Intel LAM Will Try Again For Linux 6.4

Intel's Linear Address Masking (LAM) functionality to make use of untranslated address bits of 64-bit linear addresses for arbitrary metadata is aiming to be mainlined with the upcoming Linux 6.4 cycle.

Intel LAM is similar to Arm's Top Byte Ignore (TBI) and AMD Zen 4's UAI for Upper Address Ignore. The intent is that with making use of the untranslated address bits for arbitrary metadata it can be used for optimization strategies by Just-In-Time (JIT) compilers, more efficient address santizer (ASAN) mechanisms, VMs, and other innovative purposes.

Going back to 2020 Intel's been working on the LAM Linux support and went through many rounds of revisions for the kernel patches as well as bringing up the associated toolchain support. Intel LAM was submitted for Linux 6.2 but then Linus Torvalds ended up rejecting the code over concerns of some aspects being "fundamentally broken" and raising related technical concerns.

Torvalds also isn't fond of the "LAM" acroynmn but alas it appears too late for Intel to realistically change it. Following the rejection from the Linux 6.2 kernel merge window, Intel engineers updated the kernel patches and after a total of sixteen revisions to it over the course of the patch series, it looks like it's now ready again to try for mainline.

The LAM patches have been picked up by TIP's x86/mm branch. Barring any new issues from being raised, this LAM support will be sent in then for the Linux 6.4 merge window as part of the x86 memory management updates. Unless Linus Torvalds finds new reasons to object, this Intel hardware feature is thus finally scratched off the list come v6.4 this summer.