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 /* Disable profiling for userspace code: */
21 #define DISABLE_BRANCH_PROFILING
23 #include <linux/time.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/timer.h>
27 #include <linux/seqlock.h>
28 #include <linux/jiffies.h>
29 #include <linux/sysctl.h>
30 #include <linux/clocksource.h>
31 #include <linux/getcpu.h>
32 #include <linux/cpu.h>
33 #include <linux/smp.h>
34 #include <linux/notifier.h>
36 #include <asm/vsyscall.h>
37 #include <asm/pgtable.h>
39 #include <asm/unistd.h>
40 #include <asm/fixmap.h>
41 #include <asm/errno.h>
43 #include <asm/segment.h>
45 #include <asm/topology.h>
46 #include <asm/vgtod.h>
48 #define __vsyscall(nr) \
49 __attribute__ ((unused, __section__(".vsyscall_" #nr))) notrace
50 #define __syscall_clobber "r11","cx","memory"
52 DEFINE_VVAR(int, vgetcpu_mode);
53 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data) =
55 .lock = __SEQLOCK_UNLOCKED(__vsyscall_gtod_data.lock),
58 void update_vsyscall_tz(void)
62 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
63 /* sys_tz has changed */
64 vsyscall_gtod_data.sys_tz = sys_tz;
65 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
68 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
69 struct clocksource *clock, u32 mult)
73 write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
74 /* copy vsyscall data */
75 vsyscall_gtod_data.clock.vread = clock->vread;
76 vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
77 vsyscall_gtod_data.clock.mask = clock->mask;
78 vsyscall_gtod_data.clock.mult = mult;
79 vsyscall_gtod_data.clock.shift = clock->shift;
80 vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
81 vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
82 vsyscall_gtod_data.wall_to_monotonic = *wtm;
83 vsyscall_gtod_data.wall_time_coarse = __current_kernel_time();
84 write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
87 /* RED-PEN may want to readd seq locking, but then the variable should be
90 static __always_inline void do_get_tz(struct timezone * tz)
92 *tz = VVAR(vsyscall_gtod_data).sys_tz;
95 static __always_inline int gettimeofday(struct timeval *tv, struct timezone *tz)
98 asm volatile("syscall"
100 : "0" (__NR_gettimeofday),"D" (tv),"S" (tz)
101 : __syscall_clobber );
105 static __always_inline void do_vgettimeofday(struct timeval * tv)
107 cycle_t now, base, mask, cycle_delta;
109 unsigned long mult, shift, nsec;
110 cycle_t (*vread)(void);
112 seq = read_seqbegin(&VVAR(vsyscall_gtod_data).lock);
114 vread = VVAR(vsyscall_gtod_data).clock.vread;
115 if (unlikely(!vread)) {
116 gettimeofday(tv,NULL);
121 base = VVAR(vsyscall_gtod_data).clock.cycle_last;
122 mask = VVAR(vsyscall_gtod_data).clock.mask;
123 mult = VVAR(vsyscall_gtod_data).clock.mult;
124 shift = VVAR(vsyscall_gtod_data).clock.shift;
126 tv->tv_sec = VVAR(vsyscall_gtod_data).wall_time_sec;
127 nsec = VVAR(vsyscall_gtod_data).wall_time_nsec;
128 } while (read_seqretry(&VVAR(vsyscall_gtod_data).lock, seq));
130 /* calculate interval: */
131 cycle_delta = (now - base) & mask;
132 /* convert to nsecs: */
133 nsec += (cycle_delta * mult) >> shift;
135 while (nsec >= NSEC_PER_SEC) {
137 nsec -= NSEC_PER_SEC;
139 tv->tv_usec = nsec / NSEC_PER_USEC;
142 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
145 do_vgettimeofday(tv);
151 /* This will break when the xtime seconds get inaccurate, but that is
153 time_t __vsyscall(1) vtime(time_t *t)
159 seq = read_seqbegin(&VVAR(vsyscall_gtod_data).lock);
161 result = VVAR(vsyscall_gtod_data).wall_time_sec;
163 } while (read_seqretry(&VVAR(vsyscall_gtod_data).lock, seq));
170 /* Fast way to get current CPU and node.
171 This helps to do per node and per CPU caches in user space.
172 The result is not guaranteed without CPU affinity, but usually
173 works out because the scheduler tries to keep a thread on the same
176 tcache must point to a two element sized long array.
177 All arguments can be NULL. */
179 vgetcpu(unsigned *cpu, unsigned *node, struct getcpu_cache *tcache)
184 /* Fast cache - only recompute value once per jiffies and avoid
185 relatively costly rdtscp/cpuid otherwise.
186 This works because the scheduler usually keeps the process
187 on the same CPU and this syscall doesn't guarantee its
189 We do this here because otherwise user space would do it on
190 its own in a likely inferior way (no access to jiffies).
191 If you don't like it pass NULL. */
192 if (tcache && tcache->blob[0] == (j = VVAR(jiffies))) {
194 } else if (VVAR(vgetcpu_mode) == VGETCPU_RDTSCP) {
195 /* Load per CPU data from RDTSCP */
196 native_read_tscp(&p);
198 /* Load per CPU data from GDT */
199 asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
212 /* Assume __initcall executes before all user space. Hopefully kmod
213 doesn't violate that. We'll find out if it does. */
214 static void __cpuinit vsyscall_set_cpu(int cpu)
217 unsigned long node = 0;
219 node = cpu_to_node(cpu);
221 if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
222 write_rdtscp_aux((node << 12) | cpu);
224 /* Store cpu number in limit so that it can be loaded quickly
225 in user space in vgetcpu.
226 12 bits for the CPU and 8 bits for the node. */
227 d = 0x0f40000000000ULL;
229 d |= (node & 0xf) << 12;
230 d |= (node >> 4) << 48;
231 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
234 static void __cpuinit cpu_vsyscall_init(void *arg)
236 /* preemption should be already off */
237 vsyscall_set_cpu(raw_smp_processor_id());
241 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
243 long cpu = (long)arg;
244 if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
245 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
249 void __init map_vsyscall(void)
251 extern char __vsyscall_0;
252 unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0);
253 extern char __vvar_page;
254 unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
256 /* Note that VSYSCALL_MAPPED_PAGES must agree with the code below. */
257 __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL);
258 __set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
259 BUILD_BUG_ON((unsigned long)__fix_to_virt(VVAR_PAGE) !=
260 (unsigned long)VVAR_ADDRESS);
263 static int __init vsyscall_init(void)
265 BUG_ON(((unsigned long) &vgettimeofday !=
266 VSYSCALL_ADDR(__NR_vgettimeofday)));
267 BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime));
268 BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE)));
269 BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu));
270 on_each_cpu(cpu_vsyscall_init, NULL, 1);
271 /* notifier priority > KVM */
272 hotcpu_notifier(cpu_vsyscall_notifier, 30);
276 __initcall(vsyscall_init);