[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH v2] xen: arm64: Add support for Renesas RCar Gen3 H3 Salvator-X platform



Hi,

On 11.07.2016 11:25, Andre Przywara wrote:
Hi Dirk,

On 08/07/16 12:38, Dirk Behme wrote:
Hi Andre,

On 05.07.2016 16:29, Andre Przywara wrote:
Hi,

On 05/07/16 15:22, Dirk Behme wrote:
On 05.07.2016 15:45, Andre Przywara wrote:
Hi,

On 05/07/16 14:34, Julien Grall wrote:
(CC Andre)

On 05/07/16 14:04, Dirk Behme wrote:
On 05.07.2016 14:45, Julien Grall wrote:


On 05/07/16 13:09, Dirk Behme wrote:
Hi Julien,

On 05.07.2016 13:39, Julien Grall wrote:
Hi Dirk,

On 05/07/16 07:37, Dirk Behme wrote:
Signed-off-by: Dirk Behme <dirk.behme@xxxxxxxxxxxx>

This patch looks good to me, however I would like to see the
documentation on the wiki page (see [1]) before giving any formal
ack.


Ok, many thanks for your review!

Yes, I understood that something more is needed.


I just wonder if we keep the patch on the mailing list for a
moment.
With this it's publically available and we can see how's the public
interest for this board.

Additionally, getting Xen running on this board and describe all
this in
the wiki isn't that easy. You either need to modify the flashed
firmware. Or, like me, load everything via a JTAG debugger ...

Can you details why you think it is not easy? Why do you have to
modify
the firmware? Is it because it does not boot the hypervisor in EL2?


The board boots via ATF, which starts U-Boot in EL1, then. You
have to
find a way to load Xen into memory (e.g. U-Boot TFTP) and then
switch to
EL2 to start Xen.

But ATF is running in EL3, right? If so, can ATF just starts U-boot
in EL2?

I have a board at home (pine64) which also uses ATF and U-Boot and is
able to boot Xen without any SMC call because the U-Boot is entered
in EL2.

 From having a very quick look into the rcar ATF port on github[1] I
see:

+#elif (RCAR_BL33_EXECUTION_EL == BL33_EL2)
+    return (uint32_t)SPSR_64(MODE_EL2, MODE_SP_ELX,
DISABLE_ALL_EXCEPTIONS);

So if you rebuild ATF with RCAR_BL33_EXECUTION_EL set to BL33_EL2 that
should enter U-Boot in EL2.

The fix for the Pine64 was equally simple and works fine since then.


This is an other solution most probably I meant when talking about
"modifying the firmware" ;)

I'm somehow unsure about the pros and cons running U-Boot at EL1 versus
EL2, though.

It shouldn't matter, really. U-Boot on AArch64 (called armv8 here) is
able to run in any exception level (in contrast to ARMv7).
Actually running in EL2 is the architecturally recommended way, even
with ATF (by default it drops into EL2 when returning to non-secure
world). On Juno (and the Pine64) is works fine this way.

As U-Boot only uses an identity mapping, many differences between EL2
and EL1 don't matter.

An understanding question regarding this: We found that running U-Boot
at EL2 works fine and starting Xen works fine, then, too. So many thanks
for that hint!

But we found that the Linux native boot case (without Xen) does fail,
then. I.e. running U-Boot in EL1 starts the native Linux kernel
successfully. While running U-Boot in EL2 starting the native Linux
kernel fails, then.

That's rather odd, can you provide some debug information?

Any idea regarding this?

Who is responsible to switch to EL1 for the native Linux boot case if
U-Boot runs at EL2?

That's well supported kernel code in arch/arm64/kernel/head.S. If it
sees the kernel to be entered in EL2 (which is highly recommended, btw),
it installs the KVM HYP stub and drops into EL1 (after some register setup).
You might want to check the code in there to see if something springs to
mind or use some debug UART output to see where it gets stuck (if it
gets stuck). I usually do something like "mov x1, #UART_TR; mov w0,
#'1'; str w0, [x1]" to get an idea where low level code hangs.


Just for the logs, it seems that the kernel I was using missed

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/arch/arm64/boot/dts/renesas?id=4c811edf65e809e6ac6ec35f4818efba2b1c6163

resulting in EL2 interrupt misbehavior.


Thanks!

Best regards

Dirk

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.