1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * AArch64 code
4  *
5  * Copyright (C) 2018, Red Hat, Inc.
6  */
7 
8 #include <linux/compiler.h>
9 #include <assert.h>
10 
11 #include "kvm_util.h"
12 #include "../kvm_util_internal.h"
13 #include "processor.h"
14 
15 #define DEFAULT_ARM64_GUEST_STACK_VADDR_MIN	0xac0000
16 
17 static vm_vaddr_t exception_handlers;
18 
19 static uint64_t page_align(struct kvm_vm *vm, uint64_t v)
20 {
21 	return (v + vm->page_size) & ~(vm->page_size - 1);
22 }
23 
24 static uint64_t pgd_index(struct kvm_vm *vm, vm_vaddr_t gva)
25 {
26 	unsigned int shift = (vm->pgtable_levels - 1) * (vm->page_shift - 3) + vm->page_shift;
27 	uint64_t mask = (1UL << (vm->va_bits - shift)) - 1;
28 
29 	return (gva >> shift) & mask;
30 }
31 
32 static uint64_t pud_index(struct kvm_vm *vm, vm_vaddr_t gva)
33 {
34 	unsigned int shift = 2 * (vm->page_shift - 3) + vm->page_shift;
35 	uint64_t mask = (1UL << (vm->page_shift - 3)) - 1;
36 
37 	TEST_ASSERT(vm->pgtable_levels == 4,
38 		"Mode %d does not have 4 page table levels", vm->mode);
39 
40 	return (gva >> shift) & mask;
41 }
42 
43 static uint64_t pmd_index(struct kvm_vm *vm, vm_vaddr_t gva)
44 {
45 	unsigned int shift = (vm->page_shift - 3) + vm->page_shift;
46 	uint64_t mask = (1UL << (vm->page_shift - 3)) - 1;
47 
48 	TEST_ASSERT(vm->pgtable_levels >= 3,
49 		"Mode %d does not have >= 3 page table levels", vm->mode);
50 
51 	return (gva >> shift) & mask;
52 }
53 
54 static uint64_t pte_index(struct kvm_vm *vm, vm_vaddr_t gva)
55 {
56 	uint64_t mask = (1UL << (vm->page_shift - 3)) - 1;
57 	return (gva >> vm->page_shift) & mask;
58 }
59 
60 static uint64_t pte_addr(struct kvm_vm *vm, uint64_t entry)
61 {
62 	uint64_t mask = ((1UL << (vm->va_bits - vm->page_shift)) - 1) << vm->page_shift;
63 	return entry & mask;
64 }
65 
66 static uint64_t ptrs_per_pgd(struct kvm_vm *vm)
67 {
68 	unsigned int shift = (vm->pgtable_levels - 1) * (vm->page_shift - 3) + vm->page_shift;
69 	return 1 << (vm->va_bits - shift);
70 }
71 
72 static uint64_t __maybe_unused ptrs_per_pte(struct kvm_vm *vm)
73 {
74 	return 1 << (vm->page_shift - 3);
75 }
76 
77 void virt_pgd_alloc(struct kvm_vm *vm)
78 {
79 	if (!vm->pgd_created) {
80 		vm_paddr_t paddr = vm_phy_pages_alloc(vm,
81 			page_align(vm, ptrs_per_pgd(vm) * 8) / vm->page_size,
82 			KVM_GUEST_PAGE_TABLE_MIN_PADDR, 0);
83 		vm->pgd = paddr;
84 		vm->pgd_created = true;
85 	}
86 }
87 
88 static void _virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
89 			 uint64_t flags)
90 {
91 	uint8_t attr_idx = flags & 7;
92 	uint64_t *ptep;
93 
94 	TEST_ASSERT((vaddr % vm->page_size) == 0,
95 		"Virtual address not on page boundary,\n"
96 		"  vaddr: 0x%lx vm->page_size: 0x%x", vaddr, vm->page_size);
97 	TEST_ASSERT(sparsebit_is_set(vm->vpages_valid,
98 		(vaddr >> vm->page_shift)),
99 		"Invalid virtual address, vaddr: 0x%lx", vaddr);
100 	TEST_ASSERT((paddr % vm->page_size) == 0,
101 		"Physical address not on page boundary,\n"
102 		"  paddr: 0x%lx vm->page_size: 0x%x", paddr, vm->page_size);
103 	TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn,
104 		"Physical address beyond beyond maximum supported,\n"
105 		"  paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
106 		paddr, vm->max_gfn, vm->page_size);
107 
108 	ptep = addr_gpa2hva(vm, vm->pgd) + pgd_index(vm, vaddr) * 8;
109 	if (!*ptep)
110 		*ptep = vm_alloc_page_table(vm) | 3;
111 
112 	switch (vm->pgtable_levels) {
113 	case 4:
114 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pud_index(vm, vaddr) * 8;
115 		if (!*ptep)
116 			*ptep = vm_alloc_page_table(vm) | 3;
117 		/* fall through */
118 	case 3:
119 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pmd_index(vm, vaddr) * 8;
120 		if (!*ptep)
121 			*ptep = vm_alloc_page_table(vm) | 3;
122 		/* fall through */
123 	case 2:
124 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pte_index(vm, vaddr) * 8;
125 		break;
126 	default:
127 		TEST_FAIL("Page table levels must be 2, 3, or 4");
128 	}
129 
130 	*ptep = paddr | 3;
131 	*ptep |= (attr_idx << 2) | (1 << 10) /* Access Flag */;
132 }
133 
134 void virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr)
135 {
136 	uint64_t attr_idx = 4; /* NORMAL (See DEFAULT_MAIR_EL1) */
137 
138 	_virt_pg_map(vm, vaddr, paddr, attr_idx);
139 }
140 
141 vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
142 {
143 	uint64_t *ptep;
144 
145 	if (!vm->pgd_created)
146 		goto unmapped_gva;
147 
148 	ptep = addr_gpa2hva(vm, vm->pgd) + pgd_index(vm, gva) * 8;
149 	if (!ptep)
150 		goto unmapped_gva;
151 
152 	switch (vm->pgtable_levels) {
153 	case 4:
154 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pud_index(vm, gva) * 8;
155 		if (!ptep)
156 			goto unmapped_gva;
157 		/* fall through */
158 	case 3:
159 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pmd_index(vm, gva) * 8;
160 		if (!ptep)
161 			goto unmapped_gva;
162 		/* fall through */
163 	case 2:
164 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pte_index(vm, gva) * 8;
165 		if (!ptep)
166 			goto unmapped_gva;
167 		break;
168 	default:
169 		TEST_FAIL("Page table levels must be 2, 3, or 4");
170 	}
171 
172 	return pte_addr(vm, *ptep) + (gva & (vm->page_size - 1));
173 
174 unmapped_gva:
175 	TEST_FAIL("No mapping for vm virtual address, gva: 0x%lx", gva);
176 	exit(1);
177 }
178 
179 static void pte_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent, uint64_t page, int level)
180 {
181 #ifdef DEBUG
182 	static const char * const type[] = { "", "pud", "pmd", "pte" };
183 	uint64_t pte, *ptep;
184 
185 	if (level == 4)
186 		return;
187 
188 	for (pte = page; pte < page + ptrs_per_pte(vm) * 8; pte += 8) {
189 		ptep = addr_gpa2hva(vm, pte);
190 		if (!*ptep)
191 			continue;
192 		fprintf(stream, "%*s%s: %lx: %lx at %p\n", indent, "", type[level], pte, *ptep, ptep);
193 		pte_dump(stream, vm, indent + 1, pte_addr(vm, *ptep), level + 1);
194 	}
195 #endif
196 }
197 
198 void virt_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
199 {
200 	int level = 4 - (vm->pgtable_levels - 1);
201 	uint64_t pgd, *ptep;
202 
203 	if (!vm->pgd_created)
204 		return;
205 
206 	for (pgd = vm->pgd; pgd < vm->pgd + ptrs_per_pgd(vm) * 8; pgd += 8) {
207 		ptep = addr_gpa2hva(vm, pgd);
208 		if (!*ptep)
209 			continue;
210 		fprintf(stream, "%*spgd: %lx: %lx at %p\n", indent, "", pgd, *ptep, ptep);
211 		pte_dump(stream, vm, indent + 1, pte_addr(vm, *ptep), level);
212 	}
213 }
214 
215 void aarch64_vcpu_setup(struct kvm_vm *vm, uint32_t vcpuid, struct kvm_vcpu_init *init)
216 {
217 	struct kvm_vcpu_init default_init = { .target = -1, };
218 	uint64_t sctlr_el1, tcr_el1;
219 
220 	if (!init)
221 		init = &default_init;
222 
223 	if (init->target == -1) {
224 		struct kvm_vcpu_init preferred;
225 		vm_ioctl(vm, KVM_ARM_PREFERRED_TARGET, &preferred);
226 		init->target = preferred.target;
227 	}
228 
229 	vcpu_ioctl(vm, vcpuid, KVM_ARM_VCPU_INIT, init);
230 
231 	/*
232 	 * Enable FP/ASIMD to avoid trapping when accessing Q0-Q15
233 	 * registers, which the variable argument list macros do.
234 	 */
235 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_CPACR_EL1), 3 << 20);
236 
237 	get_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_SCTLR_EL1), &sctlr_el1);
238 	get_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_TCR_EL1), &tcr_el1);
239 
240 	switch (vm->mode) {
241 	case VM_MODE_P52V48_4K:
242 		TEST_FAIL("AArch64 does not support 4K sized pages "
243 			  "with 52-bit physical address ranges");
244 	case VM_MODE_PXXV48_4K:
245 		TEST_FAIL("AArch64 does not support 4K sized pages "
246 			  "with ANY-bit physical address ranges");
247 	case VM_MODE_P52V48_64K:
248 		tcr_el1 |= 1ul << 14; /* TG0 = 64KB */
249 		tcr_el1 |= 6ul << 32; /* IPS = 52 bits */
250 		break;
251 	case VM_MODE_P48V48_4K:
252 		tcr_el1 |= 0ul << 14; /* TG0 = 4KB */
253 		tcr_el1 |= 5ul << 32; /* IPS = 48 bits */
254 		break;
255 	case VM_MODE_P48V48_64K:
256 		tcr_el1 |= 1ul << 14; /* TG0 = 64KB */
257 		tcr_el1 |= 5ul << 32; /* IPS = 48 bits */
258 		break;
259 	case VM_MODE_P40V48_4K:
260 		tcr_el1 |= 0ul << 14; /* TG0 = 4KB */
261 		tcr_el1 |= 2ul << 32; /* IPS = 40 bits */
262 		break;
263 	case VM_MODE_P40V48_64K:
264 		tcr_el1 |= 1ul << 14; /* TG0 = 64KB */
265 		tcr_el1 |= 2ul << 32; /* IPS = 40 bits */
266 		break;
267 	default:
268 		TEST_FAIL("Unknown guest mode, mode: 0x%x", vm->mode);
269 	}
270 
271 	sctlr_el1 |= (1 << 0) | (1 << 2) | (1 << 12) /* M | C | I */;
272 	/* TCR_EL1 |= IRGN0:WBWA | ORGN0:WBWA | SH0:Inner-Shareable */;
273 	tcr_el1 |= (1 << 8) | (1 << 10) | (3 << 12);
274 	tcr_el1 |= (64 - vm->va_bits) /* T0SZ */;
275 
276 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_SCTLR_EL1), sctlr_el1);
277 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_TCR_EL1), tcr_el1);
278 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_MAIR_EL1), DEFAULT_MAIR_EL1);
279 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_TTBR0_EL1), vm->pgd);
280 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_TPIDR_EL1), vcpuid);
281 }
282 
283 void vcpu_dump(FILE *stream, struct kvm_vm *vm, uint32_t vcpuid, uint8_t indent)
284 {
285 	uint64_t pstate, pc;
286 
287 	get_reg(vm, vcpuid, ARM64_CORE_REG(regs.pstate), &pstate);
288 	get_reg(vm, vcpuid, ARM64_CORE_REG(regs.pc), &pc);
289 
290 	fprintf(stream, "%*spstate: 0x%.16lx pc: 0x%.16lx\n",
291 		indent, "", pstate, pc);
292 }
293 
294 void aarch64_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid,
295 			      struct kvm_vcpu_init *init, void *guest_code)
296 {
297 	size_t stack_size = vm->page_size == 4096 ?
298 					DEFAULT_STACK_PGS * vm->page_size :
299 					vm->page_size;
300 	uint64_t stack_vaddr = vm_vaddr_alloc(vm, stack_size,
301 					      DEFAULT_ARM64_GUEST_STACK_VADDR_MIN);
302 
303 	vm_vcpu_add(vm, vcpuid);
304 	aarch64_vcpu_setup(vm, vcpuid, init);
305 
306 	set_reg(vm, vcpuid, ARM64_CORE_REG(sp_el1), stack_vaddr + stack_size);
307 	set_reg(vm, vcpuid, ARM64_CORE_REG(regs.pc), (uint64_t)guest_code);
308 }
309 
310 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code)
311 {
312 	aarch64_vcpu_add_default(vm, vcpuid, NULL, guest_code);
313 }
314 
315 void vcpu_args_set(struct kvm_vm *vm, uint32_t vcpuid, unsigned int num, ...)
316 {
317 	va_list ap;
318 	int i;
319 
320 	TEST_ASSERT(num >= 1 && num <= 8, "Unsupported number of args,\n"
321 		    "  num: %u\n", num);
322 
323 	va_start(ap, num);
324 
325 	for (i = 0; i < num; i++) {
326 		set_reg(vm, vcpuid, ARM64_CORE_REG(regs.regs[i]),
327 			va_arg(ap, uint64_t));
328 	}
329 
330 	va_end(ap);
331 }
332 
333 void kvm_exit_unexpected_exception(int vector, uint64_t ec, bool valid_ec)
334 {
335 	ucall(UCALL_UNHANDLED, 3, vector, ec, valid_ec);
336 	while (1)
337 		;
338 }
339 
340 void assert_on_unhandled_exception(struct kvm_vm *vm, uint32_t vcpuid)
341 {
342 	struct ucall uc;
343 
344 	if (get_ucall(vm, vcpuid, &uc) != UCALL_UNHANDLED)
345 		return;
346 
347 	if (uc.args[2]) /* valid_ec */ {
348 		assert(VECTOR_IS_SYNC(uc.args[0]));
349 		TEST_FAIL("Unexpected exception (vector:0x%lx, ec:0x%lx)",
350 			  uc.args[0], uc.args[1]);
351 	} else {
352 		assert(!VECTOR_IS_SYNC(uc.args[0]));
353 		TEST_FAIL("Unexpected exception (vector:0x%lx)",
354 			  uc.args[0]);
355 	}
356 }
357 
358 struct handlers {
359 	handler_fn exception_handlers[VECTOR_NUM][ESR_EC_NUM];
360 };
361 
362 void vcpu_init_descriptor_tables(struct kvm_vm *vm, uint32_t vcpuid)
363 {
364 	extern char vectors;
365 
366 	set_reg(vm, vcpuid, KVM_ARM64_SYS_REG(SYS_VBAR_EL1), (uint64_t)&vectors);
367 }
368 
369 void route_exception(struct ex_regs *regs, int vector)
370 {
371 	struct handlers *handlers = (struct handlers *)exception_handlers;
372 	bool valid_ec;
373 	int ec = 0;
374 
375 	switch (vector) {
376 	case VECTOR_SYNC_CURRENT:
377 	case VECTOR_SYNC_LOWER_64:
378 		ec = (read_sysreg(esr_el1) >> ESR_EC_SHIFT) & ESR_EC_MASK;
379 		valid_ec = true;
380 		break;
381 	case VECTOR_IRQ_CURRENT:
382 	case VECTOR_IRQ_LOWER_64:
383 	case VECTOR_FIQ_CURRENT:
384 	case VECTOR_FIQ_LOWER_64:
385 	case VECTOR_ERROR_CURRENT:
386 	case VECTOR_ERROR_LOWER_64:
387 		ec = 0;
388 		valid_ec = false;
389 		break;
390 	default:
391 		valid_ec = false;
392 		goto unexpected_exception;
393 	}
394 
395 	if (handlers && handlers->exception_handlers[vector][ec])
396 		return handlers->exception_handlers[vector][ec](regs);
397 
398 unexpected_exception:
399 	kvm_exit_unexpected_exception(vector, ec, valid_ec);
400 }
401 
402 void vm_init_descriptor_tables(struct kvm_vm *vm)
403 {
404 	vm->handlers = vm_vaddr_alloc(vm, sizeof(struct handlers),
405 			vm->page_size);
406 
407 	*(vm_vaddr_t *)addr_gva2hva(vm, (vm_vaddr_t)(&exception_handlers)) = vm->handlers;
408 }
409 
410 void vm_install_sync_handler(struct kvm_vm *vm, int vector, int ec,
411 			 void (*handler)(struct ex_regs *))
412 {
413 	struct handlers *handlers = addr_gva2hva(vm, vm->handlers);
414 
415 	assert(VECTOR_IS_SYNC(vector));
416 	assert(vector < VECTOR_NUM);
417 	assert(ec < ESR_EC_NUM);
418 	handlers->exception_handlers[vector][ec] = handler;
419 }
420 
421 void vm_install_exception_handler(struct kvm_vm *vm, int vector,
422 			 void (*handler)(struct ex_regs *))
423 {
424 	struct handlers *handlers = addr_gva2hva(vm, vm->handlers);
425 
426 	assert(!VECTOR_IS_SYNC(vector));
427 	assert(vector < VECTOR_NUM);
428 	handlers->exception_handlers[vector][0] = handler;
429 }
430 
431 uint32_t guest_get_vcpuid(void)
432 {
433 	return read_sysreg(tpidr_el1);
434 }
435