[RFC 0/5] replace sdhci-bcm2835 with bcm2835-mmc

Stephen Warren swarren at wwwdotorg.org
Wed Oct 29 20:51:06 PDT 2014

On 10/29/2014 03:10 PM, Piotr Król wrote:
> On Tue, Oct 28, 2014 at 07:42:31PM -0600, Stephen Warren wrote:
>> On 10/28/2014 06:00 PM, Piotr Król wrote:
>>> Hi all, this patch series introduce bcm2835-mmc driver, which is
>>> used in Raspberry Pi Foundation kernel. It also adds support for
>>> slave_sg transfer mode in bcm2835-dma driver which is used by
>>> bcm2835-mmc to decrease CPU utilization. This driver significantly
>>> improves throughput and system CPU utilization in comparison with
>>> sdhci-bcm2835. On my Kingston SDC10/8GB I have ~11MB/s writes with
>>> 15s of sys CPU load, instead of ~700KB/s and 43s of sys CPU
>>> utilization.
>>> Almost all code coming from raspberrypi.org, so I'm looking for
>>> feedback about missing/broken stuff and things that should be
>>> redesigned.
>> None of the patches have their git authorship set to the original
>> authors, nor Signed-off-by lines for them. This information should
>> match up with the original patches from the RPI Foundation tree,
>> although sometimes that information is missing there too:-( Please
>> take a look at Documentation/SubmittingPatches section "12) Sign your
>> work".
> Stephen,
> thanks for your comments. Unfortunately there was no "Sign-off-by"
> information in bcm2835-mmc.

Unfortunately, that's often the case in the RPi Foundation tree. What's
worked in the past is to track down the author of the patch in question,
and get them to send an email with their Signed-off-by line for the
patch you want to upstream. Be aware that often the git author of
patches in the RPi tree doesn't actually match the person who wrote the
code, and that often multiple contributions have been squashed into one
commit, so it can be hard to track down who actually wrote what code.
Care is required. Of course, perhaps things have changed since the last
time I looked at the RPi Foundation code, so things /might/ more
carefully managed now.

> I would like to follow your suggestion from
> first patch review and enhance sdhci-bcm2835 with coordination with
> Scott. 
> Do I still have to get Gellert sign-off  when sending small incremental
> changes based on his work ? If yes the what if I can't get this sign-off
> from Gellert ?

If you use extract parts of Gellert's work, or perhaps even just use it
as an information source or for inspiration, you really need to get
his(?) Signed-off-by line in order to send the code upstream. Hopefully
you can just ask via email.

An alternative is to find another source for the information. Suppose
you need to write some new registers that upstream doesn't know about
yet. If these are documented well enough in some public documentation
(or perhaps you reverse engineer that information) that you can write a
completely new patch without incorporating part of Gellert's work (even
if you independently arrive at a similar result), then that should be OK.

What you're trying to ensure is that all the code and information it
embodies has an open heritage, and you're allowed to license the code
under the GPLv2 when you upstream it. This is true if the information
came from public documentation rather than leaks, and either you fully
own the copyright on the code, or there's a Signed-off-by path all the
way back to the original authors.

More information about the linux-rpi-kernel mailing list