[U-Boot] u-boot on raspberry 2: booting in SVC secure mode

Fri Feb 27 10:41:15 CET 2015

Hi Vincent,

On 26/02/15 09:32, Vincent wrote:
> I tried what Stephen suggested, and just changing CONFIG_SYS_TEXT_BASE to
> 0x0 (with kernel_old=1) does not work: the board display some garbage on
> the uart then hangs. The content of the garbage makes me thinks that
> nothing is done to handle the four cores in this setting which ends up
> badly. I expected this since raspberry's "firmware" only let one core run
> free. I think I need to configure U-boot so that it deals with this SMP
> scenario (I just don't know how).

Ah yes, it seems like there should be code to branch off on secondary
codes based on MPIDR reading in start.S early. I try to take a look at
it later.

> My goal is to use U-boot to load some small homemade baremetal kernels on
> the raspberry (using serial or tftp loading), in *secure mode*, so I need
> U-boot to stay in secure mode, or at least let me install my own
> secure_monitor code before switching to non-secure mode.

So you should set ARMV7_BOOT_SEC_DEFAULT in this case then. Do you
desperately need the other three cores? If not you could get away with
disabling the SMP parts of the code for the time being, putting the
other cores in WFI or WFE.

> I made a couple attempts of adding secure mode support by adding options in
> configs/rpi_2_defconfig but they don't seem to be taken into account. Where
> do you suggest I put ARMV7_virt and ARMV7_NONSEC ?

adding:
------
config ARMV7_NONSEC
	default y
config ARMV7_VIRT
	default y
------
to board/raspberrypi/rpi_2/Kconfig did the trick for me, though it
(correctly) complains about some missing bits. I will not find time
until tonight to work on this, so you may want to take a look at commits
fafbc6c000db and e261c83aa04c meanwhile to see what needs to be done in
this regard for a new board.
I couldn't get u-boot to output anything at all with kernel_old=1 on my
quick try yesterday, what is your current setup to get to the UART
garbage? Did you enable some kind of early debug or did you put any
extra code at 0x0? I tried setting SYS_TEXT_BASE to 0x0, but that didn't
change anything.

Cheers,
Andre.

>> Hi Vincent,
>>
>> On 26/02/15 08:27, Vincent wrote:
>>> Hi,
>>> I finally hacked my way through U-boot and I managed to add raspberry's
>>> boot code inside U-boot so that it can start as usual when using
>> kernel_old
>>> = 1. I don't think
>>> we want this as a final solution but it made me understand a few things
>>> about U-boot architecture (in short: I added a new section located at 0x0
>>> which executes raspberry's code, and
>>> then jump to the usual _start entry point). I didn't try to modify
>>> CONFIG_SYS_TEXT_BASE yet, I'll try this morning.
>>>
>>> From what I gathered from the source code, I think I have to activate
>> some
>>> options (like the ones in arch/arm/cpu/armv7/Kconfig) so that U-boot
>> starts
>>> in secure mode,
>>> but adding them to rpi_2_defconfig doesn't seem to change anything in
>>> .config, so I'm not sure that my current U-boot is "secure boot aware".
>>> Should I add ARMV7_BOOT_SEC_DEFAULT by hand in .config or something like
>>> that ?
>>
>> AFAIK you don't need to tell U-Boot that it runs in secure: unless it
>> accesses any secure/non-secure specific peripherals it should work fine
>> in both modes.
>> Setting ARMV7_BOOT_SEC_DEFAULT just prevents U-Boot to switch into
>> non-sec, but there is no real reason to do so - at least not for Linux.
>>
>> You should do away with the original Raspi firmware snippet - not only
>> from a legal point of view.
>> Thanks to Marc U-Boot has now a much better implementation than my
>> original HYP-mode switcher, also this gives you PSCI support basically
>> for free. So just select ARMV7_VIRT and ARMV7_NONSEC.
>>
>> Do you know if there is a TrustZone Controller in that SoC? That would
>> be needed to guard the resident U-Boot code from the OS. Some SoCs have
>> secure on-chip SRAM usable for that purpose, that would do it, too.
>> But skimming through the BCM2835 .pdf I don't spot any of those,
>> unfortunately.
>>
>> Cheers,
>> Andre.
>>
>>>
>>> 2015-02-25 19:38 GMT+01:00 Stephen Warren <swarren at wwwdotorg.org>:
>>>
>>>> On 02/25/2015 02:30 AM, Vincent wrote:
>>>>
>>>>> Hi,
>>>>> as explained here http://community.arm.com/message/25127, it is
>> possible
>>>>> to
>>>>> boot the raspberry 2 in secure mode, by adding the kernel_old=1 option
>> in
>>>>> config.txt. The main effects of this option are:
>>>>> - all 4 cores start executing in secure SVC mode instead of non-secure
>> SVC
>>>>> mode
>>>>> - all 4 cores start at 0x0000 instead of 0x8000
>>>>> - the initial boot code that setup SMP and exits secure mode is not
>>>>> executed
>>>>>
>>>>> After browsing u-boot's source code, it seems that their boot code is
>> more
>>>>> or less extracted from what u-boot is doing. However I didn't manage to
>>>>> compile u-boot for the raspberry 2 supporting this secure mode.
>>>>>
>>>>> Could anyone explain me what options I need to configure in
>>>>> rpi_2_defconfig
>>>>> so that u-boot supports secure boot for the raspberry 2 and what the
>> boot
>>>>> sequence will be in this case ? I don't mind fixing the code if
>> necessary
>>>>> but I'm a bit lost in the order of events in the initialization.
>>>>>
>>>>
>>>> (Luckily I just happened to notice this message while looking at another
>>>> one nearby. CCing the relevant board maintainer(s) explicitly would help
>>>> your messages be noticed)
>>>>
>>>> To modify U-Boot to support the alternate entry point/load address,
>> you'd
>>>> hopefully only need to change the definition of CONFIG_SYS_TEXT_BASE in
>>>> include/configs/rpi*.h.
>>>>
>>>> I wasn't aware of the thread/option you mention, so I have not attempted
>>>> to boot the RPi2 U-Boot in secure mode. If you're lucky, U-Boot itself
>> will
>>>> "just work" once TEXT_BASE is fixed.
>>>>
>>>> To boot a kernel, you'll probably need to at least configure the ARM
>>>> architected timers CNTFRQ register for the kernel. Perhaps there are a
>> few
>>>> other things like that missing?
>>>>
>>>> It might be interesting to enable U-Boot's PSCI support on the RPi2, so
>>>> that an upstream kernel could gain SMP support without the need for
>>>> explicit BCM2836 SMP support code.
>>>>
>>>> So far, I haven't attempted anything with an (upstream) kernel on RPi2,
>>>> just U-Boot.
>>>>
>>> _______________________________________________
>>> U-Boot mailing list
>>> U-Boot at lists.denx.de
>>> http://lists.denx.de/mailman/listinfo/u-boot
>>>
>>
> _______________________________________________
> U-Boot mailing list
> U-Boot at lists.denx.de
> http://lists.denx.de/mailman/listinfo/u-boot
> 