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