arch/x86/kernel/vsyscall_64.c

   1 /*
   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 /* Disable profiling for userspace code: */
  21 #define DISABLE_BRANCH_PROFILING
  22
  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>
  35
  36 #include <asm/vsyscall.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/page.h>
  39 #include <asm/unistd.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/errno.h>
  42 #include <asm/io.h>
  43 #include <asm/segment.h>
  44 #include <asm/desc.h>
  45 #include <asm/topology.h>
  46 #include <asm/vgtod.h>
  47
  48 #define __vsyscall(nr) \
  49                 __attribute__ ((unused, __section__(".vsyscall_" #nr))) notrace
  50 #define __syscall_clobber "r11","cx","memory"
  51
  52 DEFINE_VVAR(int, vgetcpu_mode);
  53 DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data) =
  54 {
  55         .lock = __SEQLOCK_UNLOCKED(__vsyscall_gtod_data.lock),
  56         .sysctl_enabled = 1,
  57 };
  58
  59 void update_vsyscall_tz(void)
  60 {
  61         unsigned long flags;
  62
  63         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  64         /* sys_tz has changed */
  65         vsyscall_gtod_data.sys_tz = sys_tz;
  66         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
  67 }
  68
  69 void update_vsyscall(struct timespec *wall_time, struct timespec *wtm,
  70                         struct clocksource *clock, u32 mult)
  71 {
  72         unsigned long flags;
  73
  74         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  75         /* copy vsyscall data */
  76         vsyscall_gtod_data.clock.vread = clock->vread;
  77         vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
  78         vsyscall_gtod_data.clock.mask = clock->mask;
  79         vsyscall_gtod_data.clock.mult = mult;
  80         vsyscall_gtod_data.clock.shift = clock->shift;
  81         vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
  82         vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
  83         vsyscall_gtod_data.wall_to_monotonic = *wtm;
  84         vsyscall_gtod_data.wall_time_coarse = __current_kernel_time();
  85         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
  86 }
  87
  88 /* RED-PEN may want to readd seq locking, but then the variable should be
  89  * write-once.
  90  */
  91 static __always_inline void do_get_tz(struct timezone * tz)
  92 {
  93         *tz = VVAR(vsyscall_gtod_data).sys_tz;
  94 }
  95
  96 static __always_inline int gettimeofday(struct timeval *tv, struct timezone *tz)
  97 {
  98         int ret;
  99         asm volatile("syscall"
 100                 : "=a" (ret)
 101                 : "0" (__NR_gettimeofday),"D" (tv),"S" (tz)
 102                 : __syscall_clobber );
 103         return ret;
 104 }
 105
 106 static __always_inline long time_syscall(long *t)
 107 {
 108         long secs;
 109         asm volatile("syscall"
 110                 : "=a" (secs)
 111                 : "0" (__NR_time),"D" (t) : __syscall_clobber);
 112         return secs;
 113 }
 114
 115 static __always_inline void do_vgettimeofday(struct timeval * tv)
 116 {
 117         cycle_t now, base, mask, cycle_delta;
 118         unsigned seq;
 119         unsigned long mult, shift, nsec;
 120         cycle_t (*vread)(void);
 121         do {
 122                 seq = read_seqbegin(&VVAR(vsyscall_gtod_data).lock);
 123
 124                 vread = VVAR(vsyscall_gtod_data).clock.vread;
 125                 if (unlikely(!VVAR(vsyscall_gtod_data).sysctl_enabled ||
 126                              !vread)) {
 127                         gettimeofday(tv,NULL);
 128                         return;
 129                 }
 130
 131                 now = vread();
 132                 base = VVAR(vsyscall_gtod_data).clock.cycle_last;
 133                 mask = VVAR(vsyscall_gtod_data).clock.mask;
 134                 mult = VVAR(vsyscall_gtod_data).clock.mult;
 135                 shift = VVAR(vsyscall_gtod_data).clock.shift;
 136
 137                 tv->tv_sec = VVAR(vsyscall_gtod_data).wall_time_sec;
 138                 nsec = VVAR(vsyscall_gtod_data).wall_time_nsec;
 139         } while (read_seqretry(&VVAR(vsyscall_gtod_data).lock, seq));
 140
 141         /* calculate interval: */
 142         cycle_delta = (now - base) & mask;
 143         /* convert to nsecs: */
 144         nsec += (cycle_delta * mult) >> shift;
 145
 146         while (nsec >= NSEC_PER_SEC) {
 147                 tv->tv_sec += 1;
 148                 nsec -= NSEC_PER_SEC;
 149         }
 150         tv->tv_usec = nsec / NSEC_PER_USEC;
 151 }
 152
 153 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
 154 {
 155         if (tv)
 156                 do_vgettimeofday(tv);
 157         if (tz)
 158                 do_get_tz(tz);
 159         return 0;
 160 }
 161
 162 /* This will break when the xtime seconds get inaccurate, but that is
 163  * unlikely */
 164 time_t __vsyscall(1) vtime(time_t *t)
 165 {
 166         unsigned seq;
 167         time_t result;
 168         if (unlikely(!VVAR(vsyscall_gtod_data).sysctl_enabled))
 169                 return time_syscall(t);
 170
 171         do {
 172                 seq = read_seqbegin(&VVAR(vsyscall_gtod_data).lock);
 173
 174                 result = VVAR(vsyscall_gtod_data).wall_time_sec;
 175
 176         } while (read_seqretry(&VVAR(vsyscall_gtod_data).lock, seq));
 177
 178         if (t)
 179                 *t = result;
 180         return result;
 181 }
 182
 183 /* Fast way to get current CPU and node.
 184    This helps to do per node and per CPU caches in user space.
 185    The result is not guaranteed without CPU affinity, but usually
 186    works out because the scheduler tries to keep a thread on the same
 187    CPU.
 188
 189    tcache must point to a two element sized long array.
 190    All arguments can be NULL. */
 191 long __vsyscall(2)
 192 vgetcpu(unsigned *cpu, unsigned *node, struct getcpu_cache *tcache)
 193 {
 194         unsigned int p;
 195         unsigned long j = 0;
 196
 197         /* Fast cache - only recompute value once per jiffies and avoid
 198            relatively costly rdtscp/cpuid otherwise.
 199            This works because the scheduler usually keeps the process
 200            on the same CPU and this syscall doesn't guarantee its
 201            results anyways.
 202            We do this here because otherwise user space would do it on
 203            its own in a likely inferior way (no access to jiffies).
 204            If you don't like it pass NULL. */
 205         if (tcache && tcache->blob[0] == (j = VVAR(jiffies))) {
 206                 p = tcache->blob[1];
 207         } else if (VVAR(vgetcpu_mode) == VGETCPU_RDTSCP) {
 208                 /* Load per CPU data from RDTSCP */
 209                 native_read_tscp(&p);
 210         } else {
 211                 /* Load per CPU data from GDT */
 212                 asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
 213         }
 214         if (tcache) {
 215                 tcache->blob[0] = j;
 216                 tcache->blob[1] = p;
 217         }
 218         if (cpu)
 219                 *cpu = p & 0xfff;
 220         if (node)
 221                 *node = p >> 12;
 222         return 0;
 223 }
 224
 225 static long __vsyscall(3) venosys_1(void)
 226 {
 227         return -ENOSYS;
 228 }
 229
 230 #ifdef CONFIG_SYSCTL
 231 static ctl_table kernel_table2[] = {
 232         { .procname = "vsyscall64",
 233           .data = &vsyscall_gtod_data.sysctl_enabled, .maxlen = sizeof(int),
 234           .mode = 0644,
 235           .proc_handler = proc_dointvec },
 236         {}
 237 };
 238
 239 static ctl_table kernel_root_table2[] = {
 240         { .procname = "kernel", .mode = 0555,
 241           .child = kernel_table2 },
 242         {}
 243 };
 244 #endif
 245
 246 /* Assume __initcall executes before all user space. Hopefully kmod
 247    doesn't violate that. We'll find out if it does. */
 248 static void __cpuinit vsyscall_set_cpu(int cpu)
 249 {
 250         unsigned long d;
 251         unsigned long node = 0;
 252 #ifdef CONFIG_NUMA
 253         node = cpu_to_node(cpu);
 254 #endif
 255         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 256                 write_rdtscp_aux((node << 12) | cpu);
 257
 258         /* Store cpu number in limit so that it can be loaded quickly
 259            in user space in vgetcpu.
 260            12 bits for the CPU and 8 bits for the node. */
 261         d = 0x0f40000000000ULL;
 262         d |= cpu;
 263         d |= (node & 0xf) << 12;
 264         d |= (node >> 4) << 48;
 265         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 266 }
 267
 268 static void __cpuinit cpu_vsyscall_init(void *arg)
 269 {
 270         /* preemption should be already off */
 271         vsyscall_set_cpu(raw_smp_processor_id());
 272 }
 273
 274 static int __cpuinit
 275 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 276 {
 277         long cpu = (long)arg;
 278         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 279                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
 280         return NOTIFY_DONE;
 281 }
 282
 283 void __init map_vsyscall(void)
 284 {
 285         extern char __vsyscall_0;
 286         unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0);
 287
 288         /* Note that VSYSCALL_MAPPED_PAGES must agree with the code below. */
 289         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL);
 290 }
 291
 292 static int __init vsyscall_init(void)
 293 {
 294         BUG_ON(((unsigned long) &vgettimeofday !=
 295                         VSYSCALL_ADDR(__NR_vgettimeofday)));
 296         BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime));
 297         BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE)));
 298         BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu));
 299 #ifdef CONFIG_SYSCTL
 300         register_sysctl_table(kernel_root_table2);
 301 #endif
 302         on_each_cpu(cpu_vsyscall_init, NULL, 1);
 303         /* notifier priority > KVM */
 304         hotcpu_notifier(cpu_vsyscall_notifier, 30);
 305         return 0;
 306 }
 307
 308 __initcall(vsyscall_init);