[dm-devel] [PATCH v9 2/7] x86/mce: relocate set{clear}_mce_nospec() functions

[Dan Williams]

On Fri, Apr 22, 2022 at 3:46 PM Jane Chu <jane.chu at oracle.com> wrote:
>
> Relocate the twin mce functions to arch/x86/mm/pat/set_memory.c
> file where they belong.
>
> While at it, fixup a function name in a comment.
>
> Reviewed-by: Christoph Hellwig <hch at lst.de>
> Reviewed-by: Dan Williams <dan.j.williams at intel.com>
> Signed-off-by: Jane Chu <jane.chu at oracle.com>
> ---
>  arch/x86/include/asm/set_memory.h | 52 -------------------------------
>  arch/x86/mm/pat/set_memory.c      | 49 ++++++++++++++++++++++++++++-
>  include/linux/set_memory.h        |  8 ++---
>  3 files changed, 52 insertions(+), 57 deletions(-)
>
> diff --git a/arch/x86/include/asm/set_memory.h b/arch/x86/include/asm/set_memory.h
> index 78ca53512486..b45c4d27fd46 100644
> --- a/arch/x86/include/asm/set_memory.h
> +++ b/arch/x86/include/asm/set_memory.h
> @@ -86,56 +86,4 @@ bool kernel_page_present(struct page *page);
>
>  extern int kernel_set_to_readonly;
>
> -#ifdef CONFIG_X86_64
> -/*
> - * Prevent speculative access to the page by either unmapping
> - * it (if we do not require access to any part of the page) or
> - * marking it uncacheable (if we want to try to retrieve data
> - * from non-poisoned lines in the page).
> - */
> -static inline int set_mce_nospec(unsigned long pfn, bool unmap)
> -{
> -       unsigned long decoy_addr;
> -       int rc;
> -
> -       /* SGX pages are not in the 1:1 map */
> -       if (arch_is_platform_page(pfn << PAGE_SHIFT))
> -               return 0;
> -       /*
> -        * We would like to just call:
> -        *      set_memory_XX((unsigned long)pfn_to_kaddr(pfn), 1);
> -        * but doing that would radically increase the odds of a
> -        * speculative access to the poison page because we'd have
> -        * the virtual address of the kernel 1:1 mapping sitting
> -        * around in registers.
> -        * Instead we get tricky.  We create a non-canonical address
> -        * that looks just like the one we want, but has bit 63 flipped.
> -        * This relies on set_memory_XX() properly sanitizing any __pa()
> -        * results with __PHYSICAL_MASK or PTE_PFN_MASK.
> -        */
> -       decoy_addr = (pfn << PAGE_SHIFT) + (PAGE_OFFSET ^ BIT(63));
> -
> -       if (unmap)
> -               rc = set_memory_np(decoy_addr, 1);
> -       else
> -               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;
> -}
> -#define set_mce_nospec set_mce_nospec
> -
> -/* Restore full speculative operation to the pfn. */
> -static inline int clear_mce_nospec(unsigned long pfn)
> -{
> -       return set_memory_wb((unsigned long) pfn_to_kaddr(pfn), 1);
> -}
> -#define clear_mce_nospec clear_mce_nospec
> -#else
> -/*
> - * Few people would run a 32-bit kernel on a machine that supports
> - * recoverable errors because they have too much memory to boot 32-bit.
> - */
> -#endif
> -
>  #endif /* _ASM_X86_SET_MEMORY_H */
> diff --git a/arch/x86/mm/pat/set_memory.c b/arch/x86/mm/pat/set_memory.c
> index abf5ed76e4b7..978cf5bd2ab6 100644
> --- a/arch/x86/mm/pat/set_memory.c
> +++ b/arch/x86/mm/pat/set_memory.c
> @@ -1816,7 +1816,7 @@ static inline int cpa_clear_pages_array(struct page **pages, int numpages,
>  }
>
>  /*
> - * _set_memory_prot is an internal helper for callers that have been passed
> + * __set_memory_prot is an internal helper for callers that have been passed
>   * a pgprot_t value from upper layers and a reservation has already been taken.
>   * If you want to set the pgprot to a specific page protocol, use the
>   * set_memory_xx() functions.
> @@ -1925,6 +1925,53 @@ int set_memory_wb(unsigned long addr, int numpages)
>  }
>  EXPORT_SYMBOL(set_memory_wb);
>
> +/*
> + * Prevent speculative access to the page by either unmapping
> + * it (if we do not require access to any part of the page) or
> + * marking it uncacheable (if we want to try to retrieve data
> + * from non-poisoned lines in the page).
> + */
> +int set_mce_nospec(unsigned long pfn, bool unmap)
> +{
> +       unsigned long decoy_addr;
> +       int rc;
> +
> +       if (!IS_ENABLED(CONFIG_64BIT))
> +               return 0;
> +
> +       /* SGX pages are not in the 1:1 map */
> +       if (arch_is_platform_page(pfn << PAGE_SHIFT))
> +               return 0;
> +       /*
> +        * We would like to just call:
> +        *      set_memory_XX((unsigned long)pfn_to_kaddr(pfn), 1);
> +        * but doing that would radically increase the odds of a
> +        * speculative access to the poison page because we'd have
> +        * the virtual address of the kernel 1:1 mapping sitting
> +        * around in registers.
> +        * Instead we get tricky.  We create a non-canonical address
> +        * that looks just like the one we want, but has bit 63 flipped.
> +        * This relies on set_memory_XX() properly sanitizing any __pa()
> +        * results with __PHYSICAL_MASK or PTE_PFN_MASK.
> +        */
> +       decoy_addr = (pfn << PAGE_SHIFT) + (PAGE_OFFSET ^ BIT(63));
> +
> +       if (unmap)
> +               rc = set_memory_np(decoy_addr, 1);
> +       else
> +               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;
> +}
> +
> +/* Restore full speculative operation to the pfn. */
> +int clear_mce_nospec(unsigned long pfn)
> +{
> +       return set_memory_wb((unsigned long) pfn_to_kaddr(pfn), 1);
> +}
> +EXPORT_SYMBOL_GPL(clear_mce_nospec);
> +
>  int set_memory_x(unsigned long addr, int numpages)
>  {
>         if (!(__supported_pte_mask & _PAGE_NX))
> diff --git a/include/linux/set_memory.h b/include/linux/set_memory.h
> index f36be5166c19..683a6c3f7179 100644
> --- a/include/linux/set_memory.h
> +++ b/include/linux/set_memory.h
> @@ -42,14 +42,14 @@ static inline bool can_set_direct_map(void)
>  #endif
>  #endif /* CONFIG_ARCH_HAS_SET_DIRECT_MAP */
>
> -#ifndef set_mce_nospec
> +#ifdef CONFIG_X86_64

Jane,

I just noticed that this makes set_mce_nospec() and clear_mce_nospec()
x86_64-only. If / when more architectures add support for these
helpers they will need to go back to the "#ifndef $symbol" scheme to
allow asm/set_memory.h to indicate the availability of the arch-local
helper.