News: 0001530752

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Intel Posts Linux Kernel Patches For Supporting APX - Advanced Performance Extensions

([Intel] 2 Hours Ago Linux Kernel + Intel AMX)


Intel initially detailed [1]Advanced Performance Extensions (APX) back in [2]mid-2023 as extensions to double the number of general purpose registers and add other performance features. In the time since they have done a lot preparing the GCC and LLVM/Clang compilers for the new APX ISA features while finally today they sent out an initial set of patches for preparing the Linux kernel for APX processors.

Intel doesn't intend for Advanced Performance Extensions to be used in kernel-mode but rather just user-space applications. However, the Linux kernel needs to make accommodations around Advanced Performance Extensions for the new register state managed by the XSAVE instruction set, adjusting the XSAVE buffer offset, and ensuring all kernel assumptions are safe/relevant in an APX world.

A set of patches from Intel engineers today establish a "request for comments" baseline on Linux kernel support for Advanced Performance Extensions. The patches make kernel adjustments to the XSTATE code, go ahead in enabling APX support, and dropping MPX support as it collides with APX. Finishing up the Linux kernel removal of the Intel Memory Protection Extensions (MPX) support is needed since the new XSTATE component for APX is used in the position previously used by MPX - thus a fundamental conflict between MPX and APX although no processor exists that would support both features.

[3]This RFC patch series is what lays out that initial APX enablement by the Linux kernel. As it's in an RFC state, it may still take some time before the design elements are all settled down and ultimately upstreamed.

I don't believe I've seen Intel publicly indicate what generation of processors will first feature APX. However, given that we are only seeing these kernel-side APX patches now and it will be some months before the patches are all worked through and upstreamed, and potentially a couple extra months before it's in turn with a stable kernel and beginning to appear in prominent Linux distributions, this is still a significant affair. As we already know for example that Diamond Rapids will have AVX10.2/512 support and compiler patches have made known ISA capabilities already for both Diamond Rapids and Clearwater Forest, given all things considered we're likely not going to see APX support until after that "Xeon 7" generation. So perhaps for "Xeon 8" is where APX will make its debut.

From these APX enablement patches for the Linux kernel further pointing to APX not coming with Granite Rapids or Clearwater Forest is that it notes testing was done just on an "internal Intel emulator." With Intel confirming in the recent [4]Xeon 6500P and 6700P announcement that Clearwater Forest is already operational and fully functional, plus the fact of seeing Intel engineers bringing up Granite Rapids and Clearwater Forest hardware testing in other patches, the fact they are still relying on a simulator/emulator for APX testing punts its debut out past that point.



[1] https://www.phoronix.com/search/Advanced+Performance+Extensions

[2] https://www.phoronix.com/news/Intel-APX

[3] https://lore.kernel.org/lkml/20250227184502.10288-1-chang.seok.bae@intel.com/T/

[4] https://www.phoronix.com/review/intel-xeon-6500p-6700p



coder

An Hacker there was, one of the finest sort
Who controlled the system; graphics was his sport.
A manly man, to be a wizard able;
Many a protected file he had sitting on his table.
His console, when he typed, a man might hear
Clicking and feeping wind as clear,
Aye, and as loud as does the machine room bell
Where my lord Hacker was Prior of the cell.
The Rule of good St Savage or St Doeppnor
As old and strict he tended to ignore;
He let go by the things of yesterday
And took the modern world's more spacious way.
He did not rate that text as a plucked hen
Which says that Hackers are not holy men.
And that a hacker underworked is a mere
Fish out of water, flapping on the pier.
That is to say, a hacker out of his cloister.
That was a text he held not worth an oyster.
And I agreed and said his views were sound;
Was he to study till his head wend round
Poring over books in the cloisters? Must he toil
As Andy bade and till the very soil?
Was he to leave the world upon the shelf?
Let Andy have his labor to himself!
-- Chaucer
[well, almost. Ed.]