Linux Sees Fix For "Critical" Scheduler Bug Introduced A Few Months Ago
Coming out Friday night was a series to address a "critical" scheduler issue with the Linux kernel.
Intel's Peter Zijlstra sent out a patch series with what he describes as a critical scheduler fix along with a set of optimizations/improvements stemming from the change amount to seven patches in total.
The main fix is for addressing a race condition from a previously unexplored dependency within the kernel's scheduler code in the pick_next_task() function. The race condition was introduced in the upstream kernel back in early August. The full severity of the impact wasn't disclosed in the patch series but were tipped off to the patch-set as being important.
As part of the follow-up patches to this "0-day", there is a performance fix in the will-it-scale/sched_yield code by around 2%.
As of writing this fix and subsequent improvements aren't yet in the mainline kernel but at least the main fix is being sent in as an urgent fix while the "optimizations" may end up being held off until the Linux 5.5 cycle.
Intel's Peter Zijlstra sent out a patch series with what he describes as a critical scheduler fix along with a set of optimizations/improvements stemming from the change amount to seven patches in total.
The main fix is for addressing a race condition from a previously unexplored dependency within the kernel's scheduler code in the pick_next_task() function. The race condition was introduced in the upstream kernel back in early August. The full severity of the impact wasn't disclosed in the patch series but were tipped off to the patch-set as being important.
As part of the follow-up patches to this "0-day", there is a performance fix in the will-it-scale/sched_yield code by around 2%.
As of writing this fix and subsequent improvements aren't yet in the mainline kernel but at least the main fix is being sent in as an urgent fix while the "optimizations" may end up being held off until the Linux 5.5 cycle.
34 Comments