Re: [tip: locking/core] lockdep: Fix lockdep recursion

From: Qian Cai
Date: Fri Oct 09 2020 - 13:37:12 EST


On Fri, 2020-10-09 at 18:23 +0200, Peter Zijlstra wrote:
> On Fri, Oct 09, 2020 at 06:58:37AM -0700, Paul E. McKenney wrote:
> > On Fri, Oct 09, 2020 at 09:41:24AM -0400, Qian Cai wrote:
> > > On Fri, 2020-10-09 at 07:58 +0000, tip-bot2 for Peter Zijlstra wrote:
> > > > The following commit has been merged into the locking/core branch of
> > > > tip:
> > > >
> > > > Commit-ID: 4d004099a668c41522242aa146a38cc4eb59cb1e
> > > > Gitweb:
> > > > https://git.kernel.org/tip/4d004099a668c41522242aa146a38cc4eb59cb1e
> > > > Author: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > > > AuthorDate: Fri, 02 Oct 2020 11:04:21 +02:00
> > > > Committer: Ingo Molnar <mingo@xxxxxxxxxx>
> > > > CommitterDate: Fri, 09 Oct 2020 08:53:30 +02:00
> > > >
> > > > lockdep: Fix lockdep recursion
> > > >
> > > > Steve reported that lockdep_assert*irq*(), when nested inside lockdep
> > > > itself, will trigger a false-positive.
> > > >
> > > > One example is the stack-trace code, as called from inside lockdep,
> > > > triggering tracing, which in turn calls RCU, which then uses
> > > > lockdep_assert_irqs_disabled().
> > > >
> > > > Fixes: a21ee6055c30 ("lockdep: Change hardirq{s_enabled,_context} to
> > > > per-cpu
> > > > variables")
> > > > Reported-by: Steven Rostedt <rostedt@xxxxxxxxxxx>
> > > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > > Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
> > >
> > > Reverting this linux-next commit fixed booting RCU-list warnings
> > > everywhere.
> >
> > Is it possible that the RCU-list warnings were being wrongly suppressed
> > without a21ee6055c30? As in are you certain that these RCU-list warnings
> > are in fact false positives?
> > > [ 4.002695][ T0] init_timer_key+0x29/0x220
> > > [ 4.002695][ T0] identify_cpu+0xfcb/0x1980
> > > [ 4.002695][ T0] identify_secondary_cpu+0x1d/0x190
> > > [ 4.002695][ T0] smp_store_cpu_info+0x167/0x1f0
> > > [ 4.002695][ T0] start_secondary+0x5b/0x290
> > > [ 4.002695][ T0] secondary_startup_64_no_verify+0xb8/0xbb
>
> They're actually correct warnings, this is trying to use RCU before that
> CPU is reported to RCU.
>
> Possibly something like the below works, but I've not tested it, nor
> have I really thought hard about it, bring up tricky and this is just
> moving code.

I don't think this will always work. Basically, anything like printk() would
trigger the warning because it tries to acquire a lock. For example, on arm64:

[ 0.418627] lockdep_rcu_suspicious+0x134/0x14c
[ 0.418629] __lock_acquire+0x1c30/0x2600
[ 0.418631] lock_acquire+0x274/0xc48
[ 0.418632] _raw_spin_lock+0xc8/0x140
[ 0.418634] vprintk_emit+0x90/0x3d0
[ 0.418636] vprintk_default+0x34/0x40
[ 0.418638] vprintk_func+0x378/0x590
[ 0.418640] printk+0xa8/0xd4
[ 0.418642] __cpuinfo_store_cpu+0x71c/0x868
[ 0.418644] cpuinfo_store_cpu+0x2c/0xc8
[ 0.418645] secondary_start_kernel+0x244/0x318

Back to x86, we have:

start_secondary()
smp_callin()
apic_ap_setup()
setup_local_APIC()
printk() in certain conditions.

which is before smp_store_cpu_info().

Can't we add a rcu_cpu_starting() at the very top for each start_secondary(),
secondary_start_kernel(), smp_start_secondary() etc, so we don't worry about any
printk() later?

>
> ---
>
> diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
> index 35ad8480c464..9173d64ee69d 100644
> --- a/arch/x86/kernel/cpu/common.c
> +++ b/arch/x86/kernel/cpu/common.c
> @@ -1670,6 +1670,9 @@ void __init identify_boot_cpu(void)
> void identify_secondary_cpu(struct cpuinfo_x86 *c)
> {
> BUG_ON(c == &boot_cpu_data);
> +
> + rcu_cpu_starting(smp_processor_id());
> +
> identify_cpu(c);
> #ifdef CONFIG_X86_32
> enable_sep_cpu();
> diff --git a/arch/x86/kernel/cpu/mtrr/mtrr.c b/arch/x86/kernel/cpu/mtrr/mtrr.c
> index 6a80f36b5d59..5f436cb4f7c4 100644
> --- a/arch/x86/kernel/cpu/mtrr/mtrr.c
> +++ b/arch/x86/kernel/cpu/mtrr/mtrr.c
> @@ -794,8 +794,6 @@ void mtrr_ap_init(void)
> if (!use_intel() || mtrr_aps_delayed_init)
> return;
>
> - rcu_cpu_starting(smp_processor_id());
> -
> /*
> * Ideally we should hold mtrr_mutex here to avoid mtrr entries
> * changed, but this routine will be called in cpu boot time,
>