runtime: support -buildmode=c-shared on linux/mips64le

For golang#43264

Signed-off-by: Xiaodong Liu <liuxiaodong@loongson.cn>

src/runtime/asm_mips64x.s

@@ -446,10 +446,12 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	MOVV	m_g0(R5), R6
 	BEQ	R6, g, g0
 
+	MOVV    R25, R15 // call to save_g clobbers R25 when buildmode=c-shared 
 	JAL	gosave_systemstack_switch<>(SB)
 	MOVV	R6, g
 	JAL	runtime·save_g(SB)
 	MOVV	(g_sched+gobuf_sp)(g), R29
+	MOVV    R15, R25
 
 	// Now on a scheduling stack (a pthread-created stack).
 g0:

src/runtime/cgo/asm_mips64x.s

@@ -47,11 +47,17 @@ TEXT crosscall2(SB),NOSPLIT|NOFRAME,$0
 	MOVD	F30, (8*21)(R29)
 	MOVD	F31, (8*22)(R29)
 #endif
+	MOVV    0(R25), R14
+	SRLV    $16, R14, R14
+	MOVV    $0x3c1c, R15  //crosscall2 begain with lui instruction under c-shared mode
+	AND     R14, R15, R14
+	BEQ     R14, R15, shared
 	// Initialize Go ABI environment
 	// prepare SB register = PC & 0xffffffff00000000
 	BGEZAL	R0, 1(PC)
 	SRLV	$32, R31, RSB
 	SLLV	$32, RSB
+shared:
 	JAL	runtime·load_g(SB)
 
 	JAL	runtime·cgocallback(SB)

src/runtime/cgo/gcc_mips64x.S

@@ -15,9 +15,9 @@
 .set noat
 crosscall1:
 #ifndef __mips_soft_float
-	daddiu	$29, $29, -160
+	daddiu	$29, $29, -168
 #else
-	daddiu	$29, $29, -96 // For soft-float, no need to make room for FP registers
+	daddiu	$29, $29, -104 // For soft-float, no need to make room for FP registers
 #endif
 	sd	$31, 0($29)
 	sd	$16, 8($29)
@@ -49,9 +49,21 @@ crosscall1:
 
 	move	$20, $4 // save R4
 	move	$1, $6
-	jalr	$5	// call setg_gcc (clobbers R4)
-	jalr	$20	// call fn
-
+	// buildmode=c-shared call function by jalr t9
+#ifndef __mips_soft_float
+	sd	$25, 152($29)
+#else
+	sd	$25, 88($29)
+#endif
+	move	$25, $5
+	jalr	$25	// call setg_gcc (clobbers R4)
+	move	$25, $20
+	jalr	$25	// call fn
+#ifndef __mips_soft_float
+	ld	$25, 152($29)
+#else
+	ld	$25, 88($29)
+#endif
 	ld	$16, 8($29)
 	ld	$17, 16($29)
 	ld	$18, 24($29)
@@ -74,9 +86,9 @@ crosscall1:
 #endif
 	ld	$31, 0($29)
 #ifndef __mips_soft_float
-	daddiu	$29, $29, 160
+	daddiu	$29, $29, 168
 #else
-	daddiu	$29, $29, 96
+	daddiu	$29, $29, 104
 #endif
 	jr	$31
 

src/runtime/rt0_linux_mips64x.s

@@ -27,6 +27,110 @@ TEXT _main<>(SB),NOSPLIT|NOFRAME,$0
 	ADDV	$8, R29, R5 // argv
 	JMP	main(SB)
 
+// When building with -buildmode=c-shared, this symbol is called when the shared
+// library is loaded.
+#ifdef GOMIPS64_hardfloat
+TEXT _rt0_mips64le_linux_lib(SB),NOSPLIT,$232
+#else
+TEXT _rt0_mips64le_linux_lib(SB),NOSPLIT,$136
+#endif
+	// Preserve callee-save registers.
+	MOVV R16, 24(R29)
+	MOVV R17, 32(R29)
+	MOVV R18, 40(R29)
+	MOVV R19, 48(R29)
+	MOVV R20, 56(R29)
+	MOVV R21, 64(R29)
+	MOVV R22, 72(R29)
+	MOVV R23, 80(R29)
+	MOVV RSB, 88(R29)
+	MOVV R29, 96(R29)
+	MOVV g,   104(R29)
+	MOVV R31, 112(R29)
+#ifdef GOMIPS64_hardfloat
+	MOVF F20, 120(R29)
+	MOVF F21, 128(R29)
+	MOVF F22, 136(R29)
+	MOVF F23, 144(R29)
+	MOVF F24, 152(R29)
+	MOVF F25, 160(R29)
+	MOVF F26, 168(R29)
+	MOVF F27, 176(R29)
+	MOVF F28, 184(R29)
+	MOVF F29, 192(R29)
+	MOVF F30, 200(R29)
+	MOVF F31, 208(R29)
+#endif
+
+	// Initialize g as null in case of using g later e.g. sigaction in cgo_sigaction.go
+	MOVV	R0, g
+
+	MOVV	R4, _rt0_mips64le_linux_lib_argc<>(SB)
+	MOVV	R5, _rt0_mips64le_linux_lib_argv<>(SB)
+
+	// Synchronous initialization.
+	MOVV	$runtime·libpreinit(SB), R1
+	JAL	(R1)
+
+	// Create a new thread to do the runtime initialization and return.
+	MOVV	_cgo_sys_thread_create(SB), R1
+	BEQ	R1, nocgo
+	MOVV	$_rt0_mips64le_linux_lib_go(SB), R4
+	MOVV	$0, R5
+//	ADDV	$-16, R29		// reserve 16 bytes for sp-8 where fp may be saved.
+	JAL	(R1)
+//	ADDV	$16, R29
+	JMP	restore
+
+nocgo:
+	MOVV	$0x800000, R4                     // stacksize = 8192KB
+	MOVV	$_rt0_mips64le_linux_lib_go(SB), R5
+	MOVV	R4, 8(R29)
+	MOVV	R5, 16(R29)
+	MOVV	$runtime·newosproc0(SB),R1
+	JAL	(R1)
+
+restore:
+	// Restore callee-save registers.
+	MOVV 24(R29), R16
+	MOVV 32(R29), R17
+	MOVV 40(R29), R18
+	MOVV 48(R29), R19
+	MOVV 56(R29), R20
+	MOVV 64(R29), R21
+	MOVV 72(R29), R22
+	MOVV 80(R29), R23
+	MOVV 88(R29), RSB
+	MOVV 96(R29), R29
+	MOVV 104(R29), g
+	MOVV 112(R29), R31
+#ifdef GOMIPS64_hardfloat
+	MOVF 120(R29), F20
+	MOVF 128(R29), F21
+	MOVF 136(R29), F22
+	MOVF 144(R29), F23
+	MOVF 152(R29), F24
+	MOVF 160(R29), F25
+	MOVF 168(R29), F26
+	MOVF 176(R29), F27
+	MOVF 184(R29), F28
+	MOVF 192(R29), F29
+	MOVF 200(R29), F30
+	MOVF 208(R29), F31
+#endif
+	RET
+
+TEXT _rt0_mips64le_linux_lib_go(SB),NOSPLIT,$0
+	MOVV	_rt0_mips64le_linux_lib_argc<>(SB), R4
+	MOVV	_rt0_mips64le_linux_lib_argv<>(SB), R5
+	MOVV	$runtime·rt0_go(SB),R1
+	JMP     (R1)
+
+DATA _rt0_mips64le_linux_lib_argc<>(SB)/8, $0
+GLOBL _rt0_mips64le_linux_lib_argc<>(SB),NOPTR, $8
+DATA _rt0_mips64le_linux_lib_argv<>(SB)/8, $0
+GLOBL _rt0_mips64le_linux_lib_argv<>(SB),NOPTR, $8
+
 TEXT main(SB),NOSPLIT|NOFRAME,$0
 	// in external linking, glibc jumps to main with argc in R4
 	// and argv in R5

src/runtime/tls_mips64x.s

@@ -17,15 +17,37 @@ TEXT runtime·save_g(SB),NOSPLIT|NOFRAME,$0-0
 	MOVB	runtime·iscgo(SB), R23
 	BEQ	R23, nocgo
 
+	// buildmode=c-shared clobbers R2,R4,R25
+	MOVV	R31,-32(R29)
+	ADDV	$-32, R29
+	MOVV	R2, 8(R29)
+	MOVV	R4, 16(R29)
+	MOVV	R25, 24(R29)
 	MOVV	R3, R23	// save R3
 	MOVV	g, runtime·tls_g(SB) // TLS relocation clobbers R3
 	MOVV	R23, R3	// restore R3
+	MOVV	8(R29),R2
+	MOVV	16(R29),R4
+	MOVV	24(R29),R25
+	MOVV	0(R29), R31
+	ADDV	$32, R29
 
 nocgo:
 	RET
 
 TEXT runtime·load_g(SB),NOSPLIT|NOFRAME,$0-0
+	// buildmode=c-shared clobbers R2,R4,R25
+        MOVV	R31,-32(R29)
+        ADDV	$-32, R29
+        MOVV	R2, 8(R29)
+        MOVV	R4, 16(R29)
+        MOVV	R25, 24(R29)
 	MOVV	runtime·tls_g(SB), g // TLS relocation clobbers R3
+	MOVV	8(R29),R2
+        MOVV	16(R29),R4
+        MOVV	24(R29),R25
+        MOVV	0(R29), R31
+        ADDV	$32, R29
 	RET
 
 GLOBL runtime·tls_g(SB), TLSBSS, $8