Slashdot Mirror
Kmscon Project Seeks To Replace Linux Virtual Terminal
An anonymous reader writes "Phoronix reports on the progress of kmscon, David Herrmann's virtual console project that aims to supersede the Linux kernel's virtual terminal. kmscon takes advantage of modern Linux features such as kernel mode setting, direct rendering, and udev to provide hardware-accelerated rendering, full internationalization, monitor hot-plugging, and proper multi-seat support. A recent blog post by Herrmann addresses some of his frequently heard questions and criticisms about the kmscon project."
OTOH, please make it configurable to have a default of other than 80x24. Back in the days of 9" terminals, 80x24 is great. Today, when I have a 20" monitor attached, capable of displaying 1600x1280, please let me display more lines and columns.
you're aware that the BIOS and the linux boot screen are currently using HARDWARE ACCELERATED RENDERING for SIMPLE TEXT?
They call it "text mode"
Because monitors aren't VGA anymore. We now have small laptops with 2880x1880 screens built in that are plugged into multiple monitors. If you want to drive that at a snappy response, or if you want to select a primary monitor and switch between them, you should take advantage of the hardware.
Heck, a simple one: I have a laptop with a dead built-in monitor. I use an external monitor, which works fine with X, but the console is on the internal screen.
"$30 for the One True Ring. $10 each additional ring!" -- JRR "Bob" Tolkien
Dumb question, but what do they mean by proper multi-seat support?
http://www.x.org/wiki/Development/Documentation/Multiseat
> Multiseat is a model of computing that supports multiple local users using their sessions in a totally independent way. This looks quite similar to the old mainframe computer model, but with the "terminals" connected directly to single PC box. There's a lot of people that use and sell multiseat Linux systems due its low cost which qualifies it as a wonderful "techno-social" model of computing.
Have you ever used 80x24 on a 22" monitor with 1680x1050 native res? The letters are so huge as to be unreadable. Ubuntu et al handle it correctly by letting the X driver do KMS to the native res, which carries over to the console.
I'd be happy with defaulting to whatever the video hardware can handle and then having an easy way to configure it for other resolutions.
Hail Eris, full of mischief...
E pluribus sanguinem
Framebuffer consoles are (or were), in some cases, surprisingly slow. Huge amounts of output at least used to overwhelm them, only catching up if you swapped back and forth between VTs -- this didn't happen in non-framebuffer consoles or in terminal emulators in X.
I think it has improved significantly over the past few years, but that's probably partly from already working some hardware-accelerated rendering into some of the framebuffer console drivers. Though I'm not sure why they're not just working on improving the framebuffer drivers further -- the project seems to aim to put every damn piece into userspace, which seems like a terrible idea for something so essential as the basic console. If you break one of the things it depends on (such as xkb -- yes, really, a piece of X11 for a console system), then you lose any ability to interact with your system, or even view the boot output.
I'm really not sure who it's aimed at. If you break your install and you don't have an old kernel-based VT fallback, you're screwed. If you don't break it and it works fine, what benefit is it? You'll more than likely just be starting up X11 momentarily anyway.
no it is worse
text mode performance is unimportant to graphics vendors, they do the bare minimum to make it work, they don't lift a finger to make it faster.
nursing the dynamic data flow to potentially slow I/O devices (like USB displays) is a job for user space. Data is queued to be sent to the device for display but it might be modified while in the queue by graphics events. You don't want to put this stuff in the kernel.
you are actually making good arguments to remove this functionality from the kernel
you don't lose the ability to interact with your system or view the boot output, if you have SSH enabled, or if you dump the log messages to the serial port.
embedded people have been working on screenless systems for years, they are not necessary for interacting with the computer. There are plenty of other ways.
If you read the article (or was it the blog post?), you'll find answers.
First, kmscon has only one dependency - libudev. It can be compiled to use Pango for font rendering, or EGL for hardware-accelerated rendering, but those are optional. X11 is never used.
Second, the rationale for putting this in userspace was mainly for internationalization. In particular, some character tables (for Asian languages, especially) can be rather large. If it was in the kernel, that memory could not be swapped out. In userspace, you can.
Third, this seems aimed at regular console users, not primarily-X-users-that-use-a-terminal. There are plenty of good X11 terminal emulators (and just as many bad ones). This is aimed at people who don't need (or at least, don't want) X, but still want to use all the features of modern hardware. It also claims to interact well with both the kernel VT system, and with X - you can keep an X session on one virtual screen, keep the kernel terminal on another (for those few cases where it is needed, like kernel error messages), and put kmscon on the rest.
I will probably try this out, because I at least used to fit into that group of regular console users. I have on several occasions run out of virtual terminals. I'm not so much of one anymore, but maybe this will get me back into it.
In principle, or in fact?
I've had a multi-seat linux system for many years
Fedora 17 documentation:
For the first time Fedora 17 provides completely automatic multi-seat configuration.
To use this feature, simply plug in a USB dock such as the Plugable Docking Station, with a monitor, mouse and keyboard, and a new desktop appears.
I would be delirious with joy.
well I guess I just made your day
Thats already possible..
cat image.jpg
I don't even see the code. All I see is blonde, brunette, redhead..
Totally unrelated... except to show how relative a term "slow" can be.
This reminds me of reading the docs for a Perl DNS module once when I was writting an internal app that needed to do a lot of DNS lookups. The docs said it was "pure perl" and "slow". So, if the docs said it was slow....I figured I would use the system resolver instead.
I wrote my app, came to test it.... and DOSd my own system as my program began trying to slam the system resolver with 6 parallel workers (batch resolving IPs based on logs)...each of which had to open several files (nsswitch.conf, hosts, hosts.allow, resolv.conf.....and I think a few others if memory servers) for EACH LOOKUP.... my poor system was no match for it (this was back in the single core days)
I switched to the "slow" dns module, and maybe it was slow by some standards, but compared to the system resolver...it was lightweight and fast.
"Slow" is always a relative term.
"I opened my eyes, and everything went dark again"
> and how many real laptops have serial ports these days?
Don't remind me. Two weeks ago, I spent the better part of a day troubleshooting a nonexistent bug in an embedded hardware project that was caused by a malfunctioning USB virtual fucking serial port on the computer I was using to view the debugging info I was outputting to the MCU's serial port.
Now I know why so many guys have built their own serial terminals with a 4x20 LCD and a microcontroller. It's the only way left to reliably and infallably render serial input straight to a display without risking stupid errors caused by Windows or its USB subsystem.
When you're dealing with bare metal embedded hardware, USB anywhere along the signal chain really sucks.
this has nothing to do with gfx vendors by is piss-poor text mode EMULATION inside the kernel. your "text mode" console is actually graphics mode and emulated to be text by the kernel splatting out pixels that look like text chars - just like x does (and x apps).
it's piss-poor at this because it's poor unoptimized code that does some royally stupid things.
1. "blit" (copy region) to scroll 1 line at a time... and this ends up being almost always done by your cpu.
2. this blit will be done by cpu to/from video memory which often is on a discrete video card across a pcie bus or such
3. writes across this bus are fast, but reads can be in the order of just 10mb/sec. it's piss-poor at reads.
4. the kernel text console doesn't try and play catch-up and scroll multiple lines at a time. it scrolls 1 line at a time thus making the overhead of copying enforced per line scrolled, as opposed to internally skipping a bunch of lines and drawing 2, 5, 20 new lines and copying up only those that are not new.
it's entirely poorly performing because of poor kernel code. how poorly performing? well.
i recently decided to write a terminal emulator (for x). of course i went and used a toolkit i knew to help, and it so happens this toolkit can automatically work in the linux framebuffer just like it does in x11. it can use opengl, but doesn't need it and works fine on a dumb fbcon text console. so i now have the EXACT same terminal in bot x11 and fbcon. hell it even manages to provide me a mouse pointer in fbcon as it detects it's in fb and the toolkit emulates one itself. it looks the same pixel-for-pixel too. so i did some performance comparisons of it in fbcon vs kernel text console. it clocks in at being 55 TIMES faster than the kernel. 55 TIMES. that's how poor the kernel is. going through scads of toolkit layers and abstractions, code to alpha blend the text on top of backgrounds, blend and scale lighting overlays and more... and its still clocking in at 55 TIMES faster. (it happens to run at about the same speed in x11 - on the same machine).
it's not fast because it's amazing code or i'm a genius at all - it's fast because it just is NOT being royally stupid, which is what the kernel text console emulation is.
personally i was quite happy to know my terminal works both in x and in fb.. so if'm i'm ever stuck there.. i can use it and not be 100% grumpy that my text console is piss-slow anymore.
--------------- Codito, ergo sum - "I code, therefore I am" --------------------