Re: [PATCH v4 0/5] Add support for the ARMv8.2 Statistical Profiling Extension

From: Kim Phillips
Date: Thu Jul 06 2017 - 13:09:00 EST


On Thu, 29 Jun 2017 12:11:02 +0100
Mark Rutland <mark.rutland@xxxxxxx> wrote:

> On Wed, Jun 28, 2017 at 07:59:53PM -0500, Kim Phillips wrote:
> > arm-trusted-firmware, btw, has just been updated to enable SPE at lower
> > ELs, so I don't have to use a hacked-up version anymore.
> >
> > I also updated my BL33 to the latest upstream u-boot
> > vexpress_aemv8a_dram_defconfig, and at least now the kernel continues
> > to boot, even though it can't bring up 6 of the 7 secondary CPUs.
>
> Do you mean that you replaced the bootwrapper with u-boot?

no, sorry, arm-trusted-firmware wants a BL33 image, which u-boot
provides.

Sorry but I guess I'm not using the bootwrapper, and we are launching
the model in completely different manners.

The bootwrapper input is a kernel and a dtb, and it emits a dtb and a
linux-system.axf file, the latter of which I don't see how to launch
the model with: The model script I'm using uses a kernel, dtb, and an
fip.bin and bl1.bin.

Can you share how you invoke the model, presumably with the .axf file?

> The --with-cpu-ids option *adds* CPU nodes, but leaves the broken ones,
> and your CPU phandles (and PPI partitions for the SPE node(s)) will all
> be wrong. Linux is still seeing those erroneous CPU nodes (presumably
> taking Linux CPU ids 2-7).
>
> Generally, --with-cpu-ids doesn't work as you'd expect, which is why it
> got removed in favour of assuming an initally correct DT.
>
> Please fix the DT instead. With a fixed DT, and commit ccdc936924b3682d,
> the bootwrapper won't further mangle your DT.

OK, changing the CPU IDs alone didn't work (kernel didn't even say hi),
but taking what commit ccdc936924b3682d does to the cpu_on/off
properties makes it work for my arm-trusted-firmware (non-boot-wrapper)
invocation, so I have to use the wrapper if I change my DT CPUs for the
time being.

So I'm OK now for at least the two-partition, four CPUs each setup, but
for topologies as described in Marc/Will's fvp-base.dts commit, I don't
see how to run without knowing how to make the axf file work with the
model, i.e., solely with the boot-wrapper.

Thanks,

Kim