xref: /linux/arch/arm64/kvm/hyp/nvhe/hyp-main.c (revision a69283ae1db8dd416870d931caa9e2d3d2c1cd8b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2020 - Google Inc
4  * Author: Andrew Scull <ascull@google.com>
5  */
6 
7 #include <hyp/adjust_pc.h>
8 
9 #include <asm/pgtable-types.h>
10 #include <asm/kvm_asm.h>
11 #include <asm/kvm_emulate.h>
12 #include <asm/kvm_host.h>
13 #include <asm/kvm_hyp.h>
14 #include <asm/kvm_mmu.h>
15 
16 #include <nvhe/ffa.h>
17 #include <nvhe/mem_protect.h>
18 #include <nvhe/mm.h>
19 #include <nvhe/pkvm.h>
20 #include <nvhe/trap_handler.h>
21 
22 DEFINE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
23 
24 void __kvm_hyp_host_forward_smc(struct kvm_cpu_context *host_ctxt);
25 
26 static void __hyp_sve_save_guest(struct kvm_vcpu *vcpu)
27 {
28 	__vcpu_sys_reg(vcpu, ZCR_EL1) = read_sysreg_el1(SYS_ZCR);
29 	/*
30 	 * On saving/restoring guest sve state, always use the maximum VL for
31 	 * the guest. The layout of the data when saving the sve state depends
32 	 * on the VL, so use a consistent (i.e., the maximum) guest VL.
33 	 */
34 	sve_cond_update_zcr_vq(vcpu_sve_max_vq(vcpu) - 1, SYS_ZCR_EL2);
35 	__sve_save_state(vcpu_sve_pffr(vcpu), &vcpu->arch.ctxt.fp_regs.fpsr, true);
36 	write_sysreg_s(ZCR_ELx_LEN_MASK, SYS_ZCR_EL2);
37 }
38 
39 static void __hyp_sve_restore_host(void)
40 {
41 	struct cpu_sve_state *sve_state = *host_data_ptr(sve_state);
42 
43 	/*
44 	 * On saving/restoring host sve state, always use the maximum VL for
45 	 * the host. The layout of the data when saving the sve state depends
46 	 * on the VL, so use a consistent (i.e., the maximum) host VL.
47 	 *
48 	 * Setting ZCR_EL2 to ZCR_ELx_LEN_MASK sets the effective length
49 	 * supported by the system (or limited at EL3).
50 	 */
51 	write_sysreg_s(ZCR_ELx_LEN_MASK, SYS_ZCR_EL2);
52 	__sve_restore_state(sve_state->sve_regs + sve_ffr_offset(kvm_host_sve_max_vl),
53 			    &sve_state->fpsr,
54 			    true);
55 	write_sysreg_el1(sve_state->zcr_el1, SYS_ZCR);
56 }
57 
58 static void fpsimd_sve_flush(void)
59 {
60 	*host_data_ptr(fp_owner) = FP_STATE_HOST_OWNED;
61 }
62 
63 static void fpsimd_sve_sync(struct kvm_vcpu *vcpu)
64 {
65 	if (!guest_owns_fp_regs())
66 		return;
67 
68 	cpacr_clear_set(0, CPACR_ELx_FPEN | CPACR_ELx_ZEN);
69 	isb();
70 
71 	if (vcpu_has_sve(vcpu))
72 		__hyp_sve_save_guest(vcpu);
73 	else
74 		__fpsimd_save_state(&vcpu->arch.ctxt.fp_regs);
75 
76 	if (system_supports_sve())
77 		__hyp_sve_restore_host();
78 	else
79 		__fpsimd_restore_state(*host_data_ptr(fpsimd_state));
80 
81 	*host_data_ptr(fp_owner) = FP_STATE_HOST_OWNED;
82 }
83 
84 static void flush_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
85 {
86 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
87 
88 	fpsimd_sve_flush();
89 
90 	hyp_vcpu->vcpu.arch.ctxt	= host_vcpu->arch.ctxt;
91 
92 	hyp_vcpu->vcpu.arch.sve_state	= kern_hyp_va(host_vcpu->arch.sve_state);
93 	/* Limit guest vector length to the maximum supported by the host.  */
94 	hyp_vcpu->vcpu.arch.sve_max_vl	= min(host_vcpu->arch.sve_max_vl, kvm_host_sve_max_vl);
95 
96 	hyp_vcpu->vcpu.arch.hw_mmu	= host_vcpu->arch.hw_mmu;
97 
98 	hyp_vcpu->vcpu.arch.hcr_el2	= host_vcpu->arch.hcr_el2;
99 	hyp_vcpu->vcpu.arch.mdcr_el2	= host_vcpu->arch.mdcr_el2;
100 
101 	hyp_vcpu->vcpu.arch.iflags	= host_vcpu->arch.iflags;
102 
103 	hyp_vcpu->vcpu.arch.debug_ptr	= kern_hyp_va(host_vcpu->arch.debug_ptr);
104 
105 	hyp_vcpu->vcpu.arch.vsesr_el2	= host_vcpu->arch.vsesr_el2;
106 
107 	hyp_vcpu->vcpu.arch.vgic_cpu.vgic_v3 = host_vcpu->arch.vgic_cpu.vgic_v3;
108 }
109 
110 static void sync_hyp_vcpu(struct pkvm_hyp_vcpu *hyp_vcpu)
111 {
112 	struct kvm_vcpu *host_vcpu = hyp_vcpu->host_vcpu;
113 	struct vgic_v3_cpu_if *hyp_cpu_if = &hyp_vcpu->vcpu.arch.vgic_cpu.vgic_v3;
114 	struct vgic_v3_cpu_if *host_cpu_if = &host_vcpu->arch.vgic_cpu.vgic_v3;
115 	unsigned int i;
116 
117 	fpsimd_sve_sync(&hyp_vcpu->vcpu);
118 
119 	host_vcpu->arch.ctxt		= hyp_vcpu->vcpu.arch.ctxt;
120 
121 	host_vcpu->arch.hcr_el2		= hyp_vcpu->vcpu.arch.hcr_el2;
122 
123 	host_vcpu->arch.fault		= hyp_vcpu->vcpu.arch.fault;
124 
125 	host_vcpu->arch.iflags		= hyp_vcpu->vcpu.arch.iflags;
126 
127 	host_cpu_if->vgic_hcr		= hyp_cpu_if->vgic_hcr;
128 	for (i = 0; i < hyp_cpu_if->used_lrs; ++i)
129 		host_cpu_if->vgic_lr[i] = hyp_cpu_if->vgic_lr[i];
130 }
131 
132 static void handle___kvm_vcpu_run(struct kvm_cpu_context *host_ctxt)
133 {
134 	DECLARE_REG(struct kvm_vcpu *, host_vcpu, host_ctxt, 1);
135 	int ret;
136 
137 	host_vcpu = kern_hyp_va(host_vcpu);
138 
139 	if (unlikely(is_protected_kvm_enabled())) {
140 		struct pkvm_hyp_vcpu *hyp_vcpu;
141 		struct kvm *host_kvm;
142 
143 		host_kvm = kern_hyp_va(host_vcpu->kvm);
144 		hyp_vcpu = pkvm_load_hyp_vcpu(host_kvm->arch.pkvm.handle,
145 					      host_vcpu->vcpu_idx);
146 		if (!hyp_vcpu) {
147 			ret = -EINVAL;
148 			goto out;
149 		}
150 
151 		flush_hyp_vcpu(hyp_vcpu);
152 
153 		ret = __kvm_vcpu_run(&hyp_vcpu->vcpu);
154 
155 		sync_hyp_vcpu(hyp_vcpu);
156 		pkvm_put_hyp_vcpu(hyp_vcpu);
157 	} else {
158 		/* The host is fully trusted, run its vCPU directly. */
159 		ret = __kvm_vcpu_run(host_vcpu);
160 	}
161 
162 out:
163 	cpu_reg(host_ctxt, 1) =  ret;
164 }
165 
166 static void handle___kvm_adjust_pc(struct kvm_cpu_context *host_ctxt)
167 {
168 	DECLARE_REG(struct kvm_vcpu *, vcpu, host_ctxt, 1);
169 
170 	__kvm_adjust_pc(kern_hyp_va(vcpu));
171 }
172 
173 static void handle___kvm_flush_vm_context(struct kvm_cpu_context *host_ctxt)
174 {
175 	__kvm_flush_vm_context();
176 }
177 
178 static void handle___kvm_tlb_flush_vmid_ipa(struct kvm_cpu_context *host_ctxt)
179 {
180 	DECLARE_REG(struct kvm_s2_mmu *, mmu, host_ctxt, 1);
181 	DECLARE_REG(phys_addr_t, ipa, host_ctxt, 2);
182 	DECLARE_REG(int, level, host_ctxt, 3);
183 
184 	__kvm_tlb_flush_vmid_ipa(kern_hyp_va(mmu), ipa, level);
185 }
186 
187 static void handle___kvm_tlb_flush_vmid_ipa_nsh(struct kvm_cpu_context *host_ctxt)
188 {
189 	DECLARE_REG(struct kvm_s2_mmu *, mmu, host_ctxt, 1);
190 	DECLARE_REG(phys_addr_t, ipa, host_ctxt, 2);
191 	DECLARE_REG(int, level, host_ctxt, 3);
192 
193 	__kvm_tlb_flush_vmid_ipa_nsh(kern_hyp_va(mmu), ipa, level);
194 }
195 
196 static void
197 handle___kvm_tlb_flush_vmid_range(struct kvm_cpu_context *host_ctxt)
198 {
199 	DECLARE_REG(struct kvm_s2_mmu *, mmu, host_ctxt, 1);
200 	DECLARE_REG(phys_addr_t, start, host_ctxt, 2);
201 	DECLARE_REG(unsigned long, pages, host_ctxt, 3);
202 
203 	__kvm_tlb_flush_vmid_range(kern_hyp_va(mmu), start, pages);
204 }
205 
206 static void handle___kvm_tlb_flush_vmid(struct kvm_cpu_context *host_ctxt)
207 {
208 	DECLARE_REG(struct kvm_s2_mmu *, mmu, host_ctxt, 1);
209 
210 	__kvm_tlb_flush_vmid(kern_hyp_va(mmu));
211 }
212 
213 static void handle___kvm_flush_cpu_context(struct kvm_cpu_context *host_ctxt)
214 {
215 	DECLARE_REG(struct kvm_s2_mmu *, mmu, host_ctxt, 1);
216 
217 	__kvm_flush_cpu_context(kern_hyp_va(mmu));
218 }
219 
220 static void handle___kvm_timer_set_cntvoff(struct kvm_cpu_context *host_ctxt)
221 {
222 	__kvm_timer_set_cntvoff(cpu_reg(host_ctxt, 1));
223 }
224 
225 static void handle___kvm_enable_ssbs(struct kvm_cpu_context *host_ctxt)
226 {
227 	u64 tmp;
228 
229 	tmp = read_sysreg_el2(SYS_SCTLR);
230 	tmp |= SCTLR_ELx_DSSBS;
231 	write_sysreg_el2(tmp, SYS_SCTLR);
232 }
233 
234 static void handle___vgic_v3_get_gic_config(struct kvm_cpu_context *host_ctxt)
235 {
236 	cpu_reg(host_ctxt, 1) = __vgic_v3_get_gic_config();
237 }
238 
239 static void handle___vgic_v3_init_lrs(struct kvm_cpu_context *host_ctxt)
240 {
241 	__vgic_v3_init_lrs();
242 }
243 
244 static void handle___kvm_get_mdcr_el2(struct kvm_cpu_context *host_ctxt)
245 {
246 	cpu_reg(host_ctxt, 1) = __kvm_get_mdcr_el2();
247 }
248 
249 static void handle___vgic_v3_save_vmcr_aprs(struct kvm_cpu_context *host_ctxt)
250 {
251 	DECLARE_REG(struct vgic_v3_cpu_if *, cpu_if, host_ctxt, 1);
252 
253 	__vgic_v3_save_vmcr_aprs(kern_hyp_va(cpu_if));
254 }
255 
256 static void handle___vgic_v3_restore_vmcr_aprs(struct kvm_cpu_context *host_ctxt)
257 {
258 	DECLARE_REG(struct vgic_v3_cpu_if *, cpu_if, host_ctxt, 1);
259 
260 	__vgic_v3_restore_vmcr_aprs(kern_hyp_va(cpu_if));
261 }
262 
263 static void handle___pkvm_init(struct kvm_cpu_context *host_ctxt)
264 {
265 	DECLARE_REG(phys_addr_t, phys, host_ctxt, 1);
266 	DECLARE_REG(unsigned long, size, host_ctxt, 2);
267 	DECLARE_REG(unsigned long, nr_cpus, host_ctxt, 3);
268 	DECLARE_REG(unsigned long *, per_cpu_base, host_ctxt, 4);
269 	DECLARE_REG(u32, hyp_va_bits, host_ctxt, 5);
270 
271 	/*
272 	 * __pkvm_init() will return only if an error occurred, otherwise it
273 	 * will tail-call in __pkvm_init_finalise() which will have to deal
274 	 * with the host context directly.
275 	 */
276 	cpu_reg(host_ctxt, 1) = __pkvm_init(phys, size, nr_cpus, per_cpu_base,
277 					    hyp_va_bits);
278 }
279 
280 static void handle___pkvm_cpu_set_vector(struct kvm_cpu_context *host_ctxt)
281 {
282 	DECLARE_REG(enum arm64_hyp_spectre_vector, slot, host_ctxt, 1);
283 
284 	cpu_reg(host_ctxt, 1) = pkvm_cpu_set_vector(slot);
285 }
286 
287 static void handle___pkvm_host_share_hyp(struct kvm_cpu_context *host_ctxt)
288 {
289 	DECLARE_REG(u64, pfn, host_ctxt, 1);
290 
291 	cpu_reg(host_ctxt, 1) = __pkvm_host_share_hyp(pfn);
292 }
293 
294 static void handle___pkvm_host_unshare_hyp(struct kvm_cpu_context *host_ctxt)
295 {
296 	DECLARE_REG(u64, pfn, host_ctxt, 1);
297 
298 	cpu_reg(host_ctxt, 1) = __pkvm_host_unshare_hyp(pfn);
299 }
300 
301 static void handle___pkvm_create_private_mapping(struct kvm_cpu_context *host_ctxt)
302 {
303 	DECLARE_REG(phys_addr_t, phys, host_ctxt, 1);
304 	DECLARE_REG(size_t, size, host_ctxt, 2);
305 	DECLARE_REG(enum kvm_pgtable_prot, prot, host_ctxt, 3);
306 
307 	/*
308 	 * __pkvm_create_private_mapping() populates a pointer with the
309 	 * hypervisor start address of the allocation.
310 	 *
311 	 * However, handle___pkvm_create_private_mapping() hypercall crosses the
312 	 * EL1/EL2 boundary so the pointer would not be valid in this context.
313 	 *
314 	 * Instead pass the allocation address as the return value (or return
315 	 * ERR_PTR() on failure).
316 	 */
317 	unsigned long haddr;
318 	int err = __pkvm_create_private_mapping(phys, size, prot, &haddr);
319 
320 	if (err)
321 		haddr = (unsigned long)ERR_PTR(err);
322 
323 	cpu_reg(host_ctxt, 1) = haddr;
324 }
325 
326 static void handle___pkvm_prot_finalize(struct kvm_cpu_context *host_ctxt)
327 {
328 	cpu_reg(host_ctxt, 1) = __pkvm_prot_finalize();
329 }
330 
331 static void handle___pkvm_vcpu_init_traps(struct kvm_cpu_context *host_ctxt)
332 {
333 	DECLARE_REG(struct kvm_vcpu *, vcpu, host_ctxt, 1);
334 
335 	__pkvm_vcpu_init_traps(kern_hyp_va(vcpu));
336 }
337 
338 static void handle___pkvm_init_vm(struct kvm_cpu_context *host_ctxt)
339 {
340 	DECLARE_REG(struct kvm *, host_kvm, host_ctxt, 1);
341 	DECLARE_REG(unsigned long, vm_hva, host_ctxt, 2);
342 	DECLARE_REG(unsigned long, pgd_hva, host_ctxt, 3);
343 
344 	host_kvm = kern_hyp_va(host_kvm);
345 	cpu_reg(host_ctxt, 1) = __pkvm_init_vm(host_kvm, vm_hva, pgd_hva);
346 }
347 
348 static void handle___pkvm_init_vcpu(struct kvm_cpu_context *host_ctxt)
349 {
350 	DECLARE_REG(pkvm_handle_t, handle, host_ctxt, 1);
351 	DECLARE_REG(struct kvm_vcpu *, host_vcpu, host_ctxt, 2);
352 	DECLARE_REG(unsigned long, vcpu_hva, host_ctxt, 3);
353 
354 	host_vcpu = kern_hyp_va(host_vcpu);
355 	cpu_reg(host_ctxt, 1) = __pkvm_init_vcpu(handle, host_vcpu, vcpu_hva);
356 }
357 
358 static void handle___pkvm_teardown_vm(struct kvm_cpu_context *host_ctxt)
359 {
360 	DECLARE_REG(pkvm_handle_t, handle, host_ctxt, 1);
361 
362 	cpu_reg(host_ctxt, 1) = __pkvm_teardown_vm(handle);
363 }
364 
365 typedef void (*hcall_t)(struct kvm_cpu_context *);
366 
367 #define HANDLE_FUNC(x)	[__KVM_HOST_SMCCC_FUNC_##x] = (hcall_t)handle_##x
368 
369 static const hcall_t host_hcall[] = {
370 	/* ___kvm_hyp_init */
371 	HANDLE_FUNC(__kvm_get_mdcr_el2),
372 	HANDLE_FUNC(__pkvm_init),
373 	HANDLE_FUNC(__pkvm_create_private_mapping),
374 	HANDLE_FUNC(__pkvm_cpu_set_vector),
375 	HANDLE_FUNC(__kvm_enable_ssbs),
376 	HANDLE_FUNC(__vgic_v3_init_lrs),
377 	HANDLE_FUNC(__vgic_v3_get_gic_config),
378 	HANDLE_FUNC(__pkvm_prot_finalize),
379 
380 	HANDLE_FUNC(__pkvm_host_share_hyp),
381 	HANDLE_FUNC(__pkvm_host_unshare_hyp),
382 	HANDLE_FUNC(__kvm_adjust_pc),
383 	HANDLE_FUNC(__kvm_vcpu_run),
384 	HANDLE_FUNC(__kvm_flush_vm_context),
385 	HANDLE_FUNC(__kvm_tlb_flush_vmid_ipa),
386 	HANDLE_FUNC(__kvm_tlb_flush_vmid_ipa_nsh),
387 	HANDLE_FUNC(__kvm_tlb_flush_vmid),
388 	HANDLE_FUNC(__kvm_tlb_flush_vmid_range),
389 	HANDLE_FUNC(__kvm_flush_cpu_context),
390 	HANDLE_FUNC(__kvm_timer_set_cntvoff),
391 	HANDLE_FUNC(__vgic_v3_save_vmcr_aprs),
392 	HANDLE_FUNC(__vgic_v3_restore_vmcr_aprs),
393 	HANDLE_FUNC(__pkvm_vcpu_init_traps),
394 	HANDLE_FUNC(__pkvm_init_vm),
395 	HANDLE_FUNC(__pkvm_init_vcpu),
396 	HANDLE_FUNC(__pkvm_teardown_vm),
397 };
398 
399 static void handle_host_hcall(struct kvm_cpu_context *host_ctxt)
400 {
401 	DECLARE_REG(unsigned long, id, host_ctxt, 0);
402 	unsigned long hcall_min = 0;
403 	hcall_t hfn;
404 
405 	/*
406 	 * If pKVM has been initialised then reject any calls to the
407 	 * early "privileged" hypercalls. Note that we cannot reject
408 	 * calls to __pkvm_prot_finalize for two reasons: (1) The static
409 	 * key used to determine initialisation must be toggled prior to
410 	 * finalisation and (2) finalisation is performed on a per-CPU
411 	 * basis. This is all fine, however, since __pkvm_prot_finalize
412 	 * returns -EPERM after the first call for a given CPU.
413 	 */
414 	if (static_branch_unlikely(&kvm_protected_mode_initialized))
415 		hcall_min = __KVM_HOST_SMCCC_FUNC___pkvm_prot_finalize;
416 
417 	id &= ~ARM_SMCCC_CALL_HINTS;
418 	id -= KVM_HOST_SMCCC_ID(0);
419 
420 	if (unlikely(id < hcall_min || id >= ARRAY_SIZE(host_hcall)))
421 		goto inval;
422 
423 	hfn = host_hcall[id];
424 	if (unlikely(!hfn))
425 		goto inval;
426 
427 	cpu_reg(host_ctxt, 0) = SMCCC_RET_SUCCESS;
428 	hfn(host_ctxt);
429 
430 	return;
431 inval:
432 	cpu_reg(host_ctxt, 0) = SMCCC_RET_NOT_SUPPORTED;
433 }
434 
435 static void default_host_smc_handler(struct kvm_cpu_context *host_ctxt)
436 {
437 	__kvm_hyp_host_forward_smc(host_ctxt);
438 }
439 
440 static void handle_host_smc(struct kvm_cpu_context *host_ctxt)
441 {
442 	DECLARE_REG(u64, func_id, host_ctxt, 0);
443 	bool handled;
444 
445 	func_id &= ~ARM_SMCCC_CALL_HINTS;
446 
447 	handled = kvm_host_psci_handler(host_ctxt, func_id);
448 	if (!handled)
449 		handled = kvm_host_ffa_handler(host_ctxt, func_id);
450 	if (!handled)
451 		default_host_smc_handler(host_ctxt);
452 
453 	/* SMC was trapped, move ELR past the current PC. */
454 	kvm_skip_host_instr();
455 }
456 
457 void handle_trap(struct kvm_cpu_context *host_ctxt)
458 {
459 	u64 esr = read_sysreg_el2(SYS_ESR);
460 
461 	switch (ESR_ELx_EC(esr)) {
462 	case ESR_ELx_EC_HVC64:
463 		handle_host_hcall(host_ctxt);
464 		break;
465 	case ESR_ELx_EC_SMC64:
466 		handle_host_smc(host_ctxt);
467 		break;
468 	case ESR_ELx_EC_SVE:
469 		cpacr_clear_set(0, CPACR_ELx_ZEN);
470 		isb();
471 		sve_cond_update_zcr_vq(ZCR_ELx_LEN_MASK, SYS_ZCR_EL2);
472 		break;
473 	case ESR_ELx_EC_IABT_LOW:
474 	case ESR_ELx_EC_DABT_LOW:
475 		handle_host_mem_abort(host_ctxt);
476 		break;
477 	default:
478 		BUG();
479 	}
480 }
481