On 16 September 2018 at 16:53, Richard Henderson <richard.hender...@linaro.org> wrote: > On 9/15/18 6:32 PM, Peter Maydell wrote: >> We want to arrange to have SWP work anyway on linux-user, >> I think, since the kernel will typically trap-and-emulate >> it assuming it was built with CONFIG_SWP_EMULATE. (I don't >> know if those kernels will advertise swp in the hwcaps, >> but I guess not.) > > Ah, I did not know about SWP_EMULATE. It appears to be > specific to armv7+ (though we don't support the pre-v4 > cpus for which it might otherwise be relevant).
Yes, it's intended to allow older userspace binaries to continue to work on newer CPUs without SWP, not to try to run new binaries on older CPUs. (Anything compiled for a CPU new enough for SWP probably uses other newer insns that some pre-v4 CPU doesn't have anyway.) > It does appear that HWCAP_SWP is advertised anyway: > > mm/proc-v7.S: .long HWCAP_SWP | HWCAP_HALF | HWCAP_THUMB | HWCAP_FAST_MULT Interesting. I might ask some random kernel person what the semantics of hwcap are -- is it "will work even if a terrible plan" or "it makes sense to use this"? thanks -- PMM
