Re: [patch 2/5] x86/idt: Add comments about early #PF handling

[Peter Zijlstra]

On Thu, May 28, 2020 at 04:53:17PM +0200, Thomas Gleixner wrote:
> The difference between 32 and 64 bit vs. early #PF handling is not
> documented. Replace the FIXME at idt_setup_early_pf() with proper comments.
> 
> Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> ---
>  arch/x86/kernel/idt.c |   10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> --- a/arch/x86/kernel/idt.c
> +++ b/arch/x86/kernel/idt.c
> @@ -61,7 +61,11 @@ static bool idt_setup_done __initdata;
>  static const __initconst struct idt_data early_idts[] = {
>  	INTG(X86_TRAP_DB,		asm_exc_debug),
>  	SYSG(X86_TRAP_BP,		asm_exc_int3),
> +
>  #ifdef CONFIG_X86_32
> +	/*
> +	 * Not possible on 64-bit. See idt_setup_early_pf() for details.
> +	 */
>  	INTG(X86_TRAP_PF,		asm_exc_page_fault),
>  #endif
>  };
> @@ -276,8 +280,10 @@ void __init idt_setup_traps(void)
>   * cpu_init() is invoked and sets up TSS. The IST variant is installed
>   * after that.
>   *
> - * FIXME: Why is 32bit and 64bit installing the PF handler at different
> - * places in the early setup code?
> + * Note, that X86_64 cannot install the real #PF handler in
> + * idt_setup_early_traps() because the memory intialization needs the #PF
> + * handler from the early_idt_handler_array to initialize the early page
> + * tables.

+ "See early_make_pgtables()" ?

I had to frob around in head_64.S to find wth that early handler
actually did.

>   */
>  void __init idt_setup_early_pf(void)
>  {
>