Linux Stable Updates Are Dropping The Performance-Pounding STIBP

[phoronix]

Phoronix: Linux Stable Updates Are Dropping The Performance-Pounding STIBP

The Linux stable trees that recently received STIBP "Single Thread Indirect Branch Predictors" after back-porting from Linux 4.20 are seeing the code reverted. This is the change that recently caused major slowdowns in Linux performance for workloads like Python, PHP, Java, code compilation, and other workloads like some games...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-Stable-Dropping-STIBP

[cl333r]

Michael sometimes I get the feeling that someone at Microsoft would like to poison you for playing such an important role in removing such atrocities from the Linux kernel.

[ms178]

Well done, Michael! If it weren't for your thorough investigation into that performance regression, this could have taken much longer to get the attention it deserves. Great tech journalism!

[Templar82]

Great news!

I'm not usually a LTS person, but I feel like I am going to be using 4.19 for a long time.

[dkasak]

Interesting. I'd like to have been around to witness whatever "discussions" lead to this. It looks a lot like the consensus is that the primary goal is maintaining performance, at the expensive of security, if necessary.

[Guest]

Interesting. I'd like to have been around to witness whatever "discussions" lead to this. It looks a lot like the consensus is that the primary goal is maintaining performance, at the expensive of security, if necessary.

There's always a trade-off between security and usability. Such large performance hits are pretty bad, for all users.

[anarki2]

Michael sometimes I get the feeling that someone at Microsoft would like to poison you for playing such an important role in removing such atrocities from the Linux kernel.

It's statements like this because of which the Linux community will never be taken seriously. You guys need to grow out of this fantasy of Gates eating Linux developers' children in the black forest. It's immature, unjustified, and just plain stupid.

For starters, half the Azure VMs run Linux. .NET and PowerShell run on Linux. SQL Server runs on Linux. VS Code, the most popular development environment according to the latest Stack Overflow survey, runs on Linux. Heck, Windows 10 runs 5 different distros with the click of a button. Microsoft is a platinum (highest level) member of the Linux Foundation. Will you ever realize it's no longer 1995 and Microsoft no longer consists of just Windows and Office?

Will you?

[tildearrow]

.NET runs on Linux.

But not WinForms, which almost every single .NET/C# application I've come across uses. And no, I didn't get Wine or Mono to work with my .NET applications.

[Space Heater]

Interesting. I'd like to have been around to witness whatever "discussions" lead to this. It looks a lot like the consensus is that the primary goal is maintaining performance, at the expensive of security, if necessary.

What do you mean? Anyone can read or participate in the public mailing list discussions.

Are you aware that the actual mitigation is to disable hyper threading, and that the STIBP mitigation was not a complete solution? So the very security-conscious users who are using those affected processors would not even consider STIBP.

[boxie]

But not WinForms, which almost every single .NET/C# application I've come across uses. And no, I didn't get Wine or Mono to work with my .NET applications.

I might suggest that even if WinForms worked that the apps themselves would fail in magical ways, there are so many "windowsisms" and assumptions in desktop application code made for windows that you might find it takes a very long time for developers to make the appropriate changes.