Intel Linear Address Masking "LAM" Merged Into Linux 6.4

[phoronix]

Phoronix: Intel Linear Address Masking "LAM" Merged Into Linux 6.4

Since 2020 Intel engineers have been working on Linear Address Masking (LAM) as a feature similar to Arm's Top Byte Ignore (TBI) for letting user-space store metadata within some bits of pointers without masking it out before use. This can be of use to virtual machines, profiling / sanitizers / tagging, and other applications. The Intel LAM kernel support has finally been merged with Linux 6.4...

https://www.phoronix.com/news/Intel-LAM-Merged-Linux-6.4

[colejohnson66]

While my initial reaction to Intel LAM / AMD UAI was, "did we not learn the problems with these 'features' when '32-bit clean' programs became a requirement a few decades ago?" However, Intel's approach to, not remove canonical address requirements, but modify it in the way they did, seems like it's a decent implementation, actually. I'm not too thrilled about AMD's approach also ignoring the MSB (bit 63), as it will require more work than just checking the sign bit for user/kernel address checking, but maybe that's just me.

Also, now that virtual memory has been a thing for decades, even if processors get more than 52 physical address lines, the page table walk can easily be modified to still support this implementation on a per-thread basis.