[PATCH 0/1] MR9194: ntdll: Use ReadPointerAcquire to get pointer value in RtlRunOnceComplete.

[Jinoh Kang (＠iamahuman)]

This is needed on architectures with weak memory models like ARM64.

Complements !9178

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/9194

[PATCH 1/1] ntdll: Use ReadPointerAcquire to get pointer value in RtlRunOnceComplete.

[Jinoh Kang]

From: Jinoh Kang <jinoh.kang.kr(a)gmail.com>

This is needed on architectures with weak memory models like ARM64.
---
 dlls/ntdll/sync.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/dlls/ntdll/sync.c b/dlls/ntdll/sync.c
index 3f8aa0b399d..1bd62612efe 100644
--- a/dlls/ntdll/sync.c
+++ b/dlls/ntdll/sync.c
@@ -116,7 +116,7 @@ DWORD WINAPI RtlRunOnceComplete( RTL_RUN_ONCE *once, ULONG flags, void *context
 
     for (;;)
     {
-        ULONG_PTR val = (ULONG_PTR)once->Ptr;
+        ULONG_PTR val = (ULONG_PTR)ReadPointerAcquire( &once->Ptr );
 
         switch (val & 3)
         {
-- 
GitLab

https://gitlab.winehq.org/wine/wine/-/merge_requests/9194

[Luca Bacci (＠lb90)]

Thanks a lot!

Out of curiosity, maybe this could be a `ReadPointerNoFence`? There are three cases:

1. `STATUS_INVALID_PARAMETER` is returned: this is a programmer error, and no pointer is returned.
2. `InterlockedCompareExchangePointer` is called, which inserts a full memory fence.
3. `STATUS_UNSUCCESSFUL` is returned: the caller should retrieve the value that "won" by calling `RtlRunOnceBeginInitialize` with `RTL_RUN_ONCE_CHECK_ONLY`, which uses acquire semantics.

That said, it's _safer_ to ensure acquire semantics in a central place and not in each code branch. So probably this is best for long-term code maintenance?

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/9194#note_118883

[Jinoh Kang (＠iamahuman)]

On Fri Oct 17 12:51:04 2025 +0000, Luca Bacci wrote:
> Thanks a lot!
> Out of curiosity, maybe this could be a `ReadPointerNoFence`? There are
> three cases:
> 1. `STATUS_INVALID_PARAMETER` is returned: this is a programmer error,
> and no pointer is returned.
> 2. `InterlockedCompareExchangePointer` is called, which inserts a full
> memory fence.
> 3. `STATUS_UNSUCCESSFUL` is returned: the caller should retrieve the
> value that "won" by calling `RtlRunOnceBeginInitialize` with
> `RTL_RUN_ONCE_CHECK_ONLY`, which uses acquire semantics.
> That said, it's _safer_ to ensure acquire semantics in a central place
> and not in each code branch. So probably `ReadPointerAcquire` is better
> for code maintenance in the long term?
> Thanks for your comment!

Ideally the context should contain sll that is needed, but what if the "won" value is stored somewhere else?

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/9194#note_118884