xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision cff11abeca78aa782378401ca2800bd2194aa14e)
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 ACTLR_EL1:		*val = read_sysreg_s(SYS_ACTLR_EL1);	break;
82 	case CPACR_EL1:		*val = read_sysreg_s(SYS_CPACR_EL12);	break;
83 	case TTBR0_EL1:		*val = read_sysreg_s(SYS_TTBR0_EL12);	break;
84 	case TTBR1_EL1:		*val = read_sysreg_s(SYS_TTBR1_EL12);	break;
85 	case TCR_EL1:		*val = read_sysreg_s(SYS_TCR_EL12);	break;
86 	case ESR_EL1:		*val = read_sysreg_s(SYS_ESR_EL12);	break;
87 	case AFSR0_EL1:		*val = read_sysreg_s(SYS_AFSR0_EL12);	break;
88 	case AFSR1_EL1:		*val = read_sysreg_s(SYS_AFSR1_EL12);	break;
89 	case FAR_EL1:		*val = read_sysreg_s(SYS_FAR_EL12);	break;
90 	case MAIR_EL1:		*val = read_sysreg_s(SYS_MAIR_EL12);	break;
91 	case VBAR_EL1:		*val = read_sysreg_s(SYS_VBAR_EL12);	break;
92 	case CONTEXTIDR_EL1:	*val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
93 	case TPIDR_EL0:		*val = read_sysreg_s(SYS_TPIDR_EL0);	break;
94 	case TPIDRRO_EL0:	*val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
95 	case TPIDR_EL1:		*val = read_sysreg_s(SYS_TPIDR_EL1);	break;
96 	case AMAIR_EL1:		*val = read_sysreg_s(SYS_AMAIR_EL12);	break;
97 	case CNTKCTL_EL1:	*val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
98 	case PAR_EL1:		*val = read_sysreg_s(SYS_PAR_EL1);	break;
99 	case DACR32_EL2:	*val = read_sysreg_s(SYS_DACR32_EL2);	break;
100 	case IFSR32_EL2:	*val = read_sysreg_s(SYS_IFSR32_EL2);	break;
101 	case DBGVCR32_EL2:	*val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
102 	default:		return false;
103 	}
104 
105 	return true;
106 }
107 
108 static bool __vcpu_write_sys_reg_to_cpu(u64 val, int reg)
109 {
110 	/*
111 	 * System registers listed in the switch are not restored on every
112 	 * entry to the guest but are only restored on vcpu_load.
113 	 *
114 	 * Note that MPIDR_EL1 for the guest is set by KVM via VMPIDR_EL2 but
115 	 * should never be listed below, because the MPIDR should only be set
116 	 * once, before running the VCPU, and never changed later.
117 	 */
118 	switch (reg) {
119 	case CSSELR_EL1:	write_sysreg_s(val, SYS_CSSELR_EL1);	break;
120 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
121 	case ACTLR_EL1:		write_sysreg_s(val, SYS_ACTLR_EL1);	break;
122 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
123 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
124 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
125 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
126 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
127 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
128 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
129 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
130 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
131 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
132 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
133 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
134 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
135 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
136 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
137 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
138 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
139 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
140 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
141 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
142 	default:		return false;
143 	}
144 
145 	return true;
146 }
147 
148 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, int reg)
149 {
150 	u64 val = 0x8badf00d8badf00d;
151 
152 	if (vcpu->arch.sysregs_loaded_on_cpu &&
153 	    __vcpu_read_sys_reg_from_cpu(reg, &val))
154 		return val;
155 
156 	return __vcpu_sys_reg(vcpu, reg);
157 }
158 
159 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, int reg)
160 {
161 	if (vcpu->arch.sysregs_loaded_on_cpu &&
162 	    __vcpu_write_sys_reg_to_cpu(val, reg))
163 		return;
164 
165 	 __vcpu_sys_reg(vcpu, reg) = val;
166 }
167 
168 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
169 static u32 cache_levels;
170 
171 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
172 #define CSSELR_MAX 12
173 
174 /* Which cache CCSIDR represents depends on CSSELR value. */
175 static u32 get_ccsidr(u32 csselr)
176 {
177 	u32 ccsidr;
178 
179 	/* Make sure noone else changes CSSELR during this! */
180 	local_irq_disable();
181 	write_sysreg(csselr, csselr_el1);
182 	isb();
183 	ccsidr = read_sysreg(ccsidr_el1);
184 	local_irq_enable();
185 
186 	return ccsidr;
187 }
188 
189 /*
190  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
191  */
192 static bool access_dcsw(struct kvm_vcpu *vcpu,
193 			struct sys_reg_params *p,
194 			const struct sys_reg_desc *r)
195 {
196 	if (!p->is_write)
197 		return read_from_write_only(vcpu, p, r);
198 
199 	/*
200 	 * Only track S/W ops if we don't have FWB. It still indicates
201 	 * that the guest is a bit broken (S/W operations should only
202 	 * be done by firmware, knowing that there is only a single
203 	 * CPU left in the system, and certainly not from non-secure
204 	 * software).
205 	 */
206 	if (!cpus_have_const_cap(ARM64_HAS_STAGE2_FWB))
207 		kvm_set_way_flush(vcpu);
208 
209 	return true;
210 }
211 
212 /*
213  * Generic accessor for VM registers. Only called as long as HCR_TVM
214  * is set. If the guest enables the MMU, we stop trapping the VM
215  * sys_regs and leave it in complete control of the caches.
216  */
217 static bool access_vm_reg(struct kvm_vcpu *vcpu,
218 			  struct sys_reg_params *p,
219 			  const struct sys_reg_desc *r)
220 {
221 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
222 	u64 val;
223 	int reg = r->reg;
224 
225 	BUG_ON(!p->is_write);
226 
227 	/* See the 32bit mapping in kvm_host.h */
228 	if (p->is_aarch32)
229 		reg = r->reg / 2;
230 
231 	if (!p->is_aarch32 || !p->is_32bit) {
232 		val = p->regval;
233 	} else {
234 		val = vcpu_read_sys_reg(vcpu, reg);
235 		if (r->reg % 2)
236 			val = (p->regval << 32) | (u64)lower_32_bits(val);
237 		else
238 			val = ((u64)upper_32_bits(val) << 32) |
239 				lower_32_bits(p->regval);
240 	}
241 	vcpu_write_sys_reg(vcpu, val, reg);
242 
243 	kvm_toggle_cache(vcpu, was_enabled);
244 	return true;
245 }
246 
247 /*
248  * Trap handler for the GICv3 SGI generation system register.
249  * Forward the request to the VGIC emulation.
250  * The cp15_64 code makes sure this automatically works
251  * for both AArch64 and AArch32 accesses.
252  */
253 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
254 			   struct sys_reg_params *p,
255 			   const struct sys_reg_desc *r)
256 {
257 	bool g1;
258 
259 	if (!p->is_write)
260 		return read_from_write_only(vcpu, p, r);
261 
262 	/*
263 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
264 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
265 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
266 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
267 	 * group.
268 	 */
269 	if (p->is_aarch32) {
270 		switch (p->Op1) {
271 		default:		/* Keep GCC quiet */
272 		case 0:			/* ICC_SGI1R */
273 			g1 = true;
274 			break;
275 		case 1:			/* ICC_ASGI1R */
276 		case 2:			/* ICC_SGI0R */
277 			g1 = false;
278 			break;
279 		}
280 	} else {
281 		switch (p->Op2) {
282 		default:		/* Keep GCC quiet */
283 		case 5:			/* ICC_SGI1R_EL1 */
284 			g1 = true;
285 			break;
286 		case 6:			/* ICC_ASGI1R_EL1 */
287 		case 7:			/* ICC_SGI0R_EL1 */
288 			g1 = false;
289 			break;
290 		}
291 	}
292 
293 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
294 
295 	return true;
296 }
297 
298 static bool access_gic_sre(struct kvm_vcpu *vcpu,
299 			   struct sys_reg_params *p,
300 			   const struct sys_reg_desc *r)
301 {
302 	if (p->is_write)
303 		return ignore_write(vcpu, p);
304 
305 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
306 	return true;
307 }
308 
309 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
310 			struct sys_reg_params *p,
311 			const struct sys_reg_desc *r)
312 {
313 	if (p->is_write)
314 		return ignore_write(vcpu, p);
315 	else
316 		return read_zero(vcpu, p);
317 }
318 
319 /*
320  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
321  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
322  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
323  * treat it separately.
324  */
325 static bool trap_loregion(struct kvm_vcpu *vcpu,
326 			  struct sys_reg_params *p,
327 			  const struct sys_reg_desc *r)
328 {
329 	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
330 	u32 sr = sys_reg((u32)r->Op0, (u32)r->Op1,
331 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
332 
333 	if (!(val & (0xfUL << ID_AA64MMFR1_LOR_SHIFT))) {
334 		kvm_inject_undefined(vcpu);
335 		return false;
336 	}
337 
338 	if (p->is_write && sr == SYS_LORID_EL1)
339 		return write_to_read_only(vcpu, p, r);
340 
341 	return trap_raz_wi(vcpu, p, r);
342 }
343 
344 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
345 			   struct sys_reg_params *p,
346 			   const struct sys_reg_desc *r)
347 {
348 	if (p->is_write) {
349 		return ignore_write(vcpu, p);
350 	} else {
351 		p->regval = (1 << 3);
352 		return true;
353 	}
354 }
355 
356 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
357 				   struct sys_reg_params *p,
358 				   const struct sys_reg_desc *r)
359 {
360 	if (p->is_write) {
361 		return ignore_write(vcpu, p);
362 	} else {
363 		p->regval = read_sysreg(dbgauthstatus_el1);
364 		return true;
365 	}
366 }
367 
368 /*
369  * We want to avoid world-switching all the DBG registers all the
370  * time:
371  *
372  * - If we've touched any debug register, it is likely that we're
373  *   going to touch more of them. It then makes sense to disable the
374  *   traps and start doing the save/restore dance
375  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
376  *   then mandatory to save/restore the registers, as the guest
377  *   depends on them.
378  *
379  * For this, we use a DIRTY bit, indicating the guest has modified the
380  * debug registers, used as follow:
381  *
382  * On guest entry:
383  * - If the dirty bit is set (because we're coming back from trapping),
384  *   disable the traps, save host registers, restore guest registers.
385  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
386  *   set the dirty bit, disable the traps, save host registers,
387  *   restore guest registers.
388  * - Otherwise, enable the traps
389  *
390  * On guest exit:
391  * - If the dirty bit is set, save guest registers, restore host
392  *   registers and clear the dirty bit. This ensure that the host can
393  *   now use the debug registers.
394  */
395 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
396 			    struct sys_reg_params *p,
397 			    const struct sys_reg_desc *r)
398 {
399 	if (p->is_write) {
400 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
401 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
402 	} else {
403 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
404 	}
405 
406 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
407 
408 	return true;
409 }
410 
411 /*
412  * reg_to_dbg/dbg_to_reg
413  *
414  * A 32 bit write to a debug register leave top bits alone
415  * A 32 bit read from a debug register only returns the bottom bits
416  *
417  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
418  * hyp.S code switches between host and guest values in future.
419  */
420 static void reg_to_dbg(struct kvm_vcpu *vcpu,
421 		       struct sys_reg_params *p,
422 		       u64 *dbg_reg)
423 {
424 	u64 val = p->regval;
425 
426 	if (p->is_32bit) {
427 		val &= 0xffffffffUL;
428 		val |= ((*dbg_reg >> 32) << 32);
429 	}
430 
431 	*dbg_reg = val;
432 	vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
433 }
434 
435 static void dbg_to_reg(struct kvm_vcpu *vcpu,
436 		       struct sys_reg_params *p,
437 		       u64 *dbg_reg)
438 {
439 	p->regval = *dbg_reg;
440 	if (p->is_32bit)
441 		p->regval &= 0xffffffffUL;
442 }
443 
444 static bool trap_bvr(struct kvm_vcpu *vcpu,
445 		     struct sys_reg_params *p,
446 		     const struct sys_reg_desc *rd)
447 {
448 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
449 
450 	if (p->is_write)
451 		reg_to_dbg(vcpu, p, dbg_reg);
452 	else
453 		dbg_to_reg(vcpu, p, dbg_reg);
454 
455 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
456 
457 	return true;
458 }
459 
460 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
461 		const struct kvm_one_reg *reg, void __user *uaddr)
462 {
463 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
464 
465 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
466 		return -EFAULT;
467 	return 0;
468 }
469 
470 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
471 	const struct kvm_one_reg *reg, void __user *uaddr)
472 {
473 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
474 
475 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
476 		return -EFAULT;
477 	return 0;
478 }
479 
480 static void reset_bvr(struct kvm_vcpu *vcpu,
481 		      const struct sys_reg_desc *rd)
482 {
483 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
484 }
485 
486 static bool trap_bcr(struct kvm_vcpu *vcpu,
487 		     struct sys_reg_params *p,
488 		     const struct sys_reg_desc *rd)
489 {
490 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
491 
492 	if (p->is_write)
493 		reg_to_dbg(vcpu, p, dbg_reg);
494 	else
495 		dbg_to_reg(vcpu, p, dbg_reg);
496 
497 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
498 
499 	return true;
500 }
501 
502 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
503 		const struct kvm_one_reg *reg, void __user *uaddr)
504 {
505 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
506 
507 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
508 		return -EFAULT;
509 
510 	return 0;
511 }
512 
513 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
514 	const struct kvm_one_reg *reg, void __user *uaddr)
515 {
516 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
517 
518 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
519 		return -EFAULT;
520 	return 0;
521 }
522 
523 static void reset_bcr(struct kvm_vcpu *vcpu,
524 		      const struct sys_reg_desc *rd)
525 {
526 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
527 }
528 
529 static bool trap_wvr(struct kvm_vcpu *vcpu,
530 		     struct sys_reg_params *p,
531 		     const struct sys_reg_desc *rd)
532 {
533 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
534 
535 	if (p->is_write)
536 		reg_to_dbg(vcpu, p, dbg_reg);
537 	else
538 		dbg_to_reg(vcpu, p, dbg_reg);
539 
540 	trace_trap_reg(__func__, rd->reg, p->is_write,
541 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
542 
543 	return true;
544 }
545 
546 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
547 		const struct kvm_one_reg *reg, void __user *uaddr)
548 {
549 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
550 
551 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
552 		return -EFAULT;
553 	return 0;
554 }
555 
556 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
557 	const struct kvm_one_reg *reg, void __user *uaddr)
558 {
559 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
560 
561 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
562 		return -EFAULT;
563 	return 0;
564 }
565 
566 static void reset_wvr(struct kvm_vcpu *vcpu,
567 		      const struct sys_reg_desc *rd)
568 {
569 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
570 }
571 
572 static bool trap_wcr(struct kvm_vcpu *vcpu,
573 		     struct sys_reg_params *p,
574 		     const struct sys_reg_desc *rd)
575 {
576 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
577 
578 	if (p->is_write)
579 		reg_to_dbg(vcpu, p, dbg_reg);
580 	else
581 		dbg_to_reg(vcpu, p, dbg_reg);
582 
583 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
584 
585 	return true;
586 }
587 
588 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
589 		const struct kvm_one_reg *reg, void __user *uaddr)
590 {
591 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
592 
593 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
594 		return -EFAULT;
595 	return 0;
596 }
597 
598 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
599 	const struct kvm_one_reg *reg, void __user *uaddr)
600 {
601 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
602 
603 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
604 		return -EFAULT;
605 	return 0;
606 }
607 
608 static void reset_wcr(struct kvm_vcpu *vcpu,
609 		      const struct sys_reg_desc *rd)
610 {
611 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
612 }
613 
614 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
615 {
616 	u64 amair = read_sysreg(amair_el1);
617 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
618 }
619 
620 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
621 {
622 	u64 mpidr;
623 
624 	/*
625 	 * Map the vcpu_id into the first three affinity level fields of
626 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
627 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
628 	 * of the GICv3 to be able to address each CPU directly when
629 	 * sending IPIs.
630 	 */
631 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
632 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
633 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
634 	vcpu_write_sys_reg(vcpu, (1ULL << 31) | mpidr, MPIDR_EL1);
635 }
636 
637 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
638 {
639 	u64 pmcr, val;
640 
641 	pmcr = read_sysreg(pmcr_el0);
642 	/*
643 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
644 	 * except PMCR.E resetting to zero.
645 	 */
646 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
647 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
648 	if (!system_supports_32bit_el0())
649 		val |= ARMV8_PMU_PMCR_LC;
650 	__vcpu_sys_reg(vcpu, r->reg) = val;
651 }
652 
653 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
654 {
655 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
656 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
657 
658 	if (!enabled)
659 		kvm_inject_undefined(vcpu);
660 
661 	return !enabled;
662 }
663 
664 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
665 {
666 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
667 }
668 
669 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
670 {
671 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
672 }
673 
674 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
675 {
676 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
677 }
678 
679 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
680 {
681 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
682 }
683 
684 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
685 			const struct sys_reg_desc *r)
686 {
687 	u64 val;
688 
689 	if (!kvm_arm_pmu_v3_ready(vcpu))
690 		return trap_raz_wi(vcpu, p, r);
691 
692 	if (pmu_access_el0_disabled(vcpu))
693 		return false;
694 
695 	if (p->is_write) {
696 		/* Only update writeable bits of PMCR */
697 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
698 		val &= ~ARMV8_PMU_PMCR_MASK;
699 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
700 		if (!system_supports_32bit_el0())
701 			val |= ARMV8_PMU_PMCR_LC;
702 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
703 		kvm_pmu_handle_pmcr(vcpu, val);
704 		kvm_vcpu_pmu_restore_guest(vcpu);
705 	} else {
706 		/* PMCR.P & PMCR.C are RAZ */
707 		val = __vcpu_sys_reg(vcpu, PMCR_EL0)
708 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
709 		p->regval = val;
710 	}
711 
712 	return true;
713 }
714 
715 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
716 			  const struct sys_reg_desc *r)
717 {
718 	if (!kvm_arm_pmu_v3_ready(vcpu))
719 		return trap_raz_wi(vcpu, p, r);
720 
721 	if (pmu_access_event_counter_el0_disabled(vcpu))
722 		return false;
723 
724 	if (p->is_write)
725 		__vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
726 	else
727 		/* return PMSELR.SEL field */
728 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
729 			    & ARMV8_PMU_COUNTER_MASK;
730 
731 	return true;
732 }
733 
734 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
735 			  const struct sys_reg_desc *r)
736 {
737 	u64 pmceid;
738 
739 	if (!kvm_arm_pmu_v3_ready(vcpu))
740 		return trap_raz_wi(vcpu, p, r);
741 
742 	BUG_ON(p->is_write);
743 
744 	if (pmu_access_el0_disabled(vcpu))
745 		return false;
746 
747 	if (!(p->Op2 & 1))
748 		pmceid = read_sysreg(pmceid0_el0);
749 	else
750 		pmceid = read_sysreg(pmceid1_el0);
751 
752 	p->regval = pmceid;
753 
754 	return true;
755 }
756 
757 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
758 {
759 	u64 pmcr, val;
760 
761 	pmcr = __vcpu_sys_reg(vcpu, PMCR_EL0);
762 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
763 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
764 		kvm_inject_undefined(vcpu);
765 		return false;
766 	}
767 
768 	return true;
769 }
770 
771 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
772 			      struct sys_reg_params *p,
773 			      const struct sys_reg_desc *r)
774 {
775 	u64 idx;
776 
777 	if (!kvm_arm_pmu_v3_ready(vcpu))
778 		return trap_raz_wi(vcpu, p, r);
779 
780 	if (r->CRn == 9 && r->CRm == 13) {
781 		if (r->Op2 == 2) {
782 			/* PMXEVCNTR_EL0 */
783 			if (pmu_access_event_counter_el0_disabled(vcpu))
784 				return false;
785 
786 			idx = __vcpu_sys_reg(vcpu, PMSELR_EL0)
787 			      & ARMV8_PMU_COUNTER_MASK;
788 		} else if (r->Op2 == 0) {
789 			/* PMCCNTR_EL0 */
790 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
791 				return false;
792 
793 			idx = ARMV8_PMU_CYCLE_IDX;
794 		} else {
795 			return false;
796 		}
797 	} else if (r->CRn == 0 && r->CRm == 9) {
798 		/* PMCCNTR */
799 		if (pmu_access_event_counter_el0_disabled(vcpu))
800 			return false;
801 
802 		idx = ARMV8_PMU_CYCLE_IDX;
803 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
804 		/* PMEVCNTRn_EL0 */
805 		if (pmu_access_event_counter_el0_disabled(vcpu))
806 			return false;
807 
808 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
809 	} else {
810 		return false;
811 	}
812 
813 	if (!pmu_counter_idx_valid(vcpu, idx))
814 		return false;
815 
816 	if (p->is_write) {
817 		if (pmu_access_el0_disabled(vcpu))
818 			return false;
819 
820 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
821 	} else {
822 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
823 	}
824 
825 	return true;
826 }
827 
828 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
829 			       const struct sys_reg_desc *r)
830 {
831 	u64 idx, reg;
832 
833 	if (!kvm_arm_pmu_v3_ready(vcpu))
834 		return trap_raz_wi(vcpu, p, r);
835 
836 	if (pmu_access_el0_disabled(vcpu))
837 		return false;
838 
839 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
840 		/* PMXEVTYPER_EL0 */
841 		idx = __vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
842 		reg = PMEVTYPER0_EL0 + idx;
843 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
844 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
845 		if (idx == ARMV8_PMU_CYCLE_IDX)
846 			reg = PMCCFILTR_EL0;
847 		else
848 			/* PMEVTYPERn_EL0 */
849 			reg = PMEVTYPER0_EL0 + idx;
850 	} else {
851 		BUG();
852 	}
853 
854 	if (!pmu_counter_idx_valid(vcpu, idx))
855 		return false;
856 
857 	if (p->is_write) {
858 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
859 		__vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
860 		kvm_vcpu_pmu_restore_guest(vcpu);
861 	} else {
862 		p->regval = __vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
863 	}
864 
865 	return true;
866 }
867 
868 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
869 			   const struct sys_reg_desc *r)
870 {
871 	u64 val, mask;
872 
873 	if (!kvm_arm_pmu_v3_ready(vcpu))
874 		return trap_raz_wi(vcpu, p, r);
875 
876 	if (pmu_access_el0_disabled(vcpu))
877 		return false;
878 
879 	mask = kvm_pmu_valid_counter_mask(vcpu);
880 	if (p->is_write) {
881 		val = p->regval & mask;
882 		if (r->Op2 & 0x1) {
883 			/* accessing PMCNTENSET_EL0 */
884 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
885 			kvm_pmu_enable_counter_mask(vcpu, val);
886 			kvm_vcpu_pmu_restore_guest(vcpu);
887 		} else {
888 			/* accessing PMCNTENCLR_EL0 */
889 			__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
890 			kvm_pmu_disable_counter_mask(vcpu, val);
891 		}
892 	} else {
893 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
894 	}
895 
896 	return true;
897 }
898 
899 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
900 			   const struct sys_reg_desc *r)
901 {
902 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
903 
904 	if (!kvm_arm_pmu_v3_ready(vcpu))
905 		return trap_raz_wi(vcpu, p, r);
906 
907 	if (!vcpu_mode_priv(vcpu)) {
908 		kvm_inject_undefined(vcpu);
909 		return false;
910 	}
911 
912 	if (p->is_write) {
913 		u64 val = p->regval & mask;
914 
915 		if (r->Op2 & 0x1)
916 			/* accessing PMINTENSET_EL1 */
917 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
918 		else
919 			/* accessing PMINTENCLR_EL1 */
920 			__vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
921 	} else {
922 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
923 	}
924 
925 	return true;
926 }
927 
928 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
929 			 const struct sys_reg_desc *r)
930 {
931 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
932 
933 	if (!kvm_arm_pmu_v3_ready(vcpu))
934 		return trap_raz_wi(vcpu, p, r);
935 
936 	if (pmu_access_el0_disabled(vcpu))
937 		return false;
938 
939 	if (p->is_write) {
940 		if (r->CRm & 0x2)
941 			/* accessing PMOVSSET_EL0 */
942 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= (p->regval & mask);
943 		else
944 			/* accessing PMOVSCLR_EL0 */
945 			__vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
946 	} else {
947 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
948 	}
949 
950 	return true;
951 }
952 
953 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
954 			   const struct sys_reg_desc *r)
955 {
956 	u64 mask;
957 
958 	if (!kvm_arm_pmu_v3_ready(vcpu))
959 		return trap_raz_wi(vcpu, p, r);
960 
961 	if (!p->is_write)
962 		return read_from_write_only(vcpu, p, r);
963 
964 	if (pmu_write_swinc_el0_disabled(vcpu))
965 		return false;
966 
967 	mask = kvm_pmu_valid_counter_mask(vcpu);
968 	kvm_pmu_software_increment(vcpu, p->regval & mask);
969 	return true;
970 }
971 
972 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
973 			     const struct sys_reg_desc *r)
974 {
975 	if (!kvm_arm_pmu_v3_ready(vcpu))
976 		return trap_raz_wi(vcpu, p, r);
977 
978 	if (p->is_write) {
979 		if (!vcpu_mode_priv(vcpu)) {
980 			kvm_inject_undefined(vcpu);
981 			return false;
982 		}
983 
984 		__vcpu_sys_reg(vcpu, PMUSERENR_EL0) =
985 			       p->regval & ARMV8_PMU_USERENR_MASK;
986 	} else {
987 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
988 			    & ARMV8_PMU_USERENR_MASK;
989 	}
990 
991 	return true;
992 }
993 
994 #define reg_to_encoding(x)						\
995 	sys_reg((u32)(x)->Op0, (u32)(x)->Op1,				\
996 		(u32)(x)->CRn, (u32)(x)->CRm, (u32)(x)->Op2);
997 
998 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
999 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1000 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1001 	  trap_bvr, reset_bvr, 0, 0, get_bvr, set_bvr },		\
1002 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1003 	  trap_bcr, reset_bcr, 0, 0, get_bcr, set_bcr },		\
1004 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1005 	  trap_wvr, reset_wvr, 0, 0,  get_wvr, set_wvr },		\
1006 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1007 	  trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
1008 
1009 /* Macro to expand the PMEVCNTRn_EL0 register */
1010 #define PMU_PMEVCNTR_EL0(n)						\
1011 	{ SYS_DESC(SYS_PMEVCNTRn_EL0(n)),					\
1012 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
1013 
1014 /* Macro to expand the PMEVTYPERn_EL0 register */
1015 #define PMU_PMEVTYPER_EL0(n)						\
1016 	{ SYS_DESC(SYS_PMEVTYPERn_EL0(n)),					\
1017 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
1018 
1019 static bool access_amu(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1020 			     const struct sys_reg_desc *r)
1021 {
1022 	kvm_inject_undefined(vcpu);
1023 
1024 	return false;
1025 }
1026 
1027 /* Macro to expand the AMU counter and type registers*/
1028 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), access_amu }
1029 #define AMU_AMEVTYPE0_EL0(n) { SYS_DESC(SYS_AMEVTYPE0_EL0(n)), access_amu }
1030 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), access_amu }
1031 #define AMU_AMEVTYPE1_EL0(n) { SYS_DESC(SYS_AMEVTYPE1_EL0(n)), access_amu }
1032 
1033 static bool trap_ptrauth(struct kvm_vcpu *vcpu,
1034 			 struct sys_reg_params *p,
1035 			 const struct sys_reg_desc *rd)
1036 {
1037 	kvm_arm_vcpu_ptrauth_trap(vcpu);
1038 
1039 	/*
1040 	 * Return false for both cases as we never skip the trapped
1041 	 * instruction:
1042 	 *
1043 	 * - Either we re-execute the same key register access instruction
1044 	 *   after enabling ptrauth.
1045 	 * - Or an UNDEF is injected as ptrauth is not supported/enabled.
1046 	 */
1047 	return false;
1048 }
1049 
1050 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1051 			const struct sys_reg_desc *rd)
1052 {
1053 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1054 }
1055 
1056 #define __PTRAUTH_KEY(k)						\
1057 	{ SYS_DESC(SYS_## k), trap_ptrauth, reset_unknown, k,		\
1058 	.visibility = ptrauth_visibility}
1059 
1060 #define PTRAUTH_KEY(k)							\
1061 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1062 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1063 
1064 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1065 			      struct sys_reg_params *p,
1066 			      const struct sys_reg_desc *r)
1067 {
1068 	enum kvm_arch_timers tmr;
1069 	enum kvm_arch_timer_regs treg;
1070 	u64 reg = reg_to_encoding(r);
1071 
1072 	switch (reg) {
1073 	case SYS_CNTP_TVAL_EL0:
1074 	case SYS_AARCH32_CNTP_TVAL:
1075 		tmr = TIMER_PTIMER;
1076 		treg = TIMER_REG_TVAL;
1077 		break;
1078 	case SYS_CNTP_CTL_EL0:
1079 	case SYS_AARCH32_CNTP_CTL:
1080 		tmr = TIMER_PTIMER;
1081 		treg = TIMER_REG_CTL;
1082 		break;
1083 	case SYS_CNTP_CVAL_EL0:
1084 	case SYS_AARCH32_CNTP_CVAL:
1085 		tmr = TIMER_PTIMER;
1086 		treg = TIMER_REG_CVAL;
1087 		break;
1088 	default:
1089 		BUG();
1090 	}
1091 
1092 	if (p->is_write)
1093 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1094 	else
1095 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1096 
1097 	return true;
1098 }
1099 
1100 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1101 static u64 read_id_reg(const struct kvm_vcpu *vcpu,
1102 		struct sys_reg_desc const *r, bool raz)
1103 {
1104 	u32 id = sys_reg((u32)r->Op0, (u32)r->Op1,
1105 			 (u32)r->CRn, (u32)r->CRm, (u32)r->Op2);
1106 	u64 val = raz ? 0 : read_sanitised_ftr_reg(id);
1107 
1108 	if (id == SYS_ID_AA64PFR0_EL1) {
1109 		if (!vcpu_has_sve(vcpu))
1110 			val &= ~(0xfUL << ID_AA64PFR0_SVE_SHIFT);
1111 		val &= ~(0xfUL << ID_AA64PFR0_AMU_SHIFT);
1112 	} else if (id == SYS_ID_AA64ISAR1_EL1 && !vcpu_has_ptrauth(vcpu)) {
1113 		val &= ~((0xfUL << ID_AA64ISAR1_APA_SHIFT) |
1114 			 (0xfUL << ID_AA64ISAR1_API_SHIFT) |
1115 			 (0xfUL << ID_AA64ISAR1_GPA_SHIFT) |
1116 			 (0xfUL << ID_AA64ISAR1_GPI_SHIFT));
1117 	} else if (id == SYS_ID_AA64DFR0_EL1) {
1118 		/* Limit guests to PMUv3 for ARMv8.1 */
1119 		val = cpuid_feature_cap_perfmon_field(val,
1120 						ID_AA64DFR0_PMUVER_SHIFT,
1121 						ID_AA64DFR0_PMUVER_8_1);
1122 	} else if (id == SYS_ID_DFR0_EL1) {
1123 		/* Limit guests to PMUv3 for ARMv8.1 */
1124 		val = cpuid_feature_cap_perfmon_field(val,
1125 						ID_DFR0_PERFMON_SHIFT,
1126 						ID_DFR0_PERFMON_8_1);
1127 	}
1128 
1129 	return val;
1130 }
1131 
1132 /* cpufeature ID register access trap handlers */
1133 
1134 static bool __access_id_reg(struct kvm_vcpu *vcpu,
1135 			    struct sys_reg_params *p,
1136 			    const struct sys_reg_desc *r,
1137 			    bool raz)
1138 {
1139 	if (p->is_write)
1140 		return write_to_read_only(vcpu, p, r);
1141 
1142 	p->regval = read_id_reg(vcpu, r, raz);
1143 	return true;
1144 }
1145 
1146 static bool access_id_reg(struct kvm_vcpu *vcpu,
1147 			  struct sys_reg_params *p,
1148 			  const struct sys_reg_desc *r)
1149 {
1150 	return __access_id_reg(vcpu, p, r, false);
1151 }
1152 
1153 static bool access_raz_id_reg(struct kvm_vcpu *vcpu,
1154 			      struct sys_reg_params *p,
1155 			      const struct sys_reg_desc *r)
1156 {
1157 	return __access_id_reg(vcpu, p, r, true);
1158 }
1159 
1160 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id);
1161 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id);
1162 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
1163 
1164 /* Visibility overrides for SVE-specific control registers */
1165 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1166 				   const struct sys_reg_desc *rd)
1167 {
1168 	if (vcpu_has_sve(vcpu))
1169 		return 0;
1170 
1171 	return REG_HIDDEN_USER | REG_HIDDEN_GUEST;
1172 }
1173 
1174 /* Visibility overrides for SVE-specific ID registers */
1175 static unsigned int sve_id_visibility(const struct kvm_vcpu *vcpu,
1176 				      const struct sys_reg_desc *rd)
1177 {
1178 	if (vcpu_has_sve(vcpu))
1179 		return 0;
1180 
1181 	return REG_HIDDEN_USER;
1182 }
1183 
1184 /* Generate the emulated ID_AA64ZFR0_EL1 value exposed to the guest */
1185 static u64 guest_id_aa64zfr0_el1(const struct kvm_vcpu *vcpu)
1186 {
1187 	if (!vcpu_has_sve(vcpu))
1188 		return 0;
1189 
1190 	return read_sanitised_ftr_reg(SYS_ID_AA64ZFR0_EL1);
1191 }
1192 
1193 static bool access_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1194 				   struct sys_reg_params *p,
1195 				   const struct sys_reg_desc *rd)
1196 {
1197 	if (p->is_write)
1198 		return write_to_read_only(vcpu, p, rd);
1199 
1200 	p->regval = guest_id_aa64zfr0_el1(vcpu);
1201 	return true;
1202 }
1203 
1204 static int get_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1205 		const struct sys_reg_desc *rd,
1206 		const struct kvm_one_reg *reg, void __user *uaddr)
1207 {
1208 	u64 val;
1209 
1210 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1211 		return -ENOENT;
1212 
1213 	val = guest_id_aa64zfr0_el1(vcpu);
1214 	return reg_to_user(uaddr, &val, reg->id);
1215 }
1216 
1217 static int set_id_aa64zfr0_el1(struct kvm_vcpu *vcpu,
1218 		const struct sys_reg_desc *rd,
1219 		const struct kvm_one_reg *reg, void __user *uaddr)
1220 {
1221 	const u64 id = sys_reg_to_index(rd);
1222 	int err;
1223 	u64 val;
1224 
1225 	if (WARN_ON(!vcpu_has_sve(vcpu)))
1226 		return -ENOENT;
1227 
1228 	err = reg_from_user(&val, uaddr, id);
1229 	if (err)
1230 		return err;
1231 
1232 	/* This is what we mean by invariant: you can't change it. */
1233 	if (val != guest_id_aa64zfr0_el1(vcpu))
1234 		return -EINVAL;
1235 
1236 	return 0;
1237 }
1238 
1239 /*
1240  * cpufeature ID register user accessors
1241  *
1242  * For now, these registers are immutable for userspace, so no values
1243  * are stored, and for set_id_reg() we don't allow the effective value
1244  * to be changed.
1245  */
1246 static int __get_id_reg(const struct kvm_vcpu *vcpu,
1247 			const struct sys_reg_desc *rd, void __user *uaddr,
1248 			bool raz)
1249 {
1250 	const u64 id = sys_reg_to_index(rd);
1251 	const u64 val = read_id_reg(vcpu, rd, raz);
1252 
1253 	return reg_to_user(uaddr, &val, id);
1254 }
1255 
1256 static int __set_id_reg(const struct kvm_vcpu *vcpu,
1257 			const struct sys_reg_desc *rd, void __user *uaddr,
1258 			bool raz)
1259 {
1260 	const u64 id = sys_reg_to_index(rd);
1261 	int err;
1262 	u64 val;
1263 
1264 	err = reg_from_user(&val, uaddr, id);
1265 	if (err)
1266 		return err;
1267 
1268 	/* This is what we mean by invariant: you can't change it. */
1269 	if (val != read_id_reg(vcpu, rd, raz))
1270 		return -EINVAL;
1271 
1272 	return 0;
1273 }
1274 
1275 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1276 		      const struct kvm_one_reg *reg, void __user *uaddr)
1277 {
1278 	return __get_id_reg(vcpu, rd, uaddr, false);
1279 }
1280 
1281 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1282 		      const struct kvm_one_reg *reg, void __user *uaddr)
1283 {
1284 	return __set_id_reg(vcpu, rd, uaddr, false);
1285 }
1286 
1287 static int get_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1288 			  const struct kvm_one_reg *reg, void __user *uaddr)
1289 {
1290 	return __get_id_reg(vcpu, rd, uaddr, true);
1291 }
1292 
1293 static int set_raz_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
1294 			  const struct kvm_one_reg *reg, void __user *uaddr)
1295 {
1296 	return __set_id_reg(vcpu, rd, uaddr, true);
1297 }
1298 
1299 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1300 		       const struct sys_reg_desc *r)
1301 {
1302 	if (p->is_write)
1303 		return write_to_read_only(vcpu, p, r);
1304 
1305 	p->regval = read_sanitised_ftr_reg(SYS_CTR_EL0);
1306 	return true;
1307 }
1308 
1309 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1310 			 const struct sys_reg_desc *r)
1311 {
1312 	if (p->is_write)
1313 		return write_to_read_only(vcpu, p, r);
1314 
1315 	p->regval = read_sysreg(clidr_el1);
1316 	return true;
1317 }
1318 
1319 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1320 			  const struct sys_reg_desc *r)
1321 {
1322 	if (p->is_write)
1323 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
1324 	else
1325 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
1326 	return true;
1327 }
1328 
1329 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1330 			  const struct sys_reg_desc *r)
1331 {
1332 	u32 csselr;
1333 
1334 	if (p->is_write)
1335 		return write_to_read_only(vcpu, p, r);
1336 
1337 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
1338 	p->regval = get_ccsidr(csselr);
1339 
1340 	/*
1341 	 * Guests should not be doing cache operations by set/way at all, and
1342 	 * for this reason, we trap them and attempt to infer the intent, so
1343 	 * that we can flush the entire guest's address space at the appropriate
1344 	 * time.
1345 	 * To prevent this trapping from causing performance problems, let's
1346 	 * expose the geometry of all data and unified caches (which are
1347 	 * guaranteed to be PIPT and thus non-aliasing) as 1 set and 1 way.
1348 	 * [If guests should attempt to infer aliasing properties from the
1349 	 * geometry (which is not permitted by the architecture), they would
1350 	 * only do so for virtually indexed caches.]
1351 	 */
1352 	if (!(csselr & 1)) // data or unified cache
1353 		p->regval &= ~GENMASK(27, 3);
1354 	return true;
1355 }
1356 
1357 /* sys_reg_desc initialiser for known cpufeature ID registers */
1358 #define ID_SANITISED(name) {			\
1359 	SYS_DESC(SYS_##name),			\
1360 	.access	= access_id_reg,		\
1361 	.get_user = get_id_reg,			\
1362 	.set_user = set_id_reg,			\
1363 }
1364 
1365 /*
1366  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
1367  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
1368  * (1 <= crm < 8, 0 <= Op2 < 8).
1369  */
1370 #define ID_UNALLOCATED(crm, op2) {			\
1371 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
1372 	.access = access_raz_id_reg,			\
1373 	.get_user = get_raz_id_reg,			\
1374 	.set_user = set_raz_id_reg,			\
1375 }
1376 
1377 /*
1378  * sys_reg_desc initialiser for known ID registers that we hide from guests.
1379  * For now, these are exposed just like unallocated ID regs: they appear
1380  * RAZ for the guest.
1381  */
1382 #define ID_HIDDEN(name) {			\
1383 	SYS_DESC(SYS_##name),			\
1384 	.access = access_raz_id_reg,		\
1385 	.get_user = get_raz_id_reg,		\
1386 	.set_user = set_raz_id_reg,		\
1387 }
1388 
1389 /*
1390  * Architected system registers.
1391  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
1392  *
1393  * Debug handling: We do trap most, if not all debug related system
1394  * registers. The implementation is good enough to ensure that a guest
1395  * can use these with minimal performance degradation. The drawback is
1396  * that we don't implement any of the external debug, none of the
1397  * OSlock protocol. This should be revisited if we ever encounter a
1398  * more demanding guest...
1399  */
1400 static const struct sys_reg_desc sys_reg_descs[] = {
1401 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
1402 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
1403 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
1404 
1405 	DBG_BCR_BVR_WCR_WVR_EL1(0),
1406 	DBG_BCR_BVR_WCR_WVR_EL1(1),
1407 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
1408 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
1409 	DBG_BCR_BVR_WCR_WVR_EL1(2),
1410 	DBG_BCR_BVR_WCR_WVR_EL1(3),
1411 	DBG_BCR_BVR_WCR_WVR_EL1(4),
1412 	DBG_BCR_BVR_WCR_WVR_EL1(5),
1413 	DBG_BCR_BVR_WCR_WVR_EL1(6),
1414 	DBG_BCR_BVR_WCR_WVR_EL1(7),
1415 	DBG_BCR_BVR_WCR_WVR_EL1(8),
1416 	DBG_BCR_BVR_WCR_WVR_EL1(9),
1417 	DBG_BCR_BVR_WCR_WVR_EL1(10),
1418 	DBG_BCR_BVR_WCR_WVR_EL1(11),
1419 	DBG_BCR_BVR_WCR_WVR_EL1(12),
1420 	DBG_BCR_BVR_WCR_WVR_EL1(13),
1421 	DBG_BCR_BVR_WCR_WVR_EL1(14),
1422 	DBG_BCR_BVR_WCR_WVR_EL1(15),
1423 
1424 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
1425 	{ SYS_DESC(SYS_OSLAR_EL1), trap_raz_wi },
1426 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1 },
1427 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
1428 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
1429 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
1430 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
1431 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
1432 
1433 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
1434 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
1435 	// DBGDTR[TR]X_EL0 share the same encoding
1436 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
1437 
1438 	{ SYS_DESC(SYS_DBGVCR32_EL2), NULL, reset_val, DBGVCR32_EL2, 0 },
1439 
1440 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
1441 
1442 	/*
1443 	 * ID regs: all ID_SANITISED() entries here must have corresponding
1444 	 * entries in arm64_ftr_regs[].
1445 	 */
1446 
1447 	/* AArch64 mappings of the AArch32 ID registers */
1448 	/* CRm=1 */
1449 	ID_SANITISED(ID_PFR0_EL1),
1450 	ID_SANITISED(ID_PFR1_EL1),
1451 	ID_SANITISED(ID_DFR0_EL1),
1452 	ID_HIDDEN(ID_AFR0_EL1),
1453 	ID_SANITISED(ID_MMFR0_EL1),
1454 	ID_SANITISED(ID_MMFR1_EL1),
1455 	ID_SANITISED(ID_MMFR2_EL1),
1456 	ID_SANITISED(ID_MMFR3_EL1),
1457 
1458 	/* CRm=2 */
1459 	ID_SANITISED(ID_ISAR0_EL1),
1460 	ID_SANITISED(ID_ISAR1_EL1),
1461 	ID_SANITISED(ID_ISAR2_EL1),
1462 	ID_SANITISED(ID_ISAR3_EL1),
1463 	ID_SANITISED(ID_ISAR4_EL1),
1464 	ID_SANITISED(ID_ISAR5_EL1),
1465 	ID_SANITISED(ID_MMFR4_EL1),
1466 	ID_SANITISED(ID_ISAR6_EL1),
1467 
1468 	/* CRm=3 */
1469 	ID_SANITISED(MVFR0_EL1),
1470 	ID_SANITISED(MVFR1_EL1),
1471 	ID_SANITISED(MVFR2_EL1),
1472 	ID_UNALLOCATED(3,3),
1473 	ID_SANITISED(ID_PFR2_EL1),
1474 	ID_HIDDEN(ID_DFR1_EL1),
1475 	ID_SANITISED(ID_MMFR5_EL1),
1476 	ID_UNALLOCATED(3,7),
1477 
1478 	/* AArch64 ID registers */
1479 	/* CRm=4 */
1480 	ID_SANITISED(ID_AA64PFR0_EL1),
1481 	ID_SANITISED(ID_AA64PFR1_EL1),
1482 	ID_UNALLOCATED(4,2),
1483 	ID_UNALLOCATED(4,3),
1484 	{ 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 },
1485 	ID_UNALLOCATED(4,5),
1486 	ID_UNALLOCATED(4,6),
1487 	ID_UNALLOCATED(4,7),
1488 
1489 	/* CRm=5 */
1490 	ID_SANITISED(ID_AA64DFR0_EL1),
1491 	ID_SANITISED(ID_AA64DFR1_EL1),
1492 	ID_UNALLOCATED(5,2),
1493 	ID_UNALLOCATED(5,3),
1494 	ID_HIDDEN(ID_AA64AFR0_EL1),
1495 	ID_HIDDEN(ID_AA64AFR1_EL1),
1496 	ID_UNALLOCATED(5,6),
1497 	ID_UNALLOCATED(5,7),
1498 
1499 	/* CRm=6 */
1500 	ID_SANITISED(ID_AA64ISAR0_EL1),
1501 	ID_SANITISED(ID_AA64ISAR1_EL1),
1502 	ID_UNALLOCATED(6,2),
1503 	ID_UNALLOCATED(6,3),
1504 	ID_UNALLOCATED(6,4),
1505 	ID_UNALLOCATED(6,5),
1506 	ID_UNALLOCATED(6,6),
1507 	ID_UNALLOCATED(6,7),
1508 
1509 	/* CRm=7 */
1510 	ID_SANITISED(ID_AA64MMFR0_EL1),
1511 	ID_SANITISED(ID_AA64MMFR1_EL1),
1512 	ID_SANITISED(ID_AA64MMFR2_EL1),
1513 	ID_UNALLOCATED(7,3),
1514 	ID_UNALLOCATED(7,4),
1515 	ID_UNALLOCATED(7,5),
1516 	ID_UNALLOCATED(7,6),
1517 	ID_UNALLOCATED(7,7),
1518 
1519 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
1520 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
1521 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
1522 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
1523 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
1524 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
1525 
1526 	PTRAUTH_KEY(APIA),
1527 	PTRAUTH_KEY(APIB),
1528 	PTRAUTH_KEY(APDA),
1529 	PTRAUTH_KEY(APDB),
1530 	PTRAUTH_KEY(APGA),
1531 
1532 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
1533 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
1534 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
1535 
1536 	{ SYS_DESC(SYS_ERRIDR_EL1), trap_raz_wi },
1537 	{ SYS_DESC(SYS_ERRSELR_EL1), trap_raz_wi },
1538 	{ SYS_DESC(SYS_ERXFR_EL1), trap_raz_wi },
1539 	{ SYS_DESC(SYS_ERXCTLR_EL1), trap_raz_wi },
1540 	{ SYS_DESC(SYS_ERXSTATUS_EL1), trap_raz_wi },
1541 	{ SYS_DESC(SYS_ERXADDR_EL1), trap_raz_wi },
1542 	{ SYS_DESC(SYS_ERXMISC0_EL1), trap_raz_wi },
1543 	{ SYS_DESC(SYS_ERXMISC1_EL1), trap_raz_wi },
1544 
1545 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
1546 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
1547 
1548 	{ SYS_DESC(SYS_PMINTENSET_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1549 	{ SYS_DESC(SYS_PMINTENCLR_EL1), access_pminten, reset_unknown, PMINTENSET_EL1 },
1550 
1551 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
1552 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
1553 
1554 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
1555 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
1556 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
1557 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
1558 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
1559 
1560 	{ SYS_DESC(SYS_VBAR_EL1), NULL, reset_val, VBAR_EL1, 0 },
1561 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
1562 
1563 	{ SYS_DESC(SYS_ICC_IAR0_EL1), write_to_read_only },
1564 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), read_from_write_only },
1565 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), write_to_read_only },
1566 	{ SYS_DESC(SYS_ICC_DIR_EL1), read_from_write_only },
1567 	{ SYS_DESC(SYS_ICC_RPR_EL1), write_to_read_only },
1568 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
1569 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
1570 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
1571 	{ SYS_DESC(SYS_ICC_IAR1_EL1), write_to_read_only },
1572 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), read_from_write_only },
1573 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), write_to_read_only },
1574 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
1575 
1576 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
1577 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
1578 
1579 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
1580 
1581 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
1582 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr },
1583 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
1584 	{ SYS_DESC(SYS_CTR_EL0), access_ctr },
1585 
1586 	{ SYS_DESC(SYS_PMCR_EL0), access_pmcr, reset_pmcr, PMCR_EL0 },
1587 	{ SYS_DESC(SYS_PMCNTENSET_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1588 	{ SYS_DESC(SYS_PMCNTENCLR_EL0), access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
1589 	{ SYS_DESC(SYS_PMOVSCLR_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1590 	{ SYS_DESC(SYS_PMSWINC_EL0), access_pmswinc, reset_unknown, PMSWINC_EL0 },
1591 	{ SYS_DESC(SYS_PMSELR_EL0), access_pmselr, reset_unknown, PMSELR_EL0 },
1592 	{ SYS_DESC(SYS_PMCEID0_EL0), access_pmceid },
1593 	{ SYS_DESC(SYS_PMCEID1_EL0), access_pmceid },
1594 	{ SYS_DESC(SYS_PMCCNTR_EL0), access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1595 	{ SYS_DESC(SYS_PMXEVTYPER_EL0), access_pmu_evtyper },
1596 	{ SYS_DESC(SYS_PMXEVCNTR_EL0), access_pmu_evcntr },
1597 	/*
1598 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
1599 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1600 	 */
1601 	{ SYS_DESC(SYS_PMUSERENR_EL0), access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1602 	{ SYS_DESC(SYS_PMOVSSET_EL0), access_pmovs, reset_unknown, PMOVSSET_EL0 },
1603 
1604 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
1605 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
1606 
1607 	{ SYS_DESC(SYS_AMCR_EL0), access_amu },
1608 	{ SYS_DESC(SYS_AMCFGR_EL0), access_amu },
1609 	{ SYS_DESC(SYS_AMCGCR_EL0), access_amu },
1610 	{ SYS_DESC(SYS_AMUSERENR_EL0), access_amu },
1611 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), access_amu },
1612 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), access_amu },
1613 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), access_amu },
1614 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), access_amu },
1615 	AMU_AMEVCNTR0_EL0(0),
1616 	AMU_AMEVCNTR0_EL0(1),
1617 	AMU_AMEVCNTR0_EL0(2),
1618 	AMU_AMEVCNTR0_EL0(3),
1619 	AMU_AMEVCNTR0_EL0(4),
1620 	AMU_AMEVCNTR0_EL0(5),
1621 	AMU_AMEVCNTR0_EL0(6),
1622 	AMU_AMEVCNTR0_EL0(7),
1623 	AMU_AMEVCNTR0_EL0(8),
1624 	AMU_AMEVCNTR0_EL0(9),
1625 	AMU_AMEVCNTR0_EL0(10),
1626 	AMU_AMEVCNTR0_EL0(11),
1627 	AMU_AMEVCNTR0_EL0(12),
1628 	AMU_AMEVCNTR0_EL0(13),
1629 	AMU_AMEVCNTR0_EL0(14),
1630 	AMU_AMEVCNTR0_EL0(15),
1631 	AMU_AMEVTYPE0_EL0(0),
1632 	AMU_AMEVTYPE0_EL0(1),
1633 	AMU_AMEVTYPE0_EL0(2),
1634 	AMU_AMEVTYPE0_EL0(3),
1635 	AMU_AMEVTYPE0_EL0(4),
1636 	AMU_AMEVTYPE0_EL0(5),
1637 	AMU_AMEVTYPE0_EL0(6),
1638 	AMU_AMEVTYPE0_EL0(7),
1639 	AMU_AMEVTYPE0_EL0(8),
1640 	AMU_AMEVTYPE0_EL0(9),
1641 	AMU_AMEVTYPE0_EL0(10),
1642 	AMU_AMEVTYPE0_EL0(11),
1643 	AMU_AMEVTYPE0_EL0(12),
1644 	AMU_AMEVTYPE0_EL0(13),
1645 	AMU_AMEVTYPE0_EL0(14),
1646 	AMU_AMEVTYPE0_EL0(15),
1647 	AMU_AMEVCNTR1_EL0(0),
1648 	AMU_AMEVCNTR1_EL0(1),
1649 	AMU_AMEVCNTR1_EL0(2),
1650 	AMU_AMEVCNTR1_EL0(3),
1651 	AMU_AMEVCNTR1_EL0(4),
1652 	AMU_AMEVCNTR1_EL0(5),
1653 	AMU_AMEVCNTR1_EL0(6),
1654 	AMU_AMEVCNTR1_EL0(7),
1655 	AMU_AMEVCNTR1_EL0(8),
1656 	AMU_AMEVCNTR1_EL0(9),
1657 	AMU_AMEVCNTR1_EL0(10),
1658 	AMU_AMEVCNTR1_EL0(11),
1659 	AMU_AMEVCNTR1_EL0(12),
1660 	AMU_AMEVCNTR1_EL0(13),
1661 	AMU_AMEVCNTR1_EL0(14),
1662 	AMU_AMEVCNTR1_EL0(15),
1663 	AMU_AMEVTYPE1_EL0(0),
1664 	AMU_AMEVTYPE1_EL0(1),
1665 	AMU_AMEVTYPE1_EL0(2),
1666 	AMU_AMEVTYPE1_EL0(3),
1667 	AMU_AMEVTYPE1_EL0(4),
1668 	AMU_AMEVTYPE1_EL0(5),
1669 	AMU_AMEVTYPE1_EL0(6),
1670 	AMU_AMEVTYPE1_EL0(7),
1671 	AMU_AMEVTYPE1_EL0(8),
1672 	AMU_AMEVTYPE1_EL0(9),
1673 	AMU_AMEVTYPE1_EL0(10),
1674 	AMU_AMEVTYPE1_EL0(11),
1675 	AMU_AMEVTYPE1_EL0(12),
1676 	AMU_AMEVTYPE1_EL0(13),
1677 	AMU_AMEVTYPE1_EL0(14),
1678 	AMU_AMEVTYPE1_EL0(15),
1679 
1680 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
1681 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
1682 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
1683 
1684 	/* PMEVCNTRn_EL0 */
1685 	PMU_PMEVCNTR_EL0(0),
1686 	PMU_PMEVCNTR_EL0(1),
1687 	PMU_PMEVCNTR_EL0(2),
1688 	PMU_PMEVCNTR_EL0(3),
1689 	PMU_PMEVCNTR_EL0(4),
1690 	PMU_PMEVCNTR_EL0(5),
1691 	PMU_PMEVCNTR_EL0(6),
1692 	PMU_PMEVCNTR_EL0(7),
1693 	PMU_PMEVCNTR_EL0(8),
1694 	PMU_PMEVCNTR_EL0(9),
1695 	PMU_PMEVCNTR_EL0(10),
1696 	PMU_PMEVCNTR_EL0(11),
1697 	PMU_PMEVCNTR_EL0(12),
1698 	PMU_PMEVCNTR_EL0(13),
1699 	PMU_PMEVCNTR_EL0(14),
1700 	PMU_PMEVCNTR_EL0(15),
1701 	PMU_PMEVCNTR_EL0(16),
1702 	PMU_PMEVCNTR_EL0(17),
1703 	PMU_PMEVCNTR_EL0(18),
1704 	PMU_PMEVCNTR_EL0(19),
1705 	PMU_PMEVCNTR_EL0(20),
1706 	PMU_PMEVCNTR_EL0(21),
1707 	PMU_PMEVCNTR_EL0(22),
1708 	PMU_PMEVCNTR_EL0(23),
1709 	PMU_PMEVCNTR_EL0(24),
1710 	PMU_PMEVCNTR_EL0(25),
1711 	PMU_PMEVCNTR_EL0(26),
1712 	PMU_PMEVCNTR_EL0(27),
1713 	PMU_PMEVCNTR_EL0(28),
1714 	PMU_PMEVCNTR_EL0(29),
1715 	PMU_PMEVCNTR_EL0(30),
1716 	/* PMEVTYPERn_EL0 */
1717 	PMU_PMEVTYPER_EL0(0),
1718 	PMU_PMEVTYPER_EL0(1),
1719 	PMU_PMEVTYPER_EL0(2),
1720 	PMU_PMEVTYPER_EL0(3),
1721 	PMU_PMEVTYPER_EL0(4),
1722 	PMU_PMEVTYPER_EL0(5),
1723 	PMU_PMEVTYPER_EL0(6),
1724 	PMU_PMEVTYPER_EL0(7),
1725 	PMU_PMEVTYPER_EL0(8),
1726 	PMU_PMEVTYPER_EL0(9),
1727 	PMU_PMEVTYPER_EL0(10),
1728 	PMU_PMEVTYPER_EL0(11),
1729 	PMU_PMEVTYPER_EL0(12),
1730 	PMU_PMEVTYPER_EL0(13),
1731 	PMU_PMEVTYPER_EL0(14),
1732 	PMU_PMEVTYPER_EL0(15),
1733 	PMU_PMEVTYPER_EL0(16),
1734 	PMU_PMEVTYPER_EL0(17),
1735 	PMU_PMEVTYPER_EL0(18),
1736 	PMU_PMEVTYPER_EL0(19),
1737 	PMU_PMEVTYPER_EL0(20),
1738 	PMU_PMEVTYPER_EL0(21),
1739 	PMU_PMEVTYPER_EL0(22),
1740 	PMU_PMEVTYPER_EL0(23),
1741 	PMU_PMEVTYPER_EL0(24),
1742 	PMU_PMEVTYPER_EL0(25),
1743 	PMU_PMEVTYPER_EL0(26),
1744 	PMU_PMEVTYPER_EL0(27),
1745 	PMU_PMEVTYPER_EL0(28),
1746 	PMU_PMEVTYPER_EL0(29),
1747 	PMU_PMEVTYPER_EL0(30),
1748 	/*
1749 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
1750 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1751 	 */
1752 	{ SYS_DESC(SYS_PMCCFILTR_EL0), access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1753 
1754 	{ SYS_DESC(SYS_DACR32_EL2), NULL, reset_unknown, DACR32_EL2 },
1755 	{ SYS_DESC(SYS_IFSR32_EL2), NULL, reset_unknown, IFSR32_EL2 },
1756 	{ SYS_DESC(SYS_FPEXC32_EL2), NULL, reset_val, FPEXC32_EL2, 0x700 },
1757 };
1758 
1759 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1760 			struct sys_reg_params *p,
1761 			const struct sys_reg_desc *r)
1762 {
1763 	if (p->is_write) {
1764 		return ignore_write(vcpu, p);
1765 	} else {
1766 		u64 dfr = read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1);
1767 		u64 pfr = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1768 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1769 
1770 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1771 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1772 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1773 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1774 		return true;
1775 	}
1776 }
1777 
1778 static bool trap_debug32(struct kvm_vcpu *vcpu,
1779 			 struct sys_reg_params *p,
1780 			 const struct sys_reg_desc *r)
1781 {
1782 	if (p->is_write) {
1783 		vcpu_cp14(vcpu, r->reg) = p->regval;
1784 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1785 	} else {
1786 		p->regval = vcpu_cp14(vcpu, r->reg);
1787 	}
1788 
1789 	return true;
1790 }
1791 
1792 /* AArch32 debug register mappings
1793  *
1794  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1795  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1796  *
1797  * All control registers and watchpoint value registers are mapped to
1798  * the lower 32 bits of their AArch64 equivalents. We share the trap
1799  * handlers with the above AArch64 code which checks what mode the
1800  * system is in.
1801  */
1802 
1803 static bool trap_xvr(struct kvm_vcpu *vcpu,
1804 		     struct sys_reg_params *p,
1805 		     const struct sys_reg_desc *rd)
1806 {
1807 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1808 
1809 	if (p->is_write) {
1810 		u64 val = *dbg_reg;
1811 
1812 		val &= 0xffffffffUL;
1813 		val |= p->regval << 32;
1814 		*dbg_reg = val;
1815 
1816 		vcpu->arch.flags |= KVM_ARM64_DEBUG_DIRTY;
1817 	} else {
1818 		p->regval = *dbg_reg >> 32;
1819 	}
1820 
1821 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1822 
1823 	return true;
1824 }
1825 
1826 #define DBG_BCR_BVR_WCR_WVR(n)						\
1827 	/* DBGBVRn */							\
1828 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1829 	/* DBGBCRn */							\
1830 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1831 	/* DBGWVRn */							\
1832 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1833 	/* DBGWCRn */							\
1834 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1835 
1836 #define DBGBXVR(n)							\
1837 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1838 
1839 /*
1840  * Trapped cp14 registers. We generally ignore most of the external
1841  * debug, on the principle that they don't really make sense to a
1842  * guest. Revisit this one day, would this principle change.
1843  */
1844 static const struct sys_reg_desc cp14_regs[] = {
1845 	/* DBGIDR */
1846 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1847 	/* DBGDTRRXext */
1848 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1849 
1850 	DBG_BCR_BVR_WCR_WVR(0),
1851 	/* DBGDSCRint */
1852 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1853 	DBG_BCR_BVR_WCR_WVR(1),
1854 	/* DBGDCCINT */
1855 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1856 	/* DBGDSCRext */
1857 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1858 	DBG_BCR_BVR_WCR_WVR(2),
1859 	/* DBGDTR[RT]Xint */
1860 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1861 	/* DBGDTR[RT]Xext */
1862 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1863 	DBG_BCR_BVR_WCR_WVR(3),
1864 	DBG_BCR_BVR_WCR_WVR(4),
1865 	DBG_BCR_BVR_WCR_WVR(5),
1866 	/* DBGWFAR */
1867 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1868 	/* DBGOSECCR */
1869 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1870 	DBG_BCR_BVR_WCR_WVR(6),
1871 	/* DBGVCR */
1872 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1873 	DBG_BCR_BVR_WCR_WVR(7),
1874 	DBG_BCR_BVR_WCR_WVR(8),
1875 	DBG_BCR_BVR_WCR_WVR(9),
1876 	DBG_BCR_BVR_WCR_WVR(10),
1877 	DBG_BCR_BVR_WCR_WVR(11),
1878 	DBG_BCR_BVR_WCR_WVR(12),
1879 	DBG_BCR_BVR_WCR_WVR(13),
1880 	DBG_BCR_BVR_WCR_WVR(14),
1881 	DBG_BCR_BVR_WCR_WVR(15),
1882 
1883 	/* DBGDRAR (32bit) */
1884 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1885 
1886 	DBGBXVR(0),
1887 	/* DBGOSLAR */
1888 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1889 	DBGBXVR(1),
1890 	/* DBGOSLSR */
1891 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1892 	DBGBXVR(2),
1893 	DBGBXVR(3),
1894 	/* DBGOSDLR */
1895 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1896 	DBGBXVR(4),
1897 	/* DBGPRCR */
1898 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1899 	DBGBXVR(5),
1900 	DBGBXVR(6),
1901 	DBGBXVR(7),
1902 	DBGBXVR(8),
1903 	DBGBXVR(9),
1904 	DBGBXVR(10),
1905 	DBGBXVR(11),
1906 	DBGBXVR(12),
1907 	DBGBXVR(13),
1908 	DBGBXVR(14),
1909 	DBGBXVR(15),
1910 
1911 	/* DBGDSAR (32bit) */
1912 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1913 
1914 	/* DBGDEVID2 */
1915 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1916 	/* DBGDEVID1 */
1917 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1918 	/* DBGDEVID */
1919 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1920 	/* DBGCLAIMSET */
1921 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1922 	/* DBGCLAIMCLR */
1923 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1924 	/* DBGAUTHSTATUS */
1925 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1926 };
1927 
1928 /* Trapped cp14 64bit registers */
1929 static const struct sys_reg_desc cp14_64_regs[] = {
1930 	/* DBGDRAR (64bit) */
1931 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1932 
1933 	/* DBGDSAR (64bit) */
1934 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1935 };
1936 
1937 /* Macro to expand the PMEVCNTRn register */
1938 #define PMU_PMEVCNTR(n)							\
1939 	/* PMEVCNTRn */							\
1940 	{ Op1(0), CRn(0b1110),						\
1941 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1942 	  access_pmu_evcntr }
1943 
1944 /* Macro to expand the PMEVTYPERn register */
1945 #define PMU_PMEVTYPER(n)						\
1946 	/* PMEVTYPERn */						\
1947 	{ Op1(0), CRn(0b1110),						\
1948 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1949 	  access_pmu_evtyper }
1950 
1951 /*
1952  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1953  * depending on the way they are accessed (as a 32bit or a 64bit
1954  * register).
1955  */
1956 static const struct sys_reg_desc cp15_regs[] = {
1957 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
1958 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1959 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1960 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1961 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1962 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1963 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1964 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1965 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1966 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1967 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1968 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1969 
1970 	/*
1971 	 * DC{C,I,CI}SW operations:
1972 	 */
1973 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1974 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1975 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1976 
1977 	/* PMU */
1978 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1979 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1980 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1981 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1982 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1983 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1984 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1985 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1986 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1987 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1988 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1989 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1990 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1991 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1992 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1993 
1994 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
1995 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
1996 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
1997 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
1998 
1999 	/* ICC_SRE */
2000 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
2001 
2002 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
2003 
2004 	/* Arch Tmers */
2005 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
2006 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
2007 
2008 	/* PMEVCNTRn */
2009 	PMU_PMEVCNTR(0),
2010 	PMU_PMEVCNTR(1),
2011 	PMU_PMEVCNTR(2),
2012 	PMU_PMEVCNTR(3),
2013 	PMU_PMEVCNTR(4),
2014 	PMU_PMEVCNTR(5),
2015 	PMU_PMEVCNTR(6),
2016 	PMU_PMEVCNTR(7),
2017 	PMU_PMEVCNTR(8),
2018 	PMU_PMEVCNTR(9),
2019 	PMU_PMEVCNTR(10),
2020 	PMU_PMEVCNTR(11),
2021 	PMU_PMEVCNTR(12),
2022 	PMU_PMEVCNTR(13),
2023 	PMU_PMEVCNTR(14),
2024 	PMU_PMEVCNTR(15),
2025 	PMU_PMEVCNTR(16),
2026 	PMU_PMEVCNTR(17),
2027 	PMU_PMEVCNTR(18),
2028 	PMU_PMEVCNTR(19),
2029 	PMU_PMEVCNTR(20),
2030 	PMU_PMEVCNTR(21),
2031 	PMU_PMEVCNTR(22),
2032 	PMU_PMEVCNTR(23),
2033 	PMU_PMEVCNTR(24),
2034 	PMU_PMEVCNTR(25),
2035 	PMU_PMEVCNTR(26),
2036 	PMU_PMEVCNTR(27),
2037 	PMU_PMEVCNTR(28),
2038 	PMU_PMEVCNTR(29),
2039 	PMU_PMEVCNTR(30),
2040 	/* PMEVTYPERn */
2041 	PMU_PMEVTYPER(0),
2042 	PMU_PMEVTYPER(1),
2043 	PMU_PMEVTYPER(2),
2044 	PMU_PMEVTYPER(3),
2045 	PMU_PMEVTYPER(4),
2046 	PMU_PMEVTYPER(5),
2047 	PMU_PMEVTYPER(6),
2048 	PMU_PMEVTYPER(7),
2049 	PMU_PMEVTYPER(8),
2050 	PMU_PMEVTYPER(9),
2051 	PMU_PMEVTYPER(10),
2052 	PMU_PMEVTYPER(11),
2053 	PMU_PMEVTYPER(12),
2054 	PMU_PMEVTYPER(13),
2055 	PMU_PMEVTYPER(14),
2056 	PMU_PMEVTYPER(15),
2057 	PMU_PMEVTYPER(16),
2058 	PMU_PMEVTYPER(17),
2059 	PMU_PMEVTYPER(18),
2060 	PMU_PMEVTYPER(19),
2061 	PMU_PMEVTYPER(20),
2062 	PMU_PMEVTYPER(21),
2063 	PMU_PMEVTYPER(22),
2064 	PMU_PMEVTYPER(23),
2065 	PMU_PMEVTYPER(24),
2066 	PMU_PMEVTYPER(25),
2067 	PMU_PMEVTYPER(26),
2068 	PMU_PMEVTYPER(27),
2069 	PMU_PMEVTYPER(28),
2070 	PMU_PMEVTYPER(29),
2071 	PMU_PMEVTYPER(30),
2072 	/* PMCCFILTR */
2073 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
2074 
2075 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
2076 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
2077 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, c0_CSSELR },
2078 };
2079 
2080 static const struct sys_reg_desc cp15_64_regs[] = {
2081 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
2082 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
2083 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
2084 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
2085 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
2086 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
2087 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
2088 };
2089 
2090 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
2091 			      bool is_32)
2092 {
2093 	unsigned int i;
2094 
2095 	for (i = 0; i < n; i++) {
2096 		if (!is_32 && table[i].reg && !table[i].reset) {
2097 			kvm_err("sys_reg table %p entry %d has lacks reset\n",
2098 				table, i);
2099 			return 1;
2100 		}
2101 
2102 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2103 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2104 			return 1;
2105 		}
2106 	}
2107 
2108 	return 0;
2109 }
2110 
2111 /* Target specific emulation tables */
2112 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
2113 
2114 void kvm_register_target_sys_reg_table(unsigned int target,
2115 				       struct kvm_sys_reg_target_table *table)
2116 {
2117 	if (check_sysreg_table(table->table64.table, table->table64.num, false) ||
2118 	    check_sysreg_table(table->table32.table, table->table32.num, true))
2119 		return;
2120 
2121 	target_tables[target] = table;
2122 }
2123 
2124 /* Get specific register table for this target. */
2125 static const struct sys_reg_desc *get_target_table(unsigned target,
2126 						   bool mode_is_64,
2127 						   size_t *num)
2128 {
2129 	struct kvm_sys_reg_target_table *table;
2130 
2131 	table = target_tables[target];
2132 	if (mode_is_64) {
2133 		*num = table->table64.num;
2134 		return table->table64.table;
2135 	} else {
2136 		*num = table->table32.num;
2137 		return table->table32.table;
2138 	}
2139 }
2140 
2141 static int match_sys_reg(const void *key, const void *elt)
2142 {
2143 	const unsigned long pval = (unsigned long)key;
2144 	const struct sys_reg_desc *r = elt;
2145 
2146 	return pval - reg_to_encoding(r);
2147 }
2148 
2149 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
2150 					 const struct sys_reg_desc table[],
2151 					 unsigned int num)
2152 {
2153 	unsigned long pval = reg_to_encoding(params);
2154 
2155 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
2156 }
2157 
2158 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
2159 {
2160 	kvm_inject_undefined(vcpu);
2161 	return 1;
2162 }
2163 
2164 static void perform_access(struct kvm_vcpu *vcpu,
2165 			   struct sys_reg_params *params,
2166 			   const struct sys_reg_desc *r)
2167 {
2168 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
2169 
2170 	/* Check for regs disabled by runtime config */
2171 	if (sysreg_hidden_from_guest(vcpu, r)) {
2172 		kvm_inject_undefined(vcpu);
2173 		return;
2174 	}
2175 
2176 	/*
2177 	 * Not having an accessor means that we have configured a trap
2178 	 * that we don't know how to handle. This certainly qualifies
2179 	 * as a gross bug that should be fixed right away.
2180 	 */
2181 	BUG_ON(!r->access);
2182 
2183 	/* Skip instruction if instructed so */
2184 	if (likely(r->access(vcpu, params, r)))
2185 		kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
2186 }
2187 
2188 /*
2189  * emulate_cp --  tries to match a sys_reg access in a handling table, and
2190  *                call the corresponding trap handler.
2191  *
2192  * @params: pointer to the descriptor of the access
2193  * @table: array of trap descriptors
2194  * @num: size of the trap descriptor array
2195  *
2196  * Return 0 if the access has been handled, and -1 if not.
2197  */
2198 static int emulate_cp(struct kvm_vcpu *vcpu,
2199 		      struct sys_reg_params *params,
2200 		      const struct sys_reg_desc *table,
2201 		      size_t num)
2202 {
2203 	const struct sys_reg_desc *r;
2204 
2205 	if (!table)
2206 		return -1;	/* Not handled */
2207 
2208 	r = find_reg(params, table, num);
2209 
2210 	if (r) {
2211 		perform_access(vcpu, params, r);
2212 		return 0;
2213 	}
2214 
2215 	/* Not handled */
2216 	return -1;
2217 }
2218 
2219 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
2220 				struct sys_reg_params *params)
2221 {
2222 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
2223 	int cp = -1;
2224 
2225 	switch(hsr_ec) {
2226 	case ESR_ELx_EC_CP15_32:
2227 	case ESR_ELx_EC_CP15_64:
2228 		cp = 15;
2229 		break;
2230 	case ESR_ELx_EC_CP14_MR:
2231 	case ESR_ELx_EC_CP14_64:
2232 		cp = 14;
2233 		break;
2234 	default:
2235 		WARN_ON(1);
2236 	}
2237 
2238 	print_sys_reg_msg(params,
2239 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
2240 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2241 	kvm_inject_undefined(vcpu);
2242 }
2243 
2244 /**
2245  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
2246  * @vcpu: The VCPU pointer
2247  * @run:  The kvm_run struct
2248  */
2249 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
2250 			    const struct sys_reg_desc *global,
2251 			    size_t nr_global,
2252 			    const struct sys_reg_desc *target_specific,
2253 			    size_t nr_specific)
2254 {
2255 	struct sys_reg_params params;
2256 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
2257 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2258 	int Rt2 = (hsr >> 10) & 0x1f;
2259 
2260 	params.is_aarch32 = true;
2261 	params.is_32bit = false;
2262 	params.CRm = (hsr >> 1) & 0xf;
2263 	params.is_write = ((hsr & 1) == 0);
2264 
2265 	params.Op0 = 0;
2266 	params.Op1 = (hsr >> 16) & 0xf;
2267 	params.Op2 = 0;
2268 	params.CRn = 0;
2269 
2270 	/*
2271 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
2272 	 * backends between AArch32 and AArch64, we get away with it.
2273 	 */
2274 	if (params.is_write) {
2275 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
2276 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
2277 	}
2278 
2279 	/*
2280 	 * Try to emulate the coprocessor access using the target
2281 	 * specific table first, and using the global table afterwards.
2282 	 * If either of the tables contains a handler, handle the
2283 	 * potential register operation in the case of a read and return
2284 	 * with success.
2285 	 */
2286 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
2287 	    !emulate_cp(vcpu, &params, global, nr_global)) {
2288 		/* Split up the value between registers for the read side */
2289 		if (!params.is_write) {
2290 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
2291 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
2292 		}
2293 
2294 		return 1;
2295 	}
2296 
2297 	unhandled_cp_access(vcpu, &params);
2298 	return 1;
2299 }
2300 
2301 /**
2302  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
2303  * @vcpu: The VCPU pointer
2304  * @run:  The kvm_run struct
2305  */
2306 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
2307 			    const struct sys_reg_desc *global,
2308 			    size_t nr_global,
2309 			    const struct sys_reg_desc *target_specific,
2310 			    size_t nr_specific)
2311 {
2312 	struct sys_reg_params params;
2313 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
2314 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
2315 
2316 	params.is_aarch32 = true;
2317 	params.is_32bit = true;
2318 	params.CRm = (hsr >> 1) & 0xf;
2319 	params.regval = vcpu_get_reg(vcpu, Rt);
2320 	params.is_write = ((hsr & 1) == 0);
2321 	params.CRn = (hsr >> 10) & 0xf;
2322 	params.Op0 = 0;
2323 	params.Op1 = (hsr >> 14) & 0x7;
2324 	params.Op2 = (hsr >> 17) & 0x7;
2325 
2326 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
2327 	    !emulate_cp(vcpu, &params, global, nr_global)) {
2328 		if (!params.is_write)
2329 			vcpu_set_reg(vcpu, Rt, params.regval);
2330 		return 1;
2331 	}
2332 
2333 	unhandled_cp_access(vcpu, &params);
2334 	return 1;
2335 }
2336 
2337 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
2338 {
2339 	const struct sys_reg_desc *target_specific;
2340 	size_t num;
2341 
2342 	target_specific = get_target_table(vcpu->arch.target, false, &num);
2343 	return kvm_handle_cp_64(vcpu,
2344 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
2345 				target_specific, num);
2346 }
2347 
2348 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
2349 {
2350 	const struct sys_reg_desc *target_specific;
2351 	size_t num;
2352 
2353 	target_specific = get_target_table(vcpu->arch.target, false, &num);
2354 	return kvm_handle_cp_32(vcpu,
2355 				cp15_regs, ARRAY_SIZE(cp15_regs),
2356 				target_specific, num);
2357 }
2358 
2359 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
2360 {
2361 	return kvm_handle_cp_64(vcpu,
2362 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
2363 				NULL, 0);
2364 }
2365 
2366 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
2367 {
2368 	return kvm_handle_cp_32(vcpu,
2369 				cp14_regs, ARRAY_SIZE(cp14_regs),
2370 				NULL, 0);
2371 }
2372 
2373 static bool is_imp_def_sys_reg(struct sys_reg_params *params)
2374 {
2375 	// See ARM DDI 0487E.a, section D12.3.2
2376 	return params->Op0 == 3 && (params->CRn & 0b1011) == 0b1011;
2377 }
2378 
2379 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
2380 			   struct sys_reg_params *params)
2381 {
2382 	size_t num;
2383 	const struct sys_reg_desc *table, *r;
2384 
2385 	table = get_target_table(vcpu->arch.target, true, &num);
2386 
2387 	/* Search target-specific then generic table. */
2388 	r = find_reg(params, table, num);
2389 	if (!r)
2390 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2391 
2392 	if (likely(r)) {
2393 		perform_access(vcpu, params, r);
2394 	} else if (is_imp_def_sys_reg(params)) {
2395 		kvm_inject_undefined(vcpu);
2396 	} else {
2397 		print_sys_reg_msg(params,
2398 				  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
2399 				  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
2400 		kvm_inject_undefined(vcpu);
2401 	}
2402 	return 1;
2403 }
2404 
2405 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
2406 				const struct sys_reg_desc *table, size_t num)
2407 {
2408 	unsigned long i;
2409 
2410 	for (i = 0; i < num; i++)
2411 		if (table[i].reset)
2412 			table[i].reset(vcpu, &table[i]);
2413 }
2414 
2415 /**
2416  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
2417  * @vcpu: The VCPU pointer
2418  * @run:  The kvm_run struct
2419  */
2420 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
2421 {
2422 	struct sys_reg_params params;
2423 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
2424 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
2425 	int ret;
2426 
2427 	trace_kvm_handle_sys_reg(esr);
2428 
2429 	params.is_aarch32 = false;
2430 	params.is_32bit = false;
2431 	params.Op0 = (esr >> 20) & 3;
2432 	params.Op1 = (esr >> 14) & 0x7;
2433 	params.CRn = (esr >> 10) & 0xf;
2434 	params.CRm = (esr >> 1) & 0xf;
2435 	params.Op2 = (esr >> 17) & 0x7;
2436 	params.regval = vcpu_get_reg(vcpu, Rt);
2437 	params.is_write = !(esr & 1);
2438 
2439 	ret = emulate_sys_reg(vcpu, &params);
2440 
2441 	if (!params.is_write)
2442 		vcpu_set_reg(vcpu, Rt, params.regval);
2443 	return ret;
2444 }
2445 
2446 /******************************************************************************
2447  * Userspace API
2448  *****************************************************************************/
2449 
2450 static bool index_to_params(u64 id, struct sys_reg_params *params)
2451 {
2452 	switch (id & KVM_REG_SIZE_MASK) {
2453 	case KVM_REG_SIZE_U64:
2454 		/* Any unused index bits means it's not valid. */
2455 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
2456 			      | KVM_REG_ARM_COPROC_MASK
2457 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
2458 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
2459 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
2460 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
2461 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
2462 			return false;
2463 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
2464 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
2465 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
2466 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
2467 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
2468 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
2469 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
2470 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
2471 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
2472 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
2473 		return true;
2474 	default:
2475 		return false;
2476 	}
2477 }
2478 
2479 const struct sys_reg_desc *find_reg_by_id(u64 id,
2480 					  struct sys_reg_params *params,
2481 					  const struct sys_reg_desc table[],
2482 					  unsigned int num)
2483 {
2484 	if (!index_to_params(id, params))
2485 		return NULL;
2486 
2487 	return find_reg(params, table, num);
2488 }
2489 
2490 /* Decode an index value, and find the sys_reg_desc entry. */
2491 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
2492 						    u64 id)
2493 {
2494 	size_t num;
2495 	const struct sys_reg_desc *table, *r;
2496 	struct sys_reg_params params;
2497 
2498 	/* We only do sys_reg for now. */
2499 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
2500 		return NULL;
2501 
2502 	if (!index_to_params(id, &params))
2503 		return NULL;
2504 
2505 	table = get_target_table(vcpu->arch.target, true, &num);
2506 	r = find_reg(&params, table, num);
2507 	if (!r)
2508 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2509 
2510 	/* Not saved in the sys_reg array and not otherwise accessible? */
2511 	if (r && !(r->reg || r->get_user))
2512 		r = NULL;
2513 
2514 	return r;
2515 }
2516 
2517 /*
2518  * These are the invariant sys_reg registers: we let the guest see the
2519  * host versions of these, so they're part of the guest state.
2520  *
2521  * A future CPU may provide a mechanism to present different values to
2522  * the guest, or a future kvm may trap them.
2523  */
2524 
2525 #define FUNCTION_INVARIANT(reg)						\
2526 	static void get_##reg(struct kvm_vcpu *v,			\
2527 			      const struct sys_reg_desc *r)		\
2528 	{								\
2529 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
2530 	}
2531 
2532 FUNCTION_INVARIANT(midr_el1)
2533 FUNCTION_INVARIANT(revidr_el1)
2534 FUNCTION_INVARIANT(clidr_el1)
2535 FUNCTION_INVARIANT(aidr_el1)
2536 
2537 static void get_ctr_el0(struct kvm_vcpu *v, const struct sys_reg_desc *r)
2538 {
2539 	((struct sys_reg_desc *)r)->val = read_sanitised_ftr_reg(SYS_CTR_EL0);
2540 }
2541 
2542 /* ->val is filled in by kvm_sys_reg_table_init() */
2543 static struct sys_reg_desc invariant_sys_regs[] = {
2544 	{ SYS_DESC(SYS_MIDR_EL1), NULL, get_midr_el1 },
2545 	{ SYS_DESC(SYS_REVIDR_EL1), NULL, get_revidr_el1 },
2546 	{ SYS_DESC(SYS_CLIDR_EL1), NULL, get_clidr_el1 },
2547 	{ SYS_DESC(SYS_AIDR_EL1), NULL, get_aidr_el1 },
2548 	{ SYS_DESC(SYS_CTR_EL0), NULL, get_ctr_el0 },
2549 };
2550 
2551 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
2552 {
2553 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
2554 		return -EFAULT;
2555 	return 0;
2556 }
2557 
2558 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
2559 {
2560 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
2561 		return -EFAULT;
2562 	return 0;
2563 }
2564 
2565 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
2566 {
2567 	struct sys_reg_params params;
2568 	const struct sys_reg_desc *r;
2569 
2570 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2571 			   ARRAY_SIZE(invariant_sys_regs));
2572 	if (!r)
2573 		return -ENOENT;
2574 
2575 	return reg_to_user(uaddr, &r->val, id);
2576 }
2577 
2578 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
2579 {
2580 	struct sys_reg_params params;
2581 	const struct sys_reg_desc *r;
2582 	int err;
2583 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
2584 
2585 	r = find_reg_by_id(id, &params, invariant_sys_regs,
2586 			   ARRAY_SIZE(invariant_sys_regs));
2587 	if (!r)
2588 		return -ENOENT;
2589 
2590 	err = reg_from_user(&val, uaddr, id);
2591 	if (err)
2592 		return err;
2593 
2594 	/* This is what we mean by invariant: you can't change it. */
2595 	if (r->val != val)
2596 		return -EINVAL;
2597 
2598 	return 0;
2599 }
2600 
2601 static bool is_valid_cache(u32 val)
2602 {
2603 	u32 level, ctype;
2604 
2605 	if (val >= CSSELR_MAX)
2606 		return false;
2607 
2608 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
2609 	level = (val >> 1);
2610 	ctype = (cache_levels >> (level * 3)) & 7;
2611 
2612 	switch (ctype) {
2613 	case 0: /* No cache */
2614 		return false;
2615 	case 1: /* Instruction cache only */
2616 		return (val & 1);
2617 	case 2: /* Data cache only */
2618 	case 4: /* Unified cache */
2619 		return !(val & 1);
2620 	case 3: /* Separate instruction and data caches */
2621 		return true;
2622 	default: /* Reserved: we can't know instruction or data. */
2623 		return false;
2624 	}
2625 }
2626 
2627 static int demux_c15_get(u64 id, void __user *uaddr)
2628 {
2629 	u32 val;
2630 	u32 __user *uval = uaddr;
2631 
2632 	/* Fail if we have unknown bits set. */
2633 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2634 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2635 		return -ENOENT;
2636 
2637 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2638 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2639 		if (KVM_REG_SIZE(id) != 4)
2640 			return -ENOENT;
2641 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2642 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2643 		if (!is_valid_cache(val))
2644 			return -ENOENT;
2645 
2646 		return put_user(get_ccsidr(val), uval);
2647 	default:
2648 		return -ENOENT;
2649 	}
2650 }
2651 
2652 static int demux_c15_set(u64 id, void __user *uaddr)
2653 {
2654 	u32 val, newval;
2655 	u32 __user *uval = uaddr;
2656 
2657 	/* Fail if we have unknown bits set. */
2658 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2659 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2660 		return -ENOENT;
2661 
2662 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2663 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2664 		if (KVM_REG_SIZE(id) != 4)
2665 			return -ENOENT;
2666 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2667 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2668 		if (!is_valid_cache(val))
2669 			return -ENOENT;
2670 
2671 		if (get_user(newval, uval))
2672 			return -EFAULT;
2673 
2674 		/* This is also invariant: you can't change it. */
2675 		if (newval != get_ccsidr(val))
2676 			return -EINVAL;
2677 		return 0;
2678 	default:
2679 		return -ENOENT;
2680 	}
2681 }
2682 
2683 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2684 {
2685 	const struct sys_reg_desc *r;
2686 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2687 
2688 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2689 		return demux_c15_get(reg->id, uaddr);
2690 
2691 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2692 		return -ENOENT;
2693 
2694 	r = index_to_sys_reg_desc(vcpu, reg->id);
2695 	if (!r)
2696 		return get_invariant_sys_reg(reg->id, uaddr);
2697 
2698 	/* Check for regs disabled by runtime config */
2699 	if (sysreg_hidden_from_user(vcpu, r))
2700 		return -ENOENT;
2701 
2702 	if (r->get_user)
2703 		return (r->get_user)(vcpu, r, reg, uaddr);
2704 
2705 	return reg_to_user(uaddr, &__vcpu_sys_reg(vcpu, r->reg), reg->id);
2706 }
2707 
2708 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2709 {
2710 	const struct sys_reg_desc *r;
2711 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2712 
2713 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2714 		return demux_c15_set(reg->id, uaddr);
2715 
2716 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2717 		return -ENOENT;
2718 
2719 	r = index_to_sys_reg_desc(vcpu, reg->id);
2720 	if (!r)
2721 		return set_invariant_sys_reg(reg->id, uaddr);
2722 
2723 	/* Check for regs disabled by runtime config */
2724 	if (sysreg_hidden_from_user(vcpu, r))
2725 		return -ENOENT;
2726 
2727 	if (r->set_user)
2728 		return (r->set_user)(vcpu, r, reg, uaddr);
2729 
2730 	return reg_from_user(&__vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2731 }
2732 
2733 static unsigned int num_demux_regs(void)
2734 {
2735 	unsigned int i, count = 0;
2736 
2737 	for (i = 0; i < CSSELR_MAX; i++)
2738 		if (is_valid_cache(i))
2739 			count++;
2740 
2741 	return count;
2742 }
2743 
2744 static int write_demux_regids(u64 __user *uindices)
2745 {
2746 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2747 	unsigned int i;
2748 
2749 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2750 	for (i = 0; i < CSSELR_MAX; i++) {
2751 		if (!is_valid_cache(i))
2752 			continue;
2753 		if (put_user(val | i, uindices))
2754 			return -EFAULT;
2755 		uindices++;
2756 	}
2757 	return 0;
2758 }
2759 
2760 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2761 {
2762 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2763 		KVM_REG_ARM64_SYSREG |
2764 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2765 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2766 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2767 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2768 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2769 }
2770 
2771 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2772 {
2773 	if (!*uind)
2774 		return true;
2775 
2776 	if (put_user(sys_reg_to_index(reg), *uind))
2777 		return false;
2778 
2779 	(*uind)++;
2780 	return true;
2781 }
2782 
2783 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
2784 			    const struct sys_reg_desc *rd,
2785 			    u64 __user **uind,
2786 			    unsigned int *total)
2787 {
2788 	/*
2789 	 * Ignore registers we trap but don't save,
2790 	 * and for which no custom user accessor is provided.
2791 	 */
2792 	if (!(rd->reg || rd->get_user))
2793 		return 0;
2794 
2795 	if (sysreg_hidden_from_user(vcpu, rd))
2796 		return 0;
2797 
2798 	if (!copy_reg_to_user(rd, uind))
2799 		return -EFAULT;
2800 
2801 	(*total)++;
2802 	return 0;
2803 }
2804 
2805 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2806 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2807 {
2808 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
2809 	unsigned int total = 0;
2810 	size_t num;
2811 	int err;
2812 
2813 	/* We check for duplicates here, to allow arch-specific overrides. */
2814 	i1 = get_target_table(vcpu->arch.target, true, &num);
2815 	end1 = i1 + num;
2816 	i2 = sys_reg_descs;
2817 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2818 
2819 	BUG_ON(i1 == end1 || i2 == end2);
2820 
2821 	/* Walk carefully, as both tables may refer to the same register. */
2822 	while (i1 || i2) {
2823 		int cmp = cmp_sys_reg(i1, i2);
2824 		/* target-specific overrides generic entry. */
2825 		if (cmp <= 0)
2826 			err = walk_one_sys_reg(vcpu, i1, &uind, &total);
2827 		else
2828 			err = walk_one_sys_reg(vcpu, i2, &uind, &total);
2829 
2830 		if (err)
2831 			return err;
2832 
2833 		if (cmp <= 0 && ++i1 == end1)
2834 			i1 = NULL;
2835 		if (cmp >= 0 && ++i2 == end2)
2836 			i2 = NULL;
2837 	}
2838 	return total;
2839 }
2840 
2841 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2842 {
2843 	return ARRAY_SIZE(invariant_sys_regs)
2844 		+ num_demux_regs()
2845 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2846 }
2847 
2848 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2849 {
2850 	unsigned int i;
2851 	int err;
2852 
2853 	/* Then give them all the invariant registers' indices. */
2854 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2855 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2856 			return -EFAULT;
2857 		uindices++;
2858 	}
2859 
2860 	err = walk_sys_regs(vcpu, uindices);
2861 	if (err < 0)
2862 		return err;
2863 	uindices += err;
2864 
2865 	return write_demux_regids(uindices);
2866 }
2867 
2868 void kvm_sys_reg_table_init(void)
2869 {
2870 	unsigned int i;
2871 	struct sys_reg_desc clidr;
2872 
2873 	/* Make sure tables are unique and in order. */
2874 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), false));
2875 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), true));
2876 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), true));
2877 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), true));
2878 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), true));
2879 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs), false));
2880 
2881 	/* We abuse the reset function to overwrite the table itself. */
2882 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2883 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2884 
2885 	/*
2886 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2887 	 *
2888 	 *   If software reads the Cache Type fields from Ctype1
2889 	 *   upwards, once it has seen a value of 0b000, no caches
2890 	 *   exist at further-out levels of the hierarchy. So, for
2891 	 *   example, if Ctype3 is the first Cache Type field with a
2892 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2893 	 *   ignored.
2894 	 */
2895 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2896 	cache_levels = clidr.val;
2897 	for (i = 0; i < 7; i++)
2898 		if (((cache_levels >> (i*3)) & 7) == 0)
2899 			break;
2900 	/* Clear all higher bits. */
2901 	cache_levels &= (1 << (i*3))-1;
2902 }
2903 
2904 /**
2905  * kvm_reset_sys_regs - sets system registers to reset value
2906  * @vcpu: The VCPU pointer
2907  *
2908  * This function finds the right table above and sets the registers on the
2909  * virtual CPU struct to their architecturally defined reset values.
2910  */
2911 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2912 {
2913 	size_t num;
2914 	const struct sys_reg_desc *table;
2915 
2916 	/* Generic chip reset first (so target could override). */
2917 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2918 
2919 	table = get_target_table(vcpu->arch.target, true, &num);
2920 	reset_sys_reg_descs(vcpu, table, num);
2921 }
2922