construct/ircd/allocator.cc

789 lines
15 KiB
C++

// 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.
#include <RB_INC_SYS_RESOURCE_H
#include <RB_INC_SYS_MMAN_H
// Uncomment or -D this #define to enable our own crude but simple ability to
// profile dynamic memory usage. Global `new` and `delete` will be captured
// here by this definition file into thread_local counters accessible via
// ircd::allocator::profile. This feature allows the developer to find out if
// allocations are occurring during some scope by sampling the counters.
//
// #define RB_PROF_ALLOC
namespace ircd::allocator
{
static void advise_hugepage(void *const &, const size_t &alignment, const size_t &size);
}
#if defined(MADV_NORMAL) && defined(POSIX_MADV_NORMAL)
static_assert(MADV_NORMAL == POSIX_MADV_NORMAL);
#endif
#if defined(MADV_SEQUENTIAL) && defined(POSIX_MADV_SEQUENTIAL)
static_assert(MADV_SEQUENTIAL == POSIX_MADV_SEQUENTIAL);
#endif
#if defined(MADV_RANDOM) && defined(POSIX_MADV_RANDOM)
static_assert(MADV_RANDOM == POSIX_MADV_RANDOM);
#endif
#if defined(MADV_WILLNEED) && defined(POSIX_MADV_WILLNEED)
static_assert(MADV_WILLNEED == POSIX_MADV_WILLNEED);
#endif
#if defined(MADV_DONTNEED) && defined(POSIX_MADV_DONTNEED)
static_assert(MADV_DONTNEED == POSIX_MADV_DONTNEED);
#endif
[[gnu::hot]]
char *
ircd::allocator::allocate(const size_t alignment,
const size_t size)
{
assume(alignment > 0);
assume(size % alignment == 0);
assume(alignment % sizeof(void *) == 0);
void *ret;
switch(int errc(::posix_memalign(&ret, alignment, size)); errc)
{
[[likely]]
case 0:
break;
[[unlikely]]
case int(std::errc::not_enough_memory):
throw std::bad_alloc{};
[[unlikely]]
default:
throw_system_error();
__builtin_unreachable();
}
assert(ret != nullptr);
assert(uintptr_t(ret) % alignment == 0);
if(likely(info::thp_size))
advise_hugepage(ret, alignment, size);
#ifdef RB_PROF_ALLOC
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.alloc_bytes += size;
this_thread.alloc_count++;
#endif
return reinterpret_cast<char *>(ret);
}
void
ircd::allocator::advise_hugepage(void *const &ptr,
const size_t &alignment,
const size_t &size)
#if defined(MADV_HUGEPAGE)
try
{
if(likely(alignment < info::thp_size))
return;
if(likely(alignment % size_t(info::thp_size) != 0))
return;
if(!has(info::thp_enable, "[madvise]"))
return;
sys::call(::madvise, ptr, size, MADV_HUGEPAGE);
}
catch(const std::exception &e)
{
log::critical
{
"Failed to madvise(%p, %zu, MADV_HUGEPAGE) :%s",
ptr,
size,
e.what(),
};
}
#else
{
}
#endif
void
ircd::allocator::readonly(const mutable_buffer &buf,
const bool enable)
#if defined(HAVE_MPROTECT)
{
const int prot
{
enable?
PROT_READ:
PROT_READ|PROT_WRITE
};
void *const ptr(mutable_cast(data(buf)));
sys::call(::mprotect, ptr, size(buf), prot);
}
#else
{
#warning "mprotect(2) not available for this compilation."
}
#endif
#if defined(HAVE_MPROTECT)
void
ircd::allocator::protect(const const_buffer &buf,
const bool enable)
{
const int prot
{
enable?
PROT_NONE:
PROT_READ|PROT_WRITE
};
void *const ptr(mutable_cast(data(buf)));
sys::call(::mprotect, ptr, size(buf), prot);
}
#else
{
#warning "mprotect(2) not available for this compilation."
}
#endif
void
ircd::allocator::lock(const const_buffer &buf,
const bool enable)
#if defined(HAVE_MLOCK2) && defined(MLOCK_ONFAULT)
{
int flags {0};
flags |= MLOCK_ONFAULT;
// can't mlock w/ valgrind
if(unlikely(vg::active))
return;
if(enable)
syscall(::mlock2, data(buf), size(buf), flags);
else
syscall(::munlock, data(buf), size(buf));
}
#else
{
#warning "mlock2(2) not available for this compilation."
}
#endif
size_t
ircd::allocator::sync(const const_buffer &buf,
const bool invd)
{
assert(aligned(data(buf), info::page_size));
const prof::syscall_usage_warning message
{
"msync(2) MS_SYNC MS_INVALIDATE:%b", invd
};
#if defined(HAVE_MSYNC)
int flags {MS_SYNC};
flags |= invd? MS_INVALIDATE: 0;
sys::call(::msync, mutable_cast(data(buf)), size(buf), flags);
return size(buf);
#else
return 0;
#endif
}
size_t
ircd::allocator::flush(const const_buffer &buf,
const bool invd)
{
assert(aligned(data(buf), info::page_size));
const prof::syscall_usage_warning message
{
"msync(2) MS_ASYNC MS_INVALIDATE:%b", invd
};
#if defined(HAVE_MSYNC)
int flags {MS_ASYNC};
flags |= invd? MS_INVALIDATE: 0;
sys::call(::msync, mutable_cast(data(buf)), size(buf), flags);
return size(buf);
#else
return 0;
#endif
}
size_t
ircd::allocator::evict(const const_buffer &buf)
{
#if defined(POSIX_MADV_DONTNEED)
return advise(buf, POSIX_MADV_DONTNEED);
#else
return 0;
#endif
}
size_t
ircd::allocator::prefetch(const const_buffer &buf)
{
#if defined(POSIX_MADV_WILLNEED)
return advise(buf, POSIX_MADV_WILLNEED);
#else
return 0;
#endif
}
#if defined(HAVE_MADVISE)
size_t
ircd::allocator::advise(const const_buffer &buf,
const int &advice)
{
assert(aligned(data(buf), info::page_size));
switch(const auto r(::madvise(mutable_cast(data(buf)), size(buf), advice)); r)
{
[[likely]]
case 0:
return size(buf); // success
[[unlikely]]
default:
throw_system_error(r); // error
}
__builtin_unreachable();
}
#elif defined(HAVE_POSIX_MADVISE)
size_t
ircd::allocator::advise(const const_buffer &buf,
const int &advice)
{
const auto res
{
syscall(::posix_madvise, mutable_cast(data(buf)), size(buf), advice)
};
return size(buf);
}
#else
#warning "posix_madvise(2) not available for this compilation."
size_t
ircd::allocator::advise(const const_buffer &buf,
const int &advice)
{
return 0;
}
#endif
size_t
ircd::allocator::incore(const const_buffer &buf)
{
const auto base
{
align(begin(buf), info::page_size)
};
if(unlikely(!base))
return 0;
const auto top
{
align_up(end(buf), info::page_size)
};
assert(base <= data(buf));
const auto below
{
std::distance(base, begin(buf))
};
assert(top >= data(buf) + size(buf));
const auto span
{
std::distance(base, top)
};
const auto above
{
std::distance(end(buf), top)
};
assert(span >= 0);
assert(above >= 0);
assert(below >= 0);
assert(above < ssize_t(info::page_size));
assert(below < ssize_t(info::page_size));
assert(below + ssize_t(size(buf)) + above == span);
auto remain(span), ret(0L);
thread_local uint8_t vec alignas(64) [4096];
for(auto i(0); i < span / ssizeof(vec) && remain > 0; ++i)
{
const auto len
{
std::min(ssizeof(vec) * ssize_t(info::page_size), remain)
};
assert(len > 0);
assert(len <= span);
const ssize_t vec_size
(
std::ceil(len / double(info::page_size))
);
assert(vec_size > 0);
assert(vec_size <= ssizeof(vec));
syscall(::mincore, mutable_cast(base), len, vec);
for(auto j(0); j < vec_size; ++j)
ret += (vec[j] & 0x01) * info::page_size;
remain -= len;
assert(remain >= 0);
if(!remain && (vec[vec_size - 1] & 0x01)) // last iteration
ret -= above;
assert(ret >= 0);
if(i == 0 && (vec[0] & 0x01)) // first iteration
ret -= below;
assert(ret >= 0);
}
assert(remain == 0);
assert(ret <= ssize_t(size(buf)));
assert(ret >= 0);
return ret;
}
//
// control panel
//
bool
__attribute__((weak))
ircd::allocator::trim(const size_t &pad)
noexcept
{
return false;
}
ircd::string_view
__attribute__((weak))
ircd::allocator::get(const string_view &key,
const mutable_buffer &buf)
{
return {};
}
ircd::string_view
__attribute__((weak))
ircd::allocator::set(const string_view &key,
const string_view &val,
const mutable_buffer &cur)
{
return {};
}
//
// allocator::state
//
void
ircd::allocator::state::deallocate(const uint &pos,
const size_type &n)
{
for(size_t i(0); i < n; ++i)
{
assert(test(pos + i));
btc(pos + i);
}
last = pos;
}
uint
ircd::allocator::state::allocate(const size_type &n,
const uint &hint)
{
const auto ret
{
allocate(std::nothrow, n, hint)
};
if(unlikely(ret >= size))
throw std::bad_alloc();
return ret;
}
uint
ircd::allocator::state::allocate(std::nothrow_t,
const size_type &n,
const uint &hint)
{
const auto next(this->next(n));
if(unlikely(next >= size)) // No block of n was found anywhere (next is past-the-end)
return next;
for(size_t i(0); i < n; ++i)
{
assert(!test(next + i));
bts(next + i);
}
last = next + n;
return next;
}
uint
ircd::allocator::state::next(const size_t &n)
const
{
uint ret(last), rem(n);
for(; ret < size && rem; ++ret)
if(test(ret))
rem = n;
else
--rem;
if(likely(!rem))
return ret - n;
for(ret = 0, rem = n; ret < last && rem; ++ret)
if(test(ret))
rem = n;
else
--rem;
if(unlikely(rem)) // The allocator should throw std::bad_alloc if !rem
return size;
return ret - n;
}
bool
ircd::allocator::state::available(const size_t &n)
const
{
return this->next(n) < size;
}
//
// allocator::scope
//
decltype(ircd::allocator::scope::current)
ircd::allocator::scope::current;
ircd::allocator::scope::scope(alloc_closure ac,
realloc_closure rc,
free_closure fc)
:theirs
{
current
}
,user_alloc
{
std::move(ac)
}
,user_realloc
{
std::move(rc)
}
,user_free
{
std::move(fc)
}
{
// If an allocator::scope instance already exists somewhere
// up the stack, *current will already be set. We only install
// our global hook handlers at the first instance ctor and
// uninstall it after that first instance dtors.
if(!current)
hook_init();
current = this;
}
ircd::allocator::scope::~scope()
noexcept
{
assert(current);
current = theirs;
// Reinstall the pre-existing hooks after our last scope instance
// has destructed (the first to have constructed). We know this when
// current becomes null.
if(!current)
hook_fini();
}
void
__attribute__((weak))
ircd::allocator::scope::hook_init()
noexcept
{
}
void
__attribute__((weak))
ircd::allocator::scope::hook_fini()
noexcept
{
}
//
// allocator::profile
//
thread_local ircd::allocator::profile
ircd::allocator::profile::this_thread
{};
ircd::allocator::profile
ircd::allocator::operator-(const profile &a,
const profile &b)
{
profile ret(a);
ret -= b;
return ret;
}
ircd::allocator::profile
ircd::allocator::operator+(const profile &a,
const profile &b)
{
profile ret(a);
ret += b;
return ret;
}
ircd::allocator::profile &
ircd::allocator::operator-=(profile &a,
const profile &b)
{
a.alloc_count -= b.alloc_count;
a.free_count -= b.free_count;
a.alloc_bytes -= b.alloc_bytes;
a.free_bytes -= b.free_bytes;
return a;
}
ircd::allocator::profile &
ircd::allocator::operator+=(profile &a,
const profile &b)
{
a.alloc_count += b.alloc_count;
a.free_count += b.free_count;
a.alloc_bytes += b.alloc_bytes;
a.free_bytes += b.free_bytes;
return a;
}
//
// resource limits
//
#if defined(HAVE_SYS_RESOURCE_H) && defined(RLIMIT_MEMLOCK)
size_t
ircd::allocator::rlimit_memlock(const size_t &req)
try
{
rlimit rlim {0};
rlim.rlim_cur = req;
syscall(setrlimit, RLIMIT_MEMLOCK, &rlim);
char pbuf[48];
log::info
{
"Raised resource limit for locked memory to %s",
req != -1UL?
pretty(pbuf, iec(req)):
"unlimited"_sv,
};
return rlim.rlim_cur;
}
catch(const std::system_error &e)
{
char pbuf[48];
log::warning
{
"Failed to raise resource limit for locked memory to %s :%s",
req != -1UL?
pretty(pbuf, iec(req)):
"unlimited"_sv,
e.what(),
};
return rlimit_memlock();
}
#else
size_t
ircd::allocator::rlimit_memlock(const size_t &req)
{
return 0;
}
#endif
#if defined(HAVE_SYS_RESOURCE_H) && defined(RLIMIT_MEMLOCK)
size_t
ircd::allocator::rlimit_memlock()
{
rlimit rlim;
syscall(getrlimit, RLIMIT_MEMLOCK, &rlim);
return rlim.rlim_cur;
}
#else
size_t
ircd::allocator::rlimit_memlock()
{
return 0;
}
#endif
#if defined(HAVE_SYS_RESOURCE_H) && defined(RLIMIT_DATA)
size_t
ircd::allocator::rlimit_data()
{
rlimit rlim;
syscall(getrlimit, RLIMIT_DATA, &rlim);
return rlim.rlim_cur;
}
#else
size_t
ircd::allocator::rlimit_data()
{
return 0;
}
#endif
#if defined(HAVE_SYS_RESOURCE_H) && defined(RLIMIT_AS)
size_t
ircd::allocator::rlimit_as()
{
rlimit rlim;
syscall(getrlimit, RLIMIT_AS, &rlim);
return rlim.rlim_cur;
}
#else
size_t
ircd::allocator::rlimit_as()
{
return 0;
}
#endif
//
// Developer profiling
//
#ifdef RB_PROF_ALLOC // --------------------------------------------------
void *
__attribute__((alloc_size(1), malloc, returns_nonnull))
operator new(const size_t size)
{
void *const &ptr(::malloc(size));
if(unlikely(!ptr))
throw std::bad_alloc();
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.alloc_bytes += size;
this_thread.alloc_count++;
return ptr;
}
void
operator delete(void *const ptr)
noexcept
{
::free(ptr);
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.free_count++;
}
void
operator delete(void *const ptr,
const size_t size)
noexcept
{
::free(ptr);
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.free_bytes += size;
this_thread.free_count++;
}
#endif // RB_PROF_ALLOC --------------------------------------------------
//
// Linker symbol wrapping hook
//
extern "C" [[gnu::weak]] void *__real_malloc(size_t size);
extern "C" [[gnu::weak]] void *__real_calloc(size_t nmemb, size_t size);
extern "C" [[gnu::weak]] void *__real_realloc(void *ptr, size_t size);
extern "C" [[gnu::weak]] void __real_free(void *ptr);
extern "C" void *
__wrap_malloc(size_t size)
{
void *const &ptr(::__real_malloc(size));
if(unlikely(!ptr))
throw std::bad_alloc();
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.alloc_bytes += size;
this_thread.alloc_count++;
return ptr;
}
extern "C" void *
__wrap_calloc(size_t nmemb, size_t size)
{
void *const &ptr(::__real_calloc(nmemb, size));
if(unlikely(!ptr))
throw std::bad_alloc();
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.alloc_bytes += nmemb * size;
this_thread.alloc_count++;
return __real_calloc(nmemb, size);
}
extern "C" void *
__wrap_realloc(void *ptr, size_t size)
{
void *const &ret(::__real_realloc(ptr, size));
if(unlikely(!ret))
throw std::bad_alloc();
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.alloc_bytes += size;
this_thread.alloc_count++;
return ret;
}
extern "C" void
__wrap_free(void *ptr)
{
__real_free(ptr);
auto &this_thread(ircd::allocator::profile::this_thread);
this_thread.free_bytes += 0UL; //TODO: XXX
this_thread.free_count++;
}