| 1 | #include <asm.h> |
| 2 | #include <console/printf.h> |
| 3 | #include <elf.h> |
| 4 | #include <mem/page.h> |
| 5 | #include <mem/pmm.h> |
| 6 | #include <mem/vmm.h> |
| 7 | #include <stdbool.h> |
| 8 | #include <stdint.h> |
| 9 | #include <string.h> |
| 10 | |
| 11 | #define ELF_MAGIC 0x464C457F // "\x7FELF" |
| 12 | |
| 13 | #define USER_STACK_TOP 0x7FFFFFF000 |
| 14 | #define USER_STACK_SIZE (16 * PAGE_SIZE) |
| 15 | |
| 16 | uint64_t elf_load(const void *elf_data) { |
| 17 | struct elf64_ehdr *ehdr = (struct elf64_ehdr *) elf_data; |
| 18 | |
| 19 | if (ehdr->ident.magic != ELF_MAGIC || ehdr->ident.class != 2) { |
| 20 | printf(format: "Invalid ELF64\n" ); |
| 21 | return 0; |
| 22 | } |
| 23 | |
| 24 | struct elf64_phdr *phdrs = |
| 25 | (struct elf64_phdr *) ((uint8_t *) elf_data + ehdr->phoff); |
| 26 | |
| 27 | for (int i = 0; i < ehdr->phnum; i++) { |
| 28 | struct elf64_phdr *ph = &phdrs[i]; |
| 29 | if (ph->type != 1) |
| 30 | continue; |
| 31 | |
| 32 | uint64_t va_start = PAGE_ALIGN_DOWN(ph->vaddr); |
| 33 | uint64_t va_end = PAGE_ALIGN_UP(ph->vaddr + ph->memsz); |
| 34 | |
| 35 | for (uint64_t va = va_start; va < va_end; va += 0x1000) { |
| 36 | uint64_t phys = pmm_alloc_page(); |
| 37 | vmm_map_page(va, phys, |
| 38 | PAGE_PRESENT | PAGE_USER_ALLOWED | PAGE_WRITE, |
| 39 | VMM_FLAG_NONE); |
| 40 | } |
| 41 | } |
| 42 | |
| 43 | return ehdr->entry; |
| 44 | } |
| 45 | |
| 46 | void elf_map(uintptr_t user_pml4_phys, void *elf_data) { |
| 47 | struct elf64_ehdr *ehdr = (struct elf64_ehdr *) elf_data; |
| 48 | struct elf64_phdr *phdrs = |
| 49 | (struct elf64_phdr *) ((uint8_t *) elf_data + ehdr->phoff); |
| 50 | |
| 51 | for (int i = 0; i < ehdr->phnum; i++) { |
| 52 | struct elf64_phdr *ph = &phdrs[i]; |
| 53 | if (ph->type != PT_LOAD) |
| 54 | continue; |
| 55 | |
| 56 | uintptr_t seg_vaddr_start = PAGE_ALIGN_DOWN(ph->vaddr); |
| 57 | uintptr_t seg_vaddr_end = PAGE_ALIGN_UP(ph->vaddr + ph->memsz); |
| 58 | uintptr_t file_start = ph->offset; |
| 59 | uintptr_t file_end = ph->offset + ph->filesz; |
| 60 | |
| 61 | uint64_t flags = PAGE_USER_ALLOWED | PAGE_PRESENT; |
| 62 | if (ph->flags & PF_W) |
| 63 | flags |= PAGE_WRITE; |
| 64 | if (!(ph->flags & PF_X)) |
| 65 | flags |= PAGE_XD; |
| 66 | |
| 67 | for (vaddr_t vaddr = seg_vaddr_start; vaddr < seg_vaddr_end; |
| 68 | vaddr += PAGE_SIZE) { |
| 69 | |
| 70 | uintptr_t phys = pmm_alloc_page(); |
| 71 | if (!phys) |
| 72 | panic("Failed to allocate page for user ELF segment" ); |
| 73 | |
| 74 | void *phys_mapped = vmm_map_bump(phys, PAGE_SIZE, 0); |
| 75 | memset(phys_mapped, 0, PAGE_SIZE); |
| 76 | |
| 77 | uintptr_t offset_in_seg = vaddr - seg_vaddr_start; |
| 78 | uintptr_t file_offset_in_seg = |
| 79 | offset_in_seg + (ph->vaddr - seg_vaddr_start); |
| 80 | uintptr_t file_pos = file_start + file_offset_in_seg; |
| 81 | |
| 82 | uintptr_t page_offset = 0; |
| 83 | if (vaddr == seg_vaddr_start) |
| 84 | page_offset = ph->vaddr & (PAGE_SIZE - 1); |
| 85 | |
| 86 | uint64_t to_copy = 0; |
| 87 | if (file_pos < file_end) { |
| 88 | uint64_t bytes_remaining = file_end - file_pos; |
| 89 | to_copy = PAGE_SIZE - page_offset; |
| 90 | if (bytes_remaining < to_copy) |
| 91 | to_copy = bytes_remaining; |
| 92 | |
| 93 | memcpy((uint8_t *) phys_mapped + page_offset, |
| 94 | (uint8_t *) elf_data + file_pos, to_copy); |
| 95 | } |
| 96 | |
| 97 | vmm_map_page_user(vmm_phys_to_pml4(user_pml4_phys), vaddr, phys, |
| 98 | flags, VMM_FLAG_USER); |
| 99 | } |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | uintptr_t map_user_stack(uintptr_t user_pml4_phys) { |
| 104 | uintptr_t stack_bottom = USER_STACK_TOP - USER_STACK_SIZE; |
| 105 | |
| 106 | for (uintptr_t v = stack_bottom; v < USER_STACK_TOP; v += PAGE_SIZE) { |
| 107 | uintptr_t phys = pmm_alloc_page(); |
| 108 | if (!phys) |
| 109 | panic("Failed to alloc user stack" ); |
| 110 | |
| 111 | vmm_map_page_user(vmm_phys_to_pml4(user_pml4_phys), v, phys, |
| 112 | PAGE_WRITE | PAGE_USER_ALLOWED | PAGE_PRESENT, |
| 113 | VMM_FLAG_USER); |
| 114 | } |
| 115 | |
| 116 | return USER_STACK_TOP - 0x2000; |
| 117 | } |
| 118 | |
| 119 | void syscall_setup(void *syscall_entry) { |
| 120 | uint64_t efer = rdmsr(msr: 0xC0000080); |
| 121 | efer |= (1 << 0); // SCE: Enable syscall/sysret |
| 122 | wrmsr(msr: 0xC0000080, value: efer); |
| 123 | |
| 124 | wrmsr(msr: 0xC0000082, value: (uint64_t) syscall_entry); |
| 125 | |
| 126 | wrmsr(msr: 0xC0000084, value: (1 << 9)); |
| 127 | |
| 128 | uint64_t star = ((uint64_t) 0x08 << 32) | ((uint64_t) 0x28 << 48); |
| 129 | wrmsr(msr: 0xC0000081, value: star); |
| 130 | } |
| 131 | |
| 132 | __attribute__((noreturn)) void |
| 133 | enter_userspace(uintptr_t entry_point, uintptr_t user_stack_top, |
| 134 | uint16_t user_cs, uint16_t user_ss, uintptr_t user_pml4_phys) { |
| 135 | asm volatile("mov %0, %%cr3" : : "r" (user_pml4_phys) : "memory" ); |
| 136 | |
| 137 | uint64_t rflags; |
| 138 | asm volatile("pushfq; popq %0" : "=r" (rflags)); |
| 139 | rflags |= (1 << 9); |
| 140 | |
| 141 | asm volatile("cli\n" |
| 142 | "pushq %0\n" // SS |
| 143 | "pushq %1\n" // RSP |
| 144 | "pushq %2\n" // RFLAGS |
| 145 | "pushq %3\n" // CS |
| 146 | "pushq %4\n" // RIP |
| 147 | "iretq\n" |
| 148 | : |
| 149 | : "r" ((uint64_t) user_ss), "r" (user_stack_top), "r" (rflags), |
| 150 | "r" ((uint64_t) user_cs), "r" (entry_point) |
| 151 | : "memory" ); |
| 152 | |
| 153 | __builtin_unreachable(); |
| 154 | } |
| 155 | |