Thanks Andrew, I agree. The DMA mapping calls are required to maintain
portability since there are architectures that have different requirements
around both cache coherency and DMA address mapping. It would be useful to get
this as an errata item to EFI 1.10 as well as all later revisions. :)
Feng, can you share some details on the timing of your plan to include
Map/AllocateBuffer calls for XhciDxe?
Thanks,
Eugene
From: Andrew Fish [mailto:af...@apple.com]
Sent: Wednesday, July 24, 2013 7:04 PM
To: Tian, Feng
Cc: Cohen, Eugene; edk2-devel@lists.sourceforge.net
Subject: Re: [edk2] XHCI DMA Buffer Issues
On Jul 24, 2013, at 5:59 PM, "Tian, Feng"
<feng.t...@intel.com<mailto:feng.t...@intel.com>> wrote:
Sorry, Andrew.
I don't understand your point, where of XhciDxe current implementation is
disobeying UEFI spec? Do I miss something?
2.6.3 Driver-Specific Elements:
12. If a driver is written for a PCI root bridge, then the
EFI_PCI_ROOT_BRIDGE_IO_PROTOCOL and the EFI_PCI_IO_PROTOCOL must be implemented.
This sentence is from UEFI spec. My understanding is it defines the behavior of
PCI_IO & PCI_ROOT_BRIDGE's producer rather than the user.
I think there may be a missing requirement in the UEFI section 2.6.3 that if a
driver uses PCI it must use EFI_PCI_IO_PROTOCOL for IO. I was surprised there
was not a stronger statement about this that I could find. I'll take this issue
up with the UEFI Spec committee.
I agree we should use PciIo->map schematic, it's in our plan.
OK Thanks.
Andrew Fish
Thanks
Feng
From: Andrew Fish [mailto:af...@apple.com<http://apple.com>]
Sent: Thursday, July 25, 2013 8:41 AM
To: Tian, Feng
Cc: Cohen, Eugene;
edk2-devel@lists.sourceforge.net<mailto:edk2-devel@lists.sourceforge.net>
Subject: Re: [edk2] XHCI DMA Buffer Issues
On Jul 24, 2013, at 5:27 PM, "Tian, Feng"
<feng.t...@intel.com<mailto:feng.t...@intel.com>> wrote:
Hi, Eugene
The EDKII XhciDxe driver doesn't use PciIo.AllocateBuffer()/Map() to allocate
DMA buffers. It's mainly because EDKII is 1:1 mapping on Pci physical address
and Cpu logical address for IA architecture. AllocateBuffer() is enough for
current case.
So as long as code runs an edk2 Intel/AMD CPU system it does not need to follow
the UEFI spec? So if you had a uninitialized variable or a buffer that was just
a random memory address you would not fix it if you were lucky and it worked on
an edk2 Intel/AMD test system?
We only implement this feature at EHCI driver as Andrew from Apple was trying
to use this driver at an ARM based board "Beagle Board".
For which buffers to be mapped, I think all MMIO buffers specified in XHCI spec
are required. For how to do this, please refer to EhciDxe driver.
We ever constructed an non 1:1 mapping env at DUET, we mapped 0~1M to
FFF00000~FFFFFFFF, other memory regions are shifted downwards with 1M offset.
You can try this way.
This is a violation of the UEFI 2.4 (all the way back to EFI 1.10)
specification. Seems like we need to add an SCT for this.... Since folks seem
to confuse passing the SCTs with following the spec.
At a minimum you should update the comments in the driver to tell people that
this is not a UEFI 2.4 (or even EFI 1.10) compatible driver.
2.6.3 Driver-Specific Elements:
12. If a driver is written for a PCI root bridge, then the
EFI_PCI_ROOT_BRIDGE_IO_PROTOCOL and the EFI_PCI_IO_PROTOCOL must be implemented.
The driver needs to use the DMA scheme defined in the EFI_PCI_IO_PROTOCOL, it
is not legal for a driver to makeup its own way to do DMA.
Thanks,
Andrew Fish
Thanks
Feng
From: Cohen, Eugene [mailto:eug...@hp.com<http://hp.com>]
Sent: Thursday, July 25, 2013 4:29 AM
To: edk2-devel@lists.sourceforge.net<mailto:edk2-devel@lists.sourceforge.net>
Subject: [edk2] XHCI DMA Buffer Issues
Dear XhciDxe Maintainer,
I'm currently reviewing the XhciDxe driver and I'm trying to figure out how DMA
buffers are allocated. I see a number of pool and page allocations but I do
not see any called to PCI_IO Map()/Unmap() or to AllocateBuffer()/FreeBuffer().
This appears to be violating the rules for PCI DMA buffers since they are not
being mapped (and if common buffers are desired then they are also not being
allocated with AllocateBuffer as required).
Can someone more familiar with XHCI help me determine which buffers need to be
mapped and how (BusMasterRead, BusMasterWrite, CommonBuffer)?
It would be useful if we had a test environment that could catch driver DMA
buffer mapping issues earlier. I think one way to do this would be set up the
MMU in a non-identity mode so that anyone trying to use a processor virtual
address for DMA would see a failure.
Thanks,
Eugene
------------------------------------------------------------------------------
See everything from the browser to the database with AppDynamics
Get end-to-end visibility with application monitoring from AppDynamics
Isolate bottlenecks and diagnose root cause in seconds.
Start your free trial of AppDynamics Pro today!
http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk_______________________________________________
edk2-devel mailing list
edk2-devel@lists.sourceforge.net<mailto:edk2-devel@lists.sourceforge.net>
https://lists.sourceforge.net/lists/listinfo/edk2-devel
------------------------------------------------------------------------------
See everything from the browser to the database with AppDynamics
Get end-to-end visibility with application monitoring from AppDynamics
Isolate bottlenecks and diagnose root cause in seconds.
Start your free trial of AppDynamics Pro today!
http://pubads.g.doubleclick.net/gampad/clk?id=48808831&iu=/4140/ostg.clktrk
_______________________________________________
edk2-devel mailing list
edk2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/edk2-devel
