http://codereview.chromium.org/6170001/diff/125002/src/arm/macro-assembler-arm.cc File src/arm/macro-assembler-arm.cc (right):
http://codereview.chromium.org/6170001/diff/125002/src/arm/macro-assembler-arm.cc#newcode1520 src/arm/macro-assembler-arm.cc:1520: // return address pushed on stack (could have moved after GC). On 2011/02/02 13:56:28, antonm wrote:
On 2011/02/02 13:24:38, Søren Gjesse wrote: > As far as I can see this relies on DirectCEntryStub itself never
moving. It is
> the same assumption we have for the CEntryStub (and the
RegExpCEntryStub I
> think). Please add a comment on this. > > For the CEntryStub we have been safe so far as it is generated quite
early
(with > crankshaft this is actually not the case for the variant that saves
doubles).
> How about this will it be generated early, or can a test case where
this
> actually moves be crafted?
Søren, yes, that's exactly the reason we call goes indirectly via
stub. Sure, and as discussed offline having DirectCEntryStub created in Heap::CreateFixedStubs will in practice ensure that DirectCEntryStub itself will not move even though there is no explicit code in the GC to ensure that. http://codereview.chromium.org/6170001/ -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
