0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-11-29 18:22:50 +01:00
construct/include/ircd/ctx/this_ctx.h

245 lines
9 KiB
C
Raw Normal View History

// Matrix Construct
//
// Copyright (C) Matrix Construct Developers, Authors & Contributors
// Copyright (C) 2016-2018 Jason Volk <jason@zemos.net>
//
// Permission to use, copy, modify, and/or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright notice and this permission notice is present in all copies. The
// full license for this software is available in the LICENSE file.
#pragma once
#define HAVE_IRCD_CTX_THIS_CTX_H
namespace ircd::ctx {
/// Interface to the currently running context
inline namespace this_ctx
{
struct critical_indicator; // Indicates if yielding happened for a section
struct critical_assertion; // Assert no yielding for a section
struct exception_handler; // Must be present to yield in a handler
struct uninterruptible; // Scope convenience for interruptible()
struct ctx &cur(); ///< Assumptional reference to *current
const uint64_t &id(); // Unique ID for cur ctx
string_view name(); // Optional label for cur ctx
void wait(); // Returns when context is woken up.
void yield(); // Allow other contexts to run before returning.
void interruption_point(); // throws if interruption_requested()
bool interruption_requested(); // interruption(cur())
void interruptible(const bool &); // interruptible(cur(), bool) +INTERRUPTION POINT
void interruptible(const bool &, std::nothrow_t) noexcept;
struct stack_usage_assertion; // Assert safety factor (see ctx/prof.h)
size_t stack_at_here() __attribute__((noinline));
ulong cycles_here();
// Return remaining time if notified; or <= 0 if not, and timeout thrown on throw overloads
microseconds wait(const microseconds &, const std::nothrow_t &);
template<class E, class duration> nothrow_overload<E, duration> wait(const duration &);
template<class E = timeout, class duration> throw_overload<E, duration> wait(const duration &);
// Returns false if notified; true if time point reached, timeout thrown on throw_overloads
bool wait_until(const time_point &tp, const std::nothrow_t &);
template<class E> nothrow_overload<E, bool> wait_until(const time_point &tp);
template<class E = timeout> throw_overload<E> wait_until(const time_point &tp);
// Ignores notes. Throws if interrupted.
void sleep_until(const time_point &tp);
template<class duration> void sleep(const duration &);
void sleep(const int &secs);
}}
namespace ircd::ctx
{
/// Points to the currently running context or null for main stack (do not modify)
extern __thread ctx *current;
}
/// An instance of stack_usage_assertion is placed on a ctx stack where one
/// wants to test the stack usage at both construction and destruction points
/// to ensure it is less than the value set in ctx::prof::settings which is
/// generally some engineering safety factor of 2-3 etc. This should not be
/// entirely relied upon except during debug builds, however we may try to
/// provide an optimized build mode enabling these to account for any possible
/// differences in the stack between the environments.
///
struct ircd::ctx::this_ctx::stack_usage_assertion
{
#ifndef NDEBUG
stack_usage_assertion();
~stack_usage_assertion() noexcept;
#endif
};
/// An instance of critical_assertion detects an attempt to context switch.
///
/// For when the developer specifically does not want any yielding in a
/// section or anywhere up the stack from it. This device does not prevent
/// a switch and may carry no meaning outside of debug-mode compilation. It is
/// good practice to use this device even when it appears obvious the
/// section's callgraph has no chance of yielding: code changes, and everything
/// up the graph can change without taking notice of your section.
///
class ircd::ctx::this_ctx::critical_assertion
{
#ifndef NDEBUG
bool theirs;
public:
critical_assertion();
~critical_assertion() noexcept;
#endif
};
/// An instance of critical_indicator reports if context switching happened.
///
/// A critical_indicator remains true after construction until a context switch
/// has occurred. It then becomes false. This is not an assertion and is
/// available in optimized builds for real use. For example, a context may
/// want to recompute some value after a context switch and opportunistically
/// skip this effort when this indicator shows no switch occurred.
///
class ircd::ctx::this_ctx::critical_indicator
{
uint64_t state;
public:
uint64_t count() const { return yields(cur()) - state; }
operator bool() const { return yields(cur()) == state; }
critical_indicator()
:state{yields(cur())}
{}
};
/// An instance of exception_handler must be present to allow a context
/// switch inside a catch block. This is due to ABI limitations that stack
/// exceptions with thread-local assumptions and don't expect catch blocks
/// on the same thread to interleave when we switch the stack.
///
/// We first increment the refcount for the caught exception so it remains
/// intuitively accessible for the rest of the catch block. Then the presence
/// of this object makes the ABI believe the catch block has ended.
///
/// The exception cannot then be rethrown. DO NOT RETHROW THE EXCEPTION.
///
struct ircd::ctx::this_ctx::exception_handler
:std::exception_ptr
{
exception_handler() noexcept;
exception_handler(exception_handler &&) = delete;
exception_handler(const exception_handler &) = delete;
exception_handler &operator=(exception_handler &&) = delete;
exception_handler &operator=(const exception_handler &) = delete;
};
/// An instance of uninterruptible will suppress interrupts sent to the
/// context for the scope. Suppression does not discard any interrupt,
/// it merely ignores it at all interruption points until the suppression
/// ends, after which it will be thrown.
///
struct ircd::ctx::this_ctx::uninterruptible
{
struct nothrow;
uninterruptible();
uninterruptible(uninterruptible &&) = delete;
uninterruptible(const uninterruptible &) = delete;
~uninterruptible() noexcept(false);
};
/// A variant of uinterruptible for users that must guarantee the ending of
/// the suppression scope will not be an interruption point. The default
/// behavior for uninterruptible is to throw, even from its destructor, to
/// fulfill the interruption request without any more delay.
///
struct ircd::ctx::this_ctx::uninterruptible::nothrow
{
nothrow() noexcept;
nothrow(nothrow &&) = delete;
nothrow(const nothrow &) = delete;
~nothrow() noexcept;
};
/// This overload matches ::sleep() and acts as a drop-in for ircd contexts.
/// interruption point.
inline void
ircd::ctx::this_ctx::sleep(const int &secs)
{
sleep(seconds(secs));
}
/// Yield the context for a period of time and ignore notifications. sleep()
/// is like wait() but it only returns after the timeout and not because of a
/// note.
/// interruption point.
template<class duration>
void
ircd::ctx::this_ctx::sleep(const duration &d)
{
sleep_until(steady_clock::now() + d);
}
/// Wait for a notification until a point in time. If there is a notification
/// then context continues normally. If there's never a notification then an
/// exception (= timeout) is thrown.
/// interruption point.
template<class E>
ircd::throw_overload<E>
ircd::ctx::this_ctx::wait_until(const time_point &tp)
{
if(wait_until<std::nothrow_t>(tp))
throw E();
}
/// Wait for a notification until a point in time. If there is a notification
/// then returns true. If there's never a notification then returns false.
/// interruption point. this is not noexcept.
template<class E>
ircd::nothrow_overload<E, bool>
ircd::ctx::this_ctx::wait_until(const time_point &tp)
{
return wait_until(tp, std::nothrow);
}
/// Wait for a notification for at most some amount of time. If the duration is
/// reached without a notification then E (= timeout) is thrown. Otherwise,
/// returns the time remaining on the duration.
/// interruption point
template<class E,
class duration>
ircd::throw_overload<E, duration>
ircd::ctx::this_ctx::wait(const duration &d)
{
const auto ret(wait<std::nothrow_t>(d));
return ret <= duration(0)? throw E() : ret;
}
/// Wait for a notification for some amount of time. This function returns
/// when a context is notified. It always returns the duration remaining which
/// will be <= 0 to indicate a timeout without notification.
/// interruption point. this is not noexcept.
template<class E,
class duration>
ircd::nothrow_overload<E, duration>
ircd::ctx::this_ctx::wait(const duration &d)
{
using std::chrono::duration_cast;
const auto ret(wait(duration_cast<microseconds>(d), std::nothrow));
return duration_cast<duration>(ret);
}
/// Reference to the currently running context. Call if you expect to be in a
/// context. Otherwise use the ctx::current pointer.
inline ircd::ctx::ctx &
ircd::ctx::this_ctx::cur()
{
assert(current);
return *current;
}