Greg Kroah-Hartman

About

Posts

Post has attachment

Public

Latest upstream kernel status for meltdown and spectre from my point of view.

That's enough writing for today, back to patches...

Meltdown and Spectre Linux kernel status - Linux Kernel Monkey Log

By now, everyone knows that something “big” just got announced regarding computer security. Heck, when the Daily Mail does a report on it …

kroah.com

Show all 11 comments

Vasilis Lourdas: Ok, first of all, thank you all for this hard work. It surely mustn't have been easy to deploy a fix like that in such a short time frame.

Besides this, I suppose that Linux is the first OS to widely offer a patch to fix these issues, am I right? So, another thumb down for closed OSes.

I'm running Gentoo and I was reluctant to upgrade from 4.14.10 to 4.14.11 and later 4.14.12, because of all this publicity about reduced performance with the KPTI patches. To be honest, as a desktop (mostly) user, I haven't felt any performance degradation. On the other hand, my vps does feel slower (when comparing the build times for a few packages, before and after the 4.14.11 upgrade). Oh well.

Thanks!

Henrik Danielsson: I'd be happy to contribute a beverage of choice if you ever pass by up north near Örnsköldsvik too. Thank you all, not "only" for the patches but the great writeups, I've learned so much from this.

Dan Williams: Thanks for all your, and other's, hard work!

Add a comment...

Greg Kroah-Hartman

Public

`nproc`? Didn't know that was a thing, I've been doing `grep -c "processor" /proc/cpuinfo` for so long, time to go update a bunch of build scripts of mine...

Show all 16 comments

Ricardo J. Barberis: +David Alan Gilbert probably what +Kris Shannon said

Willy Tarreau: I used to use the "grep" approach. Regarding getconf, what does it return if a CPU is turned offline while the process is running? Also tracing nproc indeed shows a sched_getaffinity() so I guess it passes it through popcount(). Again I have no idea what it does if a CPU is turned offline then online. But for most use cases it could be useful.

Lilian Moraru: Everybody uses nproc :)

Add a comment...

Post has attachment

Greg Kroah-Hartman

Public

Took me long enough to get my act together, don't know why I didn't do this years ago...

Show all 7 comments

Chris Ahlstrom: +Bert Vermeulen You're right, the GPL was nothing worthwhile, nor is the support for dealing with GPL violators, or the information about the value of Free Software (which is now present in an enormous number of commercial products). C'mon man!

Bert Vermeulen: +Chris Ahlstrom The FSFE has done none of that. They are not the FSF.

They are just one more lobby group riding the gravy train of EU institutions.

Mark K Post: +Bert Vermeulen BS. They do quite a lot of that. I've seen it first hand. You are right that they are not the FSF. But they started out as that, and then amicably separated themselves from the FSF.

Add a comment...

Post has attachment

Greg Kroah-Hartman

Public

Best quote of the article: "We didn’t make a single kernel modification to get this to boot either."

I Booted Stock Android Oreo on the Huawei Mate 9 thanks to Project Treble

xda-developers.com

Show all 6 comments

Morgaine Fowle (de la faye): Good news bad news right? Because now no one is ever ever ever ever going to update the kernel version. As opposed to the status quo today, of almost never upgrading the kernel version. :/

Greg Kroah-Hartman: +Morgaine Fowle heh, no, not at all. For it should be obvious (but people seem to be totally ignoring it for some reason), that if you can replace the upper layer with a totally new one with everything still working, you could also replace the lower layer (i.e. the kernel) with a new one and still have the upper layers all work just fine.

In other words, it is now easier to update the kernel than ever before now that the "tight bond" is broken...

/me cackles evilly

Martin Peres: +Greg Kroah-Hartman Indeed! I was super excited to talk about this with Google engineers at XDC 2017. It was tying in quite well with the talk I gave about how to support products using upstream code.

We'll see what we can do inside Intel with this new capability. Hopefully we'll get the same sort of CI as we currently have, but for Android, but I wonder what kind of technical hurdles we will have to face :)

Add a comment...

Post has attachment

Greg Kroah-Hartman

Public

Which version of GPL version 2 are you using for your project? There's so many to choose from!

Great bunch of research here from Philippe Ombredanne of nexB.

Re: [patch V2 02/11] LICENSES: Add the GPL 2.0 license

lkml.org

Add a comment...

Post has shared content

Greg Kroah-Hartman

Public

Originally shared by David Miller (dokebi)

My netdev2.2 dietary restriction keynote in Seoul is posted, go check it out: https://www.youtube.com/watch?v=pSaXfQKDCB4

Add a comment...

Post has attachment

Greg Kroah-Hartman

Public

TIL: malloc was short for "map allocate", not "memory allocate".

Also, gotta love how pointers and integers used to be used interchangeably in UNIX, we still do that in Linux in a few places with unsigned long and pointers, and it's one of the "rules" for a processor running Linux that the two be equal in size.

blog dds: 2017.09.14 - The Origins of Malloc

spinellis.gr

Show all 12 comments

Willy Tarreau: It's very convenient to keep in mind that pointers == integers. I use this property a lot in my ebtree implementation for example where it allows me to "tag" the pointers by using the lower unused bits since they always point to aligned structures (ie: left/right, node/leaf).

Sadly with todays compilers, it becomes harder at every release to do such operations without casting and cheating all the time because the compiler constantly believes you made mistakes and emits loud warnings until you make your code unreadable to shut it up :-/

Alan Cox: +Willy Tarreau pointers are not always really integers, and the compiler does try very hard therefore to persuade you to at least use unions if you want to be crazy

There are some fun gotchas otherwise
- platforms whose pointers include keys, property tags etc
- platforms where function pointers are not convertible to void * (eg PDP/11 Split I/D)
- platforms where pointers have a 'normalised' form, so that doing simple maths on them may produce invalid pointers. Classic example being 8086.

The use of low bits for tags is of course an old trick and fairly safe. Less safe are the more insane uses for it I've seen - like an interpreted system that does the initial type checking by doing the asm version of

ptr -= (3 - type_expected);
foo = *ptr;

with alignment traps enabled...

Works great until you take signal in a tiny bit of code that has the alignment trap off and longjmp to a handler 8)

Willy Tarreau: +Alan Cox I'm indeed aware of PDP/11 and 16-bit x86's segmented mode being not so funny (ie anyone remembers this horrible MK_FP macro to make a pointer in Turbo-C ?). But most of my code today requiring these tricks targets more or less modern platforms (modern for me, coworkers have fun of me when I'm saying that armv5 and i386 must work).

I was not aware of the implementation you're mentioning, I find it rather odd... and not portable. Do that on armv5 with alignment traps off and you'll get unexpectedly rotated data in foo depending on the alignment :-)

Add a comment...

Post has attachment

Greg Kroah-Hartman

Public

So many good quotes in this paper, just go read the whole thing:

"Some Were Meant for C, The Endurance of an Unmanageable Language" by Stephen Kell.

www.cl.cam.ac.uk/~srk31/research/papers/kell17some-preprint.pdf

cl.cam.ac.uk

Add a comment...

Post has shared content

Greg Kroah-Hartman

Public

What he said :)

Originally shared by Konstantin Ryabitsev – 5 comments

Despite what various news sites out there may have told you, kernel 4.14 LTS is not planned to be supported for 6 years. Just because +Greg Kroah-Hartman is doing it for 4.4 does not mean that all LTS kernels from now on are going to be maintained for that long.

It is possible that someone may pick up maintainership of 4.14 after Greg is done with it (it's happened in the past on multiple occasions), but you should emphatically not plan on that.

Your ultimate guide for the projected EOL for all LTS kernels is this page:

https://www.kernel.org/releases.html

Any resource that contradicts that page is wrong.

Active kernel releases

In Releases. There are several main categories into which kernel releases may fall: Prepatch: Prepatch or "RC" kernels are mainline kernel pre-releases that are mostly aimed at other kernel developers and Linux enthusiasts. They must be compiled from source and usually contain new features that ...

kernel.org

Show all 9 comments

Nuno Gonçalves: 2 years is plenty of time to finally finish the kernel, no need for further updates.

Frank Wang: The security/bug fixes during the entire mobile device life cycle is what all the device makers that I talked to in the past think important to have. For consumers like me - it is true as well. Once a device is launched in the market, it is not likely to upgrade to newer kernel versions in order to get the latest fixes.

Jorge Solórzano: +Greg Kroah-Hartman how
Ben Hutchings 3.2/3.16 compares? As far I can see those are the first long LTS kernels available, those have not crashed and burn yet ;-)

Add a comment...

Post has attachment

Greg Kroah-Hartman

Public

Sometimes a name is just too good to pass up using...

Stable kernels are now released, go upgrade!

Le Petit Gorille

Add a comment...

Wait while more posts are being loaded

Press question mark to see shortcut keys available