xref: /linux/arch/arm64/kvm/sys_regs.c (revision a05e4c0af490ca7c22fc77120aafebebdeaaf537)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/coproc.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
9  *          Christoffer Dall <c.dall@virtualopensystems.com>
10  */
11 
12 #include <linux/bitfield.h>
13 #include <linux/bsearch.h>
14 #include <linux/kvm_host.h>
15 #include <linux/mm.h>
16 #include <linux/printk.h>
17 #include <linux/uaccess.h>
18 
19 #include <asm/cacheflush.h>
20 #include <asm/cputype.h>
21 #include <asm/debug-monitors.h>
22 #include <asm/esr.h>
23 #include <asm/kvm_arm.h>
24 #include <asm/kvm_emulate.h>
25 #include <asm/kvm_hyp.h>
26 #include <asm/kvm_mmu.h>
27 #include <asm/perf_event.h>
28 #include <asm/sysreg.h>
29 
30 #include <trace/events/kvm.h>
31 
32 #include "sys_regs.h"
33 
34 #include "trace.h"
35 
36 /*
37  * All of this file is extremely similar to the ARM coproc.c, but the
38  * types are different. My gut feeling is that it should be pretty
39  * easy to merge, but that would be an ABI breakage -- again. VFP
40  * would also need to be abstracted.
41  *
42  * For AArch32, we only take care of what is being trapped. Anything
43  * that has to do with init and userspace access has to go via the
44  * 64bit interface.
45  */
46 
47 static bool read_from_write_only(struct kvm_vcpu *vcpu,
48 				 struct sys_reg_params *params,
49 				 const struct sys_reg_desc *r)
50 {
51 	WARN_ONCE(1, "Unexpected sys_reg read to write-only register\n");
52 	print_sys_reg_instr(params);
53 	kvm_inject_undefined(vcpu);
54 	return false;
55 }
56 
57 static bool write_to_read_only(struct kvm_vcpu *vcpu,
58 			       struct sys_reg_params *params,
59 			       const struct sys_reg_desc *r)
60 {
61 	WARN_ONCE(1, "Unexpected sys_reg write to read-only register\n");
62 	print_sys_reg_instr(params);
63 	kvm_inject_undefined(vcpu);
64 	return false;
65 }
66 
67 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
68 {
69 	u64 val = 0x8badf00d8badf00d;
70 
71 	if (vcpu->arch.sysregs_loaded_on_cpu &&
72 	    __vcpu_read_sys_reg_from_cpu(reg, &val))
73 		return val;
74 
75 	return __vcpu_sys_reg(vcpu, reg);
76 }
77 
78 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
79 {
80 	if (vcpu->arch.sysregs_loaded_on_cpu &&
81 	    __vcpu_write_sys_reg_to_cpu(val, reg))
82 		return;
83 
84 	 __vcpu_sys_reg(vcpu, reg) = val;
85 }
86 
87 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
88 static u32 cache_levels;
89 
90 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
91 #define CSSELR_MAX 14
92 
93 /* Which cache CCSIDR represents depends on CSSELR value. */
94 static u32 get_ccsidr(u32 csselr)
95 {
96 	u32 ccsidr;
97 
98 	/* Make sure noone else changes CSSELR during this! */
99 	local_irq_disable();
100 	write_sysreg(csselr, csselr_el1);
101 	isb();
102 	ccsidr = read_sysreg(ccsidr_el1);
103 	local_irq_enable();
104 
105 	return ccsidr;
106 }
107 
108 /*
109  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
110  */
111 static bool access_dcsw(struct kvm_vcpu *vcpu,
112 			struct sys_reg_params *p,
113 			const struct sys_reg_desc *r)
114 {
115 	if (!p->is_write)
116 		return read_from_write_only(vcpu, p, r);
117 
118 	/*
119 	 * Only track S/W ops if we don't have FWB. It still indicates
120 	 * that the guest is a bit broken (S/W operations should only
121 	 * be done by firmware, knowing that there is only a single
122 	 * CPU left in the system, and certainly not from non-secure
123 	 * software).
124 	 */
125 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
126 		kvm_set_way_flush(vcpu);
127 
128 	return true;
129 }
130 
131 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
132 {
133 	switch (r->aarch32_map) {
134 	case AA32_LO:
135 		*mask = GENMASK_ULL(31, 0);
136 		*shift = 0;
137 		break;
138 	case AA32_HI:
139 		*mask = GENMASK_ULL(63, 32);
140 		*shift = 32;
141 		break;
142 	default:
143 		*mask = GENMASK_ULL(63, 0);
144 		*shift = 0;
145 		break;
146 	}
147 }
148 
149 /*
150  * Generic accessor for VM registers. Only called as long as HCR_TVM
151  * is set. If the guest enables the MMU, we stop trapping the VM
152  * sys_regs and leave it in complete control of the caches.
153  */
154 static bool access_vm_reg(struct kvm_vcpu *vcpu,
155 			  struct sys_reg_params *p,
156 			  const struct sys_reg_desc *r)
157 {
158 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
159 	u64 val, mask, shift;
160 
161 	BUG_ON(!p->is_write);
162 
163 	get_access_mask(r, &mask, &shift);
164 
165 	if (~mask) {
166 		val = vcpu_read_sys_reg(vcpu, r->reg);
167 		val &= ~mask;
168 	} else {
169 		val = 0;
170 	}
171 
172 	val |= (p->regval & (mask >> shift)) << shift;
173 	vcpu_write_sys_reg(vcpu, val, r->reg);
174 
175 	kvm_toggle_cache(vcpu, was_enabled);
176 	return true;
177 }
178 
179 static bool access_actlr(struct kvm_vcpu *vcpu,
180 			 struct sys_reg_params *p,
181 			 const struct sys_reg_desc *r)
182 {
183 	u64 mask, shift;
184 
185 	if (p->is_write)
186 		return ignore_write(vcpu, p);
187 
188 	get_access_mask(r, &mask, &shift);
189 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
190 
191 	return true;
192 }
193 
194 /*
195  * Trap handler for the GICv3 SGI generation system register.
196  * Forward the request to the VGIC emulation.
197  * The cp15_64 code makes sure this automatically works
198  * for both AArch64 and AArch32 accesses.
199  */
200 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
201 			   struct sys_reg_params *p,
202 			   const struct sys_reg_desc *r)
203 {
204 	bool g1;
205 
206 	if (!p->is_write)
207 		return read_from_write_only(vcpu, p, r);
208 
209 	/*
210 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
211 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
212 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
213 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
214 	 * group.
215 	 */
216 	if (p->Op0 == 0) {		/* AArch32 */
217 		switch (p->Op1) {
218 		default:		/* Keep GCC quiet */
219 		case 0:			/* ICC_SGI1R */
220 			g1 = true;
221 			break;
222 		case 1:			/* ICC_ASGI1R */
223 		case 2:			/* ICC_SGI0R */
224 			g1 = false;
225 			break;
226 		}
227 	} else {			/* AArch64 */
228 		switch (p->Op2) {
229 		default:		/* Keep GCC quiet */
230 		case 5:			/* ICC_SGI1R_EL1 */
231 			g1 = true;
232 			break;
233 		case 6:			/* ICC_ASGI1R_EL1 */
234 		case 7:			/* ICC_SGI0R_EL1 */
235 			g1 = false;
236 			break;
237 		}
238 	}
239 
240 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
241 
242 	return true;
243 }
244 
245 static bool access_gic_sre(struct kvm_vcpu *vcpu,
246 			   struct sys_reg_params *p,
247 			   const struct sys_reg_desc *r)
248 {
249 	if (p->is_write)
250 		return ignore_write(vcpu, p);
251 
252 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
253 	return true;
254 }
255 
256 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
257 			struct sys_reg_params *p,
258 			const struct sys_reg_desc *r)
259 {
260 	if (p->is_write)
261 		return ignore_write(vcpu, p);
262 	else
263 		return read_zero(vcpu, p);
264 }
265 
266 /*
267  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
268  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
269  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
270  * treat it separately.
271  */
272 static bool trap_loregion(struct kvm_vcpu *vcpu,
273 			  struct sys_reg_params *p,
274 			  const struct sys_reg_desc *r)
275 {
276 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
277 	u32 sr = reg_to_encoding(r);
278 
279 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
280 		kvm_inject_undefined(vcpu);
281 		return false;
282 	}
283 
284 	if (p->is_write && sr == SYS_LORID_EL1)
285 		return write_to_read_only(vcpu, p, r);
286 
287 	return trap_raz_wi(vcpu, p, r);
288 }
289 
290 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
291 			   struct sys_reg_params *p,
292 			   const struct sys_reg_desc *r)
293 {
294 	if (p->is_write) {
295 		return ignore_write(vcpu, p);
296 	} else {
297 		p->regval = (1 << 3);
298 		return true;
299 	}
300 }
301 
302 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
303 				   struct sys_reg_params *p,
304 				   const struct sys_reg_desc *r)
305 {
306 	if (p->is_write) {
307 		return ignore_write(vcpu, p);
308 	} else {
309 		p->regval = read_sysreg(dbgauthstatus_el1);
310 		return true;
311 	}
312 }
313 
314 /*
315  * We want to avoid world-switching all the DBG registers all the
316  * time:
317  *
318  * - If we've touched any debug register, it is likely that we're
319  *   going to touch more of them. It then makes sense to disable the
320  *   traps and start doing the save/restore dance
321  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
322  *   then mandatory to save/restore the registers, as the guest
323  *   depends on them.
324  *
325  * For this, we use a DIRTY bit, indicating the guest has modified the
326  * debug registers, used as follow:
327  *
328  * On guest entry:
329  * - If the dirty bit is set (because we're coming back from trapping),
330  *   disable the traps, save host registers, restore guest registers.
331  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
332  *   set the dirty bit, disable the traps, save host registers,
333  *   restore guest registers.
334  * - Otherwise, enable the traps
335  *
336  * On guest exit:
337  * - If the dirty bit is set, save guest registers, restore host
338  *   registers and clear the dirty bit. This ensure that the host can
339  *   now use the debug registers.
340  */
341 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
342 			    struct sys_reg_params *p,
343 			    const struct sys_reg_desc *r)
344 {
345 	if (p->is_write) {
346 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
347 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
348 	} else {
349 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
350 	}
351 
352 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
353 
354 	return true;
355 }
356 
357 /*
358  * reg_to_dbg/dbg_to_reg
359  *
360  * A 32 bit write to a debug register leave top bits alone
361  * A 32 bit read from a debug register only returns the bottom bits
362  *
363  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
364  * hyp.S code switches between host and guest values in future.
365  */
366 static void reg_to_dbg(struct kvm_vcpu *vcpu,
367 		       struct sys_reg_params *p,
368 		       const struct sys_reg_desc *rd,
369 		       u64 *dbg_reg)
370 {
371 	u64 mask, shift, val;
372 
373 	get_access_mask(rd, &mask, &shift);
374 
375 	val = *dbg_reg;
376 	val &= ~mask;
377 	val |= (p->regval & (mask >> shift)) << shift;
378 	*dbg_reg = val;
379 
380 	vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
381 }
382 
383 static void dbg_to_reg(struct kvm_vcpu *vcpu,
384 		       struct sys_reg_params *p,
385 		       const struct sys_reg_desc *rd,
386 		       u64 *dbg_reg)
387 {
388 	u64 mask, shift;
389 
390 	get_access_mask(rd, &mask, &shift);
391 	p->regval = (*dbg_reg & mask) >> shift;
392 }
393 
394 static bool trap_bvr(struct kvm_vcpu *vcpu,
395 		     struct sys_reg_params *p,
396 		     const struct sys_reg_desc *rd)
397 {
398 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
399 
400 	if (p->is_write)
401 		reg_to_dbg(vcpu, p, rd, dbg_reg);
402 	else
403 		dbg_to_reg(vcpu, p, rd, dbg_reg);
404 
405 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
406 
407 	return true;
408 }
409 
410 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
411 		const struct kvm_one_reg *reg, void __user *uaddr)
412 {
413 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
414 
415 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
416 		return -EFAULT;
417 	return 0;
418 }
419 
420 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
421 	const struct kvm_one_reg *reg, void __user *uaddr)
422 {
423 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm];
424 
425 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
426 		return -EFAULT;
427 	return 0;
428 }
429 
430 static void reset_bvr(struct kvm_vcpu *vcpu,
431 		      const struct sys_reg_desc *rd)
432 {
433 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->CRm] = rd->val;
434 }
435 
436 static bool trap_bcr(struct kvm_vcpu *vcpu,
437 		     struct sys_reg_params *p,
438 		     const struct sys_reg_desc *rd)
439 {
440 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
441 
442 	if (p->is_write)
443 		reg_to_dbg(vcpu, p, rd, dbg_reg);
444 	else
445 		dbg_to_reg(vcpu, p, rd, dbg_reg);
446 
447 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
448 
449 	return true;
450 }
451 
452 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
453 		const struct kvm_one_reg *reg, void __user *uaddr)
454 {
455 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
456 
457 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
458 		return -EFAULT;
459 
460 	return 0;
461 }
462 
463 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
464 	const struct kvm_one_reg *reg, void __user *uaddr)
465 {
466 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm];
467 
468 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
469 		return -EFAULT;
470 	return 0;
471 }
472 
473 static void reset_bcr(struct kvm_vcpu *vcpu,
474 		      const struct sys_reg_desc *rd)
475 {
476 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->CRm] = rd->val;
477 }
478 
479 static bool trap_wvr(struct kvm_vcpu *vcpu,
480 		     struct sys_reg_params *p,
481 		     const struct sys_reg_desc *rd)
482 {
483 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
484 
485 	if (p->is_write)
486 		reg_to_dbg(vcpu, p, rd, dbg_reg);
487 	else
488 		dbg_to_reg(vcpu, p, rd, dbg_reg);
489 
490 	trace_trap_reg(__func__, rd->CRm, p->is_write,
491 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm]);
492 
493 	return true;
494 }
495 
496 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
497 		const struct kvm_one_reg *reg, void __user *uaddr)
498 {
499 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
500 
501 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
502 		return -EFAULT;
503 	return 0;
504 }
505 
506 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
507 	const struct kvm_one_reg *reg, void __user *uaddr)
508 {
509 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm];
510 
511 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
512 		return -EFAULT;
513 	return 0;
514 }
515 
516 static void reset_wvr(struct kvm_vcpu *vcpu,
517 		      const struct sys_reg_desc *rd)
518 {
519 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->CRm] = rd->val;
520 }
521 
522 static bool trap_wcr(struct kvm_vcpu *vcpu,
523 		     struct sys_reg_params *p,
524 		     const struct sys_reg_desc *rd)
525 {
526 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
527 
528 	if (p->is_write)
529 		reg_to_dbg(vcpu, p, rd, dbg_reg);
530 	else
531 		dbg_to_reg(vcpu, p, rd, dbg_reg);
532 
533 	trace_trap_reg(__func__, rd->CRm, p->is_write, *dbg_reg);
534 
535 	return true;
536 }
537 
538 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
539 		const struct kvm_one_reg *reg, void __user *uaddr)
540 {
541 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
542 
543 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
544 		return -EFAULT;
545 	return 0;
546 }
547 
548 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
549 	const struct kvm_one_reg *reg, void __user *uaddr)
550 {
551 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm];
552 
553 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
554 		return -EFAULT;
555 	return 0;
556 }
557 
558 static void reset_wcr(struct kvm_vcpu *vcpu,
559 		      const struct sys_reg_desc *rd)
560 {
561 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->CRm] = rd->val;
562 }
563 
564 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
565 {
566 	u64 amair = read_sysreg(amair_el1);
567 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
568 }
569 
570 static void reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
571 {
572 	u64 actlr = read_sysreg(actlr_el1);
573 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
574 }
575 
576 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
577 {
578 	u64 mpidr;
579 
580 	/*
581 	 * Map the vcpu_id into the first three affinity level fields of
582 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
583 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
584 	 * of the GICv3 to be able to address each CPU directly when
585 	 * sending IPIs.
586 	 */
587 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
588 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
589 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
590 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
591 }
592 
593 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
594 				   const struct sys_reg_desc *r)
595 {
596 	if (kvm_vcpu_has_pmu(vcpu))
597 		return 0;
598 
599 	return REG_HIDDEN;
600 }
601 
602 static void reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
603 {
604 	u64 n, mask = BIT(ARMV8_PMU_CYCLE_IDX);
605 
606 	/* No PMU available, any PMU reg may UNDEF... */
607 	if (!kvm_arm_support_pmu_v3())
608 		return;
609 
610 	n = read_sysreg(pmcr_el0) >> ARMV8_PMU_PMCR_N_SHIFT;
611 	n &= ARMV8_PMU_PMCR_N_MASK;
612 	if (n)
613 		mask |= GENMASK(n - 1, 0);
614 
615 	reset_unknown(vcpu, r);
616 	__vcpu_sys_reg(vcpu, r->reg) &= mask;
617 }
618 
619 static void reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
620 {
621 	reset_unknown(vcpu, r);
622 	__vcpu_sys_reg(vcpu, r->reg) &= GENMASK(31, 0);
623 }
624 
625 static void reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
626 {
627 	reset_unknown(vcpu, r);
628 	__vcpu_sys_reg(vcpu, r->reg) &= ARMV8_PMU_EVTYPE_MASK;
629 }
630 
631 static void reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
632 {
633 	reset_unknown(vcpu, r);
634 	__vcpu_sys_reg(vcpu, r->reg) &= ARMV8_PMU_COUNTER_MASK;
635 }
636 
637 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
638 {
639 	u64 pmcr, val;
640 
641 	/* No PMU available, PMCR_EL0 may UNDEF... */
642 	if (!kvm_arm_support_pmu_v3())
643 		return;
644 
645 	pmcr = read_sysreg(pmcr_el0);
646 	/*
647 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
648 	 * except PMCR.E resetting to zero.
649 	 */
650 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
651 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
652 	if (!system_supports_32bit_el0())
653 		val |= ARMV8_PMU_PMCR_LC;
654 	__vcpu_sys_reg(vcpu, r->reg) = val;
655 }
656 
657 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
658 {
659 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
660 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
661 
662 	if (!enabled)
663 		kvm_inject_undefined(vcpu);
664 
665 	return !enabled;
666 }
667 
668 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
669 {
670 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
671 }
672 
673 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
674 {
675 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
676 }
677 
678 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
679 {
680 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
681 }
682 
683 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
684 {
685 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
686 }
687 
688 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
689 			const struct sys_reg_desc *r)
690 {
691 	u64 val;
692 
693 	if (pmu_access_el0_disabled(vcpu))
694 		return false;
695 
696 	if (p->is_write) {
697 		/* Only update writeable bits of PMCR */
698 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
699 		val &= ~ARMV8_PMU_PMCR_MASK;
700 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
701 		if (!system_supports_32bit_el0())
702 			val |= ARMV8_PMU_PMCR_LC;
703 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
704 		kvm_pmu_handle_pmcr(vcpu, val);
705 		kvm_vcpu_pmu_restore_guest(vcpu);
706 	} else {
707 		/* PMCR.P & PMCR.C are RAZ */
708 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
709 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
710 		p->regval = val;
711 	}
712 
713 	return true;
714 }
715 
716 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
717 			  const struct sys_reg_desc *r)
718 {
719 	if (pmu_access_event_counter_el0_disabled(vcpu))
720 		return false;
721 
722 	if (p->is_write)
723 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
724 	else
725 		/* return PMSELR.SEL field */
726 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
727 			    & ARMV8_PMU_COUNTER_MASK;
728 
729 	return true;
730 }
731 
732 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
733 			  const struct sys_reg_desc *r)
734 {
735 	u64 pmceid, mask, shift;
736 
737 	BUG_ON(p->is_write);
738 
739 	if (pmu_access_el0_disabled(vcpu))
740 		return false;
741 
742 	get_access_mask(r, &mask, &shift);
743 
744 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
745 	pmceid &= mask;
746 	pmceid >>= shift;
747 
748 	p->regval = pmceid;
749 
750 	return true;
751 }
752 
753 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
754 {
755 	u64 pmcr, val;
756 
757 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
758 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
759 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
760 		kvm_inject_undefined(vcpu);
761 		return false;
762 	}
763 
764 	return true;
765 }
766 
767 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
768 			      struct sys_reg_params *p,
769 			      const struct sys_reg_desc *r)
770 {
771 	u64 idx = ~0UL;
772 
773 	if (r->CRn == 9 && r->CRm == 13) {
774 		if (r->Op2 == 2) {
775 			/* PMXEVCNTR_EL0 */
776 			if (pmu_access_event_counter_el0_disabled(vcpu))
777 				return false;
778 
779 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
780 			      & ARMV8_PMU_COUNTER_MASK;
781 		} else if (r->Op2 == 0) {
782 			/* PMCCNTR_EL0 */
783 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
784 				return false;
785 
786 			idx = ARMV8_PMU_CYCLE_IDX;
787 		}
788 	} else if (r->CRn == 0 && r->CRm == 9) {
789 		/* PMCCNTR */
790 		if (pmu_access_event_counter_el0_disabled(vcpu))
791 			return false;
792 
793 		idx = ARMV8_PMU_CYCLE_IDX;
794 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
795 		/* PMEVCNTRn_EL0 */
796 		if (pmu_access_event_counter_el0_disabled(vcpu))
797 			return false;
798 
799 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
800 	}
801 
802 	/* Catch any decoding mistake */
803 	WARN_ON(idx == ~0UL);
804 
805 	if (!pmu_counter_idx_valid(vcpu, idx))
806 		return false;
807 
808 	if (p->is_write) {
809 		if (pmu_access_el0_disabled(vcpu))
810 			return false;
811 
812 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
813 	} else {
814 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
815 	}
816 
817 	return true;
818 }
819 
820 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
821 			       const struct sys_reg_desc *r)
822 {
823 	u64 idx, reg;
824 
825 	if (pmu_access_el0_disabled(vcpu))
826 		return false;
827 
828 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
829 		/* PMXEVTYPER_EL0 */
830 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
831 		reg = PMEVTYPER0_EL0 + idx;
832 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
833 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
834 		if (idx == ARMV8_PMU_CYCLE_IDX)
835 			reg = PMCCFILTR_EL0;
836 		else
837 			/* PMEVTYPERn_EL0 */
838 			reg = PMEVTYPER0_EL0 + idx;
839 	} else {
840 		BUG();
841 	}
842 
843 	if (!pmu_counter_idx_valid(vcpu, idx))
844 		return false;
845 
846 	if (p->is_write) {
847 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
848 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
849 		kvm_vcpu_pmu_restore_guest(vcpu);
850 	} else {
851 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
852 	}
853 
854 	return true;
855 }
856 
857 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
858 			   const struct sys_reg_desc *r)
859 {
860 	u64 val, mask;
861 
862 	if (pmu_access_el0_disabled(vcpu))
863 		return false;
864 
865 	mask = kvm_pmu_valid_counter_mask(vcpu);
866 	if (p->is_write) {
867 		val = p->regval & mask;
868 		if (r->Op2 & 0x1) {
869 			/* accessing PMCNTENSET_EL0 */
870 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
871 			kvm_pmu_enable_counter_mask(vcpu, val);
872 			kvm_vcpu_pmu_restore_guest(vcpu);
873 		} else {
874 			/* accessing PMCNTENCLR_EL0 */
875 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
876 			kvm_pmu_disable_counter_mask(vcpu, val);
877 		}
878 	} else {
879 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
880 	}
881 
882 	return true;
883 }
884 
885 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
886 			   const struct sys_reg_desc *r)
887 {
888 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
889 
890 	if (check_pmu_access_disabled(vcpu, 0))
891 		return false;
892 
893 	if (p->is_write) {
894 		u64 val = p->regval & mask;
895 
896 		if (r->Op2 & 0x1)
897 			/* accessing PMINTENSET_EL1 */
898 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
899 		else
900 			/* accessing PMINTENCLR_EL1 */
901 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
902 	} else {
903 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
904 	}
905 
906 	return true;
907 }
908 
909 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
910 			 const struct sys_reg_desc *r)
911 {
912 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
913 
914 	if (pmu_access_el0_disabled(vcpu))
915 		return false;
916 
917 	if (p->is_write) {
918 		if (r->CRm & 0x2)
919 			/* accessing PMOVSSET_EL0 */
920 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
921 		else
922 			/* accessing PMOVSCLR_EL0 */
923 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
924 	} else {
925 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
926 	}
927 
928 	return true;
929 }
930 
931 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
932 			   const struct sys_reg_desc *r)
933 {
934 	u64 mask;
935 
936 	if (!p->is_write)
937 		return read_from_write_only(vcpu, p, r);
938 
939 	if (pmu_write_swinc_el0_disabled(vcpu))
940 		return false;
941 
942 	mask = kvm_pmu_valid_counter_mask(vcpu);
943 	kvm_pmu_software_increment(vcpu, p->regval & mask);
944 	return true;
945 }
946 
947 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
948 			     const struct sys_reg_desc *r)
949 {
950 	if (p->is_write) {
951 		if (!vcpu_mode_priv(vcpu)) {
952 			kvm_inject_undefined(vcpu);
953 			return false;
954 		}
955 
956 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
957 			       p->regval & ARMV8_PMU_USERENR_MASK;
958 	} else {
959 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
960 			    & ARMV8_PMU_USERENR_MASK;
961 	}
962 
963 	return true;
964 }
965 
966 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
967 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
968 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
969 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
970 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
971 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
972 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
973 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
974 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
975 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
976 
977 #define PMU_SYS_REG(r)						\
978 	SYS_DESC(r), .reset = reset_pmu_reg, .visibility = pmu_visibility
979 
980 /* Macro to expand the PMEVCNTRn_EL0 register */
981 #define PMU_PMEVCNTR_EL0(n)						\
982 	{ PMU_SYS_REG(SYS_PMEVCNTRn_EL0(n)),				\
983 	  .reset = reset_pmevcntr,					\
984 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
985 
986 /* Macro to expand the PMEVTYPERn_EL0 register */
987 #define PMU_PMEVTYPER_EL0(n)						\
988 	{ PMU_SYS_REG(SYS_PMEVTYPERn_EL0(n)),				\
989 	  .reset = reset_pmevtyper,					\
990 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
991 
992 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
993 			 const struct sys_reg_desc *r)
994 {
995 	kvm_inject_undefined(vcpu);
996 
997 	return false;
998 }
999 
1000 /* Macro to expand the AMU counter and type registers*/
1001 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1002 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1003 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1004 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1005 
1006 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1007 			const struct sys_reg_desc *rd)
1008 {
1009 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1010 }
1011 
1012 /*
1013  * If we land here on a PtrAuth access, that is because we didn't
1014  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1015  * way this happens is when the guest does not have PtrAuth support
1016  * enabled.
1017  */
1018 #define __PTRAUTH_KEY(k)						\
1019 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1020 	.visibility = ptrauth_visibility}
1021 
1022 #define PTRAUTH_KEY(k)							\
1023 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1024 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1025 
1026 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1027 			      struct sys_reg_params *p,
1028 			      const struct sys_reg_desc *r)
1029 {
1030 	enum kvm_arch_timers tmr;
1031 	enum kvm_arch_timer_regs treg;
1032 	u64 reg = reg_to_encoding(r);
1033 
1034 	switch (reg) {
1035 	case SYS_CNTP_TVAL_EL0:
1036 	case SYS_AARCH32_CNTP_TVAL:
1037 		tmr = TIMER_PTIMER;
1038 		treg = TIMER_REG_TVAL;
1039 		break;
1040 	case SYS_CNTP_CTL_EL0:
1041 	case SYS_AARCH32_CNTP_CTL:
1042 		tmr = TIMER_PTIMER;
1043 		treg = TIMER_REG_CTL;
1044 		break;
1045 	case SYS_CNTP_CVAL_EL0:
1046 	case SYS_AARCH32_CNTP_CVAL:
1047 		tmr = TIMER_PTIMER;
1048 		treg = TIMER_REG_CVAL;
1049 		break;
1050 	default:
1051 		BUG();
1052 	}
1053 
1054 	if (p->is_write)
1055 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1056 	else
1057 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1058 
1059 	return true;
1060 }
1061 
1062 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1063 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1064 		struct sys_reg_desc const *r, bool raz)
1065 {
1066 	u32 id = reg_to_encoding(r);
1067 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1068 
1069 	switch (id) {
1070 	case SYS_ID_AA64PFR0_EL1:
1071 		if (!vcpu_has_sve(vcpu))
1072 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_SVE);
1073 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_AMU);
1074 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2);
1075 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2), (u64)vcpu->kvm->arch.pfr0_csv2);
1076 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3);
1077 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3), (u64)vcpu->kvm->arch.pfr0_csv3);
1078 		break;
1079 	case SYS_ID_AA64PFR1_EL1:
1080 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_MTE);
1081 		if (kvm_has_mte(vcpu->kvm)) {
1082 			u64 pfr, mte;
1083 
1084 			pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1085 			mte = cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR1_MTE_SHIFT);
1086 			val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR1_MTE), mte);
1087 		}
1088 		break;
1089 	case SYS_ID_AA64ISAR1_EL1:
1090 		if (!vcpu_has_ptrauth(vcpu))
1091 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_APA) |
1092 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_API) |
1093 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_GPA) |
1094 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_GPI));
1095 		break;
1096 	case SYS_ID_AA64DFR0_EL1:
1097 		/* Limit debug to ARMv8.0 */
1098 		val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER);
1099 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER), 6);
1100 		/* Limit guests to PMUv3 for ARMv8.4 */
1101 		val = cpuid_feature_cap_perfmon_field(val,
1102 						      ID_AA64DFR0_PMUVER_SHIFT,
1103 						      kvm_vcpu_has_pmu(vcpu) ? ID_AA64DFR0_PMUVER_8_4 : 0);
1104 		/* Hide SPE from guests */
1105 		val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_PMSVER);
1106 		break;
1107 	case SYS_ID_DFR0_EL1:
1108 		/* Limit guests to PMUv3 for ARMv8.4 */
1109 		val = cpuid_feature_cap_perfmon_field(val,
1110 						      ID_DFR0_PERFMON_SHIFT,
1111 						      kvm_vcpu_has_pmu(vcpu) ? ID_DFR0_PERFMON_8_4 : 0);
1112 		break;
1113 	}
1114 
1115 	return val;
1116 }
1117 
1118 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1119 				  const struct sys_reg_desc *r)
1120 {
1121 	u32 id = reg_to_encoding(r);
1122 
1123 	switch (id) {
1124 	case SYS_ID_AA64ZFR0_EL1:
1125 		if (!vcpu_has_sve(vcpu))
1126 			return REG_RAZ;
1127 		break;
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 /* cpufeature ID register access trap handlers */
1134 
1135 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1136 			    struct sys_reg_params *p,
1137 			    const struct sys_reg_desc *r,
1138 			    bool raz)
1139 {
1140 	if (p->is_write)
1141 		return write_to_read_only(vcpu, p, r);
1142 
1143 	p->regval = read_id_reg(vcpu, r, raz);
1144 	return true;
1145 }
1146 
1147 static bool access_id_reg(struct kvm_vcpu *vcpu,
1148 			  struct sys_reg_params *p,
1149 			  const struct sys_reg_desc *r)
1150 {
1151 	bool raz = sysreg_visible_as_raz(vcpu, r);
1152 
1153 	return __access_id_reg(vcpu, p, r, raz);
1154 }
1155 
1156 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1157 			      struct sys_reg_params *p,
1158 			      const struct sys_reg_desc *r)
1159 {
1160 	return __access_id_reg(vcpu, p, r, true);
1161 }
1162 
1163 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1164 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1165 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1166 
1167 /* Visibility overrides for SVE-specific control registers */
1168 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1169 				   const struct sys_reg_desc *rd)
1170 {
1171 	if (vcpu_has_sve(vcpu))
1172 		return 0;
1173 
1174 	return REG_HIDDEN;
1175 }
1176 
1177 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1178 			       const struct sys_reg_desc *rd,
1179 			       const struct kvm_one_reg *reg, void __user *uaddr)
1180 {
1181 	const u64 id = sys_reg_to_index(rd);
1182 	u8 csv2, csv3;
1183 	int err;
1184 	u64 val;
1185 
1186 	err = reg_from_user(&val, uaddr, id);
1187 	if (err)
1188 		return err;
1189 
1190 	/*
1191 	 * Allow AA64PFR0_EL1.CSV2 to be set from userspace as long as
1192 	 * it doesn't promise more than what is actually provided (the
1193 	 * guest could otherwise be covered in ectoplasmic residue).
1194 	 */
1195 	csv2 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV2_SHIFT);
1196 	if (csv2 > 1 ||
1197 	    (csv2 && arm64_get_spectre_v2_state() != SPECTRE_UNAFFECTED))
1198 		return -EINVAL;
1199 
1200 	/* Same thing for CSV3 */
1201 	csv3 = cpuid_feature_extract_unsigned_field(val, ID_AA64PFR0_CSV3_SHIFT);
1202 	if (csv3 > 1 ||
1203 	    (csv3 && arm64_get_meltdown_state() != SPECTRE_UNAFFECTED))
1204 		return -EINVAL;
1205 
1206 	/* We can only differ with CSV[23], and anything else is an error */
1207 	val ^= read_id_reg(vcpu, rd, false);
1208 	val &= ~((0xFUL << ID_AA64PFR0_CSV2_SHIFT) |
1209 		 (0xFUL << ID_AA64PFR0_CSV3_SHIFT));
1210 	if (val)
1211 		return -EINVAL;
1212 
1213 	vcpu->kvm->arch.pfr0_csv2 = csv2;
1214 	vcpu->kvm->arch.pfr0_csv3 = csv3 ;
1215 
1216 	return 0;
1217 }
1218 
1219 /*
1220  * cpufeature ID register user accessors
1221  *
1222  * For now, these registers are immutable for userspace, so no values
1223  * are stored, and for set_id_reg() we don't allow the effective value
1224  * to be changed.
1225  */
1226 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1227 			const struct sys_reg_desc *rd, void __user *uaddr,
1228 			bool raz)
1229 {
1230 	const u64 id = sys_reg_to_index(rd);
1231 	const u64 val = read_id_reg(vcpu, rd, raz);
1232 
1233 	return reg_to_user(uaddr, &val, id);
1234 }
1235 
1236 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1237 			const struct sys_reg_desc *rd, void __user *uaddr,
1238 			bool raz)
1239 {
1240 	const u64 id = sys_reg_to_index(rd);
1241 	int err;
1242 	u64 val;
1243 
1244 	err = reg_from_user(&val, uaddr, id);
1245 	if (err)
1246 		return err;
1247 
1248 	/* This is what we mean by invariant: you can't change it. */
1249 	if (val != read_id_reg(vcpu, rd, raz))
1250 		return -EINVAL;
1251 
1252 	return 0;
1253 }
1254 
1255 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1256 		      const struct kvm_one_reg *reg, void __user *uaddr)
1257 {
1258 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1259 
1260 	return __get_id_reg(vcpu, rd, uaddr, raz);
1261 }
1262 
1263 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1264 		      const struct kvm_one_reg *reg, void __user *uaddr)
1265 {
1266 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1267 
1268 	return __set_id_reg(vcpu, rd, uaddr, raz);
1269 }
1270 
1271 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1272 			  const struct kvm_one_reg *reg, void __user *uaddr)
1273 {
1274 	return __get_id_reg(vcpu, rd, uaddr, true);
1275 }
1276 
1277 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1278 			  const struct kvm_one_reg *reg, void __user *uaddr)
1279 {
1280 	return __set_id_reg(vcpu, rd, uaddr, true);
1281 }
1282 
1283 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1284 		      const struct kvm_one_reg *reg, void __user *uaddr)
1285 {
1286 	int err;
1287 	u64 val;
1288 
1289 	/* Perform the access even if we are going to ignore the value */
1290 	err = reg_from_user(&val, uaddr, sys_reg_to_index(rd));
1291 	if (err)
1292 		return err;
1293 
1294 	return 0;
1295 }
1296 
1297 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1298 		       const struct sys_reg_desc *r)
1299 {
1300 	if (p->is_write)
1301 		return write_to_read_only(vcpu, p, r);
1302 
1303 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1304 	return true;
1305 }
1306 
1307 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1308 			 const struct sys_reg_desc *r)
1309 {
1310 	if (p->is_write)
1311 		return write_to_read_only(vcpu, p, r);
1312 
1313 	p->regval = read_sysreg(clidr_el1);
1314 	return true;
1315 }
1316 
1317 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1318 			  const struct sys_reg_desc *r)
1319 {
1320 	int reg = r->reg;
1321 
1322 	if (p->is_write)
1323 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1324 	else
1325 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1326 	return true;
1327 }
1328 
1329 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1330 			  const struct sys_reg_desc *r)
1331 {
1332 	u32 csselr;
1333 
1334 	if (p->is_write)
1335 		return write_to_read_only(vcpu, p, r);
1336 
1337 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1338 	p->regval = get_ccsidr(csselr);
1339 
1340 	/*
1341 	 * Guests should not be doing cache operations by set/way at all, and
1342 	 * for this reason, we trap them and attempt to infer the intent, so
1343 	 * that we can flush the entire guest's address space at the appropriate
1344 	 * time.
1345 	 * To prevent this trapping from causing performance problems, let's
1346 	 * expose the geometry of all data and unified caches (which are
1347 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1348 	 * [If guests should attempt to infer aliasing properties from the
1349 	 * geometry (which is not permitted by the architecture), they would
1350 	 * only do so for virtually indexed caches.]
1351 	 */
1352 	if (!(csselr & 1)) // data or unified cache
1353 		p->regval &= ~GENMASK(27, 3);
1354 	return true;
1355 }
1356 
1357 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
1358 				   const struct sys_reg_desc *rd)
1359 {
1360 	if (kvm_has_mte(vcpu->kvm))
1361 		return 0;
1362 
1363 	return REG_HIDDEN;
1364 }
1365 
1366 #define MTE_REG(name) {				\
1367 	SYS_DESC(SYS_##name),			\
1368 	.access = undef_access,			\
1369 	.reset = reset_unknown,			\
1370 	.reg = name,				\
1371 	.visibility = mte_visibility,		\
1372 }
1373 
1374 /* sys_reg_desc initialiser for known cpufeature ID registers */
1375 #define ID_SANITISED(name) {			\
1376 	SYS_DESC(SYS_##name),			\
1377 	.access	= access_id_reg,		\
1378 	.get_user = get_id_reg,			\
1379 	.set_user = set_id_reg,			\
1380 	.visibility = id_visibility,		\
1381 }
1382 
1383 /*
1384  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1385  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1386  * (1 <= crm < 8, 0 <= Op2 < 8).
1387  */
1388 #define ID_UNALLOCATED(crm, op2) {			\
1389 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1390 	.access = access_raz_id_reg,			\
1391 	.get_user = get_raz_id_reg,			\
1392 	.set_user = set_raz_id_reg,			\
1393 }
1394 
1395 /*
1396  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1397  * For now, these are exposed just like unallocated ID regs: they appear
1398  * RAZ for the guest.
1399  */
1400 #define ID_HIDDEN(name) {			\
1401 	SYS_DESC(SYS_##name),			\
1402 	.access = access_raz_id_reg,		\
1403 	.get_user = get_raz_id_reg,		\
1404 	.set_user = set_raz_id_reg,		\
1405 }
1406 
1407 /*
1408  * Architected system registers.
1409  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1410  *
1411  * Debug handling: We do trap most, if not all debug related system
1412  * registers. The implementation is good enough to ensure that a guest
1413  * can use these with minimal performance degradation. The drawback is
1414  * that we don't implement any of the external debug, none of the
1415  * OSlock protocol. This should be revisited if we ever encounter a
1416  * more demanding guest...
1417  */
1418 static const struct sys_reg_desc sys_reg_descs[] = {
1419 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1420 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1421 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1422 
1423 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1424 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1425 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1426 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1427 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1428 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1429 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1430 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1431 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1432 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1433 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1434 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1435 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1436 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1437 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1438 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1439 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1440 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1441 
1442 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1443 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1444 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1445 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1446 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1447 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1448 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1449 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1450 
1451 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1452 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1453 	// DBGDTR[TR]X_EL0 share the same encoding
1454 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1455 
1456 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1457 
1458 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1459 
1460 	/*
1461 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1462 	 * entries in arm64_ftr_regs[].
1463 	 */
1464 
1465 	/* AArch64 mappings of the AArch32 ID registers */
1466 	/* CRm=1 */
1467 	ID_SANITISED(ID_PFR0_EL1),
1468 	ID_SANITISED(ID_PFR1_EL1),
1469 	ID_SANITISED(ID_DFR0_EL1),
1470 	ID_HIDDEN(ID_AFR0_EL1),
1471 	ID_SANITISED(ID_MMFR0_EL1),
1472 	ID_SANITISED(ID_MMFR1_EL1),
1473 	ID_SANITISED(ID_MMFR2_EL1),
1474 	ID_SANITISED(ID_MMFR3_EL1),
1475 
1476 	/* CRm=2 */
1477 	ID_SANITISED(ID_ISAR0_EL1),
1478 	ID_SANITISED(ID_ISAR1_EL1),
1479 	ID_SANITISED(ID_ISAR2_EL1),
1480 	ID_SANITISED(ID_ISAR3_EL1),
1481 	ID_SANITISED(ID_ISAR4_EL1),
1482 	ID_SANITISED(ID_ISAR5_EL1),
1483 	ID_SANITISED(ID_MMFR4_EL1),
1484 	ID_SANITISED(ID_ISAR6_EL1),
1485 
1486 	/* CRm=3 */
1487 	ID_SANITISED(MVFR0_EL1),
1488 	ID_SANITISED(MVFR1_EL1),
1489 	ID_SANITISED(MVFR2_EL1),
1490 	ID_UNALLOCATED(3,3),
1491 	ID_SANITISED(ID_PFR2_EL1),
1492 	ID_HIDDEN(ID_DFR1_EL1),
1493 	ID_SANITISED(ID_MMFR5_EL1),
1494 	ID_UNALLOCATED(3,7),
1495 
1496 	/* AArch64 ID registers */
1497 	/* CRm=4 */
1498 	{ SYS_DESC(SYS_ID_AA64PFR0_EL1), .access = access_id_reg,
1499 	  .get_user = get_id_reg, .set_user = set_id_aa64pfr0_el1, },
1500 	ID_SANITISED(ID_AA64PFR1_EL1),
1501 	ID_UNALLOCATED(4,2),
1502 	ID_UNALLOCATED(4,3),
1503 	ID_SANITISED(ID_AA64ZFR0_EL1),
1504 	ID_UNALLOCATED(4,5),
1505 	ID_UNALLOCATED(4,6),
1506 	ID_UNALLOCATED(4,7),
1507 
1508 	/* CRm=5 */
1509 	ID_SANITISED(ID_AA64DFR0_EL1),
1510 	ID_SANITISED(ID_AA64DFR1_EL1),
1511 	ID_UNALLOCATED(5,2),
1512 	ID_UNALLOCATED(5,3),
1513 	ID_HIDDEN(ID_AA64AFR0_EL1),
1514 	ID_HIDDEN(ID_AA64AFR1_EL1),
1515 	ID_UNALLOCATED(5,6),
1516 	ID_UNALLOCATED(5,7),
1517 
1518 	/* CRm=6 */
1519 	ID_SANITISED(ID_AA64ISAR0_EL1),
1520 	ID_SANITISED(ID_AA64ISAR1_EL1),
1521 	ID_UNALLOCATED(6,2),
1522 	ID_UNALLOCATED(6,3),
1523 	ID_UNALLOCATED(6,4),
1524 	ID_UNALLOCATED(6,5),
1525 	ID_UNALLOCATED(6,6),
1526 	ID_UNALLOCATED(6,7),
1527 
1528 	/* CRm=7 */
1529 	ID_SANITISED(ID_AA64MMFR0_EL1),
1530 	ID_SANITISED(ID_AA64MMFR1_EL1),
1531 	ID_SANITISED(ID_AA64MMFR2_EL1),
1532 	ID_UNALLOCATED(7,3),
1533 	ID_UNALLOCATED(7,4),
1534 	ID_UNALLOCATED(7,5),
1535 	ID_UNALLOCATED(7,6),
1536 	ID_UNALLOCATED(7,7),
1537 
1538 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1539 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1540 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1541 
1542 	MTE_REG(RGSR_EL1),
1543 	MTE_REG(GCR_EL1),
1544 
1545 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1546 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
1547 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1548 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1549 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1550 
1551 	PTRAUTH_KEY(APIA),
1552 	PTRAUTH_KEY(APIB),
1553 	PTRAUTH_KEY(APDA),
1554 	PTRAUTH_KEY(APDB),
1555 	PTRAUTH_KEY(APGA),
1556 
1557 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1558 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1559 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1560 
1561 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1562 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1563 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1564 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1565 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1566 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1567 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1568 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1569 
1570 	MTE_REG(TFSR_EL1),
1571 	MTE_REG(TFSRE0_EL1),
1572 
1573 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1574 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1575 
1576 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
1577 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
1578 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
1579 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
1580 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
1581 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
1582 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
1583 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
1584 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
1585 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
1586 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
1587 	/* PMBIDR_EL1 is not trapped */
1588 
1589 	{ PMU_SYS_REG(SYS_PMINTENSET_EL1),
1590 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1591 	{ PMU_SYS_REG(SYS_PMINTENCLR_EL1),
1592 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1593 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
1594 
1595 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1596 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1597 
1598 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1599 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1600 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1601 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1602 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1603 
1604 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1605 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1606 
1607 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1608 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1609 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1610 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1611 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1612 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1613 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1614 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1615 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1616 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1617 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1618 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1619 
1620 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1621 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1622 
1623 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
1624 
1625 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1626 
1627 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1628 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1629 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1630 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1631 
1632 	{ PMU_SYS_REG(SYS_PMCR_EL0), .access = access_pmcr,
1633 	  .reset = reset_pmcr, .reg = PMCR_EL0 },
1634 	{ PMU_SYS_REG(SYS_PMCNTENSET_EL0),
1635 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1636 	{ PMU_SYS_REG(SYS_PMCNTENCLR_EL0),
1637 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1638 	{ PMU_SYS_REG(SYS_PMOVSCLR_EL0),
1639 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1640 	/*
1641 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
1642 	 * previously (and pointlessly) advertised in the past...
1643 	 */
1644 	{ PMU_SYS_REG(SYS_PMSWINC_EL0),
1645 	  .get_user = get_raz_id_reg, .set_user = set_wi_reg,
1646 	  .access = access_pmswinc, .reset = NULL },
1647 	{ PMU_SYS_REG(SYS_PMSELR_EL0),
1648 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
1649 	{ PMU_SYS_REG(SYS_PMCEID0_EL0),
1650 	  .access = access_pmceid, .reset = NULL },
1651 	{ PMU_SYS_REG(SYS_PMCEID1_EL0),
1652 	  .access = access_pmceid, .reset = NULL },
1653 	{ PMU_SYS_REG(SYS_PMCCNTR_EL0),
1654 	  .access = access_pmu_evcntr, .reset = reset_unknown, .reg = PMCCNTR_EL0 },
1655 	{ PMU_SYS_REG(SYS_PMXEVTYPER_EL0),
1656 	  .access = access_pmu_evtyper, .reset = NULL },
1657 	{ PMU_SYS_REG(SYS_PMXEVCNTR_EL0),
1658 	  .access = access_pmu_evcntr, .reset = NULL },
1659 	/*
1660 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1661 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1662 	 */
1663 	{ PMU_SYS_REG(SYS_PMUSERENR_EL0), .access = access_pmuserenr,
1664 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
1665 	{ PMU_SYS_REG(SYS_PMOVSSET_EL0),
1666 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1667 
1668 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1669 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1670 
1671 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
1672 
1673 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
1674 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
1675 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
1676 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
1677 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
1678 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
1679 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
1680 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
1681 	AMU_AMEVCNTR0_EL0(0),
1682 	AMU_AMEVCNTR0_EL0(1),
1683 	AMU_AMEVCNTR0_EL0(2),
1684 	AMU_AMEVCNTR0_EL0(3),
1685 	AMU_AMEVCNTR0_EL0(4),
1686 	AMU_AMEVCNTR0_EL0(5),
1687 	AMU_AMEVCNTR0_EL0(6),
1688 	AMU_AMEVCNTR0_EL0(7),
1689 	AMU_AMEVCNTR0_EL0(8),
1690 	AMU_AMEVCNTR0_EL0(9),
1691 	AMU_AMEVCNTR0_EL0(10),
1692 	AMU_AMEVCNTR0_EL0(11),
1693 	AMU_AMEVCNTR0_EL0(12),
1694 	AMU_AMEVCNTR0_EL0(13),
1695 	AMU_AMEVCNTR0_EL0(14),
1696 	AMU_AMEVCNTR0_EL0(15),
1697 	AMU_AMEVTYPER0_EL0(0),
1698 	AMU_AMEVTYPER0_EL0(1),
1699 	AMU_AMEVTYPER0_EL0(2),
1700 	AMU_AMEVTYPER0_EL0(3),
1701 	AMU_AMEVTYPER0_EL0(4),
1702 	AMU_AMEVTYPER0_EL0(5),
1703 	AMU_AMEVTYPER0_EL0(6),
1704 	AMU_AMEVTYPER0_EL0(7),
1705 	AMU_AMEVTYPER0_EL0(8),
1706 	AMU_AMEVTYPER0_EL0(9),
1707 	AMU_AMEVTYPER0_EL0(10),
1708 	AMU_AMEVTYPER0_EL0(11),
1709 	AMU_AMEVTYPER0_EL0(12),
1710 	AMU_AMEVTYPER0_EL0(13),
1711 	AMU_AMEVTYPER0_EL0(14),
1712 	AMU_AMEVTYPER0_EL0(15),
1713 	AMU_AMEVCNTR1_EL0(0),
1714 	AMU_AMEVCNTR1_EL0(1),
1715 	AMU_AMEVCNTR1_EL0(2),
1716 	AMU_AMEVCNTR1_EL0(3),
1717 	AMU_AMEVCNTR1_EL0(4),
1718 	AMU_AMEVCNTR1_EL0(5),
1719 	AMU_AMEVCNTR1_EL0(6),
1720 	AMU_AMEVCNTR1_EL0(7),
1721 	AMU_AMEVCNTR1_EL0(8),
1722 	AMU_AMEVCNTR1_EL0(9),
1723 	AMU_AMEVCNTR1_EL0(10),
1724 	AMU_AMEVCNTR1_EL0(11),
1725 	AMU_AMEVCNTR1_EL0(12),
1726 	AMU_AMEVCNTR1_EL0(13),
1727 	AMU_AMEVCNTR1_EL0(14),
1728 	AMU_AMEVCNTR1_EL0(15),
1729 	AMU_AMEVTYPER1_EL0(0),
1730 	AMU_AMEVTYPER1_EL0(1),
1731 	AMU_AMEVTYPER1_EL0(2),
1732 	AMU_AMEVTYPER1_EL0(3),
1733 	AMU_AMEVTYPER1_EL0(4),
1734 	AMU_AMEVTYPER1_EL0(5),
1735 	AMU_AMEVTYPER1_EL0(6),
1736 	AMU_AMEVTYPER1_EL0(7),
1737 	AMU_AMEVTYPER1_EL0(8),
1738 	AMU_AMEVTYPER1_EL0(9),
1739 	AMU_AMEVTYPER1_EL0(10),
1740 	AMU_AMEVTYPER1_EL0(11),
1741 	AMU_AMEVTYPER1_EL0(12),
1742 	AMU_AMEVTYPER1_EL0(13),
1743 	AMU_AMEVTYPER1_EL0(14),
1744 	AMU_AMEVTYPER1_EL0(15),
1745 
1746 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1747 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1748 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1749 
1750 	/* PMEVCNTRn_EL0 */
1751 	PMU_PMEVCNTR_EL0(0),
1752 	PMU_PMEVCNTR_EL0(1),
1753 	PMU_PMEVCNTR_EL0(2),
1754 	PMU_PMEVCNTR_EL0(3),
1755 	PMU_PMEVCNTR_EL0(4),
1756 	PMU_PMEVCNTR_EL0(5),
1757 	PMU_PMEVCNTR_EL0(6),
1758 	PMU_PMEVCNTR_EL0(7),
1759 	PMU_PMEVCNTR_EL0(8),
1760 	PMU_PMEVCNTR_EL0(9),
1761 	PMU_PMEVCNTR_EL0(10),
1762 	PMU_PMEVCNTR_EL0(11),
1763 	PMU_PMEVCNTR_EL0(12),
1764 	PMU_PMEVCNTR_EL0(13),
1765 	PMU_PMEVCNTR_EL0(14),
1766 	PMU_PMEVCNTR_EL0(15),
1767 	PMU_PMEVCNTR_EL0(16),
1768 	PMU_PMEVCNTR_EL0(17),
1769 	PMU_PMEVCNTR_EL0(18),
1770 	PMU_PMEVCNTR_EL0(19),
1771 	PMU_PMEVCNTR_EL0(20),
1772 	PMU_PMEVCNTR_EL0(21),
1773 	PMU_PMEVCNTR_EL0(22),
1774 	PMU_PMEVCNTR_EL0(23),
1775 	PMU_PMEVCNTR_EL0(24),
1776 	PMU_PMEVCNTR_EL0(25),
1777 	PMU_PMEVCNTR_EL0(26),
1778 	PMU_PMEVCNTR_EL0(27),
1779 	PMU_PMEVCNTR_EL0(28),
1780 	PMU_PMEVCNTR_EL0(29),
1781 	PMU_PMEVCNTR_EL0(30),
1782 	/* PMEVTYPERn_EL0 */
1783 	PMU_PMEVTYPER_EL0(0),
1784 	PMU_PMEVTYPER_EL0(1),
1785 	PMU_PMEVTYPER_EL0(2),
1786 	PMU_PMEVTYPER_EL0(3),
1787 	PMU_PMEVTYPER_EL0(4),
1788 	PMU_PMEVTYPER_EL0(5),
1789 	PMU_PMEVTYPER_EL0(6),
1790 	PMU_PMEVTYPER_EL0(7),
1791 	PMU_PMEVTYPER_EL0(8),
1792 	PMU_PMEVTYPER_EL0(9),
1793 	PMU_PMEVTYPER_EL0(10),
1794 	PMU_PMEVTYPER_EL0(11),
1795 	PMU_PMEVTYPER_EL0(12),
1796 	PMU_PMEVTYPER_EL0(13),
1797 	PMU_PMEVTYPER_EL0(14),
1798 	PMU_PMEVTYPER_EL0(15),
1799 	PMU_PMEVTYPER_EL0(16),
1800 	PMU_PMEVTYPER_EL0(17),
1801 	PMU_PMEVTYPER_EL0(18),
1802 	PMU_PMEVTYPER_EL0(19),
1803 	PMU_PMEVTYPER_EL0(20),
1804 	PMU_PMEVTYPER_EL0(21),
1805 	PMU_PMEVTYPER_EL0(22),
1806 	PMU_PMEVTYPER_EL0(23),
1807 	PMU_PMEVTYPER_EL0(24),
1808 	PMU_PMEVTYPER_EL0(25),
1809 	PMU_PMEVTYPER_EL0(26),
1810 	PMU_PMEVTYPER_EL0(27),
1811 	PMU_PMEVTYPER_EL0(28),
1812 	PMU_PMEVTYPER_EL0(29),
1813 	PMU_PMEVTYPER_EL0(30),
1814 	/*
1815 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1816 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1817 	 */
1818 	{ PMU_SYS_REG(SYS_PMCCFILTR_EL0), .access = access_pmu_evtyper,
1819 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
1820 
1821 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1822 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1823 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1824 };
1825 
1826 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
1827 			struct sys_reg_params *p,
1828 			const struct sys_reg_desc *r)
1829 {
1830 	if (p->is_write) {
1831 		return ignore_write(vcpu, p);
1832 	} else {
1833 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1834 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1835 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1836 
1837 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1838 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1839 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1840 			     | (6 << 16) | (1 << 15) | (el3 << 14) | (el3 << 12));
1841 		return true;
1842 	}
1843 }
1844 
1845 /*
1846  * AArch32 debug register mappings
1847  *
1848  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1849  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1850  *
1851  * None of the other registers share their location, so treat them as
1852  * if they were 64bit.
1853  */
1854 #define DBG_BCR_BVR_WCR_WVR(n)						      \
1855 	/* DBGBVRn */							      \
1856 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
1857 	/* DBGBCRn */							      \
1858 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
1859 	/* DBGWVRn */							      \
1860 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
1861 	/* DBGWCRn */							      \
1862 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1863 
1864 #define DBGBXVR(n)							      \
1865 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
1866 
1867 /*
1868  * Trapped cp14 registers. We generally ignore most of the external
1869  * debug, on the principle that they don't really make sense to a
1870  * guest. Revisit this one day, would this principle change.
1871  */
1872 static const struct sys_reg_desc cp14_regs[] = {
1873 	/* DBGDIDR */
1874 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
1875 	/* DBGDTRRXext */
1876 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1877 
1878 	DBG_BCR_BVR_WCR_WVR(0),
1879 	/* DBGDSCRint */
1880 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1881 	DBG_BCR_BVR_WCR_WVR(1),
1882 	/* DBGDCCINT */
1883 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
1884 	/* DBGDSCRext */
1885 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
1886 	DBG_BCR_BVR_WCR_WVR(2),
1887 	/* DBGDTR[RT]Xint */
1888 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1889 	/* DBGDTR[RT]Xext */
1890 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1891 	DBG_BCR_BVR_WCR_WVR(3),
1892 	DBG_BCR_BVR_WCR_WVR(4),
1893 	DBG_BCR_BVR_WCR_WVR(5),
1894 	/* DBGWFAR */
1895 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1896 	/* DBGOSECCR */
1897 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1898 	DBG_BCR_BVR_WCR_WVR(6),
1899 	/* DBGVCR */
1900 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
1901 	DBG_BCR_BVR_WCR_WVR(7),
1902 	DBG_BCR_BVR_WCR_WVR(8),
1903 	DBG_BCR_BVR_WCR_WVR(9),
1904 	DBG_BCR_BVR_WCR_WVR(10),
1905 	DBG_BCR_BVR_WCR_WVR(11),
1906 	DBG_BCR_BVR_WCR_WVR(12),
1907 	DBG_BCR_BVR_WCR_WVR(13),
1908 	DBG_BCR_BVR_WCR_WVR(14),
1909 	DBG_BCR_BVR_WCR_WVR(15),
1910 
1911 	/* DBGDRAR (32bit) */
1912 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1913 
1914 	DBGBXVR(0),
1915 	/* DBGOSLAR */
1916 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1917 	DBGBXVR(1),
1918 	/* DBGOSLSR */
1919 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1920 	DBGBXVR(2),
1921 	DBGBXVR(3),
1922 	/* DBGOSDLR */
1923 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1924 	DBGBXVR(4),
1925 	/* DBGPRCR */
1926 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1927 	DBGBXVR(5),
1928 	DBGBXVR(6),
1929 	DBGBXVR(7),
1930 	DBGBXVR(8),
1931 	DBGBXVR(9),
1932 	DBGBXVR(10),
1933 	DBGBXVR(11),
1934 	DBGBXVR(12),
1935 	DBGBXVR(13),
1936 	DBGBXVR(14),
1937 	DBGBXVR(15),
1938 
1939 	/* DBGDSAR (32bit) */
1940 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1941 
1942 	/* DBGDEVID2 */
1943 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1944 	/* DBGDEVID1 */
1945 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1946 	/* DBGDEVID */
1947 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1948 	/* DBGCLAIMSET */
1949 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1950 	/* DBGCLAIMCLR */
1951 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1952 	/* DBGAUTHSTATUS */
1953 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1954 };
1955 
1956 /* Trapped cp14 64bit registers */
1957 static const struct sys_reg_desc cp14_64_regs[] = {
1958 	/* DBGDRAR (64bit) */
1959 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1960 
1961 	/* DBGDSAR (64bit) */
1962 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1963 };
1964 
1965 /* Macro to expand the PMEVCNTRn register */
1966 #define PMU_PMEVCNTR(n)							\
1967 	/* PMEVCNTRn */							\
1968 	{ Op1(0), CRn(0b1110),						\
1969 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1970 	  access_pmu_evcntr }
1971 
1972 /* Macro to expand the PMEVTYPERn register */
1973 #define PMU_PMEVTYPER(n)						\
1974 	/* PMEVTYPERn */						\
1975 	{ Op1(0), CRn(0b1110),						\
1976 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1977 	  access_pmu_evtyper }
1978 
1979 /*
1980  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1981  * depending on the way they are accessed (as a 32bit or a 64bit
1982  * register).
1983  */
1984 static const struct sys_reg_desc cp15_regs[] = {
1985 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1986 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
1987 	/* ACTLR */
1988 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
1989 	/* ACTLR2 */
1990 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
1991 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
1992 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
1993 	/* TTBCR */
1994 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
1995 	/* TTBCR2 */
1996 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
1997 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
1998 	/* DFSR */
1999 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
2000 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
2001 	/* ADFSR */
2002 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
2003 	/* AIFSR */
2004 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
2005 	/* DFAR */
2006 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
2007 	/* IFAR */
2008 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
2009 
2010 	/*
2011 	 * DC{C,I,CI}SW operations:
2012 	 */
2013 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
2014 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
2015 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
2016 
2017 	/* PMU */
2018 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
2019 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
2020 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
2021 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
2022 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
2023 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
2024 	{ AA32(LO), Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
2025 	{ AA32(LO), Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
2026 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
2027 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
2028 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
2029 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
2030 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
2031 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
2032 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
2033 	{ AA32(HI), Op1( 0), CRn( 9), CRm(14), Op2( 4), access_pmceid },
2034 	{ AA32(HI), Op1( 0), CRn( 9), CRm(14), Op2( 5), access_pmceid },
2035 	/* PMMIR */
2036 	{ Op1( 0), CRn( 9), CRm(14), Op2( 6), trap_raz_wi },
2037 
2038 	/* PRRR/MAIR0 */
2039 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
2040 	/* NMRR/MAIR1 */
2041 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
2042 	/* AMAIR0 */
2043 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
2044 	/* AMAIR1 */
2045 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
2046 
2047 	/* ICC_SRE */
2048 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2049 
2050 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
2051 
2052 	/* Arch Tmers */
2053 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2054 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2055 
2056 	/* PMEVCNTRn */
2057 	PMU_PMEVCNTR(0),
2058 	PMU_PMEVCNTR(1),
2059 	PMU_PMEVCNTR(2),
2060 	PMU_PMEVCNTR(3),
2061 	PMU_PMEVCNTR(4),
2062 	PMU_PMEVCNTR(5),
2063 	PMU_PMEVCNTR(6),
2064 	PMU_PMEVCNTR(7),
2065 	PMU_PMEVCNTR(8),
2066 	PMU_PMEVCNTR(9),
2067 	PMU_PMEVCNTR(10),
2068 	PMU_PMEVCNTR(11),
2069 	PMU_PMEVCNTR(12),
2070 	PMU_PMEVCNTR(13),
2071 	PMU_PMEVCNTR(14),
2072 	PMU_PMEVCNTR(15),
2073 	PMU_PMEVCNTR(16),
2074 	PMU_PMEVCNTR(17),
2075 	PMU_PMEVCNTR(18),
2076 	PMU_PMEVCNTR(19),
2077 	PMU_PMEVCNTR(20),
2078 	PMU_PMEVCNTR(21),
2079 	PMU_PMEVCNTR(22),
2080 	PMU_PMEVCNTR(23),
2081 	PMU_PMEVCNTR(24),
2082 	PMU_PMEVCNTR(25),
2083 	PMU_PMEVCNTR(26),
2084 	PMU_PMEVCNTR(27),
2085 	PMU_PMEVCNTR(28),
2086 	PMU_PMEVCNTR(29),
2087 	PMU_PMEVCNTR(30),
2088 	/* PMEVTYPERn */
2089 	PMU_PMEVTYPER(0),
2090 	PMU_PMEVTYPER(1),
2091 	PMU_PMEVTYPER(2),
2092 	PMU_PMEVTYPER(3),
2093 	PMU_PMEVTYPER(4),
2094 	PMU_PMEVTYPER(5),
2095 	PMU_PMEVTYPER(6),
2096 	PMU_PMEVTYPER(7),
2097 	PMU_PMEVTYPER(8),
2098 	PMU_PMEVTYPER(9),
2099 	PMU_PMEVTYPER(10),
2100 	PMU_PMEVTYPER(11),
2101 	PMU_PMEVTYPER(12),
2102 	PMU_PMEVTYPER(13),
2103 	PMU_PMEVTYPER(14),
2104 	PMU_PMEVTYPER(15),
2105 	PMU_PMEVTYPER(16),
2106 	PMU_PMEVTYPER(17),
2107 	PMU_PMEVTYPER(18),
2108 	PMU_PMEVTYPER(19),
2109 	PMU_PMEVTYPER(20),
2110 	PMU_PMEVTYPER(21),
2111 	PMU_PMEVTYPER(22),
2112 	PMU_PMEVTYPER(23),
2113 	PMU_PMEVTYPER(24),
2114 	PMU_PMEVTYPER(25),
2115 	PMU_PMEVTYPER(26),
2116 	PMU_PMEVTYPER(27),
2117 	PMU_PMEVTYPER(28),
2118 	PMU_PMEVTYPER(29),
2119 	PMU_PMEVTYPER(30),
2120 	/* PMCCFILTR */
2121 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2122 
2123 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2124 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2125 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
2126 };
2127 
2128 static const struct sys_reg_desc cp15_64_regs[] = {
2129 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
2130 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2131 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2132 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
2133 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2134 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2135 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2136 };
2137 
2138 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2139 			      bool is_32)
2140 {
2141 	unsigned int i;
2142 
2143 	for (i = 0; i < n; i++) {
2144 		if (!is_32 && table[i].reg && !table[i].reset) {
2145 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2146 				table, i);
2147 			return 1;
2148 		}
2149 
2150 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2151 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2152 			return 1;
2153 		}
2154 	}
2155 
2156 	return 0;
2157 }
2158 
2159 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2160 {
2161 	kvm_inject_undefined(vcpu);
2162 	return 1;
2163 }
2164 
2165 static void perform_access(struct kvm_vcpu *vcpu,
2166 			   struct sys_reg_params *params,
2167 			   const struct sys_reg_desc *r)
2168 {
2169 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2170 
2171 	/* Check for regs disabled by runtime config */
2172 	if (sysreg_hidden(vcpu, r)) {
2173 		kvm_inject_undefined(vcpu);
2174 		return;
2175 	}
2176 
2177 	/*
2178 	 * Not having an accessor means that we have configured a trap
2179 	 * that we don't know how to handle. This certainly qualifies
2180 	 * as a gross bug that should be fixed right away.
2181 	 */
2182 	BUG_ON(!r->access);
2183 
2184 	/* Skip instruction if instructed so */
2185 	if (likely(r->access(vcpu, params, r)))
2186 		kvm_incr_pc(vcpu);
2187 }
2188 
2189 /*
2190  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2191  *                call the corresponding trap handler.
2192  *
2193  * @params: pointer to the descriptor of the access
2194  * @table: array of trap descriptors
2195  * @num: size of the trap descriptor array
2196  *
2197  * Return 0 if the access has been handled, and -1 if not.
2198  */
2199 static int emulate_cp(struct kvm_vcpu *vcpu,
2200 		      struct sys_reg_params *params,
2201 		      const struct sys_reg_desc *table,
2202 		      size_t num)
2203 {
2204 	const struct sys_reg_desc *r;
2205 
2206 	if (!table)
2207 		return -1;	/* Not handled */
2208 
2209 	r = find_reg(params, table, num);
2210 
2211 	if (r) {
2212 		perform_access(vcpu, params, r);
2213 		return 0;
2214 	}
2215 
2216 	/* Not handled */
2217 	return -1;
2218 }
2219 
2220 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2221 				struct sys_reg_params *params)
2222 {
2223 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2224 	int cp = -1;
2225 
2226 	switch (esr_ec) {
2227 	case ESR_ELx_EC_CP15_32:
2228 	case ESR_ELx_EC_CP15_64:
2229 		cp = 15;
2230 		break;
2231 	case ESR_ELx_EC_CP14_MR:
2232 	case ESR_ELx_EC_CP14_64:
2233 		cp = 14;
2234 		break;
2235 	default:
2236 		WARN_ON(1);
2237 	}
2238 
2239 	print_sys_reg_msg(params,
2240 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2241 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2242 	kvm_inject_undefined(vcpu);
2243 }
2244 
2245 /**
2246  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2247  * @vcpu: The VCPU pointer
2248  * @run:  The kvm_run struct
2249  */
2250 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2251 			    const struct sys_reg_desc *global,
2252 			    size_t nr_global)
2253 {
2254 	struct sys_reg_params params;
2255 	u32 esr = kvm_vcpu_get_esr(vcpu);
2256 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2257 	int Rt2 = (esr >> 10) & 0x1f;
2258 
2259 	params.CRm = (esr >> 1) & 0xf;
2260 	params.is_write = ((esr & 1) == 0);
2261 
2262 	params.Op0 = 0;
2263 	params.Op1 = (esr >> 16) & 0xf;
2264 	params.Op2 = 0;
2265 	params.CRn = 0;
2266 
2267 	/*
2268 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2269 	 * backends between AArch32 and AArch64, we get away with it.
2270 	 */
2271 	if (params.is_write) {
2272 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2273 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2274 	}
2275 
2276 	/*
2277 	 * If the table contains a handler, handle the
2278 	 * potential register operation in the case of a read and return
2279 	 * with success.
2280 	 */
2281 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2282 		/* Split up the value between registers for the read side */
2283 		if (!params.is_write) {
2284 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2285 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2286 		}
2287 
2288 		return 1;
2289 	}
2290 
2291 	unhandled_cp_access(vcpu, &params);
2292 	return 1;
2293 }
2294 
2295 /**
2296  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2297  * @vcpu: The VCPU pointer
2298  * @run:  The kvm_run struct
2299  */
2300 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2301 			    const struct sys_reg_desc *global,
2302 			    size_t nr_global)
2303 {
2304 	struct sys_reg_params params;
2305 	u32 esr = kvm_vcpu_get_esr(vcpu);
2306 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2307 
2308 	params.CRm = (esr >> 1) & 0xf;
2309 	params.regval = vcpu_get_reg(vcpu, Rt);
2310 	params.is_write = ((esr & 1) == 0);
2311 	params.CRn = (esr >> 10) & 0xf;
2312 	params.Op0 = 0;
2313 	params.Op1 = (esr >> 14) & 0x7;
2314 	params.Op2 = (esr >> 17) & 0x7;
2315 
2316 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2317 		if (!params.is_write)
2318 			vcpu_set_reg(vcpu, Rt, params.regval);
2319 		return 1;
2320 	}
2321 
2322 	unhandled_cp_access(vcpu, &params);
2323 	return 1;
2324 }
2325 
2326 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2327 {
2328 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2329 }
2330 
2331 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2332 {
2333 	return kvm_handle_cp_32(vcpu, cp15_regs, ARRAY_SIZE(cp15_regs));
2334 }
2335 
2336 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2337 {
2338 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2339 }
2340 
2341 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2342 {
2343 	return kvm_handle_cp_32(vcpu, cp14_regs, ARRAY_SIZE(cp14_regs));
2344 }
2345 
2346 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2347 {
2348 	// See ARM DDI 0487E.a, section D12.3.2
2349 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2350 }
2351 
2352 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2353 			   struct sys_reg_params *params)
2354 {
2355 	const struct sys_reg_desc *r;
2356 
2357 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2358 
2359 	if (likely(r)) {
2360 		perform_access(vcpu, params, r);
2361 	} else if (is_imp_def_sys_reg(params)) {
2362 		kvm_inject_undefined(vcpu);
2363 	} else {
2364 		print_sys_reg_msg(params,
2365 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2366 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2367 		kvm_inject_undefined(vcpu);
2368 	}
2369 	return 1;
2370 }
2371 
2372 /**
2373  * kvm_reset_sys_regs - sets system registers to reset value
2374  * @vcpu: The VCPU pointer
2375  *
2376  * This function finds the right table above and sets the registers on the
2377  * virtual CPU struct to their architecturally defined reset values.
2378  */
2379 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2380 {
2381 	unsigned long i;
2382 
2383 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2384 		if (sys_reg_descs[i].reset)
2385 			sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2386 }
2387 
2388 /**
2389  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2390  * @vcpu: The VCPU pointer
2391  */
2392 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2393 {
2394 	struct sys_reg_params params;
2395 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
2396 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2397 	int ret;
2398 
2399 	trace_kvm_handle_sys_reg(esr);
2400 
2401 	params = esr_sys64_to_params(esr);
2402 	params.regval = vcpu_get_reg(vcpu, Rt);
2403 
2404 	ret = emulate_sys_reg(vcpu, &params);
2405 
2406 	if (!params.is_write)
2407 		vcpu_set_reg(vcpu, Rt, params.regval);
2408 	return ret;
2409 }
2410 
2411 /******************************************************************************
2412  * Userspace API
2413  *****************************************************************************/
2414 
2415 static bool index_to_params(u64 id, struct sys_reg_params *params)
2416 {
2417 	switch (id & KVM_REG_SIZE_MASK) {
2418 	case KVM_REG_SIZE_U64:
2419 		/* Any unused index bits means it's not valid. */
2420 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2421 			      | KVM_REG_ARM_COPROC_MASK
2422 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2423 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2424 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2425 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2426 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2427 			return false;
2428 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2429 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2430 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2431 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2432 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2433 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2434 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2435 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2436 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2437 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2438 		return true;
2439 	default:
2440 		return false;
2441 	}
2442 }
2443 
2444 const struct sys_reg_desc *find_reg_by_id(u64 id,
2445 					  struct sys_reg_params *params,
2446 					  const struct sys_reg_desc table[],
2447 					  unsigned int num)
2448 {
2449 	if (!index_to_params(id, params))
2450 		return NULL;
2451 
2452 	return find_reg(params, table, num);
2453 }
2454 
2455 /* Decode an index value, and find the sys_reg_desc entry. */
2456 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2457 						    u64 id)
2458 {
2459 	const struct sys_reg_desc *r;
2460 	struct sys_reg_params params;
2461 
2462 	/* We only do sys_reg for now. */
2463 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2464 		return NULL;
2465 
2466 	if (!index_to_params(id, &params))
2467 		return NULL;
2468 
2469 	r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2470 
2471 	/* Not saved in the sys_reg array and not otherwise accessible? */
2472 	if (r && !(r->reg || r->get_user))
2473 		r = NULL;
2474 
2475 	return r;
2476 }
2477 
2478 /*
2479  * These are the invariant sys_reg registers: we let the guest see the
2480  * host versions of these, so they're part of the guest state.
2481  *
2482  * A future CPU may provide a mechanism to present different values to
2483  * the guest, or a future kvm may trap them.
2484  */
2485 
2486 #define FUNCTION_INVARIANT(reg)						\
2487 	static void get_##reg(struct kvm_vcpu *v,			\
2488 			      const struct sys_reg_desc *r)		\
2489 	{								\
2490 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2491 	}
2492 
2493 FUNCTION_INVARIANT(midr_el1)
2494 FUNCTION_INVARIANT(revidr_el1)
2495 FUNCTION_INVARIANT(clidr_el1)
2496 FUNCTION_INVARIANT(aidr_el1)
2497 
2498 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2499 {
2500 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2501 }
2502 
2503 /* ->val is filled in by kvm_sys_reg_table_init() */
2504 static struct sys_reg_desc invariant_sys_regs[] = {
2505 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2506 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2507 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2508 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2509 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2510 };
2511 
2512 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2513 {
2514 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2515 		return -EFAULT;
2516 	return 0;
2517 }
2518 
2519 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2520 {
2521 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2522 		return -EFAULT;
2523 	return 0;
2524 }
2525 
2526 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2527 {
2528 	struct sys_reg_params params;
2529 	const struct sys_reg_desc *r;
2530 
2531 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2532 			   ARRAY_SIZE(invariant_sys_regs));
2533 	if (!r)
2534 		return -ENOENT;
2535 
2536 	return reg_to_user(uaddr, &r->val, id);
2537 }
2538 
2539 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2540 {
2541 	struct sys_reg_params params;
2542 	const struct sys_reg_desc *r;
2543 	int err;
2544 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2545 
2546 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2547 			   ARRAY_SIZE(invariant_sys_regs));
2548 	if (!r)
2549 		return -ENOENT;
2550 
2551 	err = reg_from_user(&val, uaddr, id);
2552 	if (err)
2553 		return err;
2554 
2555 	/* This is what we mean by invariant: you can't change it. */
2556 	if (r->val != val)
2557 		return -EINVAL;
2558 
2559 	return 0;
2560 }
2561 
2562 static bool is_valid_cache(u32 val)
2563 {
2564 	u32 level, ctype;
2565 
2566 	if (val >= CSSELR_MAX)
2567 		return false;
2568 
2569 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2570 	level = (val >> 1);
2571 	ctype = (cache_levels >> (level * 3)) & 7;
2572 
2573 	switch (ctype) {
2574 	case 0: /* No cache */
2575 		return false;
2576 	case 1: /* Instruction cache only */
2577 		return (val & 1);
2578 	case 2: /* Data cache only */
2579 	case 4: /* Unified cache */
2580 		return !(val & 1);
2581 	case 3: /* Separate instruction and data caches */
2582 		return true;
2583 	default: /* Reserved: we can't know instruction or data. */
2584 		return false;
2585 	}
2586 }
2587 
2588 static int demux_c15_get(u64 id, void __user *uaddr)
2589 {
2590 	u32 val;
2591 	u32 __user *uval = uaddr;
2592 
2593 	/* Fail if we have unknown bits set. */
2594 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2595 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2596 		return -ENOENT;
2597 
2598 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2599 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2600 		if (KVM_REG_SIZE(id) != 4)
2601 			return -ENOENT;
2602 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2603 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2604 		if (!is_valid_cache(val))
2605 			return -ENOENT;
2606 
2607 		return put_user(get_ccsidr(val), uval);
2608 	default:
2609 		return -ENOENT;
2610 	}
2611 }
2612 
2613 static int demux_c15_set(u64 id, void __user *uaddr)
2614 {
2615 	u32 val, newval;
2616 	u32 __user *uval = uaddr;
2617 
2618 	/* Fail if we have unknown bits set. */
2619 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2620 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2621 		return -ENOENT;
2622 
2623 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2624 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2625 		if (KVM_REG_SIZE(id) != 4)
2626 			return -ENOENT;
2627 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2628 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2629 		if (!is_valid_cache(val))
2630 			return -ENOENT;
2631 
2632 		if (get_user(newval, uval))
2633 			return -EFAULT;
2634 
2635 		/* This is also invariant: you can't change it. */
2636 		if (newval != get_ccsidr(val))
2637 			return -EINVAL;
2638 		return 0;
2639 	default:
2640 		return -ENOENT;
2641 	}
2642 }
2643 
2644 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2645 {
2646 	const struct sys_reg_desc *r;
2647 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2648 
2649 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2650 		return demux_c15_get(reg->id, uaddr);
2651 
2652 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2653 		return -ENOENT;
2654 
2655 	r = index_to_sys_reg_desc(vcpu, reg->id);
2656 	if (!r)
2657 		return get_invariant_sys_reg(reg->id, uaddr);
2658 
2659 	/* Check for regs disabled by runtime config */
2660 	if (sysreg_hidden(vcpu, r))
2661 		return -ENOENT;
2662 
2663 	if (r->get_user)
2664 		return (r->get_user)(vcpu, r, reg, uaddr);
2665 
2666 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2667 }
2668 
2669 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2670 {
2671 	const struct sys_reg_desc *r;
2672 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2673 
2674 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2675 		return demux_c15_set(reg->id, uaddr);
2676 
2677 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2678 		return -ENOENT;
2679 
2680 	r = index_to_sys_reg_desc(vcpu, reg->id);
2681 	if (!r)
2682 		return set_invariant_sys_reg(reg->id, uaddr);
2683 
2684 	/* Check for regs disabled by runtime config */
2685 	if (sysreg_hidden(vcpu, r))
2686 		return -ENOENT;
2687 
2688 	if (r->set_user)
2689 		return (r->set_user)(vcpu, r, reg, uaddr);
2690 
2691 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2692 }
2693 
2694 static unsigned int num_demux_regs(void)
2695 {
2696 	unsigned int i, count = 0;
2697 
2698 	for (i = 0; i < CSSELR_MAX; i++)
2699 		if (is_valid_cache(i))
2700 			count++;
2701 
2702 	return count;
2703 }
2704 
2705 static int write_demux_regids(u64 __user *uindices)
2706 {
2707 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2708 	unsigned int i;
2709 
2710 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2711 	for (i = 0; i < CSSELR_MAX; i++) {
2712 		if (!is_valid_cache(i))
2713 			continue;
2714 		if (put_user(val | i, uindices))
2715 			return -EFAULT;
2716 		uindices++;
2717 	}
2718 	return 0;
2719 }
2720 
2721 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2722 {
2723 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2724 		KVM_REG_ARM64_SYSREG |
2725 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2726 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2727 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2728 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2729 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2730 }
2731 
2732 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2733 {
2734 	if (!*uind)
2735 		return true;
2736 
2737 	if (put_user(sys_reg_to_index(reg), *uind))
2738 		return false;
2739 
2740 	(*uind)++;
2741 	return true;
2742 }
2743 
2744 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2745 			    const struct sys_reg_desc *rd,
2746 			    u64 __user **uind,
2747 			    unsigned int *total)
2748 {
2749 	/*
2750 	 * Ignore registers we trap but don't save,
2751 	 * and for which no custom user accessor is provided.
2752 	 */
2753 	if (!(rd->reg || rd->get_user))
2754 		return 0;
2755 
2756 	if (sysreg_hidden(vcpu, rd))
2757 		return 0;
2758 
2759 	if (!copy_reg_to_user(rd, uind))
2760 		return -EFAULT;
2761 
2762 	(*total)++;
2763 	return 0;
2764 }
2765 
2766 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2767 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2768 {
2769 	const struct sys_reg_desc *i2, *end2;
2770 	unsigned int total = 0;
2771 	int err;
2772 
2773 	i2 = sys_reg_descs;
2774 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2775 
2776 	while (i2 != end2) {
2777 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2778 		if (err)
2779 			return err;
2780 	}
2781 	return total;
2782 }
2783 
2784 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2785 {
2786 	return ARRAY_SIZE(invariant_sys_regs)
2787 		+ num_demux_regs()
2788 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2789 }
2790 
2791 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2792 {
2793 	unsigned int i;
2794 	int err;
2795 
2796 	/* Then give them all the invariant registers' indices. */
2797 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2798 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2799 			return -EFAULT;
2800 		uindices++;
2801 	}
2802 
2803 	err = walk_sys_regs(vcpu, uindices);
2804 	if (err < 0)
2805 		return err;
2806 	uindices += err;
2807 
2808 	return write_demux_regids(uindices);
2809 }
2810 
2811 void kvm_sys_reg_table_init(void)
2812 {
2813 	unsigned int i;
2814 	struct sys_reg_desc clidr;
2815 
2816 	/* Make sure tables are unique and in order. */
2817 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2818 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2819 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2820 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2821 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2822 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2823 
2824 	/* We abuse the reset function to overwrite the table itself. */
2825 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2826 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2827 
2828 	/*
2829 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2830 	 *
2831 	 *   If software reads the Cache Type fields from Ctype1
2832 	 *   upwards, once it has seen a value of 0b000, no caches
2833 	 *   exist at further-out levels of the hierarchy. So, for
2834 	 *   example, if Ctype3 is the first Cache Type field with a
2835 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2836 	 *   ignored.
2837 	 */
2838 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2839 	cache_levels = clidr.val;
2840 	for (i = 0; i < 7; i++)
2841 		if (((cache_levels >> (i*3)) & 7) == 0)
2842 			break;
2843 	/* Clear all higher bits. */
2844 	cache_levels &= (1 << (i*3))-1;
2845 }
2846