xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 8730046c)
1 /*
2  * Copyright (C) 2012,2013 - ARM Ltd
3  * Author: Marc Zyngier <marc.zyngier@arm.com>
4  *
5  * Derived from arch/arm/kvm/coproc.c:
6  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
7  * Authors: Rusty Russell <rusty@rustcorp.com.au>
8  *          Christoffer Dall <c.dall@virtualopensystems.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License, version 2, as
12  * published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License
20  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21  */
22 
23 #include <linux/bsearch.h>
24 #include <linux/kvm_host.h>
25 #include <linux/mm.h>
26 #include <linux/uaccess.h>
27 
28 #include <asm/cacheflush.h>
29 #include <asm/cputype.h>
30 #include <asm/debug-monitors.h>
31 #include <asm/esr.h>
32 #include <asm/kvm_arm.h>
33 #include <asm/kvm_asm.h>
34 #include <asm/kvm_coproc.h>
35 #include <asm/kvm_emulate.h>
36 #include <asm/kvm_host.h>
37 #include <asm/kvm_mmu.h>
38 #include <asm/perf_event.h>
39 #include <asm/sysreg.h>
40 
41 #include <trace/events/kvm.h>
42 
43 #include "sys_regs.h"
44 
45 #include "trace.h"
46 
47 /*
48  * All of this file is extremly similar to the ARM coproc.c, but the
49  * types are different. My gut feeling is that it should be pretty
50  * easy to merge, but that would be an ABI breakage -- again. VFP
51  * would also need to be abstracted.
52  *
53  * For AArch32, we only take care of what is being trapped. Anything
54  * that has to do with init and userspace access has to go via the
55  * 64bit interface.
56  */
57 
58 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
59 static u32 cache_levels;
60 
61 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
62 #define CSSELR_MAX 12
63 
64 /* Which cache CCSIDR represents depends on CSSELR value. */
65 static u32 get_ccsidr(u32 csselr)
66 {
67 	u32 ccsidr;
68 
69 	/* Make sure noone else changes CSSELR during this! */
70 	local_irq_disable();
71 	write_sysreg(csselr, csselr_el1);
72 	isb();
73 	ccsidr = read_sysreg(ccsidr_el1);
74 	local_irq_enable();
75 
76 	return ccsidr;
77 }
78 
79 /*
80  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
81  */
82 static bool access_dcsw(struct kvm_vcpu *vcpu,
83 			struct sys_reg_params *p,
84 			const struct sys_reg_desc *r)
85 {
86 	if (!p->is_write)
87 		return read_from_write_only(vcpu, p);
88 
89 	kvm_set_way_flush(vcpu);
90 	return true;
91 }
92 
93 /*
94  * Generic accessor for VM registers. Only called as long as HCR_TVM
95  * is set. If the guest enables the MMU, we stop trapping the VM
96  * sys_regs and leave it in complete control of the caches.
97  */
98 static bool access_vm_reg(struct kvm_vcpu *vcpu,
99 			  struct sys_reg_params *p,
100 			  const struct sys_reg_desc *r)
101 {
102 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
103 
104 	BUG_ON(!p->is_write);
105 
106 	if (!p->is_aarch32) {
107 		vcpu_sys_reg(vcpu, r->reg) = p->regval;
108 	} else {
109 		if (!p->is_32bit)
110 			vcpu_cp15_64_high(vcpu, r->reg) = upper_32_bits(p->regval);
111 		vcpu_cp15_64_low(vcpu, r->reg) = lower_32_bits(p->regval);
112 	}
113 
114 	kvm_toggle_cache(vcpu, was_enabled);
115 	return true;
116 }
117 
118 /*
119  * Trap handler for the GICv3 SGI generation system register.
120  * Forward the request to the VGIC emulation.
121  * The cp15_64 code makes sure this automatically works
122  * for both AArch64 and AArch32 accesses.
123  */
124 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
125 			   struct sys_reg_params *p,
126 			   const struct sys_reg_desc *r)
127 {
128 	if (!p->is_write)
129 		return read_from_write_only(vcpu, p);
130 
131 	vgic_v3_dispatch_sgi(vcpu, p->regval);
132 
133 	return true;
134 }
135 
136 static bool access_gic_sre(struct kvm_vcpu *vcpu,
137 			   struct sys_reg_params *p,
138 			   const struct sys_reg_desc *r)
139 {
140 	if (p->is_write)
141 		return ignore_write(vcpu, p);
142 
143 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
144 	return true;
145 }
146 
147 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
148 			struct sys_reg_params *p,
149 			const struct sys_reg_desc *r)
150 {
151 	if (p->is_write)
152 		return ignore_write(vcpu, p);
153 	else
154 		return read_zero(vcpu, p);
155 }
156 
157 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
158 			   struct sys_reg_params *p,
159 			   const struct sys_reg_desc *r)
160 {
161 	if (p->is_write) {
162 		return ignore_write(vcpu, p);
163 	} else {
164 		p->regval = (1 << 3);
165 		return true;
166 	}
167 }
168 
169 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
170 				   struct sys_reg_params *p,
171 				   const struct sys_reg_desc *r)
172 {
173 	if (p->is_write) {
174 		return ignore_write(vcpu, p);
175 	} else {
176 		p->regval = read_sysreg(dbgauthstatus_el1);
177 		return true;
178 	}
179 }
180 
181 /*
182  * We want to avoid world-switching all the DBG registers all the
183  * time:
184  *
185  * - If we've touched any debug register, it is likely that we're
186  *   going to touch more of them. It then makes sense to disable the
187  *   traps and start doing the save/restore dance
188  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
189  *   then mandatory to save/restore the registers, as the guest
190  *   depends on them.
191  *
192  * For this, we use a DIRTY bit, indicating the guest has modified the
193  * debug registers, used as follow:
194  *
195  * On guest entry:
196  * - If the dirty bit is set (because we're coming back from trapping),
197  *   disable the traps, save host registers, restore guest registers.
198  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
199  *   set the dirty bit, disable the traps, save host registers,
200  *   restore guest registers.
201  * - Otherwise, enable the traps
202  *
203  * On guest exit:
204  * - If the dirty bit is set, save guest registers, restore host
205  *   registers and clear the dirty bit. This ensure that the host can
206  *   now use the debug registers.
207  */
208 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
209 			    struct sys_reg_params *p,
210 			    const struct sys_reg_desc *r)
211 {
212 	if (p->is_write) {
213 		vcpu_sys_reg(vcpu, r->reg) = p->regval;
214 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
215 	} else {
216 		p->regval = vcpu_sys_reg(vcpu, r->reg);
217 	}
218 
219 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
220 
221 	return true;
222 }
223 
224 /*
225  * reg_to_dbg/dbg_to_reg
226  *
227  * A 32 bit write to a debug register leave top bits alone
228  * A 32 bit read from a debug register only returns the bottom bits
229  *
230  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
231  * hyp.S code switches between host and guest values in future.
232  */
233 static void reg_to_dbg(struct kvm_vcpu *vcpu,
234 		       struct sys_reg_params *p,
235 		       u64 *dbg_reg)
236 {
237 	u64 val = p->regval;
238 
239 	if (p->is_32bit) {
240 		val &= 0xffffffffUL;
241 		val |= ((*dbg_reg >> 32) << 32);
242 	}
243 
244 	*dbg_reg = val;
245 	vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
246 }
247 
248 static void dbg_to_reg(struct kvm_vcpu *vcpu,
249 		       struct sys_reg_params *p,
250 		       u64 *dbg_reg)
251 {
252 	p->regval = *dbg_reg;
253 	if (p->is_32bit)
254 		p->regval &= 0xffffffffUL;
255 }
256 
257 static bool trap_bvr(struct kvm_vcpu *vcpu,
258 		     struct sys_reg_params *p,
259 		     const struct sys_reg_desc *rd)
260 {
261 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
262 
263 	if (p->is_write)
264 		reg_to_dbg(vcpu, p, dbg_reg);
265 	else
266 		dbg_to_reg(vcpu, p, dbg_reg);
267 
268 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
269 
270 	return true;
271 }
272 
273 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
274 		const struct kvm_one_reg *reg, void __user *uaddr)
275 {
276 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
277 
278 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
279 		return -EFAULT;
280 	return 0;
281 }
282 
283 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
284 	const struct kvm_one_reg *reg, void __user *uaddr)
285 {
286 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
287 
288 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
289 		return -EFAULT;
290 	return 0;
291 }
292 
293 static void reset_bvr(struct kvm_vcpu *vcpu,
294 		      const struct sys_reg_desc *rd)
295 {
296 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
297 }
298 
299 static bool trap_bcr(struct kvm_vcpu *vcpu,
300 		     struct sys_reg_params *p,
301 		     const struct sys_reg_desc *rd)
302 {
303 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
304 
305 	if (p->is_write)
306 		reg_to_dbg(vcpu, p, dbg_reg);
307 	else
308 		dbg_to_reg(vcpu, p, dbg_reg);
309 
310 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
311 
312 	return true;
313 }
314 
315 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
316 		const struct kvm_one_reg *reg, void __user *uaddr)
317 {
318 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
319 
320 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
321 		return -EFAULT;
322 
323 	return 0;
324 }
325 
326 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
327 	const struct kvm_one_reg *reg, void __user *uaddr)
328 {
329 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
330 
331 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
332 		return -EFAULT;
333 	return 0;
334 }
335 
336 static void reset_bcr(struct kvm_vcpu *vcpu,
337 		      const struct sys_reg_desc *rd)
338 {
339 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
340 }
341 
342 static bool trap_wvr(struct kvm_vcpu *vcpu,
343 		     struct sys_reg_params *p,
344 		     const struct sys_reg_desc *rd)
345 {
346 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
347 
348 	if (p->is_write)
349 		reg_to_dbg(vcpu, p, dbg_reg);
350 	else
351 		dbg_to_reg(vcpu, p, dbg_reg);
352 
353 	trace_trap_reg(__func__, rd->reg, p->is_write,
354 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
355 
356 	return true;
357 }
358 
359 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
360 		const struct kvm_one_reg *reg, void __user *uaddr)
361 {
362 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
363 
364 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
365 		return -EFAULT;
366 	return 0;
367 }
368 
369 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
370 	const struct kvm_one_reg *reg, void __user *uaddr)
371 {
372 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
373 
374 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
375 		return -EFAULT;
376 	return 0;
377 }
378 
379 static void reset_wvr(struct kvm_vcpu *vcpu,
380 		      const struct sys_reg_desc *rd)
381 {
382 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
383 }
384 
385 static bool trap_wcr(struct kvm_vcpu *vcpu,
386 		     struct sys_reg_params *p,
387 		     const struct sys_reg_desc *rd)
388 {
389 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
390 
391 	if (p->is_write)
392 		reg_to_dbg(vcpu, p, dbg_reg);
393 	else
394 		dbg_to_reg(vcpu, p, dbg_reg);
395 
396 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
397 
398 	return true;
399 }
400 
401 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
402 		const struct kvm_one_reg *reg, void __user *uaddr)
403 {
404 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
405 
406 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
407 		return -EFAULT;
408 	return 0;
409 }
410 
411 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
412 	const struct kvm_one_reg *reg, void __user *uaddr)
413 {
414 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
415 
416 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
417 		return -EFAULT;
418 	return 0;
419 }
420 
421 static void reset_wcr(struct kvm_vcpu *vcpu,
422 		      const struct sys_reg_desc *rd)
423 {
424 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
425 }
426 
427 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
428 {
429 	vcpu_sys_reg(vcpu, AMAIR_EL1) = read_sysreg(amair_el1);
430 }
431 
432 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
433 {
434 	u64 mpidr;
435 
436 	/*
437 	 * Map the vcpu_id into the first three affinity level fields of
438 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
439 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
440 	 * of the GICv3 to be able to address each CPU directly when
441 	 * sending IPIs.
442 	 */
443 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
444 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
445 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
446 	vcpu_sys_reg(vcpu, MPIDR_EL1) = (1ULL << 31) | mpidr;
447 }
448 
449 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
450 {
451 	u64 pmcr, val;
452 
453 	pmcr = read_sysreg(pmcr_el0);
454 	/*
455 	 * Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) are reset to UNKNOWN
456 	 * except PMCR.E resetting to zero.
457 	 */
458 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
459 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
460 	vcpu_sys_reg(vcpu, PMCR_EL0) = val;
461 }
462 
463 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
464 {
465 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
466 
467 	return !((reg & ARMV8_PMU_USERENR_EN) || vcpu_mode_priv(vcpu));
468 }
469 
470 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
471 {
472 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
473 
474 	return !((reg & (ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN))
475 		 || vcpu_mode_priv(vcpu));
476 }
477 
478 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
479 {
480 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
481 
482 	return !((reg & (ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN))
483 		 || vcpu_mode_priv(vcpu));
484 }
485 
486 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
487 {
488 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
489 
490 	return !((reg & (ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN))
491 		 || vcpu_mode_priv(vcpu));
492 }
493 
494 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
495 			const struct sys_reg_desc *r)
496 {
497 	u64 val;
498 
499 	if (!kvm_arm_pmu_v3_ready(vcpu))
500 		return trap_raz_wi(vcpu, p, r);
501 
502 	if (pmu_access_el0_disabled(vcpu))
503 		return false;
504 
505 	if (p->is_write) {
506 		/* Only update writeable bits of PMCR */
507 		val = vcpu_sys_reg(vcpu, PMCR_EL0);
508 		val &= ~ARMV8_PMU_PMCR_MASK;
509 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
510 		vcpu_sys_reg(vcpu, PMCR_EL0) = val;
511 		kvm_pmu_handle_pmcr(vcpu, val);
512 	} else {
513 		/* PMCR.P & PMCR.C are RAZ */
514 		val = vcpu_sys_reg(vcpu, PMCR_EL0)
515 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
516 		p->regval = val;
517 	}
518 
519 	return true;
520 }
521 
522 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
523 			  const struct sys_reg_desc *r)
524 {
525 	if (!kvm_arm_pmu_v3_ready(vcpu))
526 		return trap_raz_wi(vcpu, p, r);
527 
528 	if (pmu_access_event_counter_el0_disabled(vcpu))
529 		return false;
530 
531 	if (p->is_write)
532 		vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
533 	else
534 		/* return PMSELR.SEL field */
535 		p->regval = vcpu_sys_reg(vcpu, PMSELR_EL0)
536 			    & ARMV8_PMU_COUNTER_MASK;
537 
538 	return true;
539 }
540 
541 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
542 			  const struct sys_reg_desc *r)
543 {
544 	u64 pmceid;
545 
546 	if (!kvm_arm_pmu_v3_ready(vcpu))
547 		return trap_raz_wi(vcpu, p, r);
548 
549 	BUG_ON(p->is_write);
550 
551 	if (pmu_access_el0_disabled(vcpu))
552 		return false;
553 
554 	if (!(p->Op2 & 1))
555 		pmceid = read_sysreg(pmceid0_el0);
556 	else
557 		pmceid = read_sysreg(pmceid1_el0);
558 
559 	p->regval = pmceid;
560 
561 	return true;
562 }
563 
564 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
565 {
566 	u64 pmcr, val;
567 
568 	pmcr = vcpu_sys_reg(vcpu, PMCR_EL0);
569 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
570 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX)
571 		return false;
572 
573 	return true;
574 }
575 
576 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
577 			      struct sys_reg_params *p,
578 			      const struct sys_reg_desc *r)
579 {
580 	u64 idx;
581 
582 	if (!kvm_arm_pmu_v3_ready(vcpu))
583 		return trap_raz_wi(vcpu, p, r);
584 
585 	if (r->CRn == 9 && r->CRm == 13) {
586 		if (r->Op2 == 2) {
587 			/* PMXEVCNTR_EL0 */
588 			if (pmu_access_event_counter_el0_disabled(vcpu))
589 				return false;
590 
591 			idx = vcpu_sys_reg(vcpu, PMSELR_EL0)
592 			      & ARMV8_PMU_COUNTER_MASK;
593 		} else if (r->Op2 == 0) {
594 			/* PMCCNTR_EL0 */
595 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
596 				return false;
597 
598 			idx = ARMV8_PMU_CYCLE_IDX;
599 		} else {
600 			return false;
601 		}
602 	} else if (r->CRn == 0 && r->CRm == 9) {
603 		/* PMCCNTR */
604 		if (pmu_access_event_counter_el0_disabled(vcpu))
605 			return false;
606 
607 		idx = ARMV8_PMU_CYCLE_IDX;
608 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
609 		/* PMEVCNTRn_EL0 */
610 		if (pmu_access_event_counter_el0_disabled(vcpu))
611 			return false;
612 
613 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
614 	} else {
615 		return false;
616 	}
617 
618 	if (!pmu_counter_idx_valid(vcpu, idx))
619 		return false;
620 
621 	if (p->is_write) {
622 		if (pmu_access_el0_disabled(vcpu))
623 			return false;
624 
625 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
626 	} else {
627 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
628 	}
629 
630 	return true;
631 }
632 
633 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
634 			       const struct sys_reg_desc *r)
635 {
636 	u64 idx, reg;
637 
638 	if (!kvm_arm_pmu_v3_ready(vcpu))
639 		return trap_raz_wi(vcpu, p, r);
640 
641 	if (pmu_access_el0_disabled(vcpu))
642 		return false;
643 
644 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
645 		/* PMXEVTYPER_EL0 */
646 		idx = vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
647 		reg = PMEVTYPER0_EL0 + idx;
648 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
649 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
650 		if (idx == ARMV8_PMU_CYCLE_IDX)
651 			reg = PMCCFILTR_EL0;
652 		else
653 			/* PMEVTYPERn_EL0 */
654 			reg = PMEVTYPER0_EL0 + idx;
655 	} else {
656 		BUG();
657 	}
658 
659 	if (!pmu_counter_idx_valid(vcpu, idx))
660 		return false;
661 
662 	if (p->is_write) {
663 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
664 		vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
665 	} else {
666 		p->regval = vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
667 	}
668 
669 	return true;
670 }
671 
672 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
673 			   const struct sys_reg_desc *r)
674 {
675 	u64 val, mask;
676 
677 	if (!kvm_arm_pmu_v3_ready(vcpu))
678 		return trap_raz_wi(vcpu, p, r);
679 
680 	if (pmu_access_el0_disabled(vcpu))
681 		return false;
682 
683 	mask = kvm_pmu_valid_counter_mask(vcpu);
684 	if (p->is_write) {
685 		val = p->regval & mask;
686 		if (r->Op2 & 0x1) {
687 			/* accessing PMCNTENSET_EL0 */
688 			vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
689 			kvm_pmu_enable_counter(vcpu, val);
690 		} else {
691 			/* accessing PMCNTENCLR_EL0 */
692 			vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
693 			kvm_pmu_disable_counter(vcpu, val);
694 		}
695 	} else {
696 		p->regval = vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
697 	}
698 
699 	return true;
700 }
701 
702 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
703 			   const struct sys_reg_desc *r)
704 {
705 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
706 
707 	if (!kvm_arm_pmu_v3_ready(vcpu))
708 		return trap_raz_wi(vcpu, p, r);
709 
710 	if (!vcpu_mode_priv(vcpu))
711 		return false;
712 
713 	if (p->is_write) {
714 		u64 val = p->regval & mask;
715 
716 		if (r->Op2 & 0x1)
717 			/* accessing PMINTENSET_EL1 */
718 			vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
719 		else
720 			/* accessing PMINTENCLR_EL1 */
721 			vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
722 	} else {
723 		p->regval = vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
724 	}
725 
726 	return true;
727 }
728 
729 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
730 			 const struct sys_reg_desc *r)
731 {
732 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
733 
734 	if (!kvm_arm_pmu_v3_ready(vcpu))
735 		return trap_raz_wi(vcpu, p, r);
736 
737 	if (pmu_access_el0_disabled(vcpu))
738 		return false;
739 
740 	if (p->is_write) {
741 		if (r->CRm & 0x2)
742 			/* accessing PMOVSSET_EL0 */
743 			kvm_pmu_overflow_set(vcpu, p->regval & mask);
744 		else
745 			/* accessing PMOVSCLR_EL0 */
746 			vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
747 	} else {
748 		p->regval = vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
749 	}
750 
751 	return true;
752 }
753 
754 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
755 			   const struct sys_reg_desc *r)
756 {
757 	u64 mask;
758 
759 	if (!kvm_arm_pmu_v3_ready(vcpu))
760 		return trap_raz_wi(vcpu, p, r);
761 
762 	if (pmu_write_swinc_el0_disabled(vcpu))
763 		return false;
764 
765 	if (p->is_write) {
766 		mask = kvm_pmu_valid_counter_mask(vcpu);
767 		kvm_pmu_software_increment(vcpu, p->regval & mask);
768 		return true;
769 	}
770 
771 	return false;
772 }
773 
774 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
775 			     const struct sys_reg_desc *r)
776 {
777 	if (!kvm_arm_pmu_v3_ready(vcpu))
778 		return trap_raz_wi(vcpu, p, r);
779 
780 	if (p->is_write) {
781 		if (!vcpu_mode_priv(vcpu))
782 			return false;
783 
784 		vcpu_sys_reg(vcpu, PMUSERENR_EL0) = p->regval
785 						    & ARMV8_PMU_USERENR_MASK;
786 	} else {
787 		p->regval = vcpu_sys_reg(vcpu, PMUSERENR_EL0)
788 			    & ARMV8_PMU_USERENR_MASK;
789 	}
790 
791 	return true;
792 }
793 
794 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
795 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
796 	/* DBGBVRn_EL1 */						\
797 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b100),	\
798 	  trap_bvr, reset_bvr, n, 0, get_bvr, set_bvr },		\
799 	/* DBGBCRn_EL1 */						\
800 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b101),	\
801 	  trap_bcr, reset_bcr, n, 0, get_bcr, set_bcr },		\
802 	/* DBGWVRn_EL1 */						\
803 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b110),	\
804 	  trap_wvr, reset_wvr, n, 0,  get_wvr, set_wvr },		\
805 	/* DBGWCRn_EL1 */						\
806 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b111),	\
807 	  trap_wcr, reset_wcr, n, 0,  get_wcr, set_wcr }
808 
809 /* Macro to expand the PMEVCNTRn_EL0 register */
810 #define PMU_PMEVCNTR_EL0(n)						\
811 	/* PMEVCNTRn_EL0 */						\
812 	{ Op0(0b11), Op1(0b011), CRn(0b1110),				\
813 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
814 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
815 
816 /* Macro to expand the PMEVTYPERn_EL0 register */
817 #define PMU_PMEVTYPER_EL0(n)						\
818 	/* PMEVTYPERn_EL0 */						\
819 	{ Op0(0b11), Op1(0b011), CRn(0b1110),				\
820 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
821 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
822 
823 /*
824  * Architected system registers.
825  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
826  *
827  * Debug handling: We do trap most, if not all debug related system
828  * registers. The implementation is good enough to ensure that a guest
829  * can use these with minimal performance degradation. The drawback is
830  * that we don't implement any of the external debug, none of the
831  * OSlock protocol. This should be revisited if we ever encounter a
832  * more demanding guest...
833  */
834 static const struct sys_reg_desc sys_reg_descs[] = {
835 	/* DC ISW */
836 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
837 	  access_dcsw },
838 	/* DC CSW */
839 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
840 	  access_dcsw },
841 	/* DC CISW */
842 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
843 	  access_dcsw },
844 
845 	DBG_BCR_BVR_WCR_WVR_EL1(0),
846 	DBG_BCR_BVR_WCR_WVR_EL1(1),
847 	/* MDCCINT_EL1 */
848 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
849 	  trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
850 	/* MDSCR_EL1 */
851 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
852 	  trap_debug_regs, reset_val, MDSCR_EL1, 0 },
853 	DBG_BCR_BVR_WCR_WVR_EL1(2),
854 	DBG_BCR_BVR_WCR_WVR_EL1(3),
855 	DBG_BCR_BVR_WCR_WVR_EL1(4),
856 	DBG_BCR_BVR_WCR_WVR_EL1(5),
857 	DBG_BCR_BVR_WCR_WVR_EL1(6),
858 	DBG_BCR_BVR_WCR_WVR_EL1(7),
859 	DBG_BCR_BVR_WCR_WVR_EL1(8),
860 	DBG_BCR_BVR_WCR_WVR_EL1(9),
861 	DBG_BCR_BVR_WCR_WVR_EL1(10),
862 	DBG_BCR_BVR_WCR_WVR_EL1(11),
863 	DBG_BCR_BVR_WCR_WVR_EL1(12),
864 	DBG_BCR_BVR_WCR_WVR_EL1(13),
865 	DBG_BCR_BVR_WCR_WVR_EL1(14),
866 	DBG_BCR_BVR_WCR_WVR_EL1(15),
867 
868 	/* MDRAR_EL1 */
869 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
870 	  trap_raz_wi },
871 	/* OSLAR_EL1 */
872 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
873 	  trap_raz_wi },
874 	/* OSLSR_EL1 */
875 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
876 	  trap_oslsr_el1 },
877 	/* OSDLR_EL1 */
878 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
879 	  trap_raz_wi },
880 	/* DBGPRCR_EL1 */
881 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
882 	  trap_raz_wi },
883 	/* DBGCLAIMSET_EL1 */
884 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
885 	  trap_raz_wi },
886 	/* DBGCLAIMCLR_EL1 */
887 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
888 	  trap_raz_wi },
889 	/* DBGAUTHSTATUS_EL1 */
890 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
891 	  trap_dbgauthstatus_el1 },
892 
893 	/* MDCCSR_EL1 */
894 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
895 	  trap_raz_wi },
896 	/* DBGDTR_EL0 */
897 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
898 	  trap_raz_wi },
899 	/* DBGDTR[TR]X_EL0 */
900 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
901 	  trap_raz_wi },
902 
903 	/* DBGVCR32_EL2 */
904 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
905 	  NULL, reset_val, DBGVCR32_EL2, 0 },
906 
907 	/* MPIDR_EL1 */
908 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
909 	  NULL, reset_mpidr, MPIDR_EL1 },
910 	/* SCTLR_EL1 */
911 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
912 	  access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
913 	/* CPACR_EL1 */
914 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
915 	  NULL, reset_val, CPACR_EL1, 0 },
916 	/* TTBR0_EL1 */
917 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
918 	  access_vm_reg, reset_unknown, TTBR0_EL1 },
919 	/* TTBR1_EL1 */
920 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
921 	  access_vm_reg, reset_unknown, TTBR1_EL1 },
922 	/* TCR_EL1 */
923 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
924 	  access_vm_reg, reset_val, TCR_EL1, 0 },
925 
926 	/* AFSR0_EL1 */
927 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
928 	  access_vm_reg, reset_unknown, AFSR0_EL1 },
929 	/* AFSR1_EL1 */
930 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
931 	  access_vm_reg, reset_unknown, AFSR1_EL1 },
932 	/* ESR_EL1 */
933 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
934 	  access_vm_reg, reset_unknown, ESR_EL1 },
935 	/* FAR_EL1 */
936 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
937 	  access_vm_reg, reset_unknown, FAR_EL1 },
938 	/* PAR_EL1 */
939 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
940 	  NULL, reset_unknown, PAR_EL1 },
941 
942 	/* PMINTENSET_EL1 */
943 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
944 	  access_pminten, reset_unknown, PMINTENSET_EL1 },
945 	/* PMINTENCLR_EL1 */
946 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
947 	  access_pminten, NULL, PMINTENSET_EL1 },
948 
949 	/* MAIR_EL1 */
950 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
951 	  access_vm_reg, reset_unknown, MAIR_EL1 },
952 	/* AMAIR_EL1 */
953 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
954 	  access_vm_reg, reset_amair_el1, AMAIR_EL1 },
955 
956 	/* VBAR_EL1 */
957 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
958 	  NULL, reset_val, VBAR_EL1, 0 },
959 
960 	/* ICC_SGI1R_EL1 */
961 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1011), Op2(0b101),
962 	  access_gic_sgi },
963 	/* ICC_SRE_EL1 */
964 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1100), Op2(0b101),
965 	  access_gic_sre },
966 
967 	/* CONTEXTIDR_EL1 */
968 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
969 	  access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
970 	/* TPIDR_EL1 */
971 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
972 	  NULL, reset_unknown, TPIDR_EL1 },
973 
974 	/* CNTKCTL_EL1 */
975 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
976 	  NULL, reset_val, CNTKCTL_EL1, 0},
977 
978 	/* CSSELR_EL1 */
979 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
980 	  NULL, reset_unknown, CSSELR_EL1 },
981 
982 	/* PMCR_EL0 */
983 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
984 	  access_pmcr, reset_pmcr, },
985 	/* PMCNTENSET_EL0 */
986 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
987 	  access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
988 	/* PMCNTENCLR_EL0 */
989 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
990 	  access_pmcnten, NULL, PMCNTENSET_EL0 },
991 	/* PMOVSCLR_EL0 */
992 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
993 	  access_pmovs, NULL, PMOVSSET_EL0 },
994 	/* PMSWINC_EL0 */
995 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
996 	  access_pmswinc, reset_unknown, PMSWINC_EL0 },
997 	/* PMSELR_EL0 */
998 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
999 	  access_pmselr, reset_unknown, PMSELR_EL0 },
1000 	/* PMCEID0_EL0 */
1001 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
1002 	  access_pmceid },
1003 	/* PMCEID1_EL0 */
1004 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
1005 	  access_pmceid },
1006 	/* PMCCNTR_EL0 */
1007 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
1008 	  access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1009 	/* PMXEVTYPER_EL0 */
1010 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
1011 	  access_pmu_evtyper },
1012 	/* PMXEVCNTR_EL0 */
1013 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
1014 	  access_pmu_evcntr },
1015 	/* PMUSERENR_EL0
1016 	 * This register resets as unknown in 64bit mode while it resets as zero
1017 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1018 	 */
1019 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
1020 	  access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1021 	/* PMOVSSET_EL0 */
1022 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
1023 	  access_pmovs, reset_unknown, PMOVSSET_EL0 },
1024 
1025 	/* TPIDR_EL0 */
1026 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
1027 	  NULL, reset_unknown, TPIDR_EL0 },
1028 	/* TPIDRRO_EL0 */
1029 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
1030 	  NULL, reset_unknown, TPIDRRO_EL0 },
1031 
1032 	/* PMEVCNTRn_EL0 */
1033 	PMU_PMEVCNTR_EL0(0),
1034 	PMU_PMEVCNTR_EL0(1),
1035 	PMU_PMEVCNTR_EL0(2),
1036 	PMU_PMEVCNTR_EL0(3),
1037 	PMU_PMEVCNTR_EL0(4),
1038 	PMU_PMEVCNTR_EL0(5),
1039 	PMU_PMEVCNTR_EL0(6),
1040 	PMU_PMEVCNTR_EL0(7),
1041 	PMU_PMEVCNTR_EL0(8),
1042 	PMU_PMEVCNTR_EL0(9),
1043 	PMU_PMEVCNTR_EL0(10),
1044 	PMU_PMEVCNTR_EL0(11),
1045 	PMU_PMEVCNTR_EL0(12),
1046 	PMU_PMEVCNTR_EL0(13),
1047 	PMU_PMEVCNTR_EL0(14),
1048 	PMU_PMEVCNTR_EL0(15),
1049 	PMU_PMEVCNTR_EL0(16),
1050 	PMU_PMEVCNTR_EL0(17),
1051 	PMU_PMEVCNTR_EL0(18),
1052 	PMU_PMEVCNTR_EL0(19),
1053 	PMU_PMEVCNTR_EL0(20),
1054 	PMU_PMEVCNTR_EL0(21),
1055 	PMU_PMEVCNTR_EL0(22),
1056 	PMU_PMEVCNTR_EL0(23),
1057 	PMU_PMEVCNTR_EL0(24),
1058 	PMU_PMEVCNTR_EL0(25),
1059 	PMU_PMEVCNTR_EL0(26),
1060 	PMU_PMEVCNTR_EL0(27),
1061 	PMU_PMEVCNTR_EL0(28),
1062 	PMU_PMEVCNTR_EL0(29),
1063 	PMU_PMEVCNTR_EL0(30),
1064 	/* PMEVTYPERn_EL0 */
1065 	PMU_PMEVTYPER_EL0(0),
1066 	PMU_PMEVTYPER_EL0(1),
1067 	PMU_PMEVTYPER_EL0(2),
1068 	PMU_PMEVTYPER_EL0(3),
1069 	PMU_PMEVTYPER_EL0(4),
1070 	PMU_PMEVTYPER_EL0(5),
1071 	PMU_PMEVTYPER_EL0(6),
1072 	PMU_PMEVTYPER_EL0(7),
1073 	PMU_PMEVTYPER_EL0(8),
1074 	PMU_PMEVTYPER_EL0(9),
1075 	PMU_PMEVTYPER_EL0(10),
1076 	PMU_PMEVTYPER_EL0(11),
1077 	PMU_PMEVTYPER_EL0(12),
1078 	PMU_PMEVTYPER_EL0(13),
1079 	PMU_PMEVTYPER_EL0(14),
1080 	PMU_PMEVTYPER_EL0(15),
1081 	PMU_PMEVTYPER_EL0(16),
1082 	PMU_PMEVTYPER_EL0(17),
1083 	PMU_PMEVTYPER_EL0(18),
1084 	PMU_PMEVTYPER_EL0(19),
1085 	PMU_PMEVTYPER_EL0(20),
1086 	PMU_PMEVTYPER_EL0(21),
1087 	PMU_PMEVTYPER_EL0(22),
1088 	PMU_PMEVTYPER_EL0(23),
1089 	PMU_PMEVTYPER_EL0(24),
1090 	PMU_PMEVTYPER_EL0(25),
1091 	PMU_PMEVTYPER_EL0(26),
1092 	PMU_PMEVTYPER_EL0(27),
1093 	PMU_PMEVTYPER_EL0(28),
1094 	PMU_PMEVTYPER_EL0(29),
1095 	PMU_PMEVTYPER_EL0(30),
1096 	/* PMCCFILTR_EL0
1097 	 * This register resets as unknown in 64bit mode while it resets as zero
1098 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1099 	 */
1100 	{ Op0(0b11), Op1(0b011), CRn(0b1110), CRm(0b1111), Op2(0b111),
1101 	  access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1102 
1103 	/* DACR32_EL2 */
1104 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
1105 	  NULL, reset_unknown, DACR32_EL2 },
1106 	/* IFSR32_EL2 */
1107 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
1108 	  NULL, reset_unknown, IFSR32_EL2 },
1109 	/* FPEXC32_EL2 */
1110 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
1111 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
1112 };
1113 
1114 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1115 			struct sys_reg_params *p,
1116 			const struct sys_reg_desc *r)
1117 {
1118 	if (p->is_write) {
1119 		return ignore_write(vcpu, p);
1120 	} else {
1121 		u64 dfr = read_system_reg(SYS_ID_AA64DFR0_EL1);
1122 		u64 pfr = read_system_reg(SYS_ID_AA64PFR0_EL1);
1123 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1124 
1125 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1126 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1127 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1128 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1129 		return true;
1130 	}
1131 }
1132 
1133 static bool trap_debug32(struct kvm_vcpu *vcpu,
1134 			 struct sys_reg_params *p,
1135 			 const struct sys_reg_desc *r)
1136 {
1137 	if (p->is_write) {
1138 		vcpu_cp14(vcpu, r->reg) = p->regval;
1139 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
1140 	} else {
1141 		p->regval = vcpu_cp14(vcpu, r->reg);
1142 	}
1143 
1144 	return true;
1145 }
1146 
1147 /* AArch32 debug register mappings
1148  *
1149  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1150  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1151  *
1152  * All control registers and watchpoint value registers are mapped to
1153  * the lower 32 bits of their AArch64 equivalents. We share the trap
1154  * handlers with the above AArch64 code which checks what mode the
1155  * system is in.
1156  */
1157 
1158 static bool trap_xvr(struct kvm_vcpu *vcpu,
1159 		     struct sys_reg_params *p,
1160 		     const struct sys_reg_desc *rd)
1161 {
1162 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1163 
1164 	if (p->is_write) {
1165 		u64 val = *dbg_reg;
1166 
1167 		val &= 0xffffffffUL;
1168 		val |= p->regval << 32;
1169 		*dbg_reg = val;
1170 
1171 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
1172 	} else {
1173 		p->regval = *dbg_reg >> 32;
1174 	}
1175 
1176 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1177 
1178 	return true;
1179 }
1180 
1181 #define DBG_BCR_BVR_WCR_WVR(n)						\
1182 	/* DBGBVRn */							\
1183 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1184 	/* DBGBCRn */							\
1185 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1186 	/* DBGWVRn */							\
1187 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1188 	/* DBGWCRn */							\
1189 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1190 
1191 #define DBGBXVR(n)							\
1192 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1193 
1194 /*
1195  * Trapped cp14 registers. We generally ignore most of the external
1196  * debug, on the principle that they don't really make sense to a
1197  * guest. Revisit this one day, would this principle change.
1198  */
1199 static const struct sys_reg_desc cp14_regs[] = {
1200 	/* DBGIDR */
1201 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1202 	/* DBGDTRRXext */
1203 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1204 
1205 	DBG_BCR_BVR_WCR_WVR(0),
1206 	/* DBGDSCRint */
1207 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1208 	DBG_BCR_BVR_WCR_WVR(1),
1209 	/* DBGDCCINT */
1210 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1211 	/* DBGDSCRext */
1212 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1213 	DBG_BCR_BVR_WCR_WVR(2),
1214 	/* DBGDTR[RT]Xint */
1215 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1216 	/* DBGDTR[RT]Xext */
1217 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1218 	DBG_BCR_BVR_WCR_WVR(3),
1219 	DBG_BCR_BVR_WCR_WVR(4),
1220 	DBG_BCR_BVR_WCR_WVR(5),
1221 	/* DBGWFAR */
1222 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1223 	/* DBGOSECCR */
1224 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1225 	DBG_BCR_BVR_WCR_WVR(6),
1226 	/* DBGVCR */
1227 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1228 	DBG_BCR_BVR_WCR_WVR(7),
1229 	DBG_BCR_BVR_WCR_WVR(8),
1230 	DBG_BCR_BVR_WCR_WVR(9),
1231 	DBG_BCR_BVR_WCR_WVR(10),
1232 	DBG_BCR_BVR_WCR_WVR(11),
1233 	DBG_BCR_BVR_WCR_WVR(12),
1234 	DBG_BCR_BVR_WCR_WVR(13),
1235 	DBG_BCR_BVR_WCR_WVR(14),
1236 	DBG_BCR_BVR_WCR_WVR(15),
1237 
1238 	/* DBGDRAR (32bit) */
1239 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1240 
1241 	DBGBXVR(0),
1242 	/* DBGOSLAR */
1243 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1244 	DBGBXVR(1),
1245 	/* DBGOSLSR */
1246 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1247 	DBGBXVR(2),
1248 	DBGBXVR(3),
1249 	/* DBGOSDLR */
1250 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1251 	DBGBXVR(4),
1252 	/* DBGPRCR */
1253 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1254 	DBGBXVR(5),
1255 	DBGBXVR(6),
1256 	DBGBXVR(7),
1257 	DBGBXVR(8),
1258 	DBGBXVR(9),
1259 	DBGBXVR(10),
1260 	DBGBXVR(11),
1261 	DBGBXVR(12),
1262 	DBGBXVR(13),
1263 	DBGBXVR(14),
1264 	DBGBXVR(15),
1265 
1266 	/* DBGDSAR (32bit) */
1267 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1268 
1269 	/* DBGDEVID2 */
1270 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1271 	/* DBGDEVID1 */
1272 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1273 	/* DBGDEVID */
1274 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1275 	/* DBGCLAIMSET */
1276 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1277 	/* DBGCLAIMCLR */
1278 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1279 	/* DBGAUTHSTATUS */
1280 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1281 };
1282 
1283 /* Trapped cp14 64bit registers */
1284 static const struct sys_reg_desc cp14_64_regs[] = {
1285 	/* DBGDRAR (64bit) */
1286 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1287 
1288 	/* DBGDSAR (64bit) */
1289 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1290 };
1291 
1292 /* Macro to expand the PMEVCNTRn register */
1293 #define PMU_PMEVCNTR(n)							\
1294 	/* PMEVCNTRn */							\
1295 	{ Op1(0), CRn(0b1110),						\
1296 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1297 	  access_pmu_evcntr }
1298 
1299 /* Macro to expand the PMEVTYPERn register */
1300 #define PMU_PMEVTYPER(n)						\
1301 	/* PMEVTYPERn */						\
1302 	{ Op1(0), CRn(0b1110),						\
1303 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1304 	  access_pmu_evtyper }
1305 
1306 /*
1307  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1308  * depending on the way they are accessed (as a 32bit or a 64bit
1309  * register).
1310  */
1311 static const struct sys_reg_desc cp15_regs[] = {
1312 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
1313 
1314 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1315 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1316 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1317 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1318 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1319 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1320 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1321 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1322 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1323 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1324 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1325 
1326 	/*
1327 	 * DC{C,I,CI}SW operations:
1328 	 */
1329 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1330 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1331 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1332 
1333 	/* PMU */
1334 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1335 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1336 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1337 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1338 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1339 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1340 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1341 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1342 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1343 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1344 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1345 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1346 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1347 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1348 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1349 
1350 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
1351 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
1352 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
1353 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
1354 
1355 	/* ICC_SRE */
1356 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
1357 
1358 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
1359 
1360 	/* PMEVCNTRn */
1361 	PMU_PMEVCNTR(0),
1362 	PMU_PMEVCNTR(1),
1363 	PMU_PMEVCNTR(2),
1364 	PMU_PMEVCNTR(3),
1365 	PMU_PMEVCNTR(4),
1366 	PMU_PMEVCNTR(5),
1367 	PMU_PMEVCNTR(6),
1368 	PMU_PMEVCNTR(7),
1369 	PMU_PMEVCNTR(8),
1370 	PMU_PMEVCNTR(9),
1371 	PMU_PMEVCNTR(10),
1372 	PMU_PMEVCNTR(11),
1373 	PMU_PMEVCNTR(12),
1374 	PMU_PMEVCNTR(13),
1375 	PMU_PMEVCNTR(14),
1376 	PMU_PMEVCNTR(15),
1377 	PMU_PMEVCNTR(16),
1378 	PMU_PMEVCNTR(17),
1379 	PMU_PMEVCNTR(18),
1380 	PMU_PMEVCNTR(19),
1381 	PMU_PMEVCNTR(20),
1382 	PMU_PMEVCNTR(21),
1383 	PMU_PMEVCNTR(22),
1384 	PMU_PMEVCNTR(23),
1385 	PMU_PMEVCNTR(24),
1386 	PMU_PMEVCNTR(25),
1387 	PMU_PMEVCNTR(26),
1388 	PMU_PMEVCNTR(27),
1389 	PMU_PMEVCNTR(28),
1390 	PMU_PMEVCNTR(29),
1391 	PMU_PMEVCNTR(30),
1392 	/* PMEVTYPERn */
1393 	PMU_PMEVTYPER(0),
1394 	PMU_PMEVTYPER(1),
1395 	PMU_PMEVTYPER(2),
1396 	PMU_PMEVTYPER(3),
1397 	PMU_PMEVTYPER(4),
1398 	PMU_PMEVTYPER(5),
1399 	PMU_PMEVTYPER(6),
1400 	PMU_PMEVTYPER(7),
1401 	PMU_PMEVTYPER(8),
1402 	PMU_PMEVTYPER(9),
1403 	PMU_PMEVTYPER(10),
1404 	PMU_PMEVTYPER(11),
1405 	PMU_PMEVTYPER(12),
1406 	PMU_PMEVTYPER(13),
1407 	PMU_PMEVTYPER(14),
1408 	PMU_PMEVTYPER(15),
1409 	PMU_PMEVTYPER(16),
1410 	PMU_PMEVTYPER(17),
1411 	PMU_PMEVTYPER(18),
1412 	PMU_PMEVTYPER(19),
1413 	PMU_PMEVTYPER(20),
1414 	PMU_PMEVTYPER(21),
1415 	PMU_PMEVTYPER(22),
1416 	PMU_PMEVTYPER(23),
1417 	PMU_PMEVTYPER(24),
1418 	PMU_PMEVTYPER(25),
1419 	PMU_PMEVTYPER(26),
1420 	PMU_PMEVTYPER(27),
1421 	PMU_PMEVTYPER(28),
1422 	PMU_PMEVTYPER(29),
1423 	PMU_PMEVTYPER(30),
1424 	/* PMCCFILTR */
1425 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
1426 };
1427 
1428 static const struct sys_reg_desc cp15_64_regs[] = {
1429 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1430 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
1431 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
1432 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
1433 };
1434 
1435 /* Target specific emulation tables */
1436 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
1437 
1438 void kvm_register_target_sys_reg_table(unsigned int target,
1439 				       struct kvm_sys_reg_target_table *table)
1440 {
1441 	target_tables[target] = table;
1442 }
1443 
1444 /* Get specific register table for this target. */
1445 static const struct sys_reg_desc *get_target_table(unsigned target,
1446 						   bool mode_is_64,
1447 						   size_t *num)
1448 {
1449 	struct kvm_sys_reg_target_table *table;
1450 
1451 	table = target_tables[target];
1452 	if (mode_is_64) {
1453 		*num = table->table64.num;
1454 		return table->table64.table;
1455 	} else {
1456 		*num = table->table32.num;
1457 		return table->table32.table;
1458 	}
1459 }
1460 
1461 #define reg_to_match_value(x)						\
1462 	({								\
1463 		unsigned long val;					\
1464 		val  = (x)->Op0 << 14;					\
1465 		val |= (x)->Op1 << 11;					\
1466 		val |= (x)->CRn << 7;					\
1467 		val |= (x)->CRm << 3;					\
1468 		val |= (x)->Op2;					\
1469 		val;							\
1470 	 })
1471 
1472 static int match_sys_reg(const void *key, const void *elt)
1473 {
1474 	const unsigned long pval = (unsigned long)key;
1475 	const struct sys_reg_desc *r = elt;
1476 
1477 	return pval - reg_to_match_value(r);
1478 }
1479 
1480 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
1481 					 const struct sys_reg_desc table[],
1482 					 unsigned int num)
1483 {
1484 	unsigned long pval = reg_to_match_value(params);
1485 
1486 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
1487 }
1488 
1489 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
1490 {
1491 	kvm_inject_undefined(vcpu);
1492 	return 1;
1493 }
1494 
1495 /*
1496  * emulate_cp --  tries to match a sys_reg access in a handling table, and
1497  *                call the corresponding trap handler.
1498  *
1499  * @params: pointer to the descriptor of the access
1500  * @table: array of trap descriptors
1501  * @num: size of the trap descriptor array
1502  *
1503  * Return 0 if the access has been handled, and -1 if not.
1504  */
1505 static int emulate_cp(struct kvm_vcpu *vcpu,
1506 		      struct sys_reg_params *params,
1507 		      const struct sys_reg_desc *table,
1508 		      size_t num)
1509 {
1510 	const struct sys_reg_desc *r;
1511 
1512 	if (!table)
1513 		return -1;	/* Not handled */
1514 
1515 	r = find_reg(params, table, num);
1516 
1517 	if (r) {
1518 		/*
1519 		 * Not having an accessor means that we have
1520 		 * configured a trap that we don't know how to
1521 		 * handle. This certainly qualifies as a gross bug
1522 		 * that should be fixed right away.
1523 		 */
1524 		BUG_ON(!r->access);
1525 
1526 		if (likely(r->access(vcpu, params, r))) {
1527 			/* Skip instruction, since it was emulated */
1528 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1529 			/* Handled */
1530 			return 0;
1531 		}
1532 	}
1533 
1534 	/* Not handled */
1535 	return -1;
1536 }
1537 
1538 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
1539 				struct sys_reg_params *params)
1540 {
1541 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
1542 	int cp = -1;
1543 
1544 	switch(hsr_ec) {
1545 	case ESR_ELx_EC_CP15_32:
1546 	case ESR_ELx_EC_CP15_64:
1547 		cp = 15;
1548 		break;
1549 	case ESR_ELx_EC_CP14_MR:
1550 	case ESR_ELx_EC_CP14_64:
1551 		cp = 14;
1552 		break;
1553 	default:
1554 		WARN_ON(1);
1555 	}
1556 
1557 	kvm_err("Unsupported guest CP%d access at: %08lx\n",
1558 		cp, *vcpu_pc(vcpu));
1559 	print_sys_reg_instr(params);
1560 	kvm_inject_undefined(vcpu);
1561 }
1562 
1563 /**
1564  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
1565  * @vcpu: The VCPU pointer
1566  * @run:  The kvm_run struct
1567  */
1568 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
1569 			    const struct sys_reg_desc *global,
1570 			    size_t nr_global,
1571 			    const struct sys_reg_desc *target_specific,
1572 			    size_t nr_specific)
1573 {
1574 	struct sys_reg_params params;
1575 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1576 	int Rt = (hsr >> 5) & 0xf;
1577 	int Rt2 = (hsr >> 10) & 0xf;
1578 
1579 	params.is_aarch32 = true;
1580 	params.is_32bit = false;
1581 	params.CRm = (hsr >> 1) & 0xf;
1582 	params.is_write = ((hsr & 1) == 0);
1583 
1584 	params.Op0 = 0;
1585 	params.Op1 = (hsr >> 16) & 0xf;
1586 	params.Op2 = 0;
1587 	params.CRn = 0;
1588 
1589 	/*
1590 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
1591 	 * backends between AArch32 and AArch64, we get away with it.
1592 	 */
1593 	if (params.is_write) {
1594 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
1595 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
1596 	}
1597 
1598 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
1599 		goto out;
1600 	if (!emulate_cp(vcpu, &params, global, nr_global))
1601 		goto out;
1602 
1603 	unhandled_cp_access(vcpu, &params);
1604 
1605 out:
1606 	/* Split up the value between registers for the read side */
1607 	if (!params.is_write) {
1608 		vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
1609 		vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
1610 	}
1611 
1612 	return 1;
1613 }
1614 
1615 /**
1616  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
1617  * @vcpu: The VCPU pointer
1618  * @run:  The kvm_run struct
1619  */
1620 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
1621 			    const struct sys_reg_desc *global,
1622 			    size_t nr_global,
1623 			    const struct sys_reg_desc *target_specific,
1624 			    size_t nr_specific)
1625 {
1626 	struct sys_reg_params params;
1627 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1628 	int Rt  = (hsr >> 5) & 0xf;
1629 
1630 	params.is_aarch32 = true;
1631 	params.is_32bit = true;
1632 	params.CRm = (hsr >> 1) & 0xf;
1633 	params.regval = vcpu_get_reg(vcpu, Rt);
1634 	params.is_write = ((hsr & 1) == 0);
1635 	params.CRn = (hsr >> 10) & 0xf;
1636 	params.Op0 = 0;
1637 	params.Op1 = (hsr >> 14) & 0x7;
1638 	params.Op2 = (hsr >> 17) & 0x7;
1639 
1640 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
1641 	    !emulate_cp(vcpu, &params, global, nr_global)) {
1642 		if (!params.is_write)
1643 			vcpu_set_reg(vcpu, Rt, params.regval);
1644 		return 1;
1645 	}
1646 
1647 	unhandled_cp_access(vcpu, &params);
1648 	return 1;
1649 }
1650 
1651 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1652 {
1653 	const struct sys_reg_desc *target_specific;
1654 	size_t num;
1655 
1656 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1657 	return kvm_handle_cp_64(vcpu,
1658 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
1659 				target_specific, num);
1660 }
1661 
1662 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1663 {
1664 	const struct sys_reg_desc *target_specific;
1665 	size_t num;
1666 
1667 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1668 	return kvm_handle_cp_32(vcpu,
1669 				cp15_regs, ARRAY_SIZE(cp15_regs),
1670 				target_specific, num);
1671 }
1672 
1673 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1674 {
1675 	return kvm_handle_cp_64(vcpu,
1676 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
1677 				NULL, 0);
1678 }
1679 
1680 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1681 {
1682 	return kvm_handle_cp_32(vcpu,
1683 				cp14_regs, ARRAY_SIZE(cp14_regs),
1684 				NULL, 0);
1685 }
1686 
1687 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
1688 			   struct sys_reg_params *params)
1689 {
1690 	size_t num;
1691 	const struct sys_reg_desc *table, *r;
1692 
1693 	table = get_target_table(vcpu->arch.target, true, &num);
1694 
1695 	/* Search target-specific then generic table. */
1696 	r = find_reg(params, table, num);
1697 	if (!r)
1698 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1699 
1700 	if (likely(r)) {
1701 		/*
1702 		 * Not having an accessor means that we have
1703 		 * configured a trap that we don't know how to
1704 		 * handle. This certainly qualifies as a gross bug
1705 		 * that should be fixed right away.
1706 		 */
1707 		BUG_ON(!r->access);
1708 
1709 		if (likely(r->access(vcpu, params, r))) {
1710 			/* Skip instruction, since it was emulated */
1711 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1712 			return 1;
1713 		}
1714 		/* If access function fails, it should complain. */
1715 	} else {
1716 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
1717 			*vcpu_pc(vcpu));
1718 		print_sys_reg_instr(params);
1719 	}
1720 	kvm_inject_undefined(vcpu);
1721 	return 1;
1722 }
1723 
1724 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
1725 			      const struct sys_reg_desc *table, size_t num)
1726 {
1727 	unsigned long i;
1728 
1729 	for (i = 0; i < num; i++)
1730 		if (table[i].reset)
1731 			table[i].reset(vcpu, &table[i]);
1732 }
1733 
1734 /**
1735  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
1736  * @vcpu: The VCPU pointer
1737  * @run:  The kvm_run struct
1738  */
1739 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1740 {
1741 	struct sys_reg_params params;
1742 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1743 	int Rt = (esr >> 5) & 0x1f;
1744 	int ret;
1745 
1746 	trace_kvm_handle_sys_reg(esr);
1747 
1748 	params.is_aarch32 = false;
1749 	params.is_32bit = false;
1750 	params.Op0 = (esr >> 20) & 3;
1751 	params.Op1 = (esr >> 14) & 0x7;
1752 	params.CRn = (esr >> 10) & 0xf;
1753 	params.CRm = (esr >> 1) & 0xf;
1754 	params.Op2 = (esr >> 17) & 0x7;
1755 	params.regval = vcpu_get_reg(vcpu, Rt);
1756 	params.is_write = !(esr & 1);
1757 
1758 	ret = emulate_sys_reg(vcpu, &params);
1759 
1760 	if (!params.is_write)
1761 		vcpu_set_reg(vcpu, Rt, params.regval);
1762 	return ret;
1763 }
1764 
1765 /******************************************************************************
1766  * Userspace API
1767  *****************************************************************************/
1768 
1769 static bool index_to_params(u64 id, struct sys_reg_params *params)
1770 {
1771 	switch (id & KVM_REG_SIZE_MASK) {
1772 	case KVM_REG_SIZE_U64:
1773 		/* Any unused index bits means it's not valid. */
1774 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1775 			      | KVM_REG_ARM_COPROC_MASK
1776 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
1777 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
1778 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
1779 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
1780 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
1781 			return false;
1782 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1783 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1784 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1785 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1786 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1787 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1788 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1789 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1790 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1791 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1792 		return true;
1793 	default:
1794 		return false;
1795 	}
1796 }
1797 
1798 /* Decode an index value, and find the sys_reg_desc entry. */
1799 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1800 						    u64 id)
1801 {
1802 	size_t num;
1803 	const struct sys_reg_desc *table, *r;
1804 	struct sys_reg_params params;
1805 
1806 	/* We only do sys_reg for now. */
1807 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1808 		return NULL;
1809 
1810 	if (!index_to_params(id, &params))
1811 		return NULL;
1812 
1813 	table = get_target_table(vcpu->arch.target, true, &num);
1814 	r = find_reg(&params, table, num);
1815 	if (!r)
1816 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1817 
1818 	/* Not saved in the sys_reg array? */
1819 	if (r && !r->reg)
1820 		r = NULL;
1821 
1822 	return r;
1823 }
1824 
1825 /*
1826  * These are the invariant sys_reg registers: we let the guest see the
1827  * host versions of these, so they're part of the guest state.
1828  *
1829  * A future CPU may provide a mechanism to present different values to
1830  * the guest, or a future kvm may trap them.
1831  */
1832 
1833 #define FUNCTION_INVARIANT(reg)						\
1834 	static void get_##reg(struct kvm_vcpu *v,			\
1835 			      const struct sys_reg_desc *r)		\
1836 	{								\
1837 		((struct sys_reg_desc *)r)->val = read_sysreg(reg);	\
1838 	}
1839 
1840 FUNCTION_INVARIANT(midr_el1)
1841 FUNCTION_INVARIANT(ctr_el0)
1842 FUNCTION_INVARIANT(revidr_el1)
1843 FUNCTION_INVARIANT(id_pfr0_el1)
1844 FUNCTION_INVARIANT(id_pfr1_el1)
1845 FUNCTION_INVARIANT(id_dfr0_el1)
1846 FUNCTION_INVARIANT(id_afr0_el1)
1847 FUNCTION_INVARIANT(id_mmfr0_el1)
1848 FUNCTION_INVARIANT(id_mmfr1_el1)
1849 FUNCTION_INVARIANT(id_mmfr2_el1)
1850 FUNCTION_INVARIANT(id_mmfr3_el1)
1851 FUNCTION_INVARIANT(id_isar0_el1)
1852 FUNCTION_INVARIANT(id_isar1_el1)
1853 FUNCTION_INVARIANT(id_isar2_el1)
1854 FUNCTION_INVARIANT(id_isar3_el1)
1855 FUNCTION_INVARIANT(id_isar4_el1)
1856 FUNCTION_INVARIANT(id_isar5_el1)
1857 FUNCTION_INVARIANT(clidr_el1)
1858 FUNCTION_INVARIANT(aidr_el1)
1859 
1860 /* ->val is filled in by kvm_sys_reg_table_init() */
1861 static struct sys_reg_desc invariant_sys_regs[] = {
1862 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1863 	  NULL, get_midr_el1 },
1864 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1865 	  NULL, get_revidr_el1 },
1866 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1867 	  NULL, get_id_pfr0_el1 },
1868 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1869 	  NULL, get_id_pfr1_el1 },
1870 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1871 	  NULL, get_id_dfr0_el1 },
1872 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1873 	  NULL, get_id_afr0_el1 },
1874 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1875 	  NULL, get_id_mmfr0_el1 },
1876 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1877 	  NULL, get_id_mmfr1_el1 },
1878 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1879 	  NULL, get_id_mmfr2_el1 },
1880 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1881 	  NULL, get_id_mmfr3_el1 },
1882 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1883 	  NULL, get_id_isar0_el1 },
1884 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1885 	  NULL, get_id_isar1_el1 },
1886 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1887 	  NULL, get_id_isar2_el1 },
1888 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1889 	  NULL, get_id_isar3_el1 },
1890 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1891 	  NULL, get_id_isar4_el1 },
1892 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1893 	  NULL, get_id_isar5_el1 },
1894 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1895 	  NULL, get_clidr_el1 },
1896 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1897 	  NULL, get_aidr_el1 },
1898 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1899 	  NULL, get_ctr_el0 },
1900 };
1901 
1902 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1903 {
1904 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1905 		return -EFAULT;
1906 	return 0;
1907 }
1908 
1909 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1910 {
1911 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1912 		return -EFAULT;
1913 	return 0;
1914 }
1915 
1916 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1917 {
1918 	struct sys_reg_params params;
1919 	const struct sys_reg_desc *r;
1920 
1921 	if (!index_to_params(id, &params))
1922 		return -ENOENT;
1923 
1924 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1925 	if (!r)
1926 		return -ENOENT;
1927 
1928 	return reg_to_user(uaddr, &r->val, id);
1929 }
1930 
1931 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1932 {
1933 	struct sys_reg_params params;
1934 	const struct sys_reg_desc *r;
1935 	int err;
1936 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1937 
1938 	if (!index_to_params(id, &params))
1939 		return -ENOENT;
1940 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1941 	if (!r)
1942 		return -ENOENT;
1943 
1944 	err = reg_from_user(&val, uaddr, id);
1945 	if (err)
1946 		return err;
1947 
1948 	/* This is what we mean by invariant: you can't change it. */
1949 	if (r->val != val)
1950 		return -EINVAL;
1951 
1952 	return 0;
1953 }
1954 
1955 static bool is_valid_cache(u32 val)
1956 {
1957 	u32 level, ctype;
1958 
1959 	if (val >= CSSELR_MAX)
1960 		return false;
1961 
1962 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1963 	level = (val >> 1);
1964 	ctype = (cache_levels >> (level * 3)) & 7;
1965 
1966 	switch (ctype) {
1967 	case 0: /* No cache */
1968 		return false;
1969 	case 1: /* Instruction cache only */
1970 		return (val & 1);
1971 	case 2: /* Data cache only */
1972 	case 4: /* Unified cache */
1973 		return !(val & 1);
1974 	case 3: /* Separate instruction and data caches */
1975 		return true;
1976 	default: /* Reserved: we can't know instruction or data. */
1977 		return false;
1978 	}
1979 }
1980 
1981 static int demux_c15_get(u64 id, void __user *uaddr)
1982 {
1983 	u32 val;
1984 	u32 __user *uval = uaddr;
1985 
1986 	/* Fail if we have unknown bits set. */
1987 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1988 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1989 		return -ENOENT;
1990 
1991 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1992 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1993 		if (KVM_REG_SIZE(id) != 4)
1994 			return -ENOENT;
1995 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1996 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1997 		if (!is_valid_cache(val))
1998 			return -ENOENT;
1999 
2000 		return put_user(get_ccsidr(val), uval);
2001 	default:
2002 		return -ENOENT;
2003 	}
2004 }
2005 
2006 static int demux_c15_set(u64 id, void __user *uaddr)
2007 {
2008 	u32 val, newval;
2009 	u32 __user *uval = uaddr;
2010 
2011 	/* Fail if we have unknown bits set. */
2012 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2013 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2014 		return -ENOENT;
2015 
2016 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2017 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2018 		if (KVM_REG_SIZE(id) != 4)
2019 			return -ENOENT;
2020 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2021 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2022 		if (!is_valid_cache(val))
2023 			return -ENOENT;
2024 
2025 		if (get_user(newval, uval))
2026 			return -EFAULT;
2027 
2028 		/* This is also invariant: you can't change it. */
2029 		if (newval != get_ccsidr(val))
2030 			return -EINVAL;
2031 		return 0;
2032 	default:
2033 		return -ENOENT;
2034 	}
2035 }
2036 
2037 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2038 {
2039 	const struct sys_reg_desc *r;
2040 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2041 
2042 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2043 		return demux_c15_get(reg->id, uaddr);
2044 
2045 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2046 		return -ENOENT;
2047 
2048 	r = index_to_sys_reg_desc(vcpu, reg->id);
2049 	if (!r)
2050 		return get_invariant_sys_reg(reg->id, uaddr);
2051 
2052 	if (r->get_user)
2053 		return (r->get_user)(vcpu, r, reg, uaddr);
2054 
2055 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
2056 }
2057 
2058 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2059 {
2060 	const struct sys_reg_desc *r;
2061 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2062 
2063 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2064 		return demux_c15_set(reg->id, uaddr);
2065 
2066 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2067 		return -ENOENT;
2068 
2069 	r = index_to_sys_reg_desc(vcpu, reg->id);
2070 	if (!r)
2071 		return set_invariant_sys_reg(reg->id, uaddr);
2072 
2073 	if (r->set_user)
2074 		return (r->set_user)(vcpu, r, reg, uaddr);
2075 
2076 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2077 }
2078 
2079 static unsigned int num_demux_regs(void)
2080 {
2081 	unsigned int i, count = 0;
2082 
2083 	for (i = 0; i < CSSELR_MAX; i++)
2084 		if (is_valid_cache(i))
2085 			count++;
2086 
2087 	return count;
2088 }
2089 
2090 static int write_demux_regids(u64 __user *uindices)
2091 {
2092 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2093 	unsigned int i;
2094 
2095 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2096 	for (i = 0; i < CSSELR_MAX; i++) {
2097 		if (!is_valid_cache(i))
2098 			continue;
2099 		if (put_user(val | i, uindices))
2100 			return -EFAULT;
2101 		uindices++;
2102 	}
2103 	return 0;
2104 }
2105 
2106 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2107 {
2108 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2109 		KVM_REG_ARM64_SYSREG |
2110 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2111 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2112 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2113 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2114 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2115 }
2116 
2117 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2118 {
2119 	if (!*uind)
2120 		return true;
2121 
2122 	if (put_user(sys_reg_to_index(reg), *uind))
2123 		return false;
2124 
2125 	(*uind)++;
2126 	return true;
2127 }
2128 
2129 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2130 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2131 {
2132 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
2133 	unsigned int total = 0;
2134 	size_t num;
2135 
2136 	/* We check for duplicates here, to allow arch-specific overrides. */
2137 	i1 = get_target_table(vcpu->arch.target, true, &num);
2138 	end1 = i1 + num;
2139 	i2 = sys_reg_descs;
2140 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2141 
2142 	BUG_ON(i1 == end1 || i2 == end2);
2143 
2144 	/* Walk carefully, as both tables may refer to the same register. */
2145 	while (i1 || i2) {
2146 		int cmp = cmp_sys_reg(i1, i2);
2147 		/* target-specific overrides generic entry. */
2148 		if (cmp <= 0) {
2149 			/* Ignore registers we trap but don't save. */
2150 			if (i1->reg) {
2151 				if (!copy_reg_to_user(i1, &uind))
2152 					return -EFAULT;
2153 				total++;
2154 			}
2155 		} else {
2156 			/* Ignore registers we trap but don't save. */
2157 			if (i2->reg) {
2158 				if (!copy_reg_to_user(i2, &uind))
2159 					return -EFAULT;
2160 				total++;
2161 			}
2162 		}
2163 
2164 		if (cmp <= 0 && ++i1 == end1)
2165 			i1 = NULL;
2166 		if (cmp >= 0 && ++i2 == end2)
2167 			i2 = NULL;
2168 	}
2169 	return total;
2170 }
2171 
2172 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2173 {
2174 	return ARRAY_SIZE(invariant_sys_regs)
2175 		+ num_demux_regs()
2176 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2177 }
2178 
2179 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2180 {
2181 	unsigned int i;
2182 	int err;
2183 
2184 	/* Then give them all the invariant registers' indices. */
2185 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2186 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2187 			return -EFAULT;
2188 		uindices++;
2189 	}
2190 
2191 	err = walk_sys_regs(vcpu, uindices);
2192 	if (err < 0)
2193 		return err;
2194 	uindices += err;
2195 
2196 	return write_demux_regids(uindices);
2197 }
2198 
2199 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
2200 {
2201 	unsigned int i;
2202 
2203 	for (i = 1; i < n; i++) {
2204 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2205 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2206 			return 1;
2207 		}
2208 	}
2209 
2210 	return 0;
2211 }
2212 
2213 void kvm_sys_reg_table_init(void)
2214 {
2215 	unsigned int i;
2216 	struct sys_reg_desc clidr;
2217 
2218 	/* Make sure tables are unique and in order. */
2219 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
2220 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
2221 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
2222 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
2223 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
2224 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
2225 
2226 	/* We abuse the reset function to overwrite the table itself. */
2227 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2228 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2229 
2230 	/*
2231 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2232 	 *
2233 	 *   If software reads the Cache Type fields from Ctype1
2234 	 *   upwards, once it has seen a value of 0b000, no caches
2235 	 *   exist at further-out levels of the hierarchy. So, for
2236 	 *   example, if Ctype3 is the first Cache Type field with a
2237 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2238 	 *   ignored.
2239 	 */
2240 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2241 	cache_levels = clidr.val;
2242 	for (i = 0; i < 7; i++)
2243 		if (((cache_levels >> (i*3)) & 7) == 0)
2244 			break;
2245 	/* Clear all higher bits. */
2246 	cache_levels &= (1 << (i*3))-1;
2247 }
2248 
2249 /**
2250  * kvm_reset_sys_regs - sets system registers to reset value
2251  * @vcpu: The VCPU pointer
2252  *
2253  * This function finds the right table above and sets the registers on the
2254  * virtual CPU struct to their architecturally defined reset values.
2255  */
2256 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2257 {
2258 	size_t num;
2259 	const struct sys_reg_desc *table;
2260 
2261 	/* Catch someone adding a register without putting in reset entry. */
2262 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
2263 
2264 	/* Generic chip reset first (so target could override). */
2265 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2266 
2267 	table = get_target_table(vcpu->arch.target, true, &num);
2268 	reset_sys_reg_descs(vcpu, table, num);
2269 
2270 	for (num = 1; num < NR_SYS_REGS; num++)
2271 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
2272 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
2273 }
2274