Hi, On Mon, Aug 10, 2026 at 12:56:57PM +0300, Laurent Pinchart wrote: > On Fri, Jul 17, 2026 at 03:25:30PM +0300, Laurent Pinchart wrote: > > On Tue, Jul 07, 2026 at 11:01:43AM +0300, Laurent Pinchart wrote: > > > On Tue, Jul 07, 2026 at 09:46:52AM +0200, Maxime Ripard wrote: > > > > On Mon, Jul 06, 2026 at 07:02:43PM +0300, Laurent Pinchart wrote: > > > > > On Mon, Jul 06, 2026 at 05:40:30PM +0200, Maxime Ripard wrote: > > > > > > On Mon, Jul 06, 2026 at 12:45:22PM +0300, Laurent Pinchart wrote: > > > > > > > On Mon, Jul 06, 2026 at 10:52:32AM +0200, Maxime Ripard wrote: > > > > > > > > On Mon, Jul 06, 2026 at 12:35:39AM +0300, Laurent Pinchart > > > > > > > > wrote: > > > > > > > > > Since DDC version 2, introduced in 1996, VGA monitors have > > > > > > > > > exposed EDID > > > > > > > > > data over an I2C bus. The bus is also used to detect the > > > > > > > > > presence of a > > > > > > > > > connected monitor by trying to read the EDID data. > > > > > > > > > > > > > > > > > > Some devices where the VGA display is integrated in the > > > > > > > > > device and > > > > > > > > > always connected do not connect the DDC pins. Some > > > > > > > > > development boards, > > > > > > > > > such as the Renesas M3N Salvator-XS, also do not connect the > > > > > > > > > DDC pins. > > > > > > > > > > > > > > > > > > To support those, add the ability to provide hardcoded EDID > > > > > > > > > data in the > > > > > > > > > device tree. This is mutually exclusive with specifying a DDC > > > > > > > > > bus, and > > > > > > > > > can only be done when the VGA display is guaranteed to be > > > > > > > > > always > > > > > > > > > connected. > > > > > > > > > > > > > > > > > > Signed-off-by: Laurent Pinchart > > > > > > > > > <[email protected]> > > > > > > > > > > > > > > > > This feels redundant with the EDID firmware loading mechanism > > > > > > > > we have in > > > > > > > > KMS already. It should at least be mentioned why we would need > > > > > > > > to set it > > > > > > > > in the device tree at all. > > > > > > > > > > > > > > Very good question. > > > > > > > > > > > > > > I assume you're talking about CONFIG_DRM_LOAD_EDID_FIRMWARE, as > > > > > > > the > > > > > > > debugfs override_edid feature isn't meant for production. If > > > > > > > there are > > > > > > > other relevant mechanisms I'm not aware of, please let me know. > > > > > > > > > > > > No, that's indeed what I meant. > > > > > > > > > > > > > In the use case at hand, the VGA display is an integral part of > > > > > > > the > > > > > > > device, the same way an LVDS or DSI panel would be. Using > > > > > > > CONFIG_DRM_LOAD_EDID_FIRMWARE, the manufacturer would need to set > > > > > > > the > > > > > > > drm.edid_firmware command line parameter, and provide EDID as a > > > > > > > file in > > > > > > > /lib/firmware/ (possibly in an initramfs). Beside the complexity, > > > > > > > and > > > > > > > the fact it won't be very friendly to people who run a different > > > > > > > userspace on the device, I think EDID counts in this case as > > > > > > > system > > > > > > > description, the same way we support specifying panel timings in > > > > > > > device > > > > > > > tree. > > > > > > > > > > > > It's *some* hardware description, but you have no idea whether it's > > > > > > actually the hardware you're running from. What would be in that > > > > > > EDID > > > > > > anyway? > > > > > > > > > > It would be the EDID corresponding to the connected display :-) I see > > > > > it > > > > > as identical to how a user of the simple panel driver is responsible > > > > > for > > > > > listing the timings corresponding to the hardware in the device tree. > > > > > > > > What's the expected user-experience for someone with that board > > > > unplugging a display and plugging another then? Update the device tree, > > > > rebuild it, rebuild the firmware and / or resign it and reboot? > > > > > > The whole point is that they can't easily unplug the display. This patch > > > series is meant to support devices where the display is an integral part > > > of the device. The user experience is expected to be the same as for a > > > user who will take a phone, remove the display panel, and plug a > > > different one. It's certainly possible, and they would have to update > > > the DT indeed. > > > > > > > That doesn't sound very practical, and even impossible in some cases. > > > > > > > > > > There's another alternative we've used several times already, in > > > > > > simple-bridge for example: just register any VESA mode up to a given > > > > > > resolution: > > > > > > https://elixir.bootlin.com/linux/v7.1.2/source/drivers/gpu/drm/bridge/simple-bridge.c#L66 > > > > > > > > > > I don't see how that would help. The goal is to provide the kernel > > > > > with > > > > > the data corresponding to the display integrated in the device. > > > > > > > > How is that not helping? Your problem is that you have a display you > > > > can't get the supported resolutions from. What I'm suggesting allows you > > > > to create all the possible resolutions for that display, allowing the > > > > user to use the display, which is what we did for multiple drivers with > > > > the same problem. > > > > > > How would the possible resolutions be provided to the kernel ? Isn't it > > > the whole point of EDID ?
We have multiple way to provide EDIDs to the kernel. In this thread alone, we discussed EDIDs (from DDC or the firmware loading mechanism) and drm_add_modes_noedid. And then you have panel-style fixed modes. > > > > You seem to be set on providing the EDID to the kernel, but it's a > > > > terrible idea to begin with, so I'd rather discuss alternatives. > > > > > > Could you elaborate on why it is a terrible idea ? > > > > I'd be happy to continue this discussion when you'll have time. Because I've yet to see a *good* EDID in the wild. They will need to be updated, and possibly changed by the users, which might or might not be possible depending on the system the user would use. Maxime
signature.asc
Description: PGP signature
