The way to Unity, pt.3

[polybiusproxy]

This should be the last of my PR series regarding Unity. The last main thing to implement is exception support.

To emulate POSIX's behavior, we will use special user APCs, which will allow us to 'inject' code on a remote thread, regardless of what the thread was executing—when the APC returns, the thread goes back to what it was executing.
The only caveat is that APCs won't execute if the thread is on a non-alertable wait at the time of the APC raise. This is why we did a rework of sync primitives in part 2, using wrappers that use Win32 sync primitives and WaitForSingleObjectEx.
To make use of APCs, we use NtQueueApcThreadEx and rely on its undocumented behavior to have x86 context on the APC routine as well (thanks to red_prig and Roamic).

In addition, this PR does some additional fixes on the linker to improve Unity games further—if sceKernelLoadStartModule is called more than once for a module, a handle will be returned instead of loading and calling the module's module_start function again. This fixes both crashes and race conditions on later Unity games.
Event flag cancelling is also implemented as seen on Unity (makes no effect, but it's nice to have).

[ElBread3]

So far most of my Unity games either crash at net calls, PromoteFormatToDepth unreachable, or "existing mapping does not contain requested unmap range". Again, could you please add the missing mappings for the pthread sem functions?
Interesting, one of them also spams "Stub: sigprocmask (nid: aPcyptbOiZs)" while booting.

[ElBread3]

Now that that's fixed, immediately crashes on JoinRegionsAfterUnmap.

[Kernel.Vmm] memory.cpp:Free:141: Unmaping direct mapping 0x1f77c0000 with size 0x100000
[Core] stubs.cpp:CommonStub:42: Stub: sceMouseRead (nid: x8qnXqh-tiM) called, returning zero to 0x900dc50ee
[Kernel.Vmm] memory.cpp:sceKernelMunmap:493: addr = 0x1f77c0000, len = 0x100000
[Debug] address_space.cpp:operator():268: Assertion Failed!
Invalid address/size given to unmap.

Additionally, a different variant of the error from the Unmap function:

[Kernel.Vmm] memory.cpp:Free:141: Unmaping direct mapping 0x1fe100000 with size 0xd40000
[Kernel.Vmm] memory.cpp:sceKernelMunmap:493: addr = 0x1fe100000, len = 0xd40000
[Debug] address_space.cpp:operator():182: Assertion Failed!
Unmap operation on virtual_addr=0X1FE100000 failed: Attempt to access invalid address.

[Randomuser8219]

we should rebase this because of the recent huge changes

[GHU7924]

Build f496c6b

CUSA05297 INSIDE

Unlike last time, the game is now starting to load
The game crashed
shad_log_1.txt

The game crashed with an error
shad_log_2.txt

CUSA23081 Untitled Goose Game

Unlike last time, the game is now starting to load
I went to the menu, selected a slot to save, the game started loading, but it looks like either it's an endless download or a crash
shad_log_3.txt
shad_log_4.txt

[5jd]

Build f496c6b

Race The Sun CUSA00708 01.00

Now boots but loading is stuck on

[Core] <Error> stubs.cpp:CommonStub:42: Stub: sceMoveReadStateRecent (nid: f2bcpK6kJfg) called, returning zero to 0x9000009c2

shad_log.txt

[GHU7924]

I am updating the report on my two games on the Unity engine.

Build 5be807f

CUSA05297 INSIDE

[Debug] liverpool_to_vk.h:PromoteFormatToDepth:107: Unreachable code!

CUSA23081 Untitled Goose Game

Now you can play.

[Inui-senpai]

Unmap operation on virtual_addr=0X1FE100000 failed: Attempt to access invalid address.

DDLC+ crashes with the same error (but with the other virtual address). Maybe we should create an another "[Aggregate Issue]" issue. 🤔

Just in case (used build 6d728ec):

[Debug] <Critical> address_space.cpp:operator():182: Assertion Failed!
Unmap operation on virtual_addr=0X1F5C80000 failed: Attempt to access invalid address.

shad_log.txt