[XRay][compiler-rt] Cleanup CFI/CFA annotations on trampolines

Summary:
This is a follow-up to D32202.

While the previous change (D32202) did fix the stack alignment issue, we
were still at a weird state in terms of the CFI/CFA directives (as the
offsets were wrong). This change cleans up the SAVE/RESTORE macros for
the trampoline, accounting the stack pointer adjustments with less
instructions and with some clearer math. We note that the offsets will
be different on the exit trampolines, because we don't typically 'call'
into this trampoline and we only ever jump into them (i.e. treated as a
tail call that's patched in at runtime).

Reviewers: eugenis, kpw, pelikan

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D32214

llvm-svn: 300815

1 files changed, 14 insertions, 19 deletions

diff --git a/compiler-rt/lib/xray/xray_trampoline_x86_64.S b/compiler-rt/lib/xray/xray_trampoline_x86_64.S
index 772eafbecca..847ecef8d42 100644
--- a/compiler-rt/lib/xray/xray_trampoline_x86_64.S
+++ b/compiler-rt/lib/xray/xray_trampoline_x86_64.S
@@ -16,7 +16,12 @@
 #include "../builtins/assembly.h"
 
 .macro SAVE_REGISTERS
-	subq $184, %rsp
+	subq $192, %rsp
+	.cfi_def_cfa_offset 200
+	// At this point, the stack pointer should be aligned to an 8-byte boundary,
+	// because any call instructions that come after this will add another 8
+	// bytes and therefore align it to 16-bytes.
+	movq %rbp, 184(%rsp)
 	movupd	%xmm0, 168(%rsp)
 	movupd	%xmm1, 152(%rsp)
 	movupd	%xmm2, 136(%rsp)
@@ -35,6 +40,7 @@
 .endm
 
 .macro RESTORE_REGISTERS
+	movq  184(%rsp), %rbp
 	movupd	168(%rsp), %xmm0
 	movupd	152(%rsp), %xmm1
 	movupd	136(%rsp), %xmm2
@@ -50,7 +56,8 @@
 	movq	16(%rsp), %rcx
 	movq	8(%rsp), %r8
 	movq	0(%rsp), %r9
-	addq	$184, %rsp
+	addq	$192, %rsp
+	.cfi_def_cfa_offset 8
 .endm
 
 	.text
@@ -64,10 +71,7 @@
 
 __xray_FunctionEntry:
 	.cfi_startproc
-	pushq %rbp
-	.cfi_def_cfa_offset 16
 	SAVE_REGISTERS
-	.cfi_def_cfa_offset 200
 
 	// This load has to be atomic, it's concurrent with __xray_patch().
 	// On x86/amd64, a simple (type-aligned) MOV instruction is enough.
@@ -81,7 +85,6 @@ __xray_FunctionEntry:
 	callq	*%rax
 .Ltmp0:
 	RESTORE_REGISTERS
-	popq	%rbp
 	retq
 .Ltmp1:
 	.size __xray_FunctionEntry, .Ltmp1-__xray_FunctionEntry
@@ -97,10 +100,9 @@ __xray_FunctionExit:
 	// Save the important registers first. Since we're assuming that this
 	// function is only jumped into, we only preserve the registers for
 	// returning.
-	pushq	%rbp
-	.cfi_def_cfa_offset 16
-	subq	$48, %rsp
+	subq	$56, %rsp
 	.cfi_def_cfa_offset 64
+	movq  %rbp, 48(%rsp)
 	movupd	%xmm0, 32(%rsp)
 	movupd	%xmm1, 16(%rsp)
 	movq	%rax, 8(%rsp)
@@ -114,12 +116,13 @@ __xray_FunctionExit:
 	callq	*%rax
 .Ltmp2:
 	// Restore the important registers.
+	movq  48(%rsp), %rbp
 	movupd	32(%rsp), %xmm0
 	movupd	16(%rsp), %xmm1
 	movq	8(%rsp), %rax
 	movq	0(%rsp), %rdx
-	addq	$48, %rsp
-	popq	%rbp
+	addq	$56, %rsp
+	.cfi_def_cfa_offset 8
 	retq
 .Ltmp3:
 	.size __xray_FunctionExit, .Ltmp3-__xray_FunctionExit
@@ -136,10 +139,7 @@ __xray_FunctionTailExit:
 	// this is an exit. In the future, we will introduce a new entry type that
 	// differentiates between a normal exit and a tail exit, but we'd have to do
 	// this and increment the version number for the header.
-	pushq %rbp
-	.cfi_def_cfa_offset 16
 	SAVE_REGISTERS
-	.cfi_def_cfa_offset 200
 
 	movq	_ZN6__xray19XRayPatchedFunctionE(%rip), %rax
 	testq %rax,%rax
@@ -151,7 +151,6 @@ __xray_FunctionTailExit:
 
 .Ltmp4:
 	RESTORE_REGISTERS
-	popq	%rbp
 	retq
 .Ltmp5:
 	.size __xray_FunctionTailExit, .Ltmp5-__xray_FunctionTailExit
@@ -164,10 +163,7 @@ __xray_FunctionTailExit:
 	.type __xray_ArgLoggerEntry,@function
 __xray_ArgLoggerEntry:
 	.cfi_startproc
-	pushq	%rbp
-	.cfi_def_cfa_offset 16
 	SAVE_REGISTERS
-	.cfi_def_cfa_offset 200
 
 	// Again, these function pointer loads must be atomic; MOV is fine.
 	movq	_ZN6__xray13XRayArgLoggerE(%rip), %rax
@@ -187,7 +183,6 @@ __xray_ArgLoggerEntry:
 
 .Larg1entryFail:
 	RESTORE_REGISTERS
-	popq	%rbp
 	retq
 
 .Larg1entryEnd: