Re: [PATCH v3 0/5] x86: SME: BSP/SME microcode update fix

[Tom Lendacky <thomas.lendacky@amd.com>]

On 1/19/2018 10:02 AM, Gabriel C wrote:
> 2018-01-19 16:56 GMT+01:00 Tom Lendacky <thomas.lendacky@amd.com>:
>> On 1/19/2018 9:35 AM, Greg Kroah-Hartman wrote:
>>> On Fri, Jan 19, 2018 at 09:27:47AM -0600, Tom Lendacky wrote:
>>>> On 1/19/2018 9:11 AM, Greg Kroah-Hartman wrote:
>>>>> On Fri, Jan 19, 2018 at 09:03:52AM -0600, Tom Lendacky wrote:
>>>>>> On 1/15/2018 4:47 PM, Gabriel C wrote:
>>>>>>> On 11.01.2018 19:33, Borislav Petkov wrote:
>>>>>>>> On Wed, Jan 10, 2018 at 01:25:45PM -0600, Tom Lendacky wrote:
>>>>>>>>> This patch series addresses an issue when SME is active and the BSP
>>>>>>>>> is attempting to check for and load microcode during load_ucode_bsp().
>>>>>>>>> Since the initrd has not been decrypted (yet) and the virtual address
>>>>>>>>> of the initrd treats the memory as encrypted, the CPIO archive parsing
>>>>>>>>> fails to locate the microcode.
>>>>>>>>>
>>>>>>>>> This series moves the encryption of the initrd into the early boot code
>>>>>>>>> and encrypts it at the same time that the kernel is encrypted.  Since
>>>>>>>>> the initrd is now encrypted, the CPIO archive parsing succeeds in
>>>>>>>>> properly locating the microcode.
>>>>>>>>>
>>>>>>>>> The following patches are included in this fix:
>>>>>>>>> - Cleanup register saving in arch/x86/mm/mem_encrypt_boot.S
>>>>>>>>> - Reduce parameters and complexity for creating the SME PGD mappings
>>>>>>>>> - Centralize the use of the PMD flags used in sme_encrypt_kernel() in
>>>>>>>>>    preparation for using PTE flags also.
>>>>>>>>> - Prepare sme_encrypt_kernel() to handle PAGE aligned encryption, not
>>>>>>>>>    just 2MB large page aligned encryption.
>>>>>>>>> - Encrypt the initrd in sme_encrypt_kernel() when the kernel is being
>>>>>>>>>    encrypted.
>>>>>>>>>
>>>>>>>>> This patch series is based on tip/master.
>>>>>>>>>
>>>>>>>>> ---
>>>>>>>>>
>>>>>>>>> Changes from v2:
>>>>>>>>> - General code cleanup based on feedback.
>>>>>>>>>
>>>>>>>>> Changes from v1:
>>>>>>>>> - Additional patch to cleanup the register saving performed in
>>>>>>>>>    arch/x86/mm/mem_encrypt_boot.S in prep for changes made in the
>>>>>>>>>    remainder of the patchset.
>>>>>>>>> - Additional patch to reduce parameters and complexity for creating the
>>>>>>>>>    SME PGD mappings by introducing and using a structure for referencing
>>>>>>>>>    the PGD to populate, the pagetable allocation area, the
>>>>>>>>> virtual/physical
>>>>>>>>>    addresses being mapped and the pagetable flags to be used.
>>>>>>>>> - Consolidate PMD/PTE mapping code to reduce duplication.
>>>>>>>>>
>>>>>>>>> Tom Lendacky (5):
>>>>>>>>>        x86/mm: Cleanup register saving in mem_encrypt_boot.S
>>>>>>>>>        x86/mm: Use a struct to reduce parameters for SME PGD mapping
>>>>>>>>>        x86/mm: Centralize PMD flags in sme_encrypt_kernel()
>>>>>>>>>        x86/mm: Prepare sme_encrypt_kernel() for PAGE aligned encryption
>>>>>>>>>        x86/mm: Encrypt the initrd earlier for BSP microcode update
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>   arch/x86/include/asm/mem_encrypt.h |    4
>>>>>>>>>   arch/x86/kernel/head64.c           |    4
>>>>>>>>>   arch/x86/kernel/setup.c            |   10 -
>>>>>>>>>   arch/x86/mm/mem_encrypt.c          |  356
>>>>>>>>> ++++++++++++++++++++++++++----------
>>>>>>>>>   arch/x86/mm/mem_encrypt_boot.S     |   80 ++++----
>>>>>>>>>   5 files changed, 308 insertions(+), 146 deletions(-)
>>>>>>>>
>>>>>>>> All 5:
>>>>>>>>
>>>>>>>> Reviewed-by: Borislav Petkov <bp@suse.de>
>>>>>>>>
>>>>>>>
>>>>>>> Guys , are these patches going to be part of 4.15 ?
>>>>>>>
>>>>>>> With mem_encrypt=on without these patches microcode loading doesn't
>>>>>>> work right. Also @stable 4.14 would need the fixes too.
>>>>>>
>>>>>> It looks like these patches have been pulled into 4.15.  I did forget
>>>>>> to cc stable, so I'll follow-up with a separate email to have these
>>>>>> back-ported to the 4.14 stable tree.
>>>>>
>>>>> What are the git commit ids?   That's all I need :)
>>>>
>>>> Hi Greg,
>>>>
>>>> Here are the commit ids:
>>>>   1303880179e6 (“x86/mm: Clean up register saving in the __enc_copy() assembly code”)
>>>>   bacf6b499e11 (“x86/mm: Use a struct to reduce parameters for SME PGD mapping”)
>>>>   2b5d00b6c2cd (“x86/mm: Centralize PMD flags in sme_encrypt_kernel()”)
>>>>   cc5f01e28d6c (“x86/mm: Prepare sme_encrypt_kernel() for PAGE aligned encryption”)
>>>>   107cd2532181 (“x86/mm: Encrypt the initrd earlier for BSP microcode update”)
>>>>
>>>> The last commit won't apply cleanly on 4.14.  There was a change in
>>>> arch/x86/kernel/setup.c for SEV support.  The actual patch to that file
>>>> is very small it just removes the call to sme_early_encrypt() and the
>>>> associated comment.  I can submit a new version of that patch if you
>>>> want, just let me know.
>>>
>>> A backported version of that would be great, thanks.
>>
>> Ok, I'll send that out as soon as possible.  Since it is a changed patch
>> I was planning to remove the Tested-by, Signed-off-by (except for my sign
>> off), etc. or would you prefer I leave them in this case?
>>
> 
> I tested the series on top 4.14.13/.14 already , the conflict is
> trivial and easy to fix.
> 
> If you wish you can keep my Tested-by.

Ok, a backported patch for 4.14 has been sent to stable@vger.kernel.org
(with everyone on this thread cc'd).  Look for the subject:

[PATCH] x86/mm: Encrypt the initrd earlier for BSP microcode update

Thanks,
Tom

> 
> Regards,
> 
> Gabriel C
>