Re: [PATCH v2 2/3] ArmPkg/Drivers: Add EFI_MM_COMMUNICATION_PROTOCOL DXE driver.

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On 13 November 2017 at 11:40, Supreeth Venkatesh
<supreeth.venkatesh@arm.com> wrote:
> On Mon, 2017-11-13 at 10:30 +0000, Ard Biesheuvel wrote:
>> On 25 October 2017 at 17:32, Supreeth Venkatesh
>> <supreeth.venkatesh@arm.com> wrote:
>> >
>> > PI v1.5 Specification Volume 4 defines Management Mode Core
>> > Interface
>> > and defines EFI_MM_COMMUNICATION_PROTOCOL. This protocol provides a
>> > means of communicating between drivers outside of MM and MMI
>> > handlers inside of MM.
>> >
>> > This patch implements the EFI_MM_COMMUNICATION_PROTOCOL DXE runtime
>> > driver for AARCH64 platforms. It uses SMCs allocated from the
>> > standard
>> > SMC range defined in
>> > http://infocenter.arm.com/help/topic/com.arm.doc.den0060a/DEN0060A_
>> > ARM_MM_Interface_Specification.pdf
>> > to communicate with the standalone MM environment in the secure
>> > world.
>> >
>> > Contributed-under: TianoCore Contribution Agreement 1.1
>> > Signed-off-by: Achin Gupta <achin.gupta@arm.com>
>> > Signed-off-by: Supreeth Venkatesh <supreeth.venkatesh@arm.com>
>> > ---
>> >  .../Drivers/MmCommunicationDxe/MmCommunication.c   | 339
>> > +++++++++++++++++++++
>> No need to separate the .inf and .c changes, so this patch can be
>> merged with the previous one.
> Thanks for more comments/feedback.
> Ok. Please wait for v3.
>>
>> >
>> >  1 file changed, 339 insertions(+)
>> >  create mode 100644
>> > ArmPkg/Drivers/MmCommunicationDxe/MmCommunication.c
>> >
>> > diff --git a/ArmPkg/Drivers/MmCommunicationDxe/MmCommunication.c
>> > b/ArmPkg/Drivers/MmCommunicationDxe/MmCommunication.c
>> > new file mode 100644
>> > index 0000000000..ecf70e666c
>> > --- /dev/null
>> > +++ b/ArmPkg/Drivers/MmCommunicationDxe/MmCommunication.c
>> > @@ -0,0 +1,339 @@
>> > +/** @file
>> > +
>> > +  Copyright (c) 2016-2017, ARM Limited. All rights reserved.
>> > +
>> > +  This program and the accompanying materials
>> > +  are licensed and made available under the terms and conditions
>> > of the BSD License
>> > +  which accompanies this distribution.  The full text of the
>> > license may be found at
>> > +  http://opensource.org/licenses/bsd-license.php
>> > +
>> > +  THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS"
>> > BASIS,
>> > +  WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER
>> > EXPRESS OR IMPLIED.
>> > +
>> > +**/
>> > +
>> > +#include <Library/ArmLib.h>
>> > +#include <Library/ArmSmcLib.h>
>> > +#include <Library/BaseMemoryLib.h>
>> > +#include <Library/DebugLib.h>
>> > +#include <Library/DxeServicesTableLib.h>
>> > +#include <Library/HobLib.h>
>> > +#include <Library/PcdLib.h>
>> > +#include <Library/UefiBootServicesTableLib.h>
>> > +#include <Library/UefiRuntimeServicesTableLib.h>
>> > +
>> > +#include <Protocol/MmCommunication.h>
>> > +
>> > +#include <IndustryStandard/ArmStdSmc.h>
>> > +
>> > +/**
>> > +  Communicates with a registered handler.
>> > +
>> > +  This function provides an interface to send and receive messages
>> > to the
>> > +  Standalone MM environment on behalf of UEFI services.  This
>> > function is part
>> > +  of the MM Communication Protocol that may be called in physical
>> > mode prior to
>> > +  SetVirtualAddressMap() and in virtual mode after
>> > SetVirtualAddressMap().
>> > +
>> > +  @param[in]      This                The
>> > EFI_MM_COMMUNICATION_PROTOCOL instance.
>> > +  @param[in, out] CommBuffer          A pointer to the buffer to
>> > convey into MMRAM.
>> > +  @param[in, out] CommSize            The size of the data buffer
>> > being passed in.On exit, the size of data
>> > +                                      being returned. Zero if the
>> > handler does not wish to reply with any data.
>> > +
>> > +  @retval EFI_SUCCESS                 The message was successfully
>> > posted.
>> > +  @retval EFI_INVALID_PARAMETER       The CommBuffer was NULL.
>> > +  @retval EFI_BAD_BUFFER_SIZE         The buffer is too large for
>> > the MM implementation. If this error is
>> > +                                      returned, the MessageLength
>> > field in the CommBuffer header or the integer
>> > +                                      pointed by CommSize are
>> > updated to reflect the maximum payload size the
>> > +                                      implementation can
>> > accommodate.
>> > +  @retval EFI_ACCESS_DENIED           The CommunicateBuffer
>> > parameter or CommSize parameter, if not omitted,
>> > +                                      are in address range that
>> > cannot be accessed by the MM environment
>> > +**/
>> Please don't use lines > 80 characters.
> Ok.
>>
>> >
>> > +EFI_STATUS
>> > +EFIAPI
>> > +MmCommunicationCommunicate (
>> > +  IN CONST EFI_MM_COMMUNICATION_PROTOCOL  *This,
>> > +  IN OUT VOID                             *CommBuffer,
>> > +  IN OUT UINTN                            *CommSize    OPTIONAL
>> > +  );
>> > +
>> > +//
>> > +// Address, Length of the pre-allocated buffer for communication
>> > with the secure
>> > +// world.
>> > +//
>> > +STATIC ARM_MEMORY_REGION_DESCRIPTOR  mNsCommBuffMemRegion;
>> > +
>> > +// Notification event when virtual address map is set.
>> > +STATIC EFI_EVENT  mSetVirtualAddressMapEvent;
>> > +
>> > +//
>> > +// Handle to install the MM Communication Protocol
>> > +//
>> > +STATIC EFI_HANDLE  mMmCommunicateHandle;
>> > +
>> > +//
>> > +// MM Communication Protocol instance
>> > +//
>> > +EFI_MM_COMMUNICATION_PROTOCOL  mMmCommunication = {
>> > +  MmCommunicationCommunicate
>> > +};
>> > +
>> Please move this below the definition of
>> MmCommunicationCommunicate(),
>> and drop the forward declaration.
> Ok.
>>
>> >
>> > +/**
>> > +  Communicates with a registered handler.
>> > +
>> > +  This function provides an interface to send and receive messages
>> > to the
>> > +  Standalone MM environment on behalf of UEFI services.  This
>> > function is part
>> > +  of the MM Communication Protocol that may be called in physical
>> > mode prior to
>> > +  SetVirtualAddressMap() and in virtual mode after
>> > SetVirtualAddressMap().
>> > +
>> > +  @param[in]      This                The
>> > EFI_MM_COMMUNICATION_PROTOCOL instance.
>> > +  @param[in, out] CommBuffer          A pointer to the buffer to
>> > convey into SMRAM.
>> What does 'convey into SMRAM' mean?
>>
>> >
>> > +  @param[in, out] CommSize            The size of the data buffer
>> > being passed in.On exit, the size of data
>> > +                                      being returned. Zero if the
>> > handler does not wish to reply with any data.
>> > +
>> CommSize is a pointer so it could be NULL but not zero. Or do you
>> mean
>> *CommSize?
>>
> CommSize is an optional parameter.

OK, please add OPTIONAL the prototype

> Yeah, if Commsize is used, then its
> value (*CommSize) will be zero, if there is no data. This is something
> the specification implicitly states.

OK


>> >
>> > +  @retval EFI_SUCCESS                 The message was successfully
>> > posted.
>> > +  @retval EFI_INVALID_PARAMETER       The CommBuffer was NULL.
>> > +  @retval EFI_BAD_BUFFER_SIZE         The buffer is too large for
>> > the MM implementation. If this error is
>> > +                                      returned, the MessageLength
>> > field in the CommBuffer header or the integer
>> > +                                      pointed by CommSize are
>> > updated to reflect the maximum payload size the
>> > +                                      implementation can
>> > accommodate.
>> > +  @retval EFI_ACCESS_DENIED           The CommunicateBuffer
>> > parameter or CommSize parameter, if not omitted,
>> > +                                      are in address range that
>> > cannot be accessed by the MM environment
>> > +**/
>> STATIC ?
>>
> This implements an EFIAPI for the protocol MM_COMMUNICATION_PROTOCOL in
> MdePkg.Not sure why you want this function static?

Any function that is only callable via function pointers should be static.

>> >
>> > +EFI_STATUS
>> > +EFIAPI
>> > +MmCommunicationCommunicate (
>> > +  IN CONST EFI_MM_COMMUNICATION_PROTOCOL  *This,
>> > +  IN OUT VOID                             *CommBuffer,
>> > +  IN OUT UINTN                            *CommSize
>> > +  )
>> > +{
>> > +  EFI_MM_COMMUNICATE_HEADER   *CommunicateHeader;
>> > +  ARM_SMC_ARGS                CommunicateSmcArgs;
>> > +  EFI_STATUS                  Status;
>> > +  UINTN                       BufferSize;
>> > +
>> > +  CommunicateHeader = CommBuffer;
>> > +  Status = EFI_SUCCESS;
>> > +  BufferSize = 0;
>> > +
>> > +  ZeroMem (&CommunicateSmcArgs, sizeof (ARM_SMC_ARGS));
>> > +
>> > +  //
>> > +  // Check parameters
>> > +  //
>> > +  if (CommBuffer == NULL) {
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  // If the length of the CommBuffer is 0 then return the expected
>> > length.
>> > +  if (CommSize) {
>> > +    if (*CommSize == 0) {
>> > +      *CommSize = mNsCommBuffMemRegion.Length;
>> > +      return EFI_BAD_BUFFER_SIZE;
>> Here you return EFI_BAD_BUFFER_SIZE for a 0 sized buffer, while the
>> comment says is it returned when the buffer is too large.
>>
> Specification states this:
> "If the CommSize parameter is passed into the call, but the integer it
> points to, has a value of 0, then this must be updated to reflect the
> maximum size of the CommBuffer that the implementation can tolerate."
> Also, Return Values are
> EFI_INVALID_PARAMETER - The buffer was NULL.
> EFI_BAD_BUFFER_SIZE  - The buffer is too large for the MM
> implementation. If this error is returned, the MessageLength field in
> the CommBuffer header or the integer pointed by CommSize, are updated
> to reflect the maximum payload size the implementation can accommodate.
> See the function description above for more details
>
> So, it does not specify what should be the return value, when *CommSize
> is zero. It is open to interpretation between EFI_INVALID_PARAMETER and
> EFI_BAD_BUFFER_SIZE. I choose EFI_BAD_BUFFER_SIZE.
> Let me know what you think.
>

If the specification does not state that an error should be returned,
we should clarify that first.

>> >
>> > +    }
>> > +    //
>> > +    // CommSize must hold HeaderGuid and MessageLength
>> > +    //
>> > +    if (*CommSize < sizeof (EFI_MM_COMMUNICATE_HEADER)) {
>> > +        return EFI_INVALID_PARAMETER;
>> > +    }
>> > +    BufferSize = *CommSize;
>> > +  } else {
>> > +    BufferSize = CommunicateHeader->MessageLength +
>> > +                 sizeof (CommunicateHeader->HeaderGuid) +
>> > +                 sizeof (CommunicateHeader->MessageLength);
>> > +  }
>> > +
>> > +  //
>> > +  // If the buffer size is 0 or greater than what can be tolerated
>> > by the MM
>> > +  // environment then return the expected size.
>> > +  //
>> > +  if ((BufferSize == 0) ||
>> > +      (BufferSize > mNsCommBuffMemRegion.Length)) {
>> > +    CommunicateHeader->MessageLength = mNsCommBuffMemRegion.Length
>> > -
>> > +                                       sizeof (CommunicateHeader-
>> > >HeaderGuid) -
>> > +                                       sizeof (CommunicateHeader-
>> > >MessageLength);
>> > +    return EFI_BAD_BUFFER_SIZE;
>> > +  }
>> > +
>> > +  // SMC Function ID
>> > +  CommunicateSmcArgs.Arg0 = ARM_SMC_ID_MM_COMMUNICATE_AARCH64;
>> > +
>> > +  // Reserved for Future. Must be Zero.
>> > +  CommunicateSmcArgs.Arg1 = 0;
>> > +
>> > +  if (mNsCommBuffMemRegion.VirtualBase) {
>> > +    CopyMem ((VOID *)mNsCommBuffMemRegion.VirtualBase, CommBuffer,
>> > BufferSize);
>> > +  }
>> > +
>> else? What happens if VirtualBase == NULL? Can we still proceed
>> without copying?
> Yup. This was Achin's comment as well. Please wait for v3.
>>
>> >
>> > +  // For the SMC64 version, this parameter is a 64-bit Physical
>> > Address (PA)
>> > +  // or Intermediate Physical Address (IPA).
>> > +  // For the SMC32 version, this parameter is a 32-bit PA or IPA.
>> > +  CommunicateSmcArgs.Arg2 =
>> > (UINTN)mNsCommBuffMemRegion.PhysicalBase;
>> > +
>> > +  // comm_size_address is a PA or an IPA that holds the size of
>> > the
>> > +  // communication buffer being passed in. This parameter is
>> > optional
>> > +  // and can be omitted by passing a zero.
>> > +  // ARM does not recommend using it since this might require the
>> > +  // implementation to create a separate memory mapping for the
>> > parameter.
>> > +  // ARM recommends storing the buffer size in the buffer itself.
>> > +  CommunicateSmcArgs.Arg3 = 0;
>> > +
>> > +  // Call the Standalone MM environment.
>> > +  ArmCallSmc (&CommunicateSmcArgs);
>> > +
>> > +  Status = CommunicateSmcArgs.Arg0;
>> > +  switch (Status) {
>> I think I mentioned this before, but please don't use Status for the
>> ARM_SMC_MM_xxx enum space. Use a separate, correctly typed variable
>> instead.
> Oh. your last email asks to replace EFI_SUCCESS with MM Specific RETURN
> SUCCESS. Status is still "UINTN" type variable for both MM and general
> UEFI status codes.
> Let me know if you feel strongly about it, I will have two different
> "UINTN" type variables.

Status should be EFI_STATUS not UINTN.

The ARM_SMC_MM_xxx symbolic constants are fundamentally a different
type, and you are relying on the fact that EFI_SUCCESS and
ARM_SMC_MM_RET_SUCCESS happen to be #defined to the same numeric
value. That is very bad coding style.

>>
>> >
>> > +  case ARM_SMC_MM_RET_SUCCESS:
>> > +    // On exit, the size of data being returned is inferred from
>> > +    // CommSize or MessageLength + Header.
>> > +    CopyMem (CommBuffer, (const VOID
>> > *)mNsCommBuffMemRegion.VirtualBase, BufferSize);
>> > +    break;
>> > +
>> > +  case ARM_SMC_MM_RET_NOT_SUPPORTED:
>> > +  case ARM_SMC_MM_RET_INVALID_PARAMS:
>> > +    Status = EFI_INVALID_PARAMETER;
>> > +    break;
>> > +
>> > +  case ARM_SMC_MM_RET_DENIED:
>> > +    Status = EFI_ACCESS_DENIED;
>> > +    break;
>> > +
>> > +  case ARM_SMC_MM_RET_NO_MEMORY:
>> > +    // Unexpected error since the CommSize was checked for zero
>> > length
>> > +    // prior to issuing the SMC
>> > +  default:
>> > +    Status = EFI_ACCESS_DENIED;
>> > +    ASSERT (0);
>> > +  }
>> > +
>> > +  return Status;
>> > +}
>> > +
>> > +/**
>> > +  Notification callback on SetVirtualAddressMap event.
>> > +
>> > +  This function notifies the MM communication protocol interface
>> > on
>> > +  SetVirtualAddressMap event and converts pointers used in this
>> > driver
>> > +  from physical to virtual address.
>> > +
>> > +  @param  Event          SetVirtualAddressMap event.
>> > +  @param  Context        A context when the SetVirtualAddressMap
>> > triggered.
>> > +
>> > +  @retval EFI_SUCCESS    The function executed successfully.
>> > +  @retval Other          Some error occurred when executing this
>> > function.
>> > +
>> > +**/
>> > +STATIC
>> > +VOID
>> > +EFIAPI
>> > +NotifySetVirtualAddressMap (
>> > +  IN EFI_EVENT  Event,
>> > +  IN VOID      *Context
>> > +  )
>> > +{
>> > +  EFI_STATUS  Status;
>> > +
>> > +  Status = gRT->ConvertPointer (EFI_OPTIONAL_PTR,
>> > +                                (VOID
>> > **)&mNsCommBuffMemRegion.VirtualBase
>> > +                               );
>> > +  if (EFI_ERROR (Status)) {
>> > +    DEBUG ((DEBUG_ERROR, "NotifySetVirtualAddressMap(): Unable to
>> > convert MM runtime pointer. Status:0x%x\n", Status));
>> Please use %r for reporting EFI_STATUS values.
>>
> Ok.
>> >
>> > +  }
>> > +
>> > +}
>> > +
>> > +/**
>> > +  The Entry Point for MM Communication
>> > +
>> > +  This function installs the MM communication protocol interface
>> > and finds out
>> > +  what type of buffer management will be required prior to
>> > invoking the
>> > +  communication SMC.
>> > +
>> > +  @param  ImageHandle    The firmware allocated handle for the EFI
>> > image.
>> > +  @param  SystemTable    A pointer to the EFI System Table.
>> > +
>> > +  @retval EFI_SUCCESS    The entry point is executed successfully.
>> > +  @retval Other          Some error occurred when executing this
>> > entry point.
>> > +
>> > +**/
>> > +EFI_STATUS
>> > +EFIAPI
>> > +MmCommunicationInitialize (
>> > +  IN EFI_HANDLE         ImageHandle,
>> > +  IN EFI_SYSTEM_TABLE  *SystemTable
>> > +  )
>> > +{
>> > +  EFI_STATUS                 Status;
>> > +
>> > +  mNsCommBuffMemRegion.PhysicalBase = 0;
>> > +  mNsCommBuffMemRegion.VirtualBase = 0;
>> > +  mNsCommBuffMemRegion.Length = 0;
>> > +
>> Please drop the 0 assignmentsm since they are overridden immediately.
> Ok.
>>
>> >
>> > +  mNsCommBuffMemRegion.PhysicalBase = PcdGet64 (PcdMmBufferBase);
>> > +  // During boot , Virtual and Physical are same
>> > +  mNsCommBuffMemRegion.VirtualBase =
>> > mNsCommBuffMemRegion.PhysicalBase;
>> > +  mNsCommBuffMemRegion.Length = PcdGet64 (PcdMmBufferSize);
>> > +
>> > +  if (mNsCommBuffMemRegion.PhysicalBase == 0) {
>> > +    DEBUG ((DEBUG_ERROR, "MmCommunicateInitialize: Invalid MM
>> > Buffer Base Address.\n"));
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  if (mNsCommBuffMemRegion.Length == 0) {
>> > +    DEBUG ((DEBUG_ERROR, "MmCommunicateInitialize: Maximum Buffer
>> > Size is zero.\n"));
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  Status = gDS->AddMemorySpace (EfiGcdMemoryTypeSystemMemory,
>> > +                                mNsCommBuffMemRegion.PhysicalBase,
>> > +                                mNsCommBuffMemRegion.Length,
>> > +                                EFI_MEMORY_WB |
>> > +                                EFI_MEMORY_XP |
>> > +                                EFI_MEMORY_RUNTIME);
>> > +  if (EFI_ERROR (Status)) {
>> > +    DEBUG ((DEBUG_ERROR, "MmCommunicateInitialize: Failed to add
>> > MM-NS Buffer Memory Space\n"));
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  Status = gDS-
>> > >SetMemorySpaceAttributes(mNsCommBuffMemRegion.PhysicalBase,
>> > +                                         mNsCommBuffMemRegion.Leng
>> > th,
>> > +                                         EFI_MEMORY_WB |
>> > EFI_MEMORY_XP);
>> > +  if (EFI_ERROR (Status)) {
>> > +    DEBUG ((DEBUG_ERROR, "MmCommunicateInitialize: Failed to set
>> > MM-NS Buffer Memory attributes\n"));
>> cleanup?
> Yup. Wait for v3.
>>
>> >
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  Status = gBS->AllocatePages (AllocateAddress,
>> > +                               EfiRuntimeServicesData,
>> > +                               EFI_SIZE_TO_PAGES
>> > (mNsCommBuffMemRegion.Length),
>> > +                               &mNsCommBuffMemRegion.PhysicalBase)
>> > ;
>> > +  if (EFI_ERROR (Status)) {
>> > +    DEBUG ((DEBUG_ERROR, "MmCommunicateInitialize: Failed to
>> > allocate MM-NS Buffer Memory Space\n"));
>> cleanup?
>>
> This was Udit's comment as well. Wait for v3.
>> >
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  // Install the communication protocol
>> > +  Status = gBS->InstallProtocolInterface (&mMmCommunicateHandle,
>> > +                                          &gEfiMmCommunicationProt
>> > ocolGuid,
>> > +                                          EFI_NATIVE_INTERFACE,
>> > +                                          &mMmCommunication);
>> > +  if (EFI_ERROR(Status)) {
>> > +    DEBUG ((DEBUG_ERROR, "MmCommunicationInitialize: Failed to
>> > install MM communication protocol\n"));
>> cleanup?
> Yup. Wait for v3.
>>
>> >
>> > +    return EFI_INVALID_PARAMETER;
>> > +  }
>> > +
>> > +  // Register notification callback when  virtual address is
>> > associated
>> > +  // with the physical address.
>> > +  // Create a Set Virtual Address Map event.
>> > +  //
>> > +  Status = gBS->CreateEvent
>> > (EVT_SIGNAL_VIRTUAL_ADDRESS_CHANGE,  // Type
>> > +                             TPL_NOTIFY,                         /
>> > / NotifyTpl
>> > +                             NotifySetVirtualAddressMap,         /
>> > / NotifyFunction
>> > +                             NULL,                               /
>> > / NotifyContext
>> > +                             &mSetVirtualAddressMapEvent         /
>> > / Event
>> > +                            );
>> > +  ASSERT_EFI_ERROR (Status);
>> > +
>> cleanup?
> Not required. Since it will still work at boot time.

If gBS->CreateEvent () fails in a RELEASE build, it will ignore the
ASSERT () and return the error code, which will result in the driver
to be unloaded. Without cleanup, that will leave protocol pointers
pointing into memory which has now been freed. Also, the buffer
allocation will not be freed either.