runtime: remove unnecessary NOFRAME flags on windows

This CL removes some NOFRAME flags on Windows assembly files
for several reasons:

- windows/386 does not use a frame pointer
- Leaf frameless functions already skip the frame pointer
- Some non-leaf functions do not contain enough dragons to justify
not using the frame pointer

Updates #58378

Change-Id: I31e71bf7f769e1957a4adba91778da5af66ce1e4
Reviewed-on: https://go-review.googlesource.com/c/go/+/466835
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Quim Muntal <quimmuntal@gmail.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>

diff --git a/src/runtime/rt0_windows_amd64.s b/src/runtime/rt0_windows_amd64.s
index 9c60337ddc1d54ae797b9824be077995d2a4af55..d5f0940540b0c72d2038d4f15d9b026e93d7584c 100644 (file)
--- a/src/runtime/rt0_windows_amd64.s
+++ b/src/runtime/rt0_windows_amd64.s
@@ -6,7 +6,7 @@
 #include "go_tls.h"
 #include "textflag.h"
 
-TEXT _rt0_amd64_windows(SB),NOSPLIT,$-8
+TEXT _rt0_amd64_windows(SB),NOSPLIT|NOFRAME,$-8
        JMP     _rt0_amd64(SB)
 
 // When building with -buildmode=(c-shared or c-archive), this
@@ -24,7 +24,7 @@ TEXT _rt0_amd64_windows_lib(SB),NOSPLIT|NOFRAME,$0x20
        CALL    AX
        RET
 
-TEXT _rt0_amd64_windows_lib_go(SB),NOSPLIT,$0
+TEXT _rt0_amd64_windows_lib_go(SB),NOSPLIT|NOFRAME,$0
        MOVQ  $0, DI
        MOVQ    $0, SI
        MOVQ    $runtime·rt0_go(SB), AX

diff --git a/src/runtime/rt0_windows_arm64.s b/src/runtime/rt0_windows_arm64.s
index bad85c28ac5ce9f421693be2a953d8ff260c0aac..8802c2b82e0afdde8668eb78edd7ad97003661f3 100644 (file)
--- a/src/runtime/rt0_windows_arm64.s
+++ b/src/runtime/rt0_windows_arm64.s
@@ -23,7 +23,7 @@ TEXT _rt0_arm64_windows_lib_go(SB),NOSPLIT|NOFRAME,$0
        MOVD    $runtime·rt0_go(SB), R2
        B       (R2)
 
-TEXT main(SB),NOSPLIT,$0
+TEXT main(SB),NOSPLIT|NOFRAME,$0
        MOVD    $runtime·rt0_go(SB), R2
        B       (R2)
 

diff --git a/src/runtime/sys_windows_386.s b/src/runtime/sys_windows_386.s
index d29049899e8f8c63e981eabce0b4f92609951ea8..818f1b4d7982c3787b16a651a9005e7b1b0a5444 100644 (file)
--- a/src/runtime/sys_windows_386.s
+++ b/src/runtime/sys_windows_386.s
@@ -50,7 +50,7 @@ TEXT runtime·getlasterror(SB),NOSPLIT,$0
        MOVL    AX, ret+0(FP)
        RET
 
-TEXT runtime·sigFetchGSafe<ABIInternal>(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·sigFetchGSafe<ABIInternal>(SB),NOSPLIT,$0
        get_tls(AX)
        CMPL    AX, $0
        JE      2(PC)
@@ -93,7 +93,7 @@ TEXT sigtramp<>(SB),NOSPLIT,$0-0
 // It switches stacks and jumps to the continuation address.
 // DX and CX are set above at the end of sigtrampgo
 // in the context that starts executing at sigresume.
-TEXT runtime·sigresume(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·sigresume(SB),NOSPLIT,$0
        MOVL    DX, SP
        JMP     CX
 
@@ -311,7 +311,7 @@ useQPC:
 
 // This is called from rt0_go, which runs on the system stack
 // using the initial stack allocated by the OS.
-TEXT runtime·wintls(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·wintls(SB),NOSPLIT,$0
        // Allocate a TLS slot to hold g across calls to external code
        MOVL    SP, BP
        MOVL    runtime·_TlsAlloc(SB), AX

diff --git a/src/runtime/sys_windows_amd64.s b/src/runtime/sys_windows_amd64.s
index 8780c45b1aa02ab9162258e9e6237fc2506da54d..5eb03b014ed37c5927f63493f1c551ed8ff1cfb5 100644 (file)
--- a/src/runtime/sys_windows_amd64.s
+++ b/src/runtime/sys_windows_amd64.s
@@ -12,9 +12,12 @@
 #define TEB_TlsSlots 0x1480
 
 // void runtime·asmstdcall(void *c);
-TEXT runtime·asmstdcall(SB),NOSPLIT|NOFRAME,$0
-       // asmcgocall will put first argument into CX.
-       PUSHQ   CX                      // save for later
+TEXT runtime·asmstdcall(SB),NOSPLIT,$16
+       MOVQ    SP, AX
+       ANDQ    $~15, SP        // alignment as per Windows requirement
+       MOVQ    AX, 8(SP)
+       MOVQ    CX, 0(SP)       // asmcgocall will put first argument into CX.
+
        MOVQ    libcall_fn(CX), AX
        MOVQ    libcall_args(CX), SI
        MOVQ    libcall_n(CX), CX
@@ -61,7 +64,8 @@ loadregs:
        ADDQ    $(const_maxArgs*8), SP
 
        // Return result.
-       POPQ    CX
+       MOVQ    0(SP), CX
+       MOVQ    8(SP), SP
        MOVQ    AX, libcall_r1(CX)
        // Floating point return values are returned in XMM0. Setting r2 to this
        // value in case this call returned a floating point value. For details,
@@ -233,7 +237,7 @@ TEXT runtime·settls(SB),NOSPLIT,$0
 // g may be nil.
 // The function leaves room for 4 syscall parameters
 // (as per windows amd64 calling convention).
-TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$48-4
+TEXT runtime·usleep2(SB),NOSPLIT,$48-4
        MOVLQSX dt+0(FP), BX
        MOVQ    SP, AX
        ANDQ    $~15, SP        // alignment as per Windows requirement
@@ -249,7 +253,7 @@ TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$48-4
 
 // Runs on OS stack. duration (in -100ns units) is in dt+0(FP).
 // g is valid.
-TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$72-4
+TEXT runtime·usleep2HighRes(SB),NOSPLIT,$72-4
        MOVLQSX dt+0(FP), BX
        get_tls(CX)
 
@@ -281,7 +285,7 @@ TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$72-4
        RET
 
 // Runs on OS stack.
-TEXT runtime·switchtothread(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·switchtothread(SB),NOSPLIT,$0
        MOVQ    SP, AX
        ANDQ    $~15, SP        // alignment as per Windows requirement
        SUBQ    $(48), SP       // room for SP and 4 args as per Windows requirement
@@ -306,7 +310,7 @@ useQPC:
 
 // func osSetupTLS(mp *m)
 // Setup TLS. for use by needm on Windows.
-TEXT runtime·osSetupTLS(SB),NOSPLIT|NOFRAME,$0-8
+TEXT runtime·osSetupTLS(SB),NOSPLIT,$0-8
        MOVQ    mp+0(FP), AX
        LEAQ    m_tls(AX), DI
        CALL    runtime·settls(SB)
@@ -314,7 +318,7 @@ TEXT runtime·osSetupTLS(SB),NOSPLIT|NOFRAME,$0-8
 
 // This is called from rt0_go, which runs on the system stack
 // using the initial stack allocated by the OS.
-TEXT runtime·wintls(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·wintls(SB),NOSPLIT,$0
        // Allocate a TLS slot to hold g across calls to external code
        MOVQ    SP, AX
        ANDQ    $~15, SP        // alignment as per Windows requirement

diff --git a/src/runtime/sys_windows_arm.s b/src/runtime/sys_windows_arm.s
index f6496b8d054db690f38f5d12a1a9ff156f18e07a..0af19db352ceabb5d034d22696a95c2057910809 100644 (file)
--- a/src/runtime/sys_windows_arm.s
+++ b/src/runtime/sys_windows_arm.s
@@ -232,12 +232,12 @@ TEXT ·publicationBarrier(SB),NOSPLIT|NOFRAME,$0-0
        B       runtime·armPublicationBarrier(SB)
 
 // never called (this is a GOARM=7 platform)
-TEXT runtime·read_tls_fallback(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·read_tls_fallback(SB),NOSPLIT,$0
        MOVW    $0xabcd, R0
        MOVW    R0, (R0)
        RET
 
-TEXT runtime·nanotime1(SB),NOSPLIT|NOFRAME,$0-8
+TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
        MOVW    $0, R0
        MOVB    runtime·useQPCTime(SB), R0
        CMP     $0, R0
@@ -262,7 +262,7 @@ loop:
        MOVW    R4, ret_hi+4(FP)
        RET
 useQPC:
-       B       runtime·nanotimeQPC(SB)                // tail call
+       RET     runtime·nanotimeQPC(SB)                // tail call
 
 // save_g saves the g register (R10) into thread local memory
 // so that we can call externally compiled
@@ -273,7 +273,7 @@ useQPC:
 // Save the value in the _TEB->TlsSlots array.
 // Effectively implements TlsSetValue().
 // tls_g stores the TLS slot allocated TlsAlloc().
-TEXT runtime·save_g(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·save_g(SB),NOSPLIT,$0
        MRC     15, 0, R0, C13, C0, 2
        ADD     $0xe10, R0
        MOVW    $runtime·tls_g(SB), R11
@@ -287,7 +287,7 @@ TEXT runtime·save_g(SB),NOSPLIT|NOFRAME,$0
 // ARM code that overwrote those registers.
 // Get the value from the _TEB->TlsSlots array.
 // Effectively implements TlsGetValue().
-TEXT runtime·load_g(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·load_g(SB),NOSPLIT,$0
        MRC     15, 0, R0, C13, C0, 2
        ADD     $0xe10, R0
        MOVW    $runtime·tls_g(SB), g

diff --git a/src/runtime/sys_windows_arm64.s b/src/runtime/sys_windows_arm64.s
index c78d37314ac3278ca3c2cf466d2acff7821f6bb7..6da866ac889a7884f006ed3480a09105ae9d2ddd 100644 (file)
--- a/src/runtime/sys_windows_arm64.s
+++ b/src/runtime/sys_windows_arm64.s
@@ -19,8 +19,7 @@
 // load_g and save_g (in tls_arm64.s) clobber R27 (REGTMP) and R0.
 
 // void runtime·asmstdcall(void *c);
-TEXT runtime·asmstdcall(SB),NOSPLIT|NOFRAME,$0
-       STP.W   (R29, R30), -32(RSP)    // allocate C ABI stack frame
+TEXT runtime·asmstdcall(SB),NOSPLIT,$16
        STP     (R19, R20), 16(RSP) // save old R19, R20
        MOVD    R0, R19 // save libcall pointer
        MOVD    RSP, R20        // save stack pointer
@@ -96,10 +95,9 @@ _0args:
 
        // Restore callee-saved registers.
        LDP     16(RSP), (R19, R20)
-       LDP.P   32(RSP), (R29, R30)
        RET
 
-TEXT runtime·getlasterror(SB),NOSPLIT|NOFRAME,$0
+TEXT runtime·getlasterror(SB),NOSPLIT,$0
        MOVD    TEB_error(R18_PLATFORM), R0
        MOVD    R0, ret+0(FP)
        RET
@@ -245,7 +243,7 @@ TEXT runtime·usleep2(SB),NOSPLIT,$32-4
 // duration (in -100ns units) is in dt+0(FP).
 // g is valid.
 // TODO: needs to be implemented properly.
-TEXT runtime·usleep2HighRes(SB),NOSPLIT,$0-4
+TEXT runtime·usleep2HighRes(SB),NOSPLIT|NOFRAME,$0-4
        B       runtime·abort(SB)
 
 // Runs on OS stack.
@@ -256,7 +254,7 @@ TEXT runtime·switchtothread(SB),NOSPLIT,$16-0
        ADD     $16, RSP
        RET
 
-TEXT runtime·nanotime1(SB),NOSPLIT|NOFRAME,$0-8
+TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
        MOVB    runtime·useQPCTime(SB), R0
        CMP     $0, R0
        BNE     useQPC
@@ -267,7 +265,7 @@ TEXT runtime·nanotime1(SB),NOSPLIT|NOFRAME,$0-8
        MOVD    R0, ret+0(FP)
        RET
 useQPC:
-       B       runtime·nanotimeQPC(SB)                // tail call
+       RET     runtime·nanotimeQPC(SB)                // tail call
 
 // This is called from rt0_go, which runs on the system stack
 // using the initial stack allocated by the OS.

diff --git a/src/runtime/time_windows_arm.s b/src/runtime/time_windows_arm.s
index 711af88307207355846c5093555f8f20070d65fa..8d4469f9933d01565df80991aa0db2d57be41ba5 100644 (file)
--- a/src/runtime/time_windows_arm.s
+++ b/src/runtime/time_windows_arm.s
@@ -8,7 +8,7 @@
 #include "textflag.h"
 #include "time_windows.h"
 
-TEXT time·now(SB),NOSPLIT|NOFRAME,$0-20
+TEXT time·now(SB),NOSPLIT,$0-20
        MOVW    $0, R0
        MOVB    runtime·useQPCTime(SB), R0
        CMP     $0, R0
@@ -86,5 +86,5 @@ wall:
        MOVW    R1,nsec+8(FP)
        RET
 useQPC:
-       B       runtime·nowQPC(SB)             // tail call
+       RET     runtime·nowQPC(SB)             // tail call
 

diff --git a/src/runtime/time_windows_arm64.s b/src/runtime/time_windows_arm64.s
index e0c7d28e153f72912fdc4221bf533412ef278b29..7943d6b46dcbcce34e4f4a5d64f7b597b157c151 100644 (file)
--- a/src/runtime/time_windows_arm64.s
+++ b/src/runtime/time_windows_arm64.s
@@ -8,7 +8,7 @@
 #include "textflag.h"
 #include "time_windows.h"
 
-TEXT time·now(SB),NOSPLIT|NOFRAME,$0-24
+TEXT time·now(SB),NOSPLIT,$0-24
        MOVB    runtime·useQPCTime(SB), R0
        CMP     $0, R0
        BNE     useQPC
@@ -43,5 +43,5 @@ TEXT time·now(SB),NOSPLIT|NOFRAME,$0-24
        MOVW    R0, nsec+8(FP)
        RET
 useQPC:
-       B       runtime·nowQPC(SB)             // tail call
+       RET     runtime·nowQPC(SB)             // tail call