[lkml]   [2021]   [Nov]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] Reset task stack state in bringup_cpu()
On Mon, Nov 15, 2021 at 11:33:10AM +0000, Mark Rutland wrote:
> To hot unplug a CPU, the idle task on that CPU calls a few layers of C
> code before finally leaving the kernel. When KASAN is in use, poisoned
> shadow is left around for each of the active stack frames, and when
> shadow call stacks are in use. When shadow call stacks are in use the
> task's SCS SP is left pointing at an arbitrary point within the task's
> shadow call stack.
> When an offlines CPU is hotlpugged back into the kernel, this stale
> state can adversely affect the newly onlined CPU. Stale KASAN shadow can
> alias new stackframes and result in bogus KASAN warnings. A stale SCS SP
> is effectively a memory leak, and prevents a portion of the shadow call
> stack being used. Across a number of hotplug cycles the task's entire
> shadow call stack can become unusable.
> We previously fixed the KASAN issue in commit:
> e1b77c92981a5222 ("sched/kasan: remove stale KASAN poison after hotplug")
> In commit:
> f1a0a376ca0c4ef1 ("sched/core: Initialize the idle task with preemption disabled")
> ... we broke both KASAN and SCS, with SCS being fixed up in commit:
> 63acd42c0d4942f7 ("sched/scs: Reset the shadow stack when idle_task_exit")
> ... but as this runs in the context of the idle task being offlines it's
> potentially fragile.
> Fix both of these consistently and more robustly by resetting the SCS SP
> and KASAN shadow immediately before we online a CPU. This ensures the
> idle task always has a consistent state, and removes the need to do so
> when initializing an idle task or when unplugging an idle task.
> I've tested this with both GCC and clang, with reelvant options enabled,
> offlining and online CPUs with:
> | while true; do
> | for C in /sys/devices/system/cpu/cpu*/online; do
> | echo 0 > $C;
> | echo 1 > $C;
> | done
> | done
> Link:
> Link:
> Fixes: 1a0a376ca0c4ef1 ("sched/core: Initialize the idle task with preemption disabled")
> Reported-by: Qian Cai <>
> Signed-off-by: Mark Rutland <>

Thanks for fixing this quickly, Mark. Triggering an user-after-free in
user namespace but don't think it is related. I'll investigate that
first since it is blocking the rest of regression testing.

 \ /
  Last update: 2021-11-16 17:33    [W:0.098 / U:1.088 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site