Linux's vmalloc Seeing "Large Performance Benefits" With 5.2 Kernel Changes

[phoronix]

Phoronix: Linux's vmalloc Seeing "Large Performance Benefits" With 5.2 Kernel Changes

On top of all the changes queued for Linux 5.2 is an interesting last-minute performance improvement for the vmalloc code...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-5.2-vmalloc-Performance

[tildearrow]

Typo:

can take up to 67% less time compared to tbe behavior on Linux 5.1 and prior, at least

[milkylainen]

Now that is a beautiful change! Core functionality speedup to the benefit of just about everybody.
Interesting times for 5.2 indeed.

[nuetzel]

Wasn't there 'another' performance (memory/Lx cache/TLB?) optimization in the works? - Can't remember the right announcement.
Ugh, found one, but the 'slower':

https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.2-TLB-Info-Optimization

I think there was a much better one in the works for multi CPU/core systems...

[Widefox]

Linus on vmalloc "There are basically no valid new uses of it. There's a few valid legacy users (I think the file descriptor array), and there are some drivers that use it (which is crap, but drivers are drivers), and it's _really_ valid only for modules. Nothing else." https://lwn.net/Articles/57800/

[Widefox]

Linus on vmalloc "There are basically no valid new uses of it. There's a few valid legacy users (I think the file descriptor array), and there are some drivers that use it (which is crap, but drivers are drivers), and it's _really_ valid only for modules. Nothing else." https://lwn.net/Articles/57800/

Saying that, "Android graphics pipeline uses vmalloc allocations which can lead to delays of UI content to GPU. " https://lkml.org/lkml/2019/3/22/1911 so noticable improvements from time sensitive vmalloc users like Android.

[Volta]

I think there was a much better one in the works for multi CPU/core systems...

Yes, rwsem as far as I remember. Part of it is included in 5.2, but for full implementation we have to wait for 5.3.

[Guest]

Linus on vmalloc "There are basically no valid new uses of it. There's a few valid legacy users (I think the file descriptor array), and there are some drivers that use it (which is crap, but drivers are drivers), and it's _really_ valid only for modules. Nothing else." https://lwn.net/Articles/57800/

It's still a win, especially for mentioned embedded devices. It is unfortunate that there are plenty of crappy proprietary kernels, but they can still benefit from improvements to the mainline, this will apply especially to new devices where the manufacturer is still working on the kernel.

[reavertm]

Wow, never expected kernel to have linear lookup of anything in 2019...

[Mikachu]

Linus on vmalloc "There are basically no valid new uses of it. There's a few valid legacy users (I think the file descriptor array), and there are some drivers that use it (which is crap, but drivers are drivers), and it's _really_ valid only for modules. Nothing else." https://lwn.net/Articles/57800/

This quote is 16 years old, a lot has changed since then. For example the kernel stack is now allocated via vmalloc[1], and every process that starts has one of these.
[1]https://lwn.net/Articles/692208/