• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 	/* Initialization mode: linear map everything. */
63 	bool linear_pages;
64 	int cpu_pgd; /* Which pgd this cpu is currently using */
65 
66 	/* If a hypercall was asked for, this points to the arguments. */
67 	struct hcall_args *hcall;
68 	u32 next_hcall;
69 
70 	/* Virtual clock device */
71 	struct hrtimer hrt;
72 
73 	/* Did the Guest tell us to halt? */
74 	int halted;
75 
76 	/* Pending virtual interrupts */
77 	DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
78 
79 	struct lg_cpu_arch arch;
80 };
81 
82 struct lg_eventfd {
83 	unsigned long addr;
84 	struct eventfd_ctx *event;
85 };
86 
87 struct lg_eventfd_map {
88 	unsigned int num;
89 	struct lg_eventfd map[];
90 };
91 
92 /* The private info the thread maintains about the guest. */
93 struct lguest {
94 	struct lguest_data __user *lguest_data;
95 	struct lg_cpu cpus[NR_CPUS];
96 	unsigned int nr_cpus;
97 
98 	u32 pfn_limit;
99 
100 	/*
101 	 * This provides the offset to the base of guest-physical memory in the
102 	 * Launcher.
103 	 */
104 	void __user *mem_base;
105 	unsigned long kernel_address;
106 
107 	struct pgdir pgdirs[4];
108 
109 	unsigned long noirq_start, noirq_end;
110 
111 	unsigned int stack_pages;
112 	u32 tsc_khz;
113 
114 	struct lg_eventfd_map *eventfds;
115 
116 	/* Dead? */
117 	const char *dead;
118 };
119 
120 extern struct mutex lguest_lock;
121 
122 /* core.c: */
123 bool lguest_address_ok(const struct lguest *lg,
124 		       unsigned long addr, unsigned long len);
125 void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
126 void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
127 
128 /*H:035
129  * Using memory-copy operations like that is usually inconvient, so we
130  * have the following helper macros which read and write a specific type (often
131  * an unsigned long).
132  *
133  * This reads into a variable of the given type then returns that.
134  */
135 #define lgread(cpu, addr, type)						\
136 	({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; })
137 
138 /* This checks that the variable is of the given type, then writes it out. */
139 #define lgwrite(cpu, addr, type, val)				\
140 	do {							\
141 		typecheck(type, val);				\
142 		__lgwrite((cpu), (addr), &(val), sizeof(val));	\
143 	} while(0)
144 /* (end of memory access helper routines) :*/
145 
146 int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
147 
148 /*
149  * Helper macros to obtain the first 12 or the last 20 bits, this is only the
150  * first step in the migration to the kernel types.  pte_pfn is already defined
151  * in the kernel.
152  */
153 #define pgd_flags(x)	(pgd_val(x) & ~PAGE_MASK)
154 #define pgd_pfn(x)	(pgd_val(x) >> PAGE_SHIFT)
155 #define pmd_flags(x)    (pmd_val(x) & ~PAGE_MASK)
156 #define pmd_pfn(x)	(pmd_val(x) >> PAGE_SHIFT)
157 
158 /* interrupts_and_traps.c: */
159 unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more);
160 void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more);
161 void set_interrupt(struct lg_cpu *cpu, unsigned int irq);
162 bool deliver_trap(struct lg_cpu *cpu, unsigned int num);
163 void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i,
164 			  u32 low, u32 hi);
165 void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages);
166 void pin_stack_pages(struct lg_cpu *cpu);
167 void setup_default_idt_entries(struct lguest_ro_state *state,
168 			       const unsigned long *def);
169 void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
170 		const unsigned long *def);
171 void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta);
172 bool send_notify_to_eventfd(struct lg_cpu *cpu);
173 void init_clockdev(struct lg_cpu *cpu);
174 bool check_syscall_vector(struct lguest *lg);
175 int init_interrupts(void);
176 void free_interrupts(void);
177 
178 /* segments.c: */
179 void setup_default_gdt_entries(struct lguest_ro_state *state);
180 void setup_guest_gdt(struct lg_cpu *cpu);
181 void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i,
182 			  u32 low, u32 hi);
183 void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array);
184 void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt);
185 void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt);
186 
187 /* page_tables.c: */
188 int init_guest_pagetable(struct lguest *lg);
189 void free_guest_pagetable(struct lguest *lg);
190 void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable);
191 void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i);
192 #ifdef CONFIG_X86_PAE
193 void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
194 #endif
195 void guest_pagetable_clear_all(struct lg_cpu *cpu);
196 void guest_pagetable_flush_user(struct lg_cpu *cpu);
197 void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir,
198 		   unsigned long vaddr, pte_t val);
199 void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages);
200 bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode);
201 void pin_page(struct lg_cpu *cpu, unsigned long vaddr);
202 unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr);
203 void page_table_guest_data_init(struct lg_cpu *cpu);
204 
205 /* <arch>/core.c: */
206 void lguest_arch_host_init(void);
207 void lguest_arch_host_fini(void);
208 void lguest_arch_run_guest(struct lg_cpu *cpu);
209 void lguest_arch_handle_trap(struct lg_cpu *cpu);
210 int lguest_arch_init_hypercalls(struct lg_cpu *cpu);
211 int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args);
212 void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start);
213 
214 /* <arch>/switcher.S: */
215 extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
216 
217 /* lguest_user.c: */
218 int lguest_device_init(void);
219 void lguest_device_remove(void);
220 
221 /* hypercalls.c: */
222 void do_hypercalls(struct lg_cpu *cpu);
223 void write_timestamp(struct lg_cpu *cpu);
224 
225 /*L:035
226  * Let's step aside for the moment, to study one important routine that's used
227  * widely in the Host code.
228  *
229  * There are many cases where the Guest can do something invalid, like pass crap
230  * to a hypercall.  Since only the Guest kernel can make hypercalls, it's quite
231  * acceptable to simply terminate the Guest and give the Launcher a nicely
232  * formatted reason.  It's also simpler for the Guest itself, which doesn't
233  * need to check most hypercalls for "success"; if you're still running, it
234  * succeeded.
235  *
236  * Once this is called, the Guest will never run again, so most Host code can
237  * call this then continue as if nothing had happened.  This means many
238  * functions don't have to explicitly return an error code, which keeps the
239  * code simple.
240  *
241  * It also means that this can be called more than once: only the first one is
242  * remembered.  The only trick is that we still need to kill the Guest even if
243  * we can't allocate memory to store the reason.  Linux has a neat way of
244  * packing error codes into invalid pointers, so we use that here.
245  *
246  * Like any macro which uses an "if", it is safely wrapped in a run-once "do {
247  * } while(0)".
248  */
249 #define kill_guest(cpu, fmt...)					\
250 do {								\
251 	if (!(cpu)->lg->dead) {					\
252 		(cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt);	\
253 		if (!(cpu)->lg->dead)				\
254 			(cpu)->lg->dead = ERR_PTR(-ENOMEM);	\
255 	}							\
256 } while(0)
257 /* (End of aside) :*/
258 
259 #endif	/* __ASSEMBLY__ */
260 #endif	/* _LGUEST_H */
261