Re: [PATCH v4 14/45] mm: kmsan: maintain KMSAN metadata for page operations

[Marco Elver]

On Fri, 1 Jul 2022 at 16:23, Alexander Potapenko <glider@xxxxxxxxxx> wrote:
>
> Insert KMSAN hooks that make the necessary bookkeeping changes:
>  - poison page shadow and origins in alloc_pages()/free_page();
>  - clear page shadow and origins in clear_page(), copy_user_highpage();
>  - copy page metadata in copy_highpage(), wp_page_copy();
>  - handle vmap()/vunmap()/iounmap();
>
> Signed-off-by: Alexander Potapenko <glider@xxxxxxxxxx>
> ---
> v2:
>  -- move page metadata hooks implementation here
>  -- remove call to kmsan_memblock_free_pages()
>
> v3:
>  -- use PAGE_SHIFT in kmsan_ioremap_page_range()
>
> v4:
>  -- change sizeof(type) to sizeof(*ptr)
>  -- replace occurrences of |var| with @var
>  -- swap mm: and kmsan: in the subject
>  -- drop __no_sanitize_memory from clear_page()
>
> Link: https://linux-review.googlesource.com/id/I6d4f53a0e7eab46fa29f0348f3095d9f2e326850
> ---
>  arch/x86/include/asm/page_64.h |  12 ++++
>  arch/x86/mm/ioremap.c          |   3 +
>  include/linux/highmem.h        |   3 +
>  include/linux/kmsan.h          | 123 +++++++++++++++++++++++++++++++++
>  mm/internal.h                  |   6 ++
>  mm/kmsan/hooks.c               |  87 +++++++++++++++++++++++
>  mm/kmsan/shadow.c              | 114 ++++++++++++++++++++++++++++++
>  mm/memory.c                    |   2 +
>  mm/page_alloc.c                |  11 +++
>  mm/vmalloc.c                   |  20 +++++-
>  10 files changed, 379 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/include/asm/page_64.h b/arch/x86/include/asm/page_64.h
> index baa70451b8df5..227dd33eb4efb 100644
> --- a/arch/x86/include/asm/page_64.h
> +++ b/arch/x86/include/asm/page_64.h
> @@ -45,14 +45,26 @@ void clear_page_orig(void *page);
>  void clear_page_rep(void *page);
>  void clear_page_erms(void *page);
>
> +/* This is an assembly header, avoid including too much of kmsan.h */

All of this code is under an "#ifndef __ASSEMBLY__" guard, does it matter?

> +#ifdef CONFIG_KMSAN
> +void kmsan_unpoison_memory(const void *addr, size_t size);
> +#endif
>  static inline void clear_page(void *page)
>  {
> +#ifdef CONFIG_KMSAN
> +       /* alternative_call_2() changes @page. */
> +       void *page_copy = page;
> +#endif
>         alternative_call_2(clear_page_orig,
>                            clear_page_rep, X86_FEATURE_REP_GOOD,
>                            clear_page_erms, X86_FEATURE_ERMS,
>                            "=D" (page),
>                            "0" (page)
>                            : "cc", "memory", "rax", "rcx");
> +#ifdef CONFIG_KMSAN
> +       /* Clear KMSAN shadow for the pages that have it. */
> +       kmsan_unpoison_memory(page_copy, PAGE_SIZE);

What happens if this is called before the alternative-call? Could this
(in the interest of simplicity) be moved above it? And if you used the
kmsan-checks.h header, it also doesn't need any "ifdef CONFIG_KMSAN"
anymore.

> +#endif
>  }