infernap12/jdk
1
0
Fork 0
mirror of https://github.com/openjdk/jdk.git synced 2026-01-28 12:09:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

8371702: ZGC: NUMA-Affinity for Worker Threads in the Relocation Phase

Browse Source 
Co-authored-by: Axel Boldt-Christmas <aboldtch@openjdk.org>
Reviewed-by: aboldtch, eosterlund
This commit is contained in:
Joel Sikström 2025-11-27 09:02:18 +00:00
parent 7cd3d7f157
commit de546d0e03
1 changed files with 27 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
src/hotspot/share/gc/z/zRelocate.cpp
View File

@ -1087,7 +1087,6 @@ private:
ZRelocateSmallAllocator _small_allocator;
ZRelocateMediumAllocator _medium_allocator;
const size_t _total_forwardings;
volatile size_t _numa_local_forwardings;
public:
ZRelocateTask(ZRelocationSet* relocation_set,
@ -1104,8 +1103,7 @@ public:
_medium_targets(medium_targets),
_small_allocator(_generation),
_medium_allocator(_generation, shared_medium_targets),
_total_forwardings(relocation_set->nforwardings()),
_numa_local_forwardings(0) {
_total_forwardings(relocation_set->nforwardings()) {
for (uint32_t i = 0; i < ZNUMA::count(); i++) {
ZRelocationSetParallelIterator* const iter = _iters->addr(i);
@ -1124,18 +1122,17 @@ public:
// Signal that we're not using the queue anymore. Used mostly for asserts.
_queue->deactivate();
if (ZNUMA::is_enabled()) {
log_debug(gc, reloc, numa)("Forwardings relocated NUMA-locally: %zu / %zu (%.0f%%)",
_numa_local_forwardings, _total_forwardings, percent_of(_numa_local_forwardings, _total_forwardings));
}
}
virtual void work() {
ZRelocateWork<ZRelocateSmallAllocator> small(&_small_allocator, _small_targets->addr(), _generation);
ZRelocateWork<ZRelocateMediumAllocator> medium(&_medium_allocator, _medium_targets->addr(), _generation);
const uint32_t num_nodes = ZNUMA::count();
uint32_t numa_local_forwardings_worker = 0;
const uint32_t start_node = ZNUMA::id();
uint32_t current_node = start_node;
bool has_affinity = false;
bool has_affinity_current_node = false;
const auto do_forwarding = [&](ZForwarding* forwarding) {
ZPage* const page = forwarding->page();
@ -1167,26 +1164,30 @@ public:
const auto do_forwarding_one_from_iter = [&]() {
ZForwarding* forwarding;
const uint32_t start_node = ZNUMA::id();
uint32_t current_node = start_node;
for (uint32_t i = 0; i < num_nodes; i++) {
for (;;) {
if (_iters->get(current_node).next_if(&forwarding, check_numa_local, current_node)) {
claim_and_do_forwarding(forwarding);
if (current_node == start_node) {
// Track if this forwarding was relocated on the local NUMA node
numa_local_forwardings_worker++;
// Set thread affinity for NUMA-local processing (if needed)
if (UseNUMA && !has_affinity_current_node) {
os::numa_set_thread_affinity(Thread::current(), ZNUMA::numa_id_to_node(current_node));
has_affinity = true;
has_affinity_current_node = true;
}
// Perform the forwarding task
claim_and_do_forwarding(forwarding);
return true;
}
// Check next node.
// No work found on the current node, move to the next node
current_node = (current_node + 1) % num_nodes;
}
has_affinity_current_node = false;
return false;
// If we've looped back to the starting node there's no more work to do
if (current_node == start_node) {
return false;
}
}
};
for (;;) {
@ -1209,11 +1210,13 @@ public:
}
}
if (ZNUMA::is_enabled()) {
AtomicAccess::add(&_numa_local_forwardings, numa_local_forwardings_worker, memory_order_relaxed);
}
_queue->leave();
if (UseNUMA && has_affinity) {
// Restore the affinity of the thread so that it isn't bound to a specific
// node any more
os::numa_set_thread_affinity(Thread::current(), -1);
}
}
virtual void resize_workers(uint nworkers) {