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 /*
  53  * vsyscall_gtod_data contains data that is :
  54  * - readonly from vsyscalls
  55  * - written by timer interrupt or systcl (/proc/sys/kernel/vsyscall64)
  56  * Try to keep this structure as small as possible to avoid cache line ping pongs
  57  */
  58 int __vgetcpu_mode __section_vgetcpu_mode;
  59
  60 struct vsyscall_gtod_data __vsyscall_gtod_data __section_vsyscall_gtod_data =
  61 {
  62         .lock = SEQLOCK_UNLOCKED,
  63         .sysctl_enabled = 1,
  64 };
  65
  66 void update_vsyscall_tz(void)
  67 {
  68         unsigned long flags;
  69
  70         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  71         /* sys_tz has changed */
  72         vsyscall_gtod_data.sys_tz = sys_tz;
  73         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
  74 }
  75
  76 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock)
  77 {
  78         unsigned long flags;
  79
  80         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  81         /* copy vsyscall data */
  82         vsyscall_gtod_data.clock.vread = clock->vread;
  83         vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
  84         vsyscall_gtod_data.clock.mask = clock->mask;
  85         vsyscall_gtod_data.clock.mult = clock->mult;
  86         vsyscall_gtod_data.clock.shift = clock->shift;
  87         vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
  88         vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
  89         vsyscall_gtod_data.wall_to_monotonic = wall_to_monotonic;
  90         vsyscall_gtod_data.wall_time_coarse = __current_kernel_time();
  91         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
  92 }
  93
  94 /* RED-PEN may want to readd seq locking, but then the variable should be
  95  * write-once.
  96  */
  97 static __always_inline void do_get_tz(struct timezone * tz)
  98 {
  99         *tz = __vsyscall_gtod_data.sys_tz;
 100 }
 101
 102 static __always_inline int gettimeofday(struct timeval *tv, struct timezone *tz)
 103 {
 104         int ret;
 105         asm volatile("syscall"
 106                 : "=a" (ret)
 107                 : "0" (__NR_gettimeofday),"D" (tv),"S" (tz)
 108                 : __syscall_clobber );
 109         return ret;
 110 }
 111
 112 static __always_inline long time_syscall(long *t)
 113 {
 114         long secs;
 115         asm volatile("syscall"
 116                 : "=a" (secs)
 117                 : "0" (__NR_time),"D" (t) : __syscall_clobber);
 118         return secs;
 119 }
 120
 121 static __always_inline void do_vgettimeofday(struct timeval * tv)
 122 {
 123         cycle_t now, base, mask, cycle_delta;
 124         unsigned seq;
 125         unsigned long mult, shift, nsec;
 126         cycle_t (*vread)(void);
 127         do {
 128                 seq = read_seqbegin(&__vsyscall_gtod_data.lock);
 129
 130                 vread = __vsyscall_gtod_data.clock.vread;
 131                 if (unlikely(!__vsyscall_gtod_data.sysctl_enabled || !vread)) {
 132                         gettimeofday(tv,NULL);
 133                         return;
 134                 }
 135
 136                 now = vread();
 137                 base = __vsyscall_gtod_data.clock.cycle_last;
 138                 mask = __vsyscall_gtod_data.clock.mask;
 139                 mult = __vsyscall_gtod_data.clock.mult;
 140                 shift = __vsyscall_gtod_data.clock.shift;
 141
 142                 tv->tv_sec = __vsyscall_gtod_data.wall_time_sec;
 143                 nsec = __vsyscall_gtod_data.wall_time_nsec;
 144         } while (read_seqretry(&__vsyscall_gtod_data.lock, seq));
 145
 146         /* calculate interval: */
 147         cycle_delta = (now - base) & mask;
 148         /* convert to nsecs: */
 149         nsec += (cycle_delta * mult) >> shift;
 150
 151         while (nsec >= NSEC_PER_SEC) {
 152                 tv->tv_sec += 1;
 153                 nsec -= NSEC_PER_SEC;
 154         }
 155         tv->tv_usec = nsec / NSEC_PER_USEC;
 156 }
 157
 158 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
 159 {
 160         if (tv)
 161                 do_vgettimeofday(tv);
 162         if (tz)
 163                 do_get_tz(tz);
 164         return 0;
 165 }
 166
 167 /* This will break when the xtime seconds get inaccurate, but that is
 168  * unlikely */
 169 time_t __vsyscall(1) vtime(time_t *t)
 170 {
 171         struct timeval tv;
 172         time_t result;
 173         if (unlikely(!__vsyscall_gtod_data.sysctl_enabled))
 174                 return time_syscall(t);
 175
 176         vgettimeofday(&tv, NULL);
 177         result = tv.tv_sec;
 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 = __jiffies)) {
 206                 p = tcache->blob[1];
 207         } else if (__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
 232 static int
 233 vsyscall_sysctl_change(ctl_table *ctl, int write, struct file * filp,
 234                        void __user *buffer, size_t *lenp, loff_t *ppos)
 235 {
 236         return proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 237 }
 238
 239 static ctl_table kernel_table2[] = {
 240         { .procname = "vsyscall64",
 241           .data = &vsyscall_gtod_data.sysctl_enabled, .maxlen = sizeof(int),
 242           .mode = 0644,
 243           .proc_handler = vsyscall_sysctl_change },
 244         {}
 245 };
 246
 247 static ctl_table kernel_root_table2[] = {
 248         { .ctl_name = CTL_KERN, .procname = "kernel", .mode = 0555,
 249           .child = kernel_table2 },
 250         {}
 251 };
 252 #endif
 253
 254 /* Assume __initcall executes before all user space. Hopefully kmod
 255    doesn't violate that. We'll find out if it does. */
 256 static void __cpuinit vsyscall_set_cpu(int cpu)
 257 {
 258         unsigned long d;
 259         unsigned long node = 0;
 260 #ifdef CONFIG_NUMA
 261         node = cpu_to_node(cpu);
 262 #endif
 263         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 264                 write_rdtscp_aux((node << 12) | cpu);
 265
 266         /* Store cpu number in limit so that it can be loaded quickly
 267            in user space in vgetcpu.
 268            12 bits for the CPU and 8 bits for the node. */
 269         d = 0x0f40000000000ULL;
 270         d |= cpu;
 271         d |= (node & 0xf) << 12;
 272         d |= (node >> 4) << 48;
 273         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 274 }
 275
 276 static void __cpuinit cpu_vsyscall_init(void *arg)
 277 {
 278         /* preemption should be already off */
 279         vsyscall_set_cpu(raw_smp_processor_id());
 280 }
 281
 282 static int __cpuinit
 283 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 284 {
 285         long cpu = (long)arg;
 286         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 287                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 1);
 288         return NOTIFY_DONE;
 289 }
 290
 291 void __init map_vsyscall(void)
 292 {
 293         extern char __vsyscall_0;
 294         unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0);
 295
 296         /* Note that VSYSCALL_MAPPED_PAGES must agree with the code below. */
 297         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL);
 298 }
 299
 300 static int __init vsyscall_init(void)
 301 {
 302         BUG_ON(((unsigned long) &vgettimeofday !=
 303                         VSYSCALL_ADDR(__NR_vgettimeofday)));
 304         BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime));
 305         BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE)));
 306         BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu));
 307 #ifdef CONFIG_SYSCTL
 308         register_sysctl_table(kernel_root_table2);
 309 #endif
 310         on_each_cpu(cpu_vsyscall_init, NULL, 1);
 311         hotcpu_notifier(cpu_vsyscall_notifier, 0);
 312         return 0;
 313 }
 314
 315 __initcall(vsyscall_init);