xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 82df5b73)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/coproc.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
9  *          Christoffer Dall <c.dall@virtualopensystems.com>
10  */
11 
12 #include <linux/bsearch.h>
13 #include <linux/kvm_host.h>
14 #include <linux/mm.h>
15 #include <linux/printk.h>
16 #include <linux/uaccess.h>
17 
18 #include <asm/cacheflush.h>
19 #include <asm/cputype.h>
20 #include <asm/debug-monitors.h>
21 #include <asm/esr.h>
22 #include <asm/kvm_arm.h>
23 #include <asm/kvm_coproc.h>
24 #include <asm/kvm_emulate.h>
25 #include <asm/kvm_hyp.h>
26 #include <asm/kvm_mmu.h>
27 #include <asm/perf_event.h>
28 #include <asm/sysreg.h>
29 
30 #include <trace/events/kvm.h>
31 
32 #include "sys_regs.h"
33 
34 #include "trace.h"
35 
36 /*
37  * All of this file is extremely similar to the ARM coproc.c, but the
38  * types are different. My gut feeling is that it should be pretty
39  * easy to merge, but that would be an ABI breakage -- again. VFP
40  * would also need to be abstracted.
41  *
42  * For AArch32, we only take care of what is being trapped. Anything
43  * that has to do with init and userspace access has to go via the
44  * 64bit interface.
45  */
46 
47 static bool read_from_write_only(struct kvm_vcpu *vcpu,
48 				 struct sys_reg_params *params,
49 				 const struct sys_reg_desc *r)
50 {
51 	WARN_ONCE(1, "Unexpected sys_reg read to write-only register\n");
52 	print_sys_reg_instr(params);
53 	kvm_inject_undefined(vcpu);
54 	return false;
55 }
56 
57 static bool write_to_read_only(struct kvm_vcpu *vcpu,
58 			       struct sys_reg_params *params,
59 			       const struct sys_reg_desc *r)
60 {
61 	WARN_ONCE(1, "Unexpected sys_reg write to read-only register\n");
62 	print_sys_reg_instr(params);
63 	kvm_inject_undefined(vcpu);
64 	return false;
65 }
66 
67 static bool __vcpu_read_sys_reg_from_cpu(int reg, u64 *val)
68 {
69 	/*
70 	 * System registers listed in the switch are not saved on every
71 	 * exit from the guest but are only saved on vcpu_put.
72 	 *
73 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
74 	 * should never be listed below, because the guest cannot modify its
75 	 * own MPIDR_EL1 and MPIDR_EL1 is accessed for VCPU A from VCPU B's
76 	 * thread when emulating cross-VCPU communication.
77 	 */
78 	switch (reg) {
79 	case CSSELR_EL1:	*val = read_sysreg_s(SYS_CSSELR_EL1);	break;
80 	case SCTLR_EL1:		*val = read_sysreg_s(SYS_SCTLR_EL12);	break;
81 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
82 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
83 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
84 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
85 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
86 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
87 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
88 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
89 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
90 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
91 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
92 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
93 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
94 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
95 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
96 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
97 	case PAR_EL1:		*val = read_sysreg_s(SYS_PAR_EL1);	break;
98 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
99 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
100 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
101 	default:		return false;
102 	}
103 
104 	return true;
105 }
106 
107 static bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
108 {
109 	/*
110 	 * System registers listed in the switch are not restored on every
111 	 * entry to the guest but are only restored on vcpu_load.
112 	 *
113 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
114 	 * should never be listed below, because the MPIDR should only be set
115 	 * once, before running the VCPU, and never changed later.
116 	 */
117 	switch (reg) {
118 	case CSSELR_EL1:	write_sysreg_s(val, SYS_CSSELR_EL1);	break;
119 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
120 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
121 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
122 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
123 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
124 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
125 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
126 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
127 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
128 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
129 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
130 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
131 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
132 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
133 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
134 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
135 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
136 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
137 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
138 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
139 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
140 	default:		return false;
141 	}
142 
143 	return true;
144 }
145 
146 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
147 {
148 	u64 val = 0x8badf00d8badf00d;
149 
150 	if (vcpu->arch.sysregs_loaded_on_cpu &&
151 	    __vcpu_read_sys_reg_from_cpu(reg, &val))
152 		return val;
153 
154 	return __vcpu_sys_reg(vcpu, reg);
155 }
156 
157 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
158 {
159 	if (vcpu->arch.sysregs_loaded_on_cpu &&
160 	    __vcpu_write_sys_reg_to_cpu(val, reg))
161 		return;
162 
163 	 __vcpu_sys_reg(vcpu, reg) = val;
164 }
165 
166 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
167 static u32 cache_levels;
168 
169 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
170 #define CSSELR_MAX 12
171 
172 /* Which cache CCSIDR represents depends on CSSELR value. */
173 static u32 get_ccsidr(u32 csselr)
174 {
175 	u32 ccsidr;
176 
177 	/* Make sure noone else changes CSSELR during this! */
178 	local_irq_disable();
179 	write_sysreg(csselr, csselr_el1);
180 	isb();
181 	ccsidr = read_sysreg(ccsidr_el1);
182 	local_irq_enable();
183 
184 	return ccsidr;
185 }
186 
187 /*
188  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
189  */
190 static bool access_dcsw(struct kvm_vcpu *vcpu,
191 			struct sys_reg_params *p,
192 			const struct sys_reg_desc *r)
193 {
194 	if (!p->is_write)
195 		return read_from_write_only(vcpu, p, r);
196 
197 	/*
198 	 * Only track S/W ops if we don't have FWB. It still indicates
199 	 * that the guest is a bit broken (S/W operations should only
200 	 * be done by firmware, knowing that there is only a single
201 	 * CPU left in the system, and certainly not from non-secure
202 	 * software).
203 	 */
204 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
205 		kvm_set_way_flush(vcpu);
206 
207 	return true;
208 }
209 
210 /*
211  * Generic accessor for VM registers. Only called as long as HCR_TVM
212  * is set. If the guest enables the MMU, we stop trapping the VM
213  * sys_regs and leave it in complete control of the caches.
214  */
215 static bool access_vm_reg(struct kvm_vcpu *vcpu,
216 			  struct sys_reg_params *p,
217 			  const struct sys_reg_desc *r)
218 {
219 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
220 	u64 val;
221 	int reg = r->reg;
222 
223 	BUG_ON(!p->is_write);
224 
225 	/* See the 32bit mapping in kvm_host.h */
226 	if (p->is_aarch32)
227 		reg = r->reg / 2;
228 
229 	if (!p->is_aarch32 || !p->is_32bit) {
230 		val = p->regval;
231 	} else {
232 		val = vcpu_read_sys_reg(vcpu, reg);
233 		if (r->reg % 2)
234 			val = (p->regval << 32) | (u64)lower_32_bits(val);
235 		else
236 			val = ((u64)upper_32_bits(val) << 32) |
237 				lower_32_bits(p->regval);
238 	}
239 	vcpu_write_sys_reg(vcpu, val, reg);
240 
241 	kvm_toggle_cache(vcpu, was_enabled);
242 	return true;
243 }
244 
245 /*
246  * Trap handler for the GICv3 SGI generation system register.
247  * Forward the request to the VGIC emulation.
248  * The cp15_64 code makes sure this automatically works
249  * for both AArch64 and AArch32 accesses.
250  */
251 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
252 			   struct sys_reg_params *p,
253 			   const struct sys_reg_desc *r)
254 {
255 	bool g1;
256 
257 	if (!p->is_write)
258 		return read_from_write_only(vcpu, p, r);
259 
260 	/*
261 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
262 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
263 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
264 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
265 	 * group.
266 	 */
267 	if (p->is_aarch32) {
268 		switch (p->Op1) {
269 		default:		/* Keep GCC quiet */
270 		case 0:			/* ICC_SGI1R */
271 			g1 = true;
272 			break;
273 		case 1:			/* ICC_ASGI1R */
274 		case 2:			/* ICC_SGI0R */
275 			g1 = false;
276 			break;
277 		}
278 	} else {
279 		switch (p->Op2) {
280 		default:		/* Keep GCC quiet */
281 		case 5:			/* ICC_SGI1R_EL1 */
282 			g1 = true;
283 			break;
284 		case 6:			/* ICC_ASGI1R_EL1 */
285 		case 7:			/* ICC_SGI0R_EL1 */
286 			g1 = false;
287 			break;
288 		}
289 	}
290 
291 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
292 
293 	return true;
294 }
295 
296 static bool access_gic_sre(struct kvm_vcpu *vcpu,
297 			   struct sys_reg_params *p,
298 			   const struct sys_reg_desc *r)
299 {
300 	if (p->is_write)
301 		return ignore_write(vcpu, p);
302 
303 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
304 	return true;
305 }
306 
307 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
308 			struct sys_reg_params *p,
309 			const struct sys_reg_desc *r)
310 {
311 	if (p->is_write)
312 		return ignore_write(vcpu, p);
313 	else
314 		return read_zero(vcpu, p);
315 }
316 
317 /*
318  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
319  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
320  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
321  * treat it separately.
322  */
323 static bool trap_loregion(struct kvm_vcpu *vcpu,
324 			  struct sys_reg_params *p,
325 			  const struct sys_reg_desc *r)
326 {
327 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
328 	u32 sr = sys_reg((u32)r->Op0, (u32)r->Op1,
329 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
330 
331 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
332 		kvm_inject_undefined(vcpu);
333 		return false;
334 	}
335 
336 	if (p->is_write && sr == SYS_LORID_EL1)
337 		return write_to_read_only(vcpu, p, r);
338 
339 	return trap_raz_wi(vcpu, p, r);
340 }
341 
342 static bool trap_oslsr_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 = (1 << 3);
350 		return true;
351 	}
352 }
353 
354 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
355 				   struct sys_reg_params *p,
356 				   const struct sys_reg_desc *r)
357 {
358 	if (p->is_write) {
359 		return ignore_write(vcpu, p);
360 	} else {
361 		p->regval = read_sysreg(dbgauthstatus_el1);
362 		return true;
363 	}
364 }
365 
366 /*
367  * We want to avoid world-switching all the DBG registers all the
368  * time:
369  *
370  * - If we've touched any debug register, it is likely that we're
371  *   going to touch more of them. It then makes sense to disable the
372  *   traps and start doing the save/restore dance
373  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
374  *   then mandatory to save/restore the registers, as the guest
375  *   depends on them.
376  *
377  * For this, we use a DIRTY bit, indicating the guest has modified the
378  * debug registers, used as follow:
379  *
380  * On guest entry:
381  * - If the dirty bit is set (because we're coming back from trapping),
382  *   disable the traps, save host registers, restore guest registers.
383  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
384  *   set the dirty bit, disable the traps, save host registers,
385  *   restore guest registers.
386  * - Otherwise, enable the traps
387  *
388  * On guest exit:
389  * - If the dirty bit is set, save guest registers, restore host
390  *   registers and clear the dirty bit. This ensure that the host can
391  *   now use the debug registers.
392  */
393 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
394 			    struct sys_reg_params *p,
395 			    const struct sys_reg_desc *r)
396 {
397 	if (p->is_write) {
398 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
399 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
400 	} else {
401 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
402 	}
403 
404 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
405 
406 	return true;
407 }
408 
409 /*
410  * reg_to_dbg/dbg_to_reg
411  *
412  * A 32 bit write to a debug register leave top bits alone
413  * A 32 bit read from a debug register only returns the bottom bits
414  *
415  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
416  * hyp.S code switches between host and guest values in future.
417  */
418 static void reg_to_dbg(struct kvm_vcpu *vcpu,
419 		       struct sys_reg_params *p,
420 		       u64 *dbg_reg)
421 {
422 	u64 val = p->regval;
423 
424 	if (p->is_32bit) {
425 		val &= 0xffffffffUL;
426 		val |= ((*dbg_reg >> 32) << 32);
427 	}
428 
429 	*dbg_reg = val;
430 	vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
431 }
432 
433 static void dbg_to_reg(struct kvm_vcpu *vcpu,
434 		       struct sys_reg_params *p,
435 		       u64 *dbg_reg)
436 {
437 	p->regval = *dbg_reg;
438 	if (p->is_32bit)
439 		p->regval &= 0xffffffffUL;
440 }
441 
442 static bool trap_bvr(struct kvm_vcpu *vcpu,
443 		     struct sys_reg_params *p,
444 		     const struct sys_reg_desc *rd)
445 {
446 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
447 
448 	if (p->is_write)
449 		reg_to_dbg(vcpu, p, dbg_reg);
450 	else
451 		dbg_to_reg(vcpu, p, dbg_reg);
452 
453 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
454 
455 	return true;
456 }
457 
458 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
459 		const struct kvm_one_reg *reg, void __user *uaddr)
460 {
461 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
462 
463 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
464 		return -EFAULT;
465 	return 0;
466 }
467 
468 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
469 	const struct kvm_one_reg *reg, void __user *uaddr)
470 {
471 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
472 
473 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
474 		return -EFAULT;
475 	return 0;
476 }
477 
478 static void reset_bvr(struct kvm_vcpu *vcpu,
479 		      const struct sys_reg_desc *rd)
480 {
481 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
482 }
483 
484 static bool trap_bcr(struct kvm_vcpu *vcpu,
485 		     struct sys_reg_params *p,
486 		     const struct sys_reg_desc *rd)
487 {
488 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
489 
490 	if (p->is_write)
491 		reg_to_dbg(vcpu, p, dbg_reg);
492 	else
493 		dbg_to_reg(vcpu, p, dbg_reg);
494 
495 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
496 
497 	return true;
498 }
499 
500 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
501 		const struct kvm_one_reg *reg, void __user *uaddr)
502 {
503 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
504 
505 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
506 		return -EFAULT;
507 
508 	return 0;
509 }
510 
511 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
512 	const struct kvm_one_reg *reg, void __user *uaddr)
513 {
514 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
515 
516 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
517 		return -EFAULT;
518 	return 0;
519 }
520 
521 static void reset_bcr(struct kvm_vcpu *vcpu,
522 		      const struct sys_reg_desc *rd)
523 {
524 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
525 }
526 
527 static bool trap_wvr(struct kvm_vcpu *vcpu,
528 		     struct sys_reg_params *p,
529 		     const struct sys_reg_desc *rd)
530 {
531 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
532 
533 	if (p->is_write)
534 		reg_to_dbg(vcpu, p, dbg_reg);
535 	else
536 		dbg_to_reg(vcpu, p, dbg_reg);
537 
538 	trace_trap_reg(__func__, rd->reg, p->is_write,
539 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
540 
541 	return true;
542 }
543 
544 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
545 		const struct kvm_one_reg *reg, void __user *uaddr)
546 {
547 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
548 
549 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
550 		return -EFAULT;
551 	return 0;
552 }
553 
554 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
555 	const struct kvm_one_reg *reg, void __user *uaddr)
556 {
557 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
558 
559 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
560 		return -EFAULT;
561 	return 0;
562 }
563 
564 static void reset_wvr(struct kvm_vcpu *vcpu,
565 		      const struct sys_reg_desc *rd)
566 {
567 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
568 }
569 
570 static bool trap_wcr(struct kvm_vcpu *vcpu,
571 		     struct sys_reg_params *p,
572 		     const struct sys_reg_desc *rd)
573 {
574 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
575 
576 	if (p->is_write)
577 		reg_to_dbg(vcpu, p, dbg_reg);
578 	else
579 		dbg_to_reg(vcpu, p, dbg_reg);
580 
581 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
582 
583 	return true;
584 }
585 
586 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
587 		const struct kvm_one_reg *reg, void __user *uaddr)
588 {
589 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
590 
591 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
592 		return -EFAULT;
593 	return 0;
594 }
595 
596 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
597 	const struct kvm_one_reg *reg, void __user *uaddr)
598 {
599 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
600 
601 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
602 		return -EFAULT;
603 	return 0;
604 }
605 
606 static void reset_wcr(struct kvm_vcpu *vcpu,
607 		      const struct sys_reg_desc *rd)
608 {
609 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
610 }
611 
612 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
613 {
614 	u64 amair = read_sysreg(amair_el1);
615 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
616 }
617 
618 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
619 {
620 	u64 mpidr;
621 
622 	/*
623 	 * Map the vcpu_id into the first three affinity level fields of
624 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
625 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
626 	 * of the GICv3 to be able to address each CPU directly when
627 	 * sending IPIs.
628 	 */
629 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
630 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
631 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
632 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
633 }
634 
635 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
636 {
637 	u64 pmcr, val;
638 
639 	pmcr = read_sysreg(pmcr_el0);
640 	/*
641 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
642 	 * except PMCR.E resetting to zero.
643 	 */
644 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
645 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
646 	if (!system_supports_32bit_el0())
647 		val |= ARMV8_PMU_PMCR_LC;
648 	__vcpu_sys_reg(vcpu, r->reg) = val;
649 }
650 
651 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
652 {
653 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
654 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
655 
656 	if (!enabled)
657 		kvm_inject_undefined(vcpu);
658 
659 	return !enabled;
660 }
661 
662 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
663 {
664 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
665 }
666 
667 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
668 {
669 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
670 }
671 
672 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
673 {
674 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
675 }
676 
677 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
678 {
679 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
680 }
681 
682 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
683 			const struct sys_reg_desc *r)
684 {
685 	u64 val;
686 
687 	if (!kvm_arm_pmu_v3_ready(vcpu))
688 		return trap_raz_wi(vcpu, p, r);
689 
690 	if (pmu_access_el0_disabled(vcpu))
691 		return false;
692 
693 	if (p->is_write) {
694 		/* Only update writeable bits of PMCR */
695 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
696 		val &= ~ARMV8_PMU_PMCR_MASK;
697 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
698 		if (!system_supports_32bit_el0())
699 			val |= ARMV8_PMU_PMCR_LC;
700 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
701 		kvm_pmu_handle_pmcr(vcpu, val);
702 		kvm_vcpu_pmu_restore_guest(vcpu);
703 	} else {
704 		/* PMCR.P & PMCR.C are RAZ */
705 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
706 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
707 		p->regval = val;
708 	}
709 
710 	return true;
711 }
712 
713 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
714 			  const struct sys_reg_desc *r)
715 {
716 	if (!kvm_arm_pmu_v3_ready(vcpu))
717 		return trap_raz_wi(vcpu, p, r);
718 
719 	if (pmu_access_event_counter_el0_disabled(vcpu))
720 		return false;
721 
722 	if (p->is_write)
723 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
724 	else
725 		/* return PMSELR.SEL field */
726 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
727 			    & ARMV8_PMU_COUNTER_MASK;
728 
729 	return true;
730 }
731 
732 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
733 			  const struct sys_reg_desc *r)
734 {
735 	u64 pmceid;
736 
737 	if (!kvm_arm_pmu_v3_ready(vcpu))
738 		return trap_raz_wi(vcpu, p, r);
739 
740 	BUG_ON(p->is_write);
741 
742 	if (pmu_access_el0_disabled(vcpu))
743 		return false;
744 
745 	if (!(p->Op2 & 1))
746 		pmceid = read_sysreg(pmceid0_el0);
747 	else
748 		pmceid = read_sysreg(pmceid1_el0);
749 
750 	p->regval = pmceid;
751 
752 	return true;
753 }
754 
755 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
756 {
757 	u64 pmcr, val;
758 
759 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
760 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
761 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
762 		kvm_inject_undefined(vcpu);
763 		return false;
764 	}
765 
766 	return true;
767 }
768 
769 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
770 			      struct sys_reg_params *p,
771 			      const struct sys_reg_desc *r)
772 {
773 	u64 idx;
774 
775 	if (!kvm_arm_pmu_v3_ready(vcpu))
776 		return trap_raz_wi(vcpu, p, r);
777 
778 	if (r->CRn == 9 && r->CRm == 13) {
779 		if (r->Op2 == 2) {
780 			/* PMXEVCNTR_EL0 */
781 			if (pmu_access_event_counter_el0_disabled(vcpu))
782 				return false;
783 
784 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
785 			      & ARMV8_PMU_COUNTER_MASK;
786 		} else if (r->Op2 == 0) {
787 			/* PMCCNTR_EL0 */
788 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
789 				return false;
790 
791 			idx = ARMV8_PMU_CYCLE_IDX;
792 		} else {
793 			return false;
794 		}
795 	} else if (r->CRn == 0 && r->CRm == 9) {
796 		/* PMCCNTR */
797 		if (pmu_access_event_counter_el0_disabled(vcpu))
798 			return false;
799 
800 		idx = ARMV8_PMU_CYCLE_IDX;
801 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
802 		/* PMEVCNTRn_EL0 */
803 		if (pmu_access_event_counter_el0_disabled(vcpu))
804 			return false;
805 
806 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
807 	} else {
808 		return false;
809 	}
810 
811 	if (!pmu_counter_idx_valid(vcpu, idx))
812 		return false;
813 
814 	if (p->is_write) {
815 		if (pmu_access_el0_disabled(vcpu))
816 			return false;
817 
818 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
819 	} else {
820 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
821 	}
822 
823 	return true;
824 }
825 
826 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
827 			       const struct sys_reg_desc *r)
828 {
829 	u64 idx, reg;
830 
831 	if (!kvm_arm_pmu_v3_ready(vcpu))
832 		return trap_raz_wi(vcpu, p, r);
833 
834 	if (pmu_access_el0_disabled(vcpu))
835 		return false;
836 
837 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
838 		/* PMXEVTYPER_EL0 */
839 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
840 		reg = PMEVTYPER0_EL0 + idx;
841 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
842 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
843 		if (idx == ARMV8_PMU_CYCLE_IDX)
844 			reg = PMCCFILTR_EL0;
845 		else
846 			/* PMEVTYPERn_EL0 */
847 			reg = PMEVTYPER0_EL0 + idx;
848 	} else {
849 		BUG();
850 	}
851 
852 	if (!pmu_counter_idx_valid(vcpu, idx))
853 		return false;
854 
855 	if (p->is_write) {
856 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
857 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
858 		kvm_vcpu_pmu_restore_guest(vcpu);
859 	} else {
860 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
861 	}
862 
863 	return true;
864 }
865 
866 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
867 			   const struct sys_reg_desc *r)
868 {
869 	u64 val, mask;
870 
871 	if (!kvm_arm_pmu_v3_ready(vcpu))
872 		return trap_raz_wi(vcpu, p, r);
873 
874 	if (pmu_access_el0_disabled(vcpu))
875 		return false;
876 
877 	mask = kvm_pmu_valid_counter_mask(vcpu);
878 	if (p->is_write) {
879 		val = p->regval & mask;
880 		if (r->Op2 & 0x1) {
881 			/* accessing PMCNTENSET_EL0 */
882 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
883 			kvm_pmu_enable_counter_mask(vcpu, val);
884 			kvm_vcpu_pmu_restore_guest(vcpu);
885 		} else {
886 			/* accessing PMCNTENCLR_EL0 */
887 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
888 			kvm_pmu_disable_counter_mask(vcpu, val);
889 		}
890 	} else {
891 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
892 	}
893 
894 	return true;
895 }
896 
897 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
898 			   const struct sys_reg_desc *r)
899 {
900 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
901 
902 	if (!kvm_arm_pmu_v3_ready(vcpu))
903 		return trap_raz_wi(vcpu, p, r);
904 
905 	if (!vcpu_mode_priv(vcpu)) {
906 		kvm_inject_undefined(vcpu);
907 		return false;
908 	}
909 
910 	if (p->is_write) {
911 		u64 val = p->regval & mask;
912 
913 		if (r->Op2 & 0x1)
914 			/* accessing PMINTENSET_EL1 */
915 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
916 		else
917 			/* accessing PMINTENCLR_EL1 */
918 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
919 	} else {
920 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
921 	}
922 
923 	return true;
924 }
925 
926 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
927 			 const struct sys_reg_desc *r)
928 {
929 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
930 
931 	if (!kvm_arm_pmu_v3_ready(vcpu))
932 		return trap_raz_wi(vcpu, p, r);
933 
934 	if (pmu_access_el0_disabled(vcpu))
935 		return false;
936 
937 	if (p->is_write) {
938 		if (r->CRm & 0x2)
939 			/* accessing PMOVSSET_EL0 */
940 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
941 		else
942 			/* accessing PMOVSCLR_EL0 */
943 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
944 	} else {
945 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
946 	}
947 
948 	return true;
949 }
950 
951 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
952 			   const struct sys_reg_desc *r)
953 {
954 	u64 mask;
955 
956 	if (!kvm_arm_pmu_v3_ready(vcpu))
957 		return trap_raz_wi(vcpu, p, r);
958 
959 	if (!p->is_write)
960 		return read_from_write_only(vcpu, p, r);
961 
962 	if (pmu_write_swinc_el0_disabled(vcpu))
963 		return false;
964 
965 	mask = kvm_pmu_valid_counter_mask(vcpu);
966 	kvm_pmu_software_increment(vcpu, p->regval & mask);
967 	return true;
968 }
969 
970 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
971 			     const struct sys_reg_desc *r)
972 {
973 	if (!kvm_arm_pmu_v3_ready(vcpu))
974 		return trap_raz_wi(vcpu, p, r);
975 
976 	if (p->is_write) {
977 		if (!vcpu_mode_priv(vcpu)) {
978 			kvm_inject_undefined(vcpu);
979 			return false;
980 		}
981 
982 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
983 			       p->regval & ARMV8_PMU_USERENR_MASK;
984 	} else {
985 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
986 			    & ARMV8_PMU_USERENR_MASK;
987 	}
988 
989 	return true;
990 }
991 
992 #define reg_to_encoding(x)						\
993 	sys_reg((u32)(x)->Op0, (u32)(x)->Op1,				\
994 		(u32)(x)->CRn, (u32)(x)->CRm, (u32)(x)->Op2);
995 
996 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
997 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
998 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
999 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1000 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1001 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1002 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1003 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1004 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1005 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1006 
1007 /* Macro to expand the PMEVCNTRn_EL0 register */
1008 #define PMU_PMEVCNTR_EL0(n)						\
1009 	{ SYS_DESC(SYS_PMEVCNTRn_EL0(n)),					\
1010 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
1011 
1012 /* Macro to expand the PMEVTYPERn_EL0 register */
1013 #define PMU_PMEVTYPER_EL0(n)						\
1014 	{ SYS_DESC(SYS_PMEVTYPERn_EL0(n)),					\
1015 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
1016 
1017 static bool access_amu(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1018 			     const struct sys_reg_desc *r)
1019 {
1020 	kvm_inject_undefined(vcpu);
1021 
1022 	return false;
1023 }
1024 
1025 /* Macro to expand the AMU counter and type registers*/
1026 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), access_amu }
1027 #define AMU_AMEVTYPE0_EL0(n) { SYS_DESC(SYS_AMEVTYPE0_EL0(n)), access_amu }
1028 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), access_amu }
1029 #define AMU_AMEVTYPE1_EL0(n) { SYS_DESC(SYS_AMEVTYPE1_EL0(n)), access_amu }
1030 
1031 static bool trap_ptrauth(struct kvm_vcpu *vcpu,
1032 			 struct sys_reg_params *p,
1033 			 const struct sys_reg_desc *rd)
1034 {
1035 	/*
1036 	 * If we land here, that is because we didn't fixup the access on exit
1037 	 * by allowing the PtrAuth sysregs. The only way this happens is when
1038 	 * the guest does not have PtrAuth support enabled.
1039 	 */
1040 	kvm_inject_undefined(vcpu);
1041 
1042 	return false;
1043 }
1044 
1045 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1046 			const struct sys_reg_desc *rd)
1047 {
1048 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1049 }
1050 
1051 #define __PTRAUTH_KEY(k)						\
1052 	{ SYS_DESC(SYS_## k), trap_ptrauth, reset_unknown, k,		\
1053 	.visibility = ptrauth_visibility}
1054 
1055 #define PTRAUTH_KEY(k)							\
1056 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1057 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1058 
1059 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1060 			      struct sys_reg_params *p,
1061 			      const struct sys_reg_desc *r)
1062 {
1063 	enum kvm_arch_timers tmr;
1064 	enum kvm_arch_timer_regs treg;
1065 	u64 reg = reg_to_encoding(r);
1066 
1067 	switch (reg) {
1068 	case SYS_CNTP_TVAL_EL0:
1069 	case SYS_AARCH32_CNTP_TVAL:
1070 		tmr = TIMER_PTIMER;
1071 		treg = TIMER_REG_TVAL;
1072 		break;
1073 	case SYS_CNTP_CTL_EL0:
1074 	case SYS_AARCH32_CNTP_CTL:
1075 		tmr = TIMER_PTIMER;
1076 		treg = TIMER_REG_CTL;
1077 		break;
1078 	case SYS_CNTP_CVAL_EL0:
1079 	case SYS_AARCH32_CNTP_CVAL:
1080 		tmr = TIMER_PTIMER;
1081 		treg = TIMER_REG_CVAL;
1082 		break;
1083 	default:
1084 		BUG();
1085 	}
1086 
1087 	if (p->is_write)
1088 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1089 	else
1090 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1091 
1092 	return true;
1093 }
1094 
1095 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1096 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1097 		struct sys_reg_desc const *r, bool raz)
1098 {
1099 	u32 id = sys_reg((u32)r->Op0, (u32)r->Op1,
1100 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
1101 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1102 
1103 	if (id == SYS_ID_AA64PFR0_EL1) {
1104 		if (!vcpu_has_sve(vcpu))
1105 			val &= ~(0xfUL << ID_AA64PFR0_SVE_SHIFT);
1106 		val &= ~(0xfUL << ID_AA64PFR0_AMU_SHIFT);
1107 	} else if (id == SYS_ID_AA64ISAR1_EL1 && !vcpu_has_ptrauth(vcpu)) {
1108 		val &= ~((0xfUL << ID_AA64ISAR1_APA_SHIFT) |
1109 			 (0xfUL << ID_AA64ISAR1_API_SHIFT) |
1110 			 (0xfUL << ID_AA64ISAR1_GPA_SHIFT) |
1111 			 (0xfUL << ID_AA64ISAR1_GPI_SHIFT));
1112 	} else if (id == SYS_ID_AA64DFR0_EL1) {
1113 		/* Limit guests to PMUv3 for ARMv8.1 */
1114 		val = cpuid_feature_cap_perfmon_field(val,
1115 						ID_AA64DFR0_PMUVER_SHIFT,
1116 						ID_AA64DFR0_PMUVER_8_1);
1117 	} else if (id == SYS_ID_DFR0_EL1) {
1118 		/* Limit guests to PMUv3 for ARMv8.1 */
1119 		val = cpuid_feature_cap_perfmon_field(val,
1120 						ID_DFR0_PERFMON_SHIFT,
1121 						ID_DFR0_PERFMON_8_1);
1122 	}
1123 
1124 	return val;
1125 }
1126 
1127 /* cpufeature ID register access trap handlers */
1128 
1129 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1130 			    struct sys_reg_params *p,
1131 			    const struct sys_reg_desc *r,
1132 			    bool raz)
1133 {
1134 	if (p->is_write)
1135 		return write_to_read_only(vcpu, p, r);
1136 
1137 	p->regval = read_id_reg(vcpu, r, raz);
1138 	return true;
1139 }
1140 
1141 static bool access_id_reg(struct kvm_vcpu *vcpu,
1142 			  struct sys_reg_params *p,
1143 			  const struct sys_reg_desc *r)
1144 {
1145 	return __access_id_reg(vcpu, p, r, false);
1146 }
1147 
1148 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1149 			      struct sys_reg_params *p,
1150 			      const struct sys_reg_desc *r)
1151 {
1152 	return __access_id_reg(vcpu, p, r, true);
1153 }
1154 
1155 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1156 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1157 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1158 
1159 /* Visibility overrides for SVE-specific control registers */
1160 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1161 				   const struct sys_reg_desc *rd)
1162 {
1163 	if (vcpu_has_sve(vcpu))
1164 		return 0;
1165 
1166 	return REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1167 }
1168 
1169 /* Visibility overrides for SVE-specific ID registers */
1170 static unsigned int sve_id_visibility(const struct kvm_vcpu *vcpu,
1171 				      const struct sys_reg_desc *rd)
1172 {
1173 	if (vcpu_has_sve(vcpu))
1174 		return 0;
1175 
1176 	return REG_HIDDEN_USER;
1177 }
1178 
1179 /* Generate the emulated ID_AA64ZFR0_EL1 value exposed to the guest */
1180 static u64 guest_id_aa64zfr0_el1(const struct kvm_vcpu *vcpu)
1181 {
1182 	if (!vcpu_has_sve(vcpu))
1183 		return 0;
1184 
1185 	return read_sanitised_ftr_reg(SYS_ID_AA64ZFR0_EL1);
1186 }
1187 
1188 static bool access_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1189 				   struct sys_reg_params *p,
1190 				   const struct sys_reg_desc *rd)
1191 {
1192 	if (p->is_write)
1193 		return write_to_read_only(vcpu, p, rd);
1194 
1195 	p->regval = guest_id_aa64zfr0_el1(vcpu);
1196 	return true;
1197 }
1198 
1199 static int get_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1200 		const struct sys_reg_desc *rd,
1201 		const struct kvm_one_reg *reg, void __user *uaddr)
1202 {
1203 	u64 val;
1204 
1205 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1206 		return -ENOENT;
1207 
1208 	val = guest_id_aa64zfr0_el1(vcpu);
1209 	return reg_to_user(uaddr, &val, reg->id);
1210 }
1211 
1212 static int set_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1213 		const struct sys_reg_desc *rd,
1214 		const struct kvm_one_reg *reg, void __user *uaddr)
1215 {
1216 	const u64 id = sys_reg_to_index(rd);
1217 	int err;
1218 	u64 val;
1219 
1220 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1221 		return -ENOENT;
1222 
1223 	err = reg_from_user(&val, uaddr, id);
1224 	if (err)
1225 		return err;
1226 
1227 	/* This is what we mean by invariant: you can't change it. */
1228 	if (val != guest_id_aa64zfr0_el1(vcpu))
1229 		return -EINVAL;
1230 
1231 	return 0;
1232 }
1233 
1234 /*
1235  * cpufeature ID register user accessors
1236  *
1237  * For now, these registers are immutable for userspace, so no values
1238  * are stored, and for set_id_reg() we don't allow the effective value
1239  * to be changed.
1240  */
1241 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1242 			const struct sys_reg_desc *rd, void __user *uaddr,
1243 			bool raz)
1244 {
1245 	const u64 id = sys_reg_to_index(rd);
1246 	const u64 val = read_id_reg(vcpu, rd, raz);
1247 
1248 	return reg_to_user(uaddr, &val, id);
1249 }
1250 
1251 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1252 			const struct sys_reg_desc *rd, void __user *uaddr,
1253 			bool raz)
1254 {
1255 	const u64 id = sys_reg_to_index(rd);
1256 	int err;
1257 	u64 val;
1258 
1259 	err = reg_from_user(&val, uaddr, id);
1260 	if (err)
1261 		return err;
1262 
1263 	/* This is what we mean by invariant: you can't change it. */
1264 	if (val != read_id_reg(vcpu, rd, raz))
1265 		return -EINVAL;
1266 
1267 	return 0;
1268 }
1269 
1270 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1271 		      const struct kvm_one_reg *reg, void __user *uaddr)
1272 {
1273 	return __get_id_reg(vcpu, rd, uaddr, false);
1274 }
1275 
1276 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1277 		      const struct kvm_one_reg *reg, void __user *uaddr)
1278 {
1279 	return __set_id_reg(vcpu, rd, uaddr, false);
1280 }
1281 
1282 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1283 			  const struct kvm_one_reg *reg, void __user *uaddr)
1284 {
1285 	return __get_id_reg(vcpu, rd, uaddr, true);
1286 }
1287 
1288 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1289 			  const struct kvm_one_reg *reg, void __user *uaddr)
1290 {
1291 	return __set_id_reg(vcpu, rd, uaddr, true);
1292 }
1293 
1294 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1295 		       const struct sys_reg_desc *r)
1296 {
1297 	if (p->is_write)
1298 		return write_to_read_only(vcpu, p, r);
1299 
1300 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1301 	return true;
1302 }
1303 
1304 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1305 			 const struct sys_reg_desc *r)
1306 {
1307 	if (p->is_write)
1308 		return write_to_read_only(vcpu, p, r);
1309 
1310 	p->regval = read_sysreg(clidr_el1);
1311 	return true;
1312 }
1313 
1314 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1315 			  const struct sys_reg_desc *r)
1316 {
1317 	int reg = r->reg;
1318 
1319 	/* See the 32bit mapping in kvm_host.h */
1320 	if (p->is_aarch32)
1321 		reg = r->reg / 2;
1322 
1323 	if (p->is_write)
1324 		vcpu_write_sys_reg(vcpu, p->regval, reg);
1325 	else
1326 		p->regval = vcpu_read_sys_reg(vcpu, reg);
1327 	return true;
1328 }
1329 
1330 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1331 			  const struct sys_reg_desc *r)
1332 {
1333 	u32 csselr;
1334 
1335 	if (p->is_write)
1336 		return write_to_read_only(vcpu, p, r);
1337 
1338 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1339 	p->regval = get_ccsidr(csselr);
1340 
1341 	/*
1342 	 * Guests should not be doing cache operations by set/way at all, and
1343 	 * for this reason, we trap them and attempt to infer the intent, so
1344 	 * that we can flush the entire guest's address space at the appropriate
1345 	 * time.
1346 	 * To prevent this trapping from causing performance problems, let's
1347 	 * expose the geometry of all data and unified caches (which are
1348 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1349 	 * [If guests should attempt to infer aliasing properties from the
1350 	 * geometry (which is not permitted by the architecture), they would
1351 	 * only do so for virtually indexed caches.]
1352 	 */
1353 	if (!(csselr & 1)) // data or unified cache
1354 		p->regval &= ~GENMASK(27, 3);
1355 	return true;
1356 }
1357 
1358 /* sys_reg_desc initialiser for known cpufeature ID registers */
1359 #define ID_SANITISED(name) {			\
1360 	SYS_DESC(SYS_##name),			\
1361 	.access	= access_id_reg,		\
1362 	.get_user = get_id_reg,			\
1363 	.set_user = set_id_reg,			\
1364 }
1365 
1366 /*
1367  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1368  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1369  * (1 <= crm < 8, 0 <= Op2 < 8).
1370  */
1371 #define ID_UNALLOCATED(crm, op2) {			\
1372 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1373 	.access = access_raz_id_reg,			\
1374 	.get_user = get_raz_id_reg,			\
1375 	.set_user = set_raz_id_reg,			\
1376 }
1377 
1378 /*
1379  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1380  * For now, these are exposed just like unallocated ID regs: they appear
1381  * RAZ for the guest.
1382  */
1383 #define ID_HIDDEN(name) {			\
1384 	SYS_DESC(SYS_##name),			\
1385 	.access = access_raz_id_reg,		\
1386 	.get_user = get_raz_id_reg,		\
1387 	.set_user = set_raz_id_reg,		\
1388 }
1389 
1390 /*
1391  * Architected system registers.
1392  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1393  *
1394  * Debug handling: We do trap most, if not all debug related system
1395  * registers. The implementation is good enough to ensure that a guest
1396  * can use these with minimal performance degradation. The drawback is
1397  * that we don't implement any of the external debug, none of the
1398  * OSlock protocol. This should be revisited if we ever encounter a
1399  * more demanding guest...
1400  */
1401 static const struct sys_reg_desc sys_reg_descs[] = {
1402 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1403 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1404 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1405 
1406 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1407 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1408 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1409 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1410 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1411 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1412 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1413 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1414 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1415 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1416 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1417 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1418 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1419 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1420 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1421 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1422 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1423 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1424 
1425 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1426 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1427 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1428 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1429 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1430 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1431 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1432 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1433 
1434 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1435 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1436 	// DBGDTR[TR]X_EL0 share the same encoding
1437 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1438 
1439 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1440 
1441 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1442 
1443 	/*
1444 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1445 	 * entries in arm64_ftr_regs[].
1446 	 */
1447 
1448 	/* AArch64 mappings of the AArch32 ID registers */
1449 	/* CRm=1 */
1450 	ID_SANITISED(ID_PFR0_EL1),
1451 	ID_SANITISED(ID_PFR1_EL1),
1452 	ID_SANITISED(ID_DFR0_EL1),
1453 	ID_HIDDEN(ID_AFR0_EL1),
1454 	ID_SANITISED(ID_MMFR0_EL1),
1455 	ID_SANITISED(ID_MMFR1_EL1),
1456 	ID_SANITISED(ID_MMFR2_EL1),
1457 	ID_SANITISED(ID_MMFR3_EL1),
1458 
1459 	/* CRm=2 */
1460 	ID_SANITISED(ID_ISAR0_EL1),
1461 	ID_SANITISED(ID_ISAR1_EL1),
1462 	ID_SANITISED(ID_ISAR2_EL1),
1463 	ID_SANITISED(ID_ISAR3_EL1),
1464 	ID_SANITISED(ID_ISAR4_EL1),
1465 	ID_SANITISED(ID_ISAR5_EL1),
1466 	ID_SANITISED(ID_MMFR4_EL1),
1467 	ID_SANITISED(ID_ISAR6_EL1),
1468 
1469 	/* CRm=3 */
1470 	ID_SANITISED(MVFR0_EL1),
1471 	ID_SANITISED(MVFR1_EL1),
1472 	ID_SANITISED(MVFR2_EL1),
1473 	ID_UNALLOCATED(3,3),
1474 	ID_SANITISED(ID_PFR2_EL1),
1475 	ID_HIDDEN(ID_DFR1_EL1),
1476 	ID_SANITISED(ID_MMFR5_EL1),
1477 	ID_UNALLOCATED(3,7),
1478 
1479 	/* AArch64 ID registers */
1480 	/* CRm=4 */
1481 	ID_SANITISED(ID_AA64PFR0_EL1),
1482 	ID_SANITISED(ID_AA64PFR1_EL1),
1483 	ID_UNALLOCATED(4,2),
1484 	ID_UNALLOCATED(4,3),
1485 	{ 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 },
1486 	ID_UNALLOCATED(4,5),
1487 	ID_UNALLOCATED(4,6),
1488 	ID_UNALLOCATED(4,7),
1489 
1490 	/* CRm=5 */
1491 	ID_SANITISED(ID_AA64DFR0_EL1),
1492 	ID_SANITISED(ID_AA64DFR1_EL1),
1493 	ID_UNALLOCATED(5,2),
1494 	ID_UNALLOCATED(5,3),
1495 	ID_HIDDEN(ID_AA64AFR0_EL1),
1496 	ID_HIDDEN(ID_AA64AFR1_EL1),
1497 	ID_UNALLOCATED(5,6),
1498 	ID_UNALLOCATED(5,7),
1499 
1500 	/* CRm=6 */
1501 	ID_SANITISED(ID_AA64ISAR0_EL1),
1502 	ID_SANITISED(ID_AA64ISAR1_EL1),
1503 	ID_UNALLOCATED(6,2),
1504 	ID_UNALLOCATED(6,3),
1505 	ID_UNALLOCATED(6,4),
1506 	ID_UNALLOCATED(6,5),
1507 	ID_UNALLOCATED(6,6),
1508 	ID_UNALLOCATED(6,7),
1509 
1510 	/* CRm=7 */
1511 	ID_SANITISED(ID_AA64MMFR0_EL1),
1512 	ID_SANITISED(ID_AA64MMFR1_EL1),
1513 	ID_SANITISED(ID_AA64MMFR2_EL1),
1514 	ID_UNALLOCATED(7,3),
1515 	ID_UNALLOCATED(7,4),
1516 	ID_UNALLOCATED(7,5),
1517 	ID_UNALLOCATED(7,6),
1518 	ID_UNALLOCATED(7,7),
1519 
1520 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1521 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1522 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1523 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1524 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1525 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1526 
1527 	PTRAUTH_KEY(APIA),
1528 	PTRAUTH_KEY(APIB),
1529 	PTRAUTH_KEY(APDA),
1530 	PTRAUTH_KEY(APDB),
1531 	PTRAUTH_KEY(APGA),
1532 
1533 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1534 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1535 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1536 
1537 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1538 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1539 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1540 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1541 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1542 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1543 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1544 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1545 
1546 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1547 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1548 
1549 	{ SYS_DESC(SYS_PMINTENSET_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1550 	{ SYS_DESC(SYS_PMINTENCLR_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1551 
1552 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1553 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1554 
1555 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1556 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1557 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1558 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1559 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1560 
1561 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1562 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1563 
1564 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1565 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1566 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1567 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1568 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1569 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1570 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1571 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1572 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1573 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1574 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1575 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1576 
1577 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1578 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1579 
1580 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1581 
1582 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1583 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1584 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1585 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1586 
1587 	{ SYS_DESC(SYS_PMCR_EL0), access_pmcr, reset_pmcr, PMCR_EL0 },
1588 	{ SYS_DESC(SYS_PMCNTENSET_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1589 	{ SYS_DESC(SYS_PMCNTENCLR_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1590 	{ SYS_DESC(SYS_PMOVSCLR_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1591 	{ SYS_DESC(SYS_PMSWINC_EL0), access_pmswinc, reset_unknown, PMSWINC_EL0 },
1592 	{ SYS_DESC(SYS_PMSELR_EL0), access_pmselr, reset_unknown, PMSELR_EL0 },
1593 	{ SYS_DESC(SYS_PMCEID0_EL0), access_pmceid },
1594 	{ SYS_DESC(SYS_PMCEID1_EL0), access_pmceid },
1595 	{ SYS_DESC(SYS_PMCCNTR_EL0), access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1596 	{ SYS_DESC(SYS_PMXEVTYPER_EL0), access_pmu_evtyper },
1597 	{ SYS_DESC(SYS_PMXEVCNTR_EL0), access_pmu_evcntr },
1598 	/*
1599 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1600 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1601 	 */
1602 	{ SYS_DESC(SYS_PMUSERENR_EL0), access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1603 	{ SYS_DESC(SYS_PMOVSSET_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1604 
1605 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1606 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1607 
1608 	{ SYS_DESC(SYS_AMCR_EL0), access_amu },
1609 	{ SYS_DESC(SYS_AMCFGR_EL0), access_amu },
1610 	{ SYS_DESC(SYS_AMCGCR_EL0), access_amu },
1611 	{ SYS_DESC(SYS_AMUSERENR_EL0), access_amu },
1612 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), access_amu },
1613 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), access_amu },
1614 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), access_amu },
1615 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), access_amu },
1616 	AMU_AMEVCNTR0_EL0(0),
1617 	AMU_AMEVCNTR0_EL0(1),
1618 	AMU_AMEVCNTR0_EL0(2),
1619 	AMU_AMEVCNTR0_EL0(3),
1620 	AMU_AMEVCNTR0_EL0(4),
1621 	AMU_AMEVCNTR0_EL0(5),
1622 	AMU_AMEVCNTR0_EL0(6),
1623 	AMU_AMEVCNTR0_EL0(7),
1624 	AMU_AMEVCNTR0_EL0(8),
1625 	AMU_AMEVCNTR0_EL0(9),
1626 	AMU_AMEVCNTR0_EL0(10),
1627 	AMU_AMEVCNTR0_EL0(11),
1628 	AMU_AMEVCNTR0_EL0(12),
1629 	AMU_AMEVCNTR0_EL0(13),
1630 	AMU_AMEVCNTR0_EL0(14),
1631 	AMU_AMEVCNTR0_EL0(15),
1632 	AMU_AMEVTYPE0_EL0(0),
1633 	AMU_AMEVTYPE0_EL0(1),
1634 	AMU_AMEVTYPE0_EL0(2),
1635 	AMU_AMEVTYPE0_EL0(3),
1636 	AMU_AMEVTYPE0_EL0(4),
1637 	AMU_AMEVTYPE0_EL0(5),
1638 	AMU_AMEVTYPE0_EL0(6),
1639 	AMU_AMEVTYPE0_EL0(7),
1640 	AMU_AMEVTYPE0_EL0(8),
1641 	AMU_AMEVTYPE0_EL0(9),
1642 	AMU_AMEVTYPE0_EL0(10),
1643 	AMU_AMEVTYPE0_EL0(11),
1644 	AMU_AMEVTYPE0_EL0(12),
1645 	AMU_AMEVTYPE0_EL0(13),
1646 	AMU_AMEVTYPE0_EL0(14),
1647 	AMU_AMEVTYPE0_EL0(15),
1648 	AMU_AMEVCNTR1_EL0(0),
1649 	AMU_AMEVCNTR1_EL0(1),
1650 	AMU_AMEVCNTR1_EL0(2),
1651 	AMU_AMEVCNTR1_EL0(3),
1652 	AMU_AMEVCNTR1_EL0(4),
1653 	AMU_AMEVCNTR1_EL0(5),
1654 	AMU_AMEVCNTR1_EL0(6),
1655 	AMU_AMEVCNTR1_EL0(7),
1656 	AMU_AMEVCNTR1_EL0(8),
1657 	AMU_AMEVCNTR1_EL0(9),
1658 	AMU_AMEVCNTR1_EL0(10),
1659 	AMU_AMEVCNTR1_EL0(11),
1660 	AMU_AMEVCNTR1_EL0(12),
1661 	AMU_AMEVCNTR1_EL0(13),
1662 	AMU_AMEVCNTR1_EL0(14),
1663 	AMU_AMEVCNTR1_EL0(15),
1664 	AMU_AMEVTYPE1_EL0(0),
1665 	AMU_AMEVTYPE1_EL0(1),
1666 	AMU_AMEVTYPE1_EL0(2),
1667 	AMU_AMEVTYPE1_EL0(3),
1668 	AMU_AMEVTYPE1_EL0(4),
1669 	AMU_AMEVTYPE1_EL0(5),
1670 	AMU_AMEVTYPE1_EL0(6),
1671 	AMU_AMEVTYPE1_EL0(7),
1672 	AMU_AMEVTYPE1_EL0(8),
1673 	AMU_AMEVTYPE1_EL0(9),
1674 	AMU_AMEVTYPE1_EL0(10),
1675 	AMU_AMEVTYPE1_EL0(11),
1676 	AMU_AMEVTYPE1_EL0(12),
1677 	AMU_AMEVTYPE1_EL0(13),
1678 	AMU_AMEVTYPE1_EL0(14),
1679 	AMU_AMEVTYPE1_EL0(15),
1680 
1681 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1682 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1683 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1684 
1685 	/* PMEVCNTRn_EL0 */
1686 	PMU_PMEVCNTR_EL0(0),
1687 	PMU_PMEVCNTR_EL0(1),
1688 	PMU_PMEVCNTR_EL0(2),
1689 	PMU_PMEVCNTR_EL0(3),
1690 	PMU_PMEVCNTR_EL0(4),
1691 	PMU_PMEVCNTR_EL0(5),
1692 	PMU_PMEVCNTR_EL0(6),
1693 	PMU_PMEVCNTR_EL0(7),
1694 	PMU_PMEVCNTR_EL0(8),
1695 	PMU_PMEVCNTR_EL0(9),
1696 	PMU_PMEVCNTR_EL0(10),
1697 	PMU_PMEVCNTR_EL0(11),
1698 	PMU_PMEVCNTR_EL0(12),
1699 	PMU_PMEVCNTR_EL0(13),
1700 	PMU_PMEVCNTR_EL0(14),
1701 	PMU_PMEVCNTR_EL0(15),
1702 	PMU_PMEVCNTR_EL0(16),
1703 	PMU_PMEVCNTR_EL0(17),
1704 	PMU_PMEVCNTR_EL0(18),
1705 	PMU_PMEVCNTR_EL0(19),
1706 	PMU_PMEVCNTR_EL0(20),
1707 	PMU_PMEVCNTR_EL0(21),
1708 	PMU_PMEVCNTR_EL0(22),
1709 	PMU_PMEVCNTR_EL0(23),
1710 	PMU_PMEVCNTR_EL0(24),
1711 	PMU_PMEVCNTR_EL0(25),
1712 	PMU_PMEVCNTR_EL0(26),
1713 	PMU_PMEVCNTR_EL0(27),
1714 	PMU_PMEVCNTR_EL0(28),
1715 	PMU_PMEVCNTR_EL0(29),
1716 	PMU_PMEVCNTR_EL0(30),
1717 	/* PMEVTYPERn_EL0 */
1718 	PMU_PMEVTYPER_EL0(0),
1719 	PMU_PMEVTYPER_EL0(1),
1720 	PMU_PMEVTYPER_EL0(2),
1721 	PMU_PMEVTYPER_EL0(3),
1722 	PMU_PMEVTYPER_EL0(4),
1723 	PMU_PMEVTYPER_EL0(5),
1724 	PMU_PMEVTYPER_EL0(6),
1725 	PMU_PMEVTYPER_EL0(7),
1726 	PMU_PMEVTYPER_EL0(8),
1727 	PMU_PMEVTYPER_EL0(9),
1728 	PMU_PMEVTYPER_EL0(10),
1729 	PMU_PMEVTYPER_EL0(11),
1730 	PMU_PMEVTYPER_EL0(12),
1731 	PMU_PMEVTYPER_EL0(13),
1732 	PMU_PMEVTYPER_EL0(14),
1733 	PMU_PMEVTYPER_EL0(15),
1734 	PMU_PMEVTYPER_EL0(16),
1735 	PMU_PMEVTYPER_EL0(17),
1736 	PMU_PMEVTYPER_EL0(18),
1737 	PMU_PMEVTYPER_EL0(19),
1738 	PMU_PMEVTYPER_EL0(20),
1739 	PMU_PMEVTYPER_EL0(21),
1740 	PMU_PMEVTYPER_EL0(22),
1741 	PMU_PMEVTYPER_EL0(23),
1742 	PMU_PMEVTYPER_EL0(24),
1743 	PMU_PMEVTYPER_EL0(25),
1744 	PMU_PMEVTYPER_EL0(26),
1745 	PMU_PMEVTYPER_EL0(27),
1746 	PMU_PMEVTYPER_EL0(28),
1747 	PMU_PMEVTYPER_EL0(29),
1748 	PMU_PMEVTYPER_EL0(30),
1749 	/*
1750 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1751 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1752 	 */
1753 	{ SYS_DESC(SYS_PMCCFILTR_EL0), access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1754 
1755 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1756 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1757 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1758 };
1759 
1760 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1761 			struct sys_reg_params *p,
1762 			const struct sys_reg_desc *r)
1763 {
1764 	if (p->is_write) {
1765 		return ignore_write(vcpu, p);
1766 	} else {
1767 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1768 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1769 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1770 
1771 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1772 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1773 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1774 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1775 		return true;
1776 	}
1777 }
1778 
1779 static bool trap_debug32(struct kvm_vcpu *vcpu,
1780 			 struct sys_reg_params *p,
1781 			 const struct sys_reg_desc *r)
1782 {
1783 	if (p->is_write) {
1784 		vcpu_cp14(vcpu, r->reg) = p->regval;
1785 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1786 	} else {
1787 		p->regval = vcpu_cp14(vcpu, r->reg);
1788 	}
1789 
1790 	return true;
1791 }
1792 
1793 /* AArch32 debug register mappings
1794  *
1795  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1796  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1797  *
1798  * All control registers and watchpoint value registers are mapped to
1799  * the lower 32 bits of their AArch64 equivalents. We share the trap
1800  * handlers with the above AArch64 code which checks what mode the
1801  * system is in.
1802  */
1803 
1804 static bool trap_xvr(struct kvm_vcpu *vcpu,
1805 		     struct sys_reg_params *p,
1806 		     const struct sys_reg_desc *rd)
1807 {
1808 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1809 
1810 	if (p->is_write) {
1811 		u64 val = *dbg_reg;
1812 
1813 		val &= 0xffffffffUL;
1814 		val |= p->regval << 32;
1815 		*dbg_reg = val;
1816 
1817 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1818 	} else {
1819 		p->regval = *dbg_reg >> 32;
1820 	}
1821 
1822 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1823 
1824 	return true;
1825 }
1826 
1827 #define DBG_BCR_BVR_WCR_WVR(n)						\
1828 	/* DBGBVRn */							\
1829 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1830 	/* DBGBCRn */							\
1831 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1832 	/* DBGWVRn */							\
1833 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1834 	/* DBGWCRn */							\
1835 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1836 
1837 #define DBGBXVR(n)							\
1838 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1839 
1840 /*
1841  * Trapped cp14 registers. We generally ignore most of the external
1842  * debug, on the principle that they don't really make sense to a
1843  * guest. Revisit this one day, would this principle change.
1844  */
1845 static const struct sys_reg_desc cp14_regs[] = {
1846 	/* DBGIDR */
1847 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1848 	/* DBGDTRRXext */
1849 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1850 
1851 	DBG_BCR_BVR_WCR_WVR(0),
1852 	/* DBGDSCRint */
1853 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1854 	DBG_BCR_BVR_WCR_WVR(1),
1855 	/* DBGDCCINT */
1856 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1857 	/* DBGDSCRext */
1858 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1859 	DBG_BCR_BVR_WCR_WVR(2),
1860 	/* DBGDTR[RT]Xint */
1861 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1862 	/* DBGDTR[RT]Xext */
1863 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1864 	DBG_BCR_BVR_WCR_WVR(3),
1865 	DBG_BCR_BVR_WCR_WVR(4),
1866 	DBG_BCR_BVR_WCR_WVR(5),
1867 	/* DBGWFAR */
1868 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1869 	/* DBGOSECCR */
1870 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1871 	DBG_BCR_BVR_WCR_WVR(6),
1872 	/* DBGVCR */
1873 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1874 	DBG_BCR_BVR_WCR_WVR(7),
1875 	DBG_BCR_BVR_WCR_WVR(8),
1876 	DBG_BCR_BVR_WCR_WVR(9),
1877 	DBG_BCR_BVR_WCR_WVR(10),
1878 	DBG_BCR_BVR_WCR_WVR(11),
1879 	DBG_BCR_BVR_WCR_WVR(12),
1880 	DBG_BCR_BVR_WCR_WVR(13),
1881 	DBG_BCR_BVR_WCR_WVR(14),
1882 	DBG_BCR_BVR_WCR_WVR(15),
1883 
1884 	/* DBGDRAR (32bit) */
1885 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1886 
1887 	DBGBXVR(0),
1888 	/* DBGOSLAR */
1889 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1890 	DBGBXVR(1),
1891 	/* DBGOSLSR */
1892 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1893 	DBGBXVR(2),
1894 	DBGBXVR(3),
1895 	/* DBGOSDLR */
1896 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1897 	DBGBXVR(4),
1898 	/* DBGPRCR */
1899 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1900 	DBGBXVR(5),
1901 	DBGBXVR(6),
1902 	DBGBXVR(7),
1903 	DBGBXVR(8),
1904 	DBGBXVR(9),
1905 	DBGBXVR(10),
1906 	DBGBXVR(11),
1907 	DBGBXVR(12),
1908 	DBGBXVR(13),
1909 	DBGBXVR(14),
1910 	DBGBXVR(15),
1911 
1912 	/* DBGDSAR (32bit) */
1913 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1914 
1915 	/* DBGDEVID2 */
1916 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1917 	/* DBGDEVID1 */
1918 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1919 	/* DBGDEVID */
1920 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1921 	/* DBGCLAIMSET */
1922 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1923 	/* DBGCLAIMCLR */
1924 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1925 	/* DBGAUTHSTATUS */
1926 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1927 };
1928 
1929 /* Trapped cp14 64bit registers */
1930 static const struct sys_reg_desc cp14_64_regs[] = {
1931 	/* DBGDRAR (64bit) */
1932 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1933 
1934 	/* DBGDSAR (64bit) */
1935 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1936 };
1937 
1938 /* Macro to expand the PMEVCNTRn register */
1939 #define PMU_PMEVCNTR(n)							\
1940 	/* PMEVCNTRn */							\
1941 	{ Op1(0), CRn(0b1110),						\
1942 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1943 	  access_pmu_evcntr }
1944 
1945 /* Macro to expand the PMEVTYPERn register */
1946 #define PMU_PMEVTYPER(n)						\
1947 	/* PMEVTYPERn */						\
1948 	{ Op1(0), CRn(0b1110),						\
1949 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1950 	  access_pmu_evtyper }
1951 
1952 /*
1953  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1954  * depending on the way they are accessed (as a 32bit or a 64bit
1955  * register).
1956  */
1957 static const struct sys_reg_desc cp15_regs[] = {
1958 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1959 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1960 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1961 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1962 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1963 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1964 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1965 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1966 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1967 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1968 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1969 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1970 
1971 	/*
1972 	 * DC{C,I,CI}SW operations:
1973 	 */
1974 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1975 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1976 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1977 
1978 	/* PMU */
1979 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1980 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1981 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1982 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1983 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1984 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1985 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1986 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1987 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1988 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1989 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1990 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1991 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1992 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1993 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1994 
1995 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
1996 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
1997 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
1998 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
1999 
2000 	/* ICC_SRE */
2001 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2002 
2003 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
2004 
2005 	/* Arch Tmers */
2006 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2007 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2008 
2009 	/* PMEVCNTRn */
2010 	PMU_PMEVCNTR(0),
2011 	PMU_PMEVCNTR(1),
2012 	PMU_PMEVCNTR(2),
2013 	PMU_PMEVCNTR(3),
2014 	PMU_PMEVCNTR(4),
2015 	PMU_PMEVCNTR(5),
2016 	PMU_PMEVCNTR(6),
2017 	PMU_PMEVCNTR(7),
2018 	PMU_PMEVCNTR(8),
2019 	PMU_PMEVCNTR(9),
2020 	PMU_PMEVCNTR(10),
2021 	PMU_PMEVCNTR(11),
2022 	PMU_PMEVCNTR(12),
2023 	PMU_PMEVCNTR(13),
2024 	PMU_PMEVCNTR(14),
2025 	PMU_PMEVCNTR(15),
2026 	PMU_PMEVCNTR(16),
2027 	PMU_PMEVCNTR(17),
2028 	PMU_PMEVCNTR(18),
2029 	PMU_PMEVCNTR(19),
2030 	PMU_PMEVCNTR(20),
2031 	PMU_PMEVCNTR(21),
2032 	PMU_PMEVCNTR(22),
2033 	PMU_PMEVCNTR(23),
2034 	PMU_PMEVCNTR(24),
2035 	PMU_PMEVCNTR(25),
2036 	PMU_PMEVCNTR(26),
2037 	PMU_PMEVCNTR(27),
2038 	PMU_PMEVCNTR(28),
2039 	PMU_PMEVCNTR(29),
2040 	PMU_PMEVCNTR(30),
2041 	/* PMEVTYPERn */
2042 	PMU_PMEVTYPER(0),
2043 	PMU_PMEVTYPER(1),
2044 	PMU_PMEVTYPER(2),
2045 	PMU_PMEVTYPER(3),
2046 	PMU_PMEVTYPER(4),
2047 	PMU_PMEVTYPER(5),
2048 	PMU_PMEVTYPER(6),
2049 	PMU_PMEVTYPER(7),
2050 	PMU_PMEVTYPER(8),
2051 	PMU_PMEVTYPER(9),
2052 	PMU_PMEVTYPER(10),
2053 	PMU_PMEVTYPER(11),
2054 	PMU_PMEVTYPER(12),
2055 	PMU_PMEVTYPER(13),
2056 	PMU_PMEVTYPER(14),
2057 	PMU_PMEVTYPER(15),
2058 	PMU_PMEVTYPER(16),
2059 	PMU_PMEVTYPER(17),
2060 	PMU_PMEVTYPER(18),
2061 	PMU_PMEVTYPER(19),
2062 	PMU_PMEVTYPER(20),
2063 	PMU_PMEVTYPER(21),
2064 	PMU_PMEVTYPER(22),
2065 	PMU_PMEVTYPER(23),
2066 	PMU_PMEVTYPER(24),
2067 	PMU_PMEVTYPER(25),
2068 	PMU_PMEVTYPER(26),
2069 	PMU_PMEVTYPER(27),
2070 	PMU_PMEVTYPER(28),
2071 	PMU_PMEVTYPER(29),
2072 	PMU_PMEVTYPER(30),
2073 	/* PMCCFILTR */
2074 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2075 
2076 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2077 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2078 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, c0_CSSELR },
2079 };
2080 
2081 static const struct sys_reg_desc cp15_64_regs[] = {
2082 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
2083 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2084 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2085 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
2086 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2087 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2088 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2089 };
2090 
2091 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2092 			      bool is_32)
2093 {
2094 	unsigned int i;
2095 
2096 	for (i = 0; i < n; i++) {
2097 		if (!is_32 && table[i].reg && !table[i].reset) {
2098 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2099 				table, i);
2100 			return 1;
2101 		}
2102 
2103 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2104 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2105 			return 1;
2106 		}
2107 	}
2108 
2109 	return 0;
2110 }
2111 
2112 /* Target specific emulation tables */
2113 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
2114 
2115 void kvm_register_target_sys_reg_table(unsigned int target,
2116 				       struct kvm_sys_reg_target_table *table)
2117 {
2118 	if (check_sysreg_table(table->table64.table, table->table64.num, false) ||
2119 	    check_sysreg_table(table->table32.table, table->table32.num, true))
2120 		return;
2121 
2122 	target_tables[target] = table;
2123 }
2124 
2125 /* Get specific register table for this target. */
2126 static const struct sys_reg_desc *get_target_table(unsigned target,
2127 						   bool mode_is_64,
2128 						   size_t *num)
2129 {
2130 	struct kvm_sys_reg_target_table *table;
2131 
2132 	table = target_tables[target];
2133 	if (mode_is_64) {
2134 		*num = table->table64.num;
2135 		return table->table64.table;
2136 	} else {
2137 		*num = table->table32.num;
2138 		return table->table32.table;
2139 	}
2140 }
2141 
2142 static int match_sys_reg(const void *key, const void *elt)
2143 {
2144 	const unsigned long pval = (unsigned long)key;
2145 	const struct sys_reg_desc *r = elt;
2146 
2147 	return pval - reg_to_encoding(r);
2148 }
2149 
2150 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
2151 					 const struct sys_reg_desc table[],
2152 					 unsigned int num)
2153 {
2154 	unsigned long pval = reg_to_encoding(params);
2155 
2156 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
2157 }
2158 
2159 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
2160 {
2161 	kvm_inject_undefined(vcpu);
2162 	return 1;
2163 }
2164 
2165 static void perform_access(struct kvm_vcpu *vcpu,
2166 			   struct sys_reg_params *params,
2167 			   const struct sys_reg_desc *r)
2168 {
2169 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2170 
2171 	/* Check for regs disabled by runtime config */
2172 	if (sysreg_hidden_from_guest(vcpu, r)) {
2173 		kvm_inject_undefined(vcpu);
2174 		return;
2175 	}
2176 
2177 	/*
2178 	 * Not having an accessor means that we have configured a trap
2179 	 * that we don't know how to handle. This certainly qualifies
2180 	 * as a gross bug that should be fixed right away.
2181 	 */
2182 	BUG_ON(!r->access);
2183 
2184 	/* Skip instruction if instructed so */
2185 	if (likely(r->access(vcpu, params, r)))
2186 		kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
2187 }
2188 
2189 /*
2190  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2191  *                call the corresponding trap handler.
2192  *
2193  * @params: pointer to the descriptor of the access
2194  * @table: array of trap descriptors
2195  * @num: size of the trap descriptor array
2196  *
2197  * Return 0 if the access has been handled, and -1 if not.
2198  */
2199 static int emulate_cp(struct kvm_vcpu *vcpu,
2200 		      struct sys_reg_params *params,
2201 		      const struct sys_reg_desc *table,
2202 		      size_t num)
2203 {
2204 	const struct sys_reg_desc *r;
2205 
2206 	if (!table)
2207 		return -1;	/* Not handled */
2208 
2209 	r = find_reg(params, table, num);
2210 
2211 	if (r) {
2212 		perform_access(vcpu, params, r);
2213 		return 0;
2214 	}
2215 
2216 	/* Not handled */
2217 	return -1;
2218 }
2219 
2220 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2221 				struct sys_reg_params *params)
2222 {
2223 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
2224 	int cp = -1;
2225 
2226 	switch(hsr_ec) {
2227 	case ESR_ELx_EC_CP15_32:
2228 	case ESR_ELx_EC_CP15_64:
2229 		cp = 15;
2230 		break;
2231 	case ESR_ELx_EC_CP14_MR:
2232 	case ESR_ELx_EC_CP14_64:
2233 		cp = 14;
2234 		break;
2235 	default:
2236 		WARN_ON(1);
2237 	}
2238 
2239 	print_sys_reg_msg(params,
2240 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2241 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2242 	kvm_inject_undefined(vcpu);
2243 }
2244 
2245 /**
2246  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2247  * @vcpu: The VCPU pointer
2248  * @run:  The kvm_run struct
2249  */
2250 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2251 			    const struct sys_reg_desc *global,
2252 			    size_t nr_global,
2253 			    const struct sys_reg_desc *target_specific,
2254 			    size_t nr_specific)
2255 {
2256 	struct sys_reg_params params;
2257 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
2258 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2259 	int Rt2 = (hsr >> 10) & 0x1f;
2260 
2261 	params.is_aarch32 = true;
2262 	params.is_32bit = false;
2263 	params.CRm = (hsr >> 1) & 0xf;
2264 	params.is_write = ((hsr & 1) == 0);
2265 
2266 	params.Op0 = 0;
2267 	params.Op1 = (hsr >> 16) & 0xf;
2268 	params.Op2 = 0;
2269 	params.CRn = 0;
2270 
2271 	/*
2272 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2273 	 * backends between AArch32 and AArch64, we get away with it.
2274 	 */
2275 	if (params.is_write) {
2276 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2277 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2278 	}
2279 
2280 	/*
2281 	 * Try to emulate the coprocessor access using the target
2282 	 * specific table first, and using the global table afterwards.
2283 	 * If either of the tables contains a handler, handle the
2284 	 * potential register operation in the case of a read and return
2285 	 * with success.
2286 	 */
2287 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
2288 	    !emulate_cp(vcpu, &params, global, nr_global)) {
2289 		/* Split up the value between registers for the read side */
2290 		if (!params.is_write) {
2291 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2292 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2293 		}
2294 
2295 		return 1;
2296 	}
2297 
2298 	unhandled_cp_access(vcpu, &params);
2299 	return 1;
2300 }
2301 
2302 /**
2303  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2304  * @vcpu: The VCPU pointer
2305  * @run:  The kvm_run struct
2306  */
2307 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2308 			    const struct sys_reg_desc *global,
2309 			    size_t nr_global,
2310 			    const struct sys_reg_desc *target_specific,
2311 			    size_t nr_specific)
2312 {
2313 	struct sys_reg_params params;
2314 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
2315 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2316 
2317 	params.is_aarch32 = true;
2318 	params.is_32bit = true;
2319 	params.CRm = (hsr >> 1) & 0xf;
2320 	params.regval = vcpu_get_reg(vcpu, Rt);
2321 	params.is_write = ((hsr & 1) == 0);
2322 	params.CRn = (hsr >> 10) & 0xf;
2323 	params.Op0 = 0;
2324 	params.Op1 = (hsr >> 14) & 0x7;
2325 	params.Op2 = (hsr >> 17) & 0x7;
2326 
2327 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
2328 	    !emulate_cp(vcpu, &params, global, nr_global)) {
2329 		if (!params.is_write)
2330 			vcpu_set_reg(vcpu, Rt, params.regval);
2331 		return 1;
2332 	}
2333 
2334 	unhandled_cp_access(vcpu, &params);
2335 	return 1;
2336 }
2337 
2338 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
2339 {
2340 	const struct sys_reg_desc *target_specific;
2341 	size_t num;
2342 
2343 	target_specific = get_target_table(vcpu->arch.target, false, &num);
2344 	return kvm_handle_cp_64(vcpu,
2345 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
2346 				target_specific, num);
2347 }
2348 
2349 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
2350 {
2351 	const struct sys_reg_desc *target_specific;
2352 	size_t num;
2353 
2354 	target_specific = get_target_table(vcpu->arch.target, false, &num);
2355 	return kvm_handle_cp_32(vcpu,
2356 				cp15_regs, ARRAY_SIZE(cp15_regs),
2357 				target_specific, num);
2358 }
2359 
2360 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
2361 {
2362 	return kvm_handle_cp_64(vcpu,
2363 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
2364 				NULL, 0);
2365 }
2366 
2367 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
2368 {
2369 	return kvm_handle_cp_32(vcpu,
2370 				cp14_regs, ARRAY_SIZE(cp14_regs),
2371 				NULL, 0);
2372 }
2373 
2374 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2375 {
2376 	// See ARM DDI 0487E.a, section D12.3.2
2377 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2378 }
2379 
2380 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2381 			   struct sys_reg_params *params)
2382 {
2383 	size_t num;
2384 	const struct sys_reg_desc *table, *r;
2385 
2386 	table = get_target_table(vcpu->arch.target, true, &num);
2387 
2388 	/* Search target-specific then generic table. */
2389 	r = find_reg(params, table, num);
2390 	if (!r)
2391 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2392 
2393 	if (likely(r)) {
2394 		perform_access(vcpu, params, r);
2395 	} else if (is_imp_def_sys_reg(params)) {
2396 		kvm_inject_undefined(vcpu);
2397 	} else {
2398 		print_sys_reg_msg(params,
2399 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2400 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2401 		kvm_inject_undefined(vcpu);
2402 	}
2403 	return 1;
2404 }
2405 
2406 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
2407 				const struct sys_reg_desc *table, size_t num)
2408 {
2409 	unsigned long i;
2410 
2411 	for (i = 0; i < num; i++)
2412 		if (table[i].reset)
2413 			table[i].reset(vcpu, &table[i]);
2414 }
2415 
2416 /**
2417  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2418  * @vcpu: The VCPU pointer
2419  * @run:  The kvm_run struct
2420  */
2421 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
2422 {
2423 	struct sys_reg_params params;
2424 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
2425 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2426 	int ret;
2427 
2428 	trace_kvm_handle_sys_reg(esr);
2429 
2430 	params.is_aarch32 = false;
2431 	params.is_32bit = false;
2432 	params.Op0 = (esr >> 20) & 3;
2433 	params.Op1 = (esr >> 14) & 0x7;
2434 	params.CRn = (esr >> 10) & 0xf;
2435 	params.CRm = (esr >> 1) & 0xf;
2436 	params.Op2 = (esr >> 17) & 0x7;
2437 	params.regval = vcpu_get_reg(vcpu, Rt);
2438 	params.is_write = !(esr & 1);
2439 
2440 	ret = emulate_sys_reg(vcpu, &params);
2441 
2442 	if (!params.is_write)
2443 		vcpu_set_reg(vcpu, Rt, params.regval);
2444 	return ret;
2445 }
2446 
2447 /******************************************************************************
2448  * Userspace API
2449  *****************************************************************************/
2450 
2451 static bool index_to_params(u64 id, struct sys_reg_params *params)
2452 {
2453 	switch (id & KVM_REG_SIZE_MASK) {
2454 	case KVM_REG_SIZE_U64:
2455 		/* Any unused index bits means it's not valid. */
2456 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2457 			      | KVM_REG_ARM_COPROC_MASK
2458 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2459 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2460 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2461 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2462 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2463 			return false;
2464 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2465 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2466 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2467 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2468 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2469 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2470 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2471 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2472 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2473 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2474 		return true;
2475 	default:
2476 		return false;
2477 	}
2478 }
2479 
2480 const struct sys_reg_desc *find_reg_by_id(u64 id,
2481 					  struct sys_reg_params *params,
2482 					  const struct sys_reg_desc table[],
2483 					  unsigned int num)
2484 {
2485 	if (!index_to_params(id, params))
2486 		return NULL;
2487 
2488 	return find_reg(params, table, num);
2489 }
2490 
2491 /* Decode an index value, and find the sys_reg_desc entry. */
2492 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2493 						    u64 id)
2494 {
2495 	size_t num;
2496 	const struct sys_reg_desc *table, *r;
2497 	struct sys_reg_params params;
2498 
2499 	/* We only do sys_reg for now. */
2500 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2501 		return NULL;
2502 
2503 	if (!index_to_params(id, &params))
2504 		return NULL;
2505 
2506 	table = get_target_table(vcpu->arch.target, true, &num);
2507 	r = find_reg(&params, table, num);
2508 	if (!r)
2509 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2510 
2511 	/* Not saved in the sys_reg array and not otherwise accessible? */
2512 	if (r && !(r->reg || r->get_user))
2513 		r = NULL;
2514 
2515 	return r;
2516 }
2517 
2518 /*
2519  * These are the invariant sys_reg registers: we let the guest see the
2520  * host versions of these, so they're part of the guest state.
2521  *
2522  * A future CPU may provide a mechanism to present different values to
2523  * the guest, or a future kvm may trap them.
2524  */
2525 
2526 #define FUNCTION_INVARIANT(reg)						\
2527 	static void get_##reg(struct kvm_vcpu *v,			\
2528 			      const struct sys_reg_desc *r)		\
2529 	{								\
2530 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2531 	}
2532 
2533 FUNCTION_INVARIANT(midr_el1)
2534 FUNCTION_INVARIANT(revidr_el1)
2535 FUNCTION_INVARIANT(clidr_el1)
2536 FUNCTION_INVARIANT(aidr_el1)
2537 
2538 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2539 {
2540 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2541 }
2542 
2543 /* ->val is filled in by kvm_sys_reg_table_init() */
2544 static struct sys_reg_desc invariant_sys_regs[] = {
2545 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2546 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2547 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2548 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2549 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2550 };
2551 
2552 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2553 {
2554 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2555 		return -EFAULT;
2556 	return 0;
2557 }
2558 
2559 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2560 {
2561 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2562 		return -EFAULT;
2563 	return 0;
2564 }
2565 
2566 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2567 {
2568 	struct sys_reg_params params;
2569 	const struct sys_reg_desc *r;
2570 
2571 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2572 			   ARRAY_SIZE(invariant_sys_regs));
2573 	if (!r)
2574 		return -ENOENT;
2575 
2576 	return reg_to_user(uaddr, &r->val, id);
2577 }
2578 
2579 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2580 {
2581 	struct sys_reg_params params;
2582 	const struct sys_reg_desc *r;
2583 	int err;
2584 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2585 
2586 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2587 			   ARRAY_SIZE(invariant_sys_regs));
2588 	if (!r)
2589 		return -ENOENT;
2590 
2591 	err = reg_from_user(&val, uaddr, id);
2592 	if (err)
2593 		return err;
2594 
2595 	/* This is what we mean by invariant: you can't change it. */
2596 	if (r->val != val)
2597 		return -EINVAL;
2598 
2599 	return 0;
2600 }
2601 
2602 static bool is_valid_cache(u32 val)
2603 {
2604 	u32 level, ctype;
2605 
2606 	if (val >= CSSELR_MAX)
2607 		return false;
2608 
2609 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2610 	level = (val >> 1);
2611 	ctype = (cache_levels >> (level * 3)) & 7;
2612 
2613 	switch (ctype) {
2614 	case 0: /* No cache */
2615 		return false;
2616 	case 1: /* Instruction cache only */
2617 		return (val & 1);
2618 	case 2: /* Data cache only */
2619 	case 4: /* Unified cache */
2620 		return !(val & 1);
2621 	case 3: /* Separate instruction and data caches */
2622 		return true;
2623 	default: /* Reserved: we can't know instruction or data. */
2624 		return false;
2625 	}
2626 }
2627 
2628 static int demux_c15_get(u64 id, void __user *uaddr)
2629 {
2630 	u32 val;
2631 	u32 __user *uval = uaddr;
2632 
2633 	/* Fail if we have unknown bits set. */
2634 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2635 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2636 		return -ENOENT;
2637 
2638 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2639 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2640 		if (KVM_REG_SIZE(id) != 4)
2641 			return -ENOENT;
2642 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2643 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2644 		if (!is_valid_cache(val))
2645 			return -ENOENT;
2646 
2647 		return put_user(get_ccsidr(val), uval);
2648 	default:
2649 		return -ENOENT;
2650 	}
2651 }
2652 
2653 static int demux_c15_set(u64 id, void __user *uaddr)
2654 {
2655 	u32 val, newval;
2656 	u32 __user *uval = uaddr;
2657 
2658 	/* Fail if we have unknown bits set. */
2659 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2660 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2661 		return -ENOENT;
2662 
2663 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2664 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2665 		if (KVM_REG_SIZE(id) != 4)
2666 			return -ENOENT;
2667 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2668 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2669 		if (!is_valid_cache(val))
2670 			return -ENOENT;
2671 
2672 		if (get_user(newval, uval))
2673 			return -EFAULT;
2674 
2675 		/* This is also invariant: you can't change it. */
2676 		if (newval != get_ccsidr(val))
2677 			return -EINVAL;
2678 		return 0;
2679 	default:
2680 		return -ENOENT;
2681 	}
2682 }
2683 
2684 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2685 {
2686 	const struct sys_reg_desc *r;
2687 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2688 
2689 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2690 		return demux_c15_get(reg->id, uaddr);
2691 
2692 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2693 		return -ENOENT;
2694 
2695 	r = index_to_sys_reg_desc(vcpu, reg->id);
2696 	if (!r)
2697 		return get_invariant_sys_reg(reg->id, uaddr);
2698 
2699 	/* Check for regs disabled by runtime config */
2700 	if (sysreg_hidden_from_user(vcpu, r))
2701 		return -ENOENT;
2702 
2703 	if (r->get_user)
2704 		return (r->get_user)(vcpu, r, reg, uaddr);
2705 
2706 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2707 }
2708 
2709 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2710 {
2711 	const struct sys_reg_desc *r;
2712 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2713 
2714 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2715 		return demux_c15_set(reg->id, uaddr);
2716 
2717 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2718 		return -ENOENT;
2719 
2720 	r = index_to_sys_reg_desc(vcpu, reg->id);
2721 	if (!r)
2722 		return set_invariant_sys_reg(reg->id, uaddr);
2723 
2724 	/* Check for regs disabled by runtime config */
2725 	if (sysreg_hidden_from_user(vcpu, r))
2726 		return -ENOENT;
2727 
2728 	if (r->set_user)
2729 		return (r->set_user)(vcpu, r, reg, uaddr);
2730 
2731 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2732 }
2733 
2734 static unsigned int num_demux_regs(void)
2735 {
2736 	unsigned int i, count = 0;
2737 
2738 	for (i = 0; i < CSSELR_MAX; i++)
2739 		if (is_valid_cache(i))
2740 			count++;
2741 
2742 	return count;
2743 }
2744 
2745 static int write_demux_regids(u64 __user *uindices)
2746 {
2747 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2748 	unsigned int i;
2749 
2750 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2751 	for (i = 0; i < CSSELR_MAX; i++) {
2752 		if (!is_valid_cache(i))
2753 			continue;
2754 		if (put_user(val | i, uindices))
2755 			return -EFAULT;
2756 		uindices++;
2757 	}
2758 	return 0;
2759 }
2760 
2761 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2762 {
2763 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2764 		KVM_REG_ARM64_SYSREG |
2765 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2766 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2767 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2768 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2769 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2770 }
2771 
2772 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2773 {
2774 	if (!*uind)
2775 		return true;
2776 
2777 	if (put_user(sys_reg_to_index(reg), *uind))
2778 		return false;
2779 
2780 	(*uind)++;
2781 	return true;
2782 }
2783 
2784 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2785 			    const struct sys_reg_desc *rd,
2786 			    u64 __user **uind,
2787 			    unsigned int *total)
2788 {
2789 	/*
2790 	 * Ignore registers we trap but don't save,
2791 	 * and for which no custom user accessor is provided.
2792 	 */
2793 	if (!(rd->reg || rd->get_user))
2794 		return 0;
2795 
2796 	if (sysreg_hidden_from_user(vcpu, rd))
2797 		return 0;
2798 
2799 	if (!copy_reg_to_user(rd, uind))
2800 		return -EFAULT;
2801 
2802 	(*total)++;
2803 	return 0;
2804 }
2805 
2806 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2807 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2808 {
2809 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
2810 	unsigned int total = 0;
2811 	size_t num;
2812 	int err;
2813 
2814 	/* We check for duplicates here, to allow arch-specific overrides. */
2815 	i1 = get_target_table(vcpu->arch.target, true, &num);
2816 	end1 = i1 + num;
2817 	i2 = sys_reg_descs;
2818 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2819 
2820 	BUG_ON(i1 == end1 || i2 == end2);
2821 
2822 	/* Walk carefully, as both tables may refer to the same register. */
2823 	while (i1 || i2) {
2824 		int cmp = cmp_sys_reg(i1, i2);
2825 		/* target-specific overrides generic entry. */
2826 		if (cmp <= 0)
2827 			err = walk_one_sys_reg(vcpu, i1, &uind, &total);
2828 		else
2829 			err = walk_one_sys_reg(vcpu, i2, &uind, &total);
2830 
2831 		if (err)
2832 			return err;
2833 
2834 		if (cmp <= 0 && ++i1 == end1)
2835 			i1 = NULL;
2836 		if (cmp >= 0 && ++i2 == end2)
2837 			i2 = NULL;
2838 	}
2839 	return total;
2840 }
2841 
2842 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2843 {
2844 	return ARRAY_SIZE(invariant_sys_regs)
2845 		+ num_demux_regs()
2846 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2847 }
2848 
2849 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2850 {
2851 	unsigned int i;
2852 	int err;
2853 
2854 	/* Then give them all the invariant registers' indices. */
2855 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2856 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2857 			return -EFAULT;
2858 		uindices++;
2859 	}
2860 
2861 	err = walk_sys_regs(vcpu, uindices);
2862 	if (err < 0)
2863 		return err;
2864 	uindices += err;
2865 
2866 	return write_demux_regids(uindices);
2867 }
2868 
2869 void kvm_sys_reg_table_init(void)
2870 {
2871 	unsigned int i;
2872 	struct sys_reg_desc clidr;
2873 
2874 	/* Make sure tables are unique and in order. */
2875 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2876 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2877 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2878 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2879 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2880 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2881 
2882 	/* We abuse the reset function to overwrite the table itself. */
2883 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2884 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2885 
2886 	/*
2887 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2888 	 *
2889 	 *   If software reads the Cache Type fields from Ctype1
2890 	 *   upwards, once it has seen a value of 0b000, no caches
2891 	 *   exist at further-out levels of the hierarchy. So, for
2892 	 *   example, if Ctype3 is the first Cache Type field with a
2893 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2894 	 *   ignored.
2895 	 */
2896 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2897 	cache_levels = clidr.val;
2898 	for (i = 0; i < 7; i++)
2899 		if (((cache_levels >> (i*3)) & 7) == 0)
2900 			break;
2901 	/* Clear all higher bits. */
2902 	cache_levels &= (1 << (i*3))-1;
2903 }
2904 
2905 /**
2906  * kvm_reset_sys_regs - sets system registers to reset value
2907  * @vcpu: The VCPU pointer
2908  *
2909  * This function finds the right table above and sets the registers on the
2910  * virtual CPU struct to their architecturally defined reset values.
2911  */
2912 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2913 {
2914 	size_t num;
2915 	const struct sys_reg_desc *table;
2916 
2917 	/* Generic chip reset first (so target could override). */
2918 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2919 
2920 	table = get_target_table(vcpu->arch.target, true, &num);
2921 	reset_sys_reg_descs(vcpu, table, num);
2922 }
2923