[RISC-V] ELT Profiler Bring-Up (#91313)

* [RISC-V] Generate profiling function callbacks

Initial implementation based on ARM64 code.

* [RISC-V] Fix asm stub for calling profiler callbacks

* Fix argument registers according to RISC-V calling convention
* Fix field offsets for PROFILE_PLATFORM_SPECIFIC_DATA
* Make sure field offsets for PROFILE_PLATFORM_SPECIFIC_DATA stay fixed by static asserting the offsets in asmconstants.h

* [RISC-V] Pass arguments for Profile(Enter|Leave|Tailcall)Naked stubs in t0 and t1 because t2 is used to store the call address of the stub

* [RISC-V] Remove unimplemented definition of EmitRet

* [RISC-V] Copy struct from registers into a buffer so that profiler can see whole struct arguments laid out in memory

* [RISC-V] Implement ProfileArgIterator::GetReturnBufferAddr()

Since the RISC-V ABI says values are returned like the first named argument, re-use the struct copying routine from argument parsing as much as possible.

* Factor out duplicated test results checking routine in SlowPathELTProfiler::Shutdown()

* [RISC-V] Clean up PROFILE_PLATFORM_SPECIFIC_DATA

* Remove unused t0 field
* Remove 'unused' field and widen 'flags' to 64 bits to maintain alignment and shave off one sw instruction

* [RISC-V] Remove commented out code

* [RISC-V] Fix formatting

* [RISC-V] Apply format patch from failed check

* [RISC-V] Post-review fixes

src/coreclr/jit/codegenriscv64.cpp

@@ -816,8 +816,8 @@ void CodeGen::genCaptureFuncletPrologEpilogInfo()
 
         if (compiler->lvaPSPSym != BAD_VAR_NUM)
         {
-            if (CallerSP_to_PSP_slot_delta !=
-                compiler->lvaGetCallerSPRelativeOffset(compiler->lvaPSPSym)) // for debugging
+            if (CallerSP_to_PSP_slot_delta != compiler->lvaGetCallerSPRelativeOffset(compiler->lvaPSPSym)) // for
+                                                                                                           // debugging
             {
                 printf("lvaGetCallerSPRelativeOffset(lvaPSPSym): %d\n",
                        compiler->lvaGetCallerSPRelativeOffset(compiler->lvaPSPSym));
@@ -7162,35 +7162,6 @@ void CodeGen::instGen_MemoryBarrier(BarrierKind barrierKind)
     GetEmitter()->emitIns_I(INS_fence, EA_4BYTE, INS_BARRIER_FULL);
 }
 
-//-----------------------------------------------------------------------------------
-// genProfilingLeaveCallback: Generate the profiling function leave or tailcall callback.
-// Technically, this is not part of the epilog; it is called when we are generating code for a GT_RETURN node.
-//
-// Arguments:
-//     helper - which helper to call. Either CORINFO_HELP_PROF_FCN_LEAVE or CORINFO_HELP_PROF_FCN_TAILCALL
-//
-// Return Value:
-//     None
-//
-void CodeGen::genProfilingLeaveCallback(unsigned helper /*= CORINFO_HELP_PROF_FCN_LEAVE*/)
-{
-    assert((helper == CORINFO_HELP_PROF_FCN_LEAVE) || (helper == CORINFO_HELP_PROF_FCN_TAILCALL));
-
-    // Only hook if profiler says it's okay.
-    if (!compiler->compIsProfilerHookNeeded())
-    {
-        return;
-    }
-
-    compiler->info.compProfilerCallback = true;
-
-    // Need to save on to the stack level, since the helper call will pop the argument
-    unsigned saveStackLvl2 = genStackLevel;
-
-    /* Restore the stack level */
-    SetStackLevel(saveStackLvl2);
-}
-
 /*-----------------------------------------------------------------------------
  *
  *  Push/Pop any callee-saved registers we have used
@@ -8102,6 +8073,7 @@ void CodeGen::genFnPrologCalleeRegArgs()
     assert(!regArgMaskLive);
 }
 
+#ifdef PROFILING_SUPPORTED
 //-----------------------------------------------------------------------------------
 // genProfilingEnterCallback: Generate the profiling function enter callback.
 //
@@ -8110,17 +8082,79 @@ void CodeGen::genFnPrologCalleeRegArgs()
 //     pInitRegZeroed - OUT parameter. *pInitRegZeroed set to 'false' if 'initReg' is
 //                      set to non-zero value after this call.
 //
-// Return Value:
-//     None
-//
 void CodeGen::genProfilingEnterCallback(regNumber initReg, bool* pInitRegZeroed)
 {
     assert(compiler->compGeneratingProlog);
 
-    // Give profiler a chance to back out of hooking this method
     if (!compiler->compIsProfilerHookNeeded())
     {
         return;
     }
+
+    ssize_t methHnd = (ssize_t)compiler->compProfilerMethHnd;
+    if (compiler->compProfilerMethHndIndirected)
+    {
+        instGen_Set_Reg_To_Imm(EA_PTR_DSP_RELOC, REG_PROFILER_ENTER_ARG_FUNC_ID, methHnd);
+        GetEmitter()->emitIns_R_R_I(INS_ld, EA_PTRSIZE, REG_PROFILER_ENTER_ARG_FUNC_ID, REG_PROFILER_ENTER_ARG_FUNC_ID,
+                                    0);
+    }
+    else
+    {
+        instGen_Set_Reg_To_Imm(EA_PTRSIZE, REG_PROFILER_ENTER_ARG_FUNC_ID, methHnd);
+    }
+
+    ssize_t callerSPOffset = -compiler->lvaToCallerSPRelativeOffset(0, isFramePointerUsed());
+    genInstrWithConstant(INS_addi, EA_PTRSIZE, REG_PROFILER_ENTER_ARG_CALLER_SP, genFramePointerReg(), callerSPOffset,
+                         REG_PROFILER_ENTER_ARG_CALLER_SP);
+
+    genEmitHelperCall(CORINFO_HELP_PROF_FCN_ENTER, 0, EA_UNKNOWN);
+
+    if ((genRegMask(initReg) & RBM_PROFILER_ENTER_TRASH))
+    {
+        *pInitRegZeroed = false;
+    }
 }
+
+//-----------------------------------------------------------------------------------
+// genProfilingLeaveCallback: Generate the profiling function leave or tailcall callback.
+// Technically, this is not part of the epilog; it is called when we are generating code for a GT_RETURN node.
+//
+// Arguments:
+//     helper - which helper to call. Either CORINFO_HELP_PROF_FCN_LEAVE or CORINFO_HELP_PROF_FCN_TAILCALL
+//
+void CodeGen::genProfilingLeaveCallback(unsigned helper /*= CORINFO_HELP_PROF_FCN_LEAVE*/)
+{
+    assert((helper == CORINFO_HELP_PROF_FCN_LEAVE) || (helper == CORINFO_HELP_PROF_FCN_TAILCALL));
+
+    if (!compiler->compIsProfilerHookNeeded())
+    {
+        return;
+    }
+
+    compiler->info.compProfilerCallback = true;
+
+    ssize_t methHnd = (ssize_t)compiler->compProfilerMethHnd;
+    if (compiler->compProfilerMethHndIndirected)
+    {
+        instGen_Set_Reg_To_Imm(EA_PTR_DSP_RELOC, REG_PROFILER_LEAVE_ARG_FUNC_ID, methHnd);
+        GetEmitter()->emitIns_R_R_I(INS_ld, EA_PTRSIZE, REG_PROFILER_LEAVE_ARG_FUNC_ID, REG_PROFILER_LEAVE_ARG_FUNC_ID,
+                                    0);
+    }
+    else
+    {
+        instGen_Set_Reg_To_Imm(EA_PTRSIZE, REG_PROFILER_LEAVE_ARG_FUNC_ID, methHnd);
+    }
+
+    gcInfo.gcMarkRegSetNpt(RBM_PROFILER_LEAVE_ARG_FUNC_ID);
+
+    ssize_t callerSPOffset = -compiler->lvaToCallerSPRelativeOffset(0, isFramePointerUsed());
+    genInstrWithConstant(INS_addi, EA_PTRSIZE, REG_PROFILER_LEAVE_ARG_CALLER_SP, genFramePointerReg(), callerSPOffset,
+                         REG_PROFILER_LEAVE_ARG_CALLER_SP);
+
+    gcInfo.gcMarkRegSetNpt(RBM_PROFILER_LEAVE_ARG_CALLER_SP);
+
+    genEmitHelperCall(helper, 0, EA_UNKNOWN);
+}
+#endif // PROFILING_SUPPORTED
+
 #endif // TARGET_RISCV64

src/coreclr/jit/targetriscv64.h

@@ -192,18 +192,18 @@
   #define REG_PREV(reg)           ((regNumber)((unsigned)(reg) - 1))
 
   // The following registers are used in emitting Enter/Leave/Tailcall profiler callbacks
-  #define REG_PROFILER_ENTER_ARG_FUNC_ID    REG_R16
-  #define RBM_PROFILER_ENTER_ARG_FUNC_ID    RBM_R16
-  #define REG_PROFILER_ENTER_ARG_CALLER_SP  REG_R17
-  #define RBM_PROFILER_ENTER_ARG_CALLER_SP  RBM_R17
-  #define REG_PROFILER_LEAVE_ARG_FUNC_ID    REG_R16
-  #define RBM_PROFILER_LEAVE_ARG_FUNC_ID    RBM_R16
-  #define REG_PROFILER_LEAVE_ARG_CALLER_SP  REG_R17
-  #define RBM_PROFILER_LEAVE_ARG_CALLER_SP  RBM_R17
+  #define REG_PROFILER_ENTER_ARG_FUNC_ID    REG_T0
+  #define RBM_PROFILER_ENTER_ARG_FUNC_ID    RBM_T0
+  #define REG_PROFILER_ENTER_ARG_CALLER_SP  REG_T1
+  #define RBM_PROFILER_ENTER_ARG_CALLER_SP  RBM_T1
+  #define REG_PROFILER_LEAVE_ARG_FUNC_ID    REG_PROFILER_ENTER_ARG_FUNC_ID
+  #define RBM_PROFILER_LEAVE_ARG_FUNC_ID    RBM_PROFILER_ENTER_ARG_FUNC_ID
+  #define REG_PROFILER_LEAVE_ARG_CALLER_SP  REG_PROFILER_ENTER_ARG_CALLER_SP
+  #define RBM_PROFILER_LEAVE_ARG_CALLER_SP  RBM_PROFILER_ENTER_ARG_CALLER_SP
 
   // The registers trashed by profiler enter/leave/tailcall hook
   #define RBM_PROFILER_ENTER_TRASH     (RBM_CALLEE_TRASH & ~(RBM_ARG_REGS|RBM_FLTARG_REGS|RBM_FP))
-  #define RBM_PROFILER_LEAVE_TRASH     (RBM_CALLEE_TRASH & ~(RBM_ARG_REGS|RBM_FLTARG_REGS|RBM_FP))
+  #define RBM_PROFILER_LEAVE_TRASH     RBM_PROFILER_ENTER_TRASH
   #define RBM_PROFILER_TAILCALL_TRASH  RBM_PROFILER_LEAVE_TRASH
 
   // Which register are int and long values returned in ?

src/coreclr/vm/proftoeeinterfaceimpl.h

@@ -56,14 +56,18 @@ class ProfileArgIterator
 private:
     void        *m_handle;
     ArgIterator  m_argIterator;
-#if defined(UNIX_AMD64_ABI) || defined(TARGET_ARM64)
+#if defined(UNIX_AMD64_ABI) || defined(TARGET_ARM64) || defined(TARGET_RISCV64)
     UINT64       m_bufferPos;
 
-#if defined(UNIX_AMD64_ABI)
+#if defined(UNIX_AMD64_ABI) || defined(TARGET_RISCV64)
     // On certain architectures we can pass args in non-sequential registers,
     // this function will copy the struct so it is laid out as it would be in memory
     // so it can be passed to the profiler
-    LPVOID CopyStructFromRegisters();
+    LPVOID CopyStructFromRegisters(
+#ifdef TARGET_RISCV64
+        const ArgLocDesc* argLocDesc
+#endif
+    );
 #endif
 
 #if defined(TARGET_ARM64)

src/coreclr/vm/riscv64/asmconstants.h

@@ -250,5 +250,34 @@ ASMCONSTANTS_C_ASSERT(CallCountingStubData__TargetForMethod == offsetof(CallCoun
 #define CallCountingStubData__TargetForThresholdReached 0x10
 ASMCONSTANTS_C_ASSERT(CallCountingStubData__TargetForThresholdReached == offsetof(CallCountingStubData, TargetForThresholdReached))
 
+#ifdef PROFILING_SUPPORTED
+#define PROFILE_ENTER    1
+#define PROFILE_LEAVE    2
+#define PROFILE_TAILCALL 4
+
+#define SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA 312
+#define PROFILE_PLATFORM_SPECIFIC_DATA__argumentRegisters 16
+#define PROFILE_PLATFORM_SPECIFIC_DATA__functionId 80
+#define PROFILE_PLATFORM_SPECIFIC_DATA__floatArgumentRegisters 88
+#define PROFILE_PLATFORM_SPECIFIC_DATA__probeSp 152
+#define PROFILE_PLATFORM_SPECIFIC_DATA__profiledSp 160
+#define PROFILE_PLATFORM_SPECIFIC_DATA__hiddenArg 168
+#define PROFILE_PLATFORM_SPECIFIC_DATA__flags 176
+
+ASMCONSTANTS_C_ASSERT(SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA == sizeof(PROFILE_PLATFORM_SPECIFIC_DATA))
+
+#define ASMCONSTANTS_C_ASSERT_OFFSET(type, field) \
+    ASMCONSTANTS_C_ASSERT(type##__##field == offsetof(type, field))
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, argumentRegisters)
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, functionId)
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, floatArgumentRegisters)
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, probeSp)
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, profiledSp)
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, hiddenArg)
+ASMCONSTANTS_C_ASSERT_OFFSET(PROFILE_PLATFORM_SPECIFIC_DATA, flags)
+#undef ASMCONSTANTS_C_ASSERT_OFFSET
+
+#endif // PROFILING_SUPPORTED
+
 #undef ASMCONSTANTS_RUNTIME_ASSERT
 #undef ASMCONSTANTS_C_ASSERT

src/coreclr/vm/riscv64/asmhelpers.S

@@ -894,43 +894,36 @@ LEAF_ENTRY JIT_ProfilerEnterLeaveTailcallStub, _TEXT
     ret
 LEAF_END JIT_ProfilerEnterLeaveTailcallStub, _TEXT
 
-// ------------------------------------------------------------------
-#define PROFILE_ENTER    1
-#define PROFILE_LEAVE    2
-#define PROFILE_TAILCALL 4
-#define SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA 272
-
 // ------------------------------------------------------------------
 .macro GenerateProfileHelper helper, flags
 NESTED_ENTRY \helper\()Naked, _TEXT, NoHandler
     // On entry:
-    //   t1 = functionIDOrClientID
-    //   t2 = profiledSp
+    //   t0 = functionIDOrClientID
+    //   t1 = profiledSp
     //   t6 = throwable
     //
     // On exit:
     //   Values of a0-a7, fa0-fa7, fp are preserved.
     //   Values of other volatile registers are not preserved.
 
+    // Fill in PROFILE_PLATFORM_SPECIFIC_DATA struct
     PROLOG_SAVE_REG_PAIR_INDEXED fp, ra, SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA // Allocate space and save Fp, Pc.
-    SAVE_ARGUMENT_REGISTERS sp, 16          // Save t0 and argument registers (a0-a7).
-    sd     zero, 88(sp)                     // Clear functionId.
-    SAVE_FLOAT_ARGUMENT_REGISTERS sp, 96    // Save floating-point/SIMD registers (fa0-fa7).
-    addi     t6, fp, SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA // Compute probeSp - initial value of Sp on entry to the helper.
-    sd  t6, 224(sp)            // Save probeSp.
-    sd  t2, 232(sp)            // Save profiledSp.
-
-    sd    zero, 240(sp)        // Clear hiddenArg.
+    SAVE_ARGUMENT_REGISTERS sp, PROFILE_PLATFORM_SPECIFIC_DATA__argumentRegisters
+    sd     zero, PROFILE_PLATFORM_SPECIFIC_DATA__functionId(sp)
+    SAVE_FLOAT_ARGUMENT_REGISTERS sp, PROFILE_PLATFORM_SPECIFIC_DATA__floatArgumentRegisters
+    addi     t6, sp, SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA // Compute probeSp - initial value of Sp on entry to the helper.
+    sd  t6, PROFILE_PLATFORM_SPECIFIC_DATA__probeSp(sp)
+    sd  t1, PROFILE_PLATFORM_SPECIFIC_DATA__profiledSp(sp)
+    sd  zero, PROFILE_PLATFORM_SPECIFIC_DATA__hiddenArg(sp)
     addi  t6, zero, \flags
-    sw  t6, 248(sp)            // Save flags.
-    sw  zero, 252(sp)          // clear unused field.
+    sd  t6, PROFILE_PLATFORM_SPECIFIC_DATA__flags(sp)
 
-    addi  a1, t1, 0
-    addi  a2, sp, 0
+    addi  a0, t0, 0
+    addi  a1, sp, 0
     call  C_FUNC(\helper)
 
-    RESTORE_ARGUMENT_REGISTERS sp, 16       // Restore t0 and argument registers.
-    RESTORE_FLOAT_ARGUMENT_REGISTERS sp, 96 // Restore floating-point/SIMD registers.
+    RESTORE_ARGUMENT_REGISTERS sp, PROFILE_PLATFORM_SPECIFIC_DATA__argumentRegisters
+    RESTORE_FLOAT_ARGUMENT_REGISTERS sp, PROFILE_PLATFORM_SPECIFIC_DATA__floatArgumentRegisters
     EPILOG_RESTORE_REG_PAIR_INDEXED fp, ra, SIZEOF__PROFILE_PLATFORM_SPECIFIC_DATA
     EPILOG_RETURN
 

src/coreclr/vm/riscv64/cgencpu.h

@@ -107,7 +107,7 @@ struct CalleeSavedRegisters {
 #define NUM_ARGUMENT_REGISTERS 8
 typedef DPTR(struct ArgumentRegisters) PTR_ArgumentRegisters;
 struct ArgumentRegisters {
-    INT64 a[8]; // a0 ....a7
+    INT64 a[NUM_ARGUMENT_REGISTERS]; // a0 ....a7
 };
 
 #define ARGUMENTREGISTERS_SIZE sizeof(ArgumentRegisters)
@@ -124,9 +124,30 @@ struct ArgumentRegisters {
 typedef DPTR(struct FloatArgumentRegisters) PTR_FloatArgumentRegisters;
 struct FloatArgumentRegisters {
     //TODO: not supports RISCV64-SIMD.
-    double  f[8];  // f0-f7
+    double  f[NUM_FLOAT_ARGUMENT_REGISTERS];  // f0-f7
 };
 
+//**********************************************************************
+// Profiling
+//**********************************************************************
+
+#ifdef PROFILING_SUPPORTED
+
+struct PROFILE_PLATFORM_SPECIFIC_DATA
+{
+    void*                  Fp;
+    void*                  Pc;
+    ArgumentRegisters      argumentRegisters;
+    FunctionID             functionId;
+    FloatArgumentRegisters floatArgumentRegisters;
+    void*                  probeSp;
+    void*                  profiledSp;
+    void*                  hiddenArg;
+    UINT64                 flags;
+    // Scratch space to reconstruct struct passed in registers
+    BYTE                   buffer[sizeof(ArgumentRegisters) + sizeof(FloatArgumentRegisters)];
+};
+#endif  // PROFILING_SUPPORTED
 
 //**********************************************************************
 // Exception handling
@@ -359,8 +380,6 @@ class StubLinkerCPU : public StubLinker
     void EmitLoad(FloatReg dest, IntReg srcAddr, int offset = 0);
     void EmitStore(IntReg src, IntReg destAddr, int offset = 0);
     void EmitStore(FloatReg src, IntReg destAddr, int offset = 0);
-
-    void EmitRet(IntReg reg);
 };
 
 extern "C" void SinglecastDelegateInvokeStub();