xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 32981ea5)
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 
40 #include <trace/events/kvm.h>
41 
42 #include "sys_regs.h"
43 
44 #include "trace.h"
45 
46 /*
47  * All of this file is extremly similar to the ARM coproc.c, but the
48  * types are different. My gut feeling is that it should be pretty
49  * easy to merge, but that would be an ABI breakage -- again. VFP
50  * would also need to be abstracted.
51  *
52  * For AArch32, we only take care of what is being trapped. Anything
53  * that has to do with init and userspace access has to go via the
54  * 64bit interface.
55  */
56 
57 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
58 static u32 cache_levels;
59 
60 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
61 #define CSSELR_MAX 12
62 
63 /* Which cache CCSIDR represents depends on CSSELR value. */
64 static u32 get_ccsidr(u32 csselr)
65 {
66 	u32 ccsidr;
67 
68 	/* Make sure noone else changes CSSELR during this! */
69 	local_irq_disable();
70 	/* Put value into CSSELR */
71 	asm volatile("msr csselr_el1, %x0" : : "r" (csselr));
72 	isb();
73 	/* Read result out of CCSIDR */
74 	asm volatile("mrs %0, ccsidr_el1" : "=r" (ccsidr));
75 	local_irq_enable();
76 
77 	return ccsidr;
78 }
79 
80 /*
81  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
82  */
83 static bool access_dcsw(struct kvm_vcpu *vcpu,
84 			struct sys_reg_params *p,
85 			const struct sys_reg_desc *r)
86 {
87 	if (!p->is_write)
88 		return read_from_write_only(vcpu, p);
89 
90 	kvm_set_way_flush(vcpu);
91 	return true;
92 }
93 
94 /*
95  * Generic accessor for VM registers. Only called as long as HCR_TVM
96  * is set. If the guest enables the MMU, we stop trapping the VM
97  * sys_regs and leave it in complete control of the caches.
98  */
99 static bool access_vm_reg(struct kvm_vcpu *vcpu,
100 			  struct sys_reg_params *p,
101 			  const struct sys_reg_desc *r)
102 {
103 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
104 
105 	BUG_ON(!p->is_write);
106 
107 	if (!p->is_aarch32) {
108 		vcpu_sys_reg(vcpu, r->reg) = p->regval;
109 	} else {
110 		if (!p->is_32bit)
111 			vcpu_cp15_64_high(vcpu, r->reg) = upper_32_bits(p->regval);
112 		vcpu_cp15_64_low(vcpu, r->reg) = lower_32_bits(p->regval);
113 	}
114 
115 	kvm_toggle_cache(vcpu, was_enabled);
116 	return true;
117 }
118 
119 /*
120  * Trap handler for the GICv3 SGI generation system register.
121  * Forward the request to the VGIC emulation.
122  * The cp15_64 code makes sure this automatically works
123  * for both AArch64 and AArch32 accesses.
124  */
125 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
126 			   struct sys_reg_params *p,
127 			   const struct sys_reg_desc *r)
128 {
129 	if (!p->is_write)
130 		return read_from_write_only(vcpu, p);
131 
132 	vgic_v3_dispatch_sgi(vcpu, p->regval);
133 
134 	return true;
135 }
136 
137 static bool access_gic_sre(struct kvm_vcpu *vcpu,
138 			   struct sys_reg_params *p,
139 			   const struct sys_reg_desc *r)
140 {
141 	if (p->is_write)
142 		return ignore_write(vcpu, p);
143 
144 	p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
145 	return true;
146 }
147 
148 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
149 			struct sys_reg_params *p,
150 			const struct sys_reg_desc *r)
151 {
152 	if (p->is_write)
153 		return ignore_write(vcpu, p);
154 	else
155 		return read_zero(vcpu, p);
156 }
157 
158 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
159 			   struct sys_reg_params *p,
160 			   const struct sys_reg_desc *r)
161 {
162 	if (p->is_write) {
163 		return ignore_write(vcpu, p);
164 	} else {
165 		p->regval = (1 << 3);
166 		return true;
167 	}
168 }
169 
170 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
171 				   struct sys_reg_params *p,
172 				   const struct sys_reg_desc *r)
173 {
174 	if (p->is_write) {
175 		return ignore_write(vcpu, p);
176 	} else {
177 		u32 val;
178 		asm volatile("mrs %0, dbgauthstatus_el1" : "=r" (val));
179 		p->regval = val;
180 		return true;
181 	}
182 }
183 
184 /*
185  * We want to avoid world-switching all the DBG registers all the
186  * time:
187  *
188  * - If we've touched any debug register, it is likely that we're
189  *   going to touch more of them. It then makes sense to disable the
190  *   traps and start doing the save/restore dance
191  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
192  *   then mandatory to save/restore the registers, as the guest
193  *   depends on them.
194  *
195  * For this, we use a DIRTY bit, indicating the guest has modified the
196  * debug registers, used as follow:
197  *
198  * On guest entry:
199  * - If the dirty bit is set (because we're coming back from trapping),
200  *   disable the traps, save host registers, restore guest registers.
201  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
202  *   set the dirty bit, disable the traps, save host registers,
203  *   restore guest registers.
204  * - Otherwise, enable the traps
205  *
206  * On guest exit:
207  * - If the dirty bit is set, save guest registers, restore host
208  *   registers and clear the dirty bit. This ensure that the host can
209  *   now use the debug registers.
210  */
211 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
212 			    struct sys_reg_params *p,
213 			    const struct sys_reg_desc *r)
214 {
215 	if (p->is_write) {
216 		vcpu_sys_reg(vcpu, r->reg) = p->regval;
217 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
218 	} else {
219 		p->regval = vcpu_sys_reg(vcpu, r->reg);
220 	}
221 
222 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
223 
224 	return true;
225 }
226 
227 /*
228  * reg_to_dbg/dbg_to_reg
229  *
230  * A 32 bit write to a debug register leave top bits alone
231  * A 32 bit read from a debug register only returns the bottom bits
232  *
233  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
234  * hyp.S code switches between host and guest values in future.
235  */
236 static void reg_to_dbg(struct kvm_vcpu *vcpu,
237 		       struct sys_reg_params *p,
238 		       u64 *dbg_reg)
239 {
240 	u64 val = p->regval;
241 
242 	if (p->is_32bit) {
243 		val &= 0xffffffffUL;
244 		val |= ((*dbg_reg >> 32) << 32);
245 	}
246 
247 	*dbg_reg = val;
248 	vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
249 }
250 
251 static void dbg_to_reg(struct kvm_vcpu *vcpu,
252 		       struct sys_reg_params *p,
253 		       u64 *dbg_reg)
254 {
255 	p->regval = *dbg_reg;
256 	if (p->is_32bit)
257 		p->regval &= 0xffffffffUL;
258 }
259 
260 static bool trap_bvr(struct kvm_vcpu *vcpu,
261 		     struct sys_reg_params *p,
262 		     const struct sys_reg_desc *rd)
263 {
264 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
265 
266 	if (p->is_write)
267 		reg_to_dbg(vcpu, p, dbg_reg);
268 	else
269 		dbg_to_reg(vcpu, p, dbg_reg);
270 
271 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
272 
273 	return true;
274 }
275 
276 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
277 		const struct kvm_one_reg *reg, void __user *uaddr)
278 {
279 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
280 
281 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
282 		return -EFAULT;
283 	return 0;
284 }
285 
286 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
287 	const struct kvm_one_reg *reg, void __user *uaddr)
288 {
289 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
290 
291 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
292 		return -EFAULT;
293 	return 0;
294 }
295 
296 static void reset_bvr(struct kvm_vcpu *vcpu,
297 		      const struct sys_reg_desc *rd)
298 {
299 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
300 }
301 
302 static bool trap_bcr(struct kvm_vcpu *vcpu,
303 		     struct sys_reg_params *p,
304 		     const struct sys_reg_desc *rd)
305 {
306 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
307 
308 	if (p->is_write)
309 		reg_to_dbg(vcpu, p, dbg_reg);
310 	else
311 		dbg_to_reg(vcpu, p, dbg_reg);
312 
313 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
314 
315 	return true;
316 }
317 
318 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
319 		const struct kvm_one_reg *reg, void __user *uaddr)
320 {
321 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
322 
323 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
324 		return -EFAULT;
325 
326 	return 0;
327 }
328 
329 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
330 	const struct kvm_one_reg *reg, void __user *uaddr)
331 {
332 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
333 
334 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
335 		return -EFAULT;
336 	return 0;
337 }
338 
339 static void reset_bcr(struct kvm_vcpu *vcpu,
340 		      const struct sys_reg_desc *rd)
341 {
342 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
343 }
344 
345 static bool trap_wvr(struct kvm_vcpu *vcpu,
346 		     struct sys_reg_params *p,
347 		     const struct sys_reg_desc *rd)
348 {
349 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
350 
351 	if (p->is_write)
352 		reg_to_dbg(vcpu, p, dbg_reg);
353 	else
354 		dbg_to_reg(vcpu, p, dbg_reg);
355 
356 	trace_trap_reg(__func__, rd->reg, p->is_write,
357 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
358 
359 	return true;
360 }
361 
362 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
363 		const struct kvm_one_reg *reg, void __user *uaddr)
364 {
365 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
366 
367 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
368 		return -EFAULT;
369 	return 0;
370 }
371 
372 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
373 	const struct kvm_one_reg *reg, void __user *uaddr)
374 {
375 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
376 
377 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
378 		return -EFAULT;
379 	return 0;
380 }
381 
382 static void reset_wvr(struct kvm_vcpu *vcpu,
383 		      const struct sys_reg_desc *rd)
384 {
385 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
386 }
387 
388 static bool trap_wcr(struct kvm_vcpu *vcpu,
389 		     struct sys_reg_params *p,
390 		     const struct sys_reg_desc *rd)
391 {
392 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
393 
394 	if (p->is_write)
395 		reg_to_dbg(vcpu, p, dbg_reg);
396 	else
397 		dbg_to_reg(vcpu, p, dbg_reg);
398 
399 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
400 
401 	return true;
402 }
403 
404 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
405 		const struct kvm_one_reg *reg, void __user *uaddr)
406 {
407 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
408 
409 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
410 		return -EFAULT;
411 	return 0;
412 }
413 
414 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
415 	const struct kvm_one_reg *reg, void __user *uaddr)
416 {
417 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
418 
419 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
420 		return -EFAULT;
421 	return 0;
422 }
423 
424 static void reset_wcr(struct kvm_vcpu *vcpu,
425 		      const struct sys_reg_desc *rd)
426 {
427 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
428 }
429 
430 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
431 {
432 	u64 amair;
433 
434 	asm volatile("mrs %0, amair_el1\n" : "=r" (amair));
435 	vcpu_sys_reg(vcpu, AMAIR_EL1) = amair;
436 }
437 
438 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
439 {
440 	u64 mpidr;
441 
442 	/*
443 	 * Map the vcpu_id into the first three affinity level fields of
444 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
445 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
446 	 * of the GICv3 to be able to address each CPU directly when
447 	 * sending IPIs.
448 	 */
449 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
450 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
451 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
452 	vcpu_sys_reg(vcpu, MPIDR_EL1) = (1ULL << 31) | mpidr;
453 }
454 
455 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
456 {
457 	u64 pmcr, val;
458 
459 	asm volatile("mrs %0, pmcr_el0\n" : "=r" (pmcr));
460 	/* Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) is reset to UNKNOWN
461 	 * except PMCR.E resetting to zero.
462 	 */
463 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
464 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
465 	vcpu_sys_reg(vcpu, PMCR_EL0) = val;
466 }
467 
468 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
469 {
470 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
471 
472 	return !((reg & ARMV8_PMU_USERENR_EN) || vcpu_mode_priv(vcpu));
473 }
474 
475 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
476 {
477 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
478 
479 	return !((reg & (ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN))
480 		 || vcpu_mode_priv(vcpu));
481 }
482 
483 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
484 {
485 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
486 
487 	return !((reg & (ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN))
488 		 || vcpu_mode_priv(vcpu));
489 }
490 
491 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
492 {
493 	u64 reg = vcpu_sys_reg(vcpu, PMUSERENR_EL0);
494 
495 	return !((reg & (ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN))
496 		 || vcpu_mode_priv(vcpu));
497 }
498 
499 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
500 			const struct sys_reg_desc *r)
501 {
502 	u64 val;
503 
504 	if (!kvm_arm_pmu_v3_ready(vcpu))
505 		return trap_raz_wi(vcpu, p, r);
506 
507 	if (pmu_access_el0_disabled(vcpu))
508 		return false;
509 
510 	if (p->is_write) {
511 		/* Only update writeable bits of PMCR */
512 		val = vcpu_sys_reg(vcpu, PMCR_EL0);
513 		val &= ~ARMV8_PMU_PMCR_MASK;
514 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
515 		vcpu_sys_reg(vcpu, PMCR_EL0) = val;
516 		kvm_pmu_handle_pmcr(vcpu, val);
517 	} else {
518 		/* PMCR.P & PMCR.C are RAZ */
519 		val = vcpu_sys_reg(vcpu, PMCR_EL0)
520 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
521 		p->regval = val;
522 	}
523 
524 	return true;
525 }
526 
527 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
528 			  const struct sys_reg_desc *r)
529 {
530 	if (!kvm_arm_pmu_v3_ready(vcpu))
531 		return trap_raz_wi(vcpu, p, r);
532 
533 	if (pmu_access_event_counter_el0_disabled(vcpu))
534 		return false;
535 
536 	if (p->is_write)
537 		vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
538 	else
539 		/* return PMSELR.SEL field */
540 		p->regval = vcpu_sys_reg(vcpu, PMSELR_EL0)
541 			    & ARMV8_PMU_COUNTER_MASK;
542 
543 	return true;
544 }
545 
546 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
547 			  const struct sys_reg_desc *r)
548 {
549 	u64 pmceid;
550 
551 	if (!kvm_arm_pmu_v3_ready(vcpu))
552 		return trap_raz_wi(vcpu, p, r);
553 
554 	BUG_ON(p->is_write);
555 
556 	if (pmu_access_el0_disabled(vcpu))
557 		return false;
558 
559 	if (!(p->Op2 & 1))
560 		asm volatile("mrs %0, pmceid0_el0\n" : "=r" (pmceid));
561 	else
562 		asm volatile("mrs %0, pmceid1_el0\n" : "=r" (pmceid));
563 
564 	p->regval = pmceid;
565 
566 	return true;
567 }
568 
569 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
570 {
571 	u64 pmcr, val;
572 
573 	pmcr = vcpu_sys_reg(vcpu, PMCR_EL0);
574 	val = (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
575 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX)
576 		return false;
577 
578 	return true;
579 }
580 
581 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
582 			      struct sys_reg_params *p,
583 			      const struct sys_reg_desc *r)
584 {
585 	u64 idx;
586 
587 	if (!kvm_arm_pmu_v3_ready(vcpu))
588 		return trap_raz_wi(vcpu, p, r);
589 
590 	if (r->CRn == 9 && r->CRm == 13) {
591 		if (r->Op2 == 2) {
592 			/* PMXEVCNTR_EL0 */
593 			if (pmu_access_event_counter_el0_disabled(vcpu))
594 				return false;
595 
596 			idx = vcpu_sys_reg(vcpu, PMSELR_EL0)
597 			      & ARMV8_PMU_COUNTER_MASK;
598 		} else if (r->Op2 == 0) {
599 			/* PMCCNTR_EL0 */
600 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
601 				return false;
602 
603 			idx = ARMV8_PMU_CYCLE_IDX;
604 		} else {
605 			BUG();
606 		}
607 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
608 		/* PMEVCNTRn_EL0 */
609 		if (pmu_access_event_counter_el0_disabled(vcpu))
610 			return false;
611 
612 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
613 	} else {
614 		BUG();
615 	}
616 
617 	if (!pmu_counter_idx_valid(vcpu, idx))
618 		return false;
619 
620 	if (p->is_write) {
621 		if (pmu_access_el0_disabled(vcpu))
622 			return false;
623 
624 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
625 	} else {
626 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
627 	}
628 
629 	return true;
630 }
631 
632 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
633 			       const struct sys_reg_desc *r)
634 {
635 	u64 idx, reg;
636 
637 	if (!kvm_arm_pmu_v3_ready(vcpu))
638 		return trap_raz_wi(vcpu, p, r);
639 
640 	if (pmu_access_el0_disabled(vcpu))
641 		return false;
642 
643 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
644 		/* PMXEVTYPER_EL0 */
645 		idx = vcpu_sys_reg(vcpu, PMSELR_EL0) & ARMV8_PMU_COUNTER_MASK;
646 		reg = PMEVTYPER0_EL0 + idx;
647 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
648 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
649 		if (idx == ARMV8_PMU_CYCLE_IDX)
650 			reg = PMCCFILTR_EL0;
651 		else
652 			/* PMEVTYPERn_EL0 */
653 			reg = PMEVTYPER0_EL0 + idx;
654 	} else {
655 		BUG();
656 	}
657 
658 	if (!pmu_counter_idx_valid(vcpu, idx))
659 		return false;
660 
661 	if (p->is_write) {
662 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
663 		vcpu_sys_reg(vcpu, reg) = p->regval & ARMV8_PMU_EVTYPE_MASK;
664 	} else {
665 		p->regval = vcpu_sys_reg(vcpu, reg) & ARMV8_PMU_EVTYPE_MASK;
666 	}
667 
668 	return true;
669 }
670 
671 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
672 			   const struct sys_reg_desc *r)
673 {
674 	u64 val, mask;
675 
676 	if (!kvm_arm_pmu_v3_ready(vcpu))
677 		return trap_raz_wi(vcpu, p, r);
678 
679 	if (pmu_access_el0_disabled(vcpu))
680 		return false;
681 
682 	mask = kvm_pmu_valid_counter_mask(vcpu);
683 	if (p->is_write) {
684 		val = p->regval & mask;
685 		if (r->Op2 & 0x1) {
686 			/* accessing PMCNTENSET_EL0 */
687 			vcpu_sys_reg(vcpu, PMCNTENSET_EL0) |= val;
688 			kvm_pmu_enable_counter(vcpu, val);
689 		} else {
690 			/* accessing PMCNTENCLR_EL0 */
691 			vcpu_sys_reg(vcpu, PMCNTENSET_EL0) &= ~val;
692 			kvm_pmu_disable_counter(vcpu, val);
693 		}
694 	} else {
695 		p->regval = vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & mask;
696 	}
697 
698 	return true;
699 }
700 
701 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
702 			   const struct sys_reg_desc *r)
703 {
704 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
705 
706 	if (!kvm_arm_pmu_v3_ready(vcpu))
707 		return trap_raz_wi(vcpu, p, r);
708 
709 	if (!vcpu_mode_priv(vcpu))
710 		return false;
711 
712 	if (p->is_write) {
713 		u64 val = p->regval & mask;
714 
715 		if (r->Op2 & 0x1)
716 			/* accessing PMINTENSET_EL1 */
717 			vcpu_sys_reg(vcpu, PMINTENSET_EL1) |= val;
718 		else
719 			/* accessing PMINTENCLR_EL1 */
720 			vcpu_sys_reg(vcpu, PMINTENSET_EL1) &= ~val;
721 	} else {
722 		p->regval = vcpu_sys_reg(vcpu, PMINTENSET_EL1) & mask;
723 	}
724 
725 	return true;
726 }
727 
728 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
729 			 const struct sys_reg_desc *r)
730 {
731 	u64 mask = kvm_pmu_valid_counter_mask(vcpu);
732 
733 	if (!kvm_arm_pmu_v3_ready(vcpu))
734 		return trap_raz_wi(vcpu, p, r);
735 
736 	if (pmu_access_el0_disabled(vcpu))
737 		return false;
738 
739 	if (p->is_write) {
740 		if (r->CRm & 0x2)
741 			/* accessing PMOVSSET_EL0 */
742 			kvm_pmu_overflow_set(vcpu, p->regval & mask);
743 		else
744 			/* accessing PMOVSCLR_EL0 */
745 			vcpu_sys_reg(vcpu, PMOVSSET_EL0) &= ~(p->regval & mask);
746 	} else {
747 		p->regval = vcpu_sys_reg(vcpu, PMOVSSET_EL0) & mask;
748 	}
749 
750 	return true;
751 }
752 
753 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
754 			   const struct sys_reg_desc *r)
755 {
756 	u64 mask;
757 
758 	if (!kvm_arm_pmu_v3_ready(vcpu))
759 		return trap_raz_wi(vcpu, p, r);
760 
761 	if (pmu_write_swinc_el0_disabled(vcpu))
762 		return false;
763 
764 	if (p->is_write) {
765 		mask = kvm_pmu_valid_counter_mask(vcpu);
766 		kvm_pmu_software_increment(vcpu, p->regval & mask);
767 		return true;
768 	}
769 
770 	return false;
771 }
772 
773 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
774 			     const struct sys_reg_desc *r)
775 {
776 	if (!kvm_arm_pmu_v3_ready(vcpu))
777 		return trap_raz_wi(vcpu, p, r);
778 
779 	if (p->is_write) {
780 		if (!vcpu_mode_priv(vcpu))
781 			return false;
782 
783 		vcpu_sys_reg(vcpu, PMUSERENR_EL0) = p->regval
784 						    & ARMV8_PMU_USERENR_MASK;
785 	} else {
786 		p->regval = vcpu_sys_reg(vcpu, PMUSERENR_EL0)
787 			    & ARMV8_PMU_USERENR_MASK;
788 	}
789 
790 	return true;
791 }
792 
793 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
794 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
795 	/* DBGBVRn_EL1 */						\
796 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b100),	\
797 	  trap_bvr, reset_bvr, n, 0, get_bvr, set_bvr },		\
798 	/* DBGBCRn_EL1 */						\
799 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b101),	\
800 	  trap_bcr, reset_bcr, n, 0, get_bcr, set_bcr },		\
801 	/* DBGWVRn_EL1 */						\
802 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b110),	\
803 	  trap_wvr, reset_wvr, n, 0,  get_wvr, set_wvr },		\
804 	/* DBGWCRn_EL1 */						\
805 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b111),	\
806 	  trap_wcr, reset_wcr, n, 0,  get_wcr, set_wcr }
807 
808 /* Macro to expand the PMEVCNTRn_EL0 register */
809 #define PMU_PMEVCNTR_EL0(n)						\
810 	/* PMEVCNTRn_EL0 */						\
811 	{ Op0(0b11), Op1(0b011), CRn(0b1110),				\
812 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
813 	  access_pmu_evcntr, reset_unknown, (PMEVCNTR0_EL0 + n), }
814 
815 /* Macro to expand the PMEVTYPERn_EL0 register */
816 #define PMU_PMEVTYPER_EL0(n)						\
817 	/* PMEVTYPERn_EL0 */						\
818 	{ Op0(0b11), Op1(0b011), CRn(0b1110),				\
819 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
820 	  access_pmu_evtyper, reset_unknown, (PMEVTYPER0_EL0 + n), }
821 
822 /*
823  * Architected system registers.
824  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
825  *
826  * We could trap ID_DFR0 and tell the guest we don't support performance
827  * monitoring.  Unfortunately the patch to make the kernel check ID_DFR0 was
828  * NAKed, so it will read the PMCR anyway.
829  *
830  * Therefore we tell the guest we have 0 counters.  Unfortunately, we
831  * must always support PMCCNTR (the cycle counter): we just RAZ/WI for
832  * all PM registers, which doesn't crash the guest kernel at least.
833  *
834  * Debug handling: We do trap most, if not all debug related system
835  * registers. The implementation is good enough to ensure that a guest
836  * can use these with minimal performance degradation. The drawback is
837  * that we don't implement any of the external debug, none of the
838  * OSlock protocol. This should be revisited if we ever encounter a
839  * more demanding guest...
840  */
841 static const struct sys_reg_desc sys_reg_descs[] = {
842 	/* DC ISW */
843 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
844 	  access_dcsw },
845 	/* DC CSW */
846 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
847 	  access_dcsw },
848 	/* DC CISW */
849 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
850 	  access_dcsw },
851 
852 	DBG_BCR_BVR_WCR_WVR_EL1(0),
853 	DBG_BCR_BVR_WCR_WVR_EL1(1),
854 	/* MDCCINT_EL1 */
855 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
856 	  trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
857 	/* MDSCR_EL1 */
858 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
859 	  trap_debug_regs, reset_val, MDSCR_EL1, 0 },
860 	DBG_BCR_BVR_WCR_WVR_EL1(2),
861 	DBG_BCR_BVR_WCR_WVR_EL1(3),
862 	DBG_BCR_BVR_WCR_WVR_EL1(4),
863 	DBG_BCR_BVR_WCR_WVR_EL1(5),
864 	DBG_BCR_BVR_WCR_WVR_EL1(6),
865 	DBG_BCR_BVR_WCR_WVR_EL1(7),
866 	DBG_BCR_BVR_WCR_WVR_EL1(8),
867 	DBG_BCR_BVR_WCR_WVR_EL1(9),
868 	DBG_BCR_BVR_WCR_WVR_EL1(10),
869 	DBG_BCR_BVR_WCR_WVR_EL1(11),
870 	DBG_BCR_BVR_WCR_WVR_EL1(12),
871 	DBG_BCR_BVR_WCR_WVR_EL1(13),
872 	DBG_BCR_BVR_WCR_WVR_EL1(14),
873 	DBG_BCR_BVR_WCR_WVR_EL1(15),
874 
875 	/* MDRAR_EL1 */
876 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
877 	  trap_raz_wi },
878 	/* OSLAR_EL1 */
879 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
880 	  trap_raz_wi },
881 	/* OSLSR_EL1 */
882 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
883 	  trap_oslsr_el1 },
884 	/* OSDLR_EL1 */
885 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
886 	  trap_raz_wi },
887 	/* DBGPRCR_EL1 */
888 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
889 	  trap_raz_wi },
890 	/* DBGCLAIMSET_EL1 */
891 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
892 	  trap_raz_wi },
893 	/* DBGCLAIMCLR_EL1 */
894 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
895 	  trap_raz_wi },
896 	/* DBGAUTHSTATUS_EL1 */
897 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
898 	  trap_dbgauthstatus_el1 },
899 
900 	/* MDCCSR_EL1 */
901 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
902 	  trap_raz_wi },
903 	/* DBGDTR_EL0 */
904 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
905 	  trap_raz_wi },
906 	/* DBGDTR[TR]X_EL0 */
907 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
908 	  trap_raz_wi },
909 
910 	/* DBGVCR32_EL2 */
911 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
912 	  NULL, reset_val, DBGVCR32_EL2, 0 },
913 
914 	/* MPIDR_EL1 */
915 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
916 	  NULL, reset_mpidr, MPIDR_EL1 },
917 	/* SCTLR_EL1 */
918 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
919 	  access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
920 	/* CPACR_EL1 */
921 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
922 	  NULL, reset_val, CPACR_EL1, 0 },
923 	/* TTBR0_EL1 */
924 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
925 	  access_vm_reg, reset_unknown, TTBR0_EL1 },
926 	/* TTBR1_EL1 */
927 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
928 	  access_vm_reg, reset_unknown, TTBR1_EL1 },
929 	/* TCR_EL1 */
930 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
931 	  access_vm_reg, reset_val, TCR_EL1, 0 },
932 
933 	/* AFSR0_EL1 */
934 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
935 	  access_vm_reg, reset_unknown, AFSR0_EL1 },
936 	/* AFSR1_EL1 */
937 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
938 	  access_vm_reg, reset_unknown, AFSR1_EL1 },
939 	/* ESR_EL1 */
940 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
941 	  access_vm_reg, reset_unknown, ESR_EL1 },
942 	/* FAR_EL1 */
943 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
944 	  access_vm_reg, reset_unknown, FAR_EL1 },
945 	/* PAR_EL1 */
946 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
947 	  NULL, reset_unknown, PAR_EL1 },
948 
949 	/* PMINTENSET_EL1 */
950 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
951 	  access_pminten, reset_unknown, PMINTENSET_EL1 },
952 	/* PMINTENCLR_EL1 */
953 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
954 	  access_pminten, NULL, PMINTENSET_EL1 },
955 
956 	/* MAIR_EL1 */
957 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
958 	  access_vm_reg, reset_unknown, MAIR_EL1 },
959 	/* AMAIR_EL1 */
960 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
961 	  access_vm_reg, reset_amair_el1, AMAIR_EL1 },
962 
963 	/* VBAR_EL1 */
964 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
965 	  NULL, reset_val, VBAR_EL1, 0 },
966 
967 	/* ICC_SGI1R_EL1 */
968 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1011), Op2(0b101),
969 	  access_gic_sgi },
970 	/* ICC_SRE_EL1 */
971 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1100), Op2(0b101),
972 	  access_gic_sre },
973 
974 	/* CONTEXTIDR_EL1 */
975 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
976 	  access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
977 	/* TPIDR_EL1 */
978 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
979 	  NULL, reset_unknown, TPIDR_EL1 },
980 
981 	/* CNTKCTL_EL1 */
982 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
983 	  NULL, reset_val, CNTKCTL_EL1, 0},
984 
985 	/* CSSELR_EL1 */
986 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
987 	  NULL, reset_unknown, CSSELR_EL1 },
988 
989 	/* PMCR_EL0 */
990 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
991 	  access_pmcr, reset_pmcr, },
992 	/* PMCNTENSET_EL0 */
993 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
994 	  access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
995 	/* PMCNTENCLR_EL0 */
996 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
997 	  access_pmcnten, NULL, PMCNTENSET_EL0 },
998 	/* PMOVSCLR_EL0 */
999 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
1000 	  access_pmovs, NULL, PMOVSSET_EL0 },
1001 	/* PMSWINC_EL0 */
1002 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
1003 	  access_pmswinc, reset_unknown, PMSWINC_EL0 },
1004 	/* PMSELR_EL0 */
1005 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
1006 	  access_pmselr, reset_unknown, PMSELR_EL0 },
1007 	/* PMCEID0_EL0 */
1008 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
1009 	  access_pmceid },
1010 	/* PMCEID1_EL0 */
1011 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
1012 	  access_pmceid },
1013 	/* PMCCNTR_EL0 */
1014 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
1015 	  access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1016 	/* PMXEVTYPER_EL0 */
1017 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
1018 	  access_pmu_evtyper },
1019 	/* PMXEVCNTR_EL0 */
1020 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
1021 	  access_pmu_evcntr },
1022 	/* PMUSERENR_EL0
1023 	 * This register resets as unknown in 64bit mode while it resets as zero
1024 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1025 	 */
1026 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
1027 	  access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1028 	/* PMOVSSET_EL0 */
1029 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
1030 	  access_pmovs, reset_unknown, PMOVSSET_EL0 },
1031 
1032 	/* TPIDR_EL0 */
1033 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
1034 	  NULL, reset_unknown, TPIDR_EL0 },
1035 	/* TPIDRRO_EL0 */
1036 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
1037 	  NULL, reset_unknown, TPIDRRO_EL0 },
1038 
1039 	/* PMEVCNTRn_EL0 */
1040 	PMU_PMEVCNTR_EL0(0),
1041 	PMU_PMEVCNTR_EL0(1),
1042 	PMU_PMEVCNTR_EL0(2),
1043 	PMU_PMEVCNTR_EL0(3),
1044 	PMU_PMEVCNTR_EL0(4),
1045 	PMU_PMEVCNTR_EL0(5),
1046 	PMU_PMEVCNTR_EL0(6),
1047 	PMU_PMEVCNTR_EL0(7),
1048 	PMU_PMEVCNTR_EL0(8),
1049 	PMU_PMEVCNTR_EL0(9),
1050 	PMU_PMEVCNTR_EL0(10),
1051 	PMU_PMEVCNTR_EL0(11),
1052 	PMU_PMEVCNTR_EL0(12),
1053 	PMU_PMEVCNTR_EL0(13),
1054 	PMU_PMEVCNTR_EL0(14),
1055 	PMU_PMEVCNTR_EL0(15),
1056 	PMU_PMEVCNTR_EL0(16),
1057 	PMU_PMEVCNTR_EL0(17),
1058 	PMU_PMEVCNTR_EL0(18),
1059 	PMU_PMEVCNTR_EL0(19),
1060 	PMU_PMEVCNTR_EL0(20),
1061 	PMU_PMEVCNTR_EL0(21),
1062 	PMU_PMEVCNTR_EL0(22),
1063 	PMU_PMEVCNTR_EL0(23),
1064 	PMU_PMEVCNTR_EL0(24),
1065 	PMU_PMEVCNTR_EL0(25),
1066 	PMU_PMEVCNTR_EL0(26),
1067 	PMU_PMEVCNTR_EL0(27),
1068 	PMU_PMEVCNTR_EL0(28),
1069 	PMU_PMEVCNTR_EL0(29),
1070 	PMU_PMEVCNTR_EL0(30),
1071 	/* PMEVTYPERn_EL0 */
1072 	PMU_PMEVTYPER_EL0(0),
1073 	PMU_PMEVTYPER_EL0(1),
1074 	PMU_PMEVTYPER_EL0(2),
1075 	PMU_PMEVTYPER_EL0(3),
1076 	PMU_PMEVTYPER_EL0(4),
1077 	PMU_PMEVTYPER_EL0(5),
1078 	PMU_PMEVTYPER_EL0(6),
1079 	PMU_PMEVTYPER_EL0(7),
1080 	PMU_PMEVTYPER_EL0(8),
1081 	PMU_PMEVTYPER_EL0(9),
1082 	PMU_PMEVTYPER_EL0(10),
1083 	PMU_PMEVTYPER_EL0(11),
1084 	PMU_PMEVTYPER_EL0(12),
1085 	PMU_PMEVTYPER_EL0(13),
1086 	PMU_PMEVTYPER_EL0(14),
1087 	PMU_PMEVTYPER_EL0(15),
1088 	PMU_PMEVTYPER_EL0(16),
1089 	PMU_PMEVTYPER_EL0(17),
1090 	PMU_PMEVTYPER_EL0(18),
1091 	PMU_PMEVTYPER_EL0(19),
1092 	PMU_PMEVTYPER_EL0(20),
1093 	PMU_PMEVTYPER_EL0(21),
1094 	PMU_PMEVTYPER_EL0(22),
1095 	PMU_PMEVTYPER_EL0(23),
1096 	PMU_PMEVTYPER_EL0(24),
1097 	PMU_PMEVTYPER_EL0(25),
1098 	PMU_PMEVTYPER_EL0(26),
1099 	PMU_PMEVTYPER_EL0(27),
1100 	PMU_PMEVTYPER_EL0(28),
1101 	PMU_PMEVTYPER_EL0(29),
1102 	PMU_PMEVTYPER_EL0(30),
1103 	/* PMCCFILTR_EL0
1104 	 * This register resets as unknown in 64bit mode while it resets as zero
1105 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1106 	 */
1107 	{ Op0(0b11), Op1(0b011), CRn(0b1110), CRm(0b1111), Op2(0b111),
1108 	  access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1109 
1110 	/* DACR32_EL2 */
1111 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
1112 	  NULL, reset_unknown, DACR32_EL2 },
1113 	/* IFSR32_EL2 */
1114 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
1115 	  NULL, reset_unknown, IFSR32_EL2 },
1116 	/* FPEXC32_EL2 */
1117 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
1118 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
1119 };
1120 
1121 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1122 			struct sys_reg_params *p,
1123 			const struct sys_reg_desc *r)
1124 {
1125 	if (p->is_write) {
1126 		return ignore_write(vcpu, p);
1127 	} else {
1128 		u64 dfr = read_system_reg(SYS_ID_AA64DFR0_EL1);
1129 		u64 pfr = read_system_reg(SYS_ID_AA64PFR0_EL1);
1130 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1131 
1132 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1133 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1134 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1135 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1136 		return true;
1137 	}
1138 }
1139 
1140 static bool trap_debug32(struct kvm_vcpu *vcpu,
1141 			 struct sys_reg_params *p,
1142 			 const struct sys_reg_desc *r)
1143 {
1144 	if (p->is_write) {
1145 		vcpu_cp14(vcpu, r->reg) = p->regval;
1146 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
1147 	} else {
1148 		p->regval = vcpu_cp14(vcpu, r->reg);
1149 	}
1150 
1151 	return true;
1152 }
1153 
1154 /* AArch32 debug register mappings
1155  *
1156  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1157  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1158  *
1159  * All control registers and watchpoint value registers are mapped to
1160  * the lower 32 bits of their AArch64 equivalents. We share the trap
1161  * handlers with the above AArch64 code which checks what mode the
1162  * system is in.
1163  */
1164 
1165 static bool trap_xvr(struct kvm_vcpu *vcpu,
1166 		     struct sys_reg_params *p,
1167 		     const struct sys_reg_desc *rd)
1168 {
1169 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1170 
1171 	if (p->is_write) {
1172 		u64 val = *dbg_reg;
1173 
1174 		val &= 0xffffffffUL;
1175 		val |= p->regval << 32;
1176 		*dbg_reg = val;
1177 
1178 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
1179 	} else {
1180 		p->regval = *dbg_reg >> 32;
1181 	}
1182 
1183 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1184 
1185 	return true;
1186 }
1187 
1188 #define DBG_BCR_BVR_WCR_WVR(n)						\
1189 	/* DBGBVRn */							\
1190 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1191 	/* DBGBCRn */							\
1192 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1193 	/* DBGWVRn */							\
1194 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1195 	/* DBGWCRn */							\
1196 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1197 
1198 #define DBGBXVR(n)							\
1199 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1200 
1201 /*
1202  * Trapped cp14 registers. We generally ignore most of the external
1203  * debug, on the principle that they don't really make sense to a
1204  * guest. Revisit this one day, would this principle change.
1205  */
1206 static const struct sys_reg_desc cp14_regs[] = {
1207 	/* DBGIDR */
1208 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1209 	/* DBGDTRRXext */
1210 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1211 
1212 	DBG_BCR_BVR_WCR_WVR(0),
1213 	/* DBGDSCRint */
1214 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1215 	DBG_BCR_BVR_WCR_WVR(1),
1216 	/* DBGDCCINT */
1217 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1218 	/* DBGDSCRext */
1219 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1220 	DBG_BCR_BVR_WCR_WVR(2),
1221 	/* DBGDTR[RT]Xint */
1222 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1223 	/* DBGDTR[RT]Xext */
1224 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1225 	DBG_BCR_BVR_WCR_WVR(3),
1226 	DBG_BCR_BVR_WCR_WVR(4),
1227 	DBG_BCR_BVR_WCR_WVR(5),
1228 	/* DBGWFAR */
1229 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1230 	/* DBGOSECCR */
1231 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1232 	DBG_BCR_BVR_WCR_WVR(6),
1233 	/* DBGVCR */
1234 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1235 	DBG_BCR_BVR_WCR_WVR(7),
1236 	DBG_BCR_BVR_WCR_WVR(8),
1237 	DBG_BCR_BVR_WCR_WVR(9),
1238 	DBG_BCR_BVR_WCR_WVR(10),
1239 	DBG_BCR_BVR_WCR_WVR(11),
1240 	DBG_BCR_BVR_WCR_WVR(12),
1241 	DBG_BCR_BVR_WCR_WVR(13),
1242 	DBG_BCR_BVR_WCR_WVR(14),
1243 	DBG_BCR_BVR_WCR_WVR(15),
1244 
1245 	/* DBGDRAR (32bit) */
1246 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1247 
1248 	DBGBXVR(0),
1249 	/* DBGOSLAR */
1250 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1251 	DBGBXVR(1),
1252 	/* DBGOSLSR */
1253 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1254 	DBGBXVR(2),
1255 	DBGBXVR(3),
1256 	/* DBGOSDLR */
1257 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1258 	DBGBXVR(4),
1259 	/* DBGPRCR */
1260 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1261 	DBGBXVR(5),
1262 	DBGBXVR(6),
1263 	DBGBXVR(7),
1264 	DBGBXVR(8),
1265 	DBGBXVR(9),
1266 	DBGBXVR(10),
1267 	DBGBXVR(11),
1268 	DBGBXVR(12),
1269 	DBGBXVR(13),
1270 	DBGBXVR(14),
1271 	DBGBXVR(15),
1272 
1273 	/* DBGDSAR (32bit) */
1274 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1275 
1276 	/* DBGDEVID2 */
1277 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1278 	/* DBGDEVID1 */
1279 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1280 	/* DBGDEVID */
1281 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1282 	/* DBGCLAIMSET */
1283 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1284 	/* DBGCLAIMCLR */
1285 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1286 	/* DBGAUTHSTATUS */
1287 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1288 };
1289 
1290 /* Trapped cp14 64bit registers */
1291 static const struct sys_reg_desc cp14_64_regs[] = {
1292 	/* DBGDRAR (64bit) */
1293 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1294 
1295 	/* DBGDSAR (64bit) */
1296 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1297 };
1298 
1299 /* Macro to expand the PMEVCNTRn register */
1300 #define PMU_PMEVCNTR(n)							\
1301 	/* PMEVCNTRn */							\
1302 	{ Op1(0), CRn(0b1110),						\
1303 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1304 	  access_pmu_evcntr }
1305 
1306 /* Macro to expand the PMEVTYPERn register */
1307 #define PMU_PMEVTYPER(n)						\
1308 	/* PMEVTYPERn */						\
1309 	{ Op1(0), CRn(0b1110),						\
1310 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1311 	  access_pmu_evtyper }
1312 
1313 /*
1314  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1315  * depending on the way they are accessed (as a 32bit or a 64bit
1316  * register).
1317  */
1318 static const struct sys_reg_desc cp15_regs[] = {
1319 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
1320 
1321 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1322 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1323 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1324 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1325 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1326 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1327 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1328 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1329 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1330 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1331 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1332 
1333 	/*
1334 	 * DC{C,I,CI}SW operations:
1335 	 */
1336 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1337 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1338 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1339 
1340 	/* PMU */
1341 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1342 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1343 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1344 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1345 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1346 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1347 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1348 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1349 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1350 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1351 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1352 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1353 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1354 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1355 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1356 
1357 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
1358 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
1359 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
1360 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
1361 
1362 	/* ICC_SRE */
1363 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), trap_raz_wi },
1364 
1365 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
1366 
1367 	/* PMEVCNTRn */
1368 	PMU_PMEVCNTR(0),
1369 	PMU_PMEVCNTR(1),
1370 	PMU_PMEVCNTR(2),
1371 	PMU_PMEVCNTR(3),
1372 	PMU_PMEVCNTR(4),
1373 	PMU_PMEVCNTR(5),
1374 	PMU_PMEVCNTR(6),
1375 	PMU_PMEVCNTR(7),
1376 	PMU_PMEVCNTR(8),
1377 	PMU_PMEVCNTR(9),
1378 	PMU_PMEVCNTR(10),
1379 	PMU_PMEVCNTR(11),
1380 	PMU_PMEVCNTR(12),
1381 	PMU_PMEVCNTR(13),
1382 	PMU_PMEVCNTR(14),
1383 	PMU_PMEVCNTR(15),
1384 	PMU_PMEVCNTR(16),
1385 	PMU_PMEVCNTR(17),
1386 	PMU_PMEVCNTR(18),
1387 	PMU_PMEVCNTR(19),
1388 	PMU_PMEVCNTR(20),
1389 	PMU_PMEVCNTR(21),
1390 	PMU_PMEVCNTR(22),
1391 	PMU_PMEVCNTR(23),
1392 	PMU_PMEVCNTR(24),
1393 	PMU_PMEVCNTR(25),
1394 	PMU_PMEVCNTR(26),
1395 	PMU_PMEVCNTR(27),
1396 	PMU_PMEVCNTR(28),
1397 	PMU_PMEVCNTR(29),
1398 	PMU_PMEVCNTR(30),
1399 	/* PMEVTYPERn */
1400 	PMU_PMEVTYPER(0),
1401 	PMU_PMEVTYPER(1),
1402 	PMU_PMEVTYPER(2),
1403 	PMU_PMEVTYPER(3),
1404 	PMU_PMEVTYPER(4),
1405 	PMU_PMEVTYPER(5),
1406 	PMU_PMEVTYPER(6),
1407 	PMU_PMEVTYPER(7),
1408 	PMU_PMEVTYPER(8),
1409 	PMU_PMEVTYPER(9),
1410 	PMU_PMEVTYPER(10),
1411 	PMU_PMEVTYPER(11),
1412 	PMU_PMEVTYPER(12),
1413 	PMU_PMEVTYPER(13),
1414 	PMU_PMEVTYPER(14),
1415 	PMU_PMEVTYPER(15),
1416 	PMU_PMEVTYPER(16),
1417 	PMU_PMEVTYPER(17),
1418 	PMU_PMEVTYPER(18),
1419 	PMU_PMEVTYPER(19),
1420 	PMU_PMEVTYPER(20),
1421 	PMU_PMEVTYPER(21),
1422 	PMU_PMEVTYPER(22),
1423 	PMU_PMEVTYPER(23),
1424 	PMU_PMEVTYPER(24),
1425 	PMU_PMEVTYPER(25),
1426 	PMU_PMEVTYPER(26),
1427 	PMU_PMEVTYPER(27),
1428 	PMU_PMEVTYPER(28),
1429 	PMU_PMEVTYPER(29),
1430 	PMU_PMEVTYPER(30),
1431 	/* PMCCFILTR */
1432 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
1433 };
1434 
1435 static const struct sys_reg_desc cp15_64_regs[] = {
1436 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1437 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
1438 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
1439 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
1440 };
1441 
1442 /* Target specific emulation tables */
1443 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
1444 
1445 void kvm_register_target_sys_reg_table(unsigned int target,
1446 				       struct kvm_sys_reg_target_table *table)
1447 {
1448 	target_tables[target] = table;
1449 }
1450 
1451 /* Get specific register table for this target. */
1452 static const struct sys_reg_desc *get_target_table(unsigned target,
1453 						   bool mode_is_64,
1454 						   size_t *num)
1455 {
1456 	struct kvm_sys_reg_target_table *table;
1457 
1458 	table = target_tables[target];
1459 	if (mode_is_64) {
1460 		*num = table->table64.num;
1461 		return table->table64.table;
1462 	} else {
1463 		*num = table->table32.num;
1464 		return table->table32.table;
1465 	}
1466 }
1467 
1468 #define reg_to_match_value(x)						\
1469 	({								\
1470 		unsigned long val;					\
1471 		val  = (x)->Op0 << 14;					\
1472 		val |= (x)->Op1 << 11;					\
1473 		val |= (x)->CRn << 7;					\
1474 		val |= (x)->CRm << 3;					\
1475 		val |= (x)->Op2;					\
1476 		val;							\
1477 	 })
1478 
1479 static int match_sys_reg(const void *key, const void *elt)
1480 {
1481 	const unsigned long pval = (unsigned long)key;
1482 	const struct sys_reg_desc *r = elt;
1483 
1484 	return pval - reg_to_match_value(r);
1485 }
1486 
1487 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
1488 					 const struct sys_reg_desc table[],
1489 					 unsigned int num)
1490 {
1491 	unsigned long pval = reg_to_match_value(params);
1492 
1493 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
1494 }
1495 
1496 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
1497 {
1498 	kvm_inject_undefined(vcpu);
1499 	return 1;
1500 }
1501 
1502 /*
1503  * emulate_cp --  tries to match a sys_reg access in a handling table, and
1504  *                call the corresponding trap handler.
1505  *
1506  * @params: pointer to the descriptor of the access
1507  * @table: array of trap descriptors
1508  * @num: size of the trap descriptor array
1509  *
1510  * Return 0 if the access has been handled, and -1 if not.
1511  */
1512 static int emulate_cp(struct kvm_vcpu *vcpu,
1513 		      struct sys_reg_params *params,
1514 		      const struct sys_reg_desc *table,
1515 		      size_t num)
1516 {
1517 	const struct sys_reg_desc *r;
1518 
1519 	if (!table)
1520 		return -1;	/* Not handled */
1521 
1522 	r = find_reg(params, table, num);
1523 
1524 	if (r) {
1525 		/*
1526 		 * Not having an accessor means that we have
1527 		 * configured a trap that we don't know how to
1528 		 * handle. This certainly qualifies as a gross bug
1529 		 * that should be fixed right away.
1530 		 */
1531 		BUG_ON(!r->access);
1532 
1533 		if (likely(r->access(vcpu, params, r))) {
1534 			/* Skip instruction, since it was emulated */
1535 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1536 			/* Handled */
1537 			return 0;
1538 		}
1539 	}
1540 
1541 	/* Not handled */
1542 	return -1;
1543 }
1544 
1545 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
1546 				struct sys_reg_params *params)
1547 {
1548 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
1549 	int cp;
1550 
1551 	switch(hsr_ec) {
1552 	case ESR_ELx_EC_CP15_32:
1553 	case ESR_ELx_EC_CP15_64:
1554 		cp = 15;
1555 		break;
1556 	case ESR_ELx_EC_CP14_MR:
1557 	case ESR_ELx_EC_CP14_64:
1558 		cp = 14;
1559 		break;
1560 	default:
1561 		WARN_ON((cp = -1));
1562 	}
1563 
1564 	kvm_err("Unsupported guest CP%d access at: %08lx\n",
1565 		cp, *vcpu_pc(vcpu));
1566 	print_sys_reg_instr(params);
1567 	kvm_inject_undefined(vcpu);
1568 }
1569 
1570 /**
1571  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
1572  * @vcpu: The VCPU pointer
1573  * @run:  The kvm_run struct
1574  */
1575 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
1576 			    const struct sys_reg_desc *global,
1577 			    size_t nr_global,
1578 			    const struct sys_reg_desc *target_specific,
1579 			    size_t nr_specific)
1580 {
1581 	struct sys_reg_params params;
1582 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1583 	int Rt = (hsr >> 5) & 0xf;
1584 	int Rt2 = (hsr >> 10) & 0xf;
1585 
1586 	params.is_aarch32 = true;
1587 	params.is_32bit = false;
1588 	params.CRm = (hsr >> 1) & 0xf;
1589 	params.is_write = ((hsr & 1) == 0);
1590 
1591 	params.Op0 = 0;
1592 	params.Op1 = (hsr >> 16) & 0xf;
1593 	params.Op2 = 0;
1594 	params.CRn = 0;
1595 
1596 	/*
1597 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
1598 	 * backends between AArch32 and AArch64, we get away with it.
1599 	 */
1600 	if (params.is_write) {
1601 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
1602 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
1603 	}
1604 
1605 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
1606 		goto out;
1607 	if (!emulate_cp(vcpu, &params, global, nr_global))
1608 		goto out;
1609 
1610 	unhandled_cp_access(vcpu, &params);
1611 
1612 out:
1613 	/* Split up the value between registers for the read side */
1614 	if (!params.is_write) {
1615 		vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
1616 		vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
1617 	}
1618 
1619 	return 1;
1620 }
1621 
1622 /**
1623  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
1624  * @vcpu: The VCPU pointer
1625  * @run:  The kvm_run struct
1626  */
1627 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
1628 			    const struct sys_reg_desc *global,
1629 			    size_t nr_global,
1630 			    const struct sys_reg_desc *target_specific,
1631 			    size_t nr_specific)
1632 {
1633 	struct sys_reg_params params;
1634 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1635 	int Rt  = (hsr >> 5) & 0xf;
1636 
1637 	params.is_aarch32 = true;
1638 	params.is_32bit = true;
1639 	params.CRm = (hsr >> 1) & 0xf;
1640 	params.regval = vcpu_get_reg(vcpu, Rt);
1641 	params.is_write = ((hsr & 1) == 0);
1642 	params.CRn = (hsr >> 10) & 0xf;
1643 	params.Op0 = 0;
1644 	params.Op1 = (hsr >> 14) & 0x7;
1645 	params.Op2 = (hsr >> 17) & 0x7;
1646 
1647 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
1648 	    !emulate_cp(vcpu, &params, global, nr_global)) {
1649 		if (!params.is_write)
1650 			vcpu_set_reg(vcpu, Rt, params.regval);
1651 		return 1;
1652 	}
1653 
1654 	unhandled_cp_access(vcpu, &params);
1655 	return 1;
1656 }
1657 
1658 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1659 {
1660 	const struct sys_reg_desc *target_specific;
1661 	size_t num;
1662 
1663 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1664 	return kvm_handle_cp_64(vcpu,
1665 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
1666 				target_specific, num);
1667 }
1668 
1669 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1670 {
1671 	const struct sys_reg_desc *target_specific;
1672 	size_t num;
1673 
1674 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1675 	return kvm_handle_cp_32(vcpu,
1676 				cp15_regs, ARRAY_SIZE(cp15_regs),
1677 				target_specific, num);
1678 }
1679 
1680 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1681 {
1682 	return kvm_handle_cp_64(vcpu,
1683 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
1684 				NULL, 0);
1685 }
1686 
1687 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1688 {
1689 	return kvm_handle_cp_32(vcpu,
1690 				cp14_regs, ARRAY_SIZE(cp14_regs),
1691 				NULL, 0);
1692 }
1693 
1694 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
1695 			   struct sys_reg_params *params)
1696 {
1697 	size_t num;
1698 	const struct sys_reg_desc *table, *r;
1699 
1700 	table = get_target_table(vcpu->arch.target, true, &num);
1701 
1702 	/* Search target-specific then generic table. */
1703 	r = find_reg(params, table, num);
1704 	if (!r)
1705 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1706 
1707 	if (likely(r)) {
1708 		/*
1709 		 * Not having an accessor means that we have
1710 		 * configured a trap that we don't know how to
1711 		 * handle. This certainly qualifies as a gross bug
1712 		 * that should be fixed right away.
1713 		 */
1714 		BUG_ON(!r->access);
1715 
1716 		if (likely(r->access(vcpu, params, r))) {
1717 			/* Skip instruction, since it was emulated */
1718 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1719 			return 1;
1720 		}
1721 		/* If access function fails, it should complain. */
1722 	} else {
1723 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
1724 			*vcpu_pc(vcpu));
1725 		print_sys_reg_instr(params);
1726 	}
1727 	kvm_inject_undefined(vcpu);
1728 	return 1;
1729 }
1730 
1731 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
1732 			      const struct sys_reg_desc *table, size_t num)
1733 {
1734 	unsigned long i;
1735 
1736 	for (i = 0; i < num; i++)
1737 		if (table[i].reset)
1738 			table[i].reset(vcpu, &table[i]);
1739 }
1740 
1741 /**
1742  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
1743  * @vcpu: The VCPU pointer
1744  * @run:  The kvm_run struct
1745  */
1746 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1747 {
1748 	struct sys_reg_params params;
1749 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1750 	int Rt = (esr >> 5) & 0x1f;
1751 	int ret;
1752 
1753 	trace_kvm_handle_sys_reg(esr);
1754 
1755 	params.is_aarch32 = false;
1756 	params.is_32bit = false;
1757 	params.Op0 = (esr >> 20) & 3;
1758 	params.Op1 = (esr >> 14) & 0x7;
1759 	params.CRn = (esr >> 10) & 0xf;
1760 	params.CRm = (esr >> 1) & 0xf;
1761 	params.Op2 = (esr >> 17) & 0x7;
1762 	params.regval = vcpu_get_reg(vcpu, Rt);
1763 	params.is_write = !(esr & 1);
1764 
1765 	ret = emulate_sys_reg(vcpu, &params);
1766 
1767 	if (!params.is_write)
1768 		vcpu_set_reg(vcpu, Rt, params.regval);
1769 	return ret;
1770 }
1771 
1772 /******************************************************************************
1773  * Userspace API
1774  *****************************************************************************/
1775 
1776 static bool index_to_params(u64 id, struct sys_reg_params *params)
1777 {
1778 	switch (id & KVM_REG_SIZE_MASK) {
1779 	case KVM_REG_SIZE_U64:
1780 		/* Any unused index bits means it's not valid. */
1781 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1782 			      | KVM_REG_ARM_COPROC_MASK
1783 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
1784 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
1785 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
1786 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
1787 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
1788 			return false;
1789 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1790 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1791 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1792 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1793 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1794 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1795 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1796 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1797 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1798 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1799 		return true;
1800 	default:
1801 		return false;
1802 	}
1803 }
1804 
1805 /* Decode an index value, and find the sys_reg_desc entry. */
1806 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1807 						    u64 id)
1808 {
1809 	size_t num;
1810 	const struct sys_reg_desc *table, *r;
1811 	struct sys_reg_params params;
1812 
1813 	/* We only do sys_reg for now. */
1814 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1815 		return NULL;
1816 
1817 	if (!index_to_params(id, &params))
1818 		return NULL;
1819 
1820 	table = get_target_table(vcpu->arch.target, true, &num);
1821 	r = find_reg(&params, table, num);
1822 	if (!r)
1823 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1824 
1825 	/* Not saved in the sys_reg array? */
1826 	if (r && !r->reg)
1827 		r = NULL;
1828 
1829 	return r;
1830 }
1831 
1832 /*
1833  * These are the invariant sys_reg registers: we let the guest see the
1834  * host versions of these, so they're part of the guest state.
1835  *
1836  * A future CPU may provide a mechanism to present different values to
1837  * the guest, or a future kvm may trap them.
1838  */
1839 
1840 #define FUNCTION_INVARIANT(reg)						\
1841 	static void get_##reg(struct kvm_vcpu *v,			\
1842 			      const struct sys_reg_desc *r)		\
1843 	{								\
1844 		u64 val;						\
1845 									\
1846 		asm volatile("mrs %0, " __stringify(reg) "\n"		\
1847 			     : "=r" (val));				\
1848 		((struct sys_reg_desc *)r)->val = val;			\
1849 	}
1850 
1851 FUNCTION_INVARIANT(midr_el1)
1852 FUNCTION_INVARIANT(ctr_el0)
1853 FUNCTION_INVARIANT(revidr_el1)
1854 FUNCTION_INVARIANT(id_pfr0_el1)
1855 FUNCTION_INVARIANT(id_pfr1_el1)
1856 FUNCTION_INVARIANT(id_dfr0_el1)
1857 FUNCTION_INVARIANT(id_afr0_el1)
1858 FUNCTION_INVARIANT(id_mmfr0_el1)
1859 FUNCTION_INVARIANT(id_mmfr1_el1)
1860 FUNCTION_INVARIANT(id_mmfr2_el1)
1861 FUNCTION_INVARIANT(id_mmfr3_el1)
1862 FUNCTION_INVARIANT(id_isar0_el1)
1863 FUNCTION_INVARIANT(id_isar1_el1)
1864 FUNCTION_INVARIANT(id_isar2_el1)
1865 FUNCTION_INVARIANT(id_isar3_el1)
1866 FUNCTION_INVARIANT(id_isar4_el1)
1867 FUNCTION_INVARIANT(id_isar5_el1)
1868 FUNCTION_INVARIANT(clidr_el1)
1869 FUNCTION_INVARIANT(aidr_el1)
1870 
1871 /* ->val is filled in by kvm_sys_reg_table_init() */
1872 static struct sys_reg_desc invariant_sys_regs[] = {
1873 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1874 	  NULL, get_midr_el1 },
1875 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1876 	  NULL, get_revidr_el1 },
1877 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1878 	  NULL, get_id_pfr0_el1 },
1879 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1880 	  NULL, get_id_pfr1_el1 },
1881 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1882 	  NULL, get_id_dfr0_el1 },
1883 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1884 	  NULL, get_id_afr0_el1 },
1885 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1886 	  NULL, get_id_mmfr0_el1 },
1887 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1888 	  NULL, get_id_mmfr1_el1 },
1889 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1890 	  NULL, get_id_mmfr2_el1 },
1891 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1892 	  NULL, get_id_mmfr3_el1 },
1893 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1894 	  NULL, get_id_isar0_el1 },
1895 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1896 	  NULL, get_id_isar1_el1 },
1897 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1898 	  NULL, get_id_isar2_el1 },
1899 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1900 	  NULL, get_id_isar3_el1 },
1901 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1902 	  NULL, get_id_isar4_el1 },
1903 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1904 	  NULL, get_id_isar5_el1 },
1905 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1906 	  NULL, get_clidr_el1 },
1907 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1908 	  NULL, get_aidr_el1 },
1909 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1910 	  NULL, get_ctr_el0 },
1911 };
1912 
1913 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1914 {
1915 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1916 		return -EFAULT;
1917 	return 0;
1918 }
1919 
1920 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1921 {
1922 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1923 		return -EFAULT;
1924 	return 0;
1925 }
1926 
1927 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1928 {
1929 	struct sys_reg_params params;
1930 	const struct sys_reg_desc *r;
1931 
1932 	if (!index_to_params(id, &params))
1933 		return -ENOENT;
1934 
1935 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1936 	if (!r)
1937 		return -ENOENT;
1938 
1939 	return reg_to_user(uaddr, &r->val, id);
1940 }
1941 
1942 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1943 {
1944 	struct sys_reg_params params;
1945 	const struct sys_reg_desc *r;
1946 	int err;
1947 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1948 
1949 	if (!index_to_params(id, &params))
1950 		return -ENOENT;
1951 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1952 	if (!r)
1953 		return -ENOENT;
1954 
1955 	err = reg_from_user(&val, uaddr, id);
1956 	if (err)
1957 		return err;
1958 
1959 	/* This is what we mean by invariant: you can't change it. */
1960 	if (r->val != val)
1961 		return -EINVAL;
1962 
1963 	return 0;
1964 }
1965 
1966 static bool is_valid_cache(u32 val)
1967 {
1968 	u32 level, ctype;
1969 
1970 	if (val >= CSSELR_MAX)
1971 		return false;
1972 
1973 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1974 	level = (val >> 1);
1975 	ctype = (cache_levels >> (level * 3)) & 7;
1976 
1977 	switch (ctype) {
1978 	case 0: /* No cache */
1979 		return false;
1980 	case 1: /* Instruction cache only */
1981 		return (val & 1);
1982 	case 2: /* Data cache only */
1983 	case 4: /* Unified cache */
1984 		return !(val & 1);
1985 	case 3: /* Separate instruction and data caches */
1986 		return true;
1987 	default: /* Reserved: we can't know instruction or data. */
1988 		return false;
1989 	}
1990 }
1991 
1992 static int demux_c15_get(u64 id, void __user *uaddr)
1993 {
1994 	u32 val;
1995 	u32 __user *uval = uaddr;
1996 
1997 	/* Fail if we have unknown bits set. */
1998 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1999 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2000 		return -ENOENT;
2001 
2002 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2003 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2004 		if (KVM_REG_SIZE(id) != 4)
2005 			return -ENOENT;
2006 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2007 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2008 		if (!is_valid_cache(val))
2009 			return -ENOENT;
2010 
2011 		return put_user(get_ccsidr(val), uval);
2012 	default:
2013 		return -ENOENT;
2014 	}
2015 }
2016 
2017 static int demux_c15_set(u64 id, void __user *uaddr)
2018 {
2019 	u32 val, newval;
2020 	u32 __user *uval = uaddr;
2021 
2022 	/* Fail if we have unknown bits set. */
2023 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2024 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2025 		return -ENOENT;
2026 
2027 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2028 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2029 		if (KVM_REG_SIZE(id) != 4)
2030 			return -ENOENT;
2031 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2032 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2033 		if (!is_valid_cache(val))
2034 			return -ENOENT;
2035 
2036 		if (get_user(newval, uval))
2037 			return -EFAULT;
2038 
2039 		/* This is also invariant: you can't change it. */
2040 		if (newval != get_ccsidr(val))
2041 			return -EINVAL;
2042 		return 0;
2043 	default:
2044 		return -ENOENT;
2045 	}
2046 }
2047 
2048 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2049 {
2050 	const struct sys_reg_desc *r;
2051 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2052 
2053 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2054 		return demux_c15_get(reg->id, uaddr);
2055 
2056 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2057 		return -ENOENT;
2058 
2059 	r = index_to_sys_reg_desc(vcpu, reg->id);
2060 	if (!r)
2061 		return get_invariant_sys_reg(reg->id, uaddr);
2062 
2063 	if (r->get_user)
2064 		return (r->get_user)(vcpu, r, reg, uaddr);
2065 
2066 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
2067 }
2068 
2069 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2070 {
2071 	const struct sys_reg_desc *r;
2072 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2073 
2074 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2075 		return demux_c15_set(reg->id, uaddr);
2076 
2077 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2078 		return -ENOENT;
2079 
2080 	r = index_to_sys_reg_desc(vcpu, reg->id);
2081 	if (!r)
2082 		return set_invariant_sys_reg(reg->id, uaddr);
2083 
2084 	if (r->set_user)
2085 		return (r->set_user)(vcpu, r, reg, uaddr);
2086 
2087 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2088 }
2089 
2090 static unsigned int num_demux_regs(void)
2091 {
2092 	unsigned int i, count = 0;
2093 
2094 	for (i = 0; i < CSSELR_MAX; i++)
2095 		if (is_valid_cache(i))
2096 			count++;
2097 
2098 	return count;
2099 }
2100 
2101 static int write_demux_regids(u64 __user *uindices)
2102 {
2103 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2104 	unsigned int i;
2105 
2106 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2107 	for (i = 0; i < CSSELR_MAX; i++) {
2108 		if (!is_valid_cache(i))
2109 			continue;
2110 		if (put_user(val | i, uindices))
2111 			return -EFAULT;
2112 		uindices++;
2113 	}
2114 	return 0;
2115 }
2116 
2117 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2118 {
2119 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2120 		KVM_REG_ARM64_SYSREG |
2121 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2122 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2123 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2124 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2125 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2126 }
2127 
2128 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2129 {
2130 	if (!*uind)
2131 		return true;
2132 
2133 	if (put_user(sys_reg_to_index(reg), *uind))
2134 		return false;
2135 
2136 	(*uind)++;
2137 	return true;
2138 }
2139 
2140 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2141 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2142 {
2143 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
2144 	unsigned int total = 0;
2145 	size_t num;
2146 
2147 	/* We check for duplicates here, to allow arch-specific overrides. */
2148 	i1 = get_target_table(vcpu->arch.target, true, &num);
2149 	end1 = i1 + num;
2150 	i2 = sys_reg_descs;
2151 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2152 
2153 	BUG_ON(i1 == end1 || i2 == end2);
2154 
2155 	/* Walk carefully, as both tables may refer to the same register. */
2156 	while (i1 || i2) {
2157 		int cmp = cmp_sys_reg(i1, i2);
2158 		/* target-specific overrides generic entry. */
2159 		if (cmp <= 0) {
2160 			/* Ignore registers we trap but don't save. */
2161 			if (i1->reg) {
2162 				if (!copy_reg_to_user(i1, &uind))
2163 					return -EFAULT;
2164 				total++;
2165 			}
2166 		} else {
2167 			/* Ignore registers we trap but don't save. */
2168 			if (i2->reg) {
2169 				if (!copy_reg_to_user(i2, &uind))
2170 					return -EFAULT;
2171 				total++;
2172 			}
2173 		}
2174 
2175 		if (cmp <= 0 && ++i1 == end1)
2176 			i1 = NULL;
2177 		if (cmp >= 0 && ++i2 == end2)
2178 			i2 = NULL;
2179 	}
2180 	return total;
2181 }
2182 
2183 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2184 {
2185 	return ARRAY_SIZE(invariant_sys_regs)
2186 		+ num_demux_regs()
2187 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2188 }
2189 
2190 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2191 {
2192 	unsigned int i;
2193 	int err;
2194 
2195 	/* Then give them all the invariant registers' indices. */
2196 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2197 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2198 			return -EFAULT;
2199 		uindices++;
2200 	}
2201 
2202 	err = walk_sys_regs(vcpu, uindices);
2203 	if (err < 0)
2204 		return err;
2205 	uindices += err;
2206 
2207 	return write_demux_regids(uindices);
2208 }
2209 
2210 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
2211 {
2212 	unsigned int i;
2213 
2214 	for (i = 1; i < n; i++) {
2215 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2216 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2217 			return 1;
2218 		}
2219 	}
2220 
2221 	return 0;
2222 }
2223 
2224 void kvm_sys_reg_table_init(void)
2225 {
2226 	unsigned int i;
2227 	struct sys_reg_desc clidr;
2228 
2229 	/* Make sure tables are unique and in order. */
2230 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
2231 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
2232 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
2233 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
2234 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
2235 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
2236 
2237 	/* We abuse the reset function to overwrite the table itself. */
2238 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2239 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2240 
2241 	/*
2242 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2243 	 *
2244 	 *   If software reads the Cache Type fields from Ctype1
2245 	 *   upwards, once it has seen a value of 0b000, no caches
2246 	 *   exist at further-out levels of the hierarchy. So, for
2247 	 *   example, if Ctype3 is the first Cache Type field with a
2248 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2249 	 *   ignored.
2250 	 */
2251 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2252 	cache_levels = clidr.val;
2253 	for (i = 0; i < 7; i++)
2254 		if (((cache_levels >> (i*3)) & 7) == 0)
2255 			break;
2256 	/* Clear all higher bits. */
2257 	cache_levels &= (1 << (i*3))-1;
2258 }
2259 
2260 /**
2261  * kvm_reset_sys_regs - sets system registers to reset value
2262  * @vcpu: The VCPU pointer
2263  *
2264  * This function finds the right table above and sets the registers on the
2265  * virtual CPU struct to their architecturally defined reset values.
2266  */
2267 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2268 {
2269 	size_t num;
2270 	const struct sys_reg_desc *table;
2271 
2272 	/* Catch someone adding a register without putting in reset entry. */
2273 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
2274 
2275 	/* Generic chip reset first (so target could override). */
2276 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2277 
2278 	table = get_target_table(vcpu->arch.target, true, &num);
2279 	reset_sys_reg_descs(vcpu, table, num);
2280 
2281 	for (num = 1; num < NR_SYS_REGS; num++)
2282 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
2283 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
2284 }
2285