roland/xo-alloc2
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

xo-alloc2: ++ documentation + threshold size for THP feature

Browse source
This commit is contained in:
Roland Conybeare 2025-12-23 21:06:38 -05:00
commit 289751d3fd
18 changed files with 642 additions and 192 deletions
Download patch file Download diff file Expand all files Collapse all files

45
src/alloc2/AllocError.cpp Normal file
View file

@ -0,0 +1,45 @@
/** @file AllocError.cpp
*
* @author Roland Conybeare, Dec 2025
**/
#include "AllocError.hpp"
namespace xo {
namespace mm {
const char *
AllocError::error_description(error x)
{
switch (x) {
case error::invalid:
break;
case error::ok:
return "ok";
case error::reserve_exhausted:
return "reserve-exhausted";
case error::commit_failed:
return "commit-failed";
case error::header_size_mask:
return "header-size-mask";
case error::orphan_sub_alloc:
return "orphan-sub-alloc";
case error::alloc_info_disabled:
return "alloc-info-disabled";
case error::alloc_info_address:
return "alloc-info-address";
case error::alloc_iterator_not_supported:
return "alloc-iterator-not-supported";
case error::alloc_iterator_deref:
return "alloc-iterator-deref";
case error::alloc_iterator_next:
return "alloc-iterator-next";
}
return "?error";
}
} /*namespace mm*/
} /*namespace xo*/
/* end AllocError.cpp */

1
src/alloc2/CMakeLists.txt
View file

@ -3,6 +3,7 @@
set(SELF_LIB xo_alloc2)
set(SELF_SRCS
AllocError.cpp
AllocInfo.cpp
cmpresult.cpp

129
src/alloc2/DArena.cpp
View file

@ -19,56 +19,30 @@ namespace xo {
using std::size_t;
namespace mm {
/** Map a contiguous uncommitted memory range comprising
* a whole multiple of @p hugepage_z, with at least
* @p req_z bytes.
*
* Memory will also be aligned on @p hugepage_z boundary
* (2MB in practice)
*
* - @p req_z is rounded up to a multiple of @p hugepage_z
* - Resulting uncommitted address range not backed by
* physical memory.
* - since hugpage-aligned, can find base of mapped range
* by masking off the bottom log2(align_z) bits.
* May rely on this for GC metadata
* - opt-in to transparent huge pages (THP).
* Reduces page-fault time by a lot, in return for
* lower VM granularity
* - rejecting inferior MAP_HUGETLB|MAP_HUGE_2MB flags on ::mmap here:
* - requires previously-reserved memory in /proc/sys/vm/nr_hugepages
* - reserved pages permenently resident in RAM, never swapped
* - memory cost incurred even if no application is using said pages
*
* TODO: for OSX -> need something else here.
* MAP_ALIGNED_SUPER with mmap() and/or
* use mach_vm_allocate()
*
* @return pair giving mapped address range [lo, hi)
**/
auto
DArena::map_aligned_range(size_t req_z, size_t hugepage_z) -> range_type
DArena::map_aligned_range(size_t req_z,
size_t align_z,
bool enable_hugepage_flag) -> range_type
{
// 1. round up to multiple of hugepage_z
size_t target_z = padding::with_padding(req_z, hugepage_z); // 4.
// 1. round up to multiple of align_z
size_t target_z = padding::with_padding(req_z, align_z); // 4.
// 2. mmap() will give us page-aligned memory,
// but not hugepage-aligned.
//
// Over-request by hugepage_z to ensure
// hugepage-aligned subrange of size target_z
// Over-request by align_z to ensure
// aligned subrange of size target_z
//
byte * base = (byte *)(::mmap(nullptr,
target_z + hugepage_z,
target_z + align_z,
PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS,
-1, 0));
// on mmap success: upper limit of mapped address range
byte * hi = base + (target_z + hugepage_z);
byte * hi = base + (target_z + align_z);
// lowest hugepage-aligned address in [base, hi)
byte * aligned_base = (byte *)(padding::with_padding((size_t)base, hugepage_z));
byte * aligned_base = (byte *)(padding::with_padding((size_t)base, align_z));
// end of hugeppage-aligned range starting at aligned_base
byte * aligned_hi = aligned_base + target_z;
@ -88,9 +62,9 @@ namespace xo {
xtag("size", req_z)));
}
assert((size_t)aligned_base % hugepage_z == 0);
assert((size_t)aligned_base % align_z == 0);
assert(aligned_base >= base);
assert(aligned_base < base + hugepage_z);
assert(aligned_base < base + align_z);
}
// 4. release unaligned prefix
@ -108,21 +82,25 @@ namespace xo {
}
#ifdef __linux__
/** linux:
* opt-in to transparent huge pages (THP)
* provided OS configured to support them.
* otherwise fallback gracefully.
*
* Huge pages -> use fewer TLB entries + faster
* shorter path through page table.
*
* When we commit (i.e. obtain physical memory on page fault),
* typically expect to pay ~1us per superpage.
* Much better than ~500us to commit 512 4k VM pages.
*
* But wasted if we don't use the memory.
**/
::madvise(aligned_base, target_z, MADV_HUGEPAGE); // 8.
if (enable_hugepage_flag) {
/** linux:
* opt-in to transparent huge pages (THP)
* provided OS configured to support them.
* otherwise fallback gracefully.
*
* Huge pages -> use fewer TLB entries + faster
* shorter path through page table.
*
* When we commit (i.e. obtain physical memory on page fault),
* typically expect to pay ~1us per superpage.
* Much better than ~500us to commit 512 4k VM pages.
*
* But wasted if we don't use the memory.
*
* Page table has a handful of levels
**/
::madvise(aligned_base, target_z, MADV_HUGEPAGE); // 8.
}
#endif
return std::make_pair(aligned_base, aligned_hi);
@ -133,7 +111,20 @@ namespace xo {
{
//scope log(XO_DEBUG(debug_flag), xtag("name", name));
auto [lo, hi] = map_aligned_range(cfg.size_, cfg.hugepage_z_);
/* vm page size. 4KB, probably */
size_t page_z = getpagesize();
bool enable_hugepage_flag = (cfg.size_ >= cfg.hugepage_z_);
/* Align start of arena memory on this boundary.
* Will use THP (transparent huge pages) if available
* and arena size is at least as large as hugepage size (2MB, probably)
*/
size_t align_z = (enable_hugepage_flag ? cfg.hugepage_z_ : page_z);
auto [lo, hi] = map_aligned_range(cfg.size_,
align_z,
enable_hugepage_flag);
if (!lo) {
// control here implies mmap() failed silently
@ -142,8 +133,6 @@ namespace xo {
xtag("size", cfg.size_)));
}
size_t page_z = getpagesize();
#ifdef NOPE
log && log(xtag("lo", (void*)lo_),
@ -151,14 +140,16 @@ namespace xo {
xtag("hugepage_z", hugepage_z_));
#endif
return DArena(cfg, page_z, lo, hi);
return DArena(cfg, page_z, align_z, lo, hi);
} /*map*/
DArena::DArena(const ArenaConfig & cfg,
size_type page_z,
size_type arena_align_z,
byte * lo,
byte * hi) : config_{cfg},
page_z_{page_z},
arena_align_z_{arena_align_z},
lo_{lo},
committed_z_{0},
free_{lo},
@ -177,6 +168,7 @@ namespace xo {
DArena::DArena(DArena && other) {
config_ = other.config_;
page_z_ = other.page_z_;
arena_align_z_ = other.arena_align_z_;
lo_ = other.lo_;
committed_z_ = other.committed_z_;
free_ = other.free_;
@ -200,6 +192,7 @@ namespace xo {
{
config_ = other.config_;
page_z_ = other.page_z_;
arena_align_z_ = other.arena_align_z_;
lo_ = other.lo_;
committed_z_ = other.committed_z_;
free_ = other.free_;
@ -529,23 +522,25 @@ namespace xo {
*
*/
std::size_t aligned_target_z = padding::with_padding(target_z, config_.hugepage_z_);
std::size_t aligned_target_z = padding::with_padding(target_z, arena_align_z_);
std::byte * commit_start = limit_; // = lo_ + committed_z_;
std::size_t add_commit_z = aligned_target_z - committed_z_;
assert(limit_ == lo_ + committed_z_);
// log && log(xtag("aligned_offset_z", aligned_offset_z),
// xtag("add_commit_z", add_commit_z));
// log && log("expand committed range",
// xtag("commit_start", commit_start),
// xtag("add_commit_z", add_commit_z),
// xtag("commit_end", commit_start + add_commit_z));
if (::mprotect(commit_start,
add_commit_z,
PROT_READ | PROT_WRITE) != 0) [[unlikely]]
{
if (log) {
log("commit failed!");
log(xtag("aligned_target_z", aligned_target_z),
xtag("commit_start", commit_start),
xtag("add_commit_z", add_commit_z),
xtag("commit_end", commit_start + add_commit_z)
);
}
capture_error(error::commit_failed, add_commit_z);
return false;
}
@ -563,8 +558,8 @@ namespace xo {
free_ += config_.header_.guard_z_;
}
assert(committed_z_ % config_.hugepage_z_ == 0);
assert(reinterpret_cast<size_t>(limit_) % config_.hugepage_z_ == 0);
assert(committed_z_ % arena_align_z_ == 0);
assert(reinterpret_cast<size_t>(limit_) % arena_align_z_ == 0);
return true;
} /*expand*/