Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1 | #ifndef _LGUEST_H |
| 2 | #define _LGUEST_H |
| 3 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 4 | #ifndef __ASSEMBLY__ |
| 5 | #include <linux/types.h> |
| 6 | #include <linux/init.h> |
| 7 | #include <linux/stringify.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 8 | #include <linux/lguest.h> |
| 9 | #include <linux/lguest_launcher.h> |
| 10 | #include <linux/wait.h> |
Glauber de Oliveira Costa | ca94f2b | 2008-01-18 23:59:07 -0200 | [diff] [blame] | 11 | #include <linux/hrtimer.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 12 | #include <linux/err.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 13 | |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 14 | #include <asm/lguest.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 15 | |
| 16 | void free_pagetables(void); |
| 17 | int init_pagetables(struct page **switcher_page, unsigned int pages); |
| 18 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 19 | struct pgdir |
| 20 | { |
Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 21 | unsigned long gpgdir; |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 22 | pgd_t *pgdir; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 23 | }; |
| 24 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 25 | /* We have two pages shared with guests, per cpu. */ |
| 26 | struct lguest_pages |
| 27 | { |
| 28 | /* This is the stack page mapped rw in guest */ |
| 29 | char spare[PAGE_SIZE - sizeof(struct lguest_regs)]; |
| 30 | struct lguest_regs regs; |
| 31 | |
| 32 | /* This is the host state & guest descriptor page, ro in guest */ |
| 33 | struct lguest_ro_state state; |
| 34 | } __attribute__((aligned(PAGE_SIZE))); |
| 35 | |
| 36 | #define CHANGED_IDT 1 |
| 37 | #define CHANGED_GDT 2 |
| 38 | #define CHANGED_GDT_TLS 4 /* Actually a subset of CHANGED_GDT */ |
| 39 | #define CHANGED_ALL 3 |
| 40 | |
Glauber de Oliveira Costa | badb1e0 | 2008-01-07 11:05:22 -0200 | [diff] [blame] | 41 | struct lguest; |
| 42 | |
| 43 | struct lg_cpu { |
| 44 | unsigned int id; |
| 45 | struct lguest *lg; |
Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 46 | struct task_struct *tsk; |
| 47 | struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */ |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 48 | |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 49 | u32 cr2; |
| 50 | int ts; |
| 51 | u32 esp1; |
| 52 | u8 ss1; |
| 53 | |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 54 | /* Bitmap of what has changed: see CHANGED_* above. */ |
| 55 | int changed; |
| 56 | |
Glauber de Oliveira Costa | 5e232f4 | 2008-01-07 11:05:36 -0200 | [diff] [blame] | 57 | unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */ |
| 58 | |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 59 | /* At end of a page shared mapped over lguest_pages in guest. */ |
| 60 | unsigned long regs_page; |
| 61 | struct lguest_regs *regs; |
| 62 | |
Glauber de Oliveira Costa | f34f8c5 | 2008-01-17 19:13:26 -0200 | [diff] [blame] | 63 | struct lguest_pages *last_pages; |
| 64 | |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 65 | int cpu_pgd; /* which pgd this cpu is currently using */ |
| 66 | |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 67 | /* If a hypercall was asked for, this points to the arguments. */ |
| 68 | struct hcall_args *hcall; |
| 69 | u32 next_hcall; |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 70 | |
| 71 | /* Virtual clock device */ |
| 72 | struct hrtimer hrt; |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 73 | |
Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 74 | /* Do we need to stop what we're doing and return to userspace? */ |
| 75 | int break_out; |
| 76 | wait_queue_head_t break_wq; |
| 77 | int halted; |
| 78 | |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 79 | /* Pending virtual interrupts */ |
| 80 | DECLARE_BITMAP(irqs_pending, LGUEST_IRQS); |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 81 | |
| 82 | struct lg_cpu_arch arch; |
Glauber de Oliveira Costa | badb1e0 | 2008-01-07 11:05:22 -0200 | [diff] [blame] | 83 | }; |
| 84 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 85 | /* The private info the thread maintains about the guest. */ |
| 86 | struct lguest |
| 87 | { |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 88 | struct lguest_data __user *lguest_data; |
Glauber de Oliveira Costa | badb1e0 | 2008-01-07 11:05:22 -0200 | [diff] [blame] | 89 | struct lg_cpu cpus[NR_CPUS]; |
| 90 | unsigned int nr_cpus; |
| 91 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 92 | u32 pfn_limit; |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 93 | /* This provides the offset to the base of guest-physical |
| 94 | * memory in the Launcher. */ |
| 95 | void __user *mem_base; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 96 | unsigned long kernel_address; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 97 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 98 | struct pgdir pgdirs[4]; |
| 99 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 100 | unsigned long noirq_start, noirq_end; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 101 | |
| 102 | unsigned int stack_pages; |
| 103 | u32 tsc_khz; |
| 104 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 105 | /* Dead? */ |
| 106 | const char *dead; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 107 | }; |
| 108 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 109 | extern struct mutex lguest_lock; |
| 110 | |
| 111 | /* core.c: */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 112 | bool lguest_address_ok(const struct lguest *lg, |
| 113 | unsigned long addr, unsigned long len); |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 114 | void __lgread(struct lg_cpu *, void *, unsigned long, unsigned); |
| 115 | void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); |
Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 116 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 117 | /*H:035 Using memory-copy operations like that is usually inconvient, so we |
Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 118 | * have the following helper macros which read and write a specific type (often |
| 119 | * an unsigned long). |
| 120 | * |
| 121 | * This reads into a variable of the given type then returns that. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 122 | #define lgread(cpu, addr, type) \ |
| 123 | ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; }) |
Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 124 | |
| 125 | /* This checks that the variable is of the given type, then writes it out. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 126 | #define lgwrite(cpu, addr, type, val) \ |
Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 127 | do { \ |
| 128 | typecheck(type, val); \ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 129 | __lgwrite((cpu), (addr), &(val), sizeof(val)); \ |
Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 130 | } while(0) |
| 131 | /* (end of memory access helper routines) :*/ |
| 132 | |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 133 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 134 | |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 135 | /* Helper macros to obtain the first 12 or the last 20 bits, this is only the |
| 136 | * first step in the migration to the kernel types. pte_pfn is already defined |
| 137 | * in the kernel. */ |
| 138 | #define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK) |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 139 | #define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 140 | |
| 141 | /* interrupts_and_traps.c: */ |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 142 | void maybe_do_interrupt(struct lg_cpu *cpu); |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 143 | bool deliver_trap(struct lg_cpu *cpu, unsigned int num); |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 144 | void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i, |
| 145 | u32 low, u32 hi); |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 146 | void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages); |
| 147 | void pin_stack_pages(struct lg_cpu *cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 148 | void setup_default_idt_entries(struct lguest_ro_state *state, |
| 149 | const unsigned long *def); |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 150 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 151 | const unsigned long *def); |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 152 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta); |
| 153 | void init_clockdev(struct lg_cpu *cpu); |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 154 | bool check_syscall_vector(struct lguest *lg); |
| 155 | int init_interrupts(void); |
| 156 | void free_interrupts(void); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 157 | |
| 158 | /* segments.c: */ |
| 159 | void setup_default_gdt_entries(struct lguest_ro_state *state); |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 160 | void setup_guest_gdt(struct lg_cpu *cpu); |
Rusty Russell | a489f0b | 2009-04-19 23:14:00 -0600 | [diff] [blame] | 161 | void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i, |
| 162 | u32 low, u32 hi); |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 163 | void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array); |
| 164 | void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt); |
| 165 | void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 166 | |
| 167 | /* page_tables.c: */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 168 | int init_guest_pagetable(struct lguest *lg); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 169 | void free_guest_pagetable(struct lguest *lg); |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 170 | void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable); |
Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 171 | void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i); |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 172 | void guest_pagetable_clear_all(struct lg_cpu *cpu); |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 173 | void guest_pagetable_flush_user(struct lg_cpu *cpu); |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 174 | void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir, |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 175 | unsigned long vaddr, pte_t val); |
Glauber de Oliveira Costa | 0c78441 | 2008-01-07 11:05:30 -0200 | [diff] [blame] | 176 | void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages); |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 177 | bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode); |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 178 | void pin_page(struct lg_cpu *cpu, unsigned long vaddr); |
| 179 | unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr); |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 180 | void page_table_guest_data_init(struct lg_cpu *cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 181 | |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 182 | /* <arch>/core.c: */ |
| 183 | void lguest_arch_host_init(void); |
| 184 | void lguest_arch_host_fini(void); |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 185 | void lguest_arch_run_guest(struct lg_cpu *cpu); |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 186 | void lguest_arch_handle_trap(struct lg_cpu *cpu); |
| 187 | int lguest_arch_init_hypercalls(struct lg_cpu *cpu); |
| 188 | int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args); |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 189 | void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 190 | |
| 191 | /* <arch>/switcher.S: */ |
| 192 | extern char start_switcher_text[], end_switcher_text[], switch_to_guest[]; |
| 193 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 194 | /* lguest_user.c: */ |
| 195 | int lguest_device_init(void); |
| 196 | void lguest_device_remove(void); |
| 197 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 198 | /* hypercalls.c: */ |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 199 | void do_hypercalls(struct lg_cpu *cpu); |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 200 | void write_timestamp(struct lg_cpu *cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 201 | |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 202 | /*L:035 |
| 203 | * Let's step aside for the moment, to study one important routine that's used |
| 204 | * widely in the Host code. |
| 205 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 206 | * There are many cases where the Guest can do something invalid, like pass crap |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 207 | * to a hypercall. Since only the Guest kernel can make hypercalls, it's quite |
| 208 | * acceptable to simply terminate the Guest and give the Launcher a nicely |
| 209 | * formatted reason. It's also simpler for the Guest itself, which doesn't |
| 210 | * need to check most hypercalls for "success"; if you're still running, it |
| 211 | * succeeded. |
| 212 | * |
| 213 | * Once this is called, the Guest will never run again, so most Host code can |
| 214 | * call this then continue as if nothing had happened. This means many |
| 215 | * functions don't have to explicitly return an error code, which keeps the |
| 216 | * code simple. |
| 217 | * |
| 218 | * It also means that this can be called more than once: only the first one is |
| 219 | * remembered. The only trick is that we still need to kill the Guest even if |
| 220 | * we can't allocate memory to store the reason. Linux has a neat way of |
| 221 | * packing error codes into invalid pointers, so we use that here. |
| 222 | * |
| 223 | * Like any macro which uses an "if", it is safely wrapped in a run-once "do { |
| 224 | * } while(0)". |
| 225 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 226 | #define kill_guest(cpu, fmt...) \ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 227 | do { \ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 228 | if (!(cpu)->lg->dead) { \ |
| 229 | (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt); \ |
| 230 | if (!(cpu)->lg->dead) \ |
| 231 | (cpu)->lg->dead = ERR_PTR(-ENOMEM); \ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 232 | } \ |
| 233 | } while(0) |
Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 234 | /* (End of aside) :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 235 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 236 | #endif /* __ASSEMBLY__ */ |
| 237 | #endif /* _LGUEST_H */ |