xref: /linux/arch/arm64/kvm/sys_regs.c (revision ed30aef3c864f99111e16d4ea5cf29488d99a278)
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_par();		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 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
773 
774 	p->regval = pmceid;
775 
776 	return true;
777 }
778 
779 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
780 {
781 	u64 pmcr, val;
782 
783 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
784 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
785 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
786 		kvm_inject_undefined(vcpu);
787 		return false;
788 	}
789 
790 	return true;
791 }
792 
793 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
794 			      struct sys_reg_params *p,
795 			      const struct sys_reg_desc *r)
796 {
797 	u64 idx;
798 
799 	if (!kvm_arm_pmu_v3_ready(vcpu))
800 		return trap_raz_wi(vcpu, p, r);
801 
802 	if (r->CRn == 9 && r->CRm == 13) {
803 		if (r->Op2 == 2) {
804 			/* PMXEVCNTR_EL0 */
805 			if (pmu_access_event_counter_el0_disabled(vcpu))
806 				return false;
807 
808 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
809 			      & ARMV8_PMU_COUNTER_MASK;
810 		} else if (r->Op2 == 0) {
811 			/* PMCCNTR_EL0 */
812 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
813 				return false;
814 
815 			idx = ARMV8_PMU_CYCLE_IDX;
816 		} else {
817 			return false;
818 		}
819 	} else if (r->CRn == 0 && r->CRm == 9) {
820 		/* PMCCNTR */
821 		if (pmu_access_event_counter_el0_disabled(vcpu))
822 			return false;
823 
824 		idx = ARMV8_PMU_CYCLE_IDX;
825 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
826 		/* PMEVCNTRn_EL0 */
827 		if (pmu_access_event_counter_el0_disabled(vcpu))
828 			return false;
829 
830 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
831 	} else {
832 		return false;
833 	}
834 
835 	if (!pmu_counter_idx_valid(vcpu, idx))
836 		return false;
837 
838 	if (p->is_write) {
839 		if (pmu_access_el0_disabled(vcpu))
840 			return false;
841 
842 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
843 	} else {
844 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
845 	}
846 
847 	return true;
848 }
849 
850 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
851 			       const struct sys_reg_desc *r)
852 {
853 	u64 idx, reg;
854 
855 	if (!kvm_arm_pmu_v3_ready(vcpu))
856 		return trap_raz_wi(vcpu, p, r);
857 
858 	if (pmu_access_el0_disabled(vcpu))
859 		return false;
860 
861 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
862 		/* PMXEVTYPER_EL0 */
863 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
864 		reg = PMEVTYPER0_EL0 + idx;
865 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
866 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
867 		if (idx == ARMV8_PMU_CYCLE_IDX)
868 			reg = PMCCFILTR_EL0;
869 		else
870 			/* PMEVTYPERn_EL0 */
871 			reg = PMEVTYPER0_EL0 + idx;
872 	} else {
873 		BUG();
874 	}
875 
876 	if (!pmu_counter_idx_valid(vcpu, idx))
877 		return false;
878 
879 	if (p->is_write) {
880 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
881 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
882 		kvm_vcpu_pmu_restore_guest(vcpu);
883 	} else {
884 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
885 	}
886 
887 	return true;
888 }
889 
890 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
891 			   const struct sys_reg_desc *r)
892 {
893 	u64 val, mask;
894 
895 	if (!kvm_arm_pmu_v3_ready(vcpu))
896 		return trap_raz_wi(vcpu, p, r);
897 
898 	if (pmu_access_el0_disabled(vcpu))
899 		return false;
900 
901 	mask = kvm_pmu_valid_counter_mask(vcpu);
902 	if (p->is_write) {
903 		val = p->regval & mask;
904 		if (r->Op2 & 0x1) {
905 			/* accessing PMCNTENSET_EL0 */
906 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
907 			kvm_pmu_enable_counter_mask(vcpu, val);
908 			kvm_vcpu_pmu_restore_guest(vcpu);
909 		} else {
910 			/* accessing PMCNTENCLR_EL0 */
911 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
912 			kvm_pmu_disable_counter_mask(vcpu, val);
913 		}
914 	} else {
915 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
916 	}
917 
918 	return true;
919 }
920 
921 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
922 			   const struct sys_reg_desc *r)
923 {
924 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
925 
926 	if (!kvm_arm_pmu_v3_ready(vcpu))
927 		return trap_raz_wi(vcpu, p, r);
928 
929 	if (!vcpu_mode_priv(vcpu)) {
930 		kvm_inject_undefined(vcpu);
931 		return false;
932 	}
933 
934 	if (p->is_write) {
935 		u64 val = p->regval & mask;
936 
937 		if (r->Op2 & 0x1)
938 			/* accessing PMINTENSET_EL1 */
939 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
940 		else
941 			/* accessing PMINTENCLR_EL1 */
942 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
943 	} else {
944 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
945 	}
946 
947 	return true;
948 }
949 
950 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
951 			 const struct sys_reg_desc *r)
952 {
953 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
954 
955 	if (!kvm_arm_pmu_v3_ready(vcpu))
956 		return trap_raz_wi(vcpu, p, r);
957 
958 	if (pmu_access_el0_disabled(vcpu))
959 		return false;
960 
961 	if (p->is_write) {
962 		if (r->CRm & 0x2)
963 			/* accessing PMOVSSET_EL0 */
964 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
965 		else
966 			/* accessing PMOVSCLR_EL0 */
967 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
968 	} else {
969 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
970 	}
971 
972 	return true;
973 }
974 
975 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
976 			   const struct sys_reg_desc *r)
977 {
978 	u64 mask;
979 
980 	if (!kvm_arm_pmu_v3_ready(vcpu))
981 		return trap_raz_wi(vcpu, p, r);
982 
983 	if (!p->is_write)
984 		return read_from_write_only(vcpu, p, r);
985 
986 	if (pmu_write_swinc_el0_disabled(vcpu))
987 		return false;
988 
989 	mask = kvm_pmu_valid_counter_mask(vcpu);
990 	kvm_pmu_software_increment(vcpu, p->regval & mask);
991 	return true;
992 }
993 
994 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
995 			     const struct sys_reg_desc *r)
996 {
997 	if (!kvm_arm_pmu_v3_ready(vcpu))
998 		return trap_raz_wi(vcpu, p, r);
999 
1000 	if (p->is_write) {
1001 		if (!vcpu_mode_priv(vcpu)) {
1002 			kvm_inject_undefined(vcpu);
1003 			return false;
1004 		}
1005 
1006 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
1007 			       p->regval & ARMV8_PMU_USERENR_MASK;
1008 	} else {
1009 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1010 			    & ARMV8_PMU_USERENR_MASK;
1011 	}
1012 
1013 	return true;
1014 }
1015 
1016 #define reg_to_encoding(x)						\
1017 	sys_reg((u32)(x)->Op0, (u32)(x)->Op1,				\
1018 		(u32)(x)->CRn, (u32)(x)->CRm, (u32)(x)->Op2);
1019 
1020 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1021 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1022 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1023 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1024 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1025 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1026 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1027 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1028 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1029 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1030 
1031 /* Macro to expand the PMEVCNTRn_EL0 register */
1032 #define PMU_PMEVCNTR_EL0(n)						\
1033 	{ SYS_DESC(SYS_PMEVCNTRn_EL0(n)),					\
1034 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
1035 
1036 /* Macro to expand the PMEVTYPERn_EL0 register */
1037 #define PMU_PMEVTYPER_EL0(n)						\
1038 	{ SYS_DESC(SYS_PMEVTYPERn_EL0(n)),					\
1039 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
1040 
1041 static bool access_amu(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1042 			     const struct sys_reg_desc *r)
1043 {
1044 	kvm_inject_undefined(vcpu);
1045 
1046 	return false;
1047 }
1048 
1049 /* Macro to expand the AMU counter and type registers*/
1050 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), access_amu }
1051 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), access_amu }
1052 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), access_amu }
1053 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), access_amu }
1054 
1055 static bool trap_ptrauth(struct kvm_vcpu *vcpu,
1056 			 struct sys_reg_params *p,
1057 			 const struct sys_reg_desc *rd)
1058 {
1059 	/*
1060 	 * If we land here, that is because we didn't fixup the access on exit
1061 	 * by allowing the PtrAuth sysregs. The only way this happens is when
1062 	 * the guest does not have PtrAuth support enabled.
1063 	 */
1064 	kvm_inject_undefined(vcpu);
1065 
1066 	return false;
1067 }
1068 
1069 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1070 			const struct sys_reg_desc *rd)
1071 {
1072 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1073 }
1074 
1075 #define __PTRAUTH_KEY(k)						\
1076 	{ SYS_DESC(SYS_## k), trap_ptrauth, reset_unknown, k,		\
1077 	.visibility = ptrauth_visibility}
1078 
1079 #define PTRAUTH_KEY(k)							\
1080 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1081 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1082 
1083 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1084 			      struct sys_reg_params *p,
1085 			      const struct sys_reg_desc *r)
1086 {
1087 	enum kvm_arch_timers tmr;
1088 	enum kvm_arch_timer_regs treg;
1089 	u64 reg = reg_to_encoding(r);
1090 
1091 	switch (reg) {
1092 	case SYS_CNTP_TVAL_EL0:
1093 	case SYS_AARCH32_CNTP_TVAL:
1094 		tmr = TIMER_PTIMER;
1095 		treg = TIMER_REG_TVAL;
1096 		break;
1097 	case SYS_CNTP_CTL_EL0:
1098 	case SYS_AARCH32_CNTP_CTL:
1099 		tmr = TIMER_PTIMER;
1100 		treg = TIMER_REG_CTL;
1101 		break;
1102 	case SYS_CNTP_CVAL_EL0:
1103 	case SYS_AARCH32_CNTP_CVAL:
1104 		tmr = TIMER_PTIMER;
1105 		treg = TIMER_REG_CVAL;
1106 		break;
1107 	default:
1108 		BUG();
1109 	}
1110 
1111 	if (p->is_write)
1112 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1113 	else
1114 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1115 
1116 	return true;
1117 }
1118 
1119 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1120 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1121 		struct sys_reg_desc const *r, bool raz)
1122 {
1123 	u32 id = sys_reg((u32)r->Op0, (u32)r->Op1,
1124 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
1125 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1126 
1127 	if (id == SYS_ID_AA64PFR0_EL1) {
1128 		if (!vcpu_has_sve(vcpu))
1129 			val &= ~(0xfUL << ID_AA64PFR0_SVE_SHIFT);
1130 		val &= ~(0xfUL << ID_AA64PFR0_AMU_SHIFT);
1131 		if (!(val & (0xfUL << ID_AA64PFR0_CSV2_SHIFT)) &&
1132 		    arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
1133 			val |= (1UL << ID_AA64PFR0_CSV2_SHIFT);
1134 	} else if (id == SYS_ID_AA64PFR1_EL1) {
1135 		val &= ~(0xfUL << ID_AA64PFR1_MTE_SHIFT);
1136 	} else if (id == SYS_ID_AA64ISAR1_EL1 && !vcpu_has_ptrauth(vcpu)) {
1137 		val &= ~((0xfUL << ID_AA64ISAR1_APA_SHIFT) |
1138 			 (0xfUL << ID_AA64ISAR1_API_SHIFT) |
1139 			 (0xfUL << ID_AA64ISAR1_GPA_SHIFT) |
1140 			 (0xfUL << ID_AA64ISAR1_GPI_SHIFT));
1141 	} else if (id == SYS_ID_AA64DFR0_EL1) {
1142 		/* Limit guests to PMUv3 for ARMv8.1 */
1143 		val = cpuid_feature_cap_perfmon_field(val,
1144 						ID_AA64DFR0_PMUVER_SHIFT,
1145 						ID_AA64DFR0_PMUVER_8_1);
1146 	} else if (id == SYS_ID_DFR0_EL1) {
1147 		/* Limit guests to PMUv3 for ARMv8.1 */
1148 		val = cpuid_feature_cap_perfmon_field(val,
1149 						ID_DFR0_PERFMON_SHIFT,
1150 						ID_DFR0_PERFMON_8_1);
1151 	}
1152 
1153 	return val;
1154 }
1155 
1156 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1157 				  const struct sys_reg_desc *r)
1158 {
1159 	u32 id = sys_reg((u32)r->Op0, (u32)r->Op1,
1160 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
1161 
1162 	switch (id) {
1163 	case SYS_ID_AA64ZFR0_EL1:
1164 		if (!vcpu_has_sve(vcpu))
1165 			return REG_RAZ;
1166 		break;
1167 	}
1168 
1169 	return 0;
1170 }
1171 
1172 /* cpufeature ID register access trap handlers */
1173 
1174 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1175 			    struct sys_reg_params *p,
1176 			    const struct sys_reg_desc *r,
1177 			    bool raz)
1178 {
1179 	if (p->is_write)
1180 		return write_to_read_only(vcpu, p, r);
1181 
1182 	p->regval = read_id_reg(vcpu, r, raz);
1183 	return true;
1184 }
1185 
1186 static bool access_id_reg(struct kvm_vcpu *vcpu,
1187 			  struct sys_reg_params *p,
1188 			  const struct sys_reg_desc *r)
1189 {
1190 	bool raz = sysreg_visible_as_raz(vcpu, r);
1191 
1192 	return __access_id_reg(vcpu, p, r, raz);
1193 }
1194 
1195 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1196 			      struct sys_reg_params *p,
1197 			      const struct sys_reg_desc *r)
1198 {
1199 	return __access_id_reg(vcpu, p, r, true);
1200 }
1201 
1202 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1203 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1204 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1205 
1206 /* Visibility overrides for SVE-specific control registers */
1207 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1208 				   const struct sys_reg_desc *rd)
1209 {
1210 	if (vcpu_has_sve(vcpu))
1211 		return 0;
1212 
1213 	return REG_HIDDEN;
1214 }
1215 
1216 /*
1217  * cpufeature ID register user accessors
1218  *
1219  * For now, these registers are immutable for userspace, so no values
1220  * are stored, and for set_id_reg() we don't allow the effective value
1221  * to be changed.
1222  */
1223 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1224 			const struct sys_reg_desc *rd, void __user *uaddr,
1225 			bool raz)
1226 {
1227 	const u64 id = sys_reg_to_index(rd);
1228 	const u64 val = read_id_reg(vcpu, rd, raz);
1229 
1230 	return reg_to_user(uaddr, &val, id);
1231 }
1232 
1233 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1234 			const struct sys_reg_desc *rd, void __user *uaddr,
1235 			bool raz)
1236 {
1237 	const u64 id = sys_reg_to_index(rd);
1238 	int err;
1239 	u64 val;
1240 
1241 	err = reg_from_user(&val, uaddr, id);
1242 	if (err)
1243 		return err;
1244 
1245 	/* This is what we mean by invariant: you can't change it. */
1246 	if (val != read_id_reg(vcpu, rd, raz))
1247 		return -EINVAL;
1248 
1249 	return 0;
1250 }
1251 
1252 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1253 		      const struct kvm_one_reg *reg, void __user *uaddr)
1254 {
1255 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1256 
1257 	return __get_id_reg(vcpu, rd, uaddr, raz);
1258 }
1259 
1260 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1261 		      const struct kvm_one_reg *reg, void __user *uaddr)
1262 {
1263 	bool raz = sysreg_visible_as_raz(vcpu, rd);
1264 
1265 	return __set_id_reg(vcpu, rd, uaddr, raz);
1266 }
1267 
1268 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1269 			  const struct kvm_one_reg *reg, void __user *uaddr)
1270 {
1271 	return __get_id_reg(vcpu, rd, uaddr, true);
1272 }
1273 
1274 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1275 			  const struct kvm_one_reg *reg, void __user *uaddr)
1276 {
1277 	return __set_id_reg(vcpu, rd, uaddr, true);
1278 }
1279 
1280 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1281 		       const struct sys_reg_desc *r)
1282 {
1283 	if (p->is_write)
1284 		return write_to_read_only(vcpu, p, r);
1285 
1286 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1287 	return true;
1288 }
1289 
1290 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1291 			 const struct sys_reg_desc *r)
1292 {
1293 	if (p->is_write)
1294 		return write_to_read_only(vcpu, p, r);
1295 
1296 	p->regval = read_sysreg(clidr_el1);
1297 	return true;
1298 }
1299 
1300 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1301 			  const struct sys_reg_desc *r)
1302 {
1303 	int reg = r->reg;
1304 
1305 	/* See the 32bit mapping in kvm_host.h */
1306 	if (p->is_aarch32)
1307 		reg = r->reg / 2;
1308 
1309 	if (p->is_write)
1310 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1311 	else
1312 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1313 	return true;
1314 }
1315 
1316 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1317 			  const struct sys_reg_desc *r)
1318 {
1319 	u32 csselr;
1320 
1321 	if (p->is_write)
1322 		return write_to_read_only(vcpu, p, r);
1323 
1324 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1325 	p->regval = get_ccsidr(csselr);
1326 
1327 	/*
1328 	 * Guests should not be doing cache operations by set/way at all, and
1329 	 * for this reason, we trap them and attempt to infer the intent, so
1330 	 * that we can flush the entire guest's address space at the appropriate
1331 	 * time.
1332 	 * To prevent this trapping from causing performance problems, let's
1333 	 * expose the geometry of all data and unified caches (which are
1334 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1335 	 * [If guests should attempt to infer aliasing properties from the
1336 	 * geometry (which is not permitted by the architecture), they would
1337 	 * only do so for virtually indexed caches.]
1338 	 */
1339 	if (!(csselr & 1)) // data or unified cache
1340 		p->regval &= ~GENMASK(27, 3);
1341 	return true;
1342 }
1343 
1344 static bool access_mte_regs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1345 			    const struct sys_reg_desc *r)
1346 {
1347 	kvm_inject_undefined(vcpu);
1348 	return false;
1349 }
1350 
1351 /* sys_reg_desc initialiser for known cpufeature ID registers */
1352 #define ID_SANITISED(name) {			\
1353 	SYS_DESC(SYS_##name),			\
1354 	.access	= access_id_reg,		\
1355 	.get_user = get_id_reg,			\
1356 	.set_user = set_id_reg,			\
1357 	.visibility = id_visibility,		\
1358 }
1359 
1360 /*
1361  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1362  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1363  * (1 <= crm < 8, 0 <= Op2 < 8).
1364  */
1365 #define ID_UNALLOCATED(crm, op2) {			\
1366 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1367 	.access = access_raz_id_reg,			\
1368 	.get_user = get_raz_id_reg,			\
1369 	.set_user = set_raz_id_reg,			\
1370 }
1371 
1372 /*
1373  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1374  * For now, these are exposed just like unallocated ID regs: they appear
1375  * RAZ for the guest.
1376  */
1377 #define ID_HIDDEN(name) {			\
1378 	SYS_DESC(SYS_##name),			\
1379 	.access = access_raz_id_reg,		\
1380 	.get_user = get_raz_id_reg,		\
1381 	.set_user = set_raz_id_reg,		\
1382 }
1383 
1384 /*
1385  * Architected system registers.
1386  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1387  *
1388  * Debug handling: We do trap most, if not all debug related system
1389  * registers. The implementation is good enough to ensure that a guest
1390  * can use these with minimal performance degradation. The drawback is
1391  * that we don't implement any of the external debug, none of the
1392  * OSlock protocol. This should be revisited if we ever encounter a
1393  * more demanding guest...
1394  */
1395 static const struct sys_reg_desc sys_reg_descs[] = {
1396 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1397 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1398 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1399 
1400 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1401 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1402 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1403 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1404 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1405 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1406 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1407 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1408 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1409 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1410 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1411 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1412 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1413 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1414 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1415 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1416 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1417 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1418 
1419 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1420 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1421 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1422 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1423 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1424 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1425 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1426 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1427 
1428 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1429 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1430 	// DBGDTR[TR]X_EL0 share the same encoding
1431 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1432 
1433 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1434 
1435 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1436 
1437 	/*
1438 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1439 	 * entries in arm64_ftr_regs[].
1440 	 */
1441 
1442 	/* AArch64 mappings of the AArch32 ID registers */
1443 	/* CRm=1 */
1444 	ID_SANITISED(ID_PFR0_EL1),
1445 	ID_SANITISED(ID_PFR1_EL1),
1446 	ID_SANITISED(ID_DFR0_EL1),
1447 	ID_HIDDEN(ID_AFR0_EL1),
1448 	ID_SANITISED(ID_MMFR0_EL1),
1449 	ID_SANITISED(ID_MMFR1_EL1),
1450 	ID_SANITISED(ID_MMFR2_EL1),
1451 	ID_SANITISED(ID_MMFR3_EL1),
1452 
1453 	/* CRm=2 */
1454 	ID_SANITISED(ID_ISAR0_EL1),
1455 	ID_SANITISED(ID_ISAR1_EL1),
1456 	ID_SANITISED(ID_ISAR2_EL1),
1457 	ID_SANITISED(ID_ISAR3_EL1),
1458 	ID_SANITISED(ID_ISAR4_EL1),
1459 	ID_SANITISED(ID_ISAR5_EL1),
1460 	ID_SANITISED(ID_MMFR4_EL1),
1461 	ID_SANITISED(ID_ISAR6_EL1),
1462 
1463 	/* CRm=3 */
1464 	ID_SANITISED(MVFR0_EL1),
1465 	ID_SANITISED(MVFR1_EL1),
1466 	ID_SANITISED(MVFR2_EL1),
1467 	ID_UNALLOCATED(3,3),
1468 	ID_SANITISED(ID_PFR2_EL1),
1469 	ID_HIDDEN(ID_DFR1_EL1),
1470 	ID_SANITISED(ID_MMFR5_EL1),
1471 	ID_UNALLOCATED(3,7),
1472 
1473 	/* AArch64 ID registers */
1474 	/* CRm=4 */
1475 	ID_SANITISED(ID_AA64PFR0_EL1),
1476 	ID_SANITISED(ID_AA64PFR1_EL1),
1477 	ID_UNALLOCATED(4,2),
1478 	ID_UNALLOCATED(4,3),
1479 	ID_SANITISED(ID_AA64ZFR0_EL1),
1480 	ID_UNALLOCATED(4,5),
1481 	ID_UNALLOCATED(4,6),
1482 	ID_UNALLOCATED(4,7),
1483 
1484 	/* CRm=5 */
1485 	ID_SANITISED(ID_AA64DFR0_EL1),
1486 	ID_SANITISED(ID_AA64DFR1_EL1),
1487 	ID_UNALLOCATED(5,2),
1488 	ID_UNALLOCATED(5,3),
1489 	ID_HIDDEN(ID_AA64AFR0_EL1),
1490 	ID_HIDDEN(ID_AA64AFR1_EL1),
1491 	ID_UNALLOCATED(5,6),
1492 	ID_UNALLOCATED(5,7),
1493 
1494 	/* CRm=6 */
1495 	ID_SANITISED(ID_AA64ISAR0_EL1),
1496 	ID_SANITISED(ID_AA64ISAR1_EL1),
1497 	ID_UNALLOCATED(6,2),
1498 	ID_UNALLOCATED(6,3),
1499 	ID_UNALLOCATED(6,4),
1500 	ID_UNALLOCATED(6,5),
1501 	ID_UNALLOCATED(6,6),
1502 	ID_UNALLOCATED(6,7),
1503 
1504 	/* CRm=7 */
1505 	ID_SANITISED(ID_AA64MMFR0_EL1),
1506 	ID_SANITISED(ID_AA64MMFR1_EL1),
1507 	ID_SANITISED(ID_AA64MMFR2_EL1),
1508 	ID_UNALLOCATED(7,3),
1509 	ID_UNALLOCATED(7,4),
1510 	ID_UNALLOCATED(7,5),
1511 	ID_UNALLOCATED(7,6),
1512 	ID_UNALLOCATED(7,7),
1513 
1514 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1515 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
1516 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1517 
1518 	{ SYS_DESC(SYS_RGSR_EL1), access_mte_regs },
1519 	{ SYS_DESC(SYS_GCR_EL1), access_mte_regs },
1520 
1521 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1522 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1523 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1524 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1525 
1526 	PTRAUTH_KEY(APIA),
1527 	PTRAUTH_KEY(APIB),
1528 	PTRAUTH_KEY(APDA),
1529 	PTRAUTH_KEY(APDB),
1530 	PTRAUTH_KEY(APGA),
1531 
1532 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1533 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1534 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1535 
1536 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1537 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1538 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1539 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1540 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1541 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1542 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1543 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1544 
1545 	{ SYS_DESC(SYS_TFSR_EL1), access_mte_regs },
1546 	{ SYS_DESC(SYS_TFSRE0_EL1), access_mte_regs },
1547 
1548 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1549 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1550 
1551 	{ SYS_DESC(SYS_PMINTENSET_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1552 	{ SYS_DESC(SYS_PMINTENCLR_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1553 
1554 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1555 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1556 
1557 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1558 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1559 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1560 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1561 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1562 
1563 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1564 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1565 
1566 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1567 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1568 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1569 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1570 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1571 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1572 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1573 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1574 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1575 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1576 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1577 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1578 
1579 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1580 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1581 
1582 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1583 
1584 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1585 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1586 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1587 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1588 
1589 	{ SYS_DESC(SYS_PMCR_EL0), access_pmcr, reset_pmcr, PMCR_EL0 },
1590 	{ SYS_DESC(SYS_PMCNTENSET_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1591 	{ SYS_DESC(SYS_PMCNTENCLR_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1592 	{ SYS_DESC(SYS_PMOVSCLR_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1593 	{ SYS_DESC(SYS_PMSWINC_EL0), access_pmswinc, reset_unknown, PMSWINC_EL0 },
1594 	{ SYS_DESC(SYS_PMSELR_EL0), access_pmselr, reset_unknown, PMSELR_EL0 },
1595 	{ SYS_DESC(SYS_PMCEID0_EL0), access_pmceid },
1596 	{ SYS_DESC(SYS_PMCEID1_EL0), access_pmceid },
1597 	{ SYS_DESC(SYS_PMCCNTR_EL0), access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1598 	{ SYS_DESC(SYS_PMXEVTYPER_EL0), access_pmu_evtyper },
1599 	{ SYS_DESC(SYS_PMXEVCNTR_EL0), access_pmu_evcntr },
1600 	/*
1601 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1602 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1603 	 */
1604 	{ SYS_DESC(SYS_PMUSERENR_EL0), access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1605 	{ SYS_DESC(SYS_PMOVSSET_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1606 
1607 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1608 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1609 
1610 	{ SYS_DESC(SYS_AMCR_EL0), access_amu },
1611 	{ SYS_DESC(SYS_AMCFGR_EL0), access_amu },
1612 	{ SYS_DESC(SYS_AMCGCR_EL0), access_amu },
1613 	{ SYS_DESC(SYS_AMUSERENR_EL0), access_amu },
1614 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), access_amu },
1615 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), access_amu },
1616 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), access_amu },
1617 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), access_amu },
1618 	AMU_AMEVCNTR0_EL0(0),
1619 	AMU_AMEVCNTR0_EL0(1),
1620 	AMU_AMEVCNTR0_EL0(2),
1621 	AMU_AMEVCNTR0_EL0(3),
1622 	AMU_AMEVCNTR0_EL0(4),
1623 	AMU_AMEVCNTR0_EL0(5),
1624 	AMU_AMEVCNTR0_EL0(6),
1625 	AMU_AMEVCNTR0_EL0(7),
1626 	AMU_AMEVCNTR0_EL0(8),
1627 	AMU_AMEVCNTR0_EL0(9),
1628 	AMU_AMEVCNTR0_EL0(10),
1629 	AMU_AMEVCNTR0_EL0(11),
1630 	AMU_AMEVCNTR0_EL0(12),
1631 	AMU_AMEVCNTR0_EL0(13),
1632 	AMU_AMEVCNTR0_EL0(14),
1633 	AMU_AMEVCNTR0_EL0(15),
1634 	AMU_AMEVTYPER0_EL0(0),
1635 	AMU_AMEVTYPER0_EL0(1),
1636 	AMU_AMEVTYPER0_EL0(2),
1637 	AMU_AMEVTYPER0_EL0(3),
1638 	AMU_AMEVTYPER0_EL0(4),
1639 	AMU_AMEVTYPER0_EL0(5),
1640 	AMU_AMEVTYPER0_EL0(6),
1641 	AMU_AMEVTYPER0_EL0(7),
1642 	AMU_AMEVTYPER0_EL0(8),
1643 	AMU_AMEVTYPER0_EL0(9),
1644 	AMU_AMEVTYPER0_EL0(10),
1645 	AMU_AMEVTYPER0_EL0(11),
1646 	AMU_AMEVTYPER0_EL0(12),
1647 	AMU_AMEVTYPER0_EL0(13),
1648 	AMU_AMEVTYPER0_EL0(14),
1649 	AMU_AMEVTYPER0_EL0(15),
1650 	AMU_AMEVCNTR1_EL0(0),
1651 	AMU_AMEVCNTR1_EL0(1),
1652 	AMU_AMEVCNTR1_EL0(2),
1653 	AMU_AMEVCNTR1_EL0(3),
1654 	AMU_AMEVCNTR1_EL0(4),
1655 	AMU_AMEVCNTR1_EL0(5),
1656 	AMU_AMEVCNTR1_EL0(6),
1657 	AMU_AMEVCNTR1_EL0(7),
1658 	AMU_AMEVCNTR1_EL0(8),
1659 	AMU_AMEVCNTR1_EL0(9),
1660 	AMU_AMEVCNTR1_EL0(10),
1661 	AMU_AMEVCNTR1_EL0(11),
1662 	AMU_AMEVCNTR1_EL0(12),
1663 	AMU_AMEVCNTR1_EL0(13),
1664 	AMU_AMEVCNTR1_EL0(14),
1665 	AMU_AMEVCNTR1_EL0(15),
1666 	AMU_AMEVTYPER1_EL0(0),
1667 	AMU_AMEVTYPER1_EL0(1),
1668 	AMU_AMEVTYPER1_EL0(2),
1669 	AMU_AMEVTYPER1_EL0(3),
1670 	AMU_AMEVTYPER1_EL0(4),
1671 	AMU_AMEVTYPER1_EL0(5),
1672 	AMU_AMEVTYPER1_EL0(6),
1673 	AMU_AMEVTYPER1_EL0(7),
1674 	AMU_AMEVTYPER1_EL0(8),
1675 	AMU_AMEVTYPER1_EL0(9),
1676 	AMU_AMEVTYPER1_EL0(10),
1677 	AMU_AMEVTYPER1_EL0(11),
1678 	AMU_AMEVTYPER1_EL0(12),
1679 	AMU_AMEVTYPER1_EL0(13),
1680 	AMU_AMEVTYPER1_EL0(14),
1681 	AMU_AMEVTYPER1_EL0(15),
1682 
1683 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1684 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1685 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1686 
1687 	/* PMEVCNTRn_EL0 */
1688 	PMU_PMEVCNTR_EL0(0),
1689 	PMU_PMEVCNTR_EL0(1),
1690 	PMU_PMEVCNTR_EL0(2),
1691 	PMU_PMEVCNTR_EL0(3),
1692 	PMU_PMEVCNTR_EL0(4),
1693 	PMU_PMEVCNTR_EL0(5),
1694 	PMU_PMEVCNTR_EL0(6),
1695 	PMU_PMEVCNTR_EL0(7),
1696 	PMU_PMEVCNTR_EL0(8),
1697 	PMU_PMEVCNTR_EL0(9),
1698 	PMU_PMEVCNTR_EL0(10),
1699 	PMU_PMEVCNTR_EL0(11),
1700 	PMU_PMEVCNTR_EL0(12),
1701 	PMU_PMEVCNTR_EL0(13),
1702 	PMU_PMEVCNTR_EL0(14),
1703 	PMU_PMEVCNTR_EL0(15),
1704 	PMU_PMEVCNTR_EL0(16),
1705 	PMU_PMEVCNTR_EL0(17),
1706 	PMU_PMEVCNTR_EL0(18),
1707 	PMU_PMEVCNTR_EL0(19),
1708 	PMU_PMEVCNTR_EL0(20),
1709 	PMU_PMEVCNTR_EL0(21),
1710 	PMU_PMEVCNTR_EL0(22),
1711 	PMU_PMEVCNTR_EL0(23),
1712 	PMU_PMEVCNTR_EL0(24),
1713 	PMU_PMEVCNTR_EL0(25),
1714 	PMU_PMEVCNTR_EL0(26),
1715 	PMU_PMEVCNTR_EL0(27),
1716 	PMU_PMEVCNTR_EL0(28),
1717 	PMU_PMEVCNTR_EL0(29),
1718 	PMU_PMEVCNTR_EL0(30),
1719 	/* PMEVTYPERn_EL0 */
1720 	PMU_PMEVTYPER_EL0(0),
1721 	PMU_PMEVTYPER_EL0(1),
1722 	PMU_PMEVTYPER_EL0(2),
1723 	PMU_PMEVTYPER_EL0(3),
1724 	PMU_PMEVTYPER_EL0(4),
1725 	PMU_PMEVTYPER_EL0(5),
1726 	PMU_PMEVTYPER_EL0(6),
1727 	PMU_PMEVTYPER_EL0(7),
1728 	PMU_PMEVTYPER_EL0(8),
1729 	PMU_PMEVTYPER_EL0(9),
1730 	PMU_PMEVTYPER_EL0(10),
1731 	PMU_PMEVTYPER_EL0(11),
1732 	PMU_PMEVTYPER_EL0(12),
1733 	PMU_PMEVTYPER_EL0(13),
1734 	PMU_PMEVTYPER_EL0(14),
1735 	PMU_PMEVTYPER_EL0(15),
1736 	PMU_PMEVTYPER_EL0(16),
1737 	PMU_PMEVTYPER_EL0(17),
1738 	PMU_PMEVTYPER_EL0(18),
1739 	PMU_PMEVTYPER_EL0(19),
1740 	PMU_PMEVTYPER_EL0(20),
1741 	PMU_PMEVTYPER_EL0(21),
1742 	PMU_PMEVTYPER_EL0(22),
1743 	PMU_PMEVTYPER_EL0(23),
1744 	PMU_PMEVTYPER_EL0(24),
1745 	PMU_PMEVTYPER_EL0(25),
1746 	PMU_PMEVTYPER_EL0(26),
1747 	PMU_PMEVTYPER_EL0(27),
1748 	PMU_PMEVTYPER_EL0(28),
1749 	PMU_PMEVTYPER_EL0(29),
1750 	PMU_PMEVTYPER_EL0(30),
1751 	/*
1752 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1753 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1754 	 */
1755 	{ SYS_DESC(SYS_PMCCFILTR_EL0), access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1756 
1757 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1758 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1759 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1760 };
1761 
1762 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1763 			struct sys_reg_params *p,
1764 			const struct sys_reg_desc *r)
1765 {
1766 	if (p->is_write) {
1767 		return ignore_write(vcpu, p);
1768 	} else {
1769 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1770 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1771 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1772 
1773 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1774 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1775 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1776 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1777 		return true;
1778 	}
1779 }
1780 
1781 static bool trap_debug32(struct kvm_vcpu *vcpu,
1782 			 struct sys_reg_params *p,
1783 			 const struct sys_reg_desc *r)
1784 {
1785 	if (p->is_write) {
1786 		vcpu_cp14(vcpu, r->reg) = p->regval;
1787 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1788 	} else {
1789 		p->regval = vcpu_cp14(vcpu, r->reg);
1790 	}
1791 
1792 	return true;
1793 }
1794 
1795 /* AArch32 debug register mappings
1796  *
1797  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1798  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1799  *
1800  * All control registers and watchpoint value registers are mapped to
1801  * the lower 32 bits of their AArch64 equivalents. We share the trap
1802  * handlers with the above AArch64 code which checks what mode the
1803  * system is in.
1804  */
1805 
1806 static bool trap_xvr(struct kvm_vcpu *vcpu,
1807 		     struct sys_reg_params *p,
1808 		     const struct sys_reg_desc *rd)
1809 {
1810 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1811 
1812 	if (p->is_write) {
1813 		u64 val = *dbg_reg;
1814 
1815 		val &= 0xffffffffUL;
1816 		val |= p->regval << 32;
1817 		*dbg_reg = val;
1818 
1819 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1820 	} else {
1821 		p->regval = *dbg_reg >> 32;
1822 	}
1823 
1824 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1825 
1826 	return true;
1827 }
1828 
1829 #define DBG_BCR_BVR_WCR_WVR(n)						\
1830 	/* DBGBVRn */							\
1831 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1832 	/* DBGBCRn */							\
1833 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1834 	/* DBGWVRn */							\
1835 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1836 	/* DBGWCRn */							\
1837 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1838 
1839 #define DBGBXVR(n)							\
1840 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1841 
1842 /*
1843  * Trapped cp14 registers. We generally ignore most of the external
1844  * debug, on the principle that they don't really make sense to a
1845  * guest. Revisit this one day, would this principle change.
1846  */
1847 static const struct sys_reg_desc cp14_regs[] = {
1848 	/* DBGIDR */
1849 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1850 	/* DBGDTRRXext */
1851 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1852 
1853 	DBG_BCR_BVR_WCR_WVR(0),
1854 	/* DBGDSCRint */
1855 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1856 	DBG_BCR_BVR_WCR_WVR(1),
1857 	/* DBGDCCINT */
1858 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32, NULL, cp14_DBGDCCINT },
1859 	/* DBGDSCRext */
1860 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32, NULL, cp14_DBGDSCRext },
1861 	DBG_BCR_BVR_WCR_WVR(2),
1862 	/* DBGDTR[RT]Xint */
1863 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1864 	/* DBGDTR[RT]Xext */
1865 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1866 	DBG_BCR_BVR_WCR_WVR(3),
1867 	DBG_BCR_BVR_WCR_WVR(4),
1868 	DBG_BCR_BVR_WCR_WVR(5),
1869 	/* DBGWFAR */
1870 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1871 	/* DBGOSECCR */
1872 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1873 	DBG_BCR_BVR_WCR_WVR(6),
1874 	/* DBGVCR */
1875 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32, NULL, cp14_DBGVCR },
1876 	DBG_BCR_BVR_WCR_WVR(7),
1877 	DBG_BCR_BVR_WCR_WVR(8),
1878 	DBG_BCR_BVR_WCR_WVR(9),
1879 	DBG_BCR_BVR_WCR_WVR(10),
1880 	DBG_BCR_BVR_WCR_WVR(11),
1881 	DBG_BCR_BVR_WCR_WVR(12),
1882 	DBG_BCR_BVR_WCR_WVR(13),
1883 	DBG_BCR_BVR_WCR_WVR(14),
1884 	DBG_BCR_BVR_WCR_WVR(15),
1885 
1886 	/* DBGDRAR (32bit) */
1887 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1888 
1889 	DBGBXVR(0),
1890 	/* DBGOSLAR */
1891 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1892 	DBGBXVR(1),
1893 	/* DBGOSLSR */
1894 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1895 	DBGBXVR(2),
1896 	DBGBXVR(3),
1897 	/* DBGOSDLR */
1898 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1899 	DBGBXVR(4),
1900 	/* DBGPRCR */
1901 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1902 	DBGBXVR(5),
1903 	DBGBXVR(6),
1904 	DBGBXVR(7),
1905 	DBGBXVR(8),
1906 	DBGBXVR(9),
1907 	DBGBXVR(10),
1908 	DBGBXVR(11),
1909 	DBGBXVR(12),
1910 	DBGBXVR(13),
1911 	DBGBXVR(14),
1912 	DBGBXVR(15),
1913 
1914 	/* DBGDSAR (32bit) */
1915 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1916 
1917 	/* DBGDEVID2 */
1918 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1919 	/* DBGDEVID1 */
1920 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1921 	/* DBGDEVID */
1922 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1923 	/* DBGCLAIMSET */
1924 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1925 	/* DBGCLAIMCLR */
1926 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1927 	/* DBGAUTHSTATUS */
1928 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1929 };
1930 
1931 /* Trapped cp14 64bit registers */
1932 static const struct sys_reg_desc cp14_64_regs[] = {
1933 	/* DBGDRAR (64bit) */
1934 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1935 
1936 	/* DBGDSAR (64bit) */
1937 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1938 };
1939 
1940 /* Macro to expand the PMEVCNTRn register */
1941 #define PMU_PMEVCNTR(n)							\
1942 	/* PMEVCNTRn */							\
1943 	{ Op1(0), CRn(0b1110),						\
1944 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1945 	  access_pmu_evcntr }
1946 
1947 /* Macro to expand the PMEVTYPERn register */
1948 #define PMU_PMEVTYPER(n)						\
1949 	/* PMEVTYPERn */						\
1950 	{ Op1(0), CRn(0b1110),						\
1951 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1952 	  access_pmu_evtyper }
1953 
1954 /*
1955  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1956  * depending on the way they are accessed (as a 32bit or a 64bit
1957  * register).
1958  */
1959 static const struct sys_reg_desc cp15_regs[] = {
1960 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1961 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1962 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr },
1963 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr },
1964 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1965 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1966 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1967 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1968 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1969 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1970 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1971 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1972 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1973 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1974 
1975 	/*
1976 	 * DC{C,I,CI}SW operations:
1977 	 */
1978 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1979 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1980 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1981 
1982 	/* PMU */
1983 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1984 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1985 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1986 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1987 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1988 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1989 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1990 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1991 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1992 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1993 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1994 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1995 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1996 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1997 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1998 
1999 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
2000 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
2001 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
2002 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
2003 
2004 	/* ICC_SRE */
2005 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2006 
2007 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
2008 
2009 	/* Arch Tmers */
2010 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2011 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2012 
2013 	/* PMEVCNTRn */
2014 	PMU_PMEVCNTR(0),
2015 	PMU_PMEVCNTR(1),
2016 	PMU_PMEVCNTR(2),
2017 	PMU_PMEVCNTR(3),
2018 	PMU_PMEVCNTR(4),
2019 	PMU_PMEVCNTR(5),
2020 	PMU_PMEVCNTR(6),
2021 	PMU_PMEVCNTR(7),
2022 	PMU_PMEVCNTR(8),
2023 	PMU_PMEVCNTR(9),
2024 	PMU_PMEVCNTR(10),
2025 	PMU_PMEVCNTR(11),
2026 	PMU_PMEVCNTR(12),
2027 	PMU_PMEVCNTR(13),
2028 	PMU_PMEVCNTR(14),
2029 	PMU_PMEVCNTR(15),
2030 	PMU_PMEVCNTR(16),
2031 	PMU_PMEVCNTR(17),
2032 	PMU_PMEVCNTR(18),
2033 	PMU_PMEVCNTR(19),
2034 	PMU_PMEVCNTR(20),
2035 	PMU_PMEVCNTR(21),
2036 	PMU_PMEVCNTR(22),
2037 	PMU_PMEVCNTR(23),
2038 	PMU_PMEVCNTR(24),
2039 	PMU_PMEVCNTR(25),
2040 	PMU_PMEVCNTR(26),
2041 	PMU_PMEVCNTR(27),
2042 	PMU_PMEVCNTR(28),
2043 	PMU_PMEVCNTR(29),
2044 	PMU_PMEVCNTR(30),
2045 	/* PMEVTYPERn */
2046 	PMU_PMEVTYPER(0),
2047 	PMU_PMEVTYPER(1),
2048 	PMU_PMEVTYPER(2),
2049 	PMU_PMEVTYPER(3),
2050 	PMU_PMEVTYPER(4),
2051 	PMU_PMEVTYPER(5),
2052 	PMU_PMEVTYPER(6),
2053 	PMU_PMEVTYPER(7),
2054 	PMU_PMEVTYPER(8),
2055 	PMU_PMEVTYPER(9),
2056 	PMU_PMEVTYPER(10),
2057 	PMU_PMEVTYPER(11),
2058 	PMU_PMEVTYPER(12),
2059 	PMU_PMEVTYPER(13),
2060 	PMU_PMEVTYPER(14),
2061 	PMU_PMEVTYPER(15),
2062 	PMU_PMEVTYPER(16),
2063 	PMU_PMEVTYPER(17),
2064 	PMU_PMEVTYPER(18),
2065 	PMU_PMEVTYPER(19),
2066 	PMU_PMEVTYPER(20),
2067 	PMU_PMEVTYPER(21),
2068 	PMU_PMEVTYPER(22),
2069 	PMU_PMEVTYPER(23),
2070 	PMU_PMEVTYPER(24),
2071 	PMU_PMEVTYPER(25),
2072 	PMU_PMEVTYPER(26),
2073 	PMU_PMEVTYPER(27),
2074 	PMU_PMEVTYPER(28),
2075 	PMU_PMEVTYPER(29),
2076 	PMU_PMEVTYPER(30),
2077 	/* PMCCFILTR */
2078 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2079 
2080 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2081 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2082 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, c0_CSSELR },
2083 };
2084 
2085 static const struct sys_reg_desc cp15_64_regs[] = {
2086 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
2087 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2088 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2089 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
2090 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2091 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2092 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2093 };
2094 
2095 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2096 			      bool is_32)
2097 {
2098 	unsigned int i;
2099 
2100 	for (i = 0; i < n; i++) {
2101 		if (!is_32 && table[i].reg && !table[i].reset) {
2102 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2103 				table, i);
2104 			return 1;
2105 		}
2106 
2107 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2108 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2109 			return 1;
2110 		}
2111 	}
2112 
2113 	return 0;
2114 }
2115 
2116 static int match_sys_reg(const void *key, const void *elt)
2117 {
2118 	const unsigned long pval = (unsigned long)key;
2119 	const struct sys_reg_desc *r = elt;
2120 
2121 	return pval - reg_to_encoding(r);
2122 }
2123 
2124 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
2125 					 const struct sys_reg_desc table[],
2126 					 unsigned int num)
2127 {
2128 	unsigned long pval = reg_to_encoding(params);
2129 
2130 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
2131 }
2132 
2133 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
2134 {
2135 	kvm_inject_undefined(vcpu);
2136 	return 1;
2137 }
2138 
2139 static void perform_access(struct kvm_vcpu *vcpu,
2140 			   struct sys_reg_params *params,
2141 			   const struct sys_reg_desc *r)
2142 {
2143 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2144 
2145 	/* Check for regs disabled by runtime config */
2146 	if (sysreg_hidden(vcpu, r)) {
2147 		kvm_inject_undefined(vcpu);
2148 		return;
2149 	}
2150 
2151 	/*
2152 	 * Not having an accessor means that we have configured a trap
2153 	 * that we don't know how to handle. This certainly qualifies
2154 	 * as a gross bug that should be fixed right away.
2155 	 */
2156 	BUG_ON(!r->access);
2157 
2158 	/* Skip instruction if instructed so */
2159 	if (likely(r->access(vcpu, params, r)))
2160 		kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
2161 }
2162 
2163 /*
2164  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2165  *                call the corresponding trap handler.
2166  *
2167  * @params: pointer to the descriptor of the access
2168  * @table: array of trap descriptors
2169  * @num: size of the trap descriptor array
2170  *
2171  * Return 0 if the access has been handled, and -1 if not.
2172  */
2173 static int emulate_cp(struct kvm_vcpu *vcpu,
2174 		      struct sys_reg_params *params,
2175 		      const struct sys_reg_desc *table,
2176 		      size_t num)
2177 {
2178 	const struct sys_reg_desc *r;
2179 
2180 	if (!table)
2181 		return -1;	/* Not handled */
2182 
2183 	r = find_reg(params, table, num);
2184 
2185 	if (r) {
2186 		perform_access(vcpu, params, r);
2187 		return 0;
2188 	}
2189 
2190 	/* Not handled */
2191 	return -1;
2192 }
2193 
2194 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2195 				struct sys_reg_params *params)
2196 {
2197 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
2198 	int cp = -1;
2199 
2200 	switch (esr_ec) {
2201 	case ESR_ELx_EC_CP15_32:
2202 	case ESR_ELx_EC_CP15_64:
2203 		cp = 15;
2204 		break;
2205 	case ESR_ELx_EC_CP14_MR:
2206 	case ESR_ELx_EC_CP14_64:
2207 		cp = 14;
2208 		break;
2209 	default:
2210 		WARN_ON(1);
2211 	}
2212 
2213 	print_sys_reg_msg(params,
2214 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2215 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2216 	kvm_inject_undefined(vcpu);
2217 }
2218 
2219 /**
2220  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2221  * @vcpu: The VCPU pointer
2222  * @run:  The kvm_run struct
2223  */
2224 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2225 			    const struct sys_reg_desc *global,
2226 			    size_t nr_global)
2227 {
2228 	struct sys_reg_params params;
2229 	u32 esr = kvm_vcpu_get_esr(vcpu);
2230 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2231 	int Rt2 = (esr >> 10) & 0x1f;
2232 
2233 	params.is_aarch32 = true;
2234 	params.is_32bit = false;
2235 	params.CRm = (esr >> 1) & 0xf;
2236 	params.is_write = ((esr & 1) == 0);
2237 
2238 	params.Op0 = 0;
2239 	params.Op1 = (esr >> 16) & 0xf;
2240 	params.Op2 = 0;
2241 	params.CRn = 0;
2242 
2243 	/*
2244 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2245 	 * backends between AArch32 and AArch64, we get away with it.
2246 	 */
2247 	if (params.is_write) {
2248 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2249 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2250 	}
2251 
2252 	/*
2253 	 * If the table contains a handler, handle the
2254 	 * potential register operation in the case of a read and return
2255 	 * with success.
2256 	 */
2257 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2258 		/* Split up the value between registers for the read side */
2259 		if (!params.is_write) {
2260 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2261 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2262 		}
2263 
2264 		return 1;
2265 	}
2266 
2267 	unhandled_cp_access(vcpu, &params);
2268 	return 1;
2269 }
2270 
2271 /**
2272  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2273  * @vcpu: The VCPU pointer
2274  * @run:  The kvm_run struct
2275  */
2276 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2277 			    const struct sys_reg_desc *global,
2278 			    size_t nr_global)
2279 {
2280 	struct sys_reg_params params;
2281 	u32 esr = kvm_vcpu_get_esr(vcpu);
2282 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2283 
2284 	params.is_aarch32 = true;
2285 	params.is_32bit = true;
2286 	params.CRm = (esr >> 1) & 0xf;
2287 	params.regval = vcpu_get_reg(vcpu, Rt);
2288 	params.is_write = ((esr & 1) == 0);
2289 	params.CRn = (esr >> 10) & 0xf;
2290 	params.Op0 = 0;
2291 	params.Op1 = (esr >> 14) & 0x7;
2292 	params.Op2 = (esr >> 17) & 0x7;
2293 
2294 	if (!emulate_cp(vcpu, &params, global, nr_global)) {
2295 		if (!params.is_write)
2296 			vcpu_set_reg(vcpu, Rt, params.regval);
2297 		return 1;
2298 	}
2299 
2300 	unhandled_cp_access(vcpu, &params);
2301 	return 1;
2302 }
2303 
2304 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
2305 {
2306 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
2307 }
2308 
2309 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
2310 {
2311 	return kvm_handle_cp_32(vcpu, cp15_regs, ARRAY_SIZE(cp15_regs));
2312 }
2313 
2314 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
2315 {
2316 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
2317 }
2318 
2319 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
2320 {
2321 	return kvm_handle_cp_32(vcpu, cp14_regs, ARRAY_SIZE(cp14_regs));
2322 }
2323 
2324 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2325 {
2326 	// See ARM DDI 0487E.a, section D12.3.2
2327 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2328 }
2329 
2330 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2331 			   struct sys_reg_params *params)
2332 {
2333 	const struct sys_reg_desc *r;
2334 
2335 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2336 
2337 	if (likely(r)) {
2338 		perform_access(vcpu, params, r);
2339 	} else if (is_imp_def_sys_reg(params)) {
2340 		kvm_inject_undefined(vcpu);
2341 	} else {
2342 		print_sys_reg_msg(params,
2343 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2344 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2345 		kvm_inject_undefined(vcpu);
2346 	}
2347 	return 1;
2348 }
2349 
2350 /**
2351  * kvm_reset_sys_regs - sets system registers to reset value
2352  * @vcpu: The VCPU pointer
2353  *
2354  * This function finds the right table above and sets the registers on the
2355  * virtual CPU struct to their architecturally defined reset values.
2356  */
2357 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2358 {
2359 	unsigned long i;
2360 
2361 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++)
2362 		if (sys_reg_descs[i].reset)
2363 			sys_reg_descs[i].reset(vcpu, &sys_reg_descs[i]);
2364 }
2365 
2366 /**
2367  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2368  * @vcpu: The VCPU pointer
2369  */
2370 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
2371 {
2372 	struct sys_reg_params params;
2373 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
2374 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2375 	int ret;
2376 
2377 	trace_kvm_handle_sys_reg(esr);
2378 
2379 	params.is_aarch32 = false;
2380 	params.is_32bit = false;
2381 	params.Op0 = (esr >> 20) & 3;
2382 	params.Op1 = (esr >> 14) & 0x7;
2383 	params.CRn = (esr >> 10) & 0xf;
2384 	params.CRm = (esr >> 1) & 0xf;
2385 	params.Op2 = (esr >> 17) & 0x7;
2386 	params.regval = vcpu_get_reg(vcpu, Rt);
2387 	params.is_write = !(esr & 1);
2388 
2389 	ret = emulate_sys_reg(vcpu, &params);
2390 
2391 	if (!params.is_write)
2392 		vcpu_set_reg(vcpu, Rt, params.regval);
2393 	return ret;
2394 }
2395 
2396 /******************************************************************************
2397  * Userspace API
2398  *****************************************************************************/
2399 
2400 static bool index_to_params(u64 id, struct sys_reg_params *params)
2401 {
2402 	switch (id & KVM_REG_SIZE_MASK) {
2403 	case KVM_REG_SIZE_U64:
2404 		/* Any unused index bits means it's not valid. */
2405 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2406 			      | KVM_REG_ARM_COPROC_MASK
2407 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2408 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2409 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2410 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2411 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2412 			return false;
2413 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2414 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2415 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2416 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2417 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2418 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2419 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2420 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2421 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2422 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2423 		return true;
2424 	default:
2425 		return false;
2426 	}
2427 }
2428 
2429 const struct sys_reg_desc *find_reg_by_id(u64 id,
2430 					  struct sys_reg_params *params,
2431 					  const struct sys_reg_desc table[],
2432 					  unsigned int num)
2433 {
2434 	if (!index_to_params(id, params))
2435 		return NULL;
2436 
2437 	return find_reg(params, table, num);
2438 }
2439 
2440 /* Decode an index value, and find the sys_reg_desc entry. */
2441 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2442 						    u64 id)
2443 {
2444 	const struct sys_reg_desc *r;
2445 	struct sys_reg_params params;
2446 
2447 	/* We only do sys_reg for now. */
2448 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2449 		return NULL;
2450 
2451 	if (!index_to_params(id, &params))
2452 		return NULL;
2453 
2454 	r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2455 
2456 	/* Not saved in the sys_reg array and not otherwise accessible? */
2457 	if (r && !(r->reg || r->get_user))
2458 		r = NULL;
2459 
2460 	return r;
2461 }
2462 
2463 /*
2464  * These are the invariant sys_reg registers: we let the guest see the
2465  * host versions of these, so they're part of the guest state.
2466  *
2467  * A future CPU may provide a mechanism to present different values to
2468  * the guest, or a future kvm may trap them.
2469  */
2470 
2471 #define FUNCTION_INVARIANT(reg)						\
2472 	static void get_##reg(struct kvm_vcpu *v,			\
2473 			      const struct sys_reg_desc *r)		\
2474 	{								\
2475 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2476 	}
2477 
2478 FUNCTION_INVARIANT(midr_el1)
2479 FUNCTION_INVARIANT(revidr_el1)
2480 FUNCTION_INVARIANT(clidr_el1)
2481 FUNCTION_INVARIANT(aidr_el1)
2482 
2483 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2484 {
2485 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2486 }
2487 
2488 /* ->val is filled in by kvm_sys_reg_table_init() */
2489 static struct sys_reg_desc invariant_sys_regs[] = {
2490 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2491 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2492 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2493 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2494 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2495 };
2496 
2497 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2498 {
2499 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2500 		return -EFAULT;
2501 	return 0;
2502 }
2503 
2504 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2505 {
2506 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2507 		return -EFAULT;
2508 	return 0;
2509 }
2510 
2511 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2512 {
2513 	struct sys_reg_params params;
2514 	const struct sys_reg_desc *r;
2515 
2516 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2517 			   ARRAY_SIZE(invariant_sys_regs));
2518 	if (!r)
2519 		return -ENOENT;
2520 
2521 	return reg_to_user(uaddr, &r->val, id);
2522 }
2523 
2524 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2525 {
2526 	struct sys_reg_params params;
2527 	const struct sys_reg_desc *r;
2528 	int err;
2529 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2530 
2531 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2532 			   ARRAY_SIZE(invariant_sys_regs));
2533 	if (!r)
2534 		return -ENOENT;
2535 
2536 	err = reg_from_user(&val, uaddr, id);
2537 	if (err)
2538 		return err;
2539 
2540 	/* This is what we mean by invariant: you can't change it. */
2541 	if (r->val != val)
2542 		return -EINVAL;
2543 
2544 	return 0;
2545 }
2546 
2547 static bool is_valid_cache(u32 val)
2548 {
2549 	u32 level, ctype;
2550 
2551 	if (val >= CSSELR_MAX)
2552 		return false;
2553 
2554 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2555 	level = (val >> 1);
2556 	ctype = (cache_levels >> (level * 3)) & 7;
2557 
2558 	switch (ctype) {
2559 	case 0: /* No cache */
2560 		return false;
2561 	case 1: /* Instruction cache only */
2562 		return (val & 1);
2563 	case 2: /* Data cache only */
2564 	case 4: /* Unified cache */
2565 		return !(val & 1);
2566 	case 3: /* Separate instruction and data caches */
2567 		return true;
2568 	default: /* Reserved: we can't know instruction or data. */
2569 		return false;
2570 	}
2571 }
2572 
2573 static int demux_c15_get(u64 id, void __user *uaddr)
2574 {
2575 	u32 val;
2576 	u32 __user *uval = uaddr;
2577 
2578 	/* Fail if we have unknown bits set. */
2579 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2580 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2581 		return -ENOENT;
2582 
2583 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2584 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2585 		if (KVM_REG_SIZE(id) != 4)
2586 			return -ENOENT;
2587 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2588 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2589 		if (!is_valid_cache(val))
2590 			return -ENOENT;
2591 
2592 		return put_user(get_ccsidr(val), uval);
2593 	default:
2594 		return -ENOENT;
2595 	}
2596 }
2597 
2598 static int demux_c15_set(u64 id, void __user *uaddr)
2599 {
2600 	u32 val, newval;
2601 	u32 __user *uval = uaddr;
2602 
2603 	/* Fail if we have unknown bits set. */
2604 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2605 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2606 		return -ENOENT;
2607 
2608 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2609 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2610 		if (KVM_REG_SIZE(id) != 4)
2611 			return -ENOENT;
2612 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2613 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2614 		if (!is_valid_cache(val))
2615 			return -ENOENT;
2616 
2617 		if (get_user(newval, uval))
2618 			return -EFAULT;
2619 
2620 		/* This is also invariant: you can't change it. */
2621 		if (newval != get_ccsidr(val))
2622 			return -EINVAL;
2623 		return 0;
2624 	default:
2625 		return -ENOENT;
2626 	}
2627 }
2628 
2629 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2630 {
2631 	const struct sys_reg_desc *r;
2632 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2633 
2634 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2635 		return demux_c15_get(reg->id, uaddr);
2636 
2637 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2638 		return -ENOENT;
2639 
2640 	r = index_to_sys_reg_desc(vcpu, reg->id);
2641 	if (!r)
2642 		return get_invariant_sys_reg(reg->id, uaddr);
2643 
2644 	/* Check for regs disabled by runtime config */
2645 	if (sysreg_hidden(vcpu, r))
2646 		return -ENOENT;
2647 
2648 	if (r->get_user)
2649 		return (r->get_user)(vcpu, r, reg, uaddr);
2650 
2651 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2652 }
2653 
2654 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2655 {
2656 	const struct sys_reg_desc *r;
2657 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2658 
2659 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2660 		return demux_c15_set(reg->id, uaddr);
2661 
2662 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2663 		return -ENOENT;
2664 
2665 	r = index_to_sys_reg_desc(vcpu, reg->id);
2666 	if (!r)
2667 		return set_invariant_sys_reg(reg->id, uaddr);
2668 
2669 	/* Check for regs disabled by runtime config */
2670 	if (sysreg_hidden(vcpu, r))
2671 		return -ENOENT;
2672 
2673 	if (r->set_user)
2674 		return (r->set_user)(vcpu, r, reg, uaddr);
2675 
2676 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2677 }
2678 
2679 static unsigned int num_demux_regs(void)
2680 {
2681 	unsigned int i, count = 0;
2682 
2683 	for (i = 0; i < CSSELR_MAX; i++)
2684 		if (is_valid_cache(i))
2685 			count++;
2686 
2687 	return count;
2688 }
2689 
2690 static int write_demux_regids(u64 __user *uindices)
2691 {
2692 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2693 	unsigned int i;
2694 
2695 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2696 	for (i = 0; i < CSSELR_MAX; i++) {
2697 		if (!is_valid_cache(i))
2698 			continue;
2699 		if (put_user(val | i, uindices))
2700 			return -EFAULT;
2701 		uindices++;
2702 	}
2703 	return 0;
2704 }
2705 
2706 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2707 {
2708 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2709 		KVM_REG_ARM64_SYSREG |
2710 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2711 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2712 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2713 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2714 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2715 }
2716 
2717 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2718 {
2719 	if (!*uind)
2720 		return true;
2721 
2722 	if (put_user(sys_reg_to_index(reg), *uind))
2723 		return false;
2724 
2725 	(*uind)++;
2726 	return true;
2727 }
2728 
2729 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2730 			    const struct sys_reg_desc *rd,
2731 			    u64 __user **uind,
2732 			    unsigned int *total)
2733 {
2734 	/*
2735 	 * Ignore registers we trap but don't save,
2736 	 * and for which no custom user accessor is provided.
2737 	 */
2738 	if (!(rd->reg || rd->get_user))
2739 		return 0;
2740 
2741 	if (sysreg_hidden(vcpu, rd))
2742 		return 0;
2743 
2744 	if (!copy_reg_to_user(rd, uind))
2745 		return -EFAULT;
2746 
2747 	(*total)++;
2748 	return 0;
2749 }
2750 
2751 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2752 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2753 {
2754 	const struct sys_reg_desc *i2, *end2;
2755 	unsigned int total = 0;
2756 	int err;
2757 
2758 	i2 = sys_reg_descs;
2759 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2760 
2761 	while (i2 != end2) {
2762 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
2763 		if (err)
2764 			return err;
2765 	}
2766 	return total;
2767 }
2768 
2769 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2770 {
2771 	return ARRAY_SIZE(invariant_sys_regs)
2772 		+ num_demux_regs()
2773 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2774 }
2775 
2776 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2777 {
2778 	unsigned int i;
2779 	int err;
2780 
2781 	/* Then give them all the invariant registers' indices. */
2782 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2783 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2784 			return -EFAULT;
2785 		uindices++;
2786 	}
2787 
2788 	err = walk_sys_regs(vcpu, uindices);
2789 	if (err < 0)
2790 		return err;
2791 	uindices += err;
2792 
2793 	return write_demux_regids(uindices);
2794 }
2795 
2796 void kvm_sys_reg_table_init(void)
2797 {
2798 	unsigned int i;
2799 	struct sys_reg_desc clidr;
2800 
2801 	/* Make sure tables are unique and in order. */
2802 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2803 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2804 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2805 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2806 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2807 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2808 
2809 	/* We abuse the reset function to overwrite the table itself. */
2810 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2811 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2812 
2813 	/*
2814 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2815 	 *
2816 	 *   If software reads the Cache Type fields from Ctype1
2817 	 *   upwards, once it has seen a value of 0b000, no caches
2818 	 *   exist at further-out levels of the hierarchy. So, for
2819 	 *   example, if Ctype3 is the first Cache Type field with a
2820 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2821 	 *   ignored.
2822 	 */
2823 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2824 	cache_levels = clidr.val;
2825 	for (i = 0; i < 7; i++)
2826 		if (((cache_levels >> (i*3)) & 7) == 0)
2827 			break;
2828 	/* Clear all higher bits. */
2829 	cache_levels &= (1 << (i*3))-1;
2830 }
2831