linux(4): Preserve fpu xsave state across signal delivery on amd64

[dchagin]

Summary

PR: 270247

Test Plan

golang and glibc test suite

[dchagin]

ugh, old commit, fix magic2 printout in error and some comments

[dchagin]

Rebase to D40443

[dchagin]

Don't leak xfpustate if copyin of magic2 failed

[kib]

sys/amd64/linux/linux_sysvec.c

329

Is it more correct to check magic before copying xfpustate?

sys/x86/linux/linux_x86_sigframe.h

42

This is somewhat weird way to use sizeof(). Usually we apply it to either type of storage, not an expression, for the purpose of defining the ABI. Is it really sizeof(uint32_t)?

[dchagin]

sys/amd64/linux/linux_sysvec.c

329

I've been thinking about it, and leaved as is, as caller will be killed on error or if magic/stack is trashed, and this path is not common. If you insist I’ll fix that, however it does not more sense in my pov

sys/x86/linux/linux_x86_sigframe.h

42

Due to U suffix of LINUX_FP_XSTATE_MAGIC2 it's unsigned int.
It really modeled like a Linux does, and I especially checked this, as I first declared an magic2 variable of the size_t type and spent some time looking for an error reason )))
Perhaps I should add a comment here to avoid confusing

[kib]

sys/x86/linux/linux_x86_sigframe.h

42

No, it is unsigned int due to suffix + the fact that it fits into unsigned. This is why was triggered by the expression. Of course, size_t is problematic, but uint32_t should be not.

I ask you to change the expression to sizeof(uint32_t), please.

[dchagin]

Yes, done, it's much better now, thank you!

[dchagin]

whoops, magic2 type changed to uint32_t too