Update Architecture to unknown machine

[clamp03]

Fix #96978 issue

Part of #84834
cc @dotnet/samsung

[jkotas]

LGTM

[am11]

Looks great!
(one less thing to worry about when adding a new platform support 😅)

[vitek-karas]

/cc @sbomer @agocke

[clamp03]

I will handle failures tomorrow. Thank you.
Or @am11 you can take the issue, close this PR and make new PR. I think you are better than me to solve this issue. :)

[am11]

I was able to locally repro the illink error we are seeing in the CI; it repros with crossbuild+release config. This patch seem to fix it:

--- a/src/coreclr/binder/assemblybindercommon.cpp
+++ b/src/coreclr/binder/assemblybindercommon.cpp
@@ -143,8 +143,7 @@ namespace BINDER_SPACE
         }
         else
         {
-            if ((CLRPeKind & peILonly) && !(CLRPeKind & pe32Plus) &&
-                !(CLRPeKind & pe32BitRequired) && dwImageType == IMAGE_FILE_MACHINE_I386)
+            if ((CLRPeKind & peILonly) && !(CLRPeKind & pe32Plus) && !(CLRPeKind & pe32BitRequired))
             {
                 // Processor-agnostic (MSIL)
                 *PeKind = peMSIL;
diff --git a/src/coreclr/vm/peassembly.cpp b/src/coreclr/vm/peassembly.cpp
index c7c618b4852..99ccb8638f9 100644
--- a/src/coreclr/vm/peassembly.cpp
+++ b/src/coreclr/vm/peassembly.cpp
@@ -39,7 +39,7 @@ static void ValidatePEFileMachineType(PEAssembly *pPEAssembly)
     DWORD actualMachineType;
     pPEAssembly->GetPEKindAndMachine(&peKind, &actualMachineType);

-    if (actualMachineType == IMAGE_FILE_MACHINE_I386 && ((peKind & (peILonly | pe32BitRequired)) == peILonly))
+    if ((peKind & (peILonly | pe32BitRequired)) == peILonly)
         return;    // Image is marked CPU-agnostic.

     if (actualMachineType != IMAGE_FILE_MACHINE_NATIVE && actualMachineType != IMAGE_FILE_MACHINE_NATIVE_NI)

[am11]

Mono has a different behavior than existing coreclr and proposed patch:

runtime/src/mono/mono/metadata/assembly.c

Line 499 in b7c3446

case COFF_MACHINE_I386:

it ignores the "unknown" and "msil" cases for non-i386 archs and simply does not set the machine. Unless there is a downside, we can align that to the proposed behavior as well? cc @jkotas, @lambdageek

[jkotas]

I was able to locally repro the illink error we are seeing in the CI; it repros with crossbuild+release config.

These failures suggests that the fix is not producing proper AnyCPU files. Have you checked that the fix is producing AnyCPU files?

AnyCPU are IMAGE_FILE_MACHINE_I386, COMIMAGE_FLAGS_ILONLY bit set, both COMIMAGE_FLAGS_32BITREQUIRED and COMIMAGE_FLAGS_32BITPREFERRED bits cleared.

I do not think we want to be changing the assembly loader as part of this fix.

[am11]

@jkotas, it is failing to read corelib:

Could not load file or assembly '/__w/1/s/artifacts/bin/trimmingTests/projects/System.ComponentModel.Primitives.TrimmingTests/VerifyCategoryNamesDontGetTrimmed/linux-x64/bin/Release/net9.0/linux-x64/publish/System.Private.CoreLib.dll'. An attempt was made to load a program with an incorrect format.

Shouldn't corelib be machine specific?

before:

Version   : v4.0.30319
CLR Header: 2.5
PE        : PE32+
CorFlags  : 0xc
ILONLY    : 0
32BITREQ  : 0
32BITPREF : 0
Signed    : 1

after:

Version   : v4.0.30319
CLR Header: 2.5
PE        : PE32
CorFlags  : 0x9
ILONLY    : 1
32BITREQ  : 0
32BITPREF : 0
Signed    : 1

If so, then maybe this PR is a no-go (revive dotnet/cecil#171)?

[jkotas]

Shouldn't corelib be machine specific?

CoreLib is machine, OS and runtime specific, but this fact is not encoded in the PE header.

The runtime should not require CoreLib to be machine specific.

[clamp03]

@am11 @jkotas Thank you so much!