Why does Box64 tracing stop after my_execv() on x86_64?

[parkjunhyung2021046762]

Hi,

I'm investigating why instruction tracing in Box64 stops after my_execv() when running Wine on an x86_64 Linux system.

I modified Box64 to collect instruction traces, and I confirmed that the trace always ends immediately after my_execv(). I also verified that this is not caused by fork().

My current understanding is:

Box64 emulates only the initial Wine launcher.
The launcher eventually calls my_execv().
After execve(), the Box64 process image is replaced, so the tracing code inside Box64 disappears.
From that point on, Wine continues to execute natively on an x86_64 host, so Box64 is no longer involved.

However, I observed something that I don't fully understand.

Even when running a 64-bit Windows executable (hello.exe, PE32+), the first executable after my_execv() is:

.../i386-unix/wine-preloader

After that, Wine eventually executes the normal 64-bit components.

My questions are:

Why does Wine execute i386-unix/wine-preloader even when the target program is a 64-bit executable?
Is this expected bootstrap behavior in modern Wine?
When i386-unix/wine-preloader is executed on an x86_64 host, is it still running completely natively (without any further Box64 involvement)?
Is it therefore expected that, after my_execv(), Box64 will never regain control on x86_64, even though i386-unix/wine-preloader appears in the execution chain?

I want to make sure my understanding of Box64's control flow is correct.

Thanks!

[ptitSeb]

This looks like a duplicate of #3947