#ifndef _LGUEST_H
#define _LGUEST_H
-#include <asm/desc.h>
-
-#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 <linux/types.h>
#include <linux/init.h>
#include <linux/stringify.h>
-#include <linux/binfmts.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 "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 <asm/lguest.h>
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;
+ struct lguest *owner;
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. */
+/*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;
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
{
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];
-};
-
/* We have two pages shared with guests, per cpu. */
struct lguest_pages
{
struct lguest_data __user *lguest_data;
struct task_struct *tsk;
struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */
- u16 guestid;
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;
u32 esp1;
u8 ss1;
+ /* If a hypercall was asked for, this points to the arguments. */
+ struct lguest_regs *hcall;
+
/* Do we need to stop what we're doing and return to userspace? */
int break_out;
wait_queue_head_t break_wq;
/* 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;
DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
};
-extern struct lguest lguests[];
extern struct mutex lguest_lock;
/* core.c: */
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);
int run_guest(struct lguest *lg, unsigned long __user *user);
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);
+
+/* <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);
/* 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 = ERR_PTR(-ENOMEM); \
} \
} while(0)
+/* (End of aside) :*/
static inline unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
{
projects / linux-2.6 / blobdiff