Re: [EXT] RE: [edk2-stable202002][edk2-devel] [PATCH v2 1/1] MdeModulePkg/Pci: Fixed Asserts in SCT PCIIO Protocol Test.

["Wu, Hao A" <hao.a.wu@intel.com>]

> -----Original Message-----
> From: devel@edk2.groups.io [mailto:devel@edk2.groups.io] On Behalf Of
> Gaurav Jain
> Sent: Monday, February 24, 2020 4:43 PM
> To: Wu, Hao A; devel@edk2.groups.io; Gao, Liming; afish@apple.com;
> lersek@redhat.com; leif@nuviainc.com; Kinney, Michael D
> Cc: Wang, Jian J; Ni, Ray; Ard Biesheuvel; Pankaj Bansal
> Subject: Re: [EXT] RE: [edk2-stable202002][edk2-devel] [PATCH v2 1/1]
> MdeModulePkg/Pci: Fixed Asserts in SCT PCIIO Protocol Test.
> 
> 
> 
> > -----Original Message-----
> > From: Wu, Hao A <hao.a.wu@intel.com>
> > Sent: Monday, February 24, 2020 1:56 PM
> > To: Gaurav Jain <gaurav.jain@nxp.com>; devel@edk2.groups.io; Gao,
> Liming
> > <liming.gao@intel.com>; afish@apple.com; lersek@redhat.com;
> > leif@nuviainc.com; Kinney, Michael D <michael.d.kinney@intel.com>
> > Cc: Wang, Jian J <jian.j.wang@intel.com>; Ni, Ray <ray.ni@intel.com>; Ard
> > Biesheuvel <ard.biesheuvel@linaro.org>; Pankaj Bansal
> > <pankaj.bansal@nxp.com>
> > Subject: RE: [EXT] RE: [edk2-stable202002][edk2-devel] [PATCH v2 1/1]
> > MdeModulePkg/Pci: Fixed Asserts in SCT PCIIO Protocol Test.
> >
> > Caution: EXT Email
> >
> > > -----Original Message-----
> > > From: Gaurav Jain [mailto:gaurav.jain@nxp.com]
> > > Sent: Monday, February 24, 2020 3:04 PM
> > > To: Wu, Hao A; devel@edk2.groups.io; Gao, Liming; afish@apple.com;
> > > lersek@redhat.com; leif@nuviainc.com; Kinney, Michael D
> > > Cc: Wang, Jian J; Ni, Ray; Ard Biesheuvel; Pankaj Bansal
> > > Subject: RE: [EXT] RE: [edk2-stable202002][edk2-devel] [PATCH v2 1/1]
> > > MdeModulePkg/Pci: Fixed Asserts in SCT PCIIO Protocol Test.
> > >
> > >
> > > > I think the above check for 'Attributes' can be dropped.
> > > > I found that the implementation of the PciIoGetBarAttributes()
> > > > function
> > > does not
> > > > expose any configurable attributes. So the logic can fall through to
> > > > the
> > > ASSERT
> > > > (for DEBUG images) and then returns EFI_UNSUPPORTED.
> > >
> > > I agree that PciIoGetBarAttributes() function sets *Supports as 0.
> > > But In SCT Test for SetBarAttributes, there is a test case for
> > > Unsupported Attribute which expects EFI_UNSUPPORTED. If I drop this
> > > check, ASSERT will come, which is not expected.
> > > Can we keep check for 'Attributes'?
> >
> >
> > Oh, I forgot that.
> >
> > I have one question, is there any special reason for you to pick the
> supported
> > bits specified by:
> > EFI_PCI_DEVICE_ENABLE |
> EFI_PCI_IO_ATTRIBUTE_DUAL_ADDRESS_CYCLE
> >
> > Is it relating with the SCT test case?
> 
> In PciIoAttributes() function, I can see the code
> #define DEV_SUPPORTED_ATTRIBUTES \
>     (EFI_PCI_DEVICE_ENABLE |
> EFI_PCI_IO_ATTRIBUTE_DUAL_ADDRESS_CYCLE)
> So I used the same bits in PciIoSetBarAttributes() to have a check for valid
> attributes.


Got it. I am fine to put the below check for 'Attributes' in PciIoAttributes():

  if ((Attributes & (~DEV_SUPPORTED_ATTRIBUTES)) != 0) {
    return EFI_UNSUPPORTED;
  }

Since the definition "DEV_SUPPORTED_ATTRIBUTES" will be used multiple times in
the driver, I suggest to remove the duplicate definitions in each function and
place it under file NonDiscoverablePciDeviceIo.h.

Hello Ard, do you have any concern for this? Thanks.

Best Regards,
Hao Wu


> 
> In SCT Test code
> First get the Bar attributes and set one of Unsupported attribute bit.
> Call PciIoSetBarAttributes() with Unsupported attribute and in return, test
> expects EFI_UNSUPPORTED.
> 
> Regards
> Gaurav Jain
> >
> > Best Regards,
> > Hao Wu
> >
> >
> > >
> > > > -----Original Message-----
> > > > From: Wu, Hao A <hao.a.wu@intel.com>
> > > > Sent: Friday, February 21, 2020 6:53 AM
> > > > To: devel@edk2.groups.io; Gaurav Jain <gaurav.jain@nxp.com>; Gao,
> > > Liming
> > > > <liming.gao@intel.com>; afish@apple.com; lersek@redhat.com;
> > > > leif@nuviainc.com; Kinney, Michael D <michael.d.kinney@intel.com>
> > > > Cc: Wang, Jian J <jian.j.wang@intel.com>; Ni, Ray
> > > > <ray.ni@intel.com>; Ard Biesheuvel <ard.biesheuvel@linaro.org>;
> > > > Pankaj Bansal <pankaj.bansal@nxp.com>
> > > > Subject: [EXT] RE: [edk2-stable202002][edk2-devel] [PATCH v2 1/1]
> > > > MdeModulePkg/Pci: Fixed Asserts in SCT PCIIO Protocol Test.
> > > >
> > > > Caution: EXT Email
> > > >
> > > > A couple of inline comments below. Please help to handle them in the
> > > > next version of patch.
> > > > With them addressed,
> > > > Reviewed-by: Hao A Wu <hao.a.wu@intel.com>
> > > >
> > > >
> > > > Hello Liming and Stewards,
> > > >
> > > > I would like to confirm with you for whether the patch should catch
> > > > the upcoming stable tag.
> > > >
> > > > My personal take is that the patch is more like a code refinement
> > > > rather
> > > than a
> > > > bug fix.
> > > >
> > > > Could you help to make a final call for this one? Thanks in advance.
> > > >
> > > > Best Regards,
> > > > Hao Wu
> > > >
> > > >
> > > > > -----Original Message-----
> > > > > From: devel@edk2.groups.io [mailto:devel@edk2.groups.io] On
> Behalf
> > > Of
> > > > > Gaurav Jain
> > > > > Sent: Thursday, February 20, 2020 11:40 PM
> > > > > To: devel@edk2.groups.io
> > > > > Cc: Wang, Jian J; Wu, Hao A; Ni, Ray; Ard Biesheuvel; Pankaj
> > > > > Bansal; Gaurav Jain
> > > > > Subject: [edk2-devel] [PATCH v2 1/1] MdeModulePkg/Pci: Fixed
> > > > > Asserts in SCT PCIIO Protocol Test.
> > > > >
> > > > > ASSERT in PollMem_Conf, CopyMem_Conf, SetBarAttributes_Conf
> > > > > Conformance Test.
> > > > > SCT Test expect return as Invalid Parameter or Unsupported.
> > > > > Added Checks for Function Parameters.
> > > > > return Invalid or Unsupported if Check fails.
> > > > >
> > > > > Added Checks in PciIoPollIo(), PciIoIoRead()
> > > > > PciIoIoWrite()
> > > > >
> > > > > Signed-off-by: Gaurav Jain <gaurav.jain@nxp.com>
> > > > > ---
> > > > >
> > > > > Notes:
> > > > >     v2
> > > > >     - Reverted ASSERT(FALSE) code.
> > > > >     - Added Checks for Width, BarIndex, Buffer,
> > > > >       Address range in PciIoIoRead, PciIoIoWrite.
> > > > >     - Added Checks for Width, BarIndex, Result,
> > > > >       Address range in PciIoPollIo, PciIoPollMem,
> > > > >       PciIoCopyMem.
> > > > >     - Added Checks for Attributes, BarIndex,
> > > > >       Address range in PciIoSetBarAttributes.
> > > > >
> > > > >  .../NonDiscoverablePciDeviceIo.c              | 180 ++++++++++++++++++
> > > > >  1 file changed, 180 insertions(+)
> > > > >
> > > > > diff --git
> > > > >
> > >
> a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverableP
> > > > > ciDeviceIo.c
> > > > >
> > >
> b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverableP
> > > > > ciDeviceIo.c
> > > > > index 2d55c9699322..4dd804356021 100644
> > > > > ---
> > > > >
> > >
> a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverableP
> > > > > ciDeviceIo.c
> > > > > +++
> > > > >
> > >
> b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverableP
> > > > > ciDeviceIo.c
> > > > > @@ -93,6 +93,35 @@ PciIoPollMem (
> > > > >    OUT UINT64                      *Result
> > > > >    )
> > > > >  {
> > > > > +  NON_DISCOVERABLE_PCI_DEVICE         *Dev;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *Desc;
> > > > > +  UINTN                               Count;
> > > > > +  EFI_STATUS                          Status;
> > > > > +
> > > > > +  if ((UINT32)Width > EfiPciIoWidthUint64) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  if (BarIndex >= PCI_MAX_BAR) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  if (Result == NULL) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO(This);
> > > > > +  Count = 1;
> > > > > +
> > > > > +  Status = GetBarResource (Dev, BarIndex, &Desc);  if (EFI_ERROR
> > > > > + (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (Offset + (Count << (Width & 0x3)) > Desc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > >    ASSERT (FALSE);
> > > > >    return EFI_UNSUPPORTED;
> > > > >  }
> > > > > @@ -126,6 +155,35 @@ PciIoPollIo (
> > > > >    OUT UINT64                      *Result
> > > > >    )
> > > > >  {
> > > > > +  NON_DISCOVERABLE_PCI_DEVICE         *Dev;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *Desc;
> > > > > +  UINTN                               Count;
> > > > > +  EFI_STATUS                          Status;
> > > > > +
> > > > > +  if ((UINT32)Width > EfiPciIoWidthUint64) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  if (BarIndex >= PCI_MAX_BAR) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  if (Result == NULL) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO(This);
> > > > > +  Count = 1;
> > > > > +
> > > > > +  Status = GetBarResource (Dev, BarIndex, &Desc);  if (EFI_ERROR
> > > > > + (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (Offset + (Count << (Width & 0x3)) > Desc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > >    ASSERT (FALSE);
> > > > >    return EFI_UNSUPPORTED;
> > > > >  }
> > > > > @@ -396,6 +454,33 @@ PciIoIoRead (
> > > > >    IN OUT VOID                         *Buffer
> > > > >    )
> > > > >  {
> > > > > +  NON_DISCOVERABLE_PCI_DEVICE         *Dev;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *Desc;
> > > > > +  EFI_STATUS                          Status;
> > > > > +
> > > > > +  if ((UINT32)Width > EfiPciIoWidthUint64) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > >
> > > >
> > > > For PciIoIoRead(), I think enum values smaller than
> > > > EfiPciIoWidthMaximum
> > > are
> > > > all valid. The above check seems to strict.
> > >
> > > Will address this in v3.
> > > >
> > > >
> > > > > +
> > > > > +  if (BarIndex >= PCI_MAX_BAR) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  if (Buffer == NULL) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO(This);
> > > > > +
> > > > > +  Status = GetBarResource (Dev, BarIndex, &Desc);  if (EFI_ERROR
> > > > > + (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (Offset + (Count << (Width & 0x3)) > Desc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > >    ASSERT (FALSE);
> > > > >    return EFI_UNSUPPORTED;
> > > > >  }
> > > > > @@ -425,6 +510,33 @@ PciIoIoWrite (
> > > > >    IN OUT VOID                         *Buffer
> > > > >    )
> > > > >  {
> > > > > +  NON_DISCOVERABLE_PCI_DEVICE         *Dev;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *Desc;
> > > > > +  EFI_STATUS                          Status;
> > > > > +
> > > > > +  if ((UINT32)Width > EfiPciIoWidthUint64) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > >
> > > >
> > > > For PciIoIoWrite(), I think enum values smaller than
> > > > EfiPciIoWidthMaximum
> > > are
> > > > all valid. The above check seems to strict.
> > >
> > > Will address this in v3.
> > > >
> > > >
> > > > > +
> > > > > +  if (BarIndex >= PCI_MAX_BAR) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  if (Buffer == NULL) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO(This);
> > > > > +
> > > > > +  Status = GetBarResource (Dev, BarIndex, &Desc);  if (EFI_ERROR
> > > > > + (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (Offset + (Count << (Width & 0x3)) > Desc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > >    ASSERT (FALSE);
> > > > >    return EFI_UNSUPPORTED;
> > > > >  }
> > > > > @@ -556,6 +668,40 @@ PciIoCopyMem (
> > > > >    IN     UINTN                        Count
> > > > >    )
> > > > >  {
> > > > > +  NON_DISCOVERABLE_PCI_DEVICE         *Dev;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *DestDesc;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *SrcDesc;
> > > > > +  EFI_STATUS                          Status;
> > > > > +
> > > > > +  if ((UINT32)Width > EfiPciIoWidthUint64) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  if (DestBarIndex >= PCI_MAX_BAR ||
> > > > > +      SrcBarIndex >= PCI_MAX_BAR) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO(This);
> > > > > +
> > > > > +  Status = GetBarResource (Dev, DestBarIndex, &DestDesc);  if
> > > > > + (EFI_ERROR (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (DestOffset + (Count << (Width & 0x3)) > DestDesc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  Status = GetBarResource (Dev, SrcBarIndex, &SrcDesc);  if
> > > > > + (EFI_ERROR (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (SrcOffset + (Count << (Width & 0x3)) > SrcDesc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > >    ASSERT (FALSE);
> > > > >    return EFI_UNSUPPORTED;
> > > > >  }
> > > > > @@ -1414,6 +1560,40 @@ PciIoSetBarAttributes (
> > > > >    IN OUT UINT64                       *Length
> > > > >    )
> > > > >  {
> > > > > +  NON_DISCOVERABLE_PCI_DEVICE         *Dev;
> > > > > +  EFI_ACPI_ADDRESS_SPACE_DESCRIPTOR   *Desc;
> > > > > +  EFI_PCI_IO_PROTOCOL_WIDTH           Width;
> > > > > +  UINTN                               Count;
> > > > > +  EFI_STATUS                          Status;
> > > > > +
> > > > > +  #define DEV_SUPPORTED_ATTRIBUTES \
> > > > > +    (EFI_PCI_DEVICE_ENABLE |
> > > > > EFI_PCI_IO_ATTRIBUTE_DUAL_ADDRESS_CYCLE)
> > > > > +
> > > > > +  if ((Attributes & (~DEV_SUPPORTED_ATTRIBUTES)) != 0) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > >
> > > >
> > > > I think the above check for 'Attributes' can be dropped.
> > > > I found that the implementation of the PciIoGetBarAttributes()
> > > > function
> > > does not
> > > > expose any configurable attributes. So the logic can fall through to
> > > > the
> > > ASSERT
> > > > (for DEBUG images) and then returns EFI_UNSUPPORTED.
> > > >
> > > > Best Regards,
> > > > HaoWu
> > > >
> > > >
> > > > > +
> > > > > +  if (BarIndex >= PCI_MAX_BAR) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > > +  if (Offset == NULL || Length == NULL) {
> > > > > +    return EFI_INVALID_PARAMETER;  }
> > > > > +
> > > > > +  Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO(This);
> > > > > +  Width = EfiPciIoWidthUint8;
> > > > > +  Count = (UINT32) *Length;
> > > > > +
> > > > > +  Status = GetBarResource(Dev, BarIndex, &Desc);  if (EFI_ERROR
> > > > > + (Status)) {
> > > > > +    return Status;
> > > > > +  }
> > > > > +
> > > > > +  if (*Offset + (Count << (Width & 0x3)) > Desc->AddrLen) {
> > > > > +    return EFI_UNSUPPORTED;
> > > > > +  }
> > > > > +
> > > > >    ASSERT (FALSE);
> > > > >    return EFI_UNSUPPORTED;
> > > > >  }
> > > > > --
> > > > > 2.17.1
> > > > >
> > > > >
> > > > >
> 
> 
>