xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision 275876e2)
1 /*
2  * Copyright (C) 2012,2013 - ARM Ltd
3  * Author: Marc Zyngier <marc.zyngier@arm.com>
4  *
5  * Derived from arch/arm/kvm/coproc.c:
6  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
7  * Authors: Rusty Russell <rusty@rustcorp.com.au>
8  *          Christoffer Dall <c.dall@virtualopensystems.com>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License, version 2, as
12  * published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License
20  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21  */
22 
23 #include <linux/mm.h>
24 #include <linux/kvm_host.h>
25 #include <linux/uaccess.h>
26 #include <asm/kvm_arm.h>
27 #include <asm/kvm_host.h>
28 #include <asm/kvm_emulate.h>
29 #include <asm/kvm_coproc.h>
30 #include <asm/kvm_mmu.h>
31 #include <asm/cacheflush.h>
32 #include <asm/cputype.h>
33 #include <asm/debug-monitors.h>
34 #include <trace/events/kvm.h>
35 
36 #include "sys_regs.h"
37 
38 /*
39  * All of this file is extremly similar to the ARM coproc.c, but the
40  * types are different. My gut feeling is that it should be pretty
41  * easy to merge, but that would be an ABI breakage -- again. VFP
42  * would also need to be abstracted.
43  *
44  * For AArch32, we only take care of what is being trapped. Anything
45  * that has to do with init and userspace access has to go via the
46  * 64bit interface.
47  */
48 
49 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
50 static u32 cache_levels;
51 
52 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
53 #define CSSELR_MAX 12
54 
55 /* Which cache CCSIDR represents depends on CSSELR value. */
56 static u32 get_ccsidr(u32 csselr)
57 {
58 	u32 ccsidr;
59 
60 	/* Make sure noone else changes CSSELR during this! */
61 	local_irq_disable();
62 	/* Put value into CSSELR */
63 	asm volatile("msr csselr_el1, %x0" : : "r" (csselr));
64 	isb();
65 	/* Read result out of CCSIDR */
66 	asm volatile("mrs %0, ccsidr_el1" : "=r" (ccsidr));
67 	local_irq_enable();
68 
69 	return ccsidr;
70 }
71 
72 static void do_dc_cisw(u32 val)
73 {
74 	asm volatile("dc cisw, %x0" : : "r" (val));
75 	dsb(ish);
76 }
77 
78 static void do_dc_csw(u32 val)
79 {
80 	asm volatile("dc csw, %x0" : : "r" (val));
81 	dsb(ish);
82 }
83 
84 /* See note at ARM ARM B1.14.4 */
85 static bool access_dcsw(struct kvm_vcpu *vcpu,
86 			const struct sys_reg_params *p,
87 			const struct sys_reg_desc *r)
88 {
89 	unsigned long val;
90 	int cpu;
91 
92 	if (!p->is_write)
93 		return read_from_write_only(vcpu, p);
94 
95 	cpu = get_cpu();
96 
97 	cpumask_setall(&vcpu->arch.require_dcache_flush);
98 	cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
99 
100 	/* If we were already preempted, take the long way around */
101 	if (cpu != vcpu->arch.last_pcpu) {
102 		flush_cache_all();
103 		goto done;
104 	}
105 
106 	val = *vcpu_reg(vcpu, p->Rt);
107 
108 	switch (p->CRm) {
109 	case 6:			/* Upgrade DCISW to DCCISW, as per HCR.SWIO */
110 	case 14:		/* DCCISW */
111 		do_dc_cisw(val);
112 		break;
113 
114 	case 10:		/* DCCSW */
115 		do_dc_csw(val);
116 		break;
117 	}
118 
119 done:
120 	put_cpu();
121 
122 	return true;
123 }
124 
125 /*
126  * Generic accessor for VM registers. Only called as long as HCR_TVM
127  * is set.
128  */
129 static bool access_vm_reg(struct kvm_vcpu *vcpu,
130 			  const struct sys_reg_params *p,
131 			  const struct sys_reg_desc *r)
132 {
133 	unsigned long val;
134 
135 	BUG_ON(!p->is_write);
136 
137 	val = *vcpu_reg(vcpu, p->Rt);
138 	if (!p->is_aarch32) {
139 		vcpu_sys_reg(vcpu, r->reg) = val;
140 	} else {
141 		if (!p->is_32bit)
142 			vcpu_cp15_64_high(vcpu, r->reg) = val >> 32;
143 		vcpu_cp15_64_low(vcpu, r->reg) = val & 0xffffffffUL;
144 	}
145 
146 	return true;
147 }
148 
149 /*
150  * SCTLR_EL1 accessor. Only called as long as HCR_TVM is set.  If the
151  * guest enables the MMU, we stop trapping the VM sys_regs and leave
152  * it in complete control of the caches.
153  */
154 static bool access_sctlr(struct kvm_vcpu *vcpu,
155 			 const struct sys_reg_params *p,
156 			 const struct sys_reg_desc *r)
157 {
158 	access_vm_reg(vcpu, p, r);
159 
160 	if (vcpu_has_cache_enabled(vcpu)) {	/* MMU+Caches enabled? */
161 		vcpu->arch.hcr_el2 &= ~HCR_TVM;
162 		stage2_flush_vm(vcpu->kvm);
163 	}
164 
165 	return true;
166 }
167 
168 static bool trap_raz_wi(struct kvm_vcpu *vcpu,
169 			const struct sys_reg_params *p,
170 			const struct sys_reg_desc *r)
171 {
172 	if (p->is_write)
173 		return ignore_write(vcpu, p);
174 	else
175 		return read_zero(vcpu, p);
176 }
177 
178 static bool trap_oslsr_el1(struct kvm_vcpu *vcpu,
179 			   const struct sys_reg_params *p,
180 			   const struct sys_reg_desc *r)
181 {
182 	if (p->is_write) {
183 		return ignore_write(vcpu, p);
184 	} else {
185 		*vcpu_reg(vcpu, p->Rt) = (1 << 3);
186 		return true;
187 	}
188 }
189 
190 static bool trap_dbgauthstatus_el1(struct kvm_vcpu *vcpu,
191 				   const struct sys_reg_params *p,
192 				   const struct sys_reg_desc *r)
193 {
194 	if (p->is_write) {
195 		return ignore_write(vcpu, p);
196 	} else {
197 		u32 val;
198 		asm volatile("mrs %0, dbgauthstatus_el1" : "=r" (val));
199 		*vcpu_reg(vcpu, p->Rt) = val;
200 		return true;
201 	}
202 }
203 
204 /*
205  * We want to avoid world-switching all the DBG registers all the
206  * time:
207  *
208  * - If we've touched any debug register, it is likely that we're
209  *   going to touch more of them. It then makes sense to disable the
210  *   traps and start doing the save/restore dance
211  * - If debug is active (DBG_MDSCR_KDE or DBG_MDSCR_MDE set), it is
212  *   then mandatory to save/restore the registers, as the guest
213  *   depends on them.
214  *
215  * For this, we use a DIRTY bit, indicating the guest has modified the
216  * debug registers, used as follow:
217  *
218  * On guest entry:
219  * - If the dirty bit is set (because we're coming back from trapping),
220  *   disable the traps, save host registers, restore guest registers.
221  * - If debug is actively in use (DBG_MDSCR_KDE or DBG_MDSCR_MDE set),
222  *   set the dirty bit, disable the traps, save host registers,
223  *   restore guest registers.
224  * - Otherwise, enable the traps
225  *
226  * On guest exit:
227  * - If the dirty bit is set, save guest registers, restore host
228  *   registers and clear the dirty bit. This ensure that the host can
229  *   now use the debug registers.
230  */
231 static bool trap_debug_regs(struct kvm_vcpu *vcpu,
232 			    const struct sys_reg_params *p,
233 			    const struct sys_reg_desc *r)
234 {
235 	if (p->is_write) {
236 		vcpu_sys_reg(vcpu, r->reg) = *vcpu_reg(vcpu, p->Rt);
237 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
238 	} else {
239 		*vcpu_reg(vcpu, p->Rt) = vcpu_sys_reg(vcpu, r->reg);
240 	}
241 
242 	return true;
243 }
244 
245 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
246 {
247 	u64 amair;
248 
249 	asm volatile("mrs %0, amair_el1\n" : "=r" (amair));
250 	vcpu_sys_reg(vcpu, AMAIR_EL1) = amair;
251 }
252 
253 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
254 {
255 	/*
256 	 * Simply map the vcpu_id into the Aff0 field of the MPIDR.
257 	 */
258 	vcpu_sys_reg(vcpu, MPIDR_EL1) = (1UL << 31) | (vcpu->vcpu_id & 0xff);
259 }
260 
261 /* Silly macro to expand the DBG{BCR,BVR,WVR,WCR}n_EL1 registers in one go */
262 #define DBG_BCR_BVR_WCR_WVR_EL1(n)					\
263 	/* DBGBVRn_EL1 */						\
264 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b100),	\
265 	  trap_debug_regs, reset_val, (DBGBVR0_EL1 + (n)), 0 },		\
266 	/* DBGBCRn_EL1 */						\
267 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b101),	\
268 	  trap_debug_regs, reset_val, (DBGBCR0_EL1 + (n)), 0 },		\
269 	/* DBGWVRn_EL1 */						\
270 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b110),	\
271 	  trap_debug_regs, reset_val, (DBGWVR0_EL1 + (n)), 0 },		\
272 	/* DBGWCRn_EL1 */						\
273 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm((n)), Op2(0b111),	\
274 	  trap_debug_regs, reset_val, (DBGWCR0_EL1 + (n)), 0 }
275 
276 /*
277  * Architected system registers.
278  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
279  *
280  * We could trap ID_DFR0 and tell the guest we don't support performance
281  * monitoring.  Unfortunately the patch to make the kernel check ID_DFR0 was
282  * NAKed, so it will read the PMCR anyway.
283  *
284  * Therefore we tell the guest we have 0 counters.  Unfortunately, we
285  * must always support PMCCNTR (the cycle counter): we just RAZ/WI for
286  * all PM registers, which doesn't crash the guest kernel at least.
287  *
288  * Debug handling: We do trap most, if not all debug related system
289  * registers. The implementation is good enough to ensure that a guest
290  * can use these with minimal performance degradation. The drawback is
291  * that we don't implement any of the external debug, none of the
292  * OSlock protocol. This should be revisited if we ever encounter a
293  * more demanding guest...
294  */
295 static const struct sys_reg_desc sys_reg_descs[] = {
296 	/* DC ISW */
297 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
298 	  access_dcsw },
299 	/* DC CSW */
300 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
301 	  access_dcsw },
302 	/* DC CISW */
303 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
304 	  access_dcsw },
305 
306 	DBG_BCR_BVR_WCR_WVR_EL1(0),
307 	DBG_BCR_BVR_WCR_WVR_EL1(1),
308 	/* MDCCINT_EL1 */
309 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
310 	  trap_debug_regs, reset_val, MDCCINT_EL1, 0 },
311 	/* MDSCR_EL1 */
312 	{ Op0(0b10), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
313 	  trap_debug_regs, reset_val, MDSCR_EL1, 0 },
314 	DBG_BCR_BVR_WCR_WVR_EL1(2),
315 	DBG_BCR_BVR_WCR_WVR_EL1(3),
316 	DBG_BCR_BVR_WCR_WVR_EL1(4),
317 	DBG_BCR_BVR_WCR_WVR_EL1(5),
318 	DBG_BCR_BVR_WCR_WVR_EL1(6),
319 	DBG_BCR_BVR_WCR_WVR_EL1(7),
320 	DBG_BCR_BVR_WCR_WVR_EL1(8),
321 	DBG_BCR_BVR_WCR_WVR_EL1(9),
322 	DBG_BCR_BVR_WCR_WVR_EL1(10),
323 	DBG_BCR_BVR_WCR_WVR_EL1(11),
324 	DBG_BCR_BVR_WCR_WVR_EL1(12),
325 	DBG_BCR_BVR_WCR_WVR_EL1(13),
326 	DBG_BCR_BVR_WCR_WVR_EL1(14),
327 	DBG_BCR_BVR_WCR_WVR_EL1(15),
328 
329 	/* MDRAR_EL1 */
330 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
331 	  trap_raz_wi },
332 	/* OSLAR_EL1 */
333 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b100),
334 	  trap_raz_wi },
335 	/* OSLSR_EL1 */
336 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0001), Op2(0b100),
337 	  trap_oslsr_el1 },
338 	/* OSDLR_EL1 */
339 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0011), Op2(0b100),
340 	  trap_raz_wi },
341 	/* DBGPRCR_EL1 */
342 	{ Op0(0b10), Op1(0b000), CRn(0b0001), CRm(0b0100), Op2(0b100),
343 	  trap_raz_wi },
344 	/* DBGCLAIMSET_EL1 */
345 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1000), Op2(0b110),
346 	  trap_raz_wi },
347 	/* DBGCLAIMCLR_EL1 */
348 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1001), Op2(0b110),
349 	  trap_raz_wi },
350 	/* DBGAUTHSTATUS_EL1 */
351 	{ Op0(0b10), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b110),
352 	  trap_dbgauthstatus_el1 },
353 
354 	/* TEECR32_EL1 */
355 	{ Op0(0b10), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
356 	  NULL, reset_val, TEECR32_EL1, 0 },
357 	/* TEEHBR32_EL1 */
358 	{ Op0(0b10), Op1(0b010), CRn(0b0001), CRm(0b0000), Op2(0b000),
359 	  NULL, reset_val, TEEHBR32_EL1, 0 },
360 
361 	/* MDCCSR_EL1 */
362 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0001), Op2(0b000),
363 	  trap_raz_wi },
364 	/* DBGDTR_EL0 */
365 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0100), Op2(0b000),
366 	  trap_raz_wi },
367 	/* DBGDTR[TR]X_EL0 */
368 	{ Op0(0b10), Op1(0b011), CRn(0b0000), CRm(0b0101), Op2(0b000),
369 	  trap_raz_wi },
370 
371 	/* DBGVCR32_EL2 */
372 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
373 	  NULL, reset_val, DBGVCR32_EL2, 0 },
374 
375 	/* MPIDR_EL1 */
376 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
377 	  NULL, reset_mpidr, MPIDR_EL1 },
378 	/* SCTLR_EL1 */
379 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
380 	  access_sctlr, reset_val, SCTLR_EL1, 0x00C50078 },
381 	/* CPACR_EL1 */
382 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
383 	  NULL, reset_val, CPACR_EL1, 0 },
384 	/* TTBR0_EL1 */
385 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
386 	  access_vm_reg, reset_unknown, TTBR0_EL1 },
387 	/* TTBR1_EL1 */
388 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
389 	  access_vm_reg, reset_unknown, TTBR1_EL1 },
390 	/* TCR_EL1 */
391 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
392 	  access_vm_reg, reset_val, TCR_EL1, 0 },
393 
394 	/* AFSR0_EL1 */
395 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
396 	  access_vm_reg, reset_unknown, AFSR0_EL1 },
397 	/* AFSR1_EL1 */
398 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
399 	  access_vm_reg, reset_unknown, AFSR1_EL1 },
400 	/* ESR_EL1 */
401 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
402 	  access_vm_reg, reset_unknown, ESR_EL1 },
403 	/* FAR_EL1 */
404 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
405 	  access_vm_reg, reset_unknown, FAR_EL1 },
406 	/* PAR_EL1 */
407 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
408 	  NULL, reset_unknown, PAR_EL1 },
409 
410 	/* PMINTENSET_EL1 */
411 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
412 	  trap_raz_wi },
413 	/* PMINTENCLR_EL1 */
414 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
415 	  trap_raz_wi },
416 
417 	/* MAIR_EL1 */
418 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
419 	  access_vm_reg, reset_unknown, MAIR_EL1 },
420 	/* AMAIR_EL1 */
421 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
422 	  access_vm_reg, reset_amair_el1, AMAIR_EL1 },
423 
424 	/* VBAR_EL1 */
425 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
426 	  NULL, reset_val, VBAR_EL1, 0 },
427 	/* CONTEXTIDR_EL1 */
428 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
429 	  access_vm_reg, reset_val, CONTEXTIDR_EL1, 0 },
430 	/* TPIDR_EL1 */
431 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
432 	  NULL, reset_unknown, TPIDR_EL1 },
433 
434 	/* CNTKCTL_EL1 */
435 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
436 	  NULL, reset_val, CNTKCTL_EL1, 0},
437 
438 	/* CSSELR_EL1 */
439 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
440 	  NULL, reset_unknown, CSSELR_EL1 },
441 
442 	/* PMCR_EL0 */
443 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
444 	  trap_raz_wi },
445 	/* PMCNTENSET_EL0 */
446 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
447 	  trap_raz_wi },
448 	/* PMCNTENCLR_EL0 */
449 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
450 	  trap_raz_wi },
451 	/* PMOVSCLR_EL0 */
452 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
453 	  trap_raz_wi },
454 	/* PMSWINC_EL0 */
455 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
456 	  trap_raz_wi },
457 	/* PMSELR_EL0 */
458 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
459 	  trap_raz_wi },
460 	/* PMCEID0_EL0 */
461 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
462 	  trap_raz_wi },
463 	/* PMCEID1_EL0 */
464 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
465 	  trap_raz_wi },
466 	/* PMCCNTR_EL0 */
467 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
468 	  trap_raz_wi },
469 	/* PMXEVTYPER_EL0 */
470 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
471 	  trap_raz_wi },
472 	/* PMXEVCNTR_EL0 */
473 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
474 	  trap_raz_wi },
475 	/* PMUSERENR_EL0 */
476 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
477 	  trap_raz_wi },
478 	/* PMOVSSET_EL0 */
479 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
480 	  trap_raz_wi },
481 
482 	/* TPIDR_EL0 */
483 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
484 	  NULL, reset_unknown, TPIDR_EL0 },
485 	/* TPIDRRO_EL0 */
486 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
487 	  NULL, reset_unknown, TPIDRRO_EL0 },
488 
489 	/* DACR32_EL2 */
490 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
491 	  NULL, reset_unknown, DACR32_EL2 },
492 	/* IFSR32_EL2 */
493 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
494 	  NULL, reset_unknown, IFSR32_EL2 },
495 	/* FPEXC32_EL2 */
496 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
497 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
498 };
499 
500 static bool trap_dbgidr(struct kvm_vcpu *vcpu,
501 			const struct sys_reg_params *p,
502 			const struct sys_reg_desc *r)
503 {
504 	if (p->is_write) {
505 		return ignore_write(vcpu, p);
506 	} else {
507 		u64 dfr = read_cpuid(ID_AA64DFR0_EL1);
508 		u64 pfr = read_cpuid(ID_AA64PFR0_EL1);
509 		u32 el3 = !!((pfr >> 12) & 0xf);
510 
511 		*vcpu_reg(vcpu, p->Rt) = ((((dfr >> 20) & 0xf) << 28) |
512 					  (((dfr >> 12) & 0xf) << 24) |
513 					  (((dfr >> 28) & 0xf) << 20) |
514 					  (6 << 16) | (el3 << 14) | (el3 << 12));
515 		return true;
516 	}
517 }
518 
519 static bool trap_debug32(struct kvm_vcpu *vcpu,
520 			 const struct sys_reg_params *p,
521 			 const struct sys_reg_desc *r)
522 {
523 	if (p->is_write) {
524 		vcpu_cp14(vcpu, r->reg) = *vcpu_reg(vcpu, p->Rt);
525 		vcpu->arch.debug_flags |= KVM_ARM64_DEBUG_DIRTY;
526 	} else {
527 		*vcpu_reg(vcpu, p->Rt) = vcpu_cp14(vcpu, r->reg);
528 	}
529 
530 	return true;
531 }
532 
533 #define DBG_BCR_BVR_WCR_WVR(n)					\
534 	/* DBGBVRn */						\
535 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 4), trap_debug32,	\
536 	  NULL, (cp14_DBGBVR0 + (n) * 2) },			\
537 	/* DBGBCRn */						\
538 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 5), trap_debug32,	\
539 	  NULL, (cp14_DBGBCR0 + (n) * 2) },			\
540 	/* DBGWVRn */						\
541 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 6), trap_debug32,	\
542 	  NULL, (cp14_DBGWVR0 + (n) * 2) },			\
543 	/* DBGWCRn */						\
544 	{ Op1( 0), CRn( 0), CRm((n)), Op2( 7), trap_debug32,	\
545 	  NULL, (cp14_DBGWCR0 + (n) * 2) }
546 
547 #define DBGBXVR(n)						\
548 	{ Op1( 0), CRn( 1), CRm((n)), Op2( 1), trap_debug32,	\
549 	  NULL, cp14_DBGBXVR0 + n * 2 }
550 
551 /*
552  * Trapped cp14 registers. We generally ignore most of the external
553  * debug, on the principle that they don't really make sense to a
554  * guest. Revisit this one day, whould this principle change.
555  */
556 static const struct sys_reg_desc cp14_regs[] = {
557 	/* DBGIDR */
558 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 0), trap_dbgidr },
559 	/* DBGDTRRXext */
560 	{ Op1( 0), CRn( 0), CRm( 0), Op2( 2), trap_raz_wi },
561 
562 	DBG_BCR_BVR_WCR_WVR(0),
563 	/* DBGDSCRint */
564 	{ Op1( 0), CRn( 0), CRm( 1), Op2( 0), trap_raz_wi },
565 	DBG_BCR_BVR_WCR_WVR(1),
566 	/* DBGDCCINT */
567 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), trap_debug32 },
568 	/* DBGDSCRext */
569 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 2), trap_debug32 },
570 	DBG_BCR_BVR_WCR_WVR(2),
571 	/* DBGDTR[RT]Xint */
572 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 0), trap_raz_wi },
573 	/* DBGDTR[RT]Xext */
574 	{ Op1( 0), CRn( 0), CRm( 3), Op2( 2), trap_raz_wi },
575 	DBG_BCR_BVR_WCR_WVR(3),
576 	DBG_BCR_BVR_WCR_WVR(4),
577 	DBG_BCR_BVR_WCR_WVR(5),
578 	/* DBGWFAR */
579 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 0), trap_raz_wi },
580 	/* DBGOSECCR */
581 	{ Op1( 0), CRn( 0), CRm( 6), Op2( 2), trap_raz_wi },
582 	DBG_BCR_BVR_WCR_WVR(6),
583 	/* DBGVCR */
584 	{ Op1( 0), CRn( 0), CRm( 7), Op2( 0), trap_debug32 },
585 	DBG_BCR_BVR_WCR_WVR(7),
586 	DBG_BCR_BVR_WCR_WVR(8),
587 	DBG_BCR_BVR_WCR_WVR(9),
588 	DBG_BCR_BVR_WCR_WVR(10),
589 	DBG_BCR_BVR_WCR_WVR(11),
590 	DBG_BCR_BVR_WCR_WVR(12),
591 	DBG_BCR_BVR_WCR_WVR(13),
592 	DBG_BCR_BVR_WCR_WVR(14),
593 	DBG_BCR_BVR_WCR_WVR(15),
594 
595 	/* DBGDRAR (32bit) */
596 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), trap_raz_wi },
597 
598 	DBGBXVR(0),
599 	/* DBGOSLAR */
600 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 4), trap_raz_wi },
601 	DBGBXVR(1),
602 	/* DBGOSLSR */
603 	{ Op1( 0), CRn( 1), CRm( 1), Op2( 4), trap_oslsr_el1 },
604 	DBGBXVR(2),
605 	DBGBXVR(3),
606 	/* DBGOSDLR */
607 	{ Op1( 0), CRn( 1), CRm( 3), Op2( 4), trap_raz_wi },
608 	DBGBXVR(4),
609 	/* DBGPRCR */
610 	{ Op1( 0), CRn( 1), CRm( 4), Op2( 4), trap_raz_wi },
611 	DBGBXVR(5),
612 	DBGBXVR(6),
613 	DBGBXVR(7),
614 	DBGBXVR(8),
615 	DBGBXVR(9),
616 	DBGBXVR(10),
617 	DBGBXVR(11),
618 	DBGBXVR(12),
619 	DBGBXVR(13),
620 	DBGBXVR(14),
621 	DBGBXVR(15),
622 
623 	/* DBGDSAR (32bit) */
624 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), trap_raz_wi },
625 
626 	/* DBGDEVID2 */
627 	{ Op1( 0), CRn( 7), CRm( 0), Op2( 7), trap_raz_wi },
628 	/* DBGDEVID1 */
629 	{ Op1( 0), CRn( 7), CRm( 1), Op2( 7), trap_raz_wi },
630 	/* DBGDEVID */
631 	{ Op1( 0), CRn( 7), CRm( 2), Op2( 7), trap_raz_wi },
632 	/* DBGCLAIMSET */
633 	{ Op1( 0), CRn( 7), CRm( 8), Op2( 6), trap_raz_wi },
634 	/* DBGCLAIMCLR */
635 	{ Op1( 0), CRn( 7), CRm( 9), Op2( 6), trap_raz_wi },
636 	/* DBGAUTHSTATUS */
637 	{ Op1( 0), CRn( 7), CRm(14), Op2( 6), trap_dbgauthstatus_el1 },
638 };
639 
640 /* Trapped cp14 64bit registers */
641 static const struct sys_reg_desc cp14_64_regs[] = {
642 	/* DBGDRAR (64bit) */
643 	{ Op1( 0), CRm( 1), .access = trap_raz_wi },
644 
645 	/* DBGDSAR (64bit) */
646 	{ Op1( 0), CRm( 2), .access = trap_raz_wi },
647 };
648 
649 /*
650  * Trapped cp15 registers. TTBR0/TTBR1 get a double encoding,
651  * depending on the way they are accessed (as a 32bit or a 64bit
652  * register).
653  */
654 static const struct sys_reg_desc cp15_regs[] = {
655 	{ Op1( 0), CRn( 1), CRm( 0), Op2( 0), access_sctlr, NULL, c1_SCTLR },
656 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
657 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 1), access_vm_reg, NULL, c2_TTBR1 },
658 	{ Op1( 0), CRn( 2), CRm( 0), Op2( 2), access_vm_reg, NULL, c2_TTBCR },
659 	{ Op1( 0), CRn( 3), CRm( 0), Op2( 0), access_vm_reg, NULL, c3_DACR },
660 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 0), access_vm_reg, NULL, c5_DFSR },
661 	{ Op1( 0), CRn( 5), CRm( 0), Op2( 1), access_vm_reg, NULL, c5_IFSR },
662 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 0), access_vm_reg, NULL, c5_ADFSR },
663 	{ Op1( 0), CRn( 5), CRm( 1), Op2( 1), access_vm_reg, NULL, c5_AIFSR },
664 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 0), access_vm_reg, NULL, c6_DFAR },
665 	{ Op1( 0), CRn( 6), CRm( 0), Op2( 2), access_vm_reg, NULL, c6_IFAR },
666 
667 	/*
668 	 * DC{C,I,CI}SW operations:
669 	 */
670 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
671 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
672 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
673 
674 	/* PMU */
675 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), trap_raz_wi },
676 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), trap_raz_wi },
677 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), trap_raz_wi },
678 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), trap_raz_wi },
679 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), trap_raz_wi },
680 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), trap_raz_wi },
681 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), trap_raz_wi },
682 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), trap_raz_wi },
683 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), trap_raz_wi },
684 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), trap_raz_wi },
685 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), trap_raz_wi },
686 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), trap_raz_wi },
687 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), trap_raz_wi },
688 
689 	{ Op1( 0), CRn(10), CRm( 2), Op2( 0), access_vm_reg, NULL, c10_PRRR },
690 	{ Op1( 0), CRn(10), CRm( 2), Op2( 1), access_vm_reg, NULL, c10_NMRR },
691 	{ Op1( 0), CRn(10), CRm( 3), Op2( 0), access_vm_reg, NULL, c10_AMAIR0 },
692 	{ Op1( 0), CRn(10), CRm( 3), Op2( 1), access_vm_reg, NULL, c10_AMAIR1 },
693 	{ Op1( 0), CRn(13), CRm( 0), Op2( 1), access_vm_reg, NULL, c13_CID },
694 };
695 
696 static const struct sys_reg_desc cp15_64_regs[] = {
697 	{ Op1( 0), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR0 },
698 	{ Op1( 1), CRn( 0), CRm( 2), Op2( 0), access_vm_reg, NULL, c2_TTBR1 },
699 };
700 
701 /* Target specific emulation tables */
702 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
703 
704 void kvm_register_target_sys_reg_table(unsigned int target,
705 				       struct kvm_sys_reg_target_table *table)
706 {
707 	target_tables[target] = table;
708 }
709 
710 /* Get specific register table for this target. */
711 static const struct sys_reg_desc *get_target_table(unsigned target,
712 						   bool mode_is_64,
713 						   size_t *num)
714 {
715 	struct kvm_sys_reg_target_table *table;
716 
717 	table = target_tables[target];
718 	if (mode_is_64) {
719 		*num = table->table64.num;
720 		return table->table64.table;
721 	} else {
722 		*num = table->table32.num;
723 		return table->table32.table;
724 	}
725 }
726 
727 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
728 					 const struct sys_reg_desc table[],
729 					 unsigned int num)
730 {
731 	unsigned int i;
732 
733 	for (i = 0; i < num; i++) {
734 		const struct sys_reg_desc *r = &table[i];
735 
736 		if (params->Op0 != r->Op0)
737 			continue;
738 		if (params->Op1 != r->Op1)
739 			continue;
740 		if (params->CRn != r->CRn)
741 			continue;
742 		if (params->CRm != r->CRm)
743 			continue;
744 		if (params->Op2 != r->Op2)
745 			continue;
746 
747 		return r;
748 	}
749 	return NULL;
750 }
751 
752 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
753 {
754 	kvm_inject_undefined(vcpu);
755 	return 1;
756 }
757 
758 /*
759  * emulate_cp --  tries to match a sys_reg access in a handling table, and
760  *                call the corresponding trap handler.
761  *
762  * @params: pointer to the descriptor of the access
763  * @table: array of trap descriptors
764  * @num: size of the trap descriptor array
765  *
766  * Return 0 if the access has been handled, and -1 if not.
767  */
768 static int emulate_cp(struct kvm_vcpu *vcpu,
769 		      const struct sys_reg_params *params,
770 		      const struct sys_reg_desc *table,
771 		      size_t num)
772 {
773 	const struct sys_reg_desc *r;
774 
775 	if (!table)
776 		return -1;	/* Not handled */
777 
778 	r = find_reg(params, table, num);
779 
780 	if (r) {
781 		/*
782 		 * Not having an accessor means that we have
783 		 * configured a trap that we don't know how to
784 		 * handle. This certainly qualifies as a gross bug
785 		 * that should be fixed right away.
786 		 */
787 		BUG_ON(!r->access);
788 
789 		if (likely(r->access(vcpu, params, r))) {
790 			/* Skip instruction, since it was emulated */
791 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
792 		}
793 
794 		/* Handled */
795 		return 0;
796 	}
797 
798 	/* Not handled */
799 	return -1;
800 }
801 
802 static void unhandled_cp_access(struct kvm_vcpu *vcpu,
803 				struct sys_reg_params *params)
804 {
805 	u8 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
806 	int cp;
807 
808 	switch(hsr_ec) {
809 	case ESR_EL2_EC_CP15_32:
810 	case ESR_EL2_EC_CP15_64:
811 		cp = 15;
812 		break;
813 	case ESR_EL2_EC_CP14_MR:
814 	case ESR_EL2_EC_CP14_64:
815 		cp = 14;
816 		break;
817 	default:
818 		WARN_ON((cp = -1));
819 	}
820 
821 	kvm_err("Unsupported guest CP%d access at: %08lx\n",
822 		cp, *vcpu_pc(vcpu));
823 	print_sys_reg_instr(params);
824 	kvm_inject_undefined(vcpu);
825 }
826 
827 /**
828  * kvm_handle_cp_64 -- handles a mrrc/mcrr trap on a guest CP15 access
829  * @vcpu: The VCPU pointer
830  * @run:  The kvm_run struct
831  */
832 static int kvm_handle_cp_64(struct kvm_vcpu *vcpu,
833 			    const struct sys_reg_desc *global,
834 			    size_t nr_global,
835 			    const struct sys_reg_desc *target_specific,
836 			    size_t nr_specific)
837 {
838 	struct sys_reg_params params;
839 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
840 	int Rt2 = (hsr >> 10) & 0xf;
841 
842 	params.is_aarch32 = true;
843 	params.is_32bit = false;
844 	params.CRm = (hsr >> 1) & 0xf;
845 	params.Rt = (hsr >> 5) & 0xf;
846 	params.is_write = ((hsr & 1) == 0);
847 
848 	params.Op0 = 0;
849 	params.Op1 = (hsr >> 16) & 0xf;
850 	params.Op2 = 0;
851 	params.CRn = 0;
852 
853 	/*
854 	 * Massive hack here. Store Rt2 in the top 32bits so we only
855 	 * have one register to deal with. As we use the same trap
856 	 * backends between AArch32 and AArch64, we get away with it.
857 	 */
858 	if (params.is_write) {
859 		u64 val = *vcpu_reg(vcpu, params.Rt);
860 		val &= 0xffffffff;
861 		val |= *vcpu_reg(vcpu, Rt2) << 32;
862 		*vcpu_reg(vcpu, params.Rt) = val;
863 	}
864 
865 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
866 		goto out;
867 	if (!emulate_cp(vcpu, &params, global, nr_global))
868 		goto out;
869 
870 	unhandled_cp_access(vcpu, &params);
871 
872 out:
873 	/* Do the opposite hack for the read side */
874 	if (!params.is_write) {
875 		u64 val = *vcpu_reg(vcpu, params.Rt);
876 		val >>= 32;
877 		*vcpu_reg(vcpu, Rt2) = val;
878 	}
879 
880 	return 1;
881 }
882 
883 /**
884  * kvm_handle_cp15_32 -- handles a mrc/mcr trap on a guest CP15 access
885  * @vcpu: The VCPU pointer
886  * @run:  The kvm_run struct
887  */
888 static int kvm_handle_cp_32(struct kvm_vcpu *vcpu,
889 			    const struct sys_reg_desc *global,
890 			    size_t nr_global,
891 			    const struct sys_reg_desc *target_specific,
892 			    size_t nr_specific)
893 {
894 	struct sys_reg_params params;
895 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
896 
897 	params.is_aarch32 = true;
898 	params.is_32bit = true;
899 	params.CRm = (hsr >> 1) & 0xf;
900 	params.Rt  = (hsr >> 5) & 0xf;
901 	params.is_write = ((hsr & 1) == 0);
902 	params.CRn = (hsr >> 10) & 0xf;
903 	params.Op0 = 0;
904 	params.Op1 = (hsr >> 14) & 0x7;
905 	params.Op2 = (hsr >> 17) & 0x7;
906 
907 	if (!emulate_cp(vcpu, &params, target_specific, nr_specific))
908 		return 1;
909 	if (!emulate_cp(vcpu, &params, global, nr_global))
910 		return 1;
911 
912 	unhandled_cp_access(vcpu, &params);
913 	return 1;
914 }
915 
916 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
917 {
918 	const struct sys_reg_desc *target_specific;
919 	size_t num;
920 
921 	target_specific = get_target_table(vcpu->arch.target, false, &num);
922 	return kvm_handle_cp_64(vcpu,
923 				cp15_64_regs, ARRAY_SIZE(cp15_64_regs),
924 				target_specific, num);
925 }
926 
927 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
928 {
929 	const struct sys_reg_desc *target_specific;
930 	size_t num;
931 
932 	target_specific = get_target_table(vcpu->arch.target, false, &num);
933 	return kvm_handle_cp_32(vcpu,
934 				cp15_regs, ARRAY_SIZE(cp15_regs),
935 				target_specific, num);
936 }
937 
938 int kvm_handle_cp14_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
939 {
940 	return kvm_handle_cp_64(vcpu,
941 				cp14_64_regs, ARRAY_SIZE(cp14_64_regs),
942 				NULL, 0);
943 }
944 
945 int kvm_handle_cp14_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
946 {
947 	return kvm_handle_cp_32(vcpu,
948 				cp14_regs, ARRAY_SIZE(cp14_regs),
949 				NULL, 0);
950 }
951 
952 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
953 			   const struct sys_reg_params *params)
954 {
955 	size_t num;
956 	const struct sys_reg_desc *table, *r;
957 
958 	table = get_target_table(vcpu->arch.target, true, &num);
959 
960 	/* Search target-specific then generic table. */
961 	r = find_reg(params, table, num);
962 	if (!r)
963 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
964 
965 	if (likely(r)) {
966 		/*
967 		 * Not having an accessor means that we have
968 		 * configured a trap that we don't know how to
969 		 * handle. This certainly qualifies as a gross bug
970 		 * that should be fixed right away.
971 		 */
972 		BUG_ON(!r->access);
973 
974 		if (likely(r->access(vcpu, params, r))) {
975 			/* Skip instruction, since it was emulated */
976 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
977 			return 1;
978 		}
979 		/* If access function fails, it should complain. */
980 	} else {
981 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
982 			*vcpu_pc(vcpu));
983 		print_sys_reg_instr(params);
984 	}
985 	kvm_inject_undefined(vcpu);
986 	return 1;
987 }
988 
989 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
990 			      const struct sys_reg_desc *table, size_t num)
991 {
992 	unsigned long i;
993 
994 	for (i = 0; i < num; i++)
995 		if (table[i].reset)
996 			table[i].reset(vcpu, &table[i]);
997 }
998 
999 /**
1000  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
1001  * @vcpu: The VCPU pointer
1002  * @run:  The kvm_run struct
1003  */
1004 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
1005 {
1006 	struct sys_reg_params params;
1007 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
1008 
1009 	params.is_aarch32 = false;
1010 	params.is_32bit = false;
1011 	params.Op0 = (esr >> 20) & 3;
1012 	params.Op1 = (esr >> 14) & 0x7;
1013 	params.CRn = (esr >> 10) & 0xf;
1014 	params.CRm = (esr >> 1) & 0xf;
1015 	params.Op2 = (esr >> 17) & 0x7;
1016 	params.Rt = (esr >> 5) & 0x1f;
1017 	params.is_write = !(esr & 1);
1018 
1019 	return emulate_sys_reg(vcpu, &params);
1020 }
1021 
1022 /******************************************************************************
1023  * Userspace API
1024  *****************************************************************************/
1025 
1026 static bool index_to_params(u64 id, struct sys_reg_params *params)
1027 {
1028 	switch (id & KVM_REG_SIZE_MASK) {
1029 	case KVM_REG_SIZE_U64:
1030 		/* Any unused index bits means it's not valid. */
1031 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
1032 			      | KVM_REG_ARM_COPROC_MASK
1033 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
1034 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
1035 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
1036 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
1037 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
1038 			return false;
1039 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
1040 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
1041 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
1042 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
1043 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
1044 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
1045 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
1046 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
1047 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
1048 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
1049 		return true;
1050 	default:
1051 		return false;
1052 	}
1053 }
1054 
1055 /* Decode an index value, and find the sys_reg_desc entry. */
1056 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
1057 						    u64 id)
1058 {
1059 	size_t num;
1060 	const struct sys_reg_desc *table, *r;
1061 	struct sys_reg_params params;
1062 
1063 	/* We only do sys_reg for now. */
1064 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
1065 		return NULL;
1066 
1067 	if (!index_to_params(id, &params))
1068 		return NULL;
1069 
1070 	table = get_target_table(vcpu->arch.target, true, &num);
1071 	r = find_reg(&params, table, num);
1072 	if (!r)
1073 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1074 
1075 	/* Not saved in the sys_reg array? */
1076 	if (r && !r->reg)
1077 		r = NULL;
1078 
1079 	return r;
1080 }
1081 
1082 /*
1083  * These are the invariant sys_reg registers: we let the guest see the
1084  * host versions of these, so they're part of the guest state.
1085  *
1086  * A future CPU may provide a mechanism to present different values to
1087  * the guest, or a future kvm may trap them.
1088  */
1089 
1090 #define FUNCTION_INVARIANT(reg)						\
1091 	static void get_##reg(struct kvm_vcpu *v,			\
1092 			      const struct sys_reg_desc *r)		\
1093 	{								\
1094 		u64 val;						\
1095 									\
1096 		asm volatile("mrs %0, " __stringify(reg) "\n"		\
1097 			     : "=r" (val));				\
1098 		((struct sys_reg_desc *)r)->val = val;			\
1099 	}
1100 
1101 FUNCTION_INVARIANT(midr_el1)
1102 FUNCTION_INVARIANT(ctr_el0)
1103 FUNCTION_INVARIANT(revidr_el1)
1104 FUNCTION_INVARIANT(id_pfr0_el1)
1105 FUNCTION_INVARIANT(id_pfr1_el1)
1106 FUNCTION_INVARIANT(id_dfr0_el1)
1107 FUNCTION_INVARIANT(id_afr0_el1)
1108 FUNCTION_INVARIANT(id_mmfr0_el1)
1109 FUNCTION_INVARIANT(id_mmfr1_el1)
1110 FUNCTION_INVARIANT(id_mmfr2_el1)
1111 FUNCTION_INVARIANT(id_mmfr3_el1)
1112 FUNCTION_INVARIANT(id_isar0_el1)
1113 FUNCTION_INVARIANT(id_isar1_el1)
1114 FUNCTION_INVARIANT(id_isar2_el1)
1115 FUNCTION_INVARIANT(id_isar3_el1)
1116 FUNCTION_INVARIANT(id_isar4_el1)
1117 FUNCTION_INVARIANT(id_isar5_el1)
1118 FUNCTION_INVARIANT(clidr_el1)
1119 FUNCTION_INVARIANT(aidr_el1)
1120 
1121 /* ->val is filled in by kvm_sys_reg_table_init() */
1122 static struct sys_reg_desc invariant_sys_regs[] = {
1123 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
1124 	  NULL, get_midr_el1 },
1125 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
1126 	  NULL, get_revidr_el1 },
1127 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
1128 	  NULL, get_id_pfr0_el1 },
1129 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
1130 	  NULL, get_id_pfr1_el1 },
1131 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
1132 	  NULL, get_id_dfr0_el1 },
1133 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
1134 	  NULL, get_id_afr0_el1 },
1135 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
1136 	  NULL, get_id_mmfr0_el1 },
1137 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
1138 	  NULL, get_id_mmfr1_el1 },
1139 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
1140 	  NULL, get_id_mmfr2_el1 },
1141 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
1142 	  NULL, get_id_mmfr3_el1 },
1143 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
1144 	  NULL, get_id_isar0_el1 },
1145 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
1146 	  NULL, get_id_isar1_el1 },
1147 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
1148 	  NULL, get_id_isar2_el1 },
1149 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
1150 	  NULL, get_id_isar3_el1 },
1151 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
1152 	  NULL, get_id_isar4_el1 },
1153 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
1154 	  NULL, get_id_isar5_el1 },
1155 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
1156 	  NULL, get_clidr_el1 },
1157 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
1158 	  NULL, get_aidr_el1 },
1159 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
1160 	  NULL, get_ctr_el0 },
1161 };
1162 
1163 static int reg_from_user(u64 *val, const void __user *uaddr, u64 id)
1164 {
1165 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
1166 		return -EFAULT;
1167 	return 0;
1168 }
1169 
1170 static int reg_to_user(void __user *uaddr, const u64 *val, u64 id)
1171 {
1172 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
1173 		return -EFAULT;
1174 	return 0;
1175 }
1176 
1177 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
1178 {
1179 	struct sys_reg_params params;
1180 	const struct sys_reg_desc *r;
1181 
1182 	if (!index_to_params(id, &params))
1183 		return -ENOENT;
1184 
1185 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1186 	if (!r)
1187 		return -ENOENT;
1188 
1189 	return reg_to_user(uaddr, &r->val, id);
1190 }
1191 
1192 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
1193 {
1194 	struct sys_reg_params params;
1195 	const struct sys_reg_desc *r;
1196 	int err;
1197 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
1198 
1199 	if (!index_to_params(id, &params))
1200 		return -ENOENT;
1201 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
1202 	if (!r)
1203 		return -ENOENT;
1204 
1205 	err = reg_from_user(&val, uaddr, id);
1206 	if (err)
1207 		return err;
1208 
1209 	/* This is what we mean by invariant: you can't change it. */
1210 	if (r->val != val)
1211 		return -EINVAL;
1212 
1213 	return 0;
1214 }
1215 
1216 static bool is_valid_cache(u32 val)
1217 {
1218 	u32 level, ctype;
1219 
1220 	if (val >= CSSELR_MAX)
1221 		return -ENOENT;
1222 
1223 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
1224 	level = (val >> 1);
1225 	ctype = (cache_levels >> (level * 3)) & 7;
1226 
1227 	switch (ctype) {
1228 	case 0: /* No cache */
1229 		return false;
1230 	case 1: /* Instruction cache only */
1231 		return (val & 1);
1232 	case 2: /* Data cache only */
1233 	case 4: /* Unified cache */
1234 		return !(val & 1);
1235 	case 3: /* Separate instruction and data caches */
1236 		return true;
1237 	default: /* Reserved: we can't know instruction or data. */
1238 		return false;
1239 	}
1240 }
1241 
1242 static int demux_c15_get(u64 id, void __user *uaddr)
1243 {
1244 	u32 val;
1245 	u32 __user *uval = uaddr;
1246 
1247 	/* Fail if we have unknown bits set. */
1248 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1249 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1250 		return -ENOENT;
1251 
1252 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1253 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1254 		if (KVM_REG_SIZE(id) != 4)
1255 			return -ENOENT;
1256 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1257 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1258 		if (!is_valid_cache(val))
1259 			return -ENOENT;
1260 
1261 		return put_user(get_ccsidr(val), uval);
1262 	default:
1263 		return -ENOENT;
1264 	}
1265 }
1266 
1267 static int demux_c15_set(u64 id, void __user *uaddr)
1268 {
1269 	u32 val, newval;
1270 	u32 __user *uval = uaddr;
1271 
1272 	/* Fail if we have unknown bits set. */
1273 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
1274 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
1275 		return -ENOENT;
1276 
1277 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
1278 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
1279 		if (KVM_REG_SIZE(id) != 4)
1280 			return -ENOENT;
1281 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
1282 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
1283 		if (!is_valid_cache(val))
1284 			return -ENOENT;
1285 
1286 		if (get_user(newval, uval))
1287 			return -EFAULT;
1288 
1289 		/* This is also invariant: you can't change it. */
1290 		if (newval != get_ccsidr(val))
1291 			return -EINVAL;
1292 		return 0;
1293 	default:
1294 		return -ENOENT;
1295 	}
1296 }
1297 
1298 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1299 {
1300 	const struct sys_reg_desc *r;
1301 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1302 
1303 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1304 		return demux_c15_get(reg->id, uaddr);
1305 
1306 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1307 		return -ENOENT;
1308 
1309 	r = index_to_sys_reg_desc(vcpu, reg->id);
1310 	if (!r)
1311 		return get_invariant_sys_reg(reg->id, uaddr);
1312 
1313 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
1314 }
1315 
1316 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
1317 {
1318 	const struct sys_reg_desc *r;
1319 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
1320 
1321 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
1322 		return demux_c15_set(reg->id, uaddr);
1323 
1324 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
1325 		return -ENOENT;
1326 
1327 	r = index_to_sys_reg_desc(vcpu, reg->id);
1328 	if (!r)
1329 		return set_invariant_sys_reg(reg->id, uaddr);
1330 
1331 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
1332 }
1333 
1334 static unsigned int num_demux_regs(void)
1335 {
1336 	unsigned int i, count = 0;
1337 
1338 	for (i = 0; i < CSSELR_MAX; i++)
1339 		if (is_valid_cache(i))
1340 			count++;
1341 
1342 	return count;
1343 }
1344 
1345 static int write_demux_regids(u64 __user *uindices)
1346 {
1347 	u64 val = KVM_REG_ARM64 | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
1348 	unsigned int i;
1349 
1350 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
1351 	for (i = 0; i < CSSELR_MAX; i++) {
1352 		if (!is_valid_cache(i))
1353 			continue;
1354 		if (put_user(val | i, uindices))
1355 			return -EFAULT;
1356 		uindices++;
1357 	}
1358 	return 0;
1359 }
1360 
1361 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
1362 {
1363 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
1364 		KVM_REG_ARM64_SYSREG |
1365 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
1366 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
1367 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
1368 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
1369 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
1370 }
1371 
1372 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
1373 {
1374 	if (!*uind)
1375 		return true;
1376 
1377 	if (put_user(sys_reg_to_index(reg), *uind))
1378 		return false;
1379 
1380 	(*uind)++;
1381 	return true;
1382 }
1383 
1384 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
1385 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
1386 {
1387 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
1388 	unsigned int total = 0;
1389 	size_t num;
1390 
1391 	/* We check for duplicates here, to allow arch-specific overrides. */
1392 	i1 = get_target_table(vcpu->arch.target, true, &num);
1393 	end1 = i1 + num;
1394 	i2 = sys_reg_descs;
1395 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
1396 
1397 	BUG_ON(i1 == end1 || i2 == end2);
1398 
1399 	/* Walk carefully, as both tables may refer to the same register. */
1400 	while (i1 || i2) {
1401 		int cmp = cmp_sys_reg(i1, i2);
1402 		/* target-specific overrides generic entry. */
1403 		if (cmp <= 0) {
1404 			/* Ignore registers we trap but don't save. */
1405 			if (i1->reg) {
1406 				if (!copy_reg_to_user(i1, &uind))
1407 					return -EFAULT;
1408 				total++;
1409 			}
1410 		} else {
1411 			/* Ignore registers we trap but don't save. */
1412 			if (i2->reg) {
1413 				if (!copy_reg_to_user(i2, &uind))
1414 					return -EFAULT;
1415 				total++;
1416 			}
1417 		}
1418 
1419 		if (cmp <= 0 && ++i1 == end1)
1420 			i1 = NULL;
1421 		if (cmp >= 0 && ++i2 == end2)
1422 			i2 = NULL;
1423 	}
1424 	return total;
1425 }
1426 
1427 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
1428 {
1429 	return ARRAY_SIZE(invariant_sys_regs)
1430 		+ num_demux_regs()
1431 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
1432 }
1433 
1434 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
1435 {
1436 	unsigned int i;
1437 	int err;
1438 
1439 	/* Then give them all the invariant registers' indices. */
1440 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
1441 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
1442 			return -EFAULT;
1443 		uindices++;
1444 	}
1445 
1446 	err = walk_sys_regs(vcpu, uindices);
1447 	if (err < 0)
1448 		return err;
1449 	uindices += err;
1450 
1451 	return write_demux_regids(uindices);
1452 }
1453 
1454 static int check_sysreg_table(const struct sys_reg_desc *table, unsigned int n)
1455 {
1456 	unsigned int i;
1457 
1458 	for (i = 1; i < n; i++) {
1459 		if (cmp_sys_reg(&table[i-1], &table[i]) >= 0) {
1460 			kvm_err("sys_reg table %p out of order (%d)\n", table, i - 1);
1461 			return 1;
1462 		}
1463 	}
1464 
1465 	return 0;
1466 }
1467 
1468 void kvm_sys_reg_table_init(void)
1469 {
1470 	unsigned int i;
1471 	struct sys_reg_desc clidr;
1472 
1473 	/* Make sure tables are unique and in order. */
1474 	BUG_ON(check_sysreg_table(sys_reg_descs, ARRAY_SIZE(sys_reg_descs)));
1475 	BUG_ON(check_sysreg_table(cp14_regs, ARRAY_SIZE(cp14_regs)));
1476 	BUG_ON(check_sysreg_table(cp14_64_regs, ARRAY_SIZE(cp14_64_regs)));
1477 	BUG_ON(check_sysreg_table(cp15_regs, ARRAY_SIZE(cp15_regs)));
1478 	BUG_ON(check_sysreg_table(cp15_64_regs, ARRAY_SIZE(cp15_64_regs)));
1479 	BUG_ON(check_sysreg_table(invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs)));
1480 
1481 	/* We abuse the reset function to overwrite the table itself. */
1482 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
1483 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
1484 
1485 	/*
1486 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
1487 	 *
1488 	 *   If software reads the Cache Type fields from Ctype1
1489 	 *   upwards, once it has seen a value of 0b000, no caches
1490 	 *   exist at further-out levels of the hierarchy. So, for
1491 	 *   example, if Ctype3 is the first Cache Type field with a
1492 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
1493 	 *   ignored.
1494 	 */
1495 	get_clidr_el1(NULL, &clidr); /* Ugly... */
1496 	cache_levels = clidr.val;
1497 	for (i = 0; i < 7; i++)
1498 		if (((cache_levels >> (i*3)) & 7) == 0)
1499 			break;
1500 	/* Clear all higher bits. */
1501 	cache_levels &= (1 << (i*3))-1;
1502 }
1503 
1504 /**
1505  * kvm_reset_sys_regs - sets system registers to reset value
1506  * @vcpu: The VCPU pointer
1507  *
1508  * This function finds the right table above and sets the registers on the
1509  * virtual CPU struct to their architecturally defined reset values.
1510  */
1511 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
1512 {
1513 	size_t num;
1514 	const struct sys_reg_desc *table;
1515 
1516 	/* Catch someone adding a register without putting in reset entry. */
1517 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
1518 
1519 	/* Generic chip reset first (so target could override). */
1520 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1521 
1522 	table = get_target_table(vcpu->arch.target, true, &num);
1523 	reset_sys_reg_descs(vcpu, table, num);
1524 
1525 	for (num = 1; num < NR_SYS_REGS; num++)
1526 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
1527 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
1528 }
1529