xref: /linux/arch/arm64/kvm/sys_regs.c (revision 1296bd62b0dcbdc8fd66908eb2d72dc4c3dba219)
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  * Debug handling: We do trap most, if not all debug related system
827  * registers. The implementation is good enough to ensure that a guest
828  * can use these with minimal performance degradation. The drawback is
829  * that we don't implement any of the external debug, none of the
830  * OSlock protocol. This should be revisited if we ever encounter a
831  * more demanding guest...
832  */
833 static const struct sys_reg_desc sys_reg_descs[] = {
834 	/* DC ISW */
835 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
836 	  access_dcsw },
837 	/* DC CSW */
838 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
839 	  access_dcsw },
840 	/* DC CISW */
841 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
842 	  access_dcsw },
843 
844 	DBG_BCR_BVR_WCR_WVR_EL1(0),
845 	DBG_BCR_BVR_WCR_WVR_EL1(1),
846 	/* MDCCINT_EL1 */
847 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
848 	  trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
849 	/* MDSCR_EL1 */
850 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
851 	  trap_debug_regs, reset_val, MDSCR_EL1, 0 },
852 	DBG_BCR_BVR_WCR_WVR_EL1(2),
853 	DBG_BCR_BVR_WCR_WVR_EL1(3),
854 	DBG_BCR_BVR_WCR_WVR_EL1(4),
855 	DBG_BCR_BVR_WCR_WVR_EL1(5),
856 	DBG_BCR_BVR_WCR_WVR_EL1(6),
857 	DBG_BCR_BVR_WCR_WVR_EL1(7),
858 	DBG_BCR_BVR_WCR_WVR_EL1(8),
859 	DBG_BCR_BVR_WCR_WVR_EL1(9),
860 	DBG_BCR_BVR_WCR_WVR_EL1(10),
861 	DBG_BCR_BVR_WCR_WVR_EL1(11),
862 	DBG_BCR_BVR_WCR_WVR_EL1(12),
863 	DBG_BCR_BVR_WCR_WVR_EL1(13),
864 	DBG_BCR_BVR_WCR_WVR_EL1(14),
865 	DBG_BCR_BVR_WCR_WVR_EL1(15),
866 
867 	/* MDRAR_EL1 */
868 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
869 	  trap_raz_wi },
870 	/* OSLAR_EL1 */
871 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
872 	  trap_raz_wi },
873 	/* OSLSR_EL1 */
874 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
875 	  trap_oslsr_el1 },
876 	/* OSDLR_EL1 */
877 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
878 	  trap_raz_wi },
879 	/* DBGPRCR_EL1 */
880 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
881 	  trap_raz_wi },
882 	/* DBGCLAIMSET_EL1 */
883 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
884 	  trap_raz_wi },
885 	/* DBGCLAIMCLR_EL1 */
886 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
887 	  trap_raz_wi },
888 	/* DBGAUTHSTATUS_EL1 */
889 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
890 	  trap_dbgauthstatus_el1 },
891 
892 	/* MDCCSR_EL1 */
893 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
894 	  trap_raz_wi },
895 	/* DBGDTR_EL0 */
896 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
897 	  trap_raz_wi },
898 	/* DBGDTR[TR]X_EL0 */
899 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
900 	  trap_raz_wi },
901 
902 	/* DBGVCR32_EL2 */
903 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
904 	  NULL, reset_val, DBGVCR32_EL2, 0 },
905 
906 	/* MPIDR_EL1 */
907 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
908 	  NULL, reset_mpidr, MPIDR_EL1 },
909 	/* SCTLR_EL1 */
910 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
911 	  access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
912 	/* CPACR_EL1 */
913 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
914 	  NULL, reset_val, CPACR_EL1, 0 },
915 	/* TTBR0_EL1 */
916 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
917 	  access_vm_reg, reset_unknown, TTBR0_EL1 },
918 	/* TTBR1_EL1 */
919 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
920 	  access_vm_reg, reset_unknown, TTBR1_EL1 },
921 	/* TCR_EL1 */
922 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
923 	  access_vm_reg, reset_val, TCR_EL1, 0 },
924 
925 	/* AFSR0_EL1 */
926 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
927 	  access_vm_reg, reset_unknown, AFSR0_EL1 },
928 	/* AFSR1_EL1 */
929 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
930 	  access_vm_reg, reset_unknown, AFSR1_EL1 },
931 	/* ESR_EL1 */
932 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
933 	  access_vm_reg, reset_unknown, ESR_EL1 },
934 	/* FAR_EL1 */
935 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
936 	  access_vm_reg, reset_unknown, FAR_EL1 },
937 	/* PAR_EL1 */
938 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
939 	  NULL, reset_unknown, PAR_EL1 },
940 
941 	/* PMINTENSET_EL1 */
942 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
943 	  access_pminten, reset_unknown, PMINTENSET_EL1 },
944 	/* PMINTENCLR_EL1 */
945 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
946 	  access_pminten, NULL, PMINTENSET_EL1 },
947 
948 	/* MAIR_EL1 */
949 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
950 	  access_vm_reg, reset_unknown, MAIR_EL1 },
951 	/* AMAIR_EL1 */
952 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
953 	  access_vm_reg, reset_amair_el1, AMAIR_EL1 },
954 
955 	/* VBAR_EL1 */
956 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
957 	  NULL, reset_val, VBAR_EL1, 0 },
958 
959 	/* ICC_SGI1R_EL1 */
960 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1011), Op2(0b101),
961 	  access_gic_sgi },
962 	/* ICC_SRE_EL1 */
963 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1100), Op2(0b101),
964 	  access_gic_sre },
965 
966 	/* CONTEXTIDR_EL1 */
967 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
968 	  access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
969 	/* TPIDR_EL1 */
970 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
971 	  NULL, reset_unknown, TPIDR_EL1 },
972 
973 	/* CNTKCTL_EL1 */
974 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
975 	  NULL, reset_val, CNTKCTL_EL1, 0},
976 
977 	/* CSSELR_EL1 */
978 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
979 	  NULL, reset_unknown, CSSELR_EL1 },
980 
981 	/* PMCR_EL0 */
982 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
983 	  access_pmcr, reset_pmcr, },
984 	/* PMCNTENSET_EL0 */
985 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
986 	  access_pmcnten, reset_unknown, PMCNTENSET_EL0 },
987 	/* PMCNTENCLR_EL0 */
988 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
989 	  access_pmcnten, NULL, PMCNTENSET_EL0 },
990 	/* PMOVSCLR_EL0 */
991 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
992 	  access_pmovs, NULL, PMOVSSET_EL0 },
993 	/* PMSWINC_EL0 */
994 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
995 	  access_pmswinc, reset_unknown, PMSWINC_EL0 },
996 	/* PMSELR_EL0 */
997 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
998 	  access_pmselr, reset_unknown, PMSELR_EL0 },
999 	/* PMCEID0_EL0 */
1000 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
1001 	  access_pmceid },
1002 	/* PMCEID1_EL0 */
1003 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
1004 	  access_pmceid },
1005 	/* PMCCNTR_EL0 */
1006 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
1007 	  access_pmu_evcntr, reset_unknown, PMCCNTR_EL0 },
1008 	/* PMXEVTYPER_EL0 */
1009 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
1010 	  access_pmu_evtyper },
1011 	/* PMXEVCNTR_EL0 */
1012 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
1013 	  access_pmu_evcntr },
1014 	/* PMUSERENR_EL0
1015 	 * This register resets as unknown in 64bit mode while it resets as zero
1016 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1017 	 */
1018 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
1019 	  access_pmuserenr, reset_val, PMUSERENR_EL0, 0 },
1020 	/* PMOVSSET_EL0 */
1021 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
1022 	  access_pmovs, reset_unknown, PMOVSSET_EL0 },
1023 
1024 	/* TPIDR_EL0 */
1025 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
1026 	  NULL, reset_unknown, TPIDR_EL0 },
1027 	/* TPIDRRO_EL0 */
1028 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
1029 	  NULL, reset_unknown, TPIDRRO_EL0 },
1030 
1031 	/* PMEVCNTRn_EL0 */
1032 	PMU_PMEVCNTR_EL0(0),
1033 	PMU_PMEVCNTR_EL0(1),
1034 	PMU_PMEVCNTR_EL0(2),
1035 	PMU_PMEVCNTR_EL0(3),
1036 	PMU_PMEVCNTR_EL0(4),
1037 	PMU_PMEVCNTR_EL0(5),
1038 	PMU_PMEVCNTR_EL0(6),
1039 	PMU_PMEVCNTR_EL0(7),
1040 	PMU_PMEVCNTR_EL0(8),
1041 	PMU_PMEVCNTR_EL0(9),
1042 	PMU_PMEVCNTR_EL0(10),
1043 	PMU_PMEVCNTR_EL0(11),
1044 	PMU_PMEVCNTR_EL0(12),
1045 	PMU_PMEVCNTR_EL0(13),
1046 	PMU_PMEVCNTR_EL0(14),
1047 	PMU_PMEVCNTR_EL0(15),
1048 	PMU_PMEVCNTR_EL0(16),
1049 	PMU_PMEVCNTR_EL0(17),
1050 	PMU_PMEVCNTR_EL0(18),
1051 	PMU_PMEVCNTR_EL0(19),
1052 	PMU_PMEVCNTR_EL0(20),
1053 	PMU_PMEVCNTR_EL0(21),
1054 	PMU_PMEVCNTR_EL0(22),
1055 	PMU_PMEVCNTR_EL0(23),
1056 	PMU_PMEVCNTR_EL0(24),
1057 	PMU_PMEVCNTR_EL0(25),
1058 	PMU_PMEVCNTR_EL0(26),
1059 	PMU_PMEVCNTR_EL0(27),
1060 	PMU_PMEVCNTR_EL0(28),
1061 	PMU_PMEVCNTR_EL0(29),
1062 	PMU_PMEVCNTR_EL0(30),
1063 	/* PMEVTYPERn_EL0 */
1064 	PMU_PMEVTYPER_EL0(0),
1065 	PMU_PMEVTYPER_EL0(1),
1066 	PMU_PMEVTYPER_EL0(2),
1067 	PMU_PMEVTYPER_EL0(3),
1068 	PMU_PMEVTYPER_EL0(4),
1069 	PMU_PMEVTYPER_EL0(5),
1070 	PMU_PMEVTYPER_EL0(6),
1071 	PMU_PMEVTYPER_EL0(7),
1072 	PMU_PMEVTYPER_EL0(8),
1073 	PMU_PMEVTYPER_EL0(9),
1074 	PMU_PMEVTYPER_EL0(10),
1075 	PMU_PMEVTYPER_EL0(11),
1076 	PMU_PMEVTYPER_EL0(12),
1077 	PMU_PMEVTYPER_EL0(13),
1078 	PMU_PMEVTYPER_EL0(14),
1079 	PMU_PMEVTYPER_EL0(15),
1080 	PMU_PMEVTYPER_EL0(16),
1081 	PMU_PMEVTYPER_EL0(17),
1082 	PMU_PMEVTYPER_EL0(18),
1083 	PMU_PMEVTYPER_EL0(19),
1084 	PMU_PMEVTYPER_EL0(20),
1085 	PMU_PMEVTYPER_EL0(21),
1086 	PMU_PMEVTYPER_EL0(22),
1087 	PMU_PMEVTYPER_EL0(23),
1088 	PMU_PMEVTYPER_EL0(24),
1089 	PMU_PMEVTYPER_EL0(25),
1090 	PMU_PMEVTYPER_EL0(26),
1091 	PMU_PMEVTYPER_EL0(27),
1092 	PMU_PMEVTYPER_EL0(28),
1093 	PMU_PMEVTYPER_EL0(29),
1094 	PMU_PMEVTYPER_EL0(30),
1095 	/* PMCCFILTR_EL0
1096 	 * This register resets as unknown in 64bit mode while it resets as zero
1097 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
1098 	 */
1099 	{ Op0(0b11), Op1(0b011), CRn(0b1110), CRm(0b1111), Op2(0b111),
1100 	  access_pmu_evtyper, reset_val, PMCCFILTR_EL0, 0 },
1101 
1102 	/* DACR32_EL2 */
1103 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
1104 	  NULL, reset_unknown, DACR32_EL2 },
1105 	/* IFSR32_EL2 */
1106 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
1107 	  NULL, reset_unknown, IFSR32_EL2 },
1108 	/* FPEXC32_EL2 */
1109 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
1110 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
1111 };
1112 
1113 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
1114 			struct sys_reg_params *p,
1115 			const struct sys_reg_desc *r)
1116 {
1117 	if (p->is_write) {
1118 		return ignore_write(vcpu, p);
1119 	} else {
1120 		u64 dfr = read_system_reg(SYS_ID_AA64DFR0_EL1);
1121 		u64 pfr = read_system_reg(SYS_ID_AA64PFR0_EL1);
1122 		u32 el3 = !!cpuid_feature_extract_unsigned_field(pfr, ID_AA64PFR0_EL3_SHIFT);
1123 
1124 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
1125 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
1126 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
1127 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
1128 		return true;
1129 	}
1130 }
1131 
1132 static bool trap_debug32(struct kvm_vcpu *vcpu,
1133 			 struct sys_reg_params *p,
1134 			 const struct sys_reg_desc *r)
1135 {
1136 	if (p->is_write) {
1137 		vcpu_cp14(vcpu, r->reg) = p->regval;
1138 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
1139 	} else {
1140 		p->regval = vcpu_cp14(vcpu, r->reg);
1141 	}
1142 
1143 	return true;
1144 }
1145 
1146 /* AArch32 debug register mappings
1147  *
1148  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
1149  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
1150  *
1151  * All control registers and watchpoint value registers are mapped to
1152  * the lower 32 bits of their AArch64 equivalents. We share the trap
1153  * handlers with the above AArch64 code which checks what mode the
1154  * system is in.
1155  */
1156 
1157 static bool trap_xvr(struct kvm_vcpu *vcpu,
1158 		     struct sys_reg_params *p,
1159 		     const struct sys_reg_desc *rd)
1160 {
1161 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
1162 
1163 	if (p->is_write) {
1164 		u64 val = *dbg_reg;
1165 
1166 		val &= 0xffffffffUL;
1167 		val |= p->regval << 32;
1168 		*dbg_reg = val;
1169 
1170 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
1171 	} else {
1172 		p->regval = *dbg_reg >> 32;
1173 	}
1174 
1175 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
1176 
1177 	return true;
1178 }
1179 
1180 #define DBG_BCR_BVR_WCR_WVR(n)						\
1181 	/* DBGBVRn */							\
1182 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
1183 	/* DBGBCRn */							\
1184 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
1185 	/* DBGWVRn */							\
1186 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
1187 	/* DBGWCRn */							\
1188 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
1189 
1190 #define DBGBXVR(n)							\
1191 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
1192 
1193 /*
1194  * Trapped cp14 registers. We generally ignore most of the external
1195  * debug, on the principle that they don't really make sense to a
1196  * guest. Revisit this one day, would this principle change.
1197  */
1198 static const struct sys_reg_desc cp14_regs[] = {
1199 	/* DBGIDR */
1200 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
1201 	/* DBGDTRRXext */
1202 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
1203 
1204 	DBG_BCR_BVR_WCR_WVR(0),
1205 	/* DBGDSCRint */
1206 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
1207 	DBG_BCR_BVR_WCR_WVR(1),
1208 	/* DBGDCCINT */
1209 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
1210 	/* DBGDSCRext */
1211 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
1212 	DBG_BCR_BVR_WCR_WVR(2),
1213 	/* DBGDTR[RT]Xint */
1214 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
1215 	/* DBGDTR[RT]Xext */
1216 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
1217 	DBG_BCR_BVR_WCR_WVR(3),
1218 	DBG_BCR_BVR_WCR_WVR(4),
1219 	DBG_BCR_BVR_WCR_WVR(5),
1220 	/* DBGWFAR */
1221 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
1222 	/* DBGOSECCR */
1223 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
1224 	DBG_BCR_BVR_WCR_WVR(6),
1225 	/* DBGVCR */
1226 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
1227 	DBG_BCR_BVR_WCR_WVR(7),
1228 	DBG_BCR_BVR_WCR_WVR(8),
1229 	DBG_BCR_BVR_WCR_WVR(9),
1230 	DBG_BCR_BVR_WCR_WVR(10),
1231 	DBG_BCR_BVR_WCR_WVR(11),
1232 	DBG_BCR_BVR_WCR_WVR(12),
1233 	DBG_BCR_BVR_WCR_WVR(13),
1234 	DBG_BCR_BVR_WCR_WVR(14),
1235 	DBG_BCR_BVR_WCR_WVR(15),
1236 
1237 	/* DBGDRAR (32bit) */
1238 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
1239 
1240 	DBGBXVR(0),
1241 	/* DBGOSLAR */
1242 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
1243 	DBGBXVR(1),
1244 	/* DBGOSLSR */
1245 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
1246 	DBGBXVR(2),
1247 	DBGBXVR(3),
1248 	/* DBGOSDLR */
1249 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
1250 	DBGBXVR(4),
1251 	/* DBGPRCR */
1252 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
1253 	DBGBXVR(5),
1254 	DBGBXVR(6),
1255 	DBGBXVR(7),
1256 	DBGBXVR(8),
1257 	DBGBXVR(9),
1258 	DBGBXVR(10),
1259 	DBGBXVR(11),
1260 	DBGBXVR(12),
1261 	DBGBXVR(13),
1262 	DBGBXVR(14),
1263 	DBGBXVR(15),
1264 
1265 	/* DBGDSAR (32bit) */
1266 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
1267 
1268 	/* DBGDEVID2 */
1269 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
1270 	/* DBGDEVID1 */
1271 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
1272 	/* DBGDEVID */
1273 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
1274 	/* DBGCLAIMSET */
1275 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
1276 	/* DBGCLAIMCLR */
1277 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
1278 	/* DBGAUTHSTATUS */
1279 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
1280 };
1281 
1282 /* Trapped cp14 64bit registers */
1283 static const struct sys_reg_desc cp14_64_regs[] = {
1284 	/* DBGDRAR (64bit) */
1285 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
1286 
1287 	/* DBGDSAR (64bit) */
1288 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
1289 };
1290 
1291 /* Macro to expand the PMEVCNTRn register */
1292 #define PMU_PMEVCNTR(n)							\
1293 	/* PMEVCNTRn */							\
1294 	{ Op1(0), CRn(0b1110),						\
1295 	  CRm((0b1000 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1296 	  access_pmu_evcntr }
1297 
1298 /* Macro to expand the PMEVTYPERn register */
1299 #define PMU_PMEVTYPER(n)						\
1300 	/* PMEVTYPERn */						\
1301 	{ Op1(0), CRn(0b1110),						\
1302 	  CRm((0b1100 | (((n) >> 3) & 0x3))), Op2(((n) & 0x7)),		\
1303 	  access_pmu_evtyper }
1304 
1305 /*
1306  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
1307  * depending on the way they are accessed (as a 32bit or a 64bit
1308  * register).
1309  */
1310 static const struct sys_reg_desc cp15_regs[] = {
1311 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
1312 
1313 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
1314 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1315 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
1316 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
1317 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
1318 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
1319 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
1320 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
1321 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
1322 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
1323 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
1324 
1325 	/*
1326 	 * DC{C,I,CI}SW operations:
1327 	 */
1328 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
1329 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
1330 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
1331 
1332 	/* PMU */
1333 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
1334 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), access_pmcnten },
1335 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), access_pmcnten },
1336 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), access_pmovs },
1337 	{ Op1( 0), CRn( 9), CRm(12), Op2( 4), access_pmswinc },
1338 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
1339 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
1340 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
1341 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), access_pmu_evcntr },
1342 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), access_pmu_evtyper },
1343 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), access_pmu_evcntr },
1344 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), access_pmuserenr },
1345 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), access_pminten },
1346 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), access_pminten },
1347 	{ Op1( 0), CRn( 9), CRm(14), Op2( 3), access_pmovs },
1348 
1349 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
1350 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
1351 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
1352 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
1353 
1354 	/* ICC_SRE */
1355 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), access_gic_sre },
1356 
1357 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
1358 
1359 	/* PMEVCNTRn */
1360 	PMU_PMEVCNTR(0),
1361 	PMU_PMEVCNTR(1),
1362 	PMU_PMEVCNTR(2),
1363 	PMU_PMEVCNTR(3),
1364 	PMU_PMEVCNTR(4),
1365 	PMU_PMEVCNTR(5),
1366 	PMU_PMEVCNTR(6),
1367 	PMU_PMEVCNTR(7),
1368 	PMU_PMEVCNTR(8),
1369 	PMU_PMEVCNTR(9),
1370 	PMU_PMEVCNTR(10),
1371 	PMU_PMEVCNTR(11),
1372 	PMU_PMEVCNTR(12),
1373 	PMU_PMEVCNTR(13),
1374 	PMU_PMEVCNTR(14),
1375 	PMU_PMEVCNTR(15),
1376 	PMU_PMEVCNTR(16),
1377 	PMU_PMEVCNTR(17),
1378 	PMU_PMEVCNTR(18),
1379 	PMU_PMEVCNTR(19),
1380 	PMU_PMEVCNTR(20),
1381 	PMU_PMEVCNTR(21),
1382 	PMU_PMEVCNTR(22),
1383 	PMU_PMEVCNTR(23),
1384 	PMU_PMEVCNTR(24),
1385 	PMU_PMEVCNTR(25),
1386 	PMU_PMEVCNTR(26),
1387 	PMU_PMEVCNTR(27),
1388 	PMU_PMEVCNTR(28),
1389 	PMU_PMEVCNTR(29),
1390 	PMU_PMEVCNTR(30),
1391 	/* PMEVTYPERn */
1392 	PMU_PMEVTYPER(0),
1393 	PMU_PMEVTYPER(1),
1394 	PMU_PMEVTYPER(2),
1395 	PMU_PMEVTYPER(3),
1396 	PMU_PMEVTYPER(4),
1397 	PMU_PMEVTYPER(5),
1398 	PMU_PMEVTYPER(6),
1399 	PMU_PMEVTYPER(7),
1400 	PMU_PMEVTYPER(8),
1401 	PMU_PMEVTYPER(9),
1402 	PMU_PMEVTYPER(10),
1403 	PMU_PMEVTYPER(11),
1404 	PMU_PMEVTYPER(12),
1405 	PMU_PMEVTYPER(13),
1406 	PMU_PMEVTYPER(14),
1407 	PMU_PMEVTYPER(15),
1408 	PMU_PMEVTYPER(16),
1409 	PMU_PMEVTYPER(17),
1410 	PMU_PMEVTYPER(18),
1411 	PMU_PMEVTYPER(19),
1412 	PMU_PMEVTYPER(20),
1413 	PMU_PMEVTYPER(21),
1414 	PMU_PMEVTYPER(22),
1415 	PMU_PMEVTYPER(23),
1416 	PMU_PMEVTYPER(24),
1417 	PMU_PMEVTYPER(25),
1418 	PMU_PMEVTYPER(26),
1419 	PMU_PMEVTYPER(27),
1420 	PMU_PMEVTYPER(28),
1421 	PMU_PMEVTYPER(29),
1422 	PMU_PMEVTYPER(30),
1423 	/* PMCCFILTR */
1424 	{ Op1(0), CRn(14), CRm(15), Op2(7), access_pmu_evtyper },
1425 };
1426 
1427 static const struct sys_reg_desc cp15_64_regs[] = {
1428 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
1429 	{ Op1( 0), CRn( 0), CRm( 9), Op2( 0), access_pmu_evcntr },
1430 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
1431 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
1432 };
1433 
1434 /* Target specific emulation tables */
1435 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
1436 
1437 void kvm_register_target_sys_reg_table(unsigned int target,
1438 				       struct kvm_sys_reg_target_table *table)
1439 {
1440 	target_tables[target] = table;
1441 }
1442 
1443 /* Get specific register table for this target. */
1444 static const struct sys_reg_desc *get_target_table(unsigned target,
1445 						   bool mode_is_64,
1446 						   size_t *num)
1447 {
1448 	struct kvm_sys_reg_target_table *table;
1449 
1450 	table = target_tables[target];
1451 	if (mode_is_64) {
1452 		*num = table->table64.num;
1453 		return table->table64.table;
1454 	} else {
1455 		*num = table->table32.num;
1456 		return table->table32.table;
1457 	}
1458 }
1459 
1460 #define reg_to_match_value(x)						\
1461 	({								\
1462 		unsigned long val;					\
1463 		val  = (x)->Op0 << 14;					\
1464 		val |= (x)->Op1 << 11;					\
1465 		val |= (x)->CRn << 7;					\
1466 		val |= (x)->CRm << 3;					\
1467 		val |= (x)->Op2;					\
1468 		val;							\
1469 	 })
1470 
1471 static int match_sys_reg(const void *key, const void *elt)
1472 {
1473 	const unsigned long pval = (unsigned long)key;
1474 	const struct sys_reg_desc *r = elt;
1475 
1476 	return pval - reg_to_match_value(r);
1477 }
1478 
1479 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
1480 					 const struct sys_reg_desc table[],
1481 					 unsigned int num)
1482 {
1483 	unsigned long pval = reg_to_match_value(params);
1484 
1485 	return bsearch((void *)pval, table, num, sizeof(table[0]), match_sys_reg);
1486 }
1487 
1488 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
1489 {
1490 	kvm_inject_undefined(vcpu);
1491 	return 1;
1492 }
1493 
1494 /*
1495  * emulate_cp --  tries to match a sys_reg access in a handling table, and
1496  *                call the corresponding trap handler.
1497  *
1498  * @params: pointer to the descriptor of the access
1499  * @table: array of trap descriptors
1500  * @num: size of the trap descriptor array
1501  *
1502  * Return 0 if the access has been handled, and -1 if not.
1503  */
1504 static int emulate_cp(struct kvm_vcpu *vcpu,
1505 		      struct sys_reg_params *params,
1506 		      const struct sys_reg_desc *table,
1507 		      size_t num)
1508 {
1509 	const struct sys_reg_desc *r;
1510 
1511 	if (!table)
1512 		return -1;	/* Not handled */
1513 
1514 	r = find_reg(params, table, num);
1515 
1516 	if (r) {
1517 		/*
1518 		 * Not having an accessor means that we have
1519 		 * configured a trap that we don't know how to
1520 		 * handle. This certainly qualifies as a gross bug
1521 		 * that should be fixed right away.
1522 		 */
1523 		BUG_ON(!r->access);
1524 
1525 		if (likely(r->access(vcpu, params, r))) {
1526 			/* Skip instruction, since it was emulated */
1527 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1528 			/* Handled */
1529 			return 0;
1530 		}
1531 	}
1532 
1533 	/* Not handled */
1534 	return -1;
1535 }
1536 
1537 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
1538 				struct sys_reg_params *params)
1539 {
1540 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
1541 	int cp = -1;
1542 
1543 	switch(hsr_ec) {
1544 	case ESR_ELx_EC_CP15_32:
1545 	case ESR_ELx_EC_CP15_64:
1546 		cp = 15;
1547 		break;
1548 	case ESR_ELx_EC_CP14_MR:
1549 	case ESR_ELx_EC_CP14_64:
1550 		cp = 14;
1551 		break;
1552 	default:
1553 		WARN_ON(1);
1554 	}
1555 
1556 	kvm_err("Unsupported guest CP%d access at: %08lx\n",
1557 		cp, *vcpu_pc(vcpu));
1558 	print_sys_reg_instr(params);
1559 	kvm_inject_undefined(vcpu);
1560 }
1561 
1562 /**
1563  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
1564  * @vcpu: The VCPU pointer
1565  * @run:  The kvm_run struct
1566  */
1567 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
1568 			    const struct sys_reg_desc *global,
1569 			    size_t nr_global,
1570 			    const struct sys_reg_desc *target_specific,
1571 			    size_t nr_specific)
1572 {
1573 	struct sys_reg_params params;
1574 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1575 	int Rt = (hsr >> 5) & 0xf;
1576 	int Rt2 = (hsr >> 10) & 0xf;
1577 
1578 	params.is_aarch32 = true;
1579 	params.is_32bit = false;
1580 	params.CRm = (hsr >> 1) & 0xf;
1581 	params.is_write = ((hsr & 1) == 0);
1582 
1583 	params.Op0 = 0;
1584 	params.Op1 = (hsr >> 16) & 0xf;
1585 	params.Op2 = 0;
1586 	params.CRn = 0;
1587 
1588 	/*
1589 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
1590 	 * backends between AArch32 and AArch64, we get away with it.
1591 	 */
1592 	if (params.is_write) {
1593 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
1594 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
1595 	}
1596 
1597 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
1598 		goto out;
1599 	if (!emulate_cp(vcpu, &params, global, nr_global))
1600 		goto out;
1601 
1602 	unhandled_cp_access(vcpu, &params);
1603 
1604 out:
1605 	/* Split up the value between registers for the read side */
1606 	if (!params.is_write) {
1607 		vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
1608 		vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
1609 	}
1610 
1611 	return 1;
1612 }
1613 
1614 /**
1615  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
1616  * @vcpu: The VCPU pointer
1617  * @run:  The kvm_run struct
1618  */
1619 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
1620 			    const struct sys_reg_desc *global,
1621 			    size_t nr_global,
1622 			    const struct sys_reg_desc *target_specific,
1623 			    size_t nr_specific)
1624 {
1625 	struct sys_reg_params params;
1626 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1627 	int Rt  = (hsr >> 5) & 0xf;
1628 
1629 	params.is_aarch32 = true;
1630 	params.is_32bit = true;
1631 	params.CRm = (hsr >> 1) & 0xf;
1632 	params.regval = vcpu_get_reg(vcpu, Rt);
1633 	params.is_write = ((hsr & 1) == 0);
1634 	params.CRn = (hsr >> 10) & 0xf;
1635 	params.Op0 = 0;
1636 	params.Op1 = (hsr >> 14) & 0x7;
1637 	params.Op2 = (hsr >> 17) & 0x7;
1638 
1639 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
1640 	    !emulate_cp(vcpu, &params, global, nr_global)) {
1641 		if (!params.is_write)
1642 			vcpu_set_reg(vcpu, Rt, params.regval);
1643 		return 1;
1644 	}
1645 
1646 	unhandled_cp_access(vcpu, &params);
1647 	return 1;
1648 }
1649 
1650 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1651 {
1652 	const struct sys_reg_desc *target_specific;
1653 	size_t num;
1654 
1655 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1656 	return kvm_handle_cp_64(vcpu,
1657 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
1658 				target_specific, num);
1659 }
1660 
1661 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1662 {
1663 	const struct sys_reg_desc *target_specific;
1664 	size_t num;
1665 
1666 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1667 	return kvm_handle_cp_32(vcpu,
1668 				cp15_regs, ARRAY_SIZE(cp15_regs),
1669 				target_specific, num);
1670 }
1671 
1672 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1673 {
1674 	return kvm_handle_cp_64(vcpu,
1675 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
1676 				NULL, 0);
1677 }
1678 
1679 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1680 {
1681 	return kvm_handle_cp_32(vcpu,
1682 				cp14_regs, ARRAY_SIZE(cp14_regs),
1683 				NULL, 0);
1684 }
1685 
1686 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
1687 			   struct sys_reg_params *params)
1688 {
1689 	size_t num;
1690 	const struct sys_reg_desc *table, *r;
1691 
1692 	table = get_target_table(vcpu->arch.target, true, &num);
1693 
1694 	/* Search target-specific then generic table. */
1695 	r = find_reg(params, table, num);
1696 	if (!r)
1697 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1698 
1699 	if (likely(r)) {
1700 		/*
1701 		 * Not having an accessor means that we have
1702 		 * configured a trap that we don't know how to
1703 		 * handle. This certainly qualifies as a gross bug
1704 		 * that should be fixed right away.
1705 		 */
1706 		BUG_ON(!r->access);
1707 
1708 		if (likely(r->access(vcpu, params, r))) {
1709 			/* Skip instruction, since it was emulated */
1710 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1711 			return 1;
1712 		}
1713 		/* If access function fails, it should complain. */
1714 	} else {
1715 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
1716 			*vcpu_pc(vcpu));
1717 		print_sys_reg_instr(params);
1718 	}
1719 	kvm_inject_undefined(vcpu);
1720 	return 1;
1721 }
1722 
1723 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
1724 			      const struct sys_reg_desc *table, size_t num)
1725 {
1726 	unsigned long i;
1727 
1728 	for (i = 0; i < num; i++)
1729 		if (table[i].reset)
1730 			table[i].reset(vcpu, &table[i]);
1731 }
1732 
1733 /**
1734  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
1735  * @vcpu: The VCPU pointer
1736  * @run:  The kvm_run struct
1737  */
1738 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1739 {
1740 	struct sys_reg_params params;
1741 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1742 	int Rt = (esr >> 5) & 0x1f;
1743 	int ret;
1744 
1745 	trace_kvm_handle_sys_reg(esr);
1746 
1747 	params.is_aarch32 = false;
1748 	params.is_32bit = false;
1749 	params.Op0 = (esr >> 20) & 3;
1750 	params.Op1 = (esr >> 14) & 0x7;
1751 	params.CRn = (esr >> 10) & 0xf;
1752 	params.CRm = (esr >> 1) & 0xf;
1753 	params.Op2 = (esr >> 17) & 0x7;
1754 	params.regval = vcpu_get_reg(vcpu, Rt);
1755 	params.is_write = !(esr & 1);
1756 
1757 	ret = emulate_sys_reg(vcpu, &params);
1758 
1759 	if (!params.is_write)
1760 		vcpu_set_reg(vcpu, Rt, params.regval);
1761 	return ret;
1762 }
1763 
1764 /******************************************************************************
1765  * Userspace API
1766  *****************************************************************************/
1767 
1768 static bool index_to_params(u64 id, struct sys_reg_params *params)
1769 {
1770 	switch (id & KVM_REG_SIZE_MASK) {
1771 	case KVM_REG_SIZE_U64:
1772 		/* Any unused index bits means it's not valid. */
1773 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1774 			      | KVM_REG_ARM_COPROC_MASK
1775 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
1776 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
1777 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
1778 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
1779 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
1780 			return false;
1781 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1782 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1783 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1784 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1785 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1786 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1787 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1788 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1789 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1790 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1791 		return true;
1792 	default:
1793 		return false;
1794 	}
1795 }
1796 
1797 /* Decode an index value, and find the sys_reg_desc entry. */
1798 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1799 						    u64 id)
1800 {
1801 	size_t num;
1802 	const struct sys_reg_desc *table, *r;
1803 	struct sys_reg_params params;
1804 
1805 	/* We only do sys_reg for now. */
1806 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1807 		return NULL;
1808 
1809 	if (!index_to_params(id, &params))
1810 		return NULL;
1811 
1812 	table = get_target_table(vcpu->arch.target, true, &num);
1813 	r = find_reg(&params, table, num);
1814 	if (!r)
1815 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1816 
1817 	/* Not saved in the sys_reg array? */
1818 	if (r && !r->reg)
1819 		r = NULL;
1820 
1821 	return r;
1822 }
1823 
1824 /*
1825  * These are the invariant sys_reg registers: we let the guest see the
1826  * host versions of these, so they're part of the guest state.
1827  *
1828  * A future CPU may provide a mechanism to present different values to
1829  * the guest, or a future kvm may trap them.
1830  */
1831 
1832 #define FUNCTION_INVARIANT(reg)						\
1833 	static void get_##reg(struct kvm_vcpu *v,			\
1834 			      const struct sys_reg_desc *r)		\
1835 	{								\
1836 		u64 val;						\
1837 									\
1838 		asm volatile("mrs %0, " __stringify(reg) "\n"		\
1839 			     : "=r" (val));				\
1840 		((struct sys_reg_desc *)r)->val = val;			\
1841 	}
1842 
1843 FUNCTION_INVARIANT(midr_el1)
1844 FUNCTION_INVARIANT(ctr_el0)
1845 FUNCTION_INVARIANT(revidr_el1)
1846 FUNCTION_INVARIANT(id_pfr0_el1)
1847 FUNCTION_INVARIANT(id_pfr1_el1)
1848 FUNCTION_INVARIANT(id_dfr0_el1)
1849 FUNCTION_INVARIANT(id_afr0_el1)
1850 FUNCTION_INVARIANT(id_mmfr0_el1)
1851 FUNCTION_INVARIANT(id_mmfr1_el1)
1852 FUNCTION_INVARIANT(id_mmfr2_el1)
1853 FUNCTION_INVARIANT(id_mmfr3_el1)
1854 FUNCTION_INVARIANT(id_isar0_el1)
1855 FUNCTION_INVARIANT(id_isar1_el1)
1856 FUNCTION_INVARIANT(id_isar2_el1)
1857 FUNCTION_INVARIANT(id_isar3_el1)
1858 FUNCTION_INVARIANT(id_isar4_el1)
1859 FUNCTION_INVARIANT(id_isar5_el1)
1860 FUNCTION_INVARIANT(clidr_el1)
1861 FUNCTION_INVARIANT(aidr_el1)
1862 
1863 /* ->val is filled in by kvm_sys_reg_table_init() */
1864 static struct sys_reg_desc invariant_sys_regs[] = {
1865 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1866 	  NULL, get_midr_el1 },
1867 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1868 	  NULL, get_revidr_el1 },
1869 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1870 	  NULL, get_id_pfr0_el1 },
1871 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1872 	  NULL, get_id_pfr1_el1 },
1873 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1874 	  NULL, get_id_dfr0_el1 },
1875 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1876 	  NULL, get_id_afr0_el1 },
1877 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1878 	  NULL, get_id_mmfr0_el1 },
1879 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1880 	  NULL, get_id_mmfr1_el1 },
1881 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1882 	  NULL, get_id_mmfr2_el1 },
1883 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1884 	  NULL, get_id_mmfr3_el1 },
1885 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1886 	  NULL, get_id_isar0_el1 },
1887 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1888 	  NULL, get_id_isar1_el1 },
1889 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1890 	  NULL, get_id_isar2_el1 },
1891 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1892 	  NULL, get_id_isar3_el1 },
1893 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1894 	  NULL, get_id_isar4_el1 },
1895 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1896 	  NULL, get_id_isar5_el1 },
1897 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1898 	  NULL, get_clidr_el1 },
1899 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1900 	  NULL, get_aidr_el1 },
1901 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1902 	  NULL, get_ctr_el0 },
1903 };
1904 
1905 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1906 {
1907 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1908 		return -EFAULT;
1909 	return 0;
1910 }
1911 
1912 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1913 {
1914 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1915 		return -EFAULT;
1916 	return 0;
1917 }
1918 
1919 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1920 {
1921 	struct sys_reg_params params;
1922 	const struct sys_reg_desc *r;
1923 
1924 	if (!index_to_params(id, &params))
1925 		return -ENOENT;
1926 
1927 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1928 	if (!r)
1929 		return -ENOENT;
1930 
1931 	return reg_to_user(uaddr, &r->val, id);
1932 }
1933 
1934 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1935 {
1936 	struct sys_reg_params params;
1937 	const struct sys_reg_desc *r;
1938 	int err;
1939 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1940 
1941 	if (!index_to_params(id, &params))
1942 		return -ENOENT;
1943 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1944 	if (!r)
1945 		return -ENOENT;
1946 
1947 	err = reg_from_user(&val, uaddr, id);
1948 	if (err)
1949 		return err;
1950 
1951 	/* This is what we mean by invariant: you can't change it. */
1952 	if (r->val != val)
1953 		return -EINVAL;
1954 
1955 	return 0;
1956 }
1957 
1958 static bool is_valid_cache(u32 val)
1959 {
1960 	u32 level, ctype;
1961 
1962 	if (val >= CSSELR_MAX)
1963 		return false;
1964 
1965 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1966 	level = (val >> 1);
1967 	ctype = (cache_levels >> (level * 3)) & 7;
1968 
1969 	switch (ctype) {
1970 	case 0: /* No cache */
1971 		return false;
1972 	case 1: /* Instruction cache only */
1973 		return (val & 1);
1974 	case 2: /* Data cache only */
1975 	case 4: /* Unified cache */
1976 		return !(val & 1);
1977 	case 3: /* Separate instruction and data caches */
1978 		return true;
1979 	default: /* Reserved: we can't know instruction or data. */
1980 		return false;
1981 	}
1982 }
1983 
1984 static int demux_c15_get(u64 id, void __user *uaddr)
1985 {
1986 	u32 val;
1987 	u32 __user *uval = uaddr;
1988 
1989 	/* Fail if we have unknown bits set. */
1990 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1991 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1992 		return -ENOENT;
1993 
1994 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1995 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1996 		if (KVM_REG_SIZE(id) != 4)
1997 			return -ENOENT;
1998 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1999 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2000 		if (!is_valid_cache(val))
2001 			return -ENOENT;
2002 
2003 		return put_user(get_ccsidr(val), uval);
2004 	default:
2005 		return -ENOENT;
2006 	}
2007 }
2008 
2009 static int demux_c15_set(u64 id, void __user *uaddr)
2010 {
2011 	u32 val, newval;
2012 	u32 __user *uval = uaddr;
2013 
2014 	/* Fail if we have unknown bits set. */
2015 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
2016 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
2017 		return -ENOENT;
2018 
2019 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
2020 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
2021 		if (KVM_REG_SIZE(id) != 4)
2022 			return -ENOENT;
2023 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
2024 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
2025 		if (!is_valid_cache(val))
2026 			return -ENOENT;
2027 
2028 		if (get_user(newval, uval))
2029 			return -EFAULT;
2030 
2031 		/* This is also invariant: you can't change it. */
2032 		if (newval != get_ccsidr(val))
2033 			return -EINVAL;
2034 		return 0;
2035 	default:
2036 		return -ENOENT;
2037 	}
2038 }
2039 
2040 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2041 {
2042 	const struct sys_reg_desc *r;
2043 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2044 
2045 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2046 		return demux_c15_get(reg->id, uaddr);
2047 
2048 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2049 		return -ENOENT;
2050 
2051 	r = index_to_sys_reg_desc(vcpu, reg->id);
2052 	if (!r)
2053 		return get_invariant_sys_reg(reg->id, uaddr);
2054 
2055 	if (r->get_user)
2056 		return (r->get_user)(vcpu, r, reg, uaddr);
2057 
2058 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
2059 }
2060 
2061 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
2062 {
2063 	const struct sys_reg_desc *r;
2064 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
2065 
2066 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
2067 		return demux_c15_set(reg->id, uaddr);
2068 
2069 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
2070 		return -ENOENT;
2071 
2072 	r = index_to_sys_reg_desc(vcpu, reg->id);
2073 	if (!r)
2074 		return set_invariant_sys_reg(reg->id, uaddr);
2075 
2076 	if (r->set_user)
2077 		return (r->set_user)(vcpu, r, reg, uaddr);
2078 
2079 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
2080 }
2081 
2082 static unsigned int num_demux_regs(void)
2083 {
2084 	unsigned int i, count = 0;
2085 
2086 	for (i = 0; i < CSSELR_MAX; i++)
2087 		if (is_valid_cache(i))
2088 			count++;
2089 
2090 	return count;
2091 }
2092 
2093 static int write_demux_regids(u64 __user *uindices)
2094 {
2095 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
2096 	unsigned int i;
2097 
2098 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
2099 	for (i = 0; i < CSSELR_MAX; i++) {
2100 		if (!is_valid_cache(i))
2101 			continue;
2102 		if (put_user(val | i, uindices))
2103 			return -EFAULT;
2104 		uindices++;
2105 	}
2106 	return 0;
2107 }
2108 
2109 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
2110 {
2111 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
2112 		KVM_REG_ARM64_SYSREG |
2113 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
2114 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
2115 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
2116 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
2117 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
2118 }
2119 
2120 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
2121 {
2122 	if (!*uind)
2123 		return true;
2124 
2125 	if (put_user(sys_reg_to_index(reg), *uind))
2126 		return false;
2127 
2128 	(*uind)++;
2129 	return true;
2130 }
2131 
2132 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
2133 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
2134 {
2135 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
2136 	unsigned int total = 0;
2137 	size_t num;
2138 
2139 	/* We check for duplicates here, to allow arch-specific overrides. */
2140 	i1 = get_target_table(vcpu->arch.target, true, &num);
2141 	end1 = i1 + num;
2142 	i2 = sys_reg_descs;
2143 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
2144 
2145 	BUG_ON(i1 == end1 || i2 == end2);
2146 
2147 	/* Walk carefully, as both tables may refer to the same register. */
2148 	while (i1 || i2) {
2149 		int cmp = cmp_sys_reg(i1, i2);
2150 		/* target-specific overrides generic entry. */
2151 		if (cmp <= 0) {
2152 			/* Ignore registers we trap but don't save. */
2153 			if (i1->reg) {
2154 				if (!copy_reg_to_user(i1, &uind))
2155 					return -EFAULT;
2156 				total++;
2157 			}
2158 		} else {
2159 			/* Ignore registers we trap but don't save. */
2160 			if (i2->reg) {
2161 				if (!copy_reg_to_user(i2, &uind))
2162 					return -EFAULT;
2163 				total++;
2164 			}
2165 		}
2166 
2167 		if (cmp <= 0 && ++i1 == end1)
2168 			i1 = NULL;
2169 		if (cmp >= 0 && ++i2 == end2)
2170 			i2 = NULL;
2171 	}
2172 	return total;
2173 }
2174 
2175 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
2176 {
2177 	return ARRAY_SIZE(invariant_sys_regs)
2178 		+ num_demux_regs()
2179 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
2180 }
2181 
2182 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
2183 {
2184 	unsigned int i;
2185 	int err;
2186 
2187 	/* Then give them all the invariant registers' indices. */
2188 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
2189 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
2190 			return -EFAULT;
2191 		uindices++;
2192 	}
2193 
2194 	err = walk_sys_regs(vcpu, uindices);
2195 	if (err < 0)
2196 		return err;
2197 	uindices += err;
2198 
2199 	return write_demux_regids(uindices);
2200 }
2201 
2202 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
2203 {
2204 	unsigned int i;
2205 
2206 	for (i = 1; i < n; i++) {
2207 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
2208 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
2209 			return 1;
2210 		}
2211 	}
2212 
2213 	return 0;
2214 }
2215 
2216 void kvm_sys_reg_table_init(void)
2217 {
2218 	unsigned int i;
2219 	struct sys_reg_desc clidr;
2220 
2221 	/* Make sure tables are unique and in order. */
2222 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
2223 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
2224 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
2225 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
2226 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
2227 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
2228 
2229 	/* We abuse the reset function to overwrite the table itself. */
2230 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
2231 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
2232 
2233 	/*
2234 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
2235 	 *
2236 	 *   If software reads the Cache Type fields from Ctype1
2237 	 *   upwards, once it has seen a value of 0b000, no caches
2238 	 *   exist at further-out levels of the hierarchy. So, for
2239 	 *   example, if Ctype3 is the first Cache Type field with a
2240 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
2241 	 *   ignored.
2242 	 */
2243 	get_clidr_el1(NULL, &clidr); /* Ugly... */
2244 	cache_levels = clidr.val;
2245 	for (i = 0; i < 7; i++)
2246 		if (((cache_levels >> (i*3)) & 7) == 0)
2247 			break;
2248 	/* Clear all higher bits. */
2249 	cache_levels &= (1 << (i*3))-1;
2250 }
2251 
2252 /**
2253  * kvm_reset_sys_regs - sets system registers to reset value
2254  * @vcpu: The VCPU pointer
2255  *
2256  * This function finds the right table above and sets the registers on the
2257  * virtual CPU struct to their architecturally defined reset values.
2258  */
2259 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
2260 {
2261 	size_t num;
2262 	const struct sys_reg_desc *table;
2263 
2264 	/* Catch someone adding a register without putting in reset entry. */
2265 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
2266 
2267 	/* Generic chip reset first (so target could override). */
2268 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
2269 
2270 	table = get_target_table(vcpu->arch.target, true, &num);
2271 	reset_sys_reg_descs(vcpu, table, num);
2272 
2273 	for (num = 1; num < NR_SYS_REGS; num++)
2274 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
2275 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
2276 }
2277