8134630: make code and comments consistent for stack lock optimization

Reviewed-by: dholmes, coleenp

src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp

@@ -775,15 +775,33 @@ void InterpreterMacroAssembler::lock_object(Register lock_reg)
       cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
     }
 
-    // Test if the oopMark is an obvious stack pointer, i.e.,
+    // Fast check for recursive lock.
+    //
+    // Can apply the optimization only if this is a stack lock
+    // allocated in this thread. For efficiency, we can focus on
+    // recently allocated stack locks (instead of reading the stack
+    // base and checking whether 'mark' points inside the current
+    // thread stack):
     //  1) (mark & 7) == 0, and
-    //  2) rsp <= mark < mark + os::pagesize()
+    //  2) sp <= mark < mark + os::pagesize()
+    //
+    // Warning: sp + os::pagesize can overflow the stack base. We must
+    // neither apply the optimization for an inflated lock allocated
+    // just above the thread stack (this is why condition 1 matters)
+    // nor apply the optimization if the stack lock is inside the stack
+    // of another thread. The latter is avoided even in case of overflow
+    // because we have guard pages at the end of all stacks. Hence, if
+    // we go over the stack base and hit the stack of another thread,
+    // this should not be in a writeable area that could contain a
+    // stack lock allocated by that thread. As a consequence, a stack
+    // lock less than page size away from sp is guaranteed to be
+    // owned by the current thread.
     //
     // These 3 tests can be done by evaluating the following
-    // expression: ((mark - rsp) & (7 - os::vm_page_size())),
+    // expression: ((mark - sp) & (7 - os::vm_page_size())),
     // assuming both stack pointer and pagesize have their
     // least significant 3 bits clear.
-    // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg
+    // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
     // NOTE2: aarch64 does not like to subtract sp from rn so take a
     // copy
     mov(rscratch1, sp);

src/hotspot/cpu/x86/interp_masm_x86.cpp

@@ -1252,15 +1252,33 @@ void InterpreterMacroAssembler::lock_object(Register lock_reg) {
 
     const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
 
-    // Test if the oopMark is an obvious stack pointer, i.e.,
+    // Fast check for recursive lock.
+    //
+    // Can apply the optimization only if this is a stack lock
+    // allocated in this thread. For efficiency, we can focus on
+    // recently allocated stack locks (instead of reading the stack
+    // base and checking whether 'mark' points inside the current
+    // thread stack):
     //  1) (mark & zero_bits) == 0, and
     //  2) rsp <= mark < mark + os::pagesize()
     //
+    // Warning: rsp + os::pagesize can overflow the stack base. We must
+    // neither apply the optimization for an inflated lock allocated
+    // just above the thread stack (this is why condition 1 matters)
+    // nor apply the optimization if the stack lock is inside the stack
+    // of another thread. The latter is avoided even in case of overflow
+    // because we have guard pages at the end of all stacks. Hence, if
+    // we go over the stack base and hit the stack of another thread,
+    // this should not be in a writeable area that could contain a
+    // stack lock allocated by that thread. As a consequence, a stack
+    // lock less than page size away from rsp is guaranteed to be
+    // owned by the current thread.
+    //
     // These 3 tests can be done by evaluating the following
     // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
     // assuming both stack pointer and pagesize have their
     // least significant bits clear.
-    // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
+    // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
     subptr(swap_reg, rsp);
     andptr(swap_reg, zero_bits - os::vm_page_size());