Fixes to "Retract Hp *before* checking for HpLim==0"

compiler/codeGen/CgForeignCall.hs

@@ -212,7 +212,11 @@ emitLoadThreadState = do
 	                      bWord),
 	-- SpLim = tso->stack + RESERVED_STACK_WORDS;
 	CmmAssign spLim (cmmOffsetW (cmmOffset (CmmReg (CmmLocal tso)) tso_STACK)
-			            rESERVED_STACK_WORDS)
+			            rESERVED_STACK_WORDS),
+        -- HpAlloc = 0;
+        --   HpAlloc is assumed to be set to non-zero only by a failed
+        --   a heap check, see HeapStackCheck.cmm:GC_GENERIC
+        CmmAssign hpAlloc (CmmLit zeroCLit)
     ]
   emitOpenNursery
   -- and load the current cost centre stack from the TSO when profiling:
@@ -267,13 +271,14 @@ stgHp		  = CmmReg hp
 stgCurrentTSO	  = CmmReg currentTSO
 stgCurrentNursery = CmmReg currentNursery
 
-sp, spLim, hp, hpLim, currentTSO, currentNursery :: CmmReg
+sp, spLim, hp, hpLim, currentTSO, currentNursery, hpAlloc :: CmmReg
 sp		  = CmmGlobal Sp
 spLim		  = CmmGlobal SpLim
 hp		  = CmmGlobal Hp
 hpLim		  = CmmGlobal HpLim
 currentTSO	  = CmmGlobal CurrentTSO
 currentNursery 	  = CmmGlobal CurrentNursery
+hpAlloc 	  = CmmGlobal HpAlloc
 
 -- -----------------------------------------------------------------------------
 -- For certain types passed to foreign calls, we adjust the actual

rts/HeapStackCheck.cmm

@@ -26,6 +26,17 @@ import LeaveCriticalSection;
  *    - If HpLim==0, indicating that we should context-switch, we yield
  *      to the scheduler (return ThreadYielding).
  *
+ * Note that we must leave no slop in the heap (this is a requirement
+ * for LDV profiling, at least), so if we just had a heap-check
+ * failure, then we must retract Hp by HpAlloc.  How do we know
+ * whether there was a heap-check failure?  HpLim might be zero, and
+ * yet we got here as a result of a stack-check failure.  Hence, we
+ * require that HpAlloc is only non-zero if there was a heap-check
+ * failure, otherwise it is zero, so we can always safely subtract
+ * HpAlloc from Hp.
+ *
+ * Hence, HpAlloc is zeroed in LOAD_THREAD_STATE().
+ *
  *    - If the context_switch flag is set (the backup plan if setting HpLim
  *      to 0 didn't trigger a context switch), we yield to the scheduler
  *	(return ThreadYielding).
@@ -70,6 +81,7 @@ import LeaveCriticalSection;
         }						\
         if (HpAlloc <= BLOCK_SIZE				\
             && bdescr_link(CurrentNursery) != NULL) {		\
+            HpAlloc = 0;                                        \
             CLOSE_NURSERY();					\
             CurrentNursery = bdescr_link(CurrentNursery);	\
             OPEN_NURSERY();					\