[LoongArch64] Enable TLS on linux/loongarch64 only for static resolver. (#106250)

Author: shushanhf

src/coreclr/vm/loongarch64/asmhelpers.S

@@ -1088,7 +1088,7 @@ GenerateProfileHelper ProfileTailcall, PROFILE_TAILCALL
 NESTED_ENTRY OnCallCountThresholdReachedStub, _TEXT, NoHandler
     PROLOG_WITH_TRANSITION_BLOCK
 
-    addi.d     $a0, $sp, __PWTB_TransitionBlock // TransitionBlock *
+    addi.d  $a0, $sp, __PWTB_TransitionBlock // TransitionBlock *
     ori     $a1, $t1, 0 // stub-identifying token
     bl      C_FUNC(OnCallCountThresholdReached)
     ori     $t4,$a0,0
@@ -1112,3 +1112,18 @@ LEAF_ENTRY GetThreadStaticsVariableOffset, _TEXT
         EPILOG_RETURN
 LEAF_END GetThreadStaticsVariableOffset, _TEXT
 // ------------------------------------------------------------------
+
+// ------------------------------------------------------------------
+// size_t GetTLSResolverAddress()
+
+// Helper to get the TLS resolver address. This will be then used to determine if we have a static or dynamic resolver.
+LEAF_ENTRY GetTLSResolverAddress, _TEXT
+        //                           $fp,$ra
+        PROLOG_SAVE_REG_PAIR_INDEXED  22, 1, 16
+        pcalau12i  $a0, %desc_pc_hi20(t_ThreadStatics)
+        addi.d     $a0, $a0, %desc_pc_lo12(t_ThreadStatics)
+        ld.d       $a0, $a0, %desc_ld(t_ThreadStatics)
+        EPILOG_RESTORE_REG_PAIR_INDEXED  22, 1, 16
+        EPILOG_RETURN
+LEAF_END GetTLSResolverAddress, _TEXT
+// ------------------------------------------------------------------

src/coreclr/vm/threadstatics.cpp

@@ -784,9 +784,9 @@ void FreeTLSIndicesForLoaderAllocator(LoaderAllocator *pLoaderAllocator)
 
 static void* GetTlsIndexObjectAddress();
 
-#if !defined(TARGET_OSX) && defined(TARGET_UNIX) && defined(TARGET_ARM64)
+#if !defined(TARGET_OSX) && defined(TARGET_UNIX) && (defined(TARGET_ARM64) || defined(TARGET_LOONGARCH64))
 extern "C" size_t GetTLSResolverAddress();
-#endif // !TARGET_OSX && TARGET_UNIX && TARGET_ARM64
+#endif // !TARGET_OSX && TARGET_UNIX && (TARGET_ARM64 || TARGET_LOONGARCH64)
 
 bool CanJITOptimizeTLSAccess()
 {
@@ -833,6 +833,28 @@ bool CanJITOptimizeTLSAccess()
         }
 #endif
     }
+#elif defined(TARGET_LOONGARCH64)
+    // Optimization is enabled for linux/loongarch64 only for static resolver.
+    // For static resolver, the TP offset is same for all threads.
+    // For dynamic resolver, TP offset returned is for the current thread and
+    // will be different for the other threads.
+    uint32_t* resolverAddress = reinterpret_cast<uint32_t*>(GetTLSResolverAddress());
+
+    if (
+        // ld.d a0, a0, 8
+        (resolverAddress[0] == 0x28c02084) &&
+        // ret
+        (resolverAddress[1] == 0x4c000020)
+    )
+    {
+        optimizeThreadStaticAccess = true;
+#ifdef _DEBUG
+        if (CLRConfig::GetConfigValue(CLRConfig::EXTERNAL_AssertNotStaticTlsResolver) != 0)
+        {
+            _ASSERTE(!"Detected static resolver in use when not expected");
+        }
+#endif
+    }
 #else
     optimizeThreadStaticAccess = true;
 #if !defined(TARGET_OSX) && defined(TARGET_UNIX) && defined(TARGET_AMD64)