xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision e72e8bf1)
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 extremly similar to the ARM coproc.c, but the
38  * types are different. My gut feeling is that it should be pretty
39  * easy to merge, but that would be an ABI breakage -- again. VFP
40  * would also need to be abstracted.
41  *
42  * For AArch32, we only take care of what is being trapped. Anything
43  * that has to do with init and userspace access has to go via the
44  * 64bit interface.
45  */
46 
47 static bool read_from_write_only(struct kvm_vcpu *vcpu,
48 				 struct sys_reg_params *params,
49 				 const struct sys_reg_desc *r)
50 {
51 	WARN_ONCE(1, "Unexpected sys_reg read to write-only register\n");
52 	print_sys_reg_instr(params);
53 	kvm_inject_undefined(vcpu);
54 	return false;
55 }
56 
57 static bool write_to_read_only(struct kvm_vcpu *vcpu,
58 			       struct sys_reg_params *params,
59 			       const struct sys_reg_desc *r)
60 {
61 	WARN_ONCE(1, "Unexpected sys_reg write to read-only register\n");
62 	print_sys_reg_instr(params);
63 	kvm_inject_undefined(vcpu);
64 	return false;
65 }
66 
67 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
68 {
69 	if (!vcpu->arch.sysregs_loaded_on_cpu)
70 		goto immediate_read;
71 
72 	/*
73 	 * System registers listed in the switch are not saved on every
74 	 * exit from the guest but are only saved on vcpu_put.
75 	 *
76 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
77 	 * should never be listed below, because the guest cannot modify its
78 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
79 	 * thread when emulating cross-VCPU communication.
80 	 */
81 	switch (reg) {
82 	case CSSELR_EL1:	return read_sysreg_s(SYS_CSSELR_EL1);
83 	case SCTLR_EL1:		return read_sysreg_s(SYS_SCTLR_EL12);
84 	case ACTLR_EL1:		return read_sysreg_s(SYS_ACTLR_EL1);
85 	case CPACR_EL1:		return read_sysreg_s(SYS_CPACR_EL12);
86 	case TTBR0_EL1:		return read_sysreg_s(SYS_TTBR0_EL12);
87 	case TTBR1_EL1:		return read_sysreg_s(SYS_TTBR1_EL12);
88 	case TCR_EL1:		return read_sysreg_s(SYS_TCR_EL12);
89 	case ESR_EL1:		return read_sysreg_s(SYS_ESR_EL12);
90 	case AFSR0_EL1:		return read_sysreg_s(SYS_AFSR0_EL12);
91 	case AFSR1_EL1:		return read_sysreg_s(SYS_AFSR1_EL12);
92 	case FAR_EL1:		return read_sysreg_s(SYS_FAR_EL12);
93 	case MAIR_EL1:		return read_sysreg_s(SYS_MAIR_EL12);
94 	case VBAR_EL1:		return read_sysreg_s(SYS_VBAR_EL12);
95 	case CONTEXTIDR_EL1:	return read_sysreg_s(SYS_CONTEXTIDR_EL12);
96 	case TPIDR_EL0:		return read_sysreg_s(SYS_TPIDR_EL0);
97 	case TPIDRRO_EL0:	return read_sysreg_s(SYS_TPIDRRO_EL0);
98 	case TPIDR_EL1:		return read_sysreg_s(SYS_TPIDR_EL1);
99 	case AMAIR_EL1:		return read_sysreg_s(SYS_AMAIR_EL12);
100 	case CNTKCTL_EL1:	return read_sysreg_s(SYS_CNTKCTL_EL12);
101 	case PAR_EL1:		return read_sysreg_s(SYS_PAR_EL1);
102 	case DACR32_EL2:	return read_sysreg_s(SYS_DACR32_EL2);
103 	case IFSR32_EL2:	return read_sysreg_s(SYS_IFSR32_EL2);
104 	case DBGVCR32_EL2:	return read_sysreg_s(SYS_DBGVCR32_EL2);
105 	}
106 
107 immediate_read:
108 	return __vcpu_sys_reg(vcpu, reg);
109 }
110 
111 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
112 {
113 	if (!vcpu->arch.sysregs_loaded_on_cpu)
114 		goto immediate_write;
115 
116 	/*
117 	 * System registers listed in the switch are not restored on every
118 	 * entry to the guest but are only restored on vcpu_load.
119 	 *
120 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
121 	 * should never be listed below, because the the MPIDR should only be
122 	 * set once, before running the VCPU, and never changed later.
123 	 */
124 	switch (reg) {
125 	case CSSELR_EL1:	write_sysreg_s(val, SYS_CSSELR_EL1);	return;
126 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	return;
127 	case ACTLR_EL1:		write_sysreg_s(val, SYS_ACTLR_EL1);	return;
128 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	return;
129 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	return;
130 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	return;
131 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	return;
132 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	return;
133 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	return;
134 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	return;
135 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	return;
136 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	return;
137 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	return;
138 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12); return;
139 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	return;
140 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	return;
141 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	return;
142 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	return;
143 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	return;
144 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	return;
145 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	return;
146 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	return;
147 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	return;
148 	}
149 
150 immediate_write:
151 	 __vcpu_sys_reg(vcpu, reg) = val;
152 }
153 
154 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
155 static u32 cache_levels;
156 
157 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
158 #define CSSELR_MAX 12
159 
160 /* Which cache CCSIDR represents depends on CSSELR value. */
161 static u32 get_ccsidr(u32 csselr)
162 {
163 	u32 ccsidr;
164 
165 	/* Make sure noone else changes CSSELR during this! */
166 	local_irq_disable();
167 	write_sysreg(csselr, csselr_el1);
168 	isb();
169 	ccsidr = read_sysreg(ccsidr_el1);
170 	local_irq_enable();
171 
172 	return ccsidr;
173 }
174 
175 /*
176  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
177  */
178 static bool access_dcsw(struct kvm_vcpu *vcpu,
179 			struct sys_reg_params *p,
180 			const struct sys_reg_desc *r)
181 {
182 	if (!p->is_write)
183 		return read_from_write_only(vcpu, p, r);
184 
185 	/*
186 	 * Only track S/W ops if we don't have FWB. It still indicates
187 	 * that the guest is a bit broken (S/W operations should only
188 	 * be done by firmware, knowing that there is only a single
189 	 * CPU left in the system, and certainly not from non-secure
190 	 * software).
191 	 */
192 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
193 		kvm_set_way_flush(vcpu);
194 
195 	return true;
196 }
197 
198 /*
199  * Generic accessor for VM registers. Only called as long as HCR_TVM
200  * is set. If the guest enables the MMU, we stop trapping the VM
201  * sys_regs and leave it in complete control of the caches.
202  */
203 static bool access_vm_reg(struct kvm_vcpu *vcpu,
204 			  struct sys_reg_params *p,
205 			  const struct sys_reg_desc *r)
206 {
207 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
208 	u64 val;
209 	int reg = r->reg;
210 
211 	BUG_ON(!p->is_write);
212 
213 	/* See the 32bit mapping in kvm_host.h */
214 	if (p->is_aarch32)
215 		reg = r->reg / 2;
216 
217 	if (!p->is_aarch32 || !p->is_32bit) {
218 		val = p->regval;
219 	} else {
220 		val = vcpu_read_sys_reg(vcpu, reg);
221 		if (r->reg % 2)
222 			val = (p->regval << 32) | (u64)lower_32_bits(val);
223 		else
224 			val = ((u64)upper_32_bits(val) << 32) |
225 				lower_32_bits(p->regval);
226 	}
227 	vcpu_write_sys_reg(vcpu, val, reg);
228 
229 	kvm_toggle_cache(vcpu, was_enabled);
230 	return true;
231 }
232 
233 /*
234  * Trap handler for the GICv3 SGI generation system register.
235  * Forward the request to the VGIC emulation.
236  * The cp15_64 code makes sure this automatically works
237  * for both AArch64 and AArch32 accesses.
238  */
239 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
240 			   struct sys_reg_params *p,
241 			   const struct sys_reg_desc *r)
242 {
243 	bool g1;
244 
245 	if (!p->is_write)
246 		return read_from_write_only(vcpu, p, r);
247 
248 	/*
249 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
250 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
251 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
252 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
253 	 * group.
254 	 */
255 	if (p->is_aarch32) {
256 		switch (p->Op1) {
257 		default:		/* Keep GCC quiet */
258 		case 0:			/* ICC_SGI1R */
259 			g1 = true;
260 			break;
261 		case 1:			/* ICC_ASGI1R */
262 		case 2:			/* ICC_SGI0R */
263 			g1 = false;
264 			break;
265 		}
266 	} else {
267 		switch (p->Op2) {
268 		default:		/* Keep GCC quiet */
269 		case 5:			/* ICC_SGI1R_EL1 */
270 			g1 = true;
271 			break;
272 		case 6:			/* ICC_ASGI1R_EL1 */
273 		case 7:			/* ICC_SGI0R_EL1 */
274 			g1 = false;
275 			break;
276 		}
277 	}
278 
279 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
280 
281 	return true;
282 }
283 
284 static bool access_gic_sre(struct kvm_vcpu *vcpu,
285 			   struct sys_reg_params *p,
286 			   const struct sys_reg_desc *r)
287 {
288 	if (p->is_write)
289 		return ignore_write(vcpu, p);
290 
291 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
292 	return true;
293 }
294 
295 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
296 			struct sys_reg_params *p,
297 			const struct sys_reg_desc *r)
298 {
299 	if (p->is_write)
300 		return ignore_write(vcpu, p);
301 	else
302 		return read_zero(vcpu, p);
303 }
304 
305 /*
306  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
307  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
308  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
309  * treat it separately.
310  */
311 static bool trap_loregion(struct kvm_vcpu *vcpu,
312 			  struct sys_reg_params *p,
313 			  const struct sys_reg_desc *r)
314 {
315 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
316 	u32 sr = sys_reg((u32)r->Op0, (u32)r->Op1,
317 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
318 
319 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
320 		kvm_inject_undefined(vcpu);
321 		return false;
322 	}
323 
324 	if (p->is_write && sr == SYS_LORID_EL1)
325 		return write_to_read_only(vcpu, p, r);
326 
327 	return trap_raz_wi(vcpu, p, r);
328 }
329 
330 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
331 			   struct sys_reg_params *p,
332 			   const struct sys_reg_desc *r)
333 {
334 	if (p->is_write) {
335 		return ignore_write(vcpu, p);
336 	} else {
337 		p->regval = (1 << 3);
338 		return true;
339 	}
340 }
341 
342 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
343 				   struct sys_reg_params *p,
344 				   const struct sys_reg_desc *r)
345 {
346 	if (p->is_write) {
347 		return ignore_write(vcpu, p);
348 	} else {
349 		p->regval = read_sysreg(dbgauthstatus_el1);
350 		return true;
351 	}
352 }
353 
354 /*
355  * We want to avoid world-switching all the DBG registers all the
356  * time:
357  *
358  * - If we've touched any debug register, it is likely that we're
359  *   going to touch more of them. It then makes sense to disable the
360  *   traps and start doing the save/restore dance
361  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
362  *   then mandatory to save/restore the registers, as the guest
363  *   depends on them.
364  *
365  * For this, we use a DIRTY bit, indicating the guest has modified the
366  * debug registers, used as follow:
367  *
368  * On guest entry:
369  * - If the dirty bit is set (because we're coming back from trapping),
370  *   disable the traps, save host registers, restore guest registers.
371  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
372  *   set the dirty bit, disable the traps, save host registers,
373  *   restore guest registers.
374  * - Otherwise, enable the traps
375  *
376  * On guest exit:
377  * - If the dirty bit is set, save guest registers, restore host
378  *   registers and clear the dirty bit. This ensure that the host can
379  *   now use the debug registers.
380  */
381 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
382 			    struct sys_reg_params *p,
383 			    const struct sys_reg_desc *r)
384 {
385 	if (p->is_write) {
386 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
387 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
388 	} else {
389 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
390 	}
391 
392 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
393 
394 	return true;
395 }
396 
397 /*
398  * reg_to_dbg/dbg_to_reg
399  *
400  * A 32 bit write to a debug register leave top bits alone
401  * A 32 bit read from a debug register only returns the bottom bits
402  *
403  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
404  * hyp.S code switches between host and guest values in future.
405  */
406 static void reg_to_dbg(struct kvm_vcpu *vcpu,
407 		       struct sys_reg_params *p,
408 		       u64 *dbg_reg)
409 {
410 	u64 val = p->regval;
411 
412 	if (p->is_32bit) {
413 		val &= 0xffffffffUL;
414 		val |= ((*dbg_reg >> 32) << 32);
415 	}
416 
417 	*dbg_reg = val;
418 	vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
419 }
420 
421 static void dbg_to_reg(struct kvm_vcpu *vcpu,
422 		       struct sys_reg_params *p,
423 		       u64 *dbg_reg)
424 {
425 	p->regval = *dbg_reg;
426 	if (p->is_32bit)
427 		p->regval &= 0xffffffffUL;
428 }
429 
430 static bool trap_bvr(struct kvm_vcpu *vcpu,
431 		     struct sys_reg_params *p,
432 		     const struct sys_reg_desc *rd)
433 {
434 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
435 
436 	if (p->is_write)
437 		reg_to_dbg(vcpu, p, dbg_reg);
438 	else
439 		dbg_to_reg(vcpu, p, dbg_reg);
440 
441 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
442 
443 	return true;
444 }
445 
446 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
447 		const struct kvm_one_reg *reg, void __user *uaddr)
448 {
449 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
450 
451 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
452 		return -EFAULT;
453 	return 0;
454 }
455 
456 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
457 	const struct kvm_one_reg *reg, void __user *uaddr)
458 {
459 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
460 
461 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
462 		return -EFAULT;
463 	return 0;
464 }
465 
466 static void reset_bvr(struct kvm_vcpu *vcpu,
467 		      const struct sys_reg_desc *rd)
468 {
469 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
470 }
471 
472 static bool trap_bcr(struct kvm_vcpu *vcpu,
473 		     struct sys_reg_params *p,
474 		     const struct sys_reg_desc *rd)
475 {
476 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
477 
478 	if (p->is_write)
479 		reg_to_dbg(vcpu, p, dbg_reg);
480 	else
481 		dbg_to_reg(vcpu, p, dbg_reg);
482 
483 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
484 
485 	return true;
486 }
487 
488 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
489 		const struct kvm_one_reg *reg, void __user *uaddr)
490 {
491 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
492 
493 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
494 		return -EFAULT;
495 
496 	return 0;
497 }
498 
499 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
500 	const struct kvm_one_reg *reg, void __user *uaddr)
501 {
502 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
503 
504 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
505 		return -EFAULT;
506 	return 0;
507 }
508 
509 static void reset_bcr(struct kvm_vcpu *vcpu,
510 		      const struct sys_reg_desc *rd)
511 {
512 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
513 }
514 
515 static bool trap_wvr(struct kvm_vcpu *vcpu,
516 		     struct sys_reg_params *p,
517 		     const struct sys_reg_desc *rd)
518 {
519 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
520 
521 	if (p->is_write)
522 		reg_to_dbg(vcpu, p, dbg_reg);
523 	else
524 		dbg_to_reg(vcpu, p, dbg_reg);
525 
526 	trace_trap_reg(__func__, rd->reg, p->is_write,
527 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
528 
529 	return true;
530 }
531 
532 static int set_wvr(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_wvr[rd->reg];
536 
537 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
538 		return -EFAULT;
539 	return 0;
540 }
541 
542 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
543 	const struct kvm_one_reg *reg, void __user *uaddr)
544 {
545 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
546 
547 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
548 		return -EFAULT;
549 	return 0;
550 }
551 
552 static void reset_wvr(struct kvm_vcpu *vcpu,
553 		      const struct sys_reg_desc *rd)
554 {
555 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
556 }
557 
558 static bool trap_wcr(struct kvm_vcpu *vcpu,
559 		     struct sys_reg_params *p,
560 		     const struct sys_reg_desc *rd)
561 {
562 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
563 
564 	if (p->is_write)
565 		reg_to_dbg(vcpu, p, dbg_reg);
566 	else
567 		dbg_to_reg(vcpu, p, dbg_reg);
568 
569 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
570 
571 	return true;
572 }
573 
574 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
575 		const struct kvm_one_reg *reg, void __user *uaddr)
576 {
577 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
578 
579 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
580 		return -EFAULT;
581 	return 0;
582 }
583 
584 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
585 	const struct kvm_one_reg *reg, void __user *uaddr)
586 {
587 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
588 
589 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
590 		return -EFAULT;
591 	return 0;
592 }
593 
594 static void reset_wcr(struct kvm_vcpu *vcpu,
595 		      const struct sys_reg_desc *rd)
596 {
597 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
598 }
599 
600 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
601 {
602 	u64 amair = read_sysreg(amair_el1);
603 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
604 }
605 
606 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
607 {
608 	u64 mpidr;
609 
610 	/*
611 	 * Map the vcpu_id into the first three affinity level fields of
612 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
613 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
614 	 * of the GICv3 to be able to address each CPU directly when
615 	 * sending IPIs.
616 	 */
617 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
618 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
619 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
620 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
621 }
622 
623 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
624 {
625 	u64 pmcr, val;
626 
627 	pmcr = read_sysreg(pmcr_el0);
628 	/*
629 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
630 	 * except PMCR.E resetting to zero.
631 	 */
632 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
633 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
634 	if (!system_supports_32bit_el0())
635 		val |= ARMV8_PMU_PMCR_LC;
636 	__vcpu_sys_reg(vcpu, r->reg) = val;
637 }
638 
639 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
640 {
641 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
642 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
643 
644 	if (!enabled)
645 		kvm_inject_undefined(vcpu);
646 
647 	return !enabled;
648 }
649 
650 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
651 {
652 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
653 }
654 
655 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
656 {
657 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
658 }
659 
660 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
661 {
662 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
663 }
664 
665 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
666 {
667 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
668 }
669 
670 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
671 			const struct sys_reg_desc *r)
672 {
673 	u64 val;
674 
675 	if (!kvm_arm_pmu_v3_ready(vcpu))
676 		return trap_raz_wi(vcpu, p, r);
677 
678 	if (pmu_access_el0_disabled(vcpu))
679 		return false;
680 
681 	if (p->is_write) {
682 		/* Only update writeable bits of PMCR */
683 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
684 		val &= ~ARMV8_PMU_PMCR_MASK;
685 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
686 		if (!system_supports_32bit_el0())
687 			val |= ARMV8_PMU_PMCR_LC;
688 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
689 		kvm_pmu_handle_pmcr(vcpu, val);
690 		kvm_vcpu_pmu_restore_guest(vcpu);
691 	} else {
692 		/* PMCR.P & PMCR.C are RAZ */
693 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
694 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
695 		p->regval = val;
696 	}
697 
698 	return true;
699 }
700 
701 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
702 			  const struct sys_reg_desc *r)
703 {
704 	if (!kvm_arm_pmu_v3_ready(vcpu))
705 		return trap_raz_wi(vcpu, p, r);
706 
707 	if (pmu_access_event_counter_el0_disabled(vcpu))
708 		return false;
709 
710 	if (p->is_write)
711 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
712 	else
713 		/* return PMSELR.SEL field */
714 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
715 			    & ARMV8_PMU_COUNTER_MASK;
716 
717 	return true;
718 }
719 
720 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
721 			  const struct sys_reg_desc *r)
722 {
723 	u64 pmceid;
724 
725 	if (!kvm_arm_pmu_v3_ready(vcpu))
726 		return trap_raz_wi(vcpu, p, r);
727 
728 	BUG_ON(p->is_write);
729 
730 	if (pmu_access_el0_disabled(vcpu))
731 		return false;
732 
733 	if (!(p->Op2 & 1))
734 		pmceid = read_sysreg(pmceid0_el0);
735 	else
736 		pmceid = read_sysreg(pmceid1_el0);
737 
738 	p->regval = pmceid;
739 
740 	return true;
741 }
742 
743 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
744 {
745 	u64 pmcr, val;
746 
747 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
748 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
749 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
750 		kvm_inject_undefined(vcpu);
751 		return false;
752 	}
753 
754 	return true;
755 }
756 
757 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
758 			      struct sys_reg_params *p,
759 			      const struct sys_reg_desc *r)
760 {
761 	u64 idx;
762 
763 	if (!kvm_arm_pmu_v3_ready(vcpu))
764 		return trap_raz_wi(vcpu, p, r);
765 
766 	if (r->CRn == 9 && r->CRm == 13) {
767 		if (r->Op2 == 2) {
768 			/* PMXEVCNTR_EL0 */
769 			if (pmu_access_event_counter_el0_disabled(vcpu))
770 				return false;
771 
772 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
773 			      & ARMV8_PMU_COUNTER_MASK;
774 		} else if (r->Op2 == 0) {
775 			/* PMCCNTR_EL0 */
776 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
777 				return false;
778 
779 			idx = ARMV8_PMU_CYCLE_IDX;
780 		} else {
781 			return false;
782 		}
783 	} else if (r->CRn == 0 && r->CRm == 9) {
784 		/* PMCCNTR */
785 		if (pmu_access_event_counter_el0_disabled(vcpu))
786 			return false;
787 
788 		idx = ARMV8_PMU_CYCLE_IDX;
789 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
790 		/* PMEVCNTRn_EL0 */
791 		if (pmu_access_event_counter_el0_disabled(vcpu))
792 			return false;
793 
794 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
795 	} else {
796 		return false;
797 	}
798 
799 	if (!pmu_counter_idx_valid(vcpu, idx))
800 		return false;
801 
802 	if (p->is_write) {
803 		if (pmu_access_el0_disabled(vcpu))
804 			return false;
805 
806 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
807 	} else {
808 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
809 	}
810 
811 	return true;
812 }
813 
814 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
815 			       const struct sys_reg_desc *r)
816 {
817 	u64 idx, reg;
818 
819 	if (!kvm_arm_pmu_v3_ready(vcpu))
820 		return trap_raz_wi(vcpu, p, r);
821 
822 	if (pmu_access_el0_disabled(vcpu))
823 		return false;
824 
825 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
826 		/* PMXEVTYPER_EL0 */
827 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
828 		reg = PMEVTYPER0_EL0 + idx;
829 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
830 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
831 		if (idx == ARMV8_PMU_CYCLE_IDX)
832 			reg = PMCCFILTR_EL0;
833 		else
834 			/* PMEVTYPERn_EL0 */
835 			reg = PMEVTYPER0_EL0 + idx;
836 	} else {
837 		BUG();
838 	}
839 
840 	if (!pmu_counter_idx_valid(vcpu, idx))
841 		return false;
842 
843 	if (p->is_write) {
844 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
845 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
846 		kvm_vcpu_pmu_restore_guest(vcpu);
847 	} else {
848 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
849 	}
850 
851 	return true;
852 }
853 
854 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
855 			   const struct sys_reg_desc *r)
856 {
857 	u64 val, mask;
858 
859 	if (!kvm_arm_pmu_v3_ready(vcpu))
860 		return trap_raz_wi(vcpu, p, r);
861 
862 	if (pmu_access_el0_disabled(vcpu))
863 		return false;
864 
865 	mask = kvm_pmu_valid_counter_mask(vcpu);
866 	if (p->is_write) {
867 		val = p->regval & mask;
868 		if (r->Op2 & 0x1) {
869 			/* accessing PMCNTENSET_EL0 */
870 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
871 			kvm_pmu_enable_counter_mask(vcpu, val);
872 			kvm_vcpu_pmu_restore_guest(vcpu);
873 		} else {
874 			/* accessing PMCNTENCLR_EL0 */
875 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
876 			kvm_pmu_disable_counter_mask(vcpu, val);
877 		}
878 	} else {
879 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
880 	}
881 
882 	return true;
883 }
884 
885 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
886 			   const struct sys_reg_desc *r)
887 {
888 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
889 
890 	if (!kvm_arm_pmu_v3_ready(vcpu))
891 		return trap_raz_wi(vcpu, p, r);
892 
893 	if (!vcpu_mode_priv(vcpu)) {
894 		kvm_inject_undefined(vcpu);
895 		return false;
896 	}
897 
898 	if (p->is_write) {
899 		u64 val = p->regval & mask;
900 
901 		if (r->Op2 & 0x1)
902 			/* accessing PMINTENSET_EL1 */
903 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
904 		else
905 			/* accessing PMINTENCLR_EL1 */
906 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
907 	} else {
908 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
909 	}
910 
911 	return true;
912 }
913 
914 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
915 			 const struct sys_reg_desc *r)
916 {
917 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
918 
919 	if (!kvm_arm_pmu_v3_ready(vcpu))
920 		return trap_raz_wi(vcpu, p, r);
921 
922 	if (pmu_access_el0_disabled(vcpu))
923 		return false;
924 
925 	if (p->is_write) {
926 		if (r->CRm & 0x2)
927 			/* accessing PMOVSSET_EL0 */
928 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
929 		else
930 			/* accessing PMOVSCLR_EL0 */
931 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
932 	} else {
933 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
934 	}
935 
936 	return true;
937 }
938 
939 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
940 			   const struct sys_reg_desc *r)
941 {
942 	u64 mask;
943 
944 	if (!kvm_arm_pmu_v3_ready(vcpu))
945 		return trap_raz_wi(vcpu, p, r);
946 
947 	if (!p->is_write)
948 		return read_from_write_only(vcpu, p, r);
949 
950 	if (pmu_write_swinc_el0_disabled(vcpu))
951 		return false;
952 
953 	mask = kvm_pmu_valid_counter_mask(vcpu);
954 	kvm_pmu_software_increment(vcpu, p->regval & mask);
955 	return true;
956 }
957 
958 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
959 			     const struct sys_reg_desc *r)
960 {
961 	if (!kvm_arm_pmu_v3_ready(vcpu))
962 		return trap_raz_wi(vcpu, p, r);
963 
964 	if (p->is_write) {
965 		if (!vcpu_mode_priv(vcpu)) {
966 			kvm_inject_undefined(vcpu);
967 			return false;
968 		}
969 
970 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
971 			       p->regval & ARMV8_PMU_USERENR_MASK;
972 	} else {
973 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
974 			    & ARMV8_PMU_USERENR_MASK;
975 	}
976 
977 	return true;
978 }
979 
980 #define reg_to_encoding(x)						\
981 	sys_reg((u32)(x)->Op0, (u32)(x)->Op1,				\
982 		(u32)(x)->CRn, (u32)(x)->CRm, (u32)(x)->Op2);
983 
984 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
985 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
986 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
987 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
988 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
989 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
990 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
991 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
992 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
993 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
994 
995 /* Macro to expand the PMEVCNTRn_EL0 register */
996 #define PMU_PMEVCNTR_EL0(n)						\
997 	{ SYS_DESC(SYS_PMEVCNTRn_EL0(n)),					\
998 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
999 
1000 /* Macro to expand the PMEVTYPERn_EL0 register */
1001 #define PMU_PMEVTYPER_EL0(n)						\
1002 	{ SYS_DESC(SYS_PMEVTYPERn_EL0(n)),					\
1003 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
1004 
1005 static bool access_amu(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1006 			     const struct sys_reg_desc *r)
1007 {
1008 	kvm_inject_undefined(vcpu);
1009 
1010 	return false;
1011 }
1012 
1013 /* Macro to expand the AMU counter and type registers*/
1014 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), access_amu }
1015 #define AMU_AMEVTYPE0_EL0(n) { SYS_DESC(SYS_AMEVTYPE0_EL0(n)), access_amu }
1016 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), access_amu }
1017 #define AMU_AMEVTYPE1_EL0(n) { SYS_DESC(SYS_AMEVTYPE1_EL0(n)), access_amu }
1018 
1019 static bool trap_ptrauth(struct kvm_vcpu *vcpu,
1020 			 struct sys_reg_params *p,
1021 			 const struct sys_reg_desc *rd)
1022 {
1023 	kvm_arm_vcpu_ptrauth_trap(vcpu);
1024 
1025 	/*
1026 	 * Return false for both cases as we never skip the trapped
1027 	 * instruction:
1028 	 *
1029 	 * - Either we re-execute the same key register access instruction
1030 	 *   after enabling ptrauth.
1031 	 * - Or an UNDEF is injected as ptrauth is not supported/enabled.
1032 	 */
1033 	return false;
1034 }
1035 
1036 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1037 			const struct sys_reg_desc *rd)
1038 {
1039 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1040 }
1041 
1042 #define __PTRAUTH_KEY(k)						\
1043 	{ SYS_DESC(SYS_## k), trap_ptrauth, reset_unknown, k,		\
1044 	.visibility = ptrauth_visibility}
1045 
1046 #define PTRAUTH_KEY(k)							\
1047 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1048 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1049 
1050 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1051 			      struct sys_reg_params *p,
1052 			      const struct sys_reg_desc *r)
1053 {
1054 	enum kvm_arch_timers tmr;
1055 	enum kvm_arch_timer_regs treg;
1056 	u64 reg = reg_to_encoding(r);
1057 
1058 	switch (reg) {
1059 	case SYS_CNTP_TVAL_EL0:
1060 	case SYS_AARCH32_CNTP_TVAL:
1061 		tmr = TIMER_PTIMER;
1062 		treg = TIMER_REG_TVAL;
1063 		break;
1064 	case SYS_CNTP_CTL_EL0:
1065 	case SYS_AARCH32_CNTP_CTL:
1066 		tmr = TIMER_PTIMER;
1067 		treg = TIMER_REG_CTL;
1068 		break;
1069 	case SYS_CNTP_CVAL_EL0:
1070 	case SYS_AARCH32_CNTP_CVAL:
1071 		tmr = TIMER_PTIMER;
1072 		treg = TIMER_REG_CVAL;
1073 		break;
1074 	default:
1075 		BUG();
1076 	}
1077 
1078 	if (p->is_write)
1079 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1080 	else
1081 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1082 
1083 	return true;
1084 }
1085 
1086 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1087 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1088 		struct sys_reg_desc const *r, bool raz)
1089 {
1090 	u32 id = sys_reg((u32)r->Op0, (u32)r->Op1,
1091 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
1092 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1093 
1094 	if (id == SYS_ID_AA64PFR0_EL1) {
1095 		if (!vcpu_has_sve(vcpu))
1096 			val &= ~(0xfUL << ID_AA64PFR0_SVE_SHIFT);
1097 		val &= ~(0xfUL << ID_AA64PFR0_AMU_SHIFT);
1098 	} else if (id == SYS_ID_AA64ISAR1_EL1 && !vcpu_has_ptrauth(vcpu)) {
1099 		val &= ~((0xfUL << ID_AA64ISAR1_APA_SHIFT) |
1100 			 (0xfUL << ID_AA64ISAR1_API_SHIFT) |
1101 			 (0xfUL << ID_AA64ISAR1_GPA_SHIFT) |
1102 			 (0xfUL << ID_AA64ISAR1_GPI_SHIFT));
1103 	} else if (id == SYS_ID_AA64DFR0_EL1) {
1104 		/* Limit guests to PMUv3 for ARMv8.1 */
1105 		val = cpuid_feature_cap_perfmon_field(val,
1106 						ID_AA64DFR0_PMUVER_SHIFT,
1107 						ID_AA64DFR0_PMUVER_8_1);
1108 	} else if (id == SYS_ID_DFR0_EL1) {
1109 		/* Limit guests to PMUv3 for ARMv8.1 */
1110 		val = cpuid_feature_cap_perfmon_field(val,
1111 						ID_DFR0_PERFMON_SHIFT,
1112 						ID_DFR0_PERFMON_8_1);
1113 	}
1114 
1115 	return val;
1116 }
1117 
1118 /* cpufeature ID register access trap handlers */
1119 
1120 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1121 			    struct sys_reg_params *p,
1122 			    const struct sys_reg_desc *r,
1123 			    bool raz)
1124 {
1125 	if (p->is_write)
1126 		return write_to_read_only(vcpu, p, r);
1127 
1128 	p->regval = read_id_reg(vcpu, r, raz);
1129 	return true;
1130 }
1131 
1132 static bool access_id_reg(struct kvm_vcpu *vcpu,
1133 			  struct sys_reg_params *p,
1134 			  const struct sys_reg_desc *r)
1135 {
1136 	return __access_id_reg(vcpu, p, r, false);
1137 }
1138 
1139 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1140 			      struct sys_reg_params *p,
1141 			      const struct sys_reg_desc *r)
1142 {
1143 	return __access_id_reg(vcpu, p, r, true);
1144 }
1145 
1146 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1147 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1148 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1149 
1150 /* Visibility overrides for SVE-specific control registers */
1151 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1152 				   const struct sys_reg_desc *rd)
1153 {
1154 	if (vcpu_has_sve(vcpu))
1155 		return 0;
1156 
1157 	return REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1158 }
1159 
1160 /* Visibility overrides for SVE-specific ID registers */
1161 static unsigned int sve_id_visibility(const struct kvm_vcpu *vcpu,
1162 				      const struct sys_reg_desc *rd)
1163 {
1164 	if (vcpu_has_sve(vcpu))
1165 		return 0;
1166 
1167 	return REG_HIDDEN_USER;
1168 }
1169 
1170 /* Generate the emulated ID_AA64ZFR0_EL1 value exposed to the guest */
1171 static u64 guest_id_aa64zfr0_el1(const struct kvm_vcpu *vcpu)
1172 {
1173 	if (!vcpu_has_sve(vcpu))
1174 		return 0;
1175 
1176 	return read_sanitised_ftr_reg(SYS_ID_AA64ZFR0_EL1);
1177 }
1178 
1179 static bool access_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1180 				   struct sys_reg_params *p,
1181 				   const struct sys_reg_desc *rd)
1182 {
1183 	if (p->is_write)
1184 		return write_to_read_only(vcpu, p, rd);
1185 
1186 	p->regval = guest_id_aa64zfr0_el1(vcpu);
1187 	return true;
1188 }
1189 
1190 static int get_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1191 		const struct sys_reg_desc *rd,
1192 		const struct kvm_one_reg *reg, void __user *uaddr)
1193 {
1194 	u64 val;
1195 
1196 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1197 		return -ENOENT;
1198 
1199 	val = guest_id_aa64zfr0_el1(vcpu);
1200 	return reg_to_user(uaddr, &val, reg->id);
1201 }
1202 
1203 static int set_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1204 		const struct sys_reg_desc *rd,
1205 		const struct kvm_one_reg *reg, void __user *uaddr)
1206 {
1207 	const u64 id = sys_reg_to_index(rd);
1208 	int err;
1209 	u64 val;
1210 
1211 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1212 		return -ENOENT;
1213 
1214 	err = reg_from_user(&val, uaddr, id);
1215 	if (err)
1216 		return err;
1217 
1218 	/* This is what we mean by invariant: you can't change it. */
1219 	if (val != guest_id_aa64zfr0_el1(vcpu))
1220 		return -EINVAL;
1221 
1222 	return 0;
1223 }
1224 
1225 /*
1226  * cpufeature ID register user accessors
1227  *
1228  * For now, these registers are immutable for userspace, so no values
1229  * are stored, and for set_id_reg() we don't allow the effective value
1230  * to be changed.
1231  */
1232 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1233 			const struct sys_reg_desc *rd, void __user *uaddr,
1234 			bool raz)
1235 {
1236 	const u64 id = sys_reg_to_index(rd);
1237 	const u64 val = read_id_reg(vcpu, rd, raz);
1238 
1239 	return reg_to_user(uaddr, &val, id);
1240 }
1241 
1242 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1243 			const struct sys_reg_desc *rd, void __user *uaddr,
1244 			bool raz)
1245 {
1246 	const u64 id = sys_reg_to_index(rd);
1247 	int err;
1248 	u64 val;
1249 
1250 	err = reg_from_user(&val, uaddr, id);
1251 	if (err)
1252 		return err;
1253 
1254 	/* This is what we mean by invariant: you can't change it. */
1255 	if (val != read_id_reg(vcpu, rd, raz))
1256 		return -EINVAL;
1257 
1258 	return 0;
1259 }
1260 
1261 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1262 		      const struct kvm_one_reg *reg, void __user *uaddr)
1263 {
1264 	return __get_id_reg(vcpu, rd, uaddr, false);
1265 }
1266 
1267 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1268 		      const struct kvm_one_reg *reg, void __user *uaddr)
1269 {
1270 	return __set_id_reg(vcpu, rd, uaddr, false);
1271 }
1272 
1273 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1274 			  const struct kvm_one_reg *reg, void __user *uaddr)
1275 {
1276 	return __get_id_reg(vcpu, rd, uaddr, true);
1277 }
1278 
1279 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1280 			  const struct kvm_one_reg *reg, void __user *uaddr)
1281 {
1282 	return __set_id_reg(vcpu, rd, uaddr, true);
1283 }
1284 
1285 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1286 		       const struct sys_reg_desc *r)
1287 {
1288 	if (p->is_write)
1289 		return write_to_read_only(vcpu, p, r);
1290 
1291 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1292 	return true;
1293 }
1294 
1295 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1296 			 const struct sys_reg_desc *r)
1297 {
1298 	if (p->is_write)
1299 		return write_to_read_only(vcpu, p, r);
1300 
1301 	p->regval = read_sysreg(clidr_el1);
1302 	return true;
1303 }
1304 
1305 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1306 			  const struct sys_reg_desc *r)
1307 {
1308 	if (p->is_write)
1309 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
1310 	else
1311 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
1312 	return true;
1313 }
1314 
1315 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1316 			  const struct sys_reg_desc *r)
1317 {
1318 	u32 csselr;
1319 
1320 	if (p->is_write)
1321 		return write_to_read_only(vcpu, p, r);
1322 
1323 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1324 	p->regval = get_ccsidr(csselr);
1325 
1326 	/*
1327 	 * Guests should not be doing cache operations by set/way at all, and
1328 	 * for this reason, we trap them and attempt to infer the intent, so
1329 	 * that we can flush the entire guest's address space at the appropriate
1330 	 * time.
1331 	 * To prevent this trapping from causing performance problems, let's
1332 	 * expose the geometry of all data and unified caches (which are
1333 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1334 	 * [If guests should attempt to infer aliasing properties from the
1335 	 * geometry (which is not permitted by the architecture), they would
1336 	 * only do so for virtually indexed caches.]
1337 	 */
1338 	if (!(csselr & 1)) // data or unified cache
1339 		p->regval &= ~GENMASK(27, 3);
1340 	return true;
1341 }
1342 
1343 /* sys_reg_desc initialiser for known cpufeature ID registers */
1344 #define ID_SANITISED(name) {			\
1345 	SYS_DESC(SYS_##name),			\
1346 	.access	= access_id_reg,		\
1347 	.get_user = get_id_reg,			\
1348 	.set_user = set_id_reg,			\
1349 }
1350 
1351 /*
1352  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1353  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1354  * (1 <= crm < 8, 0 <= Op2 < 8).
1355  */
1356 #define ID_UNALLOCATED(crm, op2) {			\
1357 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1358 	.access = access_raz_id_reg,			\
1359 	.get_user = get_raz_id_reg,			\
1360 	.set_user = set_raz_id_reg,			\
1361 }
1362 
1363 /*
1364  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1365  * For now, these are exposed just like unallocated ID regs: they appear
1366  * RAZ for the guest.
1367  */
1368 #define ID_HIDDEN(name) {			\
1369 	SYS_DESC(SYS_##name),			\
1370 	.access = access_raz_id_reg,		\
1371 	.get_user = get_raz_id_reg,		\
1372 	.set_user = set_raz_id_reg,		\
1373 }
1374 
1375 /*
1376  * Architected system registers.
1377  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1378  *
1379  * Debug handling: We do trap most, if not all debug related system
1380  * registers. The implementation is good enough to ensure that a guest
1381  * can use these with minimal performance degradation. The drawback is
1382  * that we don't implement any of the external debug, none of the
1383  * OSlock protocol. This should be revisited if we ever encounter a
1384  * more demanding guest...
1385  */
1386 static const struct sys_reg_desc sys_reg_descs[] = {
1387 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1388 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1389 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1390 
1391 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1392 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1393 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1394 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1395 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1396 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1397 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1398 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1399 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1400 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1401 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1402 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1403 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1404 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1405 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1406 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1407 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1408 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1409 
1410 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1411 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1412 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1413 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1414 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1415 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1416 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1417 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1418 
1419 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1420 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1421 	// DBGDTR[TR]X_EL0 share the same encoding
1422 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1423 
1424 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1425 
1426 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1427 
1428 	/*
1429 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1430 	 * entries in arm64_ftr_regs[].
1431 	 */
1432 
1433 	/* AArch64 mappings of the AArch32 ID registers */
1434 	/* CRm=1 */
1435 	ID_SANITISED(ID_PFR0_EL1),
1436 	ID_SANITISED(ID_PFR1_EL1),
1437 	ID_SANITISED(ID_DFR0_EL1),
1438 	ID_HIDDEN(ID_AFR0_EL1),
1439 	ID_SANITISED(ID_MMFR0_EL1),
1440 	ID_SANITISED(ID_MMFR1_EL1),
1441 	ID_SANITISED(ID_MMFR2_EL1),
1442 	ID_SANITISED(ID_MMFR3_EL1),
1443 
1444 	/* CRm=2 */
1445 	ID_SANITISED(ID_ISAR0_EL1),
1446 	ID_SANITISED(ID_ISAR1_EL1),
1447 	ID_SANITISED(ID_ISAR2_EL1),
1448 	ID_SANITISED(ID_ISAR3_EL1),
1449 	ID_SANITISED(ID_ISAR4_EL1),
1450 	ID_SANITISED(ID_ISAR5_EL1),
1451 	ID_SANITISED(ID_MMFR4_EL1),
1452 	ID_SANITISED(ID_ISAR6_EL1),
1453 
1454 	/* CRm=3 */
1455 	ID_SANITISED(MVFR0_EL1),
1456 	ID_SANITISED(MVFR1_EL1),
1457 	ID_SANITISED(MVFR2_EL1),
1458 	ID_UNALLOCATED(3,3),
1459 	ID_UNALLOCATED(3,4),
1460 	ID_UNALLOCATED(3,5),
1461 	ID_UNALLOCATED(3,6),
1462 	ID_UNALLOCATED(3,7),
1463 
1464 	/* AArch64 ID registers */
1465 	/* CRm=4 */
1466 	ID_SANITISED(ID_AA64PFR0_EL1),
1467 	ID_SANITISED(ID_AA64PFR1_EL1),
1468 	ID_UNALLOCATED(4,2),
1469 	ID_UNALLOCATED(4,3),
1470 	{ 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 },
1471 	ID_UNALLOCATED(4,5),
1472 	ID_UNALLOCATED(4,6),
1473 	ID_UNALLOCATED(4,7),
1474 
1475 	/* CRm=5 */
1476 	ID_SANITISED(ID_AA64DFR0_EL1),
1477 	ID_SANITISED(ID_AA64DFR1_EL1),
1478 	ID_UNALLOCATED(5,2),
1479 	ID_UNALLOCATED(5,3),
1480 	ID_HIDDEN(ID_AA64AFR0_EL1),
1481 	ID_HIDDEN(ID_AA64AFR1_EL1),
1482 	ID_UNALLOCATED(5,6),
1483 	ID_UNALLOCATED(5,7),
1484 
1485 	/* CRm=6 */
1486 	ID_SANITISED(ID_AA64ISAR0_EL1),
1487 	ID_SANITISED(ID_AA64ISAR1_EL1),
1488 	ID_UNALLOCATED(6,2),
1489 	ID_UNALLOCATED(6,3),
1490 	ID_UNALLOCATED(6,4),
1491 	ID_UNALLOCATED(6,5),
1492 	ID_UNALLOCATED(6,6),
1493 	ID_UNALLOCATED(6,7),
1494 
1495 	/* CRm=7 */
1496 	ID_SANITISED(ID_AA64MMFR0_EL1),
1497 	ID_SANITISED(ID_AA64MMFR1_EL1),
1498 	ID_SANITISED(ID_AA64MMFR2_EL1),
1499 	ID_UNALLOCATED(7,3),
1500 	ID_UNALLOCATED(7,4),
1501 	ID_UNALLOCATED(7,5),
1502 	ID_UNALLOCATED(7,6),
1503 	ID_UNALLOCATED(7,7),
1504 
1505 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1506 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1507 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1508 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1509 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1510 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1511 
1512 	PTRAUTH_KEY(APIA),
1513 	PTRAUTH_KEY(APIB),
1514 	PTRAUTH_KEY(APDA),
1515 	PTRAUTH_KEY(APDB),
1516 	PTRAUTH_KEY(APGA),
1517 
1518 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1519 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1520 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1521 
1522 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1523 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1524 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1525 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1526 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1527 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1528 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1529 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1530 
1531 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1532 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1533 
1534 	{ SYS_DESC(SYS_PMINTENSET_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1535 	{ SYS_DESC(SYS_PMINTENCLR_EL1), access_pminten, NULL, PMINTENSET_EL1 },
1536 
1537 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1538 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1539 
1540 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1541 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1542 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1543 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1544 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1545 
1546 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1547 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1548 
1549 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1550 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1551 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1552 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1553 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1554 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1555 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1556 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1557 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1558 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1559 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1560 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1561 
1562 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1563 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1564 
1565 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1566 
1567 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1568 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1569 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1570 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1571 
1572 	{ SYS_DESC(SYS_PMCR_EL0), access_pmcr, reset_pmcr, PMCR_EL0 },
1573 	{ SYS_DESC(SYS_PMCNTENSET_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1574 	{ SYS_DESC(SYS_PMCNTENCLR_EL0), access_pmcnten, NULL, PMCNTENSET_EL0 },
1575 	{ SYS_DESC(SYS_PMOVSCLR_EL0), access_pmovs, NULL, PMOVSSET_EL0 },
1576 	{ SYS_DESC(SYS_PMSWINC_EL0), access_pmswinc, reset_unknown, PMSWINC_EL0 },
1577 	{ SYS_DESC(SYS_PMSELR_EL0), access_pmselr, reset_unknown, PMSELR_EL0 },
1578 	{ SYS_DESC(SYS_PMCEID0_EL0), access_pmceid },
1579 	{ SYS_DESC(SYS_PMCEID1_EL0), access_pmceid },
1580 	{ SYS_DESC(SYS_PMCCNTR_EL0), access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1581 	{ SYS_DESC(SYS_PMXEVTYPER_EL0), access_pmu_evtyper },
1582 	{ SYS_DESC(SYS_PMXEVCNTR_EL0), access_pmu_evcntr },
1583 	/*
1584 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1585 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1586 	 */
1587 	{ SYS_DESC(SYS_PMUSERENR_EL0), access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1588 	{ SYS_DESC(SYS_PMOVSSET_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1589 
1590 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1591 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1592 
1593 	{ SYS_DESC(SYS_AMCR_EL0), access_amu },
1594 	{ SYS_DESC(SYS_AMCFGR_EL0), access_amu },
1595 	{ SYS_DESC(SYS_AMCGCR_EL0), access_amu },
1596 	{ SYS_DESC(SYS_AMUSERENR_EL0), access_amu },
1597 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), access_amu },
1598 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), access_amu },
1599 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), access_amu },
1600 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), access_amu },
1601 	AMU_AMEVCNTR0_EL0(0),
1602 	AMU_AMEVCNTR0_EL0(1),
1603 	AMU_AMEVCNTR0_EL0(2),
1604 	AMU_AMEVCNTR0_EL0(3),
1605 	AMU_AMEVCNTR0_EL0(4),
1606 	AMU_AMEVCNTR0_EL0(5),
1607 	AMU_AMEVCNTR0_EL0(6),
1608 	AMU_AMEVCNTR0_EL0(7),
1609 	AMU_AMEVCNTR0_EL0(8),
1610 	AMU_AMEVCNTR0_EL0(9),
1611 	AMU_AMEVCNTR0_EL0(10),
1612 	AMU_AMEVCNTR0_EL0(11),
1613 	AMU_AMEVCNTR0_EL0(12),
1614 	AMU_AMEVCNTR0_EL0(13),
1615 	AMU_AMEVCNTR0_EL0(14),
1616 	AMU_AMEVCNTR0_EL0(15),
1617 	AMU_AMEVTYPE0_EL0(0),
1618 	AMU_AMEVTYPE0_EL0(1),
1619 	AMU_AMEVTYPE0_EL0(2),
1620 	AMU_AMEVTYPE0_EL0(3),
1621 	AMU_AMEVTYPE0_EL0(4),
1622 	AMU_AMEVTYPE0_EL0(5),
1623 	AMU_AMEVTYPE0_EL0(6),
1624 	AMU_AMEVTYPE0_EL0(7),
1625 	AMU_AMEVTYPE0_EL0(8),
1626 	AMU_AMEVTYPE0_EL0(9),
1627 	AMU_AMEVTYPE0_EL0(10),
1628 	AMU_AMEVTYPE0_EL0(11),
1629 	AMU_AMEVTYPE0_EL0(12),
1630 	AMU_AMEVTYPE0_EL0(13),
1631 	AMU_AMEVTYPE0_EL0(14),
1632 	AMU_AMEVTYPE0_EL0(15),
1633 	AMU_AMEVCNTR1_EL0(0),
1634 	AMU_AMEVCNTR1_EL0(1),
1635 	AMU_AMEVCNTR1_EL0(2),
1636 	AMU_AMEVCNTR1_EL0(3),
1637 	AMU_AMEVCNTR1_EL0(4),
1638 	AMU_AMEVCNTR1_EL0(5),
1639 	AMU_AMEVCNTR1_EL0(6),
1640 	AMU_AMEVCNTR1_EL0(7),
1641 	AMU_AMEVCNTR1_EL0(8),
1642 	AMU_AMEVCNTR1_EL0(9),
1643 	AMU_AMEVCNTR1_EL0(10),
1644 	AMU_AMEVCNTR1_EL0(11),
1645 	AMU_AMEVCNTR1_EL0(12),
1646 	AMU_AMEVCNTR1_EL0(13),
1647 	AMU_AMEVCNTR1_EL0(14),
1648 	AMU_AMEVCNTR1_EL0(15),
1649 	AMU_AMEVTYPE1_EL0(0),
1650 	AMU_AMEVTYPE1_EL0(1),
1651 	AMU_AMEVTYPE1_EL0(2),
1652 	AMU_AMEVTYPE1_EL0(3),
1653 	AMU_AMEVTYPE1_EL0(4),
1654 	AMU_AMEVTYPE1_EL0(5),
1655 	AMU_AMEVTYPE1_EL0(6),
1656 	AMU_AMEVTYPE1_EL0(7),
1657 	AMU_AMEVTYPE1_EL0(8),
1658 	AMU_AMEVTYPE1_EL0(9),
1659 	AMU_AMEVTYPE1_EL0(10),
1660 	AMU_AMEVTYPE1_EL0(11),
1661 	AMU_AMEVTYPE1_EL0(12),
1662 	AMU_AMEVTYPE1_EL0(13),
1663 	AMU_AMEVTYPE1_EL0(14),
1664 	AMU_AMEVTYPE1_EL0(15),
1665 
1666 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1667 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1668 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1669 
1670 	/* PMEVCNTRn_EL0 */
1671 	PMU_PMEVCNTR_EL0(0),
1672 	PMU_PMEVCNTR_EL0(1),
1673 	PMU_PMEVCNTR_EL0(2),
1674 	PMU_PMEVCNTR_EL0(3),
1675 	PMU_PMEVCNTR_EL0(4),
1676 	PMU_PMEVCNTR_EL0(5),
1677 	PMU_PMEVCNTR_EL0(6),
1678 	PMU_PMEVCNTR_EL0(7),
1679 	PMU_PMEVCNTR_EL0(8),
1680 	PMU_PMEVCNTR_EL0(9),
1681 	PMU_PMEVCNTR_EL0(10),
1682 	PMU_PMEVCNTR_EL0(11),
1683 	PMU_PMEVCNTR_EL0(12),
1684 	PMU_PMEVCNTR_EL0(13),
1685 	PMU_PMEVCNTR_EL0(14),
1686 	PMU_PMEVCNTR_EL0(15),
1687 	PMU_PMEVCNTR_EL0(16),
1688 	PMU_PMEVCNTR_EL0(17),
1689 	PMU_PMEVCNTR_EL0(18),
1690 	PMU_PMEVCNTR_EL0(19),
1691 	PMU_PMEVCNTR_EL0(20),
1692 	PMU_PMEVCNTR_EL0(21),
1693 	PMU_PMEVCNTR_EL0(22),
1694 	PMU_PMEVCNTR_EL0(23),
1695 	PMU_PMEVCNTR_EL0(24),
1696 	PMU_PMEVCNTR_EL0(25),
1697 	PMU_PMEVCNTR_EL0(26),
1698 	PMU_PMEVCNTR_EL0(27),
1699 	PMU_PMEVCNTR_EL0(28),
1700 	PMU_PMEVCNTR_EL0(29),
1701 	PMU_PMEVCNTR_EL0(30),
1702 	/* PMEVTYPERn_EL0 */
1703 	PMU_PMEVTYPER_EL0(0),
1704 	PMU_PMEVTYPER_EL0(1),
1705 	PMU_PMEVTYPER_EL0(2),
1706 	PMU_PMEVTYPER_EL0(3),
1707 	PMU_PMEVTYPER_EL0(4),
1708 	PMU_PMEVTYPER_EL0(5),
1709 	PMU_PMEVTYPER_EL0(6),
1710 	PMU_PMEVTYPER_EL0(7),
1711 	PMU_PMEVTYPER_EL0(8),
1712 	PMU_PMEVTYPER_EL0(9),
1713 	PMU_PMEVTYPER_EL0(10),
1714 	PMU_PMEVTYPER_EL0(11),
1715 	PMU_PMEVTYPER_EL0(12),
1716 	PMU_PMEVTYPER_EL0(13),
1717 	PMU_PMEVTYPER_EL0(14),
1718 	PMU_PMEVTYPER_EL0(15),
1719 	PMU_PMEVTYPER_EL0(16),
1720 	PMU_PMEVTYPER_EL0(17),
1721 	PMU_PMEVTYPER_EL0(18),
1722 	PMU_PMEVTYPER_EL0(19),
1723 	PMU_PMEVTYPER_EL0(20),
1724 	PMU_PMEVTYPER_EL0(21),
1725 	PMU_PMEVTYPER_EL0(22),
1726 	PMU_PMEVTYPER_EL0(23),
1727 	PMU_PMEVTYPER_EL0(24),
1728 	PMU_PMEVTYPER_EL0(25),
1729 	PMU_PMEVTYPER_EL0(26),
1730 	PMU_PMEVTYPER_EL0(27),
1731 	PMU_PMEVTYPER_EL0(28),
1732 	PMU_PMEVTYPER_EL0(29),
1733 	PMU_PMEVTYPER_EL0(30),
1734 	/*
1735 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1736 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1737 	 */
1738 	{ SYS_DESC(SYS_PMCCFILTR_EL0), access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1739 
1740 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1741 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1742 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1743 };
1744 
1745 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1746 			struct sys_reg_params *p,
1747 			const struct sys_reg_desc *r)
1748 {
1749 	if (p->is_write) {
1750 		return ignore_write(vcpu, p);
1751 	} else {
1752 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1753 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1754 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1755 
1756 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1757 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1758 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1759 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1760 		return true;
1761 	}
1762 }
1763 
1764 static bool trap_debug32(struct kvm_vcpu *vcpu,
1765 			 struct sys_reg_params *p,
1766 			 const struct sys_reg_desc *r)
1767 {
1768 	if (p->is_write) {
1769 		vcpu_cp14(vcpu, r->reg) = p->regval;
1770 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1771 	} else {
1772 		p->regval = vcpu_cp14(vcpu, r->reg);
1773 	}
1774 
1775 	return true;
1776 }
1777 
1778 /* AArch32 debug register mappings
1779  *
1780  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1781  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1782  *
1783  * All control registers and watchpoint value registers are mapped to
1784  * the lower 32 bits of their AArch64 equivalents. We share the trap
1785  * handlers with the above AArch64 code which checks what mode the
1786  * system is in.
1787  */
1788 
1789 static bool trap_xvr(struct kvm_vcpu *vcpu,
1790 		     struct sys_reg_params *p,
1791 		     const struct sys_reg_desc *rd)
1792 {
1793 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1794 
1795 	if (p->is_write) {
1796 		u64 val = *dbg_reg;
1797 
1798 		val &= 0xffffffffUL;
1799 		val |= p->regval << 32;
1800 		*dbg_reg = val;
1801 
1802 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1803 	} else {
1804 		p->regval = *dbg_reg >> 32;
1805 	}
1806 
1807 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1808 
1809 	return true;
1810 }
1811 
1812 #define DBG_BCR_BVR_WCR_WVR(n)						\
1813 	/* DBGBVRn */							\
1814 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1815 	/* DBGBCRn */							\
1816 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1817 	/* DBGWVRn */							\
1818 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1819 	/* DBGWCRn */							\
1820 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1821 
1822 #define DBGBXVR(n)							\
1823 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1824 
1825 /*
1826  * Trapped cp14 registers. We generally ignore most of the external
1827  * debug, on the principle that they don't really make sense to a
1828  * guest. Revisit this one day, would this principle change.
1829  */
1830 static const struct sys_reg_desc cp14_regs[] = {
1831 	/* DBGIDR */
1832 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1833 	/* DBGDTRRXext */
1834 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1835 
1836 	DBG_BCR_BVR_WCR_WVR(0),
1837 	/* DBGDSCRint */
1838 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1839 	DBG_BCR_BVR_WCR_WVR(1),
1840 	/* DBGDCCINT */
1841 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1842 	/* DBGDSCRext */
1843 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1844 	DBG_BCR_BVR_WCR_WVR(2),
1845 	/* DBGDTR[RT]Xint */
1846 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1847 	/* DBGDTR[RT]Xext */
1848 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1849 	DBG_BCR_BVR_WCR_WVR(3),
1850 	DBG_BCR_BVR_WCR_WVR(4),
1851 	DBG_BCR_BVR_WCR_WVR(5),
1852 	/* DBGWFAR */
1853 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1854 	/* DBGOSECCR */
1855 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1856 	DBG_BCR_BVR_WCR_WVR(6),
1857 	/* DBGVCR */
1858 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1859 	DBG_BCR_BVR_WCR_WVR(7),
1860 	DBG_BCR_BVR_WCR_WVR(8),
1861 	DBG_BCR_BVR_WCR_WVR(9),
1862 	DBG_BCR_BVR_WCR_WVR(10),
1863 	DBG_BCR_BVR_WCR_WVR(11),
1864 	DBG_BCR_BVR_WCR_WVR(12),
1865 	DBG_BCR_BVR_WCR_WVR(13),
1866 	DBG_BCR_BVR_WCR_WVR(14),
1867 	DBG_BCR_BVR_WCR_WVR(15),
1868 
1869 	/* DBGDRAR (32bit) */
1870 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1871 
1872 	DBGBXVR(0),
1873 	/* DBGOSLAR */
1874 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1875 	DBGBXVR(1),
1876 	/* DBGOSLSR */
1877 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1878 	DBGBXVR(2),
1879 	DBGBXVR(3),
1880 	/* DBGOSDLR */
1881 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1882 	DBGBXVR(4),
1883 	/* DBGPRCR */
1884 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1885 	DBGBXVR(5),
1886 	DBGBXVR(6),
1887 	DBGBXVR(7),
1888 	DBGBXVR(8),
1889 	DBGBXVR(9),
1890 	DBGBXVR(10),
1891 	DBGBXVR(11),
1892 	DBGBXVR(12),
1893 	DBGBXVR(13),
1894 	DBGBXVR(14),
1895 	DBGBXVR(15),
1896 
1897 	/* DBGDSAR (32bit) */
1898 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1899 
1900 	/* DBGDEVID2 */
1901 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1902 	/* DBGDEVID1 */
1903 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1904 	/* DBGDEVID */
1905 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1906 	/* DBGCLAIMSET */
1907 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1908 	/* DBGCLAIMCLR */
1909 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1910 	/* DBGAUTHSTATUS */
1911 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1912 };
1913 
1914 /* Trapped cp14 64bit registers */
1915 static const struct sys_reg_desc cp14_64_regs[] = {
1916 	/* DBGDRAR (64bit) */
1917 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1918 
1919 	/* DBGDSAR (64bit) */
1920 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1921 };
1922 
1923 /* Macro to expand the PMEVCNTRn register */
1924 #define PMU_PMEVCNTR(n)							\
1925 	/* PMEVCNTRn */							\
1926 	{ Op1(0), CRn(0b1110),						\
1927 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1928 	  access_pmu_evcntr }
1929 
1930 /* Macro to expand the PMEVTYPERn register */
1931 #define PMU_PMEVTYPER(n)						\
1932 	/* PMEVTYPERn */						\
1933 	{ Op1(0), CRn(0b1110),						\
1934 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1935 	  access_pmu_evtyper }
1936 
1937 /*
1938  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1939  * depending on the way they are accessed (as a 32bit or a 64bit
1940  * register).
1941  */
1942 static const struct sys_reg_desc cp15_regs[] = {
1943 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1944 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1945 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1946 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1947 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1948 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1949 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1950 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1951 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1952 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1953 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1954 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1955 
1956 	/*
1957 	 * DC{C,I,CI}SW operations:
1958 	 */
1959 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1960 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1961 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1962 
1963 	/* PMU */
1964 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1965 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1966 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1967 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1968 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1969 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1970 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1971 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1972 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1973 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1974 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1975 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1976 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1977 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1978 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1979 
1980 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
1981 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
1982 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
1983 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
1984 
1985 	/* ICC_SRE */
1986 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
1987 
1988 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
1989 
1990 	/* Arch Tmers */
1991 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
1992 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
1993 
1994 	/* PMEVCNTRn */
1995 	PMU_PMEVCNTR(0),
1996 	PMU_PMEVCNTR(1),
1997 	PMU_PMEVCNTR(2),
1998 	PMU_PMEVCNTR(3),
1999 	PMU_PMEVCNTR(4),
2000 	PMU_PMEVCNTR(5),
2001 	PMU_PMEVCNTR(6),
2002 	PMU_PMEVCNTR(7),
2003 	PMU_PMEVCNTR(8),
2004 	PMU_PMEVCNTR(9),
2005 	PMU_PMEVCNTR(10),
2006 	PMU_PMEVCNTR(11),
2007 	PMU_PMEVCNTR(12),
2008 	PMU_PMEVCNTR(13),
2009 	PMU_PMEVCNTR(14),
2010 	PMU_PMEVCNTR(15),
2011 	PMU_PMEVCNTR(16),
2012 	PMU_PMEVCNTR(17),
2013 	PMU_PMEVCNTR(18),
2014 	PMU_PMEVCNTR(19),
2015 	PMU_PMEVCNTR(20),
2016 	PMU_PMEVCNTR(21),
2017 	PMU_PMEVCNTR(22),
2018 	PMU_PMEVCNTR(23),
2019 	PMU_PMEVCNTR(24),
2020 	PMU_PMEVCNTR(25),
2021 	PMU_PMEVCNTR(26),
2022 	PMU_PMEVCNTR(27),
2023 	PMU_PMEVCNTR(28),
2024 	PMU_PMEVCNTR(29),
2025 	PMU_PMEVCNTR(30),
2026 	/* PMEVTYPERn */
2027 	PMU_PMEVTYPER(0),
2028 	PMU_PMEVTYPER(1),
2029 	PMU_PMEVTYPER(2),
2030 	PMU_PMEVTYPER(3),
2031 	PMU_PMEVTYPER(4),
2032 	PMU_PMEVTYPER(5),
2033 	PMU_PMEVTYPER(6),
2034 	PMU_PMEVTYPER(7),
2035 	PMU_PMEVTYPER(8),
2036 	PMU_PMEVTYPER(9),
2037 	PMU_PMEVTYPER(10),
2038 	PMU_PMEVTYPER(11),
2039 	PMU_PMEVTYPER(12),
2040 	PMU_PMEVTYPER(13),
2041 	PMU_PMEVTYPER(14),
2042 	PMU_PMEVTYPER(15),
2043 	PMU_PMEVTYPER(16),
2044 	PMU_PMEVTYPER(17),
2045 	PMU_PMEVTYPER(18),
2046 	PMU_PMEVTYPER(19),
2047 	PMU_PMEVTYPER(20),
2048 	PMU_PMEVTYPER(21),
2049 	PMU_PMEVTYPER(22),
2050 	PMU_PMEVTYPER(23),
2051 	PMU_PMEVTYPER(24),
2052 	PMU_PMEVTYPER(25),
2053 	PMU_PMEVTYPER(26),
2054 	PMU_PMEVTYPER(27),
2055 	PMU_PMEVTYPER(28),
2056 	PMU_PMEVTYPER(29),
2057 	PMU_PMEVTYPER(30),
2058 	/* PMCCFILTR */
2059 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2060 
2061 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2062 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2063 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, c0_CSSELR },
2064 };
2065 
2066 static const struct sys_reg_desc cp15_64_regs[] = {
2067 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
2068 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2069 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2070 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
2071 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2072 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2073 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2074 };
2075 
2076 /* Target specific emulation tables */
2077 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
2078 
2079 void kvm_register_target_sys_reg_table(unsigned int target,
2080 				       struct kvm_sys_reg_target_table *table)
2081 {
2082 	target_tables[target] = table;
2083 }
2084 
2085 /* Get specific register table for this target. */
2086 static const struct sys_reg_desc *get_target_table(unsigned target,
2087 						   bool mode_is_64,
2088 						   size_t *num)
2089 {
2090 	struct kvm_sys_reg_target_table *table;
2091 
2092 	table = target_tables[target];
2093 	if (mode_is_64) {
2094 		*num = table->table64.num;
2095 		return table->table64.table;
2096 	} else {
2097 		*num = table->table32.num;
2098 		return table->table32.table;
2099 	}
2100 }
2101 
2102 static int match_sys_reg(const void *key, const void *elt)
2103 {
2104 	const unsigned long pval = (unsigned long)key;
2105 	const struct sys_reg_desc *r = elt;
2106 
2107 	return pval - reg_to_encoding(r);
2108 }
2109 
2110 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
2111 					 const struct sys_reg_desc table[],
2112 					 unsigned int num)
2113 {
2114 	unsigned long pval = reg_to_encoding(params);
2115 
2116 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
2117 }
2118 
2119 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
2120 {
2121 	kvm_inject_undefined(vcpu);
2122 	return 1;
2123 }
2124 
2125 static void perform_access(struct kvm_vcpu *vcpu,
2126 			   struct sys_reg_params *params,
2127 			   const struct sys_reg_desc *r)
2128 {
2129 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2130 
2131 	/* Check for regs disabled by runtime config */
2132 	if (sysreg_hidden_from_guest(vcpu, r)) {
2133 		kvm_inject_undefined(vcpu);
2134 		return;
2135 	}
2136 
2137 	/*
2138 	 * Not having an accessor means that we have configured a trap
2139 	 * that we don't know how to handle. This certainly qualifies
2140 	 * as a gross bug that should be fixed right away.
2141 	 */
2142 	BUG_ON(!r->access);
2143 
2144 	/* Skip instruction if instructed so */
2145 	if (likely(r->access(vcpu, params, r)))
2146 		kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
2147 }
2148 
2149 /*
2150  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2151  *                call the corresponding trap handler.
2152  *
2153  * @params: pointer to the descriptor of the access
2154  * @table: array of trap descriptors
2155  * @num: size of the trap descriptor array
2156  *
2157  * Return 0 if the access has been handled, and -1 if not.
2158  */
2159 static int emulate_cp(struct kvm_vcpu *vcpu,
2160 		      struct sys_reg_params *params,
2161 		      const struct sys_reg_desc *table,
2162 		      size_t num)
2163 {
2164 	const struct sys_reg_desc *r;
2165 
2166 	if (!table)
2167 		return -1;	/* Not handled */
2168 
2169 	r = find_reg(params, table, num);
2170 
2171 	if (r) {
2172 		perform_access(vcpu, params, r);
2173 		return 0;
2174 	}
2175 
2176 	/* Not handled */
2177 	return -1;
2178 }
2179 
2180 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2181 				struct sys_reg_params *params)
2182 {
2183 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
2184 	int cp = -1;
2185 
2186 	switch(hsr_ec) {
2187 	case ESR_ELx_EC_CP15_32:
2188 	case ESR_ELx_EC_CP15_64:
2189 		cp = 15;
2190 		break;
2191 	case ESR_ELx_EC_CP14_MR:
2192 	case ESR_ELx_EC_CP14_64:
2193 		cp = 14;
2194 		break;
2195 	default:
2196 		WARN_ON(1);
2197 	}
2198 
2199 	print_sys_reg_msg(params,
2200 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2201 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2202 	kvm_inject_undefined(vcpu);
2203 }
2204 
2205 /**
2206  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2207  * @vcpu: The VCPU pointer
2208  * @run:  The kvm_run struct
2209  */
2210 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2211 			    const struct sys_reg_desc *global,
2212 			    size_t nr_global,
2213 			    const struct sys_reg_desc *target_specific,
2214 			    size_t nr_specific)
2215 {
2216 	struct sys_reg_params params;
2217 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
2218 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2219 	int Rt2 = (hsr >> 10) & 0x1f;
2220 
2221 	params.is_aarch32 = true;
2222 	params.is_32bit = false;
2223 	params.CRm = (hsr >> 1) & 0xf;
2224 	params.is_write = ((hsr & 1) == 0);
2225 
2226 	params.Op0 = 0;
2227 	params.Op1 = (hsr >> 16) & 0xf;
2228 	params.Op2 = 0;
2229 	params.CRn = 0;
2230 
2231 	/*
2232 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2233 	 * backends between AArch32 and AArch64, we get away with it.
2234 	 */
2235 	if (params.is_write) {
2236 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2237 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2238 	}
2239 
2240 	/*
2241 	 * Try to emulate the coprocessor access using the target
2242 	 * specific table first, and using the global table afterwards.
2243 	 * If either of the tables contains a handler, handle the
2244 	 * potential register operation in the case of a read and return
2245 	 * with success.
2246 	 */
2247 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
2248 	    !emulate_cp(vcpu, &params, global, nr_global)) {
2249 		/* Split up the value between registers for the read side */
2250 		if (!params.is_write) {
2251 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2252 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2253 		}
2254 
2255 		return 1;
2256 	}
2257 
2258 	unhandled_cp_access(vcpu, &params);
2259 	return 1;
2260 }
2261 
2262 /**
2263  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2264  * @vcpu: The VCPU pointer
2265  * @run:  The kvm_run struct
2266  */
2267 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2268 			    const struct sys_reg_desc *global,
2269 			    size_t nr_global,
2270 			    const struct sys_reg_desc *target_specific,
2271 			    size_t nr_specific)
2272 {
2273 	struct sys_reg_params params;
2274 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
2275 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2276 
2277 	params.is_aarch32 = true;
2278 	params.is_32bit = true;
2279 	params.CRm = (hsr >> 1) & 0xf;
2280 	params.regval = vcpu_get_reg(vcpu, Rt);
2281 	params.is_write = ((hsr & 1) == 0);
2282 	params.CRn = (hsr >> 10) & 0xf;
2283 	params.Op0 = 0;
2284 	params.Op1 = (hsr >> 14) & 0x7;
2285 	params.Op2 = (hsr >> 17) & 0x7;
2286 
2287 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
2288 	    !emulate_cp(vcpu, &params, global, nr_global)) {
2289 		if (!params.is_write)
2290 			vcpu_set_reg(vcpu, Rt, params.regval);
2291 		return 1;
2292 	}
2293 
2294 	unhandled_cp_access(vcpu, &params);
2295 	return 1;
2296 }
2297 
2298 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
2299 {
2300 	const struct sys_reg_desc *target_specific;
2301 	size_t num;
2302 
2303 	target_specific = get_target_table(vcpu->arch.target, false, &num);
2304 	return kvm_handle_cp_64(vcpu,
2305 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
2306 				target_specific, num);
2307 }
2308 
2309 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
2310 {
2311 	const struct sys_reg_desc *target_specific;
2312 	size_t num;
2313 
2314 	target_specific = get_target_table(vcpu->arch.target, false, &num);
2315 	return kvm_handle_cp_32(vcpu,
2316 				cp15_regs, ARRAY_SIZE(cp15_regs),
2317 				target_specific, num);
2318 }
2319 
2320 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
2321 {
2322 	return kvm_handle_cp_64(vcpu,
2323 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
2324 				NULL, 0);
2325 }
2326 
2327 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
2328 {
2329 	return kvm_handle_cp_32(vcpu,
2330 				cp14_regs, ARRAY_SIZE(cp14_regs),
2331 				NULL, 0);
2332 }
2333 
2334 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2335 {
2336 	// See ARM DDI 0487E.a, section D12.3.2
2337 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2338 }
2339 
2340 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2341 			   struct sys_reg_params *params)
2342 {
2343 	size_t num;
2344 	const struct sys_reg_desc *table, *r;
2345 
2346 	table = get_target_table(vcpu->arch.target, true, &num);
2347 
2348 	/* Search target-specific then generic table. */
2349 	r = find_reg(params, table, num);
2350 	if (!r)
2351 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2352 
2353 	if (likely(r)) {
2354 		perform_access(vcpu, params, r);
2355 	} else if (is_imp_def_sys_reg(params)) {
2356 		kvm_inject_undefined(vcpu);
2357 	} else {
2358 		print_sys_reg_msg(params,
2359 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2360 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2361 		kvm_inject_undefined(vcpu);
2362 	}
2363 	return 1;
2364 }
2365 
2366 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
2367 				const struct sys_reg_desc *table, size_t num,
2368 				unsigned long *bmap)
2369 {
2370 	unsigned long i;
2371 
2372 	for (i = 0; i < num; i++)
2373 		if (table[i].reset) {
2374 			int reg = table[i].reg;
2375 
2376 			table[i].reset(vcpu, &table[i]);
2377 			if (reg > 0 && reg < NR_SYS_REGS)
2378 				set_bit(reg, bmap);
2379 		}
2380 }
2381 
2382 /**
2383  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2384  * @vcpu: The VCPU pointer
2385  * @run:  The kvm_run struct
2386  */
2387 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
2388 {
2389 	struct sys_reg_params params;
2390 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
2391 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2392 	int ret;
2393 
2394 	trace_kvm_handle_sys_reg(esr);
2395 
2396 	params.is_aarch32 = false;
2397 	params.is_32bit = false;
2398 	params.Op0 = (esr >> 20) & 3;
2399 	params.Op1 = (esr >> 14) & 0x7;
2400 	params.CRn = (esr >> 10) & 0xf;
2401 	params.CRm = (esr >> 1) & 0xf;
2402 	params.Op2 = (esr >> 17) & 0x7;
2403 	params.regval = vcpu_get_reg(vcpu, Rt);
2404 	params.is_write = !(esr & 1);
2405 
2406 	ret = emulate_sys_reg(vcpu, &params);
2407 
2408 	if (!params.is_write)
2409 		vcpu_set_reg(vcpu, Rt, params.regval);
2410 	return ret;
2411 }
2412 
2413 /******************************************************************************
2414  * Userspace API
2415  *****************************************************************************/
2416 
2417 static bool index_to_params(u64 id, struct sys_reg_params *params)
2418 {
2419 	switch (id & KVM_REG_SIZE_MASK) {
2420 	case KVM_REG_SIZE_U64:
2421 		/* Any unused index bits means it's not valid. */
2422 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2423 			      | KVM_REG_ARM_COPROC_MASK
2424 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2425 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2426 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2427 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2428 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2429 			return false;
2430 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2431 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2432 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2433 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2434 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2435 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2436 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2437 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2438 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2439 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2440 		return true;
2441 	default:
2442 		return false;
2443 	}
2444 }
2445 
2446 const struct sys_reg_desc *find_reg_by_id(u64 id,
2447 					  struct sys_reg_params *params,
2448 					  const struct sys_reg_desc table[],
2449 					  unsigned int num)
2450 {
2451 	if (!index_to_params(id, params))
2452 		return NULL;
2453 
2454 	return find_reg(params, table, num);
2455 }
2456 
2457 /* Decode an index value, and find the sys_reg_desc entry. */
2458 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2459 						    u64 id)
2460 {
2461 	size_t num;
2462 	const struct sys_reg_desc *table, *r;
2463 	struct sys_reg_params params;
2464 
2465 	/* We only do sys_reg for now. */
2466 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2467 		return NULL;
2468 
2469 	if (!index_to_params(id, &params))
2470 		return NULL;
2471 
2472 	table = get_target_table(vcpu->arch.target, true, &num);
2473 	r = find_reg(&params, table, num);
2474 	if (!r)
2475 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2476 
2477 	/* Not saved in the sys_reg array and not otherwise accessible? */
2478 	if (r && !(r->reg || r->get_user))
2479 		r = NULL;
2480 
2481 	return r;
2482 }
2483 
2484 /*
2485  * These are the invariant sys_reg registers: we let the guest see the
2486  * host versions of these, so they're part of the guest state.
2487  *
2488  * A future CPU may provide a mechanism to present different values to
2489  * the guest, or a future kvm may trap them.
2490  */
2491 
2492 #define FUNCTION_INVARIANT(reg)						\
2493 	static void get_##reg(struct kvm_vcpu *v,			\
2494 			      const struct sys_reg_desc *r)		\
2495 	{								\
2496 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2497 	}
2498 
2499 FUNCTION_INVARIANT(midr_el1)
2500 FUNCTION_INVARIANT(revidr_el1)
2501 FUNCTION_INVARIANT(clidr_el1)
2502 FUNCTION_INVARIANT(aidr_el1)
2503 
2504 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2505 {
2506 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2507 }
2508 
2509 /* ->val is filled in by kvm_sys_reg_table_init() */
2510 static struct sys_reg_desc invariant_sys_regs[] = {
2511 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2512 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2513 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2514 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2515 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2516 };
2517 
2518 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2519 {
2520 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2521 		return -EFAULT;
2522 	return 0;
2523 }
2524 
2525 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2526 {
2527 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2528 		return -EFAULT;
2529 	return 0;
2530 }
2531 
2532 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2533 {
2534 	struct sys_reg_params params;
2535 	const struct sys_reg_desc *r;
2536 
2537 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2538 			   ARRAY_SIZE(invariant_sys_regs));
2539 	if (!r)
2540 		return -ENOENT;
2541 
2542 	return reg_to_user(uaddr, &r->val, id);
2543 }
2544 
2545 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2546 {
2547 	struct sys_reg_params params;
2548 	const struct sys_reg_desc *r;
2549 	int err;
2550 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2551 
2552 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2553 			   ARRAY_SIZE(invariant_sys_regs));
2554 	if (!r)
2555 		return -ENOENT;
2556 
2557 	err = reg_from_user(&val, uaddr, id);
2558 	if (err)
2559 		return err;
2560 
2561 	/* This is what we mean by invariant: you can't change it. */
2562 	if (r->val != val)
2563 		return -EINVAL;
2564 
2565 	return 0;
2566 }
2567 
2568 static bool is_valid_cache(u32 val)
2569 {
2570 	u32 level, ctype;
2571 
2572 	if (val >= CSSELR_MAX)
2573 		return false;
2574 
2575 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2576 	level = (val >> 1);
2577 	ctype = (cache_levels >> (level * 3)) & 7;
2578 
2579 	switch (ctype) {
2580 	case 0: /* No cache */
2581 		return false;
2582 	case 1: /* Instruction cache only */
2583 		return (val & 1);
2584 	case 2: /* Data cache only */
2585 	case 4: /* Unified cache */
2586 		return !(val & 1);
2587 	case 3: /* Separate instruction and data caches */
2588 		return true;
2589 	default: /* Reserved: we can't know instruction or data. */
2590 		return false;
2591 	}
2592 }
2593 
2594 static int demux_c15_get(u64 id, void __user *uaddr)
2595 {
2596 	u32 val;
2597 	u32 __user *uval = uaddr;
2598 
2599 	/* Fail if we have unknown bits set. */
2600 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2601 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2602 		return -ENOENT;
2603 
2604 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2605 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2606 		if (KVM_REG_SIZE(id) != 4)
2607 			return -ENOENT;
2608 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2609 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2610 		if (!is_valid_cache(val))
2611 			return -ENOENT;
2612 
2613 		return put_user(get_ccsidr(val), uval);
2614 	default:
2615 		return -ENOENT;
2616 	}
2617 }
2618 
2619 static int demux_c15_set(u64 id, void __user *uaddr)
2620 {
2621 	u32 val, newval;
2622 	u32 __user *uval = uaddr;
2623 
2624 	/* Fail if we have unknown bits set. */
2625 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2626 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2627 		return -ENOENT;
2628 
2629 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2630 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2631 		if (KVM_REG_SIZE(id) != 4)
2632 			return -ENOENT;
2633 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2634 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2635 		if (!is_valid_cache(val))
2636 			return -ENOENT;
2637 
2638 		if (get_user(newval, uval))
2639 			return -EFAULT;
2640 
2641 		/* This is also invariant: you can't change it. */
2642 		if (newval != get_ccsidr(val))
2643 			return -EINVAL;
2644 		return 0;
2645 	default:
2646 		return -ENOENT;
2647 	}
2648 }
2649 
2650 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2651 {
2652 	const struct sys_reg_desc *r;
2653 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2654 
2655 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2656 		return demux_c15_get(reg->id, uaddr);
2657 
2658 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2659 		return -ENOENT;
2660 
2661 	r = index_to_sys_reg_desc(vcpu, reg->id);
2662 	if (!r)
2663 		return get_invariant_sys_reg(reg->id, uaddr);
2664 
2665 	/* Check for regs disabled by runtime config */
2666 	if (sysreg_hidden_from_user(vcpu, r))
2667 		return -ENOENT;
2668 
2669 	if (r->get_user)
2670 		return (r->get_user)(vcpu, r, reg, uaddr);
2671 
2672 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2673 }
2674 
2675 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2676 {
2677 	const struct sys_reg_desc *r;
2678 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2679 
2680 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2681 		return demux_c15_set(reg->id, uaddr);
2682 
2683 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2684 		return -ENOENT;
2685 
2686 	r = index_to_sys_reg_desc(vcpu, reg->id);
2687 	if (!r)
2688 		return set_invariant_sys_reg(reg->id, uaddr);
2689 
2690 	/* Check for regs disabled by runtime config */
2691 	if (sysreg_hidden_from_user(vcpu, r))
2692 		return -ENOENT;
2693 
2694 	if (r->set_user)
2695 		return (r->set_user)(vcpu, r, reg, uaddr);
2696 
2697 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2698 }
2699 
2700 static unsigned int num_demux_regs(void)
2701 {
2702 	unsigned int i, count = 0;
2703 
2704 	for (i = 0; i < CSSELR_MAX; i++)
2705 		if (is_valid_cache(i))
2706 			count++;
2707 
2708 	return count;
2709 }
2710 
2711 static int write_demux_regids(u64 __user *uindices)
2712 {
2713 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2714 	unsigned int i;
2715 
2716 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2717 	for (i = 0; i < CSSELR_MAX; i++) {
2718 		if (!is_valid_cache(i))
2719 			continue;
2720 		if (put_user(val | i, uindices))
2721 			return -EFAULT;
2722 		uindices++;
2723 	}
2724 	return 0;
2725 }
2726 
2727 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2728 {
2729 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2730 		KVM_REG_ARM64_SYSREG |
2731 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2732 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2733 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2734 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2735 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2736 }
2737 
2738 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2739 {
2740 	if (!*uind)
2741 		return true;
2742 
2743 	if (put_user(sys_reg_to_index(reg), *uind))
2744 		return false;
2745 
2746 	(*uind)++;
2747 	return true;
2748 }
2749 
2750 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2751 			    const struct sys_reg_desc *rd,
2752 			    u64 __user **uind,
2753 			    unsigned int *total)
2754 {
2755 	/*
2756 	 * Ignore registers we trap but don't save,
2757 	 * and for which no custom user accessor is provided.
2758 	 */
2759 	if (!(rd->reg || rd->get_user))
2760 		return 0;
2761 
2762 	if (sysreg_hidden_from_user(vcpu, rd))
2763 		return 0;
2764 
2765 	if (!copy_reg_to_user(rd, uind))
2766 		return -EFAULT;
2767 
2768 	(*total)++;
2769 	return 0;
2770 }
2771 
2772 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2773 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2774 {
2775 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
2776 	unsigned int total = 0;
2777 	size_t num;
2778 	int err;
2779 
2780 	/* We check for duplicates here, to allow arch-specific overrides. */
2781 	i1 = get_target_table(vcpu->arch.target, true, &num);
2782 	end1 = i1 + num;
2783 	i2 = sys_reg_descs;
2784 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2785 
2786 	BUG_ON(i1 == end1 || i2 == end2);
2787 
2788 	/* Walk carefully, as both tables may refer to the same register. */
2789 	while (i1 || i2) {
2790 		int cmp = cmp_sys_reg(i1, i2);
2791 		/* target-specific overrides generic entry. */
2792 		if (cmp <= 0)
2793 			err = walk_one_sys_reg(vcpu, i1, &uind, &total);
2794 		else
2795 			err = walk_one_sys_reg(vcpu, i2, &uind, &total);
2796 
2797 		if (err)
2798 			return err;
2799 
2800 		if (cmp <= 0 && ++i1 == end1)
2801 			i1 = NULL;
2802 		if (cmp >= 0 && ++i2 == end2)
2803 			i2 = NULL;
2804 	}
2805 	return total;
2806 }
2807 
2808 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2809 {
2810 	return ARRAY_SIZE(invariant_sys_regs)
2811 		+ num_demux_regs()
2812 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2813 }
2814 
2815 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2816 {
2817 	unsigned int i;
2818 	int err;
2819 
2820 	/* Then give them all the invariant registers' indices. */
2821 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2822 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2823 			return -EFAULT;
2824 		uindices++;
2825 	}
2826 
2827 	err = walk_sys_regs(vcpu, uindices);
2828 	if (err < 0)
2829 		return err;
2830 	uindices += err;
2831 
2832 	return write_demux_regids(uindices);
2833 }
2834 
2835 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
2836 {
2837 	unsigned int i;
2838 
2839 	for (i = 1; i < n; i++) {
2840 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2841 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2842 			return 1;
2843 		}
2844 	}
2845 
2846 	return 0;
2847 }
2848 
2849 void kvm_sys_reg_table_init(void)
2850 {
2851 	unsigned int i;
2852 	struct sys_reg_desc clidr;
2853 
2854 	/* Make sure tables are unique and in order. */
2855 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
2856 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
2857 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
2858 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
2859 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
2860 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
2861 
2862 	/* We abuse the reset function to overwrite the table itself. */
2863 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2864 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2865 
2866 	/*
2867 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2868 	 *
2869 	 *   If software reads the Cache Type fields from Ctype1
2870 	 *   upwards, once it has seen a value of 0b000, no caches
2871 	 *   exist at further-out levels of the hierarchy. So, for
2872 	 *   example, if Ctype3 is the first Cache Type field with a
2873 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2874 	 *   ignored.
2875 	 */
2876 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2877 	cache_levels = clidr.val;
2878 	for (i = 0; i < 7; i++)
2879 		if (((cache_levels >> (i*3)) & 7) == 0)
2880 			break;
2881 	/* Clear all higher bits. */
2882 	cache_levels &= (1 << (i*3))-1;
2883 }
2884 
2885 /**
2886  * kvm_reset_sys_regs - sets system registers to reset value
2887  * @vcpu: The VCPU pointer
2888  *
2889  * This function finds the right table above and sets the registers on the
2890  * virtual CPU struct to their architecturally defined reset values.
2891  */
2892 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2893 {
2894 	size_t num;
2895 	const struct sys_reg_desc *table;
2896 	DECLARE_BITMAP(bmap, NR_SYS_REGS) = { 0, };
2897 
2898 	/* Generic chip reset first (so target could override). */
2899 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs), bmap);
2900 
2901 	table = get_target_table(vcpu->arch.target, true, &num);
2902 	reset_sys_reg_descs(vcpu, table, num, bmap);
2903 
2904 	for (num = 1; num < NR_SYS_REGS; num++) {
2905 		if (WARN(!test_bit(num, bmap),
2906 			 "Didn't reset __vcpu_sys_reg(%zi)\n", num))
2907 			break;
2908 	}
2909 }
2910