[Openpvrsgx-devgroup] DDK 1.17 on S5PV210

[H. Nikolaus Schaller]

Hi Jonathan,

> Am 30.03.2020 um 00:43 schrieb Jonathan Bakker <xc-racer2 at live.ca>:
> 
> Hi Nikolaus,
> 
> On 2020-03-29 1:07 p.m., H. Nikolaus Schaller wrote:
>> Hi Jonathan,
>> 
>>> Am 29.03.2020 um 21:52 schrieb Jonathan Bakker <xc-racer2 at live.ca>:
>>> I've put the kernel driver part at https://github.com/xc-racer99/linux/tree/v5.5.1-all-devices-openpvr - note that I currently have the clock and power domain always enabled at probe as active power management is disabled for the blobs.  I'm not sure how the omap devices deal with this, but my "solution" would cause issues with increased power draw during suspend.
>> 
>> Great! I didn't know yet that there are Samsung SoC with SGX540...
>> 
>> Did you also take the omap pvrsrvkm?
>> 
>> If you have patches ready for integration into the openpvrsgx tree,
>> please let me know.
>> 
>> And what about device tree? I am again trying to get a set of
>> DTS patches and bindings upstream: 
>> 
>> https://patchwork.kernel.org/patch/11464133/
>> 
>> So maybe we can add the s5pv210 setup there
>> 
>> (https://github.com/xc-racer99/linux/commit/54d5952136c7d1fa6c3fc792a8497ed26bba39dd ?)
> 
> Yep, that's the DTS commit.  There's a singular controllable clock, but the SoC datasheet (available at http://dl.project-voodoo.org/archives/documentation/S5PC110_EVT1_UM10.pdf - see page 1382) makes note that there's both a bus and a core clock, a reset pin, and the irq pin connected to it (see page 1391).  The power-domains isn't upstream, but the assigned clocks appear to be required at those speeds.
> 
> I did use the omap pvrsrvkm for this port, but earlier I had also forward-ported the Android pvrsrvkm (DDK 1.8) and got things running using libhybris.
> 
> In terms of integration with the openpvrsgx tree, the biggest blocker is the power management.  I'm hesitant to post patches that always leave the GPU running even when it's not in use or the device is suspended.

This may be the same as with the jz4780 where I haven't seen any information about GPU power management. We also just enable clocks.

>  Do OMAP devices ever disable the GPU when active power management is disabled?  Otherwise, the patches should be good to go as my kernel tree is based on your repo :)

I do not exactly know but the TI tree has some power management stuff. And Tony's patches for pvr_drv (which is our wrapper around the TI code) do register pvr_pm_ops:

https://github.com/openpvrsgx-devgroup/linux_openpvrsgx/blob/letux-pvrsrvkm/drivers/gpu/drm/pvrsgx/pvr-drv.c

So these need to be effective by code outside of the pvrsrvkm.

But I think for the moment it is already good enough for integration into our repo. In worst case one has to "modprobe -r pvrsrvkm"...

> 
> In terms of Samsung SoC's with PowerVR - I believe there's two - s5pv210 (AKA s5pc110, just a different package version) present in the first-gen Galaxy S devices among others and the exynos 5 5410 (SGX 544 of some sort) which is found in the Galaxy S4 (Exynos version).

I have looked around and the Exynos 5410 is said to have a 3-core SGX544 (omap/dra chips have singe/dual core).

BR and thanks,
Nikolaus