HNHacker News
TopNewBestAskShowJobs

kijiki

787 karma · joined May 13, 2008

submissionscomments
kijiki··on Undocumented arm64 ISA extension present on the Apple M1
> It will be "historically significant" if Intel adds a feature to their chips in order to stay competitive with this development.

TFA points out that this Apple AMX is similar to the existing Intel AMX, even (confusingly) using the same name.

kijiki··on Mundane: Rust cryptography library that is difficult to misuse
The secure constant time implementations in Ring are from OpenSSL, and written in assembly, because that sort of code cannot be written safely in C or Rust.
kijiki··on AB07-USB3FMC: The $1.65k eval board they can't be bothered to test
Cumulus used to use a pile of Chromebooks with USB SSDs attached for ARM builds. One day in the lab I noticed that 2 of them had popped their cases open due to the battery expanding. Asking around, we were not the only folks to find that cheap Chromebooks were not designed to be run at 100% CPU 24/7.
kijiki··on Nvidia Unveils GeForce RTX 30 Series GPUs
They're just mistaken, RTLinux is hard realtime. It works by running a tiny real time OS on the bare metal, and then running a Linux kernel as a (lightly) paravirtualized lowest priority task. You write your hard realtime code to run on the RTOS, and then your soft/non realtime code runs on Linux.

In this model, your compositor, input pipeline and (at least the UI latency sensitive part of) your applications would have to be ported to the RTOS, which makes this pretty infeasible. But it works really well if you have a some hard-realtime control loop talking to a bunch of non-realtime network IO or UI code. Would be a fun way to build a wifi controlled quadcopter.

These days you probably want to use Xenomai, or possibly RTAI.

kijiki··on Raise Less Money
2/4 rounds (and the overwhelming majority of the $$$ raised) were years after the EMC acquisition, and immediately before the IPO. Not really what people think of when they talk about raising VC.

The 5 million and 20 million rounds in 2000 were strategic investments from partners who were going to resell ESX, and were done for reasons other than needing working capital. IBM and Dell were the 2 partner investors, IIRC.

kijiki··on Cooperative Linux – running Linux as a Windows driver (2011)
Oh, I didn't realize they'd worked around MS removing LDT access. They did it by running in an 16bit x86 emulator, which is a pretty big hammer, but I guess win16 apps are old enough that the performance is still more than fine on modern CPUs.
kijiki··on Cooperative Linux – running Linux as a Windows driver (2011)
It was just a choice Microsoft made. Wine happily runs win16 apps on x86_64 Linux.

There was a wine derived thing that used to work on Win64 to run win16 apps, but something in later windows versions broke that.

kijiki··on Western Digital's 16TB and 18TB drives: EAMR HDDs enter the retail channel
https://brad.livejournal.com/2116715.html

some enterprise SCSI drives too. This was all back in the late 90's/early 2000's, no idea if it still happens with SATA or SAS drives...

kijiki··on Western Digital's 16TB and 18TB drives: EAMR HDDs enter the retail channel
I'd been hearing rumors (possibly spread by the mfgs to scare enterprises away from cheap IDE drives) about it during the late 90's and early 2000's.

Then bfitz proved in in 2005, and it turns out SCSI drives sometimes do it too: https://brad.livejournal.com/2116715.html

kijiki··on C performance mystery: delete unused string constant
> ...a global constant would go in the TEXT section of the binary...

Wouldn't it go into the .rodata section, not .text?

kijiki··on Signal PINs
That isn't quite how it is implemented, there are no HSMs, we use SGX for that function instead. See my reply to Josh's follow up for details.

You need a random id not the phone number, so other users can't do guesses unless they get that random id somehow. The ID is stored on the phone, and if you lose it, when you reinstall Signal, the signal server will give it to you after SMS phone # verification.

A sim-jacker could nuke your key, but that seems like a lot of work for griefing. Signal of course, has the plaintext random id, so they could nuke your backup, but there are tons of ways Signal can DoS attack their own service. Stopping paying for the servers, for one.

kijiki··on Signal PINs
Yeah, there is a 32 byte random serviceid generated inside the enclave when it starts in "new" mode, so it is locked to the serviceid, not the HW, so we can migrate off failing/failed hardware. The only way to get a non-random serviceid is to have keyspace transferred to you from an existing enclave, after mutual attestation. Clients verify the serviceid after attestation verifies the enclave.

Each partition of the keyspace lives on a 5 node cluster, running RAFT inside the enclaves to ensure the pin guess count is consistent.

If a node fails, or is failing, or we need to do maintenance, we transfer the partition from the old nodes to new ones, which transfers the serviceid as well. If you start a new enclave from scratch, it'll generate a new serviceid, and clients won't talk to it, and existing partitions will refuse to transfer to it since it already has a serviceid. In fact, it won't even be listening for incoming transfers, since it thinks it owns the entire keyspace.

We can also do a partial partition transfer that leaves the source cluster live, which we use to split up the keyspace more when we need to scale up.

There is no persistence (wouldn't be secure, attackers could do many parallel enclave resumes to get extra guesses), so the 5 nodes in the cluster live in geographically separated data centers, and we're real careful about things that could kill a node or process.

kijiki··on Signal PINs
The enclaves will flush the data (much like the hardware you mention in your first post) after some number of wrong PIN guesses.

To do this securely made the whole thing quite difficult, because it means we can never persist anything outside the enclave, or an attacker could unpersist into 1000 unmodified enclaves (who would never know about each other) and get 1000*N guesses.

kijiki··on Signal PINs
Basically, SGX: https://github.com/signalapp/SecureValueRecovery

The enclave is hardened against Spectre and LVI with this (BOLT/LLVM based), and other techniques: https://github.com/signalapp/BOLT

The last build step before signing is a verifier that checks that there are no missed mitigations, built using Intel Xed, to try to avoid potential missing mitigations due to an LLVM or BOLT bug.

kijiki··on Rosetta: The Engine Behind Cray’s Slingshot Exascale-Era Interconnect
You can get 400Gbps bandwidth in commodity networking: https://www.fs.com/products/96982.html

Not going to get that sort of latency without extreme custom chips though.

kijiki··on Nvidia Acquires Cumulus
switchd living in userspace was in large part a result of the highly proprietary NDA's ASIC SDKs. Mellanox is making small (and welcome) steps in opening up by putting some functionality into their switchdev driver, but if you want a full feature set in 2020, you need their userspace SDK, which means switchd.

SAI, like OFDPA and OpenNSL before them (and a few more, can't even keep track anymore) are giant closed-source userspace binary blobs that Broadcom released to try (and fail) to mollify folks who want open-source networking. If you're serious about feature set and performance, there is unfortunately still no alternative to the proprietary NDA'd SDKs.

kijiki··on Nvidia Acquires Cumulus
Cumulus co-founder and original switchd author here.

Mellanox sells switches with Broadcom silicon, so certainly in the short and medium term there will be no changes there. Long term is harder to say. It is possible that once Mellanox/Nvidia's ASIC portfolio is sufficiently complete they'll stop adding support for newer competitor ASICs, but I have no special insight there.

Thanks for using Cumulus, glad to hear you're having a good experience with it!

kijiki··on Nvidia Acquires Cumulus
The plan of record is yes.
kijiki··on WD Sets the Record Straight: Lists All Drives That Use Slower SMR Tech
> Hard drives that look like tapes.

Tapes don't have fast (by spinning rust standards) random reads.

Hinting in ext4 or NTFS is of course a waste of time, but CoW filesystems like ZFS or btrfs are in a much better position to handle host-managed SMR.

And a generic "SMRTL" layer that utilizes (hopefully mirrored) SSDs as a write staging area could do massively better than the dumb host-managed SMR on these drives.

kijiki··on WD Sets the Record Straight: Lists All Drives That Use Slower SMR Tech
I am annoyed that they missed an opportunity here to fix SMR support in the non-GooFaceAmaSoft DC ecosystem.

Announce a trade-in for anyone who has a Red and has a problem with device-managed SMR, with a CMR Red. Many, many people won't notice or care (RAID1 users who don't do massive writes) Then reflash the SMR reds with the host-managed firmware they already have for the megascalers (who of course have their own highly proprietary software management layers to handle SMR) and sell those at some deep discount to anyone who wants them.

Announce a $100,000 bounty to anyone who adds host-managed SMR support (with some slightly ambitious perf target vs CMR) to ZFS, bcachefs, whatever.

They come out of this looking good (eventually) instead of looking like complete assholes. And end up with an expanded ecosystem that allows them to create a new product family using firmware work that they've already done years go!

kijiki··on PlayStation 5: the specs and the tech
> N64 was pretty much an upgraded PS1 polygon engine (hardware was SGI dervied like the PS1)

N64 was literally SGI derived, but PS1 was at best "SGI inspired". This comparison overstates the similarities.

> After that Nintendo started using ATI for graphics hardware

Nintendo was using the ArtX designs before ATI bought them. It was a very different design than the native ATI line.

kijiki··on PlayStation 5: the specs and the tech
The PS2 was where this peaked. It has a weird multi CPU architecture with multiple CPUs and coprocessors, but was popular enough (dominant, really) that devs really invested in some cases. Shadow of the Colossus is a good example.

They tried the same playbook again with the PS3, but paniced when they realized how hard the cell SPEs would be to program to execute shaders effectively and added an Nvidia GPU. The PS3 was not sufficiently more popular than the Xbox 360 to justify the level of investment to really capitalize on the unusual architecture.

kijiki··on New flaw in Intel chips lets attackers slip their own data into secure enclave
The driver for recalls is demand from the largest customers. In Intel's case, mostly big cloud operators.

They have a strong vested interest in pretending that sharing compute resources across customers is a safe thing to do, so they're unlikely to rock the boat.

kijiki··on New flaw in Intel chips lets attackers slip their own data into secure enclave
Signal already uses SGX for Contact Discovery, and has for many months.

In all cases, Signal only uses SGX to improve cases where data previously was stored in plaintext. The idea is to change the required attack from "compromise signal servers" to "compromise signal servers _and_ compromise SGX".

kijiki··on 82nd Airborne unit told to use Signal or Wickr on government cell phones
https://signal.org/blog/secure-value-recovery/
kijiki··on Inferring and hijacking VPN-tunneled TCP connections
This might be fixable with the kernel's VRF functionality: https://www.kernel.org/doc/Documentation/networking/vrf.txt
kijiki··on Tron: Evolution – SecuROM DRM expiration makes game unplayable 9 years later
So in this version of the story, the MCP defeats TRON.
kijiki··on KeePassXC 2.5
Keypass is written in C#, so requires a .NET runtime.

KeypassX is a rewrite in C++, using QT. KeypassXC is a fork of KeypassX, as KeypassX was felt to be unmaintained.

kijiki··on AMD Rome – is it for real? Architecture and initial HPC performance
Yes, but those are fixed function designs like Trident or Tomahawk (yes, they name their switch chips after missiles).

When Broadcom uses the term SOC, they mean "Switch on a Chip" to differentiate their single chip designs from older multi chip architectures.

SiByte was MIPS-based, more like an NPU than an (Broadcom term) SOC.

kijiki··on Fixing Getrandom()
It was never ARM. It used to be ARC, but these days it is x86.
← PreviousPage 5 of 12Next →