On Fri, Aug 17, 2012 at 15:28:51, Tony Lindgren wrote:

* Hiremath, Vaibhav [off-list ref] [120817 02:51]:

quoted

On Fri, Aug 17, 2012 at 14:10:47, Tony Lindgren wrote:

quoted

* Hiremath, Vaibhav [off-list ref] [120817 01:22]:

quoted

On Fri, Aug 17, 2012 at 13:19:34, Tony Lindgren wrote:

quoted

* Hiremath, Vaibhav [off-list ref] [120815 02:11]:

quoted

Somehow we have to have some mechanism to initialize " _mpu_rt_va" before 
core_init call, which will take DT resources and initialize accordingly.

That's for the device reset/idle? We should not do that in hwmod code
except in late_initcall for unclaimed devices as discussed earlier. We
discussed setting up the reset function in the device specific headers
so both device and hwmod code can call them as needed.

May be I didn't understand your above statement, can you please elaborate 
more on "device specific header file"?

We should have the device drivers idle and reset the devices. But also
hwmod may need to do that for the unclaimed devices for PM to work. 
As the driver may not be even compiled in, the driver specific reset and
idle functions should be static inline functions in the device specific
header file so hwmod code can still call those.
 

How about mapping the address of the devices?

That should be done based on the iorange coming from DT during the driver
probe. The range in hwmod can be populated at that point too if needed.
 

There will not be any driver probe for unclaimed devices, right? 

quoted

Some more thoughts on the similar line on unclaimed devices -
As par of late_initcall(), traverse all the DTB nodes and for each node with 
"state = disabled", we can do something like,

Late_initcall()
	-> Traverse DTB nodes and check for "state = disabled"
		-> Read the "ti,hwmod" entry and get offsets like, sys_config 
		   and reset.
		-> Map the base address
		-> Enable the module/clock
		-> Reset the device
		-> Write to sysc register offset
		-> Disable the module/clock

quoted

quoted

Just to highlight, its not only during late_initcall, the _enable and _idle 
functions are getting executed as part of every runtime_pm_get\put_sync from 
the driver.

Heh, but the unclaimed devices do not have a driver :) So there's no runtime_pm
calls for the unclaimed devicdes.

I think you mixed two responses here, this was not part of my last response.
Just only read above code execution steps, I think they still make sense and 
we may end up doing something similar.

quoted

At the current stage, I really think it would be really nice and beneficial 
for driver developers from the community if they get Linux Boot from 
linux-next/master branch. I have seen lot of community folks are struggling 
with it and they are blocked because we do not have booting kernel on 
BeagleBone.
They always end up patching kernel with HWMOD patch and submit the driver 
patches. So request to consider this cleanup as sequential patch on top of 
original HWMOD patch?

Yes I'm fine merging the data as is, just trying to figure out how to sort
out the issues for future. 
 

Thanks a lot Tony.
Paul, Care to add it in your next pull request.

quoted

quoted

quoted

May be get rid of resource overwriting for AM33xx and OMAP5 alone as of 
now??

Sorry I don't follow this part, care to explain a little more?

As per current implementation, we are using HWMOD information to fill the 
platform_device->resources. Even if you pass "reg" and "interrupt" property 
from DT, omap_device layer overwrites it with HWMOD data information.

So what I am trying to propose here is, avoid this overwrite and resources 
(address and interrupt) should strictly gets used from DT blob. Since AM33xx 
and OMAP5 devices are the new devices and supports DT boot only, we can 
implement it easily for them.

Yes I agree. And then we should also drop the data from hwmod when it's no
longer needed.
 

Exactly, I will submit RFC for this.

quoted

Only issue her is, DMA resources, I was going through some of the old email 
archives today, submitted by Benoit and later Jon, but I am still not sure 
how to pass dma_channel to driver as a resource.

Hmm I thought that got merged already? 

No, atleast didn't find in 3.6-rc1.

Anyways, the iorange and interrupts
are standard resources that should exist in every device entry in .dts files.

Agreed, I am submitting patch for this as well.

Thanks,
Vaibhav