[PATCH v2 1/2] of/fdt: export fdt blob as /sys/firmware/fdt

[Grant Likely]

On Wed, Nov 12, 2014 at 11:51 AM, Ard Biesheuvel
<ard.biesheuvel at linaro.org> wrote:
> On 12 November 2014 12:45, Grant Likely <grant.likely at linaro.org> wrote:
>> On Tue, Nov 11, 2014 at 8:34 PM, Ard Biesheuvel
>> <ard.biesheuvel at linaro.org> wrote:
>>> On 11 November 2014 21:08, Grant Likely <grant.likely at linaro.org> wrote:
>>>> On Tue, Nov 11, 2014 at 4:25 PM, Rob Herring <robherring2 at gmail.com> wrote:
>>>>> On Tue, Nov 11, 2014 at 8:42 AM, Grant Likely <grant.likely at linaro.org> wrote:
>>>>>> On Mon, 10 Nov 2014 19:47:01 +0100
>>>>>> , Ard Biesheuvel <ard.biesheuvel at linaro.org>
>>>>>>  wrote:
>>>>>>> Create a new /sys entry '/sys/firmware/fdt' to export the FDT blob
>>>>>>> that was passed to the kernel by the bootloader. This allows userland
>>>>>>> applications such as kexec to access the raw binary. The blob needs to
>>>>>>> be preserved as early as possible by calling preserve_fdt().
>>>>>>>
>>>>>>> The fact that this node does not reside under /sys/firmware/device-tree
>>>>>>> is deliberate: FDT is also used on arm64 UEFI/ACPI systems to
>>>>>>> communicate just the UEFI and ACPI entry points, but the FDT is never
>>>>>>> unflattened and used to configure the system.
>>>>>>>
>>>>>>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel at linaro.org>
>>>>>>
>>>>>> On further thought, nak. I want tree modifications to be treated as
>>>>>> bugs. Do a checksum CRC32 or check. It's simpler and it can be extended
>>>>>> to us an in-blob CRC when the file format gets upreved to include one.
>>>>>
>>>>> Why? MIPS Octeon is full of them BTW. So the rule is any modifications
>>>>> or fixups have to be on the unflattened tree?
>>>>
>>>> That, or we require the fixups to explicitly clean up the CRC after
>>>> themselves. That will force them to be visible. Anything using the
>>>> fdt_*() write functions could potentially take care of it
>>>> automatically. That would make it immediately discoverable where
>>>> changes are being made in the tree.
>>>>
>>>
>>> Doesn't that defeat the purpose of capturing the FDT as it was passed
>>> by the bootloader? Those fixups will be reapplied by the kexec'ed
>>> kernel anyway.
>>
>> We're always going to be in a grey area here. There are some things
>> that should not be passed over. For example, a framebuffer that is
>> passed in from firmware will no longer be valid on kexec. Right now on
>> arch/arm, the exynos platform fixes up the DT because the memory node
>> is outright /corrupt/. In the MIPS case, they do a bunch of stuff in
>> the kernel to figure out the hardware as configured by firmware. I
>> don't agree with that approach, but I would be very surprised if the
>> full original DT has any validity after the early boot fixups.
>>
>
> Well, yes, but those fixups will presumably be done on every boot,
> including the kexec ones.
>
>> Early boot fixups to the .dtb are always going to be oddball cases.
>>
>
> OK, so generally speaking, whether the original DTB is more suitable
> to be presented through /sys/firmware/fdt than the 'current' FDT
> (i.e., whatever is in memory at the time the sysfs node is accessed)
> is not obvious at all. So why not just do the latter?

We're now talking theoreticals. In the case you're working on
(ACPI+FDT), the FDT is generated by Grub or the EFI stub as an
intermediary format, and there isn't currently any code in your boot
path that modifies the FDT. You're arguing for a future situation
where the kernel makes a fixup that is not wanted in the second
kernel. I'm arguing that 1) I want to /know/ when that happens because
there are very few situations where it is a valid thing to do, and 2)
the likelyhood of that happening in the FDT+ACPI case is somewhere
between slim and none because you're working with an intermediary FDT,
not a real hardware description.

/If/ it ever becomes a problem we can revisit, until then I still
prefer the checksum method.

g.