It will live on for older CPU's but will be disabled for any new CPU not affected by the issue it fixes in the same manner as 4.15 has it disabled for AMD CPU's -- perhaps unless those new CPU's also invent a new feature to keep kernel pagetable isolation (kpti) in place without the performance impact but that seems unlikely. Very fast syscalls was always one of the major technical successes of the Linux kernel and a major selling point versus for example BSD on web servers and the like; this issue mostly undoes that, so I can assure you people will want to disable it whenever possible.michael louwe wrote:What will happen to this OS bug fix when we buy a non-vulnerable new 9th-gen Intel-based computer one or two years from now.?
Note that Linus only (directly) manages the 4.x kernel, 4.15 at this point. His current release is 4.15-rc6 which includes the fix but which is still a release candidate rather than a real release. The 4.x.y kernels are managed by Greg Kroah Hartman and consist of Linus' 4.x together with backported fixes from later Linus kernels, with the rule that for a patch to live in 4.x.y it must also already be accepted into Linus' kernel.michael louwe wrote:I think what has been happening is that Linus Torvald has included this KPTI bug fix in the latest Linux kernels, ie from kernel 4.14.11 onward, which are mostly meant for Alpha-testers and Beta-testers.
4.15-rc6, 4.14.11, 4.9.74, 4.4.109, 3.16.52, 3.18.91 and 3.2.97 are all patched, although AMD users should note that the AMD exception only made it in a few hours ago; is not yet in any of those. AMD users may as such want to avoid this first wave seeing as how they'd suffer a performance degradation for no reason.
Of these kernels, 4.15-rc6 is the tester-kernel but 4.x.y kernels are in fact also known as "the stable branch" and are not generally unsuitable for end users, even though general advise would indeed be for end users to wait for a patched distribution kernel.
Also note that multitasking is not the issue. The performance degradation happens at syscall entry/exit, i.e., when a process calls into the kernel to service some request, not when switching between processes. It is as such not multi-processing loads that are affected but user mode loads that call in to the kernel a lot; loads that read/write lots of small files for example.michael louwe wrote:Many computer users multi-task or run multi-processes at the same time, eg downloading stuffs, sending emails/comments, reading news, opening stored files, etc during the same session. So, I think, when applied, this KPTI bug fix will degrade performance considerably for many users.