On Fri, Apr 06, 2018 at 06:22:11PM +0100, Mark Rutland wrote: > Digging a bit, I also thing that our ct_user_exit and ct_user_enter > usage is on dodgy ground today. > > For example, in el0_dbg we call do_debug_exception() *before* calling > ct_user_exit. Which I believe means we'd use RCU while supposedly in an > extended quiescent period, which would be bad.
It seems this is the case. I can trigger the following by having GDB place a SW breakpoint: [ 51.217947] ============================= [ 51.217953] WARNING: suspicious RCU usage [ 51.217961] 4.16.0 #4 Not tainted [ 51.217966] ----------------------------- [ 51.217974] ./include/linux/rcupdate.h:632 rcu_read_lock() used illegally while idle! [ 51.217980] [ 51.217980] other info that might help us debug this: [ 51.217980] [ 51.217987] [ 51.217987] RCU used illegally from idle CPU! [ 51.217987] rcu_scheduler_active = 2, debug_locks = 1 [ 51.217992] RCU used illegally from extended quiescent state! [ 51.217999] 1 lock held by ls/2412: [ 51.218004] #0: (rcu_read_lock){....}, at: [<0000000092efbdd5>] brk_handler+0x0/0x198 [ 51.218041] [ 51.218041] stack backtrace: [ 51.218049] CPU: 2 PID: 2412 Comm: ls Not tainted 4.16.0 #4 [ 51.218055] Hardware name: ARM Juno development board (r1) (DT) [ 51.218061] Call trace: [ 51.218070] dump_backtrace+0x0/0x1c8 [ 51.218078] show_stack+0x14/0x20 [ 51.218087] dump_stack+0xac/0xe4 [ 51.218096] lockdep_rcu_suspicious+0xcc/0x110 [ 51.218103] brk_handler+0x144/0x198 [ 51.218110] do_debug_exception+0x9c/0x190 [ 51.218116] el0_dbg+0x14/0x20 We will need to fix this before we can fiddle with kick_all_cpus_sync(). Thanks, Mark.