Kernel - rowhammer migitation
 

Rohammer - Migitation
Does anybody know here, what is done in the kernel for that?

Welcome to LQ. It seems to have been addressed 6.5 years ago.

Does this help?:
https://www.google.com/search?q=Rowh...e%3Akernel.org
https://en.wikipedia.org/wiki/Row_hammer

Tell us more (about your concern, etc). (Are you a security researcher? Student?)

Also search LQ: Rowhammer site:linuxquestions.org

Here, from 2016.
https://lwn.net/Articles/704920/

Thx for your help.

Gentle asked:
(I hope i don't bore you, just for introducion)

I'm master of science
I started Linux in 1996 (SuSE 4.4)
I learned a lot of OS
All kind of UN!X flavour
There is no other OS than this.
OSx included. FreeBSD included, I don't forget about GNU the base for building the Linux kernel.
No Linux without GNU.
Stallman and the Linux project/Linus talked a lot about this.
They "agreed" not to mention GNU in the name. So Linux is Kernel AND Applications.
I began consulting/developing in 1992 during studying (Assembler, PowerFlex Building PCs (BNC), Modem was 2.400 (9.600?) ...)
I'm no developer any more. But i can read, analyze and optimize a lot of code. SAP-code (ABAP) is the biggest part for code. DB Oracle.
My part is the SAP Business Warehouse. Now is HANA (DB) on the air, in Ram beginning 10TB and so on.

Ok, enough, i think i make you bored.

Rowhammer 2015. I did a lot of research. Security, Hardware an OS, Network.
Rowhammer has a lot of variations, Half, Double, different kind of Spraying, Feng Shui, JS Hammer and so on.
And the different attemption to migitate this.
Til now is no chance to migitate this, cause as you mention that this a hardware bug in DDR-Ram.
This hardware problem is not solved til now. Double Refresh, TRR, ECC doesn't solve this problem.
Ok, the name today changed to RAMBLEED and is very active. All companys (Microsoft/Amazon/Google) are working very hard on this topic, cause of their server farms.

I read a lot of universities concerning this. Solution, than destroying (GoogleZero ...) this solutions (CATT(m), Anvil and so on).
This "procedure" (destroying solution) is very important to get a 100% solution for this.
One estimated solution (several times ago) was to have a marker for kernel and user pages. Destroyed. There are some of memory regions owning both.
The access to ram in the vfs was e.g. closed in Linux.


I hope i don't make you bored.

I'm interested in changing information not only technic, privat is also very important. We are people.

Til now im not used to forums/mailing lists for kernel ... Started now.
Few months ago i did the reasearch to find the atomic instruction (assembler) for locks in the kernel.
This instruction does a Test and Set in one instruction which is not interruptible. This is a basic feature.
It's "asm volatile lock cmpxchgl" e.g.

I hope i don't make you bored. (just a feedback from you i would appreciate, thx)

I collected a lot of links. Round about 100.

If you are interested in this, i will send a zip.

OK i finish now, cause of tl;dr ;-)

I would be glad to hear something from you.

People can learn a very lot from each other. They only have to try. ;-)

cu, english is a little bit rusty.

Best reguards
Harry.

PS:
One of my hobbies started several years ago is hackintosh.
OSX on Pc.
This is my productive machine.

.

Well, I can tell you, there is ALOT to research. Since 2016 there are so many new Kernel hardening functions, including dealing with memory, that it is quite hard to keep an overview of things.
You can read quite alot about it in the Kernel config, and then beyond as well.

Ofcourse hardware based problems are some of the worst to solve, but they have done quite well to mitigate those issues.

Ps. I didn't even know about this dram issue.