[ltt-dev] [PATCH RFC] Native ARM7L support
Mathieu Desnoyers
mathieu.desnoyers at efficios.com
Sun Jun 13 18:01:29 EDT 2010
* Paul E. McKenney (paulmck at linux.vnet.ibm.com) wrote:
> The machine I am using is cortex-a9. It may be necessary to create a
> more intelligent mapping to the many variants of ARM. Interestingly
> enough, __sync_synchronize() -still- fails, so use an explicit asm
> for the dmb instruction.
>
> Leaving off the '-O' gives freaking slow results. ;-)
Is this -O problem something we should fix at an higher level ? (e.g. for all
arch ?)
More comments below,
>
> Signed-off-by: Paul E. McKenney <paulmck at linux.vnet.ibm.com>
> ---
>
> configure.ac | 4 +++
> urcu/arch_armv7l.h | 59 +++++++++++++++++++++++++++++++++++++++++++++
> urcu/uatomic_arch_armv7l.h | 48 ++++++++++++++++++++++++++++++++++++
> 3 files changed, 111 insertions(+)
>
> diff --git a/configure.ac b/configure.ac
> index 9274337..9e2615f 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -51,6 +51,7 @@ case $host_cpu in
> s390x) ARCHTYPE="s390" ;;
> sparc64) ARCHTYPE="sparc64" ;;
> alpha*) ARCHTYPE="alpha" ;;
> + armv7l) ARCHTYPE="armv7l" ;;
> *) ARCHTYPE="unknown";;
> esac
>
> @@ -61,6 +62,9 @@ if test "x$ARCHTYPE" != xx86 -a "x$ARCHTYPE" != xppc; then
> else
> APISRC=tests/api_$ARCHTYPE.h
> fi
> +if test "x$ARCHTYPE" == xarmv7l; then
> + CFLAGS="-mcpu=cortex-a9 -mtune=cortex-a9 -O2"
> +fi
>
> AC_SUBST(ARCHTYPE)
> AC_SUBST(SUBARCHTYPE)
> diff --git a/urcu/arch_armv7l.h b/urcu/arch_armv7l.h
> new file mode 100644
> index 0000000..5920a4d
> --- /dev/null
> +++ b/urcu/arch_armv7l.h
> @@ -0,0 +1,59 @@
> +#ifndef _URCU_ARCH_ARMV7L_H
> +#define _URCU_ARCH_ARMV7L_H
> +
> +/*
> + * arch_unknown.h: trivial definitions for the "unknown" architecture.
> + *
> + * Copyright (c) 2010 Paul E. McKenney, IBM Corporation.
> + * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers at polymtl.ca>
> + *
> + * This 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.
> + *
> + * This 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 this library; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +#include <urcu/compiler.h>
> +#include <urcu/config.h>
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +/* We don't know, so guess!!! */
> +#define CACHE_LINE_SIZE 128
> +
> +#define mb() asm volatile("dmb":::"memory")
> +
> +#include <stdlib.h>
> +#include <sys/time.h>
> +
> +typedef unsigned long long cycles_t;
> +
> +static inline cycles_t get_cycles (void)
> +{
> + cycles_t thetime;
> + struct timeval tv;
> +
> + if (gettimeofday(&tv, NULL) != 0)
> + return 0;
> + thetime = ((cycles_t)tv.tv_sec) * 1000000ULL + ((cycles_t)tv.tv_usec);
> + return (cycles_t)thetime;
> +}
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#include <urcu/arch_generic.h>
> +
> +#endif /* _URCU_ARCH_ARMV7L_H */
> diff --git a/urcu/uatomic_arch_armv7l.h b/urcu/uatomic_arch_armv7l.h
> new file mode 100644
> index 0000000..5eece49
> --- /dev/null
> +++ b/urcu/uatomic_arch_armv7l.h
> @@ -0,0 +1,48 @@
> +#ifndef _URCU_ARCH_UATOMIC_ARMV7L_H
> +#define _URCU_ARCH_UATOMIC_ARMV7L_H
> +
> +/*
> + * Copyright (c) 1991-1994 by Xerox Corporation. All rights reserved.
> + * Copyright (c) 1996-1999 by Silicon Graphics. All rights reserved.
> + * Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
> + * Copyright (c) 2009 Mathieu Desnoyers
> + * Copyright (c) 2010 Paul E. McKenney, IBM Corporation
> + * (Adapted from uatomic_arch_unknown.h)
> + *
> + * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
> + * OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
> + *
> + * Permission is hereby granted to use or copy this program
> + * for any purpose, provided the above notices are retained on all copies.
> + * Permission to modify the code and to distribute modified code is granted,
> + * provided the above notices are retained, and a notice that the code was
> + * modified is included with the above copyright notice.
> + *
> + * Code inspired from libuatomic_ops-1.2, inherited in part from the
> + * Boehm-Demers-Weiser conservative garbage collector.
> + */
> +
> +#include <urcu/compiler.h>
> +#include <urcu/system.h>
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +/* xchg */
> +#define uatomic_xchg(addr, v) __sync_lock_test_and_set(addr, v);
I expect that these __sync_*() operations have missing memory barriers ? It
might be safer to do something like:
#define uatomic_cmpxchg(addr, old _new) \
{( \
__typeof__(_new) __ret; \
\
smp_mb(); \
__ret = __sync_val_compare_and_swap(addr, old, _new); \
smp_mb(); \
__ret;
)}
(similar for other primitives)
The idea is that I don't trust gcc's __sync_*() implementation if they got the
__sync_synchronize() wrong.
Thanks,
Mathieu
> +
> +/* cmpxchg */
> +#define uatomic_cmpxchg(addr, old, _new) \
> + __sync_val_compare_and_swap(addr, old, _new)
> +
> +/* uatomic_add_return */
> +#define uatomic_add_return(addr, v) __sync_add_and_fetch(addr, v)
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#include <urcu/uatomic_generic.h>
> +
> +#endif /* _URCU_ARCH_UATOMIC_ARMV7L_H */
--
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com
More information about the lttng-dev
mailing list