Hello Greg,
Following up on our discussion during the review of kernel 4.4.138-stable, I have backported a few patches that remove the remaining FPU lazy mode deadcode from 4.4.y and 4.9.y. To avoid confusion I will send the explanation on separate e-mails:
Patches for 4.9.y:
[PATCH 1/4] x86/fpu: Remove use_eager_fpu() -> I only had to solve a small conflict caused by the fact that commit b22cbe404a9c ("x86/fpu: Fix invalid FPU ptrace state after execve()") had been applied before (it was supposed to come after)
[PATCH 2/4] x86/fpu: Remove struct fpu::counter -> This is just a git cherry-pick of 3913cc350757 ("x86/fpu: Remove struct fpu::counter") which applies cleanly on 4.9.y. You can cherry-pick it instead of applying my patch.
[PATCH 3/4] Revert "perf: sync up x86/.../cpufeatures.h" -> Sorry to revert your patch ;). This was just for the next patch to apply cleanly.
[PATCH 4/4] x86/fpu: Finish excising 'eagerfpu' -> Again you can use a cherry pick of e63650840e8b ("x86/fpu: Finish excising 'eagerfpu'") instead of applying my patch.
Tested on x86_64 only with: - ./runltp -f math (all tests pass) - IEEE 754 tests (no regressions seen) Ref: http://www.math.utah.edu/~beebe/software/ieee/ [Note] I could not prepare a x86 32-bit machine to test no387 [Note] If someone knows better about testing the FPU please let me know.
Best regards, Daniel Sangorrin
PS: Using git send-email --cc-cmd="scripts/get_maintainer.pl" for the first time. Apologies if this is not the right method.