xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 36de991e)
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;
1068 
1069 	if (raz)
1070 		return 0;
1071 
1072 	val = read_sanitised_ftr_reg(id);
1073 
1074 	switch (id) {
1075 	case SYS_ID_AA64PFR0_EL1:
1076 		if (!vcpu_has_sve(vcpu))
1077 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_SVE);
1078 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_AMU);
1079 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2);
1080 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV2), (u64)vcpu->kvm->arch.pfr0_csv2);
1081 		val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3);
1082 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_CSV3), (u64)vcpu->kvm->arch.pfr0_csv3);
1083 		if (irqchip_in_kernel(vcpu->kvm) &&
1084 		    vcpu->kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3) {
1085 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR0_GIC);
1086 			val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_GIC), 1);
1087 		}
1088 		break;
1089 	case SYS_ID_AA64PFR1_EL1:
1090 		if (!kvm_has_mte(vcpu->kvm))
1091 			val &= ~ARM64_FEATURE_MASK(ID_AA64PFR1_MTE);
1092 		break;
1093 	case SYS_ID_AA64ISAR1_EL1:
1094 		if (!vcpu_has_ptrauth(vcpu))
1095 			val &= ~(ARM64_FEATURE_MASK(ID_AA64ISAR1_APA) |
1096 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_API) |
1097 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_GPA) |
1098 				 ARM64_FEATURE_MASK(ID_AA64ISAR1_GPI));
1099 		break;
1100 	case SYS_ID_AA64DFR0_EL1:
1101 		/* Limit debug to ARMv8.0 */
1102 		val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER);
1103 		val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64DFR0_DEBUGVER), 6);
1104 		/* Limit guests to PMUv3 for ARMv8.4 */
1105 		val = cpuid_feature_cap_perfmon_field(val,
1106 						      ID_AA64DFR0_PMUVER_SHIFT,
1107 						      kvm_vcpu_has_pmu(vcpu) ? ID_AA64DFR0_PMUVER_8_4 : 0);
1108 		/* Hide SPE from guests */
1109 		val &= ~ARM64_FEATURE_MASK(ID_AA64DFR0_PMSVER);
1110 		break;
1111 	case SYS_ID_DFR0_EL1:
1112 		/* Limit guests to PMUv3 for ARMv8.4 */
1113 		val = cpuid_feature_cap_perfmon_field(val,
1114 						      ID_DFR0_PERFMON_SHIFT,
1115 						      kvm_vcpu_has_pmu(vcpu) ? ID_DFR0_PERFMON_8_4 : 0);
1116 		break;
1117 	}
1118 
1119 	return val;
1120 }
1121 
1122 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1123 				  const struct sys_reg_desc *r)
1124 {
1125 	u32 id = reg_to_encoding(r);
1126 
1127 	switch (id) {
1128 	case SYS_ID_AA64ZFR0_EL1:
1129 		if (!vcpu_has_sve(vcpu))
1130 			return REG_RAZ;
1131 		break;
1132 	}
1133 
1134 	return 0;
1135 }
1136 
1137 /* cpufeature ID register access trap handlers */
1138 
1139 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1140 			    struct sys_reg_params *p,
1141 			    const struct sys_reg_desc *r,
1142 			    bool raz)
1143 {
1144 	if (p->is_write)
1145 		return write_to_read_only(vcpu, p, r);
1146 
1147 	p->regval = read_id_reg(vcpu, r, raz);
1148 	return true;
1149 }
1150 
1151 static bool access_id_reg(struct kvm_vcpu *vcpu,
1152 			  struct sys_reg_params *p,
1153 			  const struct sys_reg_desc *r)
1154 {
1155 	bool raz = sysreg_visible_as_raz(vcpu, r);
1156 
1157 	return __access_id_reg(vcpu, p, r, raz);
1158 }
1159 
1160 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1161 			      struct sys_reg_params *p,
1162 			      const struct sys_reg_desc *r)
1163 {
1164 	return __access_id_reg(vcpu, p, r, true);
1165 }
1166 
1167 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1168 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1169 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1170 
1171 /* Visibility overrides for SVE-specific control registers */
1172 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1173 				   const struct sys_reg_desc *rd)
1174 {
1175 	if (vcpu_has_sve(vcpu))
1176 		return 0;
1177 
1178 	return REG_HIDDEN;
1179 }
1180 
1181 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
1182 			       const struct sys_reg_desc *rd,
1183 			       const struct kvm_one_reg *reg, void __user *uaddr)
1184 {
1185 	const u64 id = sys_reg_to_index(rd);
1186 	u8 csv2, csv3;
1187 	int err;
1188 	u64 val;
1189 
1190 	err = reg_from_user(&val, uaddr, id);
1191 	if (err)
1192 		return err;
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, void __user *uaddr,
1232 			bool raz)
1233 {
1234 	const u64 id = sys_reg_to_index(rd);
1235 	const u64 val = read_id_reg(vcpu, rd, raz);
1236 
1237 	return reg_to_user(uaddr, &val, id);
1238 }
1239 
1240 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1241 			const struct sys_reg_desc *rd, void __user *uaddr,
1242 			bool raz)
1243 {
1244 	const u64 id = sys_reg_to_index(rd);
1245 	int err;
1246 	u64 val;
1247 
1248 	err = reg_from_user(&val, uaddr, id);
1249 	if (err)
1250 		return err;
1251 
1252 	/* This is what we mean by invariant: you can't change it. */
1253 	if (val != read_id_reg(vcpu, rd, raz))
1254 		return -EINVAL;
1255 
1256 	return 0;
1257 }
1258 
1259 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1260 		      const struct kvm_one_reg *reg, void __user *uaddr)
1261 {
1262 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1263 
1264 	return __get_id_reg(vcpu, rd, uaddr, raz);
1265 }
1266 
1267 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1268 		      const struct kvm_one_reg *reg, void __user *uaddr)
1269 {
1270 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1271 
1272 	return __set_id_reg(vcpu, rd, uaddr, raz);
1273 }
1274 
1275 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1276 			  const struct kvm_one_reg *reg, void __user *uaddr)
1277 {
1278 	return __set_id_reg(vcpu, rd, uaddr, true);
1279 }
1280 
1281 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1282 		       const struct kvm_one_reg *reg, void __user *uaddr)
1283 {
1284 	const u64 id = sys_reg_to_index(rd);
1285 	const u64 val = 0;
1286 
1287 	return reg_to_user(uaddr, &val, id);
1288 }
1289 
1290 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1291 		      const struct kvm_one_reg *reg, void __user *uaddr)
1292 {
1293 	int err;
1294 	u64 val;
1295 
1296 	/* Perform the access even if we are going to ignore the value */
1297 	err = reg_from_user(&val, uaddr, sys_reg_to_index(rd));
1298 	if (err)
1299 		return err;
1300 
1301 	return 0;
1302 }
1303 
1304 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1305 		       const struct sys_reg_desc *r)
1306 {
1307 	if (p->is_write)
1308 		return write_to_read_only(vcpu, p, r);
1309 
1310 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1311 	return true;
1312 }
1313 
1314 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1315 			 const struct sys_reg_desc *r)
1316 {
1317 	if (p->is_write)
1318 		return write_to_read_only(vcpu, p, r);
1319 
1320 	p->regval = read_sysreg(clidr_el1);
1321 	return true;
1322 }
1323 
1324 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1325 			  const struct sys_reg_desc *r)
1326 {
1327 	int reg = r->reg;
1328 
1329 	if (p->is_write)
1330 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1331 	else
1332 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1333 	return true;
1334 }
1335 
1336 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1337 			  const struct sys_reg_desc *r)
1338 {
1339 	u32 csselr;
1340 
1341 	if (p->is_write)
1342 		return write_to_read_only(vcpu, p, r);
1343 
1344 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1345 	p->regval = get_ccsidr(csselr);
1346 
1347 	/*
1348 	 * Guests should not be doing cache operations by set/way at all, and
1349 	 * for this reason, we trap them and attempt to infer the intent, so
1350 	 * that we can flush the entire guest's address space at the appropriate
1351 	 * time.
1352 	 * To prevent this trapping from causing performance problems, let's
1353 	 * expose the geometry of all data and unified caches (which are
1354 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1355 	 * [If guests should attempt to infer aliasing properties from the
1356 	 * geometry (which is not permitted by the architecture), they would
1357 	 * only do so for virtually indexed caches.]
1358 	 */
1359 	if (!(csselr & 1)) // data or unified cache
1360 		p->regval &= ~GENMASK(27, 3);
1361 	return true;
1362 }
1363 
1364 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
1365 				   const struct sys_reg_desc *rd)
1366 {
1367 	if (kvm_has_mte(vcpu->kvm))
1368 		return 0;
1369 
1370 	return REG_HIDDEN;
1371 }
1372 
1373 #define MTE_REG(name) {				\
1374 	SYS_DESC(SYS_##name),			\
1375 	.access = undef_access,			\
1376 	.reset = reset_unknown,			\
1377 	.reg = name,				\
1378 	.visibility = mte_visibility,		\
1379 }
1380 
1381 /* sys_reg_desc initialiser for known cpufeature ID registers */
1382 #define ID_SANITISED(name) {			\
1383 	SYS_DESC(SYS_##name),			\
1384 	.access	= access_id_reg,		\
1385 	.get_user = get_id_reg,			\
1386 	.set_user = set_id_reg,			\
1387 	.visibility = id_visibility,		\
1388 }
1389 
1390 /*
1391  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1392  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1393  * (1 <= crm < 8, 0 <= Op2 < 8).
1394  */
1395 #define ID_UNALLOCATED(crm, op2) {			\
1396 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1397 	.access = access_raz_id_reg,			\
1398 	.get_user = get_raz_reg,			\
1399 	.set_user = set_raz_id_reg,			\
1400 }
1401 
1402 /*
1403  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1404  * For now, these are exposed just like unallocated ID regs: they appear
1405  * RAZ for the guest.
1406  */
1407 #define ID_HIDDEN(name) {			\
1408 	SYS_DESC(SYS_##name),			\
1409 	.access = access_raz_id_reg,		\
1410 	.get_user = get_raz_reg,		\
1411 	.set_user = set_raz_id_reg,		\
1412 }
1413 
1414 /*
1415  * Architected system registers.
1416  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1417  *
1418  * Debug handling: We do trap most, if not all debug related system
1419  * registers. The implementation is good enough to ensure that a guest
1420  * can use these with minimal performance degradation. The drawback is
1421  * that we don't implement any of the external debug, none of the
1422  * OSlock protocol. This should be revisited if we ever encounter a
1423  * more demanding guest...
1424  */
1425 static const struct sys_reg_desc sys_reg_descs[] = {
1426 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1427 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1428 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1429 
1430 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1431 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1432 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1433 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1434 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1435 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1436 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1437 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1438 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1439 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1440 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1441 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1442 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1443 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1444 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1445 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1446 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1447 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1448 
1449 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1450 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1451 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1452 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1453 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1454 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1455 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1456 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1457 
1458 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1459 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1460 	// DBGDTR[TR]X_EL0 share the same encoding
1461 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1462 
1463 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1464 
1465 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1466 
1467 	/*
1468 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1469 	 * entries in arm64_ftr_regs[].
1470 	 */
1471 
1472 	/* AArch64 mappings of the AArch32 ID registers */
1473 	/* CRm=1 */
1474 	ID_SANITISED(ID_PFR0_EL1),
1475 	ID_SANITISED(ID_PFR1_EL1),
1476 	ID_SANITISED(ID_DFR0_EL1),
1477 	ID_HIDDEN(ID_AFR0_EL1),
1478 	ID_SANITISED(ID_MMFR0_EL1),
1479 	ID_SANITISED(ID_MMFR1_EL1),
1480 	ID_SANITISED(ID_MMFR2_EL1),
1481 	ID_SANITISED(ID_MMFR3_EL1),
1482 
1483 	/* CRm=2 */
1484 	ID_SANITISED(ID_ISAR0_EL1),
1485 	ID_SANITISED(ID_ISAR1_EL1),
1486 	ID_SANITISED(ID_ISAR2_EL1),
1487 	ID_SANITISED(ID_ISAR3_EL1),
1488 	ID_SANITISED(ID_ISAR4_EL1),
1489 	ID_SANITISED(ID_ISAR5_EL1),
1490 	ID_SANITISED(ID_MMFR4_EL1),
1491 	ID_SANITISED(ID_ISAR6_EL1),
1492 
1493 	/* CRm=3 */
1494 	ID_SANITISED(MVFR0_EL1),
1495 	ID_SANITISED(MVFR1_EL1),
1496 	ID_SANITISED(MVFR2_EL1),
1497 	ID_UNALLOCATED(3,3),
1498 	ID_SANITISED(ID_PFR2_EL1),
1499 	ID_HIDDEN(ID_DFR1_EL1),
1500 	ID_SANITISED(ID_MMFR5_EL1),
1501 	ID_UNALLOCATED(3,7),
1502 
1503 	/* AArch64 ID registers */
1504 	/* CRm=4 */
1505 	{ SYS_DESC(SYS_ID_AA64PFR0_EL1), .access = access_id_reg,
1506 	  .get_user = get_id_reg, .set_user = set_id_aa64pfr0_el1, },
1507 	ID_SANITISED(ID_AA64PFR1_EL1),
1508 	ID_UNALLOCATED(4,2),
1509 	ID_UNALLOCATED(4,3),
1510 	ID_SANITISED(ID_AA64ZFR0_EL1),
1511 	ID_UNALLOCATED(4,5),
1512 	ID_UNALLOCATED(4,6),
1513 	ID_UNALLOCATED(4,7),
1514 
1515 	/* CRm=5 */
1516 	ID_SANITISED(ID_AA64DFR0_EL1),
1517 	ID_SANITISED(ID_AA64DFR1_EL1),
1518 	ID_UNALLOCATED(5,2),
1519 	ID_UNALLOCATED(5,3),
1520 	ID_HIDDEN(ID_AA64AFR0_EL1),
1521 	ID_HIDDEN(ID_AA64AFR1_EL1),
1522 	ID_UNALLOCATED(5,6),
1523 	ID_UNALLOCATED(5,7),
1524 
1525 	/* CRm=6 */
1526 	ID_SANITISED(ID_AA64ISAR0_EL1),
1527 	ID_SANITISED(ID_AA64ISAR1_EL1),
1528 	ID_UNALLOCATED(6,2),
1529 	ID_UNALLOCATED(6,3),
1530 	ID_UNALLOCATED(6,4),
1531 	ID_UNALLOCATED(6,5),
1532 	ID_UNALLOCATED(6,6),
1533 	ID_UNALLOCATED(6,7),
1534 
1535 	/* CRm=7 */
1536 	ID_SANITISED(ID_AA64MMFR0_EL1),
1537 	ID_SANITISED(ID_AA64MMFR1_EL1),
1538 	ID_SANITISED(ID_AA64MMFR2_EL1),
1539 	ID_UNALLOCATED(7,3),
1540 	ID_UNALLOCATED(7,4),
1541 	ID_UNALLOCATED(7,5),
1542 	ID_UNALLOCATED(7,6),
1543 	ID_UNALLOCATED(7,7),
1544 
1545 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1546 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1547 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1548 
1549 	MTE_REG(RGSR_EL1),
1550 	MTE_REG(GCR_EL1),
1551 
1552 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1553 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
1554 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1555 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1556 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1557 
1558 	PTRAUTH_KEY(APIA),
1559 	PTRAUTH_KEY(APIB),
1560 	PTRAUTH_KEY(APDA),
1561 	PTRAUTH_KEY(APDB),
1562 	PTRAUTH_KEY(APGA),
1563 
1564 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1565 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1566 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1567 
1568 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1569 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1570 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1571 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1572 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1573 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1574 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1575 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1576 
1577 	MTE_REG(TFSR_EL1),
1578 	MTE_REG(TFSRE0_EL1),
1579 
1580 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1581 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1582 
1583 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
1584 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
1585 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
1586 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
1587 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
1588 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
1589 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
1590 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
1591 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
1592 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
1593 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
1594 	/* PMBIDR_EL1 is not trapped */
1595 
1596 	{ PMU_SYS_REG(SYS_PMINTENSET_EL1),
1597 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1598 	{ PMU_SYS_REG(SYS_PMINTENCLR_EL1),
1599 	  .access = access_pminten, .reg = PMINTENSET_EL1 },
1600 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
1601 
1602 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1603 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1604 
1605 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1606 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1607 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1608 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1609 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1610 
1611 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1612 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1613 
1614 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1615 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1616 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1617 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1618 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1619 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1620 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1621 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1622 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1623 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1624 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1625 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1626 
1627 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1628 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1629 
1630 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
1631 
1632 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1633 
1634 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1635 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1636 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1637 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1638 
1639 	{ PMU_SYS_REG(SYS_PMCR_EL0), .access = access_pmcr,
1640 	  .reset = reset_pmcr, .reg = PMCR_EL0 },
1641 	{ PMU_SYS_REG(SYS_PMCNTENSET_EL0),
1642 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1643 	{ PMU_SYS_REG(SYS_PMCNTENCLR_EL0),
1644 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0 },
1645 	{ PMU_SYS_REG(SYS_PMOVSCLR_EL0),
1646 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1647 	/*
1648 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
1649 	 * previously (and pointlessly) advertised in the past...
1650 	 */
1651 	{ PMU_SYS_REG(SYS_PMSWINC_EL0),
1652 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
1653 	  .access = access_pmswinc, .reset = NULL },
1654 	{ PMU_SYS_REG(SYS_PMSELR_EL0),
1655 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
1656 	{ PMU_SYS_REG(SYS_PMCEID0_EL0),
1657 	  .access = access_pmceid, .reset = NULL },
1658 	{ PMU_SYS_REG(SYS_PMCEID1_EL0),
1659 	  .access = access_pmceid, .reset = NULL },
1660 	{ PMU_SYS_REG(SYS_PMCCNTR_EL0),
1661 	  .access = access_pmu_evcntr, .reset = reset_unknown, .reg = PMCCNTR_EL0 },
1662 	{ PMU_SYS_REG(SYS_PMXEVTYPER_EL0),
1663 	  .access = access_pmu_evtyper, .reset = NULL },
1664 	{ PMU_SYS_REG(SYS_PMXEVCNTR_EL0),
1665 	  .access = access_pmu_evcntr, .reset = NULL },
1666 	/*
1667 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1668 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1669 	 */
1670 	{ PMU_SYS_REG(SYS_PMUSERENR_EL0), .access = access_pmuserenr,
1671 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
1672 	{ PMU_SYS_REG(SYS_PMOVSSET_EL0),
1673 	  .access = access_pmovs, .reg = PMOVSSET_EL0 },
1674 
1675 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1676 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1677 
1678 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
1679 
1680 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
1681 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
1682 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
1683 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
1684 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
1685 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
1686 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
1687 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
1688 	AMU_AMEVCNTR0_EL0(0),
1689 	AMU_AMEVCNTR0_EL0(1),
1690 	AMU_AMEVCNTR0_EL0(2),
1691 	AMU_AMEVCNTR0_EL0(3),
1692 	AMU_AMEVCNTR0_EL0(4),
1693 	AMU_AMEVCNTR0_EL0(5),
1694 	AMU_AMEVCNTR0_EL0(6),
1695 	AMU_AMEVCNTR0_EL0(7),
1696 	AMU_AMEVCNTR0_EL0(8),
1697 	AMU_AMEVCNTR0_EL0(9),
1698 	AMU_AMEVCNTR0_EL0(10),
1699 	AMU_AMEVCNTR0_EL0(11),
1700 	AMU_AMEVCNTR0_EL0(12),
1701 	AMU_AMEVCNTR0_EL0(13),
1702 	AMU_AMEVCNTR0_EL0(14),
1703 	AMU_AMEVCNTR0_EL0(15),
1704 	AMU_AMEVTYPER0_EL0(0),
1705 	AMU_AMEVTYPER0_EL0(1),
1706 	AMU_AMEVTYPER0_EL0(2),
1707 	AMU_AMEVTYPER0_EL0(3),
1708 	AMU_AMEVTYPER0_EL0(4),
1709 	AMU_AMEVTYPER0_EL0(5),
1710 	AMU_AMEVTYPER0_EL0(6),
1711 	AMU_AMEVTYPER0_EL0(7),
1712 	AMU_AMEVTYPER0_EL0(8),
1713 	AMU_AMEVTYPER0_EL0(9),
1714 	AMU_AMEVTYPER0_EL0(10),
1715 	AMU_AMEVTYPER0_EL0(11),
1716 	AMU_AMEVTYPER0_EL0(12),
1717 	AMU_AMEVTYPER0_EL0(13),
1718 	AMU_AMEVTYPER0_EL0(14),
1719 	AMU_AMEVTYPER0_EL0(15),
1720 	AMU_AMEVCNTR1_EL0(0),
1721 	AMU_AMEVCNTR1_EL0(1),
1722 	AMU_AMEVCNTR1_EL0(2),
1723 	AMU_AMEVCNTR1_EL0(3),
1724 	AMU_AMEVCNTR1_EL0(4),
1725 	AMU_AMEVCNTR1_EL0(5),
1726 	AMU_AMEVCNTR1_EL0(6),
1727 	AMU_AMEVCNTR1_EL0(7),
1728 	AMU_AMEVCNTR1_EL0(8),
1729 	AMU_AMEVCNTR1_EL0(9),
1730 	AMU_AMEVCNTR1_EL0(10),
1731 	AMU_AMEVCNTR1_EL0(11),
1732 	AMU_AMEVCNTR1_EL0(12),
1733 	AMU_AMEVCNTR1_EL0(13),
1734 	AMU_AMEVCNTR1_EL0(14),
1735 	AMU_AMEVCNTR1_EL0(15),
1736 	AMU_AMEVTYPER1_EL0(0),
1737 	AMU_AMEVTYPER1_EL0(1),
1738 	AMU_AMEVTYPER1_EL0(2),
1739 	AMU_AMEVTYPER1_EL0(3),
1740 	AMU_AMEVTYPER1_EL0(4),
1741 	AMU_AMEVTYPER1_EL0(5),
1742 	AMU_AMEVTYPER1_EL0(6),
1743 	AMU_AMEVTYPER1_EL0(7),
1744 	AMU_AMEVTYPER1_EL0(8),
1745 	AMU_AMEVTYPER1_EL0(9),
1746 	AMU_AMEVTYPER1_EL0(10),
1747 	AMU_AMEVTYPER1_EL0(11),
1748 	AMU_AMEVTYPER1_EL0(12),
1749 	AMU_AMEVTYPER1_EL0(13),
1750 	AMU_AMEVTYPER1_EL0(14),
1751 	AMU_AMEVTYPER1_EL0(15),
1752 
1753 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1754 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1755 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1756 
1757 	/* PMEVCNTRn_EL0 */
1758 	PMU_PMEVCNTR_EL0(0),
1759 	PMU_PMEVCNTR_EL0(1),
1760 	PMU_PMEVCNTR_EL0(2),
1761 	PMU_PMEVCNTR_EL0(3),
1762 	PMU_PMEVCNTR_EL0(4),
1763 	PMU_PMEVCNTR_EL0(5),
1764 	PMU_PMEVCNTR_EL0(6),
1765 	PMU_PMEVCNTR_EL0(7),
1766 	PMU_PMEVCNTR_EL0(8),
1767 	PMU_PMEVCNTR_EL0(9),
1768 	PMU_PMEVCNTR_EL0(10),
1769 	PMU_PMEVCNTR_EL0(11),
1770 	PMU_PMEVCNTR_EL0(12),
1771 	PMU_PMEVCNTR_EL0(13),
1772 	PMU_PMEVCNTR_EL0(14),
1773 	PMU_PMEVCNTR_EL0(15),
1774 	PMU_PMEVCNTR_EL0(16),
1775 	PMU_PMEVCNTR_EL0(17),
1776 	PMU_PMEVCNTR_EL0(18),
1777 	PMU_PMEVCNTR_EL0(19),
1778 	PMU_PMEVCNTR_EL0(20),
1779 	PMU_PMEVCNTR_EL0(21),
1780 	PMU_PMEVCNTR_EL0(22),
1781 	PMU_PMEVCNTR_EL0(23),
1782 	PMU_PMEVCNTR_EL0(24),
1783 	PMU_PMEVCNTR_EL0(25),
1784 	PMU_PMEVCNTR_EL0(26),
1785 	PMU_PMEVCNTR_EL0(27),
1786 	PMU_PMEVCNTR_EL0(28),
1787 	PMU_PMEVCNTR_EL0(29),
1788 	PMU_PMEVCNTR_EL0(30),
1789 	/* PMEVTYPERn_EL0 */
1790 	PMU_PMEVTYPER_EL0(0),
1791 	PMU_PMEVTYPER_EL0(1),
1792 	PMU_PMEVTYPER_EL0(2),
1793 	PMU_PMEVTYPER_EL0(3),
1794 	PMU_PMEVTYPER_EL0(4),
1795 	PMU_PMEVTYPER_EL0(5),
1796 	PMU_PMEVTYPER_EL0(6),
1797 	PMU_PMEVTYPER_EL0(7),
1798 	PMU_PMEVTYPER_EL0(8),
1799 	PMU_PMEVTYPER_EL0(9),
1800 	PMU_PMEVTYPER_EL0(10),
1801 	PMU_PMEVTYPER_EL0(11),
1802 	PMU_PMEVTYPER_EL0(12),
1803 	PMU_PMEVTYPER_EL0(13),
1804 	PMU_PMEVTYPER_EL0(14),
1805 	PMU_PMEVTYPER_EL0(15),
1806 	PMU_PMEVTYPER_EL0(16),
1807 	PMU_PMEVTYPER_EL0(17),
1808 	PMU_PMEVTYPER_EL0(18),
1809 	PMU_PMEVTYPER_EL0(19),
1810 	PMU_PMEVTYPER_EL0(20),
1811 	PMU_PMEVTYPER_EL0(21),
1812 	PMU_PMEVTYPER_EL0(22),
1813 	PMU_PMEVTYPER_EL0(23),
1814 	PMU_PMEVTYPER_EL0(24),
1815 	PMU_PMEVTYPER_EL0(25),
1816 	PMU_PMEVTYPER_EL0(26),
1817 	PMU_PMEVTYPER_EL0(27),
1818 	PMU_PMEVTYPER_EL0(28),
1819 	PMU_PMEVTYPER_EL0(29),
1820 	PMU_PMEVTYPER_EL0(30),
1821 	/*
1822 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1823 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1824 	 */
1825 	{ PMU_SYS_REG(SYS_PMCCFILTR_EL0), .access = access_pmu_evtyper,
1826 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
1827 
1828 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1829 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1830 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1831 };
1832 
1833 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
1834 			struct sys_reg_params *p,
1835 			const struct sys_reg_desc *r)
1836 {
1837 	if (p->is_write) {
1838 		return ignore_write(vcpu, p);
1839 	} else {
1840 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1841 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1842 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1843 
1844 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1845 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1846 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1847 			     | (6 << 16) | (1 << 15) | (el3 << 14) | (el3 << 12));
1848 		return true;
1849 	}
1850 }
1851 
1852 /*
1853  * AArch32 debug register mappings
1854  *
1855  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1856  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1857  *
1858  * None of the other registers share their location, so treat them as
1859  * if they were 64bit.
1860  */
1861 #define DBG_BCR_BVR_WCR_WVR(n)						      \
1862 	/* DBGBVRn */							      \
1863 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, \
1864 	/* DBGBCRn */							      \
1865 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	      \
1866 	/* DBGWVRn */							      \
1867 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	      \
1868 	/* DBGWCRn */							      \
1869 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1870 
1871 #define DBGBXVR(n)							      \
1872 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_bvr, NULL, n }
1873 
1874 /*
1875  * Trapped cp14 registers. We generally ignore most of the external
1876  * debug, on the principle that they don't really make sense to a
1877  * guest. Revisit this one day, would this principle change.
1878  */
1879 static const struct sys_reg_desc cp14_regs[] = {
1880 	/* DBGDIDR */
1881 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
1882 	/* DBGDTRRXext */
1883 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1884 
1885 	DBG_BCR_BVR_WCR_WVR(0),
1886 	/* DBGDSCRint */
1887 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1888 	DBG_BCR_BVR_WCR_WVR(1),
1889 	/* DBGDCCINT */
1890 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
1891 	/* DBGDSCRext */
1892 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
1893 	DBG_BCR_BVR_WCR_WVR(2),
1894 	/* DBGDTR[RT]Xint */
1895 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1896 	/* DBGDTR[RT]Xext */
1897 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1898 	DBG_BCR_BVR_WCR_WVR(3),
1899 	DBG_BCR_BVR_WCR_WVR(4),
1900 	DBG_BCR_BVR_WCR_WVR(5),
1901 	/* DBGWFAR */
1902 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1903 	/* DBGOSECCR */
1904 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1905 	DBG_BCR_BVR_WCR_WVR(6),
1906 	/* DBGVCR */
1907 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
1908 	DBG_BCR_BVR_WCR_WVR(7),
1909 	DBG_BCR_BVR_WCR_WVR(8),
1910 	DBG_BCR_BVR_WCR_WVR(9),
1911 	DBG_BCR_BVR_WCR_WVR(10),
1912 	DBG_BCR_BVR_WCR_WVR(11),
1913 	DBG_BCR_BVR_WCR_WVR(12),
1914 	DBG_BCR_BVR_WCR_WVR(13),
1915 	DBG_BCR_BVR_WCR_WVR(14),
1916 	DBG_BCR_BVR_WCR_WVR(15),
1917 
1918 	/* DBGDRAR (32bit) */
1919 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1920 
1921 	DBGBXVR(0),
1922 	/* DBGOSLAR */
1923 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1924 	DBGBXVR(1),
1925 	/* DBGOSLSR */
1926 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1927 	DBGBXVR(2),
1928 	DBGBXVR(3),
1929 	/* DBGOSDLR */
1930 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1931 	DBGBXVR(4),
1932 	/* DBGPRCR */
1933 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1934 	DBGBXVR(5),
1935 	DBGBXVR(6),
1936 	DBGBXVR(7),
1937 	DBGBXVR(8),
1938 	DBGBXVR(9),
1939 	DBGBXVR(10),
1940 	DBGBXVR(11),
1941 	DBGBXVR(12),
1942 	DBGBXVR(13),
1943 	DBGBXVR(14),
1944 	DBGBXVR(15),
1945 
1946 	/* DBGDSAR (32bit) */
1947 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1948 
1949 	/* DBGDEVID2 */
1950 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1951 	/* DBGDEVID1 */
1952 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1953 	/* DBGDEVID */
1954 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1955 	/* DBGCLAIMSET */
1956 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1957 	/* DBGCLAIMCLR */
1958 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1959 	/* DBGAUTHSTATUS */
1960 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1961 };
1962 
1963 /* Trapped cp14 64bit registers */
1964 static const struct sys_reg_desc cp14_64_regs[] = {
1965 	/* DBGDRAR (64bit) */
1966 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1967 
1968 	/* DBGDSAR (64bit) */
1969 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1970 };
1971 
1972 /* Macro to expand the PMEVCNTRn register */
1973 #define PMU_PMEVCNTR(n)							\
1974 	/* PMEVCNTRn */							\
1975 	{ Op1(0), CRn(0b1110),						\
1976 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1977 	  access_pmu_evcntr }
1978 
1979 /* Macro to expand the PMEVTYPERn register */
1980 #define PMU_PMEVTYPER(n)						\
1981 	/* PMEVTYPERn */						\
1982 	{ Op1(0), CRn(0b1110),						\
1983 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1984 	  access_pmu_evtyper }
1985 
1986 /*
1987  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1988  * depending on the way they are accessed (as a 32bit or a 64bit
1989  * register).
1990  */
1991 static const struct sys_reg_desc cp15_regs[] = {
1992 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1993 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
1994 	/* ACTLR */
1995 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
1996 	/* ACTLR2 */
1997 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
1998 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
1999 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
2000 	/* TTBCR */
2001 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
2002 	/* TTBCR2 */
2003 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
2004 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
2005 	/* DFSR */
2006 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
2007 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
2008 	/* ADFSR */
2009 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
2010 	/* AIFSR */
2011 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
2012 	/* DFAR */
2013 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
2014 	/* IFAR */
2015 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
2016 
2017 	/*
2018 	 * DC{C,I,CI}SW operations:
2019 	 */
2020 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
2021 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
2022 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
2023 
2024 	/* PMU */
2025 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
2026 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
2027 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
2028 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
2029 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
2030 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
2031 	{ AA32(LO), Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
2032 	{ AA32(LO), Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
2033 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
2034 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
2035 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
2036 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
2037 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
2038 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
2039 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
2040 	{ AA32(HI), Op1( 0), CRn( 9), CRm(14), Op2( 4), access_pmceid },
2041 	{ AA32(HI), Op1( 0), CRn( 9), CRm(14), Op2( 5), access_pmceid },
2042 	/* PMMIR */
2043 	{ Op1( 0), CRn( 9), CRm(14), Op2( 6), trap_raz_wi },
2044 
2045 	/* PRRR/MAIR0 */
2046 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
2047 	/* NMRR/MAIR1 */
2048 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
2049 	/* AMAIR0 */
2050 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
2051 	/* AMAIR1 */
2052 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
2053 
2054 	/* ICC_SRE */
2055 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2056 
2057 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
2058 
2059 	/* Arch Tmers */
2060 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2061 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2062 
2063 	/* PMEVCNTRn */
2064 	PMU_PMEVCNTR(0),
2065 	PMU_PMEVCNTR(1),
2066 	PMU_PMEVCNTR(2),
2067 	PMU_PMEVCNTR(3),
2068 	PMU_PMEVCNTR(4),
2069 	PMU_PMEVCNTR(5),
2070 	PMU_PMEVCNTR(6),
2071 	PMU_PMEVCNTR(7),
2072 	PMU_PMEVCNTR(8),
2073 	PMU_PMEVCNTR(9),
2074 	PMU_PMEVCNTR(10),
2075 	PMU_PMEVCNTR(11),
2076 	PMU_PMEVCNTR(12),
2077 	PMU_PMEVCNTR(13),
2078 	PMU_PMEVCNTR(14),
2079 	PMU_PMEVCNTR(15),
2080 	PMU_PMEVCNTR(16),
2081 	PMU_PMEVCNTR(17),
2082 	PMU_PMEVCNTR(18),
2083 	PMU_PMEVCNTR(19),
2084 	PMU_PMEVCNTR(20),
2085 	PMU_PMEVCNTR(21),
2086 	PMU_PMEVCNTR(22),
2087 	PMU_PMEVCNTR(23),
2088 	PMU_PMEVCNTR(24),
2089 	PMU_PMEVCNTR(25),
2090 	PMU_PMEVCNTR(26),
2091 	PMU_PMEVCNTR(27),
2092 	PMU_PMEVCNTR(28),
2093 	PMU_PMEVCNTR(29),
2094 	PMU_PMEVCNTR(30),
2095 	/* PMEVTYPERn */
2096 	PMU_PMEVTYPER(0),
2097 	PMU_PMEVTYPER(1),
2098 	PMU_PMEVTYPER(2),
2099 	PMU_PMEVTYPER(3),
2100 	PMU_PMEVTYPER(4),
2101 	PMU_PMEVTYPER(5),
2102 	PMU_PMEVTYPER(6),
2103 	PMU_PMEVTYPER(7),
2104 	PMU_PMEVTYPER(8),
2105 	PMU_PMEVTYPER(9),
2106 	PMU_PMEVTYPER(10),
2107 	PMU_PMEVTYPER(11),
2108 	PMU_PMEVTYPER(12),
2109 	PMU_PMEVTYPER(13),
2110 	PMU_PMEVTYPER(14),
2111 	PMU_PMEVTYPER(15),
2112 	PMU_PMEVTYPER(16),
2113 	PMU_PMEVTYPER(17),
2114 	PMU_PMEVTYPER(18),
2115 	PMU_PMEVTYPER(19),
2116 	PMU_PMEVTYPER(20),
2117 	PMU_PMEVTYPER(21),
2118 	PMU_PMEVTYPER(22),
2119 	PMU_PMEVTYPER(23),
2120 	PMU_PMEVTYPER(24),
2121 	PMU_PMEVTYPER(25),
2122 	PMU_PMEVTYPER(26),
2123 	PMU_PMEVTYPER(27),
2124 	PMU_PMEVTYPER(28),
2125 	PMU_PMEVTYPER(29),
2126 	PMU_PMEVTYPER(30),
2127 	/* PMCCFILTR */
2128 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2129 
2130 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2131 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2132 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
2133 };
2134 
2135 static const struct sys_reg_desc cp15_64_regs[] = {
2136 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
2137 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2138 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2139 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
2140 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2141 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2142 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2143 };
2144 
2145 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2146 			      bool is_32)
2147 {
2148 	unsigned int i;
2149 
2150 	for (i = 0; i < n; i++) {
2151 		if (!is_32 && table[i].reg && !table[i].reset) {
2152 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2153 				table, i);
2154 			return 1;
2155 		}
2156 
2157 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2158 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2159 			return 1;
2160 		}
2161 	}
2162 
2163 	return 0;
2164 }
2165 
2166 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2167 {
2168 	kvm_inject_undefined(vcpu);
2169 	return 1;
2170 }
2171 
2172 static void perform_access(struct kvm_vcpu *vcpu,
2173 			   struct sys_reg_params *params,
2174 			   const struct sys_reg_desc *r)
2175 {
2176 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2177 
2178 	/* Check for regs disabled by runtime config */
2179 	if (sysreg_hidden(vcpu, r)) {
2180 		kvm_inject_undefined(vcpu);
2181 		return;
2182 	}
2183 
2184 	/*
2185 	 * Not having an accessor means that we have configured a trap
2186 	 * that we don't know how to handle. This certainly qualifies
2187 	 * as a gross bug that should be fixed right away.
2188 	 */
2189 	BUG_ON(!r->access);
2190 
2191 	/* Skip instruction if instructed so */
2192 	if (likely(r->access(vcpu, params, r)))
2193 		kvm_incr_pc(vcpu);
2194 }
2195 
2196 /*
2197  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2198  *                call the corresponding trap handler.
2199  *
2200  * @params: pointer to the descriptor of the access
2201  * @table: array of trap descriptors
2202  * @num: size of the trap descriptor array
2203  *
2204  * Return 0 if the access has been handled, and -1 if not.
2205  */
2206 static int emulate_cp(struct kvm_vcpu *vcpu,
2207 		      struct sys_reg_params *params,
2208 		      const struct sys_reg_desc *table,
2209 		      size_t num)
2210 {
2211 	const struct sys_reg_desc *r;
2212 
2213 	if (!table)
2214 		return -1;	/* Not handled */
2215 
2216 	r = find_reg(params, table, num);
2217 
2218 	if (r) {
2219 		perform_access(vcpu, params, r);
2220 		return 0;
2221 	}
2222 
2223 	/* Not handled */
2224 	return -1;
2225 }
2226 
2227 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2228 				struct sys_reg_params *params)
2229 {
2230 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2231 	int cp = -1;
2232 
2233 	switch (esr_ec) {
2234 	case ESR_ELx_EC_CP15_32:
2235 	case ESR_ELx_EC_CP15_64:
2236 		cp = 15;
2237 		break;
2238 	case ESR_ELx_EC_CP14_MR:
2239 	case ESR_ELx_EC_CP14_64:
2240 		cp = 14;
2241 		break;
2242 	default:
2243 		WARN_ON(1);
2244 	}
2245 
2246 	print_sys_reg_msg(params,
2247 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2248 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2249 	kvm_inject_undefined(vcpu);
2250 }
2251 
2252 /**
2253  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2254  * @vcpu: The VCPU pointer
2255  * @run:  The kvm_run struct
2256  */
2257 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2258 			    const struct sys_reg_desc *global,
2259 			    size_t nr_global)
2260 {
2261 	struct sys_reg_params params;
2262 	u32 esr = kvm_vcpu_get_esr(vcpu);
2263 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2264 	int Rt2 = (esr >> 10) & 0x1f;
2265 
2266 	params.CRm = (esr >> 1) & 0xf;
2267 	params.is_write = ((esr & 1) == 0);
2268 
2269 	params.Op0 = 0;
2270 	params.Op1 = (esr >> 16) & 0xf;
2271 	params.Op2 = 0;
2272 	params.CRn = 0;
2273 
2274 	/*
2275 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2276 	 * backends between AArch32 and AArch64, we get away with it.
2277 	 */
2278 	if (params.is_write) {
2279 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2280 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2281 	}
2282 
2283 	/*
2284 	 * If the table contains a handler, handle the
2285 	 * potential register operation in the case of a read and return
2286 	 * with success.
2287 	 */
2288 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2289 		/* Split up the value between registers for the read side */
2290 		if (!params.is_write) {
2291 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2292 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2293 		}
2294 
2295 		return 1;
2296 	}
2297 
2298 	unhandled_cp_access(vcpu, &params);
2299 	return 1;
2300 }
2301 
2302 /**
2303  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2304  * @vcpu: The VCPU pointer
2305  * @run:  The kvm_run struct
2306  */
2307 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2308 			    const struct sys_reg_desc *global,
2309 			    size_t nr_global)
2310 {
2311 	struct sys_reg_params params;
2312 	u32 esr = kvm_vcpu_get_esr(vcpu);
2313 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2314 
2315 	params.CRm = (esr >> 1) & 0xf;
2316 	params.regval = vcpu_get_reg(vcpu, Rt);
2317 	params.is_write = ((esr & 1) == 0);
2318 	params.CRn = (esr >> 10) & 0xf;
2319 	params.Op0 = 0;
2320 	params.Op1 = (esr >> 14) & 0x7;
2321 	params.Op2 = (esr >> 17) & 0x7;
2322 
2323 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2324 		if (!params.is_write)
2325 			vcpu_set_reg(vcpu, Rt, params.regval);
2326 		return 1;
2327 	}
2328 
2329 	unhandled_cp_access(vcpu, &params);
2330 	return 1;
2331 }
2332 
2333 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2334 {
2335 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2336 }
2337 
2338 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2339 {
2340 	return kvm_handle_cp_32(vcpu, cp15_regs, ARRAY_SIZE(cp15_regs));
2341 }
2342 
2343 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2344 {
2345 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2346 }
2347 
2348 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2349 {
2350 	return kvm_handle_cp_32(vcpu, cp14_regs, ARRAY_SIZE(cp14_regs));
2351 }
2352 
2353 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2354 {
2355 	// See ARM DDI 0487E.a, section D12.3.2
2356 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2357 }
2358 
2359 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2360 			   struct sys_reg_params *params)
2361 {
2362 	const struct sys_reg_desc *r;
2363 
2364 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2365 
2366 	if (likely(r)) {
2367 		perform_access(vcpu, params, r);
2368 	} else if (is_imp_def_sys_reg(params)) {
2369 		kvm_inject_undefined(vcpu);
2370 	} else {
2371 		print_sys_reg_msg(params,
2372 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2373 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2374 		kvm_inject_undefined(vcpu);
2375 	}
2376 	return 1;
2377 }
2378 
2379 /**
2380  * kvm_reset_sys_regs - sets system registers to reset value
2381  * @vcpu: The VCPU pointer
2382  *
2383  * This function finds the right table above and sets the registers on the
2384  * virtual CPU struct to their architecturally defined reset values.
2385  */
2386 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2387 {
2388 	unsigned long i;
2389 
2390 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2391 		if (sys_reg_descs[i].reset)
2392 			sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2393 }
2394 
2395 /**
2396  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2397  * @vcpu: The VCPU pointer
2398  */
2399 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2400 {
2401 	struct sys_reg_params params;
2402 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
2403 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2404 	int ret;
2405 
2406 	trace_kvm_handle_sys_reg(esr);
2407 
2408 	params = esr_sys64_to_params(esr);
2409 	params.regval = vcpu_get_reg(vcpu, Rt);
2410 
2411 	ret = emulate_sys_reg(vcpu, &params);
2412 
2413 	if (!params.is_write)
2414 		vcpu_set_reg(vcpu, Rt, params.regval);
2415 	return ret;
2416 }
2417 
2418 /******************************************************************************
2419  * Userspace API
2420  *****************************************************************************/
2421 
2422 static bool index_to_params(u64 id, struct sys_reg_params *params)
2423 {
2424 	switch (id & KVM_REG_SIZE_MASK) {
2425 	case KVM_REG_SIZE_U64:
2426 		/* Any unused index bits means it's not valid. */
2427 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2428 			      | KVM_REG_ARM_COPROC_MASK
2429 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2430 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2431 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2432 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2433 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2434 			return false;
2435 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2436 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2437 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2438 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2439 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2440 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2441 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2442 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2443 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2444 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2445 		return true;
2446 	default:
2447 		return false;
2448 	}
2449 }
2450 
2451 const struct sys_reg_desc *find_reg_by_id(u64 id,
2452 					  struct sys_reg_params *params,
2453 					  const struct sys_reg_desc table[],
2454 					  unsigned int num)
2455 {
2456 	if (!index_to_params(id, params))
2457 		return NULL;
2458 
2459 	return find_reg(params, table, num);
2460 }
2461 
2462 /* Decode an index value, and find the sys_reg_desc entry. */
2463 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2464 						    u64 id)
2465 {
2466 	const struct sys_reg_desc *r;
2467 	struct sys_reg_params params;
2468 
2469 	/* We only do sys_reg for now. */
2470 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2471 		return NULL;
2472 
2473 	if (!index_to_params(id, &params))
2474 		return NULL;
2475 
2476 	r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2477 
2478 	/* Not saved in the sys_reg array and not otherwise accessible? */
2479 	if (r && !(r->reg || r->get_user))
2480 		r = NULL;
2481 
2482 	return r;
2483 }
2484 
2485 /*
2486  * These are the invariant sys_reg registers: we let the guest see the
2487  * host versions of these, so they're part of the guest state.
2488  *
2489  * A future CPU may provide a mechanism to present different values to
2490  * the guest, or a future kvm may trap them.
2491  */
2492 
2493 #define FUNCTION_INVARIANT(reg)						\
2494 	static void get_##reg(struct kvm_vcpu *v,			\
2495 			      const struct sys_reg_desc *r)		\
2496 	{								\
2497 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2498 	}
2499 
2500 FUNCTION_INVARIANT(midr_el1)
2501 FUNCTION_INVARIANT(revidr_el1)
2502 FUNCTION_INVARIANT(clidr_el1)
2503 FUNCTION_INVARIANT(aidr_el1)
2504 
2505 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2506 {
2507 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2508 }
2509 
2510 /* ->val is filled in by kvm_sys_reg_table_init() */
2511 static struct sys_reg_desc invariant_sys_regs[] = {
2512 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2513 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2514 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2515 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2516 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2517 };
2518 
2519 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2520 {
2521 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2522 		return -EFAULT;
2523 	return 0;
2524 }
2525 
2526 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2527 {
2528 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2529 		return -EFAULT;
2530 	return 0;
2531 }
2532 
2533 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2534 {
2535 	struct sys_reg_params params;
2536 	const struct sys_reg_desc *r;
2537 
2538 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2539 			   ARRAY_SIZE(invariant_sys_regs));
2540 	if (!r)
2541 		return -ENOENT;
2542 
2543 	return reg_to_user(uaddr, &r->val, id);
2544 }
2545 
2546 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2547 {
2548 	struct sys_reg_params params;
2549 	const struct sys_reg_desc *r;
2550 	int err;
2551 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2552 
2553 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2554 			   ARRAY_SIZE(invariant_sys_regs));
2555 	if (!r)
2556 		return -ENOENT;
2557 
2558 	err = reg_from_user(&val, uaddr, id);
2559 	if (err)
2560 		return err;
2561 
2562 	/* This is what we mean by invariant: you can't change it. */
2563 	if (r->val != val)
2564 		return -EINVAL;
2565 
2566 	return 0;
2567 }
2568 
2569 static bool is_valid_cache(u32 val)
2570 {
2571 	u32 level, ctype;
2572 
2573 	if (val >= CSSELR_MAX)
2574 		return false;
2575 
2576 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2577 	level = (val >> 1);
2578 	ctype = (cache_levels >> (level * 3)) & 7;
2579 
2580 	switch (ctype) {
2581 	case 0: /* No cache */
2582 		return false;
2583 	case 1: /* Instruction cache only */
2584 		return (val & 1);
2585 	case 2: /* Data cache only */
2586 	case 4: /* Unified cache */
2587 		return !(val & 1);
2588 	case 3: /* Separate instruction and data caches */
2589 		return true;
2590 	default: /* Reserved: we can't know instruction or data. */
2591 		return false;
2592 	}
2593 }
2594 
2595 static int demux_c15_get(u64 id, void __user *uaddr)
2596 {
2597 	u32 val;
2598 	u32 __user *uval = uaddr;
2599 
2600 	/* Fail if we have unknown bits set. */
2601 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2602 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2603 		return -ENOENT;
2604 
2605 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2606 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2607 		if (KVM_REG_SIZE(id) != 4)
2608 			return -ENOENT;
2609 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2610 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2611 		if (!is_valid_cache(val))
2612 			return -ENOENT;
2613 
2614 		return put_user(get_ccsidr(val), uval);
2615 	default:
2616 		return -ENOENT;
2617 	}
2618 }
2619 
2620 static int demux_c15_set(u64 id, void __user *uaddr)
2621 {
2622 	u32 val, newval;
2623 	u32 __user *uval = uaddr;
2624 
2625 	/* Fail if we have unknown bits set. */
2626 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2627 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2628 		return -ENOENT;
2629 
2630 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2631 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2632 		if (KVM_REG_SIZE(id) != 4)
2633 			return -ENOENT;
2634 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2635 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2636 		if (!is_valid_cache(val))
2637 			return -ENOENT;
2638 
2639 		if (get_user(newval, uval))
2640 			return -EFAULT;
2641 
2642 		/* This is also invariant: you can't change it. */
2643 		if (newval != get_ccsidr(val))
2644 			return -EINVAL;
2645 		return 0;
2646 	default:
2647 		return -ENOENT;
2648 	}
2649 }
2650 
2651 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2652 {
2653 	const struct sys_reg_desc *r;
2654 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2655 
2656 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2657 		return demux_c15_get(reg->id, uaddr);
2658 
2659 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2660 		return -ENOENT;
2661 
2662 	r = index_to_sys_reg_desc(vcpu, reg->id);
2663 	if (!r)
2664 		return get_invariant_sys_reg(reg->id, uaddr);
2665 
2666 	/* Check for regs disabled by runtime config */
2667 	if (sysreg_hidden(vcpu, r))
2668 		return -ENOENT;
2669 
2670 	if (r->get_user)
2671 		return (r->get_user)(vcpu, r, reg, uaddr);
2672 
2673 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2674 }
2675 
2676 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2677 {
2678 	const struct sys_reg_desc *r;
2679 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2680 
2681 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2682 		return demux_c15_set(reg->id, uaddr);
2683 
2684 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2685 		return -ENOENT;
2686 
2687 	r = index_to_sys_reg_desc(vcpu, reg->id);
2688 	if (!r)
2689 		return set_invariant_sys_reg(reg->id, uaddr);
2690 
2691 	/* Check for regs disabled by runtime config */
2692 	if (sysreg_hidden(vcpu, r))
2693 		return -ENOENT;
2694 
2695 	if (r->set_user)
2696 		return (r->set_user)(vcpu, r, reg, uaddr);
2697 
2698 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2699 }
2700 
2701 static unsigned int num_demux_regs(void)
2702 {
2703 	unsigned int i, count = 0;
2704 
2705 	for (i = 0; i < CSSELR_MAX; i++)
2706 		if (is_valid_cache(i))
2707 			count++;
2708 
2709 	return count;
2710 }
2711 
2712 static int write_demux_regids(u64 __user *uindices)
2713 {
2714 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2715 	unsigned int i;
2716 
2717 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2718 	for (i = 0; i < CSSELR_MAX; i++) {
2719 		if (!is_valid_cache(i))
2720 			continue;
2721 		if (put_user(val | i, uindices))
2722 			return -EFAULT;
2723 		uindices++;
2724 	}
2725 	return 0;
2726 }
2727 
2728 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2729 {
2730 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2731 		KVM_REG_ARM64_SYSREG |
2732 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2733 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2734 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2735 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2736 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2737 }
2738 
2739 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2740 {
2741 	if (!*uind)
2742 		return true;
2743 
2744 	if (put_user(sys_reg_to_index(reg), *uind))
2745 		return false;
2746 
2747 	(*uind)++;
2748 	return true;
2749 }
2750 
2751 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2752 			    const struct sys_reg_desc *rd,
2753 			    u64 __user **uind,
2754 			    unsigned int *total)
2755 {
2756 	/*
2757 	 * Ignore registers we trap but don't save,
2758 	 * and for which no custom user accessor is provided.
2759 	 */
2760 	if (!(rd->reg || rd->get_user))
2761 		return 0;
2762 
2763 	if (sysreg_hidden(vcpu, rd))
2764 		return 0;
2765 
2766 	if (!copy_reg_to_user(rd, uind))
2767 		return -EFAULT;
2768 
2769 	(*total)++;
2770 	return 0;
2771 }
2772 
2773 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2774 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2775 {
2776 	const struct sys_reg_desc *i2, *end2;
2777 	unsigned int total = 0;
2778 	int err;
2779 
2780 	i2 = sys_reg_descs;
2781 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2782 
2783 	while (i2 != end2) {
2784 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2785 		if (err)
2786 			return err;
2787 	}
2788 	return total;
2789 }
2790 
2791 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2792 {
2793 	return ARRAY_SIZE(invariant_sys_regs)
2794 		+ num_demux_regs()
2795 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2796 }
2797 
2798 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2799 {
2800 	unsigned int i;
2801 	int err;
2802 
2803 	/* Then give them all the invariant registers' indices. */
2804 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2805 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2806 			return -EFAULT;
2807 		uindices++;
2808 	}
2809 
2810 	err = walk_sys_regs(vcpu, uindices);
2811 	if (err < 0)
2812 		return err;
2813 	uindices += err;
2814 
2815 	return write_demux_regids(uindices);
2816 }
2817 
2818 void kvm_sys_reg_table_init(void)
2819 {
2820 	unsigned int i;
2821 	struct sys_reg_desc clidr;
2822 
2823 	/* Make sure tables are unique and in order. */
2824 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2825 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2826 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2827 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2828 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2829 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2830 
2831 	/* We abuse the reset function to overwrite the table itself. */
2832 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2833 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2834 
2835 	/*
2836 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2837 	 *
2838 	 *   If software reads the Cache Type fields from Ctype1
2839 	 *   upwards, once it has seen a value of 0b000, no caches
2840 	 *   exist at further-out levels of the hierarchy. So, for
2841 	 *   example, if Ctype3 is the first Cache Type field with a
2842 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2843 	 *   ignored.
2844 	 */
2845 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2846 	cache_levels = clidr.val;
2847 	for (i = 0; i < 7; i++)
2848 		if (((cache_levels >> (i*3)) & 7) == 0)
2849 			break;
2850 	/* Clear all higher bits. */
2851 	cache_levels &= (1 << (i*3))-1;
2852 }
2853