X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=drivers%2Flguest%2Flg.h;h=5f73ddff0e3d7ef2c0a068e8093180c18a09c2a3;hb=4dcc53da49c2387078fe8ceb7a420d125e027fc6;hp=3e2ddfbc816e9fc1396c8fe724e14e1732928502;hpb=d7e28ffe6c74416b54345d6004fd0964c115b12c;p=safe%2Fjmp%2Flinux-2.6 diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h index 3e2ddfb..5f73ddf 100644 --- a/drivers/lguest/lg.h +++ b/drivers/lguest/lg.h @@ -1,106 +1,25 @@ #ifndef _LGUEST_H #define _LGUEST_H -#include - -#define GDT_ENTRY_LGUEST_CS 10 -#define GDT_ENTRY_LGUEST_DS 11 -#define LGUEST_CS (GDT_ENTRY_LGUEST_CS * 8) -#define LGUEST_DS (GDT_ENTRY_LGUEST_DS * 8) - #ifndef __ASSEMBLY__ #include #include #include -#include -#include #include #include #include #include #include -#include "irq_vectors.h" -#define GUEST_PL 1 - -struct lguest_regs -{ - /* Manually saved part. */ - unsigned long ebx, ecx, edx; - unsigned long esi, edi, ebp; - unsigned long gs; - unsigned long eax; - unsigned long fs, ds, es; - unsigned long trapnum, errcode; - /* Trap pushed part */ - unsigned long eip; - unsigned long cs; - unsigned long eflags; - unsigned long esp; - unsigned long ss; -}; +#include void free_pagetables(void); int init_pagetables(struct page **switcher_page, unsigned int pages); -/* Full 4G segment descriptors, suitable for CS and DS. */ -#define FULL_EXEC_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9b00}) -#define FULL_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9300}) - -struct lguest_dma_info -{ - struct list_head list; - union futex_key key; - unsigned long dmas; - u16 next_dma; - u16 num_dmas; - u16 guestid; - u8 interrupt; /* 0 when not registered */ -}; - -/* We have separate types for the guest's ptes & pgds and the shadow ptes & - * pgds. Since this host might use three-level pagetables and the guest and - * shadow pagetables don't, we can't use the normal pte_t/pgd_t. */ -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; -#define mkgpte(_val) ((gpte_t){.raw.val = _val}) -#define mkgpgd(_val) ((gpgd_t){.raw.val = _val}) - struct pgdir { - unsigned long cr3; - spgd_t *pgdir; -}; - -/* This is a guest-specific page (mapped ro) into the guest. */ -struct lguest_ro_state -{ - /* Host information we need to restore when we switch back. */ - u32 host_cr3; - struct Xgt_desc_struct host_idt_desc; - struct Xgt_desc_struct host_gdt_desc; - u32 host_sp; - - /* Fields which are used when guest is running. */ - struct Xgt_desc_struct guest_idt_desc; - struct Xgt_desc_struct guest_gdt_desc; - struct i386_hw_tss guest_tss; - struct desc_struct guest_idt[IDT_ENTRIES]; - struct desc_struct guest_gdt[GDT_ENTRIES]; + unsigned long gpgdir; + pgd_t *pgdir; }; /* We have two pages shared with guests, per cpu. */ @@ -119,6 +38,13 @@ struct lguest_pages #define CHANGED_GDT_TLS 4 /* Actually a subset of CHANGED_GDT */ #define CHANGED_ALL 3 +struct lguest; + +struct lg_cpu { + unsigned int id; + struct lguest *lg; +}; + /* The private info the thread maintains about the guest. */ struct lguest { @@ -128,9 +54,14 @@ struct lguest struct lguest_data __user *lguest_data; struct task_struct *tsk; struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */ - u16 guestid; + struct lg_cpu cpus[NR_CPUS]; + unsigned int nr_cpus; + u32 pfn_limit; - u32 page_offset; + /* This provides the offset to the base of guest-physical + * memory in the Launcher. */ + void __user *mem_base; + unsigned long kernel_address; u32 cr2; int halted; int ts; @@ -138,6 +69,9 @@ struct lguest 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; @@ -150,28 +84,16 @@ struct lguest 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 long pending_notify; /* pfn from LHCALL_NOTIFY */ unsigned int stack_pages; u32 tsc_khz; - struct lguest_dma_info dma[LGUEST_MAX_DMA]; - /* Dead? */ const char *dead; - /* The GDT entries copied into lguest_ro_state when running. */ - struct desc_struct gdt[GDT_ENTRIES]; - - /* The IDT entries: some copied into lguest_ro_state when running. */ - struct desc_struct idt[FIRST_EXTERNAL_VECTOR+LGUEST_IRQS]; - struct desc_struct syscall_idt; + struct lguest_arch arch; /* Virtual clock device */ struct hrtimer hrt; @@ -180,19 +102,38 @@ struct lguest DECLARE_BITMAP(irqs_pending, LGUEST_IRQS); }; -extern struct lguest lguests[]; 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 find_free_guest(void); int lguest_address_ok(const struct lguest *lg, unsigned long addr, unsigned long len); +void __lgread(struct lguest *, void *, unsigned long, unsigned); +void __lgwrite(struct lguest *, unsigned long, const void *, unsigned); + +/*H:035 Using memory-copy operations like that is usually inconvient, so we + * have the following helper macros which read and write a specific type (often + * an unsigned long). + * + * This reads into a variable of the given type then returns that. */ +#define lgread(lg, addr, type) \ + ({ type _v; __lgread((lg), &_v, (addr), sizeof(_v)); _v; }) + +/* This checks that the variable is of the given type, then writes it out. */ +#define lgwrite(lg, addr, type, val) \ + do { \ + typecheck(type, val); \ + __lgwrite((lg), (addr), &(val), sizeof(val)); \ + } while(0) +/* (end of memory access helper routines) :*/ + int run_guest(struct lguest *lg, unsigned long __user *user); +/* Helper macros to obtain the first 12 or the last 20 bits, this is only the + * first step in the migration to the kernel types. pte_pfn is already defined + * in the kernel. */ +#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK) +#define pte_flags(x) (pte_val(x) & ~PAGE_MASK) +#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT) /* interrupts_and_traps.c: */ void maybe_do_interrupt(struct lguest *lg); @@ -206,6 +147,9 @@ 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); +bool check_syscall_vector(struct lguest *lg); +int init_interrupts(void); +void free_interrupts(void); /* segments.c: */ void setup_default_gdt_entries(struct lguest_ro_state *state); @@ -219,31 +163,61 @@ void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt); 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_set_pmd(struct lguest *lg, unsigned long gpgdir, 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 guest_set_pte(struct lguest *lg, unsigned long gpgdir, + unsigned long vaddr, pte_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); +unsigned long guest_pa(struct lguest *lg, unsigned long vaddr); +void page_table_guest_data_init(struct lguest *lg); + +/* /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); + +/* /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 can do 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) { \ @@ -252,10 +226,7 @@ do { \ (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 */