Small issue with array chunking in G1
Tony Printezis
tprintezis at twitter.com
Fri Jan 31 18:49:33 UTC 2014
Thomas
I attached the diffs (note that our repo is a bit behind yours so the
change might need to be reworked a bit; but you'll get the idea).
Tony
On 1/31/14, 8:28 AM, Thomas Schatzl wrote:
> Hi Tony,
>
> On Thu, 2014-01-30 at 11:29 -0500, Tony Printezis wrote:
>> (looking at Thomas' recent change reminded me that I've been meaning to
>> ask about this)
>>
>> Talking about array chunking in G1, we noticed a small difference in
>> that code compared to what the other GCs (ParNew / PS) do. G1 uses the
>> to-space length field to encode how much of the array has been scanned
>> (i.e., the from-space length field is always correct). ParNew and PS use
>> the from-space length field (and the to-space length field is always
>> correct). I came across that issue when I was working on a change to
>> periodically scan PLABs (for some profiling stuff I was working on) and
>> in G1 some PLABs were unparseable because the length field of arrays
>> could be incorrect (even if the PLAB is retired, it's possible that the
>> length field is still incorrect given that other threads might still be
>> working on that array).
>>
>> The patch to change G1 to do what ParNew / PS also do is very small (and
>> we've had it for a few months in our repo without any issues). Any
>> chance of convincing you to also take it? I should be clear that the
>> patch doesn't fix a correctness issue in G1; the code is correct as is.
>> But it will bring G1 in line with the other two GCs and allow PLAB
>> scanning during GC, if you want to do that...
> Where is the patch to look at?
>
> Thomas
>
>
--
Tony Printezis | JVM/GC Engineer / VM Team | Twitter
@TonyPrintezis
tprintezis at twitter.com
-------------- next part --------------
diff --git a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
index 986384c..ba2f3d0 100644
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
@@ -4756,9 +4756,9 @@ oop G1ParCopyClosure<do_gen_barrier, barrier, do_mark_object>
if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
// We keep track of the next start index in the length field of
- // the to-space object. The actual length can be found in the
- // length field of the from-space object.
- arrayOop(obj)->set_length(0);
+ // the from-space object. The actual length can be found in the
+ // length field of the to-space object.
+ arrayOop(old)->set_length(0);
oop* old_p = set_partial_array_mask(old);
_par_scan_state->push_on_queue(old_p);
} else {
@@ -4832,16 +4832,16 @@ template <class T> void G1ParScanPartialArrayClosure::do_oop_nv(T* p) {
assert(Universe::heap()->is_in_reserved(from_obj), "must be in heap.");
assert(from_obj->is_objArray(), "must be obj array");
objArrayOop from_obj_array = objArrayOop(from_obj);
- // The from-space object contains the real length.
- int length = from_obj_array->length();
+ // We keep track of the next start index in the length field of the
+ // from-space object.
+ int next_index = from_obj_array->length();
assert(from_obj->is_forwarded(), "must be forwarded");
oop to_obj = from_obj->forwardee();
assert(from_obj != to_obj, "should not be chunking self-forwarded objects");
objArrayOop to_obj_array = objArrayOop(to_obj);
- // We keep track of the next start index in the length field of the
- // to-space object.
- int next_index = to_obj_array->length();
+ // The to-space object contains the real length.
+ int length = to_obj_array->length();
assert(0 <= next_index && next_index < length,
err_msg("invariant, next index: %d, length: %d", next_index, length));
@@ -4851,7 +4851,7 @@ template <class T> void G1ParScanPartialArrayClosure::do_oop_nv(T* p) {
// We'll try not to push a range that's smaller than ParGCArrayScanChunk.
if (remainder > 2 * ParGCArrayScanChunk) {
end = start + ParGCArrayScanChunk;
- to_obj_array->set_length(end);
+ from_obj_array->set_length(end);
// Push the remainder before we process the range in case another
// worker has run out of things to do and can steal it.
oop* from_obj_p = set_partial_array_mask(from_obj);
@@ -4860,7 +4860,7 @@ template <class T> void G1ParScanPartialArrayClosure::do_oop_nv(T* p) {
assert(length == end, "sanity");
// We'll process the final range for this object. Restore the length
// so that the heap remains parsable in case of evacuation failure.
- to_obj_array->set_length(end);
+ from_obj_array->set_length(end);
}
_scanner.set_region(_g1->heap_region_containing_raw(to_obj));
// Process indexes [start,end). It will also process the header
More information about the hotspot-gc-dev
mailing list