On Fri, Sep 29, 2017 at 12:01:24PM +0200, Paolo Bonzini wrote: > On 29/09/2017 11:30, Boqun Feng wrote: > > On Thu, Sep 28, 2017 at 04:05:14PM +0000, Paul E. McKenney wrote: > > [...] > >>> __schedule+0x201/0x2240 kernel/sched/core.c:3292 > >>> schedule+0x113/0x460 kernel/sched/core.c:3421 > >>> kvm_async_pf_task_wait+0x43f/0x940 arch/x86/kernel/kvm.c:158 > >> > >> It is kvm_async_pf_task_wait() that calls schedule(), but it carefully > >> sets state to make that legal. Except... > >> > >>> do_async_page_fault+0x72/0x90 arch/x86/kernel/kvm.c:271 > >>> async_page_fault+0x22/0x30 arch/x86/entry/entry_64.S:1069 > >>> RIP: 0010:format_decode+0x240/0x830 lib/vsprintf.c:1996 > >>> RSP: 0018:ffff88003b2df520 EFLAGS: 00010283 > >>> RAX: 000000000000003f RBX: ffffffffb5d1e141 RCX: ffff88003b2df670 > >>> RDX: 0000000000000001 RSI: dffffc0000000000 RDI: ffffffffb5d1e140 > >>> RBP: ffff88003b2df560 R08: dffffc0000000000 R09: 0000000000000000 > >>> R10: ffff88003b2df718 R11: 0000000000000000 R12: ffff88003b2df5d8 > >>> R13: 0000000000000064 R14: ffffffffb5d1e140 R15: 0000000000000000 > >>> vsnprintf+0x173/0x1700 lib/vsprintf.c:2136 > >> > >> We took a page fault in vsnprintf() while doing link_path_walk(), > >> which looks to be within an RCU read-side critical section. > >> > >> Maybe the page fault confused lockdep? > >> > >> Sigh. It is going to be a real pain if all printk()s need to be > >> outside of RCU read-side critical sections due to the possibility of > >> page faults... > >> > > > > Does this mean whenever we get a page fault in a RCU read-side critical > > section, we may hit this? > > > > Could we simply avoid to schedule() in kvm_async_pf_task_wait() if the > > fault process is in a RCU read-side critical section as follow? > > > > diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c > > index aa60a08b65b1..291ea13b23d2 100644 > > --- a/arch/x86/kernel/kvm.c > > +++ b/arch/x86/kernel/kvm.c > > @@ -140,7 +140,7 @@ void kvm_async_pf_task_wait(u32 token) > > > > n.token = token; > > n.cpu = smp_processor_id(); > > - n.halted = is_idle_task(current) || preempt_count() > 1; > > + n.halted = is_idle_task(current) || preempt_count() > 1 || rcu_preempt_depth(); > > init_swait_queue_head(&n.wq); > > hlist_add_head(&n.link, &b->list); > > raw_spin_unlock(&b->lock); This works for PREEMPT=y kernels, but can silently break RCU read-side critical sections on PREEMPT=n kernels. > > (Add KVM folks and list Cced) > > Yes, that would work. Mind to send it as a proper patch? Just out of curiosity, why is printk() being passed something that can page fault? Thanx, Paul
