[lttng-dev] [PATCH 02/11] urcu/uatomic: Use atomic builtins if configured

[Mathieu Desnoyers]

On 6/21/23 19:19, Paul E. McKenney wrote:
[...]
>> diff --git a/include/urcu/uatomic/builtins-generic.h b/include/urcu/uatomic/builtins-generic.h
>> new file mode 100644
>> index 0000000..8e6a9b5
>> --- /dev/null
>> +++ b/include/urcu/uatomic/builtins-generic.h
>> @@ -0,0 +1,85 @@
>> +/*
>> + * urcu/uatomic/builtins-generic.h
>> + *
>> + * Copyright (c) 2023 Olivier Dion <odion at efficios.com>
>> + *
>> + * 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
>> + */
>> +
>> +#ifndef _URCU_UATOMIC_BUILTINS_GENERIC_H
>> +#define _URCU_UATOMIC_BUILTINS_GENERIC_H
>> +
>> +#include <urcu/system.h>
>> +
>> +#define uatomic_set(addr, v) __atomic_store_n(addr, v, __ATOMIC_RELAXED)
>> +
>> +#define uatomic_read(addr) __atomic_load_n(addr, __ATOMIC_RELAXED)
> 
> Does this lose the volatile semantics that the old-style definitions
> had?
> 

Yes.

[...]

>> +++ b/include/urcu/uatomic/builtins-x86.h
>> @@ -0,0 +1,85 @@
>> +/*
>> + * urcu/uatomic/builtins-x86.h
>> + *
>> + * Copyright (c) 2023 Olivier Dion <odion at efficios.com>
>> + *
>> + * 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
>> + */
>> +
>> +#ifndef _URCU_UATOMIC_BUILTINS_X86_H
>> +#define _URCU_UATOMIC_BUILTINS_X86_H
>> +
>> +#include <urcu/system.h>
>> +
>> +#define uatomic_set(addr, v) __atomic_store_n(addr, v, __ATOMIC_RELAXED)
>> +
>> +#define uatomic_read(addr) __atomic_load_n(addr, __ATOMIC_RELAXED)
> 
> And same question here.

Yes, this opens interesting questions:

* what semantic do we want for uatomic_read/set ?

* what semantic do we want for CMM_LOAD_SHARED/CMM_STORE_SHARED ?

* do we want to allow load/store-shared to work on variables larger than 
a word ? (e.g. on a uint64_t on a 32-bit architecture, or on a structure)

* what are the guarantees of a volatile type ?

* what are the guarantees of a load/store relaxed in C11 ?

Does the delta between volatile and C11 relaxed guarantees matter ?

Is there an advantage to use C11 load/store relaxed over volatile ? 
Should we combine both C11 load/store _and_ volatile ? Should we use 
atomic_signal_fence instead ?

Thanks,

Mathieu

> 
> 							Thanx, Paul
> 
>> +

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com