I am writing a book entitled The Linux Memory Manager.

It's intended to be a thorough overview of the memory management subsystem combining a walk through the code with a description of the key concepts as well as a chapter on practical kernel memory management.

EDIT: The first draft is complete :) Now editing the book.

It targets linux v6 with a target release date of mid-late 2024.

Check out https://linuxmemory.org/ and this account for details of my progress.

@ikonoklast well... not exactly :-) ATM, two identical sticks, one of them at my parents place. and a passphrase

Uh oh, cannot sign tags because of expiration. Need dig my cert key from the vault and bump it up... #openpgp

@mikael Thanks I bookmarked this for reference if any new issues pop up ;-)

@seve_py I was thinking the same actually today ;-)

@Zoarial94 Thanks for the great explanation of this topic! Appreciate this.

@kernellogger One more revision coming soon based on Linus' and James' feedback. I added you and Linus to the loop because the overall diff is not small, and this is coming quite late. Unfortunately, neither when bugs arrive and when they have been fully analyzed can be planned beforehand...

@Zoarial94 Yea, so is the difference that the other sites use U2F and Google uses FIDO2? I have to admit that I did not know that they are different thing so this cleared some details (i.e. according to Internet U2F is second factor authentication whatever that means).

Almost without noticing I’ve gradually migrated some of my passwords to pass:

❯ pass web
web
├── bitstamp.net
├── bitwig.com
├── element.kapsi.fi
├── fabfilter.com
│   ├── license
│   └── password
├── firefox.com
├── gitlab.com
├── google.com
├── icloud.com
│   ├── password
│   └── recovery
├── pypi.org
│   └── recovery
├── steampowered.com
├── storj.io
│   ├── password
│   ├── projects
│   │   └── my-cloud
│   └── recovery-codes
└── tuni.fi

Nice thing is that you can store hierarchically also other data than passwords to leaves. storj.io is a great example of this use.

I drew this 12 months ago, as LLM code assistants were just becoming available.

Has it held up?

Most #site’s ask #password + #FIDO2 but others satisfy with only FIDO2.

E.g. #Google #account opens up without password, if the #YubiKey is plugged in.

I’m just wondering why this difference exists…

@jani, @timojyrinki, this is how I've now sorted out my earlier rant. No idea what I'm doing but I don't neither care that much ;-)

I don’t really understand the cause and effect here but I finally got rid of “GPU HANG” issue with #i915, i.e.

Jul 02 21:20:50 suppilovahvero kernel: i915 0000:03:00.0: [drm] GPU HANG: ecode 12:1:84dfd7f7, in ChaosGate.exe >
Jul 02 21:20:50 suppilovahvero kernel: i915 0000:03:00.0: [drm] ChaosGate.exe[40919] context reset due to GPU ha>

I did the following:

1. Teared my machine into parts.
2. Cleaned up everything
3. Put some new cooling paste.
4. Bought and switched a new PSU.
5. Created /etc/modprobe.d/i915.conf and run sudo dracut --force (in Fedora).

The contents of /etc/modprobe.d/i915.conf are the following:

options i915 enable_guc=3
options i915 enable_fbc=1

I put these there because I found multiple forum posts etc. where this was the suggested solution but obviously I have zero idea what they’re doing 🤷 I’ve played Chaos Gate as a test without hangs, and it used to hang within just few minutes.

It is still a bit weird that #Fedora 40 with its default settings gives such crap results on Intel Arc A770.

#linux #kernel #intel #graphics

@kernellogger Yep, thank you, coming from the Holiday to fix this ;-)

I released v3. The previous version was already fully working and confirmed by Stefan (on the basis that he got the expected log message).

https://lore.kernel.org/linux-integrity/20240704085708.661142-1-jarkko@kernel.org/T/#t

This version I'm likely going to put to a pull request for -rc7 today even without reviewed/tested-by tags because null deference needs to be addressed, there's no excuses not to do it. I'll hold for some hours but then I'll put the PR out.

On the other hand, we can wait "forever" for tpm_ibmvtpm have the feature fully enabled. That endangers the stability of total zero systems in the wild.

@kernellogger v2 with baseline fixed into mainline tip: https://lore.kernel.org/linux-integrity/20240703182453.1580888-1-jarkko@kernel.org/T/#t

@kernellogger I.e. the symptom was correct but the proposed fix was not, and this is the correct way to fix the actual bug.

@kernellogger directly related to https://lore.kernel.org/linux-integrity/55e8331d-4682-40df-9a1b-8a08dc5f6409@leemhuis.info/

Critical fixes for TPM2 authenticated sessions, which I still hope landing to 6.10:

https://lore.kernel.org/linux-integrity/20240703170815.1494625-1-jarkko@kernel.org/

#tpm #linux #kernel

I like to start a bug fix by:

$ git commit -a -s --allow-empty -m "Fixes: 1085b8276bb4 (\"tpm: Add the rest of the session HMAC API\")"
[auth-null d5d3411b52f7] Fixes: 1085b8276bb4 ("tpm: Add the rest of the session HMAC API")

$ git show
commit d5d3411b52f741cb16cfb7180d24f454e97b4570 (HEAD -> auth-null)
Author: Jarkko Sakkinen <jarkko@kernel.org>
Date:   Wed Jul 3 18:47:46 2024 +0300

    Fixes: 1085b8276bb4 ("tpm: Add the rest of the session HMAC API")
    
    Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>

So that I get the fixes-line backed up fast :-) Then I just amend the meat.

For generating the fixes line I have:

git-fixes() {
  git --no-pager log --format='Fixes: %h ("%s")' --abbrev=12 -1 $1;
}

#git #linux #kernel

I like this site: https://commit.style/