Hi, On Fri, Nov 13, 2020 at 11:36:08AM +0100, Björn Töpel wrote: > I was running the selftest/bpf on riscv, and had a closer look at one > of the failing cases: > > #14/p valid read map access into a read-only array 2 FAIL retval > 65507 != -29 (run 1/1) > > The test does a csum_partial() call via a BPF helper. riscv uses the > generic implementation. arm64 uses the generic csum_partial() and fail > in the same way [1]. It's worse than that, because arm64, parisc, alpha and others implement do_csum(), called by the generic csum_partial(), and those all return a 16-bit value. It would be good to firstly formalize the size of the value returned by the bpf_csum_diff() helper, because it's not clear from the doc (and the helper returns a s64). Then homogenizing the csum_partial() implementations is difficult. One way forward, without having to immediately rewrite all arch-specific implementations, would be to replace csum_partial() and do_csum() with csum_partial_32(), csum_partial_16(), do_csum_32() and do_csum_16(). That way we can use a generic implementation of the 32-bit variant if the arch-specific implementation is 16-bit. Thanks, Jean > arm (32-bit) has a arch specfic implementation, > and fail in another way (FAIL retval 131042 != -29) [2]. > > I mimicked the test case in a userland program, comparing the generic > csum_partial() to the x86 implementation [3], and the generic and x86 > implementation does yield a different result. > > x86 : -29 : 0xffffffe3 > generic : 65507 : 0x0000ffe3 > arm : 131042 : 0x0001ffe2 > > Who is correct? :-) It would be nice to get rid of this failed case... > > > Thanks, > Björn > > > [1] https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20201112/testrun/3430401/suite/kselftest/test/bpf.test_verifier/log > [2] https://qa-reports.linaro.org/lkft/linux-mainline-master/build/v5.10-rc3-207-g585e5b17b92d/testrun/3432361/suite/kselftest/test/bpf.test_verifier/log > [3] https://gist.github.com/bjoto/dc22d593aa3ac63c2c90632de5ed82e0
