tri11paragon
/
BLT
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

allocator changes, allow huge pages

v1
Brett 2024-06-03 02:14:43 -04:00
parent a8b2bc2d01
commit bc68e6dd4a
3 changed files with 106 additions and 98 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CMakeLists.txt
View File

@ -1,7 +1,7 @@
cmake_minimum_required(VERSION 3.20) cmake_minimum_required(VERSION 3.20)
include(cmake/color.cmake) include(cmake/color.cmake)
set(BLT_VERSION 0.17.10) set(BLT_VERSION 0.17.11)
set(BLT_TEST_VERSION 0.0.1) set(BLT_TEST_VERSION 0.0.1)
set(BLT_TARGET BLT) set(BLT_TARGET BLT)

200
include/blt/std/allocator.h
View File

@ -24,8 +24,9 @@
#include <blt/std/ranges.h> #include <blt/std/ranges.h>
#include <blt/std/utility.h> #include <blt/std/utility.h>
#include <blt/std/types.h> #include <blt/std/types.h>
// TODO: remove #include <blt/std/assert.h>
//#include <blt/std/hashmap.h> // TODO: remove
//#include <blt/std/hashmap.h>
#include <blt/compatibility.h> #include <blt/compatibility.h>
#include <stdexcept> #include <stdexcept>
#include "logging.h" #include "logging.h"
@ -545,6 +546,106 @@ namespace blt
// size of 2mb in bytes // size of 2mb in bytes
inline constexpr blt::size_t BLT_2MB_SIZE = 4096 * 512; inline constexpr blt::size_t BLT_2MB_SIZE = 4096 * 512;
/**
* Logging function used for handling mmap errors. call after a failed mmap call.
* @param LOG_FUNC function to log with, must be a BLT_*_STREAM
*/
template<typename LOG_FUNC>
static void handle_mmap_error(LOG_FUNC func = BLT_ERROR_STREAM)
{
#define BLT_WRITE(arg) func << arg << '\n';
switch (errno)
{
case EACCES:
BLT_WRITE("fd not set to open!");
break;
case EAGAIN:
BLT_WRITE("The file has been locked, or too much memory has been locked");
break;
case EBADF:
BLT_WRITE("fd is not a valid file descriptor");
break;
case EEXIST:
BLT_WRITE("MAP_FIXED_NOREPLACE was specified in flags, and the range covered "
"by addr and length clashes with an existing mapping.");
break;
case EINVAL:
BLT_WRITE("We don't like addr, length, or offset (e.g., they are too large, or not aligned on a page boundary).");
BLT_WRITE("Or length was 0");
BLT_WRITE("Or flags contained none of MAP_PRIVATE, MAP_SHARED, or MAP_SHARED_VALIDATE.");
break;
case ENFILE:
BLT_WRITE("The system-wide limit on the total number of open files has been reached.");
break;
case ENODEV:
BLT_WRITE("The underlying filesystem of the specified file does not support memory mapping.");
break;
case ENOMEM:
BLT_WRITE("No memory is available.");
BLT_WRITE("Or The process's maximum number of mappings would have been exceeded. "
"This error can also occur for munmap(), when unmapping a region in the middle of an existing mapping, "
"since this results in two smaller mappings on either side of the region being unmapped.");
BLT_WRITE("Or The process's RLIMIT_DATA limit, described in getrlimit(2), would have been exceeded.");
BLT_WRITE("Or We don't like addr, because it exceeds the virtual address space of the CPU.");
break;
case EOVERFLOW:
BLT_WRITE("On 32-bit architecture together with the large file extension (i.e., using 64-bit off_t): "
"the number of pages used for length plus number of "
"pages used for offset would overflow unsigned long (32 bits).");
break;
case EPERM:
BLT_WRITE("The prot argument asks for PROT_EXEC but the mapped area "
"belongs to a file on a filesystem that was mounted no-exec.");
BLT_WRITE("Or The operation was prevented by a file seal");
BLT_WRITE("Or The MAP_HUGETLB flag was specified, but the caller "
"was not privileged (did not have the CAP_IPC_LOCK capability) "
"and is not a member of the sysctl_hugetlb_shm_group group; "
"see the description of /proc/sys/vm/sysctl_hugetlb_shm_group");
break;
case ETXTBSY:
BLT_WRITE("MAP_DENYWRITE was set but the object specified by fd is open for writing.");
break;
}
}
template<typename T, bool WARN_ON_FAIL = false>
static inline T* allocate_huge_page(blt::size_t BLOCK_SIZE, blt::size_t HUGE_PAGE_SIZE = BLT_2MB_SIZE)
{
BLT_ASSERT((BLOCK_SIZE & (HUGE_PAGE_SIZE - 1)) == 0 && "Must be multiple of the huge page size!");
T* buffer = static_cast<T*>(mmap(nullptr, BLOCK_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_POPULATE, -1, 0));
// if we fail to allocate a huge page we can try to allocate normally
if (buffer == MAP_FAILED)
{
if constexpr (WARN_ON_FAIL)
{
BLT_WARN_STREAM << "We failed to allocate huge pages\n";
handle_mmap_error(BLT_WARN_STREAM);
BLT_WARN_STREAM << "\033[1;31mYou should attempt to enable "
"huge pages as this will allocate normal pages and double the memory usage!\033[22m\n";
}
blt::size_t bytes = BLOCK_SIZE * 2;
buffer = static_cast<T*>(mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0));
if (buffer == MAP_FAILED)
{
BLT_ERROR_STREAM << "Failed to allocate normal pages\n";
handle_mmap_error(BLT_ERROR_STREAM);
throw std::bad_alloc();
}
if constexpr (WARN_ON_FAIL)
{
if (((size_t) buffer & (HUGE_PAGE_SIZE - 1)) != 0)
BLT_ERROR("Pointer is not aligned! %p", buffer);
}
auto* ptr = static_cast<void*>(buffer);
auto ptr_size = reinterpret_cast<blt::size_t>(ptr);
buffer = static_cast<T*>(std::align(BLOCK_SIZE, BLOCK_SIZE, ptr, bytes));
if constexpr (WARN_ON_FAIL)
BLT_ERROR("Offset by %ld pages, resulting: %p", (reinterpret_cast<blt::size_t>(buffer) - ptr_size) / 4096, buffer);
}
return buffer;
}
/** /**
* blt::bump_allocator. Allocates blocks of BLOCK_SIZE with zero reuse. When all objects from a block are fully deallocated the block will be freed * blt::bump_allocator. Allocates blocks of BLOCK_SIZE with zero reuse. When all objects from a block are fully deallocated the block will be freed
* @tparam BLOCK_SIZE size of block to use. recommended to be multiple of page size or huge page size. * @tparam BLOCK_SIZE size of block to use. recommended to be multiple of page size or huge page size.
@ -627,68 +728,6 @@ namespace blt
stats_t stats; stats_t stats;
//blt::hashset_t<void*> deletes; //blt::hashset_t<void*> deletes;
/**
* Logging function used for handling mmap errors. call after a failed mmap call.
* @param LOG_FUNC function to log with, must be a BLT_*_STREAM
*/
template<typename LOG_FUNC>
static void handle_mmap_error(LOG_FUNC func = BLT_ERROR_STREAM)
{
#define BLT_WRITE(arg) func << arg << '\n';
switch (errno)
{
case EACCES:
BLT_WRITE("fd not set to open!");
break;
case EAGAIN:
BLT_WRITE("The file has been locked, or too much memory has been locked");
break;
case EBADF:
BLT_WRITE("fd is not a valid file descriptor");
break;
case EEXIST:
BLT_WRITE("MAP_FIXED_NOREPLACE was specified in flags, and the range covered "
"by addr and length clashes with an existing mapping.");
break;
case EINVAL:
BLT_WRITE("We don't like addr, length, or offset (e.g., they are too large, or not aligned on a page boundary).");
BLT_WRITE("Or length was 0");
BLT_WRITE("Or flags contained none of MAP_PRIVATE, MAP_SHARED, or MAP_SHARED_VALIDATE.");
break;
case ENFILE:
BLT_WRITE("The system-wide limit on the total number of open files has been reached.");
break;
case ENODEV:
BLT_WRITE("The underlying filesystem of the specified file does not support memory mapping.");
break;
case ENOMEM:
BLT_WRITE("No memory is available.");
BLT_WRITE("Or The process's maximum number of mappings would have been exceeded. "
"This error can also occur for munmap(), when unmapping a region in the middle of an existing mapping, "
"since this results in two smaller mappings on either side of the region being unmapped.");
BLT_WRITE("Or The process's RLIMIT_DATA limit, described in getrlimit(2), would have been exceeded.");
BLT_WRITE("Or We don't like addr, because it exceeds the virtual address space of the CPU.");
break;
case EOVERFLOW:
BLT_WRITE("On 32-bit architecture together with the large file extension (i.e., using 64-bit off_t): "
"the number of pages used for length plus number of "
"pages used for offset would overflow unsigned long (32 bits).");
break;
case EPERM:
BLT_WRITE("The prot argument asks for PROT_EXEC but the mapped area "
"belongs to a file on a filesystem that was mounted no-exec.");
BLT_WRITE("Or The operation was prevented by a file seal");
BLT_WRITE("Or The MAP_HUGETLB flag was specified, but the caller "
"was not privileged (did not have the CAP_IPC_LOCK capability) "
"and is not a member of the sysctl_hugetlb_shm_group group; "
"see the description of /proc/sys/vm/sysctl_hugetlb_shm_group");
break;
case ETXTBSY:
BLT_WRITE("MAP_DENYWRITE was set but the object specified by fd is open for writing.");
break;
}
}
struct block struct block
{ {
struct block_metadata_t struct block_metadata_t
@ -724,38 +763,7 @@ namespace blt
#ifdef __unix__ #ifdef __unix__
if constexpr (USE_HUGE) if constexpr (USE_HUGE)
{ {
static_assert((BLOCK_SIZE & (HUGE_PAGE_SIZE - 1)) == 0 && "Must be multiple of the huge page size!"); buffer = allocate_huge_page<block, WARN_ON_FAIL>(BLOCK_SIZE, HUGE_PAGE_SIZE);
buffer = static_cast<block*>(mmap(nullptr, BLOCK_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_POPULATE, -1, 0));
// if we fail to allocate a huge page we can try to allocate normally
if (buffer == MAP_FAILED)
{
if constexpr (WARN_ON_FAIL)
{
BLT_WARN_STREAM << "We failed to allocate huge pages\n";
handle_mmap_error(BLT_WARN_STREAM);
BLT_WARN_STREAM << "\033[1;31mYou should attempt to enable "
"huge pages as this will allocate normal pages and double the memory usage!\033[22m\n";
}
blt::size_t bytes = BLOCK_SIZE * 2;
buffer = static_cast<block*>(mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0));
if (buffer == MAP_FAILED)
{
BLT_ERROR_STREAM << "Failed to allocate normal pages\n";
handle_mmap_error(BLT_ERROR_STREAM);
throw std::bad_alloc();
}
if constexpr (WARN_ON_FAIL)
{
if (((size_t) buffer & (HUGE_PAGE_SIZE - 1)) != 0)
BLT_ERROR("Pointer is not aligned! %p", buffer);
}
auto* ptr = static_cast<void*>(buffer);
auto ptr_size = reinterpret_cast<blt::size_t>(ptr);
buffer = static_cast<block*>(std::align(BLOCK_SIZE, BLOCK_SIZE, ptr, bytes));
if constexpr (WARN_ON_FAIL)
BLT_ERROR("Offset by %ld pages, resulting: %p", (reinterpret_cast<blt::size_t>(buffer) - ptr_size) / 4096, buffer);
}
} else } else
buffer = reinterpret_cast<block*>(std::aligned_alloc(BLOCK_SIZE, BLOCK_SIZE)); buffer = reinterpret_cast<block*>(std::aligned_alloc(BLOCK_SIZE, BLOCK_SIZE));
#else #else

2
libraries/parallel-hashmap

@ -1 +1 @@
Subproject commit 10368163ab1f4367d2f0685b5928b1c973ebd1ec Subproject commit d88c5e15079047777b418132ece5879e7c9aaa2b