blob: bc28745d05af9794f756a5a29e64bb0b51a230ff [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>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070013
Jes Sorensen625efab2007-10-22 11:03:28 +100014#include <asm/lguest.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070015
16void free_pagetables(void);
17int init_pagetables(struct page **switcher_page, unsigned int pages);
18
Rusty Russell1842f232009-07-30 16:03:46 -060019struct pgdir {
Rusty Russellee3db0f2007-10-22 11:03:34 +100020 unsigned long gpgdir;
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +100021 pgd_t *pgdir;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070022};
23
Rusty Russelld7e28ff2007-07-19 01:49:23 -070024/* We have two pages shared with guests, per cpu. */
Rusty Russell1842f232009-07-30 16:03:46 -060025struct lguest_pages {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070026 /* This is the stack page mapped rw in guest */
27 char spare[PAGE_SIZE - sizeof(struct lguest_regs)];
28 struct lguest_regs regs;
29
30 /* This is the host state & guest descriptor page, ro in guest */
31 struct lguest_ro_state state;
32} __attribute__((aligned(PAGE_SIZE)));
33
34#define CHANGED_IDT 1
35#define CHANGED_GDT 2
36#define CHANGED_GDT_TLS 4 /* Actually a subset of CHANGED_GDT */
37#define CHANGED_ALL 3
38
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020039struct lg_cpu {
40 unsigned int id;
41 struct lguest *lg;
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020042 struct task_struct *tsk;
43 struct mm_struct *mm; /* == tsk->mm, but that becomes NULL on exit */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -020044
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020045 u32 cr2;
46 int ts;
47 u32 esp1;
Matias Zabaljaureguif0861222009-06-12 22:27:04 -060048 u16 ss1;
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -020049
Glauber de Oliveira Costaae3749d2008-01-17 19:14:46 -020050 /* Bitmap of what has changed: see CHANGED_* above. */
51 int changed;
52
Glauber de Oliveira Costa5e232f42008-01-07 11:05:36 -020053 unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */
54
Rusty Russell2e04ef72009-07-30 16:03:45 -060055 /* At end of a page shared mapped over lguest_pages in guest. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020056 unsigned long regs_page;
57 struct lguest_regs *regs;
58
Glauber de Oliveira Costaf34f8c52008-01-17 19:13:26 -020059 struct lguest_pages *last_pages;
60
Rusty Russell2e04ef72009-07-30 16:03:45 -060061 int cpu_pgd; /* Which pgd this cpu is currently using */
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -020062
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -020063 /* If a hypercall was asked for, this points to the arguments. */
64 struct hcall_args *hcall;
65 u32 next_hcall;
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -020066
67 /* Virtual clock device */
68 struct hrtimer hrt;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020069
Rusty Russell5dac0512009-06-12 22:27:10 -060070 /* Did the Guest tell us to halt? */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -020071 int halted;
72
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020073 /* Pending virtual interrupts */
74 DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -020075
76 struct lg_cpu_arch arch;
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020077};
78
Rusty Russelldf60aee2009-06-12 22:27:09 -060079struct lg_eventfd {
80 unsigned long addr;
Davide Libenzi13389012009-06-30 11:41:11 -070081 struct eventfd_ctx *event;
Rusty Russelldf60aee2009-06-12 22:27:09 -060082};
83
84struct lg_eventfd_map {
85 unsigned int num;
86 struct lg_eventfd map[];
87};
88
Rusty Russelld7e28ff2007-07-19 01:49:23 -070089/* The private info the thread maintains about the guest. */
Rusty Russell1842f232009-07-30 16:03:46 -060090struct lguest {
Rusty Russelld7e28ff2007-07-19 01:49:23 -070091 struct lguest_data __user *lguest_data;
Glauber de Oliveira Costabadb1e02008-01-07 11:05:22 -020092 struct lg_cpu cpus[NR_CPUS];
93 unsigned int nr_cpus;
94
Rusty Russelld7e28ff2007-07-19 01:49:23 -070095 u32 pfn_limit;
Rusty Russell2e04ef72009-07-30 16:03:45 -060096
97 /*
98 * This provides the offset to the base of guest-physical memory in the
99 * Launcher.
100 */
Rusty Russell3c6b5bf2007-10-22 11:03:26 +1000101 void __user *mem_base;
Rusty Russell47436aa2007-10-22 11:03:36 +1000102 unsigned long kernel_address;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700103
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700104 struct pgdir pgdirs[4];
105
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700106 unsigned long noirq_start, noirq_end;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700107
108 unsigned int stack_pages;
109 u32 tsc_khz;
110
Rusty Russelldf60aee2009-06-12 22:27:09 -0600111 struct lg_eventfd_map *eventfds;
112
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700113 /* Dead? */
114 const char *dead;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700115};
116
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700117extern struct mutex lguest_lock;
118
119/* core.c: */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300120bool lguest_address_ok(const struct lguest *lg,
121 unsigned long addr, unsigned long len);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200122void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
123void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
Rusty Russell2d37f942007-10-22 11:24:24 +1000124
Rusty Russell2e04ef72009-07-30 16:03:45 -0600125/*H:035
126 * Using memory-copy operations like that is usually inconvient, so we
Rusty Russell2d37f942007-10-22 11:24:24 +1000127 * have the following helper macros which read and write a specific type (often
128 * an unsigned long).
129 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600130 * This reads into a variable of the given type then returns that.
131 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200132#define lgread(cpu, addr, type) \
133 ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; })
Rusty Russell2d37f942007-10-22 11:24:24 +1000134
135/* This checks that the variable is of the given type, then writes it out. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200136#define lgwrite(cpu, addr, type, val) \
Rusty Russell2d37f942007-10-22 11:24:24 +1000137 do { \
138 typecheck(type, val); \
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200139 __lgwrite((cpu), (addr), &(val), sizeof(val)); \
Rusty Russell2d37f942007-10-22 11:24:24 +1000140 } while(0)
141/* (end of memory access helper routines) :*/
142
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200143int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700144
Rusty Russell2e04ef72009-07-30 16:03:45 -0600145/*
146 * Helper macros to obtain the first 12 or the last 20 bits, this is only the
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000147 * first step in the migration to the kernel types. pte_pfn is already defined
Rusty Russell2e04ef72009-07-30 16:03:45 -0600148 * in the kernel.
149 */
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000150#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK)
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000151#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT)
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600152#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK)
153#define pmd_pfn(x) (pmd_val(x) >> PAGE_SHIFT)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700154
155/* interrupts_and_traps.c: */
Rusty Russella32a8812009-06-12 22:27:02 -0600156unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more);
157void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more);
Rusty Russell9f155a92009-06-12 22:27:08 -0600158void set_interrupt(struct lg_cpu *cpu, unsigned int irq);
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300159bool deliver_trap(struct lg_cpu *cpu, unsigned int num);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200160void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i,
161 u32 low, u32 hi);
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200162void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages);
163void pin_stack_pages(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700164void setup_default_idt_entries(struct lguest_ro_state *state,
165 const unsigned long *def);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200166void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700167 const unsigned long *def);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200168void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta);
Rusty Russelldf60aee2009-06-12 22:27:09 -0600169bool send_notify_to_eventfd(struct lg_cpu *cpu);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200170void init_clockdev(struct lg_cpu *cpu);
Rusty Russellc18acd72007-10-22 11:03:35 +1000171bool check_syscall_vector(struct lguest *lg);
172int init_interrupts(void);
173void free_interrupts(void);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700174
175/* segments.c: */
176void setup_default_gdt_entries(struct lguest_ro_state *state);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200177void setup_guest_gdt(struct lg_cpu *cpu);
Rusty Russella489f0b2009-04-19 23:14:00 -0600178void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i,
179 u32 low, u32 hi);
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200180void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array);
181void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt);
182void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700183
184/* page_tables.c: */
Matias Zabaljauregui58a24562008-09-29 01:40:07 -0300185int init_guest_pagetable(struct lguest *lg);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700186void free_guest_pagetable(struct lguest *lg);
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200187void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable);
Matias Zabaljaureguiebe0ba82009-05-30 15:48:08 -0300188void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i);
Matias Zabaljaureguiacdd0b62009-06-12 22:27:07 -0600189#ifdef CONFIG_X86_PAE
190void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
191#endif
Glauber de Oliveira Costa4665ac82008-01-07 11:05:35 -0200192void guest_pagetable_clear_all(struct lg_cpu *cpu);
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200193void guest_pagetable_flush_user(struct lg_cpu *cpu);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200194void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir,
Matias Zabaljaureguidf29f432007-10-22 11:03:33 +1000195 unsigned long vaddr, pte_t val);
Glauber de Oliveira Costa0c784412008-01-07 11:05:30 -0200196void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages);
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300197bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode);
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200198void pin_page(struct lg_cpu *cpu, unsigned long vaddr);
199unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200200void page_table_guest_data_init(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700201
Jes Sorensen625efab2007-10-22 11:03:28 +1000202/* <arch>/core.c: */
203void lguest_arch_host_init(void);
204void lguest_arch_host_fini(void);
Glauber de Oliveira Costad0953d42008-01-07 11:05:25 -0200205void lguest_arch_run_guest(struct lg_cpu *cpu);
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200206void lguest_arch_handle_trap(struct lg_cpu *cpu);
207int lguest_arch_init_hypercalls(struct lg_cpu *cpu);
208int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args);
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200209void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start);
Jes Sorensen625efab2007-10-22 11:03:28 +1000210
211/* <arch>/switcher.S: */
212extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
213
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700214/* lguest_user.c: */
215int lguest_device_init(void);
216void lguest_device_remove(void);
217
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700218/* hypercalls.c: */
Glauber de Oliveira Costa73044f02008-01-07 11:05:27 -0200219void do_hypercalls(struct lg_cpu *cpu);
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200220void write_timestamp(struct lg_cpu *cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700221
Rusty Russelldde79782007-07-26 10:41:03 -0700222/*L:035
223 * Let's step aside for the moment, to study one important routine that's used
224 * widely in the Host code.
225 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000226 * There are many cases where the Guest can do something invalid, like pass crap
Rusty Russelldde79782007-07-26 10:41:03 -0700227 * to a hypercall. Since only the Guest kernel can make hypercalls, it's quite
228 * acceptable to simply terminate the Guest and give the Launcher a nicely
229 * formatted reason. It's also simpler for the Guest itself, which doesn't
230 * need to check most hypercalls for "success"; if you're still running, it
231 * succeeded.
232 *
233 * Once this is called, the Guest will never run again, so most Host code can
234 * call this then continue as if nothing had happened. This means many
235 * functions don't have to explicitly return an error code, which keeps the
236 * code simple.
237 *
238 * It also means that this can be called more than once: only the first one is
239 * remembered. The only trick is that we still need to kill the Guest even if
240 * we can't allocate memory to store the reason. Linux has a neat way of
241 * packing error codes into invalid pointers, so we use that here.
242 *
243 * Like any macro which uses an "if", it is safely wrapped in a run-once "do {
244 * } while(0)".
245 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200246#define kill_guest(cpu, fmt...) \
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700247do { \
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200248 if (!(cpu)->lg->dead) { \
249 (cpu)->lg->dead = kasprintf(GFP_ATOMIC, fmt); \
250 if (!(cpu)->lg->dead) \
251 (cpu)->lg->dead = ERR_PTR(-ENOMEM); \
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700252 } \
253} while(0)
Rusty Russelldde79782007-07-26 10:41:03 -0700254/* (End of aside) :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700255
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700256#endif /* __ASSEMBLY__ */
257#endif /* _LGUEST_H */