On Tue, Jan 14, 2025 at 10:54:53AM -0600, Tom Lendacky wrote:
> On 1/14/25 09:06, Tom Lendacky wrote:
> > On 1/14/25 08:44, Kirill A. Shutemov wrote:
> >> On Tue, Jan 14, 2025 at 08:33:39AM -0600, Tom Lendacky wrote:
> >>> On 1/14/25 01:27, Kirill A. Shutemov wrote:
> >>>> On Mon, Jan 13, 2025 at 02:47:56PM -0600, Tom Lendacky wrote:
> >>>>> On 1/13/25 07:14, Kirill A. Shutemov wrote:
> >>>>>> Currently memremap(MEMREMAP_WB) can produce decrypted/shared mapping:
> >>>>>>
> >>>>>> memremap(MEMREMAP_WB)
> >>>>>> arch_memremap_wb()
> >>>>>> ioremap_cache()
> >>>>>> __ioremap_caller(.encrytped = false)
> >>>>>>
> >>>>>> In such cases, the IORES_MAP_ENCRYPTED flag on the memory will determine
> >>>>>> if the resulting mapping is encrypted or decrypted.
> >>>>>>
> >>>>>> Creating a decrypted mapping without explicit request from the caller is
> >>>>>> risky:
> >>>>>>
> >>>>>> - It can inadvertently expose the guest's data and compromise the
> >>>>>> guest.
> >>>>>>
> >>>>>> - Accessing private memory via shared/decrypted mapping on TDX will
> >>>>>> either trigger implicit conversion to shared or #VE (depending on
> >>>>>> VMM implementation).
> >>>>>>
> >>>>>> Implicit conversion is destructive: subsequent access to the same
> >>>>>> memory via private mapping will trigger a hard-to-debug #VE crash.
> >>>>>>
> >>>>>> The kernel already provides a way to request decrypted mapping
> >>>>>> explicitly via the MEMREMAP_DEC flag.
> >>>>>>
> >>>>>> Modify memremap(MEMREMAP_WB) to produce encrypted/private mapping by
> >>>>>> default unless MEMREMAP_DEC is specified.
> >>>>>>
> >>>>>> Fix the crash due to #VE on kexec in TDX guests if CONFIG_EISA is enabled.
> >>>>>
> >>>>> This patch causes my bare-metal system to crash during boot when using
> >>>>> mem_encrypt=on:
> >>>>>
> >>>>> [ 2.392934] efi: memattr: Entry type should be RuntimeServiceCode/Data
> >>>>> [ 2.393731] efi: memattr: ! 0x214c42f01f1162a-0xee70ac7bd1a9c629 [type=2028324321|attr=0x6590648fa4209879]
> >>>>
> >>>> Could you try if this helps?
> >>>>
> >>>> diff --git a/drivers/firmware/efi/memattr.c b/drivers/firmware/efi/memattr.c
> >>>> index c38b1a335590..b5051dcb7c1d 100644
> >>>> --- a/drivers/firmware/efi/memattr.c
> >>>> +++ b/drivers/firmware/efi/memattr.c
> >>>> @@ -160,7 +160,7 @@ int __init efi_memattr_apply_permissions(struct mm_struct *mm,
> >>>> if (WARN_ON(!efi_enabled(EFI_MEMMAP)))
> >>>> return 0;
> >>>>
> >>>> - tbl = memremap(efi_mem_attr_table, tbl_size, MEMREMAP_WB);
> >>>> + tbl = memremap(efi_mem_attr_table, tbl_size, MEMREMAP_WB | MEMREMAP_DEC);
> >>>
> >>> Well that would work for SME where EFI tables/data are not encrypted,
> >>> but will break for SEV where EFI tables/data are encrypted.
> >>
> >> Hm. Why would it break for SEV? It brings the situation back to what it
> >> was before the patch.
> >
> > Ah, true. I can try it and see how much further SME gets. Hopefully it
> > doesn't turn into a whack-a-mole thing.
>
> Unfortunately, it is turning into a whack-a-mole thing.
>
> But it looks the following works for SME:
>
> diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
> index 3c36f3f5e688..ff3cd5fc8508 100644
> --- a/arch/x86/mm/ioremap.c
> +++ b/arch/x86/mm/ioremap.c
> @@ -505,7 +505,7 @@ EXPORT_SYMBOL(iounmap);
>
> void *arch_memremap_wb(phys_addr_t phys_addr, size_t size, unsigned long flags)
> {
> - if (flags & MEMREMAP_DEC)
> + if (flags & MEMREMAP_DEC || cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
> return (void __force *)ioremap_cache(phys_addr, size);
>
> return (void __force *)ioremap_encrypted(phys_addr, size);
>
>
> I haven't had a chance to test the series on SEV, yet.
Please do.
I am okay with the change above. Borislav, is it acceptable direction for
you?
--
Kiryl Shutsemau / Kirill A. Shutemov
