On 05.06.20 16:57, Ard Biesheuvel wrote:
> On Fri, 5 Jun 2020 at 16:54, Heinrich Schuchardt <xypron.glpk@xxxxxx> wrote:
>>
>> On 05.06.20 14:39, Ard Biesheuvel wrote:
>>> On Fri, 5 Jun 2020 at 14:20, Heinrich Schuchardt <xypron.glpk@xxxxxx> wrote:
>>>>
>>>> On 05.06.20 13:52, Ard Biesheuvel wrote:
>>>>> EFI on ARM only supports short descriptors, and given that it mandates
>>>>> that the MMU and caches are on, it is implied that booting in HYP mode
>>>>> is not supported.
>>>>>
>>>>> However, implementations of EFI exist (i.e., U-Boot) that ignore this
>>>>> requirement, which is not entirely unreasonable, given that it does
>>>>> not make a lot of sense to begin with.
>>>>
>>>> Hello Ard,
>>>>
>>>> thanks for investigating the differences between EDK2 and U-Boot.
>>>>
>>>> What I still do not understand is if there is a bug in U-Boot where
>>>> U-Boot does not conform to the UEFI specification? In this case I would
>>>> expect a fix in U-Boot.
>>>>
>>>
>>> U-Boot violates the EFI spec, yes. The spec is very clear on how the
>>> MMU is configured, and this rules out booting with the caches off, or
>>> booting in HYP mode.
>>>
>>> However, given that this is the situation today, we still need to deal
>>> with it on the Linux side.
>>> In parallel, fixing it in U-boot may be appropriate. However, I think
>>> the EFI spec is too detailed here - instead of 'booting at the highest
>>> non-secure privilege mode', it tells you which exact bits to set in
>>> SCTLR, which seems overzealous to me. In other words, even though it
>>> violates the letter of the spec, I don't mind dealing with this
>>> exception in the Linux side, since the requirement is somewhat
>>> unreasonable to begin with.
>>>
>>>> Or is it simply a deficiency of Linux that it does not properly support
>>>> HYP/EL2 mode on 32-bit ARM?
>>>>
>>>
>>> No, this is definitely not the fault of Linux.
>>>
>>>> Up to now I never experience a problem booting a 32bit board via U-Boot
>>>> -> GRUB-EFI -> Linux. Where did you have a problem when booting via
>>>> U-Boot's UEFI implementation and the current Linux kernel?
>>>>
>>>
>>> I haven't managed to make it fail, but my only 32-bit HYP capable
>>> platform is QEMU. I am not 100% convinced that the EFI+HYP+U-Boot case
>>> is rock solid, and I may have gotten lucky with QEMU (which uses
>>> emulation not virtualization when running at HYP)
>>>
>>> Do you have any A7/A15 based boards that don't print 'WARNING: Caches
>>> not enabled' at boot?
>>
>> Hello Ard,
>>
>> I have no board that prints this. Where did you actually see this output?
>>
>
> In U-boot
>
> arch/arm/lib/cache.c: * Default implementation of enable_caches()
> arch/arm/lib/cache.c- * Real implementation should be in platform code
> arch/arm/lib/cache.c- */
> arch/arm/lib/cache.c:__weak void enable_caches(void)
> arch/arm/lib/cache.c-{
> arch/arm/lib/cache.c- puts("WARNING: Caches not enabled\n");
> arch/arm/lib/cache.c-}
> arch/arm/lib/cache.c-
>
> The QEMU port does not override that routine. This may be the reason
> it doesn't work for me under KVM, but only under emulation.
>
>> The string "Caches not enabled" does not exist in Linux next-20200505
>> according to "git grep -n 'ache not enabled'".
>>
>> Here is some sample output for an Orange Pi PC with a quad core
>> Allwinner H3 Soc. "ARMv7 Processor rev 5 (v7l)" according to
>> /proc/cpuinfo, compatible to "arm,cortex-a7" according to the device tree.
>>
>> $ uname -m
>> Linux orangepi-pc 5.6.0-2-armmp-lpae #1 SMP Debian 5.6.14-1 (2020-05-23)
>> armv7l GNU/Linux
>>
>
> Could you check whether it boots in HYP mode?
>
> [ 0.381460] CPU: All CPU(s) started in SVC mode.
>
> vs
>
> [ 0.135626] CPU: All CPU(s) started in HYP mode.
>
>
> ?
>
Booted via GRUB-EFI:
sudo dmesg -H | grep 'started in'
[ +0.000017] CPU: All CPU(s) started in HYP mode.
Booted via Linux stub:
$ sudo dmesg -H | grep 'started in'
[ +0.000016] CPU: All CPU(s) started in HYP mode.
Best regards
Heinrich
