| 1 | #ifndef _LGUEST_H |
| 2 | #define _LGUEST_H |
| 3 | |
| 4 | #ifndef __ASSEMBLY__ |
| 5 | #include <linux/types.h> |
| 6 | #include <linux/init.h> |
| 7 | #include <linux/stringify.h> |
| 8 | #include <linux/lguest.h> |
| 9 | #include <linux/lguest_launcher.h> |
| 10 | #include <linux/wait.h> |
| 11 | #include <linux/hrtimer.h> |
| 12 | #include <linux/err.h> |
| 13 | #include <linux/slab.h> |
| 14 | |
| 15 | #include <asm/lguest.h> |
| 16 | |
| 17 | struct pgdir { |
| 18 | unsigned long gpgdir; |
| 19 | bool switcher_mapped; |
| 20 | int last_host_cpu; |
| 21 | pgd_t *pgdir; |
| 22 | }; |
| 23 | |
| 24 | /* We have two pages shared with guests, per cpu. */ |
| 25 | struct lguest_pages { |
| 26 | /* This is the stack page mapped rw in guest */ |
| 27 | char spare[PAGE_SIZE - sizeof(struct lguest_regs)]; |
| 28 | struct lguest_regs regs; |
| 29 | |
| 30 | /* This is the host state & guest descriptor page, ro in guest */ |
| 31 | struct lguest_ro_state state; |
| 32 | } __attribute__((aligned(PAGE_SIZE))); |
| 33 | |
| 34 | #define CHANGED_IDT 1 |
| 35 | #define CHANGED_GDT 2 |
| 36 | #define CHANGED_GDT_TLS 4 /* Actually a subset of CHANGED_GDT */ |
| 37 | #define CHANGED_ALL 3 |
| 38 | |
| 39 | struct lg_cpu { |
| 40 | unsigned int id; |
| 41 | struct lguest *lg; |
| 42 | struct task_struct *tsk; |
| 43 | struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */ |
| 44 | |
| 45 | u32 cr2; |
| 46 | int ts; |
| 47 | u32 esp1; |
| 48 | u16 ss1; |
| 49 | |
| 50 | /* Bitmap of what has changed: see CHANGED_* above. */ |
| 51 | int changed; |
| 52 | |
| 53 | /* Pending operation. */ |
| 54 | struct lguest_pending pending; |
| 55 | |
| 56 | unsigned long *reg_read; /* register from LHREQ_GETREG */ |
| 57 | |
| 58 | /* At end of a page shared mapped over lguest_pages in guest. */ |
| 59 | unsigned long regs_page; |
| 60 | struct lguest_regs *regs; |
| 61 | |
| 62 | struct lguest_pages *last_pages; |
| 63 | |
| 64 | /* Initialization mode: linear map everything. */ |
| 65 | bool linear_pages; |
| 66 | int cpu_pgd; /* Which pgd this cpu is currently using */ |
| 67 | |
| 68 | /* If a hypercall was asked for, this points to the arguments. */ |
| 69 | struct hcall_args *hcall; |
| 70 | u32 next_hcall; |
| 71 | |
| 72 | /* Virtual clock device */ |
| 73 | struct hrtimer hrt; |
| 74 | |
| 75 | /* Did the Guest tell us to halt? */ |
| 76 | int halted; |
| 77 | |
| 78 | /* Pending virtual interrupts */ |
| 79 | DECLARE_BITMAP(irqs_pending, LGUEST_IRQS); |
| 80 | |
| 81 | struct lg_cpu_arch arch; |
| 82 | }; |
| 83 | |
| 84 | /* The private info the thread maintains about the guest. */ |
| 85 | struct lguest { |
| 86 | struct lguest_data __user *lguest_data; |
| 87 | struct lg_cpu cpus[NR_CPUS]; |
| 88 | unsigned int nr_cpus; |
| 89 | |
| 90 | /* Valid guest memory pages must be < this. */ |
| 91 | u32 pfn_limit; |
| 92 | |
| 93 | /* Device memory is >= pfn_limit and < device_limit. */ |
| 94 | u32 device_limit; |
| 95 | |
| 96 | /* |
| 97 | * This provides the offset to the base of guest-physical memory in the |
| 98 | * Launcher. |
| 99 | */ |
| 100 | void __user *mem_base; |
| 101 | unsigned long kernel_address; |
| 102 | |
| 103 | struct pgdir pgdirs[4]; |
| 104 | |
| 105 | unsigned long noirq_iret; |
| 106 | |
| 107 | unsigned int stack_pages; |
| 108 | u32 tsc_khz; |
| 109 | |
| 110 | /* Dead? */ |
| 111 | const char *dead; |
| 112 | }; |
| 113 | |
| 114 | extern struct mutex lguest_lock; |
| 115 | |
| 116 | /* core.c: */ |
| 117 | bool lguest_address_ok(const struct lguest *lg, |
| 118 | unsigned long addr, unsigned long len); |
| 119 | void __lgread(struct lg_cpu *, void *, unsigned long, unsigned); |
| 120 | void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); |
| 121 | extern struct page **lg_switcher_pages; |
| 122 | |
| 123 | /*H:035 |
| 124 | * Using memory-copy operations like that is usually inconvient, so we |
| 125 | * have the following helper macros which read and write a specific type (often |
| 126 | * an unsigned long). |
| 127 | * |
| 128 | * This reads into a variable of the given type then returns that. |
| 129 | */ |
| 130 | #define lgread(cpu, addr, type) \ |
| 131 | ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; }) |
| 132 | |
| 133 | /* This checks that the variable is of the given type, then writes it out. */ |
| 134 | #define lgwrite(cpu, addr, type, val) \ |
| 135 | do { \ |
| 136 | typecheck(type, val); \ |
| 137 | __lgwrite((cpu), (addr), &(val), sizeof(val)); \ |
| 138 | } while(0) |
| 139 | /* (end of memory access helper routines) :*/ |
| 140 | |
| 141 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user); |
| 142 | |
| 143 | /* |
| 144 | * Helper macros to obtain the first 12 or the last 20 bits, this is only the |
| 145 | * first step in the migration to the kernel types. pte_pfn is already defined |
| 146 | * in the kernel. |
| 147 | */ |
| 148 | #define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK) |
| 149 | #define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT) |
| 150 | #define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK) |
| 151 | #define pmd_pfn(x) (pmd_val(x) >> PAGE_SHIFT) |
| 152 | |
| 153 | /* interrupts_and_traps.c: */ |
| 154 | unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more); |
| 155 | void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more); |
| 156 | void set_interrupt(struct lg_cpu *cpu, unsigned int irq); |
| 157 | bool deliver_trap(struct lg_cpu *cpu, unsigned int num); |
| 158 | void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i, |
| 159 | u32 low, u32 hi); |
| 160 | void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages); |
| 161 | void pin_stack_pages(struct lg_cpu *cpu); |
| 162 | void setup_default_idt_entries(struct lguest_ro_state *state, |
| 163 | const unsigned long *def); |
| 164 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, |
| 165 | const unsigned long *def); |
| 166 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta); |
| 167 | bool send_notify_to_eventfd(struct lg_cpu *cpu); |
| 168 | void init_clockdev(struct lg_cpu *cpu); |
| 169 | bool check_syscall_vector(struct lguest *lg); |
| 170 | bool could_be_syscall(unsigned int num); |
| 171 | int init_interrupts(void); |
| 172 | void free_interrupts(void); |
| 173 | |
| 174 | /* segments.c: */ |
| 175 | void setup_default_gdt_entries(struct lguest_ro_state *state); |
| 176 | void setup_guest_gdt(struct lg_cpu *cpu); |
| 177 | void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i, |
| 178 | u32 low, u32 hi); |
| 179 | void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array); |
| 180 | void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt); |
| 181 | void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt); |
| 182 | |
| 183 | /* page_tables.c: */ |
| 184 | int init_guest_pagetable(struct lguest *lg); |
| 185 | void free_guest_pagetable(struct lguest *lg); |
| 186 | void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable); |
| 187 | void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i); |
| 188 | #ifdef CONFIG_X86_PAE |
| 189 | void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i); |
| 190 | #endif |
| 191 | void guest_pagetable_clear_all(struct lg_cpu *cpu); |
| 192 | void guest_pagetable_flush_user(struct lg_cpu *cpu); |
| 193 | void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir, |
| 194 | unsigned long vaddr, pte_t val); |
| 195 | void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages); |
| 196 | bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode, |
| 197 | unsigned long *iomem); |
| 198 | void pin_page(struct lg_cpu *cpu, unsigned long vaddr); |
| 199 | bool __guest_pa(struct lg_cpu *cpu, unsigned long vaddr, unsigned long *paddr); |
| 200 | unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr); |
| 201 | void page_table_guest_data_init(struct lg_cpu *cpu); |
| 202 | |
| 203 | /* <arch>/core.c: */ |
| 204 | void lguest_arch_host_init(void); |
| 205 | void lguest_arch_host_fini(void); |
| 206 | void lguest_arch_run_guest(struct lg_cpu *cpu); |
| 207 | void lguest_arch_handle_trap(struct lg_cpu *cpu); |
| 208 | int lguest_arch_init_hypercalls(struct lg_cpu *cpu); |
| 209 | int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args); |
| 210 | void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start); |
| 211 | unsigned long *lguest_arch_regptr(struct lg_cpu *cpu, size_t reg_off, bool any); |
| 212 | |
| 213 | /* <arch>/switcher.S: */ |
| 214 | extern char start_switcher_text[], end_switcher_text[], switch_to_guest[]; |
| 215 | |
| 216 | /* lguest_user.c: */ |
| 217 | int lguest_device_init(void); |
| 218 | void lguest_device_remove(void); |
| 219 | |
| 220 | /* hypercalls.c: */ |
| 221 | void do_hypercalls(struct lg_cpu *cpu); |
| 222 | void write_timestamp(struct lg_cpu *cpu); |
| 223 | |
| 224 | /*L:035 |
| 225 | * Let's step aside for the moment, to study one important routine that's used |
| 226 | * widely in the Host code. |
| 227 | * |
| 228 | * There are many cases where the Guest can do something invalid, like pass crap |
| 229 | * to a hypercall. Since only the Guest kernel can make hypercalls, it's quite |
| 230 | * acceptable to simply terminate the Guest and give the Launcher a nicely |
| 231 | * formatted reason. It's also simpler for the Guest itself, which doesn't |
| 232 | * need to check most hypercalls for "success"; if you're still running, it |
| 233 | * succeeded. |
| 234 | * |
| 235 | * Once this is called, the Guest will never run again, so most Host code can |
| 236 | * call this then continue as if nothing had happened. This means many |
| 237 | * functions don't have to explicitly return an error code, which keeps the |
| 238 | * code simple. |
| 239 | * |
| 240 | * It also means that this can be called more than once: only the first one is |
| 241 | * remembered. The only trick is that we still need to kill the Guest even if |
| 242 | * we can't allocate memory to store the reason. Linux has a neat way of |
| 243 | * packing error codes into invalid pointers, so we use that here. |
| 244 | * |
| 245 | * Like any macro which uses an "if", it is safely wrapped in a run-once "do { |
| 246 | * } while(0)". |
| 247 | */ |
| 248 | #define kill_guest(cpu, fmt...) \ |
| 249 | do { \ |
| 250 | if (!(cpu)->lg->dead) { \ |
| 251 | (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt); \ |
| 252 | if (!(cpu)->lg->dead) \ |
| 253 | (cpu)->lg->dead = ERR_PTR(-ENOMEM); \ |
| 254 | } \ |
| 255 | } while(0) |
| 256 | /* (End of aside) :*/ |
| 257 | |
| 258 | #endif /* __ASSEMBLY__ */ |
| 259 | #endif /* _LGUEST_H */ |