Mercurial > hg > truffle
diff src/cpu/x86/vm/frame_x86.inline.hpp @ 107:93b6525e3b82
6603919: Stackwalking crash on x86 -server with Sun Studio's collect -j on
Summary: Rewrite frame::safe_for_sender and friends to be safe for collector/analyzer
Reviewed-by: dcubed, kvn
author | sgoldman |
---|---|
date | Tue, 08 Apr 2008 12:23:15 -0400 |
parents | a61af66fc99e |
children | d1605aabd0a1 |
line wrap: on
line diff
--- a/src/cpu/x86/vm/frame_x86.inline.hpp Mon Apr 07 15:15:16 2008 -0700 +++ b/src/cpu/x86/vm/frame_x86.inline.hpp Tue Apr 08 12:23:15 2008 -0400 @@ -72,15 +72,20 @@ _unextended_sp = sp; _fp = fp; _pc = (address)(sp[-1]); - assert(_pc != NULL, "no pc?"); + + // Here's a sticky one. This constructor can be called via AsyncGetCallTrace + // when last_Java_sp is non-null but the pc fetched is junk. If we are truly + // unlucky the junk value could be to a zombied method and we'll die on the + // find_blob call. This is also why we can have no asserts on the validity + // of the pc we find here. AsyncGetCallTrace -> pd_get_top_frame_for_signal_handler + // -> pd_last_frame should use a specialized version of pd_last_frame which could + // call a specilaized frame constructor instead of this one. + // Then we could use the assert below. However this assert is of somewhat dubious + // value. + // assert(_pc != NULL, "no pc?"); + _cb = CodeCache::find_blob(_pc); - // In case of native stubs, the pc retreived here might be - // wrong. (the _last_native_pc will have the right value) - // So do not put add any asserts on the _pc here. - // QQQ The above comment is wrong and has been wrong for years. This constructor - // should (and MUST) not be called in that situation. In the native situation - // the pc should be supplied to the constructor. _deopt_state = not_deoptimized; if (_cb != NULL && _cb->is_nmethod() && ((nmethod*)_cb)->is_deopt_pc(_pc)) { _pc = (((nmethod*)_cb)->get_original_pc(this));