xref: /linux/arch/arm64/kvm/sys_regs.c (revision c01044cc819160323f3ca4acd44fca487c4432e6)
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/bsearch.h>
13 #include <linux/kvm_host.h>
14 #include <linux/mm.h>
15 #include <linux/printk.h>
16 #include <linux/uaccess.h>
17 
18 #include <asm/cacheflush.h>
19 #include <asm/cputype.h>
20 #include <asm/debug-monitors.h>
21 #include <asm/esr.h>
22 #include <asm/kvm_arm.h>
23 #include <asm/kvm_coproc.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 static bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
68 {
69 	/*
70 	 * System registers listed in the switch are not saved on every
71 	 * exit from the guest but are only saved on vcpu_put.
72 	 *
73 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
74 	 * should never be listed below, because the guest cannot modify its
75 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
76 	 * thread when emulating cross-VCPU communication.
77 	 */
78 	switch (reg) {
79 	case CSSELR_EL1:	*val = read_sysreg_s(SYS_CSSELR_EL1);	break;
80 	case SCTLR_EL1:		*val = read_sysreg_s(SYS_SCTLR_EL12);	break;
81 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
82 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
83 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
84 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
85 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
86 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
87 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
88 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
89 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
90 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
91 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
92 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
93 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
94 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
95 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
96 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
97 	case ELR_EL1:		*val = read_sysreg_s(SYS_ELR_EL12);	break;
98 	case PAR_EL1:		*val = read_sysreg_s(SYS_PAR_EL1);	break;
99 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
100 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
101 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
102 	default:		return false;
103 	}
104 
105 	return true;
106 }
107 
108 static bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
109 {
110 	/*
111 	 * System registers listed in the switch are not restored on every
112 	 * entry to the guest but are only restored on vcpu_load.
113 	 *
114 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
115 	 * should never be listed below, because the MPIDR should only be set
116 	 * once, before running the VCPU, and never changed later.
117 	 */
118 	switch (reg) {
119 	case CSSELR_EL1:	write_sysreg_s(val, SYS_CSSELR_EL1);	break;
120 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
121 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
122 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
123 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
124 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
125 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
126 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
127 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
128 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
129 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
130 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
131 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
132 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
133 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
134 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
135 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
136 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
137 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
138 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
139 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
140 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
141 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
142 	default:		return false;
143 	}
144 
145 	return true;
146 }
147 
148 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
149 {
150 	u64 val = 0x8badf00d8badf00d;
151 
152 	if (vcpu->arch.sysregs_loaded_on_cpu &&
153 	    __vcpu_read_sys_reg_from_cpu(reg, &val))
154 		return val;
155 
156 	return __vcpu_sys_reg(vcpu, reg);
157 }
158 
159 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
160 {
161 	if (vcpu->arch.sysregs_loaded_on_cpu &&
162 	    __vcpu_write_sys_reg_to_cpu(val, reg))
163 		return;
164 
165 	 __vcpu_sys_reg(vcpu, reg) = val;
166 }
167 
168 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
169 static u32 cache_levels;
170 
171 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
172 #define CSSELR_MAX 12
173 
174 /* Which cache CCSIDR represents depends on CSSELR value. */
175 static u32 get_ccsidr(u32 csselr)
176 {
177 	u32 ccsidr;
178 
179 	/* Make sure noone else changes CSSELR during this! */
180 	local_irq_disable();
181 	write_sysreg(csselr, csselr_el1);
182 	isb();
183 	ccsidr = read_sysreg(ccsidr_el1);
184 	local_irq_enable();
185 
186 	return ccsidr;
187 }
188 
189 /*
190  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
191  */
192 static bool access_dcsw(struct kvm_vcpu *vcpu,
193 			struct sys_reg_params *p,
194 			const struct sys_reg_desc *r)
195 {
196 	if (!p->is_write)
197 		return read_from_write_only(vcpu, p, r);
198 
199 	/*
200 	 * Only track S/W ops if we don't have FWB. It still indicates
201 	 * that the guest is a bit broken (S/W operations should only
202 	 * be done by firmware, knowing that there is only a single
203 	 * CPU left in the system, and certainly not from non-secure
204 	 * software).
205 	 */
206 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
207 		kvm_set_way_flush(vcpu);
208 
209 	return true;
210 }
211 
212 /*
213  * Generic accessor for VM registers. Only called as long as HCR_TVM
214  * is set. If the guest enables the MMU, we stop trapping the VM
215  * sys_regs and leave it in complete control of the caches.
216  */
217 static bool access_vm_reg(struct kvm_vcpu *vcpu,
218 			  struct sys_reg_params *p,
219 			  const struct sys_reg_desc *r)
220 {
221 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
222 	u64 val;
223 	int reg = r->reg;
224 
225 	BUG_ON(!p->is_write);
226 
227 	/* See the 32bit mapping in kvm_host.h */
228 	if (p->is_aarch32)
229 		reg = r->reg / 2;
230 
231 	if (!p->is_aarch32 || !p->is_32bit) {
232 		val = p->regval;
233 	} else {
234 		val = vcpu_read_sys_reg(vcpu, reg);
235 		if (r->reg % 2)
236 			val = (p->regval << 32) | (u64)lower_32_bits(val);
237 		else
238 			val = ((u64)upper_32_bits(val) << 32) |
239 				lower_32_bits(p->regval);
240 	}
241 	vcpu_write_sys_reg(vcpu, val, reg);
242 
243 	kvm_toggle_cache(vcpu, was_enabled);
244 	return true;
245 }
246 
247 static bool access_actlr(struct kvm_vcpu *vcpu,
248 			 struct sys_reg_params *p,
249 			 const struct sys_reg_desc *r)
250 {
251 	if (p->is_write)
252 		return ignore_write(vcpu, p);
253 
254 	p->regval = vcpu_read_sys_reg(vcpu, ACTLR_EL1);
255 
256 	if (p->is_aarch32) {
257 		if (r->Op2 & 2)
258 			p->regval = upper_32_bits(p->regval);
259 		else
260 			p->regval = lower_32_bits(p->regval);
261 	}
262 
263 	return true;
264 }
265 
266 /*
267  * Trap handler for the GICv3 SGI generation system register.
268  * Forward the request to the VGIC emulation.
269  * The cp15_64 code makes sure this automatically works
270  * for both AArch64 and AArch32 accesses.
271  */
272 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
273 			   struct sys_reg_params *p,
274 			   const struct sys_reg_desc *r)
275 {
276 	bool g1;
277 
278 	if (!p->is_write)
279 		return read_from_write_only(vcpu, p, r);
280 
281 	/*
282 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
283 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
284 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
285 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
286 	 * group.
287 	 */
288 	if (p->is_aarch32) {
289 		switch (p->Op1) {
290 		default:		/* Keep GCC quiet */
291 		case 0:			/* ICC_SGI1R */
292 			g1 = true;
293 			break;
294 		case 1:			/* ICC_ASGI1R */
295 		case 2:			/* ICC_SGI0R */
296 			g1 = false;
297 			break;
298 		}
299 	} else {
300 		switch (p->Op2) {
301 		default:		/* Keep GCC quiet */
302 		case 5:			/* ICC_SGI1R_EL1 */
303 			g1 = true;
304 			break;
305 		case 6:			/* ICC_ASGI1R_EL1 */
306 		case 7:			/* ICC_SGI0R_EL1 */
307 			g1 = false;
308 			break;
309 		}
310 	}
311 
312 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
313 
314 	return true;
315 }
316 
317 static bool access_gic_sre(struct kvm_vcpu *vcpu,
318 			   struct sys_reg_params *p,
319 			   const struct sys_reg_desc *r)
320 {
321 	if (p->is_write)
322 		return ignore_write(vcpu, p);
323 
324 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
325 	return true;
326 }
327 
328 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
329 			struct sys_reg_params *p,
330 			const struct sys_reg_desc *r)
331 {
332 	if (p->is_write)
333 		return ignore_write(vcpu, p);
334 	else
335 		return read_zero(vcpu, p);
336 }
337 
338 /*
339  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
340  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
341  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
342  * treat it separately.
343  */
344 static bool trap_loregion(struct kvm_vcpu *vcpu,
345 			  struct sys_reg_params *p,
346 			  const struct sys_reg_desc *r)
347 {
348 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
349 	u32 sr = sys_reg((u32)r->Op0, (u32)r->Op1,
350 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
351 
352 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
353 		kvm_inject_undefined(vcpu);
354 		return false;
355 	}
356 
357 	if (p->is_write && sr == SYS_LORID_EL1)
358 		return write_to_read_only(vcpu, p, r);
359 
360 	return trap_raz_wi(vcpu, p, r);
361 }
362 
363 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
364 			   struct sys_reg_params *p,
365 			   const struct sys_reg_desc *r)
366 {
367 	if (p->is_write) {
368 		return ignore_write(vcpu, p);
369 	} else {
370 		p->regval = (1 << 3);
371 		return true;
372 	}
373 }
374 
375 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
376 				   struct sys_reg_params *p,
377 				   const struct sys_reg_desc *r)
378 {
379 	if (p->is_write) {
380 		return ignore_write(vcpu, p);
381 	} else {
382 		p->regval = read_sysreg(dbgauthstatus_el1);
383 		return true;
384 	}
385 }
386 
387 /*
388  * We want to avoid world-switching all the DBG registers all the
389  * time:
390  *
391  * - If we've touched any debug register, it is likely that we're
392  *   going to touch more of them. It then makes sense to disable the
393  *   traps and start doing the save/restore dance
394  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
395  *   then mandatory to save/restore the registers, as the guest
396  *   depends on them.
397  *
398  * For this, we use a DIRTY bit, indicating the guest has modified the
399  * debug registers, used as follow:
400  *
401  * On guest entry:
402  * - If the dirty bit is set (because we're coming back from trapping),
403  *   disable the traps, save host registers, restore guest registers.
404  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
405  *   set the dirty bit, disable the traps, save host registers,
406  *   restore guest registers.
407  * - Otherwise, enable the traps
408  *
409  * On guest exit:
410  * - If the dirty bit is set, save guest registers, restore host
411  *   registers and clear the dirty bit. This ensure that the host can
412  *   now use the debug registers.
413  */
414 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
415 			    struct sys_reg_params *p,
416 			    const struct sys_reg_desc *r)
417 {
418 	if (p->is_write) {
419 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
420 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
421 	} else {
422 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
423 	}
424 
425 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
426 
427 	return true;
428 }
429 
430 /*
431  * reg_to_dbg/dbg_to_reg
432  *
433  * A 32 bit write to a debug register leave top bits alone
434  * A 32 bit read from a debug register only returns the bottom bits
435  *
436  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
437  * hyp.S code switches between host and guest values in future.
438  */
439 static void reg_to_dbg(struct kvm_vcpu *vcpu,
440 		       struct sys_reg_params *p,
441 		       u64 *dbg_reg)
442 {
443 	u64 val = p->regval;
444 
445 	if (p->is_32bit) {
446 		val &= 0xffffffffUL;
447 		val |= ((*dbg_reg >> 32) << 32);
448 	}
449 
450 	*dbg_reg = val;
451 	vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
452 }
453 
454 static void dbg_to_reg(struct kvm_vcpu *vcpu,
455 		       struct sys_reg_params *p,
456 		       u64 *dbg_reg)
457 {
458 	p->regval = *dbg_reg;
459 	if (p->is_32bit)
460 		p->regval &= 0xffffffffUL;
461 }
462 
463 static bool trap_bvr(struct kvm_vcpu *vcpu,
464 		     struct sys_reg_params *p,
465 		     const struct sys_reg_desc *rd)
466 {
467 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
468 
469 	if (p->is_write)
470 		reg_to_dbg(vcpu, p, dbg_reg);
471 	else
472 		dbg_to_reg(vcpu, p, dbg_reg);
473 
474 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
475 
476 	return true;
477 }
478 
479 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
480 		const struct kvm_one_reg *reg, void __user *uaddr)
481 {
482 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
483 
484 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
485 		return -EFAULT;
486 	return 0;
487 }
488 
489 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
490 	const struct kvm_one_reg *reg, void __user *uaddr)
491 {
492 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
493 
494 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
495 		return -EFAULT;
496 	return 0;
497 }
498 
499 static void reset_bvr(struct kvm_vcpu *vcpu,
500 		      const struct sys_reg_desc *rd)
501 {
502 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
503 }
504 
505 static bool trap_bcr(struct kvm_vcpu *vcpu,
506 		     struct sys_reg_params *p,
507 		     const struct sys_reg_desc *rd)
508 {
509 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
510 
511 	if (p->is_write)
512 		reg_to_dbg(vcpu, p, dbg_reg);
513 	else
514 		dbg_to_reg(vcpu, p, dbg_reg);
515 
516 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
517 
518 	return true;
519 }
520 
521 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
522 		const struct kvm_one_reg *reg, void __user *uaddr)
523 {
524 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
525 
526 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
527 		return -EFAULT;
528 
529 	return 0;
530 }
531 
532 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
533 	const struct kvm_one_reg *reg, void __user *uaddr)
534 {
535 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
536 
537 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
538 		return -EFAULT;
539 	return 0;
540 }
541 
542 static void reset_bcr(struct kvm_vcpu *vcpu,
543 		      const struct sys_reg_desc *rd)
544 {
545 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
546 }
547 
548 static bool trap_wvr(struct kvm_vcpu *vcpu,
549 		     struct sys_reg_params *p,
550 		     const struct sys_reg_desc *rd)
551 {
552 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
553 
554 	if (p->is_write)
555 		reg_to_dbg(vcpu, p, dbg_reg);
556 	else
557 		dbg_to_reg(vcpu, p, dbg_reg);
558 
559 	trace_trap_reg(__func__, rd->reg, p->is_write,
560 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
561 
562 	return true;
563 }
564 
565 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
566 		const struct kvm_one_reg *reg, void __user *uaddr)
567 {
568 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
569 
570 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
571 		return -EFAULT;
572 	return 0;
573 }
574 
575 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
576 	const struct kvm_one_reg *reg, void __user *uaddr)
577 {
578 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
579 
580 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
581 		return -EFAULT;
582 	return 0;
583 }
584 
585 static void reset_wvr(struct kvm_vcpu *vcpu,
586 		      const struct sys_reg_desc *rd)
587 {
588 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
589 }
590 
591 static bool trap_wcr(struct kvm_vcpu *vcpu,
592 		     struct sys_reg_params *p,
593 		     const struct sys_reg_desc *rd)
594 {
595 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
596 
597 	if (p->is_write)
598 		reg_to_dbg(vcpu, p, dbg_reg);
599 	else
600 		dbg_to_reg(vcpu, p, dbg_reg);
601 
602 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
603 
604 	return true;
605 }
606 
607 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
608 		const struct kvm_one_reg *reg, void __user *uaddr)
609 {
610 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
611 
612 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
613 		return -EFAULT;
614 	return 0;
615 }
616 
617 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
618 	const struct kvm_one_reg *reg, void __user *uaddr)
619 {
620 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
621 
622 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
623 		return -EFAULT;
624 	return 0;
625 }
626 
627 static void reset_wcr(struct kvm_vcpu *vcpu,
628 		      const struct sys_reg_desc *rd)
629 {
630 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
631 }
632 
633 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
634 {
635 	u64 amair = read_sysreg(amair_el1);
636 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
637 }
638 
639 static void reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
640 {
641 	u64 actlr = read_sysreg(actlr_el1);
642 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
643 }
644 
645 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
646 {
647 	u64 mpidr;
648 
649 	/*
650 	 * Map the vcpu_id into the first three affinity level fields of
651 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
652 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
653 	 * of the GICv3 to be able to address each CPU directly when
654 	 * sending IPIs.
655 	 */
656 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
657 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
658 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
659 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
660 }
661 
662 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
663 {
664 	u64 pmcr, val;
665 
666 	pmcr = read_sysreg(pmcr_el0);
667 	/*
668 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
669 	 * except PMCR.E resetting to zero.
670 	 */
671 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
672 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
673 	if (!system_supports_32bit_el0())
674 		val |= ARMV8_PMU_PMCR_LC;
675 	__vcpu_sys_reg(vcpu, r->reg) = val;
676 }
677 
678 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
679 {
680 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
681 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
682 
683 	if (!enabled)
684 		kvm_inject_undefined(vcpu);
685 
686 	return !enabled;
687 }
688 
689 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
690 {
691 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
692 }
693 
694 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
695 {
696 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
697 }
698 
699 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
700 {
701 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
702 }
703 
704 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
705 {
706 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
707 }
708 
709 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
710 			const struct sys_reg_desc *r)
711 {
712 	u64 val;
713 
714 	if (!kvm_arm_pmu_v3_ready(vcpu))
715 		return trap_raz_wi(vcpu, p, r);
716 
717 	if (pmu_access_el0_disabled(vcpu))
718 		return false;
719 
720 	if (p->is_write) {
721 		/* Only update writeable bits of PMCR */
722 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
723 		val &= ~ARMV8_PMU_PMCR_MASK;
724 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
725 		if (!system_supports_32bit_el0())
726 			val |= ARMV8_PMU_PMCR_LC;
727 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
728 		kvm_pmu_handle_pmcr(vcpu, val);
729 		kvm_vcpu_pmu_restore_guest(vcpu);
730 	} else {
731 		/* PMCR.P & PMCR.C are RAZ */
732 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
733 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
734 		p->regval = val;
735 	}
736 
737 	return true;
738 }
739 
740 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
741 			  const struct sys_reg_desc *r)
742 {
743 	if (!kvm_arm_pmu_v3_ready(vcpu))
744 		return trap_raz_wi(vcpu, p, r);
745 
746 	if (pmu_access_event_counter_el0_disabled(vcpu))
747 		return false;
748 
749 	if (p->is_write)
750 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
751 	else
752 		/* return PMSELR.SEL field */
753 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
754 			    & ARMV8_PMU_COUNTER_MASK;
755 
756 	return true;
757 }
758 
759 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
760 			  const struct sys_reg_desc *r)
761 {
762 	u64 pmceid;
763 
764 	if (!kvm_arm_pmu_v3_ready(vcpu))
765 		return trap_raz_wi(vcpu, p, r);
766 
767 	BUG_ON(p->is_write);
768 
769 	if (pmu_access_el0_disabled(vcpu))
770 		return false;
771 
772 	if (!(p->Op2 & 1))
773 		pmceid = read_sysreg(pmceid0_el0);
774 	else
775 		pmceid = read_sysreg(pmceid1_el0);
776 
777 	p->regval = pmceid;
778 
779 	return true;
780 }
781 
782 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
783 {
784 	u64 pmcr, val;
785 
786 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
787 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
788 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
789 		kvm_inject_undefined(vcpu);
790 		return false;
791 	}
792 
793 	return true;
794 }
795 
796 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
797 			      struct sys_reg_params *p,
798 			      const struct sys_reg_desc *r)
799 {
800 	u64 idx;
801 
802 	if (!kvm_arm_pmu_v3_ready(vcpu))
803 		return trap_raz_wi(vcpu, p, r);
804 
805 	if (r->CRn == 9 && r->CRm == 13) {
806 		if (r->Op2 == 2) {
807 			/* PMXEVCNTR_EL0 */
808 			if (pmu_access_event_counter_el0_disabled(vcpu))
809 				return false;
810 
811 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
812 			      & ARMV8_PMU_COUNTER_MASK;
813 		} else if (r->Op2 == 0) {
814 			/* PMCCNTR_EL0 */
815 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
816 				return false;
817 
818 			idx = ARMV8_PMU_CYCLE_IDX;
819 		} else {
820 			return false;
821 		}
822 	} else if (r->CRn == 0 && r->CRm == 9) {
823 		/* PMCCNTR */
824 		if (pmu_access_event_counter_el0_disabled(vcpu))
825 			return false;
826 
827 		idx = ARMV8_PMU_CYCLE_IDX;
828 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
829 		/* PMEVCNTRn_EL0 */
830 		if (pmu_access_event_counter_el0_disabled(vcpu))
831 			return false;
832 
833 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
834 	} else {
835 		return false;
836 	}
837 
838 	if (!pmu_counter_idx_valid(vcpu, idx))
839 		return false;
840 
841 	if (p->is_write) {
842 		if (pmu_access_el0_disabled(vcpu))
843 			return false;
844 
845 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
846 	} else {
847 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
848 	}
849 
850 	return true;
851 }
852 
853 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
854 			       const struct sys_reg_desc *r)
855 {
856 	u64 idx, reg;
857 
858 	if (!kvm_arm_pmu_v3_ready(vcpu))
859 		return trap_raz_wi(vcpu, p, r);
860 
861 	if (pmu_access_el0_disabled(vcpu))
862 		return false;
863 
864 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
865 		/* PMXEVTYPER_EL0 */
866 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
867 		reg = PMEVTYPER0_EL0 + idx;
868 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
869 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
870 		if (idx == ARMV8_PMU_CYCLE_IDX)
871 			reg = PMCCFILTR_EL0;
872 		else
873 			/* PMEVTYPERn_EL0 */
874 			reg = PMEVTYPER0_EL0 + idx;
875 	} else {
876 		BUG();
877 	}
878 
879 	if (!pmu_counter_idx_valid(vcpu, idx))
880 		return false;
881 
882 	if (p->is_write) {
883 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
884 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
885 		kvm_vcpu_pmu_restore_guest(vcpu);
886 	} else {
887 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
888 	}
889 
890 	return true;
891 }
892 
893 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
894 			   const struct sys_reg_desc *r)
895 {
896 	u64 val, mask;
897 
898 	if (!kvm_arm_pmu_v3_ready(vcpu))
899 		return trap_raz_wi(vcpu, p, r);
900 
901 	if (pmu_access_el0_disabled(vcpu))
902 		return false;
903 
904 	mask = kvm_pmu_valid_counter_mask(vcpu);
905 	if (p->is_write) {
906 		val = p->regval & mask;
907 		if (r->Op2 & 0x1) {
908 			/* accessing PMCNTENSET_EL0 */
909 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
910 			kvm_pmu_enable_counter_mask(vcpu, val);
911 			kvm_vcpu_pmu_restore_guest(vcpu);
912 		} else {
913 			/* accessing PMCNTENCLR_EL0 */
914 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
915 			kvm_pmu_disable_counter_mask(vcpu, val);
916 		}
917 	} else {
918 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
919 	}
920 
921 	return true;
922 }
923 
924 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
925 			   const struct sys_reg_desc *r)
926 {
927 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
928 
929 	if (!kvm_arm_pmu_v3_ready(vcpu))
930 		return trap_raz_wi(vcpu, p, r);
931 
932 	if (!vcpu_mode_priv(vcpu)) {
933 		kvm_inject_undefined(vcpu);
934 		return false;
935 	}
936 
937 	if (p->is_write) {
938 		u64 val = p->regval & mask;
939 
940 		if (r->Op2 & 0x1)
941 			/* accessing PMINTENSET_EL1 */
942 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
943 		else
944 			/* accessing PMINTENCLR_EL1 */
945 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
946 	} else {
947 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
948 	}
949 
950 	return true;
951 }
952 
953 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
954 			 const struct sys_reg_desc *r)
955 {
956 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
957 
958 	if (!kvm_arm_pmu_v3_ready(vcpu))
959 		return trap_raz_wi(vcpu, p, r);
960 
961 	if (pmu_access_el0_disabled(vcpu))
962 		return false;
963 
964 	if (p->is_write) {
965 		if (r->CRm & 0x2)
966 			/* accessing PMOVSSET_EL0 */
967 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
968 		else
969 			/* accessing PMOVSCLR_EL0 */
970 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
971 	} else {
972 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
973 	}
974 
975 	return true;
976 }
977 
978 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
979 			   const struct sys_reg_desc *r)
980 {
981 	u64 mask;
982 
983 	if (!kvm_arm_pmu_v3_ready(vcpu))
984 		return trap_raz_wi(vcpu, p, r);
985 
986 	if (!p->is_write)
987 		return read_from_write_only(vcpu, p, r);
988 
989 	if (pmu_write_swinc_el0_disabled(vcpu))
990 		return false;
991 
992 	mask = kvm_pmu_valid_counter_mask(vcpu);
993 	kvm_pmu_software_increment(vcpu, p->regval & mask);
994 	return true;
995 }
996 
997 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
998 			     const struct sys_reg_desc *r)
999 {
1000 	if (!kvm_arm_pmu_v3_ready(vcpu))
1001 		return trap_raz_wi(vcpu, p, r);
1002 
1003 	if (p->is_write) {
1004 		if (!vcpu_mode_priv(vcpu)) {
1005 			kvm_inject_undefined(vcpu);
1006 			return false;
1007 		}
1008 
1009 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
1010 			       p->regval & ARMV8_PMU_USERENR_MASK;
1011 	} else {
1012 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1013 			    & ARMV8_PMU_USERENR_MASK;
1014 	}
1015 
1016 	return true;
1017 }
1018 
1019 #define reg_to_encoding(x)						\
1020 	sys_reg((u32)(x)->Op0, (u32)(x)->Op1,				\
1021 		(u32)(x)->CRn, (u32)(x)->CRm, (u32)(x)->Op2);
1022 
1023 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1024 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1025 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1026 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1027 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1028 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1029 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1030 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1031 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1032 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1033 
1034 /* Macro to expand the PMEVCNTRn_EL0 register */
1035 #define PMU_PMEVCNTR_EL0(n)						\
1036 	{ SYS_DESC(SYS_PMEVCNTRn_EL0(n)),					\
1037 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
1038 
1039 /* Macro to expand the PMEVTYPERn_EL0 register */
1040 #define PMU_PMEVTYPER_EL0(n)						\
1041 	{ SYS_DESC(SYS_PMEVTYPERn_EL0(n)),					\
1042 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
1043 
1044 static bool access_amu(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1045 			     const struct sys_reg_desc *r)
1046 {
1047 	kvm_inject_undefined(vcpu);
1048 
1049 	return false;
1050 }
1051 
1052 /* Macro to expand the AMU counter and type registers*/
1053 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), access_amu }
1054 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), access_amu }
1055 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), access_amu }
1056 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), access_amu }
1057 
1058 static bool trap_ptrauth(struct kvm_vcpu *vcpu,
1059 			 struct sys_reg_params *p,
1060 			 const struct sys_reg_desc *rd)
1061 {
1062 	/*
1063 	 * If we land here, that is because we didn't fixup the access on exit
1064 	 * by allowing the PtrAuth sysregs. The only way this happens is when
1065 	 * the guest does not have PtrAuth support enabled.
1066 	 */
1067 	kvm_inject_undefined(vcpu);
1068 
1069 	return false;
1070 }
1071 
1072 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1073 			const struct sys_reg_desc *rd)
1074 {
1075 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1076 }
1077 
1078 #define __PTRAUTH_KEY(k)						\
1079 	{ SYS_DESC(SYS_## k), trap_ptrauth, reset_unknown, k,		\
1080 	.visibility = ptrauth_visibility}
1081 
1082 #define PTRAUTH_KEY(k)							\
1083 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1084 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1085 
1086 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1087 			      struct sys_reg_params *p,
1088 			      const struct sys_reg_desc *r)
1089 {
1090 	enum kvm_arch_timers tmr;
1091 	enum kvm_arch_timer_regs treg;
1092 	u64 reg = reg_to_encoding(r);
1093 
1094 	switch (reg) {
1095 	case SYS_CNTP_TVAL_EL0:
1096 	case SYS_AARCH32_CNTP_TVAL:
1097 		tmr = TIMER_PTIMER;
1098 		treg = TIMER_REG_TVAL;
1099 		break;
1100 	case SYS_CNTP_CTL_EL0:
1101 	case SYS_AARCH32_CNTP_CTL:
1102 		tmr = TIMER_PTIMER;
1103 		treg = TIMER_REG_CTL;
1104 		break;
1105 	case SYS_CNTP_CVAL_EL0:
1106 	case SYS_AARCH32_CNTP_CVAL:
1107 		tmr = TIMER_PTIMER;
1108 		treg = TIMER_REG_CVAL;
1109 		break;
1110 	default:
1111 		BUG();
1112 	}
1113 
1114 	if (p->is_write)
1115 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1116 	else
1117 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1118 
1119 	return true;
1120 }
1121 
1122 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1123 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1124 		struct sys_reg_desc const *r, bool raz)
1125 {
1126 	u32 id = sys_reg((u32)r->Op0, (u32)r->Op1,
1127 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
1128 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1129 
1130 	if (id == SYS_ID_AA64PFR0_EL1) {
1131 		if (!vcpu_has_sve(vcpu))
1132 			val &= ~(0xfUL << ID_AA64PFR0_SVE_SHIFT);
1133 		val &= ~(0xfUL << ID_AA64PFR0_AMU_SHIFT);
1134 		if (!(val & (0xfUL << ID_AA64PFR0_CSV2_SHIFT)) &&
1135 		    arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
1136 			val |= (1UL << ID_AA64PFR0_CSV2_SHIFT);
1137 	} else if (id == SYS_ID_AA64PFR1_EL1) {
1138 		val &= ~(0xfUL << ID_AA64PFR1_MTE_SHIFT);
1139 	} else if (id == SYS_ID_AA64ISAR1_EL1 && !vcpu_has_ptrauth(vcpu)) {
1140 		val &= ~((0xfUL << ID_AA64ISAR1_APA_SHIFT) |
1141 			 (0xfUL << ID_AA64ISAR1_API_SHIFT) |
1142 			 (0xfUL << ID_AA64ISAR1_GPA_SHIFT) |
1143 			 (0xfUL << ID_AA64ISAR1_GPI_SHIFT));
1144 	} else if (id == SYS_ID_AA64DFR0_EL1) {
1145 		/* Limit guests to PMUv3 for ARMv8.1 */
1146 		val = cpuid_feature_cap_perfmon_field(val,
1147 						ID_AA64DFR0_PMUVER_SHIFT,
1148 						ID_AA64DFR0_PMUVER_8_1);
1149 	} else if (id == SYS_ID_DFR0_EL1) {
1150 		/* Limit guests to PMUv3 for ARMv8.1 */
1151 		val = cpuid_feature_cap_perfmon_field(val,
1152 						ID_DFR0_PERFMON_SHIFT,
1153 						ID_DFR0_PERFMON_8_1);
1154 	}
1155 
1156 	return val;
1157 }
1158 
1159 /* cpufeature ID register access trap handlers */
1160 
1161 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1162 			    struct sys_reg_params *p,
1163 			    const struct sys_reg_desc *r,
1164 			    bool raz)
1165 {
1166 	if (p->is_write)
1167 		return write_to_read_only(vcpu, p, r);
1168 
1169 	p->regval = read_id_reg(vcpu, r, raz);
1170 	return true;
1171 }
1172 
1173 static bool access_id_reg(struct kvm_vcpu *vcpu,
1174 			  struct sys_reg_params *p,
1175 			  const struct sys_reg_desc *r)
1176 {
1177 	return __access_id_reg(vcpu, p, r, false);
1178 }
1179 
1180 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1181 			      struct sys_reg_params *p,
1182 			      const struct sys_reg_desc *r)
1183 {
1184 	return __access_id_reg(vcpu, p, r, true);
1185 }
1186 
1187 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1188 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1189 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1190 
1191 /* Visibility overrides for SVE-specific control registers */
1192 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1193 				   const struct sys_reg_desc *rd)
1194 {
1195 	if (vcpu_has_sve(vcpu))
1196 		return 0;
1197 
1198 	return REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1199 }
1200 
1201 /* Visibility overrides for SVE-specific ID registers */
1202 static unsigned int sve_id_visibility(const struct kvm_vcpu *vcpu,
1203 				      const struct sys_reg_desc *rd)
1204 {
1205 	if (vcpu_has_sve(vcpu))
1206 		return 0;
1207 
1208 	return REG_HIDDEN_USER;
1209 }
1210 
1211 /* Generate the emulated ID_AA64ZFR0_EL1 value exposed to the guest */
1212 static u64 guest_id_aa64zfr0_el1(const struct kvm_vcpu *vcpu)
1213 {
1214 	if (!vcpu_has_sve(vcpu))
1215 		return 0;
1216 
1217 	return read_sanitised_ftr_reg(SYS_ID_AA64ZFR0_EL1);
1218 }
1219 
1220 static bool access_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1221 				   struct sys_reg_params *p,
1222 				   const struct sys_reg_desc *rd)
1223 {
1224 	if (p->is_write)
1225 		return write_to_read_only(vcpu, p, rd);
1226 
1227 	p->regval = guest_id_aa64zfr0_el1(vcpu);
1228 	return true;
1229 }
1230 
1231 static int get_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1232 		const struct sys_reg_desc *rd,
1233 		const struct kvm_one_reg *reg, void __user *uaddr)
1234 {
1235 	u64 val;
1236 
1237 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1238 		return -ENOENT;
1239 
1240 	val = guest_id_aa64zfr0_el1(vcpu);
1241 	return reg_to_user(uaddr, &val, reg->id);
1242 }
1243 
1244 static int set_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1245 		const struct sys_reg_desc *rd,
1246 		const struct kvm_one_reg *reg, void __user *uaddr)
1247 {
1248 	const u64 id = sys_reg_to_index(rd);
1249 	int err;
1250 	u64 val;
1251 
1252 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1253 		return -ENOENT;
1254 
1255 	err = reg_from_user(&val, uaddr, id);
1256 	if (err)
1257 		return err;
1258 
1259 	/* This is what we mean by invariant: you can't change it. */
1260 	if (val != guest_id_aa64zfr0_el1(vcpu))
1261 		return -EINVAL;
1262 
1263 	return 0;
1264 }
1265 
1266 /*
1267  * cpufeature ID register user accessors
1268  *
1269  * For now, these registers are immutable for userspace, so no values
1270  * are stored, and for set_id_reg() we don't allow the effective value
1271  * to be changed.
1272  */
1273 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1274 			const struct sys_reg_desc *rd, void __user *uaddr,
1275 			bool raz)
1276 {
1277 	const u64 id = sys_reg_to_index(rd);
1278 	const u64 val = read_id_reg(vcpu, rd, raz);
1279 
1280 	return reg_to_user(uaddr, &val, id);
1281 }
1282 
1283 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1284 			const struct sys_reg_desc *rd, void __user *uaddr,
1285 			bool raz)
1286 {
1287 	const u64 id = sys_reg_to_index(rd);
1288 	int err;
1289 	u64 val;
1290 
1291 	err = reg_from_user(&val, uaddr, id);
1292 	if (err)
1293 		return err;
1294 
1295 	/* This is what we mean by invariant: you can't change it. */
1296 	if (val != read_id_reg(vcpu, rd, raz))
1297 		return -EINVAL;
1298 
1299 	return 0;
1300 }
1301 
1302 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1303 		      const struct kvm_one_reg *reg, void __user *uaddr)
1304 {
1305 	return __get_id_reg(vcpu, rd, uaddr, false);
1306 }
1307 
1308 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1309 		      const struct kvm_one_reg *reg, void __user *uaddr)
1310 {
1311 	return __set_id_reg(vcpu, rd, uaddr, false);
1312 }
1313 
1314 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1315 			  const struct kvm_one_reg *reg, void __user *uaddr)
1316 {
1317 	return __get_id_reg(vcpu, rd, uaddr, true);
1318 }
1319 
1320 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1321 			  const struct kvm_one_reg *reg, void __user *uaddr)
1322 {
1323 	return __set_id_reg(vcpu, rd, uaddr, true);
1324 }
1325 
1326 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1327 		       const struct sys_reg_desc *r)
1328 {
1329 	if (p->is_write)
1330 		return write_to_read_only(vcpu, p, r);
1331 
1332 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1333 	return true;
1334 }
1335 
1336 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1337 			 const struct sys_reg_desc *r)
1338 {
1339 	if (p->is_write)
1340 		return write_to_read_only(vcpu, p, r);
1341 
1342 	p->regval = read_sysreg(clidr_el1);
1343 	return true;
1344 }
1345 
1346 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1347 			  const struct sys_reg_desc *r)
1348 {
1349 	int reg = r->reg;
1350 
1351 	/* See the 32bit mapping in kvm_host.h */
1352 	if (p->is_aarch32)
1353 		reg = r->reg / 2;
1354 
1355 	if (p->is_write)
1356 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1357 	else
1358 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1359 	return true;
1360 }
1361 
1362 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1363 			  const struct sys_reg_desc *r)
1364 {
1365 	u32 csselr;
1366 
1367 	if (p->is_write)
1368 		return write_to_read_only(vcpu, p, r);
1369 
1370 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1371 	p->regval = get_ccsidr(csselr);
1372 
1373 	/*
1374 	 * Guests should not be doing cache operations by set/way at all, and
1375 	 * for this reason, we trap them and attempt to infer the intent, so
1376 	 * that we can flush the entire guest's address space at the appropriate
1377 	 * time.
1378 	 * To prevent this trapping from causing performance problems, let's
1379 	 * expose the geometry of all data and unified caches (which are
1380 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1381 	 * [If guests should attempt to infer aliasing properties from the
1382 	 * geometry (which is not permitted by the architecture), they would
1383 	 * only do so for virtually indexed caches.]
1384 	 */
1385 	if (!(csselr & 1)) // data or unified cache
1386 		p->regval &= ~GENMASK(27, 3);
1387 	return true;
1388 }
1389 
1390 static bool access_mte_regs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1391 			    const struct sys_reg_desc *r)
1392 {
1393 	kvm_inject_undefined(vcpu);
1394 	return false;
1395 }
1396 
1397 /* sys_reg_desc initialiser for known cpufeature ID registers */
1398 #define ID_SANITISED(name) {			\
1399 	SYS_DESC(SYS_##name),			\
1400 	.access	= access_id_reg,		\
1401 	.get_user = get_id_reg,			\
1402 	.set_user = set_id_reg,			\
1403 }
1404 
1405 /*
1406  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1407  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1408  * (1 <= crm < 8, 0 <= Op2 < 8).
1409  */
1410 #define ID_UNALLOCATED(crm, op2) {			\
1411 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1412 	.access = access_raz_id_reg,			\
1413 	.get_user = get_raz_id_reg,			\
1414 	.set_user = set_raz_id_reg,			\
1415 }
1416 
1417 /*
1418  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1419  * For now, these are exposed just like unallocated ID regs: they appear
1420  * RAZ for the guest.
1421  */
1422 #define ID_HIDDEN(name) {			\
1423 	SYS_DESC(SYS_##name),			\
1424 	.access = access_raz_id_reg,		\
1425 	.get_user = get_raz_id_reg,		\
1426 	.set_user = set_raz_id_reg,		\
1427 }
1428 
1429 /*
1430  * Architected system registers.
1431  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1432  *
1433  * Debug handling: We do trap most, if not all debug related system
1434  * registers. The implementation is good enough to ensure that a guest
1435  * can use these with minimal performance degradation. The drawback is
1436  * that we don't implement any of the external debug, none of the
1437  * OSlock protocol. This should be revisited if we ever encounter a
1438  * more demanding guest...
1439  */
1440 static const struct sys_reg_desc sys_reg_descs[] = {
1441 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1442 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1443 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1444 
1445 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1446 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1447 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1448 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1449 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1450 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1451 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1452 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1453 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1454 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1455 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1456 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1457 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1458 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1459 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1460 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1461 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1462 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1463 
1464 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1465 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1466 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1467 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1468 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1469 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1470 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1471 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1472 
1473 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1474 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1475 	// DBGDTR[TR]X_EL0 share the same encoding
1476 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1477 
1478 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1479 
1480 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1481 
1482 	/*
1483 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1484 	 * entries in arm64_ftr_regs[].
1485 	 */
1486 
1487 	/* AArch64 mappings of the AArch32 ID registers */
1488 	/* CRm=1 */
1489 	ID_SANITISED(ID_PFR0_EL1),
1490 	ID_SANITISED(ID_PFR1_EL1),
1491 	ID_SANITISED(ID_DFR0_EL1),
1492 	ID_HIDDEN(ID_AFR0_EL1),
1493 	ID_SANITISED(ID_MMFR0_EL1),
1494 	ID_SANITISED(ID_MMFR1_EL1),
1495 	ID_SANITISED(ID_MMFR2_EL1),
1496 	ID_SANITISED(ID_MMFR3_EL1),
1497 
1498 	/* CRm=2 */
1499 	ID_SANITISED(ID_ISAR0_EL1),
1500 	ID_SANITISED(ID_ISAR1_EL1),
1501 	ID_SANITISED(ID_ISAR2_EL1),
1502 	ID_SANITISED(ID_ISAR3_EL1),
1503 	ID_SANITISED(ID_ISAR4_EL1),
1504 	ID_SANITISED(ID_ISAR5_EL1),
1505 	ID_SANITISED(ID_MMFR4_EL1),
1506 	ID_SANITISED(ID_ISAR6_EL1),
1507 
1508 	/* CRm=3 */
1509 	ID_SANITISED(MVFR0_EL1),
1510 	ID_SANITISED(MVFR1_EL1),
1511 	ID_SANITISED(MVFR2_EL1),
1512 	ID_UNALLOCATED(3,3),
1513 	ID_SANITISED(ID_PFR2_EL1),
1514 	ID_HIDDEN(ID_DFR1_EL1),
1515 	ID_SANITISED(ID_MMFR5_EL1),
1516 	ID_UNALLOCATED(3,7),
1517 
1518 	/* AArch64 ID registers */
1519 	/* CRm=4 */
1520 	ID_SANITISED(ID_AA64PFR0_EL1),
1521 	ID_SANITISED(ID_AA64PFR1_EL1),
1522 	ID_UNALLOCATED(4,2),
1523 	ID_UNALLOCATED(4,3),
1524 	{ SYS_DESC(SYS_ID_AA64ZFR0_EL1), access_id_aa64zfr0_el1, .get_user = get_id_aa64zfr0_el1, .set_user = set_id_aa64zfr0_el1, .visibility = sve_id_visibility },
1525 	ID_UNALLOCATED(4,5),
1526 	ID_UNALLOCATED(4,6),
1527 	ID_UNALLOCATED(4,7),
1528 
1529 	/* CRm=5 */
1530 	ID_SANITISED(ID_AA64DFR0_EL1),
1531 	ID_SANITISED(ID_AA64DFR1_EL1),
1532 	ID_UNALLOCATED(5,2),
1533 	ID_UNALLOCATED(5,3),
1534 	ID_HIDDEN(ID_AA64AFR0_EL1),
1535 	ID_HIDDEN(ID_AA64AFR1_EL1),
1536 	ID_UNALLOCATED(5,6),
1537 	ID_UNALLOCATED(5,7),
1538 
1539 	/* CRm=6 */
1540 	ID_SANITISED(ID_AA64ISAR0_EL1),
1541 	ID_SANITISED(ID_AA64ISAR1_EL1),
1542 	ID_UNALLOCATED(6,2),
1543 	ID_UNALLOCATED(6,3),
1544 	ID_UNALLOCATED(6,4),
1545 	ID_UNALLOCATED(6,5),
1546 	ID_UNALLOCATED(6,6),
1547 	ID_UNALLOCATED(6,7),
1548 
1549 	/* CRm=7 */
1550 	ID_SANITISED(ID_AA64MMFR0_EL1),
1551 	ID_SANITISED(ID_AA64MMFR1_EL1),
1552 	ID_SANITISED(ID_AA64MMFR2_EL1),
1553 	ID_UNALLOCATED(7,3),
1554 	ID_UNALLOCATED(7,4),
1555 	ID_UNALLOCATED(7,5),
1556 	ID_UNALLOCATED(7,6),
1557 	ID_UNALLOCATED(7,7),
1558 
1559 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1560 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1561 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1562 
1563 	{ SYS_DESC(SYS_RGSR_EL1), access_mte_regs },
1564 	{ SYS_DESC(SYS_GCR_EL1), access_mte_regs },
1565 
1566 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1567 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1568 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1569 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1570 
1571 	PTRAUTH_KEY(APIA),
1572 	PTRAUTH_KEY(APIB),
1573 	PTRAUTH_KEY(APDA),
1574 	PTRAUTH_KEY(APDB),
1575 	PTRAUTH_KEY(APGA),
1576 
1577 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1578 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1579 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1580 
1581 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1582 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1583 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1584 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1585 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1586 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1587 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1588 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1589 
1590 	{ SYS_DESC(SYS_TFSR_EL1), access_mte_regs },
1591 	{ SYS_DESC(SYS_TFSRE0_EL1), access_mte_regs },
1592 
1593 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1594 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1595 
1596 	{ SYS_DESC(SYS_PMINTENSET_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1597 	{ SYS_DESC(SYS_PMINTENCLR_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1598 
1599 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1600 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1601 
1602 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1603 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1604 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1605 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1606 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1607 
1608 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1609 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1610 
1611 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1612 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1613 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1614 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1615 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1616 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1617 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1618 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1619 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1620 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1621 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1622 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1623 
1624 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1625 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1626 
1627 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1628 
1629 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1630 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1631 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1632 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1633 
1634 	{ SYS_DESC(SYS_PMCR_EL0), access_pmcr, reset_pmcr, PMCR_EL0 },
1635 	{ SYS_DESC(SYS_PMCNTENSET_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1636 	{ SYS_DESC(SYS_PMCNTENCLR_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1637 	{ SYS_DESC(SYS_PMOVSCLR_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1638 	{ SYS_DESC(SYS_PMSWINC_EL0), access_pmswinc, reset_unknown, PMSWINC_EL0 },
1639 	{ SYS_DESC(SYS_PMSELR_EL0), access_pmselr, reset_unknown, PMSELR_EL0 },
1640 	{ SYS_DESC(SYS_PMCEID0_EL0), access_pmceid },
1641 	{ SYS_DESC(SYS_PMCEID1_EL0), access_pmceid },
1642 	{ SYS_DESC(SYS_PMCCNTR_EL0), access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1643 	{ SYS_DESC(SYS_PMXEVTYPER_EL0), access_pmu_evtyper },
1644 	{ SYS_DESC(SYS_PMXEVCNTR_EL0), access_pmu_evcntr },
1645 	/*
1646 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1647 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1648 	 */
1649 	{ SYS_DESC(SYS_PMUSERENR_EL0), access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1650 	{ SYS_DESC(SYS_PMOVSSET_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1651 
1652 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1653 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1654 
1655 	{ SYS_DESC(SYS_AMCR_EL0), access_amu },
1656 	{ SYS_DESC(SYS_AMCFGR_EL0), access_amu },
1657 	{ SYS_DESC(SYS_AMCGCR_EL0), access_amu },
1658 	{ SYS_DESC(SYS_AMUSERENR_EL0), access_amu },
1659 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), access_amu },
1660 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), access_amu },
1661 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), access_amu },
1662 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), access_amu },
1663 	AMU_AMEVCNTR0_EL0(0),
1664 	AMU_AMEVCNTR0_EL0(1),
1665 	AMU_AMEVCNTR0_EL0(2),
1666 	AMU_AMEVCNTR0_EL0(3),
1667 	AMU_AMEVCNTR0_EL0(4),
1668 	AMU_AMEVCNTR0_EL0(5),
1669 	AMU_AMEVCNTR0_EL0(6),
1670 	AMU_AMEVCNTR0_EL0(7),
1671 	AMU_AMEVCNTR0_EL0(8),
1672 	AMU_AMEVCNTR0_EL0(9),
1673 	AMU_AMEVCNTR0_EL0(10),
1674 	AMU_AMEVCNTR0_EL0(11),
1675 	AMU_AMEVCNTR0_EL0(12),
1676 	AMU_AMEVCNTR0_EL0(13),
1677 	AMU_AMEVCNTR0_EL0(14),
1678 	AMU_AMEVCNTR0_EL0(15),
1679 	AMU_AMEVTYPER0_EL0(0),
1680 	AMU_AMEVTYPER0_EL0(1),
1681 	AMU_AMEVTYPER0_EL0(2),
1682 	AMU_AMEVTYPER0_EL0(3),
1683 	AMU_AMEVTYPER0_EL0(4),
1684 	AMU_AMEVTYPER0_EL0(5),
1685 	AMU_AMEVTYPER0_EL0(6),
1686 	AMU_AMEVTYPER0_EL0(7),
1687 	AMU_AMEVTYPER0_EL0(8),
1688 	AMU_AMEVTYPER0_EL0(9),
1689 	AMU_AMEVTYPER0_EL0(10),
1690 	AMU_AMEVTYPER0_EL0(11),
1691 	AMU_AMEVTYPER0_EL0(12),
1692 	AMU_AMEVTYPER0_EL0(13),
1693 	AMU_AMEVTYPER0_EL0(14),
1694 	AMU_AMEVTYPER0_EL0(15),
1695 	AMU_AMEVCNTR1_EL0(0),
1696 	AMU_AMEVCNTR1_EL0(1),
1697 	AMU_AMEVCNTR1_EL0(2),
1698 	AMU_AMEVCNTR1_EL0(3),
1699 	AMU_AMEVCNTR1_EL0(4),
1700 	AMU_AMEVCNTR1_EL0(5),
1701 	AMU_AMEVCNTR1_EL0(6),
1702 	AMU_AMEVCNTR1_EL0(7),
1703 	AMU_AMEVCNTR1_EL0(8),
1704 	AMU_AMEVCNTR1_EL0(9),
1705 	AMU_AMEVCNTR1_EL0(10),
1706 	AMU_AMEVCNTR1_EL0(11),
1707 	AMU_AMEVCNTR1_EL0(12),
1708 	AMU_AMEVCNTR1_EL0(13),
1709 	AMU_AMEVCNTR1_EL0(14),
1710 	AMU_AMEVCNTR1_EL0(15),
1711 	AMU_AMEVTYPER1_EL0(0),
1712 	AMU_AMEVTYPER1_EL0(1),
1713 	AMU_AMEVTYPER1_EL0(2),
1714 	AMU_AMEVTYPER1_EL0(3),
1715 	AMU_AMEVTYPER1_EL0(4),
1716 	AMU_AMEVTYPER1_EL0(5),
1717 	AMU_AMEVTYPER1_EL0(6),
1718 	AMU_AMEVTYPER1_EL0(7),
1719 	AMU_AMEVTYPER1_EL0(8),
1720 	AMU_AMEVTYPER1_EL0(9),
1721 	AMU_AMEVTYPER1_EL0(10),
1722 	AMU_AMEVTYPER1_EL0(11),
1723 	AMU_AMEVTYPER1_EL0(12),
1724 	AMU_AMEVTYPER1_EL0(13),
1725 	AMU_AMEVTYPER1_EL0(14),
1726 	AMU_AMEVTYPER1_EL0(15),
1727 
1728 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1729 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1730 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1731 
1732 	/* PMEVCNTRn_EL0 */
1733 	PMU_PMEVCNTR_EL0(0),
1734 	PMU_PMEVCNTR_EL0(1),
1735 	PMU_PMEVCNTR_EL0(2),
1736 	PMU_PMEVCNTR_EL0(3),
1737 	PMU_PMEVCNTR_EL0(4),
1738 	PMU_PMEVCNTR_EL0(5),
1739 	PMU_PMEVCNTR_EL0(6),
1740 	PMU_PMEVCNTR_EL0(7),
1741 	PMU_PMEVCNTR_EL0(8),
1742 	PMU_PMEVCNTR_EL0(9),
1743 	PMU_PMEVCNTR_EL0(10),
1744 	PMU_PMEVCNTR_EL0(11),
1745 	PMU_PMEVCNTR_EL0(12),
1746 	PMU_PMEVCNTR_EL0(13),
1747 	PMU_PMEVCNTR_EL0(14),
1748 	PMU_PMEVCNTR_EL0(15),
1749 	PMU_PMEVCNTR_EL0(16),
1750 	PMU_PMEVCNTR_EL0(17),
1751 	PMU_PMEVCNTR_EL0(18),
1752 	PMU_PMEVCNTR_EL0(19),
1753 	PMU_PMEVCNTR_EL0(20),
1754 	PMU_PMEVCNTR_EL0(21),
1755 	PMU_PMEVCNTR_EL0(22),
1756 	PMU_PMEVCNTR_EL0(23),
1757 	PMU_PMEVCNTR_EL0(24),
1758 	PMU_PMEVCNTR_EL0(25),
1759 	PMU_PMEVCNTR_EL0(26),
1760 	PMU_PMEVCNTR_EL0(27),
1761 	PMU_PMEVCNTR_EL0(28),
1762 	PMU_PMEVCNTR_EL0(29),
1763 	PMU_PMEVCNTR_EL0(30),
1764 	/* PMEVTYPERn_EL0 */
1765 	PMU_PMEVTYPER_EL0(0),
1766 	PMU_PMEVTYPER_EL0(1),
1767 	PMU_PMEVTYPER_EL0(2),
1768 	PMU_PMEVTYPER_EL0(3),
1769 	PMU_PMEVTYPER_EL0(4),
1770 	PMU_PMEVTYPER_EL0(5),
1771 	PMU_PMEVTYPER_EL0(6),
1772 	PMU_PMEVTYPER_EL0(7),
1773 	PMU_PMEVTYPER_EL0(8),
1774 	PMU_PMEVTYPER_EL0(9),
1775 	PMU_PMEVTYPER_EL0(10),
1776 	PMU_PMEVTYPER_EL0(11),
1777 	PMU_PMEVTYPER_EL0(12),
1778 	PMU_PMEVTYPER_EL0(13),
1779 	PMU_PMEVTYPER_EL0(14),
1780 	PMU_PMEVTYPER_EL0(15),
1781 	PMU_PMEVTYPER_EL0(16),
1782 	PMU_PMEVTYPER_EL0(17),
1783 	PMU_PMEVTYPER_EL0(18),
1784 	PMU_PMEVTYPER_EL0(19),
1785 	PMU_PMEVTYPER_EL0(20),
1786 	PMU_PMEVTYPER_EL0(21),
1787 	PMU_PMEVTYPER_EL0(22),
1788 	PMU_PMEVTYPER_EL0(23),
1789 	PMU_PMEVTYPER_EL0(24),
1790 	PMU_PMEVTYPER_EL0(25),
1791 	PMU_PMEVTYPER_EL0(26),
1792 	PMU_PMEVTYPER_EL0(27),
1793 	PMU_PMEVTYPER_EL0(28),
1794 	PMU_PMEVTYPER_EL0(29),
1795 	PMU_PMEVTYPER_EL0(30),
1796 	/*
1797 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1798 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1799 	 */
1800 	{ SYS_DESC(SYS_PMCCFILTR_EL0), access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1801 
1802 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1803 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1804 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1805 };
1806 
1807 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1808 			struct sys_reg_params *p,
1809 			const struct sys_reg_desc *r)
1810 {
1811 	if (p->is_write) {
1812 		return ignore_write(vcpu, p);
1813 	} else {
1814 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1815 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1816 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1817 
1818 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1819 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1820 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1821 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1822 		return true;
1823 	}
1824 }
1825 
1826 static bool trap_debug32(struct kvm_vcpu *vcpu,
1827 			 struct sys_reg_params *p,
1828 			 const struct sys_reg_desc *r)
1829 {
1830 	if (p->is_write) {
1831 		vcpu_cp14(vcpu, r->reg) = p->regval;
1832 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1833 	} else {
1834 		p->regval = vcpu_cp14(vcpu, r->reg);
1835 	}
1836 
1837 	return true;
1838 }
1839 
1840 /* AArch32 debug register mappings
1841  *
1842  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1843  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1844  *
1845  * All control registers and watchpoint value registers are mapped to
1846  * the lower 32 bits of their AArch64 equivalents. We share the trap
1847  * handlers with the above AArch64 code which checks what mode the
1848  * system is in.
1849  */
1850 
1851 static bool trap_xvr(struct kvm_vcpu *vcpu,
1852 		     struct sys_reg_params *p,
1853 		     const struct sys_reg_desc *rd)
1854 {
1855 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1856 
1857 	if (p->is_write) {
1858 		u64 val = *dbg_reg;
1859 
1860 		val &= 0xffffffffUL;
1861 		val |= p->regval << 32;
1862 		*dbg_reg = val;
1863 
1864 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1865 	} else {
1866 		p->regval = *dbg_reg >> 32;
1867 	}
1868 
1869 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1870 
1871 	return true;
1872 }
1873 
1874 #define DBG_BCR_BVR_WCR_WVR(n)						\
1875 	/* DBGBVRn */							\
1876 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1877 	/* DBGBCRn */							\
1878 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1879 	/* DBGWVRn */							\
1880 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1881 	/* DBGWCRn */							\
1882 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1883 
1884 #define DBGBXVR(n)							\
1885 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1886 
1887 /*
1888  * Trapped cp14 registers. We generally ignore most of the external
1889  * debug, on the principle that they don't really make sense to a
1890  * guest. Revisit this one day, would this principle change.
1891  */
1892 static const struct sys_reg_desc cp14_regs[] = {
1893 	/* DBGIDR */
1894 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1895 	/* DBGDTRRXext */
1896 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1897 
1898 	DBG_BCR_BVR_WCR_WVR(0),
1899 	/* DBGDSCRint */
1900 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1901 	DBG_BCR_BVR_WCR_WVR(1),
1902 	/* DBGDCCINT */
1903 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1904 	/* DBGDSCRext */
1905 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1906 	DBG_BCR_BVR_WCR_WVR(2),
1907 	/* DBGDTR[RT]Xint */
1908 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1909 	/* DBGDTR[RT]Xext */
1910 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1911 	DBG_BCR_BVR_WCR_WVR(3),
1912 	DBG_BCR_BVR_WCR_WVR(4),
1913 	DBG_BCR_BVR_WCR_WVR(5),
1914 	/* DBGWFAR */
1915 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1916 	/* DBGOSECCR */
1917 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1918 	DBG_BCR_BVR_WCR_WVR(6),
1919 	/* DBGVCR */
1920 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1921 	DBG_BCR_BVR_WCR_WVR(7),
1922 	DBG_BCR_BVR_WCR_WVR(8),
1923 	DBG_BCR_BVR_WCR_WVR(9),
1924 	DBG_BCR_BVR_WCR_WVR(10),
1925 	DBG_BCR_BVR_WCR_WVR(11),
1926 	DBG_BCR_BVR_WCR_WVR(12),
1927 	DBG_BCR_BVR_WCR_WVR(13),
1928 	DBG_BCR_BVR_WCR_WVR(14),
1929 	DBG_BCR_BVR_WCR_WVR(15),
1930 
1931 	/* DBGDRAR (32bit) */
1932 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1933 
1934 	DBGBXVR(0),
1935 	/* DBGOSLAR */
1936 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1937 	DBGBXVR(1),
1938 	/* DBGOSLSR */
1939 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1940 	DBGBXVR(2),
1941 	DBGBXVR(3),
1942 	/* DBGOSDLR */
1943 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1944 	DBGBXVR(4),
1945 	/* DBGPRCR */
1946 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1947 	DBGBXVR(5),
1948 	DBGBXVR(6),
1949 	DBGBXVR(7),
1950 	DBGBXVR(8),
1951 	DBGBXVR(9),
1952 	DBGBXVR(10),
1953 	DBGBXVR(11),
1954 	DBGBXVR(12),
1955 	DBGBXVR(13),
1956 	DBGBXVR(14),
1957 	DBGBXVR(15),
1958 
1959 	/* DBGDSAR (32bit) */
1960 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1961 
1962 	/* DBGDEVID2 */
1963 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1964 	/* DBGDEVID1 */
1965 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1966 	/* DBGDEVID */
1967 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1968 	/* DBGCLAIMSET */
1969 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1970 	/* DBGCLAIMCLR */
1971 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1972 	/* DBGAUTHSTATUS */
1973 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1974 };
1975 
1976 /* Trapped cp14 64bit registers */
1977 static const struct sys_reg_desc cp14_64_regs[] = {
1978 	/* DBGDRAR (64bit) */
1979 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1980 
1981 	/* DBGDSAR (64bit) */
1982 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1983 };
1984 
1985 /* Macro to expand the PMEVCNTRn register */
1986 #define PMU_PMEVCNTR(n)							\
1987 	/* PMEVCNTRn */							\
1988 	{ Op1(0), CRn(0b1110),						\
1989 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1990 	  access_pmu_evcntr }
1991 
1992 /* Macro to expand the PMEVTYPERn register */
1993 #define PMU_PMEVTYPER(n)						\
1994 	/* PMEVTYPERn */						\
1995 	{ Op1(0), CRn(0b1110),						\
1996 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1997 	  access_pmu_evtyper }
1998 
1999 /*
2000  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
2001  * depending on the way they are accessed (as a 32bit or a 64bit
2002  * register).
2003  */
2004 static const struct sys_reg_desc cp15_regs[] = {
2005 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
2006 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
2007 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr },
2008 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr },
2009 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
2010 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
2011 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
2012 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
2013 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
2014 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
2015 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
2016 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
2017 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
2018 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
2019 
2020 	/*
2021 	 * DC{C,I,CI}SW operations:
2022 	 */
2023 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
2024 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
2025 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
2026 
2027 	/* PMU */
2028 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
2029 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
2030 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
2031 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
2032 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
2033 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
2034 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
2035 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
2036 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
2037 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
2038 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
2039 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
2040 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
2041 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
2042 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
2043 
2044 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
2045 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
2046 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
2047 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
2048 
2049 	/* ICC_SRE */
2050 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2051 
2052 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
2053 
2054 	/* Arch Tmers */
2055 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2056 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2057 
2058 	/* PMEVCNTRn */
2059 	PMU_PMEVCNTR(0),
2060 	PMU_PMEVCNTR(1),
2061 	PMU_PMEVCNTR(2),
2062 	PMU_PMEVCNTR(3),
2063 	PMU_PMEVCNTR(4),
2064 	PMU_PMEVCNTR(5),
2065 	PMU_PMEVCNTR(6),
2066 	PMU_PMEVCNTR(7),
2067 	PMU_PMEVCNTR(8),
2068 	PMU_PMEVCNTR(9),
2069 	PMU_PMEVCNTR(10),
2070 	PMU_PMEVCNTR(11),
2071 	PMU_PMEVCNTR(12),
2072 	PMU_PMEVCNTR(13),
2073 	PMU_PMEVCNTR(14),
2074 	PMU_PMEVCNTR(15),
2075 	PMU_PMEVCNTR(16),
2076 	PMU_PMEVCNTR(17),
2077 	PMU_PMEVCNTR(18),
2078 	PMU_PMEVCNTR(19),
2079 	PMU_PMEVCNTR(20),
2080 	PMU_PMEVCNTR(21),
2081 	PMU_PMEVCNTR(22),
2082 	PMU_PMEVCNTR(23),
2083 	PMU_PMEVCNTR(24),
2084 	PMU_PMEVCNTR(25),
2085 	PMU_PMEVCNTR(26),
2086 	PMU_PMEVCNTR(27),
2087 	PMU_PMEVCNTR(28),
2088 	PMU_PMEVCNTR(29),
2089 	PMU_PMEVCNTR(30),
2090 	/* PMEVTYPERn */
2091 	PMU_PMEVTYPER(0),
2092 	PMU_PMEVTYPER(1),
2093 	PMU_PMEVTYPER(2),
2094 	PMU_PMEVTYPER(3),
2095 	PMU_PMEVTYPER(4),
2096 	PMU_PMEVTYPER(5),
2097 	PMU_PMEVTYPER(6),
2098 	PMU_PMEVTYPER(7),
2099 	PMU_PMEVTYPER(8),
2100 	PMU_PMEVTYPER(9),
2101 	PMU_PMEVTYPER(10),
2102 	PMU_PMEVTYPER(11),
2103 	PMU_PMEVTYPER(12),
2104 	PMU_PMEVTYPER(13),
2105 	PMU_PMEVTYPER(14),
2106 	PMU_PMEVTYPER(15),
2107 	PMU_PMEVTYPER(16),
2108 	PMU_PMEVTYPER(17),
2109 	PMU_PMEVTYPER(18),
2110 	PMU_PMEVTYPER(19),
2111 	PMU_PMEVTYPER(20),
2112 	PMU_PMEVTYPER(21),
2113 	PMU_PMEVTYPER(22),
2114 	PMU_PMEVTYPER(23),
2115 	PMU_PMEVTYPER(24),
2116 	PMU_PMEVTYPER(25),
2117 	PMU_PMEVTYPER(26),
2118 	PMU_PMEVTYPER(27),
2119 	PMU_PMEVTYPER(28),
2120 	PMU_PMEVTYPER(29),
2121 	PMU_PMEVTYPER(30),
2122 	/* PMCCFILTR */
2123 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2124 
2125 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2126 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2127 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, c0_CSSELR },
2128 };
2129 
2130 static const struct sys_reg_desc cp15_64_regs[] = {
2131 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
2132 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2133 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2134 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
2135 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2136 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2137 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2138 };
2139 
2140 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2141 			      bool is_32)
2142 {
2143 	unsigned int i;
2144 
2145 	for (i = 0; i < n; i++) {
2146 		if (!is_32 && table[i].reg && !table[i].reset) {
2147 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2148 				table, i);
2149 			return 1;
2150 		}
2151 
2152 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2153 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2154 			return 1;
2155 		}
2156 	}
2157 
2158 	return 0;
2159 }
2160 
2161 static int match_sys_reg(const void *key, const void *elt)
2162 {
2163 	const unsigned long pval = (unsigned long)key;
2164 	const struct sys_reg_desc *r = elt;
2165 
2166 	return pval - reg_to_encoding(r);
2167 }
2168 
2169 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
2170 					 const struct sys_reg_desc table[],
2171 					 unsigned int num)
2172 {
2173 	unsigned long pval = reg_to_encoding(params);
2174 
2175 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
2176 }
2177 
2178 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2179 {
2180 	kvm_inject_undefined(vcpu);
2181 	return 1;
2182 }
2183 
2184 static void perform_access(struct kvm_vcpu *vcpu,
2185 			   struct sys_reg_params *params,
2186 			   const struct sys_reg_desc *r)
2187 {
2188 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2189 
2190 	/* Check for regs disabled by runtime config */
2191 	if (sysreg_hidden_from_guest(vcpu, r)) {
2192 		kvm_inject_undefined(vcpu);
2193 		return;
2194 	}
2195 
2196 	/*
2197 	 * Not having an accessor means that we have configured a trap
2198 	 * that we don't know how to handle. This certainly qualifies
2199 	 * as a gross bug that should be fixed right away.
2200 	 */
2201 	BUG_ON(!r->access);
2202 
2203 	/* Skip instruction if instructed so */
2204 	if (likely(r->access(vcpu, params, r)))
2205 		kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
2206 }
2207 
2208 /*
2209  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2210  *                call the corresponding trap handler.
2211  *
2212  * @params: pointer to the descriptor of the access
2213  * @table: array of trap descriptors
2214  * @num: size of the trap descriptor array
2215  *
2216  * Return 0 if the access has been handled, and -1 if not.
2217  */
2218 static int emulate_cp(struct kvm_vcpu *vcpu,
2219 		      struct sys_reg_params *params,
2220 		      const struct sys_reg_desc *table,
2221 		      size_t num)
2222 {
2223 	const struct sys_reg_desc *r;
2224 
2225 	if (!table)
2226 		return -1;	/* Not handled */
2227 
2228 	r = find_reg(params, table, num);
2229 
2230 	if (r) {
2231 		perform_access(vcpu, params, r);
2232 		return 0;
2233 	}
2234 
2235 	/* Not handled */
2236 	return -1;
2237 }
2238 
2239 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2240 				struct sys_reg_params *params)
2241 {
2242 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2243 	int cp = -1;
2244 
2245 	switch (esr_ec) {
2246 	case ESR_ELx_EC_CP15_32:
2247 	case ESR_ELx_EC_CP15_64:
2248 		cp = 15;
2249 		break;
2250 	case ESR_ELx_EC_CP14_MR:
2251 	case ESR_ELx_EC_CP14_64:
2252 		cp = 14;
2253 		break;
2254 	default:
2255 		WARN_ON(1);
2256 	}
2257 
2258 	print_sys_reg_msg(params,
2259 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2260 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2261 	kvm_inject_undefined(vcpu);
2262 }
2263 
2264 /**
2265  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2266  * @vcpu: The VCPU pointer
2267  * @run:  The kvm_run struct
2268  */
2269 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2270 			    const struct sys_reg_desc *global,
2271 			    size_t nr_global)
2272 {
2273 	struct sys_reg_params params;
2274 	u32 esr = kvm_vcpu_get_esr(vcpu);
2275 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2276 	int Rt2 = (esr >> 10) & 0x1f;
2277 
2278 	params.is_aarch32 = true;
2279 	params.is_32bit = false;
2280 	params.CRm = (esr >> 1) & 0xf;
2281 	params.is_write = ((esr & 1) == 0);
2282 
2283 	params.Op0 = 0;
2284 	params.Op1 = (esr >> 16) & 0xf;
2285 	params.Op2 = 0;
2286 	params.CRn = 0;
2287 
2288 	/*
2289 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2290 	 * backends between AArch32 and AArch64, we get away with it.
2291 	 */
2292 	if (params.is_write) {
2293 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2294 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2295 	}
2296 
2297 	/*
2298 	 * If the table contains a handler, handle the
2299 	 * potential register operation in the case of a read and return
2300 	 * with success.
2301 	 */
2302 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2303 		/* Split up the value between registers for the read side */
2304 		if (!params.is_write) {
2305 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2306 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2307 		}
2308 
2309 		return 1;
2310 	}
2311 
2312 	unhandled_cp_access(vcpu, &params);
2313 	return 1;
2314 }
2315 
2316 /**
2317  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2318  * @vcpu: The VCPU pointer
2319  * @run:  The kvm_run struct
2320  */
2321 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2322 			    const struct sys_reg_desc *global,
2323 			    size_t nr_global)
2324 {
2325 	struct sys_reg_params params;
2326 	u32 esr = kvm_vcpu_get_esr(vcpu);
2327 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2328 
2329 	params.is_aarch32 = true;
2330 	params.is_32bit = true;
2331 	params.CRm = (esr >> 1) & 0xf;
2332 	params.regval = vcpu_get_reg(vcpu, Rt);
2333 	params.is_write = ((esr & 1) == 0);
2334 	params.CRn = (esr >> 10) & 0xf;
2335 	params.Op0 = 0;
2336 	params.Op1 = (esr >> 14) & 0x7;
2337 	params.Op2 = (esr >> 17) & 0x7;
2338 
2339 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2340 		if (!params.is_write)
2341 			vcpu_set_reg(vcpu, Rt, params.regval);
2342 		return 1;
2343 	}
2344 
2345 	unhandled_cp_access(vcpu, &params);
2346 	return 1;
2347 }
2348 
2349 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2350 {
2351 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2352 }
2353 
2354 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2355 {
2356 	return kvm_handle_cp_32(vcpu, cp15_regs, ARRAY_SIZE(cp15_regs));
2357 }
2358 
2359 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2360 {
2361 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2362 }
2363 
2364 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2365 {
2366 	return kvm_handle_cp_32(vcpu, cp14_regs, ARRAY_SIZE(cp14_regs));
2367 }
2368 
2369 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2370 {
2371 	// See ARM DDI 0487E.a, section D12.3.2
2372 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2373 }
2374 
2375 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2376 			   struct sys_reg_params *params)
2377 {
2378 	const struct sys_reg_desc *r;
2379 
2380 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2381 
2382 	if (likely(r)) {
2383 		perform_access(vcpu, params, r);
2384 	} else if (is_imp_def_sys_reg(params)) {
2385 		kvm_inject_undefined(vcpu);
2386 	} else {
2387 		print_sys_reg_msg(params,
2388 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2389 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2390 		kvm_inject_undefined(vcpu);
2391 	}
2392 	return 1;
2393 }
2394 
2395 /**
2396  * kvm_reset_sys_regs - sets system registers to reset value
2397  * @vcpu: The VCPU pointer
2398  *
2399  * This function finds the right table above and sets the registers on the
2400  * virtual CPU struct to their architecturally defined reset values.
2401  */
2402 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2403 {
2404 	unsigned long i;
2405 
2406 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2407 		if (sys_reg_descs[i].reset)
2408 			sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2409 }
2410 
2411 /**
2412  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2413  * @vcpu: The VCPU pointer
2414  */
2415 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2416 {
2417 	struct sys_reg_params params;
2418 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
2419 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2420 	int ret;
2421 
2422 	trace_kvm_handle_sys_reg(esr);
2423 
2424 	params.is_aarch32 = false;
2425 	params.is_32bit = false;
2426 	params.Op0 = (esr >> 20) & 3;
2427 	params.Op1 = (esr >> 14) & 0x7;
2428 	params.CRn = (esr >> 10) & 0xf;
2429 	params.CRm = (esr >> 1) & 0xf;
2430 	params.Op2 = (esr >> 17) & 0x7;
2431 	params.regval = vcpu_get_reg(vcpu, Rt);
2432 	params.is_write = !(esr & 1);
2433 
2434 	ret = emulate_sys_reg(vcpu, &params);
2435 
2436 	if (!params.is_write)
2437 		vcpu_set_reg(vcpu, Rt, params.regval);
2438 	return ret;
2439 }
2440 
2441 /******************************************************************************
2442  * Userspace API
2443  *****************************************************************************/
2444 
2445 static bool index_to_params(u64 id, struct sys_reg_params *params)
2446 {
2447 	switch (id & KVM_REG_SIZE_MASK) {
2448 	case KVM_REG_SIZE_U64:
2449 		/* Any unused index bits means it's not valid. */
2450 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2451 			      | KVM_REG_ARM_COPROC_MASK
2452 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2453 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2454 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2455 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2456 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2457 			return false;
2458 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2459 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2460 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2461 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2462 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2463 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2464 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2465 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2466 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2467 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2468 		return true;
2469 	default:
2470 		return false;
2471 	}
2472 }
2473 
2474 const struct sys_reg_desc *find_reg_by_id(u64 id,
2475 					  struct sys_reg_params *params,
2476 					  const struct sys_reg_desc table[],
2477 					  unsigned int num)
2478 {
2479 	if (!index_to_params(id, params))
2480 		return NULL;
2481 
2482 	return find_reg(params, table, num);
2483 }
2484 
2485 /* Decode an index value, and find the sys_reg_desc entry. */
2486 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2487 						    u64 id)
2488 {
2489 	const struct sys_reg_desc *r;
2490 	struct sys_reg_params params;
2491 
2492 	/* We only do sys_reg for now. */
2493 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2494 		return NULL;
2495 
2496 	if (!index_to_params(id, &params))
2497 		return NULL;
2498 
2499 	r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2500 
2501 	/* Not saved in the sys_reg array and not otherwise accessible? */
2502 	if (r && !(r->reg || r->get_user))
2503 		r = NULL;
2504 
2505 	return r;
2506 }
2507 
2508 /*
2509  * These are the invariant sys_reg registers: we let the guest see the
2510  * host versions of these, so they're part of the guest state.
2511  *
2512  * A future CPU may provide a mechanism to present different values to
2513  * the guest, or a future kvm may trap them.
2514  */
2515 
2516 #define FUNCTION_INVARIANT(reg)						\
2517 	static void get_##reg(struct kvm_vcpu *v,			\
2518 			      const struct sys_reg_desc *r)		\
2519 	{								\
2520 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2521 	}
2522 
2523 FUNCTION_INVARIANT(midr_el1)
2524 FUNCTION_INVARIANT(revidr_el1)
2525 FUNCTION_INVARIANT(clidr_el1)
2526 FUNCTION_INVARIANT(aidr_el1)
2527 
2528 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2529 {
2530 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2531 }
2532 
2533 /* ->val is filled in by kvm_sys_reg_table_init() */
2534 static struct sys_reg_desc invariant_sys_regs[] = {
2535 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2536 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2537 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2538 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2539 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2540 };
2541 
2542 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2543 {
2544 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2545 		return -EFAULT;
2546 	return 0;
2547 }
2548 
2549 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2550 {
2551 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2552 		return -EFAULT;
2553 	return 0;
2554 }
2555 
2556 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2557 {
2558 	struct sys_reg_params params;
2559 	const struct sys_reg_desc *r;
2560 
2561 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2562 			   ARRAY_SIZE(invariant_sys_regs));
2563 	if (!r)
2564 		return -ENOENT;
2565 
2566 	return reg_to_user(uaddr, &r->val, id);
2567 }
2568 
2569 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2570 {
2571 	struct sys_reg_params params;
2572 	const struct sys_reg_desc *r;
2573 	int err;
2574 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2575 
2576 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2577 			   ARRAY_SIZE(invariant_sys_regs));
2578 	if (!r)
2579 		return -ENOENT;
2580 
2581 	err = reg_from_user(&val, uaddr, id);
2582 	if (err)
2583 		return err;
2584 
2585 	/* This is what we mean by invariant: you can't change it. */
2586 	if (r->val != val)
2587 		return -EINVAL;
2588 
2589 	return 0;
2590 }
2591 
2592 static bool is_valid_cache(u32 val)
2593 {
2594 	u32 level, ctype;
2595 
2596 	if (val >= CSSELR_MAX)
2597 		return false;
2598 
2599 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2600 	level = (val >> 1);
2601 	ctype = (cache_levels >> (level * 3)) & 7;
2602 
2603 	switch (ctype) {
2604 	case 0: /* No cache */
2605 		return false;
2606 	case 1: /* Instruction cache only */
2607 		return (val & 1);
2608 	case 2: /* Data cache only */
2609 	case 4: /* Unified cache */
2610 		return !(val & 1);
2611 	case 3: /* Separate instruction and data caches */
2612 		return true;
2613 	default: /* Reserved: we can't know instruction or data. */
2614 		return false;
2615 	}
2616 }
2617 
2618 static int demux_c15_get(u64 id, void __user *uaddr)
2619 {
2620 	u32 val;
2621 	u32 __user *uval = uaddr;
2622 
2623 	/* Fail if we have unknown bits set. */
2624 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2625 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2626 		return -ENOENT;
2627 
2628 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2629 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2630 		if (KVM_REG_SIZE(id) != 4)
2631 			return -ENOENT;
2632 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2633 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2634 		if (!is_valid_cache(val))
2635 			return -ENOENT;
2636 
2637 		return put_user(get_ccsidr(val), uval);
2638 	default:
2639 		return -ENOENT;
2640 	}
2641 }
2642 
2643 static int demux_c15_set(u64 id, void __user *uaddr)
2644 {
2645 	u32 val, newval;
2646 	u32 __user *uval = uaddr;
2647 
2648 	/* Fail if we have unknown bits set. */
2649 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2650 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2651 		return -ENOENT;
2652 
2653 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2654 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2655 		if (KVM_REG_SIZE(id) != 4)
2656 			return -ENOENT;
2657 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2658 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2659 		if (!is_valid_cache(val))
2660 			return -ENOENT;
2661 
2662 		if (get_user(newval, uval))
2663 			return -EFAULT;
2664 
2665 		/* This is also invariant: you can't change it. */
2666 		if (newval != get_ccsidr(val))
2667 			return -EINVAL;
2668 		return 0;
2669 	default:
2670 		return -ENOENT;
2671 	}
2672 }
2673 
2674 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2675 {
2676 	const struct sys_reg_desc *r;
2677 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2678 
2679 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2680 		return demux_c15_get(reg->id, uaddr);
2681 
2682 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2683 		return -ENOENT;
2684 
2685 	r = index_to_sys_reg_desc(vcpu, reg->id);
2686 	if (!r)
2687 		return get_invariant_sys_reg(reg->id, uaddr);
2688 
2689 	/* Check for regs disabled by runtime config */
2690 	if (sysreg_hidden_from_user(vcpu, r))
2691 		return -ENOENT;
2692 
2693 	if (r->get_user)
2694 		return (r->get_user)(vcpu, r, reg, uaddr);
2695 
2696 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2697 }
2698 
2699 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2700 {
2701 	const struct sys_reg_desc *r;
2702 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2703 
2704 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2705 		return demux_c15_set(reg->id, uaddr);
2706 
2707 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2708 		return -ENOENT;
2709 
2710 	r = index_to_sys_reg_desc(vcpu, reg->id);
2711 	if (!r)
2712 		return set_invariant_sys_reg(reg->id, uaddr);
2713 
2714 	/* Check for regs disabled by runtime config */
2715 	if (sysreg_hidden_from_user(vcpu, r))
2716 		return -ENOENT;
2717 
2718 	if (r->set_user)
2719 		return (r->set_user)(vcpu, r, reg, uaddr);
2720 
2721 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2722 }
2723 
2724 static unsigned int num_demux_regs(void)
2725 {
2726 	unsigned int i, count = 0;
2727 
2728 	for (i = 0; i < CSSELR_MAX; i++)
2729 		if (is_valid_cache(i))
2730 			count++;
2731 
2732 	return count;
2733 }
2734 
2735 static int write_demux_regids(u64 __user *uindices)
2736 {
2737 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2738 	unsigned int i;
2739 
2740 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2741 	for (i = 0; i < CSSELR_MAX; i++) {
2742 		if (!is_valid_cache(i))
2743 			continue;
2744 		if (put_user(val | i, uindices))
2745 			return -EFAULT;
2746 		uindices++;
2747 	}
2748 	return 0;
2749 }
2750 
2751 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2752 {
2753 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2754 		KVM_REG_ARM64_SYSREG |
2755 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2756 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2757 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2758 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2759 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2760 }
2761 
2762 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2763 {
2764 	if (!*uind)
2765 		return true;
2766 
2767 	if (put_user(sys_reg_to_index(reg), *uind))
2768 		return false;
2769 
2770 	(*uind)++;
2771 	return true;
2772 }
2773 
2774 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2775 			    const struct sys_reg_desc *rd,
2776 			    u64 __user **uind,
2777 			    unsigned int *total)
2778 {
2779 	/*
2780 	 * Ignore registers we trap but don't save,
2781 	 * and for which no custom user accessor is provided.
2782 	 */
2783 	if (!(rd->reg || rd->get_user))
2784 		return 0;
2785 
2786 	if (sysreg_hidden_from_user(vcpu, rd))
2787 		return 0;
2788 
2789 	if (!copy_reg_to_user(rd, uind))
2790 		return -EFAULT;
2791 
2792 	(*total)++;
2793 	return 0;
2794 }
2795 
2796 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2797 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2798 {
2799 	const struct sys_reg_desc *i2, *end2;
2800 	unsigned int total = 0;
2801 	int err;
2802 
2803 	i2 = sys_reg_descs;
2804 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2805 
2806 	while (i2 != end2) {
2807 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2808 		if (err)
2809 			return err;
2810 	}
2811 	return total;
2812 }
2813 
2814 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2815 {
2816 	return ARRAY_SIZE(invariant_sys_regs)
2817 		+ num_demux_regs()
2818 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2819 }
2820 
2821 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2822 {
2823 	unsigned int i;
2824 	int err;
2825 
2826 	/* Then give them all the invariant registers' indices. */
2827 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2828 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2829 			return -EFAULT;
2830 		uindices++;
2831 	}
2832 
2833 	err = walk_sys_regs(vcpu, uindices);
2834 	if (err < 0)
2835 		return err;
2836 	uindices += err;
2837 
2838 	return write_demux_regids(uindices);
2839 }
2840 
2841 void kvm_sys_reg_table_init(void)
2842 {
2843 	unsigned int i;
2844 	struct sys_reg_desc clidr;
2845 
2846 	/* Make sure tables are unique and in order. */
2847 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2848 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2849 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2850 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2851 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2852 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2853 
2854 	/* We abuse the reset function to overwrite the table itself. */
2855 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2856 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2857 
2858 	/*
2859 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2860 	 *
2861 	 *   If software reads the Cache Type fields from Ctype1
2862 	 *   upwards, once it has seen a value of 0b000, no caches
2863 	 *   exist at further-out levels of the hierarchy. So, for
2864 	 *   example, if Ctype3 is the first Cache Type field with a
2865 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2866 	 *   ignored.
2867 	 */
2868 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2869 	cache_levels = clidr.val;
2870 	for (i = 0; i < 7; i++)
2871 		if (((cache_levels >> (i*3)) & 7) == 0)
2872 			break;
2873 	/* Clear all higher bits. */
2874 	cache_levels &= (1 << (i*3))-1;
2875 }
2876