Re: [PATCH 2/6] wl1251: Use request_firmware_prefer_user() for loading NVS calibration data

On 4-5-2017 4:28, Luis R. Rodriguez wrote:

On Wed, May 03, 2017 at 09:02:20PM +0200, Arend Van Spriel wrote:

quoted

On 3-1-2017 18:59, Luis R. Rodriguez wrote:

quoted

On Mon, Dec 26, 2016 at 05:35:59PM +0100, Pavel Machek wrote:

quoted

Right question is "should we solve it without user-space help"?

Answer is no, too. Way too complex. Yes, it would be nice if hardware
was designed in such a way that getting calibration data from kernel
is easy, and if you design hardware, please design it like that. But
N900 is not designed like that and getting the calibration through
userspace looks like only reasonable solution.

Arend seems to have a better alternative in mind possible for other
devices which *can* probably pull of doing this easily and nicely,
given the nasty history of the usermode helper crap we should not
in any way discourage such efforts.

Arend -- please look at the firmware cache, it not a hash but a hash
table for an O(1) lookups would be a welcomed change, then it could
be repurposed for what you describe, I think the only difference is
you'd perhaps want a custom driver hook to fetch the calibration data
so the driver does whatever it needs.

Hi Luis,

I let my idea catch dust on the shelf for a while. 

:) BTW did you get to check out Daniel Wagner and Tom Gundersen's firmware work
[0] ? This can provide a proper clear fallback mechanism which *also* helps
address the race between "critical mount points ready" problem we had discussed
long ago. IIRC it did this by having two modes of operation a best effort-mode
and a final-mode. The final-mode would only be used once all the real rootfs is
ready. Userspace decides when to kick / signal firmwared to switch to final-mode
as only userspace will know for sure when that is ready. The best-effort mode
would be used in initramfs. There is also no need for a "custom fallback", the
uevent fallback mechanism can be relied upon alone. Custom tools can just fork
off and do something similar to firmward then in terms of architecture. This is
a form of fallback mechanism I think I'd be happy to see enabled on the new
driver data API. But of course, I recall also liking what you had in mind as well
so would be happy to see more alternatives! The cleaner the better !

[0] https://github.com/teg/firmwared

quoted

Actually had a couple
of patches ready, but did get around testing them. So I wanted to rebase
them on your linux-next tree. I bumped into the umh lock thing and was
wondering why direct filesystem access was under that lock as well.

Indeed, it was an odd thing.

quoted

In your tree I noticed a fix for that. 

Yup!

It took a lot of git archeology to reach a sound approach forward which makes
me feel comfortable without regressing the kernel, this should not regress
the kernel.

quoted

The more reason to base my work on
top of your firmware_class changes. Now my question is what is the best
branch to choose, because you have a "few" in that repo to choose from ;-)

I have a series of topical changes, and I rebase onto the latest linux-next
regularly before posting patches, if 0-day finds issues, I dish out a next
try2 or tryX iteration until all issues are fixed. So in this case you'd
just go for the latest driver-data-$(next_date) and if there is a try
postfix use the latest tryX.

In this case 20170501-driver-data-try2, this is based on linux-next tag
next-20170501. If you have issues booting on that next tag though you
could instead try the v4.11-rc8-driver-data-try3 branch based on v4.11-rc8.
But naturally patches ultimately should be based on the latest linux-next
tag for actual submission.

PS. after my changes the fallback mechanism can easily be shoved into its
own file, not sure if that helps with how clean of a solution your work
will be.

Let me explain the idea to refresh your memory (and mine). It started
when we were working on adding driver support for OpenWrt in brcmfmac.
The driver requests for firmware calibration data, but on routers it is
stored in flash. So after failing on the firmware request we now call a
platform specific API. That was my itch, but it was not bad enough to go
and scratch. Now for N900 case there is a similar scenario alhtough it
has additional requirement to go to user-space due to need to use a
proprietary library to obtain the NVS calibration data. My thought: Why
should firmware_class care? So the idea is that firmware_class provides
a registry for modules that can produce a certain firmware "file". Those
modules can do whatever is needed. If they need to use umh so be it.
They would only register themselves with firmware_class on platforms
that need them. It would basically be replacing the fallback mechanism
and only be effective on certain platforms.

Let me know if this idea is still of interest and I will rebase what I
have for an RFC round.

Regards,
Arend