| 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> | 
|  | 13 | #include <asm/semaphore.h> | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 14 |  | 
| Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 15 | #include <asm/lguest.h> | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 16 |  | 
|  | 17 | void free_pagetables(void); | 
|  | 18 | int init_pagetables(struct page **switcher_page, unsigned int pages); | 
|  | 19 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 20 | struct pgdir | 
|  | 21 | { | 
| Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 22 | unsigned long gpgdir; | 
| Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 23 | pgd_t *pgdir; | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 24 | }; | 
|  | 25 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 26 | /* We have two pages shared with guests, per cpu.  */ | 
|  | 27 | struct lguest_pages | 
|  | 28 | { | 
|  | 29 | /* This is the stack page mapped rw in guest */ | 
|  | 30 | char spare[PAGE_SIZE - sizeof(struct lguest_regs)]; | 
|  | 31 | struct lguest_regs regs; | 
|  | 32 |  | 
|  | 33 | /* This is the host state & guest descriptor page, ro in guest */ | 
|  | 34 | struct lguest_ro_state state; | 
|  | 35 | } __attribute__((aligned(PAGE_SIZE))); | 
|  | 36 |  | 
|  | 37 | #define CHANGED_IDT		1 | 
|  | 38 | #define CHANGED_GDT		2 | 
|  | 39 | #define CHANGED_GDT_TLS		4 /* Actually a subset of CHANGED_GDT */ | 
|  | 40 | #define CHANGED_ALL	        3 | 
|  | 41 |  | 
| Glauber de Oliveira Costa | badb1e0 | 2008-01-07 11:05:22 -0200 | [diff] [blame] | 42 | struct lguest; | 
|  | 43 |  | 
|  | 44 | struct lg_cpu { | 
|  | 45 | unsigned int id; | 
|  | 46 | struct lguest *lg; | 
| Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 47 | struct task_struct *tsk; | 
|  | 48 | 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] | 49 |  | 
| Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 50 | u32 cr2; | 
|  | 51 | int ts; | 
|  | 52 | u32 esp1; | 
|  | 53 | u8 ss1; | 
|  | 54 |  | 
| Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 55 | /* Bitmap of what has changed: see CHANGED_* above. */ | 
|  | 56 | int changed; | 
|  | 57 |  | 
| Glauber de Oliveira Costa | 5e232f4 | 2008-01-07 11:05:36 -0200 | [diff] [blame] | 58 | unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */ | 
|  | 59 |  | 
| Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 60 | /* At end of a page shared mapped over lguest_pages in guest.  */ | 
|  | 61 | unsigned long regs_page; | 
|  | 62 | struct lguest_regs *regs; | 
|  | 63 |  | 
| Glauber de Oliveira Costa | f34f8c5 | 2008-01-17 19:13:26 -0200 | [diff] [blame] | 64 | struct lguest_pages *last_pages; | 
|  | 65 |  | 
| Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 66 | int cpu_pgd; /* which pgd this cpu is currently using */ | 
|  | 67 |  | 
| Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 68 | /* If a hypercall was asked for, this points to the arguments. */ | 
|  | 69 | struct hcall_args *hcall; | 
|  | 70 | u32 next_hcall; | 
| Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 71 |  | 
|  | 72 | /* Virtual clock device */ | 
|  | 73 | struct hrtimer hrt; | 
| Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 74 |  | 
| Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 75 | /* Do we need to stop what we're doing and return to userspace? */ | 
|  | 76 | int break_out; | 
|  | 77 | wait_queue_head_t break_wq; | 
|  | 78 | int halted; | 
|  | 79 |  | 
| Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 80 | /* Pending virtual interrupts */ | 
|  | 81 | DECLARE_BITMAP(irqs_pending, LGUEST_IRQS); | 
| Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 82 |  | 
|  | 83 | struct lg_cpu_arch arch; | 
| Glauber de Oliveira Costa | badb1e0 | 2008-01-07 11:05:22 -0200 | [diff] [blame] | 84 | }; | 
|  | 85 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 86 | /* The private info the thread maintains about the guest. */ | 
|  | 87 | struct lguest | 
|  | 88 | { | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 89 | struct lguest_data __user *lguest_data; | 
| Glauber de Oliveira Costa | badb1e0 | 2008-01-07 11:05:22 -0200 | [diff] [blame] | 90 | struct lg_cpu cpus[NR_CPUS]; | 
|  | 91 | unsigned int nr_cpus; | 
|  | 92 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 93 | u32 pfn_limit; | 
| Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 94 | /* This provides the offset to the base of guest-physical | 
|  | 95 | * memory in the Launcher. */ | 
|  | 96 | void __user *mem_base; | 
| Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 97 | unsigned long kernel_address; | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 98 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 99 | struct pgdir pgdirs[4]; | 
|  | 100 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 101 | unsigned long noirq_start, noirq_end; | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 102 |  | 
|  | 103 | unsigned int stack_pages; | 
|  | 104 | u32 tsc_khz; | 
|  | 105 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 106 | /* Dead? */ | 
|  | 107 | const char *dead; | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 108 | }; | 
|  | 109 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 110 | extern struct mutex lguest_lock; | 
|  | 111 |  | 
|  | 112 | /* core.c: */ | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 113 | int lguest_address_ok(const struct lguest *lg, | 
|  | 114 | unsigned long addr, unsigned long len); | 
| Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 115 | void __lgread(struct lg_cpu *, void *, unsigned long, unsigned); | 
|  | 116 | void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); | 
| Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 117 |  | 
| Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 118 | /*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] | 119 | * have the following helper macros which read and write a specific type (often | 
|  | 120 | * an unsigned long). | 
|  | 121 | * | 
|  | 122 | * 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] | 123 | #define lgread(cpu, addr, type)						\ | 
|  | 124 | ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; }) | 
| Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 125 |  | 
|  | 126 | /* 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] | 127 | #define lgwrite(cpu, addr, type, val)				\ | 
| Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 128 | do {							\ | 
|  | 129 | typecheck(type, val);				\ | 
| Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 130 | __lgwrite((cpu), (addr), &(val), sizeof(val));	\ | 
| Rusty Russell | 2d37f94 | 2007-10-22 11:24:24 +1000 | [diff] [blame] | 131 | } while(0) | 
|  | 132 | /* (end of memory access helper routines) :*/ | 
|  | 133 |  | 
| Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 134 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user); | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 135 |  | 
| Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 136 | /* Helper macros to obtain the first 12 or the last 20 bits, this is only the | 
|  | 137 | * first step in the migration to the kernel types.  pte_pfn is already defined | 
|  | 138 | * in the kernel. */ | 
|  | 139 | #define pgd_flags(x)	(pgd_val(x) & ~PAGE_MASK) | 
|  | 140 | #define pte_flags(x)	(pte_val(x) & ~PAGE_MASK) | 
|  | 141 | #define pgd_pfn(x)	(pgd_val(x) >> PAGE_SHIFT) | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 142 |  | 
|  | 143 | /* interrupts_and_traps.c: */ | 
| Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 144 | void maybe_do_interrupt(struct lg_cpu *cpu); | 
|  | 145 | int deliver_trap(struct lg_cpu *cpu, unsigned int num); | 
| Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 146 | void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i, | 
|  | 147 | u32 low, u32 hi); | 
| Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 148 | void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages); | 
|  | 149 | void pin_stack_pages(struct lg_cpu *cpu); | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 150 | void setup_default_idt_entries(struct lguest_ro_state *state, | 
|  | 151 | const unsigned long *def); | 
| Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 152 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 153 | const unsigned long *def); | 
| Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 154 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta); | 
|  | 155 | void init_clockdev(struct lg_cpu *cpu); | 
| Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 156 | bool check_syscall_vector(struct lguest *lg); | 
|  | 157 | int init_interrupts(void); | 
|  | 158 | void free_interrupts(void); | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 159 |  | 
|  | 160 | /* segments.c: */ | 
|  | 161 | void setup_default_gdt_entries(struct lguest_ro_state *state); | 
| Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 162 | void setup_guest_gdt(struct lg_cpu *cpu); | 
|  | 163 | void load_guest_gdt(struct lg_cpu *cpu, unsigned long table, u32 num); | 
|  | 164 | void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array); | 
|  | 165 | void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt); | 
|  | 166 | 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] | 167 |  | 
|  | 168 | /* page_tables.c: */ | 
|  | 169 | int init_guest_pagetable(struct lguest *lg, unsigned long pgtable); | 
|  | 170 | void free_guest_pagetable(struct lguest *lg); | 
| Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 171 | void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable); | 
| Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 172 | void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i); | 
| Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 173 | void guest_pagetable_clear_all(struct lg_cpu *cpu); | 
| Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 174 | void guest_pagetable_flush_user(struct lg_cpu *cpu); | 
| Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 175 | void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir, | 
| Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 176 | unsigned long vaddr, pte_t val); | 
| Glauber de Oliveira Costa | 0c78441 | 2008-01-07 11:05:30 -0200 | [diff] [blame] | 177 | void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages); | 
| Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 178 | int demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode); | 
|  | 179 | void pin_page(struct lg_cpu *cpu, unsigned long vaddr); | 
|  | 180 | 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] | 181 | void page_table_guest_data_init(struct lg_cpu *cpu); | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 182 |  | 
| Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 183 | /* <arch>/core.c: */ | 
|  | 184 | void lguest_arch_host_init(void); | 
|  | 185 | void lguest_arch_host_fini(void); | 
| Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 186 | void lguest_arch_run_guest(struct lg_cpu *cpu); | 
| Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 187 | void lguest_arch_handle_trap(struct lg_cpu *cpu); | 
|  | 188 | int lguest_arch_init_hypercalls(struct lg_cpu *cpu); | 
|  | 189 | 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] | 190 | void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start); | 
| Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 191 |  | 
|  | 192 | /* <arch>/switcher.S: */ | 
|  | 193 | extern char start_switcher_text[], end_switcher_text[], switch_to_guest[]; | 
|  | 194 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 195 | /* lguest_user.c: */ | 
|  | 196 | int lguest_device_init(void); | 
|  | 197 | void lguest_device_remove(void); | 
|  | 198 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 199 | /* hypercalls.c: */ | 
| Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 200 | void do_hypercalls(struct lg_cpu *cpu); | 
| Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 201 | void write_timestamp(struct lg_cpu *cpu); | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 202 |  | 
| Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 203 | /*L:035 | 
|  | 204 | * Let's step aside for the moment, to study one important routine that's used | 
|  | 205 | * widely in the Host code. | 
|  | 206 | * | 
| Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 207 | * 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] | 208 | * to a hypercall.  Since only the Guest kernel can make hypercalls, it's quite | 
|  | 209 | * acceptable to simply terminate the Guest and give the Launcher a nicely | 
|  | 210 | * formatted reason.  It's also simpler for the Guest itself, which doesn't | 
|  | 211 | * need to check most hypercalls for "success"; if you're still running, it | 
|  | 212 | * succeeded. | 
|  | 213 | * | 
|  | 214 | * Once this is called, the Guest will never run again, so most Host code can | 
|  | 215 | * call this then continue as if nothing had happened.  This means many | 
|  | 216 | * functions don't have to explicitly return an error code, which keeps the | 
|  | 217 | * code simple. | 
|  | 218 | * | 
|  | 219 | * It also means that this can be called more than once: only the first one is | 
|  | 220 | * remembered.  The only trick is that we still need to kill the Guest even if | 
|  | 221 | * we can't allocate memory to store the reason.  Linux has a neat way of | 
|  | 222 | * packing error codes into invalid pointers, so we use that here. | 
|  | 223 | * | 
|  | 224 | * Like any macro which uses an "if", it is safely wrapped in a run-once "do { | 
|  | 225 | * } while(0)". | 
|  | 226 | */ | 
| Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 227 | #define kill_guest(cpu, fmt...)					\ | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 228 | do {								\ | 
| Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 229 | if (!(cpu)->lg->dead) {					\ | 
|  | 230 | (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt);	\ | 
|  | 231 | if (!(cpu)->lg->dead)				\ | 
|  | 232 | (cpu)->lg->dead = ERR_PTR(-ENOMEM);	\ | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 233 | }							\ | 
|  | 234 | } while(0) | 
| Rusty Russell | dde7978 | 2007-07-26 10:41:03 -0700 | [diff] [blame] | 235 | /* (End of aside) :*/ | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 236 |  | 
| Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 237 | #endif	/* __ASSEMBLY__ */ | 
|  | 238 | #endif	/* _LGUEST_H */ |