Skip to contentSkip to conversation

Kernel Improvements

DISCUSSION Michael 1 participant 2 messages Sep 26, 2023 1 message not in the archive

lore
Share & tools
lore

Michael

author
#1 lore raw
In reply to a message not in the archive

The philosophy "everything is a file" is a cornerstone of UNIX and Linux design, offering a unified interface for interacting with resources, be they hardware or software. Shifting to a paradigm where "everything is a database entry" or "active record" introduces a layer of abstraction that could compromise the simplicity and directness for which Linux is renowned.

1. **Performance Overhead**: Databases introduce latency and CPU overhead, affecting system responsiveness.

2. **Complexity**: Managing schemas, indices, and transactions adds unnecessary complexity for basic operations.

3. **Scalability**: File operations in Linux are optimized for performance and can scale horizontally. A database model may not offer the same scalability without significant engineering.

4. **Compatibility**: Existing applications rely on file-based APIs. A transition would necessitate widespread code changes, breaking backward compatibility.

5. **Principle of Least Astonishment**: Linux users expect a file-based paradigm. Changing this could alienate a significant portion of the community who rely on existing interfaces and behaviors.

6. **Transparency and Control**: Databases are often opaque, whereas the "everything is a file" design gives developers direct access to resources, offering more control.

While innovation is vital, it's crucial to weigh these concerns carefully against the benefits your proposal might bring.

_______________________________________________
Kernelnewbies mailing list
[email protected]
https://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

author
#2 lore raw

Re: Kernelnewbies Digest, Vol 154, Issue 14

In reply to a message not in the archive

Apologies on my last message, I didn't respond directly to Juro,

Sorry Juro!

Quoted 359 lines · On Mon, 25 Sept 2023 at 19:45, <[email protected]> wrote:

Send Kernelnewbies mailing list submissions to
[email protected]

To subscribe or unsubscribe via the World Wide Web, visit
https://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
or, via email, send a message with subject or body 'help' to
[email protected]

You can reach the person managing the list at
[email protected]

When replying, please edit your Subject line so it is more specific than "Re: Contents of Kernelnewbies digest..."

Today's Topics:

1. Kernel improvements. (Juro Dobr?k)
2. Re: [Question] dt bindings for BeagleConnect (Krzysztof Kozlowski)

----------------------------------------------------------------------

Message: 1
Date: Thu, 21 Sep 2023 04:47:17 +0200
From: Juro Dobr?k <[email protected]>
To: [email protected], [email protected]
Subject: Kernel improvements.
Message-ID:
<CAAG+mdToFpmG48SC24i9ErMK9JExMMUzRjePwH=
[email protected]>
Content-Type: text/plain; charset="utf-8"

In UNIX, everything is a file. I am here to make that a database. Yes, in new system paradigm, everything is database entry, or active record.

Kernel wide database should have defined bytecode of request possibilities, along with pointer reference to query data.

At some space in disk, there should be hardcoded in system kernel, that we can only write there. It should be proposed to store deep logs.

There should be guy for filtering logs, they should be more of serialized entries in database than a line of text. If errors are serialized data, we can check public errors repo for possible fix, with some data like new keywords provided by user, but fix from database will be applied after showing fix and affirmation by the user.

Short range packet history of network also should be on write only space. So we can raid out possible intruders and inspect packets. However with master permission we could send packet out in invisible format or recieve them like that, but it'll require finger prints, an absence of data will be also locked.

Now whole system is database, how do we pass data from database to process? Simply passing read start and read end and call to refill bucket function. We also pass metadata on how to decode such thing.

When it comes to metadata, we live in world of machine learning, therefore we can have something like system registry of data, and also software packages, for packages, they may come in bundles, install should be rollbackable by also setting context of packet when doing something else, so like we can rollback install completely, install multiple versions alongside itself, each packege should be working standalone or they can come in preconfigured bundles. Whole configs along with data should then be made installable. We should be able to define system instance that runs under supervisor, which is copy on write copy with only data access we selected.

That thing can be solved like there's hook for copying for both sides.

There should be manipulable path when writing to filesystem, consisting only of compression with LCache fitting dictionary and encryption. Maybe compress - encrypt will make it to hardware, with huge memory for dictionary and not using only words but also patterns.

Disks should have innovative raid approach, write in parallel and on no writing copy to mirror.

Virtual networks, tunnels and firewalls and redirection should be on one section of system, with GUI that enables configuration of those properties we need and also shows each process that uses network, so we can route that. Or we can preconfigure this config on new instance, in that case config is fixed with keyword, not the address itself.

With online repository for errors, software, data, models, whole system instances, we can also have this repositories peer to peer utilizing blockchain consensus algorythm.

System should have be possible configured 8n a way that it only boots with key USB Stick.

Storing of sensitive data on the system should be done in encrypted space on hardisk that runs on main master instance and you should must give permission for other instances to access that sensitive data for system for being able to access passwords.

Passwords entered in instance should be possibly better being exported with it in sone cases.

And I am talking about all passwords entered in webbrowser and also elsewhere, it could even remember archive passwords on permission, also passwords should be savable in instance config, but they could be make to insert of ow on beggining of config

Database of passwords can be stored in cloud firstly encrypted by symmetric encryption.

Each RAM's kernels specified segment would have rwx settable permissions individually for different users and processes.

Passwords should be related to master username, different ussers could have different password for each service.

Did already said that databasea can be passed in Kernell by passing reference to write start, end and refill bucket function? Along withmetadata of course.

Filesystem operations(cp, mv, rm) could have main proccess that kewps track of them, and lets you browse and rollback those.

Option to make driver instance(the main os) just from hardware already plugged in, not keeping unnecessary drivers in living memory, while installing instances of os drivers doesn't play role.

Always rewrite memory with pattern when allocating it to some process.

Let the kernel be king of which sort algorythm to use and therefore we should make sort an kernel operation, we pass metadata and data to kernel, and also write head which writes or 3xecutes sorted data.

Application for deconstructing XWindow or Wayland, also properly sorting there. But any window should be blurable or even have different computed layer on what's under that.

Window snapping. With normalenviroments like plasma, we should also have i3 like keyboard shortcuts to do operation on windows.

Also we could hardly benefit from having guy, in which we can with keyboard controll explore system.

Live editor of memory on system, ability to stop a process in middle of execution and examine it's memory {from master down we can examine even whole kernels}.

Graphic interface for users, groups.

Graphic interface to explore recent network packets.

Special filesystem, that store all text files in database with possible undos and redos, and other files are stored normally on disk.

Graphical program that let's us engrave simething into bits of HDD, like a picture from it's magnetic potential.

Snapshots of system lower instances, that can be stored on disk and played later.

Fingerprint authentification,
Face recognition that writes log about who sits on computer, like nobody without password cannot replace you in computer chair and continue your work.

Optimize hashmaps in kernel. Like combination of checksum and skip table is good.

Brainstorm even weirdest optimizations over db you've ever heard.

Compilers will have they own package system that'll automatically install library for the compilation for header from public repository.

If library haven't got it, it sort by kernel's sorting function, compiler adds binary search when applicable, optimizes hashmaps based on benchmarked performance in specific application.

Graphic editor of bootloader menu, splash screen documentation, linux command tip on loading showed on screen.

File system for images, which use AI to categorize them and sort them.

Best of all 'script' javascript derivative for event driven asynchronous operation scripting. With all the system entries of services, possibility to change config and reload it. Having ORM system for kernel database. Having also good date and time functionality and full power to do API calls.

If you are registered at webservice, you can get your own url, which may act as webhook for this API service.

Kernel will also have some conditional loops, it's graph, along with whole instance would be parsed trought looking for malicious behaviour.

Script should be also capable to control web browser to some extent.

Also gui to browse and change all the data provided to instance, including strings, fonts, patterns, barelly anythint.

Online repository for live wallpapers out of generative art.

API SDK repository. Yea, script will have SDK's ready, that's why yoy should develop open standards for information transfer and requesting. Binnary with metadata about format is best way to go. Open API standards, datatype standards and their version.

Possible two factor authentification if computer is connected to internet or with bluetooth.

Possible authentification for wheel group by common access card along with fingerprint and password.

When reversing array itetation, just replace op code in instruction stack with one that does reduce pointer and returns lenght-current as position.

Analyze executed binaries by asynchronoys acyclic graphs with cyclic groups, in GUI, to know they are going to do something that shouldn't be there. They can be verified once saved checksum, and if no much change in checksum, then don't commit whole analysis again. Some points are going to be different, but it's whole picture that matters.

Also, have I mentioned, you can have digital well beings apps down from this everything is database thing, and with that data you can have like automatic selection of music, that gets along with your current situation.

Note-taking app, that support hypertext, keywords, autographing, and 'script' and database, would be really useful too.

And you could pick, which or what kind of data you will sync to your phone.

And when it comes to your phone, it can have pre-fetched content for offline browsing.

And you can sync it with bluetooth/wifi/VPN so you can read reports from computer on the way to work.

Also audiofiles should be database, vectorized for simillarity search, stored on specific file system, along with metadata, and if they are fresh maybe they are still seedbacking to community trought peer to peer sharing. Preffered format:FLAC with cache size of dictionary, prefferably big common dictionary, and then separate derivatives of compresion core.

Part of hard disk non encrypted bootable system, that boots only to ram and have very restricted functionallity.

Also for 16GB RAM swap with hibernate you need 32GB's of RAM.

Filesystem with rollbacks possible. Even rollback with persist something is programatically possible.

Kernel wide memento for undo button that carries inverse of any function done, or original of cow's before change on part and undoes even that you copy pasted something.

That's All Folks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <
http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20230921/93a01ffd/attachment.html

Quoted 1 line

------------------------------

Message: 2
Date: Sat, 23 Sep 2023 18:08:30 +0200
From: Krzysztof Kozlowski <[email protected]>
To: Ayush Singh <[email protected]>, [email protected]
Cc: [email protected], [email protected],
[email protected], [email protected]
Subject: Re: [Question] dt bindings for BeagleConnect
Message-ID: <[email protected]>
Content-Type: text/plain; charset=UTF-8

Quoted 49 lines · Ayush Singh

Hello everyone, I am working on writing a BeagleConnect driver for Beagleplay board. Let me first go over some terminology:

BeagleConnect is both a technology concept and a line of board designs that implement the technology. Born from Greybus, a mechanism for dynamically extending a Linux system with embedded peripherals, BeagleConnect adds two key elements: a 6LoWPAN transport and mikroBUS manifests. The 6LoWPAN transport provides for arbitrary connections, including the IEEE802.15.4g long-range wireless transport supported between BeaglePlay and BeagleConnect Freedom (the first BeagleConnect board design). The mikroBUS manifests provide for rapid prototyping and low-node-count production with sensor boards following the mikroBUS freely-licensable embedded bus standard, such that existing Linux drivers can be loaded upon Greybus discovery of the nodes.

BeaglePlay consists of a main AM62 processor and a CC1352 co-processor which is responsible for providing 6LoWPAN support along with Greybus node discovery. The AM62 processor and CC1352 are internally connected over UART. The CC1352 coprocessor is supposed to run a specific firmware as a part BeagleConnect setup. It looks as follows:

AM62 (Linux Driver) <--UART--> CC1352 (Zephyr Firmware) <--6LoWPAN--> BeagleConnect Freedom

I need a way to access this bridge UART. The most straightforward method is adding a compatible property to the device tree of this UART. But I am not sure how to go about adding a DT binding for it that can be merged upstream.

Here are a few comments I have encountered:

1. DT bindings need to be hardware

I am not sure which hardware the bindings need to target in my case. Should the bindings be serial in nature, since I need to use the UART device? Or they should be about SoC since CC1352 is the device I am actually communicating with? Or maybe there is a 3rd category for an SoC running a specialized firmware for a technology/protocol?

2. DON'T create nodes just for the sake of instantiating drivers.

I guess this means that adding a child node just to add a compatible property is not allowed? For some reason, the driver probe is not called if I simply try to override the top level compatible property of the serial device. But maybe that is just an issue with my approach?

3. DO attempt to make bindings complete even if a driver doesn't support some features.

Quoted 6 lines

This is only relevant if the answer to 1) is the SoC. Since the CC1352 device cannot be directly controlled by, the capability of this device is defined by the firmware running on top of it rather than the hardware. So I am not quite sure what a complete binding for such a device even mean. The only property I can make required would be the compatible property and everything else will be optional I guess?

Referring to some comments without the context - patch and the comments given - makes any discussion difficult. We do not work like this in upstream communities. Please respond inline, not top posting, to the comments you received.

Quoted 5 lines

I understand that strict guidelines are required since once a property is added to the Kernel device tree, it will almost be impossible to remove without breaking compatibility. So I would like some guidance or maybe some similar devices that are already a part of the kernel which I can look to for guidance.

There are plenty of other serial-attached MCUs. Just look for "mcu {" (for serial) or mcu@ (for other buses).

Best regards,
Krzysztof

------------------------------

Subject: Digest Footer

_______________________________________________
Kernelnewbies mailing list
[email protected]
https://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

Quoted 5 lines

------------------------------

End of Kernelnewbies Digest, Vol 154, Issue 14
**********************************************

_______________________________________________
Kernelnewbies mailing list
[email protected]
https://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies