[libc++abi][AIX] Use reserved slot in stack to pass the address of exception object

[xingxue]

Summary

The existing implementation of the personality for legacy IBM xlclang++ compiler generated code passes the address of exception object in r14 for the landing pad to retrieve with a call to __xlc_exception_handle(). This clobbers the content of r14 in user code (and potentially, when running cleanup actions, the address of another exception object being passed). This patch changes to use the stack slot reserved for compilers to pass the address. It has been confirmed that xlclang++-generated code does not use this slot. The following is the AIX stack frame layout.

Low Address
                                                          32-bit  64-bit

        SP----> +---------------------------------------+
                | back chain to caller                  | 0       0
                +---------------------------------------+
                | saved CR                              | 4       8 (8-11)
                +---------------------------------------+
                | saved LR                              | 8       16
                +---------------------------------------+
                | reserved for compilers                | 12      24                  <=======
                +---------------------------------------+
                | reserved for binders                  | 16      32
                +---------------------------------------+
                | saved TOC pointer                     | 20      40
                +---------------------------------------+
                | Parameter save area (+padding*) (P)   | 24      48
                +---------------------------------------+
                | Alloca space (A)                      | 24+P    etc.
                +---------------------------------------+
                | Local variable space (L)              | 24+P+A
                +---------------------------------------+
                | Float/int conversion temporary (X)    | 24+P+A+L
                +---------------------------------------+
                | Save area for AltiVec registers (W)   | 24+P+A+L+X
                +---------------------------------------+
                | AltiVec alignment padding (Y)         | 24+P+A+L+X+W
                +---------------------------------------+
                | Save area for VRSAVE register (Z)     | 24+P+A+L+X+W+Y
                +---------------------------------------+
                | Save area for GP registers (G)        | 24+P+A+X+L+X+W+Y+Z
                +---------------------------------------+
                | Save area for FP registers (F)        | 24+P+A+X+L+X+W+Y+Z+G
                +---------------------------------------+
        old SP->| back chain to caller's caller         |
                +---------------------------------------+
                | saved CR                              | 4       8 (8-11)
                +---------------------------------------+
                | saved LR                              | 8       16
                +---------------------------------------+
High Address

[hubert.reinterpretcast]

LGTM from the AIX perspective.

[xingxue]

Updated diff to include binary files.

[cebowleratibm]

libcxxabi/src/aix_state_tab_eh.inc

665

Suggestion: rename dummy to force_a_stackframe or similar. The comments are helpful but the name makes it really hard for anyone to erroneously remove it, noting that you also have tests that would fail if anyone did but the more layers of defense and readability the better.

[xingxue]

Changes in the update:

• addressed comment: rename dummy() to force_a_stackframe();
• added attribute optnone to force_a_stackframe();
• use assembly for IBM legacy compiler xlclang++ generated test cases instead of binary object files.

[xingxue]

libcxxabi/src/aix_state_tab_eh.inc

665

Renamed as suggested, thanks!

[hubert.reinterpretcast]

Confirming LGTM from the AIX side.

[cebowleratibm]

I have no further concerns with this patch and can confirm the solution works for EH support of the XL C++ V16 xlclang++ compiler. No further concerns.

[xingxue]

This patch is AIX specific and it has been thoroughly reviewed and approved by AIX compiler experts @hubert.reinterpretcast, @cebowleratibm (Thank you!). Since we need to have the patch in the coming service pack, I've committed it in rGa499051f10a2d0150b60c14493558476039f701a. Commit rGfb391e45e9cc35429f077413ca660f4f5f5212c1 modified a comment.

[ldionne]

In D136257#3889885, @xingxue wrote:

This patch is AIX specific and it has been thoroughly reviewed and approved by AIX compiler experts @hubert.reinterpretcast, @cebowleratibm (Thank you!). Since we need to have the patch in the coming service pack, I've committed it in rGa499051f10a2d0150b60c14493558476039f701a. Commit rGfb391e45e9cc35429f077413ca660f4f5f5212c1 modified a comment.

That's fine.

[xingxue]

Thanks, @ldionne!

libcxxabi/src/aix_state_tab_eh.inc

__xlcxx_personality_v0(int version, _Unwind_Action actions, uint64_t exceptionClass,
   }
   if (actions & _UA_CLEANUP_PHASE) {
     // Phase 2 cleanup:
     if (results.reason == _URC_HANDLER_FOUND) {
       // Store the address of unwind_exception in the stack field
       // reserved for compilers (SP + 3 * sizeof(uintptr_t)) in the stack of
       // the caller of the function containing the landing pad (within the link
       // area for the call to the latter) for __xlc_exception_handle()
-      // to retrieve.
+      // to retrieve when it is called by the landing pad.
       uintptr_t *currentSP = reinterpret_cast<uintptr_t*>(_Unwind_GetGR(context, 1));
       uintptr_t *callersSP = reinterpret_cast<uintptr_t*>(currentSP[0]);
       callersSP[3] = reinterpret_cast<uintptr_t>(unwind_exception);
       _LIBCXXABI_TRACE_STATETAB("Handshake: set unwind_exception=%p in stack=%p\n", reinterpret_cast<void*>(unwind_exception), reinterpret_cast<void*>(callersSP));
       // Jump to the handler.
       _Unwind_SetIP(context, results.landingPad);
       return _URC_INSTALL_CONTEXT;
     }

 // assumption holds true: 1) a call to force_a_stackframe() is made inside
 // __xlc_exception_handle() to make it non-leaf; and 2) optimizations are
 // disabled for this function with attribute 'optnone'. Note: this function
 // may not work as expected if these are changed.
 __attribute__((optnone))
 _LIBCXXABI_FUNC_VIS uintptr_t __xlc_exception_handle() {
   // Make a call to force_a_stackframe() so that the compiler creates a stack
   // frame for this function.
   force_a_stackframe();
 
   // Get the SP of this function, i.e., __xlc_exception_handle().
   uintptr_t *lastStack;
   asm("mr %0, 1" : "=r"(lastStack));
   // Get the SP of the caller of __xlc_exception_handle().
   uintptr_t *callerStack = reinterpret_cast<uintptr_t*>(lastStack[0]);
   // Get the SP of the caller of the caller.
   uintptr_t *callerStack2 = reinterpret_cast<uintptr_t*>(callerStack[0]);