[deliverable/linux.git] / drivers / lguest / lg.h

#ifndef _LGUEST_H
#define _LGUEST_H

#ifndef __ASSEMBLY__
#include <linux/types.h>
#include <linux/init.h>
#include <linux/stringify.h>
#include <linux/futex.h>
#include <linux/lguest.h>
#include <linux/lguest_launcher.h>
#include <linux/wait.h>
#include <linux/err.h>
#include <asm/semaphore.h>

#include <asm/lguest.h>

void free_pagetables(void);
int init_pagetables(struct page **switcher_page, unsigned int pages);

struct lguest_dma_info
{
	struct list_head list;
	union futex_key key;
	unsigned long dmas;
	struct lguest *owner;
	u16 next_dma;
	u16 num_dmas;
	u8 interrupt; 	/* 0 when not registered */
};

/*H:310 The page-table code owes a great debt of gratitude to Andi Kleen.  He
 * reviewed the original code which used "u32" for all page table entries, and
 * insisted that it would be far clearer with explicit typing.  I thought it
 * was overkill, but he was right: it is much clearer than it was before.
 *
 * We have separate types for the Guest's ptes & pgds and the shadow ptes &
 * pgds.  There's already a Linux type for these (pte_t and pgd_t) but they
 * change depending on kernel config options (PAE). */

/* Each entry is identical: lower 12 bits of flags and upper 20 bits for the
 * "page frame number" (0 == first physical page, etc).  They are different
 * types so the compiler will warn us if we mix them improperly. */
typedef union {
	struct { unsigned flags:12, pfn:20; };
	struct { unsigned long val; } raw;
} spgd_t;
typedef union {
	struct { unsigned flags:12, pfn:20; };
	struct { unsigned long val; } raw;
} spte_t;
typedef union {
	struct { unsigned flags:12, pfn:20; };
	struct { unsigned long val; } raw;
} gpgd_t;
typedef union {
	struct { unsigned flags:12, pfn:20; };
	struct { unsigned long val; } raw;
} gpte_t;

/* We have two convenient macros to convert a "raw" value as handed to us by
 * the Guest into the correct Guest PGD or PTE type. */
#define mkgpte(_val) ((gpte_t){.raw.val = _val})
#define mkgpgd(_val) ((gpgd_t){.raw.val = _val})
/*:*/

struct pgdir
{
	unsigned long cr3;
	spgd_t *pgdir;
};

/* We have two pages shared with guests, per cpu.  */
struct lguest_pages
{
	/* This is the stack page mapped rw in guest */
	char spare[PAGE_SIZE - sizeof(struct lguest_regs)];
	struct lguest_regs regs;

	/* This is the host state & guest descriptor page, ro in guest */
	struct lguest_ro_state state;
} __attribute__((aligned(PAGE_SIZE)));

#define CHANGED_IDT		1
#define CHANGED_GDT		2
#define CHANGED_GDT_TLS		4 /* Actually a subset of CHANGED_GDT */
#define CHANGED_ALL	        3

/* The private info the thread maintains about the guest. */
struct lguest
{
	/* At end of a page shared mapped over lguest_pages in guest.  */
	unsigned long regs_page;
	struct lguest_regs *regs;
	struct lguest_data __user *lguest_data;
	struct task_struct *tsk;
	struct mm_struct *mm; 	/* == tsk->mm, but that becomes NULL on exit */
	u32 pfn_limit;
	/* This provides the offset to the base of guest-physical
	 * memory in the Launcher. */
	void __user *mem_base;
	u32 page_offset;
	u32 cr2;
	int halted;
	int ts;
	u32 next_hcall;
	u32 esp1;
	u8 ss1;

	/* If a hypercall was asked for, this points to the arguments. */
	struct hcall_args *hcall;

	/* Do we need to stop what we're doing and return to userspace? */
	int break_out;
	wait_queue_head_t break_wq;

	/* Bitmap of what has changed: see CHANGED_* above. */
	int changed;
	struct lguest_pages *last_pages;

	/* We keep a small number of these. */
	u32 pgdidx;
	struct pgdir pgdirs[4];

	/* Cached wakeup: we hold a reference to this task. */
	struct task_struct *wake;

	unsigned long noirq_start, noirq_end;
	int dma_is_pending;
	unsigned long pending_dma; /* struct lguest_dma */
	unsigned long pending_key; /* address they're sending to */

	unsigned int stack_pages;
	u32 tsc_khz;

	struct lguest_dma_info dma[LGUEST_MAX_DMA];

	/* Dead? */
	const char *dead;

	struct lguest_arch arch;

	/* Virtual clock device */
	struct hrtimer hrt;

	/* Pending virtual interrupts */
	DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
};

extern struct mutex lguest_lock;

/* core.c: */
u32 lgread_u32(struct lguest *lg, unsigned long addr);
void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val);
void lgread(struct lguest *lg, void *buf, unsigned long addr, unsigned len);
void lgwrite(struct lguest *lg, unsigned long, const void *buf, unsigned len);
int lguest_address_ok(const struct lguest *lg,
		      unsigned long addr, unsigned long len);
int run_guest(struct lguest *lg, unsigned long __user *user);


/* interrupts_and_traps.c: */
void maybe_do_interrupt(struct lguest *lg);
int deliver_trap(struct lguest *lg, unsigned int num);
void load_guest_idt_entry(struct lguest *lg, unsigned int i, u32 low, u32 hi);
void guest_set_stack(struct lguest *lg, u32 seg, u32 esp, unsigned int pages);
void pin_stack_pages(struct lguest *lg);
void setup_default_idt_entries(struct lguest_ro_state *state,
			       const unsigned long *def);
void copy_traps(const struct lguest *lg, struct desc_struct *idt,
		const unsigned long *def);
void guest_set_clockevent(struct lguest *lg, unsigned long delta);
void init_clockdev(struct lguest *lg);

/* segments.c: */
void setup_default_gdt_entries(struct lguest_ro_state *state);
void setup_guest_gdt(struct lguest *lg);
void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num);
void guest_load_tls(struct lguest *lg, unsigned long tls_array);
void copy_gdt(const struct lguest *lg, struct desc_struct *gdt);
void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt);

/* page_tables.c: */
int init_guest_pagetable(struct lguest *lg, unsigned long pgtable);
void free_guest_pagetable(struct lguest *lg);
void guest_new_pagetable(struct lguest *lg, unsigned long pgtable);
void guest_set_pmd(struct lguest *lg, unsigned long cr3, u32 i);
void guest_pagetable_clear_all(struct lguest *lg);
void guest_pagetable_flush_user(struct lguest *lg);
void guest_set_pte(struct lguest *lg, unsigned long cr3,
		   unsigned long vaddr, gpte_t val);
void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages);
int demand_page(struct lguest *info, unsigned long cr2, int errcode);
void pin_page(struct lguest *lg, unsigned long vaddr);

/* <arch>/core.c: */
void lguest_arch_host_init(void);
void lguest_arch_host_fini(void);
void lguest_arch_run_guest(struct lguest *lg);
void lguest_arch_handle_trap(struct lguest *lg);
int lguest_arch_init_hypercalls(struct lguest *lg);
int lguest_arch_do_hcall(struct lguest *lg, struct hcall_args *args);
void lguest_arch_setup_regs(struct lguest *lg, unsigned long start);

/* <arch>/switcher.S: */
extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];

/* lguest_user.c: */
int lguest_device_init(void);
void lguest_device_remove(void);

/* io.c: */
void lguest_io_init(void);
int bind_dma(struct lguest *lg,
	     unsigned long key, unsigned long udma, u16 numdmas, u8 interrupt);
void send_dma(struct lguest *info, unsigned long key, unsigned long udma);
void release_all_dma(struct lguest *lg);
unsigned long get_dma_buffer(struct lguest *lg, unsigned long key,
			     unsigned long *interrupt);

/* hypercalls.c: */
void do_hypercalls(struct lguest *lg);
void write_timestamp(struct lguest *lg);

/*L:035
 * Let's step aside for the moment, to study one important routine that's used
 * widely in the Host code.
 *
 * There are many cases where the Guest does something invalid, like pass crap
 * to a hypercall.  Since only the Guest kernel can make hypercalls, it's quite
 * acceptable to simply terminate the Guest and give the Launcher a nicely
 * formatted reason.  It's also simpler for the Guest itself, which doesn't
 * need to check most hypercalls for "success"; if you're still running, it
 * succeeded.
 *
 * Once this is called, the Guest will never run again, so most Host code can
 * call this then continue as if nothing had happened.  This means many
 * functions don't have to explicitly return an error code, which keeps the
 * code simple.
 *
 * It also means that this can be called more than once: only the first one is
 * remembered.  The only trick is that we still need to kill the Guest even if
 * we can't allocate memory to store the reason.  Linux has a neat way of
 * packing error codes into invalid pointers, so we use that here.
 *
 * Like any macro which uses an "if", it is safely wrapped in a run-once "do {
 * } while(0)".
 */
#define kill_guest(lg, fmt...)					\
do {								\
	if (!(lg)->dead) {					\
		(lg)->dead = kasprintf(GFP_ATOMIC, fmt);	\
		if (!(lg)->dead)				\
			(lg)->dead = ERR_PTR(-ENOMEM);		\
	}							\
} while(0)
/* (End of aside) :*/

static inline unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
{
	return vaddr - lg->page_offset;
}
#endif	/* __ASSEMBLY__ */
#endif	/* _LGUEST_H */
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/futex.h>
	9	#include <linux/lguest.h>
	10	#include <linux/lguest_launcher.h>
	11	#include <linux/wait.h>
	12	#include <linux/err.h>
	13	#include <asm/semaphore.h>
d7e28ffe	14
625efab1	15	#include <asm/lguest.h>
d7e28ffe RR	16
	17	void free_pagetables(void);
	18	int init_pagetables(struct page **switcher_page, unsigned int pages);
	19
d7e28ffe RR	20	struct lguest_dma_info
	21	{
	22	struct list_head list;
	23	union futex_key key;
	24	unsigned long dmas;
48245cc0	25	struct lguest *owner;
d7e28ffe RR	26	u16 next_dma;
d7e28ffe RR	27	u16 num_dmas;
d7e28ffe RR	28	u8 interrupt; /* 0 when not registered */
	29	};
	30
bff672e6 RR	31	/*H:310 The page-table code owes a great debt of gratitude to Andi Kleen. He
	32	* reviewed the original code which used "u32" for all page table entries, and
	33	* insisted that it would be far clearer with explicit typing. I thought it
	34	* was overkill, but he was right: it is much clearer than it was before.
	35	*
	36	* We have separate types for the Guest's ptes & pgds and the shadow ptes &
	37	* pgds. There's already a Linux type for these (pte_t and pgd_t) but they
	38	* change depending on kernel config options (PAE). */
	39
	40	/* Each entry is identical: lower 12 bits of flags and upper 20 bits for the
	41	* "page frame number" (0 == first physical page, etc). They are different
	42	* types so the compiler will warn us if we mix them improperly. */
d7e28ffe RR	43	typedef union {
	44	struct { unsigned flags:12, pfn:20; };
	45	struct { unsigned long val; } raw;
	46	} spgd_t;
	47	typedef union {
	48	struct { unsigned flags:12, pfn:20; };
	49	struct { unsigned long val; } raw;
	50	} spte_t;
	51	typedef union {
	52	struct { unsigned flags:12, pfn:20; };
	53	struct { unsigned long val; } raw;
	54	} gpgd_t;
	55	typedef union {
	56	struct { unsigned flags:12, pfn:20; };
	57	struct { unsigned long val; } raw;
	58	} gpte_t;
bff672e6 RR	59
	60	/* We have two convenient macros to convert a "raw" value as handed to us by
	61	* the Guest into the correct Guest PGD or PTE type. */
d7e28ffe RR	62	#define mkgpte(_val) ((gpte_t){.raw.val = _val})
d7e28ffe RR	63	#define mkgpgd(_val) ((gpgd_t){.raw.val = _val})
bff672e6	64	/:/
d7e28ffe RR	65
	66	struct pgdir
	67	{
	68	unsigned long cr3;
	69	spgd_t *pgdir;
	70	};
	71
d7e28ffe RR	72	/* We have two pages shared with guests, per cpu. */
	73	struct lguest_pages
	74	{
	75	/* This is the stack page mapped rw in guest */
	76	char spare[PAGE_SIZE - sizeof(struct lguest_regs)];
	77	struct lguest_regs regs;
	78
	79	/* This is the host state & guest descriptor page, ro in guest */
	80	struct lguest_ro_state state;
	81	} __attribute__((aligned(PAGE_SIZE)));
	82
	83	#define CHANGED_IDT 1
	84	#define CHANGED_GDT 2
	85	#define CHANGED_GDT_TLS 4 /* Actually a subset of CHANGED_GDT */
	86	#define CHANGED_ALL 3
	87
	88	/* The private info the thread maintains about the guest. */
	89	struct lguest
	90	{
	91	/* At end of a page shared mapped over lguest_pages in guest. */
	92	unsigned long regs_page;
	93	struct lguest_regs *regs;
	94	struct lguest_data __user *lguest_data;
	95	struct task_struct *tsk;
	96	struct mm_struct mm; / == tsk->mm, but that becomes NULL on exit */
d7e28ffe	97	u32 pfn_limit;
3c6b5bfa RR	98	/* This provides the offset to the base of guest-physical
	99	* memory in the Launcher. */
	100	void __user *mem_base;
d7e28ffe RR	101	u32 page_offset;
	102	u32 cr2;
	103	int halted;
	104	int ts;
	105	u32 next_hcall;
	106	u32 esp1;
	107	u8 ss1;
	108
cc6d4fbc	109	/* If a hypercall was asked for, this points to the arguments. */
b410e7b1	110	struct hcall_args *hcall;
cc6d4fbc	111
d7e28ffe RR	112	/* Do we need to stop what we're doing and return to userspace? */
	113	int break_out;
	114	wait_queue_head_t break_wq;
	115
	116	/* Bitmap of what has changed: see CHANGED_* above. */
	117	int changed;
	118	struct lguest_pages *last_pages;
	119
	120	/* We keep a small number of these. */
	121	u32 pgdidx;
	122	struct pgdir pgdirs[4];
	123
	124	/* Cached wakeup: we hold a reference to this task. */
	125	struct task_struct *wake;
	126
	127	unsigned long noirq_start, noirq_end;
	128	int dma_is_pending;
	129	unsigned long pending_dma; /* struct lguest_dma */
	130	unsigned long pending_key; /* address they're sending to */
	131
	132	unsigned int stack_pages;
	133	u32 tsc_khz;
	134
	135	struct lguest_dma_info dma[LGUEST_MAX_DMA];
	136
	137	/* Dead? */
	138	const char *dead;
	139
625efab1	140	struct lguest_arch arch;
d7e28ffe RR	141
	142	/* Virtual clock device */
	143	struct hrtimer hrt;
	144
	145	/* Pending virtual interrupts */
	146	DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
	147	};
	148
d7e28ffe RR	149	extern struct mutex lguest_lock;
	150
	151	/* core.c: */
	152	u32 lgread_u32(struct lguest *lg, unsigned long addr);
	153	void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val);
	154	void lgread(struct lguest lg, void buf, unsigned long addr, unsigned len);
	155	void lgwrite(struct lguest lg, unsigned long, const void buf, unsigned len);
d7e28ffe RR	156	int lguest_address_ok(const struct lguest *lg,
	157	unsigned long addr, unsigned long len);
	158	int run_guest(struct lguest lg, unsigned long __user user);
	159
	160
	161	/* interrupts_and_traps.c: */
	162	void maybe_do_interrupt(struct lguest *lg);
	163	int deliver_trap(struct lguest *lg, unsigned int num);
	164	void load_guest_idt_entry(struct lguest *lg, unsigned int i, u32 low, u32 hi);
	165	void guest_set_stack(struct lguest *lg, u32 seg, u32 esp, unsigned int pages);
	166	void pin_stack_pages(struct lguest *lg);
	167	void setup_default_idt_entries(struct lguest_ro_state *state,
	168	const unsigned long *def);
	169	void copy_traps(const struct lguest lg, struct desc_struct idt,
	170	const unsigned long *def);
	171	void guest_set_clockevent(struct lguest *lg, unsigned long delta);
	172	void init_clockdev(struct lguest *lg);
	173
	174	/* segments.c: */
	175	void setup_default_gdt_entries(struct lguest_ro_state *state);
	176	void setup_guest_gdt(struct lguest *lg);
	177	void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num);
	178	void guest_load_tls(struct lguest *lg, unsigned long tls_array);
	179	void copy_gdt(const struct lguest lg, struct desc_struct gdt);
	180	void copy_gdt_tls(const struct lguest lg, struct desc_struct gdt);
	181
	182	/* page_tables.c: */
	183	int init_guest_pagetable(struct lguest *lg, unsigned long pgtable);
	184	void free_guest_pagetable(struct lguest *lg);
	185	void guest_new_pagetable(struct lguest *lg, unsigned long pgtable);
	186	void guest_set_pmd(struct lguest *lg, unsigned long cr3, u32 i);
	187	void guest_pagetable_clear_all(struct lguest *lg);
	188	void guest_pagetable_flush_user(struct lguest *lg);
	189	void guest_set_pte(struct lguest *lg, unsigned long cr3,
	190	unsigned long vaddr, gpte_t val);
	191	void map_switcher_in_guest(struct lguest lg, struct lguest_pages pages);
	192	int demand_page(struct lguest *info, unsigned long cr2, int errcode);
	193	void pin_page(struct lguest *lg, unsigned long vaddr);
	194
625efab1 JS	195	/* <arch>/core.c: */
	196	void lguest_arch_host_init(void);
	197	void lguest_arch_host_fini(void);
	198	void lguest_arch_run_guest(struct lguest *lg);
	199	void lguest_arch_handle_trap(struct lguest *lg);
b410e7b1 JS	200	int lguest_arch_init_hypercalls(struct lguest *lg);
b410e7b1 JS	201	int lguest_arch_do_hcall(struct lguest lg, struct hcall_args args);
d612cde0	202	void lguest_arch_setup_regs(struct lguest *lg, unsigned long start);
625efab1 JS	203
	204	/* <arch>/switcher.S: */
	205	extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
	206
d7e28ffe RR	207	/* lguest_user.c: */
	208	int lguest_device_init(void);
	209	void lguest_device_remove(void);
	210
	211	/* io.c: */
	212	void lguest_io_init(void);
	213	int bind_dma(struct lguest *lg,
	214	unsigned long key, unsigned long udma, u16 numdmas, u8 interrupt);
	215	void send_dma(struct lguest *info, unsigned long key, unsigned long udma);
	216	void release_all_dma(struct lguest *lg);
	217	unsigned long get_dma_buffer(struct lguest *lg, unsigned long key,
	218	unsigned long *interrupt);
	219
	220	/* hypercalls.c: */
	221	void do_hypercalls(struct lguest *lg);
6c8dca5d	222	void write_timestamp(struct lguest *lg);
d7e28ffe	223
dde79789 RR	224	/*L:035
	225	* Let's step aside for the moment, to study one important routine that's used
	226	* widely in the Host code.
	227	*
	228	* There are many cases where the Guest does something invalid, like pass crap
	229	* to a hypercall. Since only the Guest kernel can make hypercalls, it's quite
	230	* acceptable to simply terminate the Guest and give the Launcher a nicely
	231	* formatted reason. It's also simpler for the Guest itself, which doesn't
	232	* need to check most hypercalls for "success"; if you're still running, it
	233	* succeeded.
	234	*
	235	* Once this is called, the Guest will never run again, so most Host code can
	236	* call this then continue as if nothing had happened. This means many
	237	* functions don't have to explicitly return an error code, which keeps the
	238	* code simple.
	239	*
	240	* It also means that this can be called more than once: only the first one is
	241	* remembered. The only trick is that we still need to kill the Guest even if
	242	* we can't allocate memory to store the reason. Linux has a neat way of
	243	* packing error codes into invalid pointers, so we use that here.
	244	*
	245	* Like any macro which uses an "if", it is safely wrapped in a run-once "do {
	246	* } while(0)".
	247	*/
d7e28ffe RR	248	#define kill_guest(lg, fmt...) \
	249	do { \
	250	if (!(lg)->dead) { \
	251	(lg)->dead = kasprintf(GFP_ATOMIC, fmt); \
	252	if (!(lg)->dead) \
	253	(lg)->dead = ERR_PTR(-ENOMEM); \
	254	} \
	255	} while(0)
dde79789	256	/* (End of aside) :*/
d7e28ffe RR	257
	258	static inline unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
	259	{
	260	return vaddr - lg->page_offset;
	261	}
	262	#endif /* __ASSEMBLY__ */
	263	#endif /* _LGUEST_H */