xref: /linux/arch/arm64/kvm/sys_regs.c (revision c1ead4b4dfe0f643cfc66571ca7d2fa332eddd35)
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 
undef_access(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)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 
bad_trap(struct kvm_vcpu * vcpu,struct sys_reg_params * params,const struct sys_reg_desc * r,const char * msg)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 
read_from_write_only(struct kvm_vcpu * vcpu,struct sys_reg_params * params,const struct sys_reg_desc * r)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 
write_to_read_only(struct kvm_vcpu * vcpu,struct sys_reg_params * params,const struct sys_reg_desc * r)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 
locate_direct_register(const struct kvm_vcpu * vcpu,enum vcpu_sysreg reg)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 
locate_mapped_el2_register(const struct kvm_vcpu * vcpu,enum vcpu_sysreg reg,enum vcpu_sysreg map_reg,u64 (* xlate)(u64),struct sr_loc * loc)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 
locate_register(const struct kvm_vcpu * vcpu,enum vcpu_sysreg reg,struct sr_loc * loc)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(ZCR_EL2,     ZCR_EL1,     NULL	     );
207 		MAPPED_EL2_SYSREG(CONTEXTIDR_EL2, CONTEXTIDR_EL1, NULL	     );
208 		MAPPED_EL2_SYSREG(SCTLR2_EL2,  SCTLR2_EL1,  NULL	     );
209 	case CNTHCTL_EL2:
210 		/* CNTHCTL_EL2 is super special, until we support NV2.1 */
211 		loc->loc = ((is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu)) ?
212 			    SR_LOC_SPECIAL : SR_LOC_MEMORY);
213 		break;
214 	default:
215 		loc->loc = locate_direct_register(vcpu, reg);
216 	}
217 }
218 
read_sr_from_cpu(enum vcpu_sysreg reg)219 static u64 read_sr_from_cpu(enum vcpu_sysreg reg)
220 {
221 	u64 val = 0x8badf00d8badf00d;
222 
223 	switch (reg) {
224 	case SCTLR_EL1:		val = read_sysreg_s(SYS_SCTLR_EL12);	break;
225 	case CPACR_EL1:		val = read_sysreg_s(SYS_CPACR_EL12);	break;
226 	case TTBR0_EL1:		val = read_sysreg_s(SYS_TTBR0_EL12);	break;
227 	case TTBR1_EL1:		val = read_sysreg_s(SYS_TTBR1_EL12);	break;
228 	case TCR_EL1:		val = read_sysreg_s(SYS_TCR_EL12);	break;
229 	case TCR2_EL1:		val = read_sysreg_s(SYS_TCR2_EL12);	break;
230 	case PIR_EL1:		val = read_sysreg_s(SYS_PIR_EL12);	break;
231 	case PIRE0_EL1:		val = read_sysreg_s(SYS_PIRE0_EL12);	break;
232 	case POR_EL1:		val = read_sysreg_s(SYS_POR_EL12);	break;
233 	case ESR_EL1:		val = read_sysreg_s(SYS_ESR_EL12);	break;
234 	case AFSR0_EL1:		val = read_sysreg_s(SYS_AFSR0_EL12);	break;
235 	case AFSR1_EL1:		val = read_sysreg_s(SYS_AFSR1_EL12);	break;
236 	case FAR_EL1:		val = read_sysreg_s(SYS_FAR_EL12);	break;
237 	case MAIR_EL1:		val = read_sysreg_s(SYS_MAIR_EL12);	break;
238 	case VBAR_EL1:		val = read_sysreg_s(SYS_VBAR_EL12);	break;
239 	case CONTEXTIDR_EL1:	val = read_sysreg_s(SYS_CONTEXTIDR_EL12);break;
240 	case AMAIR_EL1:		val = read_sysreg_s(SYS_AMAIR_EL12);	break;
241 	case CNTKCTL_EL1:	val = read_sysreg_s(SYS_CNTKCTL_EL12);	break;
242 	case ELR_EL1:		val = read_sysreg_s(SYS_ELR_EL12);	break;
243 	case SPSR_EL1:		val = read_sysreg_s(SYS_SPSR_EL12);	break;
244 	case ZCR_EL1:		val = read_sysreg_s(SYS_ZCR_EL12);	break;
245 	case SCTLR2_EL1:	val = read_sysreg_s(SYS_SCTLR2_EL12);	break;
246 	case TPIDR_EL0:		val = read_sysreg_s(SYS_TPIDR_EL0);	break;
247 	case TPIDRRO_EL0:	val = read_sysreg_s(SYS_TPIDRRO_EL0);	break;
248 	case TPIDR_EL1:		val = read_sysreg_s(SYS_TPIDR_EL1);	break;
249 	case PAR_EL1:		val = read_sysreg_par();		break;
250 	case DACR32_EL2:	val = read_sysreg_s(SYS_DACR32_EL2);	break;
251 	case IFSR32_EL2:	val = read_sysreg_s(SYS_IFSR32_EL2);	break;
252 	case DBGVCR32_EL2:	val = read_sysreg_s(SYS_DBGVCR32_EL2);	break;
253 	default:		WARN_ON_ONCE(1);
254 	}
255 
256 	return val;
257 }
258 
write_sr_to_cpu(enum vcpu_sysreg reg,u64 val)259 static void write_sr_to_cpu(enum vcpu_sysreg reg, u64 val)
260 {
261 	switch (reg) {
262 	case SCTLR_EL1:		write_sysreg_s(val, SYS_SCTLR_EL12);	break;
263 	case CPACR_EL1:		write_sysreg_s(val, SYS_CPACR_EL12);	break;
264 	case TTBR0_EL1:		write_sysreg_s(val, SYS_TTBR0_EL12);	break;
265 	case TTBR1_EL1:		write_sysreg_s(val, SYS_TTBR1_EL12);	break;
266 	case TCR_EL1:		write_sysreg_s(val, SYS_TCR_EL12);	break;
267 	case TCR2_EL1:		write_sysreg_s(val, SYS_TCR2_EL12);	break;
268 	case PIR_EL1:		write_sysreg_s(val, SYS_PIR_EL12);	break;
269 	case PIRE0_EL1:		write_sysreg_s(val, SYS_PIRE0_EL12);	break;
270 	case POR_EL1:		write_sysreg_s(val, SYS_POR_EL12);	break;
271 	case ESR_EL1:		write_sysreg_s(val, SYS_ESR_EL12);	break;
272 	case AFSR0_EL1:		write_sysreg_s(val, SYS_AFSR0_EL12);	break;
273 	case AFSR1_EL1:		write_sysreg_s(val, SYS_AFSR1_EL12);	break;
274 	case FAR_EL1:		write_sysreg_s(val, SYS_FAR_EL12);	break;
275 	case MAIR_EL1:		write_sysreg_s(val, SYS_MAIR_EL12);	break;
276 	case VBAR_EL1:		write_sysreg_s(val, SYS_VBAR_EL12);	break;
277 	case CONTEXTIDR_EL1:	write_sysreg_s(val, SYS_CONTEXTIDR_EL12);break;
278 	case AMAIR_EL1:		write_sysreg_s(val, SYS_AMAIR_EL12);	break;
279 	case CNTKCTL_EL1:	write_sysreg_s(val, SYS_CNTKCTL_EL12);	break;
280 	case ELR_EL1:		write_sysreg_s(val, SYS_ELR_EL12);	break;
281 	case SPSR_EL1:		write_sysreg_s(val, SYS_SPSR_EL12);	break;
282 	case ZCR_EL1:		write_sysreg_s(val, SYS_ZCR_EL12);	break;
283 	case SCTLR2_EL1:	write_sysreg_s(val, SYS_SCTLR2_EL12);	break;
284 	case TPIDR_EL0:		write_sysreg_s(val, SYS_TPIDR_EL0);	break;
285 	case TPIDRRO_EL0:	write_sysreg_s(val, SYS_TPIDRRO_EL0);	break;
286 	case TPIDR_EL1:		write_sysreg_s(val, SYS_TPIDR_EL1);	break;
287 	case PAR_EL1:		write_sysreg_s(val, SYS_PAR_EL1);	break;
288 	case DACR32_EL2:	write_sysreg_s(val, SYS_DACR32_EL2);	break;
289 	case IFSR32_EL2:	write_sysreg_s(val, SYS_IFSR32_EL2);	break;
290 	case DBGVCR32_EL2:	write_sysreg_s(val, SYS_DBGVCR32_EL2);	break;
291 	default:		WARN_ON_ONCE(1);
292 	}
293 }
294 
vcpu_read_sys_reg(const struct kvm_vcpu * vcpu,enum vcpu_sysreg reg)295 u64 vcpu_read_sys_reg(const struct kvm_vcpu *vcpu, enum vcpu_sysreg reg)
296 {
297 	struct sr_loc loc = {};
298 
299 	locate_register(vcpu, reg, &loc);
300 
301 	WARN_ON_ONCE(!has_vhe() && loc.loc != SR_LOC_MEMORY);
302 
303 	if (loc.loc & SR_LOC_SPECIAL) {
304 		u64 val;
305 
306 		WARN_ON_ONCE(loc.loc & ~SR_LOC_SPECIAL);
307 
308 		/*
309 		 * CNTHCTL_EL2 requires some special treatment to account
310 		 * for the bits that can be set via CNTKCTL_EL1 when E2H==1.
311 		 */
312 		switch (reg) {
313 		case CNTHCTL_EL2:
314 			val = read_sysreg_el1(SYS_CNTKCTL);
315 			val &= CNTKCTL_VALID_BITS;
316 			val |= __vcpu_sys_reg(vcpu, reg) & ~CNTKCTL_VALID_BITS;
317 			return val;
318 		default:
319 			WARN_ON_ONCE(1);
320 		}
321 	}
322 
323 	if (loc.loc & SR_LOC_LOADED) {
324 		enum vcpu_sysreg map_reg = reg;
325 
326 		if (loc.loc & SR_LOC_MAPPED)
327 			map_reg = loc.map_reg;
328 
329 		if (!(loc.loc & SR_LOC_XLATED)) {
330 			u64 val = read_sr_from_cpu(map_reg);
331 
332 			if (reg >= __SANITISED_REG_START__)
333 				val = kvm_vcpu_apply_reg_masks(vcpu, reg, val);
334 
335 			return val;
336 		}
337 	}
338 
339 	return __vcpu_sys_reg(vcpu, reg);
340 }
341 
vcpu_write_sys_reg(struct kvm_vcpu * vcpu,u64 val,enum vcpu_sysreg reg)342 void vcpu_write_sys_reg(struct kvm_vcpu *vcpu, u64 val, enum vcpu_sysreg reg)
343 {
344 	struct sr_loc loc = {};
345 
346 	locate_register(vcpu, reg, &loc);
347 
348 	WARN_ON_ONCE(!has_vhe() && loc.loc != SR_LOC_MEMORY);
349 
350 	if (loc.loc & SR_LOC_SPECIAL) {
351 
352 		WARN_ON_ONCE(loc.loc & ~SR_LOC_SPECIAL);
353 
354 		switch (reg) {
355 		case CNTHCTL_EL2:
356 			/*
357 			 * If E2H=1, some of the bits are backed by
358 			 * CNTKCTL_EL1, while the rest is kept in memory.
359 			 * Yes, this is fun stuff.
360 			 */
361 			write_sysreg_el1(val, SYS_CNTKCTL);
362 			break;
363 		default:
364 			WARN_ON_ONCE(1);
365 		}
366 	}
367 
368 	if (loc.loc & SR_LOC_LOADED) {
369 		enum vcpu_sysreg map_reg = reg;
370 		u64 xlated_val;
371 
372 		if (reg >= __SANITISED_REG_START__)
373 			val = kvm_vcpu_apply_reg_masks(vcpu, reg, val);
374 
375 		if (loc.loc & SR_LOC_MAPPED)
376 			map_reg = loc.map_reg;
377 
378 		if (loc.loc & SR_LOC_XLATED)
379 			xlated_val = loc.xlate(val);
380 		else
381 			xlated_val = val;
382 
383 		write_sr_to_cpu(map_reg, xlated_val);
384 
385 		/*
386 		 * Fall through to write the backing store anyway, which
387 		 * allows translated registers to be directly read without a
388 		 * reverse translation.
389 		 */
390 	}
391 
392 	__vcpu_assign_sys_reg(vcpu, reg, val);
393 }
394 
395 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
396 #define CSSELR_MAX 14
397 
398 /*
399  * Returns the minimum line size for the selected cache, expressed as
400  * Log2(bytes).
401  */
get_min_cache_line_size(bool icache)402 static u8 get_min_cache_line_size(bool icache)
403 {
404 	u64 ctr = read_sanitised_ftr_reg(SYS_CTR_EL0);
405 	u8 field;
406 
407 	if (icache)
408 		field = SYS_FIELD_GET(CTR_EL0, IminLine, ctr);
409 	else
410 		field = SYS_FIELD_GET(CTR_EL0, DminLine, ctr);
411 
412 	/*
413 	 * Cache line size is represented as Log2(words) in CTR_EL0.
414 	 * Log2(bytes) can be derived with the following:
415 	 *
416 	 * Log2(words) + 2 = Log2(bytes / 4) + 2
417 	 * 		   = Log2(bytes) - 2 + 2
418 	 * 		   = Log2(bytes)
419 	 */
420 	return field + 2;
421 }
422 
423 /* Which cache CCSIDR represents depends on CSSELR value. */
get_ccsidr(struct kvm_vcpu * vcpu,u32 csselr)424 static u32 get_ccsidr(struct kvm_vcpu *vcpu, u32 csselr)
425 {
426 	u8 line_size;
427 
428 	if (vcpu->arch.ccsidr)
429 		return vcpu->arch.ccsidr[csselr];
430 
431 	line_size = get_min_cache_line_size(csselr & CSSELR_EL1_InD);
432 
433 	/*
434 	 * Fabricate a CCSIDR value as the overriding value does not exist.
435 	 * The real CCSIDR value will not be used as it can vary by the
436 	 * physical CPU which the vcpu currently resides in.
437 	 *
438 	 * The line size is determined with get_min_cache_line_size(), which
439 	 * should be valid for all CPUs even if they have different cache
440 	 * configuration.
441 	 *
442 	 * The associativity bits are cleared, meaning the geometry of all data
443 	 * and unified caches (which are guaranteed to be PIPT and thus
444 	 * non-aliasing) are 1 set and 1 way.
445 	 * Guests should not be doing cache operations by set/way at all, and
446 	 * for this reason, we trap them and attempt to infer the intent, so
447 	 * that we can flush the entire guest's address space at the appropriate
448 	 * time. The exposed geometry minimizes the number of the traps.
449 	 * [If guests should attempt to infer aliasing properties from the
450 	 * geometry (which is not permitted by the architecture), they would
451 	 * only do so for virtually indexed caches.]
452 	 *
453 	 * We don't check if the cache level exists as it is allowed to return
454 	 * an UNKNOWN value if not.
455 	 */
456 	return SYS_FIELD_PREP(CCSIDR_EL1, LineSize, line_size - 4);
457 }
458 
set_ccsidr(struct kvm_vcpu * vcpu,u32 csselr,u32 val)459 static int set_ccsidr(struct kvm_vcpu *vcpu, u32 csselr, u32 val)
460 {
461 	u8 line_size = FIELD_GET(CCSIDR_EL1_LineSize, val) + 4;
462 	u32 *ccsidr = vcpu->arch.ccsidr;
463 	u32 i;
464 
465 	if ((val & CCSIDR_EL1_RES0) ||
466 	    line_size < get_min_cache_line_size(csselr & CSSELR_EL1_InD))
467 		return -EINVAL;
468 
469 	if (!ccsidr) {
470 		if (val == get_ccsidr(vcpu, csselr))
471 			return 0;
472 
473 		ccsidr = kmalloc_array(CSSELR_MAX, sizeof(u32), GFP_KERNEL_ACCOUNT);
474 		if (!ccsidr)
475 			return -ENOMEM;
476 
477 		for (i = 0; i < CSSELR_MAX; i++)
478 			ccsidr[i] = get_ccsidr(vcpu, i);
479 
480 		vcpu->arch.ccsidr = ccsidr;
481 	}
482 
483 	ccsidr[csselr] = val;
484 
485 	return 0;
486 }
487 
access_rw(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)488 static bool access_rw(struct kvm_vcpu *vcpu,
489 		      struct sys_reg_params *p,
490 		      const struct sys_reg_desc *r)
491 {
492 	if (p->is_write)
493 		vcpu_write_sys_reg(vcpu, p->regval, r->reg);
494 	else
495 		p->regval = vcpu_read_sys_reg(vcpu, r->reg);
496 
497 	return true;
498 }
499 
500 /*
501  * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
502  */
access_dcsw(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)503 static bool access_dcsw(struct kvm_vcpu *vcpu,
504 			struct sys_reg_params *p,
505 			const struct sys_reg_desc *r)
506 {
507 	if (!p->is_write)
508 		return read_from_write_only(vcpu, p, r);
509 
510 	/*
511 	 * Only track S/W ops if we don't have FWB. It still indicates
512 	 * that the guest is a bit broken (S/W operations should only
513 	 * be done by firmware, knowing that there is only a single
514 	 * CPU left in the system, and certainly not from non-secure
515 	 * software).
516 	 */
517 	if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
518 		kvm_set_way_flush(vcpu);
519 
520 	return true;
521 }
522 
access_dcgsw(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)523 static bool access_dcgsw(struct kvm_vcpu *vcpu,
524 			 struct sys_reg_params *p,
525 			 const struct sys_reg_desc *r)
526 {
527 	if (!kvm_has_mte(vcpu->kvm))
528 		return undef_access(vcpu, p, r);
529 
530 	/* Treat MTE S/W ops as we treat the classic ones: with contempt */
531 	return access_dcsw(vcpu, p, r);
532 }
533 
get_access_mask(const struct sys_reg_desc * r,u64 * mask,u64 * shift)534 static void get_access_mask(const struct sys_reg_desc *r, u64 *mask, u64 *shift)
535 {
536 	switch (r->aarch32_map) {
537 	case AA32_LO:
538 		*mask = GENMASK_ULL(31, 0);
539 		*shift = 0;
540 		break;
541 	case AA32_HI:
542 		*mask = GENMASK_ULL(63, 32);
543 		*shift = 32;
544 		break;
545 	default:
546 		*mask = GENMASK_ULL(63, 0);
547 		*shift = 0;
548 		break;
549 	}
550 }
551 
552 /*
553  * Generic accessor for VM registers. Only called as long as HCR_TVM
554  * is set. If the guest enables the MMU, we stop trapping the VM
555  * sys_regs and leave it in complete control of the caches.
556  */
access_vm_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)557 static bool access_vm_reg(struct kvm_vcpu *vcpu,
558 			  struct sys_reg_params *p,
559 			  const struct sys_reg_desc *r)
560 {
561 	bool was_enabled = vcpu_has_cache_enabled(vcpu);
562 	u64 val, mask, shift;
563 
564 	BUG_ON(!p->is_write);
565 
566 	get_access_mask(r, &mask, &shift);
567 
568 	if (~mask) {
569 		val = vcpu_read_sys_reg(vcpu, r->reg);
570 		val &= ~mask;
571 	} else {
572 		val = 0;
573 	}
574 
575 	val |= (p->regval & (mask >> shift)) << shift;
576 	vcpu_write_sys_reg(vcpu, val, r->reg);
577 
578 	kvm_toggle_cache(vcpu, was_enabled);
579 	return true;
580 }
581 
access_actlr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)582 static bool access_actlr(struct kvm_vcpu *vcpu,
583 			 struct sys_reg_params *p,
584 			 const struct sys_reg_desc *r)
585 {
586 	u64 mask, shift;
587 
588 	if (p->is_write)
589 		return ignore_write(vcpu, p);
590 
591 	get_access_mask(r, &mask, &shift);
592 	p->regval = (vcpu_read_sys_reg(vcpu, r->reg) & mask) >> shift;
593 
594 	return true;
595 }
596 
597 /*
598  * Trap handler for the GICv3 SGI generation system register.
599  * Forward the request to the VGIC emulation.
600  * The cp15_64 code makes sure this automatically works
601  * for both AArch64 and AArch32 accesses.
602  */
access_gic_sgi(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)603 static bool access_gic_sgi(struct kvm_vcpu *vcpu,
604 			   struct sys_reg_params *p,
605 			   const struct sys_reg_desc *r)
606 {
607 	bool g1;
608 
609 	if (!kvm_has_gicv3(vcpu->kvm))
610 		return undef_access(vcpu, p, r);
611 
612 	if (!p->is_write)
613 		return read_from_write_only(vcpu, p, r);
614 
615 	/*
616 	 * In a system where GICD_CTLR.DS=1, a ICC_SGI0R_EL1 access generates
617 	 * Group0 SGIs only, while ICC_SGI1R_EL1 can generate either group,
618 	 * depending on the SGI configuration. ICC_ASGI1R_EL1 is effectively
619 	 * equivalent to ICC_SGI0R_EL1, as there is no "alternative" secure
620 	 * group.
621 	 */
622 	if (p->Op0 == 0) {		/* AArch32 */
623 		switch (p->Op1) {
624 		default:		/* Keep GCC quiet */
625 		case 0:			/* ICC_SGI1R */
626 			g1 = true;
627 			break;
628 		case 1:			/* ICC_ASGI1R */
629 		case 2:			/* ICC_SGI0R */
630 			g1 = false;
631 			break;
632 		}
633 	} else {			/* AArch64 */
634 		switch (p->Op2) {
635 		default:		/* Keep GCC quiet */
636 		case 5:			/* ICC_SGI1R_EL1 */
637 			g1 = true;
638 			break;
639 		case 6:			/* ICC_ASGI1R_EL1 */
640 		case 7:			/* ICC_SGI0R_EL1 */
641 			g1 = false;
642 			break;
643 		}
644 	}
645 
646 	vgic_v3_dispatch_sgi(vcpu, p->regval, g1);
647 
648 	return true;
649 }
650 
access_gic_sre(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)651 static bool access_gic_sre(struct kvm_vcpu *vcpu,
652 			   struct sys_reg_params *p,
653 			   const struct sys_reg_desc *r)
654 {
655 	if (!kvm_has_gicv3(vcpu->kvm))
656 		return undef_access(vcpu, p, r);
657 
658 	if (p->is_write)
659 		return ignore_write(vcpu, p);
660 
661 	if (p->Op1 == 4) {	/* ICC_SRE_EL2 */
662 		p->regval = KVM_ICC_SRE_EL2;
663 	} else {		/* ICC_SRE_EL1 */
664 		p->regval = vcpu->arch.vgic_cpu.vgic_v3.vgic_sre;
665 	}
666 
667 	return true;
668 }
669 
trap_raz_wi(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)670 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
671 			struct sys_reg_params *p,
672 			const struct sys_reg_desc *r)
673 {
674 	if (p->is_write)
675 		return ignore_write(vcpu, p);
676 	else
677 		return read_zero(vcpu, p);
678 }
679 
680 /*
681  * ARMv8.1 mandates at least a trivial LORegion implementation, where all the
682  * RW registers are RES0 (which we can implement as RAZ/WI). On an ARMv8.0
683  * system, these registers should UNDEF. LORID_EL1 being a RO register, we
684  * treat it separately.
685  */
trap_loregion(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)686 static bool trap_loregion(struct kvm_vcpu *vcpu,
687 			  struct sys_reg_params *p,
688 			  const struct sys_reg_desc *r)
689 {
690 	u32 sr = reg_to_encoding(r);
691 
692 	if (!kvm_has_feat(vcpu->kvm, ID_AA64MMFR1_EL1, LO, IMP))
693 		return undef_access(vcpu, p, r);
694 
695 	if (p->is_write && sr == SYS_LORID_EL1)
696 		return write_to_read_only(vcpu, p, r);
697 
698 	return trap_raz_wi(vcpu, p, r);
699 }
700 
trap_oslar_el1(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)701 static bool trap_oslar_el1(struct kvm_vcpu *vcpu,
702 			   struct sys_reg_params *p,
703 			   const struct sys_reg_desc *r)
704 {
705 	if (!p->is_write)
706 		return read_from_write_only(vcpu, p, r);
707 
708 	kvm_debug_handle_oslar(vcpu, p->regval);
709 	return true;
710 }
711 
trap_oslsr_el1(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)712 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
713 			   struct sys_reg_params *p,
714 			   const struct sys_reg_desc *r)
715 {
716 	if (p->is_write)
717 		return write_to_read_only(vcpu, p, r);
718 
719 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
720 	return true;
721 }
722 
set_oslsr_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)723 static int set_oslsr_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
724 			 u64 val)
725 {
726 	/*
727 	 * The only modifiable bit is the OSLK bit. Refuse the write if
728 	 * userspace attempts to change any other bit in the register.
729 	 */
730 	if ((val ^ rd->val) & ~OSLSR_EL1_OSLK)
731 		return -EINVAL;
732 
733 	__vcpu_assign_sys_reg(vcpu, rd->reg, val);
734 	return 0;
735 }
736 
trap_dbgauthstatus_el1(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)737 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
738 				   struct sys_reg_params *p,
739 				   const struct sys_reg_desc *r)
740 {
741 	if (p->is_write) {
742 		return ignore_write(vcpu, p);
743 	} else {
744 		p->regval = read_sysreg(dbgauthstatus_el1);
745 		return true;
746 	}
747 }
748 
trap_debug_regs(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)749 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
750 			    struct sys_reg_params *p,
751 			    const struct sys_reg_desc *r)
752 {
753 	access_rw(vcpu, p, r);
754 
755 	kvm_debug_set_guest_ownership(vcpu);
756 	return true;
757 }
758 
759 /*
760  * reg_to_dbg/dbg_to_reg
761  *
762  * A 32 bit write to a debug register leave top bits alone
763  * A 32 bit read from a debug register only returns the bottom bits
764  */
reg_to_dbg(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * rd,u64 * dbg_reg)765 static void reg_to_dbg(struct kvm_vcpu *vcpu,
766 		       struct sys_reg_params *p,
767 		       const struct sys_reg_desc *rd,
768 		       u64 *dbg_reg)
769 {
770 	u64 mask, shift, val;
771 
772 	get_access_mask(rd, &mask, &shift);
773 
774 	val = *dbg_reg;
775 	val &= ~mask;
776 	val |= (p->regval & (mask >> shift)) << shift;
777 	*dbg_reg = val;
778 }
779 
dbg_to_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * rd,u64 * dbg_reg)780 static void dbg_to_reg(struct kvm_vcpu *vcpu,
781 		       struct sys_reg_params *p,
782 		       const struct sys_reg_desc *rd,
783 		       u64 *dbg_reg)
784 {
785 	u64 mask, shift;
786 
787 	get_access_mask(rd, &mask, &shift);
788 	p->regval = (*dbg_reg & mask) >> shift;
789 }
790 
demux_wb_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)791 static u64 *demux_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd)
792 {
793 	struct kvm_guest_debug_arch *dbg = &vcpu->arch.vcpu_debug_state;
794 
795 	switch (rd->Op2) {
796 	case 0b100:
797 		return &dbg->dbg_bvr[rd->CRm];
798 	case 0b101:
799 		return &dbg->dbg_bcr[rd->CRm];
800 	case 0b110:
801 		return &dbg->dbg_wvr[rd->CRm];
802 	case 0b111:
803 		return &dbg->dbg_wcr[rd->CRm];
804 	default:
805 		KVM_BUG_ON(1, vcpu->kvm);
806 		return NULL;
807 	}
808 }
809 
trap_dbg_wb_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * rd)810 static bool trap_dbg_wb_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
811 			    const struct sys_reg_desc *rd)
812 {
813 	u64 *reg = demux_wb_reg(vcpu, rd);
814 
815 	if (!reg)
816 		return false;
817 
818 	if (p->is_write)
819 		reg_to_dbg(vcpu, p, rd, reg);
820 	else
821 		dbg_to_reg(vcpu, p, rd, reg);
822 
823 	kvm_debug_set_guest_ownership(vcpu);
824 	return true;
825 }
826 
set_dbg_wb_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)827 static int set_dbg_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
828 			  u64 val)
829 {
830 	u64 *reg = demux_wb_reg(vcpu, rd);
831 
832 	if (!reg)
833 		return -EINVAL;
834 
835 	*reg = val;
836 	return 0;
837 }
838 
get_dbg_wb_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 * val)839 static int get_dbg_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
840 			  u64 *val)
841 {
842 	u64 *reg = demux_wb_reg(vcpu, rd);
843 
844 	if (!reg)
845 		return -EINVAL;
846 
847 	*val = *reg;
848 	return 0;
849 }
850 
reset_dbg_wb_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)851 static u64 reset_dbg_wb_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd)
852 {
853 	u64 *reg = demux_wb_reg(vcpu, rd);
854 
855 	/*
856 	 * Bail early if we couldn't find storage for the register, the
857 	 * KVM_BUG_ON() in demux_wb_reg() will prevent this VM from ever
858 	 * being run.
859 	 */
860 	if (!reg)
861 		return 0;
862 
863 	*reg = rd->val;
864 	return rd->val;
865 }
866 
reset_amair_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)867 static u64 reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
868 {
869 	u64 amair = read_sysreg(amair_el1);
870 	vcpu_write_sys_reg(vcpu, amair, AMAIR_EL1);
871 	return amair;
872 }
873 
reset_actlr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)874 static u64 reset_actlr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
875 {
876 	u64 actlr = read_sysreg(actlr_el1);
877 	vcpu_write_sys_reg(vcpu, actlr, ACTLR_EL1);
878 	return actlr;
879 }
880 
reset_mpidr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)881 static u64 reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
882 {
883 	u64 mpidr;
884 
885 	/*
886 	 * Map the vcpu_id into the first three affinity level fields of
887 	 * the MPIDR. We limit the number of VCPUs in level 0 due to a
888 	 * limitation to 16 CPUs in that level in the ICC_SGIxR registers
889 	 * of the GICv3 to be able to address each CPU directly when
890 	 * sending IPIs.
891 	 */
892 	mpidr = (vcpu->vcpu_id & 0x0f) << MPIDR_LEVEL_SHIFT(0);
893 	mpidr |= ((vcpu->vcpu_id >> 4) & 0xff) << MPIDR_LEVEL_SHIFT(1);
894 	mpidr |= ((vcpu->vcpu_id >> 12) & 0xff) << MPIDR_LEVEL_SHIFT(2);
895 	mpidr |= (1ULL << 31);
896 	vcpu_write_sys_reg(vcpu, mpidr, MPIDR_EL1);
897 
898 	return mpidr;
899 }
900 
hidden_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)901 static unsigned int hidden_visibility(const struct kvm_vcpu *vcpu,
902 				      const struct sys_reg_desc *r)
903 {
904 	return REG_HIDDEN;
905 }
906 
pmu_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)907 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
908 				   const struct sys_reg_desc *r)
909 {
910 	if (kvm_vcpu_has_pmu(vcpu))
911 		return 0;
912 
913 	return REG_HIDDEN;
914 }
915 
reset_pmu_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)916 static u64 reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
917 {
918 	u64 mask = BIT(ARMV8_PMU_CYCLE_IDX);
919 	u8 n = vcpu->kvm->arch.nr_pmu_counters;
920 
921 	if (n)
922 		mask |= GENMASK(n - 1, 0);
923 
924 	reset_unknown(vcpu, r);
925 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, mask);
926 
927 	return __vcpu_sys_reg(vcpu, r->reg);
928 }
929 
reset_pmevcntr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)930 static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
931 {
932 	reset_unknown(vcpu, r);
933 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, GENMASK(31, 0));
934 
935 	return __vcpu_sys_reg(vcpu, r->reg);
936 }
937 
reset_pmevtyper(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)938 static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
939 {
940 	/* This thing will UNDEF, who cares about the reset value? */
941 	if (!kvm_vcpu_has_pmu(vcpu))
942 		return 0;
943 
944 	reset_unknown(vcpu, r);
945 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, kvm_pmu_evtyper_mask(vcpu->kvm));
946 
947 	return __vcpu_sys_reg(vcpu, r->reg);
948 }
949 
reset_pmselr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)950 static u64 reset_pmselr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
951 {
952 	reset_unknown(vcpu, r);
953 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, PMSELR_EL0_SEL_MASK);
954 
955 	return __vcpu_sys_reg(vcpu, r->reg);
956 }
957 
reset_pmcr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)958 static u64 reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
959 {
960 	u64 pmcr = 0;
961 
962 	if (!kvm_supports_32bit_el0())
963 		pmcr |= ARMV8_PMU_PMCR_LC;
964 
965 	/*
966 	 * The value of PMCR.N field is included when the
967 	 * vCPU register is read via kvm_vcpu_read_pmcr().
968 	 */
969 	__vcpu_assign_sys_reg(vcpu, r->reg, pmcr);
970 
971 	return __vcpu_sys_reg(vcpu, r->reg);
972 }
973 
check_pmu_access_disabled(struct kvm_vcpu * vcpu,u64 flags)974 static bool check_pmu_access_disabled(struct kvm_vcpu *vcpu, u64 flags)
975 {
976 	u64 reg = __vcpu_sys_reg(vcpu, PMUSERENR_EL0);
977 	bool enabled = (reg & flags) || vcpu_mode_priv(vcpu);
978 
979 	if (!enabled)
980 		kvm_inject_undefined(vcpu);
981 
982 	return !enabled;
983 }
984 
pmu_access_el0_disabled(struct kvm_vcpu * vcpu)985 static bool pmu_access_el0_disabled(struct kvm_vcpu *vcpu)
986 {
987 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_EN);
988 }
989 
pmu_write_swinc_el0_disabled(struct kvm_vcpu * vcpu)990 static bool pmu_write_swinc_el0_disabled(struct kvm_vcpu *vcpu)
991 {
992 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_SW | ARMV8_PMU_USERENR_EN);
993 }
994 
pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu * vcpu)995 static bool pmu_access_cycle_counter_el0_disabled(struct kvm_vcpu *vcpu)
996 {
997 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_CR | ARMV8_PMU_USERENR_EN);
998 }
999 
pmu_access_event_counter_el0_disabled(struct kvm_vcpu * vcpu)1000 static bool pmu_access_event_counter_el0_disabled(struct kvm_vcpu *vcpu)
1001 {
1002 	return check_pmu_access_disabled(vcpu, ARMV8_PMU_USERENR_ER | ARMV8_PMU_USERENR_EN);
1003 }
1004 
access_pmcr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1005 static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1006 			const struct sys_reg_desc *r)
1007 {
1008 	u64 val;
1009 
1010 	if (pmu_access_el0_disabled(vcpu))
1011 		return false;
1012 
1013 	if (p->is_write) {
1014 		/*
1015 		 * Only update writeable bits of PMCR (continuing into
1016 		 * kvm_pmu_handle_pmcr() as well)
1017 		 */
1018 		val = kvm_vcpu_read_pmcr(vcpu);
1019 		val &= ~ARMV8_PMU_PMCR_MASK;
1020 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
1021 		if (!kvm_supports_32bit_el0())
1022 			val |= ARMV8_PMU_PMCR_LC;
1023 		kvm_pmu_handle_pmcr(vcpu, val);
1024 	} else {
1025 		/* PMCR.P & PMCR.C are RAZ */
1026 		val = kvm_vcpu_read_pmcr(vcpu)
1027 		      & ~(ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C);
1028 		p->regval = val;
1029 	}
1030 
1031 	return true;
1032 }
1033 
access_pmselr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1034 static bool access_pmselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1035 			  const struct sys_reg_desc *r)
1036 {
1037 	if (pmu_access_event_counter_el0_disabled(vcpu))
1038 		return false;
1039 
1040 	if (p->is_write)
1041 		__vcpu_assign_sys_reg(vcpu, PMSELR_EL0, p->regval);
1042 	else
1043 		/* return PMSELR.SEL field */
1044 		p->regval = __vcpu_sys_reg(vcpu, PMSELR_EL0)
1045 			    & PMSELR_EL0_SEL_MASK;
1046 
1047 	return true;
1048 }
1049 
access_pmceid(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1050 static bool access_pmceid(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1051 			  const struct sys_reg_desc *r)
1052 {
1053 	u64 pmceid, mask, shift;
1054 
1055 	BUG_ON(p->is_write);
1056 
1057 	if (pmu_access_el0_disabled(vcpu))
1058 		return false;
1059 
1060 	get_access_mask(r, &mask, &shift);
1061 
1062 	pmceid = kvm_pmu_get_pmceid(vcpu, (p->Op2 & 1));
1063 	pmceid &= mask;
1064 	pmceid >>= shift;
1065 
1066 	p->regval = pmceid;
1067 
1068 	return true;
1069 }
1070 
pmu_counter_idx_valid(struct kvm_vcpu * vcpu,u64 idx)1071 static bool pmu_counter_idx_valid(struct kvm_vcpu *vcpu, u64 idx)
1072 {
1073 	u64 pmcr, val;
1074 
1075 	pmcr = kvm_vcpu_read_pmcr(vcpu);
1076 	val = FIELD_GET(ARMV8_PMU_PMCR_N, pmcr);
1077 	if (idx >= val && idx != ARMV8_PMU_CYCLE_IDX) {
1078 		kvm_inject_undefined(vcpu);
1079 		return false;
1080 	}
1081 
1082 	return true;
1083 }
1084 
get_pmu_evcntr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 * val)1085 static int get_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1086 			  u64 *val)
1087 {
1088 	u64 idx;
1089 
1090 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1091 		/* PMCCNTR_EL0 */
1092 		idx = ARMV8_PMU_CYCLE_IDX;
1093 	else
1094 		/* PMEVCNTRn_EL0 */
1095 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1096 
1097 	*val = kvm_pmu_get_counter_value(vcpu, idx);
1098 	return 0;
1099 }
1100 
set_pmu_evcntr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 val)1101 static int set_pmu_evcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1102 			  u64 val)
1103 {
1104 	u64 idx;
1105 
1106 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 0)
1107 		/* PMCCNTR_EL0 */
1108 		idx = ARMV8_PMU_CYCLE_IDX;
1109 	else
1110 		/* PMEVCNTRn_EL0 */
1111 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1112 
1113 	kvm_pmu_set_counter_value_user(vcpu, idx, val);
1114 	return 0;
1115 }
1116 
access_pmu_evcntr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1117 static bool access_pmu_evcntr(struct kvm_vcpu *vcpu,
1118 			      struct sys_reg_params *p,
1119 			      const struct sys_reg_desc *r)
1120 {
1121 	u64 idx = ~0UL;
1122 
1123 	if (r->CRn == 9 && r->CRm == 13) {
1124 		if (r->Op2 == 2) {
1125 			/* PMXEVCNTR_EL0 */
1126 			if (pmu_access_event_counter_el0_disabled(vcpu))
1127 				return false;
1128 
1129 			idx = SYS_FIELD_GET(PMSELR_EL0, SEL,
1130 					    __vcpu_sys_reg(vcpu, PMSELR_EL0));
1131 		} else if (r->Op2 == 0) {
1132 			/* PMCCNTR_EL0 */
1133 			if (pmu_access_cycle_counter_el0_disabled(vcpu))
1134 				return false;
1135 
1136 			idx = ARMV8_PMU_CYCLE_IDX;
1137 		}
1138 	} else if (r->CRn == 0 && r->CRm == 9) {
1139 		/* PMCCNTR */
1140 		if (pmu_access_event_counter_el0_disabled(vcpu))
1141 			return false;
1142 
1143 		idx = ARMV8_PMU_CYCLE_IDX;
1144 	} else if (r->CRn == 14 && (r->CRm & 12) == 8) {
1145 		/* PMEVCNTRn_EL0 */
1146 		if (pmu_access_event_counter_el0_disabled(vcpu))
1147 			return false;
1148 
1149 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1150 	}
1151 
1152 	/* Catch any decoding mistake */
1153 	WARN_ON(idx == ~0UL);
1154 
1155 	if (!pmu_counter_idx_valid(vcpu, idx))
1156 		return false;
1157 
1158 	if (p->is_write) {
1159 		if (pmu_access_el0_disabled(vcpu))
1160 			return false;
1161 
1162 		kvm_pmu_set_counter_value(vcpu, idx, p->regval);
1163 	} else {
1164 		p->regval = kvm_pmu_get_counter_value(vcpu, idx);
1165 	}
1166 
1167 	return true;
1168 }
1169 
access_pmu_evtyper(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1170 static bool access_pmu_evtyper(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1171 			       const struct sys_reg_desc *r)
1172 {
1173 	u64 idx, reg;
1174 
1175 	if (pmu_access_el0_disabled(vcpu))
1176 		return false;
1177 
1178 	if (r->CRn == 9 && r->CRm == 13 && r->Op2 == 1) {
1179 		/* PMXEVTYPER_EL0 */
1180 		idx = SYS_FIELD_GET(PMSELR_EL0, SEL, __vcpu_sys_reg(vcpu, PMSELR_EL0));
1181 		reg = PMEVTYPER0_EL0 + idx;
1182 	} else if (r->CRn == 14 && (r->CRm & 12) == 12) {
1183 		idx = ((r->CRm & 3) << 3) | (r->Op2 & 7);
1184 		if (idx == ARMV8_PMU_CYCLE_IDX)
1185 			reg = PMCCFILTR_EL0;
1186 		else
1187 			/* PMEVTYPERn_EL0 */
1188 			reg = PMEVTYPER0_EL0 + idx;
1189 	} else {
1190 		BUG();
1191 	}
1192 
1193 	if (!pmu_counter_idx_valid(vcpu, idx))
1194 		return false;
1195 
1196 	if (p->is_write) {
1197 		kvm_pmu_set_counter_event_type(vcpu, p->regval, idx);
1198 		kvm_vcpu_pmu_restore_guest(vcpu);
1199 	} else {
1200 		p->regval = __vcpu_sys_reg(vcpu, reg);
1201 	}
1202 
1203 	return true;
1204 }
1205 
set_pmreg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 val)1206 static int set_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 val)
1207 {
1208 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1209 
1210 	__vcpu_assign_sys_reg(vcpu, r->reg, val & mask);
1211 	kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
1212 
1213 	return 0;
1214 }
1215 
get_pmreg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 * val)1216 static int get_pmreg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r, u64 *val)
1217 {
1218 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1219 
1220 	*val = __vcpu_sys_reg(vcpu, r->reg) & mask;
1221 	return 0;
1222 }
1223 
access_pmcnten(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1224 static bool access_pmcnten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1225 			   const struct sys_reg_desc *r)
1226 {
1227 	u64 val, mask;
1228 
1229 	if (pmu_access_el0_disabled(vcpu))
1230 		return false;
1231 
1232 	mask = kvm_pmu_accessible_counter_mask(vcpu);
1233 	if (p->is_write) {
1234 		val = p->regval & mask;
1235 		if (r->Op2 & 0x1)
1236 			/* accessing PMCNTENSET_EL0 */
1237 			__vcpu_rmw_sys_reg(vcpu, PMCNTENSET_EL0, |=, val);
1238 		else
1239 			/* accessing PMCNTENCLR_EL0 */
1240 			__vcpu_rmw_sys_reg(vcpu, PMCNTENSET_EL0, &=, ~val);
1241 
1242 		kvm_pmu_reprogram_counter_mask(vcpu, val);
1243 	} else {
1244 		p->regval = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
1245 	}
1246 
1247 	return true;
1248 }
1249 
access_pminten(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1250 static bool access_pminten(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1251 			   const struct sys_reg_desc *r)
1252 {
1253 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1254 
1255 	if (check_pmu_access_disabled(vcpu, 0))
1256 		return false;
1257 
1258 	if (p->is_write) {
1259 		u64 val = p->regval & mask;
1260 
1261 		if (r->Op2 & 0x1)
1262 			/* accessing PMINTENSET_EL1 */
1263 			__vcpu_rmw_sys_reg(vcpu, PMINTENSET_EL1, |=, val);
1264 		else
1265 			/* accessing PMINTENCLR_EL1 */
1266 			__vcpu_rmw_sys_reg(vcpu, PMINTENSET_EL1, &=, ~val);
1267 	} else {
1268 		p->regval = __vcpu_sys_reg(vcpu, PMINTENSET_EL1);
1269 	}
1270 
1271 	return true;
1272 }
1273 
access_pmovs(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1274 static bool access_pmovs(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1275 			 const struct sys_reg_desc *r)
1276 {
1277 	u64 mask = kvm_pmu_accessible_counter_mask(vcpu);
1278 
1279 	if (pmu_access_el0_disabled(vcpu))
1280 		return false;
1281 
1282 	if (p->is_write) {
1283 		if (r->CRm & 0x2)
1284 			/* accessing PMOVSSET_EL0 */
1285 			__vcpu_rmw_sys_reg(vcpu, PMOVSSET_EL0, |=, (p->regval & mask));
1286 		else
1287 			/* accessing PMOVSCLR_EL0 */
1288 			__vcpu_rmw_sys_reg(vcpu, PMOVSSET_EL0, &=, ~(p->regval & mask));
1289 	} else {
1290 		p->regval = __vcpu_sys_reg(vcpu, PMOVSSET_EL0);
1291 	}
1292 
1293 	return true;
1294 }
1295 
access_pmswinc(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1296 static bool access_pmswinc(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1297 			   const struct sys_reg_desc *r)
1298 {
1299 	u64 mask;
1300 
1301 	if (!p->is_write)
1302 		return read_from_write_only(vcpu, p, r);
1303 
1304 	if (pmu_write_swinc_el0_disabled(vcpu))
1305 		return false;
1306 
1307 	mask = kvm_pmu_accessible_counter_mask(vcpu);
1308 	kvm_pmu_software_increment(vcpu, p->regval & mask);
1309 	return true;
1310 }
1311 
access_pmuserenr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1312 static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
1313 			     const struct sys_reg_desc *r)
1314 {
1315 	if (p->is_write) {
1316 		if (!vcpu_mode_priv(vcpu))
1317 			return undef_access(vcpu, p, r);
1318 
1319 		__vcpu_assign_sys_reg(vcpu, PMUSERENR_EL0,
1320 				      (p->regval & ARMV8_PMU_USERENR_MASK));
1321 	} else {
1322 		p->regval = __vcpu_sys_reg(vcpu, PMUSERENR_EL0)
1323 			    & ARMV8_PMU_USERENR_MASK;
1324 	}
1325 
1326 	return true;
1327 }
1328 
get_pmcr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 * val)1329 static int get_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1330 		    u64 *val)
1331 {
1332 	*val = kvm_vcpu_read_pmcr(vcpu);
1333 	return 0;
1334 }
1335 
set_pmcr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 val)1336 static int set_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
1337 		    u64 val)
1338 {
1339 	u8 new_n = FIELD_GET(ARMV8_PMU_PMCR_N, val);
1340 	struct kvm *kvm = vcpu->kvm;
1341 
1342 	mutex_lock(&kvm->arch.config_lock);
1343 
1344 	/*
1345 	 * The vCPU can't have more counters than the PMU hardware
1346 	 * implements. Ignore this error to maintain compatibility
1347 	 * with the existing KVM behavior.
1348 	 */
1349 	if (!kvm_vm_has_ran_once(kvm) &&
1350 	    !vcpu_has_nv(vcpu)	      &&
1351 	    new_n <= kvm_arm_pmu_get_max_counters(kvm))
1352 		kvm->arch.nr_pmu_counters = new_n;
1353 
1354 	mutex_unlock(&kvm->arch.config_lock);
1355 
1356 	/*
1357 	 * Ignore writes to RES0 bits, read only bits that are cleared on
1358 	 * vCPU reset, and writable bits that KVM doesn't support yet.
1359 	 * (i.e. only PMCR.N and bits [7:0] are mutable from userspace)
1360 	 * The LP bit is RES0 when FEAT_PMUv3p5 is not supported on the vCPU.
1361 	 * But, we leave the bit as it is here, as the vCPU's PMUver might
1362 	 * be changed later (NOTE: the bit will be cleared on first vCPU run
1363 	 * if necessary).
1364 	 */
1365 	val &= ARMV8_PMU_PMCR_MASK;
1366 
1367 	/* The LC bit is RES1 when AArch32 is not supported */
1368 	if (!kvm_supports_32bit_el0())
1369 		val |= ARMV8_PMU_PMCR_LC;
1370 
1371 	__vcpu_assign_sys_reg(vcpu, r->reg, val);
1372 	kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
1373 
1374 	return 0;
1375 }
1376 
1377 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
1378 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
1379 	{ SYS_DESC(SYS_DBGBVRn_EL1(n)),					\
1380 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1381 	  get_dbg_wb_reg, set_dbg_wb_reg },				\
1382 	{ SYS_DESC(SYS_DBGBCRn_EL1(n)),					\
1383 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1384 	  get_dbg_wb_reg, set_dbg_wb_reg },				\
1385 	{ SYS_DESC(SYS_DBGWVRn_EL1(n)),					\
1386 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1387 	  get_dbg_wb_reg, set_dbg_wb_reg },				\
1388 	{ SYS_DESC(SYS_DBGWCRn_EL1(n)),					\
1389 	  trap_dbg_wb_reg, reset_dbg_wb_reg, 0, 0,			\
1390 	  get_dbg_wb_reg, set_dbg_wb_reg }
1391 
1392 #define PMU_SYS_REG(name)						\
1393 	SYS_DESC(SYS_##name), .reset = reset_pmu_reg,			\
1394 	.visibility = pmu_visibility
1395 
1396 /* Macro to expand the PMEVCNTRn_EL0 register */
1397 #define PMU_PMEVCNTR_EL0(n)						\
1398 	{ PMU_SYS_REG(PMEVCNTRn_EL0(n)),				\
1399 	  .reset = reset_pmevcntr, .get_user = get_pmu_evcntr,		\
1400 	  .set_user = set_pmu_evcntr,					\
1401 	  .access = access_pmu_evcntr, .reg = (PMEVCNTR0_EL0 + n), }
1402 
1403 /* Macro to expand the PMEVTYPERn_EL0 register */
1404 #define PMU_PMEVTYPER_EL0(n)						\
1405 	{ PMU_SYS_REG(PMEVTYPERn_EL0(n)),				\
1406 	  .reset = reset_pmevtyper,					\
1407 	  .access = access_pmu_evtyper, .reg = (PMEVTYPER0_EL0 + n), }
1408 
1409 /* Macro to expand the AMU counter and type registers*/
1410 #define AMU_AMEVCNTR0_EL0(n) { SYS_DESC(SYS_AMEVCNTR0_EL0(n)), undef_access }
1411 #define AMU_AMEVTYPER0_EL0(n) { SYS_DESC(SYS_AMEVTYPER0_EL0(n)), undef_access }
1412 #define AMU_AMEVCNTR1_EL0(n) { SYS_DESC(SYS_AMEVCNTR1_EL0(n)), undef_access }
1413 #define AMU_AMEVTYPER1_EL0(n) { SYS_DESC(SYS_AMEVTYPER1_EL0(n)), undef_access }
1414 
ptrauth_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)1415 static unsigned int ptrauth_visibility(const struct kvm_vcpu *vcpu,
1416 			const struct sys_reg_desc *rd)
1417 {
1418 	return vcpu_has_ptrauth(vcpu) ? 0 : REG_HIDDEN;
1419 }
1420 
1421 /*
1422  * If we land here on a PtrAuth access, that is because we didn't
1423  * fixup the access on exit by allowing the PtrAuth sysregs. The only
1424  * way this happens is when the guest does not have PtrAuth support
1425  * enabled.
1426  */
1427 #define __PTRAUTH_KEY(k)						\
1428 	{ SYS_DESC(SYS_## k), undef_access, reset_unknown, k,		\
1429 	.visibility = ptrauth_visibility}
1430 
1431 #define PTRAUTH_KEY(k)							\
1432 	__PTRAUTH_KEY(k ## KEYLO_EL1),					\
1433 	__PTRAUTH_KEY(k ## KEYHI_EL1)
1434 
access_arch_timer(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1435 static bool access_arch_timer(struct kvm_vcpu *vcpu,
1436 			      struct sys_reg_params *p,
1437 			      const struct sys_reg_desc *r)
1438 {
1439 	enum kvm_arch_timers tmr;
1440 	enum kvm_arch_timer_regs treg;
1441 	u64 reg = reg_to_encoding(r);
1442 
1443 	switch (reg) {
1444 	case SYS_CNTP_TVAL_EL0:
1445 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1446 			tmr = TIMER_HPTIMER;
1447 		else
1448 			tmr = TIMER_PTIMER;
1449 		treg = TIMER_REG_TVAL;
1450 		break;
1451 
1452 	case SYS_CNTV_TVAL_EL0:
1453 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1454 			tmr = TIMER_HVTIMER;
1455 		else
1456 			tmr = TIMER_VTIMER;
1457 		treg = TIMER_REG_TVAL;
1458 		break;
1459 
1460 	case SYS_AARCH32_CNTP_TVAL:
1461 	case SYS_CNTP_TVAL_EL02:
1462 		tmr = TIMER_PTIMER;
1463 		treg = TIMER_REG_TVAL;
1464 		break;
1465 
1466 	case SYS_CNTV_TVAL_EL02:
1467 		tmr = TIMER_VTIMER;
1468 		treg = TIMER_REG_TVAL;
1469 		break;
1470 
1471 	case SYS_CNTHP_TVAL_EL2:
1472 		tmr = TIMER_HPTIMER;
1473 		treg = TIMER_REG_TVAL;
1474 		break;
1475 
1476 	case SYS_CNTHV_TVAL_EL2:
1477 		tmr = TIMER_HVTIMER;
1478 		treg = TIMER_REG_TVAL;
1479 		break;
1480 
1481 	case SYS_CNTP_CTL_EL0:
1482 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1483 			tmr = TIMER_HPTIMER;
1484 		else
1485 			tmr = TIMER_PTIMER;
1486 		treg = TIMER_REG_CTL;
1487 		break;
1488 
1489 	case SYS_CNTV_CTL_EL0:
1490 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1491 			tmr = TIMER_HVTIMER;
1492 		else
1493 			tmr = TIMER_VTIMER;
1494 		treg = TIMER_REG_CTL;
1495 		break;
1496 
1497 	case SYS_AARCH32_CNTP_CTL:
1498 	case SYS_CNTP_CTL_EL02:
1499 		tmr = TIMER_PTIMER;
1500 		treg = TIMER_REG_CTL;
1501 		break;
1502 
1503 	case SYS_CNTV_CTL_EL02:
1504 		tmr = TIMER_VTIMER;
1505 		treg = TIMER_REG_CTL;
1506 		break;
1507 
1508 	case SYS_CNTHP_CTL_EL2:
1509 		tmr = TIMER_HPTIMER;
1510 		treg = TIMER_REG_CTL;
1511 		break;
1512 
1513 	case SYS_CNTHV_CTL_EL2:
1514 		tmr = TIMER_HVTIMER;
1515 		treg = TIMER_REG_CTL;
1516 		break;
1517 
1518 	case SYS_CNTP_CVAL_EL0:
1519 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1520 			tmr = TIMER_HPTIMER;
1521 		else
1522 			tmr = TIMER_PTIMER;
1523 		treg = TIMER_REG_CVAL;
1524 		break;
1525 
1526 	case SYS_CNTV_CVAL_EL0:
1527 		if (is_hyp_ctxt(vcpu) && vcpu_el2_e2h_is_set(vcpu))
1528 			tmr = TIMER_HVTIMER;
1529 		else
1530 			tmr = TIMER_VTIMER;
1531 		treg = TIMER_REG_CVAL;
1532 		break;
1533 
1534 	case SYS_AARCH32_CNTP_CVAL:
1535 	case SYS_CNTP_CVAL_EL02:
1536 		tmr = TIMER_PTIMER;
1537 		treg = TIMER_REG_CVAL;
1538 		break;
1539 
1540 	case SYS_CNTV_CVAL_EL02:
1541 		tmr = TIMER_VTIMER;
1542 		treg = TIMER_REG_CVAL;
1543 		break;
1544 
1545 	case SYS_CNTHP_CVAL_EL2:
1546 		tmr = TIMER_HPTIMER;
1547 		treg = TIMER_REG_CVAL;
1548 		break;
1549 
1550 	case SYS_CNTHV_CVAL_EL2:
1551 		tmr = TIMER_HVTIMER;
1552 		treg = TIMER_REG_CVAL;
1553 		break;
1554 
1555 	case SYS_CNTPCT_EL0:
1556 	case SYS_CNTPCTSS_EL0:
1557 		if (is_hyp_ctxt(vcpu))
1558 			tmr = TIMER_HPTIMER;
1559 		else
1560 			tmr = TIMER_PTIMER;
1561 		treg = TIMER_REG_CNT;
1562 		break;
1563 
1564 	case SYS_AARCH32_CNTPCT:
1565 	case SYS_AARCH32_CNTPCTSS:
1566 		tmr = TIMER_PTIMER;
1567 		treg = TIMER_REG_CNT;
1568 		break;
1569 
1570 	case SYS_CNTVCT_EL0:
1571 	case SYS_CNTVCTSS_EL0:
1572 		if (is_hyp_ctxt(vcpu))
1573 			tmr = TIMER_HVTIMER;
1574 		else
1575 			tmr = TIMER_VTIMER;
1576 		treg = TIMER_REG_CNT;
1577 		break;
1578 
1579 	case SYS_AARCH32_CNTVCT:
1580 	case SYS_AARCH32_CNTVCTSS:
1581 		tmr = TIMER_VTIMER;
1582 		treg = TIMER_REG_CNT;
1583 		break;
1584 
1585 	default:
1586 		print_sys_reg_msg(p, "%s", "Unhandled trapped timer register");
1587 		return undef_access(vcpu, p, r);
1588 	}
1589 
1590 	if (p->is_write)
1591 		kvm_arm_timer_write_sysreg(vcpu, tmr, treg, p->regval);
1592 	else
1593 		p->regval = kvm_arm_timer_read_sysreg(vcpu, tmr, treg);
1594 
1595 	return true;
1596 }
1597 
access_hv_timer(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1598 static bool access_hv_timer(struct kvm_vcpu *vcpu,
1599 			    struct sys_reg_params *p,
1600 			    const struct sys_reg_desc *r)
1601 {
1602 	if (!vcpu_el2_e2h_is_set(vcpu))
1603 		return undef_access(vcpu, p, r);
1604 
1605 	return access_arch_timer(vcpu, p, r);
1606 }
1607 
kvm_arm64_ftr_safe_value(u32 id,const struct arm64_ftr_bits * ftrp,s64 new,s64 cur)1608 static s64 kvm_arm64_ftr_safe_value(u32 id, const struct arm64_ftr_bits *ftrp,
1609 				    s64 new, s64 cur)
1610 {
1611 	struct arm64_ftr_bits kvm_ftr = *ftrp;
1612 
1613 	/* Some features have different safe value type in KVM than host features */
1614 	switch (id) {
1615 	case SYS_ID_AA64DFR0_EL1:
1616 		switch (kvm_ftr.shift) {
1617 		case ID_AA64DFR0_EL1_PMUVer_SHIFT:
1618 			kvm_ftr.type = FTR_LOWER_SAFE;
1619 			break;
1620 		case ID_AA64DFR0_EL1_DebugVer_SHIFT:
1621 			kvm_ftr.type = FTR_LOWER_SAFE;
1622 			break;
1623 		}
1624 		break;
1625 	case SYS_ID_DFR0_EL1:
1626 		if (kvm_ftr.shift == ID_DFR0_EL1_PerfMon_SHIFT)
1627 			kvm_ftr.type = FTR_LOWER_SAFE;
1628 		break;
1629 	}
1630 
1631 	return arm64_ftr_safe_value(&kvm_ftr, new, cur);
1632 }
1633 
1634 /*
1635  * arm64_check_features() - Check if a feature register value constitutes
1636  * a subset of features indicated by the idreg's KVM sanitised limit.
1637  *
1638  * This function will check if each feature field of @val is the "safe" value
1639  * against idreg's KVM sanitised limit return from reset() callback.
1640  * If a field value in @val is the same as the one in limit, it is always
1641  * considered the safe value regardless For register fields that are not in
1642  * writable, only the value in limit is considered the safe value.
1643  *
1644  * Return: 0 if all the fields are safe. Otherwise, return negative errno.
1645  */
arm64_check_features(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)1646 static int arm64_check_features(struct kvm_vcpu *vcpu,
1647 				const struct sys_reg_desc *rd,
1648 				u64 val)
1649 {
1650 	const struct arm64_ftr_reg *ftr_reg;
1651 	const struct arm64_ftr_bits *ftrp = NULL;
1652 	u32 id = reg_to_encoding(rd);
1653 	u64 writable_mask = rd->val;
1654 	u64 limit = rd->reset(vcpu, rd);
1655 	u64 mask = 0;
1656 
1657 	/*
1658 	 * Hidden and unallocated ID registers may not have a corresponding
1659 	 * struct arm64_ftr_reg. Of course, if the register is RAZ we know the
1660 	 * only safe value is 0.
1661 	 */
1662 	if (sysreg_visible_as_raz(vcpu, rd))
1663 		return val ? -E2BIG : 0;
1664 
1665 	ftr_reg = get_arm64_ftr_reg(id);
1666 	if (!ftr_reg)
1667 		return -EINVAL;
1668 
1669 	ftrp = ftr_reg->ftr_bits;
1670 
1671 	for (; ftrp && ftrp->width; ftrp++) {
1672 		s64 f_val, f_lim, safe_val;
1673 		u64 ftr_mask;
1674 
1675 		ftr_mask = arm64_ftr_mask(ftrp);
1676 		if ((ftr_mask & writable_mask) != ftr_mask)
1677 			continue;
1678 
1679 		f_val = arm64_ftr_value(ftrp, val);
1680 		f_lim = arm64_ftr_value(ftrp, limit);
1681 		mask |= ftr_mask;
1682 
1683 		if (f_val == f_lim)
1684 			safe_val = f_val;
1685 		else
1686 			safe_val = kvm_arm64_ftr_safe_value(id, ftrp, f_val, f_lim);
1687 
1688 		if (safe_val != f_val)
1689 			return -E2BIG;
1690 	}
1691 
1692 	/* For fields that are not writable, values in limit are the safe values. */
1693 	if ((val & ~mask) != (limit & ~mask))
1694 		return -E2BIG;
1695 
1696 	return 0;
1697 }
1698 
pmuver_to_perfmon(u8 pmuver)1699 static u8 pmuver_to_perfmon(u8 pmuver)
1700 {
1701 	switch (pmuver) {
1702 	case ID_AA64DFR0_EL1_PMUVer_IMP:
1703 		return ID_DFR0_EL1_PerfMon_PMUv3;
1704 	case ID_AA64DFR0_EL1_PMUVer_IMP_DEF:
1705 		return ID_DFR0_EL1_PerfMon_IMPDEF;
1706 	default:
1707 		/* Anything ARMv8.1+ and NI have the same value. For now. */
1708 		return pmuver;
1709 	}
1710 }
1711 
1712 static u64 sanitise_id_aa64pfr0_el1(const struct kvm_vcpu *vcpu, u64 val);
1713 static u64 sanitise_id_aa64pfr1_el1(const struct kvm_vcpu *vcpu, u64 val);
1714 static u64 sanitise_id_aa64dfr0_el1(const struct kvm_vcpu *vcpu, u64 val);
1715 
1716 /* Read a sanitised cpufeature ID register by sys_reg_desc */
__kvm_read_sanitised_id_reg(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)1717 static u64 __kvm_read_sanitised_id_reg(const struct kvm_vcpu *vcpu,
1718 				       const struct sys_reg_desc *r)
1719 {
1720 	u32 id = reg_to_encoding(r);
1721 	u64 val;
1722 
1723 	if (sysreg_visible_as_raz(vcpu, r))
1724 		return 0;
1725 
1726 	val = read_sanitised_ftr_reg(id);
1727 
1728 	switch (id) {
1729 	case SYS_ID_AA64DFR0_EL1:
1730 		val = sanitise_id_aa64dfr0_el1(vcpu, val);
1731 		break;
1732 	case SYS_ID_AA64PFR0_EL1:
1733 		val = sanitise_id_aa64pfr0_el1(vcpu, val);
1734 		break;
1735 	case SYS_ID_AA64PFR1_EL1:
1736 		val = sanitise_id_aa64pfr1_el1(vcpu, val);
1737 		break;
1738 	case SYS_ID_AA64PFR2_EL1:
1739 		val &= ID_AA64PFR2_EL1_FPMR |
1740 			(kvm_has_mte(vcpu->kvm) ?
1741 			 ID_AA64PFR2_EL1_MTEFAR | ID_AA64PFR2_EL1_MTESTOREONLY :
1742 			 0);
1743 		break;
1744 	case SYS_ID_AA64ISAR1_EL1:
1745 		if (!vcpu_has_ptrauth(vcpu))
1746 			val &= ~(ID_AA64ISAR1_EL1_APA |
1747 				 ID_AA64ISAR1_EL1_API |
1748 				 ID_AA64ISAR1_EL1_GPA |
1749 				 ID_AA64ISAR1_EL1_GPI);
1750 		break;
1751 	case SYS_ID_AA64ISAR2_EL1:
1752 		if (!vcpu_has_ptrauth(vcpu))
1753 			val &= ~(ID_AA64ISAR2_EL1_APA3 |
1754 				 ID_AA64ISAR2_EL1_GPA3);
1755 		if (!cpus_have_final_cap(ARM64_HAS_WFXT) ||
1756 		    has_broken_cntvoff())
1757 			val &= ~ID_AA64ISAR2_EL1_WFxT;
1758 		break;
1759 	case SYS_ID_AA64ISAR3_EL1:
1760 		val &= ID_AA64ISAR3_EL1_FPRCVT | ID_AA64ISAR3_EL1_LSFE |
1761 			ID_AA64ISAR3_EL1_FAMINMAX;
1762 		break;
1763 	case SYS_ID_AA64MMFR2_EL1:
1764 		val &= ~ID_AA64MMFR2_EL1_CCIDX_MASK;
1765 		val &= ~ID_AA64MMFR2_EL1_NV;
1766 		break;
1767 	case SYS_ID_AA64MMFR3_EL1:
1768 		val &= ID_AA64MMFR3_EL1_TCRX |
1769 		       ID_AA64MMFR3_EL1_SCTLRX |
1770 		       ID_AA64MMFR3_EL1_S1POE |
1771 		       ID_AA64MMFR3_EL1_S1PIE;
1772 		break;
1773 	case SYS_ID_MMFR4_EL1:
1774 		val &= ~ID_MMFR4_EL1_CCIDX;
1775 		break;
1776 	}
1777 
1778 	if (vcpu_has_nv(vcpu))
1779 		val = limit_nv_id_reg(vcpu->kvm, id, val);
1780 
1781 	return val;
1782 }
1783 
kvm_read_sanitised_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)1784 static u64 kvm_read_sanitised_id_reg(struct kvm_vcpu *vcpu,
1785 				     const struct sys_reg_desc *r)
1786 {
1787 	return __kvm_read_sanitised_id_reg(vcpu, r);
1788 }
1789 
read_id_reg(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)1790 static u64 read_id_reg(const struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
1791 {
1792 	return kvm_read_vm_id_reg(vcpu->kvm, reg_to_encoding(r));
1793 }
1794 
is_feature_id_reg(u32 encoding)1795 static bool is_feature_id_reg(u32 encoding)
1796 {
1797 	return (sys_reg_Op0(encoding) == 3 &&
1798 		(sys_reg_Op1(encoding) < 2 || sys_reg_Op1(encoding) == 3) &&
1799 		sys_reg_CRn(encoding) == 0 &&
1800 		sys_reg_CRm(encoding) <= 7);
1801 }
1802 
1803 /*
1804  * Return true if the register's (Op0, Op1, CRn, CRm, Op2) is
1805  * (3, 0, 0, crm, op2), where 1<=crm<8, 0<=op2<8, which is the range of ID
1806  * registers KVM maintains on a per-VM basis.
1807  *
1808  * Additionally, the implementation ID registers and CTR_EL0 are handled as
1809  * per-VM registers.
1810  */
is_vm_ftr_id_reg(u32 id)1811 static inline bool is_vm_ftr_id_reg(u32 id)
1812 {
1813 	switch (id) {
1814 	case SYS_CTR_EL0:
1815 	case SYS_MIDR_EL1:
1816 	case SYS_REVIDR_EL1:
1817 	case SYS_AIDR_EL1:
1818 		return true;
1819 	default:
1820 		return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1821 			sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1822 			sys_reg_CRm(id) < 8);
1823 
1824 	}
1825 }
1826 
is_vcpu_ftr_id_reg(u32 id)1827 static inline bool is_vcpu_ftr_id_reg(u32 id)
1828 {
1829 	return is_feature_id_reg(id) && !is_vm_ftr_id_reg(id);
1830 }
1831 
is_aa32_id_reg(u32 id)1832 static inline bool is_aa32_id_reg(u32 id)
1833 {
1834 	return (sys_reg_Op0(id) == 3 && sys_reg_Op1(id) == 0 &&
1835 		sys_reg_CRn(id) == 0 && sys_reg_CRm(id) >= 1 &&
1836 		sys_reg_CRm(id) <= 3);
1837 }
1838 
id_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)1839 static unsigned int id_visibility(const struct kvm_vcpu *vcpu,
1840 				  const struct sys_reg_desc *r)
1841 {
1842 	u32 id = reg_to_encoding(r);
1843 
1844 	switch (id) {
1845 	case SYS_ID_AA64ZFR0_EL1:
1846 		if (!vcpu_has_sve(vcpu))
1847 			return REG_RAZ;
1848 		break;
1849 	}
1850 
1851 	return 0;
1852 }
1853 
aa32_id_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)1854 static unsigned int aa32_id_visibility(const struct kvm_vcpu *vcpu,
1855 				       const struct sys_reg_desc *r)
1856 {
1857 	/*
1858 	 * AArch32 ID registers are UNKNOWN if AArch32 isn't implemented at any
1859 	 * EL. Promote to RAZ/WI in order to guarantee consistency between
1860 	 * systems.
1861 	 */
1862 	if (!kvm_supports_32bit_el0())
1863 		return REG_RAZ | REG_USER_WI;
1864 
1865 	return id_visibility(vcpu, r);
1866 }
1867 
raz_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)1868 static unsigned int raz_visibility(const struct kvm_vcpu *vcpu,
1869 				   const struct sys_reg_desc *r)
1870 {
1871 	return REG_RAZ;
1872 }
1873 
1874 /* cpufeature ID register access trap handlers */
1875 
access_id_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)1876 static bool access_id_reg(struct kvm_vcpu *vcpu,
1877 			  struct sys_reg_params *p,
1878 			  const struct sys_reg_desc *r)
1879 {
1880 	if (p->is_write)
1881 		return write_to_read_only(vcpu, p, r);
1882 
1883 	p->regval = read_id_reg(vcpu, r);
1884 
1885 	return true;
1886 }
1887 
1888 /* Visibility overrides for SVE-specific control registers */
sve_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)1889 static unsigned int sve_visibility(const struct kvm_vcpu *vcpu,
1890 				   const struct sys_reg_desc *rd)
1891 {
1892 	if (vcpu_has_sve(vcpu))
1893 		return 0;
1894 
1895 	return REG_HIDDEN;
1896 }
1897 
sme_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)1898 static unsigned int sme_visibility(const struct kvm_vcpu *vcpu,
1899 				   const struct sys_reg_desc *rd)
1900 {
1901 	if (kvm_has_feat(vcpu->kvm, ID_AA64PFR1_EL1, SME, IMP))
1902 		return 0;
1903 
1904 	return REG_HIDDEN;
1905 }
1906 
fp8_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)1907 static unsigned int fp8_visibility(const struct kvm_vcpu *vcpu,
1908 				   const struct sys_reg_desc *rd)
1909 {
1910 	if (kvm_has_fpmr(vcpu->kvm))
1911 		return 0;
1912 
1913 	return REG_HIDDEN;
1914 }
1915 
sanitise_id_aa64pfr0_el1(const struct kvm_vcpu * vcpu,u64 val)1916 static u64 sanitise_id_aa64pfr0_el1(const struct kvm_vcpu *vcpu, u64 val)
1917 {
1918 	if (!vcpu_has_sve(vcpu))
1919 		val &= ~ID_AA64PFR0_EL1_SVE_MASK;
1920 
1921 	/*
1922 	 * The default is to expose CSV2 == 1 if the HW isn't affected.
1923 	 * Although this is a per-CPU feature, we make it global because
1924 	 * asymmetric systems are just a nuisance.
1925 	 *
1926 	 * Userspace can override this as long as it doesn't promise
1927 	 * the impossible.
1928 	 */
1929 	if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) {
1930 		val &= ~ID_AA64PFR0_EL1_CSV2_MASK;
1931 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV2, IMP);
1932 	}
1933 	if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) {
1934 		val &= ~ID_AA64PFR0_EL1_CSV3_MASK;
1935 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, CSV3, IMP);
1936 	}
1937 
1938 	if (vgic_is_v3(vcpu->kvm)) {
1939 		val &= ~ID_AA64PFR0_EL1_GIC_MASK;
1940 		val |= SYS_FIELD_PREP_ENUM(ID_AA64PFR0_EL1, GIC, IMP);
1941 	}
1942 
1943 	val &= ~ID_AA64PFR0_EL1_AMU_MASK;
1944 
1945 	/*
1946 	 * MPAM is disabled by default as KVM also needs a set of PARTID to
1947 	 * program the MPAMVPMx_EL2 PARTID remapping registers with. But some
1948 	 * older kernels let the guest see the ID bit.
1949 	 */
1950 	val &= ~ID_AA64PFR0_EL1_MPAM_MASK;
1951 
1952 	return val;
1953 }
1954 
sanitise_id_aa64pfr1_el1(const struct kvm_vcpu * vcpu,u64 val)1955 static u64 sanitise_id_aa64pfr1_el1(const struct kvm_vcpu *vcpu, u64 val)
1956 {
1957 	u64 pfr0 = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1958 
1959 	if (!kvm_has_mte(vcpu->kvm)) {
1960 		val &= ~ID_AA64PFR1_EL1_MTE;
1961 		val &= ~ID_AA64PFR1_EL1_MTE_frac;
1962 	}
1963 
1964 	if (!(cpus_have_final_cap(ARM64_HAS_RASV1P1_EXTN) &&
1965 	      SYS_FIELD_GET(ID_AA64PFR0_EL1, RAS, pfr0) == ID_AA64PFR0_EL1_RAS_IMP))
1966 		val &= ~ID_AA64PFR1_EL1_RAS_frac;
1967 
1968 	val &= ~ID_AA64PFR1_EL1_SME;
1969 	val &= ~ID_AA64PFR1_EL1_RNDR_trap;
1970 	val &= ~ID_AA64PFR1_EL1_NMI;
1971 	val &= ~ID_AA64PFR1_EL1_GCS;
1972 	val &= ~ID_AA64PFR1_EL1_THE;
1973 	val &= ~ID_AA64PFR1_EL1_MTEX;
1974 	val &= ~ID_AA64PFR1_EL1_PFAR;
1975 	val &= ~ID_AA64PFR1_EL1_MPAM_frac;
1976 
1977 	return val;
1978 }
1979 
sanitise_id_aa64dfr0_el1(const struct kvm_vcpu * vcpu,u64 val)1980 static u64 sanitise_id_aa64dfr0_el1(const struct kvm_vcpu *vcpu, u64 val)
1981 {
1982 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_AA64DFR0_EL1, DebugVer, V8P8);
1983 
1984 	/*
1985 	 * Only initialize the PMU version if the vCPU was configured with one.
1986 	 */
1987 	val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
1988 	if (kvm_vcpu_has_pmu(vcpu))
1989 		val |= SYS_FIELD_PREP(ID_AA64DFR0_EL1, PMUVer,
1990 				      kvm_arm_pmu_get_pmuver_limit());
1991 
1992 	/* Hide SPE from guests */
1993 	val &= ~ID_AA64DFR0_EL1_PMSVer_MASK;
1994 
1995 	/* Hide BRBE from guests */
1996 	val &= ~ID_AA64DFR0_EL1_BRBE_MASK;
1997 
1998 	return val;
1999 }
2000 
2001 /*
2002  * Older versions of KVM erroneously claim support for FEAT_DoubleLock with
2003  * NV-enabled VMs on unsupporting hardware. Silently ignore the incorrect
2004  * value if it is consistent with the bug.
2005  */
ignore_feat_doublelock(struct kvm_vcpu * vcpu,u64 val)2006 static bool ignore_feat_doublelock(struct kvm_vcpu *vcpu, u64 val)
2007 {
2008 	u8 host, user;
2009 
2010 	if (!vcpu_has_nv(vcpu))
2011 		return false;
2012 
2013 	host = SYS_FIELD_GET(ID_AA64DFR0_EL1, DoubleLock,
2014 			     read_sanitised_ftr_reg(SYS_ID_AA64DFR0_EL1));
2015 	user = SYS_FIELD_GET(ID_AA64DFR0_EL1, DoubleLock, val);
2016 
2017 	return host == ID_AA64DFR0_EL1_DoubleLock_NI &&
2018 	       user == ID_AA64DFR0_EL1_DoubleLock_IMP;
2019 }
2020 
set_id_aa64dfr0_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)2021 static int set_id_aa64dfr0_el1(struct kvm_vcpu *vcpu,
2022 			       const struct sys_reg_desc *rd,
2023 			       u64 val)
2024 {
2025 	u8 debugver = SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, val);
2026 	u8 pmuver = SYS_FIELD_GET(ID_AA64DFR0_EL1, PMUVer, val);
2027 
2028 	/*
2029 	 * Prior to commit 3d0dba5764b9 ("KVM: arm64: PMU: Move the
2030 	 * ID_AA64DFR0_EL1.PMUver limit to VM creation"), KVM erroneously
2031 	 * exposed an IMP_DEF PMU to userspace and the guest on systems w/
2032 	 * non-architectural PMUs. Of course, PMUv3 is the only game in town for
2033 	 * PMU virtualization, so the IMP_DEF value was rather user-hostile.
2034 	 *
2035 	 * At minimum, we're on the hook to allow values that were given to
2036 	 * userspace by KVM. Cover our tracks here and replace the IMP_DEF value
2037 	 * with a more sensible NI. The value of an ID register changing under
2038 	 * the nose of the guest is unfortunate, but is certainly no more
2039 	 * surprising than an ill-guided PMU driver poking at impdef system
2040 	 * registers that end in an UNDEF...
2041 	 */
2042 	if (pmuver == ID_AA64DFR0_EL1_PMUVer_IMP_DEF)
2043 		val &= ~ID_AA64DFR0_EL1_PMUVer_MASK;
2044 
2045 	/*
2046 	 * ID_AA64DFR0_EL1.DebugVer is one of those awkward fields with a
2047 	 * nonzero minimum safe value.
2048 	 */
2049 	if (debugver < ID_AA64DFR0_EL1_DebugVer_IMP)
2050 		return -EINVAL;
2051 
2052 	if (ignore_feat_doublelock(vcpu, val)) {
2053 		val &= ~ID_AA64DFR0_EL1_DoubleLock;
2054 		val |= SYS_FIELD_PREP_ENUM(ID_AA64DFR0_EL1, DoubleLock, NI);
2055 	}
2056 
2057 	return set_id_reg(vcpu, rd, val);
2058 }
2059 
read_sanitised_id_dfr0_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2060 static u64 read_sanitised_id_dfr0_el1(struct kvm_vcpu *vcpu,
2061 				      const struct sys_reg_desc *rd)
2062 {
2063 	u8 perfmon;
2064 	u64 val = read_sanitised_ftr_reg(SYS_ID_DFR0_EL1);
2065 
2066 	val &= ~ID_DFR0_EL1_PerfMon_MASK;
2067 	if (kvm_vcpu_has_pmu(vcpu)) {
2068 		perfmon = pmuver_to_perfmon(kvm_arm_pmu_get_pmuver_limit());
2069 		val |= SYS_FIELD_PREP(ID_DFR0_EL1, PerfMon, perfmon);
2070 	}
2071 
2072 	val = ID_REG_LIMIT_FIELD_ENUM(val, ID_DFR0_EL1, CopDbg, Debugv8p8);
2073 
2074 	return val;
2075 }
2076 
set_id_dfr0_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)2077 static int set_id_dfr0_el1(struct kvm_vcpu *vcpu,
2078 			   const struct sys_reg_desc *rd,
2079 			   u64 val)
2080 {
2081 	u8 perfmon = SYS_FIELD_GET(ID_DFR0_EL1, PerfMon, val);
2082 	u8 copdbg = SYS_FIELD_GET(ID_DFR0_EL1, CopDbg, val);
2083 
2084 	if (perfmon == ID_DFR0_EL1_PerfMon_IMPDEF) {
2085 		val &= ~ID_DFR0_EL1_PerfMon_MASK;
2086 		perfmon = 0;
2087 	}
2088 
2089 	/*
2090 	 * Allow DFR0_EL1.PerfMon to be set from userspace as long as
2091 	 * it doesn't promise more than what the HW gives us on the
2092 	 * AArch64 side (as everything is emulated with that), and
2093 	 * that this is a PMUv3.
2094 	 */
2095 	if (perfmon != 0 && perfmon < ID_DFR0_EL1_PerfMon_PMUv3)
2096 		return -EINVAL;
2097 
2098 	if (copdbg < ID_DFR0_EL1_CopDbg_Armv8)
2099 		return -EINVAL;
2100 
2101 	return set_id_reg(vcpu, rd, val);
2102 }
2103 
set_id_aa64pfr0_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 user_val)2104 static int set_id_aa64pfr0_el1(struct kvm_vcpu *vcpu,
2105 			       const struct sys_reg_desc *rd, u64 user_val)
2106 {
2107 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2108 	u64 mpam_mask = ID_AA64PFR0_EL1_MPAM_MASK;
2109 
2110 	/*
2111 	 * Commit 011e5f5bf529f ("arm64/cpufeature: Add remaining feature bits
2112 	 * in ID_AA64PFR0 register") exposed the MPAM field of AA64PFR0_EL1 to
2113 	 * guests, but didn't add trap handling. KVM doesn't support MPAM and
2114 	 * always returns an UNDEF for these registers. The guest must see 0
2115 	 * for this field.
2116 	 *
2117 	 * But KVM must also accept values from user-space that were provided
2118 	 * by KVM. On CPUs that support MPAM, permit user-space to write
2119 	 * the sanitizied value to ID_AA64PFR0_EL1.MPAM, but ignore this field.
2120 	 */
2121 	if ((hw_val & mpam_mask) == (user_val & mpam_mask))
2122 		user_val &= ~ID_AA64PFR0_EL1_MPAM_MASK;
2123 
2124 	/* Fail the guest's request to disable the AA64 ISA at EL{0,1,2} */
2125 	if (!FIELD_GET(ID_AA64PFR0_EL1_EL0, user_val) ||
2126 	    !FIELD_GET(ID_AA64PFR0_EL1_EL1, user_val) ||
2127 	    (vcpu_has_nv(vcpu) && !FIELD_GET(ID_AA64PFR0_EL1_EL2, user_val)))
2128 		return -EINVAL;
2129 
2130 	/*
2131 	 * If we are running on a GICv5 host and support FEAT_GCIE_LEGACY, then
2132 	 * we support GICv3. Fail attempts to do anything but set that to IMP.
2133 	 */
2134 	if (vgic_is_v3_compat(vcpu->kvm) &&
2135 	    FIELD_GET(ID_AA64PFR0_EL1_GIC_MASK, user_val) != ID_AA64PFR0_EL1_GIC_IMP)
2136 		return -EINVAL;
2137 
2138 	return set_id_reg(vcpu, rd, user_val);
2139 }
2140 
set_id_aa64pfr1_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 user_val)2141 static int set_id_aa64pfr1_el1(struct kvm_vcpu *vcpu,
2142 			       const struct sys_reg_desc *rd, u64 user_val)
2143 {
2144 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2145 	u64 mpam_mask = ID_AA64PFR1_EL1_MPAM_frac_MASK;
2146 	u8 mte = SYS_FIELD_GET(ID_AA64PFR1_EL1, MTE, hw_val);
2147 	u8 user_mte_frac = SYS_FIELD_GET(ID_AA64PFR1_EL1, MTE_frac, user_val);
2148 	u8 hw_mte_frac = SYS_FIELD_GET(ID_AA64PFR1_EL1, MTE_frac, hw_val);
2149 
2150 	/* See set_id_aa64pfr0_el1 for comment about MPAM */
2151 	if ((hw_val & mpam_mask) == (user_val & mpam_mask))
2152 		user_val &= ~ID_AA64PFR1_EL1_MPAM_frac_MASK;
2153 
2154 	/*
2155 	 * Previously MTE_frac was hidden from guest. However, if the
2156 	 * hardware supports MTE2 but not MTE_ASYM_FAULT then a value
2157 	 * of 0 for this field indicates that the hardware supports
2158 	 * MTE_ASYNC. Whereas, 0xf indicates MTE_ASYNC is not supported.
2159 	 *
2160 	 * As KVM must accept values from KVM provided by user-space,
2161 	 * when ID_AA64PFR1_EL1.MTE is 2 allow user-space to set
2162 	 * ID_AA64PFR1_EL1.MTE_frac to 0. However, ignore it to avoid
2163 	 * incorrectly claiming hardware support for MTE_ASYNC in the
2164 	 * guest.
2165 	 */
2166 
2167 	if (mte == ID_AA64PFR1_EL1_MTE_MTE2 &&
2168 	    hw_mte_frac == ID_AA64PFR1_EL1_MTE_frac_NI &&
2169 	    user_mte_frac == ID_AA64PFR1_EL1_MTE_frac_ASYNC) {
2170 		user_val &= ~ID_AA64PFR1_EL1_MTE_frac_MASK;
2171 		user_val |= hw_val & ID_AA64PFR1_EL1_MTE_frac_MASK;
2172 	}
2173 
2174 	return set_id_reg(vcpu, rd, user_val);
2175 }
2176 
2177 /*
2178  * Allow userspace to de-feature a stage-2 translation granule but prevent it
2179  * from claiming the impossible.
2180  */
2181 #define tgran2_val_allowed(tg, safe, user)			\
2182 ({								\
2183 	u8 __s = SYS_FIELD_GET(ID_AA64MMFR0_EL1, tg, safe);	\
2184 	u8 __u = SYS_FIELD_GET(ID_AA64MMFR0_EL1, tg, user);	\
2185 								\
2186 	__s == __u || __u == ID_AA64MMFR0_EL1_##tg##_NI;	\
2187 })
2188 
set_id_aa64mmfr0_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 user_val)2189 static int set_id_aa64mmfr0_el1(struct kvm_vcpu *vcpu,
2190 				const struct sys_reg_desc *rd, u64 user_val)
2191 {
2192 	u64 sanitized_val = kvm_read_sanitised_id_reg(vcpu, rd);
2193 
2194 	if (!vcpu_has_nv(vcpu))
2195 		return set_id_reg(vcpu, rd, user_val);
2196 
2197 	if (!tgran2_val_allowed(TGRAN4_2, sanitized_val, user_val) ||
2198 	    !tgran2_val_allowed(TGRAN16_2, sanitized_val, user_val) ||
2199 	    !tgran2_val_allowed(TGRAN64_2, sanitized_val, user_val))
2200 		return -EINVAL;
2201 
2202 	return set_id_reg(vcpu, rd, user_val);
2203 }
2204 
set_id_aa64mmfr2_el1(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 user_val)2205 static int set_id_aa64mmfr2_el1(struct kvm_vcpu *vcpu,
2206 				const struct sys_reg_desc *rd, u64 user_val)
2207 {
2208 	u64 hw_val = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2209 	u64 nv_mask = ID_AA64MMFR2_EL1_NV_MASK;
2210 
2211 	/*
2212 	 * We made the mistake to expose the now deprecated NV field,
2213 	 * so allow userspace to write it, but silently ignore it.
2214 	 */
2215 	if ((hw_val & nv_mask) == (user_val & nv_mask))
2216 		user_val &= ~nv_mask;
2217 
2218 	return set_id_reg(vcpu, rd, user_val);
2219 }
2220 
set_ctr_el0(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 user_val)2221 static int set_ctr_el0(struct kvm_vcpu *vcpu,
2222 		       const struct sys_reg_desc *rd, u64 user_val)
2223 {
2224 	u8 user_L1Ip = SYS_FIELD_GET(CTR_EL0, L1Ip, user_val);
2225 
2226 	/*
2227 	 * Both AIVIVT (0b01) and VPIPT (0b00) are documented as reserved.
2228 	 * Hence only allow to set VIPT(0b10) or PIPT(0b11) for L1Ip based
2229 	 * on what hardware reports.
2230 	 *
2231 	 * Using a VIPT software model on PIPT will lead to over invalidation,
2232 	 * but still correct. Hence, we can allow downgrading PIPT to VIPT,
2233 	 * but not the other way around. This is handled via arm64_ftr_safe_value()
2234 	 * as CTR_EL0 ftr_bits has L1Ip field with type FTR_EXACT and safe value
2235 	 * set as VIPT.
2236 	 */
2237 	switch (user_L1Ip) {
2238 	case CTR_EL0_L1Ip_RESERVED_VPIPT:
2239 	case CTR_EL0_L1Ip_RESERVED_AIVIVT:
2240 		return -EINVAL;
2241 	case CTR_EL0_L1Ip_VIPT:
2242 	case CTR_EL0_L1Ip_PIPT:
2243 		return set_id_reg(vcpu, rd, user_val);
2244 	default:
2245 		return -ENOENT;
2246 	}
2247 }
2248 
2249 /*
2250  * cpufeature ID register user accessors
2251  *
2252  * For now, these registers are immutable for userspace, so no values
2253  * are stored, and for set_id_reg() we don't allow the effective value
2254  * to be changed.
2255  */
get_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 * val)2256 static int get_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2257 		      u64 *val)
2258 {
2259 	/*
2260 	 * Avoid locking if the VM has already started, as the ID registers are
2261 	 * guaranteed to be invariant at that point.
2262 	 */
2263 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
2264 		*val = read_id_reg(vcpu, rd);
2265 		return 0;
2266 	}
2267 
2268 	mutex_lock(&vcpu->kvm->arch.config_lock);
2269 	*val = read_id_reg(vcpu, rd);
2270 	mutex_unlock(&vcpu->kvm->arch.config_lock);
2271 
2272 	return 0;
2273 }
2274 
set_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)2275 static int set_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2276 		      u64 val)
2277 {
2278 	u32 id = reg_to_encoding(rd);
2279 	int ret;
2280 
2281 	mutex_lock(&vcpu->kvm->arch.config_lock);
2282 
2283 	/*
2284 	 * Once the VM has started the ID registers are immutable. Reject any
2285 	 * write that does not match the final register value.
2286 	 */
2287 	if (kvm_vm_has_ran_once(vcpu->kvm)) {
2288 		if (val != read_id_reg(vcpu, rd))
2289 			ret = -EBUSY;
2290 		else
2291 			ret = 0;
2292 
2293 		mutex_unlock(&vcpu->kvm->arch.config_lock);
2294 		return ret;
2295 	}
2296 
2297 	ret = arm64_check_features(vcpu, rd, val);
2298 	if (!ret)
2299 		kvm_set_vm_id_reg(vcpu->kvm, id, val);
2300 
2301 	mutex_unlock(&vcpu->kvm->arch.config_lock);
2302 
2303 	/*
2304 	 * arm64_check_features() returns -E2BIG to indicate the register's
2305 	 * feature set is a superset of the maximally-allowed register value.
2306 	 * While it would be nice to precisely describe this to userspace, the
2307 	 * existing UAPI for KVM_SET_ONE_REG has it that invalid register
2308 	 * writes return -EINVAL.
2309 	 */
2310 	if (ret == -E2BIG)
2311 		ret = -EINVAL;
2312 	return ret;
2313 }
2314 
kvm_set_vm_id_reg(struct kvm * kvm,u32 reg,u64 val)2315 void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val)
2316 {
2317 	u64 *p = __vm_id_reg(&kvm->arch, reg);
2318 
2319 	lockdep_assert_held(&kvm->arch.config_lock);
2320 
2321 	if (KVM_BUG_ON(kvm_vm_has_ran_once(kvm) || !p, kvm))
2322 		return;
2323 
2324 	*p = val;
2325 }
2326 
get_raz_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 * val)2327 static int get_raz_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2328 		       u64 *val)
2329 {
2330 	*val = 0;
2331 	return 0;
2332 }
2333 
set_wi_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)2334 static int set_wi_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2335 		      u64 val)
2336 {
2337 	return 0;
2338 }
2339 
access_ctr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2340 static bool access_ctr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2341 		       const struct sys_reg_desc *r)
2342 {
2343 	if (p->is_write)
2344 		return write_to_read_only(vcpu, p, r);
2345 
2346 	p->regval = kvm_read_vm_id_reg(vcpu->kvm, SYS_CTR_EL0);
2347 	return true;
2348 }
2349 
access_clidr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2350 static bool access_clidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2351 			 const struct sys_reg_desc *r)
2352 {
2353 	if (p->is_write)
2354 		return write_to_read_only(vcpu, p, r);
2355 
2356 	p->regval = __vcpu_sys_reg(vcpu, r->reg);
2357 	return true;
2358 }
2359 
2360 /*
2361  * Fabricate a CLIDR_EL1 value instead of using the real value, which can vary
2362  * by the physical CPU which the vcpu currently resides in.
2363  */
reset_clidr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)2364 static u64 reset_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2365 {
2366 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2367 	u64 clidr;
2368 	u8 loc;
2369 
2370 	if ((ctr_el0 & CTR_EL0_IDC)) {
2371 		/*
2372 		 * Data cache clean to the PoU is not required so LoUU and LoUIS
2373 		 * will not be set and a unified cache, which will be marked as
2374 		 * LoC, will be added.
2375 		 *
2376 		 * If not DIC, let the unified cache L2 so that an instruction
2377 		 * cache can be added as L1 later.
2378 		 */
2379 		loc = (ctr_el0 & CTR_EL0_DIC) ? 1 : 2;
2380 		clidr = CACHE_TYPE_UNIFIED << CLIDR_CTYPE_SHIFT(loc);
2381 	} else {
2382 		/*
2383 		 * Data cache clean to the PoU is required so let L1 have a data
2384 		 * cache and mark it as LoUU and LoUIS. As L1 has a data cache,
2385 		 * it can be marked as LoC too.
2386 		 */
2387 		loc = 1;
2388 		clidr = 1 << CLIDR_LOUU_SHIFT;
2389 		clidr |= 1 << CLIDR_LOUIS_SHIFT;
2390 		clidr |= CACHE_TYPE_DATA << CLIDR_CTYPE_SHIFT(1);
2391 	}
2392 
2393 	/*
2394 	 * Instruction cache invalidation to the PoU is required so let L1 have
2395 	 * an instruction cache. If L1 already has a data cache, it will be
2396 	 * CACHE_TYPE_SEPARATE.
2397 	 */
2398 	if (!(ctr_el0 & CTR_EL0_DIC))
2399 		clidr |= CACHE_TYPE_INST << CLIDR_CTYPE_SHIFT(1);
2400 
2401 	clidr |= loc << CLIDR_LOC_SHIFT;
2402 
2403 	/*
2404 	 * Add tag cache unified to data cache. Allocation tags and data are
2405 	 * unified in a cache line so that it looks valid even if there is only
2406 	 * one cache line.
2407 	 */
2408 	if (kvm_has_mte(vcpu->kvm))
2409 		clidr |= 2ULL << CLIDR_TTYPE_SHIFT(loc);
2410 
2411 	__vcpu_assign_sys_reg(vcpu, r->reg, clidr);
2412 
2413 	return __vcpu_sys_reg(vcpu, r->reg);
2414 }
2415 
set_clidr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 val)2416 static int set_clidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *rd,
2417 		      u64 val)
2418 {
2419 	u64 ctr_el0 = read_sanitised_ftr_reg(SYS_CTR_EL0);
2420 	u64 idc = !CLIDR_LOC(val) || (!CLIDR_LOUIS(val) && !CLIDR_LOUU(val));
2421 
2422 	if ((val & CLIDR_EL1_RES0) || (!(ctr_el0 & CTR_EL0_IDC) && idc))
2423 		return -EINVAL;
2424 
2425 	__vcpu_assign_sys_reg(vcpu, rd->reg, val);
2426 
2427 	return 0;
2428 }
2429 
access_csselr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2430 static bool access_csselr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2431 			  const struct sys_reg_desc *r)
2432 {
2433 	int reg = r->reg;
2434 
2435 	if (p->is_write)
2436 		vcpu_write_sys_reg(vcpu, p->regval, reg);
2437 	else
2438 		p->regval = vcpu_read_sys_reg(vcpu, reg);
2439 	return true;
2440 }
2441 
access_ccsidr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2442 static bool access_ccsidr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
2443 			  const struct sys_reg_desc *r)
2444 {
2445 	u32 csselr;
2446 
2447 	if (p->is_write)
2448 		return write_to_read_only(vcpu, p, r);
2449 
2450 	csselr = vcpu_read_sys_reg(vcpu, CSSELR_EL1);
2451 	csselr &= CSSELR_EL1_Level | CSSELR_EL1_InD;
2452 	if (csselr < CSSELR_MAX)
2453 		p->regval = get_ccsidr(vcpu, csselr);
2454 
2455 	return true;
2456 }
2457 
mte_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2458 static unsigned int mte_visibility(const struct kvm_vcpu *vcpu,
2459 				   const struct sys_reg_desc *rd)
2460 {
2461 	if (kvm_has_mte(vcpu->kvm))
2462 		return 0;
2463 
2464 	return REG_HIDDEN;
2465 }
2466 
2467 #define MTE_REG(name) {				\
2468 	SYS_DESC(SYS_##name),			\
2469 	.access = undef_access,			\
2470 	.reset = reset_unknown,			\
2471 	.reg = name,				\
2472 	.visibility = mte_visibility,		\
2473 }
2474 
el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2475 static unsigned int el2_visibility(const struct kvm_vcpu *vcpu,
2476 				   const struct sys_reg_desc *rd)
2477 {
2478 	if (vcpu_has_nv(vcpu))
2479 		return 0;
2480 
2481 	return REG_HIDDEN;
2482 }
2483 
bad_vncr_trap(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2484 static bool bad_vncr_trap(struct kvm_vcpu *vcpu,
2485 			  struct sys_reg_params *p,
2486 			  const struct sys_reg_desc *r)
2487 {
2488 	/*
2489 	 * We really shouldn't be here, and this is likely the result
2490 	 * of a misconfigured trap, as this register should target the
2491 	 * VNCR page, and nothing else.
2492 	 */
2493 	return bad_trap(vcpu, p, r,
2494 			"trap of VNCR-backed register");
2495 }
2496 
bad_redir_trap(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2497 static bool bad_redir_trap(struct kvm_vcpu *vcpu,
2498 			   struct sys_reg_params *p,
2499 			   const struct sys_reg_desc *r)
2500 {
2501 	/*
2502 	 * We really shouldn't be here, and this is likely the result
2503 	 * of a misconfigured trap, as this register should target the
2504 	 * corresponding EL1, and nothing else.
2505 	 */
2506 	return bad_trap(vcpu, p, r,
2507 			"trap of EL2 register redirected to EL1");
2508 }
2509 
2510 #define EL2_REG_FILTERED(name, acc, rst, v, filter) {	\
2511 	SYS_DESC(SYS_##name),			\
2512 	.access = acc,				\
2513 	.reset = rst,				\
2514 	.reg = name,				\
2515 	.visibility = filter,			\
2516 	.val = v,				\
2517 }
2518 
2519 #define EL2_REG(name, acc, rst, v)			\
2520 	EL2_REG_FILTERED(name, acc, rst, v, el2_visibility)
2521 
2522 #define EL2_REG_VNCR(name, rst, v)	EL2_REG(name, bad_vncr_trap, rst, v)
2523 #define EL2_REG_VNCR_FILT(name, vis)			\
2524 	EL2_REG_FILTERED(name, bad_vncr_trap, reset_val, 0, vis)
2525 #define EL2_REG_VNCR_GICv3(name)			\
2526 	EL2_REG_VNCR_FILT(name, hidden_visibility)
2527 #define EL2_REG_REDIR(name, rst, v)	EL2_REG(name, bad_redir_trap, rst, v)
2528 
2529 /*
2530  * Since reset() callback and field val are not used for idregs, they will be
2531  * used for specific purposes for idregs.
2532  * The reset() would return KVM sanitised register value. The value would be the
2533  * same as the host kernel sanitised value if there is no KVM sanitisation.
2534  * The val would be used as a mask indicating writable fields for the idreg.
2535  * Only bits with 1 are writable from userspace. This mask might not be
2536  * necessary in the future whenever all ID registers are enabled as writable
2537  * from userspace.
2538  */
2539 
2540 #define ID_DESC_DEFAULT_CALLBACKS		\
2541 	.access	= access_id_reg,		\
2542 	.get_user = get_id_reg,			\
2543 	.set_user = set_id_reg,			\
2544 	.visibility = id_visibility,		\
2545 	.reset = kvm_read_sanitised_id_reg
2546 
2547 #define ID_DESC(name)				\
2548 	SYS_DESC(SYS_##name),			\
2549 	ID_DESC_DEFAULT_CALLBACKS
2550 
2551 /* sys_reg_desc initialiser for known cpufeature ID registers */
2552 #define ID_SANITISED(name) {			\
2553 	ID_DESC(name),				\
2554 	.val = 0,				\
2555 }
2556 
2557 /* sys_reg_desc initialiser for known cpufeature ID registers */
2558 #define AA32_ID_SANITISED(name) {		\
2559 	ID_DESC(name),				\
2560 	.visibility = aa32_id_visibility,	\
2561 	.val = 0,				\
2562 }
2563 
2564 /* sys_reg_desc initialiser for writable ID registers */
2565 #define ID_WRITABLE(name, mask) {		\
2566 	ID_DESC(name),				\
2567 	.val = mask,				\
2568 }
2569 
2570 /* sys_reg_desc initialiser for cpufeature ID registers that need filtering */
2571 #define ID_FILTERED(sysreg, name, mask) {	\
2572 	ID_DESC(sysreg),				\
2573 	.set_user = set_##name,				\
2574 	.val = (mask),					\
2575 }
2576 
2577 /*
2578  * sys_reg_desc initialiser for architecturally unallocated cpufeature ID
2579  * register with encoding Op0=3, Op1=0, CRn=0, CRm=crm, Op2=op2
2580  * (1 <= crm < 8, 0 <= Op2 < 8).
2581  */
2582 #define ID_UNALLOCATED(crm, op2) {			\
2583 	.name = "S3_0_0_" #crm "_" #op2,		\
2584 	Op0(3), Op1(0), CRn(0), CRm(crm), Op2(op2),	\
2585 	ID_DESC_DEFAULT_CALLBACKS,			\
2586 	.visibility = raz_visibility,			\
2587 	.val = 0,					\
2588 }
2589 
2590 /*
2591  * sys_reg_desc initialiser for known ID registers that we hide from guests.
2592  * For now, these are exposed just like unallocated ID regs: they appear
2593  * RAZ for the guest.
2594  */
2595 #define ID_HIDDEN(name) {			\
2596 	ID_DESC(name),				\
2597 	.visibility = raz_visibility,		\
2598 	.val = 0,				\
2599 }
2600 
access_sp_el1(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2601 static bool access_sp_el1(struct kvm_vcpu *vcpu,
2602 			  struct sys_reg_params *p,
2603 			  const struct sys_reg_desc *r)
2604 {
2605 	if (p->is_write)
2606 		__vcpu_assign_sys_reg(vcpu, SP_EL1, p->regval);
2607 	else
2608 		p->regval = __vcpu_sys_reg(vcpu, SP_EL1);
2609 
2610 	return true;
2611 }
2612 
access_elr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2613 static bool access_elr(struct kvm_vcpu *vcpu,
2614 		       struct sys_reg_params *p,
2615 		       const struct sys_reg_desc *r)
2616 {
2617 	if (p->is_write)
2618 		vcpu_write_sys_reg(vcpu, p->regval, ELR_EL1);
2619 	else
2620 		p->regval = vcpu_read_sys_reg(vcpu, ELR_EL1);
2621 
2622 	return true;
2623 }
2624 
access_spsr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2625 static bool access_spsr(struct kvm_vcpu *vcpu,
2626 			struct sys_reg_params *p,
2627 			const struct sys_reg_desc *r)
2628 {
2629 	if (p->is_write)
2630 		__vcpu_assign_sys_reg(vcpu, SPSR_EL1, p->regval);
2631 	else
2632 		p->regval = __vcpu_sys_reg(vcpu, SPSR_EL1);
2633 
2634 	return true;
2635 }
2636 
access_cntkctl_el12(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2637 static bool access_cntkctl_el12(struct kvm_vcpu *vcpu,
2638 				struct sys_reg_params *p,
2639 				const struct sys_reg_desc *r)
2640 {
2641 	if (p->is_write)
2642 		__vcpu_assign_sys_reg(vcpu, CNTKCTL_EL1, p->regval);
2643 	else
2644 		p->regval = __vcpu_sys_reg(vcpu, CNTKCTL_EL1);
2645 
2646 	return true;
2647 }
2648 
reset_hcr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)2649 static u64 reset_hcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2650 {
2651 	u64 val = r->val;
2652 
2653 	if (!cpus_have_final_cap(ARM64_HAS_HCR_NV1))
2654 		val |= HCR_E2H;
2655 
2656 	__vcpu_assign_sys_reg(vcpu, r->reg, val);
2657 
2658 	return __vcpu_sys_reg(vcpu, r->reg);
2659 }
2660 
__el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,unsigned int (* fn)(const struct kvm_vcpu *,const struct sys_reg_desc *))2661 static unsigned int __el2_visibility(const struct kvm_vcpu *vcpu,
2662 				     const struct sys_reg_desc *rd,
2663 				     unsigned int (*fn)(const struct kvm_vcpu *,
2664 							const struct sys_reg_desc *))
2665 {
2666 	return el2_visibility(vcpu, rd) ?: fn(vcpu, rd);
2667 }
2668 
sve_el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2669 static unsigned int sve_el2_visibility(const struct kvm_vcpu *vcpu,
2670 				       const struct sys_reg_desc *rd)
2671 {
2672 	return __el2_visibility(vcpu, rd, sve_visibility);
2673 }
2674 
vncr_el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2675 static unsigned int vncr_el2_visibility(const struct kvm_vcpu *vcpu,
2676 					const struct sys_reg_desc *rd)
2677 {
2678 	if (el2_visibility(vcpu, rd) == 0 &&
2679 	    kvm_has_feat(vcpu->kvm, ID_AA64MMFR4_EL1, NV_frac, NV2_ONLY))
2680 		return 0;
2681 
2682 	return REG_HIDDEN;
2683 }
2684 
sctlr2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2685 static unsigned int sctlr2_visibility(const struct kvm_vcpu *vcpu,
2686 				      const struct sys_reg_desc *rd)
2687 {
2688 	if (kvm_has_sctlr2(vcpu->kvm))
2689 		return 0;
2690 
2691 	return REG_HIDDEN;
2692 }
2693 
sctlr2_el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2694 static unsigned int sctlr2_el2_visibility(const struct kvm_vcpu *vcpu,
2695 					  const struct sys_reg_desc *rd)
2696 {
2697 	return __el2_visibility(vcpu, rd, sctlr2_visibility);
2698 }
2699 
access_zcr_el2(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2700 static bool access_zcr_el2(struct kvm_vcpu *vcpu,
2701 			   struct sys_reg_params *p,
2702 			   const struct sys_reg_desc *r)
2703 {
2704 	unsigned int vq;
2705 
2706 	if (guest_hyp_sve_traps_enabled(vcpu)) {
2707 		kvm_inject_nested_sve_trap(vcpu);
2708 		return true;
2709 	}
2710 
2711 	if (!p->is_write) {
2712 		p->regval = vcpu_read_sys_reg(vcpu, ZCR_EL2);
2713 		return true;
2714 	}
2715 
2716 	vq = SYS_FIELD_GET(ZCR_ELx, LEN, p->regval) + 1;
2717 	vq = min(vq, vcpu_sve_max_vq(vcpu));
2718 	vcpu_write_sys_reg(vcpu, vq - 1, ZCR_EL2);
2719 
2720 	return true;
2721 }
2722 
access_gic_vtr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2723 static bool access_gic_vtr(struct kvm_vcpu *vcpu,
2724 			   struct sys_reg_params *p,
2725 			   const struct sys_reg_desc *r)
2726 {
2727 	if (p->is_write)
2728 		return write_to_read_only(vcpu, p, r);
2729 
2730 	p->regval = kvm_get_guest_vtr_el2();
2731 
2732 	return true;
2733 }
2734 
access_gic_misr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2735 static bool access_gic_misr(struct kvm_vcpu *vcpu,
2736 			    struct sys_reg_params *p,
2737 			    const struct sys_reg_desc *r)
2738 {
2739 	if (p->is_write)
2740 		return write_to_read_only(vcpu, p, r);
2741 
2742 	p->regval = vgic_v3_get_misr(vcpu);
2743 
2744 	return true;
2745 }
2746 
access_gic_eisr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2747 static bool access_gic_eisr(struct kvm_vcpu *vcpu,
2748 			    struct sys_reg_params *p,
2749 			    const struct sys_reg_desc *r)
2750 {
2751 	if (p->is_write)
2752 		return write_to_read_only(vcpu, p, r);
2753 
2754 	p->regval = vgic_v3_get_eisr(vcpu);
2755 
2756 	return true;
2757 }
2758 
access_gic_elrsr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2759 static bool access_gic_elrsr(struct kvm_vcpu *vcpu,
2760 			     struct sys_reg_params *p,
2761 			     const struct sys_reg_desc *r)
2762 {
2763 	if (p->is_write)
2764 		return write_to_read_only(vcpu, p, r);
2765 
2766 	p->regval = vgic_v3_get_elrsr(vcpu);
2767 
2768 	return true;
2769 }
2770 
s1poe_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2771 static unsigned int s1poe_visibility(const struct kvm_vcpu *vcpu,
2772 				     const struct sys_reg_desc *rd)
2773 {
2774 	if (kvm_has_s1poe(vcpu->kvm))
2775 		return 0;
2776 
2777 	return REG_HIDDEN;
2778 }
2779 
s1poe_el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2780 static unsigned int s1poe_el2_visibility(const struct kvm_vcpu *vcpu,
2781 					 const struct sys_reg_desc *rd)
2782 {
2783 	return __el2_visibility(vcpu, rd, s1poe_visibility);
2784 }
2785 
tcr2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2786 static unsigned int tcr2_visibility(const struct kvm_vcpu *vcpu,
2787 				    const struct sys_reg_desc *rd)
2788 {
2789 	if (kvm_has_tcr2(vcpu->kvm))
2790 		return 0;
2791 
2792 	return REG_HIDDEN;
2793 }
2794 
tcr2_el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2795 static unsigned int tcr2_el2_visibility(const struct kvm_vcpu *vcpu,
2796 				    const struct sys_reg_desc *rd)
2797 {
2798 	return __el2_visibility(vcpu, rd, tcr2_visibility);
2799 }
2800 
fgt2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2801 static unsigned int fgt2_visibility(const struct kvm_vcpu *vcpu,
2802 				    const struct sys_reg_desc *rd)
2803 {
2804 	if (el2_visibility(vcpu, rd) == 0 &&
2805 	    kvm_has_feat(vcpu->kvm, ID_AA64MMFR0_EL1, FGT, FGT2))
2806 		return 0;
2807 
2808 	return REG_HIDDEN;
2809 }
2810 
fgt_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2811 static unsigned int fgt_visibility(const struct kvm_vcpu *vcpu,
2812 				   const struct sys_reg_desc *rd)
2813 {
2814 	if (el2_visibility(vcpu, rd) == 0 &&
2815 	    kvm_has_feat(vcpu->kvm, ID_AA64MMFR0_EL1, FGT, IMP))
2816 		return 0;
2817 
2818 	return REG_HIDDEN;
2819 }
2820 
s1pie_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2821 static unsigned int s1pie_visibility(const struct kvm_vcpu *vcpu,
2822 				     const struct sys_reg_desc *rd)
2823 {
2824 	if (kvm_has_s1pie(vcpu->kvm))
2825 		return 0;
2826 
2827 	return REG_HIDDEN;
2828 }
2829 
s1pie_el2_visibility(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd)2830 static unsigned int s1pie_el2_visibility(const struct kvm_vcpu *vcpu,
2831 					 const struct sys_reg_desc *rd)
2832 {
2833 	return __el2_visibility(vcpu, rd, s1pie_visibility);
2834 }
2835 
access_mdcr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2836 static bool access_mdcr(struct kvm_vcpu *vcpu,
2837 			struct sys_reg_params *p,
2838 			const struct sys_reg_desc *r)
2839 {
2840 	u64 hpmn, val, old = __vcpu_sys_reg(vcpu, MDCR_EL2);
2841 
2842 	if (!p->is_write) {
2843 		p->regval = old;
2844 		return true;
2845 	}
2846 
2847 	val = p->regval;
2848 	hpmn = FIELD_GET(MDCR_EL2_HPMN, val);
2849 
2850 	/*
2851 	 * If HPMN is out of bounds, limit it to what we actually
2852 	 * support. This matches the UNKNOWN definition of the field
2853 	 * in that case, and keeps the emulation simple. Sort of.
2854 	 */
2855 	if (hpmn > vcpu->kvm->arch.nr_pmu_counters) {
2856 		hpmn = vcpu->kvm->arch.nr_pmu_counters;
2857 		u64p_replace_bits(&val, hpmn, MDCR_EL2_HPMN);
2858 	}
2859 
2860 	__vcpu_assign_sys_reg(vcpu, MDCR_EL2, val);
2861 
2862 	/*
2863 	 * Request a reload of the PMU to enable/disable the counters
2864 	 * affected by HPME.
2865 	 */
2866 	if ((old ^ val) & MDCR_EL2_HPME)
2867 		kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
2868 
2869 	return true;
2870 }
2871 
access_ras(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2872 static bool access_ras(struct kvm_vcpu *vcpu,
2873 		       struct sys_reg_params *p,
2874 		       const struct sys_reg_desc *r)
2875 {
2876 	struct kvm *kvm = vcpu->kvm;
2877 
2878 	switch(reg_to_encoding(r)) {
2879 	case SYS_ERXPFGCDN_EL1:
2880 	case SYS_ERXPFGCTL_EL1:
2881 	case SYS_ERXPFGF_EL1:
2882 	case SYS_ERXMISC2_EL1:
2883 	case SYS_ERXMISC3_EL1:
2884 		if (!(kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, V1P1) ||
2885 		      (kvm_has_feat_enum(kvm, ID_AA64PFR0_EL1, RAS, IMP) &&
2886 		       kvm_has_feat(kvm, ID_AA64PFR1_EL1, RAS_frac, RASv1p1)))) {
2887 			kvm_inject_undefined(vcpu);
2888 			return false;
2889 		}
2890 		break;
2891 	default:
2892 		if (!kvm_has_feat(kvm, ID_AA64PFR0_EL1, RAS, IMP)) {
2893 			kvm_inject_undefined(vcpu);
2894 			return false;
2895 		}
2896 	}
2897 
2898 	return trap_raz_wi(vcpu, p, r);
2899 }
2900 
2901 /*
2902  * For historical (ahem ABI) reasons, KVM treated MIDR_EL1, REVIDR_EL1, and
2903  * AIDR_EL1 as "invariant" registers, meaning userspace cannot change them.
2904  * The values made visible to userspace were the register values of the boot
2905  * CPU.
2906  *
2907  * At the same time, reads from these registers at EL1 previously were not
2908  * trapped, allowing the guest to read the actual hardware value. On big-little
2909  * machines, this means the VM can see different values depending on where a
2910  * given vCPU got scheduled.
2911  *
2912  * These registers are now trapped as collateral damage from SME, and what
2913  * follows attempts to give a user / guest view consistent with the existing
2914  * ABI.
2915  */
access_imp_id_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)2916 static bool access_imp_id_reg(struct kvm_vcpu *vcpu,
2917 			      struct sys_reg_params *p,
2918 			      const struct sys_reg_desc *r)
2919 {
2920 	if (p->is_write)
2921 		return write_to_read_only(vcpu, p, r);
2922 
2923 	/*
2924 	 * Return the VM-scoped implementation ID register values if userspace
2925 	 * has made them writable.
2926 	 */
2927 	if (test_bit(KVM_ARCH_FLAG_WRITABLE_IMP_ID_REGS, &vcpu->kvm->arch.flags))
2928 		return access_id_reg(vcpu, p, r);
2929 
2930 	/*
2931 	 * Otherwise, fall back to the old behavior of returning the value of
2932 	 * the current CPU.
2933 	 */
2934 	switch (reg_to_encoding(r)) {
2935 	case SYS_REVIDR_EL1:
2936 		p->regval = read_sysreg(revidr_el1);
2937 		break;
2938 	case SYS_AIDR_EL1:
2939 		p->regval = read_sysreg(aidr_el1);
2940 		break;
2941 	default:
2942 		WARN_ON_ONCE(1);
2943 	}
2944 
2945 	return true;
2946 }
2947 
2948 static u64 __ro_after_init boot_cpu_midr_val;
2949 static u64 __ro_after_init boot_cpu_revidr_val;
2950 static u64 __ro_after_init boot_cpu_aidr_val;
2951 
init_imp_id_regs(void)2952 static void init_imp_id_regs(void)
2953 {
2954 	boot_cpu_midr_val = read_sysreg(midr_el1);
2955 	boot_cpu_revidr_val = read_sysreg(revidr_el1);
2956 	boot_cpu_aidr_val = read_sysreg(aidr_el1);
2957 }
2958 
reset_imp_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)2959 static u64 reset_imp_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
2960 {
2961 	switch (reg_to_encoding(r)) {
2962 	case SYS_MIDR_EL1:
2963 		return boot_cpu_midr_val;
2964 	case SYS_REVIDR_EL1:
2965 		return boot_cpu_revidr_val;
2966 	case SYS_AIDR_EL1:
2967 		return boot_cpu_aidr_val;
2968 	default:
2969 		KVM_BUG_ON(1, vcpu->kvm);
2970 		return 0;
2971 	}
2972 }
2973 
set_imp_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r,u64 val)2974 static int set_imp_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r,
2975 			  u64 val)
2976 {
2977 	struct kvm *kvm = vcpu->kvm;
2978 	u64 expected;
2979 
2980 	guard(mutex)(&kvm->arch.config_lock);
2981 
2982 	expected = read_id_reg(vcpu, r);
2983 	if (expected == val)
2984 		return 0;
2985 
2986 	if (!test_bit(KVM_ARCH_FLAG_WRITABLE_IMP_ID_REGS, &kvm->arch.flags))
2987 		return -EINVAL;
2988 
2989 	/*
2990 	 * Once the VM has started the ID registers are immutable. Reject the
2991 	 * write if userspace tries to change it.
2992 	 */
2993 	if (kvm_vm_has_ran_once(kvm))
2994 		return -EBUSY;
2995 
2996 	/*
2997 	 * Any value is allowed for the implementation ID registers so long as
2998 	 * it is within the writable mask.
2999 	 */
3000 	if ((val & r->val) != val)
3001 		return -EINVAL;
3002 
3003 	kvm_set_vm_id_reg(kvm, reg_to_encoding(r), val);
3004 	return 0;
3005 }
3006 
3007 #define IMPLEMENTATION_ID(reg, mask) {			\
3008 	SYS_DESC(SYS_##reg),				\
3009 	.access = access_imp_id_reg,			\
3010 	.get_user = get_id_reg,				\
3011 	.set_user = set_imp_id_reg,			\
3012 	.reset = reset_imp_id_reg,			\
3013 	.val = mask,					\
3014 	}
3015 
reset_mdcr(struct kvm_vcpu * vcpu,const struct sys_reg_desc * r)3016 static u64 reset_mdcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
3017 {
3018 	__vcpu_assign_sys_reg(vcpu, r->reg, vcpu->kvm->arch.nr_pmu_counters);
3019 	return vcpu->kvm->arch.nr_pmu_counters;
3020 }
3021 
3022 /*
3023  * Architected system registers.
3024  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
3025  *
3026  * Debug handling: We do trap most, if not all debug related system
3027  * registers. The implementation is good enough to ensure that a guest
3028  * can use these with minimal performance degradation. The drawback is
3029  * that we don't implement any of the external debug architecture.
3030  * This should be revisited if we ever encounter a more demanding
3031  * guest...
3032  */
3033 static const struct sys_reg_desc sys_reg_descs[] = {
3034 	DBG_BCR_BVR_WCR_WVR_EL1(0),
3035 	DBG_BCR_BVR_WCR_WVR_EL1(1),
3036 	{ SYS_DESC(SYS_MDCCINT_EL1), trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
3037 	{ SYS_DESC(SYS_MDSCR_EL1), trap_debug_regs, reset_val, MDSCR_EL1, 0 },
3038 	DBG_BCR_BVR_WCR_WVR_EL1(2),
3039 	DBG_BCR_BVR_WCR_WVR_EL1(3),
3040 	DBG_BCR_BVR_WCR_WVR_EL1(4),
3041 	DBG_BCR_BVR_WCR_WVR_EL1(5),
3042 	DBG_BCR_BVR_WCR_WVR_EL1(6),
3043 	DBG_BCR_BVR_WCR_WVR_EL1(7),
3044 	DBG_BCR_BVR_WCR_WVR_EL1(8),
3045 	DBG_BCR_BVR_WCR_WVR_EL1(9),
3046 	DBG_BCR_BVR_WCR_WVR_EL1(10),
3047 	DBG_BCR_BVR_WCR_WVR_EL1(11),
3048 	DBG_BCR_BVR_WCR_WVR_EL1(12),
3049 	DBG_BCR_BVR_WCR_WVR_EL1(13),
3050 	DBG_BCR_BVR_WCR_WVR_EL1(14),
3051 	DBG_BCR_BVR_WCR_WVR_EL1(15),
3052 
3053 	{ SYS_DESC(SYS_MDRAR_EL1), trap_raz_wi },
3054 	{ SYS_DESC(SYS_OSLAR_EL1), trap_oslar_el1 },
3055 	{ SYS_DESC(SYS_OSLSR_EL1), trap_oslsr_el1, reset_val, OSLSR_EL1,
3056 		OSLSR_EL1_OSLM_IMPLEMENTED, .set_user = set_oslsr_el1, },
3057 	{ SYS_DESC(SYS_OSDLR_EL1), trap_raz_wi },
3058 	{ SYS_DESC(SYS_DBGPRCR_EL1), trap_raz_wi },
3059 	{ SYS_DESC(SYS_DBGCLAIMSET_EL1), trap_raz_wi },
3060 	{ SYS_DESC(SYS_DBGCLAIMCLR_EL1), trap_raz_wi },
3061 	{ SYS_DESC(SYS_DBGAUTHSTATUS_EL1), trap_dbgauthstatus_el1 },
3062 
3063 	{ SYS_DESC(SYS_MDCCSR_EL0), trap_raz_wi },
3064 	{ SYS_DESC(SYS_DBGDTR_EL0), trap_raz_wi },
3065 	// DBGDTR[TR]X_EL0 share the same encoding
3066 	{ SYS_DESC(SYS_DBGDTRTX_EL0), trap_raz_wi },
3067 
3068 	{ SYS_DESC(SYS_DBGVCR32_EL2), undef_access, reset_val, DBGVCR32_EL2, 0 },
3069 
3070 	IMPLEMENTATION_ID(MIDR_EL1, GENMASK_ULL(31, 0)),
3071 	{ SYS_DESC(SYS_MPIDR_EL1), NULL, reset_mpidr, MPIDR_EL1 },
3072 	IMPLEMENTATION_ID(REVIDR_EL1, GENMASK_ULL(63, 0)),
3073 
3074 	/*
3075 	 * ID regs: all ID_SANITISED() entries here must have corresponding
3076 	 * entries in arm64_ftr_regs[].
3077 	 */
3078 
3079 	/* AArch64 mappings of the AArch32 ID registers */
3080 	/* CRm=1 */
3081 	AA32_ID_SANITISED(ID_PFR0_EL1),
3082 	AA32_ID_SANITISED(ID_PFR1_EL1),
3083 	{ SYS_DESC(SYS_ID_DFR0_EL1),
3084 	  .access = access_id_reg,
3085 	  .get_user = get_id_reg,
3086 	  .set_user = set_id_dfr0_el1,
3087 	  .visibility = aa32_id_visibility,
3088 	  .reset = read_sanitised_id_dfr0_el1,
3089 	  .val = ID_DFR0_EL1_PerfMon_MASK |
3090 		 ID_DFR0_EL1_CopDbg_MASK, },
3091 	ID_HIDDEN(ID_AFR0_EL1),
3092 	AA32_ID_SANITISED(ID_MMFR0_EL1),
3093 	AA32_ID_SANITISED(ID_MMFR1_EL1),
3094 	AA32_ID_SANITISED(ID_MMFR2_EL1),
3095 	AA32_ID_SANITISED(ID_MMFR3_EL1),
3096 
3097 	/* CRm=2 */
3098 	AA32_ID_SANITISED(ID_ISAR0_EL1),
3099 	AA32_ID_SANITISED(ID_ISAR1_EL1),
3100 	AA32_ID_SANITISED(ID_ISAR2_EL1),
3101 	AA32_ID_SANITISED(ID_ISAR3_EL1),
3102 	AA32_ID_SANITISED(ID_ISAR4_EL1),
3103 	AA32_ID_SANITISED(ID_ISAR5_EL1),
3104 	AA32_ID_SANITISED(ID_MMFR4_EL1),
3105 	AA32_ID_SANITISED(ID_ISAR6_EL1),
3106 
3107 	/* CRm=3 */
3108 	AA32_ID_SANITISED(MVFR0_EL1),
3109 	AA32_ID_SANITISED(MVFR1_EL1),
3110 	AA32_ID_SANITISED(MVFR2_EL1),
3111 	ID_UNALLOCATED(3,3),
3112 	AA32_ID_SANITISED(ID_PFR2_EL1),
3113 	ID_HIDDEN(ID_DFR1_EL1),
3114 	AA32_ID_SANITISED(ID_MMFR5_EL1),
3115 	ID_UNALLOCATED(3,7),
3116 
3117 	/* AArch64 ID registers */
3118 	/* CRm=4 */
3119 	ID_FILTERED(ID_AA64PFR0_EL1, id_aa64pfr0_el1,
3120 		    ~(ID_AA64PFR0_EL1_AMU |
3121 		      ID_AA64PFR0_EL1_MPAM |
3122 		      ID_AA64PFR0_EL1_SVE |
3123 		      ID_AA64PFR0_EL1_AdvSIMD |
3124 		      ID_AA64PFR0_EL1_FP)),
3125 	ID_FILTERED(ID_AA64PFR1_EL1, id_aa64pfr1_el1,
3126 				     ~(ID_AA64PFR1_EL1_PFAR |
3127 				       ID_AA64PFR1_EL1_MTEX |
3128 				       ID_AA64PFR1_EL1_THE |
3129 				       ID_AA64PFR1_EL1_GCS |
3130 				       ID_AA64PFR1_EL1_MTE_frac |
3131 				       ID_AA64PFR1_EL1_NMI |
3132 				       ID_AA64PFR1_EL1_RNDR_trap |
3133 				       ID_AA64PFR1_EL1_SME |
3134 				       ID_AA64PFR1_EL1_RES0 |
3135 				       ID_AA64PFR1_EL1_MPAM_frac |
3136 				       ID_AA64PFR1_EL1_MTE)),
3137 	ID_WRITABLE(ID_AA64PFR2_EL1,
3138 		    ID_AA64PFR2_EL1_FPMR |
3139 		    ID_AA64PFR2_EL1_MTEFAR |
3140 		    ID_AA64PFR2_EL1_MTESTOREONLY),
3141 	ID_UNALLOCATED(4,3),
3142 	ID_WRITABLE(ID_AA64ZFR0_EL1, ~ID_AA64ZFR0_EL1_RES0),
3143 	ID_HIDDEN(ID_AA64SMFR0_EL1),
3144 	ID_UNALLOCATED(4,6),
3145 	ID_WRITABLE(ID_AA64FPFR0_EL1, ~ID_AA64FPFR0_EL1_RES0),
3146 
3147 	/* CRm=5 */
3148 	/*
3149 	 * Prior to FEAT_Debugv8.9, the architecture defines context-aware
3150 	 * breakpoints (CTX_CMPs) as the highest numbered breakpoints (BRPs).
3151 	 * KVM does not trap + emulate the breakpoint registers, and as such
3152 	 * cannot support a layout that misaligns with the underlying hardware.
3153 	 * While it may be possible to describe a subset that aligns with
3154 	 * hardware, just prevent changes to BRPs and CTX_CMPs altogether for
3155 	 * simplicity.
3156 	 *
3157 	 * See DDI0487K.a, section D2.8.3 Breakpoint types and linking
3158 	 * of breakpoints for more details.
3159 	 */
3160 	ID_FILTERED(ID_AA64DFR0_EL1, id_aa64dfr0_el1,
3161 		    ID_AA64DFR0_EL1_DoubleLock_MASK |
3162 		    ID_AA64DFR0_EL1_WRPs_MASK |
3163 		    ID_AA64DFR0_EL1_PMUVer_MASK |
3164 		    ID_AA64DFR0_EL1_DebugVer_MASK),
3165 	ID_SANITISED(ID_AA64DFR1_EL1),
3166 	ID_UNALLOCATED(5,2),
3167 	ID_UNALLOCATED(5,3),
3168 	ID_HIDDEN(ID_AA64AFR0_EL1),
3169 	ID_HIDDEN(ID_AA64AFR1_EL1),
3170 	ID_UNALLOCATED(5,6),
3171 	ID_UNALLOCATED(5,7),
3172 
3173 	/* CRm=6 */
3174 	ID_WRITABLE(ID_AA64ISAR0_EL1, ~ID_AA64ISAR0_EL1_RES0),
3175 	ID_WRITABLE(ID_AA64ISAR1_EL1, ~(ID_AA64ISAR1_EL1_GPI |
3176 					ID_AA64ISAR1_EL1_GPA |
3177 					ID_AA64ISAR1_EL1_API |
3178 					ID_AA64ISAR1_EL1_APA)),
3179 	ID_WRITABLE(ID_AA64ISAR2_EL1, ~(ID_AA64ISAR2_EL1_RES0 |
3180 					ID_AA64ISAR2_EL1_APA3 |
3181 					ID_AA64ISAR2_EL1_GPA3)),
3182 	ID_WRITABLE(ID_AA64ISAR3_EL1, (ID_AA64ISAR3_EL1_FPRCVT |
3183 				       ID_AA64ISAR3_EL1_LSFE |
3184 				       ID_AA64ISAR3_EL1_FAMINMAX)),
3185 	ID_UNALLOCATED(6,4),
3186 	ID_UNALLOCATED(6,5),
3187 	ID_UNALLOCATED(6,6),
3188 	ID_UNALLOCATED(6,7),
3189 
3190 	/* CRm=7 */
3191 	ID_FILTERED(ID_AA64MMFR0_EL1, id_aa64mmfr0_el1,
3192 				      ~(ID_AA64MMFR0_EL1_RES0 |
3193 					ID_AA64MMFR0_EL1_ASIDBITS)),
3194 	ID_WRITABLE(ID_AA64MMFR1_EL1, ~(ID_AA64MMFR1_EL1_RES0 |
3195 					ID_AA64MMFR1_EL1_XNX |
3196 					ID_AA64MMFR1_EL1_VH |
3197 					ID_AA64MMFR1_EL1_VMIDBits)),
3198 	ID_FILTERED(ID_AA64MMFR2_EL1,
3199 		    id_aa64mmfr2_el1, ~(ID_AA64MMFR2_EL1_RES0 |
3200 					ID_AA64MMFR2_EL1_EVT |
3201 					ID_AA64MMFR2_EL1_FWB |
3202 					ID_AA64MMFR2_EL1_IDS |
3203 					ID_AA64MMFR2_EL1_NV |
3204 					ID_AA64MMFR2_EL1_CCIDX)),
3205 	ID_WRITABLE(ID_AA64MMFR3_EL1, (ID_AA64MMFR3_EL1_TCRX	|
3206 				       ID_AA64MMFR3_EL1_SCTLRX	|
3207 				       ID_AA64MMFR3_EL1_S1PIE   |
3208 				       ID_AA64MMFR3_EL1_S1POE)),
3209 	ID_WRITABLE(ID_AA64MMFR4_EL1, ID_AA64MMFR4_EL1_NV_frac),
3210 	ID_UNALLOCATED(7,5),
3211 	ID_UNALLOCATED(7,6),
3212 	ID_UNALLOCATED(7,7),
3213 
3214 	{ SYS_DESC(SYS_SCTLR_EL1), access_vm_reg, reset_val, SCTLR_EL1, 0x00C50078 },
3215 	{ SYS_DESC(SYS_ACTLR_EL1), access_actlr, reset_actlr, ACTLR_EL1 },
3216 	{ SYS_DESC(SYS_CPACR_EL1), NULL, reset_val, CPACR_EL1, 0 },
3217 	{ SYS_DESC(SYS_SCTLR2_EL1), access_vm_reg, reset_val, SCTLR2_EL1, 0,
3218 	  .visibility = sctlr2_visibility },
3219 
3220 	MTE_REG(RGSR_EL1),
3221 	MTE_REG(GCR_EL1),
3222 
3223 	{ SYS_DESC(SYS_ZCR_EL1), NULL, reset_val, ZCR_EL1, 0, .visibility = sve_visibility },
3224 	{ SYS_DESC(SYS_TRFCR_EL1), undef_access },
3225 	{ SYS_DESC(SYS_SMPRI_EL1), undef_access },
3226 	{ SYS_DESC(SYS_SMCR_EL1), undef_access },
3227 	{ SYS_DESC(SYS_TTBR0_EL1), access_vm_reg, reset_unknown, TTBR0_EL1 },
3228 	{ SYS_DESC(SYS_TTBR1_EL1), access_vm_reg, reset_unknown, TTBR1_EL1 },
3229 	{ SYS_DESC(SYS_TCR_EL1), access_vm_reg, reset_val, TCR_EL1, 0 },
3230 	{ SYS_DESC(SYS_TCR2_EL1), access_vm_reg, reset_val, TCR2_EL1, 0,
3231 	  .visibility = tcr2_visibility },
3232 
3233 	PTRAUTH_KEY(APIA),
3234 	PTRAUTH_KEY(APIB),
3235 	PTRAUTH_KEY(APDA),
3236 	PTRAUTH_KEY(APDB),
3237 	PTRAUTH_KEY(APGA),
3238 
3239 	{ SYS_DESC(SYS_SPSR_EL1), access_spsr},
3240 	{ SYS_DESC(SYS_ELR_EL1), access_elr},
3241 
3242 	{ SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
3243 
3244 	{ SYS_DESC(SYS_AFSR0_EL1), access_vm_reg, reset_unknown, AFSR0_EL1 },
3245 	{ SYS_DESC(SYS_AFSR1_EL1), access_vm_reg, reset_unknown, AFSR1_EL1 },
3246 	{ SYS_DESC(SYS_ESR_EL1), access_vm_reg, reset_unknown, ESR_EL1 },
3247 
3248 	{ SYS_DESC(SYS_ERRIDR_EL1), access_ras },
3249 	{ SYS_DESC(SYS_ERRSELR_EL1), access_ras },
3250 	{ SYS_DESC(SYS_ERXFR_EL1), access_ras },
3251 	{ SYS_DESC(SYS_ERXCTLR_EL1), access_ras },
3252 	{ SYS_DESC(SYS_ERXSTATUS_EL1), access_ras },
3253 	{ SYS_DESC(SYS_ERXADDR_EL1), access_ras },
3254 	{ SYS_DESC(SYS_ERXPFGF_EL1), access_ras },
3255 	{ SYS_DESC(SYS_ERXPFGCTL_EL1), access_ras },
3256 	{ SYS_DESC(SYS_ERXPFGCDN_EL1), access_ras },
3257 	{ SYS_DESC(SYS_ERXMISC0_EL1), access_ras },
3258 	{ SYS_DESC(SYS_ERXMISC1_EL1), access_ras },
3259 	{ SYS_DESC(SYS_ERXMISC2_EL1), access_ras },
3260 	{ SYS_DESC(SYS_ERXMISC3_EL1), access_ras },
3261 
3262 	MTE_REG(TFSR_EL1),
3263 	MTE_REG(TFSRE0_EL1),
3264 
3265 	{ SYS_DESC(SYS_FAR_EL1), access_vm_reg, reset_unknown, FAR_EL1 },
3266 	{ SYS_DESC(SYS_PAR_EL1), NULL, reset_unknown, PAR_EL1 },
3267 
3268 	{ SYS_DESC(SYS_PMSCR_EL1), undef_access },
3269 	{ SYS_DESC(SYS_PMSNEVFR_EL1), undef_access },
3270 	{ SYS_DESC(SYS_PMSICR_EL1), undef_access },
3271 	{ SYS_DESC(SYS_PMSIRR_EL1), undef_access },
3272 	{ SYS_DESC(SYS_PMSFCR_EL1), undef_access },
3273 	{ SYS_DESC(SYS_PMSEVFR_EL1), undef_access },
3274 	{ SYS_DESC(SYS_PMSLATFR_EL1), undef_access },
3275 	{ SYS_DESC(SYS_PMSIDR_EL1), undef_access },
3276 	{ SYS_DESC(SYS_PMBLIMITR_EL1), undef_access },
3277 	{ SYS_DESC(SYS_PMBPTR_EL1), undef_access },
3278 	{ SYS_DESC(SYS_PMBSR_EL1), undef_access },
3279 	{ SYS_DESC(SYS_PMSDSFR_EL1), undef_access },
3280 	/* PMBIDR_EL1 is not trapped */
3281 
3282 	{ PMU_SYS_REG(PMINTENSET_EL1),
3283 	  .access = access_pminten, .reg = PMINTENSET_EL1,
3284 	  .get_user = get_pmreg, .set_user = set_pmreg },
3285 	{ PMU_SYS_REG(PMINTENCLR_EL1),
3286 	  .access = access_pminten, .reg = PMINTENSET_EL1,
3287 	  .get_user = get_pmreg, .set_user = set_pmreg },
3288 	{ SYS_DESC(SYS_PMMIR_EL1), trap_raz_wi },
3289 
3290 	{ SYS_DESC(SYS_MAIR_EL1), access_vm_reg, reset_unknown, MAIR_EL1 },
3291 	{ SYS_DESC(SYS_PIRE0_EL1), NULL, reset_unknown, PIRE0_EL1,
3292 	  .visibility = s1pie_visibility },
3293 	{ SYS_DESC(SYS_PIR_EL1), NULL, reset_unknown, PIR_EL1,
3294 	  .visibility = s1pie_visibility },
3295 	{ SYS_DESC(SYS_POR_EL1), NULL, reset_unknown, POR_EL1,
3296 	  .visibility = s1poe_visibility },
3297 	{ SYS_DESC(SYS_AMAIR_EL1), access_vm_reg, reset_amair_el1, AMAIR_EL1 },
3298 
3299 	{ SYS_DESC(SYS_LORSA_EL1), trap_loregion },
3300 	{ SYS_DESC(SYS_LOREA_EL1), trap_loregion },
3301 	{ SYS_DESC(SYS_LORN_EL1), trap_loregion },
3302 	{ SYS_DESC(SYS_LORC_EL1), trap_loregion },
3303 	{ SYS_DESC(SYS_MPAMIDR_EL1), undef_access },
3304 	{ SYS_DESC(SYS_LORID_EL1), trap_loregion },
3305 
3306 	{ SYS_DESC(SYS_MPAM1_EL1), undef_access },
3307 	{ SYS_DESC(SYS_MPAM0_EL1), undef_access },
3308 	{ SYS_DESC(SYS_VBAR_EL1), access_rw, reset_val, VBAR_EL1, 0 },
3309 	{ SYS_DESC(SYS_DISR_EL1), NULL, reset_val, DISR_EL1, 0 },
3310 
3311 	{ SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
3312 	{ SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
3313 	{ SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
3314 	{ SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
3315 	{ SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
3316 	{ SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
3317 	{ SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
3318 	{ SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
3319 	{ SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
3320 	{ SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
3321 	{ SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
3322 	{ SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
3323 	{ SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
3324 	{ SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
3325 	{ SYS_DESC(SYS_ICC_SGI1R_EL1), access_gic_sgi },
3326 	{ SYS_DESC(SYS_ICC_ASGI1R_EL1), access_gic_sgi },
3327 	{ SYS_DESC(SYS_ICC_SGI0R_EL1), access_gic_sgi },
3328 	{ SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
3329 	{ SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
3330 	{ SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
3331 	{ SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
3332 	{ SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
3333 	{ SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
3334 	{ SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
3335 	{ SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
3336 
3337 	{ SYS_DESC(SYS_CONTEXTIDR_EL1), access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
3338 	{ SYS_DESC(SYS_TPIDR_EL1), NULL, reset_unknown, TPIDR_EL1 },
3339 
3340 	{ SYS_DESC(SYS_ACCDATA_EL1), undef_access },
3341 
3342 	{ SYS_DESC(SYS_SCXTNUM_EL1), undef_access },
3343 
3344 	{ SYS_DESC(SYS_CNTKCTL_EL1), NULL, reset_val, CNTKCTL_EL1, 0},
3345 
3346 	{ SYS_DESC(SYS_CCSIDR_EL1), access_ccsidr },
3347 	{ SYS_DESC(SYS_CLIDR_EL1), access_clidr, reset_clidr, CLIDR_EL1,
3348 	  .set_user = set_clidr, .val = ~CLIDR_EL1_RES0 },
3349 	{ SYS_DESC(SYS_CCSIDR2_EL1), undef_access },
3350 	{ SYS_DESC(SYS_SMIDR_EL1), undef_access },
3351 	IMPLEMENTATION_ID(AIDR_EL1, GENMASK_ULL(63, 0)),
3352 	{ SYS_DESC(SYS_CSSELR_EL1), access_csselr, reset_unknown, CSSELR_EL1 },
3353 	ID_FILTERED(CTR_EL0, ctr_el0,
3354 		    CTR_EL0_DIC_MASK |
3355 		    CTR_EL0_IDC_MASK |
3356 		    CTR_EL0_DminLine_MASK |
3357 		    CTR_EL0_L1Ip_MASK |
3358 		    CTR_EL0_IminLine_MASK),
3359 	{ SYS_DESC(SYS_SVCR), undef_access, reset_val, SVCR, 0, .visibility = sme_visibility  },
3360 	{ SYS_DESC(SYS_FPMR), undef_access, reset_val, FPMR, 0, .visibility = fp8_visibility },
3361 
3362 	{ PMU_SYS_REG(PMCR_EL0), .access = access_pmcr, .reset = reset_pmcr,
3363 	  .reg = PMCR_EL0, .get_user = get_pmcr, .set_user = set_pmcr },
3364 	{ PMU_SYS_REG(PMCNTENSET_EL0),
3365 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
3366 	  .get_user = get_pmreg, .set_user = set_pmreg },
3367 	{ PMU_SYS_REG(PMCNTENCLR_EL0),
3368 	  .access = access_pmcnten, .reg = PMCNTENSET_EL0,
3369 	  .get_user = get_pmreg, .set_user = set_pmreg },
3370 	{ PMU_SYS_REG(PMOVSCLR_EL0),
3371 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
3372 	  .get_user = get_pmreg, .set_user = set_pmreg },
3373 	/*
3374 	 * PM_SWINC_EL0 is exposed to userspace as RAZ/WI, as it was
3375 	 * previously (and pointlessly) advertised in the past...
3376 	 */
3377 	{ PMU_SYS_REG(PMSWINC_EL0),
3378 	  .get_user = get_raz_reg, .set_user = set_wi_reg,
3379 	  .access = access_pmswinc, .reset = NULL },
3380 	{ PMU_SYS_REG(PMSELR_EL0),
3381 	  .access = access_pmselr, .reset = reset_pmselr, .reg = PMSELR_EL0 },
3382 	{ PMU_SYS_REG(PMCEID0_EL0),
3383 	  .access = access_pmceid, .reset = NULL },
3384 	{ PMU_SYS_REG(PMCEID1_EL0),
3385 	  .access = access_pmceid, .reset = NULL },
3386 	{ PMU_SYS_REG(PMCCNTR_EL0),
3387 	  .access = access_pmu_evcntr, .reset = reset_unknown,
3388 	  .reg = PMCCNTR_EL0, .get_user = get_pmu_evcntr,
3389 	  .set_user = set_pmu_evcntr },
3390 	{ PMU_SYS_REG(PMXEVTYPER_EL0),
3391 	  .access = access_pmu_evtyper, .reset = NULL },
3392 	{ PMU_SYS_REG(PMXEVCNTR_EL0),
3393 	  .access = access_pmu_evcntr, .reset = NULL },
3394 	/*
3395 	 * PMUSERENR_EL0 resets as unknown in 64bit mode while it resets as zero
3396 	 * in 32bit mode. Here we choose to reset it as zero for consistency.
3397 	 */
3398 	{ PMU_SYS_REG(PMUSERENR_EL0), .access = access_pmuserenr,
3399 	  .reset = reset_val, .reg = PMUSERENR_EL0, .val = 0 },
3400 	{ PMU_SYS_REG(PMOVSSET_EL0),
3401 	  .access = access_pmovs, .reg = PMOVSSET_EL0,
3402 	  .get_user = get_pmreg, .set_user = set_pmreg },
3403 
3404 	{ SYS_DESC(SYS_POR_EL0), NULL, reset_unknown, POR_EL0,
3405 	  .visibility = s1poe_visibility },
3406 	{ SYS_DESC(SYS_TPIDR_EL0), NULL, reset_unknown, TPIDR_EL0 },
3407 	{ SYS_DESC(SYS_TPIDRRO_EL0), NULL, reset_unknown, TPIDRRO_EL0 },
3408 	{ SYS_DESC(SYS_TPIDR2_EL0), undef_access },
3409 
3410 	{ SYS_DESC(SYS_SCXTNUM_EL0), undef_access },
3411 
3412 	{ SYS_DESC(SYS_AMCR_EL0), undef_access },
3413 	{ SYS_DESC(SYS_AMCFGR_EL0), undef_access },
3414 	{ SYS_DESC(SYS_AMCGCR_EL0), undef_access },
3415 	{ SYS_DESC(SYS_AMUSERENR_EL0), undef_access },
3416 	{ SYS_DESC(SYS_AMCNTENCLR0_EL0), undef_access },
3417 	{ SYS_DESC(SYS_AMCNTENSET0_EL0), undef_access },
3418 	{ SYS_DESC(SYS_AMCNTENCLR1_EL0), undef_access },
3419 	{ SYS_DESC(SYS_AMCNTENSET1_EL0), undef_access },
3420 	AMU_AMEVCNTR0_EL0(0),
3421 	AMU_AMEVCNTR0_EL0(1),
3422 	AMU_AMEVCNTR0_EL0(2),
3423 	AMU_AMEVCNTR0_EL0(3),
3424 	AMU_AMEVCNTR0_EL0(4),
3425 	AMU_AMEVCNTR0_EL0(5),
3426 	AMU_AMEVCNTR0_EL0(6),
3427 	AMU_AMEVCNTR0_EL0(7),
3428 	AMU_AMEVCNTR0_EL0(8),
3429 	AMU_AMEVCNTR0_EL0(9),
3430 	AMU_AMEVCNTR0_EL0(10),
3431 	AMU_AMEVCNTR0_EL0(11),
3432 	AMU_AMEVCNTR0_EL0(12),
3433 	AMU_AMEVCNTR0_EL0(13),
3434 	AMU_AMEVCNTR0_EL0(14),
3435 	AMU_AMEVCNTR0_EL0(15),
3436 	AMU_AMEVTYPER0_EL0(0),
3437 	AMU_AMEVTYPER0_EL0(1),
3438 	AMU_AMEVTYPER0_EL0(2),
3439 	AMU_AMEVTYPER0_EL0(3),
3440 	AMU_AMEVTYPER0_EL0(4),
3441 	AMU_AMEVTYPER0_EL0(5),
3442 	AMU_AMEVTYPER0_EL0(6),
3443 	AMU_AMEVTYPER0_EL0(7),
3444 	AMU_AMEVTYPER0_EL0(8),
3445 	AMU_AMEVTYPER0_EL0(9),
3446 	AMU_AMEVTYPER0_EL0(10),
3447 	AMU_AMEVTYPER0_EL0(11),
3448 	AMU_AMEVTYPER0_EL0(12),
3449 	AMU_AMEVTYPER0_EL0(13),
3450 	AMU_AMEVTYPER0_EL0(14),
3451 	AMU_AMEVTYPER0_EL0(15),
3452 	AMU_AMEVCNTR1_EL0(0),
3453 	AMU_AMEVCNTR1_EL0(1),
3454 	AMU_AMEVCNTR1_EL0(2),
3455 	AMU_AMEVCNTR1_EL0(3),
3456 	AMU_AMEVCNTR1_EL0(4),
3457 	AMU_AMEVCNTR1_EL0(5),
3458 	AMU_AMEVCNTR1_EL0(6),
3459 	AMU_AMEVCNTR1_EL0(7),
3460 	AMU_AMEVCNTR1_EL0(8),
3461 	AMU_AMEVCNTR1_EL0(9),
3462 	AMU_AMEVCNTR1_EL0(10),
3463 	AMU_AMEVCNTR1_EL0(11),
3464 	AMU_AMEVCNTR1_EL0(12),
3465 	AMU_AMEVCNTR1_EL0(13),
3466 	AMU_AMEVCNTR1_EL0(14),
3467 	AMU_AMEVCNTR1_EL0(15),
3468 	AMU_AMEVTYPER1_EL0(0),
3469 	AMU_AMEVTYPER1_EL0(1),
3470 	AMU_AMEVTYPER1_EL0(2),
3471 	AMU_AMEVTYPER1_EL0(3),
3472 	AMU_AMEVTYPER1_EL0(4),
3473 	AMU_AMEVTYPER1_EL0(5),
3474 	AMU_AMEVTYPER1_EL0(6),
3475 	AMU_AMEVTYPER1_EL0(7),
3476 	AMU_AMEVTYPER1_EL0(8),
3477 	AMU_AMEVTYPER1_EL0(9),
3478 	AMU_AMEVTYPER1_EL0(10),
3479 	AMU_AMEVTYPER1_EL0(11),
3480 	AMU_AMEVTYPER1_EL0(12),
3481 	AMU_AMEVTYPER1_EL0(13),
3482 	AMU_AMEVTYPER1_EL0(14),
3483 	AMU_AMEVTYPER1_EL0(15),
3484 
3485 	{ SYS_DESC(SYS_CNTPCT_EL0), access_arch_timer },
3486 	{ SYS_DESC(SYS_CNTVCT_EL0), access_arch_timer },
3487 	{ SYS_DESC(SYS_CNTPCTSS_EL0), access_arch_timer },
3488 	{ SYS_DESC(SYS_CNTVCTSS_EL0), access_arch_timer },
3489 	{ SYS_DESC(SYS_CNTP_TVAL_EL0), access_arch_timer },
3490 	{ SYS_DESC(SYS_CNTP_CTL_EL0), access_arch_timer },
3491 	{ SYS_DESC(SYS_CNTP_CVAL_EL0), access_arch_timer },
3492 
3493 	{ SYS_DESC(SYS_CNTV_TVAL_EL0), access_arch_timer },
3494 	{ SYS_DESC(SYS_CNTV_CTL_EL0), access_arch_timer },
3495 	{ SYS_DESC(SYS_CNTV_CVAL_EL0), access_arch_timer },
3496 
3497 	/* PMEVCNTRn_EL0 */
3498 	PMU_PMEVCNTR_EL0(0),
3499 	PMU_PMEVCNTR_EL0(1),
3500 	PMU_PMEVCNTR_EL0(2),
3501 	PMU_PMEVCNTR_EL0(3),
3502 	PMU_PMEVCNTR_EL0(4),
3503 	PMU_PMEVCNTR_EL0(5),
3504 	PMU_PMEVCNTR_EL0(6),
3505 	PMU_PMEVCNTR_EL0(7),
3506 	PMU_PMEVCNTR_EL0(8),
3507 	PMU_PMEVCNTR_EL0(9),
3508 	PMU_PMEVCNTR_EL0(10),
3509 	PMU_PMEVCNTR_EL0(11),
3510 	PMU_PMEVCNTR_EL0(12),
3511 	PMU_PMEVCNTR_EL0(13),
3512 	PMU_PMEVCNTR_EL0(14),
3513 	PMU_PMEVCNTR_EL0(15),
3514 	PMU_PMEVCNTR_EL0(16),
3515 	PMU_PMEVCNTR_EL0(17),
3516 	PMU_PMEVCNTR_EL0(18),
3517 	PMU_PMEVCNTR_EL0(19),
3518 	PMU_PMEVCNTR_EL0(20),
3519 	PMU_PMEVCNTR_EL0(21),
3520 	PMU_PMEVCNTR_EL0(22),
3521 	PMU_PMEVCNTR_EL0(23),
3522 	PMU_PMEVCNTR_EL0(24),
3523 	PMU_PMEVCNTR_EL0(25),
3524 	PMU_PMEVCNTR_EL0(26),
3525 	PMU_PMEVCNTR_EL0(27),
3526 	PMU_PMEVCNTR_EL0(28),
3527 	PMU_PMEVCNTR_EL0(29),
3528 	PMU_PMEVCNTR_EL0(30),
3529 	/* PMEVTYPERn_EL0 */
3530 	PMU_PMEVTYPER_EL0(0),
3531 	PMU_PMEVTYPER_EL0(1),
3532 	PMU_PMEVTYPER_EL0(2),
3533 	PMU_PMEVTYPER_EL0(3),
3534 	PMU_PMEVTYPER_EL0(4),
3535 	PMU_PMEVTYPER_EL0(5),
3536 	PMU_PMEVTYPER_EL0(6),
3537 	PMU_PMEVTYPER_EL0(7),
3538 	PMU_PMEVTYPER_EL0(8),
3539 	PMU_PMEVTYPER_EL0(9),
3540 	PMU_PMEVTYPER_EL0(10),
3541 	PMU_PMEVTYPER_EL0(11),
3542 	PMU_PMEVTYPER_EL0(12),
3543 	PMU_PMEVTYPER_EL0(13),
3544 	PMU_PMEVTYPER_EL0(14),
3545 	PMU_PMEVTYPER_EL0(15),
3546 	PMU_PMEVTYPER_EL0(16),
3547 	PMU_PMEVTYPER_EL0(17),
3548 	PMU_PMEVTYPER_EL0(18),
3549 	PMU_PMEVTYPER_EL0(19),
3550 	PMU_PMEVTYPER_EL0(20),
3551 	PMU_PMEVTYPER_EL0(21),
3552 	PMU_PMEVTYPER_EL0(22),
3553 	PMU_PMEVTYPER_EL0(23),
3554 	PMU_PMEVTYPER_EL0(24),
3555 	PMU_PMEVTYPER_EL0(25),
3556 	PMU_PMEVTYPER_EL0(26),
3557 	PMU_PMEVTYPER_EL0(27),
3558 	PMU_PMEVTYPER_EL0(28),
3559 	PMU_PMEVTYPER_EL0(29),
3560 	PMU_PMEVTYPER_EL0(30),
3561 	/*
3562 	 * PMCCFILTR_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(PMCCFILTR_EL0), .access = access_pmu_evtyper,
3566 	  .reset = reset_val, .reg = PMCCFILTR_EL0, .val = 0 },
3567 
3568 	EL2_REG_VNCR(VPIDR_EL2, reset_unknown, 0),
3569 	EL2_REG_VNCR(VMPIDR_EL2, reset_unknown, 0),
3570 	EL2_REG(SCTLR_EL2, access_rw, reset_val, SCTLR_EL2_RES1),
3571 	EL2_REG(ACTLR_EL2, access_rw, reset_val, 0),
3572 	EL2_REG_FILTERED(SCTLR2_EL2, access_vm_reg, reset_val, 0,
3573 			 sctlr2_el2_visibility),
3574 	EL2_REG_VNCR(HCR_EL2, reset_hcr, 0),
3575 	EL2_REG(MDCR_EL2, access_mdcr, reset_mdcr, 0),
3576 	EL2_REG(CPTR_EL2, access_rw, reset_val, CPTR_NVHE_EL2_RES1),
3577 	EL2_REG_VNCR(HSTR_EL2, reset_val, 0),
3578 	EL2_REG_VNCR_FILT(HFGRTR_EL2, fgt_visibility),
3579 	EL2_REG_VNCR_FILT(HFGWTR_EL2, fgt_visibility),
3580 	EL2_REG_VNCR(HFGITR_EL2, reset_val, 0),
3581 	EL2_REG_VNCR(HACR_EL2, reset_val, 0),
3582 
3583 	EL2_REG_FILTERED(ZCR_EL2, access_zcr_el2, reset_val, 0,
3584 			 sve_el2_visibility),
3585 
3586 	EL2_REG_VNCR(HCRX_EL2, reset_val, 0),
3587 
3588 	EL2_REG(TTBR0_EL2, access_rw, reset_val, 0),
3589 	EL2_REG(TTBR1_EL2, access_rw, reset_val, 0),
3590 	EL2_REG(TCR_EL2, access_rw, reset_val, TCR_EL2_RES1),
3591 	EL2_REG_FILTERED(TCR2_EL2, access_rw, reset_val, TCR2_EL2_RES1,
3592 			 tcr2_el2_visibility),
3593 	EL2_REG_VNCR(VTTBR_EL2, reset_val, 0),
3594 	EL2_REG_VNCR(VTCR_EL2, reset_val, 0),
3595 	EL2_REG_FILTERED(VNCR_EL2, bad_vncr_trap, reset_val, 0,
3596 			 vncr_el2_visibility),
3597 
3598 	{ SYS_DESC(SYS_DACR32_EL2), undef_access, reset_unknown, DACR32_EL2 },
3599 	EL2_REG_VNCR_FILT(HDFGRTR2_EL2, fgt2_visibility),
3600 	EL2_REG_VNCR_FILT(HDFGWTR2_EL2, fgt2_visibility),
3601 	EL2_REG_VNCR_FILT(HFGRTR2_EL2, fgt2_visibility),
3602 	EL2_REG_VNCR_FILT(HFGWTR2_EL2, fgt2_visibility),
3603 	EL2_REG_VNCR_FILT(HDFGRTR_EL2, fgt_visibility),
3604 	EL2_REG_VNCR_FILT(HDFGWTR_EL2, fgt_visibility),
3605 	EL2_REG_VNCR_FILT(HAFGRTR_EL2, fgt_visibility),
3606 	EL2_REG_VNCR_FILT(HFGITR2_EL2, fgt2_visibility),
3607 	EL2_REG_REDIR(SPSR_EL2, reset_val, 0),
3608 	EL2_REG_REDIR(ELR_EL2, reset_val, 0),
3609 	{ SYS_DESC(SYS_SP_EL1), access_sp_el1},
3610 
3611 	/* AArch32 SPSR_* are RES0 if trapped from a NV guest */
3612 	{ SYS_DESC(SYS_SPSR_irq), .access = trap_raz_wi },
3613 	{ SYS_DESC(SYS_SPSR_abt), .access = trap_raz_wi },
3614 	{ SYS_DESC(SYS_SPSR_und), .access = trap_raz_wi },
3615 	{ SYS_DESC(SYS_SPSR_fiq), .access = trap_raz_wi },
3616 
3617 	{ SYS_DESC(SYS_IFSR32_EL2), undef_access, reset_unknown, IFSR32_EL2 },
3618 	EL2_REG(AFSR0_EL2, access_rw, reset_val, 0),
3619 	EL2_REG(AFSR1_EL2, access_rw, reset_val, 0),
3620 	EL2_REG_REDIR(ESR_EL2, reset_val, 0),
3621 	EL2_REG_VNCR(VSESR_EL2, reset_unknown, 0),
3622 	{ SYS_DESC(SYS_FPEXC32_EL2), undef_access, reset_val, FPEXC32_EL2, 0x700 },
3623 
3624 	EL2_REG_REDIR(FAR_EL2, reset_val, 0),
3625 	EL2_REG(HPFAR_EL2, access_rw, reset_val, 0),
3626 
3627 	EL2_REG(MAIR_EL2, access_rw, reset_val, 0),
3628 	EL2_REG_FILTERED(PIRE0_EL2, access_rw, reset_val, 0,
3629 			 s1pie_el2_visibility),
3630 	EL2_REG_FILTERED(PIR_EL2, access_rw, reset_val, 0,
3631 			 s1pie_el2_visibility),
3632 	EL2_REG_FILTERED(POR_EL2, access_rw, reset_val, 0,
3633 			 s1poe_el2_visibility),
3634 	EL2_REG(AMAIR_EL2, access_rw, reset_val, 0),
3635 	{ SYS_DESC(SYS_MPAMHCR_EL2), undef_access },
3636 	{ SYS_DESC(SYS_MPAMVPMV_EL2), undef_access },
3637 	{ SYS_DESC(SYS_MPAM2_EL2), undef_access },
3638 	{ SYS_DESC(SYS_MPAMVPM0_EL2), undef_access },
3639 	{ SYS_DESC(SYS_MPAMVPM1_EL2), undef_access },
3640 	{ SYS_DESC(SYS_MPAMVPM2_EL2), undef_access },
3641 	{ SYS_DESC(SYS_MPAMVPM3_EL2), undef_access },
3642 	{ SYS_DESC(SYS_MPAMVPM4_EL2), undef_access },
3643 	{ SYS_DESC(SYS_MPAMVPM5_EL2), undef_access },
3644 	{ SYS_DESC(SYS_MPAMVPM6_EL2), undef_access },
3645 	{ SYS_DESC(SYS_MPAMVPM7_EL2), undef_access },
3646 
3647 	EL2_REG(VBAR_EL2, access_rw, reset_val, 0),
3648 	{ SYS_DESC(SYS_RVBAR_EL2), undef_access },
3649 	{ SYS_DESC(SYS_RMR_EL2), undef_access },
3650 	EL2_REG_VNCR(VDISR_EL2, reset_unknown, 0),
3651 
3652 	EL2_REG_VNCR_GICv3(ICH_AP0R0_EL2),
3653 	EL2_REG_VNCR_GICv3(ICH_AP0R1_EL2),
3654 	EL2_REG_VNCR_GICv3(ICH_AP0R2_EL2),
3655 	EL2_REG_VNCR_GICv3(ICH_AP0R3_EL2),
3656 	EL2_REG_VNCR_GICv3(ICH_AP1R0_EL2),
3657 	EL2_REG_VNCR_GICv3(ICH_AP1R1_EL2),
3658 	EL2_REG_VNCR_GICv3(ICH_AP1R2_EL2),
3659 	EL2_REG_VNCR_GICv3(ICH_AP1R3_EL2),
3660 
3661 	{ SYS_DESC(SYS_ICC_SRE_EL2), access_gic_sre },
3662 
3663 	EL2_REG_VNCR_GICv3(ICH_HCR_EL2),
3664 	{ SYS_DESC(SYS_ICH_VTR_EL2), access_gic_vtr },
3665 	{ SYS_DESC(SYS_ICH_MISR_EL2), access_gic_misr },
3666 	{ SYS_DESC(SYS_ICH_EISR_EL2), access_gic_eisr },
3667 	{ SYS_DESC(SYS_ICH_ELRSR_EL2), access_gic_elrsr },
3668 	EL2_REG_VNCR_GICv3(ICH_VMCR_EL2),
3669 
3670 	EL2_REG_VNCR_GICv3(ICH_LR0_EL2),
3671 	EL2_REG_VNCR_GICv3(ICH_LR1_EL2),
3672 	EL2_REG_VNCR_GICv3(ICH_LR2_EL2),
3673 	EL2_REG_VNCR_GICv3(ICH_LR3_EL2),
3674 	EL2_REG_VNCR_GICv3(ICH_LR4_EL2),
3675 	EL2_REG_VNCR_GICv3(ICH_LR5_EL2),
3676 	EL2_REG_VNCR_GICv3(ICH_LR6_EL2),
3677 	EL2_REG_VNCR_GICv3(ICH_LR7_EL2),
3678 	EL2_REG_VNCR_GICv3(ICH_LR8_EL2),
3679 	EL2_REG_VNCR_GICv3(ICH_LR9_EL2),
3680 	EL2_REG_VNCR_GICv3(ICH_LR10_EL2),
3681 	EL2_REG_VNCR_GICv3(ICH_LR11_EL2),
3682 	EL2_REG_VNCR_GICv3(ICH_LR12_EL2),
3683 	EL2_REG_VNCR_GICv3(ICH_LR13_EL2),
3684 	EL2_REG_VNCR_GICv3(ICH_LR14_EL2),
3685 	EL2_REG_VNCR_GICv3(ICH_LR15_EL2),
3686 
3687 	EL2_REG(CONTEXTIDR_EL2, access_rw, reset_val, 0),
3688 	EL2_REG(TPIDR_EL2, access_rw, reset_val, 0),
3689 
3690 	EL2_REG_VNCR(CNTVOFF_EL2, reset_val, 0),
3691 	EL2_REG(CNTHCTL_EL2, access_rw, reset_val, 0),
3692 	{ SYS_DESC(SYS_CNTHP_TVAL_EL2), access_arch_timer },
3693 	EL2_REG(CNTHP_CTL_EL2, access_arch_timer, reset_val, 0),
3694 	EL2_REG(CNTHP_CVAL_EL2, access_arch_timer, reset_val, 0),
3695 
3696 	{ SYS_DESC(SYS_CNTHV_TVAL_EL2), access_hv_timer },
3697 	EL2_REG(CNTHV_CTL_EL2, access_hv_timer, reset_val, 0),
3698 	EL2_REG(CNTHV_CVAL_EL2, access_hv_timer, reset_val, 0),
3699 
3700 	{ SYS_DESC(SYS_CNTKCTL_EL12), access_cntkctl_el12 },
3701 
3702 	{ SYS_DESC(SYS_CNTP_TVAL_EL02), access_arch_timer },
3703 	{ SYS_DESC(SYS_CNTP_CTL_EL02), access_arch_timer },
3704 	{ SYS_DESC(SYS_CNTP_CVAL_EL02), access_arch_timer },
3705 
3706 	{ SYS_DESC(SYS_CNTV_TVAL_EL02), access_arch_timer },
3707 	{ SYS_DESC(SYS_CNTV_CTL_EL02), access_arch_timer },
3708 	{ SYS_DESC(SYS_CNTV_CVAL_EL02), access_arch_timer },
3709 
3710 	EL2_REG(SP_EL2, NULL, reset_unknown, 0),
3711 };
3712 
handle_at_s1e01(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3713 static bool handle_at_s1e01(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3714 			    const struct sys_reg_desc *r)
3715 {
3716 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3717 
3718 	__kvm_at_s1e01(vcpu, op, p->regval);
3719 
3720 	return true;
3721 }
3722 
handle_at_s1e2(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3723 static bool handle_at_s1e2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3724 			   const struct sys_reg_desc *r)
3725 {
3726 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3727 
3728 	/* There is no FGT associated with AT S1E2A :-( */
3729 	if (op == OP_AT_S1E2A &&
3730 	    !kvm_has_feat(vcpu->kvm, ID_AA64ISAR2_EL1, ATS1A, IMP)) {
3731 		kvm_inject_undefined(vcpu);
3732 		return false;
3733 	}
3734 
3735 	__kvm_at_s1e2(vcpu, op, p->regval);
3736 
3737 	return true;
3738 }
3739 
handle_at_s12(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3740 static bool handle_at_s12(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3741 			  const struct sys_reg_desc *r)
3742 {
3743 	u32 op = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3744 
3745 	__kvm_at_s12(vcpu, op, p->regval);
3746 
3747 	return true;
3748 }
3749 
kvm_supported_tlbi_s12_op(struct kvm_vcpu * vpcu,u32 instr)3750 static bool kvm_supported_tlbi_s12_op(struct kvm_vcpu *vpcu, u32 instr)
3751 {
3752 	struct kvm *kvm = vpcu->kvm;
3753 	u8 CRm = sys_reg_CRm(instr);
3754 
3755 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3756 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3757 		return false;
3758 
3759 	if (CRm == TLBI_CRm_nROS &&
3760 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3761 		return false;
3762 
3763 	return true;
3764 }
3765 
handle_alle1is(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3766 static bool handle_alle1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3767 			   const struct sys_reg_desc *r)
3768 {
3769 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3770 
3771 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3772 		return undef_access(vcpu, p, r);
3773 
3774 	write_lock(&vcpu->kvm->mmu_lock);
3775 
3776 	/*
3777 	 * Drop all shadow S2s, resulting in S1/S2 TLBIs for each of the
3778 	 * corresponding VMIDs.
3779 	 */
3780 	kvm_nested_s2_unmap(vcpu->kvm, true);
3781 
3782 	write_unlock(&vcpu->kvm->mmu_lock);
3783 
3784 	return true;
3785 }
3786 
kvm_supported_tlbi_ipas2_op(struct kvm_vcpu * vpcu,u32 instr)3787 static bool kvm_supported_tlbi_ipas2_op(struct kvm_vcpu *vpcu, u32 instr)
3788 {
3789 	struct kvm *kvm = vpcu->kvm;
3790 	u8 CRm = sys_reg_CRm(instr);
3791 	u8 Op2 = sys_reg_Op2(instr);
3792 
3793 	if (sys_reg_CRn(instr) == TLBI_CRn_nXS &&
3794 	    !kvm_has_feat(kvm, ID_AA64ISAR1_EL1, XS, IMP))
3795 		return false;
3796 
3797 	if (CRm == TLBI_CRm_IPAIS && (Op2 == 2 || Op2 == 6) &&
3798 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3799 		return false;
3800 
3801 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 0 || Op2 == 4) &&
3802 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
3803 		return false;
3804 
3805 	if (CRm == TLBI_CRm_IPAONS && (Op2 == 3 || Op2 == 7) &&
3806 	    !kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, RANGE))
3807 		return false;
3808 
3809 	return true;
3810 }
3811 
3812 /* Only defined here as this is an internal "abstraction" */
3813 union tlbi_info {
3814 	struct {
3815 		u64	start;
3816 		u64	size;
3817 	} range;
3818 
3819 	struct {
3820 		u64	addr;
3821 	} ipa;
3822 
3823 	struct {
3824 		u64	addr;
3825 		u32	encoding;
3826 	} va;
3827 };
3828 
s2_mmu_unmap_range(struct kvm_s2_mmu * mmu,const union tlbi_info * info)3829 static void s2_mmu_unmap_range(struct kvm_s2_mmu *mmu,
3830 			       const union tlbi_info *info)
3831 {
3832 	/*
3833 	 * The unmap operation is allowed to drop the MMU lock and block, which
3834 	 * means that @mmu could be used for a different context than the one
3835 	 * currently being invalidated.
3836 	 *
3837 	 * This behavior is still safe, as:
3838 	 *
3839 	 *  1) The vCPU(s) that recycled the MMU are responsible for invalidating
3840 	 *     the entire MMU before reusing it, which still honors the intent
3841 	 *     of a TLBI.
3842 	 *
3843 	 *  2) Until the guest TLBI instruction is 'retired' (i.e. increment PC
3844 	 *     and ERET to the guest), other vCPUs are allowed to use stale
3845 	 *     translations.
3846 	 *
3847 	 *  3) Accidentally unmapping an unrelated MMU context is nonfatal, and
3848 	 *     at worst may cause more aborts for shadow stage-2 fills.
3849 	 *
3850 	 * Dropping the MMU lock also implies that shadow stage-2 fills could
3851 	 * happen behind the back of the TLBI. This is still safe, though, as
3852 	 * the L1 needs to put its stage-2 in a consistent state before doing
3853 	 * the TLBI.
3854 	 */
3855 	kvm_stage2_unmap_range(mmu, info->range.start, info->range.size, true);
3856 }
3857 
handle_vmalls12e1is(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3858 static bool handle_vmalls12e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3859 				const struct sys_reg_desc *r)
3860 {
3861 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3862 	u64 limit, vttbr;
3863 
3864 	if (!kvm_supported_tlbi_s12_op(vcpu, sys_encoding))
3865 		return undef_access(vcpu, p, r);
3866 
3867 	vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3868 	limit = BIT_ULL(kvm_get_pa_bits(vcpu->kvm));
3869 
3870 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3871 				   &(union tlbi_info) {
3872 					   .range = {
3873 						   .start = 0,
3874 						   .size = limit,
3875 					   },
3876 				   },
3877 				   s2_mmu_unmap_range);
3878 
3879 	return true;
3880 }
3881 
handle_ripas2e1is(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3882 static bool handle_ripas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3883 			      const struct sys_reg_desc *r)
3884 {
3885 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3886 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3887 	u64 base, range;
3888 
3889 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3890 		return undef_access(vcpu, p, r);
3891 
3892 	/*
3893 	 * Because the shadow S2 structure doesn't necessarily reflect that
3894 	 * of the guest's S2 (different base granule size, for example), we
3895 	 * decide to ignore TTL and only use the described range.
3896 	 */
3897 	base = decode_range_tlbi(p->regval, &range, NULL);
3898 
3899 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3900 				   &(union tlbi_info) {
3901 					   .range = {
3902 						   .start = base,
3903 						   .size = range,
3904 					   },
3905 				   },
3906 				   s2_mmu_unmap_range);
3907 
3908 	return true;
3909 }
3910 
s2_mmu_unmap_ipa(struct kvm_s2_mmu * mmu,const union tlbi_info * info)3911 static void s2_mmu_unmap_ipa(struct kvm_s2_mmu *mmu,
3912 			     const union tlbi_info *info)
3913 {
3914 	unsigned long max_size;
3915 	u64 base_addr;
3916 
3917 	/*
3918 	 * We drop a number of things from the supplied value:
3919 	 *
3920 	 * - NS bit: we're non-secure only.
3921 	 *
3922 	 * - IPA[51:48]: We don't support 52bit IPA just yet...
3923 	 *
3924 	 * And of course, adjust the IPA to be on an actual address.
3925 	 */
3926 	base_addr = (info->ipa.addr & GENMASK_ULL(35, 0)) << 12;
3927 	max_size = compute_tlb_inval_range(mmu, info->ipa.addr);
3928 	base_addr &= ~(max_size - 1);
3929 
3930 	/*
3931 	 * See comment in s2_mmu_unmap_range() for why this is allowed to
3932 	 * reschedule.
3933 	 */
3934 	kvm_stage2_unmap_range(mmu, base_addr, max_size, true);
3935 }
3936 
handle_ipas2e1is(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3937 static bool handle_ipas2e1is(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3938 			     const struct sys_reg_desc *r)
3939 {
3940 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3941 	u64 vttbr = vcpu_read_sys_reg(vcpu, VTTBR_EL2);
3942 
3943 	if (!kvm_supported_tlbi_ipas2_op(vcpu, sys_encoding))
3944 		return undef_access(vcpu, p, r);
3945 
3946 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr),
3947 				   &(union tlbi_info) {
3948 					   .ipa = {
3949 						   .addr = p->regval,
3950 					   },
3951 				   },
3952 				   s2_mmu_unmap_ipa);
3953 
3954 	return true;
3955 }
3956 
s2_mmu_tlbi_s1e1(struct kvm_s2_mmu * mmu,const union tlbi_info * info)3957 static void s2_mmu_tlbi_s1e1(struct kvm_s2_mmu *mmu,
3958 			     const union tlbi_info *info)
3959 {
3960 	WARN_ON(__kvm_tlbi_s1e2(mmu, info->va.addr, info->va.encoding));
3961 }
3962 
handle_tlbi_el2(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3963 static bool handle_tlbi_el2(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3964 			    const struct sys_reg_desc *r)
3965 {
3966 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3967 
3968 	if (!kvm_supported_tlbi_s1e2_op(vcpu, sys_encoding))
3969 		return undef_access(vcpu, p, r);
3970 
3971 	kvm_handle_s1e2_tlbi(vcpu, sys_encoding, p->regval);
3972 	return true;
3973 }
3974 
handle_tlbi_el1(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)3975 static bool handle_tlbi_el1(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
3976 			    const struct sys_reg_desc *r)
3977 {
3978 	u32 sys_encoding = sys_insn(p->Op0, p->Op1, p->CRn, p->CRm, p->Op2);
3979 
3980 	/*
3981 	 * If we're here, this is because we've trapped on a EL1 TLBI
3982 	 * instruction that affects the EL1 translation regime while
3983 	 * we're running in a context that doesn't allow us to let the
3984 	 * HW do its thing (aka vEL2):
3985 	 *
3986 	 * - HCR_EL2.E2H == 0 : a non-VHE guest
3987 	 * - HCR_EL2.{E2H,TGE} == { 1, 0 } : a VHE guest in guest mode
3988 	 *
3989 	 * Another possibility is that we are invalidating the EL2 context
3990 	 * using EL1 instructions, but that we landed here because we need
3991 	 * additional invalidation for structures that are not held in the
3992 	 * CPU TLBs (such as the VNCR pseudo-TLB and its EL2 mapping). In
3993 	 * that case, we are guaranteed that HCR_EL2.{E2H,TGE} == { 1, 1 }
3994 	 * as we don't allow an NV-capable L1 in a nVHE configuration.
3995 	 *
3996 	 * We don't expect these helpers to ever be called when running
3997 	 * in a vEL1 context.
3998 	 */
3999 
4000 	WARN_ON(!vcpu_is_el2(vcpu));
4001 
4002 	if (!kvm_supported_tlbi_s1e1_op(vcpu, sys_encoding))
4003 		return undef_access(vcpu, p, r);
4004 
4005 	if (vcpu_el2_e2h_is_set(vcpu) && vcpu_el2_tge_is_set(vcpu)) {
4006 		kvm_handle_s1e2_tlbi(vcpu, sys_encoding, p->regval);
4007 		return true;
4008 	}
4009 
4010 	kvm_s2_mmu_iterate_by_vmid(vcpu->kvm,
4011 				   get_vmid(__vcpu_sys_reg(vcpu, VTTBR_EL2)),
4012 				   &(union tlbi_info) {
4013 					   .va = {
4014 						   .addr = p->regval,
4015 						   .encoding = sys_encoding,
4016 					   },
4017 				   },
4018 				   s2_mmu_tlbi_s1e1);
4019 
4020 	return true;
4021 }
4022 
4023 #define SYS_INSN(insn, access_fn)					\
4024 	{								\
4025 		SYS_DESC(OP_##insn),					\
4026 		.access = (access_fn),					\
4027 	}
4028 
4029 static struct sys_reg_desc sys_insn_descs[] = {
4030 	{ SYS_DESC(SYS_DC_ISW), access_dcsw },
4031 	{ SYS_DESC(SYS_DC_IGSW), access_dcgsw },
4032 	{ SYS_DESC(SYS_DC_IGDSW), access_dcgsw },
4033 
4034 	SYS_INSN(AT_S1E1R, handle_at_s1e01),
4035 	SYS_INSN(AT_S1E1W, handle_at_s1e01),
4036 	SYS_INSN(AT_S1E0R, handle_at_s1e01),
4037 	SYS_INSN(AT_S1E0W, handle_at_s1e01),
4038 	SYS_INSN(AT_S1E1RP, handle_at_s1e01),
4039 	SYS_INSN(AT_S1E1WP, handle_at_s1e01),
4040 
4041 	{ SYS_DESC(SYS_DC_CSW), access_dcsw },
4042 	{ SYS_DESC(SYS_DC_CGSW), access_dcgsw },
4043 	{ SYS_DESC(SYS_DC_CGDSW), access_dcgsw },
4044 	{ SYS_DESC(SYS_DC_CISW), access_dcsw },
4045 	{ SYS_DESC(SYS_DC_CIGSW), access_dcgsw },
4046 	{ SYS_DESC(SYS_DC_CIGDSW), access_dcgsw },
4047 
4048 	SYS_INSN(TLBI_VMALLE1OS, handle_tlbi_el1),
4049 	SYS_INSN(TLBI_VAE1OS, handle_tlbi_el1),
4050 	SYS_INSN(TLBI_ASIDE1OS, handle_tlbi_el1),
4051 	SYS_INSN(TLBI_VAAE1OS, handle_tlbi_el1),
4052 	SYS_INSN(TLBI_VALE1OS, handle_tlbi_el1),
4053 	SYS_INSN(TLBI_VAALE1OS, handle_tlbi_el1),
4054 
4055 	SYS_INSN(TLBI_RVAE1IS, handle_tlbi_el1),
4056 	SYS_INSN(TLBI_RVAAE1IS, handle_tlbi_el1),
4057 	SYS_INSN(TLBI_RVALE1IS, handle_tlbi_el1),
4058 	SYS_INSN(TLBI_RVAALE1IS, handle_tlbi_el1),
4059 
4060 	SYS_INSN(TLBI_VMALLE1IS, handle_tlbi_el1),
4061 	SYS_INSN(TLBI_VAE1IS, handle_tlbi_el1),
4062 	SYS_INSN(TLBI_ASIDE1IS, handle_tlbi_el1),
4063 	SYS_INSN(TLBI_VAAE1IS, handle_tlbi_el1),
4064 	SYS_INSN(TLBI_VALE1IS, handle_tlbi_el1),
4065 	SYS_INSN(TLBI_VAALE1IS, handle_tlbi_el1),
4066 
4067 	SYS_INSN(TLBI_RVAE1OS, handle_tlbi_el1),
4068 	SYS_INSN(TLBI_RVAAE1OS, handle_tlbi_el1),
4069 	SYS_INSN(TLBI_RVALE1OS, handle_tlbi_el1),
4070 	SYS_INSN(TLBI_RVAALE1OS, handle_tlbi_el1),
4071 
4072 	SYS_INSN(TLBI_RVAE1, handle_tlbi_el1),
4073 	SYS_INSN(TLBI_RVAAE1, handle_tlbi_el1),
4074 	SYS_INSN(TLBI_RVALE1, handle_tlbi_el1),
4075 	SYS_INSN(TLBI_RVAALE1, handle_tlbi_el1),
4076 
4077 	SYS_INSN(TLBI_VMALLE1, handle_tlbi_el1),
4078 	SYS_INSN(TLBI_VAE1, handle_tlbi_el1),
4079 	SYS_INSN(TLBI_ASIDE1, handle_tlbi_el1),
4080 	SYS_INSN(TLBI_VAAE1, handle_tlbi_el1),
4081 	SYS_INSN(TLBI_VALE1, handle_tlbi_el1),
4082 	SYS_INSN(TLBI_VAALE1, handle_tlbi_el1),
4083 
4084 	SYS_INSN(TLBI_VMALLE1OSNXS, handle_tlbi_el1),
4085 	SYS_INSN(TLBI_VAE1OSNXS, handle_tlbi_el1),
4086 	SYS_INSN(TLBI_ASIDE1OSNXS, handle_tlbi_el1),
4087 	SYS_INSN(TLBI_VAAE1OSNXS, handle_tlbi_el1),
4088 	SYS_INSN(TLBI_VALE1OSNXS, handle_tlbi_el1),
4089 	SYS_INSN(TLBI_VAALE1OSNXS, handle_tlbi_el1),
4090 
4091 	SYS_INSN(TLBI_RVAE1ISNXS, handle_tlbi_el1),
4092 	SYS_INSN(TLBI_RVAAE1ISNXS, handle_tlbi_el1),
4093 	SYS_INSN(TLBI_RVALE1ISNXS, handle_tlbi_el1),
4094 	SYS_INSN(TLBI_RVAALE1ISNXS, handle_tlbi_el1),
4095 
4096 	SYS_INSN(TLBI_VMALLE1ISNXS, handle_tlbi_el1),
4097 	SYS_INSN(TLBI_VAE1ISNXS, handle_tlbi_el1),
4098 	SYS_INSN(TLBI_ASIDE1ISNXS, handle_tlbi_el1),
4099 	SYS_INSN(TLBI_VAAE1ISNXS, handle_tlbi_el1),
4100 	SYS_INSN(TLBI_VALE1ISNXS, handle_tlbi_el1),
4101 	SYS_INSN(TLBI_VAALE1ISNXS, handle_tlbi_el1),
4102 
4103 	SYS_INSN(TLBI_RVAE1OSNXS, handle_tlbi_el1),
4104 	SYS_INSN(TLBI_RVAAE1OSNXS, handle_tlbi_el1),
4105 	SYS_INSN(TLBI_RVALE1OSNXS, handle_tlbi_el1),
4106 	SYS_INSN(TLBI_RVAALE1OSNXS, handle_tlbi_el1),
4107 
4108 	SYS_INSN(TLBI_RVAE1NXS, handle_tlbi_el1),
4109 	SYS_INSN(TLBI_RVAAE1NXS, handle_tlbi_el1),
4110 	SYS_INSN(TLBI_RVALE1NXS, handle_tlbi_el1),
4111 	SYS_INSN(TLBI_RVAALE1NXS, handle_tlbi_el1),
4112 
4113 	SYS_INSN(TLBI_VMALLE1NXS, handle_tlbi_el1),
4114 	SYS_INSN(TLBI_VAE1NXS, handle_tlbi_el1),
4115 	SYS_INSN(TLBI_ASIDE1NXS, handle_tlbi_el1),
4116 	SYS_INSN(TLBI_VAAE1NXS, handle_tlbi_el1),
4117 	SYS_INSN(TLBI_VALE1NXS, handle_tlbi_el1),
4118 	SYS_INSN(TLBI_VAALE1NXS, handle_tlbi_el1),
4119 
4120 	SYS_INSN(AT_S1E2R, handle_at_s1e2),
4121 	SYS_INSN(AT_S1E2W, handle_at_s1e2),
4122 	SYS_INSN(AT_S12E1R, handle_at_s12),
4123 	SYS_INSN(AT_S12E1W, handle_at_s12),
4124 	SYS_INSN(AT_S12E0R, handle_at_s12),
4125 	SYS_INSN(AT_S12E0W, handle_at_s12),
4126 	SYS_INSN(AT_S1E2A, handle_at_s1e2),
4127 
4128 	SYS_INSN(TLBI_IPAS2E1IS, handle_ipas2e1is),
4129 	SYS_INSN(TLBI_RIPAS2E1IS, handle_ripas2e1is),
4130 	SYS_INSN(TLBI_IPAS2LE1IS, handle_ipas2e1is),
4131 	SYS_INSN(TLBI_RIPAS2LE1IS, handle_ripas2e1is),
4132 
4133 	SYS_INSN(TLBI_ALLE2OS, handle_tlbi_el2),
4134 	SYS_INSN(TLBI_VAE2OS, handle_tlbi_el2),
4135 	SYS_INSN(TLBI_ALLE1OS, handle_alle1is),
4136 	SYS_INSN(TLBI_VALE2OS, handle_tlbi_el2),
4137 	SYS_INSN(TLBI_VMALLS12E1OS, handle_vmalls12e1is),
4138 
4139 	SYS_INSN(TLBI_RVAE2IS, handle_tlbi_el2),
4140 	SYS_INSN(TLBI_RVALE2IS, handle_tlbi_el2),
4141 	SYS_INSN(TLBI_ALLE2IS, handle_tlbi_el2),
4142 	SYS_INSN(TLBI_VAE2IS, handle_tlbi_el2),
4143 
4144 	SYS_INSN(TLBI_ALLE1IS, handle_alle1is),
4145 
4146 	SYS_INSN(TLBI_VALE2IS, handle_tlbi_el2),
4147 
4148 	SYS_INSN(TLBI_VMALLS12E1IS, handle_vmalls12e1is),
4149 	SYS_INSN(TLBI_IPAS2E1OS, handle_ipas2e1is),
4150 	SYS_INSN(TLBI_IPAS2E1, handle_ipas2e1is),
4151 	SYS_INSN(TLBI_RIPAS2E1, handle_ripas2e1is),
4152 	SYS_INSN(TLBI_RIPAS2E1OS, handle_ripas2e1is),
4153 	SYS_INSN(TLBI_IPAS2LE1OS, handle_ipas2e1is),
4154 	SYS_INSN(TLBI_IPAS2LE1, handle_ipas2e1is),
4155 	SYS_INSN(TLBI_RIPAS2LE1, handle_ripas2e1is),
4156 	SYS_INSN(TLBI_RIPAS2LE1OS, handle_ripas2e1is),
4157 	SYS_INSN(TLBI_RVAE2OS, handle_tlbi_el2),
4158 	SYS_INSN(TLBI_RVALE2OS, handle_tlbi_el2),
4159 	SYS_INSN(TLBI_RVAE2, handle_tlbi_el2),
4160 	SYS_INSN(TLBI_RVALE2, handle_tlbi_el2),
4161 	SYS_INSN(TLBI_ALLE2, handle_tlbi_el2),
4162 	SYS_INSN(TLBI_VAE2, handle_tlbi_el2),
4163 
4164 	SYS_INSN(TLBI_ALLE1, handle_alle1is),
4165 
4166 	SYS_INSN(TLBI_VALE2, handle_tlbi_el2),
4167 
4168 	SYS_INSN(TLBI_VMALLS12E1, handle_vmalls12e1is),
4169 
4170 	SYS_INSN(TLBI_IPAS2E1ISNXS, handle_ipas2e1is),
4171 	SYS_INSN(TLBI_RIPAS2E1ISNXS, handle_ripas2e1is),
4172 	SYS_INSN(TLBI_IPAS2LE1ISNXS, handle_ipas2e1is),
4173 	SYS_INSN(TLBI_RIPAS2LE1ISNXS, handle_ripas2e1is),
4174 
4175 	SYS_INSN(TLBI_ALLE2OSNXS, handle_tlbi_el2),
4176 	SYS_INSN(TLBI_VAE2OSNXS, handle_tlbi_el2),
4177 	SYS_INSN(TLBI_ALLE1OSNXS, handle_alle1is),
4178 	SYS_INSN(TLBI_VALE2OSNXS, handle_tlbi_el2),
4179 	SYS_INSN(TLBI_VMALLS12E1OSNXS, handle_vmalls12e1is),
4180 
4181 	SYS_INSN(TLBI_RVAE2ISNXS, handle_tlbi_el2),
4182 	SYS_INSN(TLBI_RVALE2ISNXS, handle_tlbi_el2),
4183 	SYS_INSN(TLBI_ALLE2ISNXS, handle_tlbi_el2),
4184 	SYS_INSN(TLBI_VAE2ISNXS, handle_tlbi_el2),
4185 
4186 	SYS_INSN(TLBI_ALLE1ISNXS, handle_alle1is),
4187 	SYS_INSN(TLBI_VALE2ISNXS, handle_tlbi_el2),
4188 	SYS_INSN(TLBI_VMALLS12E1ISNXS, handle_vmalls12e1is),
4189 	SYS_INSN(TLBI_IPAS2E1OSNXS, handle_ipas2e1is),
4190 	SYS_INSN(TLBI_IPAS2E1NXS, handle_ipas2e1is),
4191 	SYS_INSN(TLBI_RIPAS2E1NXS, handle_ripas2e1is),
4192 	SYS_INSN(TLBI_RIPAS2E1OSNXS, handle_ripas2e1is),
4193 	SYS_INSN(TLBI_IPAS2LE1OSNXS, handle_ipas2e1is),
4194 	SYS_INSN(TLBI_IPAS2LE1NXS, handle_ipas2e1is),
4195 	SYS_INSN(TLBI_RIPAS2LE1NXS, handle_ripas2e1is),
4196 	SYS_INSN(TLBI_RIPAS2LE1OSNXS, handle_ripas2e1is),
4197 	SYS_INSN(TLBI_RVAE2OSNXS, handle_tlbi_el2),
4198 	SYS_INSN(TLBI_RVALE2OSNXS, handle_tlbi_el2),
4199 	SYS_INSN(TLBI_RVAE2NXS, handle_tlbi_el2),
4200 	SYS_INSN(TLBI_RVALE2NXS, handle_tlbi_el2),
4201 	SYS_INSN(TLBI_ALLE2NXS, handle_tlbi_el2),
4202 	SYS_INSN(TLBI_VAE2NXS, handle_tlbi_el2),
4203 	SYS_INSN(TLBI_ALLE1NXS, handle_alle1is),
4204 	SYS_INSN(TLBI_VALE2NXS, handle_tlbi_el2),
4205 	SYS_INSN(TLBI_VMALLS12E1NXS, handle_vmalls12e1is),
4206 };
4207 
trap_dbgdidr(struct kvm_vcpu * vcpu,struct sys_reg_params * p,const struct sys_reg_desc * r)4208 static bool trap_dbgdidr(struct kvm_vcpu *vcpu,
4209 			struct sys_reg_params *p,
4210 			const struct sys_reg_desc *r)
4211 {
4212 	if (p->is_write) {
4213 		return ignore_write(vcpu, p);
4214 	} else {
4215 		u64 dfr = kvm_read_vm_id_reg(vcpu->kvm, SYS_ID_AA64DFR0_EL1);
4216 		u32 el3 = kvm_has_feat(vcpu->kvm, ID_AA64PFR0_EL1, EL3, IMP);
4217 
4218 		p->regval = ((SYS_FIELD_GET(ID_AA64DFR0_EL1, WRPs, dfr) << 28) |
4219 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, BRPs, dfr) << 24) |
4220 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, CTX_CMPs, dfr) << 20) |
4221 			     (SYS_FIELD_GET(ID_AA64DFR0_EL1, DebugVer, dfr) << 16) |
4222 			     (1 << 15) | (el3 << 14) | (el3 << 12));
4223 		return true;
4224 	}
4225 }
4226 
4227 /*
4228  * AArch32 debug register mappings
4229  *
4230  * AArch32 DBGBVRn is mapped to DBGBVRn_EL1[31:0]
4231  * AArch32 DBGBXVRn is mapped to DBGBVRn_EL1[63:32]
4232  *
4233  * None of the other registers share their location, so treat them as
4234  * if they were 64bit.
4235  */
4236 #define DBG_BCR_BVR_WCR_WVR(n)							\
4237 	/* DBGBVRn */								\
4238 	{ AA32(LO), Op1( 0), CRn( 0), CRm((n)), Op2( 4),			\
4239 	  trap_dbg_wb_reg, NULL, n },						\
4240 	/* DBGBCRn */								\
4241 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_dbg_wb_reg, NULL, n },	\
4242 	/* DBGWVRn */								\
4243 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_dbg_wb_reg, NULL, n },	\
4244 	/* DBGWCRn */								\
4245 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_dbg_wb_reg, NULL, n }
4246 
4247 #define DBGBXVR(n)								\
4248 	{ AA32(HI), Op1( 0), CRn( 1), CRm((n)), Op2( 1),			\
4249 	  trap_dbg_wb_reg, NULL, n }
4250 
4251 /*
4252  * Trapped cp14 registers. We generally ignore most of the external
4253  * debug, on the principle that they don't really make sense to a
4254  * guest. Revisit this one day, would this principle change.
4255  */
4256 static const struct sys_reg_desc cp14_regs[] = {
4257 	/* DBGDIDR */
4258 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgdidr },
4259 	/* DBGDTRRXext */
4260 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
4261 
4262 	DBG_BCR_BVR_WCR_WVR(0),
4263 	/* DBGDSCRint */
4264 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
4265 	DBG_BCR_BVR_WCR_WVR(1),
4266 	/* DBGDCCINT */
4267 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug_regs, NULL, MDCCINT_EL1 },
4268 	/* DBGDSCRext */
4269 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug_regs, NULL, MDSCR_EL1 },
4270 	DBG_BCR_BVR_WCR_WVR(2),
4271 	/* DBGDTR[RT]Xint */
4272 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
4273 	/* DBGDTR[RT]Xext */
4274 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
4275 	DBG_BCR_BVR_WCR_WVR(3),
4276 	DBG_BCR_BVR_WCR_WVR(4),
4277 	DBG_BCR_BVR_WCR_WVR(5),
4278 	/* DBGWFAR */
4279 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
4280 	/* DBGOSECCR */
4281 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
4282 	DBG_BCR_BVR_WCR_WVR(6),
4283 	/* DBGVCR */
4284 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug_regs, NULL, DBGVCR32_EL2 },
4285 	DBG_BCR_BVR_WCR_WVR(7),
4286 	DBG_BCR_BVR_WCR_WVR(8),
4287 	DBG_BCR_BVR_WCR_WVR(9),
4288 	DBG_BCR_BVR_WCR_WVR(10),
4289 	DBG_BCR_BVR_WCR_WVR(11),
4290 	DBG_BCR_BVR_WCR_WVR(12),
4291 	DBG_BCR_BVR_WCR_WVR(13),
4292 	DBG_BCR_BVR_WCR_WVR(14),
4293 	DBG_BCR_BVR_WCR_WVR(15),
4294 
4295 	/* DBGDRAR (32bit) */
4296 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
4297 
4298 	DBGBXVR(0),
4299 	/* DBGOSLAR */
4300 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_oslar_el1 },
4301 	DBGBXVR(1),
4302 	/* DBGOSLSR */
4303 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1, NULL, OSLSR_EL1 },
4304 	DBGBXVR(2),
4305 	DBGBXVR(3),
4306 	/* DBGOSDLR */
4307 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
4308 	DBGBXVR(4),
4309 	/* DBGPRCR */
4310 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
4311 	DBGBXVR(5),
4312 	DBGBXVR(6),
4313 	DBGBXVR(7),
4314 	DBGBXVR(8),
4315 	DBGBXVR(9),
4316 	DBGBXVR(10),
4317 	DBGBXVR(11),
4318 	DBGBXVR(12),
4319 	DBGBXVR(13),
4320 	DBGBXVR(14),
4321 	DBGBXVR(15),
4322 
4323 	/* DBGDSAR (32bit) */
4324 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
4325 
4326 	/* DBGDEVID2 */
4327 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
4328 	/* DBGDEVID1 */
4329 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
4330 	/* DBGDEVID */
4331 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
4332 	/* DBGCLAIMSET */
4333 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
4334 	/* DBGCLAIMCLR */
4335 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
4336 	/* DBGAUTHSTATUS */
4337 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
4338 };
4339 
4340 /* Trapped cp14 64bit registers */
4341 static const struct sys_reg_desc cp14_64_regs[] = {
4342 	/* DBGDRAR (64bit) */
4343 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
4344 
4345 	/* DBGDSAR (64bit) */
4346 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
4347 };
4348 
4349 #define CP15_PMU_SYS_REG(_map, _Op1, _CRn, _CRm, _Op2)			\
4350 	AA32(_map),							\
4351 	Op1(_Op1), CRn(_CRn), CRm(_CRm), Op2(_Op2),			\
4352 	.visibility = pmu_visibility
4353 
4354 /* Macro to expand the PMEVCNTRn register */
4355 #define PMU_PMEVCNTR(n)							\
4356 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
4357 	  (0b1000 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
4358 	  .access = access_pmu_evcntr }
4359 
4360 /* Macro to expand the PMEVTYPERn register */
4361 #define PMU_PMEVTYPER(n)						\
4362 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0b1110,				\
4363 	  (0b1100 | (((n) >> 3) & 0x3)), ((n) & 0x7)),			\
4364 	  .access = access_pmu_evtyper }
4365 /*
4366  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
4367  * depending on the way they are accessed (as a 32bit or a 64bit
4368  * register).
4369  */
4370 static const struct sys_reg_desc cp15_regs[] = {
4371 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 1), access_ctr },
4372 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_vm_reg, NULL, SCTLR_EL1 },
4373 	/* ACTLR */
4374 	{ AA32(LO), Op1( 0), CRn( 1), CRm( 0), Op2( 1), access_actlr, NULL, ACTLR_EL1 },
4375 	/* ACTLR2 */
4376 	{ AA32(HI), Op1( 0), CRn( 1), CRm( 0), Op2( 3), access_actlr, NULL, ACTLR_EL1 },
4377 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
4378 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, TTBR1_EL1 },
4379 	/* TTBCR */
4380 	{ AA32(LO), Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, TCR_EL1 },
4381 	/* TTBCR2 */
4382 	{ AA32(HI), Op1( 0), CRn( 2), CRm( 0), Op2( 3), access_vm_reg, NULL, TCR_EL1 },
4383 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, DACR32_EL2 },
4384 	{ CP15_SYS_DESC(SYS_ICC_PMR_EL1), undef_access },
4385 	/* DFSR */
4386 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, ESR_EL1 },
4387 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, IFSR32_EL2 },
4388 	/* ADFSR */
4389 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, AFSR0_EL1 },
4390 	/* AIFSR */
4391 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, AFSR1_EL1 },
4392 	/* DFAR */
4393 	{ AA32(LO), Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, FAR_EL1 },
4394 	/* IFAR */
4395 	{ AA32(HI), Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, FAR_EL1 },
4396 
4397 	/*
4398 	 * DC{C,I,CI}SW operations:
4399 	 */
4400 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
4401 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
4402 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
4403 
4404 	/* PMU */
4405 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 0), .access = access_pmcr },
4406 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 1), .access = access_pmcnten },
4407 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 2), .access = access_pmcnten },
4408 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 3), .access = access_pmovs },
4409 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 4), .access = access_pmswinc },
4410 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 12, 5), .access = access_pmselr },
4411 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 6), .access = access_pmceid },
4412 	{ CP15_PMU_SYS_REG(LO,     0, 9, 12, 7), .access = access_pmceid },
4413 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 0), .access = access_pmu_evcntr },
4414 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 1), .access = access_pmu_evtyper },
4415 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 13, 2), .access = access_pmu_evcntr },
4416 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 0), .access = access_pmuserenr },
4417 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 1), .access = access_pminten },
4418 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 2), .access = access_pminten },
4419 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 3), .access = access_pmovs },
4420 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 4), .access = access_pmceid },
4421 	{ CP15_PMU_SYS_REG(HI,     0, 9, 14, 5), .access = access_pmceid },
4422 	/* PMMIR */
4423 	{ CP15_PMU_SYS_REG(DIRECT, 0, 9, 14, 6), .access = trap_raz_wi },
4424 
4425 	/* PRRR/MAIR0 */
4426 	{ AA32(LO), Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, MAIR_EL1 },
4427 	/* NMRR/MAIR1 */
4428 	{ AA32(HI), Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, MAIR_EL1 },
4429 	/* AMAIR0 */
4430 	{ AA32(LO), Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, AMAIR_EL1 },
4431 	/* AMAIR1 */
4432 	{ AA32(HI), Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, AMAIR_EL1 },
4433 
4434 	{ CP15_SYS_DESC(SYS_ICC_IAR0_EL1), undef_access },
4435 	{ CP15_SYS_DESC(SYS_ICC_EOIR0_EL1), undef_access },
4436 	{ CP15_SYS_DESC(SYS_ICC_HPPIR0_EL1), undef_access },
4437 	{ CP15_SYS_DESC(SYS_ICC_BPR0_EL1), undef_access },
4438 	{ CP15_SYS_DESC(SYS_ICC_AP0R0_EL1), undef_access },
4439 	{ CP15_SYS_DESC(SYS_ICC_AP0R1_EL1), undef_access },
4440 	{ CP15_SYS_DESC(SYS_ICC_AP0R2_EL1), undef_access },
4441 	{ CP15_SYS_DESC(SYS_ICC_AP0R3_EL1), undef_access },
4442 	{ CP15_SYS_DESC(SYS_ICC_AP1R0_EL1), undef_access },
4443 	{ CP15_SYS_DESC(SYS_ICC_AP1R1_EL1), undef_access },
4444 	{ CP15_SYS_DESC(SYS_ICC_AP1R2_EL1), undef_access },
4445 	{ CP15_SYS_DESC(SYS_ICC_AP1R3_EL1), undef_access },
4446 	{ CP15_SYS_DESC(SYS_ICC_DIR_EL1), undef_access },
4447 	{ CP15_SYS_DESC(SYS_ICC_RPR_EL1), undef_access },
4448 	{ CP15_SYS_DESC(SYS_ICC_IAR1_EL1), undef_access },
4449 	{ CP15_SYS_DESC(SYS_ICC_EOIR1_EL1), undef_access },
4450 	{ CP15_SYS_DESC(SYS_ICC_HPPIR1_EL1), undef_access },
4451 	{ CP15_SYS_DESC(SYS_ICC_BPR1_EL1), undef_access },
4452 	{ CP15_SYS_DESC(SYS_ICC_CTLR_EL1), undef_access },
4453 	{ CP15_SYS_DESC(SYS_ICC_SRE_EL1), access_gic_sre },
4454 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN0_EL1), undef_access },
4455 	{ CP15_SYS_DESC(SYS_ICC_IGRPEN1_EL1), undef_access },
4456 
4457 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, CONTEXTIDR_EL1 },
4458 
4459 	/* Arch Tmers */
4460 	{ SYS_DESC(SYS_AARCH32_CNTP_TVAL), access_arch_timer },
4461 	{ SYS_DESC(SYS_AARCH32_CNTP_CTL), access_arch_timer },
4462 
4463 	/* PMEVCNTRn */
4464 	PMU_PMEVCNTR(0),
4465 	PMU_PMEVCNTR(1),
4466 	PMU_PMEVCNTR(2),
4467 	PMU_PMEVCNTR(3),
4468 	PMU_PMEVCNTR(4),
4469 	PMU_PMEVCNTR(5),
4470 	PMU_PMEVCNTR(6),
4471 	PMU_PMEVCNTR(7),
4472 	PMU_PMEVCNTR(8),
4473 	PMU_PMEVCNTR(9),
4474 	PMU_PMEVCNTR(10),
4475 	PMU_PMEVCNTR(11),
4476 	PMU_PMEVCNTR(12),
4477 	PMU_PMEVCNTR(13),
4478 	PMU_PMEVCNTR(14),
4479 	PMU_PMEVCNTR(15),
4480 	PMU_PMEVCNTR(16),
4481 	PMU_PMEVCNTR(17),
4482 	PMU_PMEVCNTR(18),
4483 	PMU_PMEVCNTR(19),
4484 	PMU_PMEVCNTR(20),
4485 	PMU_PMEVCNTR(21),
4486 	PMU_PMEVCNTR(22),
4487 	PMU_PMEVCNTR(23),
4488 	PMU_PMEVCNTR(24),
4489 	PMU_PMEVCNTR(25),
4490 	PMU_PMEVCNTR(26),
4491 	PMU_PMEVCNTR(27),
4492 	PMU_PMEVCNTR(28),
4493 	PMU_PMEVCNTR(29),
4494 	PMU_PMEVCNTR(30),
4495 	/* PMEVTYPERn */
4496 	PMU_PMEVTYPER(0),
4497 	PMU_PMEVTYPER(1),
4498 	PMU_PMEVTYPER(2),
4499 	PMU_PMEVTYPER(3),
4500 	PMU_PMEVTYPER(4),
4501 	PMU_PMEVTYPER(5),
4502 	PMU_PMEVTYPER(6),
4503 	PMU_PMEVTYPER(7),
4504 	PMU_PMEVTYPER(8),
4505 	PMU_PMEVTYPER(9),
4506 	PMU_PMEVTYPER(10),
4507 	PMU_PMEVTYPER(11),
4508 	PMU_PMEVTYPER(12),
4509 	PMU_PMEVTYPER(13),
4510 	PMU_PMEVTYPER(14),
4511 	PMU_PMEVTYPER(15),
4512 	PMU_PMEVTYPER(16),
4513 	PMU_PMEVTYPER(17),
4514 	PMU_PMEVTYPER(18),
4515 	PMU_PMEVTYPER(19),
4516 	PMU_PMEVTYPER(20),
4517 	PMU_PMEVTYPER(21),
4518 	PMU_PMEVTYPER(22),
4519 	PMU_PMEVTYPER(23),
4520 	PMU_PMEVTYPER(24),
4521 	PMU_PMEVTYPER(25),
4522 	PMU_PMEVTYPER(26),
4523 	PMU_PMEVTYPER(27),
4524 	PMU_PMEVTYPER(28),
4525 	PMU_PMEVTYPER(29),
4526 	PMU_PMEVTYPER(30),
4527 	/* PMCCFILTR */
4528 	{ CP15_PMU_SYS_REG(DIRECT, 0, 14, 15, 7), .access = access_pmu_evtyper },
4529 
4530 	{ Op1(1), CRn( 0), CRm( 0), Op2(0), access_ccsidr },
4531 	{ Op1(1), CRn( 0), CRm( 0), Op2(1), access_clidr },
4532 
4533 	/* CCSIDR2 */
4534 	{ Op1(1), CRn( 0), CRm( 0),  Op2(2), undef_access },
4535 
4536 	{ Op1(2), CRn( 0), CRm( 0), Op2(0), access_csselr, NULL, CSSELR_EL1 },
4537 };
4538 
4539 static const struct sys_reg_desc cp15_64_regs[] = {
4540 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR0_EL1 },
4541 	{ CP15_PMU_SYS_REG(DIRECT, 0, 0, 9, 0), .access = access_pmu_evcntr },
4542 	{ Op1( 0), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI1R */
4543 	{ SYS_DESC(SYS_AARCH32_CNTPCT),	      access_arch_timer },
4544 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, TTBR1_EL1 },
4545 	{ Op1( 1), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_ASGI1R */
4546 	{ SYS_DESC(SYS_AARCH32_CNTVCT),	      access_arch_timer },
4547 	{ Op1( 2), CRn( 0), CRm(12), Op2( 0), access_gic_sgi }, /* ICC_SGI0R */
4548 	{ SYS_DESC(SYS_AARCH32_CNTP_CVAL),    access_arch_timer },
4549 	{ SYS_DESC(SYS_AARCH32_CNTPCTSS),     access_arch_timer },
4550 	{ SYS_DESC(SYS_AARCH32_CNTVCTSS),     access_arch_timer },
4551 };
4552 
check_sysreg_table(const struct sys_reg_desc * table,unsigned int n,bool reset_check)4553 static bool check_sysreg_table(const struct sys_reg_desc *table, unsigned int n,
4554 			       bool reset_check)
4555 {
4556 	unsigned int i;
4557 
4558 	for (i = 0; i < n; i++) {
4559 		if (reset_check && table[i].reg && !table[i].reset) {
4560 			kvm_err("sys_reg table %pS entry %d (%s) lacks reset\n",
4561 				&table[i], i, table[i].name);
4562 			return false;
4563 		}
4564 
4565 		if (i && cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
4566 			kvm_err("sys_reg table %pS entry %d (%s -> %s) out of order\n",
4567 				&table[i], i, table[i - 1].name, table[i].name);
4568 			return false;
4569 		}
4570 	}
4571 
4572 	return true;
4573 }
4574 
kvm_handle_cp14_load_store(struct kvm_vcpu * vcpu)4575 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu)
4576 {
4577 	kvm_inject_undefined(vcpu);
4578 	return 1;
4579 }
4580 
perform_access(struct kvm_vcpu * vcpu,struct sys_reg_params * params,const struct sys_reg_desc * r)4581 static void perform_access(struct kvm_vcpu *vcpu,
4582 			   struct sys_reg_params *params,
4583 			   const struct sys_reg_desc *r)
4584 {
4585 	trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
4586 
4587 	/* Check for regs disabled by runtime config */
4588 	if (sysreg_hidden(vcpu, r)) {
4589 		kvm_inject_undefined(vcpu);
4590 		return;
4591 	}
4592 
4593 	/*
4594 	 * Not having an accessor means that we have configured a trap
4595 	 * that we don't know how to handle. This certainly qualifies
4596 	 * as a gross bug that should be fixed right away.
4597 	 */
4598 	BUG_ON(!r->access);
4599 
4600 	/* Skip instruction if instructed so */
4601 	if (likely(r->access(vcpu, params, r)))
4602 		kvm_incr_pc(vcpu);
4603 }
4604 
4605 /*
4606  * emulate_cp --  tries to match a sys_reg access in a handling table, and
4607  *                call the corresponding trap handler.
4608  *
4609  * @params: pointer to the descriptor of the access
4610  * @table: array of trap descriptors
4611  * @num: size of the trap descriptor array
4612  *
4613  * Return true if the access has been handled, false if not.
4614  */
emulate_cp(struct kvm_vcpu * vcpu,struct sys_reg_params * params,const struct sys_reg_desc * table,size_t num)4615 static bool emulate_cp(struct kvm_vcpu *vcpu,
4616 		       struct sys_reg_params *params,
4617 		       const struct sys_reg_desc *table,
4618 		       size_t num)
4619 {
4620 	const struct sys_reg_desc *r;
4621 
4622 	if (!table)
4623 		return false;	/* Not handled */
4624 
4625 	r = find_reg(params, table, num);
4626 
4627 	if (r) {
4628 		perform_access(vcpu, params, r);
4629 		return true;
4630 	}
4631 
4632 	/* Not handled */
4633 	return false;
4634 }
4635 
unhandled_cp_access(struct kvm_vcpu * vcpu,struct sys_reg_params * params)4636 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
4637 				struct sys_reg_params *params)
4638 {
4639 	u8 esr_ec = kvm_vcpu_trap_get_class(vcpu);
4640 	int cp = -1;
4641 
4642 	switch (esr_ec) {
4643 	case ESR_ELx_EC_CP15_32:
4644 	case ESR_ELx_EC_CP15_64:
4645 		cp = 15;
4646 		break;
4647 	case ESR_ELx_EC_CP14_MR:
4648 	case ESR_ELx_EC_CP14_64:
4649 		cp = 14;
4650 		break;
4651 	default:
4652 		WARN_ON(1);
4653 	}
4654 
4655 	print_sys_reg_msg(params,
4656 			  "Unsupported guest CP%d access at: %08lx [%08lx]\n",
4657 			  cp, *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
4658 	kvm_inject_undefined(vcpu);
4659 }
4660 
4661 /**
4662  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP14/CP15 access
4663  * @vcpu: The VCPU pointer
4664  * @global: &struct sys_reg_desc
4665  * @nr_global: size of the @global array
4666  */
kvm_handle_cp_64(struct kvm_vcpu * vcpu,const struct sys_reg_desc * global,size_t nr_global)4667 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
4668 			    const struct sys_reg_desc *global,
4669 			    size_t nr_global)
4670 {
4671 	struct sys_reg_params params;
4672 	u64 esr = kvm_vcpu_get_esr(vcpu);
4673 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4674 	int Rt2 = (esr >> 10) & 0x1f;
4675 
4676 	params.CRm = (esr >> 1) & 0xf;
4677 	params.is_write = ((esr & 1) == 0);
4678 
4679 	params.Op0 = 0;
4680 	params.Op1 = (esr >> 16) & 0xf;
4681 	params.Op2 = 0;
4682 	params.CRn = 0;
4683 
4684 	/*
4685 	 * Make a 64-bit value out of Rt and Rt2. As we use the same trap
4686 	 * backends between AArch32 and AArch64, we get away with it.
4687 	 */
4688 	if (params.is_write) {
4689 		params.regval = vcpu_get_reg(vcpu, Rt) & 0xffffffff;
4690 		params.regval |= vcpu_get_reg(vcpu, Rt2) << 32;
4691 	}
4692 
4693 	/*
4694 	 * If the table contains a handler, handle the
4695 	 * potential register operation in the case of a read and return
4696 	 * with success.
4697 	 */
4698 	if (emulate_cp(vcpu, &params, global, nr_global)) {
4699 		/* Split up the value between registers for the read side */
4700 		if (!params.is_write) {
4701 			vcpu_set_reg(vcpu, Rt, lower_32_bits(params.regval));
4702 			vcpu_set_reg(vcpu, Rt2, upper_32_bits(params.regval));
4703 		}
4704 
4705 		return 1;
4706 	}
4707 
4708 	unhandled_cp_access(vcpu, &params);
4709 	return 1;
4710 }
4711 
4712 static bool emulate_sys_reg(struct kvm_vcpu *vcpu, struct sys_reg_params *params);
4713 
4714 /*
4715  * The CP10 ID registers are architecturally mapped to AArch64 feature
4716  * registers. Abuse that fact so we can rely on the AArch64 handler for accesses
4717  * from AArch32.
4718  */
kvm_esr_cp10_id_to_sys64(u64 esr,struct sys_reg_params * params)4719 static bool kvm_esr_cp10_id_to_sys64(u64 esr, struct sys_reg_params *params)
4720 {
4721 	u8 reg_id = (esr >> 10) & 0xf;
4722 	bool valid;
4723 
4724 	params->is_write = ((esr & 1) == 0);
4725 	params->Op0 = 3;
4726 	params->Op1 = 0;
4727 	params->CRn = 0;
4728 	params->CRm = 3;
4729 
4730 	/* CP10 ID registers are read-only */
4731 	valid = !params->is_write;
4732 
4733 	switch (reg_id) {
4734 	/* MVFR0 */
4735 	case 0b0111:
4736 		params->Op2 = 0;
4737 		break;
4738 	/* MVFR1 */
4739 	case 0b0110:
4740 		params->Op2 = 1;
4741 		break;
4742 	/* MVFR2 */
4743 	case 0b0101:
4744 		params->Op2 = 2;
4745 		break;
4746 	default:
4747 		valid = false;
4748 	}
4749 
4750 	if (valid)
4751 		return true;
4752 
4753 	kvm_pr_unimpl("Unhandled cp10 register %s: %u\n",
4754 		      str_write_read(params->is_write), reg_id);
4755 	return false;
4756 }
4757 
4758 /**
4759  * kvm_handle_cp10_id() - Handles a VMRS trap on guest access to a 'Media and
4760  *			  VFP Register' from AArch32.
4761  * @vcpu: The vCPU pointer
4762  *
4763  * MVFR{0-2} are architecturally mapped to the AArch64 MVFR{0-2}_EL1 registers.
4764  * Work out the correct AArch64 system register encoding and reroute to the
4765  * AArch64 system register emulation.
4766  */
kvm_handle_cp10_id(struct kvm_vcpu * vcpu)4767 int kvm_handle_cp10_id(struct kvm_vcpu *vcpu)
4768 {
4769 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4770 	u64 esr = kvm_vcpu_get_esr(vcpu);
4771 	struct sys_reg_params params;
4772 
4773 	/* UNDEF on any unhandled register access */
4774 	if (!kvm_esr_cp10_id_to_sys64(esr, &params)) {
4775 		kvm_inject_undefined(vcpu);
4776 		return 1;
4777 	}
4778 
4779 	if (emulate_sys_reg(vcpu, &params))
4780 		vcpu_set_reg(vcpu, Rt, params.regval);
4781 
4782 	return 1;
4783 }
4784 
4785 /**
4786  * kvm_emulate_cp15_id_reg() - Handles an MRC trap on a guest CP15 access where
4787  *			       CRn=0, which corresponds to the AArch32 feature
4788  *			       registers.
4789  * @vcpu: the vCPU pointer
4790  * @params: the system register access parameters.
4791  *
4792  * Our cp15 system register tables do not enumerate the AArch32 feature
4793  * registers. Conveniently, our AArch64 table does, and the AArch32 system
4794  * register encoding can be trivially remapped into the AArch64 for the feature
4795  * registers: Append op0=3, leaving op1, CRn, CRm, and op2 the same.
4796  *
4797  * According to DDI0487G.b G7.3.1, paragraph "Behavior of VMSAv8-32 32-bit
4798  * System registers with (coproc=0b1111, CRn==c0)", read accesses from this
4799  * range are either UNKNOWN or RES0. Rerouting remains architectural as we
4800  * treat undefined registers in this range as RAZ.
4801  */
kvm_emulate_cp15_id_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * params)4802 static int kvm_emulate_cp15_id_reg(struct kvm_vcpu *vcpu,
4803 				   struct sys_reg_params *params)
4804 {
4805 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
4806 
4807 	/* Treat impossible writes to RO registers as UNDEFINED */
4808 	if (params->is_write) {
4809 		unhandled_cp_access(vcpu, params);
4810 		return 1;
4811 	}
4812 
4813 	params->Op0 = 3;
4814 
4815 	/*
4816 	 * All registers where CRm > 3 are known to be UNKNOWN/RAZ from AArch32.
4817 	 * Avoid conflicting with future expansion of AArch64 feature registers
4818 	 * and simply treat them as RAZ here.
4819 	 */
4820 	if (params->CRm > 3)
4821 		params->regval = 0;
4822 	else if (!emulate_sys_reg(vcpu, params))
4823 		return 1;
4824 
4825 	vcpu_set_reg(vcpu, Rt, params->regval);
4826 	return 1;
4827 }
4828 
4829 /**
4830  * kvm_handle_cp_32 -- handles a mrc/mcr trap on a guest CP14/CP15 access
4831  * @vcpu: The VCPU pointer
4832  * @params: &struct sys_reg_params
4833  * @global: &struct sys_reg_desc
4834  * @nr_global: size of the @global array
4835  */
kvm_handle_cp_32(struct kvm_vcpu * vcpu,struct sys_reg_params * params,const struct sys_reg_desc * global,size_t nr_global)4836 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
4837 			    struct sys_reg_params *params,
4838 			    const struct sys_reg_desc *global,
4839 			    size_t nr_global)
4840 {
4841 	int Rt  = kvm_vcpu_sys_get_rt(vcpu);
4842 
4843 	params->regval = vcpu_get_reg(vcpu, Rt);
4844 
4845 	if (emulate_cp(vcpu, params, global, nr_global)) {
4846 		if (!params->is_write)
4847 			vcpu_set_reg(vcpu, Rt, params->regval);
4848 		return 1;
4849 	}
4850 
4851 	unhandled_cp_access(vcpu, params);
4852 	return 1;
4853 }
4854 
kvm_handle_cp15_64(struct kvm_vcpu * vcpu)4855 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu)
4856 {
4857 	return kvm_handle_cp_64(vcpu, cp15_64_regs, ARRAY_SIZE(cp15_64_regs));
4858 }
4859 
kvm_handle_cp15_32(struct kvm_vcpu * vcpu)4860 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu)
4861 {
4862 	struct sys_reg_params params;
4863 
4864 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4865 
4866 	/*
4867 	 * Certain AArch32 ID registers are handled by rerouting to the AArch64
4868 	 * system register table. Registers in the ID range where CRm=0 are
4869 	 * excluded from this scheme as they do not trivially map into AArch64
4870 	 * system register encodings, except for AIDR/REVIDR.
4871 	 */
4872 	if (params.Op1 == 0 && params.CRn == 0 &&
4873 	    (params.CRm || params.Op2 == 6 /* REVIDR */))
4874 		return kvm_emulate_cp15_id_reg(vcpu, &params);
4875 	if (params.Op1 == 1 && params.CRn == 0 &&
4876 	    params.CRm == 0 && params.Op2 == 7 /* AIDR */)
4877 		return kvm_emulate_cp15_id_reg(vcpu, &params);
4878 
4879 	return kvm_handle_cp_32(vcpu, &params, cp15_regs, ARRAY_SIZE(cp15_regs));
4880 }
4881 
kvm_handle_cp14_64(struct kvm_vcpu * vcpu)4882 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu)
4883 {
4884 	return kvm_handle_cp_64(vcpu, cp14_64_regs, ARRAY_SIZE(cp14_64_regs));
4885 }
4886 
kvm_handle_cp14_32(struct kvm_vcpu * vcpu)4887 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu)
4888 {
4889 	struct sys_reg_params params;
4890 
4891 	params = esr_cp1x_32_to_params(kvm_vcpu_get_esr(vcpu));
4892 
4893 	return kvm_handle_cp_32(vcpu, &params, cp14_regs, ARRAY_SIZE(cp14_regs));
4894 }
4895 
4896 /**
4897  * emulate_sys_reg - Emulate a guest access to an AArch64 system register
4898  * @vcpu: The VCPU pointer
4899  * @params: Decoded system register parameters
4900  *
4901  * Return: true if the system register access was successful, false otherwise.
4902  */
emulate_sys_reg(struct kvm_vcpu * vcpu,struct sys_reg_params * params)4903 static bool emulate_sys_reg(struct kvm_vcpu *vcpu,
4904 			    struct sys_reg_params *params)
4905 {
4906 	const struct sys_reg_desc *r;
4907 
4908 	r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
4909 	if (likely(r)) {
4910 		perform_access(vcpu, params, r);
4911 		return true;
4912 	}
4913 
4914 	print_sys_reg_msg(params,
4915 			  "Unsupported guest sys_reg access at: %lx [%08lx]\n",
4916 			  *vcpu_pc(vcpu), *vcpu_cpsr(vcpu));
4917 	kvm_inject_undefined(vcpu);
4918 
4919 	return false;
4920 }
4921 
idregs_debug_find(struct kvm * kvm,u8 pos)4922 static const struct sys_reg_desc *idregs_debug_find(struct kvm *kvm, u8 pos)
4923 {
4924 	unsigned long i, idreg_idx = 0;
4925 
4926 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
4927 		const struct sys_reg_desc *r = &sys_reg_descs[i];
4928 
4929 		if (!is_vm_ftr_id_reg(reg_to_encoding(r)))
4930 			continue;
4931 
4932 		if (idreg_idx == pos)
4933 			return r;
4934 
4935 		idreg_idx++;
4936 	}
4937 
4938 	return NULL;
4939 }
4940 
idregs_debug_start(struct seq_file * s,loff_t * pos)4941 static void *idregs_debug_start(struct seq_file *s, loff_t *pos)
4942 {
4943 	struct kvm *kvm = s->private;
4944 	u8 *iter;
4945 
4946 	mutex_lock(&kvm->arch.config_lock);
4947 
4948 	iter = &kvm->arch.idreg_debugfs_iter;
4949 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags) &&
4950 	    *iter == (u8)~0) {
4951 		*iter = *pos;
4952 		if (!idregs_debug_find(kvm, *iter))
4953 			iter = NULL;
4954 	} else {
4955 		iter = ERR_PTR(-EBUSY);
4956 	}
4957 
4958 	mutex_unlock(&kvm->arch.config_lock);
4959 
4960 	return iter;
4961 }
4962 
idregs_debug_next(struct seq_file * s,void * v,loff_t * pos)4963 static void *idregs_debug_next(struct seq_file *s, void *v, loff_t *pos)
4964 {
4965 	struct kvm *kvm = s->private;
4966 
4967 	(*pos)++;
4968 
4969 	if (idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter + 1)) {
4970 		kvm->arch.idreg_debugfs_iter++;
4971 
4972 		return &kvm->arch.idreg_debugfs_iter;
4973 	}
4974 
4975 	return NULL;
4976 }
4977 
idregs_debug_stop(struct seq_file * s,void * v)4978 static void idregs_debug_stop(struct seq_file *s, void *v)
4979 {
4980 	struct kvm *kvm = s->private;
4981 
4982 	if (IS_ERR(v))
4983 		return;
4984 
4985 	mutex_lock(&kvm->arch.config_lock);
4986 
4987 	kvm->arch.idreg_debugfs_iter = ~0;
4988 
4989 	mutex_unlock(&kvm->arch.config_lock);
4990 }
4991 
idregs_debug_show(struct seq_file * s,void * v)4992 static int idregs_debug_show(struct seq_file *s, void *v)
4993 {
4994 	const struct sys_reg_desc *desc;
4995 	struct kvm *kvm = s->private;
4996 
4997 	desc = idregs_debug_find(kvm, kvm->arch.idreg_debugfs_iter);
4998 
4999 	if (!desc->name)
5000 		return 0;
5001 
5002 	seq_printf(s, "%20s:\t%016llx\n",
5003 		   desc->name, kvm_read_vm_id_reg(kvm, reg_to_encoding(desc)));
5004 
5005 	return 0;
5006 }
5007 
5008 static const struct seq_operations idregs_debug_sops = {
5009 	.start	= idregs_debug_start,
5010 	.next	= idregs_debug_next,
5011 	.stop	= idregs_debug_stop,
5012 	.show	= idregs_debug_show,
5013 };
5014 
5015 DEFINE_SEQ_ATTRIBUTE(idregs_debug);
5016 
kvm_sys_regs_create_debugfs(struct kvm * kvm)5017 void kvm_sys_regs_create_debugfs(struct kvm *kvm)
5018 {
5019 	kvm->arch.idreg_debugfs_iter = ~0;
5020 
5021 	debugfs_create_file("idregs", 0444, kvm->debugfs_dentry, kvm,
5022 			    &idregs_debug_fops);
5023 }
5024 
reset_vm_ftr_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * reg)5025 static void reset_vm_ftr_id_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *reg)
5026 {
5027 	u32 id = reg_to_encoding(reg);
5028 	struct kvm *kvm = vcpu->kvm;
5029 
5030 	if (test_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags))
5031 		return;
5032 
5033 	kvm_set_vm_id_reg(kvm, id, reg->reset(vcpu, reg));
5034 }
5035 
reset_vcpu_ftr_id_reg(struct kvm_vcpu * vcpu,const struct sys_reg_desc * reg)5036 static void reset_vcpu_ftr_id_reg(struct kvm_vcpu *vcpu,
5037 				  const struct sys_reg_desc *reg)
5038 {
5039 	if (kvm_vcpu_initialized(vcpu))
5040 		return;
5041 
5042 	reg->reset(vcpu, reg);
5043 }
5044 
5045 /**
5046  * kvm_reset_sys_regs - sets system registers to reset value
5047  * @vcpu: The VCPU pointer
5048  *
5049  * This function finds the right table above and sets the registers on the
5050  * virtual CPU struct to their architecturally defined reset values.
5051  */
kvm_reset_sys_regs(struct kvm_vcpu * vcpu)5052 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
5053 {
5054 	struct kvm *kvm = vcpu->kvm;
5055 	unsigned long i;
5056 
5057 	for (i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
5058 		const struct sys_reg_desc *r = &sys_reg_descs[i];
5059 
5060 		if (!r->reset)
5061 			continue;
5062 
5063 		if (is_vm_ftr_id_reg(reg_to_encoding(r)))
5064 			reset_vm_ftr_id_reg(vcpu, r);
5065 		else if (is_vcpu_ftr_id_reg(reg_to_encoding(r)))
5066 			reset_vcpu_ftr_id_reg(vcpu, r);
5067 		else
5068 			r->reset(vcpu, r);
5069 
5070 		if (r->reg >= __SANITISED_REG_START__ && r->reg < NR_SYS_REGS)
5071 			__vcpu_rmw_sys_reg(vcpu, r->reg, |=, 0);
5072 	}
5073 
5074 	set_bit(KVM_ARCH_FLAG_ID_REGS_INITIALIZED, &kvm->arch.flags);
5075 
5076 	if (kvm_vcpu_has_pmu(vcpu))
5077 		kvm_make_request(KVM_REQ_RELOAD_PMU, vcpu);
5078 }
5079 
5080 /**
5081  * kvm_handle_sys_reg -- handles a system instruction or mrs/msr instruction
5082  *			 trap on a guest execution
5083  * @vcpu: The VCPU pointer
5084  */
kvm_handle_sys_reg(struct kvm_vcpu * vcpu)5085 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu)
5086 {
5087 	const struct sys_reg_desc *desc = NULL;
5088 	struct sys_reg_params params;
5089 	unsigned long esr = kvm_vcpu_get_esr(vcpu);
5090 	int Rt = kvm_vcpu_sys_get_rt(vcpu);
5091 	int sr_idx;
5092 
5093 	trace_kvm_handle_sys_reg(esr);
5094 
5095 	if (triage_sysreg_trap(vcpu, &sr_idx))
5096 		return 1;
5097 
5098 	params = esr_sys64_to_params(esr);
5099 	params.regval = vcpu_get_reg(vcpu, Rt);
5100 
5101 	/* System registers have Op0=={2,3}, as per DDI487 J.a C5.1.2 */
5102 	if (params.Op0 == 2 || params.Op0 == 3)
5103 		desc = &sys_reg_descs[sr_idx];
5104 	else
5105 		desc = &sys_insn_descs[sr_idx];
5106 
5107 	perform_access(vcpu, &params, desc);
5108 
5109 	/* Read from system register? */
5110 	if (!params.is_write &&
5111 	    (params.Op0 == 2 || params.Op0 == 3))
5112 		vcpu_set_reg(vcpu, Rt, params.regval);
5113 
5114 	return 1;
5115 }
5116 
5117 /******************************************************************************
5118  * Userspace API
5119  *****************************************************************************/
5120 
index_to_params(u64 id,struct sys_reg_params * params)5121 static bool index_to_params(u64 id, struct sys_reg_params *params)
5122 {
5123 	switch (id & KVM_REG_SIZE_MASK) {
5124 	case KVM_REG_SIZE_U64:
5125 		/* Any unused index bits means it's not valid. */
5126 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
5127 			      | KVM_REG_ARM_COPROC_MASK
5128 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
5129 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
5130 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
5131 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
5132 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
5133 			return false;
5134 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
5135 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
5136 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
5137 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
5138 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
5139 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
5140 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
5141 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
5142 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
5143 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
5144 		return true;
5145 	default:
5146 		return false;
5147 	}
5148 }
5149 
get_reg_by_id(u64 id,const struct sys_reg_desc table[],unsigned int num)5150 const struct sys_reg_desc *get_reg_by_id(u64 id,
5151 					 const struct sys_reg_desc table[],
5152 					 unsigned int num)
5153 {
5154 	struct sys_reg_params params;
5155 
5156 	if (!index_to_params(id, &params))
5157 		return NULL;
5158 
5159 	return find_reg(&params, table, num);
5160 }
5161 
5162 /* Decode an index value, and find the sys_reg_desc entry. */
5163 static const struct sys_reg_desc *
id_to_sys_reg_desc(struct kvm_vcpu * vcpu,u64 id,const struct sys_reg_desc table[],unsigned int num)5164 id_to_sys_reg_desc(struct kvm_vcpu *vcpu, u64 id,
5165 		   const struct sys_reg_desc table[], unsigned int num)
5166 
5167 {
5168 	const struct sys_reg_desc *r;
5169 
5170 	/* We only do sys_reg for now. */
5171 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
5172 		return NULL;
5173 
5174 	r = get_reg_by_id(id, table, num);
5175 
5176 	/* Not saved in the sys_reg array and not otherwise accessible? */
5177 	if (r && (!(r->reg || r->get_user) || sysreg_hidden(vcpu, r)))
5178 		r = NULL;
5179 
5180 	return r;
5181 }
5182 
demux_c15_get(struct kvm_vcpu * vcpu,u64 id,void __user * uaddr)5183 static int demux_c15_get(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
5184 {
5185 	u32 val;
5186 	u32 __user *uval = uaddr;
5187 
5188 	/* Fail if we have unknown bits set. */
5189 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
5190 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
5191 		return -ENOENT;
5192 
5193 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
5194 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
5195 		if (KVM_REG_SIZE(id) != 4)
5196 			return -ENOENT;
5197 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
5198 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
5199 		if (val >= CSSELR_MAX)
5200 			return -ENOENT;
5201 
5202 		return put_user(get_ccsidr(vcpu, val), uval);
5203 	default:
5204 		return -ENOENT;
5205 	}
5206 }
5207 
demux_c15_set(struct kvm_vcpu * vcpu,u64 id,void __user * uaddr)5208 static int demux_c15_set(struct kvm_vcpu *vcpu, u64 id, void __user *uaddr)
5209 {
5210 	u32 val, newval;
5211 	u32 __user *uval = uaddr;
5212 
5213 	/* Fail if we have unknown bits set. */
5214 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
5215 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
5216 		return -ENOENT;
5217 
5218 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
5219 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
5220 		if (KVM_REG_SIZE(id) != 4)
5221 			return -ENOENT;
5222 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
5223 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
5224 		if (val >= CSSELR_MAX)
5225 			return -ENOENT;
5226 
5227 		if (get_user(newval, uval))
5228 			return -EFAULT;
5229 
5230 		return set_ccsidr(vcpu, val, newval);
5231 	default:
5232 		return -ENOENT;
5233 	}
5234 }
5235 
kvm_sys_reg_get_user(struct kvm_vcpu * vcpu,const struct kvm_one_reg * reg,const struct sys_reg_desc table[],unsigned int num)5236 int kvm_sys_reg_get_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
5237 			 const struct sys_reg_desc table[], unsigned int num)
5238 {
5239 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
5240 	const struct sys_reg_desc *r;
5241 	u64 val;
5242 	int ret;
5243 
5244 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
5245 	if (!r || sysreg_hidden(vcpu, r))
5246 		return -ENOENT;
5247 
5248 	if (r->get_user) {
5249 		ret = (r->get_user)(vcpu, r, &val);
5250 	} else {
5251 		val = __vcpu_sys_reg(vcpu, r->reg);
5252 		ret = 0;
5253 	}
5254 
5255 	if (!ret)
5256 		ret = put_user(val, uaddr);
5257 
5258 	return ret;
5259 }
5260 
kvm_arm_sys_reg_get_reg(struct kvm_vcpu * vcpu,const struct kvm_one_reg * reg)5261 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
5262 {
5263 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
5264 
5265 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
5266 		return demux_c15_get(vcpu, reg->id, uaddr);
5267 
5268 	return kvm_sys_reg_get_user(vcpu, reg,
5269 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
5270 }
5271 
kvm_sys_reg_set_user(struct kvm_vcpu * vcpu,const struct kvm_one_reg * reg,const struct sys_reg_desc table[],unsigned int num)5272 int kvm_sys_reg_set_user(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg,
5273 			 const struct sys_reg_desc table[], unsigned int num)
5274 {
5275 	u64 __user *uaddr = (u64 __user *)(unsigned long)reg->addr;
5276 	const struct sys_reg_desc *r;
5277 	u64 val;
5278 	int ret;
5279 
5280 	if (get_user(val, uaddr))
5281 		return -EFAULT;
5282 
5283 	r = id_to_sys_reg_desc(vcpu, reg->id, table, num);
5284 	if (!r || sysreg_hidden(vcpu, r))
5285 		return -ENOENT;
5286 
5287 	if (sysreg_user_write_ignore(vcpu, r))
5288 		return 0;
5289 
5290 	if (r->set_user) {
5291 		ret = (r->set_user)(vcpu, r, val);
5292 	} else {
5293 		__vcpu_assign_sys_reg(vcpu, r->reg, val);
5294 		ret = 0;
5295 	}
5296 
5297 	return ret;
5298 }
5299 
kvm_arm_sys_reg_set_reg(struct kvm_vcpu * vcpu,const struct kvm_one_reg * reg)5300 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
5301 {
5302 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
5303 
5304 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
5305 		return demux_c15_set(vcpu, reg->id, uaddr);
5306 
5307 	return kvm_sys_reg_set_user(vcpu, reg,
5308 				    sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
5309 }
5310 
num_demux_regs(void)5311 static unsigned int num_demux_regs(void)
5312 {
5313 	return CSSELR_MAX;
5314 }
5315 
write_demux_regids(u64 __user * uindices)5316 static int write_demux_regids(u64 __user *uindices)
5317 {
5318 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
5319 	unsigned int i;
5320 
5321 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
5322 	for (i = 0; i < CSSELR_MAX; i++) {
5323 		if (put_user(val | i, uindices))
5324 			return -EFAULT;
5325 		uindices++;
5326 	}
5327 	return 0;
5328 }
5329 
sys_reg_to_index(const struct sys_reg_desc * reg)5330 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
5331 {
5332 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
5333 		KVM_REG_ARM64_SYSREG |
5334 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
5335 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
5336 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
5337 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
5338 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
5339 }
5340 
copy_reg_to_user(const struct sys_reg_desc * reg,u64 __user ** uind)5341 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
5342 {
5343 	if (!*uind)
5344 		return true;
5345 
5346 	if (put_user(sys_reg_to_index(reg), *uind))
5347 		return false;
5348 
5349 	(*uind)++;
5350 	return true;
5351 }
5352 
walk_one_sys_reg(const struct kvm_vcpu * vcpu,const struct sys_reg_desc * rd,u64 __user ** uind,unsigned int * total)5353 static int walk_one_sys_reg(const struct kvm_vcpu *vcpu,
5354 			    const struct sys_reg_desc *rd,
5355 			    u64 __user **uind,
5356 			    unsigned int *total)
5357 {
5358 	/*
5359 	 * Ignore registers we trap but don't save,
5360 	 * and for which no custom user accessor is provided.
5361 	 */
5362 	if (!(rd->reg || rd->get_user))
5363 		return 0;
5364 
5365 	if (sysreg_hidden(vcpu, rd))
5366 		return 0;
5367 
5368 	if (!copy_reg_to_user(rd, uind))
5369 		return -EFAULT;
5370 
5371 	(*total)++;
5372 	return 0;
5373 }
5374 
5375 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
walk_sys_regs(struct kvm_vcpu * vcpu,u64 __user * uind)5376 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
5377 {
5378 	const struct sys_reg_desc *i2, *end2;
5379 	unsigned int total = 0;
5380 	int err;
5381 
5382 	i2 = sys_reg_descs;
5383 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
5384 
5385 	while (i2 != end2) {
5386 		err = walk_one_sys_reg(vcpu, i2++, &uind, &total);
5387 		if (err)
5388 			return err;
5389 	}
5390 	return total;
5391 }
5392 
kvm_arm_num_sys_reg_descs(struct kvm_vcpu * vcpu)5393 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
5394 {
5395 	return num_demux_regs()
5396 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
5397 }
5398 
kvm_arm_copy_sys_reg_indices(struct kvm_vcpu * vcpu,u64 __user * uindices)5399 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
5400 {
5401 	int err;
5402 
5403 	err = walk_sys_regs(vcpu, uindices);
5404 	if (err < 0)
5405 		return err;
5406 	uindices += err;
5407 
5408 	return write_demux_regids(uindices);
5409 }
5410 
5411 #define KVM_ARM_FEATURE_ID_RANGE_INDEX(r)			\
5412 	KVM_ARM_FEATURE_ID_RANGE_IDX(sys_reg_Op0(r),		\
5413 		sys_reg_Op1(r),					\
5414 		sys_reg_CRn(r),					\
5415 		sys_reg_CRm(r),					\
5416 		sys_reg_Op2(r))
5417 
kvm_vm_ioctl_get_reg_writable_masks(struct kvm * kvm,struct reg_mask_range * range)5418 int kvm_vm_ioctl_get_reg_writable_masks(struct kvm *kvm, struct reg_mask_range *range)
5419 {
5420 	const void *zero_page = page_to_virt(ZERO_PAGE(0));
5421 	u64 __user *masks = (u64 __user *)range->addr;
5422 
5423 	/* Only feature id range is supported, reserved[13] must be zero. */
5424 	if (range->range ||
5425 	    memcmp(range->reserved, zero_page, sizeof(range->reserved)))
5426 		return -EINVAL;
5427 
5428 	/* Wipe the whole thing first */
5429 	if (clear_user(masks, KVM_ARM_FEATURE_ID_RANGE_SIZE * sizeof(__u64)))
5430 		return -EFAULT;
5431 
5432 	for (int i = 0; i < ARRAY_SIZE(sys_reg_descs); i++) {
5433 		const struct sys_reg_desc *reg = &sys_reg_descs[i];
5434 		u32 encoding = reg_to_encoding(reg);
5435 		u64 val;
5436 
5437 		if (!is_feature_id_reg(encoding) || !reg->set_user)
5438 			continue;
5439 
5440 		if (!reg->val ||
5441 		    (is_aa32_id_reg(encoding) && !kvm_supports_32bit_el0())) {
5442 			continue;
5443 		}
5444 		val = reg->val;
5445 
5446 		if (put_user(val, (masks + KVM_ARM_FEATURE_ID_RANGE_INDEX(encoding))))
5447 			return -EFAULT;
5448 	}
5449 
5450 	return 0;
5451 }
5452 
vcpu_set_hcr(struct kvm_vcpu * vcpu)5453 static void vcpu_set_hcr(struct kvm_vcpu *vcpu)
5454 {
5455 	struct kvm *kvm = vcpu->kvm;
5456 
5457 	if (has_vhe() || has_hvhe())
5458 		vcpu->arch.hcr_el2 |= HCR_E2H;
5459 	if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN)) {
5460 		/* route synchronous external abort exceptions to EL2 */
5461 		vcpu->arch.hcr_el2 |= HCR_TEA;
5462 		/* trap error record accesses */
5463 		vcpu->arch.hcr_el2 |= HCR_TERR;
5464 	}
5465 
5466 	if (cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
5467 		vcpu->arch.hcr_el2 |= HCR_FWB;
5468 
5469 	if (cpus_have_final_cap(ARM64_HAS_EVT) &&
5470 	    !cpus_have_final_cap(ARM64_MISMATCHED_CACHE_TYPE) &&
5471 	    kvm_read_vm_id_reg(kvm, SYS_CTR_EL0) == read_sanitised_ftr_reg(SYS_CTR_EL0))
5472 		vcpu->arch.hcr_el2 |= HCR_TID4;
5473 	else
5474 		vcpu->arch.hcr_el2 |= HCR_TID2;
5475 
5476 	if (vcpu_el1_is_32bit(vcpu))
5477 		vcpu->arch.hcr_el2 &= ~HCR_RW;
5478 
5479 	if (kvm_has_mte(vcpu->kvm))
5480 		vcpu->arch.hcr_el2 |= HCR_ATA;
5481 
5482 	/*
5483 	 * In the absence of FGT, we cannot independently trap TLBI
5484 	 * Range instructions. This isn't great, but trapping all
5485 	 * TLBIs would be far worse. Live with it...
5486 	 */
5487 	if (!kvm_has_feat(kvm, ID_AA64ISAR0_EL1, TLB, OS))
5488 		vcpu->arch.hcr_el2 |= HCR_TTLBOS;
5489 }
5490 
kvm_calculate_traps(struct kvm_vcpu * vcpu)5491 void kvm_calculate_traps(struct kvm_vcpu *vcpu)
5492 {
5493 	struct kvm *kvm = vcpu->kvm;
5494 
5495 	mutex_lock(&kvm->arch.config_lock);
5496 	vcpu_set_hcr(vcpu);
5497 	vcpu_set_ich_hcr(vcpu);
5498 	vcpu_set_hcrx(vcpu);
5499 
5500 	if (test_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags))
5501 		goto out;
5502 
5503 	compute_fgu(kvm, HFGRTR_GROUP);
5504 	compute_fgu(kvm, HFGITR_GROUP);
5505 	compute_fgu(kvm, HDFGRTR_GROUP);
5506 	compute_fgu(kvm, HAFGRTR_GROUP);
5507 	compute_fgu(kvm, HFGRTR2_GROUP);
5508 	compute_fgu(kvm, HFGITR2_GROUP);
5509 	compute_fgu(kvm, HDFGRTR2_GROUP);
5510 
5511 	set_bit(KVM_ARCH_FLAG_FGU_INITIALIZED, &kvm->arch.flags);
5512 out:
5513 	mutex_unlock(&kvm->arch.config_lock);
5514 }
5515 
5516 /*
5517  * Perform last adjustments to the ID registers that are implied by the
5518  * configuration outside of the ID regs themselves, as well as any
5519  * initialisation that directly depend on these ID registers (such as
5520  * RES0/RES1 behaviours). This is not the place to configure traps though.
5521  *
5522  * Because this can be called once per CPU, changes must be idempotent.
5523  */
kvm_finalize_sys_regs(struct kvm_vcpu * vcpu)5524 int kvm_finalize_sys_regs(struct kvm_vcpu *vcpu)
5525 {
5526 	struct kvm *kvm = vcpu->kvm;
5527 
5528 	guard(mutex)(&kvm->arch.config_lock);
5529 
5530 	if (!(static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif) &&
5531 	      irqchip_in_kernel(kvm) &&
5532 	      kvm->arch.vgic.vgic_model == KVM_DEV_TYPE_ARM_VGIC_V3)) {
5533 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_AA64PFR0_EL1)] &= ~ID_AA64PFR0_EL1_GIC_MASK;
5534 		kvm->arch.id_regs[IDREG_IDX(SYS_ID_PFR1_EL1)] &= ~ID_PFR1_EL1_GIC_MASK;
5535 	}
5536 
5537 	if (vcpu_has_nv(vcpu)) {
5538 		int ret = kvm_init_nv_sysregs(vcpu);
5539 		if (ret)
5540 			return ret;
5541 	}
5542 
5543 	return 0;
5544 }
5545 
kvm_sys_reg_table_init(void)5546 int __init kvm_sys_reg_table_init(void)
5547 {
5548 	const struct sys_reg_desc *gicv3_regs;
5549 	bool valid = true;
5550 	unsigned int i, sz;
5551 	int ret = 0;
5552 
5553 	/* Make sure tables are unique and in order. */
5554 	valid &= check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs), true);
5555 	valid &= check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs), false);
5556 	valid &= check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs), false);
5557 	valid &= check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs), false);
5558 	valid &= check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs), false);
5559 	valid &= check_sysreg_table(sys_insn_descs, ARRAY_SIZE(sys_insn_descs), false);
5560 
5561 	gicv3_regs = vgic_v3_get_sysreg_table(&sz);
5562 	valid &= check_sysreg_table(gicv3_regs, sz, false);
5563 
5564 	if (!valid)
5565 		return -EINVAL;
5566 
5567 	init_imp_id_regs();
5568 
5569 	ret = populate_nv_trap_config();
5570 
5571 	check_feature_map();
5572 
5573 	for (i = 0; !ret && i < ARRAY_SIZE(sys_reg_descs); i++)
5574 		ret = populate_sysreg_config(sys_reg_descs + i, i);
5575 
5576 	for (i = 0; !ret && i < ARRAY_SIZE(sys_insn_descs); i++)
5577 		ret = populate_sysreg_config(sys_insn_descs + i, i);
5578 
5579 	return ret;
5580 }
5581