4 ms·

write-edid – writes custom EDID data into HDMI adapter over I2C on Linux

56 points by pabs3 4y ago

userbinator 4y ago
Presumably this is for HDMI-VGA adapters, but it assumes that the adapter itself has an EDID EEPROM instead of passing the data through (otherwise you'd be writing to whatever monitor it's connected to), and that the EDID EEPROM isn't write-protected.
- toast0 4y ago
  I wonder if this works with the HDMI passthroughs that force the EDID. It'd be nice to have a customizable one of those that's not a lot of money or size.
tverbeure 4y ago
It’s become more common to have the EDID stored in an I2C device that is part of the main silicon (because HDMI requires I2C devices anyway for HDCP and HDMI 2.0 SCDC support), but well designed monitors with an I2C PROM for the EDID have the write-protect pin of the PROM asserted to prevent accidental mishaps by tools like this… or by intentionally malicious tools.
russdill 4y ago
I realize it's just an EDID EEPROM so if it gets hosed, it's not a big deal, but it still gives me a not good feeling to have a bash script blindly programming it. It's also possible to do this by probing with the appropriate i2c EEPROM driver and using that to dump the EEPROM, program it, and verify the contents.
PaulDavisThe1st 4y ago
You don't have to do this. You can tell the kernel to read the EDID data from a file instead of from the device. I have two identical ASUS monitors; the EDID on one of them failed, so my boot sequence tells the kernel to fetch the EDID from a file I created from the working one.
- pabs3 4y ago
  That means the monitor is still broken if you move it to a new system, or even just reinstall your system and forget to copy the file over. Better to update the monitor itself. Also good to keep a backup in case the issue happens again.
  - washadjeffmad 4y ago
    This doesn't help if the cause is that the EEPROM has failed or is no longer writable, a not uncommon cause of EDID errors.
- toast0 4y ago
  Adding to the sibling, what if you want proper edid during boot?
  - Aissen 4y ago
    During boot, on a machine you never touched. This is absolutely useful.
- gnull 4y ago
  Also you need to identify the monitor somehow for kernel to do what you propose. If your monitor setup is not static, say, you connect monitors in different order or through different ports each time, you won't be able to tell which edid should go for which monitor. (Unless you're willing to script some heuristics to guess that.)
- detaro 4y ago
  Unless I misunderstand something, writing it this way afterwards also works on other platforms, which are not as flexible in that regard.
- eulenteufel 4y ago
  Because I have a display whith broken edid I tried setting the edid with the kernel command line, but unfortunately that did not work for me. It might be because I use a nvidia graphics card, but I'm not sure.
oktwtf 4y ago
I wonder if I can use this to finally get my Qnix 2710 monitor to finally show up without resorting to xorg.conf and a EDID dump.[0] [0]: https://wiki.archlinux.org/title/Qnix_QX2710 https://wiki.archlinux.org/title/Qnix_QX2710
pabs3 4y ago
There is an article on the Debian wiki about repairing corrupted EDIDs too: https://wiki.debian.org/RepairEDID https://wiki.debian.org/RepairEDID
IshKebab 4y ago
How many monitors support writable EDIDs?
cshein 4y ago
I've been having a weird EDID issue for a few months now. Ever since using a new laptop my external monitor's EDID will quite reliably corrupt when I try to (re)attach it using HDMI while the system is running. It's always the same, singular byte in the EDID that changes - even though the EEPROM appears to be write-protected. In fact, every time this happens I have to go into the monitor's hidden service menu and disable write-protection in order to write back the correct byte over I2C. This thankfully never happens if the monitor is connected before the laptop boots. But I have no idea if this is fundamentally a problem with the laptop or the monitor since my previous computer never triggered this issue. I'm glad EDID tools and resources exist for Linux. If I ran Windows exclusively I'm pretty sure I wouldn't ever have figured out how to write to I2C.
- userbinator 4y ago
  That doesn't sound like corruption, but something deliberate. What's the interpretation of that byte in the data? I wonder if it's some "feature" of the firmware.
  - cshein 4y ago
    It's always the byte at chip-address 0x50, data-address 0x20 that is overriden with the value 0x00. Since this byte _should_ have value 0x11 the EDID checksum doesn't match the data anymore. So then my computers reject the entire EDID and either output nothing or only output at the lowest resolution possible. This byte is part of the chroma "blue_y" information. Clearly, this byte shouldn't normally be written to when attaching HDMI. Reviewing my investigation notes from a year ago when I enabled I2C tracing, the issue is somehow caused by Xorg writing 0x00 to chip-address 0x54, data-address 0x20 when the monitor is connected. Although, I'm pretty sure I wasn't able to manually write to this location using i2cset, possibly due to the write-protection mechanism - my memory's hazy here. This is where I'd have to continue investigating if I ever get around to it again. But I'd expect this to not even be possible - the write-protection _should_ block a write through the wrong address.
rsolva 4y ago
I found a way to fetch the monitors EDID, modify it and load the file during boot, but this is port, not monitor, specific. I wrote up the process on my blog: http://rubensolvang.com/2022/custom-monitor-resolution-on-wayland/ http://rubensolvang.com/2022/custom-monitor-resolution-on-wa... If anyone knows how to apply this to a specific monitor, please tell me how!
fire 4y ago
I would love to do this for my TV honestly. Bit of a tangent rant here - The LG77CX has a native res of 3840x2160, but the damned display's edid lists 4096x2160 as the "maximum", even though it causes skewing of the image on the display. If you want to use DSR on nvidia cards, it bases it to the highest res, NOT the native res, which... causes skewing. If you then manually override the EDID data for the display via EITHER the windows registry or an unsigned monitor profile ( which requires you to disable driver signing overall! ), the nvidia control panel disables basically all configurable functionality; presumably so that you can't force DSR with a faked display resolution, because it used to work in older driver versions ( which are now too old to support 3k or 4k series cards )

56 points by pabs3 4y ago

Presumably this is for HDMI-VGA adapters, but it assumes that the adapter itself has an EDID EEPROM instead of passing the data through (otherwise you'd be writing to whatever monitor it's connected to), and that the EDID EEPROM isn't write-protected.

toast0 4y ago
I wonder if this works with the HDMI passthroughs that force the EDID. It'd be nice to have a customizable one of those that's not a lot of money or size.

tverbeure 4y ago

It’s become more common to have the EDID stored in an I2C device that is part of the main silicon (because HDMI requires I2C devices anyway for HDCP and HDMI 2.0 SCDC support), but well designed monitors with an I2C PROM for the EDID have the write-protect pin of the PROM asserted to prevent accidental mishaps by tools like this… or by intentionally malicious tools.

russdill 4y ago

I realize it's just an EDID EEPROM so if it gets hosed, it's not a big deal, but it still gives me a not good feeling to have a bash script blindly programming it. It's also possible to do this by probing with the appropriate i2c EEPROM driver and using that to dump the EEPROM, program it, and verify the contents.

PaulDavisThe1st 4y ago

You don't have to do this. You can tell the kernel to read the EDID data from a file instead of from the device. I have two identical ASUS monitors; the EDID on one of them failed, so my boot sequence tells the kernel to fetch the EDID from a file I created from the working one.

pabs3 4y ago
That means the monitor is still broken if you move it to a new system, or even just reinstall your system and forget to copy the file over. Better to update the monitor itself. Also good to keep a backup in case the issue happens again.
- washadjeffmad 4y ago
  This doesn't help if the cause is that the EEPROM has failed or is no longer writable, a not uncommon cause of EDID errors.
toast0 4y ago
Adding to the sibling, what if you want proper edid during boot?
- Aissen 4y ago
  During boot, on a machine you never touched. This is absolutely useful.
gnull 4y ago
Also you need to identify the monitor somehow for kernel to do what you propose. If your monitor setup is not static, say, you connect monitors in different order or through different ports each time, you won't be able to tell which edid should go for which monitor. (Unless you're willing to script some heuristics to guess that.)
detaro 4y ago
Unless I misunderstand something, writing it this way afterwards also works on other platforms, which are not as flexible in that regard.
eulenteufel 4y ago
Because I have a display whith broken edid I tried setting the edid with the kernel command line, but unfortunately that did not work for me. It might be because I use a nvidia graphics card, but I'm not sure.

oktwtf 4y ago

I wonder if I can use this to finally get my Qnix 2710 monitor to finally show up without resorting to xorg.conf and a EDID dump.[0] [0]: https://wiki.archlinux.org/title/Qnix_QX2710 https://wiki.archlinux.org/title/Qnix_QX2710

pabs3 4y ago

There is an article on the Debian wiki about repairing corrupted EDIDs too: https://wiki.debian.org/RepairEDID https://wiki.debian.org/RepairEDID

IshKebab 4y ago

How many monitors support writable EDIDs?

cshein 4y ago

I've been having a weird EDID issue for a few months now. Ever since using a new laptop my external monitor's EDID will quite reliably corrupt when I try to (re)attach it using HDMI while the system is running. It's always the same, singular byte in the EDID that changes - even though the EEPROM appears to be write-protected. In fact, every time this happens I have to go into the monitor's hidden service menu and disable write-protection in order to write back the correct byte over I2C. This thankfully never happens if the monitor is connected before the laptop boots. But I have no idea if this is fundamentally a problem with the laptop or the monitor since my previous computer never triggered this issue. I'm glad EDID tools and resources exist for Linux. If I ran Windows exclusively I'm pretty sure I wouldn't ever have figured out how to write to I2C.

userbinator 4y ago
That doesn't sound like corruption, but something deliberate. What's the interpretation of that byte in the data? I wonder if it's some "feature" of the firmware.
- cshein 4y ago
  It's always the byte at chip-address 0x50, data-address 0x20 that is overriden with the value 0x00. Since this byte _should_ have value 0x11 the EDID checksum doesn't match the data anymore. So then my computers reject the entire EDID and either output nothing or only output at the lowest resolution possible. This byte is part of the chroma "blue_y" information. Clearly, this byte shouldn't normally be written to when attaching HDMI. Reviewing my investigation notes from a year ago when I enabled I2C tracing, the issue is somehow caused by Xorg writing 0x00 to chip-address 0x54, data-address 0x20 when the monitor is connected. Although, I'm pretty sure I wasn't able to manually write to this location using i2cset, possibly due to the write-protection mechanism - my memory's hazy here. This is where I'd have to continue investigating if I ever get around to it again. But I'd expect this to not even be possible - the write-protection _should_ block a write through the wrong address.

rsolva 4y ago

I found a way to fetch the monitors EDID, modify it and load the file during boot, but this is port, not monitor, specific. I wrote up the process on my blog: http://rubensolvang.com/2022/custom-monitor-resolution-on-wayland/ http://rubensolvang.com/2022/custom-monitor-resolution-on-wa... If anyone knows how to apply this to a specific monitor, please tell me how!

fire 4y ago

I would love to do this for my TV honestly. Bit of a tangent rant here - The LG77CX has a native res of 3840x2160, but the damned display's edid lists 4096x2160 as the "maximum", even though it causes skewing of the image on the display. If you want to use DSR on nvidia cards, it bases it to the highest res, NOT the native res, which... causes skewing. If you then manually override the EDID data for the display via EITHER the windows registry or an unsigned monitor profile ( which requires you to disable driver signing overall! ), the nvidia control panel disables basically all configurable functionality; presumably so that you can't force DSR with a faked display resolution, because it used to work in older driver versions ( which are now too old to support 3k or 4k series cards )