xref: /linux/arch/arm64/kvm/sys_regs.c (revision 751d041a13bdc9d72bf7efdc86224da1174ff31d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012,2013 - ARM Ltd
4  * Author: Marc Zyngier <marc.zyngier@arm.com>
5  *
6  * Derived from arch/arm/kvm/coproc.c:
7  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
8  * Authors: Rusty Russell <rusty@rustcorp.com.au>
9  *          Christoffer Dall <c.dall@virtualopensystems.com>
10  */
11 
12 #include <linux/bitfield.h>
13 #include <linux/bsearch.h>
14 #include <linux/cacheinfo.h>
15 #include <linux/debugfs.h>
16 #include <linux/kvm_host.h>
17 #include <linux/mm.h>
18 #include <linux/printk.h>
19 #include <linux/uaccess.h>
20 #include <linux/irqchip/arm-gic-v3.h>
21 
22 #include <asm/arm_pmuv3.h>
23 #include <asm/cacheflush.h>
24 #include <asm/cputype.h>
25 #include <asm/debug-monitors.h>
26 #include <asm/esr.h>
27 #include <asm/kvm_arm.h>
28 #include <asm/kvm_emulate.h>
29 #include <asm/kvm_hyp.h>
30 #include <asm/kvm_mmu.h>
31 #include <asm/kvm_nested.h>
32 #include <asm/perf_event.h>
33 #include <asm/sysreg.h>
34 
35 #include <trace/events/kvm.h>
36 
37 #include "sys_regs.h"
38 #include "vgic/vgic.h"
39 
40 #include "trace.h"
41 
42 /*
43  * For AArch32, we only take care of what is being trapped. Anything
44  * that has to do with init and userspace access has to go via the
45  * 64bit interface.
46  */
47 
48 static u64 sys_reg_to_index(const struct sys_reg_desc *reg);
49 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
50 		      u64 val);
51 
52 static bool undef_access(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
53 			 const struct sys_reg_desc *r)
54 {
55 	kvm_inject_undefined(vcpu);
56 	return false;
57 }
58 
59 static bool bad_trap(struct kvm_vcpu *vcpu,
60 		     struct sys_reg_params *params,
61 		     const struct sys_reg_desc *r,
62 		     const char *msg)
63 {
64 	WARN_ONCE(1, "Unexpected %s\n", msg);
65 	print_sys_reg_instr(params);
66 	return undef_access(vcpu, params, r);
67 }
68 
69 static bool read_from_write_only(struct kvm_vcpu *vcpu,
70 				 struct sys_reg_params *params,
71 				 const struct sys_reg_desc *r)
72 {
73 	return bad_trap(vcpu, params, r,
74 			"sys_reg read to write-only register");
75 }
76 
77 static bool write_to_read_only(struct kvm_vcpu *vcpu,
78 			       struct sys_reg_params *params,
79 			       const struct sys_reg_desc *r)
80 {
81 	return bad_trap(vcpu, params, r,
82 			"sys_reg write to read-only register");
83 }
84 
85 enum sr_loc_attr {
86 	SR_LOC_MEMORY	= 0,	  /* Register definitely in memory */
87 	SR_LOC_LOADED	= BIT(0), /* Register on CPU, unless it cannot */
88 	SR_LOC_MAPPED	= BIT(1), /* Register in a different CPU register */
89 	SR_LOC_XLATED	= BIT(2), /* Register translated to fit another reg */
90 	SR_LOC_SPECIAL	= BIT(3), /* Demanding register, implies loaded */
91 };
92 
93 struct sr_loc {
94 	enum sr_loc_attr loc;
95 	enum vcpu_sysreg map_reg;
96 	u64		 (*xlate)(u64);
97 };
98 
99 static enum sr_loc_attr locate_direct_register(const struct kvm_vcpu *vcpu,
100 					       enum vcpu_sysreg reg)
101 {
102 	switch (reg) {
103 	case SCTLR_EL1:
104 	case CPACR_EL1:
105 	case TTBR0_EL1:
106 	case TTBR1_EL1:
107 	case TCR_EL1:
108 	case TCR2_EL1:
109 	case PIR_EL1:
110 	case PIRE0_EL1:
111 	case POR_EL1:
112 	case ESR_EL1:
113 	case AFSR0_EL1:
114 	case AFSR1_EL1:
115 	case FAR_EL1:
116 	case MAIR_EL1:
117 	case VBAR_EL1:
118 	case CONTEXTIDR_EL1:
119 	case AMAIR_EL1:
120 	case CNTKCTL_EL1:
121 	case ELR_EL1:
122 	case SPSR_EL1:
123 	case ZCR_EL1:
124 	case SCTLR2_EL1:
125 		/*
126 		 * EL1 registers which have an ELx2 mapping are loaded if
127 		 * we're not in hypervisor context.
128 		 */
129 		return is_hyp_ctxt(vcpu) ? SR_LOC_MEMORY : SR_LOC_LOADED;
130 
131 	case TPIDR_EL0:
132 	case TPIDRRO_EL0:
133 	case TPIDR_EL1:
134 	case PAR_EL1:
135 	case DACR32_EL2:
136 	case IFSR32_EL2:
137 	case DBGVCR32_EL2:
138 		/* These registers are always loaded, no matter what */
139 		return SR_LOC_LOADED;
140 
141 	default:
142 		/* Non-mapped EL2 registers are by definition in memory. */
143 		return SR_LOC_MEMORY;
144 	}
145 }
146 
147 static void locate_mapped_el2_register(const struct kvm_vcpu *vcpu,
148 				       enum vcpu_sysreg reg,
149 				       enum vcpu_sysreg map_reg,
150 				       u64 (*xlate)(u64),
151 				       struct sr_loc *loc)
152 {
153 	if (!is_hyp_ctxt(vcpu)) {
154 		loc->loc = SR_LOC_MEMORY;
155 		return;
156 	}
157 
158 	loc->loc = SR_LOC_LOADED | SR_LOC_MAPPED;
159 	loc->map_reg = map_reg;
160 
161 	WARN_ON(locate_direct_register(vcpu, map_reg) != SR_LOC_MEMORY);
162 
163 	if (xlate != NULL && !vcpu_el2_e2h_is_set(vcpu)) {
164 		loc->loc |= SR_LOC_XLATED;
165 		loc->xlate = xlate;
166 	}
167 }
168 
169 #define MAPPED_EL2_SYSREG(r, m, t)					\
170 	case r:	{							\
171 		locate_mapped_el2_register(vcpu, r, m, t, loc);		\
172 		break;							\
173 	}
174 
175 static void locate_register(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg,
176 			    struct sr_loc *loc)
177 {
178 	if (!vcpu_get_flag(vcpu, SYSREGS_ON_CPU)) {
179 		loc->loc = SR_LOC_MEMORY;
180 		return;
181 	}
182 
183 	switch (reg) {
184 		MAPPED_EL2_SYSREG(SCTLR_EL2,   SCTLR_EL1,
185 				  translate_sctlr_el2_to_sctlr_el1	     );
186 		MAPPED_EL2_SYSREG(CPTR_EL2,    CPACR_EL1,
187 				  translate_cptr_el2_to_cpacr_el1	     );
188 		MAPPED_EL2_SYSREG(TTBR0_EL2,   TTBR0_EL1,
189 				  translate_ttbr0_el2_to_ttbr0_el1	     );
190 		MAPPED_EL2_SYSREG(TTBR1_EL2,   TTBR1_EL1,   NULL	     );
191 		MAPPED_EL2_SYSREG(TCR_EL2,     TCR_EL1,
192 				  translate_tcr_el2_to_tcr_el1		     );
193 		MAPPED_EL2_SYSREG(VBAR_EL2,    VBAR_EL1,    NULL	     );
194 		MAPPED_EL2_SYSREG(AFSR0_EL2,   AFSR0_EL1,   NULL	     );
195 		MAPPED_EL2_SYSREG(AFSR1_EL2,   AFSR1_EL1,   NULL	     );
196 		MAPPED_EL2_SYSREG(ESR_EL2,     ESR_EL1,     NULL	     );
197 		MAPPED_EL2_SYSREG(FAR_EL2,     FAR_EL1,     NULL	     );
198 		MAPPED_EL2_SYSREG(MAIR_EL2,    MAIR_EL1,    NULL	     );
199 		MAPPED_EL2_SYSREG(TCR2_EL2,    TCR2_EL1,    NULL	     );
200 		MAPPED_EL2_SYSREG(PIR_EL2,     PIR_EL1,     NULL	     );
201 		MAPPED_EL2_SYSREG(PIRE0_EL2,   PIRE0_EL1,   NULL	     );
202 		MAPPED_EL2_SYSREG(POR_EL2,     POR_EL1,     NULL	     );
203 		MAPPED_EL2_SYSREG(AMAIR_EL2,   AMAIR_EL1,   NULL	     );
204 		MAPPED_EL2_SYSREG(ELR_EL2,     ELR_EL1,	    NULL	     );
205 		MAPPED_EL2_SYSREG(SPSR_EL2,    SPSR_EL1,    NULL	     );
206 		MAPPED_EL2_SYSREG(CONTEXTIDR_EL2, CONTEXTIDR_EL1, NULL	     );
207 		MAPPED_EL2_SYSREG(SCTLR2_EL2,  SCTLR2_EL1,  NULL	     );
208 	case CNTHCTL_EL2:
209 		/* CNTHCTL_EL2 is super special, until we support NV2.1 */
210 		loc->loc = ((is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu)) ?
211 			    SR_LOC_SPECIAL : SR_LOC_MEMORY);
212 		break;
213 	default:
214 		loc->loc = locate_direct_register(vcpu, reg);
215 	}
216 }
217 
218 static u64 read_sr_from_cpu(enum vcpu_sysreg reg)
219 {
220 	u64 val = 0x8badf00d8badf00d;
221 
222 	switch (reg) {
223 	case SCTLR_EL1:		val = read_sysreg_s(SYS_SCTLR_EL12);	break;
224 	case CPACR_EL1:		val = read_sysreg_s(SYS_CPACR_EL12);	break;
225 	case TTBR0_EL1:		val = read_sysreg_s(SYS_TTBR0_EL12);	break;
226 	case TTBR1_EL1:		val = read_sysreg_s(SYS_TTBR1_EL12);	break;
227 	case TCR_EL1:		val = read_sysreg_s(SYS_TCR_EL12);	break;
228 	case TCR2_EL1:		val = read_sysreg_s(SYS_TCR2_EL12);	break;
229 	case PIR_EL1:		val = read_sysreg_s(SYS_PIR_EL12);	break;
230 	case PIRE0_EL1:		val = read_sysreg_s(SYS_PIRE0_EL12);	break;
231 	case POR_EL1:		val = read_sysreg_s(SYS_POR_EL12);	break;
232 	case ESR_EL1:		val = read_sysreg_s(SYS_ESR_EL12);	break;
233 	case AFSR0_EL1:		val = read_sysreg_s(SYS_AFSR0_EL12);	break;
234 	case AFSR1_EL1:		val = read_sysreg_s(SYS_AFSR1_EL12);	break;
235 	case FAR_EL1:		val = read_sysreg_s(SYS_FAR_EL12);	break;
236 	case MAIR_EL1:		val = read_sysreg_s(SYS_MAIR_EL12);	break;
237 	case VBAR_EL1:		val = read_sysreg_s(SYS_VBAR_EL12);	break;
238 	case CONTEXTIDR_EL1:	val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
239 	case AMAIR_EL1:		val = read_sysreg_s(SYS_AMAIR_EL12);	break;
240 	case CNTKCTL_EL1:	val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
241 	case ELR_EL1:		val = read_sysreg_s(SYS_ELR_EL12);	break;
242 	case SPSR_EL1:		val = read_sysreg_s(SYS_SPSR_EL12);	break;
243 	case ZCR_EL1:		val = read_sysreg_s(SYS_ZCR_EL12);	break;
244 	case SCTLR2_EL1:	val = read_sysreg_s(SYS_SCTLR2_EL12);	break;
245 	case TPIDR_EL0:		val = read_sysreg_s(SYS_TPIDR_EL0);	break;
246 	case TPIDRRO_EL0:	val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
247 	case TPIDR_EL1:		val = read_sysreg_s(SYS_TPIDR_EL1);	break;
248 	case PAR_EL1:		val = read_sysreg_par();		break;
249 	case DACR32_EL2:	val = read_sysreg_s(SYS_DACR32_EL2);	break;
250 	case IFSR32_EL2:	val = read_sysreg_s(SYS_IFSR32_EL2);	break;
251 	case DBGVCR32_EL2:	val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
252 	default:		WARN_ON_ONCE(1);
253 	}
254 
255 	return val;
256 }
257 
258 static void write_sr_to_cpu(enum vcpu_sysreg reg, u64 val)
259 {
260 	switch (reg) {
261 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
262 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
263 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
264 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
265 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
266 	case TCR2_EL1:		write_sysreg_s(val, SYS_TCR2_EL12);	break;
267 	case PIR_EL1:		write_sysreg_s(val, SYS_PIR_EL12);	break;
268 	case PIRE0_EL1:		write_sysreg_s(val, SYS_PIRE0_EL12);	break;
269 	case POR_EL1:		write_sysreg_s(val, SYS_POR_EL12);	break;
270 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
271 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
272 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
273 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
274 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
275 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
276 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
277 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
278 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
279 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
280 	case SPSR_EL1:		write_sysreg_s(val, SYS_SPSR_EL12);	break;
281 	case ZCR_EL1:		write_sysreg_s(val, SYS_ZCR_EL12);	break;
282 	case SCTLR2_EL1:	write_sysreg_s(val, SYS_SCTLR2_EL12);	break;
283 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
284 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
285 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
286 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
287 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
288 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
289 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
290 	default:		WARN_ON_ONCE(1);
291 	}
292 }
293 
294 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg)
295 {
296 	struct sr_loc loc = {};
297 
298 	locate_register(vcpu, reg, &loc);
299 
300 	WARN_ON_ONCE(!has_vhe() && loc.loc != SR_LOC_MEMORY);
301 
302 	if (loc.loc & SR_LOC_SPECIAL) {
303 		u64 val;
304 
305 		WARN_ON_ONCE(loc.loc & ~SR_LOC_SPECIAL);
306 
307 		/*
308 		 * CNTHCTL_EL2 requires some special treatment to account
309 		 * for the bits that can be set via CNTKCTL_EL1 when E2H==1.
310 		 */
311 		switch (reg) {
312 		case CNTHCTL_EL2:
313 			val = read_sysreg_el1(SYS_CNTKCTL);
314 			val &= CNTKCTL_VALID_BITS;
315 			val |= __vcpu_sys_reg(vcpu, reg) & ~CNTKCTL_VALID_BITS;
316 			return val;
317 		default:
318 			WARN_ON_ONCE(1);
319 		}
320 	}
321 
322 	if (loc.loc & SR_LOC_LOADED) {
323 		enum vcpu_sysreg map_reg = reg;
324 
325 		if (loc.loc & SR_LOC_MAPPED)
326 			map_reg = loc.map_reg;
327 
328 		if (!(loc.loc & SR_LOC_XLATED)) {
329 			u64 val = read_sr_from_cpu(map_reg);
330 
331 			if (reg >= __SANITISED_REG_START__)
332 				val = kvm_vcpu_apply_reg_masks(vcpu, reg, val);
333 
334 			return val;
335 		}
336 	}
337 
338 	return __vcpu_sys_reg(vcpu, reg);
339 }
340 
341 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, enum vcpu_sysreg reg)
342 {
343 	struct sr_loc loc = {};
344 
345 	locate_register(vcpu, reg, &loc);
346 
347 	WARN_ON_ONCE(!has_vhe() && loc.loc != SR_LOC_MEMORY);
348 
349 	if (loc.loc & SR_LOC_SPECIAL) {
350 
351 		WARN_ON_ONCE(loc.loc & ~SR_LOC_SPECIAL);
352 
353 		switch (reg) {
354 		case CNTHCTL_EL2:
355 			/*
356 			 * If E2H=1, some of the bits are backed by
357 			 * CNTKCTL_EL1, while the rest is kept in memory.
358 			 * Yes, this is fun stuff.
359 			 */
360 			write_sysreg_el1(val, SYS_CNTKCTL);
361 			break;
362 		default:
363 			WARN_ON_ONCE(1);
364 		}
365 	}
366 
367 	if (loc.loc & SR_LOC_LOADED) {
368 		enum vcpu_sysreg map_reg = reg;
369 		u64 xlated_val;
370 
371 		if (reg >= __SANITISED_REG_START__)
372 			val = kvm_vcpu_apply_reg_masks(vcpu, reg, val);
373 
374 		if (loc.loc & SR_LOC_MAPPED)
375 			map_reg = loc.map_reg;
376 
377 		if (loc.loc & SR_LOC_XLATED)
378 			xlated_val = loc.xlate(val);
379 		else
380 			xlated_val = val;
381 
382 		write_sr_to_cpu(map_reg, xlated_val);
383 
384 		/*
385 		 * Fall through to write the backing store anyway, which
386 		 * allows translated registers to be directly read without a
387 		 * reverse translation.
388 		 */
389 	}
390 
391 	__vcpu_assign_sys_reg(vcpu, reg, val);
392 }
393 
394 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
395 #define CSSELR_MAX 14
396 
397 /*
398  * Returns the minimum line size for the selected cache, expressed as
399  * Log2(bytes).
400  */
401 static u8 get_min_cache_line_size(bool icache)
402 {
403 	u64 ctr = read_sanitised_ftr_reg(SYS_CTR_EL0);
404 	u8 field;
405 
406 	if (icache)
407 		field = SYS_FIELD_GET(CTR_EL0, IminLine, ctr);
408 	else
409 		field = SYS_FIELD_GET(CTR_EL0, DminLine, ctr);
410 
411 	/*
412 	 * Cache line size is represented as Log2(words) in CTR_EL0.
413 	 * Log2(bytes) can be derived with the following:
414 	 *
415 	 * Log2(words) + 2 = Log2(bytes / 4) + 2
416 	 * 		   = Log2(bytes) - 2 + 2
417 	 * 		   = Log2(bytes)
418 	 */
419 	return field + 2;
420 }
421 
422 /* Which cache CCSIDR represents depends on CSSELR value. */
423 static u32 get_ccsidr(struct kvm_vcpu *vcpu, u32 csselr)
424 {
425 	u8 line_size;
426 
427 	if (vcpu->arch.ccsidr)
428 		return vcpu->arch.ccsidr[csselr];
429 
430 	line_size = get_min_cache_line_size(csselr & CSSELR_EL1_InD);
431 
432 	/*
433 	 * Fabricate a CCSIDR value as the overriding value does not exist.
434 	 * The real CCSIDR value will not be used as it can vary by the
435 	 * physical CPU which the vcpu currently resides in.
436 	 *
437 	 * The line size is determined with get_min_cache_line_size(), which
438 	 * should be valid for all CPUs even if they have different cache
439 	 * configuration.
440 	 *
441 	 * The associativity bits are cleared, meaning the geometry of all data
442 	 * and unified caches (which are guaranteed to be PIPT and thus
443 	 * non-aliasing) are 1 set and 1 way.
444 	 * Guests should not be doing cache operations by set/way at all, and
445 	 * for this reason, we trap them and attempt to infer the intent, so
446 	 * that we can flush the entire guest's address space at the appropriate
447 	 * time. The exposed geometry minimizes the number of the traps.
448 	 * [If guests should attempt to infer aliasing properties from the
449 	 * geometry (which is not permitted by the architecture), they would
450 	 * only do so for virtually indexed caches.]
451 	 *
452 	 * We don't check if the cache level exists as it is allowed to return
453 	 * an UNKNOWN value if not.
454 	 */
455 	return SYS_FIELD_PREP(CCSIDR_EL1, LineSize, line_size - 4);
456 }
457 
458 static int set_ccsidr(struct kvm_vcpu *vcpu, u32 csselr, u32 val)
459 {
460 	u8 line_size = FIELD_GET(CCSIDR_EL1_LineSize, val) + 4;
461 	u32 *ccsidr = vcpu->arch.ccsidr;
462 	u32 i;
463 
464 	if ((val & CCSIDR_EL1_RES0) ||
465 	    line_size < get_min_cache_line_size(csselr & CSSELR_EL1_InD))
466 		return -EINVAL;
467 
468 	if (!ccsidr) {
469 		if (val == get_ccsidr(vcpu, csselr))
470 			return 0;
471 
472 		ccsidr = kmalloc_array(CSSELR_MAX, sizeof(u32), GFP_KERNEL_ACCOUNT);
473 		if (!ccsidr)
474 			return -ENOMEM;
475 
476 		for (i = 0; i < CSSELR_MAX; i++)
477 			ccsidr[i] = get_ccsidr(vcpu, i);
478 
479 		vcpu->arch.ccsidr = ccsidr;
480 	}
481 
482 	ccsidr[csselr] = val;
483 
484 	return 0;
485 }
486 
487 static bool access_rw(struct kvm_vcpu *vcpu,
488 		      struct sys_reg_params *p,
489 		      const struct sys_reg_desc *r)
490 {
491 	if (p->is_write)
492 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
493 	else
494 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
495 
496 	return true;
497 }
498 
499 /*
500  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
501  */
502 static bool access_dcsw(struct kvm_vcpu *vcpu,
503 			struct sys_reg_params *p,
504 			const struct sys_reg_desc *r)
505 {
506 	if (!p->is_write)
507 		return read_from_write_only(vcpu, p, r);
508 
509 	/*
510 	 * Only track S/W ops if we don't have FWB. It still indicates
511 	 * that the guest is a bit broken (S/W operations should only
512 	 * be done by firmware, knowing that there is only a single
513 	 * CPU left in the system, and certainly not from non-secure
514 	 * software).
515 	 */
516 	if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
517 		kvm_set_way_flush(vcpu);
518 
519 	return true;
520 }
521 
522 static bool access_dcgsw(struct kvm_vcpu *vcpu,
523 			 struct sys_reg_params *p,
524 			 const struct sys_reg_desc *r)
525 {
526 	if (!kvm_has_mte(vcpu->kvm))
527 		return undef_access(vcpu, p, r);
528 
529 	/* Treat MTE S/W ops as we treat the classic ones: with contempt */
530 	return access_dcsw(vcpu, p, r);
531 }
532 
533 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
534 {
535 	switch (r->aarch32_map) {
536 	case AA32_LO:
537 		*mask = GENMASK_ULL(31, 0);
538 		*shift = 0;
539 		break;
540 	case AA32_HI:
541 		*mask = GENMASK_ULL(63, 32);
542 		*shift = 32;
543 		break;
544 	default:
545 		*mask = GENMASK_ULL(63, 0);
546 		*shift = 0;
547 		break;
548 	}
549 }
550 
551 /*
552  * Generic accessor for VM registers. Only called as long as HCR_TVM
553  * is set. If the guest enables the MMU, we stop trapping the VM
554  * sys_regs and leave it in complete control of the caches.
555  */
556 static bool access_vm_reg(struct kvm_vcpu *vcpu,
557 			  struct sys_reg_params *p,
558 			  const struct sys_reg_desc *r)
559 {
560 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
561 	u64 val, mask, shift;
562 
563 	BUG_ON(!p->is_write);
564 
565 	get_access_mask(r, &mask, &shift);
566 
567 	if (~mask) {
568 		val = vcpu_read_sys_reg(vcpu, r->reg);
569 		val &= ~mask;
570 	} else {
571 		val = 0;
572 	}
573 
574 	val |= (p->regval & (mask >> shift)) << shift;
575 	vcpu_write_sys_reg(vcpu, val, r->reg);
576 
577 	kvm_toggle_cache(vcpu, was_enabled);
578 	return true;
579 }
580 
581 static bool access_actlr(struct kvm_vcpu *vcpu,
582 			 struct sys_reg_params *p,
583 			 const struct sys_reg_desc *r)
584 {
585 	u64 mask, shift;
586 
587 	if (p->is_write)
588 		return ignore_write(vcpu, p);
589 
590 	get_access_mask(r, &mask, &shift);
591 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
592 
593 	return true;
594 }
595 
596 /*
597  * Trap handler for the GICv3 SGI generation system register.
598  * Forward the request to the VGIC emulation.
599  * The cp15_64 code makes sure this automatically works
600  * for both AArch64 and AArch32 accesses.
601  */
602 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
603 			   struct sys_reg_params *p,
604 			   const struct sys_reg_desc *r)
605 {
606 	bool g1;
607 
608 	if (!kvm_has_gicv3(vcpu->kvm))
609 		return undef_access(vcpu, p, r);
610 
611 	if (!p->is_write)
612 		return read_from_write_only(vcpu, p, r);
613 
614 	/*
615 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
616 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
617 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
618 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
619 	 * group.
620 	 */
621 	if (p->Op0 == 0) {		/* AArch32 */
622 		switch (p->Op1) {
623 		default:		/* Keep GCC quiet */
624 		case 0:			/* ICC_SGI1R */
625 			g1 = true;
626 			break;
627 		case 1:			/* ICC_ASGI1R */
628 		case 2:			/* ICC_SGI0R */
629 			g1 = false;
630 			break;
631 		}
632 	} else {			/* AArch64 */
633 		switch (p->Op2) {
634 		default:		/* Keep GCC quiet */
635 		case 5:			/* ICC_SGI1R_EL1 */
636 			g1 = true;
637 			break;
638 		case 6:			/* ICC_ASGI1R_EL1 */
639 		case 7:			/* ICC_SGI0R_EL1 */
640 			g1 = false;
641 			break;
642 		}
643 	}
644 
645 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
646 
647 	return true;
648 }
649 
650 static bool access_gic_sre(struct kvm_vcpu *vcpu,
651 			   struct sys_reg_params *p,
652 			   const struct sys_reg_desc *r)
653 {
654 	if (!kvm_has_gicv3(vcpu->kvm))
655 		return undef_access(vcpu, p, r);
656 
657 	if (p->is_write)
658 		return ignore_write(vcpu, p);
659 
660 	if (p->Op1 == 4) {	/* ICC_SRE_EL2 */
661 		p->regval = KVM_ICC_SRE_EL2;
662 	} else {		/* ICC_SRE_EL1 */
663 		p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
664 	}
665 
666 	return true;
667 }
668 
669 static bool access_gic_dir(struct kvm_vcpu *vcpu,
670 			   struct sys_reg_params *p,
671 			   const struct sys_reg_desc *r)
672 {
673 	if (!kvm_has_gicv3(vcpu->kvm))
674 		return undef_access(vcpu, p, r);
675 
676 	if (!p->is_write)
677 		return undef_access(vcpu, p, r);
678 
679 	vgic_v3_deactivate(vcpu, p->regval);
680 
681 	return true;
682 }
683 
684 static bool access_gicv5_idr0(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
685 			      const struct sys_reg_desc *r)
686 {
687 	if (p->is_write)
688 		return undef_access(vcpu, p, r);
689 
690 	/*
691 	 * Expose KVM's priority- and ID-bits to the guest, but not GCIE_LEGACY.
692 	 *
693 	 * Note: for GICv5 the mimic the way that the num_pri_bits and
694 	 * num_id_bits fields are used with GICv3:
695 	 * - num_pri_bits stores the actual number of priority bits, whereas the
696 	 *   register field stores num_pri_bits - 1.
697 	 * - num_id_bits stores the raw field value, which is 0b0000 for 16 bits
698 	 *   and 0b0001 for 24 bits.
699 	 */
700 	p->regval = FIELD_PREP(ICC_IDR0_EL1_PRI_BITS, vcpu->arch.vgic_cpu.num_pri_bits - 1) |
701 		    FIELD_PREP(ICC_IDR0_EL1_ID_BITS, vcpu->arch.vgic_cpu.num_id_bits);
702 
703 	return true;
704 }
705 
706 static bool access_gicv5_iaffid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
707 				const struct sys_reg_desc *r)
708 {
709 	if (p->is_write)
710 		return undef_access(vcpu, p, r);
711 
712 	/*
713 	 * For GICv5 VMs, the IAFFID value is the same as the VPE ID. The VPE ID
714 	 * is the same as the VCPU's ID.
715 	 */
716 	p->regval = FIELD_PREP(ICC_IAFFIDR_EL1_IAFFID, vcpu->vcpu_id);
717 
718 	return true;
719 }
720 
721 static bool access_gicv5_ppi_enabler(struct kvm_vcpu *vcpu,
722 				     struct sys_reg_params *p,
723 				     const struct sys_reg_desc *r)
724 {
725 	unsigned long *mask = vcpu->kvm->arch.vgic.gicv5_vm.vgic_ppi_mask;
726 	struct vgic_v5_cpu_if *cpu_if = &vcpu->arch.vgic_cpu.vgic_v5;
727 	unsigned long reg = p->regval;
728 	int i;
729 
730 	/* We never expect to get here with a read! */
731 	if (WARN_ON_ONCE(!p->is_write))
732 		return undef_access(vcpu, p, r);
733 
734 	/*
735 	 * As we're only handling architected PPIs, the guest writes to the
736 	 * enable for the non-architected PPIs just return as there's
737 	 * nothing to do at all. We don't even allocate the storage for them.
738 	 */
739 	if (p->Op2 % 2)
740 		return true;
741 
742 	/*
743 	 * Merge the raw guest write into out bitmap, anded with our PPI mask.
744 	 */
745 	bitmap_and(cpu_if->vgic_ppi_enabler, &reg, mask, VGIC_V5_NR_PRIVATE_IRQS);
746 
747 	/*
748 	 * Sync the change in enable states to the vgic_irqs. We consider all
749 	 * PPIs as we don't expose many to the guest.
750 	 */
751 	for_each_visible_v5_ppi(i, vcpu->kvm) {
752 		u32 intid = vgic_v5_make_ppi(i);
753 		struct vgic_irq *irq;
754 
755 		irq = vgic_get_vcpu_irq(vcpu, intid);
756 
757 		scoped_guard(raw_spinlock_irqsave, &irq->irq_lock)
758 			irq->enabled = test_bit(i, cpu_if->vgic_ppi_enabler);
759 
760 		vgic_put_irq(vcpu->kvm, irq);
761 	}
762 
763 	return true;
764 }
765 
766 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
767 			struct sys_reg_params *p,
768 			const struct sys_reg_desc *r)
769 {
770 	if (p->is_write)
771 		return ignore_write(vcpu, p);
772 	else
773 		return read_zero(vcpu, p);
774 }
775 
776 /*
777  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
778  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
779  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
780  * treat it separately.
781  */
782 static bool trap_loregion(struct kvm_vcpu *vcpu,
783 			  struct sys_reg_params *p,
784 			  const struct sys_reg_desc *r)
785 {
786 	u32 sr = reg_to_encoding(r);
787 
788 	if (!kvm_has_feat(vcpu->kvm, ID_AA64MMFR1_EL1, LO, IMP))
789 		return undef_access(vcpu, p, r);
790 
791 	if (p->is_write && sr == SYS_LORID_EL1)
792 		return write_to_read_only(vcpu, p, r);
793 
794 	return trap_raz_wi(vcpu, p, r);
795 }
796 
797 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
798 			   struct sys_reg_params *p,
799 			   const struct sys_reg_desc *r)
800 {
801 	if (!p->is_write)
802 		return read_from_write_only(vcpu, p, r);
803 
804 	kvm_debug_handle_oslar(vcpu, p->regval);
805 	return true;
806 }
807 
808 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
809 			   struct sys_reg_params *p,
810 			   const struct sys_reg_desc *r)
811 {
812 	if (p->is_write)
813 		return write_to_read_only(vcpu, p, r);
814 
815 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
816 	return true;
817 }
818 
819 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
820 			 u64 val)
821 {
822 	/*
823 	 * The only modifiable bit is the OSLK bit. Refuse the write if
824 	 * userspace attempts to change any other bit in the register.
825 	 */
826 	if ((val ^ rd->val) & ~OSLSR_EL1_OSLK)
827 		return -EINVAL;
828 
829 	__vcpu_assign_sys_reg(vcpu, rd->reg, val);
830 	return 0;
831 }
832 
833 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
834 				   struct sys_reg_params *p,
835 				   const struct sys_reg_desc *r)
836 {
837 	if (p->is_write) {
838 		return ignore_write(vcpu, p);
839 	} else {
840 		p->regval = read_sysreg(dbgauthstatus_el1);
841 		return true;
842 	}
843 }
844 
845 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
846 			    struct sys_reg_params *p,
847 			    const struct sys_reg_desc *r)
848 {
849 	access_rw(vcpu, p, r);
850 
851 	kvm_debug_set_guest_ownership(vcpu);
852 	return true;
853 }
854 
855 /*
856  * reg_to_dbg/dbg_to_reg
857  *
858  * A 32 bit write to a debug register leave top bits alone
859  * A 32 bit read from a debug register only returns the bottom bits
860  */
861 static void reg_to_dbg(struct kvm_vcpu *vcpu,
862 		       struct sys_reg_params *p,
863 		       const struct sys_reg_desc *rd,
864 		       u64 *dbg_reg)
865 {
866 	u64 mask, shift, val;
867 
868 	get_access_mask(rd, &mask, &shift);
869 
870 	val = *dbg_reg;
871 	val &= ~mask;
872 	val |= (p->regval & (mask >> shift)) << shift;
873 	*dbg_reg = val;
874 }
875 
876 static void dbg_to_reg(struct kvm_vcpu *vcpu,
877 		       struct sys_reg_params *p,
878 		       const struct sys_reg_desc *rd,
879 		       u64 *dbg_reg)
880 {
881 	u64 mask, shift;
882 
883 	get_access_mask(rd, &mask, &shift);
884 	p->regval = (*dbg_reg & mask) >> shift;
885 }
886 
887 static u64 *demux_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd)
888 {
889 	struct kvm_guest_debug_arch *dbg = &vcpu->arch.vcpu_debug_state;
890 
891 	switch (rd->Op2) {
892 	case 0b100:
893 		return &dbg->dbg_bvr[rd->CRm];
894 	case 0b101:
895 		return &dbg->dbg_bcr[rd->CRm];
896 	case 0b110:
897 		return &dbg->dbg_wvr[rd->CRm];
898 	case 0b111:
899 		return &dbg->dbg_wcr[rd->CRm];
900 	default:
901 		KVM_BUG_ON(1, vcpu->kvm);
902 		return NULL;
903 	}
904 }
905 
906 static bool trap_dbg_wb_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
907 			    const struct sys_reg_desc *rd)
908 {
909 	u64 *reg = demux_wb_reg(vcpu, rd);
910 
911 	if (!reg)
912 		return false;
913 
914 	if (p->is_write)
915 		reg_to_dbg(vcpu, p, rd, reg);
916 	else
917 		dbg_to_reg(vcpu, p, rd, reg);
918 
919 	kvm_debug_set_guest_ownership(vcpu);
920 	return true;
921 }
922 
923 static int set_dbg_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
924 			  u64 val)
925 {
926 	u64 *reg = demux_wb_reg(vcpu, rd);
927 
928 	if (!reg)
929 		return -EINVAL;
930 
931 	*reg = val;
932 	return 0;
933 }
934 
935 static int get_dbg_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
936 			  u64 *val)
937 {
938 	u64 *reg = demux_wb_reg(vcpu, rd);
939 
940 	if (!reg)
941 		return -EINVAL;
942 
943 	*val = *reg;
944 	return 0;
945 }
946 
947 static u64 reset_dbg_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd)
948 {
949 	u64 *reg = demux_wb_reg(vcpu, rd);
950 
951 	/*
952 	 * Bail early if we couldn't find storage for the register, the
953 	 * KVM_BUG_ON() in demux_wb_reg() will prevent this VM from ever
954 	 * being run.
955 	 */
956 	if (!reg)
957 		return 0;
958 
959 	*reg = rd->val;
960 	return rd->val;
961 }
962 
963 static u64 reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
964 {
965 	u64 amair = read_sysreg(amair_el1);
966 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
967 	return amair;
968 }
969 
970 static u64 reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
971 {
972 	u64 actlr = read_sysreg(actlr_el1);
973 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
974 	return actlr;
975 }
976 
977 static u64 reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
978 {
979 	u64 mpidr;
980 
981 	/*
982 	 * Map the vcpu_id into the first three affinity level fields of
983 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
984 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
985 	 * of the GICv3 to be able to address each CPU directly when
986 	 * sending IPIs.
987 	 */
988 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
989 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
990 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
991 	mpidr |= (1ULL << 31);
992 	vcpu_write_sys_reg(vcpu, mpidr, MPIDR_EL1);
993 
994 	return mpidr;
995 }
996 
997 static unsigned int hidden_visibility(const struct kvm_vcpu *vcpu,
998 				      const struct sys_reg_desc *r)
999 {
1000 	return REG_HIDDEN;
1001 }
1002 
1003 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
1004 				   const struct sys_reg_desc *r)
1005 {
1006 	if (kvm_vcpu_has_pmu(vcpu))
1007 		return 0;
1008 
1009 	return REG_HIDDEN;
1010 }
1011 
1012 static u64 reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1013 {
1014 	u64 mask = BIT(ARMV8_PMU_CYCLE_IDX);
1015 	u8 n = vcpu->kvm->arch.nr_pmu_counters;
1016 
1017 	if (n)
1018 		mask |= GENMASK(n - 1, 0);
1019 
1020 	reset_unknown(vcpu, r);
1021 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, mask);
1022 
1023 	return __vcpu_sys_reg(vcpu, r->reg);
1024 }
1025 
1026 static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1027 {
1028 	reset_unknown(vcpu, r);
1029 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, GENMASK(31, 0));
1030 
1031 	return __vcpu_sys_reg(vcpu, r->reg);
1032 }
1033 
1034 static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1035 {
1036 	/* This thing will UNDEF, who cares about the reset value? */
1037 	if (!kvm_vcpu_has_pmu(vcpu))
1038 		return 0;
1039 
1040 	reset_unknown(vcpu, r);
1041 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, kvm_pmu_evtyper_mask(vcpu->kvm));
1042 
1043 	return __vcpu_sys_reg(vcpu, r->reg);
1044 }
1045 
1046 static u64 reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1047 {
1048 	reset_unknown(vcpu, r);
1049 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, PMSELR_EL0_SEL_MASK);
1050 
1051 	return __vcpu_sys_reg(vcpu, r->reg);
1052 }
1053 
1054 static u64 reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1055 {
1056 	u64 pmcr = 0;
1057 
1058 	if (!kvm_supports_32bit_el0())
1059 		pmcr |= ARMV8_PMU_PMCR_LC;
1060 
1061 	/*
1062 	 * The value of PMCR.N field is included when the
1063 	 * vCPU register is read via kvm_vcpu_read_pmcr().
1064 	 */
1065 	__vcpu_assign_sys_reg(vcpu, r->reg, pmcr);
1066 
1067 	return __vcpu_sys_reg(vcpu, r->reg);
1068 }
1069 
1070 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
1071 {
1072 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
1073 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
1074 
1075 	if (!enabled)
1076 		kvm_inject_undefined(vcpu);
1077 
1078 	return !enabled;
1079 }
1080 
1081 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
1082 {
1083 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
1084 }
1085 
1086 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
1087 {
1088 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
1089 }
1090 
1091 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
1092 {
1093 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
1094 }
1095 
1096 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
1097 {
1098 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
1099 }
1100 
1101 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1102 			const struct sys_reg_desc *r)
1103 {
1104 	u64 val;
1105 
1106 	if (pmu_access_el0_disabled(vcpu))
1107 		return false;
1108 
1109 	if (p->is_write) {
1110 		/*
1111 		 * Only update writeable bits of PMCR (continuing into
1112 		 * kvm_pmu_handle_pmcr() as well)
1113 		 */
1114 		val = kvm_vcpu_read_pmcr(vcpu);
1115 		val &= ~ARMV8_PMU_PMCR_MASK;
1116 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
1117 		if (!kvm_supports_32bit_el0())
1118 			val |= ARMV8_PMU_PMCR_LC;
1119 		kvm_pmu_handle_pmcr(vcpu, val);
1120 	} else {
1121 		/* PMCR.P & PMCR.C are RAZ */
1122 		val = kvm_vcpu_read_pmcr(vcpu)
1123 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
1124 		p->regval = val;
1125 	}
1126 
1127 	return true;
1128 }
1129 
1130 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1131 			  const struct sys_reg_desc *r)
1132 {
1133 	if (pmu_access_event_counter_el0_disabled(vcpu))
1134 		return false;
1135 
1136 	if (p->is_write)
1137 		__vcpu_assign_sys_reg(vcpu, PMSELR_EL0, p->regval);
1138 	else
1139 		/* return PMSELR.SEL field */
1140 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
1141 			    & PMSELR_EL0_SEL_MASK;
1142 
1143 	return true;
1144 }
1145 
1146 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1147 			  const struct sys_reg_desc *r)
1148 {
1149 	u64 pmceid, mask, shift;
1150 
1151 	BUG_ON(p->is_write);
1152 
1153 	if (pmu_access_el0_disabled(vcpu))
1154 		return false;
1155 
1156 	get_access_mask(r, &mask, &shift);
1157 
1158 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
1159 	pmceid &= mask;
1160 	pmceid >>= shift;
1161 
1162 	p->regval = pmceid;
1163 
1164 	return true;
1165 }
1166 
1167 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
1168 {
1169 	u64 pmcr, val;
1170 
1171 	pmcr = kvm_vcpu_read_pmcr(vcpu);
1172 	val = FIELD_GET(ARMV8_PMU_PMCR_N, pmcr);
1173 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
1174 		kvm_inject_undefined(vcpu);
1175 		return false;
1176 	}
1177 
1178 	return true;
1179 }
1180 
1181 static int get_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1182 			  u64 *val)
1183 {
1184 	u64 idx;
1185 
1186 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1187 		/* PMCCNTR_EL0 */
1188 		idx = ARMV8_PMU_CYCLE_IDX;
1189 	else
1190 		/* PMEVCNTRn_EL0 */
1191 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1192 
1193 	*val = kvm_pmu_get_counter_value(vcpu, idx);
1194 	return 0;
1195 }
1196 
1197 static int set_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1198 			  u64 val)
1199 {
1200 	u64 idx;
1201 
1202 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1203 		/* PMCCNTR_EL0 */
1204 		idx = ARMV8_PMU_CYCLE_IDX;
1205 	else
1206 		/* PMEVCNTRn_EL0 */
1207 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1208 
1209 	kvm_pmu_set_counter_value_user(vcpu, idx, val);
1210 	return 0;
1211 }
1212 
1213 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
1214 			      struct sys_reg_params *p,
1215 			      const struct sys_reg_desc *r)
1216 {
1217 	u64 idx = ~0UL;
1218 
1219 	if (r->CRn == 9 && r->CRm == 13) {
1220 		if (r->Op2 == 2) {
1221 			/* PMXEVCNTR_EL0 */
1222 			if (pmu_access_event_counter_el0_disabled(vcpu))
1223 				return false;
1224 
1225 			idx = SYS_FIELD_GET(PMSELR_EL0, SEL,
1226 					    __vcpu_sys_reg(vcpu, PMSELR_EL0));
1227 		} else if (r->Op2 == 0) {
1228 			/* PMCCNTR_EL0 */
1229 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
1230 				return false;
1231 
1232 			idx = ARMV8_PMU_CYCLE_IDX;
1233 		}
1234 	} else if (r->CRn == 0 && r->CRm == 9) {
1235 		/* PMCCNTR */
1236 		if (pmu_access_event_counter_el0_disabled(vcpu))
1237 			return false;
1238 
1239 		idx = ARMV8_PMU_CYCLE_IDX;
1240 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
1241 		/* PMEVCNTRn_EL0 */
1242 		if (pmu_access_event_counter_el0_disabled(vcpu))
1243 			return false;
1244 
1245 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1246 	}
1247 
1248 	/* Catch any decoding mistake */
1249 	WARN_ON(idx == ~0UL);
1250 
1251 	if (!pmu_counter_idx_valid(vcpu, idx))
1252 		return false;
1253 
1254 	if (p->is_write) {
1255 		if (pmu_access_el0_disabled(vcpu))
1256 			return false;
1257 
1258 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
1259 	} else {
1260 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
1261 	}
1262 
1263 	return true;
1264 }
1265 
1266 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1267 			       const struct sys_reg_desc *r)
1268 {
1269 	u64 idx, reg;
1270 
1271 	if (pmu_access_el0_disabled(vcpu))
1272 		return false;
1273 
1274 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
1275 		/* PMXEVTYPER_EL0 */
1276 		idx = SYS_FIELD_GET(PMSELR_EL0, SEL, __vcpu_sys_reg(vcpu, PMSELR_EL0));
1277 		reg = PMEVTYPER0_EL0 + idx;
1278 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
1279 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1280 		if (idx == ARMV8_PMU_CYCLE_IDX)
1281 			reg = PMCCFILTR_EL0;
1282 		else
1283 			/* PMEVTYPERn_EL0 */
1284 			reg = PMEVTYPER0_EL0 + idx;
1285 	} else {
1286 		BUG();
1287 	}
1288 
1289 	if (!pmu_counter_idx_valid(vcpu, idx))
1290 		return false;
1291 
1292 	if (p->is_write) {
1293 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
1294 		kvm_vcpu_pmu_restore_guest(vcpu);
1295 	} else {
1296 		p->regval = __vcpu_sys_reg(vcpu, reg);
1297 	}
1298 
1299 	return true;
1300 }
1301 
1302 static int set_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 val)
1303 {
1304 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1305 
1306 	__vcpu_assign_sys_reg(vcpu, r->reg, val & mask);
1307 	kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
1308 
1309 	return 0;
1310 }
1311 
1312 static int get_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 *val)
1313 {
1314 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1315 
1316 	*val = __vcpu_sys_reg(vcpu, r->reg) & mask;
1317 	return 0;
1318 }
1319 
1320 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1321 			   const struct sys_reg_desc *r)
1322 {
1323 	u64 val, mask;
1324 
1325 	if (pmu_access_el0_disabled(vcpu))
1326 		return false;
1327 
1328 	mask = kvm_pmu_accessible_counter_mask(vcpu);
1329 	if (p->is_write) {
1330 		val = p->regval & mask;
1331 		if (r->Op2 & 0x1)
1332 			/* accessing PMCNTENSET_EL0 */
1333 			__vcpu_rmw_sys_reg(vcpu, PMCNTENSET_EL0, |=, val);
1334 		else
1335 			/* accessing PMCNTENCLR_EL0 */
1336 			__vcpu_rmw_sys_reg(vcpu, PMCNTENSET_EL0, &=, ~val);
1337 
1338 		kvm_pmu_reprogram_counter_mask(vcpu, val);
1339 	} else {
1340 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
1341 	}
1342 
1343 	return true;
1344 }
1345 
1346 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1347 			   const struct sys_reg_desc *r)
1348 {
1349 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1350 
1351 	if (check_pmu_access_disabled(vcpu, 0))
1352 		return false;
1353 
1354 	if (p->is_write) {
1355 		u64 val = p->regval & mask;
1356 
1357 		if (r->Op2 & 0x1)
1358 			/* accessing PMINTENSET_EL1 */
1359 			__vcpu_rmw_sys_reg(vcpu, PMINTENSET_EL1, |=, val);
1360 		else
1361 			/* accessing PMINTENCLR_EL1 */
1362 			__vcpu_rmw_sys_reg(vcpu, PMINTENSET_EL1, &=, ~val);
1363 	} else {
1364 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
1365 	}
1366 
1367 	return true;
1368 }
1369 
1370 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1371 			 const struct sys_reg_desc *r)
1372 {
1373 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1374 
1375 	if (pmu_access_el0_disabled(vcpu))
1376 		return false;
1377 
1378 	if (p->is_write) {
1379 		if (r->CRm & 0x2)
1380 			/* accessing PMOVSSET_EL0 */
1381 			__vcpu_rmw_sys_reg(vcpu, PMOVSSET_EL0, |=, (p->regval & mask));
1382 		else
1383 			/* accessing PMOVSCLR_EL0 */
1384 			__vcpu_rmw_sys_reg(vcpu, PMOVSSET_EL0, &=, ~(p->regval & mask));
1385 	} else {
1386 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
1387 	}
1388 
1389 	return true;
1390 }
1391 
1392 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1393 			   const struct sys_reg_desc *r)
1394 {
1395 	u64 mask;
1396 
1397 	if (!p->is_write)
1398 		return read_from_write_only(vcpu, p, r);
1399 
1400 	if (pmu_write_swinc_el0_disabled(vcpu))
1401 		return false;
1402 
1403 	mask = kvm_pmu_accessible_counter_mask(vcpu);
1404 	kvm_pmu_software_increment(vcpu, p->regval & mask);
1405 	return true;
1406 }
1407 
1408 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1409 			     const struct sys_reg_desc *r)
1410 {
1411 	if (p->is_write) {
1412 		if (!vcpu_mode_priv(vcpu))
1413 			return undef_access(vcpu, p, r);
1414 
1415 		__vcpu_assign_sys_reg(vcpu, PMUSERENR_EL0,
1416 				      (p->regval & ARMV8_PMU_USERENR_MASK));
1417 	} else {
1418 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1419 			    & ARMV8_PMU_USERENR_MASK;
1420 	}
1421 
1422 	return true;
1423 }
1424 
1425 static int get_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1426 		    u64 *val)
1427 {
1428 	*val = kvm_vcpu_read_pmcr(vcpu);
1429 	return 0;
1430 }
1431 
1432 static int set_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1433 		    u64 val)
1434 {
1435 	u8 new_n = FIELD_GET(ARMV8_PMU_PMCR_N, val);
1436 	struct kvm *kvm = vcpu->kvm;
1437 
1438 	mutex_lock(&kvm->arch.config_lock);
1439 
1440 	/*
1441 	 * The vCPU can't have more counters than the PMU hardware
1442 	 * implements. Ignore this error to maintain compatibility
1443 	 * with the existing KVM behavior.
1444 	 */
1445 	if (!kvm_vm_has_ran_once(kvm) &&
1446 	    !vcpu_has_nv(vcpu)	      &&
1447 	    new_n <= kvm_arm_pmu_get_max_counters(kvm))
1448 		kvm->arch.nr_pmu_counters = new_n;
1449 
1450 	mutex_unlock(&kvm->arch.config_lock);
1451 
1452 	/*
1453 	 * Ignore writes to RES0 bits, read only bits that are cleared on
1454 	 * vCPU reset, and writable bits that KVM doesn't support yet.
1455 	 * (i.e. only PMCR.N and bits [7:0] are mutable from userspace)
1456 	 * The LP bit is RES0 when FEAT_PMUv3p5 is not supported on the vCPU.
1457 	 * But, we leave the bit as it is here, as the vCPU's PMUver might
1458 	 * be changed later (NOTE: the bit will be cleared on first vCPU run
1459 	 * if necessary).
1460 	 */
1461 	val &= ARMV8_PMU_PMCR_MASK;
1462 
1463 	/* The LC bit is RES1 when AArch32 is not supported */
1464 	if (!kvm_supports_32bit_el0())
1465 		val |= ARMV8_PMU_PMCR_LC;
1466 
1467 	__vcpu_assign_sys_reg(vcpu, r->reg, val);
1468 	kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
1469 
1470 	return 0;
1471 }
1472 
1473 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1474 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1475 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1476 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1477 	  get_dbg_wb_reg, set_dbg_wb_reg },				\
1478 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1479 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1480 	  get_dbg_wb_reg, set_dbg_wb_reg },				\
1481 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1482 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1483 	  get_dbg_wb_reg, set_dbg_wb_reg },				\
1484 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1485 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1486 	  get_dbg_wb_reg, set_dbg_wb_reg }
1487 
1488 #define PMU_SYS_REG(name)						\
1489 	SYS_DESC(SYS_##name), .reset = reset_pmu_reg,			\
1490 	.visibility = pmu_visibility
1491 
1492 /* Macro to expand the PMEVCNTRn_EL0 register */
1493 #define PMU_PMEVCNTR_EL0(n)						\
1494 	{ PMU_SYS_REG(PMEVCNTRn_EL0(n)),				\
1495 	  .reset = reset_pmevcntr, .get_user = get_pmu_evcntr,		\
1496 	  .set_user = set_pmu_evcntr,					\
1497 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
1498 
1499 /* Macro to expand the PMEVTYPERn_EL0 register */
1500 #define PMU_PMEVTYPER_EL0(n)						\
1501 	{ PMU_SYS_REG(PMEVTYPERn_EL0(n)),				\
1502 	  .reset = reset_pmevtyper,					\
1503 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
1504 
1505 /* Macro to expand the AMU counter and type registers*/
1506 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1507 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1508 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1509 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1510 
1511 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1512 			const struct sys_reg_desc *rd)
1513 {
1514 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1515 }
1516 
1517 /*
1518  * If we land here on a PtrAuth access, that is because we didn't
1519  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1520  * way this happens is when the guest does not have PtrAuth support
1521  * enabled.
1522  */
1523 #define __PTRAUTH_KEY(k)						\
1524 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1525 	.visibility = ptrauth_visibility}
1526 
1527 #define PTRAUTH_KEY(k)							\
1528 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1529 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1530 
1531 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1532 			      struct sys_reg_params *p,
1533 			      const struct sys_reg_desc *r)
1534 {
1535 	enum kvm_arch_timers tmr;
1536 	enum kvm_arch_timer_regs treg;
1537 	u64 reg = reg_to_encoding(r);
1538 
1539 	switch (reg) {
1540 	case SYS_CNTP_TVAL_EL0:
1541 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1542 			tmr = TIMER_HPTIMER;
1543 		else
1544 			tmr = TIMER_PTIMER;
1545 		treg = TIMER_REG_TVAL;
1546 		break;
1547 
1548 	case SYS_CNTV_TVAL_EL0:
1549 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1550 			tmr = TIMER_HVTIMER;
1551 		else
1552 			tmr = TIMER_VTIMER;
1553 		treg = TIMER_REG_TVAL;
1554 		break;
1555 
1556 	case SYS_AARCH32_CNTP_TVAL:
1557 	case SYS_CNTP_TVAL_EL02:
1558 		tmr = TIMER_PTIMER;
1559 		treg = TIMER_REG_TVAL;
1560 		break;
1561 
1562 	case SYS_CNTV_TVAL_EL02:
1563 		tmr = TIMER_VTIMER;
1564 		treg = TIMER_REG_TVAL;
1565 		break;
1566 
1567 	case SYS_CNTHP_TVAL_EL2:
1568 		tmr = TIMER_HPTIMER;
1569 		treg = TIMER_REG_TVAL;
1570 		break;
1571 
1572 	case SYS_CNTHV_TVAL_EL2:
1573 		tmr = TIMER_HVTIMER;
1574 		treg = TIMER_REG_TVAL;
1575 		break;
1576 
1577 	case SYS_CNTP_CTL_EL0:
1578 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1579 			tmr = TIMER_HPTIMER;
1580 		else
1581 			tmr = TIMER_PTIMER;
1582 		treg = TIMER_REG_CTL;
1583 		break;
1584 
1585 	case SYS_CNTV_CTL_EL0:
1586 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1587 			tmr = TIMER_HVTIMER;
1588 		else
1589 			tmr = TIMER_VTIMER;
1590 		treg = TIMER_REG_CTL;
1591 		break;
1592 
1593 	case SYS_AARCH32_CNTP_CTL:
1594 	case SYS_CNTP_CTL_EL02:
1595 		tmr = TIMER_PTIMER;
1596 		treg = TIMER_REG_CTL;
1597 		break;
1598 
1599 	case SYS_CNTV_CTL_EL02:
1600 		tmr = TIMER_VTIMER;
1601 		treg = TIMER_REG_CTL;
1602 		break;
1603 
1604 	case SYS_CNTHP_CTL_EL2:
1605 		tmr = TIMER_HPTIMER;
1606 		treg = TIMER_REG_CTL;
1607 		break;
1608 
1609 	case SYS_CNTHV_CTL_EL2:
1610 		tmr = TIMER_HVTIMER;
1611 		treg = TIMER_REG_CTL;
1612 		break;
1613 
1614 	case SYS_CNTP_CVAL_EL0:
1615 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1616 			tmr = TIMER_HPTIMER;
1617 		else
1618 			tmr = TIMER_PTIMER;
1619 		treg = TIMER_REG_CVAL;
1620 		break;
1621 
1622 	case SYS_CNTV_CVAL_EL0:
1623 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1624 			tmr = TIMER_HVTIMER;
1625 		else
1626 			tmr = TIMER_VTIMER;
1627 		treg = TIMER_REG_CVAL;
1628 		break;
1629 
1630 	case SYS_AARCH32_CNTP_CVAL:
1631 	case SYS_CNTP_CVAL_EL02:
1632 		tmr = TIMER_PTIMER;
1633 		treg = TIMER_REG_CVAL;
1634 		break;
1635 
1636 	case SYS_CNTV_CVAL_EL02:
1637 		tmr = TIMER_VTIMER;
1638 		treg = TIMER_REG_CVAL;
1639 		break;
1640 
1641 	case SYS_CNTHP_CVAL_EL2:
1642 		tmr = TIMER_HPTIMER;
1643 		treg = TIMER_REG_CVAL;
1644 		break;
1645 
1646 	case SYS_CNTHV_CVAL_EL2:
1647 		tmr = TIMER_HVTIMER;
1648 		treg = TIMER_REG_CVAL;
1649 		break;
1650 
1651 	case SYS_CNTPCT_EL0:
1652 	case SYS_CNTPCTSS_EL0:
1653 		if (is_hyp_ctxt(vcpu))
1654 			tmr = TIMER_HPTIMER;
1655 		else
1656 			tmr = TIMER_PTIMER;
1657 		treg = TIMER_REG_CNT;
1658 		break;
1659 
1660 	case SYS_AARCH32_CNTPCT:
1661 	case SYS_AARCH32_CNTPCTSS:
1662 		tmr = TIMER_PTIMER;
1663 		treg = TIMER_REG_CNT;
1664 		break;
1665 
1666 	case SYS_CNTVCT_EL0:
1667 	case SYS_CNTVCTSS_EL0:
1668 		if (is_hyp_ctxt(vcpu))
1669 			tmr = TIMER_HVTIMER;
1670 		else
1671 			tmr = TIMER_VTIMER;
1672 		treg = TIMER_REG_CNT;
1673 		break;
1674 
1675 	case SYS_AARCH32_CNTVCT:
1676 	case SYS_AARCH32_CNTVCTSS:
1677 		tmr = TIMER_VTIMER;
1678 		treg = TIMER_REG_CNT;
1679 		break;
1680 
1681 	default:
1682 		print_sys_reg_msg(p, "%s", "Unhandled trapped timer register");
1683 		return undef_access(vcpu, p, r);
1684 	}
1685 
1686 	if (p->is_write)
1687 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1688 	else
1689 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1690 
1691 	return true;
1692 }
1693 
1694 static int arch_timer_set_user(struct kvm_vcpu *vcpu,
1695 			       const struct sys_reg_desc *rd,
1696 			       u64 val)
1697 {
1698 	switch (reg_to_encoding(rd)) {
1699 	case SYS_CNTV_CTL_EL0:
1700 	case SYS_CNTP_CTL_EL0:
1701 	case SYS_CNTHV_CTL_EL2:
1702 	case SYS_CNTHP_CTL_EL2:
1703 		val &= ~ARCH_TIMER_CTRL_IT_STAT;
1704 		break;
1705 	case SYS_CNTVCT_EL0:
1706 		if (!test_bit(KVM_ARCH_FLAG_VM_COUNTER_OFFSET, &vcpu->kvm->arch.flags))
1707 			timer_set_offset(vcpu_vtimer(vcpu), kvm_phys_timer_read() - val);
1708 		return 0;
1709 	case SYS_CNTPCT_EL0:
1710 		if (!test_bit(KVM_ARCH_FLAG_VM_COUNTER_OFFSET, &vcpu->kvm->arch.flags))
1711 			timer_set_offset(vcpu_ptimer(vcpu), kvm_phys_timer_read() - val);
1712 		return 0;
1713 	}
1714 
1715 	__vcpu_assign_sys_reg(vcpu, rd->reg, val);
1716 	return 0;
1717 }
1718 
1719 static int arch_timer_get_user(struct kvm_vcpu *vcpu,
1720 			       const struct sys_reg_desc *rd,
1721 			       u64 *val)
1722 {
1723 	switch (reg_to_encoding(rd)) {
1724 	case SYS_CNTVCT_EL0:
1725 		*val = kvm_phys_timer_read() - timer_get_offset(vcpu_vtimer(vcpu));
1726 		break;
1727 	case SYS_CNTPCT_EL0:
1728 		*val = kvm_phys_timer_read() - timer_get_offset(vcpu_ptimer(vcpu));
1729 		break;
1730 	default:
1731 		*val = __vcpu_sys_reg(vcpu, rd->reg);
1732 	}
1733 
1734 	return 0;
1735 }
1736 
1737 static s64 kvm_arm64_ftr_safe_value(u32 id, const struct arm64_ftr_bits *ftrp,
1738 				    s64 new, s64 cur)
1739 {
1740 	struct arm64_ftr_bits kvm_ftr = *ftrp;
1741 
1742 	/* Some features have different safe value type in KVM than host features */
1743 	switch (id) {
1744 	case SYS_ID_AA64DFR0_EL1:
1745 		switch (kvm_ftr.shift) {
1746 		case ID_AA64DFR0_EL1_PMUVer_SHIFT:
1747 			kvm_ftr.type = FTR_LOWER_SAFE;
1748 			break;
1749 		case ID_AA64DFR0_EL1_DebugVer_SHIFT:
1750 			kvm_ftr.type = FTR_LOWER_SAFE;
1751 			break;
1752 		}
1753 		break;
1754 	case SYS_ID_DFR0_EL1:
1755 		if (kvm_ftr.shift == ID_DFR0_EL1_PerfMon_SHIFT)
1756 			kvm_ftr.type = FTR_LOWER_SAFE;
1757 		break;
1758 	}
1759 
1760 	return arm64_ftr_safe_value(&kvm_ftr, new, cur);
1761 }
1762 
1763 /*
1764  * arm64_check_features() - Check if a feature register value constitutes
1765  * a subset of features indicated by the idreg's KVM sanitised limit.
1766  *
1767  * This function will check if each feature field of @val is the "safe" value
1768  * against idreg's KVM sanitised limit return from reset() callback.
1769  * If a field value in @val is the same as the one in limit, it is always
1770  * considered the safe value regardless For register fields that are not in
1771  * writable, only the value in limit is considered the safe value.
1772  *
1773  * Return: 0 if all the fields are safe. Otherwise, return negative errno.
1774  */
1775 static int arm64_check_features(struct kvm_vcpu *vcpu,
1776 				const struct sys_reg_desc *rd,
1777 				u64 val)
1778 {
1779 	const struct arm64_ftr_reg *ftr_reg;
1780 	const struct arm64_ftr_bits *ftrp = NULL;
1781 	u32 id = reg_to_encoding(rd);
1782 	u64 writable_mask = rd->val;
1783 	u64 limit = rd->reset(vcpu, rd);
1784 	u64 mask = 0;
1785 
1786 	/*
1787 	 * Hidden and unallocated ID registers may not have a corresponding
1788 	 * struct arm64_ftr_reg. Of course, if the register is RAZ we know the
1789 	 * only safe value is 0.
1790 	 */
1791 	if (sysreg_visible_as_raz(vcpu, rd))
1792 		return val ? -E2BIG : 0;
1793 
1794 	ftr_reg = get_arm64_ftr_reg(id);
1795 	if (!ftr_reg)
1796 		return -EINVAL;
1797 
1798 	ftrp = ftr_reg->ftr_bits;
1799 
1800 	for (; ftrp && ftrp->width; ftrp++) {
1801 		s64 f_val, f_lim, safe_val;
1802 		u64 ftr_mask;
1803 
1804 		ftr_mask = arm64_ftr_mask(ftrp);
1805 		if ((ftr_mask & writable_mask) != ftr_mask)
1806 			continue;
1807 
1808 		f_val = arm64_ftr_value(ftrp, val);
1809 		f_lim = arm64_ftr_value(ftrp, limit);
1810 		mask |= ftr_mask;
1811 
1812 		if (f_val == f_lim)
1813 			safe_val = f_val;
1814 		else
1815 			safe_val = kvm_arm64_ftr_safe_value(id, ftrp, f_val, f_lim);
1816 
1817 		if (safe_val != f_val)
1818 			return -E2BIG;
1819 	}
1820 
1821 	/* For fields that are not writable, values in limit are the safe values. */
1822 	if ((val & ~mask) != (limit & ~mask))
1823 		return -E2BIG;
1824 
1825 	return 0;
1826 }
1827 
1828 static u8 pmuver_to_perfmon(u8 pmuver)
1829 {
1830 	switch (pmuver) {
1831 	case ID_AA64DFR0_EL1_PMUVer_IMP:
1832 		return ID_DFR0_EL1_PerfMon_PMUv3;
1833 	case ID_AA64DFR0_EL1_PMUVer_IMP_DEF:
1834 		return ID_DFR0_EL1_PerfMon_IMPDEF;
1835 	default:
1836 		/* Anything ARMv8.1+ and NI have the same value. For now. */
1837 		return pmuver;
1838 	}
1839 }
1840 
1841 static u64 sanitise_id_aa64pfr0_el1(const struct kvm_vcpu *vcpu, u64 val);
1842 static u64 sanitise_id_aa64pfr1_el1(const struct kvm_vcpu *vcpu, u64 val);
1843 static u64 sanitise_id_aa64pfr2_el1(const struct kvm_vcpu *vcpu, u64 val);
1844 static u64 sanitise_id_aa64dfr0_el1(const struct kvm_vcpu *vcpu, u64 val);
1845 
1846 /* Read a sanitised cpufeature ID register by sys_reg_desc */
1847 static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu,
1848 				       const struct sys_reg_desc *r)
1849 {
1850 	u32 id = reg_to_encoding(r);
1851 	u64 val;
1852 
1853 	if (sysreg_visible_as_raz(vcpu, r))
1854 		return 0;
1855 
1856 	val = read_sanitised_ftr_reg(id);
1857 
1858 	switch (id) {
1859 	case SYS_ID_AA64DFR0_EL1:
1860 		val = sanitise_id_aa64dfr0_el1(vcpu, val);
1861 		break;
1862 	case SYS_ID_AA64PFR0_EL1:
1863 		val = sanitise_id_aa64pfr0_el1(vcpu, val);
1864 		break;
1865 	case SYS_ID_AA64PFR1_EL1:
1866 		val = sanitise_id_aa64pfr1_el1(vcpu, val);
1867 		break;
1868 	case SYS_ID_AA64PFR2_EL1:
1869 		val = sanitise_id_aa64pfr2_el1(vcpu, val);
1870 		break;
1871 	case SYS_ID_AA64ISAR1_EL1:
1872 		if (!vcpu_has_ptrauth(vcpu))
1873 			val &= ~(ID_AA64ISAR1_EL1_APA |
1874 				 ID_AA64ISAR1_EL1_API |
1875 				 ID_AA64ISAR1_EL1_GPA |
1876 				 ID_AA64ISAR1_EL1_GPI);
1877 		break;
1878 	case SYS_ID_AA64ISAR2_EL1:
1879 		if (!vcpu_has_ptrauth(vcpu))
1880 			val &= ~(ID_AA64ISAR2_EL1_APA3 |
1881 				 ID_AA64ISAR2_EL1_GPA3);
1882 		if (!cpus_have_final_cap(ARM64_HAS_WFXT) ||
1883 		    has_broken_cntvoff())
1884 			val &= ~ID_AA64ISAR2_EL1_WFxT;
1885 		break;
1886 	case SYS_ID_AA64ISAR3_EL1:
1887 		val &= ID_AA64ISAR3_EL1_FPRCVT | ID_AA64ISAR3_EL1_LSFE |
1888 			ID_AA64ISAR3_EL1_FAMINMAX | ID_AA64ISAR3_EL1_LSUI;
1889 		break;
1890 	case SYS_ID_AA64MMFR2_EL1:
1891 		val &= ~ID_AA64MMFR2_EL1_CCIDX_MASK;
1892 		val &= ~ID_AA64MMFR2_EL1_NV;
1893 		break;
1894 	case SYS_ID_AA64MMFR3_EL1:
1895 		val &= ID_AA64MMFR3_EL1_TCRX |
1896 		       ID_AA64MMFR3_EL1_SCTLRX |
1897 		       ID_AA64MMFR3_EL1_S1POE |
1898 		       ID_AA64MMFR3_EL1_S1PIE;
1899 
1900 		if (!system_supports_poe())
1901 			val &= ~ID_AA64MMFR3_EL1_S1POE;
1902 		break;
1903 	case SYS_ID_MMFR4_EL1:
1904 		val &= ~ID_MMFR4_EL1_CCIDX;
1905 		break;
1906 	}
1907 
1908 	if (vcpu_has_nv(vcpu))
1909 		val = limit_nv_id_reg(vcpu->kvm, id, val);
1910 
1911 	return val;
1912 }
1913 
1914 static u64 kvm_read_sanitised_id_reg(struct kvm_vcpu *vcpu,
1915 				     const struct sys_reg_desc *r)
1916 {
1917 	return __kvm_read_sanitised_id_reg(vcpu, r);
1918 }
1919 
1920 static u64 read_id_reg(const struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1921 {
1922 	return kvm_read_vm_id_reg(vcpu->kvm, reg_to_encoding(r));
1923 }
1924 
1925 static bool is_feature_id_reg(u32 encoding)
1926 {
1927 	return (sys_reg_Op0(encoding) == 3 &&
1928 		(sys_reg_Op1(encoding) < 2 || sys_reg_Op1(encoding) == 3) &&
1929 		sys_reg_CRn(encoding) == 0 &&
1930 		sys_reg_CRm(encoding) <= 7);
1931 }
1932 
1933 /*
1934  * Return true if the register's (Op0, Op1, CRn, CRm, Op2) is
1935  * (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8, which is the range of ID
1936  * registers KVM maintains on a per-VM basis.
1937  *
1938  * Additionally, the implementation ID registers and CTR_EL0 are handled as
1939  * per-VM registers.
1940  */
1941 static inline bool is_vm_ftr_id_reg(u32 id)
1942 {
1943 	switch (id) {
1944 	case SYS_CTR_EL0:
1945 	case SYS_MIDR_EL1:
1946 	case SYS_REVIDR_EL1:
1947 	case SYS_AIDR_EL1:
1948 		return true;
1949 	default:
1950 		return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1951 			sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1952 			sys_reg_CRm(id) < 8);
1953 
1954 	}
1955 }
1956 
1957 static inline bool is_vcpu_ftr_id_reg(u32 id)
1958 {
1959 	return is_feature_id_reg(id) && !is_vm_ftr_id_reg(id);
1960 }
1961 
1962 static inline bool is_aa32_id_reg(u32 id)
1963 {
1964 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1965 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1966 		sys_reg_CRm(id) <= 3);
1967 }
1968 
1969 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1970 				  const struct sys_reg_desc *r)
1971 {
1972 	u32 id = reg_to_encoding(r);
1973 
1974 	switch (id) {
1975 	case SYS_ID_AA64ZFR0_EL1:
1976 		if (!vcpu_has_sve(vcpu))
1977 			return REG_RAZ;
1978 		break;
1979 	}
1980 
1981 	return 0;
1982 }
1983 
1984 static unsigned int aa32_id_visibility(const struct kvm_vcpu *vcpu,
1985 				       const struct sys_reg_desc *r)
1986 {
1987 	/*
1988 	 * AArch32 ID registers are UNKNOWN if AArch32 isn't implemented at any
1989 	 * EL. Promote to RAZ/WI in order to guarantee consistency between
1990 	 * systems.
1991 	 */
1992 	if (!kvm_supports_32bit_el0())
1993 		return REG_RAZ | REG_USER_WI;
1994 
1995 	return id_visibility(vcpu, r);
1996 }
1997 
1998 static unsigned int raz_visibility(const struct kvm_vcpu *vcpu,
1999 				   const struct sys_reg_desc *r)
2000 {
2001 	return REG_RAZ;
2002 }
2003 
2004 /* cpufeature ID register access trap handlers */
2005 
2006 static bool access_id_reg(struct kvm_vcpu *vcpu,
2007 			  struct sys_reg_params *p,
2008 			  const struct sys_reg_desc *r)
2009 {
2010 	if (p->is_write)
2011 		return write_to_read_only(vcpu, p, r);
2012 
2013 	p->regval = read_id_reg(vcpu, r);
2014 
2015 	return true;
2016 }
2017 
2018 /* Visibility overrides for SVE-specific control registers */
2019 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
2020 				   const struct sys_reg_desc *rd)
2021 {
2022 	if (vcpu_has_sve(vcpu))
2023 		return 0;
2024 
2025 	return REG_HIDDEN;
2026 }
2027 
2028 static unsigned int sme_visibility(const struct kvm_vcpu *vcpu,
2029 				   const struct sys_reg_desc *rd)
2030 {
2031 	if (kvm_has_feat(vcpu->kvm, ID_AA64PFR1_EL1, SME, IMP))
2032 		return 0;
2033 
2034 	return REG_HIDDEN;
2035 }
2036 
2037 static unsigned int fp8_visibility(const struct kvm_vcpu *vcpu,
2038 				   const struct sys_reg_desc *rd)
2039 {
2040 	if (kvm_has_fpmr(vcpu->kvm))
2041 		return 0;
2042 
2043 	return REG_HIDDEN;
2044 }
2045 
2046 static u64 sanitise_id_aa64pfr0_el1(const struct kvm_vcpu *vcpu, u64 val)
2047 {
2048 	if (!vcpu_has_sve(vcpu))
2049 		val &= ~ID_AA64PFR0_EL1_SVE_MASK;
2050 
2051 	/*
2052 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
2053 	 * Although this is a per-CPU feature, we make it global because
2054 	 * asymmetric systems are just a nuisance.
2055 	 *
2056 	 * Userspace can override this as long as it doesn't promise
2057 	 * the impossible.
2058 	 */
2059 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) {
2060 		val &= ~ID_AA64PFR0_EL1_CSV2_MASK;
2061 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV2, IMP);
2062 	}
2063 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) {
2064 		val &= ~ID_AA64PFR0_EL1_CSV3_MASK;
2065 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV3, IMP);
2066 	}
2067 
2068 	if (vgic_host_has_gicv3()) {
2069 		val &= ~ID_AA64PFR0_EL1_GIC_MASK;
2070 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, GIC, IMP);
2071 	}
2072 
2073 	val &= ~ID_AA64PFR0_EL1_AMU_MASK;
2074 
2075 	/*
2076 	 * MPAM is disabled by default as KVM also needs a set of PARTID to
2077 	 * program the MPAMVPMx_EL2 PARTID remapping registers with. But some
2078 	 * older kernels let the guest see the ID bit.
2079 	 */
2080 	val &= ~ID_AA64PFR0_EL1_MPAM_MASK;
2081 
2082 	return val;
2083 }
2084 
2085 static u64 sanitise_id_aa64pfr1_el1(const struct kvm_vcpu *vcpu, u64 val)
2086 {
2087 	u64 pfr0 = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2088 
2089 	if (!kvm_has_mte(vcpu->kvm)) {
2090 		val &= ~ID_AA64PFR1_EL1_MTE;
2091 		val &= ~ID_AA64PFR1_EL1_MTE_frac;
2092 	}
2093 
2094 	if (!(cpus_have_final_cap(ARM64_HAS_RASV1P1_EXTN) &&
2095 	      SYS_FIELD_GET(ID_AA64PFR0_EL1, RAS, pfr0) == ID_AA64PFR0_EL1_RAS_IMP))
2096 		val &= ~ID_AA64PFR1_EL1_RAS_frac;
2097 
2098 	val &= ~ID_AA64PFR1_EL1_SME;
2099 	val &= ~ID_AA64PFR1_EL1_RNDR_trap;
2100 	val &= ~ID_AA64PFR1_EL1_NMI;
2101 	val &= ~ID_AA64PFR1_EL1_GCS;
2102 	val &= ~ID_AA64PFR1_EL1_THE;
2103 	val &= ~ID_AA64PFR1_EL1_MTEX;
2104 	val &= ~ID_AA64PFR1_EL1_PFAR;
2105 	val &= ~ID_AA64PFR1_EL1_MPAM_frac;
2106 
2107 	return val;
2108 }
2109 
2110 static u64 sanitise_id_aa64pfr2_el1(const struct kvm_vcpu *vcpu, u64 val)
2111 {
2112 	val &= ID_AA64PFR2_EL1_FPMR |
2113 	       ID_AA64PFR2_EL1_MTEFAR |
2114 	       ID_AA64PFR2_EL1_MTESTOREONLY;
2115 
2116 	if (!kvm_has_mte(vcpu->kvm)) {
2117 		val &= ~ID_AA64PFR2_EL1_MTEFAR;
2118 		val &= ~ID_AA64PFR2_EL1_MTESTOREONLY;
2119 	}
2120 
2121 	if (vgic_host_has_gicv5())
2122 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR2_EL1, GCIE, IMP);
2123 
2124 	return val;
2125 }
2126 
2127 static u64 sanitise_id_aa64dfr0_el1(const struct kvm_vcpu *vcpu, u64 val)
2128 {
2129 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_AA64DFR0_EL1, DebugVer, V8P8);
2130 
2131 	/*
2132 	 * Only initialize the PMU version if the vCPU was configured with one.
2133 	 */
2134 	val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
2135 	if (kvm_vcpu_has_pmu(vcpu))
2136 		val |= SYS_FIELD_PREP(ID_AA64DFR0_EL1, PMUVer,
2137 				      kvm_arm_pmu_get_pmuver_limit());
2138 
2139 	/* Hide SPE from guests */
2140 	val &= ~ID_AA64DFR0_EL1_PMSVer_MASK;
2141 
2142 	/* Hide BRBE from guests */
2143 	val &= ~ID_AA64DFR0_EL1_BRBE_MASK;
2144 
2145 	return val;
2146 }
2147 
2148 /*
2149  * Older versions of KVM erroneously claim support for FEAT_DoubleLock with
2150  * NV-enabled VMs on unsupporting hardware. Silently ignore the incorrect
2151  * value if it is consistent with the bug.
2152  */
2153 static bool ignore_feat_doublelock(struct kvm_vcpu *vcpu, u64 val)
2154 {
2155 	u8 host, user;
2156 
2157 	if (!vcpu_has_nv(vcpu))
2158 		return false;
2159 
2160 	host = SYS_FIELD_GET(ID_AA64DFR0_EL1, DoubleLock,
2161 			     read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1));
2162 	user = SYS_FIELD_GET(ID_AA64DFR0_EL1, DoubleLock, val);
2163 
2164 	return host == ID_AA64DFR0_EL1_DoubleLock_NI &&
2165 	       user == ID_AA64DFR0_EL1_DoubleLock_IMP;
2166 }
2167 
2168 static int set_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
2169 			       const struct sys_reg_desc *rd,
2170 			       u64 val)
2171 {
2172 	u8 debugver = SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, val);
2173 	u8 pmuver = SYS_FIELD_GET(ID_AA64DFR0_EL1, PMUVer, val);
2174 
2175 	/*
2176 	 * Prior to commit 3d0dba5764b9 ("KVM: arm64: PMU: Move the
2177 	 * ID_AA64DFR0_EL1.PMUver limit to VM creation"), KVM erroneously
2178 	 * exposed an IMP_DEF PMU to userspace and the guest on systems w/
2179 	 * non-architectural PMUs. Of course, PMUv3 is the only game in town for
2180 	 * PMU virtualization, so the IMP_DEF value was rather user-hostile.
2181 	 *
2182 	 * At minimum, we're on the hook to allow values that were given to
2183 	 * userspace by KVM. Cover our tracks here and replace the IMP_DEF value
2184 	 * with a more sensible NI. The value of an ID register changing under
2185 	 * the nose of the guest is unfortunate, but is certainly no more
2186 	 * surprising than an ill-guided PMU driver poking at impdef system
2187 	 * registers that end in an UNDEF...
2188 	 */
2189 	if (pmuver == ID_AA64DFR0_EL1_PMUVer_IMP_DEF)
2190 		val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
2191 
2192 	/*
2193 	 * ID_AA64DFR0_EL1.DebugVer is one of those awkward fields with a
2194 	 * nonzero minimum safe value.
2195 	 */
2196 	if (debugver < ID_AA64DFR0_EL1_DebugVer_IMP)
2197 		return -EINVAL;
2198 
2199 	if (ignore_feat_doublelock(vcpu, val)) {
2200 		val &= ~ID_AA64DFR0_EL1_DoubleLock;
2201 		val |= SYS_FIELD_PREP_ENUM(ID_AA64DFR0_EL1, DoubleLock, NI);
2202 	}
2203 
2204 	return set_id_reg(vcpu, rd, val);
2205 }
2206 
2207 static u64 read_sanitised_id_dfr0_el1(struct kvm_vcpu *vcpu,
2208 				      const struct sys_reg_desc *rd)
2209 {
2210 	u8 perfmon;
2211 	u64 val = read_sanitised_ftr_reg(SYS_ID_DFR0_EL1);
2212 
2213 	val &= ~ID_DFR0_EL1_PerfMon_MASK;
2214 	if (kvm_vcpu_has_pmu(vcpu)) {
2215 		perfmon = pmuver_to_perfmon(kvm_arm_pmu_get_pmuver_limit());
2216 		val |= SYS_FIELD_PREP(ID_DFR0_EL1, PerfMon, perfmon);
2217 	}
2218 
2219 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_DFR0_EL1, CopDbg, Debugv8p8);
2220 
2221 	return val;
2222 }
2223 
2224 static int set_id_dfr0_el1(struct kvm_vcpu *vcpu,
2225 			   const struct sys_reg_desc *rd,
2226 			   u64 val)
2227 {
2228 	u8 perfmon = SYS_FIELD_GET(ID_DFR0_EL1, PerfMon, val);
2229 	u8 copdbg = SYS_FIELD_GET(ID_DFR0_EL1, CopDbg, val);
2230 
2231 	if (perfmon == ID_DFR0_EL1_PerfMon_IMPDEF) {
2232 		val &= ~ID_DFR0_EL1_PerfMon_MASK;
2233 		perfmon = 0;
2234 	}
2235 
2236 	/*
2237 	 * Allow DFR0_EL1.PerfMon to be set from userspace as long as
2238 	 * it doesn't promise more than what the HW gives us on the
2239 	 * AArch64 side (as everything is emulated with that), and
2240 	 * that this is a PMUv3.
2241 	 */
2242 	if (perfmon != 0 && perfmon < ID_DFR0_EL1_PerfMon_PMUv3)
2243 		return -EINVAL;
2244 
2245 	if (copdbg < ID_DFR0_EL1_CopDbg_Armv8)
2246 		return -EINVAL;
2247 
2248 	return set_id_reg(vcpu, rd, val);
2249 }
2250 
2251 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
2252 			       const struct sys_reg_desc *rd, u64 user_val)
2253 {
2254 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2255 	u64 mpam_mask = ID_AA64PFR0_EL1_MPAM_MASK;
2256 
2257 	/*
2258 	 * Commit 011e5f5bf529f ("arm64/cpufeature: Add remaining feature bits
2259 	 * in ID_AA64PFR0 register") exposed the MPAM field of AA64PFR0_EL1 to
2260 	 * guests, but didn't add trap handling. KVM doesn't support MPAM and
2261 	 * always returns an UNDEF for these registers. The guest must see 0
2262 	 * for this field.
2263 	 *
2264 	 * But KVM must also accept values from user-space that were provided
2265 	 * by KVM. On CPUs that support MPAM, permit user-space to write
2266 	 * the sanitizied value to ID_AA64PFR0_EL1.MPAM, but ignore this field.
2267 	 */
2268 	if ((hw_val & mpam_mask) == (user_val & mpam_mask))
2269 		user_val &= ~ID_AA64PFR0_EL1_MPAM_MASK;
2270 
2271 	/* Fail the guest's request to disable the AA64 ISA at EL{0,1,2} */
2272 	if (!FIELD_GET(ID_AA64PFR0_EL1_EL0, user_val) ||
2273 	    !FIELD_GET(ID_AA64PFR0_EL1_EL1, user_val) ||
2274 	    (vcpu_has_nv(vcpu) && !FIELD_GET(ID_AA64PFR0_EL1_EL2, user_val)))
2275 		return -EINVAL;
2276 
2277 	return set_id_reg(vcpu, rd, user_val);
2278 }
2279 
2280 static int set_id_aa64pfr1_el1(struct kvm_vcpu *vcpu,
2281 			       const struct sys_reg_desc *rd, u64 user_val)
2282 {
2283 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2284 	u64 mpam_mask = ID_AA64PFR1_EL1_MPAM_frac_MASK;
2285 	u8 mte = SYS_FIELD_GET(ID_AA64PFR1_EL1, MTE, hw_val);
2286 	u8 user_mte_frac = SYS_FIELD_GET(ID_AA64PFR1_EL1, MTE_frac, user_val);
2287 	u8 hw_mte_frac = SYS_FIELD_GET(ID_AA64PFR1_EL1, MTE_frac, hw_val);
2288 
2289 	/* See set_id_aa64pfr0_el1 for comment about MPAM */
2290 	if ((hw_val & mpam_mask) == (user_val & mpam_mask))
2291 		user_val &= ~ID_AA64PFR1_EL1_MPAM_frac_MASK;
2292 
2293 	/*
2294 	 * Previously MTE_frac was hidden from guest. However, if the
2295 	 * hardware supports MTE2 but not MTE_ASYM_FAULT then a value
2296 	 * of 0 for this field indicates that the hardware supports
2297 	 * MTE_ASYNC. Whereas, 0xf indicates MTE_ASYNC is not supported.
2298 	 *
2299 	 * As KVM must accept values from KVM provided by user-space,
2300 	 * when ID_AA64PFR1_EL1.MTE is 2 allow user-space to set
2301 	 * ID_AA64PFR1_EL1.MTE_frac to 0. However, ignore it to avoid
2302 	 * incorrectly claiming hardware support for MTE_ASYNC in the
2303 	 * guest.
2304 	 */
2305 
2306 	if (mte == ID_AA64PFR1_EL1_MTE_MTE2 &&
2307 	    hw_mte_frac == ID_AA64PFR1_EL1_MTE_frac_NI &&
2308 	    user_mte_frac == ID_AA64PFR1_EL1_MTE_frac_ASYNC) {
2309 		user_val &= ~ID_AA64PFR1_EL1_MTE_frac_MASK;
2310 		user_val |= hw_val & ID_AA64PFR1_EL1_MTE_frac_MASK;
2311 	}
2312 
2313 	return set_id_reg(vcpu, rd, user_val);
2314 }
2315 
2316 static int set_id_aa64pfr2_el1(struct kvm_vcpu *vcpu,
2317 			       const struct sys_reg_desc *rd, u64 user_val)
2318 {
2319 	return set_id_reg(vcpu, rd, user_val);
2320 }
2321 
2322 /*
2323  * Allow userspace to de-feature a stage-2 translation granule but prevent it
2324  * from claiming the impossible.
2325  */
2326 #define tgran2_val_allowed(tg, safe, user)			\
2327 ({								\
2328 	u8 __s = SYS_FIELD_GET(ID_AA64MMFR0_EL1, tg, safe);	\
2329 	u8 __u = SYS_FIELD_GET(ID_AA64MMFR0_EL1, tg, user);	\
2330 								\
2331 	__s == __u || __u == ID_AA64MMFR0_EL1_##tg##_NI;	\
2332 })
2333 
2334 static int set_id_aa64mmfr0_el1(struct kvm_vcpu *vcpu,
2335 				const struct sys_reg_desc *rd, u64 user_val)
2336 {
2337 	u64 sanitized_val = kvm_read_sanitised_id_reg(vcpu, rd);
2338 
2339 	if (!vcpu_has_nv(vcpu))
2340 		return set_id_reg(vcpu, rd, user_val);
2341 
2342 	if (!tgran2_val_allowed(TGRAN4_2, sanitized_val, user_val) ||
2343 	    !tgran2_val_allowed(TGRAN16_2, sanitized_val, user_val) ||
2344 	    !tgran2_val_allowed(TGRAN64_2, sanitized_val, user_val))
2345 		return -EINVAL;
2346 
2347 	return set_id_reg(vcpu, rd, user_val);
2348 }
2349 
2350 static int set_id_aa64mmfr2_el1(struct kvm_vcpu *vcpu,
2351 				const struct sys_reg_desc *rd, u64 user_val)
2352 {
2353 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2354 	u64 nv_mask = ID_AA64MMFR2_EL1_NV_MASK;
2355 
2356 	/*
2357 	 * We made the mistake to expose the now deprecated NV field,
2358 	 * so allow userspace to write it, but silently ignore it.
2359 	 */
2360 	if ((hw_val & nv_mask) == (user_val & nv_mask))
2361 		user_val &= ~nv_mask;
2362 
2363 	return set_id_reg(vcpu, rd, user_val);
2364 }
2365 
2366 static int set_ctr_el0(struct kvm_vcpu *vcpu,
2367 		       const struct sys_reg_desc *rd, u64 user_val)
2368 {
2369 	u8 user_L1Ip = SYS_FIELD_GET(CTR_EL0, L1Ip, user_val);
2370 
2371 	/*
2372 	 * Both AIVIVT (0b01) and VPIPT (0b00) are documented as reserved.
2373 	 * Hence only allow to set VIPT(0b10) or PIPT(0b11) for L1Ip based
2374 	 * on what hardware reports.
2375 	 *
2376 	 * Using a VIPT software model on PIPT will lead to over invalidation,
2377 	 * but still correct. Hence, we can allow downgrading PIPT to VIPT,
2378 	 * but not the other way around. This is handled via arm64_ftr_safe_value()
2379 	 * as CTR_EL0 ftr_bits has L1Ip field with type FTR_EXACT and safe value
2380 	 * set as VIPT.
2381 	 */
2382 	switch (user_L1Ip) {
2383 	case CTR_EL0_L1Ip_RESERVED_VPIPT:
2384 	case CTR_EL0_L1Ip_RESERVED_AIVIVT:
2385 		return -EINVAL;
2386 	case CTR_EL0_L1Ip_VIPT:
2387 	case CTR_EL0_L1Ip_PIPT:
2388 		return set_id_reg(vcpu, rd, user_val);
2389 	default:
2390 		return -ENOENT;
2391 	}
2392 }
2393 
2394 /*
2395  * cpufeature ID register user accessors
2396  *
2397  * For now, these registers are immutable for userspace, so no values
2398  * are stored, and for set_id_reg() we don't allow the effective value
2399  * to be changed.
2400  */
2401 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2402 		      u64 *val)
2403 {
2404 	/*
2405 	 * Avoid locking if the VM has already started, as the ID registers are
2406 	 * guaranteed to be invariant at that point.
2407 	 */
2408 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
2409 		*val = read_id_reg(vcpu, rd);
2410 		return 0;
2411 	}
2412 
2413 	mutex_lock(&vcpu->kvm->arch.config_lock);
2414 	*val = read_id_reg(vcpu, rd);
2415 	mutex_unlock(&vcpu->kvm->arch.config_lock);
2416 
2417 	return 0;
2418 }
2419 
2420 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2421 		      u64 val)
2422 {
2423 	u32 id = reg_to_encoding(rd);
2424 	int ret;
2425 
2426 	mutex_lock(&vcpu->kvm->arch.config_lock);
2427 
2428 	/*
2429 	 * Once the VM has started the ID registers are immutable. Reject any
2430 	 * write that does not match the final register value.
2431 	 */
2432 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
2433 		if (val != read_id_reg(vcpu, rd))
2434 			ret = -EBUSY;
2435 		else
2436 			ret = 0;
2437 
2438 		mutex_unlock(&vcpu->kvm->arch.config_lock);
2439 		return ret;
2440 	}
2441 
2442 	ret = arm64_check_features(vcpu, rd, val);
2443 	if (!ret)
2444 		kvm_set_vm_id_reg(vcpu->kvm, id, val);
2445 
2446 	mutex_unlock(&vcpu->kvm->arch.config_lock);
2447 
2448 	/*
2449 	 * arm64_check_features() returns -E2BIG to indicate the register's
2450 	 * feature set is a superset of the maximally-allowed register value.
2451 	 * While it would be nice to precisely describe this to userspace, the
2452 	 * existing UAPI for KVM_SET_ONE_REG has it that invalid register
2453 	 * writes return -EINVAL.
2454 	 */
2455 	if (ret == -E2BIG)
2456 		ret = -EINVAL;
2457 	return ret;
2458 }
2459 
2460 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val)
2461 {
2462 	u64 *p = __vm_id_reg(&kvm->arch, reg);
2463 
2464 	lockdep_assert_held(&kvm->arch.config_lock);
2465 
2466 	if (KVM_BUG_ON(kvm_vm_has_ran_once(kvm) || !p, kvm))
2467 		return;
2468 
2469 	*p = val;
2470 }
2471 
2472 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2473 		       u64 *val)
2474 {
2475 	*val = 0;
2476 	return 0;
2477 }
2478 
2479 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2480 		      u64 val)
2481 {
2482 	return 0;
2483 }
2484 
2485 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2486 		       const struct sys_reg_desc *r)
2487 {
2488 	if (p->is_write)
2489 		return write_to_read_only(vcpu, p, r);
2490 
2491 	p->regval = kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0);
2492 	return true;
2493 }
2494 
2495 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2496 			 const struct sys_reg_desc *r)
2497 {
2498 	if (p->is_write)
2499 		return write_to_read_only(vcpu, p, r);
2500 
2501 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
2502 	return true;
2503 }
2504 
2505 /*
2506  * Fabricate a CLIDR_EL1 value instead of using the real value, which can vary
2507  * by the physical CPU which the vcpu currently resides in.
2508  */
2509 static u64 reset_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2510 {
2511 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2512 	u64 clidr;
2513 	u8 loc;
2514 
2515 	if ((ctr_el0 & CTR_EL0_IDC)) {
2516 		/*
2517 		 * Data cache clean to the PoU is not required so LoUU and LoUIS
2518 		 * will not be set and a unified cache, which will be marked as
2519 		 * LoC, will be added.
2520 		 *
2521 		 * If not DIC, let the unified cache L2 so that an instruction
2522 		 * cache can be added as L1 later.
2523 		 */
2524 		loc = (ctr_el0 & CTR_EL0_DIC) ? 1 : 2;
2525 		clidr = CACHE_TYPE_UNIFIED << CLIDR_CTYPE_SHIFT(loc);
2526 	} else {
2527 		/*
2528 		 * Data cache clean to the PoU is required so let L1 have a data
2529 		 * cache and mark it as LoUU and LoUIS. As L1 has a data cache,
2530 		 * it can be marked as LoC too.
2531 		 */
2532 		loc = 1;
2533 		clidr = 1 << CLIDR_LOUU_SHIFT;
2534 		clidr |= 1 << CLIDR_LOUIS_SHIFT;
2535 		clidr |= CACHE_TYPE_DATA << CLIDR_CTYPE_SHIFT(1);
2536 	}
2537 
2538 	/*
2539 	 * Instruction cache invalidation to the PoU is required so let L1 have
2540 	 * an instruction cache. If L1 already has a data cache, it will be
2541 	 * CACHE_TYPE_SEPARATE.
2542 	 */
2543 	if (!(ctr_el0 & CTR_EL0_DIC))
2544 		clidr |= CACHE_TYPE_INST << CLIDR_CTYPE_SHIFT(1);
2545 
2546 	clidr |= loc << CLIDR_LOC_SHIFT;
2547 
2548 	/*
2549 	 * Add tag cache unified to data cache. Allocation tags and data are
2550 	 * unified in a cache line so that it looks valid even if there is only
2551 	 * one cache line.
2552 	 */
2553 	if (kvm_has_mte(vcpu->kvm))
2554 		clidr |= 2ULL << CLIDR_TTYPE_SHIFT(loc);
2555 
2556 	__vcpu_assign_sys_reg(vcpu, r->reg, clidr);
2557 
2558 	return __vcpu_sys_reg(vcpu, r->reg);
2559 }
2560 
2561 static int set_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2562 		      u64 val)
2563 {
2564 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2565 	u64 idc = !CLIDR_LOC(val) || (!CLIDR_LOUIS(val) && !CLIDR_LOUU(val));
2566 
2567 	if ((val & CLIDR_EL1_RES0) || (!(ctr_el0 & CTR_EL0_IDC) && idc))
2568 		return -EINVAL;
2569 
2570 	__vcpu_assign_sys_reg(vcpu, rd->reg, val);
2571 
2572 	return 0;
2573 }
2574 
2575 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2576 			  const struct sys_reg_desc *r)
2577 {
2578 	int reg = r->reg;
2579 
2580 	if (p->is_write)
2581 		vcpu_write_sys_reg(vcpu, p->regval, reg);
2582 	else
2583 		p->regval = vcpu_read_sys_reg(vcpu, reg);
2584 	return true;
2585 }
2586 
2587 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2588 			  const struct sys_reg_desc *r)
2589 {
2590 	u32 csselr;
2591 
2592 	if (p->is_write)
2593 		return write_to_read_only(vcpu, p, r);
2594 
2595 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
2596 	csselr &= CSSELR_EL1_Level | CSSELR_EL1_InD;
2597 	if (csselr < CSSELR_MAX)
2598 		p->regval = get_ccsidr(vcpu, csselr);
2599 
2600 	return true;
2601 }
2602 
2603 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
2604 				   const struct sys_reg_desc *rd)
2605 {
2606 	if (kvm_has_mte(vcpu->kvm))
2607 		return 0;
2608 
2609 	return REG_HIDDEN;
2610 }
2611 
2612 #define MTE_REG(name) {				\
2613 	SYS_DESC(SYS_##name),			\
2614 	.access = undef_access,			\
2615 	.reset = reset_unknown,			\
2616 	.reg = name,				\
2617 	.visibility = mte_visibility,		\
2618 }
2619 
2620 static unsigned int el2_visibility(const struct kvm_vcpu *vcpu,
2621 				   const struct sys_reg_desc *rd)
2622 {
2623 	if (vcpu_has_nv(vcpu))
2624 		return 0;
2625 
2626 	return REG_HIDDEN;
2627 }
2628 
2629 static bool bad_vncr_trap(struct kvm_vcpu *vcpu,
2630 			  struct sys_reg_params *p,
2631 			  const struct sys_reg_desc *r)
2632 {
2633 	/*
2634 	 * We really shouldn't be here, and this is likely the result
2635 	 * of a misconfigured trap, as this register should target the
2636 	 * VNCR page, and nothing else.
2637 	 */
2638 	return bad_trap(vcpu, p, r,
2639 			"trap of VNCR-backed register");
2640 }
2641 
2642 static bool bad_redir_trap(struct kvm_vcpu *vcpu,
2643 			   struct sys_reg_params *p,
2644 			   const struct sys_reg_desc *r)
2645 {
2646 	/*
2647 	 * We really shouldn't be here, and this is likely the result
2648 	 * of a misconfigured trap, as this register should target the
2649 	 * corresponding EL1, and nothing else.
2650 	 */
2651 	return bad_trap(vcpu, p, r,
2652 			"trap of EL2 register redirected to EL1");
2653 }
2654 
2655 #define SYS_REG_USER_FILTER(name, acc, rst, v, gu, su, filter) { \
2656 	SYS_DESC(SYS_##name),			\
2657 	.access = acc,				\
2658 	.reset = rst,				\
2659 	.reg = name,				\
2660 	.get_user = gu,				\
2661 	.set_user = su,				\
2662 	.visibility = filter,			\
2663 	.val = v,				\
2664 }
2665 
2666 #define EL2_REG_FILTERED(name, acc, rst, v, filter)	\
2667 	SYS_REG_USER_FILTER(name, acc, rst, v, NULL, NULL, filter)
2668 
2669 #define EL2_REG(name, acc, rst, v)			\
2670 	EL2_REG_FILTERED(name, acc, rst, v, el2_visibility)
2671 
2672 #define EL2_REG_VNCR(name, rst, v)	EL2_REG(name, bad_vncr_trap, rst, v)
2673 #define EL2_REG_VNCR_FILT(name, vis)			\
2674 	EL2_REG_FILTERED(name, bad_vncr_trap, reset_val, 0, vis)
2675 #define EL2_REG_VNCR_GICv3(name)			\
2676 	EL2_REG_VNCR_FILT(name, hidden_visibility)
2677 #define EL2_REG_REDIR(name, rst, v)	EL2_REG(name, bad_redir_trap, rst, v)
2678 
2679 #define TIMER_REG(name, vis)					   \
2680 	SYS_REG_USER_FILTER(name, access_arch_timer, reset_val, 0, \
2681 			    arch_timer_get_user, arch_timer_set_user, vis)
2682 
2683 /*
2684  * Since reset() callback and field val are not used for idregs, they will be
2685  * used for specific purposes for idregs.
2686  * The reset() would return KVM sanitised register value. The value would be the
2687  * same as the host kernel sanitised value if there is no KVM sanitisation.
2688  * The val would be used as a mask indicating writable fields for the idreg.
2689  * Only bits with 1 are writable from userspace. This mask might not be
2690  * necessary in the future whenever all ID registers are enabled as writable
2691  * from userspace.
2692  */
2693 
2694 #define ID_DESC_DEFAULT_CALLBACKS		\
2695 	.access	= access_id_reg,		\
2696 	.get_user = get_id_reg,			\
2697 	.set_user = set_id_reg,			\
2698 	.visibility = id_visibility,		\
2699 	.reset = kvm_read_sanitised_id_reg
2700 
2701 #define ID_DESC(name)				\
2702 	SYS_DESC(SYS_##name),			\
2703 	ID_DESC_DEFAULT_CALLBACKS
2704 
2705 /* sys_reg_desc initialiser for known cpufeature ID registers */
2706 #define ID_SANITISED(name) {			\
2707 	ID_DESC(name),				\
2708 	.val = 0,				\
2709 }
2710 
2711 /* sys_reg_desc initialiser for writable ID registers */
2712 #define ID_WRITABLE(name, mask) {		\
2713 	ID_DESC(name),				\
2714 	.val = mask,				\
2715 }
2716 
2717 /*
2718  * 32bit ID regs are fully writable when the guest is 32bit
2719  * capable. Nothing in the KVM code should rely on 32bit features
2720  * anyway, only 64bit, so let the VMM do its worse.
2721  */
2722 #define AA32_ID_WRITABLE(name) {		\
2723 	ID_DESC(name),				\
2724 	.visibility = aa32_id_visibility,	\
2725 	.val = GENMASK(31, 0),			\
2726 }
2727 
2728 /* sys_reg_desc initialiser for cpufeature ID registers that need filtering */
2729 #define ID_FILTERED(sysreg, name, mask) {	\
2730 	ID_DESC(sysreg),				\
2731 	.set_user = set_##name,				\
2732 	.val = (mask),					\
2733 }
2734 
2735 /*
2736  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
2737  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
2738  * (1 <= crm < 8, 0 <= Op2 < 8).
2739  */
2740 #define ID_UNALLOCATED(crm, op2) {			\
2741 	.name = "S3_0_0_" #crm "_" #op2,		\
2742 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
2743 	ID_DESC_DEFAULT_CALLBACKS,			\
2744 	.visibility = raz_visibility,			\
2745 	.val = 0,					\
2746 }
2747 
2748 /*
2749  * sys_reg_desc initialiser for known ID registers that we hide from guests.
2750  * For now, these are exposed just like unallocated ID regs: they appear
2751  * RAZ for the guest.
2752  */
2753 #define ID_HIDDEN(name) {			\
2754 	ID_DESC(name),				\
2755 	.visibility = raz_visibility,		\
2756 	.val = 0,				\
2757 }
2758 
2759 static bool access_sp_el1(struct kvm_vcpu *vcpu,
2760 			  struct sys_reg_params *p,
2761 			  const struct sys_reg_desc *r)
2762 {
2763 	if (p->is_write)
2764 		__vcpu_assign_sys_reg(vcpu, SP_EL1, p->regval);
2765 	else
2766 		p->regval = __vcpu_sys_reg(vcpu, SP_EL1);
2767 
2768 	return true;
2769 }
2770 
2771 static bool access_elr(struct kvm_vcpu *vcpu,
2772 		       struct sys_reg_params *p,
2773 		       const struct sys_reg_desc *r)
2774 {
2775 	if (p->is_write)
2776 		vcpu_write_sys_reg(vcpu, p->regval, ELR_EL1);
2777 	else
2778 		p->regval = vcpu_read_sys_reg(vcpu, ELR_EL1);
2779 
2780 	return true;
2781 }
2782 
2783 static bool access_spsr(struct kvm_vcpu *vcpu,
2784 			struct sys_reg_params *p,
2785 			const struct sys_reg_desc *r)
2786 {
2787 	if (p->is_write)
2788 		__vcpu_assign_sys_reg(vcpu, SPSR_EL1, p->regval);
2789 	else
2790 		p->regval = __vcpu_sys_reg(vcpu, SPSR_EL1);
2791 
2792 	return true;
2793 }
2794 
2795 static bool access_cntkctl_el12(struct kvm_vcpu *vcpu,
2796 				struct sys_reg_params *p,
2797 				const struct sys_reg_desc *r)
2798 {
2799 	if (p->is_write)
2800 		__vcpu_assign_sys_reg(vcpu, CNTKCTL_EL1, p->regval);
2801 	else
2802 		p->regval = __vcpu_sys_reg(vcpu, CNTKCTL_EL1);
2803 
2804 	return true;
2805 }
2806 
2807 static u64 reset_hcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2808 {
2809 	u64 val = r->val;
2810 
2811 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
2812 		val |= HCR_E2H;
2813 
2814 	__vcpu_assign_sys_reg(vcpu, r->reg, val);
2815 
2816 	return __vcpu_sys_reg(vcpu, r->reg);
2817 }
2818 
2819 static unsigned int __el2_visibility(const struct kvm_vcpu *vcpu,
2820 				     const struct sys_reg_desc *rd,
2821 				     unsigned int (*fn)(const struct kvm_vcpu *,
2822 							const struct sys_reg_desc *))
2823 {
2824 	return el2_visibility(vcpu, rd) ?: fn(vcpu, rd);
2825 }
2826 
2827 static unsigned int sve_el2_visibility(const struct kvm_vcpu *vcpu,
2828 				       const struct sys_reg_desc *rd)
2829 {
2830 	return __el2_visibility(vcpu, rd, sve_visibility);
2831 }
2832 
2833 static unsigned int vncr_el2_visibility(const struct kvm_vcpu *vcpu,
2834 					const struct sys_reg_desc *rd)
2835 {
2836 	if (el2_visibility(vcpu, rd) == 0 &&
2837 	    kvm_has_feat(vcpu->kvm, ID_AA64MMFR4_EL1, NV_frac, NV2_ONLY))
2838 		return 0;
2839 
2840 	return REG_HIDDEN;
2841 }
2842 
2843 static unsigned int sctlr2_visibility(const struct kvm_vcpu *vcpu,
2844 				      const struct sys_reg_desc *rd)
2845 {
2846 	if (kvm_has_sctlr2(vcpu->kvm))
2847 		return 0;
2848 
2849 	return REG_HIDDEN;
2850 }
2851 
2852 static unsigned int sctlr2_el2_visibility(const struct kvm_vcpu *vcpu,
2853 					  const struct sys_reg_desc *rd)
2854 {
2855 	return __el2_visibility(vcpu, rd, sctlr2_visibility);
2856 }
2857 
2858 static bool access_zcr_el2(struct kvm_vcpu *vcpu,
2859 			   struct sys_reg_params *p,
2860 			   const struct sys_reg_desc *r)
2861 {
2862 	if (guest_hyp_sve_traps_enabled(vcpu)) {
2863 		kvm_inject_nested_sve_trap(vcpu);
2864 		return false;
2865 	}
2866 
2867 	if (!p->is_write)
2868 		p->regval = __vcpu_sys_reg(vcpu, ZCR_EL2);
2869 	else
2870 		__vcpu_assign_sys_reg(vcpu, ZCR_EL2, p->regval);
2871 
2872 	return true;
2873 }
2874 
2875 static bool access_gic_vtr(struct kvm_vcpu *vcpu,
2876 			   struct sys_reg_params *p,
2877 			   const struct sys_reg_desc *r)
2878 {
2879 	if (p->is_write)
2880 		return write_to_read_only(vcpu, p, r);
2881 
2882 	p->regval = kvm_get_guest_vtr_el2();
2883 
2884 	return true;
2885 }
2886 
2887 static bool access_gic_misr(struct kvm_vcpu *vcpu,
2888 			    struct sys_reg_params *p,
2889 			    const struct sys_reg_desc *r)
2890 {
2891 	if (p->is_write)
2892 		return write_to_read_only(vcpu, p, r);
2893 
2894 	p->regval = vgic_v3_get_misr(vcpu);
2895 
2896 	return true;
2897 }
2898 
2899 static bool access_gic_eisr(struct kvm_vcpu *vcpu,
2900 			    struct sys_reg_params *p,
2901 			    const struct sys_reg_desc *r)
2902 {
2903 	if (p->is_write)
2904 		return write_to_read_only(vcpu, p, r);
2905 
2906 	p->regval = vgic_v3_get_eisr(vcpu);
2907 
2908 	return true;
2909 }
2910 
2911 static bool access_gic_elrsr(struct kvm_vcpu *vcpu,
2912 			     struct sys_reg_params *p,
2913 			     const struct sys_reg_desc *r)
2914 {
2915 	if (p->is_write)
2916 		return write_to_read_only(vcpu, p, r);
2917 
2918 	p->regval = vgic_v3_get_elrsr(vcpu);
2919 
2920 	return true;
2921 }
2922 
2923 static unsigned int s1poe_visibility(const struct kvm_vcpu *vcpu,
2924 				     const struct sys_reg_desc *rd)
2925 {
2926 	if (kvm_has_s1poe(vcpu->kvm))
2927 		return 0;
2928 
2929 	return REG_HIDDEN;
2930 }
2931 
2932 static unsigned int s1poe_el2_visibility(const struct kvm_vcpu *vcpu,
2933 					 const struct sys_reg_desc *rd)
2934 {
2935 	return __el2_visibility(vcpu, rd, s1poe_visibility);
2936 }
2937 
2938 static unsigned int tcr2_visibility(const struct kvm_vcpu *vcpu,
2939 				    const struct sys_reg_desc *rd)
2940 {
2941 	if (kvm_has_tcr2(vcpu->kvm))
2942 		return 0;
2943 
2944 	return REG_HIDDEN;
2945 }
2946 
2947 static unsigned int tcr2_el2_visibility(const struct kvm_vcpu *vcpu,
2948 				    const struct sys_reg_desc *rd)
2949 {
2950 	return __el2_visibility(vcpu, rd, tcr2_visibility);
2951 }
2952 
2953 static unsigned int fgt2_visibility(const struct kvm_vcpu *vcpu,
2954 				    const struct sys_reg_desc *rd)
2955 {
2956 	if (el2_visibility(vcpu, rd) == 0 &&
2957 	    kvm_has_feat(vcpu->kvm, ID_AA64MMFR0_EL1, FGT, FGT2))
2958 		return 0;
2959 
2960 	return REG_HIDDEN;
2961 }
2962 
2963 static unsigned int fgt_visibility(const struct kvm_vcpu *vcpu,
2964 				   const struct sys_reg_desc *rd)
2965 {
2966 	if (el2_visibility(vcpu, rd) == 0 &&
2967 	    kvm_has_feat(vcpu->kvm, ID_AA64MMFR0_EL1, FGT, IMP))
2968 		return 0;
2969 
2970 	return REG_HIDDEN;
2971 }
2972 
2973 static unsigned int s1pie_visibility(const struct kvm_vcpu *vcpu,
2974 				     const struct sys_reg_desc *rd)
2975 {
2976 	if (kvm_has_s1pie(vcpu->kvm))
2977 		return 0;
2978 
2979 	return REG_HIDDEN;
2980 }
2981 
2982 static unsigned int s1pie_el2_visibility(const struct kvm_vcpu *vcpu,
2983 					 const struct sys_reg_desc *rd)
2984 {
2985 	return __el2_visibility(vcpu, rd, s1pie_visibility);
2986 }
2987 
2988 static unsigned int cnthv_visibility(const struct kvm_vcpu *vcpu,
2989 				     const struct sys_reg_desc *rd)
2990 {
2991 	if (vcpu_has_nv(vcpu) &&
2992 	    !vcpu_has_feature(vcpu, KVM_ARM_VCPU_HAS_EL2_E2H0))
2993 		return 0;
2994 
2995 	return REG_HIDDEN;
2996 }
2997 
2998 static bool access_mdcr(struct kvm_vcpu *vcpu,
2999 			struct sys_reg_params *p,
3000 			const struct sys_reg_desc *r)
3001 {
3002 	u64 hpmn, val, old = __vcpu_sys_reg(vcpu, MDCR_EL2);
3003 
3004 	if (!p->is_write) {
3005 		p->regval = old;
3006 		return true;
3007 	}
3008 
3009 	val = p->regval;
3010 	hpmn = FIELD_GET(MDCR_EL2_HPMN, val);
3011 
3012 	/*
3013 	 * If HPMN is out of bounds, limit it to what we actually
3014 	 * support. This matches the UNKNOWN definition of the field
3015 	 * in that case, and keeps the emulation simple. Sort of.
3016 	 */
3017 	if (hpmn > vcpu->kvm->arch.nr_pmu_counters) {
3018 		hpmn = vcpu->kvm->arch.nr_pmu_counters;
3019 		u64p_replace_bits(&val, hpmn, MDCR_EL2_HPMN);
3020 	}
3021 
3022 	__vcpu_assign_sys_reg(vcpu, MDCR_EL2, val);
3023 
3024 	/*
3025 	 * Request a reload of the PMU to enable/disable the counters
3026 	 * affected by HPME.
3027 	 */
3028 	if ((old ^ val) & MDCR_EL2_HPME)
3029 		kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
3030 
3031 	return true;
3032 }
3033 
3034 static bool access_ras(struct kvm_vcpu *vcpu,
3035 		       struct sys_reg_params *p,
3036 		       const struct sys_reg_desc *r)
3037 {
3038 	struct kvm *kvm = vcpu->kvm;
3039 
3040 	switch(reg_to_encoding(r)) {
3041 	case SYS_ERXPFGCDN_EL1:
3042 	case SYS_ERXPFGCTL_EL1:
3043 	case SYS_ERXPFGF_EL1:
3044 	case SYS_ERXMISC2_EL1:
3045 	case SYS_ERXMISC3_EL1:
3046 		if (!(kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, V1P1) ||
3047 		      (kvm_has_feat_enum(kvm, ID_AA64PFR0_EL1, RAS, IMP) &&
3048 		       kvm_has_feat(kvm, ID_AA64PFR1_EL1, RAS_frac, RASv1p1)))) {
3049 			kvm_inject_undefined(vcpu);
3050 			return false;
3051 		}
3052 		break;
3053 	default:
3054 		if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, IMP)) {
3055 			kvm_inject_undefined(vcpu);
3056 			return false;
3057 		}
3058 	}
3059 
3060 	return trap_raz_wi(vcpu, p, r);
3061 }
3062 
3063 /*
3064  * For historical (ahem ABI) reasons, KVM treated MIDR_EL1, REVIDR_EL1, and
3065  * AIDR_EL1 as "invariant" registers, meaning userspace cannot change them.
3066  * The values made visible to userspace were the register values of the boot
3067  * CPU.
3068  *
3069  * At the same time, reads from these registers at EL1 previously were not
3070  * trapped, allowing the guest to read the actual hardware value. On big-little
3071  * machines, this means the VM can see different values depending on where a
3072  * given vCPU got scheduled.
3073  *
3074  * These registers are now trapped as collateral damage from SME, and what
3075  * follows attempts to give a user / guest view consistent with the existing
3076  * ABI.
3077  */
3078 static bool access_imp_id_reg(struct kvm_vcpu *vcpu,
3079 			      struct sys_reg_params *p,
3080 			      const struct sys_reg_desc *r)
3081 {
3082 	if (p->is_write)
3083 		return write_to_read_only(vcpu, p, r);
3084 
3085 	/*
3086 	 * Return the VM-scoped implementation ID register values if userspace
3087 	 * has made them writable.
3088 	 */
3089 	if (test_bit(KVM_ARCH_FLAG_WRITABLE_IMP_ID_REGS, &vcpu->kvm->arch.flags))
3090 		return access_id_reg(vcpu, p, r);
3091 
3092 	/*
3093 	 * Otherwise, fall back to the old behavior of returning the value of
3094 	 * the current CPU.
3095 	 */
3096 	switch (reg_to_encoding(r)) {
3097 	case SYS_REVIDR_EL1:
3098 		p->regval = read_sysreg(revidr_el1);
3099 		break;
3100 	case SYS_AIDR_EL1:
3101 		p->regval = read_sysreg(aidr_el1);
3102 		break;
3103 	default:
3104 		WARN_ON_ONCE(1);
3105 	}
3106 
3107 	return true;
3108 }
3109 
3110 static u64 __ro_after_init boot_cpu_midr_val;
3111 static u64 __ro_after_init boot_cpu_revidr_val;
3112 static u64 __ro_after_init boot_cpu_aidr_val;
3113 
3114 static void init_imp_id_regs(void)
3115 {
3116 	boot_cpu_midr_val = read_sysreg(midr_el1);
3117 	boot_cpu_revidr_val = read_sysreg(revidr_el1);
3118 	boot_cpu_aidr_val = read_sysreg(aidr_el1);
3119 }
3120 
3121 static u64 reset_imp_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
3122 {
3123 	switch (reg_to_encoding(r)) {
3124 	case SYS_MIDR_EL1:
3125 		return boot_cpu_midr_val;
3126 	case SYS_REVIDR_EL1:
3127 		return boot_cpu_revidr_val;
3128 	case SYS_AIDR_EL1:
3129 		return boot_cpu_aidr_val;
3130 	default:
3131 		KVM_BUG_ON(1, vcpu->kvm);
3132 		return 0;
3133 	}
3134 }
3135 
3136 static int set_imp_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
3137 			  u64 val)
3138 {
3139 	struct kvm *kvm = vcpu->kvm;
3140 	u64 expected;
3141 
3142 	guard(mutex)(&kvm->arch.config_lock);
3143 
3144 	expected = read_id_reg(vcpu, r);
3145 	if (expected == val)
3146 		return 0;
3147 
3148 	if (!test_bit(KVM_ARCH_FLAG_WRITABLE_IMP_ID_REGS, &kvm->arch.flags))
3149 		return -EINVAL;
3150 
3151 	/*
3152 	 * Once the VM has started the ID registers are immutable. Reject the
3153 	 * write if userspace tries to change it.
3154 	 */
3155 	if (kvm_vm_has_ran_once(kvm))
3156 		return -EBUSY;
3157 
3158 	/*
3159 	 * Any value is allowed for the implementation ID registers so long as
3160 	 * it is within the writable mask.
3161 	 */
3162 	if ((val & r->val) != val)
3163 		return -EINVAL;
3164 
3165 	kvm_set_vm_id_reg(kvm, reg_to_encoding(r), val);
3166 	return 0;
3167 }
3168 
3169 #define IMPLEMENTATION_ID(reg, mask) {			\
3170 	SYS_DESC(SYS_##reg),				\
3171 	.access = access_imp_id_reg,			\
3172 	.get_user = get_id_reg,				\
3173 	.set_user = set_imp_id_reg,			\
3174 	.reset = reset_imp_id_reg,			\
3175 	.val = mask,					\
3176 	}
3177 
3178 static u64 reset_mdcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
3179 {
3180 	__vcpu_assign_sys_reg(vcpu, r->reg, vcpu->kvm->arch.nr_pmu_counters);
3181 	return vcpu->kvm->arch.nr_pmu_counters;
3182 }
3183 
3184 /*
3185  * Architected system registers.
3186  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
3187  *
3188  * Debug handling: We do trap most, if not all debug related system
3189  * registers. The implementation is good enough to ensure that a guest
3190  * can use these with minimal performance degradation. The drawback is
3191  * that we don't implement any of the external debug architecture.
3192  * This should be revisited if we ever encounter a more demanding
3193  * guest...
3194  */
3195 static const struct sys_reg_desc sys_reg_descs[] = {
3196 	DBG_BCR_BVR_WCR_WVR_EL1(0),
3197 	DBG_BCR_BVR_WCR_WVR_EL1(1),
3198 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
3199 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
3200 	DBG_BCR_BVR_WCR_WVR_EL1(2),
3201 	DBG_BCR_BVR_WCR_WVR_EL1(3),
3202 	DBG_BCR_BVR_WCR_WVR_EL1(4),
3203 	DBG_BCR_BVR_WCR_WVR_EL1(5),
3204 	DBG_BCR_BVR_WCR_WVR_EL1(6),
3205 	DBG_BCR_BVR_WCR_WVR_EL1(7),
3206 	DBG_BCR_BVR_WCR_WVR_EL1(8),
3207 	DBG_BCR_BVR_WCR_WVR_EL1(9),
3208 	DBG_BCR_BVR_WCR_WVR_EL1(10),
3209 	DBG_BCR_BVR_WCR_WVR_EL1(11),
3210 	DBG_BCR_BVR_WCR_WVR_EL1(12),
3211 	DBG_BCR_BVR_WCR_WVR_EL1(13),
3212 	DBG_BCR_BVR_WCR_WVR_EL1(14),
3213 	DBG_BCR_BVR_WCR_WVR_EL1(15),
3214 
3215 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
3216 	{ SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
3217 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
3218 		OSLSR_EL1_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
3219 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
3220 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
3221 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
3222 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
3223 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
3224 
3225 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
3226 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
3227 	// DBGDTR[TR]X_EL0 share the same encoding
3228 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
3229 
3230 	{ SYS_DESC(SYS_DBGVCR32_EL2), undef_access, reset_val, DBGVCR32_EL2, 0 },
3231 
3232 	IMPLEMENTATION_ID(MIDR_EL1, GENMASK_ULL(31, 0)),
3233 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
3234 	IMPLEMENTATION_ID(REVIDR_EL1, GENMASK_ULL(63, 0)),
3235 
3236 	/*
3237 	 * ID regs: all ID_SANITISED() entries here must have corresponding
3238 	 * entries in arm64_ftr_regs[].
3239 	 */
3240 
3241 	/* AArch64 mappings of the AArch32 ID registers */
3242 	/* CRm=1 */
3243 	AA32_ID_WRITABLE(ID_PFR0_EL1),
3244 	AA32_ID_WRITABLE(ID_PFR1_EL1),
3245 	{ SYS_DESC(SYS_ID_DFR0_EL1),
3246 	  .access = access_id_reg,
3247 	  .get_user = get_id_reg,
3248 	  .set_user = set_id_dfr0_el1,
3249 	  .visibility = aa32_id_visibility,
3250 	  .reset = read_sanitised_id_dfr0_el1,
3251 	  .val = GENMASK(31, 0) },
3252 	ID_HIDDEN(ID_AFR0_EL1),
3253 	AA32_ID_WRITABLE(ID_MMFR0_EL1),
3254 	AA32_ID_WRITABLE(ID_MMFR1_EL1),
3255 	AA32_ID_WRITABLE(ID_MMFR2_EL1),
3256 	AA32_ID_WRITABLE(ID_MMFR3_EL1),
3257 
3258 	/* CRm=2 */
3259 	AA32_ID_WRITABLE(ID_ISAR0_EL1),
3260 	AA32_ID_WRITABLE(ID_ISAR1_EL1),
3261 	AA32_ID_WRITABLE(ID_ISAR2_EL1),
3262 	AA32_ID_WRITABLE(ID_ISAR3_EL1),
3263 	AA32_ID_WRITABLE(ID_ISAR4_EL1),
3264 	AA32_ID_WRITABLE(ID_ISAR5_EL1),
3265 	AA32_ID_WRITABLE(ID_MMFR4_EL1),
3266 	AA32_ID_WRITABLE(ID_ISAR6_EL1),
3267 
3268 	/* CRm=3 */
3269 	AA32_ID_WRITABLE(MVFR0_EL1),
3270 	AA32_ID_WRITABLE(MVFR1_EL1),
3271 	AA32_ID_WRITABLE(MVFR2_EL1),
3272 	ID_UNALLOCATED(3,3),
3273 	AA32_ID_WRITABLE(ID_PFR2_EL1),
3274 	ID_HIDDEN(ID_DFR1_EL1),
3275 	AA32_ID_WRITABLE(ID_MMFR5_EL1),
3276 	ID_UNALLOCATED(3,7),
3277 
3278 	/* AArch64 ID registers */
3279 	/* CRm=4 */
3280 	ID_FILTERED(ID_AA64PFR0_EL1, id_aa64pfr0_el1,
3281 		    ~(ID_AA64PFR0_EL1_AMU |
3282 		      ID_AA64PFR0_EL1_MPAM |
3283 		      ID_AA64PFR0_EL1_SVE |
3284 		      ID_AA64PFR0_EL1_AdvSIMD |
3285 		      ID_AA64PFR0_EL1_FP)),
3286 	ID_FILTERED(ID_AA64PFR1_EL1, id_aa64pfr1_el1,
3287 				     ~(ID_AA64PFR1_EL1_PFAR |
3288 				       ID_AA64PFR1_EL1_MTEX |
3289 				       ID_AA64PFR1_EL1_THE |
3290 				       ID_AA64PFR1_EL1_GCS |
3291 				       ID_AA64PFR1_EL1_MTE_frac |
3292 				       ID_AA64PFR1_EL1_NMI |
3293 				       ID_AA64PFR1_EL1_RNDR_trap |
3294 				       ID_AA64PFR1_EL1_SME |
3295 				       ID_AA64PFR1_EL1_RES0 |
3296 				       ID_AA64PFR1_EL1_MPAM_frac |
3297 				       ID_AA64PFR1_EL1_MTE)),
3298 	ID_FILTERED(ID_AA64PFR2_EL1, id_aa64pfr2_el1,
3299 		    (ID_AA64PFR2_EL1_FPMR		|
3300 		     ID_AA64PFR2_EL1_MTEFAR		|
3301 		     ID_AA64PFR2_EL1_MTESTOREONLY	|
3302 		     ID_AA64PFR2_EL1_GCIE)),
3303 	ID_UNALLOCATED(4,3),
3304 	ID_WRITABLE(ID_AA64ZFR0_EL1, ~ID_AA64ZFR0_EL1_RES0),
3305 	ID_HIDDEN(ID_AA64SMFR0_EL1),
3306 	ID_UNALLOCATED(4,6),
3307 	ID_WRITABLE(ID_AA64FPFR0_EL1, ~ID_AA64FPFR0_EL1_RES0),
3308 
3309 	/* CRm=5 */
3310 	/*
3311 	 * Prior to FEAT_Debugv8.9, the architecture defines context-aware
3312 	 * breakpoints (CTX_CMPs) as the highest numbered breakpoints (BRPs).
3313 	 * KVM does not trap + emulate the breakpoint registers, and as such
3314 	 * cannot support a layout that misaligns with the underlying hardware.
3315 	 * While it may be possible to describe a subset that aligns with
3316 	 * hardware, just prevent changes to BRPs and CTX_CMPs altogether for
3317 	 * simplicity.
3318 	 *
3319 	 * See DDI0487K.a, section D2.8.3 Breakpoint types and linking
3320 	 * of breakpoints for more details.
3321 	 */
3322 	ID_FILTERED(ID_AA64DFR0_EL1, id_aa64dfr0_el1,
3323 		    ID_AA64DFR0_EL1_DoubleLock_MASK |
3324 		    ID_AA64DFR0_EL1_WRPs_MASK |
3325 		    ID_AA64DFR0_EL1_PMUVer_MASK |
3326 		    ID_AA64DFR0_EL1_DebugVer_MASK),
3327 	ID_SANITISED(ID_AA64DFR1_EL1),
3328 	ID_UNALLOCATED(5,2),
3329 	ID_UNALLOCATED(5,3),
3330 	ID_HIDDEN(ID_AA64AFR0_EL1),
3331 	ID_HIDDEN(ID_AA64AFR1_EL1),
3332 	ID_UNALLOCATED(5,6),
3333 	ID_UNALLOCATED(5,7),
3334 
3335 	/* CRm=6 */
3336 	ID_WRITABLE(ID_AA64ISAR0_EL1, ~ID_AA64ISAR0_EL1_RES0),
3337 	ID_WRITABLE(ID_AA64ISAR1_EL1, ~(ID_AA64ISAR1_EL1_GPI |
3338 					ID_AA64ISAR1_EL1_GPA |
3339 					ID_AA64ISAR1_EL1_API |
3340 					ID_AA64ISAR1_EL1_APA)),
3341 	ID_WRITABLE(ID_AA64ISAR2_EL1, ~(ID_AA64ISAR2_EL1_RES0 |
3342 					ID_AA64ISAR2_EL1_APA3 |
3343 					ID_AA64ISAR2_EL1_GPA3)),
3344 	ID_WRITABLE(ID_AA64ISAR3_EL1, (ID_AA64ISAR3_EL1_FPRCVT |
3345 				       ID_AA64ISAR3_EL1_LSFE |
3346 				       ID_AA64ISAR3_EL1_LSUI |
3347 				       ID_AA64ISAR3_EL1_FAMINMAX)),
3348 	ID_UNALLOCATED(6,4),
3349 	ID_UNALLOCATED(6,5),
3350 	ID_UNALLOCATED(6,6),
3351 	ID_UNALLOCATED(6,7),
3352 
3353 	/* CRm=7 */
3354 	ID_FILTERED(ID_AA64MMFR0_EL1, id_aa64mmfr0_el1,
3355 				      ~(ID_AA64MMFR0_EL1_RES0 |
3356 					ID_AA64MMFR0_EL1_ASIDBITS)),
3357 	ID_WRITABLE(ID_AA64MMFR1_EL1, ~(ID_AA64MMFR1_EL1_RES0 |
3358 					ID_AA64MMFR1_EL1_XNX |
3359 					ID_AA64MMFR1_EL1_VH |
3360 					ID_AA64MMFR1_EL1_VMIDBits)),
3361 	ID_FILTERED(ID_AA64MMFR2_EL1,
3362 		    id_aa64mmfr2_el1, ~(ID_AA64MMFR2_EL1_RES0 |
3363 					ID_AA64MMFR2_EL1_EVT |
3364 					ID_AA64MMFR2_EL1_FWB |
3365 					ID_AA64MMFR2_EL1_IDS |
3366 					ID_AA64MMFR2_EL1_NV |
3367 					ID_AA64MMFR2_EL1_CCIDX)),
3368 	ID_WRITABLE(ID_AA64MMFR3_EL1, (ID_AA64MMFR3_EL1_TCRX	|
3369 				       ID_AA64MMFR3_EL1_SCTLRX	|
3370 				       ID_AA64MMFR3_EL1_S1PIE   |
3371 				       ID_AA64MMFR3_EL1_S1POE)),
3372 	ID_WRITABLE(ID_AA64MMFR4_EL1, ID_AA64MMFR4_EL1_NV_frac),
3373 	ID_UNALLOCATED(7,5),
3374 	ID_UNALLOCATED(7,6),
3375 	ID_UNALLOCATED(7,7),
3376 
3377 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
3378 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
3379 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
3380 	{ SYS_DESC(SYS_SCTLR2_EL1), access_vm_reg, reset_val, SCTLR2_EL1, 0,
3381 	  .visibility = sctlr2_visibility },
3382 
3383 	MTE_REG(RGSR_EL1),
3384 	MTE_REG(GCR_EL1),
3385 
3386 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
3387 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
3388 	{ SYS_DESC(SYS_SMPRI_EL1), undef_access },
3389 	{ SYS_DESC(SYS_SMCR_EL1), undef_access },
3390 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
3391 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
3392 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
3393 	{ SYS_DESC(SYS_TCR2_EL1), access_vm_reg, reset_val, TCR2_EL1, 0,
3394 	  .visibility = tcr2_visibility },
3395 
3396 	PTRAUTH_KEY(APIA),
3397 	PTRAUTH_KEY(APIB),
3398 	PTRAUTH_KEY(APDA),
3399 	PTRAUTH_KEY(APDB),
3400 	PTRAUTH_KEY(APGA),
3401 
3402 	{ SYS_DESC(SYS_SPSR_EL1), access_spsr},
3403 	{ SYS_DESC(SYS_ELR_EL1), access_elr},
3404 
3405 	{ SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
3406 
3407 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
3408 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
3409 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
3410 
3411 	{ SYS_DESC(SYS_ERRIDR_EL1), access_ras },
3412 	{ SYS_DESC(SYS_ERRSELR_EL1), access_ras },
3413 	{ SYS_DESC(SYS_ERXFR_EL1), access_ras },
3414 	{ SYS_DESC(SYS_ERXCTLR_EL1), access_ras },
3415 	{ SYS_DESC(SYS_ERXSTATUS_EL1), access_ras },
3416 	{ SYS_DESC(SYS_ERXADDR_EL1), access_ras },
3417 	{ SYS_DESC(SYS_ERXPFGF_EL1), access_ras },
3418 	{ SYS_DESC(SYS_ERXPFGCTL_EL1), access_ras },
3419 	{ SYS_DESC(SYS_ERXPFGCDN_EL1), access_ras },
3420 	{ SYS_DESC(SYS_ERXMISC0_EL1), access_ras },
3421 	{ SYS_DESC(SYS_ERXMISC1_EL1), access_ras },
3422 	{ SYS_DESC(SYS_ERXMISC2_EL1), access_ras },
3423 	{ SYS_DESC(SYS_ERXMISC3_EL1), access_ras },
3424 
3425 	MTE_REG(TFSR_EL1),
3426 	MTE_REG(TFSRE0_EL1),
3427 
3428 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
3429 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
3430 
3431 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
3432 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
3433 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
3434 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
3435 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
3436 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
3437 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
3438 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
3439 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
3440 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
3441 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
3442 	{ SYS_DESC(SYS_PMSDSFR_EL1), undef_access },
3443 	/* PMBIDR_EL1 is not trapped */
3444 
3445 	{ PMU_SYS_REG(PMINTENSET_EL1),
3446 	  .access = access_pminten, .reg = PMINTENSET_EL1,
3447 	  .get_user = get_pmreg, .set_user = set_pmreg },
3448 	{ PMU_SYS_REG(PMINTENCLR_EL1),
3449 	  .access = access_pminten, .reg = PMINTENSET_EL1,
3450 	  .get_user = get_pmreg, .set_user = set_pmreg },
3451 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
3452 
3453 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
3454 	{ SYS_DESC(SYS_PIRE0_EL1), NULL, reset_unknown, PIRE0_EL1,
3455 	  .visibility = s1pie_visibility },
3456 	{ SYS_DESC(SYS_PIR_EL1), NULL, reset_unknown, PIR_EL1,
3457 	  .visibility = s1pie_visibility },
3458 	{ SYS_DESC(SYS_POR_EL1), NULL, reset_unknown, POR_EL1,
3459 	  .visibility = s1poe_visibility },
3460 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
3461 
3462 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
3463 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
3464 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
3465 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
3466 	{ SYS_DESC(SYS_MPAMIDR_EL1), undef_access },
3467 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
3468 
3469 	{ SYS_DESC(SYS_MPAM1_EL1), undef_access },
3470 	{ SYS_DESC(SYS_MPAM0_EL1), undef_access },
3471 	{ SYS_DESC(SYS_MPAMSM_EL1), undef_access },
3472 
3473 	{ SYS_DESC(SYS_VBAR_EL1), access_rw, reset_val, VBAR_EL1, 0 },
3474 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
3475 
3476 	{ SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
3477 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
3478 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
3479 	{ SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
3480 	{ SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
3481 	{ SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
3482 	{ SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
3483 	{ SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
3484 	{ SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
3485 	{ SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
3486 	{ SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
3487 	{ SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
3488 	{ SYS_DESC(SYS_ICC_IDR0_EL1), access_gicv5_idr0 },
3489 	{ SYS_DESC(SYS_ICC_IAFFIDR_EL1), access_gicv5_iaffid },
3490 	{ SYS_DESC(SYS_ICC_PPI_ENABLER0_EL1), access_gicv5_ppi_enabler },
3491 	{ SYS_DESC(SYS_ICC_PPI_ENABLER1_EL1), access_gicv5_ppi_enabler },
3492 	{ SYS_DESC(SYS_ICC_DIR_EL1), access_gic_dir },
3493 	{ SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
3494 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
3495 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
3496 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
3497 	{ SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
3498 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
3499 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
3500 	{ SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
3501 	{ SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
3502 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
3503 	{ SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
3504 	{ SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
3505 
3506 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
3507 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
3508 
3509 	{ SYS_DESC(SYS_ACCDATA_EL1), undef_access },
3510 
3511 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
3512 
3513 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
3514 
3515 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
3516 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr, reset_clidr, CLIDR_EL1,
3517 	  .set_user = set_clidr, .val = ~CLIDR_EL1_RES0 },
3518 	IMPLEMENTATION_ID(AIDR_EL1, GENMASK_ULL(63, 0)),
3519 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
3520 	ID_FILTERED(CTR_EL0, ctr_el0,
3521 		    CTR_EL0_DIC_MASK |
3522 		    CTR_EL0_IDC_MASK |
3523 		    CTR_EL0_DminLine_MASK |
3524 		    CTR_EL0_L1Ip_MASK |
3525 		    CTR_EL0_IminLine_MASK),
3526 	{ SYS_DESC(SYS_SVCR), undef_access, reset_val, SVCR, 0, .visibility = sme_visibility  },
3527 	{ SYS_DESC(SYS_FPMR), undef_access, reset_val, FPMR, 0, .visibility = fp8_visibility },
3528 
3529 	{ PMU_SYS_REG(PMCR_EL0), .access = access_pmcr, .reset = reset_pmcr,
3530 	  .reg = PMCR_EL0, .get_user = get_pmcr, .set_user = set_pmcr },
3531 	{ PMU_SYS_REG(PMCNTENSET_EL0),
3532 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
3533 	  .get_user = get_pmreg, .set_user = set_pmreg },
3534 	{ PMU_SYS_REG(PMCNTENCLR_EL0),
3535 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
3536 	  .get_user = get_pmreg, .set_user = set_pmreg },
3537 	{ PMU_SYS_REG(PMOVSCLR_EL0),
3538 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
3539 	  .get_user = get_pmreg, .set_user = set_pmreg },
3540 	/*
3541 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
3542 	 * previously (and pointlessly) advertised in the past...
3543 	 */
3544 	{ PMU_SYS_REG(PMSWINC_EL0),
3545 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
3546 	  .access = access_pmswinc, .reset = NULL },
3547 	{ PMU_SYS_REG(PMSELR_EL0),
3548 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
3549 	{ PMU_SYS_REG(PMCEID0_EL0),
3550 	  .access = access_pmceid, .reset = NULL },
3551 	{ PMU_SYS_REG(PMCEID1_EL0),
3552 	  .access = access_pmceid, .reset = NULL },
3553 	{ PMU_SYS_REG(PMCCNTR_EL0),
3554 	  .access = access_pmu_evcntr, .reset = reset_unknown,
3555 	  .reg = PMCCNTR_EL0, .get_user = get_pmu_evcntr,
3556 	  .set_user = set_pmu_evcntr },
3557 	{ PMU_SYS_REG(PMXEVTYPER_EL0),
3558 	  .access = access_pmu_evtyper, .reset = NULL },
3559 	{ PMU_SYS_REG(PMXEVCNTR_EL0),
3560 	  .access = access_pmu_evcntr, .reset = NULL },
3561 	/*
3562 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
3563 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
3564 	 */
3565 	{ PMU_SYS_REG(PMUSERENR_EL0), .access = access_pmuserenr,
3566 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
3567 	{ PMU_SYS_REG(PMOVSSET_EL0),
3568 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
3569 	  .get_user = get_pmreg, .set_user = set_pmreg },
3570 
3571 	{ SYS_DESC(SYS_POR_EL0), NULL, reset_unknown, POR_EL0,
3572 	  .visibility = s1poe_visibility },
3573 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
3574 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
3575 	{ SYS_DESC(SYS_TPIDR2_EL0), undef_access },
3576 
3577 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
3578 
3579 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
3580 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
3581 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
3582 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
3583 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
3584 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
3585 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
3586 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
3587 	AMU_AMEVCNTR0_EL0(0),
3588 	AMU_AMEVCNTR0_EL0(1),
3589 	AMU_AMEVCNTR0_EL0(2),
3590 	AMU_AMEVCNTR0_EL0(3),
3591 	AMU_AMEVCNTR0_EL0(4),
3592 	AMU_AMEVCNTR0_EL0(5),
3593 	AMU_AMEVCNTR0_EL0(6),
3594 	AMU_AMEVCNTR0_EL0(7),
3595 	AMU_AMEVCNTR0_EL0(8),
3596 	AMU_AMEVCNTR0_EL0(9),
3597 	AMU_AMEVCNTR0_EL0(10),
3598 	AMU_AMEVCNTR0_EL0(11),
3599 	AMU_AMEVCNTR0_EL0(12),
3600 	AMU_AMEVCNTR0_EL0(13),
3601 	AMU_AMEVCNTR0_EL0(14),
3602 	AMU_AMEVCNTR0_EL0(15),
3603 	AMU_AMEVTYPER0_EL0(0),
3604 	AMU_AMEVTYPER0_EL0(1),
3605 	AMU_AMEVTYPER0_EL0(2),
3606 	AMU_AMEVTYPER0_EL0(3),
3607 	AMU_AMEVTYPER0_EL0(4),
3608 	AMU_AMEVTYPER0_EL0(5),
3609 	AMU_AMEVTYPER0_EL0(6),
3610 	AMU_AMEVTYPER0_EL0(7),
3611 	AMU_AMEVTYPER0_EL0(8),
3612 	AMU_AMEVTYPER0_EL0(9),
3613 	AMU_AMEVTYPER0_EL0(10),
3614 	AMU_AMEVTYPER0_EL0(11),
3615 	AMU_AMEVTYPER0_EL0(12),
3616 	AMU_AMEVTYPER0_EL0(13),
3617 	AMU_AMEVTYPER0_EL0(14),
3618 	AMU_AMEVTYPER0_EL0(15),
3619 	AMU_AMEVCNTR1_EL0(0),
3620 	AMU_AMEVCNTR1_EL0(1),
3621 	AMU_AMEVCNTR1_EL0(2),
3622 	AMU_AMEVCNTR1_EL0(3),
3623 	AMU_AMEVCNTR1_EL0(4),
3624 	AMU_AMEVCNTR1_EL0(5),
3625 	AMU_AMEVCNTR1_EL0(6),
3626 	AMU_AMEVCNTR1_EL0(7),
3627 	AMU_AMEVCNTR1_EL0(8),
3628 	AMU_AMEVCNTR1_EL0(9),
3629 	AMU_AMEVCNTR1_EL0(10),
3630 	AMU_AMEVCNTR1_EL0(11),
3631 	AMU_AMEVCNTR1_EL0(12),
3632 	AMU_AMEVCNTR1_EL0(13),
3633 	AMU_AMEVCNTR1_EL0(14),
3634 	AMU_AMEVCNTR1_EL0(15),
3635 	AMU_AMEVTYPER1_EL0(0),
3636 	AMU_AMEVTYPER1_EL0(1),
3637 	AMU_AMEVTYPER1_EL0(2),
3638 	AMU_AMEVTYPER1_EL0(3),
3639 	AMU_AMEVTYPER1_EL0(4),
3640 	AMU_AMEVTYPER1_EL0(5),
3641 	AMU_AMEVTYPER1_EL0(6),
3642 	AMU_AMEVTYPER1_EL0(7),
3643 	AMU_AMEVTYPER1_EL0(8),
3644 	AMU_AMEVTYPER1_EL0(9),
3645 	AMU_AMEVTYPER1_EL0(10),
3646 	AMU_AMEVTYPER1_EL0(11),
3647 	AMU_AMEVTYPER1_EL0(12),
3648 	AMU_AMEVTYPER1_EL0(13),
3649 	AMU_AMEVTYPER1_EL0(14),
3650 	AMU_AMEVTYPER1_EL0(15),
3651 
3652 	{ SYS_DESC(SYS_CNTPCT_EL0), .access = access_arch_timer,
3653 	  .get_user = arch_timer_get_user, .set_user = arch_timer_set_user },
3654 	{ SYS_DESC(SYS_CNTVCT_EL0), .access = access_arch_timer,
3655 	  .get_user = arch_timer_get_user, .set_user = arch_timer_set_user },
3656 	{ SYS_DESC(SYS_CNTPCTSS_EL0), access_arch_timer },
3657 	{ SYS_DESC(SYS_CNTVCTSS_EL0), access_arch_timer },
3658 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
3659 	TIMER_REG(CNTP_CTL_EL0, NULL),
3660 	TIMER_REG(CNTP_CVAL_EL0, NULL),
3661 
3662 	{ SYS_DESC(SYS_CNTV_TVAL_EL0), access_arch_timer },
3663 	TIMER_REG(CNTV_CTL_EL0, NULL),
3664 	TIMER_REG(CNTV_CVAL_EL0, NULL),
3665 
3666 	/* PMEVCNTRn_EL0 */
3667 	PMU_PMEVCNTR_EL0(0),
3668 	PMU_PMEVCNTR_EL0(1),
3669 	PMU_PMEVCNTR_EL0(2),
3670 	PMU_PMEVCNTR_EL0(3),
3671 	PMU_PMEVCNTR_EL0(4),
3672 	PMU_PMEVCNTR_EL0(5),
3673 	PMU_PMEVCNTR_EL0(6),
3674 	PMU_PMEVCNTR_EL0(7),
3675 	PMU_PMEVCNTR_EL0(8),
3676 	PMU_PMEVCNTR_EL0(9),
3677 	PMU_PMEVCNTR_EL0(10),
3678 	PMU_PMEVCNTR_EL0(11),
3679 	PMU_PMEVCNTR_EL0(12),
3680 	PMU_PMEVCNTR_EL0(13),
3681 	PMU_PMEVCNTR_EL0(14),
3682 	PMU_PMEVCNTR_EL0(15),
3683 	PMU_PMEVCNTR_EL0(16),
3684 	PMU_PMEVCNTR_EL0(17),
3685 	PMU_PMEVCNTR_EL0(18),
3686 	PMU_PMEVCNTR_EL0(19),
3687 	PMU_PMEVCNTR_EL0(20),
3688 	PMU_PMEVCNTR_EL0(21),
3689 	PMU_PMEVCNTR_EL0(22),
3690 	PMU_PMEVCNTR_EL0(23),
3691 	PMU_PMEVCNTR_EL0(24),
3692 	PMU_PMEVCNTR_EL0(25),
3693 	PMU_PMEVCNTR_EL0(26),
3694 	PMU_PMEVCNTR_EL0(27),
3695 	PMU_PMEVCNTR_EL0(28),
3696 	PMU_PMEVCNTR_EL0(29),
3697 	PMU_PMEVCNTR_EL0(30),
3698 	/* PMEVTYPERn_EL0 */
3699 	PMU_PMEVTYPER_EL0(0),
3700 	PMU_PMEVTYPER_EL0(1),
3701 	PMU_PMEVTYPER_EL0(2),
3702 	PMU_PMEVTYPER_EL0(3),
3703 	PMU_PMEVTYPER_EL0(4),
3704 	PMU_PMEVTYPER_EL0(5),
3705 	PMU_PMEVTYPER_EL0(6),
3706 	PMU_PMEVTYPER_EL0(7),
3707 	PMU_PMEVTYPER_EL0(8),
3708 	PMU_PMEVTYPER_EL0(9),
3709 	PMU_PMEVTYPER_EL0(10),
3710 	PMU_PMEVTYPER_EL0(11),
3711 	PMU_PMEVTYPER_EL0(12),
3712 	PMU_PMEVTYPER_EL0(13),
3713 	PMU_PMEVTYPER_EL0(14),
3714 	PMU_PMEVTYPER_EL0(15),
3715 	PMU_PMEVTYPER_EL0(16),
3716 	PMU_PMEVTYPER_EL0(17),
3717 	PMU_PMEVTYPER_EL0(18),
3718 	PMU_PMEVTYPER_EL0(19),
3719 	PMU_PMEVTYPER_EL0(20),
3720 	PMU_PMEVTYPER_EL0(21),
3721 	PMU_PMEVTYPER_EL0(22),
3722 	PMU_PMEVTYPER_EL0(23),
3723 	PMU_PMEVTYPER_EL0(24),
3724 	PMU_PMEVTYPER_EL0(25),
3725 	PMU_PMEVTYPER_EL0(26),
3726 	PMU_PMEVTYPER_EL0(27),
3727 	PMU_PMEVTYPER_EL0(28),
3728 	PMU_PMEVTYPER_EL0(29),
3729 	PMU_PMEVTYPER_EL0(30),
3730 	/*
3731 	 * PMCCFILTR_EL0 resets as unknown in 64bit mode while it resets as zero
3732 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
3733 	 */
3734 	{ PMU_SYS_REG(PMCCFILTR_EL0), .access = access_pmu_evtyper,
3735 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
3736 
3737 	EL2_REG_VNCR(VPIDR_EL2, reset_unknown, 0),
3738 	EL2_REG_VNCR(VMPIDR_EL2, reset_unknown, 0),
3739 	EL2_REG(SCTLR_EL2, access_rw, reset_val, SCTLR_EL2_RES1),
3740 	EL2_REG(ACTLR_EL2, access_rw, reset_val, 0),
3741 	EL2_REG_FILTERED(SCTLR2_EL2, access_vm_reg, reset_val, 0,
3742 			 sctlr2_el2_visibility),
3743 	EL2_REG_VNCR(HCR_EL2, reset_hcr, 0),
3744 	EL2_REG(MDCR_EL2, access_mdcr, reset_mdcr, 0),
3745 	EL2_REG(CPTR_EL2, access_rw, reset_val, CPTR_NVHE_EL2_RES1),
3746 	EL2_REG_VNCR(HSTR_EL2, reset_val, 0),
3747 	EL2_REG_VNCR_FILT(HFGRTR_EL2, fgt_visibility),
3748 	EL2_REG_VNCR_FILT(HFGWTR_EL2, fgt_visibility),
3749 	EL2_REG_VNCR(HFGITR_EL2, reset_val, 0),
3750 	EL2_REG_VNCR(HACR_EL2, reset_val, 0),
3751 
3752 	EL2_REG_FILTERED(ZCR_EL2, access_zcr_el2, reset_val, 0,
3753 			 sve_el2_visibility),
3754 
3755 	EL2_REG_VNCR(HCRX_EL2, reset_val, 0),
3756 
3757 	EL2_REG(TTBR0_EL2, access_rw, reset_val, 0),
3758 	EL2_REG(TTBR1_EL2, access_rw, reset_val, 0),
3759 	EL2_REG(TCR_EL2, access_rw, reset_val, TCR_EL2_RES1),
3760 	EL2_REG_FILTERED(TCR2_EL2, access_rw, reset_val, TCR2_EL2_RES1,
3761 			 tcr2_el2_visibility),
3762 	EL2_REG_VNCR(VTTBR_EL2, reset_val, 0),
3763 	EL2_REG_VNCR(VTCR_EL2, reset_val, 0),
3764 	EL2_REG_FILTERED(VNCR_EL2, bad_vncr_trap, reset_val, 0,
3765 			 vncr_el2_visibility),
3766 
3767 	{ SYS_DESC(SYS_DACR32_EL2), undef_access, reset_unknown, DACR32_EL2 },
3768 	EL2_REG_VNCR_FILT(HDFGRTR2_EL2, fgt2_visibility),
3769 	EL2_REG_VNCR_FILT(HDFGWTR2_EL2, fgt2_visibility),
3770 	EL2_REG_VNCR_FILT(HFGRTR2_EL2, fgt2_visibility),
3771 	EL2_REG_VNCR_FILT(HFGWTR2_EL2, fgt2_visibility),
3772 	EL2_REG_VNCR_FILT(HDFGRTR_EL2, fgt_visibility),
3773 	EL2_REG_VNCR_FILT(HDFGWTR_EL2, fgt_visibility),
3774 	EL2_REG_VNCR_FILT(HAFGRTR_EL2, fgt_visibility),
3775 	EL2_REG_VNCR_FILT(HFGITR2_EL2, fgt2_visibility),
3776 	EL2_REG_REDIR(SPSR_EL2, reset_val, 0),
3777 	EL2_REG_REDIR(ELR_EL2, reset_val, 0),
3778 	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},
3779 
3780 	/* AArch32 SPSR_* are RES0 if trapped from a NV guest */
3781 	{ SYS_DESC(SYS_SPSR_irq), .access = trap_raz_wi },
3782 	{ SYS_DESC(SYS_SPSR_abt), .access = trap_raz_wi },
3783 	{ SYS_DESC(SYS_SPSR_und), .access = trap_raz_wi },
3784 	{ SYS_DESC(SYS_SPSR_fiq), .access = trap_raz_wi },
3785 
3786 	{ SYS_DESC(SYS_IFSR32_EL2), undef_access, reset_unknown, IFSR32_EL2 },
3787 	EL2_REG(AFSR0_EL2, access_rw, reset_val, 0),
3788 	EL2_REG(AFSR1_EL2, access_rw, reset_val, 0),
3789 	EL2_REG_REDIR(ESR_EL2, reset_val, 0),
3790 	EL2_REG_VNCR(VSESR_EL2, reset_unknown, 0),
3791 	{ SYS_DESC(SYS_FPEXC32_EL2), undef_access, reset_val, FPEXC32_EL2, 0x700 },
3792 
3793 	EL2_REG_REDIR(FAR_EL2, reset_val, 0),
3794 	EL2_REG(HPFAR_EL2, access_rw, reset_val, 0),
3795 
3796 	EL2_REG(MAIR_EL2, access_rw, reset_val, 0),
3797 	EL2_REG_FILTERED(PIRE0_EL2, access_rw, reset_val, 0,
3798 			 s1pie_el2_visibility),
3799 	EL2_REG_FILTERED(PIR_EL2, access_rw, reset_val, 0,
3800 			 s1pie_el2_visibility),
3801 	EL2_REG_FILTERED(POR_EL2, access_rw, reset_val, 0,
3802 			 s1poe_el2_visibility),
3803 	EL2_REG(AMAIR_EL2, access_rw, reset_val, 0),
3804 	{ SYS_DESC(SYS_MPAMHCR_EL2), undef_access },
3805 	{ SYS_DESC(SYS_MPAMVPMV_EL2), undef_access },
3806 	{ SYS_DESC(SYS_MPAM2_EL2), undef_access },
3807 	{ SYS_DESC(SYS_MPAMVPM0_EL2), undef_access },
3808 	{ SYS_DESC(SYS_MPAMVPM1_EL2), undef_access },
3809 	{ SYS_DESC(SYS_MPAMVPM2_EL2), undef_access },
3810 	{ SYS_DESC(SYS_MPAMVPM3_EL2), undef_access },
3811 	{ SYS_DESC(SYS_MPAMVPM4_EL2), undef_access },
3812 	{ SYS_DESC(SYS_MPAMVPM5_EL2), undef_access },
3813 	{ SYS_DESC(SYS_MPAMVPM6_EL2), undef_access },
3814 	{ SYS_DESC(SYS_MPAMVPM7_EL2), undef_access },
3815 
3816 	EL2_REG(VBAR_EL2, access_rw, reset_val, 0),
3817 	{ SYS_DESC(SYS_RVBAR_EL2), undef_access },
3818 	{ SYS_DESC(SYS_RMR_EL2), undef_access },
3819 	EL2_REG_VNCR(VDISR_EL2, reset_unknown, 0),
3820 
3821 	EL2_REG_VNCR_GICv3(ICH_AP0R0_EL2),
3822 	EL2_REG_VNCR_GICv3(ICH_AP0R1_EL2),
3823 	EL2_REG_VNCR_GICv3(ICH_AP0R2_EL2),
3824 	EL2_REG_VNCR_GICv3(ICH_AP0R3_EL2),
3825 	EL2_REG_VNCR_GICv3(ICH_AP1R0_EL2),
3826 	EL2_REG_VNCR_GICv3(ICH_AP1R1_EL2),
3827 	EL2_REG_VNCR_GICv3(ICH_AP1R2_EL2),
3828 	EL2_REG_VNCR_GICv3(ICH_AP1R3_EL2),
3829 
3830 	{ SYS_DESC(SYS_ICC_SRE_EL2), access_gic_sre },
3831 
3832 	EL2_REG_VNCR_GICv3(ICH_HCR_EL2),
3833 	{ SYS_DESC(SYS_ICH_VTR_EL2), access_gic_vtr },
3834 	{ SYS_DESC(SYS_ICH_MISR_EL2), access_gic_misr },
3835 	{ SYS_DESC(SYS_ICH_EISR_EL2), access_gic_eisr },
3836 	{ SYS_DESC(SYS_ICH_ELRSR_EL2), access_gic_elrsr },
3837 	EL2_REG_VNCR_GICv3(ICH_VMCR_EL2),
3838 
3839 	EL2_REG_VNCR_GICv3(ICH_LR0_EL2),
3840 	EL2_REG_VNCR_GICv3(ICH_LR1_EL2),
3841 	EL2_REG_VNCR_GICv3(ICH_LR2_EL2),
3842 	EL2_REG_VNCR_GICv3(ICH_LR3_EL2),
3843 	EL2_REG_VNCR_GICv3(ICH_LR4_EL2),
3844 	EL2_REG_VNCR_GICv3(ICH_LR5_EL2),
3845 	EL2_REG_VNCR_GICv3(ICH_LR6_EL2),
3846 	EL2_REG_VNCR_GICv3(ICH_LR7_EL2),
3847 	EL2_REG_VNCR_GICv3(ICH_LR8_EL2),
3848 	EL2_REG_VNCR_GICv3(ICH_LR9_EL2),
3849 	EL2_REG_VNCR_GICv3(ICH_LR10_EL2),
3850 	EL2_REG_VNCR_GICv3(ICH_LR11_EL2),
3851 	EL2_REG_VNCR_GICv3(ICH_LR12_EL2),
3852 	EL2_REG_VNCR_GICv3(ICH_LR13_EL2),
3853 	EL2_REG_VNCR_GICv3(ICH_LR14_EL2),
3854 	EL2_REG_VNCR_GICv3(ICH_LR15_EL2),
3855 
3856 	EL2_REG(CONTEXTIDR_EL2, access_rw, reset_val, 0),
3857 	EL2_REG(TPIDR_EL2, access_rw, reset_val, 0),
3858 
3859 	EL2_REG_VNCR(CNTVOFF_EL2, reset_val, 0),
3860 	EL2_REG(CNTHCTL_EL2, access_rw, reset_val, 0),
3861 	{ SYS_DESC(SYS_CNTHP_TVAL_EL2), access_arch_timer },
3862 	TIMER_REG(CNTHP_CTL_EL2, el2_visibility),
3863 	TIMER_REG(CNTHP_CVAL_EL2, el2_visibility),
3864 
3865 	{ SYS_DESC(SYS_CNTHV_TVAL_EL2), access_arch_timer, .visibility = cnthv_visibility },
3866 	TIMER_REG(CNTHV_CTL_EL2, cnthv_visibility),
3867 	TIMER_REG(CNTHV_CVAL_EL2, cnthv_visibility),
3868 
3869 	{ SYS_DESC(SYS_CNTKCTL_EL12), access_cntkctl_el12 },
3870 
3871 	{ SYS_DESC(SYS_CNTP_TVAL_EL02), access_arch_timer },
3872 	{ SYS_DESC(SYS_CNTP_CTL_EL02), access_arch_timer },
3873 	{ SYS_DESC(SYS_CNTP_CVAL_EL02), access_arch_timer },
3874 
3875 	{ SYS_DESC(SYS_CNTV_TVAL_EL02), access_arch_timer },
3876 	{ SYS_DESC(SYS_CNTV_CTL_EL02), access_arch_timer },
3877 	{ SYS_DESC(SYS_CNTV_CVAL_EL02), access_arch_timer },
3878 
3879 	EL2_REG(SP_EL2, NULL, reset_unknown, 0),
3880 };
3881 
3882 static bool handle_at_s1e01(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3883 			    const struct sys_reg_desc *r)
3884 {
3885 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3886 
3887 	if (__kvm_at_s1e01(vcpu, op, p->regval))
3888 		return false;
3889 
3890 	return true;
3891 }
3892 
3893 static bool handle_at_s1e2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3894 			   const struct sys_reg_desc *r)
3895 {
3896 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3897 
3898 	/* There is no FGT associated with AT S1E2A :-( */
3899 	if (op == OP_AT_S1E2A &&
3900 	    !kvm_has_feat(vcpu->kvm, ID_AA64ISAR2_EL1, ATS1A, IMP)) {
3901 		kvm_inject_undefined(vcpu);
3902 		return false;
3903 	}
3904 
3905 	if (__kvm_at_s1e2(vcpu, op, p->regval))
3906 		return false;
3907 
3908 	return true;
3909 }
3910 
3911 static bool handle_at_s12(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3912 			  const struct sys_reg_desc *r)
3913 {
3914 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3915 
3916 	if (__kvm_at_s12(vcpu, op, p->regval))
3917 		return false;
3918 
3919 	return true;
3920 }
3921 
3922 static bool kvm_supported_tlbi_s12_op(struct kvm_vcpu *vpcu, u32 instr)
3923 {
3924 	struct kvm *kvm = vpcu->kvm;
3925 	u8 CRm = sys_reg_CRm(instr);
3926 
3927 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3928 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3929 		return false;
3930 
3931 	if (CRm == TLBI_CRm_nROS &&
3932 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3933 		return false;
3934 
3935 	return true;
3936 }
3937 
3938 static bool handle_alle1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3939 			   const struct sys_reg_desc *r)
3940 {
3941 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3942 
3943 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3944 		return undef_access(vcpu, p, r);
3945 
3946 	write_lock(&vcpu->kvm->mmu_lock);
3947 
3948 	/*
3949 	 * Drop all shadow S2s, resulting in S1/S2 TLBIs for each of the
3950 	 * corresponding VMIDs.
3951 	 */
3952 	kvm_nested_s2_unmap(vcpu->kvm, true);
3953 
3954 	write_unlock(&vcpu->kvm->mmu_lock);
3955 
3956 	return true;
3957 }
3958 
3959 static bool kvm_supported_tlbi_ipas2_op(struct kvm_vcpu *vpcu, u32 instr)
3960 {
3961 	struct kvm *kvm = vpcu->kvm;
3962 	u8 CRm = sys_reg_CRm(instr);
3963 	u8 Op2 = sys_reg_Op2(instr);
3964 
3965 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3966 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3967 		return false;
3968 
3969 	if (CRm == TLBI_CRm_IPAIS && (Op2 == 2 || Op2 == 6) &&
3970 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3971 		return false;
3972 
3973 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 0 || Op2 == 4) &&
3974 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3975 		return false;
3976 
3977 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 3 || Op2 == 7) &&
3978 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3979 		return false;
3980 
3981 	return true;
3982 }
3983 
3984 /* Only defined here as this is an internal "abstraction" */
3985 union tlbi_info {
3986 	struct {
3987 		u64	start;
3988 		u64	size;
3989 	} range;
3990 
3991 	struct {
3992 		u64	addr;
3993 	} ipa;
3994 
3995 	struct {
3996 		u64	addr;
3997 		u32	encoding;
3998 	} va;
3999 };
4000 
4001 static void s2_mmu_unmap_range(struct kvm_s2_mmu *mmu,
4002 			       const union tlbi_info *info)
4003 {
4004 	/*
4005 	 * The unmap operation is allowed to drop the MMU lock and block, which
4006 	 * means that @mmu could be used for a different context than the one
4007 	 * currently being invalidated.
4008 	 *
4009 	 * This behavior is still safe, as:
4010 	 *
4011 	 *  1) The vCPU(s) that recycled the MMU are responsible for invalidating
4012 	 *     the entire MMU before reusing it, which still honors the intent
4013 	 *     of a TLBI.
4014 	 *
4015 	 *  2) Until the guest TLBI instruction is 'retired' (i.e. increment PC
4016 	 *     and ERET to the guest), other vCPUs are allowed to use stale
4017 	 *     translations.
4018 	 *
4019 	 *  3) Accidentally unmapping an unrelated MMU context is nonfatal, and
4020 	 *     at worst may cause more aborts for shadow stage-2 fills.
4021 	 *
4022 	 * Dropping the MMU lock also implies that shadow stage-2 fills could
4023 	 * happen behind the back of the TLBI. This is still safe, though, as
4024 	 * the L1 needs to put its stage-2 in a consistent state before doing
4025 	 * the TLBI.
4026 	 */
4027 	kvm_stage2_unmap_range(mmu, info->range.start, info->range.size, true);
4028 }
4029 
4030 static bool handle_vmalls12e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
4031 				const struct sys_reg_desc *r)
4032 {
4033 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
4034 	u64 limit, vttbr;
4035 
4036 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
4037 		return undef_access(vcpu, p, r);
4038 
4039 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
4040 	limit = BIT_ULL(kvm_get_pa_bits(vcpu->kvm));
4041 
4042 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
4043 				   &(union tlbi_info) {
4044 					   .range = {
4045 						   .start = 0,
4046 						   .size = limit,
4047 					   },
4048 				   },
4049 				   s2_mmu_unmap_range);
4050 
4051 	return true;
4052 }
4053 
4054 static bool handle_ripas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
4055 			      const struct sys_reg_desc *r)
4056 {
4057 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
4058 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
4059 	u64 base, range;
4060 
4061 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
4062 		return undef_access(vcpu, p, r);
4063 
4064 	/*
4065 	 * Because the shadow S2 structure doesn't necessarily reflect that
4066 	 * of the guest's S2 (different base granule size, for example), we
4067 	 * decide to ignore TTL and only use the described range.
4068 	 */
4069 	base = decode_range_tlbi(p->regval, &range, NULL);
4070 
4071 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
4072 				   &(union tlbi_info) {
4073 					   .range = {
4074 						   .start = base,
4075 						   .size = range,
4076 					   },
4077 				   },
4078 				   s2_mmu_unmap_range);
4079 
4080 	return true;
4081 }
4082 
4083 static void s2_mmu_unmap_ipa(struct kvm_s2_mmu *mmu,
4084 			     const union tlbi_info *info)
4085 {
4086 	unsigned long max_size;
4087 	u64 base_addr;
4088 
4089 	/*
4090 	 * We drop a number of things from the supplied value:
4091 	 *
4092 	 * - NS bit: we're non-secure only.
4093 	 *
4094 	 * - IPA[51:48]: We don't support 52bit IPA just yet...
4095 	 *
4096 	 * And of course, adjust the IPA to be on an actual address.
4097 	 */
4098 	base_addr = (info->ipa.addr & GENMASK_ULL(35, 0)) << 12;
4099 	max_size = compute_tlb_inval_range(mmu, info->ipa.addr);
4100 	base_addr &= ~(max_size - 1);
4101 
4102 	/*
4103 	 * See comment in s2_mmu_unmap_range() for why this is allowed to
4104 	 * reschedule.
4105 	 */
4106 	kvm_stage2_unmap_range(mmu, base_addr, max_size, true);
4107 }
4108 
4109 static bool handle_ipas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
4110 			     const struct sys_reg_desc *r)
4111 {
4112 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
4113 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
4114 
4115 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
4116 		return undef_access(vcpu, p, r);
4117 
4118 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
4119 				   &(union tlbi_info) {
4120 					   .ipa = {
4121 						   .addr = p->regval,
4122 					   },
4123 				   },
4124 				   s2_mmu_unmap_ipa);
4125 
4126 	return true;
4127 }
4128 
4129 static void s2_mmu_tlbi_s1e1(struct kvm_s2_mmu *mmu,
4130 			     const union tlbi_info *info)
4131 {
4132 	WARN_ON(__kvm_tlbi_s1e2(mmu, info->va.addr, info->va.encoding));
4133 }
4134 
4135 static bool handle_tlbi_el2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
4136 			    const struct sys_reg_desc *r)
4137 {
4138 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
4139 
4140 	if (!kvm_supported_tlbi_s1e2_op(vcpu, sys_encoding))
4141 		return undef_access(vcpu, p, r);
4142 
4143 	kvm_handle_s1e2_tlbi(vcpu, sys_encoding, p->regval);
4144 	return true;
4145 }
4146 
4147 static bool handle_tlbi_el1(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
4148 			    const struct sys_reg_desc *r)
4149 {
4150 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
4151 
4152 	/*
4153 	 * If we're here, this is because we've trapped on a EL1 TLBI
4154 	 * instruction that affects the EL1 translation regime while
4155 	 * we're running in a context that doesn't allow us to let the
4156 	 * HW do its thing (aka vEL2):
4157 	 *
4158 	 * - HCR_EL2.E2H == 0 : a non-VHE guest
4159 	 * - HCR_EL2.{E2H,TGE} == { 1, 0 } : a VHE guest in guest mode
4160 	 *
4161 	 * Another possibility is that we are invalidating the EL2 context
4162 	 * using EL1 instructions, but that we landed here because we need
4163 	 * additional invalidation for structures that are not held in the
4164 	 * CPU TLBs (such as the VNCR pseudo-TLB and its EL2 mapping). In
4165 	 * that case, we are guaranteed that HCR_EL2.{E2H,TGE} == { 1, 1 }
4166 	 * as we don't allow an NV-capable L1 in a nVHE configuration.
4167 	 *
4168 	 * We don't expect these helpers to ever be called when running
4169 	 * in a vEL1 context.
4170 	 */
4171 
4172 	WARN_ON(!vcpu_is_el2(vcpu));
4173 
4174 	if (!kvm_supported_tlbi_s1e1_op(vcpu, sys_encoding))
4175 		return undef_access(vcpu, p, r);
4176 
4177 	if (vcpu_el2_e2h_is_set(vcpu) && vcpu_el2_tge_is_set(vcpu)) {
4178 		kvm_handle_s1e2_tlbi(vcpu, sys_encoding, p->regval);
4179 		return true;
4180 	}
4181 
4182 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm,
4183 				   get_vmid(__vcpu_sys_reg(vcpu, VTTBR_EL2)),
4184 				   &(union tlbi_info) {
4185 					   .va = {
4186 						   .addr = p->regval,
4187 						   .encoding = sys_encoding,
4188 					   },
4189 				   },
4190 				   s2_mmu_tlbi_s1e1);
4191 
4192 	return true;
4193 }
4194 
4195 #define SYS_INSN(insn, access_fn)					\
4196 	{								\
4197 		SYS_DESC(OP_##insn),					\
4198 		.access = (access_fn),					\
4199 	}
4200 
4201 static struct sys_reg_desc sys_insn_descs[] = {
4202 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
4203 	{ SYS_DESC(SYS_DC_IGSW), access_dcgsw },
4204 	{ SYS_DESC(SYS_DC_IGDSW), access_dcgsw },
4205 
4206 	SYS_INSN(AT_S1E1R, handle_at_s1e01),
4207 	SYS_INSN(AT_S1E1W, handle_at_s1e01),
4208 	SYS_INSN(AT_S1E0R, handle_at_s1e01),
4209 	SYS_INSN(AT_S1E0W, handle_at_s1e01),
4210 	SYS_INSN(AT_S1E1RP, handle_at_s1e01),
4211 	SYS_INSN(AT_S1E1WP, handle_at_s1e01),
4212 	SYS_INSN(AT_S1E1A, handle_at_s1e01),
4213 
4214 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
4215 	{ SYS_DESC(SYS_DC_CGSW), access_dcgsw },
4216 	{ SYS_DESC(SYS_DC_CGDSW), access_dcgsw },
4217 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
4218 	{ SYS_DESC(SYS_DC_CIGSW), access_dcgsw },
4219 	{ SYS_DESC(SYS_DC_CIGDSW), access_dcgsw },
4220 
4221 	SYS_INSN(TLBI_VMALLE1OS, handle_tlbi_el1),
4222 	SYS_INSN(TLBI_VAE1OS, handle_tlbi_el1),
4223 	SYS_INSN(TLBI_ASIDE1OS, handle_tlbi_el1),
4224 	SYS_INSN(TLBI_VAAE1OS, handle_tlbi_el1),
4225 	SYS_INSN(TLBI_VALE1OS, handle_tlbi_el1),
4226 	SYS_INSN(TLBI_VAALE1OS, handle_tlbi_el1),
4227 
4228 	SYS_INSN(TLBI_RVAE1IS, handle_tlbi_el1),
4229 	SYS_INSN(TLBI_RVAAE1IS, handle_tlbi_el1),
4230 	SYS_INSN(TLBI_RVALE1IS, handle_tlbi_el1),
4231 	SYS_INSN(TLBI_RVAALE1IS, handle_tlbi_el1),
4232 
4233 	SYS_INSN(TLBI_VMALLE1IS, handle_tlbi_el1),
4234 	SYS_INSN(TLBI_VAE1IS, handle_tlbi_el1),
4235 	SYS_INSN(TLBI_ASIDE1IS, handle_tlbi_el1),
4236 	SYS_INSN(TLBI_VAAE1IS, handle_tlbi_el1),
4237 	SYS_INSN(TLBI_VALE1IS, handle_tlbi_el1),
4238 	SYS_INSN(TLBI_VAALE1IS, handle_tlbi_el1),
4239 
4240 	SYS_INSN(TLBI_RVAE1OS, handle_tlbi_el1),
4241 	SYS_INSN(TLBI_RVAAE1OS, handle_tlbi_el1),
4242 	SYS_INSN(TLBI_RVALE1OS, handle_tlbi_el1),
4243 	SYS_INSN(TLBI_RVAALE1OS, handle_tlbi_el1),
4244 
4245 	SYS_INSN(TLBI_RVAE1, handle_tlbi_el1),
4246 	SYS_INSN(TLBI_RVAAE1, handle_tlbi_el1),
4247 	SYS_INSN(TLBI_RVALE1, handle_tlbi_el1),
4248 	SYS_INSN(TLBI_RVAALE1, handle_tlbi_el1),
4249 
4250 	SYS_INSN(TLBI_VMALLE1, handle_tlbi_el1),
4251 	SYS_INSN(TLBI_VAE1, handle_tlbi_el1),
4252 	SYS_INSN(TLBI_ASIDE1, handle_tlbi_el1),
4253 	SYS_INSN(TLBI_VAAE1, handle_tlbi_el1),
4254 	SYS_INSN(TLBI_VALE1, handle_tlbi_el1),
4255 	SYS_INSN(TLBI_VAALE1, handle_tlbi_el1),
4256 
4257 	SYS_INSN(TLBI_VMALLE1OSNXS, handle_tlbi_el1),
4258 	SYS_INSN(TLBI_VAE1OSNXS, handle_tlbi_el1),
4259 	SYS_INSN(TLBI_ASIDE1OSNXS, handle_tlbi_el1),
4260 	SYS_INSN(TLBI_VAAE1OSNXS, handle_tlbi_el1),
4261 	SYS_INSN(TLBI_VALE1OSNXS, handle_tlbi_el1),
4262 	SYS_INSN(TLBI_VAALE1OSNXS, handle_tlbi_el1),
4263 
4264 	SYS_INSN(TLBI_RVAE1ISNXS, handle_tlbi_el1),
4265 	SYS_INSN(TLBI_RVAAE1ISNXS, handle_tlbi_el1),
4266 	SYS_INSN(TLBI_RVALE1ISNXS, handle_tlbi_el1),
4267 	SYS_INSN(TLBI_RVAALE1ISNXS, handle_tlbi_el1),
4268 
4269 	SYS_INSN(TLBI_VMALLE1ISNXS, handle_tlbi_el1),
4270 	SYS_INSN(TLBI_VAE1ISNXS, handle_tlbi_el1),
4271 	SYS_INSN(TLBI_ASIDE1ISNXS, handle_tlbi_el1),
4272 	SYS_INSN(TLBI_VAAE1ISNXS, handle_tlbi_el1),
4273 	SYS_INSN(TLBI_VALE1ISNXS, handle_tlbi_el1),
4274 	SYS_INSN(TLBI_VAALE1ISNXS, handle_tlbi_el1),
4275 
4276 	SYS_INSN(TLBI_RVAE1OSNXS, handle_tlbi_el1),
4277 	SYS_INSN(TLBI_RVAAE1OSNXS, handle_tlbi_el1),
4278 	SYS_INSN(TLBI_RVALE1OSNXS, handle_tlbi_el1),
4279 	SYS_INSN(TLBI_RVAALE1OSNXS, handle_tlbi_el1),
4280 
4281 	SYS_INSN(TLBI_RVAE1NXS, handle_tlbi_el1),
4282 	SYS_INSN(TLBI_RVAAE1NXS, handle_tlbi_el1),
4283 	SYS_INSN(TLBI_RVALE1NXS, handle_tlbi_el1),
4284 	SYS_INSN(TLBI_RVAALE1NXS, handle_tlbi_el1),
4285 
4286 	SYS_INSN(TLBI_VMALLE1NXS, handle_tlbi_el1),
4287 	SYS_INSN(TLBI_VAE1NXS, handle_tlbi_el1),
4288 	SYS_INSN(TLBI_ASIDE1NXS, handle_tlbi_el1),
4289 	SYS_INSN(TLBI_VAAE1NXS, handle_tlbi_el1),
4290 	SYS_INSN(TLBI_VALE1NXS, handle_tlbi_el1),
4291 	SYS_INSN(TLBI_VAALE1NXS, handle_tlbi_el1),
4292 
4293 	SYS_INSN(AT_S1E2R, handle_at_s1e2),
4294 	SYS_INSN(AT_S1E2W, handle_at_s1e2),
4295 	SYS_INSN(AT_S12E1R, handle_at_s12),
4296 	SYS_INSN(AT_S12E1W, handle_at_s12),
4297 	SYS_INSN(AT_S12E0R, handle_at_s12),
4298 	SYS_INSN(AT_S12E0W, handle_at_s12),
4299 	SYS_INSN(AT_S1E2A, handle_at_s1e2),
4300 
4301 	SYS_INSN(TLBI_IPAS2E1IS, handle_ipas2e1is),
4302 	SYS_INSN(TLBI_RIPAS2E1IS, handle_ripas2e1is),
4303 	SYS_INSN(TLBI_IPAS2LE1IS, handle_ipas2e1is),
4304 	SYS_INSN(TLBI_RIPAS2LE1IS, handle_ripas2e1is),
4305 
4306 	SYS_INSN(TLBI_ALLE2OS, handle_tlbi_el2),
4307 	SYS_INSN(TLBI_VAE2OS, handle_tlbi_el2),
4308 	SYS_INSN(TLBI_ALLE1OS, handle_alle1is),
4309 	SYS_INSN(TLBI_VALE2OS, handle_tlbi_el2),
4310 	SYS_INSN(TLBI_VMALLS12E1OS, handle_vmalls12e1is),
4311 
4312 	SYS_INSN(TLBI_RVAE2IS, handle_tlbi_el2),
4313 	SYS_INSN(TLBI_RVALE2IS, handle_tlbi_el2),
4314 	SYS_INSN(TLBI_ALLE2IS, handle_tlbi_el2),
4315 	SYS_INSN(TLBI_VAE2IS, handle_tlbi_el2),
4316 
4317 	SYS_INSN(TLBI_ALLE1IS, handle_alle1is),
4318 
4319 	SYS_INSN(TLBI_VALE2IS, handle_tlbi_el2),
4320 
4321 	SYS_INSN(TLBI_VMALLS12E1IS, handle_vmalls12e1is),
4322 	SYS_INSN(TLBI_IPAS2E1OS, handle_ipas2e1is),
4323 	SYS_INSN(TLBI_IPAS2E1, handle_ipas2e1is),
4324 	SYS_INSN(TLBI_RIPAS2E1, handle_ripas2e1is),
4325 	SYS_INSN(TLBI_RIPAS2E1OS, handle_ripas2e1is),
4326 	SYS_INSN(TLBI_IPAS2LE1OS, handle_ipas2e1is),
4327 	SYS_INSN(TLBI_IPAS2LE1, handle_ipas2e1is),
4328 	SYS_INSN(TLBI_RIPAS2LE1, handle_ripas2e1is),
4329 	SYS_INSN(TLBI_RIPAS2LE1OS, handle_ripas2e1is),
4330 	SYS_INSN(TLBI_RVAE2OS, handle_tlbi_el2),
4331 	SYS_INSN(TLBI_RVALE2OS, handle_tlbi_el2),
4332 	SYS_INSN(TLBI_RVAE2, handle_tlbi_el2),
4333 	SYS_INSN(TLBI_RVALE2, handle_tlbi_el2),
4334 	SYS_INSN(TLBI_ALLE2, handle_tlbi_el2),
4335 	SYS_INSN(TLBI_VAE2, handle_tlbi_el2),
4336 
4337 	SYS_INSN(TLBI_ALLE1, handle_alle1is),
4338 
4339 	SYS_INSN(TLBI_VALE2, handle_tlbi_el2),
4340 
4341 	SYS_INSN(TLBI_VMALLS12E1, handle_vmalls12e1is),
4342 
4343 	SYS_INSN(TLBI_IPAS2E1ISNXS, handle_ipas2e1is),
4344 	SYS_INSN(TLBI_RIPAS2E1ISNXS, handle_ripas2e1is),
4345 	SYS_INSN(TLBI_IPAS2LE1ISNXS, handle_ipas2e1is),
4346 	SYS_INSN(TLBI_RIPAS2LE1ISNXS, handle_ripas2e1is),
4347 
4348 	SYS_INSN(TLBI_ALLE2OSNXS, handle_tlbi_el2),
4349 	SYS_INSN(TLBI_VAE2OSNXS, handle_tlbi_el2),
4350 	SYS_INSN(TLBI_ALLE1OSNXS, handle_alle1is),
4351 	SYS_INSN(TLBI_VALE2OSNXS, handle_tlbi_el2),
4352 	SYS_INSN(TLBI_VMALLS12E1OSNXS, handle_vmalls12e1is),
4353 
4354 	SYS_INSN(TLBI_RVAE2ISNXS, handle_tlbi_el2),
4355 	SYS_INSN(TLBI_RVALE2ISNXS, handle_tlbi_el2),
4356 	SYS_INSN(TLBI_ALLE2ISNXS, handle_tlbi_el2),
4357 	SYS_INSN(TLBI_VAE2ISNXS, handle_tlbi_el2),
4358 
4359 	SYS_INSN(TLBI_ALLE1ISNXS, handle_alle1is),
4360 	SYS_INSN(TLBI_VALE2ISNXS, handle_tlbi_el2),
4361 	SYS_INSN(TLBI_VMALLS12E1ISNXS, handle_vmalls12e1is),
4362 	SYS_INSN(TLBI_IPAS2E1OSNXS, handle_ipas2e1is),
4363 	SYS_INSN(TLBI_IPAS2E1NXS, handle_ipas2e1is),
4364 	SYS_INSN(TLBI_RIPAS2E1NXS, handle_ripas2e1is),
4365 	SYS_INSN(TLBI_RIPAS2E1OSNXS, handle_ripas2e1is),
4366 	SYS_INSN(TLBI_IPAS2LE1OSNXS, handle_ipas2e1is),
4367 	SYS_INSN(TLBI_IPAS2LE1NXS, handle_ipas2e1is),
4368 	SYS_INSN(TLBI_RIPAS2LE1NXS, handle_ripas2e1is),
4369 	SYS_INSN(TLBI_RIPAS2LE1OSNXS, handle_ripas2e1is),
4370 	SYS_INSN(TLBI_RVAE2OSNXS, handle_tlbi_el2),
4371 	SYS_INSN(TLBI_RVALE2OSNXS, handle_tlbi_el2),
4372 	SYS_INSN(TLBI_RVAE2NXS, handle_tlbi_el2),
4373 	SYS_INSN(TLBI_RVALE2NXS, handle_tlbi_el2),
4374 	SYS_INSN(TLBI_ALLE2NXS, handle_tlbi_el2),
4375 	SYS_INSN(TLBI_VAE2NXS, handle_tlbi_el2),
4376 	SYS_INSN(TLBI_ALLE1NXS, handle_alle1is),
4377 	SYS_INSN(TLBI_VALE2NXS, handle_tlbi_el2),
4378 	SYS_INSN(TLBI_VMALLS12E1NXS, handle_vmalls12e1is),
4379 };
4380 
4381 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
4382 			struct sys_reg_params *p,
4383 			const struct sys_reg_desc *r)
4384 {
4385 	if (p->is_write) {
4386 		return ignore_write(vcpu, p);
4387 	} else {
4388 		u64 dfr = kvm_read_vm_id_reg(vcpu->kvm, SYS_ID_AA64DFR0_EL1);
4389 		u32 el3 = kvm_has_feat(vcpu->kvm, ID_AA64PFR0_EL1, EL3, IMP);
4390 
4391 		p->regval = ((SYS_FIELD_GET(ID_AA64DFR0_EL1, WRPs, dfr) << 28) |
4392 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, BRPs, dfr) << 24) |
4393 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, CTX_CMPs, dfr) << 20) |
4394 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, dfr) << 16) |
4395 			     (1 << 15) | (el3 << 14) | (el3 << 12));
4396 		return true;
4397 	}
4398 }
4399 
4400 /*
4401  * AArch32 debug register mappings
4402  *
4403  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
4404  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
4405  *
4406  * None of the other registers share their location, so treat them as
4407  * if they were 64bit.
4408  */
4409 #define DBG_BCR_BVR_WCR_WVR(n)							\
4410 	/* DBGBVRn */								\
4411 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4),			\
4412 	  trap_dbg_wb_reg, NULL, n },						\
4413 	/* DBGBCRn */								\
4414 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_dbg_wb_reg, NULL, n },	\
4415 	/* DBGWVRn */								\
4416 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_dbg_wb_reg, NULL, n },	\
4417 	/* DBGWCRn */								\
4418 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_dbg_wb_reg, NULL, n }
4419 
4420 #define DBGBXVR(n)								\
4421 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1),			\
4422 	  trap_dbg_wb_reg, NULL, n }
4423 
4424 /*
4425  * Trapped cp14 registers. We generally ignore most of the external
4426  * debug, on the principle that they don't really make sense to a
4427  * guest. Revisit this one day, would this principle change.
4428  */
4429 static const struct sys_reg_desc cp14_regs[] = {
4430 	/* DBGDIDR */
4431 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
4432 	/* DBGDTRRXext */
4433 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
4434 
4435 	DBG_BCR_BVR_WCR_WVR(0),
4436 	/* DBGDSCRint */
4437 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
4438 	DBG_BCR_BVR_WCR_WVR(1),
4439 	/* DBGDCCINT */
4440 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
4441 	/* DBGDSCRext */
4442 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
4443 	DBG_BCR_BVR_WCR_WVR(2),
4444 	/* DBGDTR[RT]Xint */
4445 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
4446 	/* DBGDTR[RT]Xext */
4447 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
4448 	DBG_BCR_BVR_WCR_WVR(3),
4449 	DBG_BCR_BVR_WCR_WVR(4),
4450 	DBG_BCR_BVR_WCR_WVR(5),
4451 	/* DBGWFAR */
4452 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
4453 	/* DBGOSECCR */
4454 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
4455 	DBG_BCR_BVR_WCR_WVR(6),
4456 	/* DBGVCR */
4457 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
4458 	DBG_BCR_BVR_WCR_WVR(7),
4459 	DBG_BCR_BVR_WCR_WVR(8),
4460 	DBG_BCR_BVR_WCR_WVR(9),
4461 	DBG_BCR_BVR_WCR_WVR(10),
4462 	DBG_BCR_BVR_WCR_WVR(11),
4463 	DBG_BCR_BVR_WCR_WVR(12),
4464 	DBG_BCR_BVR_WCR_WVR(13),
4465 	DBG_BCR_BVR_WCR_WVR(14),
4466 	DBG_BCR_BVR_WCR_WVR(15),
4467 
4468 	/* DBGDRAR (32bit) */
4469 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
4470 
4471 	DBGBXVR(0),
4472 	/* DBGOSLAR */
4473 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
4474 	DBGBXVR(1),
4475 	/* DBGOSLSR */
4476 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
4477 	DBGBXVR(2),
4478 	DBGBXVR(3),
4479 	/* DBGOSDLR */
4480 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
4481 	DBGBXVR(4),
4482 	/* DBGPRCR */
4483 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
4484 	DBGBXVR(5),
4485 	DBGBXVR(6),
4486 	DBGBXVR(7),
4487 	DBGBXVR(8),
4488 	DBGBXVR(9),
4489 	DBGBXVR(10),
4490 	DBGBXVR(11),
4491 	DBGBXVR(12),
4492 	DBGBXVR(13),
4493 	DBGBXVR(14),
4494 	DBGBXVR(15),
4495 
4496 	/* DBGDSAR (32bit) */
4497 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
4498 
4499 	/* DBGDEVID2 */
4500 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
4501 	/* DBGDEVID1 */
4502 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
4503 	/* DBGDEVID */
4504 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
4505 	/* DBGCLAIMSET */
4506 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
4507 	/* DBGCLAIMCLR */
4508 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
4509 	/* DBGAUTHSTATUS */
4510 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
4511 };
4512 
4513 /* Trapped cp14 64bit registers */
4514 static const struct sys_reg_desc cp14_64_regs[] = {
4515 	/* DBGDRAR (64bit) */
4516 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
4517 
4518 	/* DBGDSAR (64bit) */
4519 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
4520 };
4521 
4522 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)			\
4523 	AA32(_map),							\
4524 	Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),			\
4525 	.visibility = pmu_visibility
4526 
4527 /* Macro to expand the PMEVCNTRn register */
4528 #define PMU_PMEVCNTR(n)							\
4529 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
4530 	  (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
4531 	  .access = access_pmu_evcntr }
4532 
4533 /* Macro to expand the PMEVTYPERn register */
4534 #define PMU_PMEVTYPER(n)						\
4535 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
4536 	  (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
4537 	  .access = access_pmu_evtyper }
4538 /*
4539  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
4540  * depending on the way they are accessed (as a 32bit or a 64bit
4541  * register).
4542  */
4543 static const struct sys_reg_desc cp15_regs[] = {
4544 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
4545 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
4546 	/* ACTLR */
4547 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
4548 	/* ACTLR2 */
4549 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
4550 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
4551 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
4552 	/* TTBCR */
4553 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
4554 	/* TTBCR2 */
4555 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
4556 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
4557 	{ CP15_SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
4558 	/* DFSR */
4559 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
4560 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
4561 	/* ADFSR */
4562 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
4563 	/* AIFSR */
4564 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
4565 	/* DFAR */
4566 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
4567 	/* IFAR */
4568 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
4569 
4570 	/*
4571 	 * DC{C,I,CI}SW operations:
4572 	 */
4573 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
4574 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
4575 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
4576 
4577 	/* PMU */
4578 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
4579 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
4580 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
4581 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
4582 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
4583 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
4584 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
4585 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
4586 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
4587 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
4588 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
4589 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
4590 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
4591 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
4592 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
4593 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
4594 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
4595 	/* PMMIR */
4596 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
4597 
4598 	/* PRRR/MAIR0 */
4599 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
4600 	/* NMRR/MAIR1 */
4601 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
4602 	/* AMAIR0 */
4603 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
4604 	/* AMAIR1 */
4605 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
4606 
4607 	{ CP15_SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
4608 	{ CP15_SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
4609 	{ CP15_SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
4610 	{ CP15_SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
4611 	{ CP15_SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
4612 	{ CP15_SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
4613 	{ CP15_SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
4614 	{ CP15_SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
4615 	{ CP15_SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
4616 	{ CP15_SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
4617 	{ CP15_SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
4618 	{ CP15_SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
4619 	{ CP15_SYS_DESC(SYS_ICC_DIR_EL1), access_gic_dir },
4620 	{ CP15_SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
4621 	{ CP15_SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
4622 	{ CP15_SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
4623 	{ CP15_SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
4624 	{ CP15_SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
4625 	{ CP15_SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
4626 	{ CP15_SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
4627 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
4628 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
4629 
4630 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
4631 
4632 	/* Arch Tmers */
4633 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
4634 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
4635 
4636 	/* PMEVCNTRn */
4637 	PMU_PMEVCNTR(0),
4638 	PMU_PMEVCNTR(1),
4639 	PMU_PMEVCNTR(2),
4640 	PMU_PMEVCNTR(3),
4641 	PMU_PMEVCNTR(4),
4642 	PMU_PMEVCNTR(5),
4643 	PMU_PMEVCNTR(6),
4644 	PMU_PMEVCNTR(7),
4645 	PMU_PMEVCNTR(8),
4646 	PMU_PMEVCNTR(9),
4647 	PMU_PMEVCNTR(10),
4648 	PMU_PMEVCNTR(11),
4649 	PMU_PMEVCNTR(12),
4650 	PMU_PMEVCNTR(13),
4651 	PMU_PMEVCNTR(14),
4652 	PMU_PMEVCNTR(15),
4653 	PMU_PMEVCNTR(16),
4654 	PMU_PMEVCNTR(17),
4655 	PMU_PMEVCNTR(18),
4656 	PMU_PMEVCNTR(19),
4657 	PMU_PMEVCNTR(20),
4658 	PMU_PMEVCNTR(21),
4659 	PMU_PMEVCNTR(22),
4660 	PMU_PMEVCNTR(23),
4661 	PMU_PMEVCNTR(24),
4662 	PMU_PMEVCNTR(25),
4663 	PMU_PMEVCNTR(26),
4664 	PMU_PMEVCNTR(27),
4665 	PMU_PMEVCNTR(28),
4666 	PMU_PMEVCNTR(29),
4667 	PMU_PMEVCNTR(30),
4668 	/* PMEVTYPERn */
4669 	PMU_PMEVTYPER(0),
4670 	PMU_PMEVTYPER(1),
4671 	PMU_PMEVTYPER(2),
4672 	PMU_PMEVTYPER(3),
4673 	PMU_PMEVTYPER(4),
4674 	PMU_PMEVTYPER(5),
4675 	PMU_PMEVTYPER(6),
4676 	PMU_PMEVTYPER(7),
4677 	PMU_PMEVTYPER(8),
4678 	PMU_PMEVTYPER(9),
4679 	PMU_PMEVTYPER(10),
4680 	PMU_PMEVTYPER(11),
4681 	PMU_PMEVTYPER(12),
4682 	PMU_PMEVTYPER(13),
4683 	PMU_PMEVTYPER(14),
4684 	PMU_PMEVTYPER(15),
4685 	PMU_PMEVTYPER(16),
4686 	PMU_PMEVTYPER(17),
4687 	PMU_PMEVTYPER(18),
4688 	PMU_PMEVTYPER(19),
4689 	PMU_PMEVTYPER(20),
4690 	PMU_PMEVTYPER(21),
4691 	PMU_PMEVTYPER(22),
4692 	PMU_PMEVTYPER(23),
4693 	PMU_PMEVTYPER(24),
4694 	PMU_PMEVTYPER(25),
4695 	PMU_PMEVTYPER(26),
4696 	PMU_PMEVTYPER(27),
4697 	PMU_PMEVTYPER(28),
4698 	PMU_PMEVTYPER(29),
4699 	PMU_PMEVTYPER(30),
4700 	/* PMCCFILTR */
4701 	{ CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
4702 
4703 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
4704 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
4705 
4706 	/* CCSIDR2 */
4707 	{ Op1(1), CRn( 0), CRm( 0),  Op2(2), undef_access },
4708 
4709 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
4710 };
4711 
4712 static const struct sys_reg_desc cp15_64_regs[] = {
4713 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
4714 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
4715 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
4716 	{ SYS_DESC(SYS_AARCH32_CNTPCT),	      access_arch_timer },
4717 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
4718 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
4719 	{ SYS_DESC(SYS_AARCH32_CNTVCT),	      access_arch_timer },
4720 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
4721 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
4722 	{ SYS_DESC(SYS_AARCH32_CNTPCTSS),     access_arch_timer },
4723 	{ SYS_DESC(SYS_AARCH32_CNTVCTSS),     access_arch_timer },
4724 };
4725 
4726 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
4727 			       bool reset_check)
4728 {
4729 	unsigned int i;
4730 
4731 	for (i = 0; i < n; i++) {
4732 		if (reset_check && table[i].reg && !table[i].reset) {
4733 			kvm_err("sys_reg table %pS entry %d (%s) lacks reset\n",
4734 				&table[i], i, table[i].name);
4735 			return false;
4736 		}
4737 
4738 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
4739 			kvm_err("sys_reg table %pS entry %d (%s -> %s) out of order\n",
4740 				&table[i], i, table[i - 1].name, table[i].name);
4741 			return false;
4742 		}
4743 	}
4744 
4745 	return true;
4746 }
4747 
4748 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
4749 {
4750 	kvm_inject_undefined(vcpu);
4751 	return 1;
4752 }
4753 
4754 static void perform_access(struct kvm_vcpu *vcpu,
4755 			   struct sys_reg_params *params,
4756 			   const struct sys_reg_desc *r)
4757 {
4758 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
4759 
4760 	/* Check for regs disabled by runtime config */
4761 	if (sysreg_hidden(vcpu, r)) {
4762 		kvm_inject_undefined(vcpu);
4763 		return;
4764 	}
4765 
4766 	/*
4767 	 * Not having an accessor means that we have configured a trap
4768 	 * that we don't know how to handle. This certainly qualifies
4769 	 * as a gross bug that should be fixed right away.
4770 	 */
4771 	if (!r->access) {
4772 		bad_trap(vcpu, params, r, "register access");
4773 		return;
4774 	}
4775 
4776 	/* Skip instruction if instructed so */
4777 	if (likely(r->access(vcpu, params, r)))
4778 		kvm_incr_pc(vcpu);
4779 }
4780 
4781 /*
4782  * emulate_cp --  tries to match a sys_reg access in a handling table, and
4783  *                call the corresponding trap handler.
4784  *
4785  * @params: pointer to the descriptor of the access
4786  * @table: array of trap descriptors
4787  * @num: size of the trap descriptor array
4788  *
4789  * Return true if the access has been handled, false if not.
4790  */
4791 static bool emulate_cp(struct kvm_vcpu *vcpu,
4792 		       struct sys_reg_params *params,
4793 		       const struct sys_reg_desc *table,
4794 		       size_t num)
4795 {
4796 	const struct sys_reg_desc *r;
4797 
4798 	if (!table)
4799 		return false;	/* Not handled */
4800 
4801 	r = find_reg(params, table, num);
4802 
4803 	if (r) {
4804 		perform_access(vcpu, params, r);
4805 		return true;
4806 	}
4807 
4808 	/* Not handled */
4809 	return false;
4810 }
4811 
4812 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
4813 				struct sys_reg_params *params)
4814 {
4815 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
4816 	int cp = -1;
4817 
4818 	switch (esr_ec) {
4819 	case ESR_ELx_EC_CP15_32:
4820 	case ESR_ELx_EC_CP15_64:
4821 		cp = 15;
4822 		break;
4823 	case ESR_ELx_EC_CP14_MR:
4824 	case ESR_ELx_EC_CP14_64:
4825 		cp = 14;
4826 		break;
4827 	default:
4828 		WARN_ON(1);
4829 	}
4830 
4831 	print_sys_reg_msg(params,
4832 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
4833 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
4834 	kvm_inject_undefined(vcpu);
4835 }
4836 
4837 /**
4838  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
4839  * @vcpu: The VCPU pointer
4840  * @global: &struct sys_reg_desc
4841  * @nr_global: size of the @global array
4842  */
4843 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
4844 			    const struct sys_reg_desc *global,
4845 			    size_t nr_global)
4846 {
4847 	struct sys_reg_params params;
4848 	u64 esr = kvm_vcpu_get_esr(vcpu);
4849 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4850 	int Rt2 = (esr >> 10) & 0x1f;
4851 
4852 	params.CRm = (esr >> 1) & 0xf;
4853 	params.is_write = ((esr & 1) == 0);
4854 
4855 	params.Op0 = 0;
4856 	params.Op1 = (esr >> 16) & 0xf;
4857 	params.Op2 = 0;
4858 	params.CRn = 0;
4859 
4860 	/*
4861 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
4862 	 * backends between AArch32 and AArch64, we get away with it.
4863 	 */
4864 	if (params.is_write) {
4865 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
4866 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
4867 	}
4868 
4869 	/*
4870 	 * If the table contains a handler, handle the
4871 	 * potential register operation in the case of a read and return
4872 	 * with success.
4873 	 */
4874 	if (emulate_cp(vcpu, &params, global, nr_global)) {
4875 		/* Split up the value between registers for the read side */
4876 		if (!params.is_write) {
4877 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
4878 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
4879 		}
4880 
4881 		return 1;
4882 	}
4883 
4884 	unhandled_cp_access(vcpu, &params);
4885 	return 1;
4886 }
4887 
4888 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
4889 
4890 /*
4891  * The CP10 ID registers are architecturally mapped to AArch64 feature
4892  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
4893  * from AArch32.
4894  */
4895 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
4896 {
4897 	u8 reg_id = (esr >> 10) & 0xf;
4898 	bool valid;
4899 
4900 	params->is_write = ((esr & 1) == 0);
4901 	params->Op0 = 3;
4902 	params->Op1 = 0;
4903 	params->CRn = 0;
4904 	params->CRm = 3;
4905 
4906 	/* CP10 ID registers are read-only */
4907 	valid = !params->is_write;
4908 
4909 	switch (reg_id) {
4910 	/* MVFR0 */
4911 	case 0b0111:
4912 		params->Op2 = 0;
4913 		break;
4914 	/* MVFR1 */
4915 	case 0b0110:
4916 		params->Op2 = 1;
4917 		break;
4918 	/* MVFR2 */
4919 	case 0b0101:
4920 		params->Op2 = 2;
4921 		break;
4922 	default:
4923 		valid = false;
4924 	}
4925 
4926 	if (valid)
4927 		return true;
4928 
4929 	kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
4930 		      str_write_read(params->is_write), reg_id);
4931 	return false;
4932 }
4933 
4934 /**
4935  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
4936  *			  VFP Register' from AArch32.
4937  * @vcpu: The vCPU pointer
4938  *
4939  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
4940  * Work out the correct AArch64 system register encoding and reroute to the
4941  * AArch64 system register emulation.
4942  */
4943 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
4944 {
4945 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4946 	u64 esr = kvm_vcpu_get_esr(vcpu);
4947 	struct sys_reg_params params;
4948 
4949 	/* UNDEF on any unhandled register access */
4950 	if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
4951 		kvm_inject_undefined(vcpu);
4952 		return 1;
4953 	}
4954 
4955 	if (emulate_sys_reg(vcpu, &params))
4956 		vcpu_set_reg(vcpu, Rt, params.regval);
4957 
4958 	return 1;
4959 }
4960 
4961 /**
4962  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
4963  *			       CRn=0, which corresponds to the AArch32 feature
4964  *			       registers.
4965  * @vcpu: the vCPU pointer
4966  * @params: the system register access parameters.
4967  *
4968  * Our cp15 system register tables do not enumerate the AArch32 feature
4969  * registers. Conveniently, our AArch64 table does, and the AArch32 system
4970  * register encoding can be trivially remapped into the AArch64 for the feature
4971  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
4972  *
4973  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
4974  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
4975  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
4976  * treat undefined registers in this range as RAZ.
4977  */
4978 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
4979 				   struct sys_reg_params *params)
4980 {
4981 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4982 
4983 	/* Treat impossible writes to RO registers as UNDEFINED */
4984 	if (params->is_write) {
4985 		unhandled_cp_access(vcpu, params);
4986 		return 1;
4987 	}
4988 
4989 	params->Op0 = 3;
4990 
4991 	/*
4992 	 * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
4993 	 * Avoid conflicting with future expansion of AArch64 feature registers
4994 	 * and simply treat them as RAZ here.
4995 	 */
4996 	if (params->CRm > 3)
4997 		params->regval = 0;
4998 	else if (!emulate_sys_reg(vcpu, params))
4999 		return 1;
5000 
5001 	vcpu_set_reg(vcpu, Rt, params->regval);
5002 	return 1;
5003 }
5004 
5005 /**
5006  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
5007  * @vcpu: The VCPU pointer
5008  * @params: &struct sys_reg_params
5009  * @global: &struct sys_reg_desc
5010  * @nr_global: size of the @global array
5011  */
5012 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
5013 			    struct sys_reg_params *params,
5014 			    const struct sys_reg_desc *global,
5015 			    size_t nr_global)
5016 {
5017 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
5018 
5019 	params->regval = vcpu_get_reg(vcpu, Rt);
5020 
5021 	if (emulate_cp(vcpu, params, global, nr_global)) {
5022 		if (!params->is_write)
5023 			vcpu_set_reg(vcpu, Rt, params->regval);
5024 		return 1;
5025 	}
5026 
5027 	unhandled_cp_access(vcpu, params);
5028 	return 1;
5029 }
5030 
5031 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
5032 {
5033 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
5034 }
5035 
5036 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
5037 {
5038 	struct sys_reg_params params;
5039 
5040 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
5041 
5042 	/*
5043 	 * Certain AArch32 ID registers are handled by rerouting to the AArch64
5044 	 * system register table. Registers in the ID range where CRm=0 are
5045 	 * excluded from this scheme as they do not trivially map into AArch64
5046 	 * system register encodings, except for AIDR/REVIDR.
5047 	 */
5048 	if (params.Op1 == 0 && params.CRn == 0 &&
5049 	    (params.CRm || params.Op2 == 6 /* REVIDR */))
5050 		return kvm_emulate_cp15_id_reg(vcpu, &params);
5051 	if (params.Op1 == 1 && params.CRn == 0 &&
5052 	    params.CRm == 0 && params.Op2 == 7 /* AIDR */)
5053 		return kvm_emulate_cp15_id_reg(vcpu, &params);
5054 
5055 	return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
5056 }
5057 
5058 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
5059 {
5060 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
5061 }
5062 
5063 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
5064 {
5065 	struct sys_reg_params params;
5066 
5067 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
5068 
5069 	return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
5070 }
5071 
5072 /**
5073  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
5074  * @vcpu: The VCPU pointer
5075  * @params: Decoded system register parameters
5076  *
5077  * Return: true if the system register access was successful, false otherwise.
5078  */
5079 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
5080 			    struct sys_reg_params *params)
5081 {
5082 	const struct sys_reg_desc *r;
5083 
5084 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
5085 	if (likely(r)) {
5086 		perform_access(vcpu, params, r);
5087 		return true;
5088 	}
5089 
5090 	print_sys_reg_msg(params,
5091 			  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
5092 			  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
5093 	kvm_inject_undefined(vcpu);
5094 
5095 	return false;
5096 }
5097 
5098 static const struct sys_reg_desc *idregs_debug_find(struct kvm *kvm, loff_t pos)
5099 {
5100 	unsigned long i, idreg_idx = 0;
5101 
5102 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
5103 		const struct sys_reg_desc *r = &sys_reg_descs[i];
5104 
5105 		if (!is_vm_ftr_id_reg(reg_to_encoding(r)))
5106 			continue;
5107 
5108 		if (idreg_idx++ == pos)
5109 			return r;
5110 	}
5111 
5112 	return NULL;
5113 }
5114 
5115 static void *idregs_debug_start(struct seq_file *s, loff_t *pos)
5116 {
5117 	struct kvm *kvm = s->private;
5118 
5119 	if (!test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags))
5120 		return NULL;
5121 
5122 	return (void *)idregs_debug_find(kvm, *pos);
5123 }
5124 
5125 static void *idregs_debug_next(struct seq_file *s, void *v, loff_t *pos)
5126 {
5127 	struct kvm *kvm = s->private;
5128 
5129 	(*pos)++;
5130 
5131 	return (void *)idregs_debug_find(kvm, *pos);
5132 }
5133 
5134 static void idregs_debug_stop(struct seq_file *s, void *v)
5135 {
5136 }
5137 
5138 static int idregs_debug_show(struct seq_file *s, void *v)
5139 {
5140 	const struct sys_reg_desc *desc = v;
5141 	struct kvm *kvm = s->private;
5142 
5143 	if (!desc)
5144 		return 0;
5145 
5146 	seq_printf(s, "%20s:\t%016llx\n",
5147 		   desc->name, kvm_read_vm_id_reg(kvm, reg_to_encoding(desc)));
5148 
5149 	return 0;
5150 }
5151 
5152 static const struct seq_operations idregs_debug_sops = {
5153 	.start	= idregs_debug_start,
5154 	.next	= idregs_debug_next,
5155 	.stop	= idregs_debug_stop,
5156 	.show	= idregs_debug_show,
5157 };
5158 
5159 DEFINE_SEQ_ATTRIBUTE(idregs_debug);
5160 
5161 static const struct sys_reg_desc *sr_resx_find(struct kvm *kvm, loff_t pos)
5162 {
5163 	unsigned long i, sr_idx = 0;
5164 
5165 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
5166 		const struct sys_reg_desc *r = &sys_reg_descs[i];
5167 
5168 		if (r->reg < __SANITISED_REG_START__)
5169 			continue;
5170 
5171 		if (sr_idx++ == pos)
5172 			return r;
5173 	}
5174 
5175 	return NULL;
5176 }
5177 
5178 static void *sr_resx_start(struct seq_file *s, loff_t *pos)
5179 {
5180 	struct kvm *kvm = s->private;
5181 
5182 	if (!kvm->arch.sysreg_masks)
5183 		return NULL;
5184 
5185 	return (void *)sr_resx_find(kvm, *pos);
5186 }
5187 
5188 static void *sr_resx_next(struct seq_file *s, void *v, loff_t *pos)
5189 {
5190 	struct kvm *kvm = s->private;
5191 
5192 	(*pos)++;
5193 
5194 	return (void *)sr_resx_find(kvm, *pos);
5195 }
5196 
5197 static void sr_resx_stop(struct seq_file *s, void *v)
5198 {
5199 }
5200 
5201 static int sr_resx_show(struct seq_file *s, void *v)
5202 {
5203 	const struct sys_reg_desc *desc = v;
5204 	struct kvm *kvm = s->private;
5205 	struct resx resx;
5206 
5207 	if (!desc)
5208 		return 0;
5209 
5210 	resx = kvm_get_sysreg_resx(kvm, desc->reg);
5211 
5212 	seq_printf(s, "%20s:\tRES0:%016llx\tRES1:%016llx\n",
5213 		   desc->name, resx.res0, resx.res1);
5214 
5215 	return 0;
5216 }
5217 
5218 static const struct seq_operations sr_resx_sops = {
5219 	.start	= sr_resx_start,
5220 	.next	= sr_resx_next,
5221 	.stop	= sr_resx_stop,
5222 	.show	= sr_resx_show,
5223 };
5224 
5225 DEFINE_SEQ_ATTRIBUTE(sr_resx);
5226 
5227 void kvm_sys_regs_create_debugfs(struct kvm *kvm)
5228 {
5229 	debugfs_create_file("idregs", 0444, kvm->debugfs_dentry, kvm,
5230 			    &idregs_debug_fops);
5231 	debugfs_create_file("resx", 0444, kvm->debugfs_dentry, kvm,
5232 			    &sr_resx_fops);
5233 }
5234 
5235 static void reset_vm_ftr_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *reg)
5236 {
5237 	u32 id = reg_to_encoding(reg);
5238 	struct kvm *kvm = vcpu->kvm;
5239 
5240 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags))
5241 		return;
5242 
5243 	kvm_set_vm_id_reg(kvm, id, reg->reset(vcpu, reg));
5244 }
5245 
5246 static void reset_vcpu_ftr_id_reg(struct kvm_vcpu *vcpu,
5247 				  const struct sys_reg_desc *reg)
5248 {
5249 	if (kvm_vcpu_initialized(vcpu))
5250 		return;
5251 
5252 	reg->reset(vcpu, reg);
5253 }
5254 
5255 /**
5256  * kvm_reset_sys_regs - sets system registers to reset value
5257  * @vcpu: The VCPU pointer
5258  *
5259  * This function finds the right table above and sets the registers on the
5260  * virtual CPU struct to their architecturally defined reset values.
5261  */
5262 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
5263 {
5264 	struct kvm *kvm = vcpu->kvm;
5265 	unsigned long i;
5266 
5267 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
5268 		const struct sys_reg_desc *r = &sys_reg_descs[i];
5269 
5270 		if (!r->reset)
5271 			continue;
5272 
5273 		if (is_vm_ftr_id_reg(reg_to_encoding(r)))
5274 			reset_vm_ftr_id_reg(vcpu, r);
5275 		else if (is_vcpu_ftr_id_reg(reg_to_encoding(r)))
5276 			reset_vcpu_ftr_id_reg(vcpu, r);
5277 		else
5278 			r->reset(vcpu, r);
5279 
5280 		if (r->reg >= __SANITISED_REG_START__ && r->reg < NR_SYS_REGS)
5281 			__vcpu_rmw_sys_reg(vcpu, r->reg, |=, 0);
5282 	}
5283 
5284 	set_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags);
5285 
5286 	if (kvm_vcpu_has_pmu(vcpu))
5287 		kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
5288 }
5289 
5290 /**
5291  * kvm_handle_sys_reg -- handles a system instruction or mrs/msr instruction
5292  *			 trap on a guest execution
5293  * @vcpu: The VCPU pointer
5294  */
5295 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
5296 {
5297 	const struct sys_reg_desc *desc = NULL;
5298 	struct sys_reg_params params;
5299 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
5300 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
5301 	int sr_idx;
5302 
5303 	trace_kvm_handle_sys_reg(esr);
5304 
5305 	if (triage_sysreg_trap(vcpu, &sr_idx))
5306 		return 1;
5307 
5308 	params = esr_sys64_to_params(esr);
5309 	params.regval = vcpu_get_reg(vcpu, Rt);
5310 
5311 	/* System registers have Op0=={2,3}, as per DDI487 J.a C5.1.2 */
5312 	if (params.Op0 == 2 || params.Op0 == 3)
5313 		desc = &sys_reg_descs[sr_idx];
5314 	else
5315 		desc = &sys_insn_descs[sr_idx];
5316 
5317 	perform_access(vcpu, &params, desc);
5318 
5319 	/* Read from system register? */
5320 	if (!params.is_write &&
5321 	    (params.Op0 == 2 || params.Op0 == 3))
5322 		vcpu_set_reg(vcpu, Rt, params.regval);
5323 
5324 	return 1;
5325 }
5326 
5327 /******************************************************************************
5328  * Userspace API
5329  *****************************************************************************/
5330 
5331 static bool index_to_params(u64 id, struct sys_reg_params *params)
5332 {
5333 	switch (id & KVM_REG_SIZE_MASK) {
5334 	case KVM_REG_SIZE_U64:
5335 		/* Any unused index bits means it's not valid. */
5336 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
5337 			      | KVM_REG_ARM_COPROC_MASK
5338 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
5339 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
5340 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
5341 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
5342 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
5343 			return false;
5344 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
5345 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
5346 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
5347 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
5348 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
5349 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
5350 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
5351 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
5352 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
5353 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
5354 		return true;
5355 	default:
5356 		return false;
5357 	}
5358 }
5359 
5360 const struct sys_reg_desc *get_reg_by_id(u64 id,
5361 					 const struct sys_reg_desc table[],
5362 					 unsigned int num)
5363 {
5364 	struct sys_reg_params params;
5365 
5366 	if (!index_to_params(id, &params))
5367 		return NULL;
5368 
5369 	return find_reg(&params, table, num);
5370 }
5371 
5372 /* Decode an index value, and find the sys_reg_desc entry. */
5373 static const struct sys_reg_desc *
5374 id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
5375 		   const struct sys_reg_desc table[], unsigned int num)
5376 
5377 {
5378 	const struct sys_reg_desc *r;
5379 
5380 	/* We only do sys_reg for now. */
5381 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
5382 		return NULL;
5383 
5384 	r = get_reg_by_id(id, table, num);
5385 
5386 	/* Not saved in the sys_reg array and not otherwise accessible? */
5387 	if (r && (!(r->reg || r->get_user) || sysreg_hidden(vcpu, r)))
5388 		r = NULL;
5389 
5390 	return r;
5391 }
5392 
5393 static int demux_c15_get(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
5394 {
5395 	u32 val;
5396 	u32 __user *uval = uaddr;
5397 
5398 	/* Fail if we have unknown bits set. */
5399 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
5400 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
5401 		return -ENOENT;
5402 
5403 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
5404 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
5405 		if (KVM_REG_SIZE(id) != 4)
5406 			return -ENOENT;
5407 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
5408 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
5409 		if (val >= CSSELR_MAX)
5410 			return -ENOENT;
5411 
5412 		return put_user(get_ccsidr(vcpu, val), uval);
5413 	default:
5414 		return -ENOENT;
5415 	}
5416 }
5417 
5418 static int demux_c15_set(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
5419 {
5420 	u32 val, newval;
5421 	u32 __user *uval = uaddr;
5422 
5423 	/* Fail if we have unknown bits set. */
5424 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
5425 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
5426 		return -ENOENT;
5427 
5428 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
5429 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
5430 		if (KVM_REG_SIZE(id) != 4)
5431 			return -ENOENT;
5432 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
5433 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
5434 		if (val >= CSSELR_MAX)
5435 			return -ENOENT;
5436 
5437 		if (get_user(newval, uval))
5438 			return -EFAULT;
5439 
5440 		return set_ccsidr(vcpu, val, newval);
5441 	default:
5442 		return -ENOENT;
5443 	}
5444 }
5445 
5446 static u64 kvm_one_reg_to_id(const struct kvm_one_reg *reg)
5447 {
5448 	switch(reg->id) {
5449 	case KVM_REG_ARM_TIMER_CVAL:
5450 		return TO_ARM64_SYS_REG(CNTV_CVAL_EL0);
5451 	case KVM_REG_ARM_TIMER_CNT:
5452 		return TO_ARM64_SYS_REG(CNTVCT_EL0);
5453 	default:
5454 		return reg->id;
5455 	}
5456 }
5457 
5458 int kvm_sys_reg_get_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
5459 			 const struct sys_reg_desc table[], unsigned int num)
5460 {
5461 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
5462 	const struct sys_reg_desc *r;
5463 	u64 id = kvm_one_reg_to_id(reg);
5464 	u64 val;
5465 	int ret;
5466 
5467 	r = id_to_sys_reg_desc(vcpu, id, table, num);
5468 	if (!r || sysreg_hidden(vcpu, r))
5469 		return -ENOENT;
5470 
5471 	if (r->get_user) {
5472 		ret = (r->get_user)(vcpu, r, &val);
5473 	} else {
5474 		val = __vcpu_sys_reg(vcpu, r->reg);
5475 		ret = 0;
5476 	}
5477 
5478 	if (!ret)
5479 		ret = put_user(val, uaddr);
5480 
5481 	return ret;
5482 }
5483 
5484 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
5485 {
5486 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
5487 
5488 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
5489 		return demux_c15_get(vcpu, reg->id, uaddr);
5490 
5491 	return kvm_sys_reg_get_user(vcpu, reg,
5492 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
5493 }
5494 
5495 int kvm_sys_reg_set_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
5496 			 const struct sys_reg_desc table[], unsigned int num)
5497 {
5498 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
5499 	const struct sys_reg_desc *r;
5500 	u64 id = kvm_one_reg_to_id(reg);
5501 	u64 val;
5502 	int ret;
5503 
5504 	if (get_user(val, uaddr))
5505 		return -EFAULT;
5506 
5507 	r = id_to_sys_reg_desc(vcpu, id, table, num);
5508 	if (!r || sysreg_hidden(vcpu, r))
5509 		return -ENOENT;
5510 
5511 	if (sysreg_user_write_ignore(vcpu, r))
5512 		return 0;
5513 
5514 	if (r->set_user) {
5515 		ret = (r->set_user)(vcpu, r, val);
5516 	} else {
5517 		__vcpu_assign_sys_reg(vcpu, r->reg, val);
5518 		ret = 0;
5519 	}
5520 
5521 	return ret;
5522 }
5523 
5524 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
5525 {
5526 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
5527 
5528 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
5529 		return demux_c15_set(vcpu, reg->id, uaddr);
5530 
5531 	return kvm_sys_reg_set_user(vcpu, reg,
5532 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
5533 }
5534 
5535 static unsigned int num_demux_regs(void)
5536 {
5537 	return CSSELR_MAX;
5538 }
5539 
5540 static int write_demux_regids(u64 __user *uindices)
5541 {
5542 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
5543 	unsigned int i;
5544 
5545 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
5546 	for (i = 0; i < CSSELR_MAX; i++) {
5547 		if (put_user(val | i, uindices))
5548 			return -EFAULT;
5549 		uindices++;
5550 	}
5551 	return 0;
5552 }
5553 
5554 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
5555 {
5556 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
5557 		KVM_REG_ARM64_SYSREG |
5558 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
5559 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
5560 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
5561 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
5562 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
5563 }
5564 
5565 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
5566 {
5567 	u64 idx;
5568 
5569 	if (!*uind)
5570 		return true;
5571 
5572 	switch (reg_to_encoding(reg)) {
5573 	case SYS_CNTV_CVAL_EL0:
5574 		idx = KVM_REG_ARM_TIMER_CVAL;
5575 		break;
5576 	case SYS_CNTVCT_EL0:
5577 		idx = KVM_REG_ARM_TIMER_CNT;
5578 		break;
5579 	default:
5580 		idx = sys_reg_to_index(reg);
5581 	}
5582 
5583 	if (put_user(idx, *uind))
5584 		return false;
5585 
5586 	(*uind)++;
5587 	return true;
5588 }
5589 
5590 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
5591 			    const struct sys_reg_desc *rd,
5592 			    u64 __user **uind,
5593 			    unsigned int *total)
5594 {
5595 	/*
5596 	 * Ignore registers we trap but don't save,
5597 	 * and for which no custom user accessor is provided.
5598 	 */
5599 	if (!(rd->reg || rd->get_user))
5600 		return 0;
5601 
5602 	if (sysreg_hidden(vcpu, rd))
5603 		return 0;
5604 
5605 	if (!copy_reg_to_user(rd, uind))
5606 		return -EFAULT;
5607 
5608 	(*total)++;
5609 	return 0;
5610 }
5611 
5612 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
5613 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
5614 {
5615 	const struct sys_reg_desc *i2, *end2;
5616 	unsigned int total = 0;
5617 	int err;
5618 
5619 	i2 = sys_reg_descs;
5620 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
5621 
5622 	while (i2 != end2) {
5623 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
5624 		if (err)
5625 			return err;
5626 	}
5627 	return total;
5628 }
5629 
5630 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
5631 {
5632 	return num_demux_regs()
5633 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
5634 }
5635 
5636 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
5637 {
5638 	int err;
5639 
5640 	err = walk_sys_regs(vcpu, uindices);
5641 	if (err < 0)
5642 		return err;
5643 	uindices += err;
5644 
5645 	return write_demux_regids(uindices);
5646 }
5647 
5648 #define KVM_ARM_FEATURE_ID_RANGE_INDEX(r)			\
5649 	KVM_ARM_FEATURE_ID_RANGE_IDX(sys_reg_Op0(r),		\
5650 		sys_reg_Op1(r),					\
5651 		sys_reg_CRn(r),					\
5652 		sys_reg_CRm(r),					\
5653 		sys_reg_Op2(r))
5654 
5655 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, struct reg_mask_range *range)
5656 {
5657 	const void *zero_page = page_to_virt(ZERO_PAGE(0));
5658 	u64 __user *masks = (u64 __user *)range->addr;
5659 
5660 	/* Only feature id range is supported, reserved[13] must be zero. */
5661 	if (range->range ||
5662 	    memcmp(range->reserved, zero_page, sizeof(range->reserved)))
5663 		return -EINVAL;
5664 
5665 	/* Wipe the whole thing first */
5666 	if (clear_user(masks, KVM_ARM_FEATURE_ID_RANGE_SIZE * sizeof(__u64)))
5667 		return -EFAULT;
5668 
5669 	for (int i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
5670 		const struct sys_reg_desc *reg = &sys_reg_descs[i];
5671 		u32 encoding = reg_to_encoding(reg);
5672 		u64 val;
5673 
5674 		if (!is_feature_id_reg(encoding) || !reg->set_user)
5675 			continue;
5676 
5677 		if (!reg->val ||
5678 		    (is_aa32_id_reg(encoding) && !kvm_supports_32bit_el0())) {
5679 			continue;
5680 		}
5681 		val = reg->val;
5682 
5683 		if (put_user(val, (masks + KVM_ARM_FEATURE_ID_RANGE_INDEX(encoding))))
5684 			return -EFAULT;
5685 	}
5686 
5687 	return 0;
5688 }
5689 
5690 static void vcpu_set_hcr(struct kvm_vcpu *vcpu)
5691 {
5692 	struct kvm *kvm = vcpu->kvm;
5693 
5694 	if (has_vhe() || has_hvhe())
5695 		vcpu->arch.hcr_el2 |= HCR_E2H;
5696 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN)) {
5697 		/* route synchronous external abort exceptions to EL2 */
5698 		vcpu->arch.hcr_el2 |= HCR_TEA;
5699 		/* trap error record accesses */
5700 		vcpu->arch.hcr_el2 |= HCR_TERR;
5701 	}
5702 
5703 	if (cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
5704 		vcpu->arch.hcr_el2 |= HCR_FWB;
5705 
5706 	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
5707 	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
5708 	    kvm_read_vm_id_reg(kvm, SYS_CTR_EL0) == read_sanitised_ftr_reg(SYS_CTR_EL0))
5709 		vcpu->arch.hcr_el2 |= HCR_TID4;
5710 	else
5711 		vcpu->arch.hcr_el2 |= HCR_TID2;
5712 
5713 	if (vcpu_el1_is_32bit(vcpu))
5714 		vcpu->arch.hcr_el2 &= ~HCR_RW;
5715 
5716 	if (kvm_has_mte(vcpu->kvm))
5717 		vcpu->arch.hcr_el2 |= HCR_ATA;
5718 	else
5719 		vcpu->arch.hcr_el2 |= HCR_TID5;
5720 
5721 	/*
5722 	 * In the absence of FGT, we cannot independently trap TLBI
5723 	 * Range instructions. This isn't great, but trapping all
5724 	 * TLBIs would be far worse. Live with it...
5725 	 */
5726 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
5727 		vcpu->arch.hcr_el2 |= HCR_TTLBOS;
5728 }
5729 
5730 void kvm_calculate_traps(struct kvm_vcpu *vcpu)
5731 {
5732 	struct kvm *kvm = vcpu->kvm;
5733 
5734 	mutex_lock(&kvm->arch.config_lock);
5735 	vcpu_set_hcr(vcpu);
5736 	vcpu_set_ich_hcr(vcpu);
5737 	vcpu_set_hcrx(vcpu);
5738 
5739 	if (test_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags))
5740 		goto out;
5741 
5742 	compute_fgu(kvm, HFGRTR_GROUP);
5743 	compute_fgu(kvm, HFGITR_GROUP);
5744 	compute_fgu(kvm, HDFGRTR_GROUP);
5745 	compute_fgu(kvm, HAFGRTR_GROUP);
5746 	compute_fgu(kvm, HFGRTR2_GROUP);
5747 	compute_fgu(kvm, HFGITR2_GROUP);
5748 	compute_fgu(kvm, HDFGRTR2_GROUP);
5749 	compute_fgu(kvm, ICH_HFGRTR_GROUP);
5750 	compute_fgu(kvm, ICH_HFGITR_GROUP);
5751 
5752 	set_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags);
5753 out:
5754 	mutex_unlock(&kvm->arch.config_lock);
5755 }
5756 
5757 /*
5758  * Perform last adjustments to the ID registers that are implied by the
5759  * configuration outside of the ID regs themselves, as well as any
5760  * initialisation that directly depend on these ID registers (such as
5761  * RES0/RES1 behaviours). This is not the place to configure traps though.
5762  *
5763  * Because this can be called once per CPU, changes must be idempotent.
5764  */
5765 int kvm_finalize_sys_regs(struct kvm_vcpu *vcpu)
5766 {
5767 	struct kvm *kvm = vcpu->kvm;
5768 
5769 	guard(mutex)(&kvm->arch.config_lock);
5770 
5771 	if (vcpu_has_nv(vcpu)) {
5772 		int ret = kvm_init_nv_sysregs(vcpu);
5773 		if (ret)
5774 			return ret;
5775 	}
5776 
5777 	if (kvm_vm_has_ran_once(kvm))
5778 		return 0;
5779 
5780 	/*
5781 	 * This hacks into the ID registers, so only perform it when the
5782 	 * first vcpu runs, or the kvm_set_vm_id_reg() helper will scream.
5783 	 */
5784 	if (!irqchip_in_kernel(kvm)) {
5785 		u64 val;
5786 
5787 		val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64PFR0_EL1) & ~ID_AA64PFR0_EL1_GIC;
5788 		kvm_set_vm_id_reg(kvm, SYS_ID_AA64PFR0_EL1, val);
5789 		val = kvm_read_vm_id_reg(kvm, SYS_ID_AA64PFR2_EL1) & ~ID_AA64PFR2_EL1_GCIE;
5790 		kvm_set_vm_id_reg(kvm, SYS_ID_AA64PFR2_EL1, val);
5791 		val = kvm_read_vm_id_reg(kvm, SYS_ID_PFR1_EL1) & ~ID_PFR1_EL1_GIC;
5792 		kvm_set_vm_id_reg(kvm, SYS_ID_PFR1_EL1, val);
5793 	} else {
5794 		/*
5795 		 * Certain userspace software - QEMU - samples the system
5796 		 * register state without creating an irqchip, then blindly
5797 		 * restores the state prior to running the final guest. This
5798 		 * means that it restores the virtualization & emulation
5799 		 * capabilities of the host system, rather than something that
5800 		 * reflects the final guest state. Moreover, it checks that the
5801 		 * state was "correctly" restored (i.e., verbatim), bailing if
5802 		 * it isn't, so masking off invalid state isn't an option.
5803 		 *
5804 		 * On GICv5 hardware that supports FEAT_GCIE_LEGACY we can run
5805 		 * both GICv3- and GICv5-based guests. Therefore, we initially
5806 		 * present both ID_AA64PFR0.GIC and ID_AA64PFR2.GCIE as IMP to
5807 		 * reflect that userspace can create EITHER a vGICv3 or a
5808 		 * vGICv5. This is an architecturally invalid combination, of
5809 		 * course. Once an in-kernel GIC is created, the sysreg state is
5810 		 * updated to reflect the actual, valid configuration.
5811 		 *
5812 		 * Setting both the GIC and GCIE features to IMP unsurprisingly
5813 		 * results in guests falling over, and hence we need to fix up
5814 		 * this mess in KVM. Before running for the first time we yet
5815 		 * again ensure that the GIC and GCIE fields accurately reflect
5816 		 * the actual hardware the guest should see.
5817 		 *
5818 		 * This hack allows legacy QEMU-based GICv3 guests to run
5819 		 * unmodified on compatible GICv5 hosts, and avoids the inverse
5820 		 * problem for GICv5-based guests in the future.
5821 		 */
5822 		kvm_vgic_finalize_idregs(kvm);
5823 	}
5824 
5825 	return 0;
5826 }
5827 
5828 int __init kvm_sys_reg_table_init(void)
5829 {
5830 	const struct sys_reg_desc *gicv3_regs;
5831 	bool valid = true;
5832 	unsigned int i, sz;
5833 	int ret = 0;
5834 
5835 	/* Make sure tables are unique and in order. */
5836 	valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), true);
5837 	valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), false);
5838 	valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), false);
5839 	valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), false);
5840 	valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), false);
5841 	valid &= check_sysreg_table(sys_insn_descs, ARRAY_SIZE(sys_insn_descs), false);
5842 
5843 	gicv3_regs = vgic_v3_get_sysreg_table(&sz);
5844 	valid &= check_sysreg_table(gicv3_regs, sz, false);
5845 
5846 	if (!valid)
5847 		return -EINVAL;
5848 
5849 	init_imp_id_regs();
5850 
5851 	ret = populate_nv_trap_config();
5852 
5853 	check_feature_map();
5854 
5855 	for (i = 0; !ret && i < ARRAY_SIZE(sys_reg_descs); i++)
5856 		ret = populate_sysreg_config(sys_reg_descs + i, i);
5857 
5858 	for (i = 0; !ret && i < ARRAY_SIZE(sys_insn_descs); i++)
5859 		ret = populate_sysreg_config(sys_insn_descs + i, i);
5860 
5861 	return ret;
5862 }
5863