A Fresh Take On Speculative Page Faulting Aims For The Mainline Kernel

Over the years there have been a number of rounds of patches published for speculative page faulting. The goal has been to support user-space page-faults without holding the memory management semaphores and to ultimately allow for better performance especially with threaded workloads. A fresh take on the speculative page faulting (SPF) functionality was recently volleyed on the kernel mailing list.

Various kernel developers have been involved with Linux's speculative page faulting patches in recent years albeit none over the finish line for the mainline kernel but some Android vendors and other third-parties have carried SPF patches in some cases.

Michel Lespinasse sent out some fresh patches on Friday for speculative page faulting that has incorporated ideas/feedback raised during earlier SPF work. Additionally, this new patch series is more bisect friendly for working through any regressions. Though initially being eyed for merging is a limited set of the patches (limiting it to anonymous VMAs at this point) and can be built upon moving forward when more of the code has been well tested.

Early test results in memory management benchmarks have been quite promising. Macro benchmarks are showing mostly neutral results but some Android vendors have been carrying SPF patches in the name of finding faster application start-up times.

More details on this latest attempt of speculative page faulting for the Linux kernel can be found on the LKML. So far there aren't many comments and would be too late to see it land for 5.13, but we'll see what happens this time around.