Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86_64/kernel/vsyscall.c | |
3 | * | |
4 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Copyright 2003 Andi Kleen, SuSE Labs. | |
6 | * | |
7 | * Thanks to hpa@transmeta.com for some useful hint. | |
8 | * Special thanks to Ingo Molnar for his early experience with | |
9 | * a different vsyscall implementation for Linux/IA32 and for the name. | |
10 | * | |
11 | * vsyscall 1 is located at -10Mbyte, vsyscall 2 is located | |
12 | * at virtual address -10Mbyte+1024bytes etc... There are at max 4 | |
13 | * vsyscalls. One vsyscall can reserve more than 1 slot to avoid | |
14 | * jumping out of line if necessary. We cannot add more with this | |
15 | * mechanism because older kernels won't return -ENOSYS. | |
16 | * If we want more than four we need a vDSO. | |
17 | * | |
18 | * Note: the concept clashes with user mode linux. If you use UML and | |
19 | * want per guest time just set the kernel.vsyscall64 sysctl to 0. | |
20 | */ | |
21 | ||
22 | #include <linux/time.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/kernel.h> | |
25 | #include <linux/timer.h> | |
26 | #include <linux/seqlock.h> | |
27 | #include <linux/jiffies.h> | |
28 | #include <linux/sysctl.h> | |
29 | ||
30 | #include <asm/vsyscall.h> | |
31 | #include <asm/pgtable.h> | |
32 | #include <asm/page.h> | |
33 | #include <asm/fixmap.h> | |
34 | #include <asm/errno.h> | |
35 | #include <asm/io.h> | |
36 | ||
37 | #define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr))) | |
38 | #define force_inline __attribute__((always_inline)) inline | |
39 | ||
40 | int __sysctl_vsyscall __section_sysctl_vsyscall = 1; | |
41 | seqlock_t __xtime_lock __section_xtime_lock = SEQLOCK_UNLOCKED; | |
42 | ||
43 | #include <asm/unistd.h> | |
44 | ||
45 | static force_inline void timeval_normalize(struct timeval * tv) | |
46 | { | |
47 | time_t __sec; | |
48 | ||
49 | __sec = tv->tv_usec / 1000000; | |
50 | if (__sec) { | |
51 | tv->tv_usec %= 1000000; | |
52 | tv->tv_sec += __sec; | |
53 | } | |
54 | } | |
55 | ||
56 | static force_inline void do_vgettimeofday(struct timeval * tv) | |
57 | { | |
58 | long sequence, t; | |
59 | unsigned long sec, usec; | |
60 | ||
61 | do { | |
62 | sequence = read_seqbegin(&__xtime_lock); | |
63 | ||
64 | sec = __xtime.tv_sec; | |
65 | usec = (__xtime.tv_nsec / 1000) + | |
66 | (__jiffies - __wall_jiffies) * (1000000 / HZ); | |
67 | ||
68 | if (__vxtime.mode == VXTIME_TSC) { | |
69 | sync_core(); | |
70 | rdtscll(t); | |
71 | if (t < __vxtime.last_tsc) | |
72 | t = __vxtime.last_tsc; | |
73 | usec += ((t - __vxtime.last_tsc) * | |
74 | __vxtime.tsc_quot) >> 32; | |
75 | /* See comment in x86_64 do_gettimeofday. */ | |
76 | } else { | |
77 | usec += ((readl((void *)fix_to_virt(VSYSCALL_HPET) + 0xf0) - | |
78 | __vxtime.last) * __vxtime.quot) >> 32; | |
79 | } | |
80 | } while (read_seqretry(&__xtime_lock, sequence)); | |
81 | ||
82 | tv->tv_sec = sec + usec / 1000000; | |
83 | tv->tv_usec = usec % 1000000; | |
84 | } | |
85 | ||
86 | /* RED-PEN may want to readd seq locking, but then the variable should be write-once. */ | |
87 | static force_inline void do_get_tz(struct timezone * tz) | |
88 | { | |
89 | *tz = __sys_tz; | |
90 | } | |
91 | ||
92 | static force_inline int gettimeofday(struct timeval *tv, struct timezone *tz) | |
93 | { | |
94 | int ret; | |
95 | asm volatile("vsysc2: syscall" | |
96 | : "=a" (ret) | |
97 | : "0" (__NR_gettimeofday),"D" (tv),"S" (tz) : __syscall_clobber ); | |
98 | return ret; | |
99 | } | |
100 | ||
101 | static force_inline long time_syscall(long *t) | |
102 | { | |
103 | long secs; | |
104 | asm volatile("vsysc1: syscall" | |
105 | : "=a" (secs) | |
106 | : "0" (__NR_time),"D" (t) : __syscall_clobber); | |
107 | return secs; | |
108 | } | |
109 | ||
110 | static int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz) | |
111 | { | |
112 | if (unlikely(!__sysctl_vsyscall)) | |
113 | return gettimeofday(tv,tz); | |
114 | if (tv) | |
115 | do_vgettimeofday(tv); | |
116 | if (tz) | |
117 | do_get_tz(tz); | |
118 | return 0; | |
119 | } | |
120 | ||
121 | /* This will break when the xtime seconds get inaccurate, but that is | |
122 | * unlikely */ | |
123 | static time_t __vsyscall(1) vtime(time_t *t) | |
124 | { | |
125 | if (unlikely(!__sysctl_vsyscall)) | |
126 | return time_syscall(t); | |
127 | else if (t) | |
128 | *t = __xtime.tv_sec; | |
129 | return __xtime.tv_sec; | |
130 | } | |
131 | ||
132 | static long __vsyscall(2) venosys_0(void) | |
133 | { | |
134 | return -ENOSYS; | |
135 | } | |
136 | ||
137 | static long __vsyscall(3) venosys_1(void) | |
138 | { | |
139 | return -ENOSYS; | |
140 | } | |
141 | ||
142 | #ifdef CONFIG_SYSCTL | |
143 | ||
144 | #define SYSCALL 0x050f | |
145 | #define NOP2 0x9090 | |
146 | ||
147 | /* | |
148 | * NOP out syscall in vsyscall page when not needed. | |
149 | */ | |
150 | static int vsyscall_sysctl_change(ctl_table *ctl, int write, struct file * filp, | |
151 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
152 | { | |
153 | extern u16 vsysc1, vsysc2; | |
154 | u16 *map1, *map2; | |
155 | int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos); | |
156 | if (!write) | |
157 | return ret; | |
158 | /* gcc has some trouble with __va(__pa()), so just do it this | |
159 | way. */ | |
160 | map1 = ioremap(__pa_symbol(&vsysc1), 2); | |
161 | if (!map1) | |
162 | return -ENOMEM; | |
163 | map2 = ioremap(__pa_symbol(&vsysc2), 2); | |
164 | if (!map2) { | |
165 | ret = -ENOMEM; | |
166 | goto out; | |
167 | } | |
168 | if (!sysctl_vsyscall) { | |
169 | *map1 = SYSCALL; | |
170 | *map2 = SYSCALL; | |
171 | } else { | |
172 | *map1 = NOP2; | |
173 | *map2 = NOP2; | |
174 | } | |
175 | iounmap(map2); | |
176 | out: | |
177 | iounmap(map1); | |
178 | return ret; | |
179 | } | |
180 | ||
181 | static int vsyscall_sysctl_nostrat(ctl_table *t, int __user *name, int nlen, | |
182 | void __user *oldval, size_t __user *oldlenp, | |
183 | void __user *newval, size_t newlen, | |
184 | void **context) | |
185 | { | |
186 | return -ENOSYS; | |
187 | } | |
188 | ||
189 | static ctl_table kernel_table2[] = { | |
190 | { .ctl_name = 99, .procname = "vsyscall64", | |
191 | .data = &sysctl_vsyscall, .maxlen = sizeof(int), .mode = 0644, | |
192 | .strategy = vsyscall_sysctl_nostrat, | |
193 | .proc_handler = vsyscall_sysctl_change }, | |
194 | { 0, } | |
195 | }; | |
196 | ||
197 | static ctl_table kernel_root_table2[] = { | |
198 | { .ctl_name = CTL_KERN, .procname = "kernel", .mode = 0555, | |
199 | .child = kernel_table2 }, | |
200 | { 0 }, | |
201 | }; | |
202 | ||
203 | #endif | |
204 | ||
205 | static void __init map_vsyscall(void) | |
206 | { | |
207 | extern char __vsyscall_0; | |
208 | unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0); | |
209 | ||
210 | __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL); | |
211 | } | |
212 | ||
213 | static int __init vsyscall_init(void) | |
214 | { | |
215 | BUG_ON(((unsigned long) &vgettimeofday != | |
216 | VSYSCALL_ADDR(__NR_vgettimeofday))); | |
217 | BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime)); | |
218 | BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE))); | |
219 | map_vsyscall(); | |
220 | sysctl_vsyscall = 1; | |
221 | register_sysctl_table(kernel_root_table2, 0); | |
222 | return 0; | |
223 | } | |
224 | ||
225 | __initcall(vsyscall_init); |