*
* ---------------------------------------------------------------------------*/
-// #include "PosixSource.h"
+#include "PosixSource.h"
#include "Rts.h"
-#include "RtsFlags.h"
+#include "HsFFI.h"
+
+#include "Storage.h"
#include "RtsUtils.h"
#include "Apply.h"
-#include "OSThreads.h"
-#include "LdvProfile.h"
#include "Updates.h"
#include "Stats.h"
#include "Schedule.h"
#include "Sanity.h"
#include "BlockAlloc.h"
-#include "MBlock.h"
#include "ProfHeap.h"
-#include "SchedAPI.h"
#include "Weak.h"
#include "Prelude.h"
-#include "ParTicky.h" // ToDo: move into Rts.h
#include "RtsSignals.h"
#include "STM.h"
-#include "HsFFI.h"
-#include "Linker.h"
#if defined(RTS_GTK_FRONTPANEL)
#include "FrontPanel.h"
#endif
#include "Trace.h"
#include "RetainerProfile.h"
+#include "LdvProfile.h"
#include "RaiseAsync.h"
#include "Papi.h"
+#include "Stable.h"
#include "GC.h"
#include "GCThread.h"
#include "Evac.h"
#include "Scav.h"
#include "GCUtils.h"
+#include "MarkStack.h"
#include "MarkWeak.h"
#include "Sparks.h"
#include "Sweep.h"
static void resize_nursery (void);
static void start_gc_threads (void);
static void scavenge_until_all_done (void);
-static nat inc_running (void);
-static nat dec_running (void);
+static StgWord inc_running (void);
+static StgWord dec_running (void);
static void wakeup_gc_threads (nat n_threads, nat me);
static void shutdown_gc_threads (nat n_threads, nat me);
#endif
/* -----------------------------------------------------------------------------
- The mark bitmap & stack.
+ The mark stack.
-------------------------------------------------------------------------- */
-#define MARK_STACK_BLOCKS 4
-
-bdescr *mark_stack_bdescr;
-StgPtr *mark_stack;
-StgPtr *mark_sp;
-StgPtr *mark_splim;
-
-// Flag and pointers used for falling back to a linear scan when the
-// mark stack overflows.
-rtsBool mark_stack_overflowed;
-bdescr *oldgen_scan_bd;
-StgPtr oldgen_scan;
+bdescr *mark_stack_top_bd; // topmost block in the mark stack
+bdescr *mark_stack_bd; // current block in the mark stack
+StgPtr mark_sp; // pointer to the next unallocated mark stack entry
/* -----------------------------------------------------------------------------
GarbageCollect: the main entry point to the garbage collector.
// tell the STM to discard any cached closures it's hoping to re-use
stmPreGCHook();
+ // lock the StablePtr table
+ stablePtrPreGC();
+
#ifdef DEBUG
mutlist_MUTVARS = 0;
mutlist_MUTARRS = 0;
n = initialise_N(force_major_gc);
#if defined(THREADED_RTS)
- work_stealing = RtsFlags.ParFlags.parGcLoadBalancing;
+ work_stealing = RtsFlags.ParFlags.parGcLoadBalancingEnabled &&
+ N >= RtsFlags.ParFlags.parGcLoadBalancingGen;
// It's not always a good idea to do load balancing in parallel
// GC. In particular, for a parallel program we don't want to
// lose locality by moving cached data into another CPU's cache
#ifdef DEBUG
// check for memory leaks if DEBUG is on
- memInventory(traceClass(DEBUG_gc));
+ memInventory(DEBUG_gc);
#endif
// check stack sanity *before* GC
/* Allocate a mark stack if we're doing a major collection.
*/
if (major_gc && oldest_gen->steps[0].mark) {
- nat mark_stack_blocks;
- mark_stack_blocks = stg_max(MARK_STACK_BLOCKS,
- oldest_gen->steps[0].n_old_blocks / 100);
- mark_stack_bdescr = allocGroup(mark_stack_blocks);
- mark_stack = (StgPtr *)mark_stack_bdescr->start;
- mark_sp = mark_stack;
- mark_splim = mark_stack + (mark_stack_blocks * BLOCK_SIZE_W);
+ mark_stack_bd = allocBlock();
+ mark_stack_top_bd = mark_stack_bd;
+ mark_stack_bd->link = NULL;
+ mark_stack_bd->u.back = NULL;
+ mark_sp = mark_stack_bd->start;
} else {
- mark_stack_bdescr = NULL;
+ mark_stack_bd = NULL;
+ mark_stack_top_bd = NULL;
+ mark_sp = NULL;
}
// this is the main thread
pinned_object_block = NULL;
// Free the mark stack.
- if (mark_stack_bdescr != NULL) {
- freeGroup(mark_stack_bdescr);
+ if (mark_stack_top_bd != NULL) {
+ debugTrace(DEBUG_gc, "mark stack: %d blocks",
+ countBlocks(mark_stack_top_bd));
+ freeChain(mark_stack_top_bd);
}
// Free any bitmaps.
#ifdef DEBUG
// check for memory leaks if DEBUG is on
- memInventory(traceClass(DEBUG_gc));
+ memInventory(DEBUG_gc);
#endif
#ifdef RTS_GTK_FRONTPANEL
slop = calcLiveBlocks() * BLOCK_SIZE_W - live;
stat_endGC(allocated, live, copied, N, max_copied, avg_copied, slop);
+ // unlock the StablePtr table
+ stablePtrPostGC();
+
// Guess which generation we'll collect *next* time
initialise_N(force_major_gc);
}
}
+void
+freeGcThreads (void)
+{
+ if (gc_threads != NULL) {
+#if defined(THREADED_RTS)
+ nat i;
+ for (i = 0; i < RtsFlags.ParFlags.nNodes; i++) {
+ stgFree (gc_threads[i]);
+ }
+ stgFree (gc_threads);
+#else
+ stgFree (gc_threads);
+#endif
+ gc_threads = NULL;
+ }
+}
+
/* ----------------------------------------------------------------------------
Start GC threads
------------------------------------------------------------------------- */
-static nat gc_running_threads;
+static volatile StgWord gc_running_threads;
-#if defined(THREADED_RTS)
-static Mutex gc_running_mutex;
-#endif
-
-static nat
+static StgWord
inc_running (void)
{
- nat n_running;
- ACQUIRE_LOCK(&gc_running_mutex);
- n_running = ++gc_running_threads;
- RELEASE_LOCK(&gc_running_mutex);
- ASSERT(n_running <= n_gc_threads);
- return n_running;
+ StgWord new;
+ new = atomic_inc(&gc_running_threads);
+ ASSERT(new <= n_gc_threads);
+ return new;
}
-static nat
+static StgWord
dec_running (void)
{
- nat n_running;
- ACQUIRE_LOCK(&gc_running_mutex);
- ASSERT(n_gc_threads != 0);
- n_running = --gc_running_threads;
- RELEASE_LOCK(&gc_running_mutex);
- return n_running;
+ ASSERT(gc_running_threads != 0);
+ return atomic_dec(&gc_running_threads);
}
static rtsBool
write_barrier();
// scavenge objects in compacted generation
- if (mark_stack_overflowed || oldgen_scan_bd != NULL ||
- (mark_stack_bdescr != NULL && !mark_stack_empty())) {
+ if (mark_stack_bd != NULL && !mark_stack_empty()) {
return rtsTrue;
}
{
nat r;
- debugTrace(DEBUG_gc, "GC thread %d working", gct->thread_index);
loop:
+ traceEvent(&capabilities[gct->thread_index], EVENT_GC_WORK);
+
#if defined(THREADED_RTS)
if (n_gc_threads > 1) {
scavenge_loop();
// scavenge_loop() only exits when there's no work to do
r = dec_running();
- debugTrace(DEBUG_gc, "GC thread %d idle (%d still running)",
- gct->thread_index, r);
+ traceEvent(&capabilities[gct->thread_index], EVENT_GC_IDLE);
+
+ debugTrace(DEBUG_gc, "%d GC threads still running", r);
while (gc_running_threads != 0) {
// usleep(1);
// scavenge_loop() to perform any pending work.
}
- // All threads are now stopped
- debugTrace(DEBUG_gc, "GC thread %d finished.", gct->thread_index);
+ traceEvent(&capabilities[gct->thread_index], EVENT_GC_DONE);
}
#if defined(THREADED_RTS)
void
gcWorkerThread (Capability *cap)
{
+ gc_thread *saved_gct;
+
+ // necessary if we stole a callee-saves register for gct:
+ saved_gct = gct;
+
cap->in_gc = rtsTrue;
gct = gc_threads[cap->no];
gct->thread_index);
ACQUIRE_SPIN_LOCK(&gct->mut_spin);
debugTrace(DEBUG_gc, "GC thread %d on my way...", gct->thread_index);
+
+ SET_GCT(saved_gct);
}
#endif
+#if defined(THREADED_RTS)
+
void
waitForGcThreads (Capability *cap USED_IF_THREADS)
{
-#if defined(THREADED_RTS)
nat n_threads = RtsFlags.ParFlags.nNodes;
nat me = cap->no;
nat i, j;
prodCapability(&capabilities[i], cap->running_task);
}
}
- for (j=0; j < 10000000; j++) {
+ for (j=0; j < 10; j++) {
retry = rtsFalse;
for (i=0; i < n_threads; i++) {
if (i == me) continue;
}
}
if (!retry) break;
+ yieldThread();
}
}
-#endif
}
+#endif // THREADED_RTS
+
static void
start_gc_threads (void)
{
#if defined(THREADED_RTS)
gc_running_threads = 0;
- initMutex(&gc_running_mutex);
#endif
}
#endif
}
+#if defined(THREADED_RTS)
void
releaseGCThreads (Capability *cap USED_IF_THREADS)
{
-#if defined(THREADED_RTS)
nat n_threads = RtsFlags.ParFlags.nNodes;
nat me = cap->no;
nat i;
ACQUIRE_SPIN_LOCK(&gc_threads[i]->gc_spin);
RELEASE_SPIN_LOCK(&gc_threads[i]->mut_spin);
}
-#endif
}
+#endif
/* ----------------------------------------------------------------------------
Initialise a generation that is to be collected
static void
resize_nursery (void)
{
+ lnat min_nursery = RtsFlags.GcFlags.minAllocAreaSize * n_capabilities;
+
if (RtsFlags.GcFlags.generations == 1)
{ // Two-space collector:
nat blocks;
else
{
blocks *= RtsFlags.GcFlags.oldGenFactor;
- if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
+ if (blocks < min_nursery)
{
- blocks = RtsFlags.GcFlags.minAllocAreaSize;
+ blocks = min_nursery;
}
}
resizeNurseries(blocks);
(((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
(100 + (long)g0s0_pcnt_kept);
- if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
- blocks = RtsFlags.GcFlags.minAllocAreaSize;
+ if (blocks < (long)min_nursery) {
+ blocks = min_nursery;
}
resizeNurseries((nat)blocks);