xref: /linux/arch/arm64/kvm/sys_regs.c (revision a86b5505304404dc5fc5e62a6dc294706e525003)
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/kvm_host.h>
24 #include <linux/mm.h>
25 #include <linux/uaccess.h>
26 
27 #include <asm/cacheflush.h>
28 #include <asm/cputype.h>
29 #include <asm/debug-monitors.h>
30 #include <asm/esr.h>
31 #include <asm/kvm_arm.h>
32 #include <asm/kvm_asm.h>
33 #include <asm/kvm_coproc.h>
34 #include <asm/kvm_emulate.h>
35 #include <asm/kvm_host.h>
36 #include <asm/kvm_mmu.h>
37 #include <asm/perf_event.h>
38 
39 #include <trace/events/kvm.h>
40 
41 #include "sys_regs.h"
42 
43 #include "trace.h"
44 
45 /*
46  * All of this file is extremly similar to the ARM coproc.c, but the
47  * types are different. My gut feeling is that it should be pretty
48  * easy to merge, but that would be an ABI breakage -- again. VFP
49  * would also need to be abstracted.
50  *
51  * For AArch32, we only take care of what is being trapped. Anything
52  * that has to do with init and userspace access has to go via the
53  * 64bit interface.
54  */
55 
56 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
57 static u32 cache_levels;
58 
59 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
60 #define CSSELR_MAX 12
61 
62 /* Which cache CCSIDR represents depends on CSSELR value. */
63 static u32 get_ccsidr(u32 csselr)
64 {
65 	u32 ccsidr;
66 
67 	/* Make sure noone else changes CSSELR during this! */
68 	local_irq_disable();
69 	/* Put value into CSSELR */
70 	asm volatile("msr csselr_el1, %x0" : : "r" (csselr));
71 	isb();
72 	/* Read result out of CCSIDR */
73 	asm volatile("mrs %0, ccsidr_el1" : "=r" (ccsidr));
74 	local_irq_enable();
75 
76 	return ccsidr;
77 }
78 
79 /*
80  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
81  */
82 static bool access_dcsw(struct kvm_vcpu *vcpu,
83 			struct sys_reg_params *p,
84 			const struct sys_reg_desc *r)
85 {
86 	if (!p->is_write)
87 		return read_from_write_only(vcpu, p);
88 
89 	kvm_set_way_flush(vcpu);
90 	return true;
91 }
92 
93 /*
94  * Generic accessor for VM registers. Only called as long as HCR_TVM
95  * is set. If the guest enables the MMU, we stop trapping the VM
96  * sys_regs and leave it in complete control of the caches.
97  */
98 static bool access_vm_reg(struct kvm_vcpu *vcpu,
99 			  struct sys_reg_params *p,
100 			  const struct sys_reg_desc *r)
101 {
102 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
103 
104 	BUG_ON(!p->is_write);
105 
106 	if (!p->is_aarch32) {
107 		vcpu_sys_reg(vcpu, r->reg) = p->regval;
108 	} else {
109 		if (!p->is_32bit)
110 			vcpu_cp15_64_high(vcpu, r->reg) = upper_32_bits(p->regval);
111 		vcpu_cp15_64_low(vcpu, r->reg) = lower_32_bits(p->regval);
112 	}
113 
114 	kvm_toggle_cache(vcpu, was_enabled);
115 	return true;
116 }
117 
118 /*
119  * Trap handler for the GICv3 SGI generation system register.
120  * Forward the request to the VGIC emulation.
121  * The cp15_64 code makes sure this automatically works
122  * for both AArch64 and AArch32 accesses.
123  */
124 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
125 			   struct sys_reg_params *p,
126 			   const struct sys_reg_desc *r)
127 {
128 	if (!p->is_write)
129 		return read_from_write_only(vcpu, p);
130 
131 	vgic_v3_dispatch_sgi(vcpu, p->regval);
132 
133 	return true;
134 }
135 
136 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
137 			struct sys_reg_params *p,
138 			const struct sys_reg_desc *r)
139 {
140 	if (p->is_write)
141 		return ignore_write(vcpu, p);
142 	else
143 		return read_zero(vcpu, p);
144 }
145 
146 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
147 			   struct sys_reg_params *p,
148 			   const struct sys_reg_desc *r)
149 {
150 	if (p->is_write) {
151 		return ignore_write(vcpu, p);
152 	} else {
153 		p->regval = (1 << 3);
154 		return true;
155 	}
156 }
157 
158 static bool trap_dbgauthstatus_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 		u32 val;
166 		asm volatile("mrs %0, dbgauthstatus_el1" : "=r" (val));
167 		p->regval = val;
168 		return true;
169 	}
170 }
171 
172 /*
173  * We want to avoid world-switching all the DBG registers all the
174  * time:
175  *
176  * - If we've touched any debug register, it is likely that we're
177  *   going to touch more of them. It then makes sense to disable the
178  *   traps and start doing the save/restore dance
179  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
180  *   then mandatory to save/restore the registers, as the guest
181  *   depends on them.
182  *
183  * For this, we use a DIRTY bit, indicating the guest has modified the
184  * debug registers, used as follow:
185  *
186  * On guest entry:
187  * - If the dirty bit is set (because we're coming back from trapping),
188  *   disable the traps, save host registers, restore guest registers.
189  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
190  *   set the dirty bit, disable the traps, save host registers,
191  *   restore guest registers.
192  * - Otherwise, enable the traps
193  *
194  * On guest exit:
195  * - If the dirty bit is set, save guest registers, restore host
196  *   registers and clear the dirty bit. This ensure that the host can
197  *   now use the debug registers.
198  */
199 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
200 			    struct sys_reg_params *p,
201 			    const struct sys_reg_desc *r)
202 {
203 	if (p->is_write) {
204 		vcpu_sys_reg(vcpu, r->reg) = p->regval;
205 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
206 	} else {
207 		p->regval = vcpu_sys_reg(vcpu, r->reg);
208 	}
209 
210 	trace_trap_reg(__func__, r->reg, p->is_write, p->regval);
211 
212 	return true;
213 }
214 
215 /*
216  * reg_to_dbg/dbg_to_reg
217  *
218  * A 32 bit write to a debug register leave top bits alone
219  * A 32 bit read from a debug register only returns the bottom bits
220  *
221  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
222  * hyp.S code switches between host and guest values in future.
223  */
224 static void reg_to_dbg(struct kvm_vcpu *vcpu,
225 		       struct sys_reg_params *p,
226 		       u64 *dbg_reg)
227 {
228 	u64 val = p->regval;
229 
230 	if (p->is_32bit) {
231 		val &= 0xffffffffUL;
232 		val |= ((*dbg_reg >> 32) << 32);
233 	}
234 
235 	*dbg_reg = val;
236 	vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
237 }
238 
239 static void dbg_to_reg(struct kvm_vcpu *vcpu,
240 		       struct sys_reg_params *p,
241 		       u64 *dbg_reg)
242 {
243 	p->regval = *dbg_reg;
244 	if (p->is_32bit)
245 		p->regval &= 0xffffffffUL;
246 }
247 
248 static bool trap_bvr(struct kvm_vcpu *vcpu,
249 		     struct sys_reg_params *p,
250 		     const struct sys_reg_desc *rd)
251 {
252 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
253 
254 	if (p->is_write)
255 		reg_to_dbg(vcpu, p, dbg_reg);
256 	else
257 		dbg_to_reg(vcpu, p, dbg_reg);
258 
259 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
260 
261 	return true;
262 }
263 
264 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
265 		const struct kvm_one_reg *reg, void __user *uaddr)
266 {
267 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
268 
269 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
270 		return -EFAULT;
271 	return 0;
272 }
273 
274 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
275 	const struct kvm_one_reg *reg, void __user *uaddr)
276 {
277 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
278 
279 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
280 		return -EFAULT;
281 	return 0;
282 }
283 
284 static void reset_bvr(struct kvm_vcpu *vcpu,
285 		      const struct sys_reg_desc *rd)
286 {
287 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
288 }
289 
290 static bool trap_bcr(struct kvm_vcpu *vcpu,
291 		     struct sys_reg_params *p,
292 		     const struct sys_reg_desc *rd)
293 {
294 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
295 
296 	if (p->is_write)
297 		reg_to_dbg(vcpu, p, dbg_reg);
298 	else
299 		dbg_to_reg(vcpu, p, dbg_reg);
300 
301 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
302 
303 	return true;
304 }
305 
306 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
307 		const struct kvm_one_reg *reg, void __user *uaddr)
308 {
309 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
310 
311 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
312 		return -EFAULT;
313 
314 	return 0;
315 }
316 
317 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
318 	const struct kvm_one_reg *reg, void __user *uaddr)
319 {
320 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
321 
322 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
323 		return -EFAULT;
324 	return 0;
325 }
326 
327 static void reset_bcr(struct kvm_vcpu *vcpu,
328 		      const struct sys_reg_desc *rd)
329 {
330 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
331 }
332 
333 static bool trap_wvr(struct kvm_vcpu *vcpu,
334 		     struct sys_reg_params *p,
335 		     const struct sys_reg_desc *rd)
336 {
337 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
338 
339 	if (p->is_write)
340 		reg_to_dbg(vcpu, p, dbg_reg);
341 	else
342 		dbg_to_reg(vcpu, p, dbg_reg);
343 
344 	trace_trap_reg(__func__, rd->reg, p->is_write,
345 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
346 
347 	return true;
348 }
349 
350 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
351 		const struct kvm_one_reg *reg, void __user *uaddr)
352 {
353 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
354 
355 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
356 		return -EFAULT;
357 	return 0;
358 }
359 
360 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
361 	const struct kvm_one_reg *reg, void __user *uaddr)
362 {
363 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
364 
365 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
366 		return -EFAULT;
367 	return 0;
368 }
369 
370 static void reset_wvr(struct kvm_vcpu *vcpu,
371 		      const struct sys_reg_desc *rd)
372 {
373 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
374 }
375 
376 static bool trap_wcr(struct kvm_vcpu *vcpu,
377 		     struct sys_reg_params *p,
378 		     const struct sys_reg_desc *rd)
379 {
380 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
381 
382 	if (p->is_write)
383 		reg_to_dbg(vcpu, p, dbg_reg);
384 	else
385 		dbg_to_reg(vcpu, p, dbg_reg);
386 
387 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
388 
389 	return true;
390 }
391 
392 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
393 		const struct kvm_one_reg *reg, void __user *uaddr)
394 {
395 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
396 
397 	if (copy_from_user(r, uaddr, KVM_REG_SIZE(reg->id)) != 0)
398 		return -EFAULT;
399 	return 0;
400 }
401 
402 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
403 	const struct kvm_one_reg *reg, void __user *uaddr)
404 {
405 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
406 
407 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
408 		return -EFAULT;
409 	return 0;
410 }
411 
412 static void reset_wcr(struct kvm_vcpu *vcpu,
413 		      const struct sys_reg_desc *rd)
414 {
415 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
416 }
417 
418 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
419 {
420 	u64 amair;
421 
422 	asm volatile("mrs %0, amair_el1\n" : "=r" (amair));
423 	vcpu_sys_reg(vcpu, AMAIR_EL1) = amair;
424 }
425 
426 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
427 {
428 	u64 mpidr;
429 
430 	/*
431 	 * Map the vcpu_id into the first three affinity level fields of
432 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
433 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
434 	 * of the GICv3 to be able to address each CPU directly when
435 	 * sending IPIs.
436 	 */
437 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
438 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
439 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
440 	vcpu_sys_reg(vcpu, MPIDR_EL1) = (1ULL << 31) | mpidr;
441 }
442 
443 static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
444 {
445 	u64 pmcr, val;
446 
447 	asm volatile("mrs %0, pmcr_el0\n" : "=r" (pmcr));
448 	/* Writable bits of PMCR_EL0 (ARMV8_PMU_PMCR_MASK) is reset to UNKNOWN
449 	 * except PMCR.E resetting to zero.
450 	 */
451 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
452 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
453 	vcpu_sys_reg(vcpu, PMCR_EL0) = val;
454 }
455 
456 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
457 			const struct sys_reg_desc *r)
458 {
459 	u64 val;
460 
461 	if (!kvm_arm_pmu_v3_ready(vcpu))
462 		return trap_raz_wi(vcpu, p, r);
463 
464 	if (p->is_write) {
465 		/* Only update writeable bits of PMCR */
466 		val = vcpu_sys_reg(vcpu, PMCR_EL0);
467 		val &= ~ARMV8_PMU_PMCR_MASK;
468 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
469 		vcpu_sys_reg(vcpu, PMCR_EL0) = val;
470 	} else {
471 		/* PMCR.P & PMCR.C are RAZ */
472 		val = vcpu_sys_reg(vcpu, PMCR_EL0)
473 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
474 		p->regval = val;
475 	}
476 
477 	return true;
478 }
479 
480 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
481 			  const struct sys_reg_desc *r)
482 {
483 	if (!kvm_arm_pmu_v3_ready(vcpu))
484 		return trap_raz_wi(vcpu, p, r);
485 
486 	if (p->is_write)
487 		vcpu_sys_reg(vcpu, PMSELR_EL0) = p->regval;
488 	else
489 		/* return PMSELR.SEL field */
490 		p->regval = vcpu_sys_reg(vcpu, PMSELR_EL0)
491 			    & ARMV8_PMU_COUNTER_MASK;
492 
493 	return true;
494 }
495 
496 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
497 			  const struct sys_reg_desc *r)
498 {
499 	u64 pmceid;
500 
501 	if (!kvm_arm_pmu_v3_ready(vcpu))
502 		return trap_raz_wi(vcpu, p, r);
503 
504 	BUG_ON(p->is_write);
505 
506 	if (!(p->Op2 & 1))
507 		asm volatile("mrs %0, pmceid0_el0\n" : "=r" (pmceid));
508 	else
509 		asm volatile("mrs %0, pmceid1_el0\n" : "=r" (pmceid));
510 
511 	p->regval = pmceid;
512 
513 	return true;
514 }
515 
516 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
517 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
518 	/* DBGBVRn_EL1 */						\
519 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b100),	\
520 	  trap_bvr, reset_bvr, n, 0, get_bvr, set_bvr },		\
521 	/* DBGBCRn_EL1 */						\
522 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b101),	\
523 	  trap_bcr, reset_bcr, n, 0, get_bcr, set_bcr },		\
524 	/* DBGWVRn_EL1 */						\
525 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b110),	\
526 	  trap_wvr, reset_wvr, n, 0,  get_wvr, set_wvr },		\
527 	/* DBGWCRn_EL1 */						\
528 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b111),	\
529 	  trap_wcr, reset_wcr, n, 0,  get_wcr, set_wcr }
530 
531 /*
532  * Architected system registers.
533  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
534  *
535  * We could trap ID_DFR0 and tell the guest we don't support performance
536  * monitoring.  Unfortunately the patch to make the kernel check ID_DFR0 was
537  * NAKed, so it will read the PMCR anyway.
538  *
539  * Therefore we tell the guest we have 0 counters.  Unfortunately, we
540  * must always support PMCCNTR (the cycle counter): we just RAZ/WI for
541  * all PM registers, which doesn't crash the guest kernel at least.
542  *
543  * Debug handling: We do trap most, if not all debug related system
544  * registers. The implementation is good enough to ensure that a guest
545  * can use these with minimal performance degradation. The drawback is
546  * that we don't implement any of the external debug, none of the
547  * OSlock protocol. This should be revisited if we ever encounter a
548  * more demanding guest...
549  */
550 static const struct sys_reg_desc sys_reg_descs[] = {
551 	/* DC ISW */
552 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
553 	  access_dcsw },
554 	/* DC CSW */
555 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
556 	  access_dcsw },
557 	/* DC CISW */
558 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
559 	  access_dcsw },
560 
561 	DBG_BCR_BVR_WCR_WVR_EL1(0),
562 	DBG_BCR_BVR_WCR_WVR_EL1(1),
563 	/* MDCCINT_EL1 */
564 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
565 	  trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
566 	/* MDSCR_EL1 */
567 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
568 	  trap_debug_regs, reset_val, MDSCR_EL1, 0 },
569 	DBG_BCR_BVR_WCR_WVR_EL1(2),
570 	DBG_BCR_BVR_WCR_WVR_EL1(3),
571 	DBG_BCR_BVR_WCR_WVR_EL1(4),
572 	DBG_BCR_BVR_WCR_WVR_EL1(5),
573 	DBG_BCR_BVR_WCR_WVR_EL1(6),
574 	DBG_BCR_BVR_WCR_WVR_EL1(7),
575 	DBG_BCR_BVR_WCR_WVR_EL1(8),
576 	DBG_BCR_BVR_WCR_WVR_EL1(9),
577 	DBG_BCR_BVR_WCR_WVR_EL1(10),
578 	DBG_BCR_BVR_WCR_WVR_EL1(11),
579 	DBG_BCR_BVR_WCR_WVR_EL1(12),
580 	DBG_BCR_BVR_WCR_WVR_EL1(13),
581 	DBG_BCR_BVR_WCR_WVR_EL1(14),
582 	DBG_BCR_BVR_WCR_WVR_EL1(15),
583 
584 	/* MDRAR_EL1 */
585 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
586 	  trap_raz_wi },
587 	/* OSLAR_EL1 */
588 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
589 	  trap_raz_wi },
590 	/* OSLSR_EL1 */
591 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
592 	  trap_oslsr_el1 },
593 	/* OSDLR_EL1 */
594 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
595 	  trap_raz_wi },
596 	/* DBGPRCR_EL1 */
597 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
598 	  trap_raz_wi },
599 	/* DBGCLAIMSET_EL1 */
600 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
601 	  trap_raz_wi },
602 	/* DBGCLAIMCLR_EL1 */
603 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
604 	  trap_raz_wi },
605 	/* DBGAUTHSTATUS_EL1 */
606 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
607 	  trap_dbgauthstatus_el1 },
608 
609 	/* MDCCSR_EL1 */
610 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
611 	  trap_raz_wi },
612 	/* DBGDTR_EL0 */
613 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
614 	  trap_raz_wi },
615 	/* DBGDTR[TR]X_EL0 */
616 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
617 	  trap_raz_wi },
618 
619 	/* DBGVCR32_EL2 */
620 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
621 	  NULL, reset_val, DBGVCR32_EL2, 0 },
622 
623 	/* MPIDR_EL1 */
624 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
625 	  NULL, reset_mpidr, MPIDR_EL1 },
626 	/* SCTLR_EL1 */
627 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
628 	  access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
629 	/* CPACR_EL1 */
630 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
631 	  NULL, reset_val, CPACR_EL1, 0 },
632 	/* TTBR0_EL1 */
633 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
634 	  access_vm_reg, reset_unknown, TTBR0_EL1 },
635 	/* TTBR1_EL1 */
636 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
637 	  access_vm_reg, reset_unknown, TTBR1_EL1 },
638 	/* TCR_EL1 */
639 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
640 	  access_vm_reg, reset_val, TCR_EL1, 0 },
641 
642 	/* AFSR0_EL1 */
643 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
644 	  access_vm_reg, reset_unknown, AFSR0_EL1 },
645 	/* AFSR1_EL1 */
646 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
647 	  access_vm_reg, reset_unknown, AFSR1_EL1 },
648 	/* ESR_EL1 */
649 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
650 	  access_vm_reg, reset_unknown, ESR_EL1 },
651 	/* FAR_EL1 */
652 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
653 	  access_vm_reg, reset_unknown, FAR_EL1 },
654 	/* PAR_EL1 */
655 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
656 	  NULL, reset_unknown, PAR_EL1 },
657 
658 	/* PMINTENSET_EL1 */
659 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
660 	  trap_raz_wi },
661 	/* PMINTENCLR_EL1 */
662 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
663 	  trap_raz_wi },
664 
665 	/* MAIR_EL1 */
666 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
667 	  access_vm_reg, reset_unknown, MAIR_EL1 },
668 	/* AMAIR_EL1 */
669 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
670 	  access_vm_reg, reset_amair_el1, AMAIR_EL1 },
671 
672 	/* VBAR_EL1 */
673 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
674 	  NULL, reset_val, VBAR_EL1, 0 },
675 
676 	/* ICC_SGI1R_EL1 */
677 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1011), Op2(0b101),
678 	  access_gic_sgi },
679 	/* ICC_SRE_EL1 */
680 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1100), Op2(0b101),
681 	  trap_raz_wi },
682 
683 	/* CONTEXTIDR_EL1 */
684 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
685 	  access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
686 	/* TPIDR_EL1 */
687 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
688 	  NULL, reset_unknown, TPIDR_EL1 },
689 
690 	/* CNTKCTL_EL1 */
691 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
692 	  NULL, reset_val, CNTKCTL_EL1, 0},
693 
694 	/* CSSELR_EL1 */
695 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
696 	  NULL, reset_unknown, CSSELR_EL1 },
697 
698 	/* PMCR_EL0 */
699 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
700 	  access_pmcr, reset_pmcr, },
701 	/* PMCNTENSET_EL0 */
702 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
703 	  trap_raz_wi },
704 	/* PMCNTENCLR_EL0 */
705 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
706 	  trap_raz_wi },
707 	/* PMOVSCLR_EL0 */
708 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
709 	  trap_raz_wi },
710 	/* PMSWINC_EL0 */
711 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
712 	  trap_raz_wi },
713 	/* PMSELR_EL0 */
714 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
715 	  access_pmselr, reset_unknown, PMSELR_EL0 },
716 	/* PMCEID0_EL0 */
717 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
718 	  access_pmceid },
719 	/* PMCEID1_EL0 */
720 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
721 	  access_pmceid },
722 	/* PMCCNTR_EL0 */
723 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
724 	  trap_raz_wi },
725 	/* PMXEVTYPER_EL0 */
726 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
727 	  trap_raz_wi },
728 	/* PMXEVCNTR_EL0 */
729 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
730 	  trap_raz_wi },
731 	/* PMUSERENR_EL0 */
732 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
733 	  trap_raz_wi },
734 	/* PMOVSSET_EL0 */
735 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
736 	  trap_raz_wi },
737 
738 	/* TPIDR_EL0 */
739 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
740 	  NULL, reset_unknown, TPIDR_EL0 },
741 	/* TPIDRRO_EL0 */
742 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
743 	  NULL, reset_unknown, TPIDRRO_EL0 },
744 
745 	/* DACR32_EL2 */
746 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
747 	  NULL, reset_unknown, DACR32_EL2 },
748 	/* IFSR32_EL2 */
749 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
750 	  NULL, reset_unknown, IFSR32_EL2 },
751 	/* FPEXC32_EL2 */
752 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
753 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
754 };
755 
756 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
757 			struct sys_reg_params *p,
758 			const struct sys_reg_desc *r)
759 {
760 	if (p->is_write) {
761 		return ignore_write(vcpu, p);
762 	} else {
763 		u64 dfr = read_system_reg(SYS_ID_AA64DFR0_EL1);
764 		u64 pfr = read_system_reg(SYS_ID_AA64PFR0_EL1);
765 		u32 el3 = !!cpuid_feature_extract_field(pfr, ID_AA64PFR0_EL3_SHIFT);
766 
767 		p->regval = ((((dfr >> ID_AA64DFR0_WRPS_SHIFT) & 0xf) << 28) |
768 			     (((dfr >> ID_AA64DFR0_BRPS_SHIFT) & 0xf) << 24) |
769 			     (((dfr >> ID_AA64DFR0_CTX_CMPS_SHIFT) & 0xf) << 20)
770 			     | (6 << 16) | (el3 << 14) | (el3 << 12));
771 		return true;
772 	}
773 }
774 
775 static bool trap_debug32(struct kvm_vcpu *vcpu,
776 			 struct sys_reg_params *p,
777 			 const struct sys_reg_desc *r)
778 {
779 	if (p->is_write) {
780 		vcpu_cp14(vcpu, r->reg) = p->regval;
781 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
782 	} else {
783 		p->regval = vcpu_cp14(vcpu, r->reg);
784 	}
785 
786 	return true;
787 }
788 
789 /* AArch32 debug register mappings
790  *
791  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
792  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
793  *
794  * All control registers and watchpoint value registers are mapped to
795  * the lower 32 bits of their AArch64 equivalents. We share the trap
796  * handlers with the above AArch64 code which checks what mode the
797  * system is in.
798  */
799 
800 static bool trap_xvr(struct kvm_vcpu *vcpu,
801 		     struct sys_reg_params *p,
802 		     const struct sys_reg_desc *rd)
803 {
804 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
805 
806 	if (p->is_write) {
807 		u64 val = *dbg_reg;
808 
809 		val &= 0xffffffffUL;
810 		val |= p->regval << 32;
811 		*dbg_reg = val;
812 
813 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
814 	} else {
815 		p->regval = *dbg_reg >> 32;
816 	}
817 
818 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
819 
820 	return true;
821 }
822 
823 #define DBG_BCR_BVR_WCR_WVR(n)						\
824 	/* DBGBVRn */							\
825 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
826 	/* DBGBCRn */							\
827 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
828 	/* DBGWVRn */							\
829 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
830 	/* DBGWCRn */							\
831 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
832 
833 #define DBGBXVR(n)							\
834 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
835 
836 /*
837  * Trapped cp14 registers. We generally ignore most of the external
838  * debug, on the principle that they don't really make sense to a
839  * guest. Revisit this one day, would this principle change.
840  */
841 static const struct sys_reg_desc cp14_regs[] = {
842 	/* DBGIDR */
843 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
844 	/* DBGDTRRXext */
845 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
846 
847 	DBG_BCR_BVR_WCR_WVR(0),
848 	/* DBGDSCRint */
849 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
850 	DBG_BCR_BVR_WCR_WVR(1),
851 	/* DBGDCCINT */
852 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
853 	/* DBGDSCRext */
854 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
855 	DBG_BCR_BVR_WCR_WVR(2),
856 	/* DBGDTR[RT]Xint */
857 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
858 	/* DBGDTR[RT]Xext */
859 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
860 	DBG_BCR_BVR_WCR_WVR(3),
861 	DBG_BCR_BVR_WCR_WVR(4),
862 	DBG_BCR_BVR_WCR_WVR(5),
863 	/* DBGWFAR */
864 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
865 	/* DBGOSECCR */
866 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
867 	DBG_BCR_BVR_WCR_WVR(6),
868 	/* DBGVCR */
869 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
870 	DBG_BCR_BVR_WCR_WVR(7),
871 	DBG_BCR_BVR_WCR_WVR(8),
872 	DBG_BCR_BVR_WCR_WVR(9),
873 	DBG_BCR_BVR_WCR_WVR(10),
874 	DBG_BCR_BVR_WCR_WVR(11),
875 	DBG_BCR_BVR_WCR_WVR(12),
876 	DBG_BCR_BVR_WCR_WVR(13),
877 	DBG_BCR_BVR_WCR_WVR(14),
878 	DBG_BCR_BVR_WCR_WVR(15),
879 
880 	/* DBGDRAR (32bit) */
881 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
882 
883 	DBGBXVR(0),
884 	/* DBGOSLAR */
885 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
886 	DBGBXVR(1),
887 	/* DBGOSLSR */
888 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
889 	DBGBXVR(2),
890 	DBGBXVR(3),
891 	/* DBGOSDLR */
892 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
893 	DBGBXVR(4),
894 	/* DBGPRCR */
895 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
896 	DBGBXVR(5),
897 	DBGBXVR(6),
898 	DBGBXVR(7),
899 	DBGBXVR(8),
900 	DBGBXVR(9),
901 	DBGBXVR(10),
902 	DBGBXVR(11),
903 	DBGBXVR(12),
904 	DBGBXVR(13),
905 	DBGBXVR(14),
906 	DBGBXVR(15),
907 
908 	/* DBGDSAR (32bit) */
909 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
910 
911 	/* DBGDEVID2 */
912 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
913 	/* DBGDEVID1 */
914 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
915 	/* DBGDEVID */
916 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
917 	/* DBGCLAIMSET */
918 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
919 	/* DBGCLAIMCLR */
920 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
921 	/* DBGAUTHSTATUS */
922 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
923 };
924 
925 /* Trapped cp14 64bit registers */
926 static const struct sys_reg_desc cp14_64_regs[] = {
927 	/* DBGDRAR (64bit) */
928 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
929 
930 	/* DBGDSAR (64bit) */
931 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
932 };
933 
934 /*
935  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
936  * depending on the way they are accessed (as a 32bit or a 64bit
937  * register).
938  */
939 static const struct sys_reg_desc cp15_regs[] = {
940 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
941 
942 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
943 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
944 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
945 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
946 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
947 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
948 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
949 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
950 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
951 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
952 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
953 
954 	/*
955 	 * DC{C,I,CI}SW operations:
956 	 */
957 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
958 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
959 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
960 
961 	/* PMU */
962 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), access_pmcr },
963 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), trap_raz_wi },
964 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), trap_raz_wi },
965 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), trap_raz_wi },
966 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), access_pmselr },
967 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), access_pmceid },
968 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), access_pmceid },
969 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), trap_raz_wi },
970 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), trap_raz_wi },
971 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), trap_raz_wi },
972 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), trap_raz_wi },
973 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), trap_raz_wi },
974 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), trap_raz_wi },
975 
976 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
977 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
978 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
979 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
980 
981 	/* ICC_SRE */
982 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), trap_raz_wi },
983 
984 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
985 };
986 
987 static const struct sys_reg_desc cp15_64_regs[] = {
988 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
989 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
990 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
991 };
992 
993 /* Target specific emulation tables */
994 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
995 
996 void kvm_register_target_sys_reg_table(unsigned int target,
997 				       struct kvm_sys_reg_target_table *table)
998 {
999 	target_tables[target] = table;
1000 }
1001 
1002 /* Get specific register table for this target. */
1003 static const struct sys_reg_desc *get_target_table(unsigned target,
1004 						   bool mode_is_64,
1005 						   size_t *num)
1006 {
1007 	struct kvm_sys_reg_target_table *table;
1008 
1009 	table = target_tables[target];
1010 	if (mode_is_64) {
1011 		*num = table->table64.num;
1012 		return table->table64.table;
1013 	} else {
1014 		*num = table->table32.num;
1015 		return table->table32.table;
1016 	}
1017 }
1018 
1019 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
1020 					 const struct sys_reg_desc table[],
1021 					 unsigned int num)
1022 {
1023 	unsigned int i;
1024 
1025 	for (i = 0; i < num; i++) {
1026 		const struct sys_reg_desc *r = &table[i];
1027 
1028 		if (params->Op0 != r->Op0)
1029 			continue;
1030 		if (params->Op1 != r->Op1)
1031 			continue;
1032 		if (params->CRn != r->CRn)
1033 			continue;
1034 		if (params->CRm != r->CRm)
1035 			continue;
1036 		if (params->Op2 != r->Op2)
1037 			continue;
1038 
1039 		return r;
1040 	}
1041 	return NULL;
1042 }
1043 
1044 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
1045 {
1046 	kvm_inject_undefined(vcpu);
1047 	return 1;
1048 }
1049 
1050 /*
1051  * emulate_cp --  tries to match a sys_reg access in a handling table, and
1052  *                call the corresponding trap handler.
1053  *
1054  * @params: pointer to the descriptor of the access
1055  * @table: array of trap descriptors
1056  * @num: size of the trap descriptor array
1057  *
1058  * Return 0 if the access has been handled, and -1 if not.
1059  */
1060 static int emulate_cp(struct kvm_vcpu *vcpu,
1061 		      struct sys_reg_params *params,
1062 		      const struct sys_reg_desc *table,
1063 		      size_t num)
1064 {
1065 	const struct sys_reg_desc *r;
1066 
1067 	if (!table)
1068 		return -1;	/* Not handled */
1069 
1070 	r = find_reg(params, table, num);
1071 
1072 	if (r) {
1073 		/*
1074 		 * Not having an accessor means that we have
1075 		 * configured a trap that we don't know how to
1076 		 * handle. This certainly qualifies as a gross bug
1077 		 * that should be fixed right away.
1078 		 */
1079 		BUG_ON(!r->access);
1080 
1081 		if (likely(r->access(vcpu, params, r))) {
1082 			/* Skip instruction, since it was emulated */
1083 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1084 			/* Handled */
1085 			return 0;
1086 		}
1087 	}
1088 
1089 	/* Not handled */
1090 	return -1;
1091 }
1092 
1093 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
1094 				struct sys_reg_params *params)
1095 {
1096 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
1097 	int cp;
1098 
1099 	switch(hsr_ec) {
1100 	case ESR_ELx_EC_CP15_32:
1101 	case ESR_ELx_EC_CP15_64:
1102 		cp = 15;
1103 		break;
1104 	case ESR_ELx_EC_CP14_MR:
1105 	case ESR_ELx_EC_CP14_64:
1106 		cp = 14;
1107 		break;
1108 	default:
1109 		WARN_ON((cp = -1));
1110 	}
1111 
1112 	kvm_err("Unsupported guest CP%d access at: %08lx\n",
1113 		cp, *vcpu_pc(vcpu));
1114 	print_sys_reg_instr(params);
1115 	kvm_inject_undefined(vcpu);
1116 }
1117 
1118 /**
1119  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
1120  * @vcpu: The VCPU pointer
1121  * @run:  The kvm_run struct
1122  */
1123 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
1124 			    const struct sys_reg_desc *global,
1125 			    size_t nr_global,
1126 			    const struct sys_reg_desc *target_specific,
1127 			    size_t nr_specific)
1128 {
1129 	struct sys_reg_params params;
1130 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1131 	int Rt = (hsr >> 5) & 0xf;
1132 	int Rt2 = (hsr >> 10) & 0xf;
1133 
1134 	params.is_aarch32 = true;
1135 	params.is_32bit = false;
1136 	params.CRm = (hsr >> 1) & 0xf;
1137 	params.is_write = ((hsr & 1) == 0);
1138 
1139 	params.Op0 = 0;
1140 	params.Op1 = (hsr >> 16) & 0xf;
1141 	params.Op2 = 0;
1142 	params.CRn = 0;
1143 
1144 	/*
1145 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
1146 	 * backends between AArch32 and AArch64, we get away with it.
1147 	 */
1148 	if (params.is_write) {
1149 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
1150 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
1151 	}
1152 
1153 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
1154 		goto out;
1155 	if (!emulate_cp(vcpu, &params, global, nr_global))
1156 		goto out;
1157 
1158 	unhandled_cp_access(vcpu, &params);
1159 
1160 out:
1161 	/* Split up the value between registers for the read side */
1162 	if (!params.is_write) {
1163 		vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
1164 		vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
1165 	}
1166 
1167 	return 1;
1168 }
1169 
1170 /**
1171  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
1172  * @vcpu: The VCPU pointer
1173  * @run:  The kvm_run struct
1174  */
1175 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
1176 			    const struct sys_reg_desc *global,
1177 			    size_t nr_global,
1178 			    const struct sys_reg_desc *target_specific,
1179 			    size_t nr_specific)
1180 {
1181 	struct sys_reg_params params;
1182 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1183 	int Rt  = (hsr >> 5) & 0xf;
1184 
1185 	params.is_aarch32 = true;
1186 	params.is_32bit = true;
1187 	params.CRm = (hsr >> 1) & 0xf;
1188 	params.regval = vcpu_get_reg(vcpu, Rt);
1189 	params.is_write = ((hsr & 1) == 0);
1190 	params.CRn = (hsr >> 10) & 0xf;
1191 	params.Op0 = 0;
1192 	params.Op1 = (hsr >> 14) & 0x7;
1193 	params.Op2 = (hsr >> 17) & 0x7;
1194 
1195 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific) ||
1196 	    !emulate_cp(vcpu, &params, global, nr_global)) {
1197 		if (!params.is_write)
1198 			vcpu_set_reg(vcpu, Rt, params.regval);
1199 		return 1;
1200 	}
1201 
1202 	unhandled_cp_access(vcpu, &params);
1203 	return 1;
1204 }
1205 
1206 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1207 {
1208 	const struct sys_reg_desc *target_specific;
1209 	size_t num;
1210 
1211 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1212 	return kvm_handle_cp_64(vcpu,
1213 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
1214 				target_specific, num);
1215 }
1216 
1217 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1218 {
1219 	const struct sys_reg_desc *target_specific;
1220 	size_t num;
1221 
1222 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1223 	return kvm_handle_cp_32(vcpu,
1224 				cp15_regs, ARRAY_SIZE(cp15_regs),
1225 				target_specific, num);
1226 }
1227 
1228 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1229 {
1230 	return kvm_handle_cp_64(vcpu,
1231 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
1232 				NULL, 0);
1233 }
1234 
1235 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1236 {
1237 	return kvm_handle_cp_32(vcpu,
1238 				cp14_regs, ARRAY_SIZE(cp14_regs),
1239 				NULL, 0);
1240 }
1241 
1242 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
1243 			   struct sys_reg_params *params)
1244 {
1245 	size_t num;
1246 	const struct sys_reg_desc *table, *r;
1247 
1248 	table = get_target_table(vcpu->arch.target, true, &num);
1249 
1250 	/* Search target-specific then generic table. */
1251 	r = find_reg(params, table, num);
1252 	if (!r)
1253 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1254 
1255 	if (likely(r)) {
1256 		/*
1257 		 * Not having an accessor means that we have
1258 		 * configured a trap that we don't know how to
1259 		 * handle. This certainly qualifies as a gross bug
1260 		 * that should be fixed right away.
1261 		 */
1262 		BUG_ON(!r->access);
1263 
1264 		if (likely(r->access(vcpu, params, r))) {
1265 			/* Skip instruction, since it was emulated */
1266 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1267 			return 1;
1268 		}
1269 		/* If access function fails, it should complain. */
1270 	} else {
1271 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
1272 			*vcpu_pc(vcpu));
1273 		print_sys_reg_instr(params);
1274 	}
1275 	kvm_inject_undefined(vcpu);
1276 	return 1;
1277 }
1278 
1279 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
1280 			      const struct sys_reg_desc *table, size_t num)
1281 {
1282 	unsigned long i;
1283 
1284 	for (i = 0; i < num; i++)
1285 		if (table[i].reset)
1286 			table[i].reset(vcpu, &table[i]);
1287 }
1288 
1289 /**
1290  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
1291  * @vcpu: The VCPU pointer
1292  * @run:  The kvm_run struct
1293  */
1294 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1295 {
1296 	struct sys_reg_params params;
1297 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1298 	int Rt = (esr >> 5) & 0x1f;
1299 	int ret;
1300 
1301 	trace_kvm_handle_sys_reg(esr);
1302 
1303 	params.is_aarch32 = false;
1304 	params.is_32bit = false;
1305 	params.Op0 = (esr >> 20) & 3;
1306 	params.Op1 = (esr >> 14) & 0x7;
1307 	params.CRn = (esr >> 10) & 0xf;
1308 	params.CRm = (esr >> 1) & 0xf;
1309 	params.Op2 = (esr >> 17) & 0x7;
1310 	params.regval = vcpu_get_reg(vcpu, Rt);
1311 	params.is_write = !(esr & 1);
1312 
1313 	ret = emulate_sys_reg(vcpu, &params);
1314 
1315 	if (!params.is_write)
1316 		vcpu_set_reg(vcpu, Rt, params.regval);
1317 	return ret;
1318 }
1319 
1320 /******************************************************************************
1321  * Userspace API
1322  *****************************************************************************/
1323 
1324 static bool index_to_params(u64 id, struct sys_reg_params *params)
1325 {
1326 	switch (id & KVM_REG_SIZE_MASK) {
1327 	case KVM_REG_SIZE_U64:
1328 		/* Any unused index bits means it's not valid. */
1329 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1330 			      | KVM_REG_ARM_COPROC_MASK
1331 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
1332 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
1333 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
1334 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
1335 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
1336 			return false;
1337 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1338 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1339 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1340 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1341 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1342 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1343 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1344 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1345 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1346 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1347 		return true;
1348 	default:
1349 		return false;
1350 	}
1351 }
1352 
1353 /* Decode an index value, and find the sys_reg_desc entry. */
1354 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1355 						    u64 id)
1356 {
1357 	size_t num;
1358 	const struct sys_reg_desc *table, *r;
1359 	struct sys_reg_params params;
1360 
1361 	/* We only do sys_reg for now. */
1362 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1363 		return NULL;
1364 
1365 	if (!index_to_params(id, &params))
1366 		return NULL;
1367 
1368 	table = get_target_table(vcpu->arch.target, true, &num);
1369 	r = find_reg(&params, table, num);
1370 	if (!r)
1371 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1372 
1373 	/* Not saved in the sys_reg array? */
1374 	if (r && !r->reg)
1375 		r = NULL;
1376 
1377 	return r;
1378 }
1379 
1380 /*
1381  * These are the invariant sys_reg registers: we let the guest see the
1382  * host versions of these, so they're part of the guest state.
1383  *
1384  * A future CPU may provide a mechanism to present different values to
1385  * the guest, or a future kvm may trap them.
1386  */
1387 
1388 #define FUNCTION_INVARIANT(reg)						\
1389 	static void get_##reg(struct kvm_vcpu *v,			\
1390 			      const struct sys_reg_desc *r)		\
1391 	{								\
1392 		u64 val;						\
1393 									\
1394 		asm volatile("mrs %0, " __stringify(reg) "\n"		\
1395 			     : "=r" (val));				\
1396 		((struct sys_reg_desc *)r)->val = val;			\
1397 	}
1398 
1399 FUNCTION_INVARIANT(midr_el1)
1400 FUNCTION_INVARIANT(ctr_el0)
1401 FUNCTION_INVARIANT(revidr_el1)
1402 FUNCTION_INVARIANT(id_pfr0_el1)
1403 FUNCTION_INVARIANT(id_pfr1_el1)
1404 FUNCTION_INVARIANT(id_dfr0_el1)
1405 FUNCTION_INVARIANT(id_afr0_el1)
1406 FUNCTION_INVARIANT(id_mmfr0_el1)
1407 FUNCTION_INVARIANT(id_mmfr1_el1)
1408 FUNCTION_INVARIANT(id_mmfr2_el1)
1409 FUNCTION_INVARIANT(id_mmfr3_el1)
1410 FUNCTION_INVARIANT(id_isar0_el1)
1411 FUNCTION_INVARIANT(id_isar1_el1)
1412 FUNCTION_INVARIANT(id_isar2_el1)
1413 FUNCTION_INVARIANT(id_isar3_el1)
1414 FUNCTION_INVARIANT(id_isar4_el1)
1415 FUNCTION_INVARIANT(id_isar5_el1)
1416 FUNCTION_INVARIANT(clidr_el1)
1417 FUNCTION_INVARIANT(aidr_el1)
1418 
1419 /* ->val is filled in by kvm_sys_reg_table_init() */
1420 static struct sys_reg_desc invariant_sys_regs[] = {
1421 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1422 	  NULL, get_midr_el1 },
1423 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1424 	  NULL, get_revidr_el1 },
1425 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1426 	  NULL, get_id_pfr0_el1 },
1427 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1428 	  NULL, get_id_pfr1_el1 },
1429 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1430 	  NULL, get_id_dfr0_el1 },
1431 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1432 	  NULL, get_id_afr0_el1 },
1433 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1434 	  NULL, get_id_mmfr0_el1 },
1435 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1436 	  NULL, get_id_mmfr1_el1 },
1437 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1438 	  NULL, get_id_mmfr2_el1 },
1439 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1440 	  NULL, get_id_mmfr3_el1 },
1441 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1442 	  NULL, get_id_isar0_el1 },
1443 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1444 	  NULL, get_id_isar1_el1 },
1445 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1446 	  NULL, get_id_isar2_el1 },
1447 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1448 	  NULL, get_id_isar3_el1 },
1449 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1450 	  NULL, get_id_isar4_el1 },
1451 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1452 	  NULL, get_id_isar5_el1 },
1453 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1454 	  NULL, get_clidr_el1 },
1455 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1456 	  NULL, get_aidr_el1 },
1457 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1458 	  NULL, get_ctr_el0 },
1459 };
1460 
1461 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1462 {
1463 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1464 		return -EFAULT;
1465 	return 0;
1466 }
1467 
1468 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1469 {
1470 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1471 		return -EFAULT;
1472 	return 0;
1473 }
1474 
1475 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1476 {
1477 	struct sys_reg_params params;
1478 	const struct sys_reg_desc *r;
1479 
1480 	if (!index_to_params(id, &params))
1481 		return -ENOENT;
1482 
1483 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1484 	if (!r)
1485 		return -ENOENT;
1486 
1487 	return reg_to_user(uaddr, &r->val, id);
1488 }
1489 
1490 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1491 {
1492 	struct sys_reg_params params;
1493 	const struct sys_reg_desc *r;
1494 	int err;
1495 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1496 
1497 	if (!index_to_params(id, &params))
1498 		return -ENOENT;
1499 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1500 	if (!r)
1501 		return -ENOENT;
1502 
1503 	err = reg_from_user(&val, uaddr, id);
1504 	if (err)
1505 		return err;
1506 
1507 	/* This is what we mean by invariant: you can't change it. */
1508 	if (r->val != val)
1509 		return -EINVAL;
1510 
1511 	return 0;
1512 }
1513 
1514 static bool is_valid_cache(u32 val)
1515 {
1516 	u32 level, ctype;
1517 
1518 	if (val >= CSSELR_MAX)
1519 		return false;
1520 
1521 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1522 	level = (val >> 1);
1523 	ctype = (cache_levels >> (level * 3)) & 7;
1524 
1525 	switch (ctype) {
1526 	case 0: /* No cache */
1527 		return false;
1528 	case 1: /* Instruction cache only */
1529 		return (val & 1);
1530 	case 2: /* Data cache only */
1531 	case 4: /* Unified cache */
1532 		return !(val & 1);
1533 	case 3: /* Separate instruction and data caches */
1534 		return true;
1535 	default: /* Reserved: we can't know instruction or data. */
1536 		return false;
1537 	}
1538 }
1539 
1540 static int demux_c15_get(u64 id, void __user *uaddr)
1541 {
1542 	u32 val;
1543 	u32 __user *uval = uaddr;
1544 
1545 	/* Fail if we have unknown bits set. */
1546 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1547 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1548 		return -ENOENT;
1549 
1550 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1551 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1552 		if (KVM_REG_SIZE(id) != 4)
1553 			return -ENOENT;
1554 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1555 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1556 		if (!is_valid_cache(val))
1557 			return -ENOENT;
1558 
1559 		return put_user(get_ccsidr(val), uval);
1560 	default:
1561 		return -ENOENT;
1562 	}
1563 }
1564 
1565 static int demux_c15_set(u64 id, void __user *uaddr)
1566 {
1567 	u32 val, newval;
1568 	u32 __user *uval = uaddr;
1569 
1570 	/* Fail if we have unknown bits set. */
1571 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1572 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1573 		return -ENOENT;
1574 
1575 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1576 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1577 		if (KVM_REG_SIZE(id) != 4)
1578 			return -ENOENT;
1579 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1580 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1581 		if (!is_valid_cache(val))
1582 			return -ENOENT;
1583 
1584 		if (get_user(newval, uval))
1585 			return -EFAULT;
1586 
1587 		/* This is also invariant: you can't change it. */
1588 		if (newval != get_ccsidr(val))
1589 			return -EINVAL;
1590 		return 0;
1591 	default:
1592 		return -ENOENT;
1593 	}
1594 }
1595 
1596 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1597 {
1598 	const struct sys_reg_desc *r;
1599 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1600 
1601 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1602 		return demux_c15_get(reg->id, uaddr);
1603 
1604 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1605 		return -ENOENT;
1606 
1607 	r = index_to_sys_reg_desc(vcpu, reg->id);
1608 	if (!r)
1609 		return get_invariant_sys_reg(reg->id, uaddr);
1610 
1611 	if (r->get_user)
1612 		return (r->get_user)(vcpu, r, reg, uaddr);
1613 
1614 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
1615 }
1616 
1617 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1618 {
1619 	const struct sys_reg_desc *r;
1620 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1621 
1622 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1623 		return demux_c15_set(reg->id, uaddr);
1624 
1625 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1626 		return -ENOENT;
1627 
1628 	r = index_to_sys_reg_desc(vcpu, reg->id);
1629 	if (!r)
1630 		return set_invariant_sys_reg(reg->id, uaddr);
1631 
1632 	if (r->set_user)
1633 		return (r->set_user)(vcpu, r, reg, uaddr);
1634 
1635 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
1636 }
1637 
1638 static unsigned int num_demux_regs(void)
1639 {
1640 	unsigned int i, count = 0;
1641 
1642 	for (i = 0; i < CSSELR_MAX; i++)
1643 		if (is_valid_cache(i))
1644 			count++;
1645 
1646 	return count;
1647 }
1648 
1649 static int write_demux_regids(u64 __user *uindices)
1650 {
1651 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
1652 	unsigned int i;
1653 
1654 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
1655 	for (i = 0; i < CSSELR_MAX; i++) {
1656 		if (!is_valid_cache(i))
1657 			continue;
1658 		if (put_user(val | i, uindices))
1659 			return -EFAULT;
1660 		uindices++;
1661 	}
1662 	return 0;
1663 }
1664 
1665 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
1666 {
1667 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
1668 		KVM_REG_ARM64_SYSREG |
1669 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
1670 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
1671 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
1672 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
1673 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
1674 }
1675 
1676 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
1677 {
1678 	if (!*uind)
1679 		return true;
1680 
1681 	if (put_user(sys_reg_to_index(reg), *uind))
1682 		return false;
1683 
1684 	(*uind)++;
1685 	return true;
1686 }
1687 
1688 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
1689 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
1690 {
1691 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
1692 	unsigned int total = 0;
1693 	size_t num;
1694 
1695 	/* We check for duplicates here, to allow arch-specific overrides. */
1696 	i1 = get_target_table(vcpu->arch.target, true, &num);
1697 	end1 = i1 + num;
1698 	i2 = sys_reg_descs;
1699 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
1700 
1701 	BUG_ON(i1 == end1 || i2 == end2);
1702 
1703 	/* Walk carefully, as both tables may refer to the same register. */
1704 	while (i1 || i2) {
1705 		int cmp = cmp_sys_reg(i1, i2);
1706 		/* target-specific overrides generic entry. */
1707 		if (cmp <= 0) {
1708 			/* Ignore registers we trap but don't save. */
1709 			if (i1->reg) {
1710 				if (!copy_reg_to_user(i1, &uind))
1711 					return -EFAULT;
1712 				total++;
1713 			}
1714 		} else {
1715 			/* Ignore registers we trap but don't save. */
1716 			if (i2->reg) {
1717 				if (!copy_reg_to_user(i2, &uind))
1718 					return -EFAULT;
1719 				total++;
1720 			}
1721 		}
1722 
1723 		if (cmp <= 0 && ++i1 == end1)
1724 			i1 = NULL;
1725 		if (cmp >= 0 && ++i2 == end2)
1726 			i2 = NULL;
1727 	}
1728 	return total;
1729 }
1730 
1731 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
1732 {
1733 	return ARRAY_SIZE(invariant_sys_regs)
1734 		+ num_demux_regs()
1735 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
1736 }
1737 
1738 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
1739 {
1740 	unsigned int i;
1741 	int err;
1742 
1743 	/* Then give them all the invariant registers' indices. */
1744 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
1745 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
1746 			return -EFAULT;
1747 		uindices++;
1748 	}
1749 
1750 	err = walk_sys_regs(vcpu, uindices);
1751 	if (err < 0)
1752 		return err;
1753 	uindices += err;
1754 
1755 	return write_demux_regids(uindices);
1756 }
1757 
1758 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
1759 {
1760 	unsigned int i;
1761 
1762 	for (i = 1; i < n; i++) {
1763 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
1764 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
1765 			return 1;
1766 		}
1767 	}
1768 
1769 	return 0;
1770 }
1771 
1772 void kvm_sys_reg_table_init(void)
1773 {
1774 	unsigned int i;
1775 	struct sys_reg_desc clidr;
1776 
1777 	/* Make sure tables are unique and in order. */
1778 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
1779 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
1780 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
1781 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
1782 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
1783 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
1784 
1785 	/* We abuse the reset function to overwrite the table itself. */
1786 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
1787 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
1788 
1789 	/*
1790 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
1791 	 *
1792 	 *   If software reads the Cache Type fields from Ctype1
1793 	 *   upwards, once it has seen a value of 0b000, no caches
1794 	 *   exist at further-out levels of the hierarchy. So, for
1795 	 *   example, if Ctype3 is the first Cache Type field with a
1796 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
1797 	 *   ignored.
1798 	 */
1799 	get_clidr_el1(NULL, &clidr); /* Ugly... */
1800 	cache_levels = clidr.val;
1801 	for (i = 0; i < 7; i++)
1802 		if (((cache_levels >> (i*3)) & 7) == 0)
1803 			break;
1804 	/* Clear all higher bits. */
1805 	cache_levels &= (1 << (i*3))-1;
1806 }
1807 
1808 /**
1809  * kvm_reset_sys_regs - sets system registers to reset value
1810  * @vcpu: The VCPU pointer
1811  *
1812  * This function finds the right table above and sets the registers on the
1813  * virtual CPU struct to their architecturally defined reset values.
1814  */
1815 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
1816 {
1817 	size_t num;
1818 	const struct sys_reg_desc *table;
1819 
1820 	/* Catch someone adding a register without putting in reset entry. */
1821 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
1822 
1823 	/* Generic chip reset first (so target could override). */
1824 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1825 
1826 	table = get_target_table(vcpu->arch.target, true, &num);
1827 	reset_sys_reg_descs(vcpu, table, num);
1828 
1829 	for (num = 1; num < NR_SYS_REGS; num++)
1830 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
1831 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
1832 }
1833