metag: Add NPTL support

Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Signed-off-by: Bernhard Reutner-Fischer <rep.dot.nop@gmail.com>

8 files changed, 293 insertions, 36 deletions

diff --git a/libc/sysdeps/linux/metag/Makefile.arch b/libc/sysdeps/linux/metag/Makefile.arch
index 0e6fbfe12..3f11d6da8 100644
--- a/libc/sysdeps/linux/metag/Makefile.arch
+++ b/libc/sysdeps/linux/metag/Makefile.arch
@@ -7,4 +7,6 @@
 
 CSRC-y := brk.c syscall.c metag.c __syscall_error.c
 
-SSRC-y := _longjmp.S clone.S setjmp.S
+SSRC-y := _longjmp.S clone.S setjmp.S vfork.S
+
+SSRC-$(UCLIBC_HAS_THREADS_NATIVE) += libc-metag_load_tp.S
diff --git a/libc/sysdeps/linux/metag/bits/atomic.h b/libc/sysdeps/linux/metag/bits/atomic.h
new file mode 100644
index 000000000..64aa50bc4
--- /dev/null
+++ b/libc/sysdeps/linux/metag/bits/atomic.h
@@ -0,0 +1,66 @@
+/*
+ * Copyrith (C) 2013 Imagination Technologies Ltd.
+ *
+ * Licensed under LGPL v2.1 or later, see the file COPYING.LIB in this tarball.
+ *
+ */
+
+#include <stdint.h>
+#include <sysdep.h>
+
+typedef int8_t atomic8_t;
+typedef uint8_t uatomic8_t;
+typedef int_fast8_t atomic_fast8_t;
+typedef uint_fast8_t uatomic_fast8_t;
+
+typedef int32_t atomic32_t;
+typedef uint32_t uatomic32_t;
+typedef int_fast32_t atomic_fast32_t;
+typedef uint_fast32_t uatomic_fast32_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+void __metag_link_error (void);
+
+#define atomic_full_barrier() \
+     __asm__ __volatile__("": : :"memory")
+
+/* Atomic compare and exchange.  This sequence relies on the kernel to
+   provide a compare and exchange operation which is atomic. */
+
+#define __arch_compare_and_exchange_val_8_acq(mem, newval, oldval) \
+  ({ __metag_link_error (); oldval; })
+
+#define __arch_compare_and_exchange_val_16_acq(mem, newval, oldval) \
+  ({ __metag_link_error (); oldval; })
+
+/* This code uses the kernel helper to do cmpxchg. It relies on the fact
+   the helper code only clobbers D0Re0. */
+#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval)     \
+  ({ register __typeof (oldval) a_current __asm__ ("D1Ar1");		\
+    register __typeof (oldval) a_newval __asm__ ("D0Ar2") = (newval);   \
+    register __typeof (mem) a_ptr __asm__ ("D1Ar3") = (mem);            \
+    register __typeof (oldval) a_oldval __asm__ ("D0Ar4") = (oldval);	\
+    __asm__ __volatile__						\
+      ("0:\n\t"								\
+       "GETD	%[cur], [%[ptr]]\n\t"					\
+       "CMP	%[cur], %[old]\n\t"					\
+       "BNE	1f\n\t"							\
+       "MOVT	D1RtP, #0x6fff\n\t"					\
+       "ADD	D1RtP, D1RtP, #0xf040\n\t"				\
+       "SWAP	D1RtP, PC\n\t"						\
+       "MOV	%[cur], %[old]\n\t"					\
+       "CMP	D0Re0, #0\n\t"						\
+       "BNE	0b\n\t"							\
+       "1:"								\
+       : [cur] "=&r" (a_current)					\
+       : [new] "r" (a_newval), [ptr] "r" (a_ptr),			\
+	 [old] "r" (a_oldval)						\
+       : "D0Re0", "D1RtP", "cc", "memory");				\
+    a_current; })
+
+#define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
+  ({ __metag_link_error (); oldval; })
diff --git a/libc/sysdeps/linux/metag/bits/syscalls.h b/libc/sysdeps/linux/metag/bits/syscalls.h
index b5c8fc58c..7ea09c2c3 100644
--- a/libc/sysdeps/linux/metag/bits/syscalls.h
+++ b/libc/sysdeps/linux/metag/bits/syscalls.h
@@ -49,8 +49,9 @@
 (__extension__                                                          \
   ({unsigned int __sys_result;						\
     {									\
+      PREP_ARGS_##nr (args);						\
       register int _result __asm__ ("D0Re0"), _nr __asm__ ("D1Re0");	\
-      LOAD_ARGS_##nr (args);						\
+      LOAD_ARGS_##nr;							\
       _nr = (name);							\
       __asm__ volatile ("SWITCH  #0x440001 ! syscall " #name		\
 			: "=r" (_result)				\
@@ -68,32 +69,52 @@
 #undef INTERNAL_SYSCALL_ERRNO
 #define INTERNAL_SYSCALL_ERRNO(val, err)	(-(val))
 
-#define LOAD_ARGS_0()
+#define PREP_ARGS_0()
+#define PREP_ARGS_1(a1)                                 \
+  int _t1 = (int) (a1);                                 \
+  PREP_ARGS_0 ()
+#define PREP_ARGS_2(a1, a2)                             \
+  int _t2 = (int) (a2);                                 \
+  PREP_ARGS_1 (a1)
+#define PREP_ARGS_3(a1, a2, a3)                         \
+  int _t3 = (int) (a3);                                 \
+  PREP_ARGS_2 (a1, a2)
+#define PREP_ARGS_4(a1, a2, a3, a4)                     \
+  int _t4 = (int) (a4);                                 \
+  PREP_ARGS_3 (a1, a2, a3)
+#define PREP_ARGS_5(a1, a2, a3, a4, a5)                 \
+  int _t5 = (int) (a5);                                 \
+  PREP_ARGS_4 (a1, a2, a3, a4)
+#define PREP_ARGS_6(a1, a2, a3, a4, a5, a6)             \
+  int _t6 = (int) (a6);                                 \
+  PREP_ARGS_5 (a1, a2, a3, a4, a5)
+
+#define LOAD_ARGS_0
 #define ASM_ARGS_0
-#define LOAD_ARGS_1(a1)					\
-  register int _a1 __asm__ ("D1Ar1") = (int) (a1);	\
-  LOAD_ARGS_0 ()
-#define ASM_ARGS_1	ASM_ARGS_0, "d" (_a1)
-#define LOAD_ARGS_2(a1, a2)				\
-  register int _a2 __asm__ ("D0Ar2") = (int) (a2);	\
-  LOAD_ARGS_1 (a1)
-#define ASM_ARGS_2	ASM_ARGS_1, "d" (_a2)
-#define LOAD_ARGS_3(a1, a2, a3)				\
-  register int _a3 __asm__ ("D1Ar3") = (int) (a3);	\
-  LOAD_ARGS_2 (a1, a2)
-#define ASM_ARGS_3	ASM_ARGS_2, "d" (_a3)
-#define LOAD_ARGS_4(a1, a2, a3, a4)			\
-  register int _a4 __asm__ ("D0Ar4") = (int) (a4);	\
-  LOAD_ARGS_3 (a1, a2, a3)
-#define ASM_ARGS_4	ASM_ARGS_3, "d" (_a4)
-#define LOAD_ARGS_5(a1, a2, a3, a4, a5)			\
-  register int _a5 __asm__ ("D1Ar5") = (int) (a5);	\
-  LOAD_ARGS_4 (a1, a2, a3, a4)
-#define ASM_ARGS_5	ASM_ARGS_4, "d" (_a5)
-#define LOAD_ARGS_6(a1, a2, a3, a4, a5, a6)		\
-  register int _a6 __asm__ ("D0Ar6") = (int) (a6);	\
-  LOAD_ARGS_5 (a1, a2, a3, a4, a5)
-#define ASM_ARGS_6	ASM_ARGS_5, "d" (_a6)
+#define LOAD_ARGS_1                                     \
+  register int _a1 __asm__ ("D1Ar1") = (int) (_t1);     \
+  LOAD_ARGS_0
+#define ASM_ARGS_1 ASM_ARGS_0, "d" (_a1)
+#define LOAD_ARGS_2                                     \
+  register int _a2 __asm__ ("D0Ar2") = (int) (_t2);     \
+  LOAD_ARGS_1
+#define ASM_ARGS_2 ASM_ARGS_1, "d" (_a2)
+#define LOAD_ARGS_3                                     \
+  register int _a3 __asm__ ("D1Ar3") = (int) (_t3);     \
+  LOAD_ARGS_2
+#define ASM_ARGS_3 ASM_ARGS_2, "d" (_a3)
+#define LOAD_ARGS_4                                     \
+  register int _a4 __asm__ ("D0Ar4") = (int) (_t4);     \
+  LOAD_ARGS_3
+#define ASM_ARGS_4 ASM_ARGS_3, "d" (_a4)
+#define LOAD_ARGS_5                                     \
+  register int _a5 __asm__ ("D1Ar5") = (int) (_t5);     \
+  LOAD_ARGS_4
+#define ASM_ARGS_5 ASM_ARGS_4, "d" (_a5)
+#define LOAD_ARGS_6                                     \
+  register int _a6 __asm__ ("D0Ar6") = (int) (_t6);     \
+  LOAD_ARGS_5
+#define ASM_ARGS_6 ASM_ARGS_5, "d" (_a6)
 
 #endif /* __ASSEMBLER__ */
 #endif /* _BITS_SYSCALLS_H */
diff --git a/libc/sysdeps/linux/metag/clone.S b/libc/sysdeps/linux/metag/clone.S
index 8fff56710..d9d836338 100644
--- a/libc/sysdeps/linux/metag/clone.S
+++ b/libc/sysdeps/linux/metag/clone.S
@@ -8,7 +8,17 @@
 #include <asm/errno.h>
 #include <asm/unistd.h>
 
-/* int clone(int (*fn)(void *arg), void *child_stack, int flags, void *arg); */
+#define CLONE_VM      0x00000100
+#define CLONE_THREAD  0x00010000
+
+#ifdef __PIC__
+#define __CLONE_METAG_LOAD_TP ___metag_load_tp@PLT
+#else
+#define __CLONE_METAG_LOAD_TP ___metag_load_tp
+#endif
+
+/* int clone(int (*fn)(void *arg), void *child_stack, int flags, void *arg,
+			pid_t *ptid, struct user_desc *tls, pid_t *ctid); */
 
 	.text
 	.global	__clone
@@ -25,8 +35,12 @@ __clone:
 	MOV D0FrT, D1Ar1
 
 	! do the system call
-	! get flags
 	MOV	D1Ar1, D1Ar3
+	MOV	D1Ar3, D1Ar5
+	MOV	D1Ar5, D0Ar6
+	MOV	D0Ar6, D0Ar4
+	GETD	D0Ar4, [A0StP+#-4]
+
 	! new sp is already in D0Ar2
 	MOV	D1Re0, #__NR_clone
 	SWITCH	#0x440001
@@ -38,14 +52,36 @@ __clone:
 	! BRKPNT
 
 	! We are the child
-	! pick the function arg and call address off the stack and execute
-	MOV     D0Ar2, D0FrT
-	MOV     D1Ar1, D0Ar4
-	MOV     D1RtP, PC
+#ifdef RESET_PID
+	SETL	[A0StP++], D0FrT, D1RtP
+	MOVT	D0FrT, #HI(CLONE_THREAD)
+	ADD	D0FrT, D0FrT, #LO(CLONE_THREAD)
+	ANDS	D0FrT, D0FrT, D1Ar1
+	BNZ	3f
+	MOVT	D0FrT, #HI(CLONE_VM)
+	ADD	D0FrT, D0FrT, #LO(CLONE_VM)
+	ANDS	D0FrT, D0FrT, D1Ar1
+	BZ	1f
+	MOV	D1Ar1, #-1
+	BA	2f
+1:	MOV	D1Re0, #__NR_getpid
+	SWITCH	#0x440001
+	MOV	D1Ar1, D0Re0
+2:	CALLR	D1RtP, __CLONE_METAG_LOAD_TP
+	SUB	D0Re0, D0Re0, #TLS_PRE_TCB_SIZE
+	SETD	[D0Re0 + #PID], D1Ar1
+	SETD	[D0Re0 + #TID], D1Ar1
+3:	GETL	D0FrT, D1RtP, [--A0StP]
+#endif
+	! Rearrange the function arg and call address from registers
+	MOV	D0Ar2, D0FrT
+	MOV	D1Ar1, D0Ar6
+	MOV	D1RtP, PC
 	ADD	D1RtP, D1RtP, #8
 	MOV	PC, D0Ar2
 
-	! and we are done, passing the return value through D0Re0
+	! and we are done, passing the return value D0Re0 through D1Ar1
+	MOV	D1Ar1, D0Re0
 #ifdef __PIC__
 	B	_exit@PLT
 #else
diff --git a/libc/sysdeps/linux/metag/crt1.S b/libc/sysdeps/linux/metag/crt1.S
index 74adc1fdb..e8561a46b 100644
--- a/libc/sysdeps/linux/metag/crt1.S
+++ b/libc/sysdeps/linux/metag/crt1.S
@@ -29,6 +29,7 @@ __start:
 
 	MOV	D1Ar3,D0Ar2			! argv
 	MOV	D0Ar2,D1Ar1			! argc
+	MOV	D0Ar6,D0Ar4			! rtld_fini
 
 #ifdef __PIC__
 	ADDT	A1LbP,CPC1,#HI(__GLOBAL_OFFSET_TABLE__)
@@ -54,8 +55,6 @@ __start:
 	ADD	D1Ar5,D1Ar5,#LO(__fini)
 #endif
 
-	MOV	D0Ar6,#0			! rtld_fini
-
 	MOVT	D1Re0,#0x80
 
 	SETL	[A0StP++],D0Re0,D1Re0		! stack_end (8Mb)
diff --git a/libc/sysdeps/linux/metag/libc-metag_load_tp.S b/libc/sysdeps/linux/metag/libc-metag_load_tp.S
new file mode 100644
index 000000000..a91f162ec
--- /dev/null
+++ b/libc/sysdeps/linux/metag/libc-metag_load_tp.S
@@ -0,0 +1,7 @@
+/*
+ * Copyright (C) 2013 Imagination Technologies Ltd.
+ *
+ * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB in this tarball.
+ */
+
+#include <ldso/ldso/metag/metag_load_tp.S>
diff --git a/libc/sysdeps/linux/metag/sysdep.h b/libc/sysdeps/linux/metag/sysdep.h
new file mode 100644
index 000000000..a12f393ce
--- /dev/null
+++ b/libc/sysdeps/linux/metag/sysdep.h
@@ -0,0 +1,59 @@
+/*
+ * Assembler macros for Meta.
+ *
+ * Licensed under LGPL v2.1 or later, see the file COPYING.LIB in this tarball.
+ */
+
+#include <common/sysdep.h>
+
+#include <features.h>
+#include <libc-internal.h>
+
+#ifdef	__ASSEMBLER__
+
+#ifdef SHARED
+#define PLTJMP(_x)	_x##@PLT
+#else
+#define PLTJMP(_x)	_x
+#endif
+
+#undef	PSEUDO_END
+#define	PSEUDO_END(name)						\
+  SYSCALL_ERROR_HANDLER							\
+  END (name)
+
+#if defined NOT_IN_libc
+# define SYSCALL_ERROR __local_syscall_error
+# ifdef RTLD_PRIVATE_ERRNO
+#  define SYSCALL_ERROR_HANDLER					\
+__local_syscall_error:						\
+       NEG	D0Re0, D0Re0;					\
+       ADDT	D1Re0, CPC1, #HI(_rtld_errno);			\
+       ADD	D1Re0, D1Re0, #LO(_rtld_errno) + 4;		\
+       SETD	[D1Re0], D0Re0;					\
+       NEG	D0Re0, #0x1;					\
+       MOV	PC, D1RtP;
+# else
+#  define SYSCALL_ERROR_HANDLER					\
+__local_syscall_error:						\
+	MOV	D1Re0, D1RtP;					\
+	SETL	[A0StP++], D0Re0, D1Re0;			\
+	CALLR	D1RtP, PLTJMP(___errno_location);		\
+	GETD	D1Re0, [A0StP+#-8];				\
+	NEG	D1Re0, D1Re0;					\
+	SETD	[D0Re0], D1Re0;					\
+	NEG	D0Re0, #0x1;					\
+	GETD	D1RtP, [A0StP+#-4];				\
+	SUB	A0StP, A0StP, #0x8;				\
+	MOV	PC, D1RtP;
+# endif
+#else
+# define SYSCALL_ERROR_HANDLER	/* Nothing here; code in sysdep.S is used.  */
+# define SYSCALL_ERROR ___syscall_error
+#endif
+
+#endif /* __ASSEMBLER __*/
+
+/* Pointer mangling is not yet supported for META.  */
+#define PTR_MANGLE(var) (void) (var)
+#define PTR_DEMANGLE(var) (void) (var)
diff --git a/libc/sysdeps/linux/metag/vfork.S b/libc/sysdeps/linux/metag/vfork.S
new file mode 100644
index 000000000..8573dedd3
--- /dev/null
+++ b/libc/sysdeps/linux/metag/vfork.S
@@ -0,0 +1,67 @@
+!    Copyright (C) 2013 Imagination Technologies Ltd.
+
+!    Licensed under LGPL v2.1 or later, see the file COPYING.LIB in this tarball.
+
+#include <asm/unistd.h>
+
+#define _ERRNO_H
+#include <bits/errno.h>
+#include <sys/syscall.h>
+
+#ifndef SAVE_PID
+#define SAVE_PID
+#endif
+
+#ifndef RESTORE_PID
+#define RESTORE_PID
+#endif
+
+#ifdef __NR_vfork
+#define __VFORK_NR __NR_vfork
+#else
+#define __VFORK_NR __NR_fork
+#endif
+
+/* Clone the calling process, but without copying the whole address space.
+   The calling process is suspended until the new process exits or is
+   replaced by a call to `execve'.  Return -1 for errors, 0 to the new process,
+   and the process ID of the new process to the old process.  */
+
+	.balign 4
+	.global	___vfork
+	.hidden	___vfork
+	.type	___vfork, @function
+___vfork:
+
+	SAVE_PID
+
+	MOV	D1Ar1, #0x4111	/* CLONE_VM | CLONE_VFORK | SIGCHLD */
+	MOV	D0Ar2, #0
+	MOV	D1Ar3, #0
+	MOV	D0Ar4, #0
+	MOV	D1Ar5, #0
+	MOV	D0Ar6, #0
+	MOV	D1Re0, #__NR_clone
+	SWITCH  #0x440001
+
+	RESTORE_PID
+
+	MOVT	D1Re0, #HI(-4096)
+	ADD	D1Re0, D1Re0, #LO(-4096)
+	CMP	D1Re0, D0Re0
+	BCS	error
+
+	/* Syscall worked. Return to child/parent */
+	MOV	PC, D1RtP
+
+error:
+	MOV	D1Ar1, D0Re0
+#ifdef __PIC__
+	B	___syscall_error@PLT
+#else
+	B	___syscall_error
+#endif
+	.size ___vfork,.-___vfork
+
+weak_alias(__vfork,vfork)
+libc_hidden_weak(vfork)