[U-Boot] [PATCH v4 19/21] sandbox: Allow to execute from RAM

[Alexander Graf <agraf@suse.de>]

On 06/21/2018 09:45 PM, Simon Glass wrote:
> Hi Alex,
>
> On 21 June 2018 at 03:44, Alexander Graf <agraf@suse.de> wrote:
>> On 06/21/2018 04:02 AM, Simon Glass wrote:
>>> Hi Alex,
>>>
>>> On 18 June 2018 at 09:23, Alexander Graf <agraf@suse.de> wrote:
>>>> With efi_loader, we may want to execute payloads from RAM. By default,
>>>> permissions on the RAM region don't allow us to execute from there
>>>> though.
>>>>
>>>> So whenever we get into the efi_loader case, let's mark RAM as
>>>> executable.
>>>> That way we still protect normal cases, but allow for efi binaries to
>>>> directly get executed from within RAM.
>>>>
>>>> For this, we hook into the already existing allow_unaligned() call which
>>>> also transitions the system over into semantics required by the UEFI
>>>> specification.
>>>>
>>>> Signed-off-by: Alexander Graf <agraf@suse.de>
>>>> ---
>>>>    arch/sandbox/cpu/cpu.c | 14 ++++++++++++++
>>>>    arch/sandbox/cpu/os.c  | 14 ++++++++++++++
>>>>    include/os.h           | 19 +++++++++++++++++++
>>>>    3 files changed, 47 insertions(+)
>>>>
>>> What is this patch actually for? Does it make something work that did
>>> not before? Where is it called?
>>
>> At least on aarch64 executing from the RAM region fails on the first
>> instruction you call inside it, because it's not mapped with PROT_EXEC. I
>> think not mapping it with PROT_EXEC is a good thing in the normal sandbox
>> use case, but for EFI we need to run from RAM ;).
>>
>> So yes, this patch makes that work. It's called from allow_unaligned() which
>> gets called from the bootefi command function.
> OK I see. This is so that sandbox running on an ARM platform can
> actually load and run an EFI application, right?
>
> Can you please make this code and commit message ARM-specific? Also,

Not, it's not ARM specific. You may hit this on any system that actually 
checks for PROT_EXEC.

> just call your code on start-up in cpu.c. We don't need to connect to
> an allow_unaligned() thing, which is a misnomer in this case.

In that case we can just tell mmap() to map RAM with PROC_EXEC.


Alex