lib: sbi_fp_save enables VS instead of FS before saving floating-point state

Summary

lib/sbi/sbi_fp.c:sbi_fp_save() enables the Vector state (mstatus.VS) instead of the Floating-Point state (mstatus.FS) before executing floating-point store instructions.

The issue is present in OpenSBI v1.9 and current master.

Affected code

In sbi_fp_save():

mstatus_orig = csr_read_set(CSR_MSTATUS, MSTATUS_VS);

The function subsequently executes fsd/fsw instructions to save f0-f31.

In contrast, sbi_fp_restore() correctly uses:

mstatus_orig = csr_read_set(CSR_MSTATUS, MSTATUS_FS);

Expected behavior

Before accessing the floating-point registers, sbi_fp_save() should enable floating-point state through mstatus.FS.

Actual behavior

sbi_fp_save() sets mstatus.VS, which controls vector state rather than floating-point state.

If mstatus.FS is Off when sbi_fp_save() is called, the subsequent floating-point save instructions cannot execute normally, causing the domain context switch to fail instead of saving the FP context.

Impact

This is a functional/reliability issue in the floating-point domain context-switching support.

A domain switch on a hart with the F or D extension can fail when the outgoing context has mstatus.FS=Off.

Proposed fix

- mstatus_orig = csr_read_set(CSR_MSTATUS, MSTATUS_VS);
+ mstatus_orig = csr_read_set(CSR_MSTATUS, MSTATUS_FS);

This also makes the save path consistent with sbi_fp_restore(), which already uses MSTATUS_FS.

Start in lib/sbi/sbi_fp.c at sbi_fp_save() and compare its mstatus handling with sbi_fp_restore(). Done means the save path enables floating-point state through mstatus.FS before executing the f0-f31 save instructions, while preserving the existing restore behavior.