xref: /linux/arch/arm64/kvm/sys_regs.c (revision 3aed61d1eb06b8b19b7bb09d49b222ebc3f83347)
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_coproc.h>
33 #include <asm/kvm_emulate.h>
34 #include <asm/kvm_host.h>
35 #include <asm/kvm_mmu.h>
36 
37 #include <trace/events/kvm.h>
38 
39 #include "sys_regs.h"
40 
41 #include "trace.h"
42 
43 /*
44  * All of this file is extremly similar to the ARM coproc.c, but the
45  * types are different. My gut feeling is that it should be pretty
46  * easy to merge, but that would be an ABI breakage -- again. VFP
47  * would also need to be abstracted.
48  *
49  * For AArch32, we only take care of what is being trapped. Anything
50  * that has to do with init and userspace access has to go via the
51  * 64bit interface.
52  */
53 
54 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
55 static u32 cache_levels;
56 
57 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
58 #define CSSELR_MAX 12
59 
60 /* Which cache CCSIDR represents depends on CSSELR value. */
61 static u32 get_ccsidr(u32 csselr)
62 {
63 	u32 ccsidr;
64 
65 	/* Make sure noone else changes CSSELR during this! */
66 	local_irq_disable();
67 	/* Put value into CSSELR */
68 	asm volatile("msr csselr_el1, %x0" : : "r" (csselr));
69 	isb();
70 	/* Read result out of CCSIDR */
71 	asm volatile("mrs %0, ccsidr_el1" : "=r" (ccsidr));
72 	local_irq_enable();
73 
74 	return ccsidr;
75 }
76 
77 /*
78  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
79  */
80 static bool access_dcsw(struct kvm_vcpu *vcpu,
81 			const struct sys_reg_params *p,
82 			const struct sys_reg_desc *r)
83 {
84 	if (!p->is_write)
85 		return read_from_write_only(vcpu, p);
86 
87 	kvm_set_way_flush(vcpu);
88 	return true;
89 }
90 
91 /*
92  * Generic accessor for VM registers. Only called as long as HCR_TVM
93  * is set. If the guest enables the MMU, we stop trapping the VM
94  * sys_regs and leave it in complete control of the caches.
95  */
96 static bool access_vm_reg(struct kvm_vcpu *vcpu,
97 			  const struct sys_reg_params *p,
98 			  const struct sys_reg_desc *r)
99 {
100 	unsigned long val;
101 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
102 
103 	BUG_ON(!p->is_write);
104 
105 	val = *vcpu_reg(vcpu, p->Rt);
106 	if (!p->is_aarch32) {
107 		vcpu_sys_reg(vcpu, r->reg) = val;
108 	} else {
109 		if (!p->is_32bit)
110 			vcpu_cp15_64_high(vcpu, r->reg) = val >> 32;
111 		vcpu_cp15_64_low(vcpu, r->reg) = val & 0xffffffffUL;
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 			   const struct sys_reg_params *p,
126 			   const struct sys_reg_desc *r)
127 {
128 	u64 val;
129 
130 	if (!p->is_write)
131 		return read_from_write_only(vcpu, p);
132 
133 	val = *vcpu_reg(vcpu, p->Rt);
134 	vgic_v3_dispatch_sgi(vcpu, val);
135 
136 	return true;
137 }
138 
139 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
140 			const struct sys_reg_params *p,
141 			const struct sys_reg_desc *r)
142 {
143 	if (p->is_write)
144 		return ignore_write(vcpu, p);
145 	else
146 		return read_zero(vcpu, p);
147 }
148 
149 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
150 			   const struct sys_reg_params *p,
151 			   const struct sys_reg_desc *r)
152 {
153 	if (p->is_write) {
154 		return ignore_write(vcpu, p);
155 	} else {
156 		*vcpu_reg(vcpu, p->Rt) = (1 << 3);
157 		return true;
158 	}
159 }
160 
161 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
162 				   const struct sys_reg_params *p,
163 				   const struct sys_reg_desc *r)
164 {
165 	if (p->is_write) {
166 		return ignore_write(vcpu, p);
167 	} else {
168 		u32 val;
169 		asm volatile("mrs %0, dbgauthstatus_el1" : "=r" (val));
170 		*vcpu_reg(vcpu, p->Rt) = val;
171 		return true;
172 	}
173 }
174 
175 /*
176  * We want to avoid world-switching all the DBG registers all the
177  * time:
178  *
179  * - If we've touched any debug register, it is likely that we're
180  *   going to touch more of them. It then makes sense to disable the
181  *   traps and start doing the save/restore dance
182  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
183  *   then mandatory to save/restore the registers, as the guest
184  *   depends on them.
185  *
186  * For this, we use a DIRTY bit, indicating the guest has modified the
187  * debug registers, used as follow:
188  *
189  * On guest entry:
190  * - If the dirty bit is set (because we're coming back from trapping),
191  *   disable the traps, save host registers, restore guest registers.
192  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
193  *   set the dirty bit, disable the traps, save host registers,
194  *   restore guest registers.
195  * - Otherwise, enable the traps
196  *
197  * On guest exit:
198  * - If the dirty bit is set, save guest registers, restore host
199  *   registers and clear the dirty bit. This ensure that the host can
200  *   now use the debug registers.
201  */
202 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
203 			    const struct sys_reg_params *p,
204 			    const struct sys_reg_desc *r)
205 {
206 	if (p->is_write) {
207 		vcpu_sys_reg(vcpu, r->reg) = *vcpu_reg(vcpu, p->Rt);
208 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
209 	} else {
210 		*vcpu_reg(vcpu, p->Rt) = vcpu_sys_reg(vcpu, r->reg);
211 	}
212 
213 	trace_trap_reg(__func__, r->reg, p->is_write, *vcpu_reg(vcpu, p->Rt));
214 
215 	return true;
216 }
217 
218 /*
219  * reg_to_dbg/dbg_to_reg
220  *
221  * A 32 bit write to a debug register leave top bits alone
222  * A 32 bit read from a debug register only returns the bottom bits
223  *
224  * All writes will set the KVM_ARM64_DEBUG_DIRTY flag to ensure the
225  * hyp.S code switches between host and guest values in future.
226  */
227 static inline void reg_to_dbg(struct kvm_vcpu *vcpu,
228 			      const struct sys_reg_params *p,
229 			      u64 *dbg_reg)
230 {
231 	u64 val = *vcpu_reg(vcpu, p->Rt);
232 
233 	if (p->is_32bit) {
234 		val &= 0xffffffffUL;
235 		val |= ((*dbg_reg >> 32) << 32);
236 	}
237 
238 	*dbg_reg = val;
239 	vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
240 }
241 
242 static inline void dbg_to_reg(struct kvm_vcpu *vcpu,
243 			      const struct sys_reg_params *p,
244 			      u64 *dbg_reg)
245 {
246 	u64 val = *dbg_reg;
247 
248 	if (p->is_32bit)
249 		val &= 0xffffffffUL;
250 
251 	*vcpu_reg(vcpu, p->Rt) = val;
252 }
253 
254 static inline bool trap_bvr(struct kvm_vcpu *vcpu,
255 			    const struct sys_reg_params *p,
256 			    const struct sys_reg_desc *rd)
257 {
258 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
259 
260 	if (p->is_write)
261 		reg_to_dbg(vcpu, p, dbg_reg);
262 	else
263 		dbg_to_reg(vcpu, p, dbg_reg);
264 
265 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
266 
267 	return true;
268 }
269 
270 static int set_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
271 		const struct kvm_one_reg *reg, void __user *uaddr)
272 {
273 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
274 
275 	if (copy_from_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
276 		return -EFAULT;
277 	return 0;
278 }
279 
280 static int get_bvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
281 	const struct kvm_one_reg *reg, void __user *uaddr)
282 {
283 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
284 
285 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
286 		return -EFAULT;
287 	return 0;
288 }
289 
290 static inline void reset_bvr(struct kvm_vcpu *vcpu,
291 			     const struct sys_reg_desc *rd)
292 {
293 	vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg] = rd->val;
294 }
295 
296 static inline bool trap_bcr(struct kvm_vcpu *vcpu,
297 			    const struct sys_reg_params *p,
298 			    const struct sys_reg_desc *rd)
299 {
300 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
301 
302 	if (p->is_write)
303 		reg_to_dbg(vcpu, p, dbg_reg);
304 	else
305 		dbg_to_reg(vcpu, p, dbg_reg);
306 
307 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
308 
309 	return true;
310 }
311 
312 static int set_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
313 		const struct kvm_one_reg *reg, void __user *uaddr)
314 {
315 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
316 
317 	if (copy_from_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
318 		return -EFAULT;
319 
320 	return 0;
321 }
322 
323 static int get_bcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
324 	const struct kvm_one_reg *reg, void __user *uaddr)
325 {
326 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg];
327 
328 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
329 		return -EFAULT;
330 	return 0;
331 }
332 
333 static inline void reset_bcr(struct kvm_vcpu *vcpu,
334 			     const struct sys_reg_desc *rd)
335 {
336 	vcpu->arch.vcpu_debug_state.dbg_bcr[rd->reg] = rd->val;
337 }
338 
339 static inline bool trap_wvr(struct kvm_vcpu *vcpu,
340 			    const struct sys_reg_params *p,
341 			    const struct sys_reg_desc *rd)
342 {
343 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
344 
345 	if (p->is_write)
346 		reg_to_dbg(vcpu, p, dbg_reg);
347 	else
348 		dbg_to_reg(vcpu, p, dbg_reg);
349 
350 	trace_trap_reg(__func__, rd->reg, p->is_write,
351 		vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg]);
352 
353 	return true;
354 }
355 
356 static int set_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
357 		const struct kvm_one_reg *reg, void __user *uaddr)
358 {
359 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
360 
361 	if (copy_from_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
362 		return -EFAULT;
363 	return 0;
364 }
365 
366 static int get_wvr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
367 	const struct kvm_one_reg *reg, void __user *uaddr)
368 {
369 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg];
370 
371 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
372 		return -EFAULT;
373 	return 0;
374 }
375 
376 static inline void reset_wvr(struct kvm_vcpu *vcpu,
377 			     const struct sys_reg_desc *rd)
378 {
379 	vcpu->arch.vcpu_debug_state.dbg_wvr[rd->reg] = rd->val;
380 }
381 
382 static inline bool trap_wcr(struct kvm_vcpu *vcpu,
383 			    const struct sys_reg_params *p,
384 			    const struct sys_reg_desc *rd)
385 {
386 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
387 
388 	if (p->is_write)
389 		reg_to_dbg(vcpu, p, dbg_reg);
390 	else
391 		dbg_to_reg(vcpu, p, dbg_reg);
392 
393 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
394 
395 	return true;
396 }
397 
398 static int set_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
399 		const struct kvm_one_reg *reg, void __user *uaddr)
400 {
401 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
402 
403 	if (copy_from_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
404 		return -EFAULT;
405 	return 0;
406 }
407 
408 static int get_wcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
409 	const struct kvm_one_reg *reg, void __user *uaddr)
410 {
411 	__u64 *r = &vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg];
412 
413 	if (copy_to_user(uaddr, r, KVM_REG_SIZE(reg->id)) != 0)
414 		return -EFAULT;
415 	return 0;
416 }
417 
418 static inline void reset_wcr(struct kvm_vcpu *vcpu,
419 			     const struct sys_reg_desc *rd)
420 {
421 	vcpu->arch.vcpu_debug_state.dbg_wcr[rd->reg] = rd->val;
422 }
423 
424 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
425 {
426 	u64 amair;
427 
428 	asm volatile("mrs %0, amair_el1\n" : "=r" (amair));
429 	vcpu_sys_reg(vcpu, AMAIR_EL1) = amair;
430 }
431 
432 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
433 {
434 	u64 mpidr;
435 
436 	/*
437 	 * Map the vcpu_id into the first three affinity level fields of
438 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
439 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
440 	 * of the GICv3 to be able to address each CPU directly when
441 	 * sending IPIs.
442 	 */
443 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
444 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
445 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
446 	vcpu_sys_reg(vcpu, MPIDR_EL1) = (1ULL << 31) | mpidr;
447 }
448 
449 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
450 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
451 	/* DBGBVRn_EL1 */						\
452 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b100),	\
453 	  trap_bvr, reset_bvr, n, 0, get_bvr, set_bvr },		\
454 	/* DBGBCRn_EL1 */						\
455 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b101),	\
456 	  trap_bcr, reset_bcr, n, 0, get_bcr, set_bcr },		\
457 	/* DBGWVRn_EL1 */						\
458 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b110),	\
459 	  trap_wvr, reset_wvr, n, 0,  get_wvr, set_wvr },		\
460 	/* DBGWCRn_EL1 */						\
461 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b111),	\
462 	  trap_wcr, reset_wcr, n, 0,  get_wcr, set_wcr }
463 
464 /*
465  * Architected system registers.
466  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
467  *
468  * We could trap ID_DFR0 and tell the guest we don't support performance
469  * monitoring.  Unfortunately the patch to make the kernel check ID_DFR0 was
470  * NAKed, so it will read the PMCR anyway.
471  *
472  * Therefore we tell the guest we have 0 counters.  Unfortunately, we
473  * must always support PMCCNTR (the cycle counter): we just RAZ/WI for
474  * all PM registers, which doesn't crash the guest kernel at least.
475  *
476  * Debug handling: We do trap most, if not all debug related system
477  * registers. The implementation is good enough to ensure that a guest
478  * can use these with minimal performance degradation. The drawback is
479  * that we don't implement any of the external debug, none of the
480  * OSlock protocol. This should be revisited if we ever encounter a
481  * more demanding guest...
482  */
483 static const struct sys_reg_desc sys_reg_descs[] = {
484 	/* DC ISW */
485 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
486 	  access_dcsw },
487 	/* DC CSW */
488 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
489 	  access_dcsw },
490 	/* DC CISW */
491 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
492 	  access_dcsw },
493 
494 	DBG_BCR_BVR_WCR_WVR_EL1(0),
495 	DBG_BCR_BVR_WCR_WVR_EL1(1),
496 	/* MDCCINT_EL1 */
497 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
498 	  trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
499 	/* MDSCR_EL1 */
500 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
501 	  trap_debug_regs, reset_val, MDSCR_EL1, 0 },
502 	DBG_BCR_BVR_WCR_WVR_EL1(2),
503 	DBG_BCR_BVR_WCR_WVR_EL1(3),
504 	DBG_BCR_BVR_WCR_WVR_EL1(4),
505 	DBG_BCR_BVR_WCR_WVR_EL1(5),
506 	DBG_BCR_BVR_WCR_WVR_EL1(6),
507 	DBG_BCR_BVR_WCR_WVR_EL1(7),
508 	DBG_BCR_BVR_WCR_WVR_EL1(8),
509 	DBG_BCR_BVR_WCR_WVR_EL1(9),
510 	DBG_BCR_BVR_WCR_WVR_EL1(10),
511 	DBG_BCR_BVR_WCR_WVR_EL1(11),
512 	DBG_BCR_BVR_WCR_WVR_EL1(12),
513 	DBG_BCR_BVR_WCR_WVR_EL1(13),
514 	DBG_BCR_BVR_WCR_WVR_EL1(14),
515 	DBG_BCR_BVR_WCR_WVR_EL1(15),
516 
517 	/* MDRAR_EL1 */
518 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
519 	  trap_raz_wi },
520 	/* OSLAR_EL1 */
521 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
522 	  trap_raz_wi },
523 	/* OSLSR_EL1 */
524 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
525 	  trap_oslsr_el1 },
526 	/* OSDLR_EL1 */
527 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
528 	  trap_raz_wi },
529 	/* DBGPRCR_EL1 */
530 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
531 	  trap_raz_wi },
532 	/* DBGCLAIMSET_EL1 */
533 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
534 	  trap_raz_wi },
535 	/* DBGCLAIMCLR_EL1 */
536 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
537 	  trap_raz_wi },
538 	/* DBGAUTHSTATUS_EL1 */
539 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
540 	  trap_dbgauthstatus_el1 },
541 
542 	/* TEECR32_EL1 */
543 	{ Op0(0b10), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
544 	  NULL, reset_val, TEECR32_EL1, 0 },
545 	/* TEEHBR32_EL1 */
546 	{ Op0(0b10), Op1(0b010), CRn(0b0001), CRm(0b0000), Op2(0b000),
547 	  NULL, reset_val, TEEHBR32_EL1, 0 },
548 
549 	/* MDCCSR_EL1 */
550 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
551 	  trap_raz_wi },
552 	/* DBGDTR_EL0 */
553 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
554 	  trap_raz_wi },
555 	/* DBGDTR[TR]X_EL0 */
556 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
557 	  trap_raz_wi },
558 
559 	/* DBGVCR32_EL2 */
560 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
561 	  NULL, reset_val, DBGVCR32_EL2, 0 },
562 
563 	/* MPIDR_EL1 */
564 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
565 	  NULL, reset_mpidr, MPIDR_EL1 },
566 	/* SCTLR_EL1 */
567 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
568 	  access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
569 	/* CPACR_EL1 */
570 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
571 	  NULL, reset_val, CPACR_EL1, 0 },
572 	/* TTBR0_EL1 */
573 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
574 	  access_vm_reg, reset_unknown, TTBR0_EL1 },
575 	/* TTBR1_EL1 */
576 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
577 	  access_vm_reg, reset_unknown, TTBR1_EL1 },
578 	/* TCR_EL1 */
579 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
580 	  access_vm_reg, reset_val, TCR_EL1, 0 },
581 
582 	/* AFSR0_EL1 */
583 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
584 	  access_vm_reg, reset_unknown, AFSR0_EL1 },
585 	/* AFSR1_EL1 */
586 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
587 	  access_vm_reg, reset_unknown, AFSR1_EL1 },
588 	/* ESR_EL1 */
589 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
590 	  access_vm_reg, reset_unknown, ESR_EL1 },
591 	/* FAR_EL1 */
592 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
593 	  access_vm_reg, reset_unknown, FAR_EL1 },
594 	/* PAR_EL1 */
595 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
596 	  NULL, reset_unknown, PAR_EL1 },
597 
598 	/* PMINTENSET_EL1 */
599 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
600 	  trap_raz_wi },
601 	/* PMINTENCLR_EL1 */
602 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
603 	  trap_raz_wi },
604 
605 	/* MAIR_EL1 */
606 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
607 	  access_vm_reg, reset_unknown, MAIR_EL1 },
608 	/* AMAIR_EL1 */
609 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
610 	  access_vm_reg, reset_amair_el1, AMAIR_EL1 },
611 
612 	/* VBAR_EL1 */
613 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
614 	  NULL, reset_val, VBAR_EL1, 0 },
615 
616 	/* ICC_SGI1R_EL1 */
617 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1011), Op2(0b101),
618 	  access_gic_sgi },
619 	/* ICC_SRE_EL1 */
620 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b1100), Op2(0b101),
621 	  trap_raz_wi },
622 
623 	/* CONTEXTIDR_EL1 */
624 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
625 	  access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
626 	/* TPIDR_EL1 */
627 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
628 	  NULL, reset_unknown, TPIDR_EL1 },
629 
630 	/* CNTKCTL_EL1 */
631 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
632 	  NULL, reset_val, CNTKCTL_EL1, 0},
633 
634 	/* CSSELR_EL1 */
635 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
636 	  NULL, reset_unknown, CSSELR_EL1 },
637 
638 	/* PMCR_EL0 */
639 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
640 	  trap_raz_wi },
641 	/* PMCNTENSET_EL0 */
642 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
643 	  trap_raz_wi },
644 	/* PMCNTENCLR_EL0 */
645 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
646 	  trap_raz_wi },
647 	/* PMOVSCLR_EL0 */
648 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
649 	  trap_raz_wi },
650 	/* PMSWINC_EL0 */
651 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
652 	  trap_raz_wi },
653 	/* PMSELR_EL0 */
654 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
655 	  trap_raz_wi },
656 	/* PMCEID0_EL0 */
657 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
658 	  trap_raz_wi },
659 	/* PMCEID1_EL0 */
660 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
661 	  trap_raz_wi },
662 	/* PMCCNTR_EL0 */
663 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
664 	  trap_raz_wi },
665 	/* PMXEVTYPER_EL0 */
666 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
667 	  trap_raz_wi },
668 	/* PMXEVCNTR_EL0 */
669 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
670 	  trap_raz_wi },
671 	/* PMUSERENR_EL0 */
672 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
673 	  trap_raz_wi },
674 	/* PMOVSSET_EL0 */
675 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
676 	  trap_raz_wi },
677 
678 	/* TPIDR_EL0 */
679 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
680 	  NULL, reset_unknown, TPIDR_EL0 },
681 	/* TPIDRRO_EL0 */
682 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
683 	  NULL, reset_unknown, TPIDRRO_EL0 },
684 
685 	/* DACR32_EL2 */
686 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
687 	  NULL, reset_unknown, DACR32_EL2 },
688 	/* IFSR32_EL2 */
689 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
690 	  NULL, reset_unknown, IFSR32_EL2 },
691 	/* FPEXC32_EL2 */
692 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
693 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
694 };
695 
696 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
697 			const struct sys_reg_params *p,
698 			const struct sys_reg_desc *r)
699 {
700 	if (p->is_write) {
701 		return ignore_write(vcpu, p);
702 	} else {
703 		u64 dfr = read_cpuid(ID_AA64DFR0_EL1);
704 		u64 pfr = read_cpuid(ID_AA64PFR0_EL1);
705 		u32 el3 = !!((pfr >> 12) & 0xf);
706 
707 		*vcpu_reg(vcpu, p->Rt) = ((((dfr >> 20) & 0xf) << 28) |
708 					  (((dfr >> 12) & 0xf) << 24) |
709 					  (((dfr >> 28) & 0xf) << 20) |
710 					  (6 << 16) | (el3 << 14) | (el3 << 12));
711 		return true;
712 	}
713 }
714 
715 static bool trap_debug32(struct kvm_vcpu *vcpu,
716 			 const struct sys_reg_params *p,
717 			 const struct sys_reg_desc *r)
718 {
719 	if (p->is_write) {
720 		vcpu_cp14(vcpu, r->reg) = *vcpu_reg(vcpu, p->Rt);
721 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
722 	} else {
723 		*vcpu_reg(vcpu, p->Rt) = vcpu_cp14(vcpu, r->reg);
724 	}
725 
726 	return true;
727 }
728 
729 /* AArch32 debug register mappings
730  *
731  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
732  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
733  *
734  * All control registers and watchpoint value registers are mapped to
735  * the lower 32 bits of their AArch64 equivalents. We share the trap
736  * handlers with the above AArch64 code which checks what mode the
737  * system is in.
738  */
739 
740 static inline bool trap_xvr(struct kvm_vcpu *vcpu,
741 			    const struct sys_reg_params *p,
742 			    const struct sys_reg_desc *rd)
743 {
744 	u64 *dbg_reg = &vcpu->arch.vcpu_debug_state.dbg_bvr[rd->reg];
745 
746 	if (p->is_write) {
747 		u64 val = *dbg_reg;
748 
749 		val &= 0xffffffffUL;
750 		val |= *vcpu_reg(vcpu, p->Rt) << 32;
751 		*dbg_reg = val;
752 
753 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
754 	} else {
755 		*vcpu_reg(vcpu, p->Rt) = *dbg_reg >> 32;
756 	}
757 
758 	trace_trap_reg(__func__, rd->reg, p->is_write, *dbg_reg);
759 
760 	return true;
761 }
762 
763 #define DBG_BCR_BVR_WCR_WVR(n)						\
764 	/* DBGBVRn */							\
765 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_bvr, NULL, n }, 	\
766 	/* DBGBCRn */							\
767 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_bcr, NULL, n },	\
768 	/* DBGWVRn */							\
769 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_wvr, NULL, n },	\
770 	/* DBGWCRn */							\
771 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_wcr, NULL, n }
772 
773 #define DBGBXVR(n)							\
774 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_xvr, NULL, n }
775 
776 /*
777  * Trapped cp14 registers. We generally ignore most of the external
778  * debug, on the principle that they don't really make sense to a
779  * guest. Revisit this one day, would this principle change.
780  */
781 static const struct sys_reg_desc cp14_regs[] = {
782 	/* DBGIDR */
783 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
784 	/* DBGDTRRXext */
785 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
786 
787 	DBG_BCR_BVR_WCR_WVR(0),
788 	/* DBGDSCRint */
789 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
790 	DBG_BCR_BVR_WCR_WVR(1),
791 	/* DBGDCCINT */
792 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
793 	/* DBGDSCRext */
794 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
795 	DBG_BCR_BVR_WCR_WVR(2),
796 	/* DBGDTR[RT]Xint */
797 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
798 	/* DBGDTR[RT]Xext */
799 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
800 	DBG_BCR_BVR_WCR_WVR(3),
801 	DBG_BCR_BVR_WCR_WVR(4),
802 	DBG_BCR_BVR_WCR_WVR(5),
803 	/* DBGWFAR */
804 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
805 	/* DBGOSECCR */
806 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
807 	DBG_BCR_BVR_WCR_WVR(6),
808 	/* DBGVCR */
809 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
810 	DBG_BCR_BVR_WCR_WVR(7),
811 	DBG_BCR_BVR_WCR_WVR(8),
812 	DBG_BCR_BVR_WCR_WVR(9),
813 	DBG_BCR_BVR_WCR_WVR(10),
814 	DBG_BCR_BVR_WCR_WVR(11),
815 	DBG_BCR_BVR_WCR_WVR(12),
816 	DBG_BCR_BVR_WCR_WVR(13),
817 	DBG_BCR_BVR_WCR_WVR(14),
818 	DBG_BCR_BVR_WCR_WVR(15),
819 
820 	/* DBGDRAR (32bit) */
821 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
822 
823 	DBGBXVR(0),
824 	/* DBGOSLAR */
825 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
826 	DBGBXVR(1),
827 	/* DBGOSLSR */
828 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
829 	DBGBXVR(2),
830 	DBGBXVR(3),
831 	/* DBGOSDLR */
832 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
833 	DBGBXVR(4),
834 	/* DBGPRCR */
835 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
836 	DBGBXVR(5),
837 	DBGBXVR(6),
838 	DBGBXVR(7),
839 	DBGBXVR(8),
840 	DBGBXVR(9),
841 	DBGBXVR(10),
842 	DBGBXVR(11),
843 	DBGBXVR(12),
844 	DBGBXVR(13),
845 	DBGBXVR(14),
846 	DBGBXVR(15),
847 
848 	/* DBGDSAR (32bit) */
849 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
850 
851 	/* DBGDEVID2 */
852 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
853 	/* DBGDEVID1 */
854 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
855 	/* DBGDEVID */
856 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
857 	/* DBGCLAIMSET */
858 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
859 	/* DBGCLAIMCLR */
860 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
861 	/* DBGAUTHSTATUS */
862 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
863 };
864 
865 /* Trapped cp14 64bit registers */
866 static const struct sys_reg_desc cp14_64_regs[] = {
867 	/* DBGDRAR (64bit) */
868 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
869 
870 	/* DBGDSAR (64bit) */
871 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
872 };
873 
874 /*
875  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
876  * depending on the way they are accessed (as a 32bit or a 64bit
877  * register).
878  */
879 static const struct sys_reg_desc cp15_regs[] = {
880 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
881 
882 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, c1_SCTLR },
883 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
884 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
885 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
886 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
887 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
888 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
889 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
890 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
891 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
892 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
893 
894 	/*
895 	 * DC{C,I,CI}SW operations:
896 	 */
897 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
898 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
899 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
900 
901 	/* PMU */
902 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), trap_raz_wi },
903 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), trap_raz_wi },
904 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), trap_raz_wi },
905 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), trap_raz_wi },
906 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), trap_raz_wi },
907 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), trap_raz_wi },
908 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), trap_raz_wi },
909 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), trap_raz_wi },
910 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), trap_raz_wi },
911 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), trap_raz_wi },
912 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), trap_raz_wi },
913 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), trap_raz_wi },
914 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), trap_raz_wi },
915 
916 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
917 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
918 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
919 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
920 
921 	/* ICC_SRE */
922 	{ Op1( 0), CRn(12), CRm(12), Op2( 5), trap_raz_wi },
923 
924 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
925 };
926 
927 static const struct sys_reg_desc cp15_64_regs[] = {
928 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
929 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi },
930 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
931 };
932 
933 /* Target specific emulation tables */
934 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
935 
936 void kvm_register_target_sys_reg_table(unsigned int target,
937 				       struct kvm_sys_reg_target_table *table)
938 {
939 	target_tables[target] = table;
940 }
941 
942 /* Get specific register table for this target. */
943 static const struct sys_reg_desc *get_target_table(unsigned target,
944 						   bool mode_is_64,
945 						   size_t *num)
946 {
947 	struct kvm_sys_reg_target_table *table;
948 
949 	table = target_tables[target];
950 	if (mode_is_64) {
951 		*num = table->table64.num;
952 		return table->table64.table;
953 	} else {
954 		*num = table->table32.num;
955 		return table->table32.table;
956 	}
957 }
958 
959 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
960 					 const struct sys_reg_desc table[],
961 					 unsigned int num)
962 {
963 	unsigned int i;
964 
965 	for (i = 0; i < num; i++) {
966 		const struct sys_reg_desc *r = &table[i];
967 
968 		if (params->Op0 != r->Op0)
969 			continue;
970 		if (params->Op1 != r->Op1)
971 			continue;
972 		if (params->CRn != r->CRn)
973 			continue;
974 		if (params->CRm != r->CRm)
975 			continue;
976 		if (params->Op2 != r->Op2)
977 			continue;
978 
979 		return r;
980 	}
981 	return NULL;
982 }
983 
984 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
985 {
986 	kvm_inject_undefined(vcpu);
987 	return 1;
988 }
989 
990 /*
991  * emulate_cp --  tries to match a sys_reg access in a handling table, and
992  *                call the corresponding trap handler.
993  *
994  * @params: pointer to the descriptor of the access
995  * @table: array of trap descriptors
996  * @num: size of the trap descriptor array
997  *
998  * Return 0 if the access has been handled, and -1 if not.
999  */
1000 static int emulate_cp(struct kvm_vcpu *vcpu,
1001 		      const struct sys_reg_params *params,
1002 		      const struct sys_reg_desc *table,
1003 		      size_t num)
1004 {
1005 	const struct sys_reg_desc *r;
1006 
1007 	if (!table)
1008 		return -1;	/* Not handled */
1009 
1010 	r = find_reg(params, table, num);
1011 
1012 	if (r) {
1013 		/*
1014 		 * Not having an accessor means that we have
1015 		 * configured a trap that we don't know how to
1016 		 * handle. This certainly qualifies as a gross bug
1017 		 * that should be fixed right away.
1018 		 */
1019 		BUG_ON(!r->access);
1020 
1021 		if (likely(r->access(vcpu, params, r))) {
1022 			/* Skip instruction, since it was emulated */
1023 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1024 		}
1025 
1026 		/* Handled */
1027 		return 0;
1028 	}
1029 
1030 	/* Not handled */
1031 	return -1;
1032 }
1033 
1034 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
1035 				struct sys_reg_params *params)
1036 {
1037 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
1038 	int cp;
1039 
1040 	switch(hsr_ec) {
1041 	case ESR_ELx_EC_CP15_32:
1042 	case ESR_ELx_EC_CP15_64:
1043 		cp = 15;
1044 		break;
1045 	case ESR_ELx_EC_CP14_MR:
1046 	case ESR_ELx_EC_CP14_64:
1047 		cp = 14;
1048 		break;
1049 	default:
1050 		WARN_ON((cp = -1));
1051 	}
1052 
1053 	kvm_err("Unsupported guest CP%d access at: %08lx\n",
1054 		cp, *vcpu_pc(vcpu));
1055 	print_sys_reg_instr(params);
1056 	kvm_inject_undefined(vcpu);
1057 }
1058 
1059 /**
1060  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP15 access
1061  * @vcpu: The VCPU pointer
1062  * @run:  The kvm_run struct
1063  */
1064 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
1065 			    const struct sys_reg_desc *global,
1066 			    size_t nr_global,
1067 			    const struct sys_reg_desc *target_specific,
1068 			    size_t nr_specific)
1069 {
1070 	struct sys_reg_params params;
1071 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1072 	int Rt2 = (hsr >> 10) & 0xf;
1073 
1074 	params.is_aarch32 = true;
1075 	params.is_32bit = false;
1076 	params.CRm = (hsr >> 1) & 0xf;
1077 	params.Rt = (hsr >> 5) & 0xf;
1078 	params.is_write = ((hsr & 1) == 0);
1079 
1080 	params.Op0 = 0;
1081 	params.Op1 = (hsr >> 16) & 0xf;
1082 	params.Op2 = 0;
1083 	params.CRn = 0;
1084 
1085 	/*
1086 	 * Massive hack here. Store Rt2 in the top 32bits so we only
1087 	 * have one register to deal with. As we use the same trap
1088 	 * backends between AArch32 and AArch64, we get away with it.
1089 	 */
1090 	if (params.is_write) {
1091 		u64 val = *vcpu_reg(vcpu, params.Rt);
1092 		val &= 0xffffffff;
1093 		val |= *vcpu_reg(vcpu, Rt2) << 32;
1094 		*vcpu_reg(vcpu, params.Rt) = val;
1095 	}
1096 
1097 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
1098 		goto out;
1099 	if (!emulate_cp(vcpu, &params, global, nr_global))
1100 		goto out;
1101 
1102 	unhandled_cp_access(vcpu, &params);
1103 
1104 out:
1105 	/* Do the opposite hack for the read side */
1106 	if (!params.is_write) {
1107 		u64 val = *vcpu_reg(vcpu, params.Rt);
1108 		val >>= 32;
1109 		*vcpu_reg(vcpu, Rt2) = val;
1110 	}
1111 
1112 	return 1;
1113 }
1114 
1115 /**
1116  * kvm_handle_cp15_32 -- handles a mrc/mcr trap on a guest CP15 access
1117  * @vcpu: The VCPU pointer
1118  * @run:  The kvm_run struct
1119  */
1120 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
1121 			    const struct sys_reg_desc *global,
1122 			    size_t nr_global,
1123 			    const struct sys_reg_desc *target_specific,
1124 			    size_t nr_specific)
1125 {
1126 	struct sys_reg_params params;
1127 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
1128 
1129 	params.is_aarch32 = true;
1130 	params.is_32bit = true;
1131 	params.CRm = (hsr >> 1) & 0xf;
1132 	params.Rt  = (hsr >> 5) & 0xf;
1133 	params.is_write = ((hsr & 1) == 0);
1134 	params.CRn = (hsr >> 10) & 0xf;
1135 	params.Op0 = 0;
1136 	params.Op1 = (hsr >> 14) & 0x7;
1137 	params.Op2 = (hsr >> 17) & 0x7;
1138 
1139 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
1140 		return 1;
1141 	if (!emulate_cp(vcpu, &params, global, nr_global))
1142 		return 1;
1143 
1144 	unhandled_cp_access(vcpu, &params);
1145 	return 1;
1146 }
1147 
1148 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1149 {
1150 	const struct sys_reg_desc *target_specific;
1151 	size_t num;
1152 
1153 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1154 	return kvm_handle_cp_64(vcpu,
1155 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
1156 				target_specific, num);
1157 }
1158 
1159 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1160 {
1161 	const struct sys_reg_desc *target_specific;
1162 	size_t num;
1163 
1164 	target_specific = get_target_table(vcpu->arch.target, false, &num);
1165 	return kvm_handle_cp_32(vcpu,
1166 				cp15_regs, ARRAY_SIZE(cp15_regs),
1167 				target_specific, num);
1168 }
1169 
1170 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
1171 {
1172 	return kvm_handle_cp_64(vcpu,
1173 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
1174 				NULL, 0);
1175 }
1176 
1177 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
1178 {
1179 	return kvm_handle_cp_32(vcpu,
1180 				cp14_regs, ARRAY_SIZE(cp14_regs),
1181 				NULL, 0);
1182 }
1183 
1184 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
1185 			   const struct sys_reg_params *params)
1186 {
1187 	size_t num;
1188 	const struct sys_reg_desc *table, *r;
1189 
1190 	table = get_target_table(vcpu->arch.target, true, &num);
1191 
1192 	/* Search target-specific then generic table. */
1193 	r = find_reg(params, table, num);
1194 	if (!r)
1195 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1196 
1197 	if (likely(r)) {
1198 		/*
1199 		 * Not having an accessor means that we have
1200 		 * configured a trap that we don't know how to
1201 		 * handle. This certainly qualifies as a gross bug
1202 		 * that should be fixed right away.
1203 		 */
1204 		BUG_ON(!r->access);
1205 
1206 		if (likely(r->access(vcpu, params, r))) {
1207 			/* Skip instruction, since it was emulated */
1208 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
1209 			return 1;
1210 		}
1211 		/* If access function fails, it should complain. */
1212 	} else {
1213 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
1214 			*vcpu_pc(vcpu));
1215 		print_sys_reg_instr(params);
1216 	}
1217 	kvm_inject_undefined(vcpu);
1218 	return 1;
1219 }
1220 
1221 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
1222 			      const struct sys_reg_desc *table, size_t num)
1223 {
1224 	unsigned long i;
1225 
1226 	for (i = 0; i < num; i++)
1227 		if (table[i].reset)
1228 			table[i].reset(vcpu, &table[i]);
1229 }
1230 
1231 /**
1232  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
1233  * @vcpu: The VCPU pointer
1234  * @run:  The kvm_run struct
1235  */
1236 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1237 {
1238 	struct sys_reg_params params;
1239 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1240 
1241 	trace_kvm_handle_sys_reg(esr);
1242 
1243 	params.is_aarch32 = false;
1244 	params.is_32bit = false;
1245 	params.Op0 = (esr >> 20) & 3;
1246 	params.Op1 = (esr >> 14) & 0x7;
1247 	params.CRn = (esr >> 10) & 0xf;
1248 	params.CRm = (esr >> 1) & 0xf;
1249 	params.Op2 = (esr >> 17) & 0x7;
1250 	params.Rt = (esr >> 5) & 0x1f;
1251 	params.is_write = !(esr & 1);
1252 
1253 	return emulate_sys_reg(vcpu, &params);
1254 }
1255 
1256 /******************************************************************************
1257  * Userspace API
1258  *****************************************************************************/
1259 
1260 static bool index_to_params(u64 id, struct sys_reg_params *params)
1261 {
1262 	switch (id & KVM_REG_SIZE_MASK) {
1263 	case KVM_REG_SIZE_U64:
1264 		/* Any unused index bits means it's not valid. */
1265 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1266 			      | KVM_REG_ARM_COPROC_MASK
1267 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
1268 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
1269 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
1270 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
1271 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
1272 			return false;
1273 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1274 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1275 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1276 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1277 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1278 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1279 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1280 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1281 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1282 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1283 		return true;
1284 	default:
1285 		return false;
1286 	}
1287 }
1288 
1289 /* Decode an index value, and find the sys_reg_desc entry. */
1290 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1291 						    u64 id)
1292 {
1293 	size_t num;
1294 	const struct sys_reg_desc *table, *r;
1295 	struct sys_reg_params params;
1296 
1297 	/* We only do sys_reg for now. */
1298 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1299 		return NULL;
1300 
1301 	if (!index_to_params(id, &params))
1302 		return NULL;
1303 
1304 	table = get_target_table(vcpu->arch.target, true, &num);
1305 	r = find_reg(&params, table, num);
1306 	if (!r)
1307 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1308 
1309 	/* Not saved in the sys_reg array? */
1310 	if (r && !r->reg)
1311 		r = NULL;
1312 
1313 	return r;
1314 }
1315 
1316 /*
1317  * These are the invariant sys_reg registers: we let the guest see the
1318  * host versions of these, so they're part of the guest state.
1319  *
1320  * A future CPU may provide a mechanism to present different values to
1321  * the guest, or a future kvm may trap them.
1322  */
1323 
1324 #define FUNCTION_INVARIANT(reg)						\
1325 	static void get_##reg(struct kvm_vcpu *v,			\
1326 			      const struct sys_reg_desc *r)		\
1327 	{								\
1328 		u64 val;						\
1329 									\
1330 		asm volatile("mrs %0, " __stringify(reg) "\n"		\
1331 			     : "=r" (val));				\
1332 		((struct sys_reg_desc *)r)->val = val;			\
1333 	}
1334 
1335 FUNCTION_INVARIANT(midr_el1)
1336 FUNCTION_INVARIANT(ctr_el0)
1337 FUNCTION_INVARIANT(revidr_el1)
1338 FUNCTION_INVARIANT(id_pfr0_el1)
1339 FUNCTION_INVARIANT(id_pfr1_el1)
1340 FUNCTION_INVARIANT(id_dfr0_el1)
1341 FUNCTION_INVARIANT(id_afr0_el1)
1342 FUNCTION_INVARIANT(id_mmfr0_el1)
1343 FUNCTION_INVARIANT(id_mmfr1_el1)
1344 FUNCTION_INVARIANT(id_mmfr2_el1)
1345 FUNCTION_INVARIANT(id_mmfr3_el1)
1346 FUNCTION_INVARIANT(id_isar0_el1)
1347 FUNCTION_INVARIANT(id_isar1_el1)
1348 FUNCTION_INVARIANT(id_isar2_el1)
1349 FUNCTION_INVARIANT(id_isar3_el1)
1350 FUNCTION_INVARIANT(id_isar4_el1)
1351 FUNCTION_INVARIANT(id_isar5_el1)
1352 FUNCTION_INVARIANT(clidr_el1)
1353 FUNCTION_INVARIANT(aidr_el1)
1354 
1355 /* ->val is filled in by kvm_sys_reg_table_init() */
1356 static struct sys_reg_desc invariant_sys_regs[] = {
1357 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1358 	  NULL, get_midr_el1 },
1359 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1360 	  NULL, get_revidr_el1 },
1361 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1362 	  NULL, get_id_pfr0_el1 },
1363 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1364 	  NULL, get_id_pfr1_el1 },
1365 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1366 	  NULL, get_id_dfr0_el1 },
1367 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1368 	  NULL, get_id_afr0_el1 },
1369 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1370 	  NULL, get_id_mmfr0_el1 },
1371 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1372 	  NULL, get_id_mmfr1_el1 },
1373 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1374 	  NULL, get_id_mmfr2_el1 },
1375 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1376 	  NULL, get_id_mmfr3_el1 },
1377 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1378 	  NULL, get_id_isar0_el1 },
1379 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1380 	  NULL, get_id_isar1_el1 },
1381 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1382 	  NULL, get_id_isar2_el1 },
1383 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1384 	  NULL, get_id_isar3_el1 },
1385 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1386 	  NULL, get_id_isar4_el1 },
1387 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1388 	  NULL, get_id_isar5_el1 },
1389 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1390 	  NULL, get_clidr_el1 },
1391 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1392 	  NULL, get_aidr_el1 },
1393 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1394 	  NULL, get_ctr_el0 },
1395 };
1396 
1397 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1398 {
1399 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1400 		return -EFAULT;
1401 	return 0;
1402 }
1403 
1404 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1405 {
1406 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1407 		return -EFAULT;
1408 	return 0;
1409 }
1410 
1411 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1412 {
1413 	struct sys_reg_params params;
1414 	const struct sys_reg_desc *r;
1415 
1416 	if (!index_to_params(id, &params))
1417 		return -ENOENT;
1418 
1419 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1420 	if (!r)
1421 		return -ENOENT;
1422 
1423 	return reg_to_user(uaddr, &r->val, id);
1424 }
1425 
1426 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1427 {
1428 	struct sys_reg_params params;
1429 	const struct sys_reg_desc *r;
1430 	int err;
1431 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1432 
1433 	if (!index_to_params(id, &params))
1434 		return -ENOENT;
1435 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1436 	if (!r)
1437 		return -ENOENT;
1438 
1439 	err = reg_from_user(&val, uaddr, id);
1440 	if (err)
1441 		return err;
1442 
1443 	/* This is what we mean by invariant: you can't change it. */
1444 	if (r->val != val)
1445 		return -EINVAL;
1446 
1447 	return 0;
1448 }
1449 
1450 static bool is_valid_cache(u32 val)
1451 {
1452 	u32 level, ctype;
1453 
1454 	if (val >= CSSELR_MAX)
1455 		return false;
1456 
1457 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1458 	level = (val >> 1);
1459 	ctype = (cache_levels >> (level * 3)) & 7;
1460 
1461 	switch (ctype) {
1462 	case 0: /* No cache */
1463 		return false;
1464 	case 1: /* Instruction cache only */
1465 		return (val & 1);
1466 	case 2: /* Data cache only */
1467 	case 4: /* Unified cache */
1468 		return !(val & 1);
1469 	case 3: /* Separate instruction and data caches */
1470 		return true;
1471 	default: /* Reserved: we can't know instruction or data. */
1472 		return false;
1473 	}
1474 }
1475 
1476 static int demux_c15_get(u64 id, void __user *uaddr)
1477 {
1478 	u32 val;
1479 	u32 __user *uval = uaddr;
1480 
1481 	/* Fail if we have unknown bits set. */
1482 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1483 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1484 		return -ENOENT;
1485 
1486 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1487 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1488 		if (KVM_REG_SIZE(id) != 4)
1489 			return -ENOENT;
1490 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1491 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1492 		if (!is_valid_cache(val))
1493 			return -ENOENT;
1494 
1495 		return put_user(get_ccsidr(val), uval);
1496 	default:
1497 		return -ENOENT;
1498 	}
1499 }
1500 
1501 static int demux_c15_set(u64 id, void __user *uaddr)
1502 {
1503 	u32 val, newval;
1504 	u32 __user *uval = uaddr;
1505 
1506 	/* Fail if we have unknown bits set. */
1507 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1508 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1509 		return -ENOENT;
1510 
1511 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1512 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1513 		if (KVM_REG_SIZE(id) != 4)
1514 			return -ENOENT;
1515 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1516 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1517 		if (!is_valid_cache(val))
1518 			return -ENOENT;
1519 
1520 		if (get_user(newval, uval))
1521 			return -EFAULT;
1522 
1523 		/* This is also invariant: you can't change it. */
1524 		if (newval != get_ccsidr(val))
1525 			return -EINVAL;
1526 		return 0;
1527 	default:
1528 		return -ENOENT;
1529 	}
1530 }
1531 
1532 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1533 {
1534 	const struct sys_reg_desc *r;
1535 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1536 
1537 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1538 		return demux_c15_get(reg->id, uaddr);
1539 
1540 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1541 		return -ENOENT;
1542 
1543 	r = index_to_sys_reg_desc(vcpu, reg->id);
1544 	if (!r)
1545 		return get_invariant_sys_reg(reg->id, uaddr);
1546 
1547 	if (r->get_user)
1548 		return (r->get_user)(vcpu, r, reg, uaddr);
1549 
1550 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
1551 }
1552 
1553 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1554 {
1555 	const struct sys_reg_desc *r;
1556 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1557 
1558 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1559 		return demux_c15_set(reg->id, uaddr);
1560 
1561 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1562 		return -ENOENT;
1563 
1564 	r = index_to_sys_reg_desc(vcpu, reg->id);
1565 	if (!r)
1566 		return set_invariant_sys_reg(reg->id, uaddr);
1567 
1568 	if (r->set_user)
1569 		return (r->set_user)(vcpu, r, reg, uaddr);
1570 
1571 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
1572 }
1573 
1574 static unsigned int num_demux_regs(void)
1575 {
1576 	unsigned int i, count = 0;
1577 
1578 	for (i = 0; i < CSSELR_MAX; i++)
1579 		if (is_valid_cache(i))
1580 			count++;
1581 
1582 	return count;
1583 }
1584 
1585 static int write_demux_regids(u64 __user *uindices)
1586 {
1587 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
1588 	unsigned int i;
1589 
1590 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
1591 	for (i = 0; i < CSSELR_MAX; i++) {
1592 		if (!is_valid_cache(i))
1593 			continue;
1594 		if (put_user(val | i, uindices))
1595 			return -EFAULT;
1596 		uindices++;
1597 	}
1598 	return 0;
1599 }
1600 
1601 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
1602 {
1603 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
1604 		KVM_REG_ARM64_SYSREG |
1605 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
1606 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
1607 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
1608 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
1609 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
1610 }
1611 
1612 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
1613 {
1614 	if (!*uind)
1615 		return true;
1616 
1617 	if (put_user(sys_reg_to_index(reg), *uind))
1618 		return false;
1619 
1620 	(*uind)++;
1621 	return true;
1622 }
1623 
1624 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
1625 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
1626 {
1627 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
1628 	unsigned int total = 0;
1629 	size_t num;
1630 
1631 	/* We check for duplicates here, to allow arch-specific overrides. */
1632 	i1 = get_target_table(vcpu->arch.target, true, &num);
1633 	end1 = i1 + num;
1634 	i2 = sys_reg_descs;
1635 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
1636 
1637 	BUG_ON(i1 == end1 || i2 == end2);
1638 
1639 	/* Walk carefully, as both tables may refer to the same register. */
1640 	while (i1 || i2) {
1641 		int cmp = cmp_sys_reg(i1, i2);
1642 		/* target-specific overrides generic entry. */
1643 		if (cmp <= 0) {
1644 			/* Ignore registers we trap but don't save. */
1645 			if (i1->reg) {
1646 				if (!copy_reg_to_user(i1, &uind))
1647 					return -EFAULT;
1648 				total++;
1649 			}
1650 		} else {
1651 			/* Ignore registers we trap but don't save. */
1652 			if (i2->reg) {
1653 				if (!copy_reg_to_user(i2, &uind))
1654 					return -EFAULT;
1655 				total++;
1656 			}
1657 		}
1658 
1659 		if (cmp <= 0 && ++i1 == end1)
1660 			i1 = NULL;
1661 		if (cmp >= 0 && ++i2 == end2)
1662 			i2 = NULL;
1663 	}
1664 	return total;
1665 }
1666 
1667 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
1668 {
1669 	return ARRAY_SIZE(invariant_sys_regs)
1670 		+ num_demux_regs()
1671 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
1672 }
1673 
1674 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
1675 {
1676 	unsigned int i;
1677 	int err;
1678 
1679 	/* Then give them all the invariant registers' indices. */
1680 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
1681 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
1682 			return -EFAULT;
1683 		uindices++;
1684 	}
1685 
1686 	err = walk_sys_regs(vcpu, uindices);
1687 	if (err < 0)
1688 		return err;
1689 	uindices += err;
1690 
1691 	return write_demux_regids(uindices);
1692 }
1693 
1694 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
1695 {
1696 	unsigned int i;
1697 
1698 	for (i = 1; i < n; i++) {
1699 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
1700 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
1701 			return 1;
1702 		}
1703 	}
1704 
1705 	return 0;
1706 }
1707 
1708 void kvm_sys_reg_table_init(void)
1709 {
1710 	unsigned int i;
1711 	struct sys_reg_desc clidr;
1712 
1713 	/* Make sure tables are unique and in order. */
1714 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
1715 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
1716 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
1717 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
1718 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
1719 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
1720 
1721 	/* We abuse the reset function to overwrite the table itself. */
1722 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
1723 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
1724 
1725 	/*
1726 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
1727 	 *
1728 	 *   If software reads the Cache Type fields from Ctype1
1729 	 *   upwards, once it has seen a value of 0b000, no caches
1730 	 *   exist at further-out levels of the hierarchy. So, for
1731 	 *   example, if Ctype3 is the first Cache Type field with a
1732 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
1733 	 *   ignored.
1734 	 */
1735 	get_clidr_el1(NULL, &clidr); /* Ugly... */
1736 	cache_levels = clidr.val;
1737 	for (i = 0; i < 7; i++)
1738 		if (((cache_levels >> (i*3)) & 7) == 0)
1739 			break;
1740 	/* Clear all higher bits. */
1741 	cache_levels &= (1 << (i*3))-1;
1742 }
1743 
1744 /**
1745  * kvm_reset_sys_regs - sets system registers to reset value
1746  * @vcpu: The VCPU pointer
1747  *
1748  * This function finds the right table above and sets the registers on the
1749  * virtual CPU struct to their architecturally defined reset values.
1750  */
1751 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
1752 {
1753 	size_t num;
1754 	const struct sys_reg_desc *table;
1755 
1756 	/* Catch someone adding a register without putting in reset entry. */
1757 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
1758 
1759 	/* Generic chip reset first (so target could override). */
1760 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1761 
1762 	table = get_target_table(vcpu->arch.target, true, &num);
1763 	reset_sys_reg_descs(vcpu, table, num);
1764 
1765 	for (num = 1; num < NR_SYS_REGS; num++)
1766 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
1767 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
1768 }
1769