Re: [Xen-devel] [PATCH] x86: re-add stack alignment check

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 14/11/16 13:25, Jan Beulich wrote:
> Commit 279840d5ea ("x86/boot: install trap handlers much earlier on
> boot"), perhaps not really intentionally, removed this check. Add it

No - that was deliberate.  The check isn't useful (see below).

> back,
> - preventing it to trigger before any output is set up,

"preventing it from triggering ..."

> - accompanying it with a (weaker, due to its open coding of what
>   get_stack_bottom() does) build time check.
>
> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> ---
> While not strictly needed for 4.8, I thought I'd still submit it as a
> possible candidate.
>
> --- a/xen/arch/x86/cpu/common.c
> +++ b/xen/arch/x86/cpu/common.c
> @@ -643,6 +643,11 @@ void load_system_tables(void)
>               .limit = (IDT_ENTRIES * sizeof(idt_entry_t)) - 1,
>       };
>  
> +     /* Bottom-of-stack must be 16-byte aligned! */
> +     BUILD_BUG_ON((sizeof(struct cpu_info) -
> +                   offsetof(struct cpu_info, guest_cpu_user_regs.es)) & 0xf);
> +     BUG_ON(system_state != SYS_STATE_early_boot && (stack_bottom & 0xf));

This will still triple fault the system if it triggers on an AP. 
Exceptions aren't hooked up yet.

The reason I dropped the check was that there was no way it was going to
fail on the BSP (because of the alignment of cpu0_stack) or APs (because
of the alloc_xenheap_pages(STACK_ORDER, memflags) allocation).

The BUILD_BUG_ON() is useful to retain, but I would suggest making
get_stack_bottom() a static inline and putting the BUILD_BUG_ON() there.

~Andrew

> +
>       /* Main stack for interrupts/exceptions. */
>       tss->rsp0 = stack_bottom;
>       tss->bitmap = IOBMP_INVALID_OFFSET;
>
>
>


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel