2 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
3 * Copyright 2003 Andi Kleen, SuSE Labs.
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.
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.
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.
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>
27 #include <linux/clocksource.h>
28 #include <linux/getcpu.h>
29 #include <linux/cpu.h>
30 #include <linux/smp.h>
31 #include <linux/notifier.h>
33 #include <asm/vsyscall.h>
34 #include <asm/pgtable.h>
36 #include <asm/unistd.h>
37 #include <asm/fixmap.h>
38 #include <asm/errno.h>
40 #include <asm/segment.h>
42 #include <asm/topology.h>
43 #include <asm/vgtod.h>
45 #define __vsyscall(nr) \
46 __attribute__ ((unused, __section__(".vsyscall_" #nr))) notrace
47 #define __syscall_clobber "r11","cx","memory"
50 * vsyscall_gtod_data contains data that is :
51 * - readonly from vsyscalls
52 * - written by timer interrupt or systcl (/proc/sys/kernel/vsyscall64)
53 * Try to keep this structure as small as possible to avoid cache line ping pongs
55 int __vgetcpu_mode __section_vgetcpu_mode;
57 struct vsyscall_gtod_data __vsyscall_gtod_data __section_vsyscall_gtod_data =
59 .lock = SEQLOCK_UNLOCKED,
63 void update_vsyscall_tz(void)
67 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
68 /* sys_tz has changed */
69 vsyscall_gtod_data.sys_tz = sys_tz;
70 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
73 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock)
77 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
78 /* copy vsyscall data */
79 vsyscall_gtod_data.clock.vread = clock->vread;
80 vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
81 vsyscall_gtod_data.clock.mask = clock->mask;
82 vsyscall_gtod_data.clock.mult = clock->mult;
83 vsyscall_gtod_data.clock.shift = clock->shift;
84 vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
85 vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
86 vsyscall_gtod_data.wall_to_monotonic = wall_to_monotonic;
87 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
90 /* RED-PEN may want to readd seq locking, but then the variable should be
93 static __always_inline void do_get_tz(struct timezone * tz)
95 *tz = __vsyscall_gtod_data.sys_tz;
98 static __always_inline int gettimeofday(struct timeval *tv, struct timezone *tz)
101 asm volatile("syscall"
103 : "0" (__NR_gettimeofday),"D" (tv),"S" (tz)
104 : __syscall_clobber );
108 static __always_inline long time_syscall(long *t)
111 asm volatile("syscall"
113 : "0" (__NR_time),"D" (t) : __syscall_clobber);
117 static __always_inline void do_vgettimeofday(struct timeval * tv)
119 cycle_t now, base, mask, cycle_delta;
121 unsigned long mult, shift, nsec;
122 cycle_t (*vread)(void);
124 seq = read_seqbegin(&__vsyscall_gtod_data.lock);
126 vread = __vsyscall_gtod_data.clock.vread;
127 if (unlikely(!__vsyscall_gtod_data.sysctl_enabled || !vread)) {
128 gettimeofday(tv,NULL);
132 base = __vsyscall_gtod_data.clock.cycle_last;
133 mask = __vsyscall_gtod_data.clock.mask;
134 mult = __vsyscall_gtod_data.clock.mult;
135 shift = __vsyscall_gtod_data.clock.shift;
137 tv->tv_sec = __vsyscall_gtod_data.wall_time_sec;
138 nsec = __vsyscall_gtod_data.wall_time_nsec;
139 } while (read_seqretry(&__vsyscall_gtod_data.lock, seq));
141 /* calculate interval: */
142 cycle_delta = (now - base) & mask;
143 /* convert to nsecs: */
144 nsec += (cycle_delta * mult) >> shift;
146 while (nsec >= NSEC_PER_SEC) {
148 nsec -= NSEC_PER_SEC;
150 tv->tv_usec = nsec / NSEC_PER_USEC;
153 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
156 do_vgettimeofday(tv);
162 /* This will break when the xtime seconds get inaccurate, but that is
164 time_t __vsyscall(1) vtime(time_t *t)
168 if (unlikely(!__vsyscall_gtod_data.sysctl_enabled))
169 return time_syscall(t);
171 vgettimeofday(&tv, NULL);
178 /* Fast way to get current CPU and node.
179 This helps to do per node and per CPU caches in user space.
180 The result is not guaranteed without CPU affinity, but usually
181 works out because the scheduler tries to keep a thread on the same
184 tcache must point to a two element sized long array.
185 All arguments can be NULL. */
187 vgetcpu(unsigned *cpu, unsigned *node, struct getcpu_cache *tcache)
192 /* Fast cache - only recompute value once per jiffies and avoid
193 relatively costly rdtscp/cpuid otherwise.
194 This works because the scheduler usually keeps the process
195 on the same CPU and this syscall doesn't guarantee its
197 We do this here because otherwise user space would do it on
198 its own in a likely inferior way (no access to jiffies).
199 If you don't like it pass NULL. */
200 if (tcache && tcache->blob[0] == (j = __jiffies)) {
202 } else if (__vgetcpu_mode == VGETCPU_RDTSCP) {
203 /* Load per CPU data from RDTSCP */
204 native_read_tscp(&p);
206 /* Load per CPU data from GDT */
207 asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
220 static long __vsyscall(3) venosys_1(void)
228 vsyscall_sysctl_change(ctl_table *ctl, int write, struct file * filp,
229 void __user *buffer, size_t *lenp, loff_t *ppos)
231 return proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
234 static ctl_table kernel_table2[] = {
235 { .procname = "vsyscall64",
236 .data = &vsyscall_gtod_data.sysctl_enabled, .maxlen = sizeof(int),
238 .proc_handler = vsyscall_sysctl_change },
242 static ctl_table kernel_root_table2[] = {
243 { .ctl_name = CTL_KERN, .procname = "kernel", .mode = 0555,
244 .child = kernel_table2 },
249 /* Assume __initcall executes before all user space. Hopefully kmod
250 doesn't violate that. We'll find out if it does. */
251 static void __cpuinit vsyscall_set_cpu(int cpu)
254 unsigned long node = 0;
256 node = cpu_to_node(cpu);
258 if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
259 write_rdtscp_aux((node << 12) | cpu);
261 /* Store cpu number in limit so that it can be loaded quickly
262 in user space in vgetcpu.
263 12 bits for the CPU and 8 bits for the node. */
264 d = 0x0f40000000000ULL;
266 d |= (node & 0xf) << 12;
267 d |= (node >> 4) << 48;
268 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
271 static void __cpuinit cpu_vsyscall_init(void *arg)
273 /* preemption should be already off */
274 vsyscall_set_cpu(raw_smp_processor_id());
278 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
280 long cpu = (long)arg;
281 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
282 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
286 void __init map_vsyscall(void)
288 extern char __vsyscall_0;
289 unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0);
291 /* Note that VSYSCALL_MAPPED_PAGES must agree with the code below. */
292 __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL);
295 static int __init vsyscall_init(void)
297 BUG_ON(((unsigned long) &vgettimeofday !=
298 VSYSCALL_ADDR(__NR_vgettimeofday)));
299 BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime));
300 BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE)));
301 BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu));
303 register_sysctl_table(kernel_root_table2);
305 on_each_cpu(cpu_vsyscall_init, NULL, 1);
306 hotcpu_notifier(cpu_vsyscall_notifier, 0);
310 __initcall(vsyscall_init);