Re: [PATCH] eal: add notes to SMP memory barrier APIs

[Mattias Rönnblom]

On 2023-06-25 10:17, Ruifeng Wang wrote:
-----Original Message-----
From: Mattias Rönnblom <hof...@lysator.liu.se>
Sent: Friday, June 23, 2023 2:20 AM
To: Ruifeng Wang <ruifeng.w...@arm.com>; tho...@monjalon.net; 
david.march...@redhat.com
Cc: dev@dpdk.org; konstantin.v.anan...@yandex.ru; Honnappa Nagarahalli
<honnappa.nagaraha...@arm.com>; nd <n...@arm.com>
Subject: Re: [PATCH] eal: add notes to SMP memory barrier APIs
On 2023-06-21 08:44, Ruifeng Wang wrote:
The rte_smp_xx() APIs are deprecated. But it is not mentioned in the
function header.
Added notes in function header for clarification.

Signed-off-by: Ruifeng Wang <ruifeng.w...@arm.com>
---
   lib/eal/include/generic/rte_atomic.h | 15 +++++++++++++++
   1 file changed, 15 insertions(+)

diff --git a/lib/eal/include/generic/rte_atomic.h
b/lib/eal/include/generic/rte_atomic.h
index 58df843c54..542a2c16ff 100644
--- a/lib/eal/include/generic/rte_atomic.h
+++ b/lib/eal/include/generic/rte_atomic.h
@@ -55,6 +55,11 @@ static inline void rte_rmb(void);
    * Guarantees that the LOAD and STORE operations that precede the
    * rte_smp_mb() call are globally visible across the lcores
    * before the LOAD and STORE operations that follows it.
+ *
+ * @note
+ *  This function is deprecated. It adds complexity to the memory
+ model
+ *  used by this project. C11 memory model should always be used.
+ *  rte_atomic_thread_fence() should be used instead.

It's somewhat confusing to learn I should use the C11 memory model, and then in 
the next
sentence that I should call a function which is not in C11.

I should say "memory order semantics". It will be more specific.
The wrapper function rte_atomic_thread_fence is a special case. It provides an 
optimized implementation
for __ATOMIC_SEQ_CST for x86:
https://www.dpdk.org/blog/2021/03/26/dpdk-adopts-the-c11-memory-model/


I think it would be helpful to say which memory_model parameters should be used 
to replace
the rte_smp_*mb() calls, and if there are any difference in semantics between 
the Linux
kernel-style barriers and their C11 (near-)equivalents.

As compiler atomic built-ins are being used. The memory model parameters should 
be the ones listed in:
https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html
We are not taking Linux kernel-style barriers. So no need to mention that.


Yeah, sure. But which one of the C11 memory models, for respective 
legacy barrier?

What you are moving from is Linux kernel-style barriers, so if you are 
to recommend a migration path, their semantics will matter.


Is there some particular reason these functions aren't marked __rte_deprecated? 
Too many
warnings?

Yes, warnings will come up. Some occurrences still remain in the project.


    */
   static inline void rte_smp_mb(void);

@@ -64,6 +69,11 @@ static inline void rte_smp_mb(void);
    * Guarantees that the STORE operations that precede the
    * rte_smp_wmb() call are globally visible across the lcores
    * before the STORE operations that follows it.
+ *
+ * @note
+ *  This function is deprecated. It adds complexity to the memory
+ model
+ *  used by this project. C11 memory model should always be used.
+ *  rte_atomic_thread_fence() should be used instead.
    */
   static inline void rte_smp_wmb(void);

@@ -73,6 +83,11 @@ static inline void rte_smp_wmb(void);
    * Guarantees that the LOAD operations that precede the
    * rte_smp_rmb() call are globally visible across the lcores
    * before the LOAD operations that follows it.
+ *
+ * @note
+ *  This function is deprecated. It adds complexity to the memory
+ model
+ *  used by this project. C11 memory model should always be used.
+ *  rte_atomic_thread_fence() should be used instead.
    */
   static inline void rte_smp_rmb(void);
   ///@}