blob: 9136411fadd53e10c47c0624cf8576e7ae9d70b2 [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 Russell2e04ef72009-07-30 16:03:45 -060062 int cpu_pgd; /* Which pgd this cpu is currently using */
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -020063
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -020064 /* If a hypercall was asked for, this points to the arguments. */
65 struct hcall_args *hcall;
66 u32 next_hcall;
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -020067
68 /* Virtual clock device */
69 struct hrtimer hrt;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020070
Rusty Russell5dac0512009-06-12 22:27:10 -060071 /* Did the Guest tell us to halt? */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020072 int halted;
73
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020074 /* Pending virtual interrupts */
75 DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -020076
77 struct lg_cpu_arch arch;
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020078};
79
Rusty Russelldf60aee2009-06-12 22:27:09 -060080struct lg_eventfd {
81 unsigned long addr;
Davide Libenzi13389012009-06-30 11:41:11 -070082 struct eventfd_ctx *event;
Rusty Russelldf60aee2009-06-12 22:27:09 -060083};
84
85struct lg_eventfd_map {
86 unsigned int num;
87 struct lg_eventfd map[];
88};
89
Rusty Russelld7e28ff2007-07-19 01:49:23 -070090/* The private info the thread maintains about the guest. */
Rusty Russell1842f232009-07-30 16:03:46 -060091struct lguest {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070092 struct lguest_data __user *lguest_data;
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020093 struct lg_cpu cpus[NR_CPUS];
94 unsigned int nr_cpus;
95
Rusty Russelld7e28ff2007-07-19 01:49:23 -070096 u32 pfn_limit;
Rusty Russell2e04ef72009-07-30 16:03:45 -060097
98 /*
99 * This provides the offset to the base of guest-physical memory in the
100 * Launcher.
101 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000102 void __user *mem_base;
Rusty Russell47436aa2007-10-22 11:03:36 +1000103 unsigned long kernel_address;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700104
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700105 struct pgdir pgdirs[4];
106
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700107 unsigned long noirq_start, noirq_end;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700108
109 unsigned int stack_pages;
110 u32 tsc_khz;
111
Rusty Russelldf60aee2009-06-12 22:27:09 -0600112 struct lg_eventfd_map *eventfds;
113
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700114 /* Dead? */
115 const char *dead;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700116};
117
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700118extern struct mutex lguest_lock;
119
120/* core.c: */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300121bool lguest_address_ok(const struct lguest *lg,
122 unsigned long addr, unsigned long len);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200123void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
124void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
Rusty Russell2d37f942007-10-22 11:24:24 +1000125
Rusty Russell2e04ef72009-07-30 16:03:45 -0600126/*H:035
127 * Using memory-copy operations like that is usually inconvient, so we
Rusty Russell2d37f942007-10-22 11:24:24 +1000128 * have the following helper macros which read and write a specific type (often
129 * an unsigned long).
130 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600131 * This reads into a variable of the given type then returns that.
132 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200133#define lgread(cpu, addr, type) \
134 ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; })
Rusty Russell2d37f942007-10-22 11:24:24 +1000135
136/* This checks that the variable is of the given type, then writes it out. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200137#define lgwrite(cpu, addr, type, val) \
Rusty Russell2d37f942007-10-22 11:24:24 +1000138 do { \
139 typecheck(type, val); \
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200140 __lgwrite((cpu), (addr), &(val), sizeof(val)); \
Rusty Russell2d37f942007-10-22 11:24:24 +1000141 } while(0)
142/* (end of memory access helper routines) :*/
143
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200144int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700145
Rusty Russell2e04ef72009-07-30 16:03:45 -0600146/*
147 * Helper macros to obtain the first 12 or the last 20 bits, this is only the
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000148 * first step in the migration to the kernel types. pte_pfn is already defined
Rusty Russell2e04ef72009-07-30 16:03:45 -0600149 * in the kernel.
150 */
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000151#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK)
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000152#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT)
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600153#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK)
154#define pmd_pfn(x) (pmd_val(x) >> PAGE_SHIFT)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700155
156/* interrupts_and_traps.c: */
Rusty Russella32a8812009-06-12 22:27:02 -0600157unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more);
158void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more);
Rusty Russell9f155a92009-06-12 22:27:08 -0600159void set_interrupt(struct lg_cpu *cpu, unsigned int irq);
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300160bool deliver_trap(struct lg_cpu *cpu, unsigned int num);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200161void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i,
162 u32 low, u32 hi);
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200163void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages);
164void pin_stack_pages(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700165void setup_default_idt_entries(struct lguest_ro_state *state,
166 const unsigned long *def);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200167void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700168 const unsigned long *def);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200169void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta);
Rusty Russelldf60aee2009-06-12 22:27:09 -0600170bool send_notify_to_eventfd(struct lg_cpu *cpu);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200171void init_clockdev(struct lg_cpu *cpu);
Rusty Russellc18acd72007-10-22 11:03:35 +1000172bool check_syscall_vector(struct lguest *lg);
173int init_interrupts(void);
174void free_interrupts(void);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700175
176/* segments.c: */
177void setup_default_gdt_entries(struct lguest_ro_state *state);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200178void setup_guest_gdt(struct lg_cpu *cpu);
Rusty Russella489f0b2009-04-19 23:14:00 -0600179void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i,
180 u32 low, u32 hi);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200181void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array);
182void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt);
183void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700184
185/* page_tables.c: */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300186int init_guest_pagetable(struct lguest *lg);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700187void free_guest_pagetable(struct lguest *lg);
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200188void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable);
Matias Zabaljaureguiebe0ba82009-05-30 15:48:08 -0300189void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i);
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600190#ifdef CONFIG_X86_PAE
191void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
192#endif
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200193void guest_pagetable_clear_all(struct lg_cpu *cpu);
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200194void guest_pagetable_flush_user(struct lg_cpu *cpu);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200195void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir,
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000196 unsigned long vaddr, pte_t val);
Glauber de Oliveira Costa0c784412008-01-07 11:05:30 -0200197void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages);
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300198bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode);
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200199void pin_page(struct lg_cpu *cpu, unsigned long vaddr);
200unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200201void page_table_guest_data_init(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700202
Jes Sorensen625efab2007-10-22 11:03:28 +1000203/* <arch>/core.c: */
204void lguest_arch_host_init(void);
205void lguest_arch_host_fini(void);
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200206void lguest_arch_run_guest(struct lg_cpu *cpu);
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200207void lguest_arch_handle_trap(struct lg_cpu *cpu);
208int lguest_arch_init_hypercalls(struct lg_cpu *cpu);
209int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args);
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200210void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start);
Jes Sorensen625efab2007-10-22 11:03:28 +1000211
212/* <arch>/switcher.S: */
213extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
214
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700215/* lguest_user.c: */
216int lguest_device_init(void);
217void lguest_device_remove(void);
218
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700219/* hypercalls.c: */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200220void do_hypercalls(struct lg_cpu *cpu);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200221void write_timestamp(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700222
Rusty Russelldde79782007-07-26 10:41:03 -0700223/*L:035
224 * Let's step aside for the moment, to study one important routine that's used
225 * widely in the Host code.
226 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000227 * There are many cases where the Guest can do something invalid, like pass crap
Rusty Russelldde79782007-07-26 10:41:03 -0700228 * to a hypercall. Since only the Guest kernel can make hypercalls, it's quite
229 * acceptable to simply terminate the Guest and give the Launcher a nicely
230 * formatted reason. It's also simpler for the Guest itself, which doesn't
231 * need to check most hypercalls for "success"; if you're still running, it
232 * succeeded.
233 *
234 * Once this is called, the Guest will never run again, so most Host code can
235 * call this then continue as if nothing had happened. This means many
236 * functions don't have to explicitly return an error code, which keeps the
237 * code simple.
238 *
239 * It also means that this can be called more than once: only the first one is
240 * remembered. The only trick is that we still need to kill the Guest even if
241 * we can't allocate memory to store the reason. Linux has a neat way of
242 * packing error codes into invalid pointers, so we use that here.
243 *
244 * Like any macro which uses an "if", it is safely wrapped in a run-once "do {
245 * } while(0)".
246 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200247#define kill_guest(cpu, fmt...) \
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700248do { \
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200249 if (!(cpu)->lg->dead) { \
250 (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt); \
251 if (!(cpu)->lg->dead) \
252 (cpu)->lg->dead = ERR_PTR(-ENOMEM); \
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700253 } \
254} while(0)
Rusty Russelldde79782007-07-26 10:41:03 -0700255/* (End of aside) :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700256
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700257#endif /* __ASSEMBLY__ */
258#endif /* _LGUEST_H */