riscv: Fix fork() system call

[pussuw]

Summary

When executing fork() via a system call, the parent's stack gets corrupted
by the child, as during exception return the child loads the parent's
stack pointer from the context save area.

This happens because the full parent stack (including what has been pushed
during the system call) is copied to the child. What should be copied, is
only the user stack of the parent (the kernel stack is not interesting).

Fix this by only copying the parent's user stack to the child; and make
the child return directly to userspace (not via dispatch_syscall).

Impact

Fixes fork() when LIB_SYSCALL=y

Testing

rv-virt:pnsh64

[pussuw]

Forgot to save FPU, fill fix it asap

[pussuw]

This is now tested to work in protected mode, just like before, and SHOULD work with kernel stack as well, but could not test it yet.

I will still test it but vfork() has some kind of a bug for kernel mode which prevents doing this. I'll look into it tomorrow.

[yf13]

I tried this locally but vfork_test() got illegal instruction exception:

user_main: vfork() test
riscv_exception: EXCEPTION: Illegal instruction. MCAUSE: 00000002, EPC: 80203a40, MTVAL: 00000000
riscv_exception: PANIC!!! Exception = 00000002

[pussuw]

Yes I noticed late last night there is an error; the child returns to userspace with kernel privileges. I did a fix but did not have time to push.

FPU state is also not copied to the child. The patch is incomplete.

[pussuw]

@yf13 thanks for testing, I fixed the aforementioned issues; the child now returns to user space with user privileges and FPU registers are also copied to the child.

[yf13]

@pussuw, thanks for the updates, it works here. Also I added two in-place comemnts.

[pussuw]

@pussuw, thanks for the updates, it works here. Also I added two in-place comemnts.

Nice! Your comments are not visible to me, maybe you need to "submit review" ?
"Files changed" -> "Review changes" -> "Submit review" ?

[yf13]

@pussuw, thanks for teaching, I submitted the reviews. Please see if they are visibile now? Sorry for inconveninces as I never added in-place comments before.

[yf13]

Seems that the warning with maix-bit/knsh_smp is gone for the riscv_cpustart.c in master branch, so we are not too far to pass the CI checks.

[xiaoxiang781216]

@pussuw let's rebase to the latest.master.

[pkarashchenko]

Seems like some defconfigs need to be normalized after adding new Kconfig dependency