Re: smm lock query

["Zeng, Star" <star.zeng@intel.com>]

I do not see issue according to the spec.
Platform should know when to signal DxeSmmReadyToLock (after EndOfDxe).
DxeSmmReadyToLock event is to notify DXE handlers.
Modules are responsible to lock or protect their resource and effect the appropriate protections in their notification handlers.
SmmIplGuidedEventNotify is used to inform SMM environment to signal SmmReadyToLock.
SmmReadyToLock event is to notify SMM handlers.

DXE handlers could not touch SMRAM (after SMRAM is locked or even after SMRR is configured in PiSmmCpuDxeSmm if I know it is correct).

“This protocol in tandem with the End of DXE Event facilitates transition of the platform from the environment where all of the components are under the authority of the platform manufacturer to the environment where third party extensible modules such as UEFI drivers and UEFI applications are executed.

The protocol is published immediately after signaling of the End of DXE Event.
PI modules that need to lock or protect their resources in anticipation of the invocation of 3rd party extensible modules should register for notification on installation of this protocol and effect the appropriate protections in their notification handlers. For example, PI platform code may choose to use notification handler to lock MM by invoking EFI_MM_ACCESS_PROTOCOL.Lock() function.”


SMM environment is a *runtime* environment. SMRAM will be even changed after SmmReadyToLock, for example, by SMM handler for SMM communication from DXE.


Thanks,
Star
From: Abhishek Singh [mailto:abh@cs.unc.edu]
Sent: Tuesday, May 29, 2018 2:02 AM
To: Marvin Häuser <Marvin.Haeuser@outlook.com>
Cc: edk2-devel@lists.01.org; Laszlo Ersek <lersek@redhat.com>; afish@apple.com; Ni, Ruiyu <ruiyu.ni@intel.com>; Dong, Eric <eric.dong@intel.com>; Zeng, Star <star.zeng@intel.com>
Subject: Re: [edk2] smm lock query

Thank you everyone for your inputs and clarifications. They are helping me to better understand the uefi code, to which I am very new. I do not mean to hijack the thread: so please continue your discussions about whether the implementation matches the spec.

However, I want to state why I am interested in the IPL code. For my research, I wish to dump the contents of SMRAM when it has reached steady state, i.e., all the drivers have made changes to smram and it has been locked. The current implementation (smm ipl) locks smram when it receives the SmmReadyToLock event and then propagates the event to the smm drivers that make further changes to smram. Unfortunately, I cannot take a snapshot of smram after it has been locked! Thus, I have solved this issue by propagating the event to the smm drivers first (using SmmIplGuidedEventNotify), then opening access to and dumping the contents of SMRAM, and finally closing access to and locking smram. Would it be fair to say that this would give me the fully initialized contents of smram?



My second observation is that despite opening access to smram, I am unable to access its contents, which is a violation of the EFI_MM_ACCESS_PROTOCOL.Open() description in the spec, which says: "This function “opens” MMRAM so that it is visible while not inside of MM." Note that I am working with minnowboard firmware release 0.97. So some of the binaries like SmmAccess.efi are provided by Intel, while others have been built from the edk source tree: thus, this may not be an EDK issue. Please suggest further steps and/or workarounds. Should I contact edk2-platforms maintainers, or start a new thread here for this issue?

-Abhishek


On Mon, May 28, 2018 at 10:03 AM, Marvin Häuser <Marvin.Haeuser@outlook.com<mailto:Marvin.Haeuser@outlook.com>> wrote:
Hello Andrew and Laszlo,

Thanks for your comments!
Of course I'm with you that it is the platform that signals the SmmReadyToLock event and therefor is aware.
However, they might rely on the protocol's description that the resources are about(!) to be locked and code accordingly, not considering the event characteristic of the handler in PiSmmIpl.
The code might be written by different people, not especially reviewed against edk2's actions, or additional code might be supplied by third parties that do not have tree code access (which, by integration, would be "platform binaries" by the definition applying here).

Therefor I would still ask everyone to consider figuring out a solution to this discrepancy from the specification, such as the internal "dummy event" I proposed.

Thanks,
Marvin.

> -----Original Message-----
> From: Laszlo Ersek <lersek@redhat.com<mailto:lersek@redhat.com>>
> Sent: Monday, May 28, 2018 12:58 PM
> To: Andrew Fish <afish@apple.com<mailto:afish@apple.com>>; Marvin H?user
> <Marvin.Haeuser@outlook.com<mailto:Marvin.Haeuser@outlook.com>>
> Cc: edk2-devel@lists.01.org<mailto:edk2-devel@lists.01.org>; Abhishek Singh <abh@cs.unc.edu<mailto:abh@cs.unc.edu>>;
> ruiyu.ni@intel.com<mailto:ruiyu.ni@intel.com>; eric.dong@intel.com<mailto:eric.dong@intel.com>; star.zeng@intel.com<mailto:star.zeng@intel.com>
> Subject: Re: [edk2] smm lock query
>
> On 05/27/18 22:44, Andrew Fish wrote:
> >
> >
> >> On May 27, 2018, at 9:47 AM, Marvin H?user
> <Marvin.Haeuser@outlook.com<mailto:Marvin.Haeuser@outlook.com>> wrote:
> >>
> >> Good day Abhishek,
> >>
> >> I CC'd the MdeModulePkg maintainers, Ruiyu for the Platform BDS aspect
> (exposes the ReadyToLock protocol) and Laszlo for his high-quality answers.
> >>
> >> Strictly speaking you are, right, because of the description for the MM
> protocol:
> >> "Indicates that MM resources and services that should not be used by the
> third party code are about[Marvin: (!)] to be locked."
> >> Practically however, I don't see any issue with the current
> implementation. Code inside MMRAM is not affected directly by the lock, it is
> just notified.
> >> However, either the code or the specification should be slightly updated
> to be in sync. A code update might require review of the caller assumptions,
> just to be sure.
> >>
> >> I have a different concern though and hope I'm actually overlooking
> something.
> >> If I understand the code correctly, it is the Platform BDS that exposes the
> (S)MmReadyToLock protocol. PiSmmIpl seems to consume that event and
> lock SMM resources based on the event.
> >> Because of latter being an event however, I don't think it is, or can be,
> guaranteed to be the last event group member executing.
> >> When it is not the last, the "about to be locked" part is not true for any
> subsequent callbacks, that could actually be a risky break of the specification
> - if it is.
> >> If it is a break of the specification, I can only think of letting Platform BDS
> expose an "internal" event group, which is only caught by PiSmmIpl, which
> then drives the actual SmmReadyToLock flow.
> >> This would require updates to all platform trees and hence I would
> propose a temporary backwards-compatibility.
> >>
> >> Any comments? Did I overlook something (I hope)?
> >>
> >
> > Mavvin,
> >
> > You are correct there is no guarantee of order in events. Thanks for cc'ing
> the right folks, as I don't remember all the low level details...
> >
> > In general the idea behind the MM code is it only comes from the platform,
> then by definition that code should be aware when the platform was going
> to lock MM. In a practical sense any MM module that had a depex evaluate
> to true would have dispatched in DXE prior to BDS being launched. In general
> BDS is the code that enumerates PCI and connects devices, thus there is no
> chance for 3rd party software to run before that point in the boot. So in an
> abstract sense that lock represents the end of DXE dispatch.
>
> This is my understanding as well. gEfiDxeSmmReadyToLockProtocolGuid is
> installed by Plaform BDS before any non-platform binaries get a chance to
> run. In terms of the current PlatformBootManagerLib interfaces, that means
> the protocol should be installed from
> PlatformBootManagerBeforeConsole() (as noted on the API declaration
> itself).
>
> Thanks
> Laszlo