i386 vsyscall support (vdso-provided sysenter/syscall instruction based)

this doubles the performance of the fastest syscalls on the atom I
tested it on; improvement is reportedly much more dramatic on
worst-case cpus. cannot be used for cancellable syscalls.

2 files changed, 59 insertions, 16 deletions

diff --git a/src/internal/i386/syscall.s b/src/internal/i386/syscall.s
index 327d4cc0..79296ba0 100644
--- a/src/internal/i386/syscall.s
+++ b/src/internal/i386/syscall.s
@@ -1,20 +1,62 @@
+.hidden __sysinfo
+
+.global __vsyscall
+.type __vsyscall,@function
+__vsyscall:
+	push %edi
+	push %ebx
+	mov %edx,%ebx
+	mov %edi,%edx
+	mov 12(%esp),%edi
+	push %eax
+	call 1f
+2:	pop %ebx
+	pop %ebx
+	pop %edi
+	ret
+
+1:	mov (%esp),%eax
+	add $[__sysinfo-2b],%eax
+	mov (%eax),%eax
+	test %eax,%eax
+	jz 1f
+	push %eax
+	mov 8(%esp),%eax
+	ret
+1:	mov 4(%esp),%eax
+	int $128
+	ret
+
+.global __vsyscall6
+.type __vsyscall6,@function
+__vsyscall6:
+	push %ebp
+	push %eax
+	mov 12(%esp), %ebp
+	mov (%ebp), %eax
+	mov 4(%ebp), %ebp
+	push %eax
+	mov 4(%esp),%eax
+	call __vsyscall
+	pop %ebp
+	pop %ebp
+	pop %ebp
+	ret
+
 .global __syscall
 .type __syscall,@function
 __syscall:
-	pushl %ebx
-	pushl %esi
-	pushl %edi
-	pushl %ebp
-	movl 20(%esp),%eax
-	movl 24(%esp),%ebx
-	movl 28(%esp),%ecx
-	movl 32(%esp),%edx
-	movl 36(%esp),%esi
-	movl 40(%esp),%edi
-	movl 44(%esp),%ebp
-	int $128
-	popl %ebp
-	popl %edi
-	popl %esi
-	popl %ebx
+	lea 24(%esp),%eax
+	push %esi
+	push %edi
+	push %eax
+	mov 16(%esp),%eax
+	mov 20(%esp),%edx
+	mov 24(%esp),%ecx
+	mov 28(%esp),%edi
+	mov 32(%esp),%esi
+	call __vsyscall6
+	pop %edi
+	pop %edi
+	pop %esi
 	ret
diff --git a/src/internal/libc.c b/src/internal/libc.c
index 004c1a33..d22d2747 100644
--- a/src/internal/libc.c
+++ b/src/internal/libc.c
@@ -15,3 +15,4 @@ __asm__(".hidden __libc");
 #endif
 
 size_t __hwcap;
+size_t __sysinfo;