Reserving memory using efi_mem_reserve() calls into the x86
efi_arch_mem_reserve() function. This function will insert a new EFI
memory descriptor into the EFI memory map representing the area of
memory to be reserved and marking it as EFI runtime memory.
As part of adding this new entry, a new EFI memory map is allocated and
mapped. The mapping is where a problem can occur. This new EFI memory map
is mapped using early_memremap(). If the allocated memory comes from an
area that is marked as EFI_BOOT_SERVICES_DATA memory in the current EFI
memory map, then it will be mapped unencrypted (see memremap_is_efi_data()
and the call to efi_mem_type()).
However, during replacement of the old EFI memory map with the new EFI
memory map, efi_mem_type() is disabled, resulting in the new EFI memory
map always being mapped encrypted in efi.memmap. This will cause a kernel
crash later in the boot.
Since it is known that the new EFI memory map will always be mapped
encrypted when efi_memmap_install() is called, explicitly map the new EFI
memory map as encrypted (using early_memremap_prot()) when inserting the
new memory map entry.
Cc: <[email protected]> # 4.14.x
Fixes: 8f716c9b5feb ("x86/mm: Add support to access boot related data in the clear")
Acked-by: Ard Biesheuvel <[email protected]>
Signed-off-by: Tom Lendacky <[email protected]>
---
Changes for v2:
- Update/expand commit message to (hopefully) make it easier to read and
understand
- Add a comment around the use of the early_memremap_prot() call
---
arch/x86/platform/efi/quirks.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
index b15ebfe40a73..14f8f20d727a 100644
--- a/arch/x86/platform/efi/quirks.c
+++ b/arch/x86/platform/efi/quirks.c
@@ -277,7 +277,19 @@ void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)
return;
}
- new = early_memremap(data.phys_map, data.size);
+ /*
+ * When SME is active, early_memremap() can map the memory unencrypted
+ * if the allocation came from EFI_BOOT_SERVICES_DATA (see
+ * memremap_is_efi_data() and the call to efi_mem_type()). However,
+ * when efi_memmap_install() is called to replace the memory map,
+ * efi_mem_type() is "disabled" and so the memory will always be mapped
+ * encrypted. To avoid this possible mismatch between the mappings,
+ * always map the newly allocated memmap memory as encrypted.
+ *
+ * When SME is not active, this behaves just like early_memremap().
+ */
+ new = early_memremap_prot(data.phys_map, data.size,
+ pgprot_val(pgprot_encrypted(FIXMAP_PAGE_NORMAL)));
if (!new) {
pr_err("Failed to map new boot services memmap\n");
return;
--
2.33.1
On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
> I could take it, but since it will ultimately go through -tip anyway,
> perhaps better if they just take it directly? (This will change after
> the next -rc1 though)
>
> Boris?
Yeah, I'm being told this is not urgent enough to rush in now so you
could queue it into your fixes branch for 5.16 once -rc1 is out and send
it to Linus then. The stable tag is just so it gets backported to the
respective trees.
But if you prefer I should take it, then I can queue it after -rc1.
It'll boil down to the same thing though.
Thx.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
On Wed, 27 Oct 2021 at 18:56, Borislav Petkov <[email protected]> wrote:
>
> On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
> > I could take it, but since it will ultimately go through -tip anyway,
> > perhaps better if they just take it directly? (This will change after
> > the next -rc1 though)
> >
> > Boris?
>
> Yeah, I'm being told this is not urgent enough to rush in now so you
> could queue it into your fixes branch for 5.16 once -rc1 is out and send
> it to Linus then. The stable tag is just so it gets backported to the
> respective trees.
>
> But if you prefer I should take it, then I can queue it after -rc1.
> It'll boil down to the same thing though.
>
No, in that case, I can take it myself.
Tom, does that work for you?
On Wed, 27 Oct 2021 at 17:11, Tom Lendacky <[email protected]> wrote:
>
> On 10/22/21 12:02 PM, Tom Lendacky wrote:
> > Reserving memory using efi_mem_reserve() calls into the x86
> > efi_arch_mem_reserve() function. This function will insert a new EFI
> > memory descriptor into the EFI memory map representing the area of
> > memory to be reserved and marking it as EFI runtime memory.
> >
> > As part of adding this new entry, a new EFI memory map is allocated and
> > mapped. The mapping is where a problem can occur. This new EFI memory map
> > is mapped using early_memremap(). If the allocated memory comes from an
> > area that is marked as EFI_BOOT_SERVICES_DATA memory in the current EFI
> > memory map, then it will be mapped unencrypted (see memremap_is_efi_data()
> > and the call to efi_mem_type()).
> >
> > However, during replacement of the old EFI memory map with the new EFI
> > memory map, efi_mem_type() is disabled, resulting in the new EFI memory
> > map always being mapped encrypted in efi.memmap. This will cause a kernel
> > crash later in the boot.
> >
> > Since it is known that the new EFI memory map will always be mapped
> > encrypted when efi_memmap_install() is called, explicitly map the new EFI
> > memory map as encrypted (using early_memremap_prot()) when inserting the
> > new memory map entry.
> >
> > Cc: <[email protected]> # 4.14.x
> > Fixes: 8f716c9b5feb ("x86/mm: Add support to access boot related data in the clear")
> > Acked-by: Ard Biesheuvel <[email protected]>
> > Signed-off-by: Tom Lendacky <[email protected]>
>
> Ard, are you going to take this through the EFI tree or does it need to go
> through another tree?
>
I could take it, but since it will ultimately go through -tip anyway,
perhaps better if they just take it directly? (This will change after
the next -rc1 though)
Boris?
On 10/22/21 12:02 PM, Tom Lendacky wrote:
> Reserving memory using efi_mem_reserve() calls into the x86
> efi_arch_mem_reserve() function. This function will insert a new EFI
> memory descriptor into the EFI memory map representing the area of
> memory to be reserved and marking it as EFI runtime memory.
>
> As part of adding this new entry, a new EFI memory map is allocated and
> mapped. The mapping is where a problem can occur. This new EFI memory map
> is mapped using early_memremap(). If the allocated memory comes from an
> area that is marked as EFI_BOOT_SERVICES_DATA memory in the current EFI
> memory map, then it will be mapped unencrypted (see memremap_is_efi_data()
> and the call to efi_mem_type()).
>
> However, during replacement of the old EFI memory map with the new EFI
> memory map, efi_mem_type() is disabled, resulting in the new EFI memory
> map always being mapped encrypted in efi.memmap. This will cause a kernel
> crash later in the boot.
>
> Since it is known that the new EFI memory map will always be mapped
> encrypted when efi_memmap_install() is called, explicitly map the new EFI
> memory map as encrypted (using early_memremap_prot()) when inserting the
> new memory map entry.
>
> Cc: <[email protected]> # 4.14.x
> Fixes: 8f716c9b5feb ("x86/mm: Add support to access boot related data in the clear")
> Acked-by: Ard Biesheuvel <[email protected]>
> Signed-off-by: Tom Lendacky <[email protected]>
Ard, are you going to take this through the EFI tree or does it need to go
through another tree?
Thanks,
Tom
>
> ---
> Changes for v2:
> - Update/expand commit message to (hopefully) make it easier to read and
> understand
> - Add a comment around the use of the early_memremap_prot() call
> ---
> arch/x86/platform/efi/quirks.c | 14 +++++++++++++-
> 1 file changed, 13 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/platform/efi/quirks.c b/arch/x86/platform/efi/quirks.c
> index b15ebfe40a73..14f8f20d727a 100644
> --- a/arch/x86/platform/efi/quirks.c
> +++ b/arch/x86/platform/efi/quirks.c
> @@ -277,7 +277,19 @@ void __init efi_arch_mem_reserve(phys_addr_t addr, u64 size)
> return;
> }
>
> - new = early_memremap(data.phys_map, data.size);
> + /*
> + * When SME is active, early_memremap() can map the memory unencrypted
> + * if the allocation came from EFI_BOOT_SERVICES_DATA (see
> + * memremap_is_efi_data() and the call to efi_mem_type()). However,
> + * when efi_memmap_install() is called to replace the memory map,
> + * efi_mem_type() is "disabled" and so the memory will always be mapped
> + * encrypted. To avoid this possible mismatch between the mappings,
> + * always map the newly allocated memmap memory as encrypted.
> + *
> + * When SME is not active, this behaves just like early_memremap().
> + */
> + new = early_memremap_prot(data.phys_map, data.size,
> + pgprot_val(pgprot_encrypted(FIXMAP_PAGE_NORMAL)));
> if (!new) {
> pr_err("Failed to map new boot services memmap\n");
> return;
>
On 10/27/21 11:59 AM, Ard Biesheuvel wrote:
> On Wed, 27 Oct 2021 at 18:56, Borislav Petkov <[email protected]> wrote:
>>
>> On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
>>> I could take it, but since it will ultimately go through -tip anyway,
>>> perhaps better if they just take it directly? (This will change after
>>> the next -rc1 though)
>>>
>>> Boris?
>>
>> Yeah, I'm being told this is not urgent enough to rush in now so you
>> could queue it into your fixes branch for 5.16 once -rc1 is out and send
>> it to Linus then. The stable tag is just so it gets backported to the
>> respective trees.
>>
>> But if you prefer I should take it, then I can queue it after -rc1.
>> It'll boil down to the same thing though.
>>
>
> No, in that case, I can take it myself.
>
> Tom, does that work for you?
Yup, that works for me. Thanks guys!
Tom
>
On 10/27/21 12:04 PM, Tom Lendacky wrote:
>
>
> On 10/27/21 11:59 AM, Ard Biesheuvel wrote:
>> On Wed, 27 Oct 2021 at 18:56, Borislav Petkov <[email protected]> wrote:
>>>
>>> On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
>>>> I could take it, but since it will ultimately go through -tip anyway,
>>>> perhaps better if they just take it directly? (This will change after
>>>> the next -rc1 though)
>>>>
>>>> Boris?
>>>
>>> Yeah, I'm being told this is not urgent enough to rush in now so you
>>> could queue it into your fixes branch for 5.16 once -rc1 is out and send
>>> it to Linus then. The stable tag is just so it gets backported to the
>>> respective trees.
>>>
>>> But if you prefer I should take it, then I can queue it after -rc1.
>>> It'll boil down to the same thing though.
>>>
>>
>> No, in that case, I can take it myself.
>>
>> Tom, does that work for you?
>
> Yup, that works for me. Thanks guys!
I don't see this in any tree yet, so just a gentle reminder in case it
dropped off the radar.
Thanks,
Tom
>
> Tom
>
>>
On Wed, 1 Dec 2021 at 15:06, Tom Lendacky <[email protected]> wrote:
>
> On 10/27/21 12:04 PM, Tom Lendacky wrote:
> >
> >
> > On 10/27/21 11:59 AM, Ard Biesheuvel wrote:
> >> On Wed, 27 Oct 2021 at 18:56, Borislav Petkov <[email protected]> wrote:
> >>>
> >>> On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
> >>>> I could take it, but since it will ultimately go through -tip anyway,
> >>>> perhaps better if they just take it directly? (This will change after
> >>>> the next -rc1 though)
> >>>>
> >>>> Boris?
> >>>
> >>> Yeah, I'm being told this is not urgent enough to rush in now so you
> >>> could queue it into your fixes branch for 5.16 once -rc1 is out and send
> >>> it to Linus then. The stable tag is just so it gets backported to the
> >>> respective trees.
> >>>
> >>> But if you prefer I should take it, then I can queue it after -rc1.
> >>> It'll boil down to the same thing though.
> >>>
> >>
> >> No, in that case, I can take it myself.
> >>
> >> Tom, does that work for you?
> >
> > Yup, that works for me. Thanks guys!
>
> I don't see this in any tree yet, so just a gentle reminder in case it
> dropped off the radar.
>
Apologies for the delay, I've pushed this out to -next now.
Before I send it to Linus, can you please confirm (for my peace of
mind) how this only affects systems that have memory encryption
available and enabled in the first place?
On 12/3/21 4:30 AM, Ard Biesheuvel wrote:
> On Wed, 1 Dec 2021 at 15:06, Tom Lendacky <[email protected]> wrote:
>>
>> On 10/27/21 12:04 PM, Tom Lendacky wrote:
>>>
>>>
>>> On 10/27/21 11:59 AM, Ard Biesheuvel wrote:
>>>> On Wed, 27 Oct 2021 at 18:56, Borislav Petkov <[email protected]> wrote:
>>>>>
>>>>> On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
>>>>>> I could take it, but since it will ultimately go through -tip anyway,
>>>>>> perhaps better if they just take it directly? (This will change after
>>>>>> the next -rc1 though)
>>>>>>
>>>>>> Boris?
>>>>>
>>>>> Yeah, I'm being told this is not urgent enough to rush in now so you
>>>>> could queue it into your fixes branch for 5.16 once -rc1 is out and send
>>>>> it to Linus then. The stable tag is just so it gets backported to the
>>>>> respective trees.
>>>>>
>>>>> But if you prefer I should take it, then I can queue it after -rc1.
>>>>> It'll boil down to the same thing though.
>>>>>
>>>>
>>>> No, in that case, I can take it myself.
>>>>
>>>> Tom, does that work for you?
>>>
>>> Yup, that works for me. Thanks guys!
>>
>> I don't see this in any tree yet, so just a gentle reminder in case it
>> dropped off the radar.
>>
>
> Apologies for the delay, I've pushed this out to -next now.
>
> Before I send it to Linus, can you please confirm (for my peace of
> mind) how this only affects systems that have memory encryption
> available and enabled in the first place?
Certainly.
An early_memremap() call uses the FIXMAP_PAGE_NORMAL protection value for
performing the mapping. Prior to performing the actual mapping though, a
call to early_memremap_pgprot_adjust() is made to possibly alter the
protection value, but only if memory encryption is active.
Changing the call to early_memremap_prot() and providing
pgprot_encrypted(FIXMAP_PAGE_NORMAL) as the protection value results in an
equivalent call to early_memremap() when memory encryption is not active.
This is because the pgprot_encrypted() is, in effect, a NOP when memory
encryption is not active.
So when memory encryption is not available or active, the result of an
early_memremap_prot() call with a protection value of
pgprot_encrypted(FIXMAP_PAGE_NORMAL) is equivalent to an early_memremap()
call.
Let me know if that answers your question.
Thanks,
Tom
>
On Fri, 3 Dec 2021 at 17:22, Tom Lendacky <[email protected]> wrote:
>
> On 12/3/21 4:30 AM, Ard Biesheuvel wrote:
> > On Wed, 1 Dec 2021 at 15:06, Tom Lendacky <[email protected]> wrote:
> >>
> >> On 10/27/21 12:04 PM, Tom Lendacky wrote:
> >>>
> >>>
> >>> On 10/27/21 11:59 AM, Ard Biesheuvel wrote:
> >>>> On Wed, 27 Oct 2021 at 18:56, Borislav Petkov <[email protected]> wrote:
> >>>>>
> >>>>> On Wed, Oct 27, 2021 at 05:14:35PM +0200, Ard Biesheuvel wrote:
> >>>>>> I could take it, but since it will ultimately go through -tip anyway,
> >>>>>> perhaps better if they just take it directly? (This will change after
> >>>>>> the next -rc1 though)
> >>>>>>
> >>>>>> Boris?
> >>>>>
> >>>>> Yeah, I'm being told this is not urgent enough to rush in now so you
> >>>>> could queue it into your fixes branch for 5.16 once -rc1 is out and send
> >>>>> it to Linus then. The stable tag is just so it gets backported to the
> >>>>> respective trees.
> >>>>>
> >>>>> But if you prefer I should take it, then I can queue it after -rc1.
> >>>>> It'll boil down to the same thing though.
> >>>>>
> >>>>
> >>>> No, in that case, I can take it myself.
> >>>>
> >>>> Tom, does that work for you?
> >>>
> >>> Yup, that works for me. Thanks guys!
> >>
> >> I don't see this in any tree yet, so just a gentle reminder in case it
> >> dropped off the radar.
> >>
> >
> > Apologies for the delay, I've pushed this out to -next now.
> >
> > Before I send it to Linus, can you please confirm (for my peace of
> > mind) how this only affects systems that have memory encryption
> > available and enabled in the first place?
>
> Certainly.
>
> An early_memremap() call uses the FIXMAP_PAGE_NORMAL protection value for
> performing the mapping. Prior to performing the actual mapping though, a
> call to early_memremap_pgprot_adjust() is made to possibly alter the
> protection value, but only if memory encryption is active.
>
> Changing the call to early_memremap_prot() and providing
> pgprot_encrypted(FIXMAP_PAGE_NORMAL) as the protection value results in an
> equivalent call to early_memremap() when memory encryption is not active.
> This is because the pgprot_encrypted() is, in effect, a NOP when memory
> encryption is not active.
>
> So when memory encryption is not available or active, the result of an
> early_memremap_prot() call with a protection value of
> pgprot_encrypted(FIXMAP_PAGE_NORMAL) is equivalent to an early_memremap()
> call.
>
> Let me know if that answers your question.
>
It does, thanks.