...
Run Format

Source file src/runtime/os_linux.go

Documentation: runtime

     1  // Copyright 2009 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package runtime
     6  
     7  import (
     8  	"runtime/internal/sys"
     9  	"unsafe"
    10  )
    11  
    12  type mOS struct{}
    13  
    14  //go:noescape
    15  func futex(addr unsafe.Pointer, op int32, val uint32, ts, addr2 unsafe.Pointer, val3 uint32) int32
    16  
    17  // Linux futex.
    18  //
    19  //	futexsleep(uint32 *addr, uint32 val)
    20  //	futexwakeup(uint32 *addr)
    21  //
    22  // Futexsleep atomically checks if *addr == val and if so, sleeps on addr.
    23  // Futexwakeup wakes up threads sleeping on addr.
    24  // Futexsleep is allowed to wake up spuriously.
    25  
    26  const (
    27  	_FUTEX_PRIVATE_FLAG = 128
    28  	_FUTEX_WAIT_PRIVATE = 0 | _FUTEX_PRIVATE_FLAG
    29  	_FUTEX_WAKE_PRIVATE = 1 | _FUTEX_PRIVATE_FLAG
    30  )
    31  
    32  // Atomically,
    33  //	if(*addr == val) sleep
    34  // Might be woken up spuriously; that's allowed.
    35  // Don't sleep longer than ns; ns < 0 means forever.
    36  //go:nosplit
    37  func futexsleep(addr *uint32, val uint32, ns int64) {
    38  	var ts timespec
    39  
    40  	// Some Linux kernels have a bug where futex of
    41  	// FUTEX_WAIT returns an internal error code
    42  	// as an errno. Libpthread ignores the return value
    43  	// here, and so can we: as it says a few lines up,
    44  	// spurious wakeups are allowed.
    45  	if ns < 0 {
    46  		futex(unsafe.Pointer(addr), _FUTEX_WAIT_PRIVATE, val, nil, nil, 0)
    47  		return
    48  	}
    49  
    50  	// It's difficult to live within the no-split stack limits here.
    51  	// On ARM and 386, a 64-bit divide invokes a general software routine
    52  	// that needs more stack than we can afford. So we use timediv instead.
    53  	// But on real 64-bit systems, where words are larger but the stack limit
    54  	// is not, even timediv is too heavy, and we really need to use just an
    55  	// ordinary machine instruction.
    56  	if sys.PtrSize == 8 {
    57  		ts.set_sec(ns / 1000000000)
    58  		ts.set_nsec(int32(ns % 1000000000))
    59  	} else {
    60  		ts.tv_nsec = 0
    61  		ts.set_sec(int64(timediv(ns, 1000000000, (*int32)(unsafe.Pointer(&ts.tv_nsec)))))
    62  	}
    63  	futex(unsafe.Pointer(addr), _FUTEX_WAIT_PRIVATE, val, unsafe.Pointer(&ts), nil, 0)
    64  }
    65  
    66  // If any procs are sleeping on addr, wake up at most cnt.
    67  //go:nosplit
    68  func futexwakeup(addr *uint32, cnt uint32) {
    69  	ret := futex(unsafe.Pointer(addr), _FUTEX_WAKE_PRIVATE, cnt, nil, nil, 0)
    70  	if ret >= 0 {
    71  		return
    72  	}
    73  
    74  	// I don't know that futex wakeup can return
    75  	// EAGAIN or EINTR, but if it does, it would be
    76  	// safe to loop and call futex again.
    77  	systemstack(func() {
    78  		print("futexwakeup addr=", addr, " returned ", ret, "\n")
    79  	})
    80  
    81  	*(*int32)(unsafe.Pointer(uintptr(0x1006))) = 0x1006
    82  }
    83  
    84  func getproccount() int32 {
    85  	// This buffer is huge (8 kB) but we are on the system stack
    86  	// and there should be plenty of space (64 kB).
    87  	// Also this is a leaf, so we're not holding up the memory for long.
    88  	// See golang.org/issue/11823.
    89  	// The suggested behavior here is to keep trying with ever-larger
    90  	// buffers, but we don't have a dynamic memory allocator at the
    91  	// moment, so that's a bit tricky and seems like overkill.
    92  	const maxCPUs = 64 * 1024
    93  	var buf [maxCPUs / 8]byte
    94  	r := sched_getaffinity(0, unsafe.Sizeof(buf), &buf[0])
    95  	if r < 0 {
    96  		return 1
    97  	}
    98  	n := int32(0)
    99  	for _, v := range buf[:r] {
   100  		for v != 0 {
   101  			n += int32(v & 1)
   102  			v >>= 1
   103  		}
   104  	}
   105  	if n == 0 {
   106  		n = 1
   107  	}
   108  	return n
   109  }
   110  
   111  // Clone, the Linux rfork.
   112  const (
   113  	_CLONE_VM             = 0x100
   114  	_CLONE_FS             = 0x200
   115  	_CLONE_FILES          = 0x400
   116  	_CLONE_SIGHAND        = 0x800
   117  	_CLONE_PTRACE         = 0x2000
   118  	_CLONE_VFORK          = 0x4000
   119  	_CLONE_PARENT         = 0x8000
   120  	_CLONE_THREAD         = 0x10000
   121  	_CLONE_NEWNS          = 0x20000
   122  	_CLONE_SYSVSEM        = 0x40000
   123  	_CLONE_SETTLS         = 0x80000
   124  	_CLONE_PARENT_SETTID  = 0x100000
   125  	_CLONE_CHILD_CLEARTID = 0x200000
   126  	_CLONE_UNTRACED       = 0x800000
   127  	_CLONE_CHILD_SETTID   = 0x1000000
   128  	_CLONE_STOPPED        = 0x2000000
   129  	_CLONE_NEWUTS         = 0x4000000
   130  	_CLONE_NEWIPC         = 0x8000000
   131  
   132  	cloneFlags = _CLONE_VM | /* share memory */
   133  		_CLONE_FS | /* share cwd, etc */
   134  		_CLONE_FILES | /* share fd table */
   135  		_CLONE_SIGHAND | /* share sig handler table */
   136  		_CLONE_SYSVSEM | /* share SysV semaphore undo lists (see issue #20763) */
   137  		_CLONE_THREAD /* revisit - okay for now */
   138  )
   139  
   140  //go:noescape
   141  func clone(flags int32, stk, mp, gp, fn unsafe.Pointer) int32
   142  
   143  // May run with m.p==nil, so write barriers are not allowed.
   144  //go:nowritebarrier
   145  func newosproc(mp *m) {
   146  	stk := unsafe.Pointer(mp.g0.stack.hi)
   147  	/*
   148  	 * note: strace gets confused if we use CLONE_PTRACE here.
   149  	 */
   150  	if false {
   151  		print("newosproc stk=", stk, " m=", mp, " g=", mp.g0, " clone=", funcPC(clone), " id=", mp.id, " ostk=", &mp, "\n")
   152  	}
   153  
   154  	// Disable signals during clone, so that the new thread starts
   155  	// with signals disabled. It will enable them in minit.
   156  	var oset sigset
   157  	sigprocmask(_SIG_SETMASK, &sigset_all, &oset)
   158  	ret := clone(cloneFlags, stk, unsafe.Pointer(mp), unsafe.Pointer(mp.g0), unsafe.Pointer(funcPC(mstart)))
   159  	sigprocmask(_SIG_SETMASK, &oset, nil)
   160  
   161  	if ret < 0 {
   162  		print("runtime: failed to create new OS thread (have ", mcount(), " already; errno=", -ret, ")\n")
   163  		if ret == -_EAGAIN {
   164  			println("runtime: may need to increase max user processes (ulimit -u)")
   165  		}
   166  		throw("newosproc")
   167  	}
   168  }
   169  
   170  // Version of newosproc that doesn't require a valid G.
   171  //go:nosplit
   172  func newosproc0(stacksize uintptr, fn unsafe.Pointer) {
   173  	stack := sysAlloc(stacksize, &memstats.stacks_sys)
   174  	if stack == nil {
   175  		write(2, unsafe.Pointer(&failallocatestack[0]), int32(len(failallocatestack)))
   176  		exit(1)
   177  	}
   178  	ret := clone(cloneFlags, unsafe.Pointer(uintptr(stack)+stacksize), nil, nil, fn)
   179  	if ret < 0 {
   180  		write(2, unsafe.Pointer(&failthreadcreate[0]), int32(len(failthreadcreate)))
   181  		exit(1)
   182  	}
   183  }
   184  
   185  var failallocatestack = []byte("runtime: failed to allocate stack for the new OS thread\n")
   186  var failthreadcreate = []byte("runtime: failed to create new OS thread\n")
   187  
   188  const (
   189  	_AT_NULL   = 0  // End of vector
   190  	_AT_PAGESZ = 6  // System physical page size
   191  	_AT_HWCAP  = 16 // hardware capability bit vector
   192  	_AT_RANDOM = 25 // introduced in 2.6.29
   193  	_AT_HWCAP2 = 26 // hardware capability bit vector 2
   194  )
   195  
   196  var procAuxv = []byte("/proc/self/auxv\x00")
   197  
   198  var addrspace_vec [1]byte
   199  
   200  func mincore(addr unsafe.Pointer, n uintptr, dst *byte) int32
   201  
   202  func sysargs(argc int32, argv **byte) {
   203  	n := argc + 1
   204  
   205  	// skip over argv, envp to get to auxv
   206  	for argv_index(argv, n) != nil {
   207  		n++
   208  	}
   209  
   210  	// skip NULL separator
   211  	n++
   212  
   213  	// now argv+n is auxv
   214  	auxv := (*[1 << 28]uintptr)(add(unsafe.Pointer(argv), uintptr(n)*sys.PtrSize))
   215  	if sysauxv(auxv[:]) != 0 {
   216  		return
   217  	}
   218  	// In some situations we don't get a loader-provided
   219  	// auxv, such as when loaded as a library on Android.
   220  	// Fall back to /proc/self/auxv.
   221  	fd := open(&procAuxv[0], 0 /* O_RDONLY */, 0)
   222  	if fd < 0 {
   223  		// On Android, /proc/self/auxv might be unreadable (issue 9229), so we fallback to
   224  		// try using mincore to detect the physical page size.
   225  		// mincore should return EINVAL when address is not a multiple of system page size.
   226  		const size = 256 << 10 // size of memory region to allocate
   227  		p, err := mmap(nil, size, _PROT_READ|_PROT_WRITE, _MAP_ANON|_MAP_PRIVATE, -1, 0)
   228  		if err != 0 {
   229  			return
   230  		}
   231  		var n uintptr
   232  		for n = 4 << 10; n < size; n <<= 1 {
   233  			err := mincore(unsafe.Pointer(uintptr(p)+n), 1, &addrspace_vec[0])
   234  			if err == 0 {
   235  				physPageSize = n
   236  				break
   237  			}
   238  		}
   239  		if physPageSize == 0 {
   240  			physPageSize = size
   241  		}
   242  		munmap(p, size)
   243  		return
   244  	}
   245  	var buf [128]uintptr
   246  	n = read(fd, noescape(unsafe.Pointer(&buf[0])), int32(unsafe.Sizeof(buf)))
   247  	closefd(fd)
   248  	if n < 0 {
   249  		return
   250  	}
   251  	// Make sure buf is terminated, even if we didn't read
   252  	// the whole file.
   253  	buf[len(buf)-2] = _AT_NULL
   254  	sysauxv(buf[:])
   255  }
   256  
   257  func sysauxv(auxv []uintptr) int {
   258  	var i int
   259  	for ; auxv[i] != _AT_NULL; i += 2 {
   260  		tag, val := auxv[i], auxv[i+1]
   261  		switch tag {
   262  		case _AT_RANDOM:
   263  			// The kernel provides a pointer to 16-bytes
   264  			// worth of random data.
   265  			startupRandomData = (*[16]byte)(unsafe.Pointer(val))[:]
   266  
   267  		case _AT_PAGESZ:
   268  			physPageSize = val
   269  		}
   270  
   271  		archauxv(tag, val)
   272  		vdsoauxv(tag, val)
   273  	}
   274  	return i / 2
   275  }
   276  
   277  func osinit() {
   278  	ncpu = getproccount()
   279  }
   280  
   281  var urandom_dev = []byte("/dev/urandom\x00")
   282  
   283  func getRandomData(r []byte) {
   284  	if startupRandomData != nil {
   285  		n := copy(r, startupRandomData)
   286  		extendRandom(r, n)
   287  		return
   288  	}
   289  	fd := open(&urandom_dev[0], 0 /* O_RDONLY */, 0)
   290  	n := read(fd, unsafe.Pointer(&r[0]), int32(len(r)))
   291  	closefd(fd)
   292  	extendRandom(r, int(n))
   293  }
   294  
   295  func goenvs() {
   296  	goenvs_unix()
   297  }
   298  
   299  // Called to do synchronous initialization of Go code built with
   300  // -buildmode=c-archive or -buildmode=c-shared.
   301  // None of the Go runtime is initialized.
   302  //go:nosplit
   303  //go:nowritebarrierrec
   304  func libpreinit() {
   305  	initsig(true)
   306  }
   307  
   308  // Called to initialize a new m (including the bootstrap m).
   309  // Called on the parent thread (main thread in case of bootstrap), can allocate memory.
   310  func mpreinit(mp *m) {
   311  	mp.gsignal = malg(32 * 1024) // Linux wants >= 2K
   312  	mp.gsignal.m = mp
   313  }
   314  
   315  func gettid() uint32
   316  
   317  // Called to initialize a new m (including the bootstrap m).
   318  // Called on the new thread, cannot allocate memory.
   319  func minit() {
   320  	minitSignals()
   321  
   322  	// for debuggers, in case cgo created the thread
   323  	getg().m.procid = uint64(gettid())
   324  }
   325  
   326  // Called from dropm to undo the effect of an minit.
   327  //go:nosplit
   328  func unminit() {
   329  	unminitSignals()
   330  }
   331  
   332  //#ifdef GOARCH_386
   333  //#define sa_handler k_sa_handler
   334  //#endif
   335  
   336  func sigreturn()
   337  func sigtramp(sig uint32, info *siginfo, ctx unsafe.Pointer)
   338  func cgoSigtramp()
   339  
   340  //go:noescape
   341  func sigaltstack(new, old *stackt)
   342  
   343  //go:noescape
   344  func setitimer(mode int32, new, old *itimerval)
   345  
   346  //go:noescape
   347  func rtsigprocmask(how int32, new, old *sigset, size int32)
   348  
   349  //go:nosplit
   350  //go:nowritebarrierrec
   351  func sigprocmask(how int32, new, old *sigset) {
   352  	rtsigprocmask(how, new, old, int32(unsafe.Sizeof(*new)))
   353  }
   354  
   355  func raise(sig uint32)
   356  func raiseproc(sig uint32)
   357  
   358  //go:noescape
   359  func sched_getaffinity(pid, len uintptr, buf *byte) int32
   360  func osyield()
   361  
   362  //go:nosplit
   363  //go:nowritebarrierrec
   364  func setsig(i uint32, fn uintptr) {
   365  	var sa sigactiont
   366  	sa.sa_flags = _SA_SIGINFO | _SA_ONSTACK | _SA_RESTORER | _SA_RESTART
   367  	sigfillset(&sa.sa_mask)
   368  	// Although Linux manpage says "sa_restorer element is obsolete and
   369  	// should not be used". x86_64 kernel requires it. Only use it on
   370  	// x86.
   371  	if GOARCH == "386" || GOARCH == "amd64" {
   372  		sa.sa_restorer = funcPC(sigreturn)
   373  	}
   374  	if fn == funcPC(sighandler) {
   375  		if iscgo {
   376  			fn = funcPC(cgoSigtramp)
   377  		} else {
   378  			fn = funcPC(sigtramp)
   379  		}
   380  	}
   381  	sa.sa_handler = fn
   382  	sigaction(i, &sa, nil)
   383  }
   384  
   385  //go:nosplit
   386  //go:nowritebarrierrec
   387  func setsigstack(i uint32) {
   388  	var sa sigactiont
   389  	sigaction(i, nil, &sa)
   390  	if sa.sa_flags&_SA_ONSTACK != 0 {
   391  		return
   392  	}
   393  	sa.sa_flags |= _SA_ONSTACK
   394  	sigaction(i, &sa, nil)
   395  }
   396  
   397  //go:nosplit
   398  //go:nowritebarrierrec
   399  func getsig(i uint32) uintptr {
   400  	var sa sigactiont
   401  	sigaction(i, nil, &sa)
   402  	return sa.sa_handler
   403  }
   404  
   405  // setSignaltstackSP sets the ss_sp field of a stackt.
   406  //go:nosplit
   407  func setSignalstackSP(s *stackt, sp uintptr) {
   408  	*(*uintptr)(unsafe.Pointer(&s.ss_sp)) = sp
   409  }
   410  
   411  func (c *sigctxt) fixsigcode(sig uint32) {
   412  }
   413  
   414  // sysSigaction calls the rt_sigaction system call.
   415  //go:nosplit
   416  func sysSigaction(sig uint32, new, old *sigactiont) {
   417  	if rt_sigaction(uintptr(sig), new, old, unsafe.Sizeof(sigactiont{}.sa_mask)) != 0 {
   418  		// Workaround for bugs in QEMU user mode emulation.
   419  		//
   420  		// QEMU turns calls to the sigaction system call into
   421  		// calls to the C library sigaction call; the C
   422  		// library call rejects attempts to call sigaction for
   423  		// SIGCANCEL (32) or SIGSETXID (33).
   424  		//
   425  		// QEMU rejects calling sigaction on SIGRTMAX (64).
   426  		//
   427  		// Just ignore the error in these case. There isn't
   428  		// anything we can do about it anyhow.
   429  		if sig != 32 && sig != 33 && sig != 64 {
   430  			// Use system stack to avoid split stack overflow on ppc64/ppc64le.
   431  			systemstack(func() {
   432  				throw("sigaction failed")
   433  			})
   434  		}
   435  	}
   436  }
   437  
   438  // rt_sigaction is implemented in assembly.
   439  //go:noescape
   440  func rt_sigaction(sig uintptr, new, old *sigactiont, size uintptr) int32
   441  

View as plain text