Re: [edk2-devel] [PATCH] UefiCpuPkg PiSmmCpuDxeSmm: Only support IN/OUT IO save state read (CVE-2018-12182)

["Wang, Jian J" <jian.j.wang@intel.com>]

Laszlo,

> -----Original Message-----
> From: devel@edk2.groups.io [mailto:devel@edk2.groups.io] On Behalf Of
> Laszlo Ersek
> Sent: Tuesday, May 14, 2019 12:46 AM
> To: devel@edk2.groups.io; Wang, Jian J <jian.j.wang@intel.com>
> Cc: Zeng, Star <star.zeng@intel.com>; Dong, Eric <eric.dong@intel.com>; Ni,
> Ray <ray.ni@intel.com>
> Subject: Re: [edk2-devel] [PATCH] UefiCpuPkg PiSmmCpuDxeSmm: Only support
> IN/OUT IO save state read (CVE-2018-12182)
> 
> On 05/10/19 07:16, Wang, Jian J wrote:
> > From: Star Zeng <star.zeng@intel.com>
> >
> > BZ: https://bugzilla.tianocore.org/show_bug.cgi?id=1136
> > CVE: CVE-2018-12182
> >
> > Customer met system hang-up during serial port loopback test in OS.
> > It is a corner case happened with one CPU core doing "out dx,al" and
> > another CPU core(s) doing "rep outs dx,byte ptr [rsi]".
> >
> > Detailed code flow is as below.
> >
> > 1. Serial port loopback test in OS.
> > One CPU core: "out dx,al" -> Writing B2h, SMI will happen.
> > Another CPU core(s): "rep outs dx,byte ptr [rsi]".
> >
> > 2. SMI happens to enter SMM.
> > "out dx" (SMM_IO_TYPE_OUT_DX) is saved as I/O instruction type in
> > SMRAM save state for CPU doing "out dx,al".
> > "rep outs dx" (SMM_IO_TYPE_REP_OUTS) is saved as I/O instruction
> > type and rsi is save as I/O Memory Address in SMRAM save state for
> > CPU doing "rep outs dx, byte ptr [rsi]".
> >
> > NOTE: I/O Memory Address (rsi) is a virtual address mapped by
> > OS/Virtual Machine.
> >
> > 3. Some SMM code calls EFI_SMM_CPU_PROTOCOL.ReadSaveState() with
> > EFI_SMM_SAVE_STATE_REGISTER_IO and parse data returned.
> >
> > For example:
> > https://github.com/tianocore/edk2/blob/master/QuarkSocPkg/
> >
> QuarkNorthCluster/Smm/DxeSmm/QncSmmDispatcher/QNC/QNCSmmSw.c#L7
> 6
> >
> > 4. SmmReadSaveState() is executed to read save state for
> > EFI_SMM_SAVE_STATE_REGISTER_IO.
> >
> > - The SmmReadSaveState() function in
> >   "UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.c" calls the
> >   SmmCpuFeaturesReadSaveStateRegister() function, from the platform's
> >   SmmCpuFeaturesLib instance.
> >
> > - If that platform-specific function returns EFI_UNSUPPORTED, then
> >   PiSmmCpuDxeSmm falls back to the common function
> >   ReadSaveStateRegister(), defined in file
> >   "UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c".
> >
> > Current ReadSaveStateRegister() in
> > UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c is trying to copy data
> > from I/O Memory Address for EFI_SMM_SAVE_STATE_IO_TYPE_REP_PREFIX,
> > PF will happen as SMM page table does not know and cover this
> > OS/Virtual Machine virtual address.
> >
> > Same case is for SmmCpuFeaturesReadSaveStateRegister() in platform-
> > specific SmmCpuFeaturesLib instance if it has similar implementation
> > to read save state for EFI_SMM_SAVE_STATE_REGISTER_IO with
> > EFI_SMM_SAVE_STATE_IO_TYPE_REP_PREFIX.
> >
> > Same case is for "ins", 'outs' and 'rep ins'.
> >
> > So to fix the problem, this patch updates the code to only support
> > IN/OUT, but not INS/OUTS/REP INS/REP OUTS for SmmReadSaveState().
> >
> > Cc: Eric Dong <eric.dong@intel.com>
> > Cc: Ray Ni <ray.ni@intel.com>
> > Cc: Laszlo Ersek <lersek@redhat.com>
> > Signed-off-by: Star Zeng <star.zeng@intel.com>
> > ---
> >  UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c | 17 +++++++++--------
> >  1 file changed, 9 insertions(+), 8 deletions(-)
> >
> > diff --git a/UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c
> b/UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c
> > index 26e365eabc..08cb9c05cf 100644
> > --- a/UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c
> > +++ b/UefiCpuPkg/PiSmmCpuDxeSmm/SmramSaveState.c
> > @@ -360,7 +360,6 @@ ReadSaveStateRegister (
> >    UINT32                      SmmRevId;
> >    SMRAM_SAVE_STATE_IOMISC     IoMisc;
> >    EFI_SMM_SAVE_STATE_IO_INFO  *IoInfo;
> > -  VOID                        *IoMemAddr;
> >
> >    //
> >    // Check for special EFI_SMM_SAVE_STATE_REGISTER_LMA
> > @@ -406,6 +405,14 @@ ReadSaveStateRegister (
> >        return EFI_NOT_FOUND;
> >      }
> >
> > +    //
> > +    // Only support IN/OUT, but not INS/OUTS/REP INS/REP OUTS.
> > +    //
> > +    if ((mSmmCpuIoType[IoMisc.Bits.Type] !=
> EFI_SMM_SAVE_STATE_IO_TYPE_INPUT) &&
> > +        (mSmmCpuIoType[IoMisc.Bits.Type] !=
> EFI_SMM_SAVE_STATE_IO_TYPE_OUTPUT)) {
> > +      return EFI_UNSUPPORTED;
> 
> I think this return value (EFI_UNSUPPORTED) should be replaced with
> EFI_NOT_FOUND, here.
> 
> The return value from this function will be propagated to the caller,
> through SmmReadSaveState()
> [UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.c].
> 
> The latter function implements EFI_MM_CPU_PROTOCOL.ReadSaveState(), and
> the PI-1.7 spec writes,
> 
>     If the CPU does not support the specified register Register, then
>     EFI_NOT_FOUND should be returned. If the CPU does not support the
>     specified register width Width , then EFI_INVALID_PARAMETER is
>     returned.
> 
> I don't feel too strongly about this, but I think it's worth
> considering. If others think EFI_UNSUPPORTED is better, I'm OK with
> that, in the end.
> 
> Either way,
> 
> Reviewed-by: Laszlo Ersek <lersek@redhat.com>
> 

I agree. EFI_NOT_FOUND is better than EFI_UNSUPPORTED. At least returning
EFI_UNSUPPORTED will somewhat break spec. Thanks for catching this.

Regards,
Jian

> Thanks
> Laszlo
> 
> 
> 
> > +    }
> > +
> >      //
> >      // Compute index for the I/O Length and I/O Type lookup tables
> >      //
> > @@ -425,13 +432,7 @@ ReadSaveStateRegister (
> >      IoInfo->IoPort = (UINT16)IoMisc.Bits.Port;
> >      IoInfo->IoWidth = mSmmCpuIoWidth[IoMisc.Bits.Length].IoWidth;
> >      IoInfo->IoType = mSmmCpuIoType[IoMisc.Bits.Type];
> > -    if (IoInfo->IoType == EFI_SMM_SAVE_STATE_IO_TYPE_INPUT || IoInfo-
> >IoType == EFI_SMM_SAVE_STATE_IO_TYPE_OUTPUT) {
> > -      ReadSaveStateRegister (CpuIndex, EFI_SMM_SAVE_STATE_REGISTER_RAX,
> mSmmCpuIoWidth[IoMisc.Bits.Length].Width, &IoInfo->IoData);
> > -    }
> > -    else {
> > -      ReadSaveStateRegisterByIndex(CpuIndex,
> SMM_SAVE_STATE_REGISTER_IOMEMADDR_INDEX, sizeof(IoMemAddr),
> &IoMemAddr);
> > -      CopyMem(&IoInfo->IoData, IoMemAddr,
> mSmmCpuIoWidth[IoMisc.Bits.Length].Width);
> > -    }
> > +    ReadSaveStateRegister (CpuIndex, EFI_SMM_SAVE_STATE_REGISTER_RAX,
> mSmmCpuIoWidth[IoMisc.Bits.Length].Width, &IoInfo->IoData);
> >      return EFI_SUCCESS;
> >    }
> >
> >
> 
> 
>