In D23193#531527, @asomers wrote:

I agree with Renato. There's no way to test in_cksum on both endiannesses without using something complicated like qemu. The right approach is to checksum the same input on different platforms, and assert that the result is identical on all. That's what Renato's change does. @renato.riolino_eldorado.org.br I don't think you have a commit bit, right? If not, I can commit the change for you. But I don't have any BE hardware. Can you first confirm that with your change the tests pass on BE hardware on a newish build?

Adjust code style

Moved exception type check from trap.c to fpu.c

Passing SIGFPE on emulated FPU as well.

Fixed order for calling save_fpu() and enable_fpu()

Treating fpe exceptions as integers instead of bitfields and calling save_fpu() when detected the signal as been FPE.

Original test wasn't working on BE. Just swaping the input data and testing on the same platform is not the same as testing the same input on a different platform. The best prove that in_cksum is enadiannes independent is that the same input outputs the same value on both platforms as long as you test the output byte-a-byte.

Using a uint64_t instead of a union as suggested by luporl

Fixes pointed by jhibbits and luporl and a test of endianess when checking the type of fp exception.

Ok, now it makes sense. But then something is wrong with the test or with the in_cksum implementation, since it is failing on all tests on BE. For been independent of the endianess, I suppose the result checksum for {0x12, 0x34, 0x56, 0x78} and {0x34, 0x12, 0x78, 0x56} should be the same, right?

Diff now is against freebsd tree

Fix for #192839 - Duplicate entries in an mtree file cause nmtree to coredump

Perfect, thanks! =-) May I ask which big-endian arch you're running these tests on? This is just my curiosity- the tests are indisputably broken as-is.

Fix speeling "be" and "le" to "eb" and "el"