Re: [edk2-devel] [PATCH RFC v2 04/28] MdePkg: Define the Page State Change VMGEXIT structures

["Laszlo Ersek" <lersek@redhat.com>]

On 04/30/21 13:51, Brijesh Singh wrote:
> BZ: https://bugzilla.tianocore.org/show_bug.cgi?id=3275
> 
> The Page State Change NAE exit will be used by the SEV-SNP guest to
> request a page state change using the GHCB protocol. See the GHCB
> spec section 4.1.6 and 2.3.1 for more detail on the structure
> definitions.
> 
> Cc: James Bottomley <jejb@linux.ibm.com>
> Cc: Min Xu <min.m.xu@intel.com>
> Cc: Jiewen Yao <jiewen.yao@intel.com>
> Cc: Tom Lendacky <thomas.lendacky@amd.com>
> Cc: Jordan Justen <jordan.l.justen@intel.com>
> Cc: Ard Biesheuvel <ardb+tianocore@kernel.org>
> Cc: Laszlo Ersek <lersek@redhat.com>
> Cc: Erdem Aktas <erdemaktas@google.com>
> Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
> ---
>  MdePkg/Include/Register/Amd/Fam17Msr.h | 15 ++++++++++
>  MdePkg/Include/Register/Amd/Ghcb.h     | 29 ++++++++++++++++++++
>  2 files changed, 44 insertions(+)
> 
> diff --git a/MdePkg/Include/Register/Amd/Fam17Msr.h b/MdePkg/Include/Register/Amd/Fam17Msr.h
> index e19bd04b6c..432cee2feb 100644
> --- a/MdePkg/Include/Register/Amd/Fam17Msr.h
> +++ b/MdePkg/Include/Register/Amd/Fam17Msr.h
> @@ -58,6 +58,19 @@ typedef union {
>      UINT64  GuestFrameNumber:52;
>    } GhcbGpaRegister;
>  
> +  struct {
> +    UINT64 Function:12;
> +    UINT64 GuestFrameNumber:40;
> +    UINT64 Operation:4;
> +    UINT64 Reserved:8;
> +  } SnpPageStateChangeRequest;
> +
> +  struct {
> +    UINT32 Function:12;
> +    UINT32 Reserved:20;
> +    UINT32 ErrorCode;
> +  } SnpPageStateChangeResponse;
> +
>    VOID    *Ghcb;
>  
This matches section 2.3.1 in rev 2.00.

>    UINT64  GhcbPhysicalAddress;
> @@ -69,6 +82,8 @@ typedef union {
>  #define GHCB_INFO_CPUID_RESPONSE           5
>  #define GHCB_INFO_GHCB_GPA_REGISTER_REQUEST   18
>  #define GHCB_INFO_GHCB_GPA_REGISTER_RESPONSE  19
> +#define GHCB_INFO_SNP_PAGE_STATE_CHANGE_REQUEST   20
> +#define GHCB_INFO_SNP_PAGE_STATE_CHANGE_RESPONSE  21
>  #define GHCB_HYPERVISOR_FEATURES_REQUEST   128
>  #define GHCB_HYPERVISOR_FEATURES_RESPONSE  129
>  #define GHCB_INFO_TERMINATE_REQUEST        256

Matches section 2.3.1.

> diff --git a/MdePkg/Include/Register/Amd/Ghcb.h b/MdePkg/Include/Register/Amd/Ghcb.h
> index 2d64a4c28f..1e7c0daed3 100644
> --- a/MdePkg/Include/Register/Amd/Ghcb.h
> +++ b/MdePkg/Include/Register/Amd/Ghcb.h
> @@ -54,6 +54,7 @@
>  #define SVM_EXIT_NMI_COMPLETE   0x80000003ULL
>  #define SVM_EXIT_AP_RESET_HOLD  0x80000004ULL
>  #define SVM_EXIT_AP_JUMP_TABLE  0x80000005ULL
> +#define SVM_EXIT_SNP_PAGE_STATE_CHANGE    0x80000010ULL
>  #define SVM_EXIT_HYPERVISOR_FEATURES  0x8000FFFDULL
>  #define SVM_EXIT_UNSUPPORTED    0x8000FFFFULL
>  

Matches "Table 5. List of Supported Non-Automatic Events".

> @@ -160,4 +161,32 @@ typedef union {
>  #define GHCB_HV_FEATURES_SNP_AP_CREATE                    (GHCB_HV_FEATURES_SNP | BIT1)
>  #define GHCB_HV_FEATURES_SNP_RESTRICTED_INJECTION         (GHCB_HV_FEATURES_SNP_AP_CREATE | BIT2)
>  #define GHCB_HV_FEATURES_SNP_RESTRICTED_INJECTION_TIMER   (GHCB_HV_FEATURES_SNP_RESTRICTED_INJECTION | BIT3)
> +
> +// SNP Page State Change

(1) Comment style.

> +#define SNP_PAGE_STATE_MAX_NPAGES           4095
> +#define SNP_PAGE_STATE_MAX_ENTRY            253
> +#define SNP_PAGE_STATE_PRIVATE              1
> +#define SNP_PAGE_STATE_SHARED               2
> +#define SNP_PAGE_STATE_PSMASH               3
> +#define SNP_PAGE_STATE_UNSMASH              4

(2) The PSMASH and UNSMASH operations are not documented in the rev 2.00
spec, in the GHCB MSR protocol. That's probably because PSMASH and
UNSMASH can only be defined in terms of 2MB pages, and
GHCB_INFO_SNP_PAGE_STATE_CHANGE_REQUEST is suitable only for individual,
4KB pages. I think it would be useful to point out somehow here that
PSMASH and UNSMASH are restricted to the GHCB shared area protocol
(perhaps extend the leading comment on this block of macros).

(3) I don't understand what "MAX_NPAGES" stands for (4095). The rest of
the series never uses the macro, and I can't associate it with anything
from the spec. If the macro is supposed to relate to the 4KB / 2MB page
smashing / splitting, then its replacement text should be 512. Unless
the macro corresponds to a definition in the spec, I think we should
drop it.

> +
> +typedef PACKED struct {
> +  UINT64  CurrentPage:12;
> +  UINT64  GuestFrameNumber:40;
> +  UINT64  Op:4;
> +  UINT64  PageSize:1;
> +  UINT64  Rsvd: 7;
> +} SNP_PAGE_STATE_ENTRY;
> +
> +typedef PACKED struct {
> +  UINT16 CurrentEntry;
> +  UINT16 EndEntry;
> +  UINT32 Rsvd;
> +} SNP_PAGE_STATE_HEADER;

(4) We tend to write

#pragma pack (1)
...
#pragma pack ()

rather than PACKED -- but anyway, is packing really necessary? "Natural
alignment" is required in edk2. I'm OK with packing, but I think the
pragma is the preferred form.

(5) Please spell out both "Rsvd" fields above as "Reserved".

(6) Stray space character in "Rsvd: 7".

(7) The field name "Op" is inconsistent with the other field name
"Operation".

(8) I think there is a bug (typo) in the rev 2.00 spec, in 4.1.6 "SNP
Page State Change": it says

  ... calculated from the supplied guest physical frame number (GFN) for
  the requested page size (GPA = GFN << 12).

But, if you can choose 2MB page size in the request, then the (GPA = GFN
<< 12) formula is not g

(9) If my understanding of the spec is correct, "EndEntry" has
*inclusive* meaning. That's unusual. Any particular reason for not
making "EndEntry" exclusive (in the spec)?

> +
> +typedef struct {
> +  SNP_PAGE_STATE_HEADER  Header;
> +  SNP_PAGE_STATE_ENTRY   Entry[SNP_PAGE_STATE_MAX_ENTRY];
> +} SNP_PAGE_STATE_CHANGE_INFO;
> +
>  #endif
> 

Yes, this looks OK. Size is 2+2+4+253*8 = 2032 bytes, which matches the
size of GHCB.SharedBuffer.

(10) However, *if* you decide to declare SNP_PAGE_STATE_ENTRY and
SNP_PAGE_STATE_HEADER explicitly as packed, then you should do the same
for SNP_PAGE_STATE_CHANGE_INFO.

(11) Like I mentioned earlier, it's probably helpful if you start the
subject line with

  MdePkg/Register/Amd: ...

on all of these MdePkg patches. If that becomes too tight, for some of
the MdePkg patches, then I suggest "MdePkg/Amd: ..." (i.e., drop
"Register").

Thanks
Laszlo