Re: RuntimeDXE unable to save variables to flash

[Hristo Mihaylov <hristo.mihaylov@prodrive-technologies.com>]

Thanks Laszlo,

> Accessing invalid MSRs may raise injections. How do you know the MSR 0x1FE is valid (and the Data32 value is valid)?

I'll investigate this further.

> That's an Fvb->Write() call. Do you have access to the source of the flash driver (which produces the FVB protocol instance)?

Yes, I found that the writing is done in a function called SendSpiCmd, here's the signature (I can't disclose the full source).

```
/**
  This function sends the programmed SPI command to the slave device.

  @param[in] This                 Pointer to the PCH_SPI_PROTOCOL instance.
  @param[in] SpiRegionType        The SPI Region type for flash cycle which is listed in the Descriptor
  @param[in] FlashCycleType       The Flash SPI cycle type list in HSFC (Hardware Sequencing Flash Control Register) register
  @param[in] Address              The Flash Linear Address must fall within a region for which BIOS has access permissions.
  @param[in] ByteCount            Number of bytes in the data portion of the SPI cycle.
  @param[in,out] Buffer           Pointer to caller-allocated buffer containing the dada received or sent during the SPI cycle.

  @retval EFI_SUCCESS             SPI command completes successfully.
  @retval EFI_DEVICE_ERROR        Device error, the command aborts abnormally.
  @retval EFI_ACCESS_DENIED       Some unrecognized command encountered in hardware sequencing mode
  @retval EFI_INVALID_PARAMETER   The parameters specified are not valid.
**/
EFI_STATUS
SendSpiCmd (
  IN     EFI_SPI_PROTOCOL   *This,
  IN     FLASH_REGION_TYPE  FlashRegionType,
  IN     FLASH_CYCLE_TYPE   FlashCycleType,
  IN     UINT32             Address,
  IN     UINT32             ByteCount,
  IN OUT UINT8              *Buffer
  )
```

It fails when it does a sanity check. It checks `if (Address + ByteCount) > FlashRegionSize`. The FlashRegionSize 
seems to be estimated correctly, because it's used to write other values to that region, which succeed.

When I log the sanity check like for the incorrect value I see: `SendSpiCmd: (46334052 + 60 = 46334112) > 20971520` 
the address seems to be way off.

The Address is calculated in FvbGetLbaAddress. Almost a 1 to 1 copy of 
https://github.com/tianocore/edk2/blob/75b7aa9528bdd05a7ecf4e64a6beb478d31b402c/OvmfPkg/QemuFlashFvbServicesRuntimeDxe/FwBlockService.c#L237

I thought that the issue is either because of invalid FV block or the size of the flash region is calculated
 incorrectly, but this only happens in SMM. Now I'm leaning more towards permissions.

Any advice?

Regards,
Hristo Mihaylov

-----Original Message-----
From: Laszlo Ersek [mailto:lersek@redhat.com] 
Sent: 04 October 2018 19:00
To: Hristo Mihaylov <hristo.mihaylov@prodrive-technologies.com>; edk2-devel@lists.01.org
Subject: Re: [edk2] RuntimeDXE unable to save variables to flash

On 10/04/18 12:56, Hristo Mihaylov wrote:
> Hello,
> 
> I have an issue that I don't know how to fix. I'm building a BIOS for 
> a custom
> x86_64 platform.
> 
> The first part of the problem is: the BIOS boots to the end of DXE 
> where it crashes with a General Protection exception.
> 
> ```
> !!!! X64 Exception Type - 0D(#GP - General Protection)  CPU Apic ID - 00000027 !!!!
> RIP  - 0000000077ADE823, CS  - 0000000000000038, RFLAGS - 
> 0000000000010047 ExceptionData - 0000000000000000 RAX  - 
> 00000000FFFFFFFF, RCX - 00000000000001FE, RDX - 0000000000000000 RBX  
> - 00000000FFFFFFFF, RSP - 0000000075B443C0, RBP - 00000000800FD000 RSI  
> - 0000000000000001, RDI - 0000000077AFC520
> R8   - 0000000077AFC618, R9  - 0000000000000001, R10 - 0000000070000000
> R11  - 00000000758B7F10, R12 - 0000000077AFC510, R13 - 
> 0000000000000001
> R14  - 0000000080000000, R15 - 0000000002C30064
> DS   - 0000000000000020, ES  - 0000000000000020, FS  - 0000000000000020
> GS   - 0000000000000020, SS  - 0000000000000020
> CR0  - 0000000080000033, CR2 - 0000000000000000, CR3 - 
> 0000000075977000
> CR4  - 0000000000000668, CR8 - 0000000000000000
> DR0  - 0000000000000000, DR1 - 0000000000000000, DR2 - 
> 0000000000000000
> DR3  - 0000000000000000, DR6 - 00000000FFFF0FF0, DR7 - 
> 0000000000000400 GDTR - 0000000075986928 000000000000004F, LDTR - 0000000000000000
> IDTR - 0000000077ACF400 00000000000001FF,   TR - 0000000000000040
> FXSAVE_STATE - 0000000075B44020
> ```
> 
> This exception occurs when the BIOS is enabling or disabling the write 
> protection of the SPI. It only crashes when it's doing the protection in SMM.
> The code responsible for that is:
> 
> ```
> EFI_STATUS
> EFIAPI
> DisableBiosWriteProtect (
>   VOID
>   )
> {
>   UINTN     SpiBaseAddress;
>   UINT32    Data32;
> 
>   SpiBaseAddress = MmPciBase (
>                      DEFAULT_PCI_BUS_NUMBER_PCH,
>                      PCI_DEVICE_NUMBER_PCH_SPI,
>                      PCI_FUNCTION_NUMBER_PCH_SPI
>                      );
>   // Write clear BC_SYNC_SS prior to change WPD from 0 to 1.
>   //
>   MmioOr8 (
>    SpiBaseAddress + R_PCH_SPI_BC + 1,
>     (B_PCH_SPI_BC_SYNC_SS >> 8)
>     );
>   ///
>   /// Set BIOSWE bit (SPI PCI Offset DCh [0]) = 1b
>   /// Enable the access to the BIOS space for both read and write cycles
>   ///
>   MmioOr8 (
>     SpiBaseAddress + R_PCH_SPI_BC,
>     B_PCH_SPI_BC_WPD
>     );
>   ///
>   /// PCH BIOS Spec Section 3.7 BIOS Region SMM Protection Enabling
>   /// If the following steps are implemented:
>   ///  - Set the EISS bit (SPI PCI Offset DCh [5]) = 1b
>   ///  - Follow the 1st recommendation in section 3.6
>   /// the BIOS Region can only be updated by following the steps bellow:
>   ///  - Once all threads enter SMM
>   ///  - Read memory location FED30880h OR with 00000001h, place the result in EAX,
>   ///    and write data to lower 32 bits of MSR 1FEh (sample code available)
>   ///  - Set BIOSWE bit (SPI PCI Offset DCh [0]) = 1b
>   ///  - Modify BIOS Region
>   ///  - Clear BIOSWE bit (SPI PCI Offset DCh [0]) = 0b
>   ///
>   if ((MmioRead8 (SpiBaseAddress + R_PCH_SPI_BC) & B_PCH_SPI_BC_EISS) != 0) {
>     ///
>     /// Read memory location FED30880h OR with 00000001h, place the result in EAX,
>     /// and write data to lower 32 bits of MSR 1FEh (sample code available)
>     ///
>     Data32 = MmioRead32 ((UINTN) (0xFED30880)) | BIT0;
> 
>     // crash occurs here
>     AsmWriteMsr32 (0x1FE, Data32);
>   }
> 
>   return EFI_SUCCESS;
> }
> 
> VOID
> EFIAPI
> EnableBiosWriteProtect (
>   VOID
>   )
> {
>   UINTN     SpiBaseAddress;
>   UINT32    Data32;
> 
>   SpiBaseAddress = MmPciBase (
>                      DEFAULT_PCI_BUS_NUMBER_PCH,
>                      PCI_DEVICE_NUMBER_PCH_SPI,
>                      PCI_FUNCTION_NUMBER_PCH_SPI
>                      );
>   ///
>   /// Clear BIOSWE bit (SPI PCI Offset DCh [0]) = 0b
>   /// Disable the access to the BIOS space for write cycles
>   ///
>   MmioAnd8 (
>     SpiBaseAddress + R_PCH_SPI_BC,
>     (UINT8) (~B_PCH_SPI_BC_WPD)
>     );
> 
>   ///
>   /// Check if EISS bit is set
>   ///
>   if (((MmioRead8 (SpiBaseAddress + R_PCH_SPI_BC)) & B_PCH_SPI_BC_EISS) == B_PCH_SPI_BC_EISS) {
>     ///
>     /// Read memory location FED30880h AND with FFFFFFFEh, place the result in EAX,
>     /// and write data to lower 32 bits of MSR 1FEh (sample code available)
>     ///
>     Data32 = MmioRead32 ((UINTN) (0xFED30880)) & (~BIT0);
> 
>     // crash occurs here
>     //AsmWriteMsr32 (0x1FE, Data32);
>   }
> }
> ```
> 
> I can work around the crash by hardcoding the values 0x0 and 0x1, or commenting out the writes.

Accessing invalid MSRs may raise injections. How do you know the MSR 0x1FE is valid (and the Data32 value is valid)?

> 
> The second part is that variables cannot be saved or updated:
> 
> ```
> Variable driver flush the HOB variable to flash: {guid} 
> TCG2_CONFIGURATION Invalid Parameter Variable driver flush the HOB 
> variable to flash: {guid} Setup Invalid Parameter Variable driver 
> flush the HOB variable to flash: {guid} PchRcConfiguration Invalid 
> Parameter Variable driver flush the HOB variable to flash: {guid} 
> MeRcConfiguration Invalid Parameter Variable driver flush the HOB 
> variable to flash: {guid} IeRcConfiguration Invalid Parameter Variable 
> driver flush the HOB variable to flash: {guid} SocketIioConfig Invalid 
> Parameter Variable driver flush the HOB variable to flash: {guid} 
> SocketCommonRcConfig Invalid Parameter Variable driver flush the HOB 
> variable to flash: {guid} SocketMpLinkConfig Invalid Parameter 
> Variable driver flush the HOB variable to flash: {guid} 
> SocketMemoryConfig Invalid Parameter Variable driver flush the HOB 
> variable to flash: {guid} SocketPowerManagementConfig Invalid 
> Parameter Variable driver flush the HOB variable to flash: {guid} 
> SocketProcessorCoreConfig Invalid Parameter Variable driver flush the 
> HOB variable to flash: {guid} FpgaSocketConfig Invalid Parameter ```
> 
> I can't save variables to non-volatile storage, I can save some to 
> volatile storage, but others I cannot save at all.
> 
> I tracked the failure down to here: 
> https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Universal/V
> ariable/RuntimeDxe/Variable.c#L338
> and I'm not sure how to proceed. The platform is based on UDK2015.

That's an Fvb->Write() call. Do you have access to the source of the flash driver (which produces the FVB protocol instance)?

(FWIW, EFI_INVALID_PARAMETER doesn't seem a spec-compliant return value for EFI_FIRMWARE_VOLUME_BLOCK2_PROTOCOL.Write()).

Thanks
Laszlo