[U-Boot] [PATCH v2 1/1] core/uclass: iterate over all devices of a uclass

[Andreas Färber]

Am 19.04.2017 um 23:34 schrieb Simon Glass:
> On 19 April 2017 at 15:06, Andreas Färber <afaerber at suse.de> wrote:
>> Am 19.04.2017 um 13:26 schrieb Heinrich Schuchardt:
>>> When iterating over the devices of an uclass the iteration stops
>>> at the first device that cannot be probed.
>>> When calling booefi this will result in no block device being
>>
>> "bootefi"
>>
>>> passed to the EFI executable if the first device cannot be probed.
>>>
>>> The problem was reported by Andreas Färber in
>>> https://lists.denx.de/pipermail/u-boot/2017-April/287432.html
>>>
>>> For testing I used an odroid-c2 with a dts including
>>> &sd_emmc_a {
>>>       status = "okay";
>>> }
>>> This device does not exist on the board and cannot be initialized.
>>>
>>> With the patch uclass_first_device and uclass_next_device
>>> iterate internally until they find the first device that can be
>>> probed or the end of the device list is reached.
>>>
>>> Debug output is provided for the two functions.
>>>
>>> Reported-by: Andreas Färber <afaerber at suse.de>
>>> Cc: Simon Glass <sjg at chromium.org>
>>> Signed-off-by: Heinrich Schuchardt <xypron.glpk at gmx.de>
>>> ---
>>> v2:
>>>   As suggested by Simon Glass correct uclass_first_device() and
>>>   uclass_next_device() instead of uclass_get_device_tail() to
>>>   avoid side effects.
>>> v1:
>>>   The original patch was posted as
>>>   core/uclass: uclass_get_device_tail: always set devp
>>>   https://lists.denx.de/pipermail/u-boot/2017-April/288068.html
>>> ---
>>>  drivers/core/uclass.c | 44 +++++++++++++++++++++++++++++++++-----------
>>>  1 file changed, 33 insertions(+), 11 deletions(-)
>>
>> Reviewed-by: Andreas Färber <afaerber at suse.de>
>> Tested-by: Andreas Färber <afaerber at suse.de>
>>
>> Confirming that on my Vega S95 Telos this results in a full GRUB menu,
>> and GRUB sees two disks.
>>
>> Many thanks,
>>
>> Andreas
> 
> Just to be clear, I am NAKing this patch. I do not want to change the
> existing semantics as it requires existing code to check the function
> return value.

And to be clear I still think you are mistaken in holding on to an
implementation that assumes that all devices will probe okay. We might
be able to fix the issue at hand in different ways - be it at the
callsite or by adding a new API - but this is papering over other
potential problems in the codebase. Hardware might malfunction, drivers
might not be implemented in U-Boot or be disabled by user's config, etc.

The caller might notice that some error has occurred by checking the
return code, but what do you seriously expect it to do then? If it's
supposed to call an alternative iteration function to continue, it could
just call that in the first place.

Regards,
Andreas

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)