Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE |
3 | * Copyright 2003 Andi Kleen, SuSE Labs. | |
4 | * | |
5 | * Thanks to hpa@transmeta.com for some useful hint. | |
6 | * Special thanks to Ingo Molnar for his early experience with | |
7 | * a different vsyscall implementation for Linux/IA32 and for the name. | |
8 | * | |
9 | * vsyscall 1 is located at -10Mbyte, vsyscall 2 is located | |
10 | * at virtual address -10Mbyte+1024bytes etc... There are at max 4 | |
11 | * vsyscalls. One vsyscall can reserve more than 1 slot to avoid | |
12 | * jumping out of line if necessary. We cannot add more with this | |
13 | * mechanism because older kernels won't return -ENOSYS. | |
14 | * If we want more than four we need a vDSO. | |
15 | * | |
16 | * Note: the concept clashes with user mode linux. If you use UML and | |
17 | * want per guest time just set the kernel.vsyscall64 sysctl to 0. | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/init.h> | |
22 | #include <linux/kernel.h> | |
23 | #include <linux/timer.h> | |
24 | #include <linux/seqlock.h> | |
25 | #include <linux/jiffies.h> | |
26 | #include <linux/sysctl.h> | |
7460ed28 | 27 | #include <linux/clocksource.h> |
c08c8205 | 28 | #include <linux/getcpu.h> |
8c131af1 AK |
29 | #include <linux/cpu.h> |
30 | #include <linux/smp.h> | |
31 | #include <linux/notifier.h> | |
1da177e4 LT |
32 | |
33 | #include <asm/vsyscall.h> | |
34 | #include <asm/pgtable.h> | |
35 | #include <asm/page.h> | |
7460ed28 | 36 | #include <asm/unistd.h> |
1da177e4 LT |
37 | #include <asm/fixmap.h> |
38 | #include <asm/errno.h> | |
39 | #include <asm/io.h> | |
c08c8205 VP |
40 | #include <asm/segment.h> |
41 | #include <asm/desc.h> | |
42 | #include <asm/topology.h> | |
2aae950b | 43 | #include <asm/vgtod.h> |
1da177e4 LT |
44 | |
45 | #define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr))) | |
f5738cee | 46 | #define __syscall_clobber "r11","rcx","memory" |
0dbf7028 VG |
47 | #define __pa_vsymbol(x) \ |
48 | ({unsigned long v; \ | |
49 | extern char __vsyscall_0; \ | |
50 | asm("" : "=r" (v) : "0" (x)); \ | |
51 | ((v - VSYSCALL_FIRST_PAGE) + __pa_symbol(&__vsyscall_0)); }) | |
1da177e4 | 52 | |
c8118c6c ED |
53 | /* |
54 | * vsyscall_gtod_data contains data that is : | |
55 | * - readonly from vsyscalls | |
56 | * - writen by timer interrupt or systcl (/proc/sys/kernel/vsyscall64) | |
57 | * Try to keep this structure as small as possible to avoid cache line ping pongs | |
58 | */ | |
c08c8205 | 59 | int __vgetcpu_mode __section_vgetcpu_mode; |
1da177e4 | 60 | |
2aae950b | 61 | struct vsyscall_gtod_data __vsyscall_gtod_data __section_vsyscall_gtod_data = |
1da177e4 | 62 | { |
7460ed28 | 63 | .lock = SEQLOCK_UNLOCKED, |
64 | .sysctl_enabled = 1, | |
65 | }; | |
1da177e4 | 66 | |
7460ed28 | 67 | void update_vsyscall(struct timespec *wall_time, struct clocksource *clock) |
1da177e4 | 68 | { |
7460ed28 | 69 | unsigned long flags; |
1da177e4 | 70 | |
7460ed28 | 71 | write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags); |
72 | /* copy vsyscall data */ | |
c8118c6c ED |
73 | vsyscall_gtod_data.clock.vread = clock->vread; |
74 | vsyscall_gtod_data.clock.cycle_last = clock->cycle_last; | |
75 | vsyscall_gtod_data.clock.mask = clock->mask; | |
76 | vsyscall_gtod_data.clock.mult = clock->mult; | |
77 | vsyscall_gtod_data.clock.shift = clock->shift; | |
78 | vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec; | |
79 | vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec; | |
7460ed28 | 80 | vsyscall_gtod_data.sys_tz = sys_tz; |
2aae950b AK |
81 | vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec; |
82 | vsyscall_gtod_data.wall_to_monotonic = wall_to_monotonic; | |
7460ed28 | 83 | write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags); |
1da177e4 LT |
84 | } |
85 | ||
7460ed28 | 86 | /* RED-PEN may want to readd seq locking, but then the variable should be |
87 | * write-once. | |
88 | */ | |
2c8bc944 | 89 | static __always_inline void do_get_tz(struct timezone * tz) |
1da177e4 | 90 | { |
7460ed28 | 91 | *tz = __vsyscall_gtod_data.sys_tz; |
1da177e4 LT |
92 | } |
93 | ||
2c8bc944 | 94 | static __always_inline int gettimeofday(struct timeval *tv, struct timezone *tz) |
1da177e4 LT |
95 | { |
96 | int ret; | |
97 | asm volatile("vsysc2: syscall" | |
98 | : "=a" (ret) | |
7460ed28 | 99 | : "0" (__NR_gettimeofday),"D" (tv),"S" (tz) |
100 | : __syscall_clobber ); | |
1da177e4 LT |
101 | return ret; |
102 | } | |
103 | ||
2c8bc944 | 104 | static __always_inline long time_syscall(long *t) |
1da177e4 LT |
105 | { |
106 | long secs; | |
107 | asm volatile("vsysc1: syscall" | |
108 | : "=a" (secs) | |
109 | : "0" (__NR_time),"D" (t) : __syscall_clobber); | |
110 | return secs; | |
111 | } | |
112 | ||
7460ed28 | 113 | static __always_inline void do_vgettimeofday(struct timeval * tv) |
114 | { | |
115 | cycle_t now, base, mask, cycle_delta; | |
c8118c6c ED |
116 | unsigned seq; |
117 | unsigned long mult, shift, nsec; | |
7460ed28 | 118 | cycle_t (*vread)(void); |
119 | do { | |
120 | seq = read_seqbegin(&__vsyscall_gtod_data.lock); | |
121 | ||
122 | vread = __vsyscall_gtod_data.clock.vread; | |
123 | if (unlikely(!__vsyscall_gtod_data.sysctl_enabled || !vread)) { | |
89952d13 | 124 | gettimeofday(tv,NULL); |
7460ed28 | 125 | return; |
126 | } | |
127 | now = vread(); | |
128 | base = __vsyscall_gtod_data.clock.cycle_last; | |
129 | mask = __vsyscall_gtod_data.clock.mask; | |
130 | mult = __vsyscall_gtod_data.clock.mult; | |
131 | shift = __vsyscall_gtod_data.clock.shift; | |
132 | ||
c8118c6c ED |
133 | tv->tv_sec = __vsyscall_gtod_data.wall_time_sec; |
134 | nsec = __vsyscall_gtod_data.wall_time_nsec; | |
7460ed28 | 135 | } while (read_seqretry(&__vsyscall_gtod_data.lock, seq)); |
136 | ||
137 | /* calculate interval: */ | |
138 | cycle_delta = (now - base) & mask; | |
139 | /* convert to nsecs: */ | |
c8118c6c | 140 | nsec += (cycle_delta * mult) >> shift; |
7460ed28 | 141 | |
c8118c6c | 142 | while (nsec >= NSEC_PER_SEC) { |
7460ed28 | 143 | tv->tv_sec += 1; |
c8118c6c | 144 | nsec -= NSEC_PER_SEC; |
7460ed28 | 145 | } |
c8118c6c | 146 | tv->tv_usec = nsec / NSEC_PER_USEC; |
7460ed28 | 147 | } |
148 | ||
2e8ad43e | 149 | int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz) |
1da177e4 | 150 | { |
1da177e4 LT |
151 | if (tv) |
152 | do_vgettimeofday(tv); | |
153 | if (tz) | |
154 | do_get_tz(tz); | |
155 | return 0; | |
156 | } | |
157 | ||
158 | /* This will break when the xtime seconds get inaccurate, but that is | |
159 | * unlikely */ | |
2e8ad43e | 160 | time_t __vsyscall(1) vtime(time_t *t) |
1da177e4 | 161 | { |
d0aff6e6 | 162 | struct timeval tv; |
272a3713 | 163 | time_t result; |
7460ed28 | 164 | if (unlikely(!__vsyscall_gtod_data.sysctl_enabled)) |
1da177e4 | 165 | return time_syscall(t); |
d0aff6e6 | 166 | |
167 | vgettimeofday(&tv, 0); | |
168 | result = tv.tv_sec; | |
272a3713 ED |
169 | if (t) |
170 | *t = result; | |
171 | return result; | |
1da177e4 LT |
172 | } |
173 | ||
c08c8205 VP |
174 | /* Fast way to get current CPU and node. |
175 | This helps to do per node and per CPU caches in user space. | |
176 | The result is not guaranteed without CPU affinity, but usually | |
177 | works out because the scheduler tries to keep a thread on the same | |
178 | CPU. | |
179 | ||
180 | tcache must point to a two element sized long array. | |
181 | All arguments can be NULL. */ | |
182 | long __vsyscall(2) | |
183 | vgetcpu(unsigned *cpu, unsigned *node, struct getcpu_cache *tcache) | |
1da177e4 | 184 | { |
c08c8205 VP |
185 | unsigned int dummy, p; |
186 | unsigned long j = 0; | |
187 | ||
188 | /* Fast cache - only recompute value once per jiffies and avoid | |
189 | relatively costly rdtscp/cpuid otherwise. | |
190 | This works because the scheduler usually keeps the process | |
191 | on the same CPU and this syscall doesn't guarantee its | |
192 | results anyways. | |
193 | We do this here because otherwise user space would do it on | |
194 | its own in a likely inferior way (no access to jiffies). | |
195 | If you don't like it pass NULL. */ | |
34596dc9 AK |
196 | if (tcache && tcache->blob[0] == (j = __jiffies)) { |
197 | p = tcache->blob[1]; | |
c08c8205 VP |
198 | } else if (__vgetcpu_mode == VGETCPU_RDTSCP) { |
199 | /* Load per CPU data from RDTSCP */ | |
200 | rdtscp(dummy, dummy, p); | |
201 | } else { | |
202 | /* Load per CPU data from GDT */ | |
203 | asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG)); | |
204 | } | |
205 | if (tcache) { | |
34596dc9 AK |
206 | tcache->blob[0] = j; |
207 | tcache->blob[1] = p; | |
c08c8205 VP |
208 | } |
209 | if (cpu) | |
210 | *cpu = p & 0xfff; | |
211 | if (node) | |
212 | *node = p >> 12; | |
213 | return 0; | |
1da177e4 LT |
214 | } |
215 | ||
2e8ad43e | 216 | long __vsyscall(3) venosys_1(void) |
1da177e4 LT |
217 | { |
218 | return -ENOSYS; | |
219 | } | |
220 | ||
221 | #ifdef CONFIG_SYSCTL | |
222 | ||
223 | #define SYSCALL 0x050f | |
224 | #define NOP2 0x9090 | |
225 | ||
226 | /* | |
227 | * NOP out syscall in vsyscall page when not needed. | |
228 | */ | |
229 | static int vsyscall_sysctl_change(ctl_table *ctl, int write, struct file * filp, | |
230 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
231 | { | |
232 | extern u16 vsysc1, vsysc2; | |
131cfd7b AK |
233 | u16 __iomem *map1; |
234 | u16 __iomem *map2; | |
1da177e4 LT |
235 | int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos); |
236 | if (!write) | |
237 | return ret; | |
238 | /* gcc has some trouble with __va(__pa()), so just do it this | |
239 | way. */ | |
0dbf7028 | 240 | map1 = ioremap(__pa_vsymbol(&vsysc1), 2); |
1da177e4 LT |
241 | if (!map1) |
242 | return -ENOMEM; | |
0dbf7028 | 243 | map2 = ioremap(__pa_vsymbol(&vsysc2), 2); |
1da177e4 LT |
244 | if (!map2) { |
245 | ret = -ENOMEM; | |
246 | goto out; | |
247 | } | |
7460ed28 | 248 | if (!vsyscall_gtod_data.sysctl_enabled) { |
131cfd7b AK |
249 | writew(SYSCALL, map1); |
250 | writew(SYSCALL, map2); | |
1da177e4 | 251 | } else { |
131cfd7b AK |
252 | writew(NOP2, map1); |
253 | writew(NOP2, map2); | |
1da177e4 LT |
254 | } |
255 | iounmap(map2); | |
256 | out: | |
257 | iounmap(map1); | |
258 | return ret; | |
259 | } | |
260 | ||
261 | static int vsyscall_sysctl_nostrat(ctl_table *t, int __user *name, int nlen, | |
262 | void __user *oldval, size_t __user *oldlenp, | |
1f29bcd7 | 263 | void __user *newval, size_t newlen) |
1da177e4 LT |
264 | { |
265 | return -ENOSYS; | |
266 | } | |
267 | ||
268 | static ctl_table kernel_table2[] = { | |
269 | { .ctl_name = 99, .procname = "vsyscall64", | |
7460ed28 | 270 | .data = &vsyscall_gtod_data.sysctl_enabled, .maxlen = sizeof(int), |
271 | .mode = 0644, | |
1da177e4 LT |
272 | .strategy = vsyscall_sysctl_nostrat, |
273 | .proc_handler = vsyscall_sysctl_change }, | |
7a44d37d | 274 | {} |
1da177e4 LT |
275 | }; |
276 | ||
277 | static ctl_table kernel_root_table2[] = { | |
278 | { .ctl_name = CTL_KERN, .procname = "kernel", .mode = 0555, | |
279 | .child = kernel_table2 }, | |
7a44d37d | 280 | {} |
1da177e4 LT |
281 | }; |
282 | ||
283 | #endif | |
284 | ||
8c131af1 AK |
285 | /* Assume __initcall executes before all user space. Hopefully kmod |
286 | doesn't violate that. We'll find out if it does. */ | |
287 | static void __cpuinit vsyscall_set_cpu(int cpu) | |
c08c8205 VP |
288 | { |
289 | unsigned long *d; | |
290 | unsigned long node = 0; | |
291 | #ifdef CONFIG_NUMA | |
292 | node = cpu_to_node[cpu]; | |
293 | #endif | |
8c131af1 AK |
294 | if (cpu_has(&cpu_data[cpu], X86_FEATURE_RDTSCP)) |
295 | write_rdtscp_aux((node << 12) | cpu); | |
c08c8205 VP |
296 | |
297 | /* Store cpu number in limit so that it can be loaded quickly | |
298 | in user space in vgetcpu. | |
299 | 12 bits for the CPU and 8 bits for the node. */ | |
300 | d = (unsigned long *)(cpu_gdt(cpu) + GDT_ENTRY_PER_CPU); | |
301 | *d = 0x0f40000000000ULL; | |
302 | *d |= cpu; | |
303 | *d |= (node & 0xf) << 12; | |
304 | *d |= (node >> 4) << 48; | |
305 | } | |
306 | ||
8c131af1 AK |
307 | static void __cpuinit cpu_vsyscall_init(void *arg) |
308 | { | |
309 | /* preemption should be already off */ | |
310 | vsyscall_set_cpu(raw_smp_processor_id()); | |
311 | } | |
312 | ||
313 | static int __cpuinit | |
314 | cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg) | |
315 | { | |
316 | long cpu = (long)arg; | |
8bb78442 | 317 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) |
8c131af1 AK |
318 | smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 0, 1); |
319 | return NOTIFY_DONE; | |
320 | } | |
321 | ||
1da177e4 LT |
322 | static void __init map_vsyscall(void) |
323 | { | |
324 | extern char __vsyscall_0; | |
325 | unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0); | |
326 | ||
103efcd9 | 327 | /* Note that VSYSCALL_MAPPED_PAGES must agree with the code below. */ |
1da177e4 LT |
328 | __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL); |
329 | } | |
330 | ||
331 | static int __init vsyscall_init(void) | |
332 | { | |
333 | BUG_ON(((unsigned long) &vgettimeofday != | |
334 | VSYSCALL_ADDR(__NR_vgettimeofday))); | |
335 | BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime)); | |
336 | BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE))); | |
c08c8205 | 337 | BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu)); |
1da177e4 | 338 | map_vsyscall(); |
f3c5f5e7 | 339 | #ifdef CONFIG_SYSCTL |
0b4d4147 | 340 | register_sysctl_table(kernel_root_table2); |
f3c5f5e7 | 341 | #endif |
8c131af1 AK |
342 | on_each_cpu(cpu_vsyscall_init, NULL, 0, 1); |
343 | hotcpu_notifier(cpu_vsyscall_notifier, 0); | |
1da177e4 LT |
344 | return 0; |
345 | } | |
346 | ||
347 | __initcall(vsyscall_init); |