Hello Laurent, Marek, On Fri Jun 5, 2026 at 7:31 PM CEST, Laurent Pinchart wrote: > On Fri, Jun 05, 2026 at 06:56:54PM +0200, Luca Ceresoli wrote: >> On Fri Jun 5, 2026 at 6:09 PM CEST, Laurent Pinchart wrote: >> > On Fri, Jun 05, 2026 at 05:18:43PM +0200, Luca Ceresoli wrote: >> >> Hello Sean, >> >> >> >> +Cc Marek, Maxime. >> >> >> >> On Sat May 30, 2026 at 8:51 PM CEST, Sean Nyekjaer wrote: >> >> > On Wed, May 27, 2026 at 02:27:36PM +0100, Sudarshan Shetty wrote: >> >> >> The current DSI configuration enables MIPI_DSI_MODE_VIDEO_BURST. >> >> >> while burst mode is supported by the hardware, its use >> >> >> depends on continuous clock behavior from the DSI host. In practice, >> >> >> burst mode may introduce instability depending on the host controller >> >> >> implementation, as the DSI link may transition to low-power state >> >> >> between bursts. >> >> >> >> >> >> Testing showed improved display stability when using non-burst mode on >> >> >> affected panels. >> >> >> >> >> >> Remove MIPI_DSI_MODE_VIDEO_BURST and use non-burst video mode. >> >> > >> >> > We briefly talked about this at Embedded Recipes >> >> > I promised to sent a link: >> >> > https://lore.kernel.org/all/[email protected]/ >> >> >> >> Thanks for the discussion at ER and for this follow-up e-mail. >> >> >> >> > When burst mode is enabled, the LVDS clock gets way to high for my >> >> > panel. I don't know if it's the DSI controller in the STM32MP1 or >> >> > something not supported on the TI side. >> >> > >> >> > We have been running with this fix for 2 years :) >> >> >> >> If I can summarize the situation in the last 4 years: >> >> >> >> * Several users reported the same trouble >> >> * Those users patch their kernel out of tree to disable burst mode as a >> >> workaround >> >> * According to Marek the correct way to make burst mode work is >> >> implementing link negotiation >> >> * Nobody is willing to implement link negotiation as of now >> >> >> >> And this leads me to some questions. >> >> >> >> * Do we want to keep the current situation (everybody beats their head on >> >> the wall until they discover disabling burst mode "fixes" their panel, >> >> and keep an out of tree patch)? >> >> >> >> * Assuming the priority is getting a screen working (and not saving power >> >> on a black screen), would it make sense to apply this patch, and let >> >> people improve in the future by implementing link negotiation? >> >> >> >> Let's pretend for a moment this is a new driver being developed: would >> >> it be OK to have a basic working driver, without some power >> >> optimization >> >> features which can be added later on? The only valid answer to this >> >> question is obviously "yes". Doesn't the same principle apply here? If >> >> it doesn't, why? >> >> >> >> * What is the expected power saving with burst mode? >> >> >> >> I'm afraid I don't have precise numbers but I measured the total board >> >> consumption with or without burst mode (the former with a black screen >> >> but backlight enabled) and found no difference: exactly 12.74 W in both >> >> cases. >> >> >> >> Thanks for you rpatience in reading this. I hope it helps in finding a >> >> better solution. >> > >> > Rephrasing this a bit, is the discussion about dropping support for a >> > supported feature (burst mode) because users who suffer from the lack of >> > another feature (link negotiation) are not willing to spend time >> > implementing it, >> >> That's the question I have, more or less. I have no answer yet, I'm mostly >> trying to clarify the situation in the first place, for myself and anyone >> interested. >> >> Maybe it's worth pointing out that AFAICU any driver enabling burst mode is >> buggy because in lack of link negotiation it may work or not, based on pure >> luck. >> >> > and would prefer if users of burst mode were forced to >> > do the work instead ? That doesn't seem very fair to me. >> >> Link negotiation is not just "another feature" w.r.t. burts mode. It's a >> prerequisite for burst mode to work reliably. So hard-enabling burst mode >> was building a roof without solid walls (link negotiation). >> >> So I'm rephrasing your the question :) as: shouldn't users of burst mode be >> forced to implement link negotiation, since _they_ need it? > > It's quite annoying when both positions have compelling arguments :-) > > Has anyone analyzed what work would be needed to implement the link > negotiation ?
Marek did multiple attempts in the past, I think it's the best analysis available in public as of now. Following the link provided by Sean Nyekjaer in this thread on May 30 leads to them: * https://lore.kernel.org/dri-devel/[email protected]/ * https://lore.kernel.org/all/[email protected]/ > Dropping burst mode without any plan to support it will be > demotivating for some people. If asking for link negotiation support to > support non-burst mode is too much yak shaving, would researching a > technical plan be an acceptable middleground ? On my side I haven't gone into the details. But I'm not sure having a plan would help: Marek had one, but it didn't fly. Maybe Marek can comment on this? Luca -- Luca Ceresoli, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
