[Letux-kernel] [PATCH 00/20] A bunch of JZ4730 fixups for letux-kernel

[Paul Boddie]

On Thursday, 14 January 2021 19:02:12 CET H. Nikolaus Schaller wrote:
> Am 14.01.2021 um 11:55 schrieb H. Nikolaus Schaller <hns at goldelico.com>:
> > 
> > So we have a problem with the pinmux function selection.
> 
> Found it. Can now boot to blinking cursor without any user-space
> help. Unless I try to start X11. Maybe X11 needs too much memory
> or is influenced by the schedluer timer.
> 
> The problem was that there are new macros INGENIC_PIN_GROUP() and
> INGENIC_PIN_GROUP_FUNCS() in pinctrl-ingenic to optimize memory
> useage. So I had updated the earlier jz4730 pinmux group tables.
> 
> And I defined one of the pinmux arrays as int[] instead of u8[].
> So 3 bytes were 0x00 until the next valid function state was there.
> This made the pin functions for the LCDC not properly initialized
> and so there was no HSYNC/VSYNC output to the panel.

That was rather unfortunate and potentially difficult to spot, so it was a 
nice piece of detective work!

> Running my devmem2 script did overwrite and the display worked.
> 
> > This means the LCDC hacks are already working and the different
> > lcd and pix clocks are not that critical.
> 
> BTW, I think I have found a patch for the DMA descriptor number
> which should not harm the SoC with OSD and two LCDs but still do
> the right thing for jz4730 and jz4740. So this is potentially
> upstreamable.

I'll make a note about that below.

> So what remains to be fixed? Only the issue with the connector
> type and bus format. I think the hack should not disturb other
> Letux devices so we can merge it temporarily into the Letux tree.

OK. It would be good to figure that out and get it tested by CI20 users.

> So the Display should work now. I will consolidate the patches
> so that my printk stuff is removed... So we will have a 5.11-rc4
> on Monday with working LCD.
> 
> Then we can take another look at the I2C issues and the sluggish
> scheduler timer and DMA. And finally on sound and Ethernet and why
> X11 is not running.

A quick note about the timer: I found that in my L4 environment, sleep 
operations seem to be accurate. However, that environment employs the OST for 
scheduling (unless I am mistaken), and might not be instructive about the 
timing-related issues within Linux.

[...]

> commit 7f4e6eda5bc9b634900ff33ae0fa3e7a7ce832ea
> Author: H. Nikolaus Schaller <hns at goldelico.com>
> Date:   Thu Jan 14 17:41:00 2021 +0100
> 
>     drm/ingenic: fix assignment of DMA channels
>     for SoC without OSD like jz4730 and jz4740. On those, the first DMA
>     channel is for the single LCD and the second for the palette. If we
>     leave the first uninitialized, the LCDC will not become enabled.

Just to clarify this, the second channel is for a second panel (where STN/
passive matrix panels are used). As I noted earlier, the palette details are 
actually "chained" to framebuffer details using two descriptors associated 
with a single channel.

Or in a simple diagram:

LCDDA0 -> palette descriptor <- next
          ...                   ...
          next ---------------> framebuffer descriptor

LCDDA1 -> (descriptors for the second STN panel, which we don't have)

Paul