xref: /openbmc/linux/arch/arm64/kvm/sys_regs.c (revision e23feb16)
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/cacheflush.h>
31 #include <asm/cputype.h>
32 #include <trace/events/kvm.h>
33 
34 #include "sys_regs.h"
35 
36 /*
37  * All of this file is extremly similar to the ARM coproc.c, but the
38  * types are different. My gut feeling is that it should be pretty
39  * easy to merge, but that would be an ABI breakage -- again. VFP
40  * would also need to be abstracted.
41  *
42  * For AArch32, we only take care of what is being trapped. Anything
43  * that has to do with init and userspace access has to go via the
44  * 64bit interface.
45  */
46 
47 /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */
48 static u32 cache_levels;
49 
50 /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */
51 #define CSSELR_MAX 12
52 
53 /* Which cache CCSIDR represents depends on CSSELR value. */
54 static u32 get_ccsidr(u32 csselr)
55 {
56 	u32 ccsidr;
57 
58 	/* Make sure noone else changes CSSELR during this! */
59 	local_irq_disable();
60 	/* Put value into CSSELR */
61 	asm volatile("msr csselr_el1, %x0" : : "r" (csselr));
62 	isb();
63 	/* Read result out of CCSIDR */
64 	asm volatile("mrs %0, ccsidr_el1" : "=r" (ccsidr));
65 	local_irq_enable();
66 
67 	return ccsidr;
68 }
69 
70 static void do_dc_cisw(u32 val)
71 {
72 	asm volatile("dc cisw, %x0" : : "r" (val));
73 	dsb();
74 }
75 
76 static void do_dc_csw(u32 val)
77 {
78 	asm volatile("dc csw, %x0" : : "r" (val));
79 	dsb();
80 }
81 
82 /* See note at ARM ARM B1.14.4 */
83 static bool access_dcsw(struct kvm_vcpu *vcpu,
84 			const struct sys_reg_params *p,
85 			const struct sys_reg_desc *r)
86 {
87 	unsigned long val;
88 	int cpu;
89 
90 	if (!p->is_write)
91 		return read_from_write_only(vcpu, p);
92 
93 	cpu = get_cpu();
94 
95 	cpumask_setall(&vcpu->arch.require_dcache_flush);
96 	cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
97 
98 	/* If we were already preempted, take the long way around */
99 	if (cpu != vcpu->arch.last_pcpu) {
100 		flush_cache_all();
101 		goto done;
102 	}
103 
104 	val = *vcpu_reg(vcpu, p->Rt);
105 
106 	switch (p->CRm) {
107 	case 6:			/* Upgrade DCISW to DCCISW, as per HCR.SWIO */
108 	case 14:		/* DCCISW */
109 		do_dc_cisw(val);
110 		break;
111 
112 	case 10:		/* DCCSW */
113 		do_dc_csw(val);
114 		break;
115 	}
116 
117 done:
118 	put_cpu();
119 
120 	return true;
121 }
122 
123 /*
124  * We could trap ID_DFR0 and tell the guest we don't support performance
125  * monitoring.  Unfortunately the patch to make the kernel check ID_DFR0 was
126  * NAKed, so it will read the PMCR anyway.
127  *
128  * Therefore we tell the guest we have 0 counters.  Unfortunately, we
129  * must always support PMCCNTR (the cycle counter): we just RAZ/WI for
130  * all PM registers, which doesn't crash the guest kernel at least.
131  */
132 static bool pm_fake(struct kvm_vcpu *vcpu,
133 		    const struct sys_reg_params *p,
134 		    const struct sys_reg_desc *r)
135 {
136 	if (p->is_write)
137 		return ignore_write(vcpu, p);
138 	else
139 		return read_zero(vcpu, p);
140 }
141 
142 static void reset_amair_el1(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
143 {
144 	u64 amair;
145 
146 	asm volatile("mrs %0, amair_el1\n" : "=r" (amair));
147 	vcpu_sys_reg(vcpu, AMAIR_EL1) = amair;
148 }
149 
150 static void reset_mpidr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
151 {
152 	/*
153 	 * Simply map the vcpu_id into the Aff0 field of the MPIDR.
154 	 */
155 	vcpu_sys_reg(vcpu, MPIDR_EL1) = (1UL << 31) | (vcpu->vcpu_id & 0xff);
156 }
157 
158 /*
159  * Architected system registers.
160  * Important: Must be sorted ascending by Op0, Op1, CRn, CRm, Op2
161  */
162 static const struct sys_reg_desc sys_reg_descs[] = {
163 	/* DC ISW */
164 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b0110), Op2(0b010),
165 	  access_dcsw },
166 	/* DC CSW */
167 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1010), Op2(0b010),
168 	  access_dcsw },
169 	/* DC CISW */
170 	{ Op0(0b01), Op1(0b000), CRn(0b0111), CRm(0b1110), Op2(0b010),
171 	  access_dcsw },
172 
173 	/* TEECR32_EL1 */
174 	{ Op0(0b10), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
175 	  NULL, reset_val, TEECR32_EL1, 0 },
176 	/* TEEHBR32_EL1 */
177 	{ Op0(0b10), Op1(0b010), CRn(0b0001), CRm(0b0000), Op2(0b000),
178 	  NULL, reset_val, TEEHBR32_EL1, 0 },
179 	/* DBGVCR32_EL2 */
180 	{ Op0(0b10), Op1(0b100), CRn(0b0000), CRm(0b0111), Op2(0b000),
181 	  NULL, reset_val, DBGVCR32_EL2, 0 },
182 
183 	/* MPIDR_EL1 */
184 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b101),
185 	  NULL, reset_mpidr, MPIDR_EL1 },
186 	/* SCTLR_EL1 */
187 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b000),
188 	  NULL, reset_val, SCTLR_EL1, 0x00C50078 },
189 	/* CPACR_EL1 */
190 	{ Op0(0b11), Op1(0b000), CRn(0b0001), CRm(0b0000), Op2(0b010),
191 	  NULL, reset_val, CPACR_EL1, 0 },
192 	/* TTBR0_EL1 */
193 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b000),
194 	  NULL, reset_unknown, TTBR0_EL1 },
195 	/* TTBR1_EL1 */
196 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b001),
197 	  NULL, reset_unknown, TTBR1_EL1 },
198 	/* TCR_EL1 */
199 	{ Op0(0b11), Op1(0b000), CRn(0b0010), CRm(0b0000), Op2(0b010),
200 	  NULL, reset_val, TCR_EL1, 0 },
201 
202 	/* AFSR0_EL1 */
203 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b000),
204 	  NULL, reset_unknown, AFSR0_EL1 },
205 	/* AFSR1_EL1 */
206 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0001), Op2(0b001),
207 	  NULL, reset_unknown, AFSR1_EL1 },
208 	/* ESR_EL1 */
209 	{ Op0(0b11), Op1(0b000), CRn(0b0101), CRm(0b0010), Op2(0b000),
210 	  NULL, reset_unknown, ESR_EL1 },
211 	/* FAR_EL1 */
212 	{ Op0(0b11), Op1(0b000), CRn(0b0110), CRm(0b0000), Op2(0b000),
213 	  NULL, reset_unknown, FAR_EL1 },
214 	/* PAR_EL1 */
215 	{ Op0(0b11), Op1(0b000), CRn(0b0111), CRm(0b0100), Op2(0b000),
216 	  NULL, reset_unknown, PAR_EL1 },
217 
218 	/* PMINTENSET_EL1 */
219 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b001),
220 	  pm_fake },
221 	/* PMINTENCLR_EL1 */
222 	{ Op0(0b11), Op1(0b000), CRn(0b1001), CRm(0b1110), Op2(0b010),
223 	  pm_fake },
224 
225 	/* MAIR_EL1 */
226 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0010), Op2(0b000),
227 	  NULL, reset_unknown, MAIR_EL1 },
228 	/* AMAIR_EL1 */
229 	{ Op0(0b11), Op1(0b000), CRn(0b1010), CRm(0b0011), Op2(0b000),
230 	  NULL, reset_amair_el1, AMAIR_EL1 },
231 
232 	/* VBAR_EL1 */
233 	{ Op0(0b11), Op1(0b000), CRn(0b1100), CRm(0b0000), Op2(0b000),
234 	  NULL, reset_val, VBAR_EL1, 0 },
235 	/* CONTEXTIDR_EL1 */
236 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b001),
237 	  NULL, reset_val, CONTEXTIDR_EL1, 0 },
238 	/* TPIDR_EL1 */
239 	{ Op0(0b11), Op1(0b000), CRn(0b1101), CRm(0b0000), Op2(0b100),
240 	  NULL, reset_unknown, TPIDR_EL1 },
241 
242 	/* CNTKCTL_EL1 */
243 	{ Op0(0b11), Op1(0b000), CRn(0b1110), CRm(0b0001), Op2(0b000),
244 	  NULL, reset_val, CNTKCTL_EL1, 0},
245 
246 	/* CSSELR_EL1 */
247 	{ Op0(0b11), Op1(0b010), CRn(0b0000), CRm(0b0000), Op2(0b000),
248 	  NULL, reset_unknown, CSSELR_EL1 },
249 
250 	/* PMCR_EL0 */
251 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b000),
252 	  pm_fake },
253 	/* PMCNTENSET_EL0 */
254 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b001),
255 	  pm_fake },
256 	/* PMCNTENCLR_EL0 */
257 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b010),
258 	  pm_fake },
259 	/* PMOVSCLR_EL0 */
260 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b011),
261 	  pm_fake },
262 	/* PMSWINC_EL0 */
263 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b100),
264 	  pm_fake },
265 	/* PMSELR_EL0 */
266 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b101),
267 	  pm_fake },
268 	/* PMCEID0_EL0 */
269 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b110),
270 	  pm_fake },
271 	/* PMCEID1_EL0 */
272 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1100), Op2(0b111),
273 	  pm_fake },
274 	/* PMCCNTR_EL0 */
275 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b000),
276 	  pm_fake },
277 	/* PMXEVTYPER_EL0 */
278 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b001),
279 	  pm_fake },
280 	/* PMXEVCNTR_EL0 */
281 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1101), Op2(0b010),
282 	  pm_fake },
283 	/* PMUSERENR_EL0 */
284 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b000),
285 	  pm_fake },
286 	/* PMOVSSET_EL0 */
287 	{ Op0(0b11), Op1(0b011), CRn(0b1001), CRm(0b1110), Op2(0b011),
288 	  pm_fake },
289 
290 	/* TPIDR_EL0 */
291 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b010),
292 	  NULL, reset_unknown, TPIDR_EL0 },
293 	/* TPIDRRO_EL0 */
294 	{ Op0(0b11), Op1(0b011), CRn(0b1101), CRm(0b0000), Op2(0b011),
295 	  NULL, reset_unknown, TPIDRRO_EL0 },
296 
297 	/* DACR32_EL2 */
298 	{ Op0(0b11), Op1(0b100), CRn(0b0011), CRm(0b0000), Op2(0b000),
299 	  NULL, reset_unknown, DACR32_EL2 },
300 	/* IFSR32_EL2 */
301 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0000), Op2(0b001),
302 	  NULL, reset_unknown, IFSR32_EL2 },
303 	/* FPEXC32_EL2 */
304 	{ Op0(0b11), Op1(0b100), CRn(0b0101), CRm(0b0011), Op2(0b000),
305 	  NULL, reset_val, FPEXC32_EL2, 0x70 },
306 };
307 
308 /* Trapped cp15 registers */
309 static const struct sys_reg_desc cp15_regs[] = {
310 	/*
311 	 * DC{C,I,CI}SW operations:
312 	 */
313 	{ Op1( 0), CRn( 7), CRm( 6), Op2( 2), access_dcsw },
314 	{ Op1( 0), CRn( 7), CRm(10), Op2( 2), access_dcsw },
315 	{ Op1( 0), CRn( 7), CRm(14), Op2( 2), access_dcsw },
316 	{ Op1( 0), CRn( 9), CRm(12), Op2( 0), pm_fake },
317 	{ Op1( 0), CRn( 9), CRm(12), Op2( 1), pm_fake },
318 	{ Op1( 0), CRn( 9), CRm(12), Op2( 2), pm_fake },
319 	{ Op1( 0), CRn( 9), CRm(12), Op2( 3), pm_fake },
320 	{ Op1( 0), CRn( 9), CRm(12), Op2( 5), pm_fake },
321 	{ Op1( 0), CRn( 9), CRm(12), Op2( 6), pm_fake },
322 	{ Op1( 0), CRn( 9), CRm(12), Op2( 7), pm_fake },
323 	{ Op1( 0), CRn( 9), CRm(13), Op2( 0), pm_fake },
324 	{ Op1( 0), CRn( 9), CRm(13), Op2( 1), pm_fake },
325 	{ Op1( 0), CRn( 9), CRm(13), Op2( 2), pm_fake },
326 	{ Op1( 0), CRn( 9), CRm(14), Op2( 0), pm_fake },
327 	{ Op1( 0), CRn( 9), CRm(14), Op2( 1), pm_fake },
328 	{ Op1( 0), CRn( 9), CRm(14), Op2( 2), pm_fake },
329 };
330 
331 /* Target specific emulation tables */
332 static struct kvm_sys_reg_target_table *target_tables[KVM_ARM_NUM_TARGETS];
333 
334 void kvm_register_target_sys_reg_table(unsigned int target,
335 				       struct kvm_sys_reg_target_table *table)
336 {
337 	target_tables[target] = table;
338 }
339 
340 /* Get specific register table for this target. */
341 static const struct sys_reg_desc *get_target_table(unsigned target,
342 						   bool mode_is_64,
343 						   size_t *num)
344 {
345 	struct kvm_sys_reg_target_table *table;
346 
347 	table = target_tables[target];
348 	if (mode_is_64) {
349 		*num = table->table64.num;
350 		return table->table64.table;
351 	} else {
352 		*num = table->table32.num;
353 		return table->table32.table;
354 	}
355 }
356 
357 static const struct sys_reg_desc *find_reg(const struct sys_reg_params *params,
358 					 const struct sys_reg_desc table[],
359 					 unsigned int num)
360 {
361 	unsigned int i;
362 
363 	for (i = 0; i < num; i++) {
364 		const struct sys_reg_desc *r = &table[i];
365 
366 		if (params->Op0 != r->Op0)
367 			continue;
368 		if (params->Op1 != r->Op1)
369 			continue;
370 		if (params->CRn != r->CRn)
371 			continue;
372 		if (params->CRm != r->CRm)
373 			continue;
374 		if (params->Op2 != r->Op2)
375 			continue;
376 
377 		return r;
378 	}
379 	return NULL;
380 }
381 
382 int kvm_handle_cp14_load_store(struct kvm_vcpu *vcpu, struct kvm_run *run)
383 {
384 	kvm_inject_undefined(vcpu);
385 	return 1;
386 }
387 
388 int kvm_handle_cp14_access(struct kvm_vcpu *vcpu, struct kvm_run *run)
389 {
390 	kvm_inject_undefined(vcpu);
391 	return 1;
392 }
393 
394 static void emulate_cp15(struct kvm_vcpu *vcpu,
395 			 const struct sys_reg_params *params)
396 {
397 	size_t num;
398 	const struct sys_reg_desc *table, *r;
399 
400 	table = get_target_table(vcpu->arch.target, false, &num);
401 
402 	/* Search target-specific then generic table. */
403 	r = find_reg(params, table, num);
404 	if (!r)
405 		r = find_reg(params, cp15_regs, ARRAY_SIZE(cp15_regs));
406 
407 	if (likely(r)) {
408 		/*
409 		 * Not having an accessor means that we have
410 		 * configured a trap that we don't know how to
411 		 * handle. This certainly qualifies as a gross bug
412 		 * that should be fixed right away.
413 		 */
414 		BUG_ON(!r->access);
415 
416 		if (likely(r->access(vcpu, params, r))) {
417 			/* Skip instruction, since it was emulated */
418 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
419 			return;
420 		}
421 		/* If access function fails, it should complain. */
422 	}
423 
424 	kvm_err("Unsupported guest CP15 access at: %08lx\n", *vcpu_pc(vcpu));
425 	print_sys_reg_instr(params);
426 	kvm_inject_undefined(vcpu);
427 }
428 
429 /**
430  * kvm_handle_cp15_64 -- handles a mrrc/mcrr trap on a guest CP15 access
431  * @vcpu: The VCPU pointer
432  * @run:  The kvm_run struct
433  */
434 int kvm_handle_cp15_64(struct kvm_vcpu *vcpu, struct kvm_run *run)
435 {
436 	struct sys_reg_params params;
437 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
438 	int Rt2 = (hsr >> 10) & 0xf;
439 
440 	params.CRm = (hsr >> 1) & 0xf;
441 	params.Rt = (hsr >> 5) & 0xf;
442 	params.is_write = ((hsr & 1) == 0);
443 
444 	params.Op0 = 0;
445 	params.Op1 = (hsr >> 16) & 0xf;
446 	params.Op2 = 0;
447 	params.CRn = 0;
448 
449 	/*
450 	 * Massive hack here. Store Rt2 in the top 32bits so we only
451 	 * have one register to deal with. As we use the same trap
452 	 * backends between AArch32 and AArch64, we get away with it.
453 	 */
454 	if (params.is_write) {
455 		u64 val = *vcpu_reg(vcpu, params.Rt);
456 		val &= 0xffffffff;
457 		val |= *vcpu_reg(vcpu, Rt2) << 32;
458 		*vcpu_reg(vcpu, params.Rt) = val;
459 	}
460 
461 	emulate_cp15(vcpu, &params);
462 
463 	/* Do the opposite hack for the read side */
464 	if (!params.is_write) {
465 		u64 val = *vcpu_reg(vcpu, params.Rt);
466 		val >>= 32;
467 		*vcpu_reg(vcpu, Rt2) = val;
468 	}
469 
470 	return 1;
471 }
472 
473 /**
474  * kvm_handle_cp15_32 -- handles a mrc/mcr trap on a guest CP15 access
475  * @vcpu: The VCPU pointer
476  * @run:  The kvm_run struct
477  */
478 int kvm_handle_cp15_32(struct kvm_vcpu *vcpu, struct kvm_run *run)
479 {
480 	struct sys_reg_params params;
481 	u32 hsr = kvm_vcpu_get_hsr(vcpu);
482 
483 	params.CRm = (hsr >> 1) & 0xf;
484 	params.Rt  = (hsr >> 5) & 0xf;
485 	params.is_write = ((hsr & 1) == 0);
486 	params.CRn = (hsr >> 10) & 0xf;
487 	params.Op0 = 0;
488 	params.Op1 = (hsr >> 14) & 0x7;
489 	params.Op2 = (hsr >> 17) & 0x7;
490 
491 	emulate_cp15(vcpu, &params);
492 	return 1;
493 }
494 
495 static int emulate_sys_reg(struct kvm_vcpu *vcpu,
496 			   const struct sys_reg_params *params)
497 {
498 	size_t num;
499 	const struct sys_reg_desc *table, *r;
500 
501 	table = get_target_table(vcpu->arch.target, true, &num);
502 
503 	/* Search target-specific then generic table. */
504 	r = find_reg(params, table, num);
505 	if (!r)
506 		r = find_reg(params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
507 
508 	if (likely(r)) {
509 		/*
510 		 * Not having an accessor means that we have
511 		 * configured a trap that we don't know how to
512 		 * handle. This certainly qualifies as a gross bug
513 		 * that should be fixed right away.
514 		 */
515 		BUG_ON(!r->access);
516 
517 		if (likely(r->access(vcpu, params, r))) {
518 			/* Skip instruction, since it was emulated */
519 			kvm_skip_instr(vcpu, kvm_vcpu_trap_il_is32bit(vcpu));
520 			return 1;
521 		}
522 		/* If access function fails, it should complain. */
523 	} else {
524 		kvm_err("Unsupported guest sys_reg access at: %lx\n",
525 			*vcpu_pc(vcpu));
526 		print_sys_reg_instr(params);
527 	}
528 	kvm_inject_undefined(vcpu);
529 	return 1;
530 }
531 
532 static void reset_sys_reg_descs(struct kvm_vcpu *vcpu,
533 			      const struct sys_reg_desc *table, size_t num)
534 {
535 	unsigned long i;
536 
537 	for (i = 0; i < num; i++)
538 		if (table[i].reset)
539 			table[i].reset(vcpu, &table[i]);
540 }
541 
542 /**
543  * kvm_handle_sys_reg -- handles a mrs/msr trap on a guest sys_reg access
544  * @vcpu: The VCPU pointer
545  * @run:  The kvm_run struct
546  */
547 int kvm_handle_sys_reg(struct kvm_vcpu *vcpu, struct kvm_run *run)
548 {
549 	struct sys_reg_params params;
550 	unsigned long esr = kvm_vcpu_get_hsr(vcpu);
551 
552 	params.Op0 = (esr >> 20) & 3;
553 	params.Op1 = (esr >> 14) & 0x7;
554 	params.CRn = (esr >> 10) & 0xf;
555 	params.CRm = (esr >> 1) & 0xf;
556 	params.Op2 = (esr >> 17) & 0x7;
557 	params.Rt = (esr >> 5) & 0x1f;
558 	params.is_write = !(esr & 1);
559 
560 	return emulate_sys_reg(vcpu, &params);
561 }
562 
563 /******************************************************************************
564  * Userspace API
565  *****************************************************************************/
566 
567 static bool index_to_params(u64 id, struct sys_reg_params *params)
568 {
569 	switch (id & KVM_REG_SIZE_MASK) {
570 	case KVM_REG_SIZE_U64:
571 		/* Any unused index bits means it's not valid. */
572 		if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK
573 			      | KVM_REG_ARM_COPROC_MASK
574 			      | KVM_REG_ARM64_SYSREG_OP0_MASK
575 			      | KVM_REG_ARM64_SYSREG_OP1_MASK
576 			      | KVM_REG_ARM64_SYSREG_CRN_MASK
577 			      | KVM_REG_ARM64_SYSREG_CRM_MASK
578 			      | KVM_REG_ARM64_SYSREG_OP2_MASK))
579 			return false;
580 		params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK)
581 			       >> KVM_REG_ARM64_SYSREG_OP0_SHIFT);
582 		params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK)
583 			       >> KVM_REG_ARM64_SYSREG_OP1_SHIFT);
584 		params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK)
585 			       >> KVM_REG_ARM64_SYSREG_CRN_SHIFT);
586 		params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK)
587 			       >> KVM_REG_ARM64_SYSREG_CRM_SHIFT);
588 		params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK)
589 			       >> KVM_REG_ARM64_SYSREG_OP2_SHIFT);
590 		return true;
591 	default:
592 		return false;
593 	}
594 }
595 
596 /* Decode an index value, and find the sys_reg_desc entry. */
597 static const struct sys_reg_desc *index_to_sys_reg_desc(struct kvm_vcpu *vcpu,
598 						    u64 id)
599 {
600 	size_t num;
601 	const struct sys_reg_desc *table, *r;
602 	struct sys_reg_params params;
603 
604 	/* We only do sys_reg for now. */
605 	if ((id & KVM_REG_ARM_COPROC_MASK) != KVM_REG_ARM64_SYSREG)
606 		return NULL;
607 
608 	if (!index_to_params(id, &params))
609 		return NULL;
610 
611 	table = get_target_table(vcpu->arch.target, true, &num);
612 	r = find_reg(&params, table, num);
613 	if (!r)
614 		r = find_reg(&params, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
615 
616 	/* Not saved in the sys_reg array? */
617 	if (r && !r->reg)
618 		r = NULL;
619 
620 	return r;
621 }
622 
623 /*
624  * These are the invariant sys_reg registers: we let the guest see the
625  * host versions of these, so they're part of the guest state.
626  *
627  * A future CPU may provide a mechanism to present different values to
628  * the guest, or a future kvm may trap them.
629  */
630 
631 #define FUNCTION_INVARIANT(reg)						\
632 	static void get_##reg(struct kvm_vcpu *v,			\
633 			      const struct sys_reg_desc *r)		\
634 	{								\
635 		u64 val;						\
636 									\
637 		asm volatile("mrs %0, " __stringify(reg) "\n"		\
638 			     : "=r" (val));				\
639 		((struct sys_reg_desc *)r)->val = val;			\
640 	}
641 
642 FUNCTION_INVARIANT(midr_el1)
643 FUNCTION_INVARIANT(ctr_el0)
644 FUNCTION_INVARIANT(revidr_el1)
645 FUNCTION_INVARIANT(id_pfr0_el1)
646 FUNCTION_INVARIANT(id_pfr1_el1)
647 FUNCTION_INVARIANT(id_dfr0_el1)
648 FUNCTION_INVARIANT(id_afr0_el1)
649 FUNCTION_INVARIANT(id_mmfr0_el1)
650 FUNCTION_INVARIANT(id_mmfr1_el1)
651 FUNCTION_INVARIANT(id_mmfr2_el1)
652 FUNCTION_INVARIANT(id_mmfr3_el1)
653 FUNCTION_INVARIANT(id_isar0_el1)
654 FUNCTION_INVARIANT(id_isar1_el1)
655 FUNCTION_INVARIANT(id_isar2_el1)
656 FUNCTION_INVARIANT(id_isar3_el1)
657 FUNCTION_INVARIANT(id_isar4_el1)
658 FUNCTION_INVARIANT(id_isar5_el1)
659 FUNCTION_INVARIANT(clidr_el1)
660 FUNCTION_INVARIANT(aidr_el1)
661 
662 /* ->val is filled in by kvm_sys_reg_table_init() */
663 static struct sys_reg_desc invariant_sys_regs[] = {
664 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b000),
665 	  NULL, get_midr_el1 },
666 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0000), Op2(0b110),
667 	  NULL, get_revidr_el1 },
668 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b000),
669 	  NULL, get_id_pfr0_el1 },
670 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b001),
671 	  NULL, get_id_pfr1_el1 },
672 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b010),
673 	  NULL, get_id_dfr0_el1 },
674 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b011),
675 	  NULL, get_id_afr0_el1 },
676 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b100),
677 	  NULL, get_id_mmfr0_el1 },
678 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b101),
679 	  NULL, get_id_mmfr1_el1 },
680 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b110),
681 	  NULL, get_id_mmfr2_el1 },
682 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0001), Op2(0b111),
683 	  NULL, get_id_mmfr3_el1 },
684 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b000),
685 	  NULL, get_id_isar0_el1 },
686 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b001),
687 	  NULL, get_id_isar1_el1 },
688 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b010),
689 	  NULL, get_id_isar2_el1 },
690 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b011),
691 	  NULL, get_id_isar3_el1 },
692 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b100),
693 	  NULL, get_id_isar4_el1 },
694 	{ Op0(0b11), Op1(0b000), CRn(0b0000), CRm(0b0010), Op2(0b101),
695 	  NULL, get_id_isar5_el1 },
696 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b001),
697 	  NULL, get_clidr_el1 },
698 	{ Op0(0b11), Op1(0b001), CRn(0b0000), CRm(0b0000), Op2(0b111),
699 	  NULL, get_aidr_el1 },
700 	{ Op0(0b11), Op1(0b011), CRn(0b0000), CRm(0b0000), Op2(0b001),
701 	  NULL, get_ctr_el0 },
702 };
703 
704 static int reg_from_user(void *val, const void __user *uaddr, u64 id)
705 {
706 	/* This Just Works because we are little endian. */
707 	if (copy_from_user(val, uaddr, KVM_REG_SIZE(id)) != 0)
708 		return -EFAULT;
709 	return 0;
710 }
711 
712 static int reg_to_user(void __user *uaddr, const void *val, u64 id)
713 {
714 	/* This Just Works because we are little endian. */
715 	if (copy_to_user(uaddr, val, KVM_REG_SIZE(id)) != 0)
716 		return -EFAULT;
717 	return 0;
718 }
719 
720 static int get_invariant_sys_reg(u64 id, void __user *uaddr)
721 {
722 	struct sys_reg_params params;
723 	const struct sys_reg_desc *r;
724 
725 	if (!index_to_params(id, &params))
726 		return -ENOENT;
727 
728 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
729 	if (!r)
730 		return -ENOENT;
731 
732 	return reg_to_user(uaddr, &r->val, id);
733 }
734 
735 static int set_invariant_sys_reg(u64 id, void __user *uaddr)
736 {
737 	struct sys_reg_params params;
738 	const struct sys_reg_desc *r;
739 	int err;
740 	u64 val = 0; /* Make sure high bits are 0 for 32-bit regs */
741 
742 	if (!index_to_params(id, &params))
743 		return -ENOENT;
744 	r = find_reg(&params, invariant_sys_regs, ARRAY_SIZE(invariant_sys_regs));
745 	if (!r)
746 		return -ENOENT;
747 
748 	err = reg_from_user(&val, uaddr, id);
749 	if (err)
750 		return err;
751 
752 	/* This is what we mean by invariant: you can't change it. */
753 	if (r->val != val)
754 		return -EINVAL;
755 
756 	return 0;
757 }
758 
759 static bool is_valid_cache(u32 val)
760 {
761 	u32 level, ctype;
762 
763 	if (val >= CSSELR_MAX)
764 		return -ENOENT;
765 
766 	/* Bottom bit is Instruction or Data bit.  Next 3 bits are level. */
767 	level = (val >> 1);
768 	ctype = (cache_levels >> (level * 3)) & 7;
769 
770 	switch (ctype) {
771 	case 0: /* No cache */
772 		return false;
773 	case 1: /* Instruction cache only */
774 		return (val & 1);
775 	case 2: /* Data cache only */
776 	case 4: /* Unified cache */
777 		return !(val & 1);
778 	case 3: /* Separate instruction and data caches */
779 		return true;
780 	default: /* Reserved: we can't know instruction or data. */
781 		return false;
782 	}
783 }
784 
785 static int demux_c15_get(u64 id, void __user *uaddr)
786 {
787 	u32 val;
788 	u32 __user *uval = uaddr;
789 
790 	/* Fail if we have unknown bits set. */
791 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
792 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
793 		return -ENOENT;
794 
795 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
796 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
797 		if (KVM_REG_SIZE(id) != 4)
798 			return -ENOENT;
799 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
800 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
801 		if (!is_valid_cache(val))
802 			return -ENOENT;
803 
804 		return put_user(get_ccsidr(val), uval);
805 	default:
806 		return -ENOENT;
807 	}
808 }
809 
810 static int demux_c15_set(u64 id, void __user *uaddr)
811 {
812 	u32 val, newval;
813 	u32 __user *uval = uaddr;
814 
815 	/* Fail if we have unknown bits set. */
816 	if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
817 		   | ((1 << KVM_REG_ARM_COPROC_SHIFT)-1)))
818 		return -ENOENT;
819 
820 	switch (id & KVM_REG_ARM_DEMUX_ID_MASK) {
821 	case KVM_REG_ARM_DEMUX_ID_CCSIDR:
822 		if (KVM_REG_SIZE(id) != 4)
823 			return -ENOENT;
824 		val = (id & KVM_REG_ARM_DEMUX_VAL_MASK)
825 			>> KVM_REG_ARM_DEMUX_VAL_SHIFT;
826 		if (!is_valid_cache(val))
827 			return -ENOENT;
828 
829 		if (get_user(newval, uval))
830 			return -EFAULT;
831 
832 		/* This is also invariant: you can't change it. */
833 		if (newval != get_ccsidr(val))
834 			return -EINVAL;
835 		return 0;
836 	default:
837 		return -ENOENT;
838 	}
839 }
840 
841 int kvm_arm_sys_reg_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
842 {
843 	const struct sys_reg_desc *r;
844 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
845 
846 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
847 		return demux_c15_get(reg->id, uaddr);
848 
849 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
850 		return -ENOENT;
851 
852 	r = index_to_sys_reg_desc(vcpu, reg->id);
853 	if (!r)
854 		return get_invariant_sys_reg(reg->id, uaddr);
855 
856 	return reg_to_user(uaddr, &vcpu_sys_reg(vcpu, r->reg), reg->id);
857 }
858 
859 int kvm_arm_sys_reg_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
860 {
861 	const struct sys_reg_desc *r;
862 	void __user *uaddr = (void __user *)(unsigned long)reg->addr;
863 
864 	if ((reg->id & KVM_REG_ARM_COPROC_MASK) == KVM_REG_ARM_DEMUX)
865 		return demux_c15_set(reg->id, uaddr);
866 
867 	if (KVM_REG_SIZE(reg->id) != sizeof(__u64))
868 		return -ENOENT;
869 
870 	r = index_to_sys_reg_desc(vcpu, reg->id);
871 	if (!r)
872 		return set_invariant_sys_reg(reg->id, uaddr);
873 
874 	return reg_from_user(&vcpu_sys_reg(vcpu, r->reg), uaddr, reg->id);
875 }
876 
877 static unsigned int num_demux_regs(void)
878 {
879 	unsigned int i, count = 0;
880 
881 	for (i = 0; i < CSSELR_MAX; i++)
882 		if (is_valid_cache(i))
883 			count++;
884 
885 	return count;
886 }
887 
888 static int write_demux_regids(u64 __user *uindices)
889 {
890 	u64 val = KVM_REG_ARM | KVM_REG_SIZE_U32 | KVM_REG_ARM_DEMUX;
891 	unsigned int i;
892 
893 	val |= KVM_REG_ARM_DEMUX_ID_CCSIDR;
894 	for (i = 0; i < CSSELR_MAX; i++) {
895 		if (!is_valid_cache(i))
896 			continue;
897 		if (put_user(val | i, uindices))
898 			return -EFAULT;
899 		uindices++;
900 	}
901 	return 0;
902 }
903 
904 static u64 sys_reg_to_index(const struct sys_reg_desc *reg)
905 {
906 	return (KVM_REG_ARM64 | KVM_REG_SIZE_U64 |
907 		KVM_REG_ARM64_SYSREG |
908 		(reg->Op0 << KVM_REG_ARM64_SYSREG_OP0_SHIFT) |
909 		(reg->Op1 << KVM_REG_ARM64_SYSREG_OP1_SHIFT) |
910 		(reg->CRn << KVM_REG_ARM64_SYSREG_CRN_SHIFT) |
911 		(reg->CRm << KVM_REG_ARM64_SYSREG_CRM_SHIFT) |
912 		(reg->Op2 << KVM_REG_ARM64_SYSREG_OP2_SHIFT));
913 }
914 
915 static bool copy_reg_to_user(const struct sys_reg_desc *reg, u64 __user **uind)
916 {
917 	if (!*uind)
918 		return true;
919 
920 	if (put_user(sys_reg_to_index(reg), *uind))
921 		return false;
922 
923 	(*uind)++;
924 	return true;
925 }
926 
927 /* Assumed ordered tables, see kvm_sys_reg_table_init. */
928 static int walk_sys_regs(struct kvm_vcpu *vcpu, u64 __user *uind)
929 {
930 	const struct sys_reg_desc *i1, *i2, *end1, *end2;
931 	unsigned int total = 0;
932 	size_t num;
933 
934 	/* We check for duplicates here, to allow arch-specific overrides. */
935 	i1 = get_target_table(vcpu->arch.target, true, &num);
936 	end1 = i1 + num;
937 	i2 = sys_reg_descs;
938 	end2 = sys_reg_descs + ARRAY_SIZE(sys_reg_descs);
939 
940 	BUG_ON(i1 == end1 || i2 == end2);
941 
942 	/* Walk carefully, as both tables may refer to the same register. */
943 	while (i1 || i2) {
944 		int cmp = cmp_sys_reg(i1, i2);
945 		/* target-specific overrides generic entry. */
946 		if (cmp <= 0) {
947 			/* Ignore registers we trap but don't save. */
948 			if (i1->reg) {
949 				if (!copy_reg_to_user(i1, &uind))
950 					return -EFAULT;
951 				total++;
952 			}
953 		} else {
954 			/* Ignore registers we trap but don't save. */
955 			if (i2->reg) {
956 				if (!copy_reg_to_user(i2, &uind))
957 					return -EFAULT;
958 				total++;
959 			}
960 		}
961 
962 		if (cmp <= 0 && ++i1 == end1)
963 			i1 = NULL;
964 		if (cmp >= 0 && ++i2 == end2)
965 			i2 = NULL;
966 	}
967 	return total;
968 }
969 
970 unsigned long kvm_arm_num_sys_reg_descs(struct kvm_vcpu *vcpu)
971 {
972 	return ARRAY_SIZE(invariant_sys_regs)
973 		+ num_demux_regs()
974 		+ walk_sys_regs(vcpu, (u64 __user *)NULL);
975 }
976 
977 int kvm_arm_copy_sys_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices)
978 {
979 	unsigned int i;
980 	int err;
981 
982 	/* Then give them all the invariant registers' indices. */
983 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++) {
984 		if (put_user(sys_reg_to_index(&invariant_sys_regs[i]), uindices))
985 			return -EFAULT;
986 		uindices++;
987 	}
988 
989 	err = walk_sys_regs(vcpu, uindices);
990 	if (err < 0)
991 		return err;
992 	uindices += err;
993 
994 	return write_demux_regids(uindices);
995 }
996 
997 void kvm_sys_reg_table_init(void)
998 {
999 	unsigned int i;
1000 	struct sys_reg_desc clidr;
1001 
1002 	/* Make sure tables are unique and in order. */
1003 	for (i = 1; i < ARRAY_SIZE(sys_reg_descs); i++)
1004 		BUG_ON(cmp_sys_reg(&sys_reg_descs[i-1], &sys_reg_descs[i]) >= 0);
1005 
1006 	/* We abuse the reset function to overwrite the table itself. */
1007 	for (i = 0; i < ARRAY_SIZE(invariant_sys_regs); i++)
1008 		invariant_sys_regs[i].reset(NULL, &invariant_sys_regs[i]);
1009 
1010 	/*
1011 	 * CLIDR format is awkward, so clean it up.  See ARM B4.1.20:
1012 	 *
1013 	 *   If software reads the Cache Type fields from Ctype1
1014 	 *   upwards, once it has seen a value of 0b000, no caches
1015 	 *   exist at further-out levels of the hierarchy. So, for
1016 	 *   example, if Ctype3 is the first Cache Type field with a
1017 	 *   value of 0b000, the values of Ctype4 to Ctype7 must be
1018 	 *   ignored.
1019 	 */
1020 	get_clidr_el1(NULL, &clidr); /* Ugly... */
1021 	cache_levels = clidr.val;
1022 	for (i = 0; i < 7; i++)
1023 		if (((cache_levels >> (i*3)) & 7) == 0)
1024 			break;
1025 	/* Clear all higher bits. */
1026 	cache_levels &= (1 << (i*3))-1;
1027 }
1028 
1029 /**
1030  * kvm_reset_sys_regs - sets system registers to reset value
1031  * @vcpu: The VCPU pointer
1032  *
1033  * This function finds the right table above and sets the registers on the
1034  * virtual CPU struct to their architecturally defined reset values.
1035  */
1036 void kvm_reset_sys_regs(struct kvm_vcpu *vcpu)
1037 {
1038 	size_t num;
1039 	const struct sys_reg_desc *table;
1040 
1041 	/* Catch someone adding a register without putting in reset entry. */
1042 	memset(&vcpu->arch.ctxt.sys_regs, 0x42, sizeof(vcpu->arch.ctxt.sys_regs));
1043 
1044 	/* Generic chip reset first (so target could override). */
1045 	reset_sys_reg_descs(vcpu, sys_reg_descs, ARRAY_SIZE(sys_reg_descs));
1046 
1047 	table = get_target_table(vcpu->arch.target, true, &num);
1048 	reset_sys_reg_descs(vcpu, table, num);
1049 
1050 	for (num = 1; num < NR_SYS_REGS; num++)
1051 		if (vcpu_sys_reg(vcpu, num) == 0x4242424242424242)
1052 			panic("Didn't reset vcpu_sys_reg(%zi)", num);
1053 }
1054