Re: [PATCH v2 20/23] x86/pv: Exception handling in FRED mode

[Jan Beulich <jbeulich@xxxxxxxx>]

On 28.08.2025 17:04, Andrew Cooper wrote:
> --- a/xen/arch/x86/traps.c
> +++ b/xen/arch/x86/traps.c
> @@ -2265,9 +2265,83 @@ void asmlinkage check_ist_exit(const struct 
> cpu_user_regs *regs, bool ist_exit)
>  
>  void asmlinkage entry_from_pv(struct cpu_user_regs *regs)
>  {
> +    struct fred_info *fi = cpu_regs_fred_info(regs);
> +    uint8_t type = regs->fred_ss.type;
> +    uint8_t vec = regs->fred_ss.vector;
> +
>      /* Copy fred_ss.vector into entry_vector as IDT delivery would have 
> done. */
> -    regs->entry_vector = regs->fred_ss.vector;
> +    regs->entry_vector = vec;
> +
> +    if ( !IS_ENABLED(CONFIG_PV) )
> +        goto fatal;
> +
> +    /*
> +     * First, handle the asynchronous or fatal events.  These are either
> +     * unrelated to the interrupted context, or may not have valid context
> +     * recorded, and all have special rules on how/whether to re-enable IRQs.
> +     */
> +    switch ( type )
> +    {
> +    case X86_ET_EXT_INTR:
> +        return do_IRQ(regs);
>  
> +    case X86_ET_NMI:
> +        return do_nmi(regs);
> +
> +    case X86_ET_HW_EXC:
> +        switch ( vec )
> +        {
> +        case X86_EXC_DF: return do_double_fault(regs);
> +        case X86_EXC_MC: return do_machine_check(regs);
> +        }
> +        break;
> +    }

This switch() is identical to entry_from_xen()'s. Fold into a helper?

> +    /*
> +     * With the asynchronous events handled, what remains are the synchronous
> +     * ones.  Guest context always had interrupts enabled.
> +     */
> +    local_irq_enable();

In the comment, maybe s/Guest/PV guest/?

> +    switch ( type )
> +    {
> +    case X86_ET_HW_EXC:
> +    case X86_ET_PRIV_SW_EXC:
> +    case X86_ET_SW_EXC:
> +        switch ( vec )
> +        {
> +        case X86_EXC_PF:  handle_PF(regs, fi->edata); break;
> +        case X86_EXC_GP:  do_general_protection(regs); break;
> +        case X86_EXC_UD:  do_invalid_op(regs); break;
> +        case X86_EXC_NM:  do_device_not_available(regs); break;
> +        case X86_EXC_BP:  do_int3(regs); break;
> +        case X86_EXC_DB:  handle_DB(regs, fi->edata); break;
> +
> +        case X86_EXC_DE:
> +        case X86_EXC_OF:
> +        case X86_EXC_BR:
> +        case X86_EXC_NP:
> +        case X86_EXC_SS:
> +        case X86_EXC_MF:
> +        case X86_EXC_AC:
> +        case X86_EXC_XM:
> +            do_trap(regs);
> +            break;
> +
> +        case X86_EXC_CP:  do_entry_CP(regs); break;
> +
> +        default:
> +            goto fatal;
> +        }
> +        break;

This again looks identical to when entry_from_xen() has. Maybe, instead of
a helper for each switch(), we could have a common always-inline function
(with all necessary parametrization) that both invoke?

> --- a/xen/arch/x86/x86_64/entry.S
> +++ b/xen/arch/x86/x86_64/entry.S
> @@ -63,7 +63,7 @@ UNLIKELY_END(syscall_no_callback)
>          /* Conditionally clear DF */
>          and   %esi, UREGS_eflags(%rsp)
>  /* %rbx: struct vcpu */
> -test_all_events:
> +LABEL(test_all_events, 0)
>          ASSERT_NOT_IN_ATOMIC
>          cli                             # tests must not race interrupts
>  /*test_softirqs:*/
> @@ -152,6 +152,8 @@ END(switch_to_kernel)
>  FUNC_LOCAL(restore_all_guest)
>          ASSERT_INTERRUPTS_DISABLED
>  
> +        ALTERNATIVE "", "jmp eretu_exit_to_guest", X86_FEATURE_XEN_FRED
> +
>          /* Stash guest SPEC_CTRL value while we can read struct vcpu. */
>          mov VCPU_arch_msrs(%rbx), %rdx

I assume it's deliberate that you don't "consume" this insn into the
alternative, but without the description saying anything it's not quite
clear why.

Jan