1 |
62 |
marcus.erl |
#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/err.h>
|
12 |
|
|
#include <asm/semaphore.h>
|
13 |
|
|
|
14 |
|
|
#include <asm/lguest.h>
|
15 |
|
|
|
16 |
|
|
void free_pagetables(void);
|
17 |
|
|
int init_pagetables(struct page **switcher_page, unsigned int pages);
|
18 |
|
|
|
19 |
|
|
struct pgdir
|
20 |
|
|
{
|
21 |
|
|
unsigned long gpgdir;
|
22 |
|
|
pgd_t *pgdir;
|
23 |
|
|
};
|
24 |
|
|
|
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 |
|
|
|
41 |
|
|
/* The private info the thread maintains about the guest. */
|
42 |
|
|
struct lguest
|
43 |
|
|
{
|
44 |
|
|
/* At end of a page shared mapped over lguest_pages in guest. */
|
45 |
|
|
unsigned long regs_page;
|
46 |
|
|
struct lguest_regs *regs;
|
47 |
|
|
struct lguest_data __user *lguest_data;
|
48 |
|
|
struct task_struct *tsk;
|
49 |
|
|
struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */
|
50 |
|
|
u32 pfn_limit;
|
51 |
|
|
/* This provides the offset to the base of guest-physical
|
52 |
|
|
* memory in the Launcher. */
|
53 |
|
|
void __user *mem_base;
|
54 |
|
|
unsigned long kernel_address;
|
55 |
|
|
u32 cr2;
|
56 |
|
|
int halted;
|
57 |
|
|
int ts;
|
58 |
|
|
u32 next_hcall;
|
59 |
|
|
u32 esp1;
|
60 |
|
|
u8 ss1;
|
61 |
|
|
|
62 |
|
|
/* If a hypercall was asked for, this points to the arguments. */
|
63 |
|
|
struct hcall_args *hcall;
|
64 |
|
|
|
65 |
|
|
/* Do we need to stop what we're doing and return to userspace? */
|
66 |
|
|
int break_out;
|
67 |
|
|
wait_queue_head_t break_wq;
|
68 |
|
|
|
69 |
|
|
/* Bitmap of what has changed: see CHANGED_* above. */
|
70 |
|
|
int changed;
|
71 |
|
|
struct lguest_pages *last_pages;
|
72 |
|
|
|
73 |
|
|
/* We keep a small number of these. */
|
74 |
|
|
u32 pgdidx;
|
75 |
|
|
struct pgdir pgdirs[4];
|
76 |
|
|
|
77 |
|
|
unsigned long noirq_start, noirq_end;
|
78 |
|
|
unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */
|
79 |
|
|
|
80 |
|
|
unsigned int stack_pages;
|
81 |
|
|
u32 tsc_khz;
|
82 |
|
|
|
83 |
|
|
/* Dead? */
|
84 |
|
|
const char *dead;
|
85 |
|
|
|
86 |
|
|
struct lguest_arch arch;
|
87 |
|
|
|
88 |
|
|
/* Virtual clock device */
|
89 |
|
|
struct hrtimer hrt;
|
90 |
|
|
|
91 |
|
|
/* Pending virtual interrupts */
|
92 |
|
|
DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
|
93 |
|
|
};
|
94 |
|
|
|
95 |
|
|
extern struct mutex lguest_lock;
|
96 |
|
|
|
97 |
|
|
/* core.c: */
|
98 |
|
|
int lguest_address_ok(const struct lguest *lg,
|
99 |
|
|
unsigned long addr, unsigned long len);
|
100 |
|
|
void __lgread(struct lguest *, void *, unsigned long, unsigned);
|
101 |
|
|
void __lgwrite(struct lguest *, unsigned long, const void *, unsigned);
|
102 |
|
|
|
103 |
|
|
/*H:035 Using memory-copy operations like that is usually inconvient, so we
|
104 |
|
|
* have the following helper macros which read and write a specific type (often
|
105 |
|
|
* an unsigned long).
|
106 |
|
|
*
|
107 |
|
|
* This reads into a variable of the given type then returns that. */
|
108 |
|
|
#define lgread(lg, addr, type) \
|
109 |
|
|
({ type _v; __lgread((lg), &_v, (addr), sizeof(_v)); _v; })
|
110 |
|
|
|
111 |
|
|
/* This checks that the variable is of the given type, then writes it out. */
|
112 |
|
|
#define lgwrite(lg, addr, type, val) \
|
113 |
|
|
do { \
|
114 |
|
|
typecheck(type, val); \
|
115 |
|
|
__lgwrite((lg), (addr), &(val), sizeof(val)); \
|
116 |
|
|
} while(0)
|
117 |
|
|
/* (end of memory access helper routines) :*/
|
118 |
|
|
|
119 |
|
|
int run_guest(struct lguest *lg, unsigned long __user *user);
|
120 |
|
|
|
121 |
|
|
/* Helper macros to obtain the first 12 or the last 20 bits, this is only the
|
122 |
|
|
* first step in the migration to the kernel types. pte_pfn is already defined
|
123 |
|
|
* in the kernel. */
|
124 |
|
|
#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK)
|
125 |
|
|
#define pte_flags(x) (pte_val(x) & ~PAGE_MASK)
|
126 |
|
|
#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT)
|
127 |
|
|
|
128 |
|
|
/* interrupts_and_traps.c: */
|
129 |
|
|
void maybe_do_interrupt(struct lguest *lg);
|
130 |
|
|
int deliver_trap(struct lguest *lg, unsigned int num);
|
131 |
|
|
void load_guest_idt_entry(struct lguest *lg, unsigned int i, u32 low, u32 hi);
|
132 |
|
|
void guest_set_stack(struct lguest *lg, u32 seg, u32 esp, unsigned int pages);
|
133 |
|
|
void pin_stack_pages(struct lguest *lg);
|
134 |
|
|
void setup_default_idt_entries(struct lguest_ro_state *state,
|
135 |
|
|
const unsigned long *def);
|
136 |
|
|
void copy_traps(const struct lguest *lg, struct desc_struct *idt,
|
137 |
|
|
const unsigned long *def);
|
138 |
|
|
void guest_set_clockevent(struct lguest *lg, unsigned long delta);
|
139 |
|
|
void init_clockdev(struct lguest *lg);
|
140 |
|
|
bool check_syscall_vector(struct lguest *lg);
|
141 |
|
|
int init_interrupts(void);
|
142 |
|
|
void free_interrupts(void);
|
143 |
|
|
|
144 |
|
|
/* segments.c: */
|
145 |
|
|
void setup_default_gdt_entries(struct lguest_ro_state *state);
|
146 |
|
|
void setup_guest_gdt(struct lguest *lg);
|
147 |
|
|
void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num);
|
148 |
|
|
void guest_load_tls(struct lguest *lg, unsigned long tls_array);
|
149 |
|
|
void copy_gdt(const struct lguest *lg, struct desc_struct *gdt);
|
150 |
|
|
void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt);
|
151 |
|
|
|
152 |
|
|
/* page_tables.c: */
|
153 |
|
|
int init_guest_pagetable(struct lguest *lg, unsigned long pgtable);
|
154 |
|
|
void free_guest_pagetable(struct lguest *lg);
|
155 |
|
|
void guest_new_pagetable(struct lguest *lg, unsigned long pgtable);
|
156 |
|
|
void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
|
157 |
|
|
void guest_pagetable_clear_all(struct lguest *lg);
|
158 |
|
|
void guest_pagetable_flush_user(struct lguest *lg);
|
159 |
|
|
void guest_set_pte(struct lguest *lg, unsigned long gpgdir,
|
160 |
|
|
unsigned long vaddr, pte_t val);
|
161 |
|
|
void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages);
|
162 |
|
|
int demand_page(struct lguest *info, unsigned long cr2, int errcode);
|
163 |
|
|
void pin_page(struct lguest *lg, unsigned long vaddr);
|
164 |
|
|
unsigned long guest_pa(struct lguest *lg, unsigned long vaddr);
|
165 |
|
|
void page_table_guest_data_init(struct lguest *lg);
|
166 |
|
|
|
167 |
|
|
/* <arch>/core.c: */
|
168 |
|
|
void lguest_arch_host_init(void);
|
169 |
|
|
void lguest_arch_host_fini(void);
|
170 |
|
|
void lguest_arch_run_guest(struct lguest *lg);
|
171 |
|
|
void lguest_arch_handle_trap(struct lguest *lg);
|
172 |
|
|
int lguest_arch_init_hypercalls(struct lguest *lg);
|
173 |
|
|
int lguest_arch_do_hcall(struct lguest *lg, struct hcall_args *args);
|
174 |
|
|
void lguest_arch_setup_regs(struct lguest *lg, unsigned long start);
|
175 |
|
|
|
176 |
|
|
/* <arch>/switcher.S: */
|
177 |
|
|
extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
|
178 |
|
|
|
179 |
|
|
/* lguest_user.c: */
|
180 |
|
|
int lguest_device_init(void);
|
181 |
|
|
void lguest_device_remove(void);
|
182 |
|
|
|
183 |
|
|
/* hypercalls.c: */
|
184 |
|
|
void do_hypercalls(struct lguest *lg);
|
185 |
|
|
void write_timestamp(struct lguest *lg);
|
186 |
|
|
|
187 |
|
|
/*L:035
|
188 |
|
|
* Let's step aside for the moment, to study one important routine that's used
|
189 |
|
|
* widely in the Host code.
|
190 |
|
|
*
|
191 |
|
|
* There are many cases where the Guest can do something invalid, like pass crap
|
192 |
|
|
* to a hypercall. Since only the Guest kernel can make hypercalls, it's quite
|
193 |
|
|
* acceptable to simply terminate the Guest and give the Launcher a nicely
|
194 |
|
|
* formatted reason. It's also simpler for the Guest itself, which doesn't
|
195 |
|
|
* need to check most hypercalls for "success"; if you're still running, it
|
196 |
|
|
* succeeded.
|
197 |
|
|
*
|
198 |
|
|
* Once this is called, the Guest will never run again, so most Host code can
|
199 |
|
|
* call this then continue as if nothing had happened. This means many
|
200 |
|
|
* functions don't have to explicitly return an error code, which keeps the
|
201 |
|
|
* code simple.
|
202 |
|
|
*
|
203 |
|
|
* It also means that this can be called more than once: only the first one is
|
204 |
|
|
* remembered. The only trick is that we still need to kill the Guest even if
|
205 |
|
|
* we can't allocate memory to store the reason. Linux has a neat way of
|
206 |
|
|
* packing error codes into invalid pointers, so we use that here.
|
207 |
|
|
*
|
208 |
|
|
* Like any macro which uses an "if", it is safely wrapped in a run-once "do {
|
209 |
|
|
* } while(0)".
|
210 |
|
|
*/
|
211 |
|
|
#define kill_guest(lg, fmt...) \
|
212 |
|
|
do { \
|
213 |
|
|
if (!(lg)->dead) { \
|
214 |
|
|
(lg)->dead = kasprintf(GFP_ATOMIC, fmt); \
|
215 |
|
|
if (!(lg)->dead) \
|
216 |
|
|
(lg)->dead = ERR_PTR(-ENOMEM); \
|
217 |
|
|
} \
|
218 |
|
|
} while(0)
|
219 |
|
|
/* (End of aside) :*/
|
220 |
|
|
|
221 |
|
|
#endif /* __ASSEMBLY__ */
|
222 |
|
|
#endif /* _LGUEST_H */
|