kvx: add support for kvx arch to uClibc-ng

This commit adds support for Kalray VLIW family (kvx)

Kalray kv3 core is embedded in Kalray Coolidge SoC. This core which is the
third of the KV family has the following features:

    32/64 bits execution mode
    6-issue VLIW architecture
    64 x 64bits general purpose registers
    SIMD instructions
    little-endian

In order to build a usable toolchain, build scripts are provided at the
following address: https://github.com/kalray/build-scripts.

Kalray uses FOSS which is available at https://github.com/kalray
This includes Linux kernel, uClibc-ng, gcc, binutils, etc.

Signed-off-by: Clément Léger <cleger@kalray.eu>
Signed-off-by: Guillaume Thouvenin <gthouvenin@kalray.eu>
Signed-off-by: Laurent Thevenoux <lthevenoux@kalray.eu>
Signed-off-by: Marc Poulhies <mpoulhies@kalray.eu>
Signed-off-by: Marius Gligor <mgligor@kalray.eu>
Signed-off-by: Yann Sionneau <ysionneau@kalray.eu>

36 files changed, 2080 insertions, 0 deletions

diff --git a/libc/string/kvx/Makefile b/libc/string/kvx/Makefile
new file mode 100644
index 000000000..0a95346fd
--- /dev/null
+++ b/libc/string/kvx/Makefile
@@ -0,0 +1,13 @@
+# Makefile for uClibc
+#
+# Copyright (C) 2000-2005 Erik Andersen <andersen@uclibc.org>
+#
+# Licensed under the LGPL v2.1, see the file COPYING.LIB in this tarball.
+#
+
+top_srcdir:=../../../
+top_builddir:=../../../
+all: objs
+include $(top_builddir)Rules.mak
+include ../Makefile.in
+include $(top_srcdir)Makerules
diff --git a/libc/string/kvx/memcpy.S b/libc/string/kvx/memcpy.S
new file mode 100644
index 000000000..290e705b4
--- /dev/null
+++ b/libc/string/kvx/memcpy.S
@@ -0,0 +1,221 @@
+/*
+ * Copyright (C) 2020 Kalray Inc.
+ *
+ * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB
+ * in this tarball.
+ */
+
+#include <sysdep.h>
+
+.align 16
+ENTRY(memcpy)
+	cb.deqz $r2? .Lreturn
+	compd.geu $r3 = $r2, 256
+	copyd $r6 = $r0
+	;;
+	cb.deqz $r3? .Lremaining_256
+	;;
+	lq.u $r32r33 = 0[$r1]
+	addd $r2 = $r2, -256
+	;;
+	lq.u $r34r35 = 16[$r1]
+	;;
+	lq.u $r36r37 = 32[$r1]
+	srld $r7 = $r2, 8
+	;;
+	lq.u $r38r39 = 48[$r1]
+	;;
+	lq.u $r40r41 = 64[$r1]
+	;;
+	lq.u $r42r43 = 80[$r1]
+	;;
+	lq.u $r44r45 = 96[$r1]
+	;;
+	lq.u $r46r47 = 112[$r1]
+	;;
+	lq.u $r48r49 = 128[$r1]
+	;;
+	lq.u $r50r51 = 144[$r1]
+	;;
+	lq.u $r52r53 = 160[$r1]
+	;;
+	lq.u $r54r55 = 176[$r1]
+	;;
+	lq.u $r56r57 = 192[$r1]
+	;;
+	lq.u $r58r59 = 208[$r1]
+	compd.geu $r3 = $r2, 256
+	;;
+	lq.u $r60r61 = 224[$r1]
+	;;
+	lq.u $r62r63 = 240[$r1]
+	addd $r1 = $r1, 256
+	;;
+	cb.deqz $r7? .Lstreaming_loop_end
+	;;
+	loopdo $r7? .Lstreaming_loop_end
+		;;
+		sq 0[$r0] = $r32r33
+		addd $r2 = $r2, -256
+		;;
+		lq.u $r32r33 = 0[$r1]
+		;;
+		sq 16[$r0] = $r34r35
+		;;
+		lq.u $r34r35 = 16[$r1]
+		;;
+		sq 32[$r0] = $r36r37
+		;;
+		lq.u $r36r37 = 32[$r1]
+		;;
+		sq 48[$r0] = $r38r39
+		;;
+		lq.u $r38r39 = 48[$r1]
+		;;
+		sq 64[$r0] = $r40r41
+		;;
+		lq.u $r40r41 = 64[$r1]
+		;;
+		sq 80[$r0] = $r42r43
+		;;
+		lq.u $r42r43 = 80[$r1]
+		;;
+		sq 96[$r0] = $r44r45
+		;;
+		lq.u $r44r45 = 96[$r1]
+		;;
+		sq 112[$r0] = $r46r47
+		;;
+		lq.u $r46r47 = 112[$r1]
+		;;
+		sq 128[$r0] = $r48r49
+		;;
+		lq.u $r48r49 = 128[$r1]
+		;;
+		sq 144[$r0] = $r50r51
+		;;
+		lq.u $r50r51 = 144[$r1]
+		;;
+		sq 160[$r0] = $r52r53
+		;;
+		lq.u $r52r53 = 160[$r1]
+		;;
+		sq 176[$r0] = $r54r55
+		;;
+		lq.u $r54r55 = 176[$r1]
+		;;
+		sq 192[$r0] = $r56r57
+		;;
+		lq.u $r56r57 = 192[$r1]
+		;;
+		sq 208[$r0] = $r58r59
+		;;
+		lq.u $r58r59 = 208[$r1]
+		;;
+		sq 224[$r0] = $r60r61
+		;;
+		lq.u $r60r61 = 224[$r1]
+		;;
+		sq 240[$r0] = $r62r63
+		addd $r0 = $r0, 256
+		;;
+		lq.u $r62r63 = 240[$r1]
+		addd $r1 = $r1, 256
+		;;
+	.Lstreaming_loop_end:
+	sq 0[$r0] = $r32r33
+	;;
+	sq 16[$r0] = $r34r35
+	;;
+	sq 32[$r0] = $r36r37
+	;;
+	sq 48[$r0] = $r38r39
+	;;
+	sq 64[$r0] = $r40r41
+	;;
+	sq 80[$r0] = $r42r43
+	;;
+	sq 96[$r0] = $r44r45
+	;;
+	sq 112[$r0] = $r46r47
+	;;
+	sq 128[$r0] = $r48r49
+	;;
+	sq 144[$r0] = $r50r51
+	;;
+	sq 160[$r0] = $r52r53
+	;;
+	sq 176[$r0] = $r54r55
+	;;
+	sq 192[$r0] = $r56r57
+	;;
+	sq 208[$r0] = $r58r59
+	;;
+	sq 224[$r0] = $r60r61
+	;;
+	sq 240[$r0] = $r62r63
+	addd $r0 = $r0, 256
+	;;
+.Lremaining_256:
+	andd $r11 = $r2, 16
+	srld $r7 = $r2, 5
+	;;
+	cb.deqz $r7? .Lloop_32_end
+	;;
+	loopdo $r7? .Lloop_32_end
+		;;
+		lo $r32r33r34r35 = 0[$r1]
+		addd $r1 = $r1, 32
+		addd $r2 = $r2, -32
+		;;
+		so 0[$r0] = $r32r33r34r35
+		addd $r0 = $r0, 32
+		;;
+	.Lloop_32_end:
+	andd $r10 = $r2, 8
+	andd $r9 = $r2, 4
+	cb.deqz $r11? .Lloop_remaining_16
+	lq.u.dnez $r11? $r32r33 = 0[$r1]
+	;;
+	sq 0[$r0] = $r32r33
+	addd $r1 = $r1, 16
+	addd $r0 = $r0, 16
+	;;
+.Lloop_remaining_16:
+	andd $r8 = $r2, 2
+	andd $r7 = $r2, 1
+	cb.deqz $r10? .Lloop_remaining_8
+	ld.dnez $r10? $r32 = 0[$r1]
+	;;
+	sd 0[$r0] = $r32
+	addd $r1 = $r1, 8
+	addd $r0 = $r0, 8
+	;;
+.Lloop_remaining_8:
+	cb.deqz $r9? .Lloop_remaining_4
+	lwz.dnez $r9? $r32 = 0[$r1]
+	;;
+	sw 0[$r0] = $r32
+	addd $r1 = $r1, 4
+	addd $r0 = $r0, 4
+	;;
+.Lloop_remaining_4:
+	cb.deqz $r8? .Lloop_remaining_2
+	lhz.dnez $r8? $r32 = 0[$r1]
+	;;
+	sh 0[$r0] = $r32
+	addd $r1 = $r1, 2
+	addd $r0 = $r0, 2
+	;;
+.Lloop_remaining_2:
+	lbz.dnez $r7? $r32 = 0[$r1]
+	;;
+	sb.dnez $r7? 0[$r0] = $r32
+	;;
+.Lreturn:
+	copyd $r0 = $r6
+	ret
+	;;
+END(memcpy)
+
+libc_hidden_def(memcpy)
diff --git a/libc/string/kvx/memset.S b/libc/string/kvx/memset.S
new file mode 100644
index 000000000..45023a68f
--- /dev/null
+++ b/libc/string/kvx/memset.S
@@ -0,0 +1,146 @@
+/*
+ * Copyright (C) 2019 Kalray Inc.
+ *
+ * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB
+ * in this tarball.
+ */
+
+#define REPLICATE_BYTE_MASK	0x0101010101010101
+#define MIN_SIZE_FOR_ALIGN	128
+
+/*
+ * Optimized memset for kvx architecture
+ *
+ * In order to optimize memset on kvx, we can use various things:
+ * - conditionnal store which avoid branch penalty
+ * - store half/word/double/quad/octuple to store up to 16 bytes at a time
+ * - hardware loop for steady cases.
+ *
+ * First,  we start by checking if the size is below a minimum size. If so, we
+ * skip the alignment part. Indeed, the kvx supports misalignment and the
+ * penalty for letting it do unaligned accesses is lower than trying to
+ * realigning us. So for small sizes, we don't even bother to realign.
+ * In order to create the 64 bits pattern, we use sbmm to replicate the pattern
+ * on all bits on a register in one call.
+ * Once alignment has been reached, we can do the hardware loop using store
+ * octuple in order to optimize throughput. Care must be taken to align hardware
+ * loops on at least 8 bytes for performances.
+ * Once the main loop has been done, we finish the copy by checking length to do
+ * the necessary calls to store remaining bytes.
+ */
+
+#include <sysdep.h>
+
+.align 16
+ENTRY(memset)
+	/* Preserve return value */
+	copyd $r3 = $r0
+	/* Replicate the first pattern byte on all bytes */
+	sbmm8 $r32 = $r1, REPLICATE_BYTE_MASK
+	/* Check if length < MIN_SIZE_FOR_ALIGN */
+	compd.geu $r7 = $r2, MIN_SIZE_FOR_ALIGN
+	/* Invert address to compute what we need to copy to be aligned on 32 bytes */
+	negd $r5 = $r0
+	;;
+	/* Check if we are aligned on 32 bytes */
+	andw $r9 = $r0, 0x1F
+	/* Compute the length that will be copied to align on 32 bytes boundary */
+	andw $r6 = $r5, 0x1F
+	/*
+	 * If size < MIN_SIZE_FOR_ALIGN bits, directly go to so, it will be done
+	 * unaligned but that is still better that what we can do with sb
+	 */
+	cb.deqz $r7? .Laligned_32
+	;;
+	/* Remove unaligned part from length */
+	sbfd $r2 = $r6, $r2
+	/* If we are already aligned on 32 bytes, jump to main "so" loop */
+	cb.deqz $r9? .Laligned_32
+	/* Check if we need to copy 1 byte */
+	andw $r4 = $r5, (1 << 0)
+	;;
+	/* If we are not aligned, store byte */
+	sb.dnez $r4? [$r0] = $r32
+	/* Check if we need to copy 2 bytes */
+	andw $r4 = $r5, (1 << 1)
+	/* Add potentially copied part for next store offset */
+	addd $r0 = $r0, $r4
+	;;
+	sh.dnez $r4? [$r0] = $r32
+	/* Check if we need to copy 4 bytes */
+	andw $r4 = $r5, (1 << 2)
+	addd $r0 = $r0, $r4
+	;;
+	sw.dnez $r4? [$r0] = $r32
+	/* Check if we need to copy 8 bytes */
+	andw $r4 = $r5, (1 << 3)
+	addd $r0 = $r0, $r4
+	/* Copy second part of pattern for sq */
+	copyd $r33 = $r32
+	;;
+	sd.dnez $r4? [$r0] = $r32
+	/* Check if we need to copy 16 bytes */
+	andw $r4 = $r5, (1 << 4)
+	addd $r0 = $r0, $r4
+	;;
+	sq.dnez $r4? [$r0] = $r32r33
+	addd $r0 = $r0, $r4
+	;;
+.Laligned_32:
+	/* Copy second part of pattern for sq */
+	copyd $r33 = $r32
+	/* Prepare amount of data for 32 bytes store */
+	srld $r10 = $r2, 5
+	nop
+	nop
+	;;
+	copyq $r34r35 = $r32, $r33
+	/* Remaining bytes for 16 bytes store */
+	andw $r8 = $r2, (1 << 4)
+	make $r11 = 32
+	/* Check if there are enough data for 32 bytes store */
+	cb.deqz $r10? .Laligned_32_done
+	;;
+	loopdo $r10, .Laligned_32_done
+		;;
+		so 0[$r0] = $r32r33r34r35
+		addd $r0 = $r0, $r11
+		;;
+	.Laligned_32_done:
+	/*
+	 * Now that we have handled every aligned bytes using 'so', we can
+	 * handled the remainder of length using store by decrementing size
+	 * We also exploit the fact we are aligned to simply check remaining
+	 * size */
+	sq.dnez $r8? [$r0] = $r32r33
+	addd $r0 = $r0, $r8
+	/* Remaining bytes for 8 bytes store */
+	andw $r8 = $r2, (1 << 3)
+	cb.deqz $r2? .Lmemset_done
+	;;
+	sd.dnez $r8? [$r0] = $r32
+	addd $r0 = $r0, $r8
+	/* Remaining bytes for 4 bytes store */
+	andw $r8 = $r2, (1 << 2)
+	;;
+	sw.dnez $r8? [$r0] = $r32
+	addd $r0 = $r0, $r8
+	/* Remaining bytes for 2 bytes store */
+	andw $r8 = $r2, (1 << 1)
+	;;
+	sh.dnez $r8? [$r0] = $r32
+	addd $r0 = $r0, $r8
+	;;
+	sb.odd $r2? [$r0] = $r32
+	/* Restore original value */
+	copyd $r0 = $r3
+	ret
+	;;
+.Lmemset_done:
+	/* Restore original value */
+	copyd $r0 = $r3
+	ret
+	;;
+END(memset)
+
+libc_hidden_def(memset)
diff --git a/libc/sysdeps/linux/kvx/Makefile b/libc/sysdeps/linux/kvx/Makefile
new file mode 100644
index 000000000..633c91f3e
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/Makefile
@@ -0,0 +1,13 @@
+# Makefile for uClibc
+#
+# Copyright (C) 2000-2005 Erik Andersen <andersen@uclibc.org>
+#
+# Licensed under the LGPL v2.1, see the file COPYING.LIB in this tarball.
+#
+
+top_srcdir=../../../../
+top_builddir=../../../../
+all: objs
+include $(top_builddir)Rules.mak
+include Makefile.arch
+include $(top_srcdir)Makerules
diff --git a/libc/sysdeps/linux/kvx/Makefile.arch b/libc/sysdeps/linux/kvx/Makefile.arch
new file mode 100644
index 000000000..3ad290915
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/Makefile.arch
@@ -0,0 +1,10 @@
+# Makefile for uClibc
+#
+# Copyright (C) 2000-2005 Erik Andersen <andersen@uclibc.org>
+#
+# Licensed under the LGPL v2.1, see the file COPYING.LIB in this tarball.
+#
+
+CSRC-y := __syscall_error.c
+CSRC-$(UCLIBC_LINUX_SPECIFIC) += cachectl.c
+SSRC-y := setjmp.S bsd-setjmp.S bsd-_setjmp.S __longjmp.S clone.S vfork.S
diff --git a/libc/sysdeps/linux/kvx/__longjmp.S b/libc/sysdeps/linux/kvx/__longjmp.S
new file mode 100644
index 000000000..fbfefe81c
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/__longjmp.S
@@ -0,0 +1,53 @@
+/*
+ * This file is subject to the terms and conditions of the LGPL V2.1
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2018 Kalray Inc.
+ */
+
+#include <sysdep.h>
+#define _SETJMP_H
+#define _ASM
+#include <bits/setjmp.h>
+#include <libc-symbols.h>
+
+/**
+ * void __longjmp(__jmp_buf __env, int __val)
+ */
+ENTRY(__longjmp)
+	/* Load $ra and $csinto r40r41 */
+	lq $r40r41 = JMPBUF_RA_CS_OFFSET[$r0]
+	;;
+	/* Load $r36r37r38r39 with r12(sp) r14 r18 r19 */
+	lo $r36r37r38r39 = (JMPBUF_REGS_OFFSET)[$r0]
+	set $ra = $r40
+	;;
+	/* Load $lc, $le and $ls */
+	lo $r32r33r34r35 = JMPBUF_LC_LE_LS_OFFSET[$r0]
+	copyd $sp = $r36
+	copyd $r14 = $r37
+	set $cs = $r41
+	;;
+	/* Load r20r21r22r23 */
+	lo $r20r21r22r23 = (JMPBUF_REGS_OFFSET + QUAD_REG_SIZE)[$r0]
+	copyd $r18 = $r38
+	copyd $r19 = $r39
+	set $lc = $r32
+	;;
+	/* Load r24r25r26r27 */
+	lo $r24r25r26r27 = (JMPBUF_REGS_OFFSET + 2 * QUAD_REG_SIZE)[$r0]
+	set $le = $r33
+	;;
+	/* Load r28r29r30r31 */
+	lo $r28r29r30r31 = (JMPBUF_REGS_OFFSET + 3 * QUAD_REG_SIZE)[$r0]
+	set $ls = $r34
+	/* Copy retval */
+	copyd $r0 = $r1
+	;;
+	/* According to man, if retval is equal to 0, then we should return 1 */
+	cmoved.deqz $r0? $r0 = 1
+	ret
+	;;
+END(__longjmp)
+libc_hidden_def(__longjmp)
diff --git a/libc/sysdeps/linux/kvx/__syscall_error.c b/libc/sysdeps/linux/kvx/__syscall_error.c
new file mode 100644
index 000000000..d534ee7d4
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/__syscall_error.c
@@ -0,0 +1,19 @@
+/*
+ * This file is subject to the terms and conditions of the LGPL V2.1
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2018 Kalray Inc.
+ */
+
+#include <errno.h>
+#include <features.h>
+
+/* This routine is jumped to by all the syscall handlers, to stash
+ * an error number into errno.  */
+long __syscall_error(int err_no) attribute_hidden;
+long __syscall_error(int err_no)
+{
+	__set_errno(-err_no);
+	return -1;
+}
diff --git a/libc/sysdeps/linux/kvx/bits/atomic.h b/libc/sysdeps/linux/kvx/bits/atomic.h
new file mode 100644
index 000000000..3c423e9ba
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/bits/atomic.h
@@ -0,0 +1,141 @@
+/* Copyright (C) 2010-2012 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Maxim Kuvyrkov <maxim@codesourcery.com>, 2010.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _KVX_BITS_ATOMIC_H
+#define _KVX_BITS_ATOMIC_H
+
+#include <stdint.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 int16_t atomic16_t;
+typedef uint16_t uatomic16_t;
+typedef int_fast16_t atomic_fast16_t;
+typedef uint_fast16_t uatomic_fast16_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 int64_t atomic64_t;
+typedef uint64_t uatomic64_t;
+typedef int_fast64_t atomic_fast64_t;
+typedef uint_fast64_t uatomic_fast64_t;
+
+typedef intptr_t atomicptr_t;
+typedef uintptr_t uatomicptr_t;
+typedef intmax_t atomic_max_t;
+typedef uintmax_t uatomic_max_t;
+
+
+#ifndef atomic_full_barrier
+# define atomic_full_barrier() do { atomic_read_barrier();		\
+					atomic_write_barrier(); } while(0)
+#endif
+
+#ifndef atomic_read_barrier
+# define atomic_read_barrier() __builtin_kvx_dinval()
+#endif
+
+#ifndef atomic_write_barrier
+# define atomic_write_barrier() __builtin_kvx_fence()
+#endif
+
+/*
+ * On kvx, we have a boolean compare and swap which means that the operation
+ * returns only the success of operation.
+ * If operation succeeds, this is simple, we just need to return the provided
+ * old value. However, if it fails, we need to load the value to return it for
+ * the caller. If the loaded value is different from the "old" provided by the
+ * caller, we can return it since it will mean it failed.
+ * However, if for some reason the value we read is equal to the old value
+ * provided by the caller, we can't simply return it or the caller will think it
+ * succeeded. So if the value we read is the same as the "old" provided by
+ * the caller, we try again until either we succeed or we fail with a different
+ * value than the provided one.
+ */
+#define __cmpxchg(ptr, old, new, op_suffix, load_suffix)		\
+({									\
+	register unsigned long __rn __asm__("r62");			\
+	register unsigned long __ro __asm__("r63");			\
+	__asm__ __volatile__ (						\
+		/* Fence to guarantee previous store to be committed */	\
+		"fence\n"						\
+		/* Init "expect" with previous value */			\
+		"copyd $r63 = %[rOld]\n"				\
+		";;\n"							\
+		"1:\n"							\
+		/* Init "update" value with new */			\
+		"copyd $r62 = %[rNew]\n"				\
+		";;\n"							\
+		"acswap" #op_suffix " 0[%[rPtr]], $r62r63\n"		\
+		";;\n"							\
+		/* if acswap succeeds, simply return */			\
+		"cb.dnez $r62? 2f\n"					\
+		";;\n"							\
+		/* We failed, load old value */				\
+		"l"  #op_suffix  #load_suffix" $r63 = 0[%[rPtr]]\n"	\
+		";;\n"							\
+		/* Check if equal to "old" one */			\
+		"compd.ne $r62 = $r63, %[rOld]\n"			\
+		";;\n"							\
+		/* If different from "old", return it to caller */	\
+		"cb.deqz $r62? 1b\n"					\
+		";;\n"							\
+		"2:\n"							\
+		: "+r" (__rn), "+r" (__ro)				\
+		: [rPtr] "r" (ptr), [rOld] "r" (old), [rNew] "r" (new)	\
+		: "memory");						\
+	(__ro);								\
+})
+
+#define cmpxchg(ptr, o, n)						\
+({									\
+	unsigned long __cmpxchg__ret;					\
+	switch (sizeof(*(ptr))) {					\
+	case 4:								\
+		__cmpxchg__ret = __cmpxchg((ptr), (o), (n), w, s);	\
+		break;							\
+	case 8:								\
+		__cmpxchg__ret = __cmpxchg((ptr), (o), (n), d, );	\
+		break;							\
+	}								\
+	(__typeof(*(ptr))) (__cmpxchg__ret);				\
+})
+
+#define atomic_compare_and_exchange_val_acq(mem, newval, oldval)	\
+	cmpxchg((mem), (oldval), (newval))
+
+
+#define atomic_exchange_acq(mem, newval)				\
+({									\
+	unsigned long __aea__ret, __aea__old;					\
+	volatile __typeof((mem)) __aea__m = (mem);				\
+	do {								\
+		__aea__old = *__aea__m;						\
+		__aea__ret = atomic_compare_and_exchange_val_acq((mem),	\
+						(newval), (__aea__old));\
+	} while (__aea__old != __aea__ret);					\
+	(__aea__old);							\
+})
+
+#endif
diff --git a/libc/sysdeps/linux/kvx/bits/endian.h b/libc/sysdeps/linux/kvx/bits/endian.h
new file mode 100644
index 000000000..03a1b7f0c
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/bits/endian.h
@@ -0,0 +1,13 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2018 Kalray Inc.
+ */
+
+#ifndef _ENDIAN_H
+# error "Never use <bits/endian.h> directly; include <endian.h> instead."
+#endif
+
+#define __BYTE_ORDER __LITTLE_ENDIAN
diff --git a/libc/sysdeps/linux/kvx/bits/fcntl.h b/libc/sysdeps/linux/kvx/bits/fcntl.h
new file mode 100644
index 000000000..c1815b44f
--- /dev/null
+++ b/libc/sysdeps/linux/kvx/bits/fcntl.h
@@ -0,0 +1,226 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2018 Kalray Inc.
+ */
+
+#ifndef	_FCNTL_H
+# error "Never use <bits/fcntl.h> directly; include <fcntl.h> instead."
+#endif
+
+#include <sys/types.h>
+#ifdef __USE_GNU
+# include <bits/uio.h>
+#endif
+
+/* open/fcntl - O_SYNC is only implemented on blocks devices and on files
+   located on an ext2 file system */
+#define O_ACCMODE	  0003
+#define O_RDONLY	    00
+#define O_WRONLY	    01
+#define O_RDWR		    02
+#define O_CREAT		  0100	/* not fcntl */
+#define O_EXCL		  0200	/* not fcntl */
+#define O_NOCTTY	  0400	/* not fcntl */
+#define O_TRUNC		 01000	/* not fcntl */
+#define O_APPEND	 02000
+#define O_NONBLOCK	 04000
+#define O_NDELAY	O_NONBLOCK
+#define O_SYNC		010000
+#define O_FSYNC		O_SYNC
+#define O_ASYNC		020000
+
+#ifdef __USE_XOPEN2K8
+# define O_DIRECTORY   0200000	/* Must be a directory.	 */
+# define O_NOFOLLOW    0400000	/* Do not follow links.	 */
+# define O_CLOEXEC    02000000	/* Set close_on_exec.  */
+#endif
+
+#ifdef __USE_GNU
+# define O_DIRECT	040000	/* Direct disk access.	*/
+# define O_NOATIME    01000000	/* Do not set atime.  */
+# define O_PATH	     010000000  /* Resolve pathname but do not open file.  */
+#endif
+
+#ifdef __USE_LARGEFILE64
+# define O_LARGEFILE   0100000
+#endif
+
+/* For now Linux has synchronisity options for data and read operations.
+   We define the symbols here but let them do the same as O_SYNC since
+   this is a superset.	*/
+#if defined __USE_POSIX199309 || defined __USE_UNIX98
+# define O_DSYNC	O_SYNC	/* Synchronize data.  */
+# define O_RSYNC	O_SYNC	/* Synchronize read operations.	 */
+#endif
+
+/* Values for the second argument to `fcntl'.  */
+#define F_DUPFD		0	/* Duplicate file descriptor.  */
+#define F_GETFD		1	/* Get file descriptor flags.  */
+#define F_SETFD		2	/* Set file descriptor flags.  */
+#define F_GETFL		3	/* Get file status flags.  */
+#define F_SETFL		4	/* Set file status flags.  */
+
+#define F_GETLK		5	/* Get record locking info.  */
+#define F_SETLK		6	/* Set record locking info (non-blocking).  */
+#define F_SETLKW	7	/* Set record locking info (blocking).	*/
+
+/* Same as standard, since we always have 64-bit offsets.  */
+#define F_GETLK64	F_GETLK		/* Get record locking info.  */
+#define F_SETLK64	F_SETLK		/* Set record locking info (non-blocking).  */
+#define F_SETLKW64	F_SETLKW	/* Set record locking info (blocking).	*/
+
+#if defined __USE_BSD || defined __USE_XOPEN2K
+# define F_SETOWN	8	/* Get owner of socket (receiver of SIGIO).  */
+# define F_GETOWN	9	/* Set owner of socket (receiver of SIGIO).  */
+#endif
+
+#ifdef __USE_GNU
+# define F_SETSIG	10	/* Set number of signal to be sent.  */
+# define F_GETSIG	11	/* Get number of signal to be sent.  */
+#endif
+
+#ifdef __USE_GNU
+# define F_SETLEASE	1024	/* Set a lease.	 */
+# define F_GETLEASE	1025	/* Enquire what lease is active.  */
+# define F_NOTIFY	1026	/* Request notfications on a directory.	 */
+# define F_DUPFD_CLOEXEC 1030	/* Duplicate file descriptor with
+				   close-on-exit set on new fd.  */
+# define F_SETPIPE_SZ	1031    /* Set pipe page size array.  */
+# define F_GETPIPE_SZ	1032    /* Get pipe page size array.  */
+#endif
+
+/* For F_[GET|SET]FL.  */
+#define FD_CLOEXEC	1	/* actually anything with low bit set goes */
+
+/* For posix fcntl() and `l_type' field of a `struct flock' for lockf().  */
+#define F_RDLCK		0	/* Read lock.  */
+#define F_WRLCK		1	/* Write lock.	*/
+#define F_UNLCK		2	/* Remove lock.	 */
+
+/* For old implementation of bsd flock().  */
+#define F_EXLCK		4	/* or 3 */
+#define F_SHLCK		8	/* or 4 */
+
+#ifdef __USE_BSD
+/* Operations for bsd flock(), also used by the kernel implementation.	*/
+# define LOCK_SH	1	/* shared lock */
+# define LOCK_EX	2	/* exclusive lock */
+# define LOCK_NB	4	/* or'd with one of the above to prevent
+				   blocking */
+# define LOCK_UN	8	/* remove lock */
+#endif
+
+#ifdef __USE_GNU
+# define LOCK_MAND	32	/* This is a mandatory flock:	*/
+# define LOCK_READ	64	/* ... which allows concurrent read operations.	 */
+# define LOCK_WRITE	128	/* ... which allows concurrent write operations.  */
+# define LOCK_RW	192	/* ... Which allows concurrent read & write operations.	 */
+#endif
+
+#ifdef __USE_GNU
+/* Types of directory notifications that may be requested with F_NOTIFY.  */
+# define DN_ACCESS	0x00000001	/* File accessed.  */
+# define DN_MODIFY	0x00000002	/* File modified.  */
+# define DN_CREATE	0x00000004	/* File created.  */
+# define DN_DELETE	0x00000008	/* File removed.  */
+# define DN_RENAME	0x00000010	/* File renamed.  */
+# define DN_ATTRIB	0x00000020	/* File changed attibutes.  */
+# define DN_MULTISHOT	0x80000000	/* Don't remove notifier.  */
+#endif
+
+struct flock
+  {
+    short int l_type;	/* Type of lock: F_RDLCK, F_WRLCK, or F_UNLCK.	*/
+    short int l_whence;	/* Where `l_start' is relative to (like `lseek').  */
+#ifndef __USE_FILE_OFFSET64
+    __off_t l_start;	/* Offset where the lock begins.  */
+    __off_t l_