Comment # 1 on bug 1085958 from Jiri Slaby

Ugh, oh my. The interaction of upstream vs. non-upstream spectre v2 handling is
in start_kernel as follows:

> parse_early_param
>       nospec <- if passed "nospec" parameter
> check_bugs
>       identify_boot_cpu
>               identify_cpu(boot_cpu_data)
>                       init_intel
>                               x86_spec_check -> non-upstream IBRS
>                                       setup_force_cpu_cap(X86_FEATURE_SPEC_CTRL)
>                       apply_forced_caps
>       spectre_v2_select_mitigation
>               spectre_v2_parse_cmdline
>                       nospec -> set by "nospectre_v2" param
>                               setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL)
> rest_init
>       kernel_init
>       kernel_init_freeable
>       smp_init -> distribute

So non-upstream spec v2 sets the forced flag, but later call to nospec won't
help, as it does not remove forced flag to be cleared.  So nospectre_v2 removes
the flag only from the boot cpu.

This should fix it:
--- a/arch/x86/kernel/cpu/spec_ctrl.c
+++ b/arch/x86/kernel/cpu/spec_ctrl.c
@@ -100,3 +100,4 @@ int nospec(char *str)
        return 0;
 }
 early_param("nospec", nospec);
+early_param("nospectre_v2", nospec);