]>
Commit | Line | Data |
---|---|---|
d7e28ffe RR |
1 | #ifndef _LGUEST_H |
2 | #define _LGUEST_H | |
3 | ||
d7e28ffe RR |
4 | #ifndef __ASSEMBLY__ |
5 | #include <linux/types.h> | |
6 | #include <linux/init.h> | |
7 | #include <linux/stringify.h> | |
d7e28ffe RR |
8 | #include <linux/lguest.h> |
9 | #include <linux/lguest_launcher.h> | |
10 | #include <linux/wait.h> | |
ca94f2bd | 11 | #include <linux/hrtimer.h> |
d7e28ffe | 12 | #include <linux/err.h> |
d7e28ffe | 13 | |
625efab1 | 14 | #include <asm/lguest.h> |
d7e28ffe RR |
15 | |
16 | void free_pagetables(void); | |
17 | int init_pagetables(struct page **switcher_page, unsigned int pages); | |
18 | ||
1842f23c | 19 | struct pgdir { |
ee3db0f2 | 20 | unsigned long gpgdir; |
df29f43e | 21 | pgd_t *pgdir; |
d7e28ffe RR |
22 | }; |
23 | ||
d7e28ffe | 24 | /* We have two pages shared with guests, per cpu. */ |
1842f23c | 25 | struct lguest_pages { |
d7e28ffe RR |
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 | ||
badb1e04 GOC |
39 | struct lg_cpu { |
40 | unsigned int id; | |
41 | struct lguest *lg; | |
66686c2a GOC |
42 | struct task_struct *tsk; |
43 | struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */ | |
73044f05 | 44 | |
4665ac8e GOC |
45 | u32 cr2; |
46 | int ts; | |
47 | u32 esp1; | |
f086122b | 48 | u16 ss1; |
4665ac8e | 49 | |
ae3749dc GOC |
50 | /* Bitmap of what has changed: see CHANGED_* above. */ |
51 | int changed; | |
52 | ||
5e232f4f GOC |
53 | unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */ |
54 | ||
2e04ef76 | 55 | /* At end of a page shared mapped over lguest_pages in guest. */ |
a53a35a8 GOC |
56 | unsigned long regs_page; |
57 | struct lguest_regs *regs; | |
58 | ||
f34f8c5f GOC |
59 | struct lguest_pages *last_pages; |
60 | ||
2e04ef76 | 61 | int cpu_pgd; /* Which pgd this cpu is currently using */ |
1713608f | 62 | |
73044f05 GOC |
63 | /* If a hypercall was asked for, this points to the arguments. */ |
64 | struct hcall_args *hcall; | |
65 | u32 next_hcall; | |
ad8d8f3b GOC |
66 | |
67 | /* Virtual clock device */ | |
68 | struct hrtimer hrt; | |
177e449d | 69 | |
5dac051b | 70 | /* Did the Guest tell us to halt? */ |
66686c2a GOC |
71 | int halted; |
72 | ||
177e449d GOC |
73 | /* Pending virtual interrupts */ |
74 | DECLARE_BITMAP(irqs_pending, LGUEST_IRQS); | |
fc708b3e GOC |
75 | |
76 | struct lg_cpu_arch arch; | |
badb1e04 GOC |
77 | }; |
78 | ||
df60aeef RR |
79 | struct lg_eventfd { |
80 | unsigned long addr; | |
13389010 | 81 | struct eventfd_ctx *event; |
df60aeef RR |
82 | }; |
83 | ||
84 | struct lg_eventfd_map { | |
85 | unsigned int num; | |
86 | struct lg_eventfd map[]; | |
87 | }; | |
88 | ||
d7e28ffe | 89 | /* The private info the thread maintains about the guest. */ |
1842f23c | 90 | struct lguest { |
d7e28ffe | 91 | struct lguest_data __user *lguest_data; |
badb1e04 GOC |
92 | struct lg_cpu cpus[NR_CPUS]; |
93 | unsigned int nr_cpus; | |
94 | ||
d7e28ffe | 95 | u32 pfn_limit; |
2e04ef76 RR |
96 | |
97 | /* | |
98 | * This provides the offset to the base of guest-physical memory in the | |
99 | * Launcher. | |
100 | */ | |
3c6b5bfa | 101 | void __user *mem_base; |
47436aa4 | 102 | unsigned long kernel_address; |
d7e28ffe | 103 | |
d7e28ffe RR |
104 | struct pgdir pgdirs[4]; |
105 | ||
d7e28ffe | 106 | unsigned long noirq_start, noirq_end; |
d7e28ffe RR |
107 | |
108 | unsigned int stack_pages; | |
109 | u32 tsc_khz; | |
110 | ||
df60aeef RR |
111 | struct lg_eventfd_map *eventfds; |
112 | ||
d7e28ffe RR |
113 | /* Dead? */ |
114 | const char *dead; | |
d7e28ffe RR |
115 | }; |
116 | ||
d7e28ffe RR |
117 | extern struct mutex lguest_lock; |
118 | ||
119 | /* core.c: */ | |
df1693ab MZ |
120 | bool lguest_address_ok(const struct lguest *lg, |
121 | unsigned long addr, unsigned long len); | |
382ac6b3 GOC |
122 | void __lgread(struct lg_cpu *, void *, unsigned long, unsigned); |
123 | void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); | |
2d37f94a | 124 | |
2e04ef76 RR |
125 | /*H:035 |
126 | * Using memory-copy operations like that is usually inconvient, so we | |
2d37f94a RR |
127 | * have the following helper macros which read and write a specific type (often |
128 | * an unsigned long). | |
129 | * | |
2e04ef76 RR |
130 | * This reads into a variable of the given type then returns that. |
131 | */ | |
382ac6b3 GOC |
132 | #define lgread(cpu, addr, type) \ |
133 | ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; }) | |
2d37f94a RR |
134 | |
135 | /* This checks that the variable is of the given type, then writes it out. */ | |
382ac6b3 | 136 | #define lgwrite(cpu, addr, type, val) \ |
2d37f94a RR |
137 | do { \ |
138 | typecheck(type, val); \ | |
382ac6b3 | 139 | __lgwrite((cpu), (addr), &(val), sizeof(val)); \ |
2d37f94a RR |
140 | } while(0) |
141 | /* (end of memory access helper routines) :*/ | |
142 | ||
d0953d42 | 143 | int run_guest(struct lg_cpu *cpu, unsigned long __user *user); |
d7e28ffe | 144 | |
2e04ef76 RR |
145 | /* |
146 | * Helper macros to obtain the first 12 or the last 20 bits, this is only the | |
df29f43e | 147 | * first step in the migration to the kernel types. pte_pfn is already defined |
2e04ef76 RR |
148 | * in the kernel. |
149 | */ | |
df29f43e | 150 | #define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK) |
df29f43e | 151 | #define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT) |
acdd0b62 MZ |
152 | #define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK) |
153 | #define pmd_pfn(x) (pmd_val(x) >> PAGE_SHIFT) | |
d7e28ffe RR |
154 | |
155 | /* interrupts_and_traps.c: */ | |
a32a8813 RR |
156 | unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more); |
157 | void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more); | |
9f155a9b | 158 | void set_interrupt(struct lg_cpu *cpu, unsigned int irq); |
df1693ab | 159 | bool deliver_trap(struct lg_cpu *cpu, unsigned int num); |
fc708b3e GOC |
160 | void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i, |
161 | u32 low, u32 hi); | |
4665ac8e GOC |
162 | void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages); |
163 | void pin_stack_pages(struct lg_cpu *cpu); | |
d7e28ffe RR |
164 | void setup_default_idt_entries(struct lguest_ro_state *state, |
165 | const unsigned long *def); | |
fc708b3e | 166 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, |
d7e28ffe | 167 | const unsigned long *def); |
ad8d8f3b | 168 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta); |
df60aeef | 169 | bool send_notify_to_eventfd(struct lg_cpu *cpu); |
ad8d8f3b | 170 | void init_clockdev(struct lg_cpu *cpu); |
c18acd73 RR |
171 | bool check_syscall_vector(struct lguest *lg); |
172 | int init_interrupts(void); | |
173 | void free_interrupts(void); | |
d7e28ffe RR |
174 | |
175 | /* segments.c: */ | |
176 | void setup_default_gdt_entries(struct lguest_ro_state *state); | |
fc708b3e | 177 | void setup_guest_gdt(struct lg_cpu *cpu); |
a489f0b5 RR |
178 | void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i, |
179 | u32 low, u32 hi); | |
fc708b3e GOC |
180 | void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array); |
181 | void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt); | |
182 | void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt); | |
d7e28ffe RR |
183 | |
184 | /* page_tables.c: */ | |
58a24566 | 185 | int init_guest_pagetable(struct lguest *lg); |
d7e28ffe | 186 | void free_guest_pagetable(struct lguest *lg); |
4665ac8e | 187 | void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable); |
ebe0ba84 | 188 | void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i); |
acdd0b62 MZ |
189 | #ifdef CONFIG_X86_PAE |
190 | void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i); | |
191 | #endif | |
4665ac8e | 192 | void guest_pagetable_clear_all(struct lg_cpu *cpu); |
1713608f | 193 | void guest_pagetable_flush_user(struct lg_cpu *cpu); |
382ac6b3 | 194 | void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir, |
df29f43e | 195 | unsigned long vaddr, pte_t val); |
0c78441c | 196 | void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages); |
df1693ab | 197 | bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode); |
1713608f GOC |
198 | void pin_page(struct lg_cpu *cpu, unsigned long vaddr); |
199 | unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr); | |
382ac6b3 | 200 | void page_table_guest_data_init(struct lg_cpu *cpu); |
d7e28ffe | 201 | |
625efab1 JS |
202 | /* <arch>/core.c: */ |
203 | void lguest_arch_host_init(void); | |
204 | void lguest_arch_host_fini(void); | |
d0953d42 | 205 | void lguest_arch_run_guest(struct lg_cpu *cpu); |
73044f05 GOC |
206 | void lguest_arch_handle_trap(struct lg_cpu *cpu); |
207 | int lguest_arch_init_hypercalls(struct lg_cpu *cpu); | |
208 | int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args); | |
a53a35a8 | 209 | void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start); |
625efab1 JS |
210 | |
211 | /* <arch>/switcher.S: */ | |
212 | extern char start_switcher_text[], end_switcher_text[], switch_to_guest[]; | |
213 | ||
d7e28ffe RR |
214 | /* lguest_user.c: */ |
215 | int lguest_device_init(void); | |
216 | void lguest_device_remove(void); | |
217 | ||
d7e28ffe | 218 | /* hypercalls.c: */ |
73044f05 | 219 | void do_hypercalls(struct lg_cpu *cpu); |
382ac6b3 | 220 | void write_timestamp(struct lg_cpu *cpu); |
d7e28ffe | 221 | |
dde79789 RR |
222 | /*L:035 |
223 | * Let's step aside for the moment, to study one important routine that's used | |
224 | * widely in the Host code. | |
225 | * | |
e1e72965 | 226 | * There are many cases where the Guest can do something invalid, like pass crap |
dde79789 RR |
227 | * to a hypercall. Since only the Guest kernel can make hypercalls, it's quite |
228 | * acceptable to simply terminate the Guest and give the Launcher a nicely | |
229 | * formatted reason. It's also simpler for the Guest itself, which doesn't | |
230 | * need to check most hypercalls for "success"; if you're still running, it | |
231 | * succeeded. | |
232 | * | |
233 | * Once this is called, the Guest will never run again, so most Host code can | |
234 | * call this then continue as if nothing had happened. This means many | |
235 | * functions don't have to explicitly return an error code, which keeps the | |
236 | * code simple. | |
237 | * | |
238 | * It also means that this can be called more than once: only the first one is | |
239 | * remembered. The only trick is that we still need to kill the Guest even if | |
240 | * we can't allocate memory to store the reason. Linux has a neat way of | |
241 | * packing error codes into invalid pointers, so we use that here. | |
242 | * | |
243 | * Like any macro which uses an "if", it is safely wrapped in a run-once "do { | |
244 | * } while(0)". | |
245 | */ | |
382ac6b3 | 246 | #define kill_guest(cpu, fmt...) \ |
d7e28ffe | 247 | do { \ |
382ac6b3 GOC |
248 | if (!(cpu)->lg->dead) { \ |
249 | (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt); \ | |
250 | if (!(cpu)->lg->dead) \ | |
251 | (cpu)->lg->dead = ERR_PTR(-ENOMEM); \ | |
d7e28ffe RR |
252 | } \ |
253 | } while(0) | |
dde79789 | 254 | /* (End of aside) :*/ |
d7e28ffe | 255 | |
d7e28ffe RR |
256 | #endif /* __ASSEMBLY__ */ |
257 | #endif /* _LGUEST_H */ |