overhaul i386 syscall mechanism not to depend on external asm source

this is the first part of a series of patches intended to make
__syscall fully self-contained in the object file produced using
syscall.h, which will make it possible for crt1 code to perform
syscalls.

the (confusingly named) i386 __vsyscall mechanism, which this commit
removes, was introduced before the presence of a valid thread pointer
was mandatory; back then the thread pointer was setup lazily only if
threads were used. the intent was to be able to perform syscalls using
the kernel's fast entry point in the VDSO, which can use the sysenter
(Intel) or syscall (AMD) instruction instead of int $128, but without
inlining an access to the __syscall global at the point of each
syscall, which would incur a significant size cost from PIC setup
everywhere. the mechanism also shuffled registers/calling convention
around to avoid spills of call-saved registers, and to avoid
allocating ebx or ebp via asm constraints, since there are plenty of
broken-but-supported compiler versions which are incapable of
allocating ebx with -fPIC or ebp with -fno-omit-frame-pointer.

the new mechanism preserves the properties of avoiding spills and
avoiding allocation of ebx/ebp in constraints, but does it inline,
using some fairly simple register shuffling, and uses a field of the
thread structure rather than global data for the vdso-provided syscall
code address.

for now, the external __syscall function is refactored not to use the
old __vsyscall so it can be kept, but the intent is to remove it too.

arch/i386/syscall_arch.h

@@ -3,52 +3,63 @@
 ((union { long long ll; long l[2]; }){ .ll = x }).l[1]
 #define __SYSCALL_LL_O(x) __SYSCALL_LL_E((x))
 
+#if SYSCALL_NO_TLS
+#define SYSCALL_INSNS "int $128"
+#else
+#define SYSCALL_INSNS "call *%%gs:16"
+#endif
+
+#define SYSCALL_INSNS_12 "xchg %%ebx,%%edx ; " SYSCALL_INSNS " ; xchg %%ebx,%%edx"
+#define SYSCALL_INSNS_34 "xchg %%ebx,%%edi ; " SYSCALL_INSNS " ; xchg %%ebx,%%edi"
+
 static inline long __syscall0(long n)
 {
 	unsigned long __ret;
-	__asm__ __volatile__ (".hidden __vsyscall ; call __vsyscall" : "=a"(__ret) : "a"(n) : "memory");
+	__asm__ __volatile__ (SYSCALL_INSNS : "=a"(__ret) : "a"(n) : "memory");
 	return __ret;
 }
 
 static inline long __syscall1(long n, long a1)
 {
 	unsigned long __ret;
-	__asm__ __volatile__ (".hidden __vsyscall ; call __vsyscall" : "=a"(__ret) : "a"(n), "d"(a1) : "memory");
+	__asm__ __volatile__ (SYSCALL_INSNS_12 : "=a"(__ret) : "a"(n), "d"(a1) : "memory");
 	return __ret;
 }
 
 static inline long __syscall2(long n, long a1, long a2)
 {
 	unsigned long __ret;
-	__asm__ __volatile__ (".hidden __vsyscall ; call __vsyscall" : "=a"(__ret) : "a"(n), "d"(a1), "c"(a2) : "memory");
+	__asm__ __volatile__ (SYSCALL_INSNS_12 : "=a"(__ret) : "a"(n), "d"(a1), "c"(a2) : "memory");
 	return __ret;
 }
 
 static inline long __syscall3(long n, long a1, long a2, long a3)
 {
 	unsigned long __ret;
-	__asm__ __volatile__ (".hidden __vsyscall ; call __vsyscall" : "=a"(__ret) : "a"(n), "d"(a1), "c"(a2), "D"(a3) : "memory");
+	__asm__ __volatile__ (SYSCALL_INSNS_34 : "=a"(__ret) : "a"(n), "D"(a1), "c"(a2), "d"(a3) : "memory");
 	return __ret;
 }
 
 static inline long __syscall4(long n, long a1, long a2, long a3, long a4)
 {
 	unsigned long __ret;
-	__asm__ __volatile__ (".hidden __vsyscall ; call __vsyscall" : "=a"(__ret) : "a"(n), "d"(a1), "c"(a2), "D"(a3), "S"(a4) : "memory");
+	__asm__ __volatile__ (SYSCALL_INSNS_34 : "=a"(__ret) : "a"(n), "D"(a1), "c"(a2), "d"(a3), "S"(a4) : "memory");
 	return __ret;
 }
 
 static inline long __syscall5(long n, long a1, long a2, long a3, long a4, long a5)
 {
-	unsigned long __ret;
-	__asm__ __volatile__ ("push %6 ; .hidden __vsyscall ; call __vsyscall ; add $4,%%esp" : "=a"(__ret) : "a"(n), "d"(a1), "c"(a2), "D"(a3), "S"(a4), "g"(a5) : "memory");
+	unsigned long __ret, __tmp;
+	__asm__ __volatile__ ("mov %%ebx,%1 ; mov %3,%%ebx ; " SYSCALL_INSNS " ; mov %1,%%ebx"
+		: "=a"(__ret), "=m"(__tmp) : "a"(n), "g"(a1), "c"(a2), "d"(a3), "S"(a4), "D"(a5) : "memory");
 	return __ret;
 }
 
 static inline long __syscall6(long n, long a1, long a2, long a3, long a4, long a5, long a6)
 {
-	unsigned long __ret;
-	__asm__ __volatile__ ("push %6 ; .hidden __vsyscall6 ; call __vsyscall6 ; add $4,%%esp" : "=a"(__ret) : "a"(n), "d"(a1), "c"(a2), "D"(a3), "S"(a4), "g"(0+(long[]){a5, a6}) : "memory");
+	unsigned long __ret, __tmp1, __tmp2;
+	__asm__ __volatile__ ("mov %%ebx,%1 ; mov %%ebp,%2 ; mov %4,%%ebx ; mov %9,%%ebp ; " SYSCALL_INSNS " ; mov %2,%%ebp ; mov %1,%%ebx"
+		: "=a"(__ret), "=m"(__tmp1), "=m"(__tmp2) : "a"(n), "g"(a1), "c"(a2), "d"(a3), "S"(a4), "D"(a5), "g"(a6) : "memory");
 	return __ret;
 }
 

ldso/dynlink.c

@@ -1,4 +1,5 @@
 #define _GNU_SOURCE
+#define SYSCALL_NO_TLS 1
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdarg.h>
@@ -1685,6 +1686,8 @@ _Noreturn void __dls3(size_t *sp)
 	libc.auxv = auxv = (void *)(argv+i+1);
 	decode_vec(auxv, aux, AUX_CNT);
 	__hwcap = aux[AT_HWCAP];
+	search_vec(auxv, &__sysinfo, AT_SYSINFO);
+	__pthread_self()->sysinfo = __sysinfo;
 	libc.page_size = aux[AT_PAGESZ];
 	libc.secure = ((aux[0]&0x7800)!=0x7800 || aux[AT_UID]!=aux[AT_EUID]
 		|| aux[AT_GID]!=aux[AT_EGID] || aux[AT_SECURE]);

src/env/__init_tls.c

@@ -1,3 +1,4 @@
+#define SYSCALL_NO_TLS 1
 #include <elf.h>
 #include <limits.h>
 #include <sys/mman.h>
@@ -21,6 +22,7 @@ int __init_tp(void *p)
 	td->tid = __syscall(SYS_set_tid_address, &__thread_list_lock);
 	td->locale = &libc.global_locale;
 	td->robust_list.head = &td->robust_list.head;
+	td->sysinfo = __sysinfo;
 	td->next = td->prev = td;
 	return 0;
 }

src/env/__libc_start_main.c

@@ -28,7 +28,7 @@ void __init_libc(char **envp, char *pn)
 	libc.auxv = auxv = (void *)(envp+i+1);
 	for (i=0; auxv[i]; i+=2) if (auxv[i]<AUX_CNT) aux[auxv[i]] = auxv[i+1];
 	__hwcap = aux[AT_HWCAP];
-	__sysinfo = aux[AT_SYSINFO];
+	if (aux[AT_SYSINFO]) __sysinfo = aux[AT_SYSINFO];
 	libc.page_size = aux[AT_PAGESZ];
 
 	if (!pn) pn = (void*)aux[AT_EXECFN];

src/internal/defsysinfo.c

@@ -0,0 +1,3 @@
+#include "libc.h"
+
+size_t __sysinfo;

src/internal/i386/defsysinfo.s

@@ -0,0 +1,9 @@
+1:	int $128
+	ret
+
+.data
+.align 4
+.hidden __sysinfo
+.global __sysinfo
+__sysinfo:
+	.long 1b

src/internal/i386/syscall.s

@@ -1,78 +1,21 @@
-.hidden __sysinfo
-
-# The calling convention for __vsyscall has the syscall number
-# and 5 args arriving as:  eax, edx, ecx, edi, esi, 4(%esp).
-# This ensures that the inline asm in the C code never has to touch
-# ebx or ebp (which are unavailable in PIC and frame-pointer-using
-# code, respectively), and optimizes for size/simplicity in the caller.
-
-.global __vsyscall
-.hidden __vsyscall
-.type __vsyscall,@function
-__vsyscall:
-	push %edi
-	push %ebx
-	mov %edx,%ebx
-	mov %edi,%edx
-	mov 12(%esp),%edi
-	push %eax
-	call 1f
-2:	mov %ebx,%edx
-	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                     # tail call to kernel vsyscall entry
-1:	mov 4(%esp),%eax
-	int $128
-	ret
-
-# The __vsyscall6 entry point is used only for 6-argument syscalls.
-# Instead of passing the 5th argument on the stack, a pointer to the
-# 5th and 6th arguments is passed. This is ugly, but there are no
-# register constraints the inline asm could use that would make it
-# possible to pass two arguments on the stack.
-
-.global __vsyscall6
-.hidden __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
 .hidden __syscall
 .type __syscall,@function
 __syscall:
-	lea 24(%esp),%eax
+	push %ebx
 	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
+	push %ebp
+	mov 20(%esp),%eax
+	mov 24(%esp),%ebx
+	mov 28(%esp),%ecx
+	mov 32(%esp),%edx
+	mov 36(%esp),%esi
+	mov 40(%esp),%edi
+	mov 44(%esp),%ebp
+	call *%gs:16
+	pop %ebp
 	pop %edi
 	pop %esi
+	pop %ebx
 	ret

src/internal/libc.c

@@ -3,7 +3,6 @@
 struct __libc __libc;
 
 size_t __hwcap;
-size_t __sysinfo;
 char *__progname=0, *__progname_full=0;
 
 weak_alias(__progname, program_invocation_short_name);

src/thread/pthread_create.c

@@ -315,6 +315,7 @@ int __pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict att
 	}
 	new->robust_list.head = &new->robust_list.head;
 	new->CANARY = self->CANARY;
+	new->sysinfo = self->sysinfo;
 
 	/* Setup argument structure for the new thread on its stack.
 	 * It's safe to access from the caller only until the thread