Re: [PATCH v2] Cygwin: autoload: Add AArch64 implementation

On 04/06/2026 08:35, Chandru Kumaresan wrote:
> Hi Evgeny,
> 
>> LoadDLLprime contains only data. It might make sense to keep only
>> one version for x86_64 and aarch64, and use WORD64 for .quad/.xword.
> 
> In v2, we introduced a WORD64 macro:
> 
> #if defined(__x86_64__)
> # define WORD64 ".quad"
> #elif defined(__aarch64__)
> # define WORD64 ".xword"
> #else
> # error unimplemented for this target
> #endif
> 
> And unified LoadDLLprime into a single shared definition using WORD64,
> removing the duplicate x86_64/aarch64 versions.
> Also added inline comments throughout AArch64 assembly blocks
> documenting execution flow.
> 
> Please let me know if anything needs further clarification.

Thanks very much!

(and thanks for adding copious comments on the assembly! that really helps!)

I've applied this patch.

I did a little burnishing to update the introductory comments (dropping 
the obsolete 32-bit sizings for the fields), but this is all 
sufficiently convoluted I'm not 100% sure I've got it all right! :)

I have to say, I'm not 100% convinced that the "elide the call to 
LoadLibrary() for subsequent symbols in the same DLL" functionality 
alluded to in the comment is actually there, but perhaps I need to do a 
bit more staring at the code...


Just so I'm doing my due diligence, I'd like you confirm that you are 
submitting this under an open source license as per [1], and you and 
your colleagues at multicoreware are authorized to do so.

[1] https://cygwin.com/contrib/dll.html



Some notes on possible future work I made in the course of my review, 
saved here:

* Looking at the git history, the comment on the problem the 
no_resolve_on_fork flag is working around has disappeared, but the 
functionality is still there.  Is it still needed?

* Since it's just data, it seems to me that the initializations of the 
various instances of struct dll_info could actually all be written in C.