Re: [PATCH 0/2] Implement NULL pointer detection feature

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

Updated version according to some internal discussions:

a) Use one PCD with bitmask to enable/disable this feature for different boot phases. This PCD is a "fixed" PCD with type of UINT8, and all bits are cleared by default. Please note that setting BIT7 with BIT0 set is used to "DISABLE" NULL pointer detection for code after EndOfDxe as a workaround for non-fixable NULL access in OpROM or boot loaders.

     gEfiMdeModulePkgTokenSpaceGuid.PcdNullPointerDetectionPropertyMask
     #  BIT0 - Enable NULL pointer detection for UEFI.<BR>
     #  BIT1 - Enable NULL pointer detection for SMM.<BR>
     #  BIT2-6 - Reserved for PEI and other possible uses.<BR>
     #  BIT7 - Disable NULL pointer detection after EndOfDxe.<BR>

b) Add PEI support. Since PEI requirement has been covered by following new bug tracker, it won't be included by current patch update but a separate one.

    https://bugzilla.tianocore.org/show_bug.cgi?id=687

c) CSM code lines which access page 0 will be enclosed by <enable_page_0> and then <disable_page_0> code. This is subject to change if critical performance issue is found.


-----Original Message-----
From: edk2-devel [mailto:edk2-devel-bounces@lists.01.org] On Behalf Of Wang, Jian J
Sent: Friday, September 01, 2017 10:27 AM
To: Yao, Jiewen <jiewen.yao@intel.com>; Johnson, Brian (EXL - Eagan) <brian.johnson@hpe.com>; afish@apple.com
Cc: edk2-devel@lists.01.org
Subject: Re: [edk2] [PATCH 0/2] Implement NULL pointer detection feature

Let me summarize the changes to this feature. If no new comments, I'll start to make changes to current implementation.

a) Use one PCD with bit set/clear to enable/disable this feature for different boot phases. To make it easier to use, I'd suggest to use the bit sequence to follow the boot phase sequence (bit0->bit1->bit2......bit7 ===> PEI->DXE->PostDxe......SMM), just like below. SMM is a little bit special so it consumes the MSB.

     gEfiMdeModulePkgTokenSpaceGuid.PcdNullPointerDetectionPropertyMask
     #  BIT0 - Enable NULL pointer detection for PEI.<BR>
     #  BIT1 - Enable NULL pointer detection for DXE.<BR>
     #  BIT2 - Enable NULL pointer detection after EndOfDxe.<BR>
     #  BIT7 - Disable NULL pointer detection for SMM.<BR>

    This PCD is a "fixed" PCD and all bits are cleared by default. Since this is a feature for debug purpose, I see no reason to make it dynamic. 

b) Add PEI and post DXE phase support. Since PEI phase requirement has been covered by following new bug tracker, it won't be included by current patch update but a separate one. Changes for Post DXE phase will be included.

    https://bugzilla.tianocore.org/show_bug.cgi?id=687

Thanks,
Wang, Jian J

-----Original Message-----
From: edk2-devel [mailto:edk2-devel-bounces@lists.01.org] On Behalf Of Yao, Jiewen
Sent: Thursday, August 31, 2017 9:16 AM
To: Johnson, Brian (EXL - Eagan) <brian.johnson@hpe.com>; afish@apple.com
Cc: edk2-devel@lists.01.org
Subject: Re: [edk2] [PATCH 0/2] Implement NULL pointer detection feature

Thanks.
I have filed https://bugzilla.tianocore.org/show_bug.cgi?id=687 for PEI phase, with suggestion from Brian Johnson and Andrew Fish.

If you share some of your PEI code to elimination the duplicated effort, that would be great.

Yes, we can calculate the perf data in CSM to see what is the best way to enable this feature. Good suggestion.

Thank you
Yao Jiewen

From: edk2-devel [mailto:edk2-devel-bounces@lists.01.org] On Behalf Of Johnson, Brian (EXL - Eagan)
Sent: Thursday, August 31, 2017 5:36 AM
To: afish@apple.com; Yao, Jiewen <jiewen.yao@intel.com>
Cc: edk2-devel@lists.01.org
Subject: Re: [edk2] [PATCH 0/2] Implement NULL pointer detection feature

Jiewen,

Certainly PEI could be done later.  There's no need to get all the code in at once, and you have the DXE/SMM part pretty much ready to go.

I'm willing to go with a single PCD with a bitmap, or multiple PCDs.  Whatever seems best.

IIRC protecting page 0 in PEI involves modifying the main GCD descriptor to be a grows-downward variety, with a limit above page 0.  That way you get an exception if page 0 is accessed.  I'd have to check on the steps needed to release the actual code, which may take quite a while.  You may be better off just doing it yourself.  I'll take another look at the code to make sure I'm explaining this properly, though.

CSM is one of the few pieces of BIOS which has a legitimate need to access page 0.  We get the best protection by having it enable, access, and then disable page 0.  But if that affects performance too badly, we may need to have it leave page 0 enabled.  I don't enable CSM on the platforms I work on, so it's not something I have much to say about.  Those who use CSM would want to weigh in.

Brian Johnson

-----Original Message-----
From: afish@apple.com<mailto:afish@apple.com> [mailto:afish@apple.com]
Sent: Wednesday, August 30, 2017 11:27 AM
To: Yao, Jiewen <jiewen.yao@intel.com<mailto:jiewen.yao@intel.com>>
Cc: Johnson, Brian (EXL - Eagan) <brian.johnson@hpe.com<mailto:brian.johnson@hpe.com>>; Wang, Jian J <jian.j.wang@intel.com<mailto:jian.j.wang@intel.com>>; edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org>
Subject: Re: [edk2] [PATCH 0/2] Implement NULL pointer detection feature


> On Aug 30, 2017, at 7:55 AM, Yao, Jiewen <jiewen.yao@intel.com<mailto:jiewen.yao@intel.com>> wrote:
>
> Hi Brian
> Good feedback.
> Comment in line.
>
> From: Johnson, Brian (EXL - Eagan) [mailto:brian.johnson@hpe.com]
> Sent: Tuesday, August 29, 2017 11:17 PM
> To: Yao, Jiewen <jiewen.yao@intel.com<mailto:jiewen.yao@intel.com>>; Wang, Jian J <jian.j.wang@intel.com<mailto:jian.j.wang@intel.com>>; edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org>
> Subject: RE: [edk2] [PATCH 0/2] Implement NULL pointer detection feature
>
> Thank you for implementing this feature!  It is very  helpful for catching pointer-related problems.  We have used a similar scheme on our systems for years, and caught several important bugs.  Some comments:
>
> * It's possible to implement similar protections in PEI (IA32) by modifying the GDT descriptors.
> [Jiewen] Enabling PEI is interesting.
> Since we are working on DXE and SMM as our first priority, can we treat the PEI enabling as a separate task?
> If you can help to create a patch for PEI and submit, that will be even better.
>

Jiewen,

This is slightly off topic but I've noticed in early PEI the issue on some x86 chipsets is not that a NULL pointer reference is bad, but any reference to an address space that is not decoded will triple bus fault the processor and that is really hard to debug. If the GDT trick could be used to cause an exception that drops into the debugger vs. a triple bus fault that would be a real win for debugging. While the memory ranges would be platform specific it would be awesome to have some plumbing in the edk2 to make this easier to implement.

Thanks,

Andrew Fish

>
>
> * For flexibility, I'd like NULL pointer protection to be controlled independently in PEI, DXE, and SMM, using separate PCDs.
> [Jiewen] I think we can existing example on using one PCD to control all phase or using one PCD to control different phase.
> For example, we use below PCD to control all phases, PEI/DXE/SMM
>  ## The mask is used to control PerformanceLib behavior.<BR><BR>
>  #  BIT0 - Enable Performance Measurement.<BR>
>  # @Prompt Performance Measurement Property.
>  # @Expression  0x80000002 | (gEfiMdePkgTokenSpaceGuid.PcdPerformanceLibraryPropertyMask & 0xFE) == 0
>  gEfiMdePkgTokenSpaceGuid.PcdPerformanceLibraryPropertyMask|0|UINT8|0x00000009
>
> And we use below PCD to control different phase, DXE/SMM
>  ## The mask is used to control memory profile behavior.<BR><BR>
>  #  BIT0 - Enable UEFI memory profile.<BR>
>  #  BIT1 - Enable SMRAM profile.<BR>
>  #  BIT7 - Disable recording at the start.<BR>
>  # @Prompt Memory Profile Property.
>  # @Expression  0x80000002 | (gEfiMdeModulePkgTokenSpaceGuid.PcdMemoryProfilePropertyMask & 0x7C) == 0
>  gEfiMdeModulePkgTokenSpaceGuid.PcdMemoryProfilePropertyMask|0x0|UINT8|0x30001041
>
> In order to make the code consistent with the existing PCD, I am thinking to use one PCD named PcdNullPointerDetectionPropertyMask
>  #  BIT0 - Enable NULL pointer detection for UEFI.<BR>
>  #  BIT1 - Enable NULL pointer detection for SMM.<BR>
>  #  BIT2 - Enable NULL pointer detection for PEI.<BR>
>  #  BIT7 - Disable NULL pointer detection after EndOfDxe.<BR>
>
> I am not so worried about pre-memory initialization PEI phase, because page 0 is always invalid.
> For post-memory initialization PEI phase, we can modify the page table in memory, and dynamic PCD can be used at that time.
>
>
> * We have seen various option ROMs and OS boot loaders which have NULL pointer issues, but are outside of our control.  It is useful to enable NULL pointer protection during as much of the boot as possible, but disable it before running these other executables.  So I'd suggest adding another PCD, perhaps PcdNullPointerDetectionPostDxe, to control NULL pointer protection late in boot.  If PcdNullPointerDetection != PcdNullPointerDetectionPostDxe, install an end-of-DXE event (gEfiEndOfDxeEventGroupGuid) which changes the protection of page 0 using a call to EFI_CPU_ARCH_PROTOCOL. SetMemoryAttributes(CpuArch, 0, EFI_PAGE_SIZE, 0).
> [Jiewen] Good point for the compatibility consideration.
> Another thing is that CSM module may need access address 0. We have 2 choices:
> 1) CSM32 can call CpuArch->SetMemoryAttribute() to enable page 0, access page 0, then disable page 0.
> 2) CSM32 can call CpuArch->SetMemoryAttribute() to enable page 0 at the beginning, and always keep it enabled.
> Any though on that?
>
>
> So ideally I'd like to have 4 PCDs:
>
> PcdNullPointerDetectionPei
> PcdNullPointerDetectionDxe
> PcdNullPointerDetectionSmm
> PcdNullPointerDetectionPostDxe
>
> Thanks,
> Brian Johnson
> HPE
>
> -----Original Message-----
> From: edk2-devel [mailto:edk2-devel-bounces@lists.01.org] On Behalf Of Yao, Jiewen
> Sent: Sunday, August 27, 2017 10:39 PM
> To: Wang, Jian J <jian.j.wang@intel.com<mailto:jian.j.wang@intel.com<mailto:jian.j.wang@intel.com%3cmailto:jian.j.wang@intel.com>>>; edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org>>
> Subject: Re: [edk2] [PATCH 0/2] Implement NULL pointer detection feature
>
> Comment in line.
>
> From: Wang, Jian J
> Sent: Monday, August 28, 2017 11:24 AM
> To: Yao, Jiewen <jiewen.yao@intel.com<mailto:jiewen.yao@intel.com<mailto:jiewen.yao@intel.com%3cmailto:jiewen.yao@intel.com>>>; edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org>>
> Subject: RE: [edk2] [PATCH 0/2] Implement NULL pointer detection feature
>
>
> 1)      I think this feature should be 'FALSE' by default. I forgot to reset its default value. This feature makes use of page mechanism to detect NULL pointer. So any ARCH doesn't support paging in EDK-II can't use it. Currently validated platform includes VLV2 and Denlow. Let me know if all platform must be validated or not.
>
> [Jiewen] Yes, I am OK to set to be FALSE to provide best compatibility.
> I suggest we validate all open source IA platforms, such as Quark, and OVMF with TRUE.
>
>
> 2)      It's hard to make it a dynamic feature because we need to setup page table for physical address 0-4095 in advance. If there's no memory alloc/free action after enabling this feature, there's no chance to make those change in page table. Then the usage of feature will be limited in such case.
>
> [Jiewen] Dynamic means the initial value is dynamic. I am not saying we need modify the page table.
>
> An implementation could be:
> A platform can set this PCD in PEI phase based upon a setup variable to choose CSM enable or disable.
>
> The IPL sets page table based upon this PCD value. The DXE Core cannot consume PCD directly, because it might be dynamic. But we can pass the information from IPL via HOB. All the DXE module just checks the value based upon HOB.
>
>
>
>
> From: Yao, Jiewen
> Sent: Monday, August 28, 2017 11:10 AM
> To: Wang, Jian J <jian.j.wang@intel.com<mailto:jian.j.wang@intel.com<mailto:jian.j.wang@intel.com%3cmailto:jian.j.wang@intel.com<mailto:jian.j.wang@intel.com%3cmailto:jian.j.wang@intel.com%3cmailto:jian.j.wang@intel.com%3cmailto:jian.j.wang@intel.com>>>>; edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org>>>
> Subject: RE: [edk2] [PATCH 0/2] Implement NULL pointer detection feature
>
> Thank you  to enable this feature.
>
> I have 2 comments, after a very quick review.
>
>
> 1)      I notice it is enabled by default "gEfiMdeModulePkgTokenSpaceGuid.PcdNullPointerDetection|TRUE".
>
> Would you please provide the information on how many open source platforms are validated?
> Such as, IA platform (VLV2, Quark), emu platform (OVMF, NT32)?
> Or do we need validate any ARM platform?
>
>
>
> 2)      I am thinking about CSM platform. Do you think we can make it dynamic, as such, a platform may set the validate based upon CSM enable/disable?
>
>
> Or if we need update the CSM module to patch the page table automatically? Once this is feature is ON.
>
>
> Thank you
> Yao Jiewen
>
>
>> -----Original Message-----
>> From: edk2-devel [mailto:edk2-devel-bounces@lists.01.org] On Behalf Of Wang,
>> Jian J
>> Sent: Monday, August 28, 2017 10:51 AM
>> To: edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org>>>
>> Subject: [edk2] [PATCH 0/2] Implement NULL pointer detection feature
>>
>> This patch is the implementation of NULL pointer detection feature,
>> which is one of the small features of Special Pool.
>>
>> Wang, Jian J (2):
>>  Implement NULL pointer detection for EDK-II Core
>>  Implement NULL pointer detection for EDK-II SMM Core and driver
>>
>> MdeModulePkg/Core/Dxe/DxeMain.inf                |  3 ++-
>> MdeModulePkg/Core/Dxe/Mem/Page.c                 |  5 +++--
>> MdeModulePkg/Core/DxeIplPeim/DxeIpl.inf          |  1 +
>> MdeModulePkg/Core/DxeIplPeim/Ia32/DxeLoadFunc.c  |  6 ++++--
>> MdeModulePkg/Core/DxeIplPeim/X64/VirtualMemory.c | 26
>> ++++++++++++++++--------
>> MdeModulePkg/MdeModulePkg.dec                    |  7 +++++++
>> UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/PageTbl.c         | 12 +++++++++++
>> UefiCpuPkg/PiSmmCpuDxeSmm/MpService.c            | 25
>> ++++++++++++++++++++++-
>> UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.inf     |  1 +
>> UefiCpuPkg/PiSmmCpuDxeSmm/X64/PageTbl.c          | 12 +++++++++++
>> 10 files changed, 84 insertions(+), 14 deletions(-)
>>
>> --
>> 2.11.0.windows.1
>>
>> _______________________________________________
>> edk2-devel mailing list
>> edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org>>>
>> https://lists.01.org/mailman/listinfo/edk2-devel
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org%3cmailto:edk2-devel@lists.01.org>>
> https://lists.01.org/mailman/listinfo/edk2-devel
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org>
> https://lists.01.org/mailman/listinfo/edk2-devel

_______________________________________________
edk2-devel mailing list
edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org>
https://lists.01.org/mailman/listinfo/edk2-devel
_______________________________________________
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel
_______________________________________________
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel