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2497 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2498 		kvm_inject_undefined(vcpu);
2499 	}
2500 	return false;
2501 }
2502 
2503 /**
2504  * kvm_reset_sys_regs - sets system registers to reset value
2505  * @vcpu: The VCPU pointer
2506  *
2507  * This function finds the right table above and sets the registers on the
2508  * virtual CPU struct to their architecturally defined reset values.
2509  */
2510 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2511 {
2512 	unsigned long i;
2513 
2514 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2515 		if (sys_reg_descs[i].reset)
2516 			sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2517 }
2518 
2519 /**
2520  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2521  * @vcpu: The VCPU pointer
2522  */
2523 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2524 {
2525 	struct sys_reg_params params;
2526 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
2527 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2528 
2529 	trace_kvm_handle_sys_reg(esr);
2530 
2531 	params = esr_sys64_to_params(esr);
2532 	params.regval = vcpu_get_reg(vcpu, Rt);
2533 
2534 	if (!emulate_sys_reg(vcpu, &params))
2535 		return 1;
2536 
2537 	if (!params.is_write)
2538 		vcpu_set_reg(vcpu, Rt, params.regval);
2539 	return 1;
2540 }
2541 
2542 /******************************************************************************
2543  * Userspace API
2544  *****************************************************************************/
2545 
2546 static bool index_to_params(u64 id, struct sys_reg_params *params)
2547 {
2548 	switch (id & KVM_REG_SIZE_MASK) {
2549 	case KVM_REG_SIZE_U64:
2550 		/* Any unused index bits means it's not valid. */
2551 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2552 			      | KVM_REG_ARM_COPROC_MASK
2553 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2554 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2555 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2556 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2557 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2558 			return false;
2559 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2560 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2561 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2562 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2563 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2564 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2565 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2566 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2567 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2568 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2569 		return true;
2570 	default:
2571 		return false;
2572 	}
2573 }
2574 
2575 const struct sys_reg_desc *get_reg_by_id(u64 id,
2576 					 const struct sys_reg_desc table[],
2577 					 unsigned int num)
2578 {
2579 	struct sys_reg_params params;
2580 
2581 	if (!index_to_params(id, &params))
2582 		return NULL;
2583 
2584 	return find_reg(&params, table, num);
2585 }
2586 
2587 /* Decode an index value, and find the sys_reg_desc entry. */
2588 static const struct sys_reg_desc *
2589 id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
2590 		   const struct sys_reg_desc table[], unsigned int num)
2591 
2592 {
2593 	const struct sys_reg_desc *r;
2594 
2595 	/* We only do sys_reg for now. */
2596 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2597 		return NULL;
2598 
2599 	r = get_reg_by_id(id, table, num);
2600 
2601 	/* Not saved in the sys_reg array and not otherwise accessible? */
2602 	if (r && (!(r->reg || r->get_user) || sysreg_hidden(vcpu, r)))
2603 		r = NULL;
2604 
2605 	return r;
2606 }
2607 
2608 /*
2609  * These are the invariant sys_reg registers: we let the guest see the
2610  * host versions of these, so they're part of the guest state.
2611  *
2612  * A future CPU may provide a mechanism to present different values to
2613  * the guest, or a future kvm may trap them.
2614  */
2615 
2616 #define FUNCTION_INVARIANT(reg)						\
2617 	static void get_##reg(struct kvm_vcpu *v,			\
2618 			      const struct sys_reg_desc *r)		\
2619 	{								\
2620 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2621 	}
2622 
2623 FUNCTION_INVARIANT(midr_el1)
2624 FUNCTION_INVARIANT(revidr_el1)
2625 FUNCTION_INVARIANT(clidr_el1)
2626 FUNCTION_INVARIANT(aidr_el1)
2627 
2628 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2629 {
2630 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2631 }
2632 
2633 /* ->val is filled in by kvm_sys_reg_table_init() */
2634 static struct sys_reg_desc invariant_sys_regs[] = {
2635 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2636 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2637 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2638 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2639 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2640 };
2641 
2642 static int get_invariant_sys_reg(u64 id, u64 __user *uaddr)
2643 {
2644 	const struct sys_reg_desc *r;
2645 
2646 	r = get_reg_by_id(id, invariant_sys_regs,
2647 			  ARRAY_SIZE(invariant_sys_regs));
2648 	if (!r)
2649 		return -ENOENT;
2650 
2651 	return put_user(r->val, uaddr);
2652 }
2653 
2654 static int set_invariant_sys_reg(u64 id, u64 __user *uaddr)
2655 {
2656 	const struct sys_reg_desc *r;
2657 	u64 val;
2658 
2659 	r = get_reg_by_id(id, invariant_sys_regs,
2660 			  ARRAY_SIZE(invariant_sys_regs));
2661 	if (!r)
2662 		return -ENOENT;
2663 
2664 	if (get_user(val, uaddr))
2665 		return -EFAULT;
2666 
2667 	/* This is what we mean by invariant: you can't change it. */
2668 	if (r->val != val)
2669 		return -EINVAL;
2670 
2671 	return 0;
2672 }
2673 
2674 static bool is_valid_cache(u32 val)
2675 {
2676 	u32 level, ctype;
2677 
2678 	if (val >= CSSELR_MAX)
2679 		return false;
2680 
2681 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2682 	level = (val >> 1);
2683 	ctype = (cache_levels >> (level * 3)) & 7;
2684 
2685 	switch (ctype) {
2686 	case 0: /* No cache */
2687 		return false;
2688 	case 1: /* Instruction cache only */
2689 		return (val & 1);
2690 	case 2: /* Data cache only */
2691 	case 4: /* Unified cache */
2692 		return !(val & 1);
2693 	case 3: /* Separate instruction and data caches */
2694 		return true;
2695 	default: /* Reserved: we can't know instruction or data. */
2696 		return false;
2697 	}
2698 }
2699 
2700 static int demux_c15_get(u64 id, void __user *uaddr)
2701 {
2702 	u32 val;
2703 	u32 __user *uval = uaddr;
2704 
2705 	/* Fail if we have unknown bits set. */
2706 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2707 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2708 		return -ENOENT;
2709 
2710 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2711 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2712 		if (KVM_REG_SIZE(id) != 4)
2713 			return -ENOENT;
2714 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2715 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2716 		if (!is_valid_cache(val))
2717 			return -ENOENT;
2718 
2719 		return put_user(get_ccsidr(val), uval);
2720 	default:
2721 		return -ENOENT;
2722 	}
2723 }
2724 
2725 static int demux_c15_set(u64 id, void __user *uaddr)
2726 {
2727 	u32 val, newval;
2728 	u32 __user *uval = uaddr;
2729 
2730 	/* Fail if we have unknown bits set. */
2731 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2732 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2733 		return -ENOENT;
2734 
2735 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2736 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2737 		if (KVM_REG_SIZE(id) != 4)
2738 			return -ENOENT;
2739 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2740 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2741 		if (!is_valid_cache(val))
2742 			return -ENOENT;
2743 
2744 		if (get_user(newval, uval))
2745 			return -EFAULT;
2746 
2747 		/* This is also invariant: you can't change it. */
2748 		if (newval != get_ccsidr(val))
2749 			return -EINVAL;
2750 		return 0;
2751 	default:
2752 		return -ENOENT;
2753 	}
2754 }
2755 
2756 int kvm_sys_reg_get_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
2757 			 const struct sys_reg_desc table[], unsigned int num)
2758 {
2759 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
2760 	const struct sys_reg_desc *r;
2761 	u64 val;
2762 	int ret;
2763 
2764 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
2765 	if (!r)
2766 		return -ENOENT;
2767 
2768 	if (r->get_user) {
2769 		ret = (r->get_user)(vcpu, r, &val);
2770 	} else {
2771 		val = __vcpu_sys_reg(vcpu, r->reg);
2772 		ret = 0;
2773 	}
2774 
2775 	if (!ret)
2776 		ret = put_user(val, uaddr);
2777 
2778 	return ret;
2779 }
2780 
2781 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2782 {
2783 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2784 	int err;
2785 
2786 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2787 		return demux_c15_get(reg->id, uaddr);
2788 
2789 	err = get_invariant_sys_reg(reg->id, uaddr);
2790 	if (err != -ENOENT)
2791 		return err;
2792 
2793 	return kvm_sys_reg_get_user(vcpu, reg,
2794 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2795 }
2796 
2797 int kvm_sys_reg_set_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
2798 			 const struct sys_reg_desc table[], unsigned int num)
2799 {
2800 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
2801 	const struct sys_reg_desc *r;
2802 	u64 val;
2803 	int ret;
2804 
2805 	if (get_user(val, uaddr))
2806 		return -EFAULT;
2807 
2808 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
2809 	if (!r)
2810 		return -ENOENT;
2811 
2812 	if (r->set_user) {
2813 		ret = (r->set_user)(vcpu, r, val);
2814 	} else {
2815 		__vcpu_sys_reg(vcpu, r->reg) = val;
2816 		ret = 0;
2817 	}
2818 
2819 	return ret;
2820 }
2821 
2822 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2823 {
2824 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2825 	int err;
2826 
2827 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2828 		return demux_c15_set(reg->id, uaddr);
2829 
2830 	err = set_invariant_sys_reg(reg->id, uaddr);
2831 	if (err != -ENOENT)
2832 		return err;
2833 
2834 	return kvm_sys_reg_set_user(vcpu, reg,
2835 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2836 }
2837 
2838 static unsigned int num_demux_regs(void)
2839 {
2840 	unsigned int i, count = 0;
2841 
2842 	for (i = 0; i < CSSELR_MAX; i++)
2843 		if (is_valid_cache(i))
2844 			count++;
2845 
2846 	return count;
2847 }
2848 
2849 static int write_demux_regids(u64 __user *uindices)
2850 {
2851 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2852 	unsigned int i;
2853 
2854 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2855 	for (i = 0; i < CSSELR_MAX; i++) {
2856 		if (!is_valid_cache(i))
2857 			continue;
2858 		if (put_user(val | i, uindices))
2859 			return -EFAULT;
2860 		uindices++;
2861 	}
2862 	return 0;
2863 }
2864 
2865 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2866 {
2867 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2868 		KVM_REG_ARM64_SYSREG |
2869 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2870 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2871 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2872 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2873 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2874 }
2875 
2876 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2877 {
2878 	if (!*uind)
2879 		return true;
2880 
2881 	if (put_user(sys_reg_to_index(reg), *uind))
2882 		return false;
2883 
2884 	(*uind)++;
2885 	return true;
2886 }
2887 
2888 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2889 			    const struct sys_reg_desc *rd,
2890 			    u64 __user **uind,
2891 			    unsigned int *total)
2892 {
2893 	/*
2894 	 * Ignore registers we trap but don't save,
2895 	 * and for which no custom user accessor is provided.
2896 	 */
2897 	if (!(rd->reg || rd->get_user))
2898 		return 0;
2899 
2900 	if (sysreg_hidden(vcpu, rd))
2901 		return 0;
2902 
2903 	if (!copy_reg_to_user(rd, uind))
2904 		return -EFAULT;
2905 
2906 	(*total)++;
2907 	return 0;
2908 }
2909 
2910 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2911 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2912 {
2913 	const struct sys_reg_desc *i2, *end2;
2914 	unsigned int total = 0;
2915 	int err;
2916 
2917 	i2 = sys_reg_descs;
2918 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2919 
2920 	while (i2 != end2) {
2921 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2922 		if (err)
2923 			return err;
2924 	}
2925 	return total;
2926 }
2927 
2928 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2929 {
2930 	return ARRAY_SIZE(invariant_sys_regs)
2931 		+ num_demux_regs()
2932 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2933 }
2934 
2935 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2936 {
2937 	unsigned int i;
2938 	int err;
2939 
2940 	/* Then give them all the invariant registers' indices. */
2941 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2942 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2943 			return -EFAULT;
2944 		uindices++;
2945 	}
2946 
2947 	err = walk_sys_regs(vcpu, uindices);
2948 	if (err < 0)
2949 		return err;
2950 	uindices += err;
2951 
2952 	return write_demux_regids(uindices);
2953 }
2954 
2955 int kvm_sys_reg_table_init(void)
2956 {
2957 	bool valid = true;
2958 	unsigned int i;
2959 	struct sys_reg_desc clidr;
2960 
2961 	/* Make sure tables are unique and in order. */
2962 	valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false);
2963 	valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true);
2964 	valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true);
2965 	valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true);
2966 	valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true);
2967 	valid &= check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false);
2968 
2969 	if (!valid)
2970 		return -EINVAL;
2971 
2972 	/* We abuse the reset function to overwrite the table itself. */
2973 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2974 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2975 
2976 	/*
2977 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2978 	 *
2979 	 *   If software reads the Cache Type fields from Ctype1
2980 	 *   upwards, once it has seen a value of 0b000, no caches
2981 	 *   exist at further-out levels of the hierarchy. So, for
2982 	 *   example, if Ctype3 is the first Cache Type field with a
2983 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2984 	 *   ignored.
2985 	 */
2986 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2987 	cache_levels = clidr.val;
2988 	for (i = 0; i < 7; i++)
2989 		if (((cache_levels >> (i*3)) & 7) == 0)
2990 			break;
2991 	/* Clear all higher bits. */
2992 	cache_levels &= (1 << (i*3))-1;
2993 
2994 	return 0;
2995 }
2996