gfp_kh さんがブースト

Electronics in Shenzhen, China be like: Ready-made FPGA development boards that plug straight into a PCIe slot to access host memory via DMA. What do they do? Stealing crypto keys for espionage, as featured on Black Hat? No, it's for the kids to cheat in video games like PUBG. #electronics

gfp_kh さんがブースト
Heh, you're trying to convince a "Russian identified person" in one of the senior positions in the Linux Foundation that the Linux Foundation is being unfair towards "Russian identified people." It's pretty hilarious.

I'm not a lawyer and I don't speak for the LF, so I won't give you any kind of "official comment." But here's my view of it.

The people removed from maintainer positions were identified as employed by companies on the US and EU sanctions list. These companies are directly involved in the Russian military complex and therefore are directly complicit in war crimes being committed daily in Ukraine. If these maintainers want to think that they are "just techies helping improve the Linux kernel," or that "they are outside of politics," then they are fucking wrong. If they work for companies that develop weaponry or logistics used by the Russian military, they are complicit in Russia's war crimes, and I hold them responsible at a very personal level -- and that's my official comment on the situation.
gfp_kh さんがブースト
gfp_kh さんがブースト

What the fuck is SBAT, why did it break your dual-boot setup, it's not strictly my fault but if you need a scapegoat whatever: dreamwidth.org/captcha?returnt

gfp_kh さんがブースト

fun Linux fact:

There are at least two places in the Linux kernel where performance-sensitive code paths use a, uh, retrospective approach to memory safety, where you do the memory read first (and suppress the occasional kernel page fault) and worry about making sure it was memory-safe later.

One of the two is in the SLUB allocator's lockless allocation fastpath: elixir.bootlin.com/linux/v6.10
It (1) first loads the pointer to the current top freelist element, then (2) reads the next freelist pointer out of that element, and (3) does a double-wide atomic compare-exchange to check if it raced and replace the freelist head. The read in step (2) can return garbage or fault if someone else concurrently modified the freelist (basically it can be UAF), but in that case step (3) that uses the result of the read is guaranteed to fail, so it's fine.

The other one is in the Virtual File System layer, used for printing pathnames (like for getcwd()):
elixir.bootlin.com/linux/v6.10
As the comment there explains:

 * The source of the prepend data can be an optimistic load
* of a dentry name and length. And because we don't hold any
* locks, the length and the pointer to the name may not be
* in sync if a concurrent rename happens, and the kernel
* copy might fault as a result.
*
* The end result will correct itself when we check the
* rename sequence count, but we need to be able to handle
* the fault gracefully.

So this one can race so that OOB memory is copied into the path buffer, but the kernel will afterwards notice that the "how many rename operations have happened" counter has gone up (or that the "we are currently in the middle of a rename" bit is set), discard the result, and try again.

gfp_kh さんがブースト

The new screen capture Wayland protocol has been merged! It will unlock neat features such as window capture and better performance. Many thanks to for pushing this forward!

gitlab.freedesktop.org/wayland

gfp_kh さんがブースト

don't let anyone ruin your day

it's YOUR day!

ruin it yourself by attempting a gentoo install

gfp_kh さんがブースト
gfp_kh さんがブースト

Our first price for a laptop with the Snapdragon Elite X just leaked:

14" Galaxy Book Edge: 1800 EUR with 16+512GB.

The same laptop with Intel Core Ultra apparently starts at 1899 and a similar memory spec 13.6" MacBook Air goes for 1749 (though you have the option for lower specs).

A bit disappointed tbh. Galaxy Books always launch weirdly expensive & I guess this might include touch + OLED + pen, but Windows really needs to at least match prices IMO

winfuture.de/news,141636.html

gfp_kh さんがブースト
Ikakonbu  
招待制解除で話題のBlueskyがどういうSNSなのかはこの画像でわかります
gfp_kh さんがブースト
gfp_kh さんがブースト

Good news for those wanting to build out a custom Arm64 desktop:

ASRock Rack has a mobo+Ampere CPU combo that restocks every few weeks on Newegg: newegg.com/p/N82E16813140134

And Noctua's prototyping two new LGA 4926 coolers (pictured: NH-D9 AMP-4926 4U)

I am building out my 45Drives HL15 server with this board and will hopefully have a video up on it soon!

Full review will take longer though—need to test out software, get an idea of power draw, test performance/networking... but so far so good!

gfp_kh さんがブースト

btrfs on linux 6.7さん、metadataを使いもしないのにバカ食い(pre-allocしちゃってる)の問題もあるので、新しい各位は6.6にいといたほうがいいです

gfp_kh さんがブースト

Dumping the ROM of a GBA game by crashing it - YouTube
youtube.com/watch?v=0-7PSmYYHF
GBAのゲームクラッシュさせるとメモリを音として出力するからROMをコピーできるの意味分からなさすぎる

gfp_kh さんがブースト

GitHub feels extremely down right now. Can't even ping any part of it 👀 (image from xkcd.com/303/)

gfp_kh さんがブースト

I'm so happy to announce that the first NPU driver has been accepted in Mesa.

Have written a bigger update to celebrate this and properly thank everybody who has been supporting me in this amazing trip so far.

blog.tomeuvizoso.net/2024/01/e

#vivante #npu #etnaviv #mesa

gfp_kh さんがブースト
gfp_kh さんがブースト

We're done here, contention totally gone, improvement now +300% for both cases. I believe it's weekend time.

スレッドを表示
gfp_kh さんがブースト

Samsung、NVMe SSD「990 EVO」発表、前モデルから最大43%高速化 | マイナビニュース news.mynavi.jp/article/20240124-2868988/

古いものを表示
Fedibird

様々な目的に使える、日本の汎用マストドンサーバーです。安定した利用環境と、多数の独自機能を提供しています。