Re: Cannot create mini-os guest for arm64

[Chen Baozi <baozich@gmail.com>]

On Mon, May 20, 2013 at 09:19:21AM +0800, Chen Baozi wrote:
> On Fri, May 17, 2013 at 03:30:24PM +0100, Ian Campbell wrote:
> > On Fri, 2013-05-17 at 17:58 +0800, Chen Baozi wrote:
> > > On Fri, May 17, 2013 at 10:26:08AM +0100, Ian Campbell wrote:
> > > > On Fri, 2013-05-17 at 08:00 +0800, Chen Baozi wrote:
> > > > > On May 17, 2013, at 12:20 AM, Ian Campbell <Ian.Campbell@citrix.com> wrote:
> > > > > 
> > > > > > On Thu, 2013-05-16 at 23:34 +0800, Chen Baozi wrote:
> > > > > >> On May 16, 2013, at 10:41 PM, Ian Campbell <ian.campbell@citrix.com> wrote:
> > > > > >> 
> > > > > >>> On Thu, 2013-05-16 at 17:49 +0800, Chen Baozi wrote:
> > > > > >>>> Hi Ian,
> > > > > >>>> 
> > > > > >>>> I've done the basic interfaces/framework to pass mini-os build on arm64,
> > > > > >>>> which still needs lots of implementation to work properly.
> > > > > >>>> 
> > > > > >>>> Then I tried to create the mini-os guest for arm64 on Fast Model environment
> > > > > >>>> to drive the following porting. (I know it must not work at present, but it
> > > > > >>>> would give some feedback through the test.) However, when I tried to "xl
> > > > > >>>> create" the mini-os, it reported the following errors:
> > > > > >>> 
> > > > > >>> There are some issues with loading guests from ELF (rather than zImage)
> > > > > >>> kernels on ARM.
> > > > > >> Is it possible because of there is no arm64 "guest_type" of
> > > > > >> xc_dom_arch struct? I can only find "den-3.0-armv7l" guest_type
> > > > > >> related codes in libxc files.
> > > > > > 
> > > > > > Oh, yes, we haven't done arm64 guests yet... Sorry I forgot about this
> > > > > > before. Hrm, that does rather block you from doing arm64 mini-os.
> > > > > > 
> > > > > > I posted a series to allow 64-bit dom0 a while back but we are currently
> > > > > > frozen so it has been committed, but that's the obvious starting point
> > > > > > for 64-bit guest support.
> > > > > > 
> > > > > > I don't recall why you are looking at 64-bit minios rather than 32-bit,
> > > > > > is that an option? A lot of the code should end up common.
> > > > > I did plan to start the work on arm32 at the very beginning. But The
> > > > > only platform of cortex-a15 I've got is the samsung's chromebook. I
> > > > > tried hard to run Xen on it but didn't succeed in the end due to the
> > > > > lack of debug method. At the same time, I've got an chance getting
> > > > > involved in a arm64 project of our lab so it is easy to have a
> > > > > FastModel license, which at least is a debugable platform for my at
> > > > > present. 
> > > > 
> > > > You should be able to run arm32 guests on that platform.
> > > Ok, I'll try it next.
> > > 
> > > > 
> > > > > PS, I ordered an OMAP5 development board, but it is stilling on shipping. 
> > > > 
> > > > Which devboard did you go for?
> > > It is OMAP5432 uEVM from SVTronics.
> > > 
> > > BTW, I found I can only run Xen hypervisor of your arm-cross2 git branch on
> > > Fast Model. For the version built from xen mainline, the program would
> > > run to an unknown address after paging enabled. (more exactly, from the
> > > 'isb' instruction which in line 267 of arch/arm/arm64/head.S.)
> > > 
> > > Any ideas?
> > 
> > That ISB is effectively the point at which paging gets enabled.
> > 
> > I'm afraid I've not seen anything like that for a very long time, commit
> > 56bbdea (which is pretty recent) works OK for me on both 32 and 64 bit
> > models.
> > 
> > What version of the model do you use? How are you invoking it?
> 
> I'm using Fast Model (Model Debugger - 8.0.027, Model Shell 8.0.16). To
> invoking it, I use the following command:
> 
> modeldebugger/model_shell64 $ATLAS_MODEL /path/to/xen-syms -C
> motherboard.flashloader0.fname=~/path/to/zImage -C
> motherboard.mmc.p_mmc_file=/path/to/rootfs

Sorry, I think the problem described above only happened if I use "xen-syms"
for debugging. If I switch to "xen", there is no such problem on paging though
it still not work (phenomenons: couldn't see any output).

So the situation is that xen-syms doesn't work for either arm-cross2 branch
or mainline, (I used it by mistake),  while xen doesn't work in mainline but 
arm-cross2 branch.

> 
> $ATLAS_MODEL=~/ARM/FastModelsPortfolio_8.0/examples/RTSM_VE/Build_Atlasx4/Linux-Release-GCC-4.1/cadi_system_Linux-Release-GCC-4.1.so
> 
> > 
> > Ian.
> >