On Tue, 19 Nov 2024, Corinna Vinschen wrote: > On Nov 18 10:10, Jeremy Drake via Cygwin-patches wrote: > > On Mon, 18 Nov 2024, Corinna Vinschen wrote: > > > > > Neat, but if this only affects the ARM64 emulation, shouldn't this only > > > be called under wincap.cpu_arch() == PROCESSOR_ARCHITECTURE_AMD64? > > > > Wouldn't this always be true though? > > Copy/Paste and not thinking straight is doing that for me, sorry. > > The above was supposed to be PROCESSOR_ARCHITECTURE_ARM64, but this is > obviously dumb either way. What I *meant* was checking if we're running > in an emulation vias GetNativeSystemInfo or IsWow64Process2. We're > using the latter in find_fast_cwd() already.
GetNativeSystemInfo also lies, following the Microsoft compatibility theory of: make existing APIs lie, add a new API that gives the real answer, repeat for new versions. That's why we need to use IsWow64Process2: that's the API that (currently) doesn't lie. > > (Except that I backported this to > > 3.3.6 for i686 support, where I'd have to check for that possibility as > > well). > > Have I missed something? I thought this only occurs in an > AMD64-on-ARM64 environment. What is it used for on i686? This also occurs on i686-on-ARM64, but since Cygwin no longer supports i686 I didn't get into that here. Actually, I was only able to get a debugger to give me any info about where this thread was stuck on the i686-on-ARM64 case. In retrospect, this was presumably so hard to debug for the same reason as this workaround works: SuspendThread/GetThreadContext tries to get out of emulation before returning, but the emulator is waiting on a lock that will never be released because the thread that owned it was terminated. > > > Is this just to avoid the code when a native ARM64 Cygwin appears? > > No, the idea was running the code only if we know we're being emulated > on ARM64. I just screwed up :} > > > I have been sort of considering if the results from IsWow64Process2 should > > be cached in wincap, then we could check that here if we are running on > > ARM64 under emulation, and also used the cached value in the check around > > FAST_CWD. In addition, I was thining it might make sense to include this > > info in `uname -s` like -WOW64 used to be when i686 was supported. > > That might be a good idea in the long run and patches are welcome. > Right now we have this isolated usage of IsWow64Process2 in find_fast_cwd(). > Given this function is called at least once in a process tree anyway, > wincap would be a natural place to keep the info. > > And yeah, maybe we can just attach "-ARM64" to the -s info or something. I'll save that for a future patch, and for now send v3 without an additional check for architecture.
