Cover photo
Arnd Bergmann
Works at Linaro
1,417 followers|227,593 views


Arnd Bergmann

Shared publicly  - 
#bearwithme +Bear Hands concert in Las Vegas

Add a comment...

Arnd Bergmann

Shared publicly  - 
Yes, there is now a "fake" short fingerprint for my kernel signing key out there on the key servers, and yes, it's not really mine, and yes, we know who did it, and yes, it's revoked, and no, it wasn't just targeted at kernel developers, but at all 24000 keys in the "strong" ring of PGP trust, and yes something like this has been possible for a very long time now so it's not really that much news, and yes, gpg really is horrible to use and almost impossible to use correctly.

See the top comment here for more details:

And of course, read the site for loads of details.

I guess I should be happy that people are checking the signature of my kernel releases, and emailing me that something is "wrong" on their system, that's nice to see. Too bad their scripts are "wrong" as they pull in all keys with a possible 32bit signature and things go boom.

Short answer, always use "long" keys when using gpg, and never auto-refresh keys from the keyservers.
GPG usage has grown steadily while the tooling that supports it remains stagnant despite staggering hardware advancement. 32bit key ids were reasonable 15 years ago but are obsolete now. Using modern GPUs, we have found collisions for every 32bit key id in the WOT's (Web of Trust) strong set.
15 comments on original post
Linus Torvalds's profile photoJohannes Berg's profile photo
Right, I'm aware of the details, was just wondering why it was now a story rather than the ~three years ago they generated them.

But if they were only now added to the keyservers and thus causing problems, that explains why we're seeing so many stories etc. about it now :)
Add a comment...

Arnd Bergmann

Shared publicly  - 
+Dirk Herrendoerfer No idea what I'd do with this, but now I want one.

Nice article!
 ·  Translate
Koen Kooi's profile photoDavid “” Tweed's profile photoDirk Herrendoerfer's profile photoCarsten Otte's profile photo
Nice article indeed, +Dirk Herrendoerfer​. Well done
Add a comment...

Arnd Bergmann

Shared publicly  - 
Nice to see today's doodle about Lotte Reiniger, arguably the most famous citizen of Dettenhausen, where I now live (I walk past her grave every day on the way to daycare). has an article about her life and her contributions to the art of animation.
Lotte Reiniger’s 117th birthday! #GoogleDoodle
Heiko Stübner's profile photo
Also the Making-of for the Doodle is pretty amazing, showing all the non-digital effort that went into it.
Add a comment...

Arnd Bergmann

Shared publicly  - 
Not everyone reads all the merge commit logs that go into the Linux kernel (why not?), so for those who haven't seen these but are interested in the ARM changes anyway, here are some links to what is going in for v4.7 from the arm-soc tree.

This is less exciting than the last two merge windows with the ARMv6/v7 multiplatform work concluding in v4.5 and lots of new platforms getting merged in v4.6.

We have had seven pull requests so far (one more coming, plus fixes), so see below for the introductory mail, and the comments under it for the individual merges.
Hi Linus

These are the first seven pull requests for arm-soc this time, there is one small follow-up coming once the dependencies for that have landed in your tree.

Things remain relatively calm for us. We merged 122 pull requests with 822 patches, plus 35 individual patches, which is less than what we used to have. We are shifting even more towards changing only dts files,
for the most part adding new functionality on existing boards, and adding new machines, as shown clearly by the dirstat:

3.4% Documentation/devicetree/bindings/
56.3% arch/arm/boot/dts/
0.7% arch/arm/configs/
0.0% arch/arm/mach-aspeed/
0.2% arch/arm/mach-at91/
0.0% arch/arm/mach-bcm/
0.5% arch/arm/mach-davinci/
0.0% arch/arm/mach-dove/
0.1% arch/arm/mach-exynos/
0.0% arch/arm/mach-imx/
0.1% arch/arm/mach-integrator/
0.1% arch/arm/mach-lpc32xx/
0.0% arch/arm/mach-mediatek/
0.0% arch/arm/mach-mv78xx0/
1.3% arch/arm/mach-omap2/
0.0% arch/arm/mach-orion5x/
0.0% arch/arm/mach-oxnas/
0.0% arch/arm/mach-realview/
0.0% arch/arm/mach-rockchip/
0.3% arch/arm/mach-shmobile/
0.0% arch/arm/mach-sti/
0.0% arch/arm/mach-tegra/
0.0% arch/arm/mach-uniphier/
0.0% arch/arm/mach-versatile/
0.0% arch/arm/mach-vexpress/
0.0% arch/arm/plat-samsung/
0.9% arch/arm/
13.1% arch/arm64/boot/dts/
0.0% arch/arm64/configs/
0.0% arch/arm64/
0.0% drivers/bus/
0.9% drivers/clk/
0.0% drivers/firmware/
0.0% drivers/gpu/drm/tegra/
0.0% drivers/gpu/drm/vc4/
0.0% drivers/irqchip/
0.4% drivers/memory/samsung/
0.0% drivers/memory/
0.4% drivers/mtd/maps/
0.0% drivers/of/
0.3% drivers/pci/host/
9.2% drivers/phy/tegra/
0.1% drivers/phy/
0.0% drivers/pinctrl/tegra/
0.6% drivers/reset/
0.1% drivers/soc/brcmstb/
1.0% drivers/soc/mediatek/
0.4% drivers/soc/qcom/
1.2% drivers/soc/renesas/
0.4% drivers/soc/rockchip/
0.7% drivers/soc/tegra/
730 files changed, 47909 insertions(+), 7916 deletions(-)

My feeling is that the number of dts changes remain constant, but everything else gets less every time, and that's good.

One part that sticks out this time in the dirstat this time is the NVIDIA rework for the USB PHY, which had complex interdependencies so we are merging all the subsystem specific change through arm-soc.

I currently don't see any conflicts against other trees, or between our branches.

The most active individual developers (15 patches or more) out of the 165 people that contributed this time were:

51 Geert Uytterhoeven
42 Simon Horman
23 Heiko Stuebner
22 Hans de Goede
21 Jon Hunter
21 Chanwoo Choi
20 Thierry Reding
17 Vladimir Zapolskiy
17 Linus Walleij
17 Krzysztof Kozlowski
16 Lee Jones
15 Srinivas Kandagatla
15 Roger Quadros
15 Fabio Estevam

Arnd Bergmann's profile photoAlison Chaiken's profile photoAnton Fosselius's profile photo
Thanks indeed! Interesting read!
Add a comment...

Arnd Bergmann

Shared publicly  - 
Final preparations for Linaro Connect: drove over to the Ritter Sport factory to get chocolate.
Julian Calaby's profile photoRoy Franz's profile photoArnd Bergmann's profile photo
+Roy Franz just find me in the kernel team hacking room.
Add a comment...

Arnd Bergmann

Shared publicly  - 
Grant Grundler's profile photoGeert Uytterhoeven's profile photoArnd Bergmann's profile photo
it was the time I needed to see most of the world (I missed the vulcano lair).
Add a comment...

Arnd Bergmann

Shared publicly  - 
I got the scenic flight and saw Las Vegas from above coming from AMS, still stuck in LAX for a few more hours before I get the connection back to LAS.
Andrew Pinski's profile photo
I am in sjc for a few hours. Coming from pek.
Add a comment...

Arnd Bergmann

Shared publicly  - 
Linus merged the arm-soc changes for this merge window, after +Olof Johansson posted the pull requests. Here is the summary for everyone not following the mailing lists or kernel commit log. See comments below for the individual pull request comments.
Here's our set of branches for this release. As usual, each pull request describes the conflicts that will show up and brief descriptions on how to resolve. I've also pushed a sample-merge branch to the repo to compare final results with.

When it comes to contents, this is a release cycle that's on the small side for us at "only" 858 patches merged that haven't already gone in through other trees.

New platforms for this cycle are:

- Broadcom BCM23550
- Freescale i.MX7Solo
- Qualcomm MDM9615
- Renesas r8a7792

- Broadcom BCM2837
- Renesas r8a7796

Out of those, BCM2837 might be worth pointing out since it is the SoC used on Raspberry Pi 3, so it's great to start seeing support there.

As contributors go, the top 15 authors of patches are:

+Geert Uytterhoeven (73)
+Krzysztof Kozłowski (41)
+Javier Martinez Canillas (36)
+Ben Dooks (33)
+Hans De Goede (33)
+Alexandre Belloni (29)
+Arnd Bergmann (25)
+Chen-Yu Tsai (23)
+Srinivas Kandagatla (21)
+Thierry Reding (20)
Jon Hunter (19)
+Andy Gross (18)
Alexander Shiyan (17)
Florian Fainelli (17)
+Sergei Shtylyov (17)

And dirstat speaks pretty clearly about where the bulk of changes are for us now: 32-bit device-tree contents. 64-bit platforms are slowly picking up but have nowhere near the rate of change:

61.8% arch/arm/boot/dts/
4.3% arch/arm/mach-ux500/
8.0% arch/arm/
9.2% arch/arm64/boot/dts/
11.4% drivers/

Driver stand out this release cycle in part due to an active time for reset driver conversions, and that tree is normally merged through us.
Arnd Bergmann's profile photo

"Just as the 32-bit contents, the 64-bit device tree branch also contains a number of additions this release cycle.

New platforms:
- LG LG1313
- Mediatek MT6755
- Renesas r8a7796
- Broadcom 2837

Other platforms with larger updates are:
- Nvidia X1 platforms (USB 3.0, regulators,
display subsystem)
- Mediatek MT8173 (display subsystem added)
- Rockchip RK3399 (a lot of new peripherals)
- ARM Juno reference implementation (SCPI
power domains, coresight, thermal)"
Add a comment...

Arnd Bergmann

Shared publicly  - 
The new top500 list is out, here is what I found interesting:

- Compute accelerators startig to die: there are a couple of new NVIDIA Tesla systems, but overall the numbers are shrinking: Xeon PHI is down from 29 to 23, NVIDIA is down from 68 to 63, everything else is stable but irrelevant.

- Xeon PHI as a processor (rather than an accelerator) has finally appeared in one system.

- As expected, the latest Intel Xeon E5 systems are the ones taking over almost all of the spots of the systems that drop out, but there are also a couple of new Xeon E7.

- The only other CPU architecture that gained new systems is ShenWei/SunWay, which doubled the number of installed systems (now there are two) and holds the top spot but apparently missed the 100 petaflops goal they set by a few percent.

- The last Intel Core system (Xeon E5450) is gone, as expected.

- The seven SPARC systems are all still there, most of them were installed in 2015.

- Similarly, the 19 BlueGene systems from last time are all still there (and fairly high up the list), which is surprising given their age.

- Non-Linux systems are down from six to three. All of them run AIX on POWER7 (one additional P7 runs Linux) and are currently at #282, 283 and 417, so it will still take some time before the first two (both installed 2014) fall off the bottom of the list and everything runs Linux.

- While almost everything is a cluster, Cray managed to sell five new MPP machines that made it into the list, and the new #1 is also MPP.

- No POWER8 and no ARM to be seen anywhere, or any other architecture.
Peter Robinson's profile photoMåns Rullgård's profile photoRiku Voipio's profile photoGeert Uytterhoeven's profile photo
Where's the SW26010 upstream Linux support?
Add a comment...

Arnd Bergmann

Shared publicly  - 
I saw this on today. I've seen m.2 adapters before, but this comes in a convenient USB stick form factor. If your job involves working with files on a SATA-less ARM development board, e.g. doing native compiles or installing distros frequently, you probably want one of them (I'm sure you can find a similar thing at your favourite retailer).

Even if you have a "fast" eMMC on-board.
Even if you only have USB2.
Even if you get the cheapest m.2 SSD you can find (Transcend MTS400 32GB costs EUR 23).

Almost all SSDs have the DRAM cache and the controller necessary to give you good throughput on small I/Os, while almost all eMMC and SD cards do not.

If someone has one of these or similar, please let me know how they are in terms of build quality and throughput.
I like this from DX. Find the cool gadgets at a incredibly low price with worldwide free shipping here.
Felipe Balbi's profile photoArnd Bergmann's profile photo
+Felipe Balbi no, sync() only forces out data from the page cache to get sent to the device. The GC state is in the device internally, and even survives a poweroff of the device in many cases. For eMMC, you can trigger the GC through software (but we don't do that for a sync), but not for USB.
Note that the 'dd' command already uses oflag=sync, so an extra sync is not needed.
Add a comment...
Basic Information
  • Linaro
    2010 - present
  • IBM
    2002 - 2012