How can I find out which processor architectures are supported via emulation by the current system?
A customer was writing a debugging tool and wanted to know in their installer which processor architectures are supported by the current system, both the native architecture as well as anything supported by emulation. That way, the tool could install the versions that apply to those architectures.
Okay, getting the native architecture is easy. You can call GetÂNativeÂSystemÂInfo to find out what the native system architecture is.
Getting the emulated ones is a little trickier. Back in the old days, you just hard-coded some knowledge. “Well, Windows on x86-64 can emulate x86-32. And Windows on AArch64 can emulate x86-32 as well as T32.” But that broke down with the introduction of HoloLens 2, which is natively AArch64 and omits the x86-32 emulation, and then broke down further with the introduction of x86-64 emulation on native AArch64 systems.
To find out whether a particular architecture is supported in emulation, you can pass it to IsÂWow64ÂGuestÂMachineÂSupported and see what the answer is. Repeat for each architecture you are curious about.
There is no way to get a list of all the architectures supported by emulation, primarily because there’s no real use for it. I mean, suppose you learn that Windows now supports architecture number 31415. You have no idea what that number means, because it’s not on your list of known architectures. Even if you somehow figured out that 31415 is IMAGE_, that still does you no good because you don’t have any binaries that support the SuperMegaGreat. Just enumerate through all the ones you support and intersect that with the ones that the operating system supports, and that gives you the ones for which you should install your architecture-specific packages.
Bonus chatter: Sometimes, the kernel folks get a little cute when picking the ID numbers for new architectures. Here are the ones I was able to guess at:
| Architecture | Value (hex) | Proposed explanation |
|---|---|---|
| CEF | 0x0CEF |
Hex digits spell out name |
| CEE | 0xC0EE |
Why not 0x0CEE? |
| EBC | 0x0EBC |
Hex digits spell out name |
| AMD64 | 0x8664 |
Also known as x86–64 |
| ARM64 | 0xAA64 |
Also known as AArch64 |
| MIPS16 | 0x0266 |
There’s clearly something going on here but I don’t know what |
| MIPSFPU | 0x0366 |
|
| MIPSFPU16 | 0x0466 |
Light
Dark
17 comments
This is sadly incorrect, Microsoft did a little too much compatibility fixing.
GetÂNativeÂSystemÂInfo lies on ARM64.
IsÂWow64ÂGuestÂMachineÂSupported only returns 32-bit guests, not AMD64 on ARM64.
IsWow64Process2 tells the truth.
Compatibility has always been #1 priority for Microsoft, for example the location of system dlls:
16-bit Windows –> C:\Windows\System
32-bit Windows –> C:\Windows\System32
64-bit Windows –> C:\Windows\System32
32-bit dlls on 64-bit Windows -> C:\Windows\SysWOW64
I think any other company would have used C:\Windows\System64 for 64-bit Windows dlls…
If Microsoft had done that, they would have gotten lots of complains from users:
“Why doesn’t my app work on the new 64-bit Windows?”
“Sorry, the app has a bug and incorrectly hard-codes the wrong file path that happened to work on 32-bit Windows but doesn’t work any more on 64-bit WIndows. You should ask the developer to fix that.”
“They went out of business 5 years ago.”
Microsoft did they right thing for compatibility.
I’ve never quite understood what scenario would break here. I mean, it would have to be a 32-bit app, right? So c:\windows\system32 would still be the right path for it to find a DLL, even if the 64-bit apps were using c:\windows\system64.
Perhaps it was more about making it easier to port apps from 32-bit to 64-bit?
There are places that store command lines without any associated bitness. E.g. batch files, command lines in the registry. A batch file that runs C:\Windows\System32\notepad.exe, expects to get the native notepad.exe (not an emulated one that may suffer registry or file system redirection).
Note that notepad.exe (the native one) is always also stored in C:\Windows I think for this purpose.
Anyways, I think it’s tricky to explain for the layman why 64-bit stuff is stored in a xxx32 directory and 32-bit stuff is stored in a xxx64 directory.
this comment has been deleted.
It helps explain to users (if they even notice) that the xxx placeholders in your example paths are not the same and thus the example isn’t useful. Better to use xxx32 and yyy64 as the start of the discussion.
Explorer is a 64-bit app and you’re looking at the 64 bit view of the world, yada yada. System files are just in this thing called System32 for back compat reasons. The 32-bit ones need to be somewhere so they’re in the “windows (32) on Windows 64” system files folder, etc.
Moving from win16 to win32 was plainly necessary and made dev lives easier.
Moving from 32 to 64 for regular apps isn’t as necessary so better side-by-side and shim stuff was required.
It will even lie, if you are x86-64 program running on AArch64 and will attempt to track architecture (.Machine) word through HMODULE -> IMAGE_DOS_HEADER -> IMAGE_NT_HEADERS -> .Machine. It will lie that it’s IMAGE_FILE_MACHINE_AMD64 even if it actually is ARM64 kernel32.dll or any other system DLL. Likely an artifact of how the ARM64X/ARM64EC works, but still.
There is no lie there. The machine member is the PE file declaring which CPU it was compiled for. This member will always match the .exe. A WoW64 process for example will have a 32-bit ntdll.dll loaded but it will also have a hidden 64-bit native ntdll.dll in the process. ARM64 blurs the lines a little bit with CHPE but the application should not notice.
X64 emulation on ARM64 doesn’t come with it’s own set of DLLs. Also it’s not WoW64, therefore no fs/registry redirections etc.
Instead there are extra sections in the native AArch64 DLLs that assist the emulator when jumping to/from native code. The loader probably patches the header in memory, if I query the module path, open it as a file, then I get the proper 0xAA64.
No of course it’s not WoW64 but that is the only example I know of where the machine member does not match the .exe. And yes, on ARM 64 the PE header and import table are patched when loaded into a AMD64 process.
Speculation…
i386 = 0x014C = 332, as in “386, 32 bits”
From then on, 0x016x is MIPS, 0x018x is Alpha, 0x01Ax is SH3, 0x01Cx is ARM…
Then 64 bits is introduced with 0x0200, where 0x02xx means 64-bits. So 0x026x is MIPS-64, 0x028x is Alpha-64, etc.
Then they got creative. 😉
What are CEE and CEF? I’ve tried Googling but I don’t see anything that looks plausible.
I’m pretty sure they’re .NET image formats. There’s some reference to CEE being the “CLR pure MSIL” format early on.
Without context… there was CEF as in Common Executable Format. This was just before DotNet came around, so early 2000s. A WinCE sort of thing, where you had mips, shx, arm, x86? and who wants a binary for each? CEF was scraped for DotNet. Google should have something.
The Windows 11 API
is the new way to check. It is the only way to detect AMD64 support on ARM64 devices, but can also be used to detect all of the cases that
supports. And if it is important to you, you can even learn whether the given architecture is supported in user/kernel/via WoW64 mechanism.