1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * AArch64 code
4  *
5  * Copyright (C) 2018, Red Hat, Inc.
6  */
7 
8 #define _GNU_SOURCE /* for program_invocation_name */
9 
10 #include <linux/compiler.h>
11 
12 #include "kvm_util.h"
13 #include "../kvm_util_internal.h"
14 #include "processor.h"
15 
16 #define KVM_GUEST_PAGE_TABLE_MIN_PADDR		0x180000
17 #define DEFAULT_ARM64_GUEST_STACK_VADDR_MIN	0xac0000
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, uint32_t pgd_memslot)
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, pgd_memslot);
83 		vm->pgd = paddr;
84 		vm->pgd_created = true;
85 	}
86 }
87 
88 void _virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
89 		  uint32_t pgd_memslot, 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_phy_page_alloc(vm, KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot);
111 		*ptep |= 3;
112 	}
113 
114 	switch (vm->pgtable_levels) {
115 	case 4:
116 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pud_index(vm, vaddr) * 8;
117 		if (!*ptep) {
118 			*ptep = vm_phy_page_alloc(vm, KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot);
119 			*ptep |= 3;
120 		}
121 		/* fall through */
122 	case 3:
123 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pmd_index(vm, vaddr) * 8;
124 		if (!*ptep) {
125 			*ptep = vm_phy_page_alloc(vm, KVM_GUEST_PAGE_TABLE_MIN_PADDR, pgd_memslot);
126 			*ptep |= 3;
127 		}
128 		/* fall through */
129 	case 2:
130 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pte_index(vm, vaddr) * 8;
131 		break;
132 	default:
133 		TEST_ASSERT(false, "Page table levels must be 2, 3, or 4");
134 	}
135 
136 	*ptep = paddr | 3;
137 	*ptep |= (attr_idx << 2) | (1 << 10) /* Access Flag */;
138 }
139 
140 void virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
141 		 uint32_t pgd_memslot)
142 {
143 	uint64_t attr_idx = 4; /* NORMAL (See DEFAULT_MAIR_EL1) */
144 
145 	_virt_pg_map(vm, vaddr, paddr, pgd_memslot, attr_idx);
146 }
147 
148 vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
149 {
150 	uint64_t *ptep;
151 
152 	if (!vm->pgd_created)
153 		goto unmapped_gva;
154 
155 	ptep = addr_gpa2hva(vm, vm->pgd) + pgd_index(vm, gva) * 8;
156 	if (!ptep)
157 		goto unmapped_gva;
158 
159 	switch (vm->pgtable_levels) {
160 	case 4:
161 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pud_index(vm, gva) * 8;
162 		if (!ptep)
163 			goto unmapped_gva;
164 		/* fall through */
165 	case 3:
166 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pmd_index(vm, gva) * 8;
167 		if (!ptep)
168 			goto unmapped_gva;
169 		/* fall through */
170 	case 2:
171 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) + pte_index(vm, gva) * 8;
172 		if (!ptep)
173 			goto unmapped_gva;
174 		break;
175 	default:
176 		TEST_ASSERT(false, "Page table levels must be 2, 3, or 4");
177 	}
178 
179 	return pte_addr(vm, *ptep) + (gva & (vm->page_size - 1));
180 
181 unmapped_gva:
182 	TEST_ASSERT(false, "No mapping for vm virtual address, "
183 		    "gva: 0x%lx", gva);
184 	exit(1);
185 }
186 
187 static void pte_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent, uint64_t page, int level)
188 {
189 #ifdef DEBUG_VM
190 	static const char * const type[] = { "", "pud", "pmd", "pte" };
191 	uint64_t pte, *ptep;
192 
193 	if (level == 4)
194 		return;
195 
196 	for (pte = page; pte < page + ptrs_per_pte(vm) * 8; pte += 8) {
197 		ptep = addr_gpa2hva(vm, pte);
198 		if (!*ptep)
199 			continue;
200 		printf("%*s%s: %lx: %lx at %p\n", indent, "", type[level], pte, *ptep, ptep);
201 		pte_dump(stream, vm, indent + 1, pte_addr(vm, *ptep), level + 1);
202 	}
203 #endif
204 }
205 
206 void virt_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
207 {
208 	int level = 4 - (vm->pgtable_levels - 1);
209 	uint64_t pgd, *ptep;
210 
211 	if (!vm->pgd_created)
212 		return;
213 
214 	for (pgd = vm->pgd; pgd < vm->pgd + ptrs_per_pgd(vm) * 8; pgd += 8) {
215 		ptep = addr_gpa2hva(vm, pgd);
216 		if (!*ptep)
217 			continue;
218 		printf("%*spgd: %lx: %lx at %p\n", indent, "", pgd, *ptep, ptep);
219 		pte_dump(stream, vm, indent + 1, pte_addr(vm, *ptep), level);
220 	}
221 }
222 
223 struct kvm_vm *vm_create_default(uint32_t vcpuid, uint64_t extra_mem_pages,
224 				 void *guest_code)
225 {
226 	uint64_t ptrs_per_4k_pte = 512;
227 	uint64_t extra_pg_pages = (extra_mem_pages / ptrs_per_4k_pte) * 2;
228 	struct kvm_vm *vm;
229 
230 	vm = vm_create(VM_MODE_P40V48_4K, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages, O_RDWR);
231 
232 	kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
233 	vm_vcpu_add_default(vm, vcpuid, guest_code);
234 
235 	return vm;
236 }
237 
238 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code)
239 {
240 	size_t stack_size = vm->page_size == 4096 ?
241 					DEFAULT_STACK_PGS * vm->page_size :
242 					vm->page_size;
243 	uint64_t stack_vaddr = vm_vaddr_alloc(vm, stack_size,
244 					DEFAULT_ARM64_GUEST_STACK_VADDR_MIN, 0, 0);
245 
246 	vm_vcpu_add(vm, vcpuid, 0, 0);
247 
248 	set_reg(vm, vcpuid, ARM64_CORE_REG(sp_el1), stack_vaddr + stack_size);
249 	set_reg(vm, vcpuid, ARM64_CORE_REG(regs.pc), (uint64_t)guest_code);
250 }
251 
252 void vcpu_setup(struct kvm_vm *vm, int vcpuid, int pgd_memslot, int gdt_memslot)
253 {
254 	struct kvm_vcpu_init init;
255 	uint64_t sctlr_el1, tcr_el1;
256 
257 	memset(&init, 0, sizeof(init));
258 	init.target = KVM_ARM_TARGET_GENERIC_V8;
259 	vcpu_ioctl(vm, vcpuid, KVM_ARM_VCPU_INIT, &init);
260 
261 	/*
262 	 * Enable FP/ASIMD to avoid trapping when accessing Q0-Q15
263 	 * registers, which the variable argument list macros do.
264 	 */
265 	set_reg(vm, vcpuid, ARM64_SYS_REG(CPACR_EL1), 3 << 20);
266 
267 	get_reg(vm, vcpuid, ARM64_SYS_REG(SCTLR_EL1), &sctlr_el1);
268 	get_reg(vm, vcpuid, ARM64_SYS_REG(TCR_EL1), &tcr_el1);
269 
270 	switch (vm->mode) {
271 	case VM_MODE_P52V48_4K:
272 		TEST_ASSERT(false, "AArch64 does not support 4K sized pages "
273 				   "with 52-bit physical address ranges");
274 	case VM_MODE_P52V48_64K:
275 		tcr_el1 |= 1ul << 14; /* TG0 = 64KB */
276 		tcr_el1 |= 6ul << 32; /* IPS = 52 bits */
277 		break;
278 	case VM_MODE_P48V48_4K:
279 		tcr_el1 |= 0ul << 14; /* TG0 = 4KB */
280 		tcr_el1 |= 5ul << 32; /* IPS = 48 bits */
281 		break;
282 	case VM_MODE_P48V48_64K:
283 		tcr_el1 |= 1ul << 14; /* TG0 = 64KB */
284 		tcr_el1 |= 5ul << 32; /* IPS = 48 bits */
285 		break;
286 	case VM_MODE_P40V48_4K:
287 		tcr_el1 |= 0ul << 14; /* TG0 = 4KB */
288 		tcr_el1 |= 2ul << 32; /* IPS = 40 bits */
289 		break;
290 	case VM_MODE_P40V48_64K:
291 		tcr_el1 |= 1ul << 14; /* TG0 = 64KB */
292 		tcr_el1 |= 2ul << 32; /* IPS = 40 bits */
293 		break;
294 	default:
295 		TEST_ASSERT(false, "Unknown guest mode, mode: 0x%x", vm->mode);
296 	}
297 
298 	sctlr_el1 |= (1 << 0) | (1 << 2) | (1 << 12) /* M | C | I */;
299 	/* TCR_EL1 |= IRGN0:WBWA | ORGN0:WBWA | SH0:Inner-Shareable */;
300 	tcr_el1 |= (1 << 8) | (1 << 10) | (3 << 12);
301 	tcr_el1 |= (64 - vm->va_bits) /* T0SZ */;
302 
303 	set_reg(vm, vcpuid, ARM64_SYS_REG(SCTLR_EL1), sctlr_el1);
304 	set_reg(vm, vcpuid, ARM64_SYS_REG(TCR_EL1), tcr_el1);
305 	set_reg(vm, vcpuid, ARM64_SYS_REG(MAIR_EL1), DEFAULT_MAIR_EL1);
306 	set_reg(vm, vcpuid, ARM64_SYS_REG(TTBR0_EL1), vm->pgd);
307 }
308 
309 void vcpu_dump(FILE *stream, struct kvm_vm *vm, uint32_t vcpuid, uint8_t indent)
310 {
311 	uint64_t pstate, pc;
312 
313 	get_reg(vm, vcpuid, ARM64_CORE_REG(regs.pstate), &pstate);
314 	get_reg(vm, vcpuid, ARM64_CORE_REG(regs.pc), &pc);
315 
316 	fprintf(stream, "%*spstate: 0x%.16lx pc: 0x%.16lx\n",
317 		indent, "", pstate, pc);
318 }
319