[PATCH v2 3/4] bloblist: add support for CONFIG_BLOBLIST_PASSAGE

[Harrison Mutai]

On 31/01/2025 00:10, Tom Rini wrote:
> On Mon, Jan 27, 2025 at 12:42:30PM +0000, Harrison Mutai wrote:
>> Hi Tom,
>>
>> Thanks for the additional details, that helped!
>>
>> On 23/01/2025 17:12, Tom Rini wrote:
>>> On Thu, Jan 23, 2025 at 05:03:40PM +0000, Harrison Mutai wrote:
>>>> On 22/01/2025 17:21, Tom Rini wrote:
>>>>> On Wed, Jan 15, 2025 at 01:52:06PM +0000, Harrison Mutai wrote:
>>>>>
>>>>>> When the configuration option CONFIG_BLOBLIST_PASSAGE is selected, the
>>>>>> bloblist present in the incoming standard passage is utilised in-place.
>>>>>> There is no need to specify the size of the bloblist as the system
>>>>>> automatically detects it using the header information.
>>>>>>
>>>>>> Signed-off-by: Harrison Mutai <harrison.mutai at arm.com>
>>>>>> ---
>>>>>>     common/Kconfig    | 8 +++++++-
>>>>>>     common/bloblist.c | 5 ++++-
>>>>>>     2 files changed, 11 insertions(+), 2 deletions(-)
>>>>>
>>>>> This leads to failure to boot on qemu-arm-sbsa as seen in CI:
>>>>> https://dev.azure.com/u-boot/u-boot/_build/results?buildId=10392&view=logs&j=ac5e904b-a30c-5e1d-9570-462b8181c990&t=863544fb-ed04-5d34-fdac-10f5bbb057bd&l=423
>>>>>
>>>> Sorry, I'm unfamilliar with U-Boot's CI but I can't make out what the source
>>>> of the failure is. All I see in the logs is what looks like an intermittent
>>>> error.
>>>>
>>>> https://dev.azure.com/u-boot/u-boot/_build/results?buildId=10392&view=logs&j=ac5e904b-a30c-5e1d-9570-462b8181c990&t=863544fb-ed04-5d34-fdac-10f5bbb057bd&l=423
>>>>
>>>> Would you be able to share the results XML?
>>>
>>> The problem is that qemu doesn't start anymore, sorry for being unclear.
>>> Looking at
>>> https://source.denx.de/u-boot/u-boot-test-hooks/-/blob/master/bin/travis-ci/conf.qemu-arm-sbsa_na?ref_type=heads
>>> we can see how qemu is invoked.
>>>
>>
>> I've successfully reproduced and resolved the error. The issue occurred
>> because I relaxed the condition for calling the architecture-specific hook
>> xferlist_from_boot_arg. Previously, this hook was only called if the
>> bloblist was at a fixed address and U-Boot was in the first phase. However,
>> since we also need the hook for BLOBLIST_PASSAGE, I've added a condition to
>> prevent calling the hook if BLOBLIST_ALLOC is enabled, assuming we don't
>> expect to receive a bloblist from the previous stage in this mode.
>>
>> However, I'm slightly unsure if this assumption is correct. Generally, when
>> BLOBLIST_ALLOC is enabled, is it expected that U-Boot will allocate memory
>> for the bloblist, ignoring the bloblist from previous stages? In the case of
>> BLOBLIST_FIXED, we relocate any bloblist received from the prior stage to
>> the fixed address. I'm surprised we don't do the same with BLOBLIST_ALLOC.
> 
> The problem is that today at least yes, BLOBLIST_ALLOC is only viable
> for full U-Boot as we don't have a mechanism to pass our own from one
> stage to the next unless it's fixed.
> 

To summarize, BLOBLIST_ALLOC is meant to serve as a fallback buffer for 
internal use when a prior-stage bootloader hasn't provided a bloblist at 
a fixed address (BLOBLIST_FIXED) or a pointer to a bloblist 
(BLOBLIST_PASSAGE). From the perspective of a prior-stage bootloader, 
these are the only relevant options. BLOBLIST_ALLOC is also entirely 
irrelevant when U-Boot is staged. Have a I missed anything?