2 * linux/arch/x86_64/kernel/vsyscall.c
4 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Copyright 2003 Andi Kleen, SuSE Labs.
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.
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.
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.
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 #include <linux/getcpu.h>
31 #include <asm/vsyscall.h>
32 #include <asm/pgtable.h>
34 #include <asm/fixmap.h>
35 #include <asm/errno.h>
37 #include <asm/segment.h>
39 #include <asm/topology.h>
41 #define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr)))
43 int __sysctl_vsyscall __section_sysctl_vsyscall = 1;
44 seqlock_t __xtime_lock __section_xtime_lock = SEQLOCK_UNLOCKED;
45 int __vgetcpu_mode __section_vgetcpu_mode;
47 #include <asm/unistd.h>
49 static __always_inline void timeval_normalize(struct timeval * tv)
53 __sec = tv->tv_usec / 1000000;
55 tv->tv_usec %= 1000000;
60 static __always_inline void do_vgettimeofday(struct timeval * tv)
63 unsigned long sec, usec;
66 sequence = read_seqbegin(&__xtime_lock);
69 usec = __xtime.tv_nsec / 1000;
71 if (__vxtime.mode != VXTIME_HPET) {
72 t = get_cycles_sync();
73 if (t < __vxtime.last_tsc)
74 t = __vxtime.last_tsc;
75 usec += ((t - __vxtime.last_tsc) *
76 __vxtime.tsc_quot) >> 32;
77 /* See comment in x86_64 do_gettimeofday. */
79 usec += ((readl((void __iomem *)
80 fix_to_virt(VSYSCALL_HPET) + 0xf0) -
81 __vxtime.last) * __vxtime.quot) >> 32;
83 } while (read_seqretry(&__xtime_lock, sequence));
85 tv->tv_sec = sec + usec / 1000000;
86 tv->tv_usec = usec % 1000000;
89 /* RED-PEN may want to readd seq locking, but then the variable should be write-once. */
90 static __always_inline void do_get_tz(struct timezone * tz)
95 static __always_inline int gettimeofday(struct timeval *tv, struct timezone *tz)
98 asm volatile("vsysc2: syscall"
100 : "0" (__NR_gettimeofday),"D" (tv),"S" (tz) : __syscall_clobber );
104 static __always_inline long time_syscall(long *t)
107 asm volatile("vsysc1: syscall"
109 : "0" (__NR_time),"D" (t) : __syscall_clobber);
113 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
115 if (!__sysctl_vsyscall)
116 return gettimeofday(tv,tz);
118 do_vgettimeofday(tv);
124 /* This will break when the xtime seconds get inaccurate, but that is
126 time_t __vsyscall(1) vtime(time_t *t)
128 if (!__sysctl_vsyscall)
129 return time_syscall(t);
132 return __xtime.tv_sec;
135 /* Fast way to get current CPU and node.
136 This helps to do per node and per CPU caches in user space.
137 The result is not guaranteed without CPU affinity, but usually
138 works out because the scheduler tries to keep a thread on the same
141 tcache must point to a two element sized long array.
142 All arguments can be NULL. */
144 vgetcpu(unsigned *cpu, unsigned *node, struct getcpu_cache *tcache)
146 unsigned int dummy, p;
149 /* Fast cache - only recompute value once per jiffies and avoid
150 relatively costly rdtscp/cpuid otherwise.
151 This works because the scheduler usually keeps the process
152 on the same CPU and this syscall doesn't guarantee its
154 We do this here because otherwise user space would do it on
155 its own in a likely inferior way (no access to jiffies).
156 If you don't like it pass NULL. */
157 if (tcache && tcache->blob[0] == (j = __jiffies)) {
159 } else if (__vgetcpu_mode == VGETCPU_RDTSCP) {
160 /* Load per CPU data from RDTSCP */
161 rdtscp(dummy, dummy, p);
163 /* Load per CPU data from GDT */
164 asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
177 long __vsyscall(3) venosys_1(void)
184 #define SYSCALL 0x050f
188 * NOP out syscall in vsyscall page when not needed.
190 static int vsyscall_sysctl_change(ctl_table *ctl, int write, struct file * filp,
191 void __user *buffer, size_t *lenp, loff_t *ppos)
193 extern u16 vsysc1, vsysc2;
196 int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
199 /* gcc has some trouble with __va(__pa()), so just do it this
201 map1 = ioremap(__pa_symbol(&vsysc1), 2);
204 map2 = ioremap(__pa_symbol(&vsysc2), 2);
209 if (!sysctl_vsyscall) {
210 writew(SYSCALL, map1);
211 writew(SYSCALL, map2);
222 static int vsyscall_sysctl_nostrat(ctl_table *t, int __user *name, int nlen,
223 void __user *oldval, size_t __user *oldlenp,
224 void __user *newval, size_t newlen,
230 static ctl_table kernel_table2[] = {
231 { .ctl_name = 99, .procname = "vsyscall64",
232 .data = &sysctl_vsyscall, .maxlen = sizeof(int), .mode = 0644,
233 .strategy = vsyscall_sysctl_nostrat,
234 .proc_handler = vsyscall_sysctl_change },
238 static ctl_table kernel_root_table2[] = {
239 { .ctl_name = CTL_KERN, .procname = "kernel", .mode = 0555,
240 .child = kernel_table2 },
246 static void __cpuinit write_rdtscp_cb(void *info)
248 write_rdtscp_aux((unsigned long)info);
251 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 void *info = (void *)((node << 12) | cpu);
260 /* Can happen on preemptive kernel */
261 if (get_cpu() == cpu)
262 write_rdtscp_cb(info);
265 /* the notifier is unfortunately not executed on the
267 smp_call_function_single(cpu,write_rdtscp_cb,info,0,1);
273 /* Store cpu number in limit so that it can be loaded quickly
274 in user space in vgetcpu.
275 12 bits for the CPU and 8 bits for the node. */
276 d = (unsigned long *)(cpu_gdt(cpu) + GDT_ENTRY_PER_CPU);
277 *d = 0x0f40000000000ULL;
279 *d |= (node & 0xf) << 12;
280 *d |= (node >> 4) << 48;
283 static void __init map_vsyscall(void)
285 extern char __vsyscall_0;
286 unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0);
288 __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL);
291 static int __init vsyscall_init(void)
293 BUG_ON(((unsigned long) &vgettimeofday !=
294 VSYSCALL_ADDR(__NR_vgettimeofday)));
295 BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime));
296 BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE)));
297 BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu));
300 register_sysctl_table(kernel_root_table2, 0);
305 __initcall(vsyscall_init);