Add a way to manage thread signal mask using shared word, instead of syscall.

A new syscall sigfastblock(2) is added which registers a uint32_t
variable as containing the count of blocks for signal delivery.  Its
content is read by kernel on each syscall entry and on AST processing,
non-zero count of blocks is interpreted same as the signal mask
blocking all signals.

The biggest downside of the feature that I see is that memory
corruption that affects the registered fast sigblock location, would
cause quite strange application misbehavior. For instance, the process
would be immune to ^C (but killable by SIGKILL).

With consumers (rtld and libthr added), benchmarks do not show a
slow-down of the syscalls in micro-measurements, and macro benchmarks
like buildworld do not demonstrate a difference. Part of the reason is
that buildworld time is dominated by compiler, and clang already links
to libthr. On the other hand, small utilities typically used by shell
scripts have the total number of syscalls cut by half.

The syscall is not exported from the stable libc version namespace on
purpose.  It is intended to be used only by our C runtime
implementation internals.

Tested by:	pho
Disscussed with:	cem, emaste, jilles
Sponsored by:	The FreeBSD Foundation
Differential revision:	https://reviews.freebsd.org/D12773

1 files changed, 166 insertions, 0 deletions

diff --git a/lib/libc/sys/sigfastblock.2 b/lib/libc/sys/sigfastblock.2
new file mode 100644
index 000000000000..da835abdf9aa
--- /dev/null
+++ b/lib/libc/sys/sigfastblock.2
@@ -0,0 +1,166 @@
+.\" Copyright (c) 2016 The FreeBSD Foundation, Inc.
+.\"
+.\" This documentation was written by
+.\" Konstantin Belousov <kib@FreeBSD.org> under sponsorship
+.\" from the FreeBSD Foundation.
+.\"
+.\" Redistribution and use in source and binary forms, with or without
+.\" modification, are permitted provided that the following conditions
+.\" are met:
+.\" 1. Redistributions of source code must retain the above copyright
+.\"    notice, this list of conditions and the following disclaimer.
+.\" 2. Redistributions in binary form must reproduce the above copyright
+.\"    notice, this list of conditions and the following disclaimer in the
+.\"    documentation and/or other materials provided with the distribution.
+.\"
+.\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
+.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
+.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+.\" SUCH DAMAGE.
+.\"
+.\" $FreeBSD$
+.\"
+.Dd December 13, 2019
+.Dt SIGFASTBLOCK 2
+.Os
+.Sh NAME
+.Nm sigfastblock
+.Nd controls signals blocking with a simple memory write
+.Sh LIBRARY
+.Lb libc
+.Sh SYNOPSIS
+.In sys/signalvar.h
+.Ft int
+.Fn sigfastblock "int cmd" "void *ptr"
+.Sh DESCRIPTION
+.Bf -symbolic
+This function is not intended for a direct usage by applications.
+The functionality is provided for implementing some optimizations in
+.Xr ld-elf.so.1 8
+and
+.Lb libthr .
+.Ef
+.Pp
+The function configures the kernel facility that allows a thread to
+block asynchronous signals delivery with a single write to userspace
+memory, avoiding overhead of system calls like
+.Xr sigprocmask 2
+for establishing critical sections.
+The C runtime uses it to optimize implementation of async-signal-safe
+functionality.
+.Pp
+A thread might register a
+.Dv sigblock
+variable of type
+.Vt int
+as a location which is consulted by kernel when calculating the
+blocked signal mask for delivery of asynchronous signals.
+If the variable indicates that blocking is requested, then the kernel
+effectively operates as if the mask containing all blockable signals was
+supplied to
+.Xr sigprocmask 2 .
+.Pp
+The variable is supposed to be modified only from the owning thread,
+there is no way to guarantee visibility of update from other thread
+to kernel when signals are delivered.
+.Pp
+Lower bits of the sigblock variable are reserved as flags,
+which might be set or cleared by kernel at arbitrary moments.
+Userspace code should use
+.Xr atomic 9
+operations of incrementing and decrementing by
+.Dv SIGFASTBLOCK_INC
+quantity to recursively block or unblock signals delivery.
+.Pp
+If a signal would be delivered when unmasked, kernel might set the
+.Dv SIGFASTBLOCK_PEND
+.Dq pending signal
+flag in the sigblock variable.
+Userspace should perform
+.Dv SIGFASTBLOCK_UNBLOCK
+operation when clearing the variable if it notes the pending signal
+bit is set, which would deliver the pending signals immediately.
+Otherwise, signals delivery might be postponed.
+.Pp
+The
+.Fa cmd
+argument specifies one of the following operations:
+.Bl -tag -width SIGFASTBLOCK_UNSETPTR
+.It Dv SIGFASTBLOCK_SETPTR
+Register the variable of type
+.Vt int
+at location pointed to by the
+.Fa ptr
+argument as sigblock variable for the calling thread.
+.It Dv SIGFASTBLOCK_UNSETPTR
+Unregister the currently registered sigblock location.
+Kernel stops inferring the blocked mask from non-zero value of its
+blocked count.
+New location can be registered after previous one is deregistered.
+.It Dv SIGFASTBLOCK_UNBLOCK
+If there are pending signals which should be delivered to the calling
+thread, they are delivered before returning from the call.
+The sigblock variable should have zero blocking count, and indicate
+that the pending signal exists.
+Effectively this means that the variable should have the value
+.Dv SIGFASTBLOCK_PEND .
+.El
+.Sh RETURN VALUES
+.Rv -std
+.Sh ERRORS
+The operation may fail with the following errors:
+.Bl -tag -width Er
+.It Bq Er EBUSY
+The
+.Dv SIGFASTBLOCK_SETPTR
+attempted while the sigblock address was already registered.
+The
+.Dv SIGFASTBLOCK_UNBLOCK
+was called while sigblock variable value is not equal to
+.Dv SIGFASTBLOCK_PEND .
+.It Bq Er EINVAL
+The variable address passed to
+.Dv SIGFASTBLOCK_SETPTR
+is not aligned naturally.
+The
+.Dv SIGFASTBLOCK_UNSETPTR
+operation was attempted without prior successfull call to
+.Dv SIGFASTBLOCK_SETPTR .
+.It Bq Er EFAULT
+Attempt to read or write to the sigblock variable failed.
+Note that kernel generates the
+.Dv SIGSEGV
+signal if an attempt to read from the sigblock variable faulted
+during implicit accesses from syscall entry.
+.El
+.Sh SEE ALSO
+.Xr kill 2 ,
+.Xr signal 2 ,
+.Xr sigprocmask 2 ,
+.Xr libthr 3 ,
+.Xr ld-elf.so.1 8
+.Sh STANDARDS
+The
+.Nm
+function is non-standard, although a similar functionality is a common
+optimization provided by several other systems.
+.Sh HISTORY
+The
+.Nm
+function was introduced in
+.Fx 13.0 .
+.Sh BUGS
+The
+.Nm
+symbol is currently not exported by libc, on purpose.
+Consumers should either use the
+.Dv __sys_fast_sigblock
+symbol from the private libc namespace, or utilize
+.Xr syscall 2 .