blob: 295df06e65903cc15524f8a36b45014ce1000f27 [file] [log] [blame]
Rusty Russelld7e28ff2007-07-19 01:49:23 -07001#ifndef _LGUEST_H
2#define _LGUEST_H
3
Rusty Russelld7e28ff2007-07-19 01:49:23 -07004#ifndef __ASSEMBLY__
5#include <linux/types.h>
6#include <linux/init.h>
7#include <linux/stringify.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -07008#include <linux/lguest.h>
9#include <linux/lguest_launcher.h>
10#include <linux/wait.h>
Glauber de Oliveira Costaca94f2b2008-01-18 23:59:07 -020011#include <linux/hrtimer.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070012#include <linux/err.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/slab.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070014
Jes Sorensen625efab2007-10-22 11:03:28 +100015#include <asm/lguest.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070016
17void free_pagetables(void);
18int init_pagetables(struct page **switcher_page, unsigned int pages);
19
Rusty Russell1842f232009-07-30 16:03:46 -060020struct pgdir {
Rusty Russellee3db0f2007-10-22 11:03:34 +100021 unsigned long gpgdir;
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +100022 pgd_t *pgdir;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070023};
24
Rusty Russelld7e28ff2007-07-19 01:49:23 -070025/* We have two pages shared with guests, per cpu. */
Rusty Russell1842f232009-07-30 16:03:46 -060026struct lguest_pages {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070027 /* 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
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020040struct lg_cpu {
41 unsigned int id;
42 struct lguest *lg;
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020043 struct task_struct *tsk;
44 struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -020045
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020046 u32 cr2;
47 int ts;
48 u32 esp1;
Matias Zabaljaureguif0861222009-06-12 22:27:04 -060049 u16 ss1;
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020050
Glauber de Oliveira Costaae3749d2008-01-17 19:14:46 -020051 /* Bitmap of what has changed: see CHANGED_* above. */
52 int changed;
53
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -020054 unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */
55
Rusty Russell2e04ef72009-07-30 16:03:45 -060056 /* At end of a page shared mapped over lguest_pages in guest. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020057 unsigned long regs_page;
58 struct lguest_regs *regs;
59
Glauber de Oliveira Costaf34f8c52008-01-17 19:13:26 -020060 struct lguest_pages *last_pages;
61
Rusty Russell5dea1c82011-07-22 14:39:48 +093062 /* Initialization mode: linear map everything. */
63 bool linear_pages;
Rusty Russell2e04ef72009-07-30 16:03:45 -060064 int cpu_pgd; /* Which pgd this cpu is currently using */
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -020065
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -020066 /* If a hypercall was asked for, this points to the arguments. */
67 struct hcall_args *hcall;
68 u32 next_hcall;
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -020069
70 /* Virtual clock device */
71 struct hrtimer hrt;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020072
Rusty Russell5dac0512009-06-12 22:27:10 -060073 /* Did the Guest tell us to halt? */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020074 int halted;
75
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020076 /* Pending virtual interrupts */
77 DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -020078
79 struct lg_cpu_arch arch;
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020080};
81
Rusty Russelldf60aee2009-06-12 22:27:09 -060082struct lg_eventfd {
83 unsigned long addr;
Davide Libenzi13389012009-06-30 11:41:11 -070084 struct eventfd_ctx *event;
Rusty Russelldf60aee2009-06-12 22:27:09 -060085};
86
87struct lg_eventfd_map {
88 unsigned int num;
89 struct lg_eventfd map[];
90};
91
Rusty Russelld7e28ff2007-07-19 01:49:23 -070092/* The private info the thread maintains about the guest. */
Rusty Russell1842f232009-07-30 16:03:46 -060093struct lguest {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070094 struct lguest_data __user *lguest_data;
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020095 struct lg_cpu cpus[NR_CPUS];
96 unsigned int nr_cpus;
97
Rusty Russelld7e28ff2007-07-19 01:49:23 -070098 u32 pfn_limit;
Rusty Russell2e04ef72009-07-30 16:03:45 -060099
100 /*
101 * This provides the offset to the base of guest-physical memory in the
102 * Launcher.
103 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000104 void __user *mem_base;
Rusty Russell47436aa2007-10-22 11:03:36 +1000105 unsigned long kernel_address;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700106
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700107 struct pgdir pgdirs[4];
108
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700109 unsigned long noirq_start, noirq_end;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700110
111 unsigned int stack_pages;
112 u32 tsc_khz;
113
Rusty Russelldf60aee2009-06-12 22:27:09 -0600114 struct lg_eventfd_map *eventfds;
115
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700116 /* Dead? */
117 const char *dead;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700118};
119
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700120extern struct mutex lguest_lock;
121
122/* core.c: */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300123bool lguest_address_ok(const struct lguest *lg,
124 unsigned long addr, unsigned long len);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200125void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
126void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
Rusty Russell2d37f942007-10-22 11:24:24 +1000127
Rusty Russell2e04ef72009-07-30 16:03:45 -0600128/*H:035
129 * Using memory-copy operations like that is usually inconvient, so we
Rusty Russell2d37f942007-10-22 11:24:24 +1000130 * have the following helper macros which read and write a specific type (often
131 * an unsigned long).
132 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600133 * This reads into a variable of the given type then returns that.
134 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200135#define lgread(cpu, addr, type) \
136 ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; })
Rusty Russell2d37f942007-10-22 11:24:24 +1000137
138/* This checks that the variable is of the given type, then writes it out. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200139#define lgwrite(cpu, addr, type, val) \
Rusty Russell2d37f942007-10-22 11:24:24 +1000140 do { \
141 typecheck(type, val); \
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200142 __lgwrite((cpu), (addr), &(val), sizeof(val)); \
Rusty Russell2d37f942007-10-22 11:24:24 +1000143 } while(0)
144/* (end of memory access helper routines) :*/
145
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200146int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700147
Rusty Russell2e04ef72009-07-30 16:03:45 -0600148/*
149 * Helper macros to obtain the first 12 or the last 20 bits, this is only the
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000150 * first step in the migration to the kernel types. pte_pfn is already defined
Rusty Russell2e04ef72009-07-30 16:03:45 -0600151 * in the kernel.
152 */
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000153#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK)
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000154#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT)
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600155#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK)
156#define pmd_pfn(x) (pmd_val(x) >> PAGE_SHIFT)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700157
158/* interrupts_and_traps.c: */
Rusty Russella32a8812009-06-12 22:27:02 -0600159unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more);
160void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more);
Rusty Russell9f155a92009-06-12 22:27:08 -0600161void set_interrupt(struct lg_cpu *cpu, unsigned int irq);
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300162bool deliver_trap(struct lg_cpu *cpu, unsigned int num);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200163void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i,
164 u32 low, u32 hi);
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200165void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages);
166void pin_stack_pages(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700167void setup_default_idt_entries(struct lguest_ro_state *state,
168 const unsigned long *def);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200169void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700170 const unsigned long *def);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200171void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta);
Rusty Russelldf60aee2009-06-12 22:27:09 -0600172bool send_notify_to_eventfd(struct lg_cpu *cpu);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200173void init_clockdev(struct lg_cpu *cpu);
Rusty Russellc18acd72007-10-22 11:03:35 +1000174bool check_syscall_vector(struct lguest *lg);
175int init_interrupts(void);
176void free_interrupts(void);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700177
178/* segments.c: */
179void setup_default_gdt_entries(struct lguest_ro_state *state);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200180void setup_guest_gdt(struct lg_cpu *cpu);
Rusty Russella489f0b2009-04-19 23:14:00 -0600181void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i,
182 u32 low, u32 hi);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200183void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array);
184void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt);
185void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700186
187/* page_tables.c: */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300188int init_guest_pagetable(struct lguest *lg);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700189void free_guest_pagetable(struct lguest *lg);
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200190void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable);
Matias Zabaljaureguiebe0ba82009-05-30 15:48:08 -0300191void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i);
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600192#ifdef CONFIG_X86_PAE
193void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
194#endif
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200195void guest_pagetable_clear_all(struct lg_cpu *cpu);
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200196void guest_pagetable_flush_user(struct lg_cpu *cpu);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200197void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir,
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000198 unsigned long vaddr, pte_t val);
Glauber de Oliveira Costa0c784412008-01-07 11:05:30 -0200199void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages);
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300200bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode);
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200201void pin_page(struct lg_cpu *cpu, unsigned long vaddr);
202unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200203void page_table_guest_data_init(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700204
Jes Sorensen625efab2007-10-22 11:03:28 +1000205/* <arch>/core.c: */
206void lguest_arch_host_init(void);
207void lguest_arch_host_fini(void);
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200208void lguest_arch_run_guest(struct lg_cpu *cpu);
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200209void lguest_arch_handle_trap(struct lg_cpu *cpu);
210int lguest_arch_init_hypercalls(struct lg_cpu *cpu);
211int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args);
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200212void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start);
Jes Sorensen625efab2007-10-22 11:03:28 +1000213
214/* <arch>/switcher.S: */
215extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
216
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700217/* lguest_user.c: */
218int lguest_device_init(void);
219void lguest_device_remove(void);
220
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700221/* hypercalls.c: */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200222void do_hypercalls(struct lg_cpu *cpu);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200223void write_timestamp(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700224
Rusty Russelldde79782007-07-26 10:41:03 -0700225/*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 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000229 * There are many cases where the Guest can do something invalid, like pass crap
Rusty Russelldde79782007-07-26 10:41:03 -0700230 * 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 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200249#define kill_guest(cpu, fmt...) \
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700250do { \
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200251 if (!(cpu)->lg->dead) { \
252 (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt); \
253 if (!(cpu)->lg->dead) \
254 (cpu)->lg->dead = ERR_PTR(-ENOMEM); \
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700255 } \
256} while(0)
Rusty Russelldde79782007-07-26 10:41:03 -0700257/* (End of aside) :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700258
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700259#endif /* __ASSEMBLY__ */
260#endif /* _LGUEST_H */