Re: [PATCH] arm64: dts: qcom: Split sdm845-db845c to add headless support

[Bryan O'Donoghue]

On 05/06/2023 10:47, Amit Pundir wrote:
> This is a follow-up of the upstream discussion,
> https://lore.kernel.org/linux-kernel/20230124182857.1524912-1-amit.pundir@xxxxxxxxxx/T/#u,
> around adding a reserved memory region in sdm845-db845c
> for the framebuffer memory (the splash region set up by
> the bootloader) but the general opinion was to avoid
> adding that reserved memory for the headless DB845c
> usecase.
>
> So this patch splits the sdm845-db845c into a common dtsi,
> a new sdm845-db845-headless DT, which disables the mdss
> display subsystem, and a new sdm845-db845c DT with an
> additional reserved-memory region for the framebuffer.
>
> The default sdm845-db845c.dtb remains pretty much the same
> (with an exception of additional reserved-memory region),
> while others can use sdm845-db845c-headless.dtb for their
> headless systems.
>
> Signed-off-by: Amit Pundir <amit.pundir@xxxxxxxxxx>

Walk me through this one again.

We only need this reserved memory for as long as the splash screen 
persists - correct ?

If there is no HDMI or DSI connected then XBL will never allocate this 
memory.

If the MDPSS? driver runs, the display hardware gets pointed at 
different memory and the splashscreen memory is either marked as 
reserved - and never usable again or its not marked as reserved and we 
have system race but in the racy case, the memory is usable.

Is the solution then not to make the MDSS driver always liberate this 
memory - irrespective of whether or not a display is connected ?

It seems wrong to sacrifice ~ 37 megabytes of memory permanently. If I 
were building an embedded system based on db845 I would want to have 
that memory available to me either way.

---
bod