Re: [Ikvm-developers] Java -> .NET using IKVM, x86 code works x64 doesn't

Hi Nadim,

Thanks for the repro. As I suspected, it is indeed caused by an x64 JIT bug.

Here's a small C# repro:

class Program {
    static void Dummy() { }

    static void Method(int dummy1, int dummy2, int dummy3, int dummy4, int counter) {
        for (int i = 0; i < 10; i++) {
            try {
                Dummy();
            } catch {
            }
            System.Console.WriteLine(counter++);
        }
    }

    public static void Main() {
        Method(0, 0, 0, 0, 0);
    }
}

It no longer repros in the .NET 4.5 developer preview and looking through connect I'm pretty sure this is the same bug:
https://connect.microsoft.com/VisualStudio/feedback/details/636466/variable-is-not-incrementing-in-c-release-x64#details

I could add another workaround to IKVM, but at this point I'd rather just advise people to stay away from the x64 JIT :-(

If you want a workaround for the specific problem you are seeing, you could build IKVM from source and tweak the mgf1 method in 
openjdk7-b147\jdk\src\share\classes\sun\security\rsa\RSAPadding.java to avoid this pattern (e.g. copy the outOfs parameter into a local variable) or add an annotation to the method to disable JIT optimizations:
@cli.System.Runtime.CompilerServices.MethodImplAttribute.Annotation({ cli.System.Runtime.CompilerServices.MethodImplOptions.__Enum.NoOptimization })

Regards,
Jeroen

> -----Original Message-----
> From: Nadim Ratani [mailto:nr...@go...]
> Sent: Tuesday, October 18, 2011 19:40
> To: Jeroen Frijters
> Cc: ikv...@li...
> Subject: Re: [Ikvm-developers] Java -> .NET using IKVM, x86 code works
> x64 doesn't
> 
> Here is a reduced test case. It works as expected in x86 (exits
> successfully) but gives the exception I mentioned in x64.
> 
> 3 attachments in the zip:
> Program.cs (the test case)
> dsa (dsa key in keyczar format)
> rsa (rsa key in keyczar format)
> 
> Regards,
> 
> +n
> Nadim Ratani | Google Developer Relations | nr...@go... | Google
> Inc.
> 
> 
> On Mon, Oct 17, 2011 at 10:56 AM, Nadim Ratani <nr...@go...>
> wrote:
> 
> 
> 	Hi Jeroen,
> 
> 	I am doing something a bit more complicated than the
> straightforward example in the KeyczarDotNet wiki. I am trying to
> reproduce this error in a smaller test case that I can share. FYI I am
> using:
> 	IKVM.NET Compiler version 0.46.0.1
> 
> 	Thank you for the quick response.
> 
> 
> 	+n
> 	Nadim Ratani | Google Developer Relations | nr...@go... |
> Google Inc.
> 
> 
> 	On Sat, Oct 15, 2011 at 1:33 AM, Jeroen Frijters
> <je...@su...> wrote:
> 
> 
> 		Hi Nadim,
> 
> 		That sample on the wiki works for me.
> 
> 		I tried the following combinations:
> 		IKVM 0.46.0.1   .NET 2.0.50727.5448   x86
> 		IKVM 0.46.0.1   .NET 2.0.50727.5448   x64
> 		IKVM 0.46.0.1   .NET 4.0.30319.488    x86
> 		IKVM 0.46.0.1   .NET 4.0.30319.488    x64
> 		IKVM cvs head   .NET 2.0.50727.5448   x86
> 		IKVM cvs head   .NET 2.0.50727.5448   x64
> 		IKVM cvs head   .NET 4.0.30319.488    x86
> 		IKVM cvs head   .NET 4.0.30319.488    x64
> 
> 		It could be a JIT bug. I have encounted many x64 JIT bugs and
> IKVM contains workarounds for a few of them.
> 
> 		What exact versions (of IKVM and .NET) are you using? On what
> OS?
> 
> 		Regards,
> 		Jeroen
> 
> 		> -----Original Message-----
> 		> From: Nadim Ratani [mailto:nr...@go...]
> 		> Sent: Friday, October 14, 2011 23:55
> 		> To: ikv...@li...
> 		> Subject: [Ikvm-developers] Java -> .NET using IKVM, x86
> code works x64
> 		> doesn't
> 
> 		>
> 		> I am using a library that uses javax.crypto to perform RSA
> decryption.
> 		> When I compile to a x86 DLL and use this library everything
> works fine.
> 		> When I switch to x64 (with no other changes that I am aware
> of) I get an
> 		> exception:
> 		> javax.crypto.BadPaddingException: lHash mismatch
> 		>
> 		> I am roughly following the instructions here to create a
> DLL:
> 		> http://code.google.com/p/keyczar/wiki/KeyczarDotNet
> 		>
> 		> And then using this DLL in a .NET project targeting x64.
> 		>
> 		> Does anyone have suggestions on where I can begin looking
> to get to the
> 		> bottom of this?
> 		>
> 		> +n
> 		> Nadim Ratani | Google Developer Relations |
> nr...@go... | Google
> 		> Inc.
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
> 
>