[PATCH] drm/sun4i: tcon: Support keeping dclk rate upon ancestor clock changes

[Maxime Ripard]

Hi,

On Sun, Mar 10, 2024 at 02:32:29PM +0100, Frank Oltmanns wrote:
> Allow the dclk to reset its rate when a rate change is initiated from an
> ancestor clock. This makes it possible to no longer to get an exclusive
> lock. As a consequence, it is now possible to set new rates if
> necessary, e.g. when an external display is connected.
> 
> The first user of this functionality is the A64 because PLL-VIDEO0 is an
> ancestor for both HDMI and TCON0. This allows to select an optimal rate
> for TCON0 as long as there is no external HDMI connection. Once a change
> in PLL-VIDEO0 is performed when an HDMI connection is established, TCON0
> can react gracefully and select an optimal rate based on this the new
> constraint.
> 
> Signed-off-by: Frank Oltmanns <frank at oltmanns.dev>
> ---
> I would like to make the Allwinner A64's data-clock keep its rate
> when its ancestor's (pll-video0) rate changes. Keeping data-clock's rate
> is required, to let the A64 drive both an LCD and HDMI display at the
> same time, because both have pll-video0 as an ancestor.
> 
> TCONs that use this flag store the ideal rate for their data-clock and
> subscribe to be notified when data-clock changes. When rate setting has
> finished (indicated by a POST_RATE_CHANGE event) the call back function
> schedules delayed work to set the data-clock's rate to the initial value
> after 100 ms. Using delayed work maks sure that the clock setting is
> finished.
> 
> I've implemented this functionality as a quirk, so that it is possible
> to use it only for the A64.
> 
> This patch supersedes [1].
> 
> This work is inspired by an out-of-tree patchset [2] [3] [4].
> Unfortunately, the patchset uses clk_set_rate() directly in a notifier
> callback, which the following comment on clk_notifier_register()
> forbids: "The callbacks associated with the notifier must not re-enter
> into the clk framework by calling any top-level clk APIs." [5]
> Furthermore, that out-of-tree patchset no longer works since 6.6,
> because setting pll-mipi is now also resetting pll-video0 and therefore
> causes a race condition.

Workqueues don't have an upper boundary on when they execute. As we
discussed multiple times, this should be solved in the clock framework
itself, not bypassing it.

Maxime
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20240314/633c4520/attachment.sig>