1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * RISC-V code
4  *
5  * Copyright (C) 2021 Western Digital Corporation or its affiliates.
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_RISCV_GUEST_STACK_VADDR_MIN	0xac0000
16 
17 static uint64_t page_align(struct kvm_vm *vm, uint64_t v)
18 {
19 	return (v + vm->page_size) & ~(vm->page_size - 1);
20 }
21 
22 static uint64_t pte_addr(struct kvm_vm *vm, uint64_t entry)
23 {
24 	return ((entry & PGTBL_PTE_ADDR_MASK) >> PGTBL_PTE_ADDR_SHIFT) <<
25 		PGTBL_PAGE_SIZE_SHIFT;
26 }
27 
28 static uint64_t ptrs_per_pte(struct kvm_vm *vm)
29 {
30 	return PGTBL_PAGE_SIZE / sizeof(uint64_t);
31 }
32 
33 static uint64_t pte_index_mask[] = {
34 	PGTBL_L0_INDEX_MASK,
35 	PGTBL_L1_INDEX_MASK,
36 	PGTBL_L2_INDEX_MASK,
37 	PGTBL_L3_INDEX_MASK,
38 };
39 
40 static uint32_t pte_index_shift[] = {
41 	PGTBL_L0_INDEX_SHIFT,
42 	PGTBL_L1_INDEX_SHIFT,
43 	PGTBL_L2_INDEX_SHIFT,
44 	PGTBL_L3_INDEX_SHIFT,
45 };
46 
47 static uint64_t pte_index(struct kvm_vm *vm, vm_vaddr_t gva, int level)
48 {
49 	TEST_ASSERT(level > -1,
50 		"Negative page table level (%d) not possible", level);
51 	TEST_ASSERT(level < vm->pgtable_levels,
52 		"Invalid page table level (%d)", level);
53 
54 	return (gva & pte_index_mask[level]) >> pte_index_shift[level];
55 }
56 
57 void virt_pgd_alloc(struct kvm_vm *vm)
58 {
59 	if (!vm->pgd_created) {
60 		vm_paddr_t paddr = vm_phy_pages_alloc(vm,
61 			page_align(vm, ptrs_per_pte(vm) * 8) / vm->page_size,
62 			KVM_GUEST_PAGE_TABLE_MIN_PADDR, 0);
63 		vm->pgd = paddr;
64 		vm->pgd_created = true;
65 	}
66 }
67 
68 void virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr)
69 {
70 	uint64_t *ptep, next_ppn;
71 	int level = vm->pgtable_levels - 1;
72 
73 	TEST_ASSERT((vaddr % vm->page_size) == 0,
74 		"Virtual address not on page boundary,\n"
75 		"  vaddr: 0x%lx vm->page_size: 0x%x", vaddr, vm->page_size);
76 	TEST_ASSERT(sparsebit_is_set(vm->vpages_valid,
77 		(vaddr >> vm->page_shift)),
78 		"Invalid virtual address, vaddr: 0x%lx", vaddr);
79 	TEST_ASSERT((paddr % vm->page_size) == 0,
80 		"Physical address not on page boundary,\n"
81 		"  paddr: 0x%lx vm->page_size: 0x%x", paddr, vm->page_size);
82 	TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn,
83 		"Physical address beyond maximum supported,\n"
84 		"  paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
85 		paddr, vm->max_gfn, vm->page_size);
86 
87 	ptep = addr_gpa2hva(vm, vm->pgd) + pte_index(vm, vaddr, level) * 8;
88 	if (!*ptep) {
89 		next_ppn = vm_alloc_page_table(vm) >> PGTBL_PAGE_SIZE_SHIFT;
90 		*ptep = (next_ppn << PGTBL_PTE_ADDR_SHIFT) |
91 			PGTBL_PTE_VALID_MASK;
92 	}
93 	level--;
94 
95 	while (level > -1) {
96 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) +
97 		       pte_index(vm, vaddr, level) * 8;
98 		if (!*ptep && level > 0) {
99 			next_ppn = vm_alloc_page_table(vm) >>
100 				   PGTBL_PAGE_SIZE_SHIFT;
101 			*ptep = (next_ppn << PGTBL_PTE_ADDR_SHIFT) |
102 				PGTBL_PTE_VALID_MASK;
103 		}
104 		level--;
105 	}
106 
107 	paddr = paddr >> PGTBL_PAGE_SIZE_SHIFT;
108 	*ptep = (paddr << PGTBL_PTE_ADDR_SHIFT) |
109 		PGTBL_PTE_PERM_MASK | PGTBL_PTE_VALID_MASK;
110 }
111 
112 vm_paddr_t addr_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
113 {
114 	uint64_t *ptep;
115 	int level = vm->pgtable_levels - 1;
116 
117 	if (!vm->pgd_created)
118 		goto unmapped_gva;
119 
120 	ptep = addr_gpa2hva(vm, vm->pgd) + pte_index(vm, gva, level) * 8;
121 	if (!ptep)
122 		goto unmapped_gva;
123 	level--;
124 
125 	while (level > -1) {
126 		ptep = addr_gpa2hva(vm, pte_addr(vm, *ptep)) +
127 		       pte_index(vm, gva, level) * 8;
128 		if (!ptep)
129 			goto unmapped_gva;
130 		level--;
131 	}
132 
133 	return pte_addr(vm, *ptep) + (gva & (vm->page_size - 1));
134 
135 unmapped_gva:
136 	TEST_FAIL("No mapping for vm virtual address gva: 0x%lx level: %d",
137 		  gva, level);
138 	exit(1);
139 }
140 
141 static void pte_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent,
142 		     uint64_t page, int level)
143 {
144 #ifdef DEBUG
145 	static const char *const type[] = { "pte", "pmd", "pud", "p4d"};
146 	uint64_t pte, *ptep;
147 
148 	if (level < 0)
149 		return;
150 
151 	for (pte = page; pte < page + ptrs_per_pte(vm) * 8; pte += 8) {
152 		ptep = addr_gpa2hva(vm, pte);
153 		if (!*ptep)
154 			continue;
155 		fprintf(stream, "%*s%s: %lx: %lx at %p\n", indent, "",
156 			type[level], pte, *ptep, ptep);
157 		pte_dump(stream, vm, indent + 1,
158 			 pte_addr(vm, *ptep), level - 1);
159 	}
160 #endif
161 }
162 
163 void virt_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
164 {
165 	int level = vm->pgtable_levels - 1;
166 	uint64_t pgd, *ptep;
167 
168 	if (!vm->pgd_created)
169 		return;
170 
171 	for (pgd = vm->pgd; pgd < vm->pgd + ptrs_per_pte(vm) * 8; pgd += 8) {
172 		ptep = addr_gpa2hva(vm, pgd);
173 		if (!*ptep)
174 			continue;
175 		fprintf(stream, "%*spgd: %lx: %lx at %p\n", indent, "",
176 			pgd, *ptep, ptep);
177 		pte_dump(stream, vm, indent + 1,
178 			 pte_addr(vm, *ptep), level - 1);
179 	}
180 }
181 
182 void riscv_vcpu_mmu_setup(struct kvm_vm *vm, int vcpuid)
183 {
184 	unsigned long satp;
185 
186 	/*
187 	 * The RISC-V Sv48 MMU mode supports 56-bit physical address
188 	 * for 48-bit virtual address with 4KB last level page size.
189 	 */
190 	switch (vm->mode) {
191 	case VM_MODE_P52V48_4K:
192 	case VM_MODE_P48V48_4K:
193 	case VM_MODE_P40V48_4K:
194 		break;
195 	default:
196 		TEST_FAIL("Unknown guest mode, mode: 0x%x", vm->mode);
197 	}
198 
199 	satp = (vm->pgd >> PGTBL_PAGE_SIZE_SHIFT) & SATP_PPN;
200 	satp |= SATP_MODE_48;
201 
202 	set_reg(vm, vcpuid, RISCV_CSR_REG(satp), satp);
203 }
204 
205 void vcpu_dump(FILE *stream, struct kvm_vm *vm, uint32_t vcpuid, uint8_t indent)
206 {
207 	struct kvm_riscv_core core;
208 
209 	get_reg(vm, vcpuid, RISCV_CORE_REG(mode), &core.mode);
210 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.pc), &core.regs.pc);
211 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.ra), &core.regs.ra);
212 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.sp), &core.regs.sp);
213 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.gp), &core.regs.gp);
214 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.tp), &core.regs.tp);
215 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t0), &core.regs.t0);
216 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t1), &core.regs.t1);
217 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t2), &core.regs.t2);
218 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s0), &core.regs.s0);
219 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s1), &core.regs.s1);
220 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a0), &core.regs.a0);
221 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a1), &core.regs.a1);
222 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a2), &core.regs.a2);
223 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a3), &core.regs.a3);
224 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a4), &core.regs.a4);
225 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a5), &core.regs.a5);
226 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a6), &core.regs.a6);
227 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.a7), &core.regs.a7);
228 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s2), &core.regs.s2);
229 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s3), &core.regs.s3);
230 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s4), &core.regs.s4);
231 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s5), &core.regs.s5);
232 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s6), &core.regs.s6);
233 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s7), &core.regs.s7);
234 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s8), &core.regs.s8);
235 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s9), &core.regs.s9);
236 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s10), &core.regs.s10);
237 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.s11), &core.regs.s11);
238 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t3), &core.regs.t3);
239 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t4), &core.regs.t4);
240 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t5), &core.regs.t5);
241 	get_reg(vm, vcpuid, RISCV_CORE_REG(regs.t6), &core.regs.t6);
242 
243 	fprintf(stream,
244 		" MODE:  0x%lx\n", core.mode);
245 	fprintf(stream,
246 		" PC: 0x%016lx   RA: 0x%016lx SP: 0x%016lx GP: 0x%016lx\n",
247 		core.regs.pc, core.regs.ra, core.regs.sp, core.regs.gp);
248 	fprintf(stream,
249 		" TP: 0x%016lx   T0: 0x%016lx T1: 0x%016lx T2: 0x%016lx\n",
250 		core.regs.tp, core.regs.t0, core.regs.t1, core.regs.t2);
251 	fprintf(stream,
252 		" S0: 0x%016lx   S1: 0x%016lx A0: 0x%016lx A1: 0x%016lx\n",
253 		core.regs.s0, core.regs.s1, core.regs.a0, core.regs.a1);
254 	fprintf(stream,
255 		" A2: 0x%016lx   A3: 0x%016lx A4: 0x%016lx A5: 0x%016lx\n",
256 		core.regs.a2, core.regs.a3, core.regs.a4, core.regs.a5);
257 	fprintf(stream,
258 		" A6: 0x%016lx   A7: 0x%016lx S2: 0x%016lx S3: 0x%016lx\n",
259 		core.regs.a6, core.regs.a7, core.regs.s2, core.regs.s3);
260 	fprintf(stream,
261 		" S4: 0x%016lx   S5: 0x%016lx S6: 0x%016lx S7: 0x%016lx\n",
262 		core.regs.s4, core.regs.s5, core.regs.s6, core.regs.s7);
263 	fprintf(stream,
264 		" S8: 0x%016lx   S9: 0x%016lx S10: 0x%016lx S11: 0x%016lx\n",
265 		core.regs.s8, core.regs.s9, core.regs.s10, core.regs.s11);
266 	fprintf(stream,
267 		" T3: 0x%016lx   T4: 0x%016lx T5: 0x%016lx T6: 0x%016lx\n",
268 		core.regs.t3, core.regs.t4, core.regs.t5, core.regs.t6);
269 }
270 
271 static void __aligned(16) guest_unexp_trap(void)
272 {
273 	sbi_ecall(KVM_RISCV_SELFTESTS_SBI_EXT,
274 		  KVM_RISCV_SELFTESTS_SBI_UNEXP,
275 		  0, 0, 0, 0, 0, 0);
276 }
277 
278 void vm_vcpu_add_default(struct kvm_vm *vm, uint32_t vcpuid, void *guest_code)
279 {
280 	int r;
281 	size_t stack_size = vm->page_size == 4096 ?
282 					DEFAULT_STACK_PGS * vm->page_size :
283 					vm->page_size;
284 	unsigned long stack_vaddr = vm_vaddr_alloc(vm, stack_size,
285 					DEFAULT_RISCV_GUEST_STACK_VADDR_MIN);
286 	unsigned long current_gp = 0;
287 	struct kvm_mp_state mps;
288 
289 	vm_vcpu_add(vm, vcpuid);
290 	riscv_vcpu_mmu_setup(vm, vcpuid);
291 
292 	/*
293 	 * With SBI HSM support in KVM RISC-V, all secondary VCPUs are
294 	 * powered-off by default so we ensure that all secondary VCPUs
295 	 * are powered-on using KVM_SET_MP_STATE ioctl().
296 	 */
297 	mps.mp_state = KVM_MP_STATE_RUNNABLE;
298 	r = _vcpu_ioctl(vm, vcpuid, KVM_SET_MP_STATE, &mps);
299 	TEST_ASSERT(!r, "IOCTL KVM_SET_MP_STATE failed (error %d)", r);
300 
301 	/* Setup global pointer of guest to be same as the host */
302 	asm volatile (
303 		"add %0, gp, zero" : "=r" (current_gp) : : "memory");
304 	set_reg(vm, vcpuid, RISCV_CORE_REG(regs.gp), current_gp);
305 
306 	/* Setup stack pointer and program counter of guest */
307 	set_reg(vm, vcpuid, RISCV_CORE_REG(regs.sp),
308 		stack_vaddr + stack_size);
309 	set_reg(vm, vcpuid, RISCV_CORE_REG(regs.pc),
310 		(unsigned long)guest_code);
311 
312 	/* Setup default exception vector of guest */
313 	set_reg(vm, vcpuid, RISCV_CSR_REG(stvec),
314 		(unsigned long)guest_unexp_trap);
315 }
316 
317 void vcpu_args_set(struct kvm_vm *vm, uint32_t vcpuid, unsigned int num, ...)
318 {
319 	va_list ap;
320 	uint64_t id = RISCV_CORE_REG(regs.a0);
321 	int i;
322 
323 	TEST_ASSERT(num >= 1 && num <= 8, "Unsupported number of args,\n"
324 		    "  num: %u\n", num);
325 
326 	va_start(ap, num);
327 
328 	for (i = 0; i < num; i++) {
329 		switch (i) {
330 		case 0:
331 			id = RISCV_CORE_REG(regs.a0);
332 			break;
333 		case 1:
334 			id = RISCV_CORE_REG(regs.a1);
335 			break;
336 		case 2:
337 			id = RISCV_CORE_REG(regs.a2);
338 			break;
339 		case 3:
340 			id = RISCV_CORE_REG(regs.a3);
341 			break;
342 		case 4:
343 			id = RISCV_CORE_REG(regs.a4);
344 			break;
345 		case 5:
346 			id = RISCV_CORE_REG(regs.a5);
347 			break;
348 		case 6:
349 			id = RISCV_CORE_REG(regs.a6);
350 			break;
351 		case 7:
352 			id = RISCV_CORE_REG(regs.a7);
353 			break;
354 		}
355 		set_reg(vm, vcpuid, id, va_arg(ap, uint64_t));
356 	}
357 
358 	va_end(ap);
359 }
360 
361 void assert_on_unhandled_exception(struct kvm_vm *vm, uint32_t vcpuid)
362 {
363 }
364