Song noticed switch_mm_irqs_off taking a lot of CPU time in recent kernels, using 1.9% of a 48 CPU system during a netperf run. Digging into the profile, the atomic operations in cpumask_clear_cpu and cpumask_set_cpu are responsible for about half of that CPU use.
However, the CPUs running netperf are simply switching back and forth between netperf and the idle task, which does not require any changes to the mm_cpumask if lazy TLB mode were used. Additionally, the init_mm cpumask ends up being the most heavily contended one in the system, for no reason at all. Making really lazy TLB mode work again on modern kernels, by sending a shootdown IPI only when page table pages are being unmapped, we get back some performance. v6 addresses the comment and Signed-off-by issues pointed out by Ingo. On memcache workloads on 2 socket systems, this patch series seems to reduce total system CPU use by 1-2%. On Song's netbench tests, CPU use in the context switch time is about cut in half. These patches also provide a little memory savings by shrinking the size of mm_struct, especially on distro kernels compiled with a gigantically large NR_CPUS.