Re: [RFT PATCH] x86/pat: Fix set_mce_nospec() for pmem

[Dan Williams <dan.j.williams@intel.com>]

On Tue, Jul 6, 2021 at 6:01 PM Dan Williams <dan.j.williams@intel.com> wrote:
>
> When poison is discovered and triggers memory_failure() the physical
> page is unmapped from all process address space. However, it is not
> unmapped from kernel address space. Unlike a typical memory page that
> can be retired from use in the page allocator and marked 'not present',
> pmem needs to remain accessible given it can not be physically remapped
> or retired. set_memory_uc() tries to maintain consistent nominal memtype
> mappings for a given pfn, but memory_failure() is an exceptional
> condition.
>
> For the same reason that set_memory_np() bypasses memtype checks
> because they do not apply in the memory failure case, memtype validation
> is not applicable for marking the pmem pfn uncacheable. Use
> _set_memory_uc().
>
> Reported-by: Jane Chu <jane.chu@oracle.com>
> Fixes: 284ce4011ba6 ("x86/memory_failure: Introduce {set,clear}_mce_nospec()")
> Cc: Luis Chamberlain <mcgrof@suse.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: Tony Luck <tony.luck@intel.com>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
> Jane, can you give this a try and see if it cleans up the error you are
> seeing?
>
> Thanks for the help.

Jane, does this resolve the failure you reported [1]?

[1]: https://lore.kernel.org/r/327f9156-9b28-d20e-2850-21c125ece8c7@oracle.com

>
>  arch/x86/include/asm/set_memory.h |    9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/include/asm/set_memory.h b/arch/x86/include/asm/set_memory.h
> index 43fa081a1adb..0bf2274c5186 100644
> --- a/arch/x86/include/asm/set_memory.h
> +++ b/arch/x86/include/asm/set_memory.h
> @@ -114,8 +114,13 @@ static inline int set_mce_nospec(unsigned long pfn, bool unmap)
>
>         if (unmap)
>                 rc = set_memory_np(decoy_addr, 1);
> -       else
> -               rc = set_memory_uc(decoy_addr, 1);
> +       else {
> +               /*
> +                * Bypass memtype checks since memory-failure has shot
> +                * down mappings.
> +                */
> +               rc = _set_memory_uc(decoy_addr, 1);
> +       }
>         if (rc)
>                 pr_warn("Could not invalidate pfn=0x%lx from 1:1 map\n", pfn);
>         return rc;
>