Re: [PATCH 4.4 00/37] 4.4.110-stable review

[Steven Sistare]

On 1/11/2018 3:46 PM, Linus Torvalds wrote:
> On Thu, Jan 11, 2018 at 12:37 PM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>>
>> 67a9108ed431 ("x86/efi: Build our own page table structures")
>>
>> got rid of EFI depending on real_mode_header->trampoline_pgd
> 
> So I think it only got rid of by default - the codepath is still
> there, the allocation is still there, it's just that it's not actually
> used unless somebody does that "efi=old_mmap" thing.
> 
> Looking around, there's at least one quirk for the SGI UV1 system that
> enables EFI_OLD_MMAP automatically. There might be others that I
> missed, but I think that's it.
> 
> So it *can* trigger without "efi=old_mmap", but not on any normal machines.
> 
> And as Pavel points out, even when the bug is active, it's pretty hard
> to actually trigger.
> 
> But yeah, there may be other EFI patches that I didn't notice that
> changed things in other ways too.
> 
>                Linus

The bug is not present in the latest upstream kernel because the efi_pgd is
correctly aligned:

  arch/x86/platform/efi/efi_64.c
    int __init efi_alloc_page_tables(void)
      efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);

  arch/x86/include/asm/pgalloc.h
    +#ifdef CONFIG_PAGE_TABLE_ISOLATION
    +#define PGD_ALLOCATION_ORDER 1
    +#else
    +#define PGD_ALLOCATION_ORDER 0
    +#endif

Pavel's patch fixes kernels prior to
  67a9108ed431 ("x86/efi: Build our own page table structures")

where the efi pgd allocation looks like:

  arch/x86/realmode/init.c
    void __init reserve_real_mode(void)
       mem = memblock_find_in_range(0, 1<<20, size, PAGE_SIZE);
       base = __va(mem);
       real_mode_header = (struct real_mode_header *) base;

  void __init setup_real_mode(void)
    trampoline_pgd = (u64 *) __va(real_mode_header->trampoline_pgd);

Kernel versions between 67a9108ed431 and the latest also have the bug and
need a similar fix:

  arch/x86/platform/efi/efi_64.c

    int __init efi_alloc_page_tables(void)
      efi_pgd = (pgd_t *)__get_free_page(gfp_mask);

    int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
      pgd = efi_pgd;
      efi_scratch.efi_pgt = (pgd_t *)__pa(efi_pgd);

All of the code paths above are taken when *not* EFI_OLD_MMAP.

- Steve