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 #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>
  32
  33 #include <asm/vsyscall.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/page.h>
  36 #include <asm/unistd.h>
  37 #include <asm/fixmap.h>
  38 #include <asm/errno.h>
  39 #include <asm/io.h>
  40 #include <asm/segment.h>
  41 #include <asm/desc.h>
  42 #include <asm/topology.h>
  43 #include <asm/vgtod.h>
  44
  45 #define __vsyscall(nr) __attribute__ ((unused,__section__(".vsyscall_" #nr)))
  46 #define __syscall_clobber "r11","cx","memory"
  47 #define __pa_vsymbol(x)                 \
  48         ({unsigned long v;              \
  49         extern char __vsyscall_0;       \
  50           asm("" : "=r" (v) : "0" (x)); \
  51           ((v - VSYSCALL_START) + __pa_symbol(&__vsyscall_0)); })
  52
  53 /*
  54  * vsyscall_gtod_data contains data that is :
  55  * - readonly from vsyscalls
  56  * - written by timer interrupt or systcl (/proc/sys/kernel/vsyscall64)
  57  * Try to keep this structure as small as possible to avoid cache line ping pongs
  58  */
  59 int __vgetcpu_mode __section_vgetcpu_mode;
  60
  61 struct vsyscall_gtod_data __vsyscall_gtod_data __section_vsyscall_gtod_data =
  62 {
  63         .lock = SEQLOCK_UNLOCKED,
  64         .sysctl_enabled = 1,
  65 };
  66
  67 void update_vsyscall_tz(void)
  68 {
  69         unsigned long flags;
  70
  71         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  72         /* sys_tz has changed */
  73         vsyscall_gtod_data.sys_tz = sys_tz;
  74         write_sequnlock_irqrestore(&vsyscall_gtod_data.lock, flags);
  75 }
  76
  77 void update_vsyscall(struct timespec *wall_time, struct clocksource *clock)
  78 {
  79         unsigned long flags;
  80
  81         write_seqlock_irqsave(&vsyscall_gtod_data.lock, flags);
  82         /* copy vsyscall data */
  83         vsyscall_gtod_data.clock.vread = clock->vread;
  84         vsyscall_gtod_data.clock.cycle_last = clock->cycle_last;
  85         vsyscall_gtod_data.clock.mask = clock->mask;
  86         vsyscall_gtod_data.clock.mult = clock->mult;
  87         vsyscall_gtod_data.clock.shift = clock->shift;
  88         vsyscall_gtod_data.wall_time_sec = wall_time->tv_sec;
  89         vsyscall_gtod_data.wall_time_nsec = wall_time->tv_nsec;
  90         vsyscall_gtod_data.wall_to_monotonic = wall_to_monotonic;
  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("vsysc2: 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("vsysc1: 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                 now = vread();
 136                 base = __vsyscall_gtod_data.clock.cycle_last;
 137                 mask = __vsyscall_gtod_data.clock.mask;
 138                 mult = __vsyscall_gtod_data.clock.mult;
 139                 shift = __vsyscall_gtod_data.clock.shift;
 140
 141                 tv->tv_sec = __vsyscall_gtod_data.wall_time_sec;
 142                 nsec = __vsyscall_gtod_data.wall_time_nsec;
 143         } while (read_seqretry(&__vsyscall_gtod_data.lock, seq));
 144
 145         /* calculate interval: */
 146         cycle_delta = (now - base) & mask;
 147         /* convert to nsecs: */
 148         nsec += (cycle_delta * mult) >> shift;
 149
 150         while (nsec >= NSEC_PER_SEC) {
 151                 tv->tv_sec += 1;
 152                 nsec -= NSEC_PER_SEC;
 153         }
 154         tv->tv_usec = nsec / NSEC_PER_USEC;
 155 }
 156
 157 int __vsyscall(0) vgettimeofday(struct timeval * tv, struct timezone * tz)
 158 {
 159         if (tv)
 160                 do_vgettimeofday(tv);
 161         if (tz)
 162                 do_get_tz(tz);
 163         return 0;
 164 }
 165
 166 /* This will break when the xtime seconds get inaccurate, but that is
 167  * unlikely */
 168 time_t __vsyscall(1) vtime(time_t *t)
 169 {
 170         struct timeval tv;
 171         time_t result;
 172         if (unlikely(!__vsyscall_gtod_data.sysctl_enabled))
 173                 return time_syscall(t);
 174
 175         vgettimeofday(&tv, NULL);
 176         result = tv.tv_sec;
 177         if (t)
 178                 *t = result;
 179         return result;
 180 }
 181
 182 /* Fast way to get current CPU and node.
 183    This helps to do per node and per CPU caches in user space.
 184    The result is not guaranteed without CPU affinity, but usually
 185    works out because the scheduler tries to keep a thread on the same
 186    CPU.
 187
 188    tcache must point to a two element sized long array.
 189    All arguments can be NULL. */
 190 long __vsyscall(2)
 191 vgetcpu(unsigned *cpu, unsigned *node, struct getcpu_cache *tcache)
 192 {
 193         unsigned int p;
 194         unsigned long j = 0;
 195
 196         /* Fast cache - only recompute value once per jiffies and avoid
 197            relatively costly rdtscp/cpuid otherwise.
 198            This works because the scheduler usually keeps the process
 199            on the same CPU and this syscall doesn't guarantee its
 200            results anyways.
 201            We do this here because otherwise user space would do it on
 202            its own in a likely inferior way (no access to jiffies).
 203            If you don't like it pass NULL. */
 204         if (tcache && tcache->blob[0] == (j = __jiffies)) {
 205                 p = tcache->blob[1];
 206         } else if (__vgetcpu_mode == VGETCPU_RDTSCP) {
 207                 /* Load per CPU data from RDTSCP */
 208                 native_read_tscp(&p);
 209         } else {
 210                 /* Load per CPU data from GDT */
 211                 asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
 212         }
 213         if (tcache) {
 214                 tcache->blob[0] = j;
 215                 tcache->blob[1] = p;
 216         }
 217         if (cpu)
 218                 *cpu = p & 0xfff;
 219         if (node)
 220                 *node = p >> 12;
 221         return 0;
 222 }
 223
 224 long __vsyscall(3) venosys_1(void)
 225 {
 226         return -ENOSYS;
 227 }
 228
 229 #ifdef CONFIG_SYSCTL
 230
 231 #define SYSCALL 0x050f
 232 #define NOP2    0x9090
 233
 234 /*
 235  * NOP out syscall in vsyscall page when not needed.
 236  */
 237 static int vsyscall_sysctl_change(ctl_table *ctl, int write, struct file * filp,
 238                         void __user *buffer, size_t *lenp, loff_t *ppos)
 239 {
 240         extern u16 vsysc1, vsysc2;
 241         u16 __iomem *map1;
 242         u16 __iomem *map2;
 243         int ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
 244         if (!write)
 245                 return ret;
 246         /* gcc has some trouble with __va(__pa()), so just do it this
 247            way. */
 248         map1 = ioremap(__pa_vsymbol(&vsysc1), 2);
 249         if (!map1)
 250                 return -ENOMEM;
 251         map2 = ioremap(__pa_vsymbol(&vsysc2), 2);
 252         if (!map2) {
 253                 ret = -ENOMEM;
 254                 goto out;
 255         }
 256         if (!vsyscall_gtod_data.sysctl_enabled) {
 257                 writew(SYSCALL, map1);
 258                 writew(SYSCALL, map2);
 259         } else {
 260                 writew(NOP2, map1);
 261                 writew(NOP2, map2);
 262         }
 263         iounmap(map2);
 264 out:
 265         iounmap(map1);
 266         return ret;
 267 }
 268
 269 static ctl_table kernel_table2[] = {
 270         { .procname = "vsyscall64",
 271           .data = &vsyscall_gtod_data.sysctl_enabled, .maxlen = sizeof(int),
 272           .mode = 0644,
 273           .proc_handler = vsyscall_sysctl_change },
 274         {}
 275 };
 276
 277 static ctl_table kernel_root_table2[] = {
 278         { .ctl_name = CTL_KERN, .procname = "kernel", .mode = 0555,
 279           .child = kernel_table2 },
 280         {}
 281 };
 282
 283 #endif
 284
 285 /* Assume __initcall executes before all user space. Hopefully kmod
 286    doesn't violate that. We'll find out if it does. */
 287 static void __cpuinit vsyscall_set_cpu(int cpu)
 288 {
 289         unsigned long *d;
 290         unsigned long node = 0;
 291 #ifdef CONFIG_NUMA
 292         node = cpu_to_node(cpu);
 293 #endif
 294         if (cpu_has(&cpu_data(cpu), X86_FEATURE_RDTSCP))
 295                 write_rdtscp_aux((node << 12) | cpu);
 296
 297         /* Store cpu number in limit so that it can be loaded quickly
 298            in user space in vgetcpu.
 299            12 bits for the CPU and 8 bits for the node. */
 300         d = (unsigned long *)(get_cpu_gdt_table(cpu) + GDT_ENTRY_PER_CPU);
 301         *d = 0x0f40000000000ULL;
 302         *d |= cpu;
 303         *d |= (node & 0xf) << 12;
 304         *d |= (node >> 4) << 48;
 305 }
 306
 307 static void __cpuinit cpu_vsyscall_init(void *arg)
 308 {
 309         /* preemption should be already off */
 310         vsyscall_set_cpu(raw_smp_processor_id());
 311 }
 312
 313 static int __cpuinit
 314 cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg)
 315 {
 316         long cpu = (long)arg;
 317         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 318                 smp_call_function_single(cpu, cpu_vsyscall_init, NULL, 0, 1);
 319         return NOTIFY_DONE;
 320 }
 321
 322 static void __init map_vsyscall(void)
 323 {
 324         extern char __vsyscall_0;
 325         unsigned long physaddr_page0 = __pa_symbol(&__vsyscall_0);
 326
 327         /* Note that VSYSCALL_MAPPED_PAGES must agree with the code below. */
 328         __set_fixmap(VSYSCALL_FIRST_PAGE, physaddr_page0, PAGE_KERNEL_VSYSCALL);
 329 }
 330
 331 static int __init vsyscall_init(void)
 332 {
 333         BUG_ON(((unsigned long) &vgettimeofday !=
 334                         VSYSCALL_ADDR(__NR_vgettimeofday)));
 335         BUG_ON((unsigned long) &vtime != VSYSCALL_ADDR(__NR_vtime));
 336         BUG_ON((VSYSCALL_ADDR(0) != __fix_to_virt(VSYSCALL_FIRST_PAGE)));
 337         BUG_ON((unsigned long) &vgetcpu != VSYSCALL_ADDR(__NR_vgetcpu));
 338         map_vsyscall();
 339 #ifdef CONFIG_SYSCTL
 340         register_sysctl_table(kernel_root_table2);
 341 #endif
 342         on_each_cpu(cpu_vsyscall_init, NULL, 0, 1);
 343         hotcpu_notifier(cpu_vsyscall_notifier, 0);
 344         return 0;
 345 }
 346
 347 __initcall(vsyscall_init);