Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Message-ID: <dacc7a27f5545714f1f9ab51510e7856d5118b88.camel@gmx.de>
Subject: Re: [PATCH] kasan: Fix sleeping function called from invalid
 context in PREEMPT_RT
From:   Mike Galbraith <efault@gmx.de>
To:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        Zqiang <qiang1.zhang@intel.com>
Cc:     ryabinin.a.a@gmail.com, glider@google.com, andreyknvl@gmail.com,
        dvyukov@google.com, akpm@linux-foundation.org,
        kasan-dev@googlegroups.com, linux-kernel@vger.kernel.org,
        linux-rt-users@vger.kernel.org
Date:   Mon, 04 Apr 2022 14:12:27 +0200
In-Reply-To: <YkbFhgN1jZPTMfnS@linutronix.de>
References: <20220401091006.2100058-1-qiang1.zhang@intel.com>
         <YkbFhgN1jZPTMfnS@linutronix.de>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.42.4 
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, 2022-04-01 at 11:27 +0200, Sebastian Andrzej Siewior wrote:
> On 2022-04-01 17:10:06 [+0800], Zqiang wrote:
> > BUG: sleeping function called from invalid context at kernel/locking/s=
pinlock_rt.c:46
> > in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 1, name: swappe=
r/0
> > preempt_count: 1, expected: 0
> > ...........
> > CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.17.1-rt16-yocto-preempt-rt=
 #22
> > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009),
> > BIOS rel-1.15.0-0-g2dd4b9b3f840-prebuilt.qemu.org 04/01/2014
> > Call Trace:
> > <TASK>
> > dump_stack_lvl+0x60/0x8c
> > dump_stack+0x10/0x12
> > =C2=A0__might_resched.cold+0x13b/0x173
> > rt_spin_lock+0x5b/0xf0
> > =C2=A0___cache_free+0xa5/0x180
> > qlist_free_all+0x7a/0x160
> > per_cpu_remove_cache+0x5f/0x70
> > smp_call_function_many_cond+0x4c4/0x4f0
> > on_each_cpu_cond_mask+0x49/0xc0
> > kasan_quarantine_remove_cache+0x54/0xf0
> > kasan_cache_shrink+0x9/0x10
> > kmem_cache_shrink+0x13/0x20
> > acpi_os_purge_cache+0xe/0x20
> > acpi_purge_cached_objects+0x21/0x6d
> > acpi_initialize_objects+0x15/0x3b
> > acpi_init+0x130/0x5ba
> > do_one_initcall+0xe5/0x5b0
> > kernel_init_freeable+0x34f/0x3ad
> > kernel_init+0x1e/0x140
> > ret_from_fork+0x22/0x30
> >
> > When the kmem_cache_shrink() be called, the IPI was triggered, the
> > ___cache_free() is called in IPI interrupt context, the local lock
> > or spin lock will be acquired. on PREEMPT_RT kernel, these lock is
> > replaced with sleepbale rt spin lock, so the above problem is triggere=
d.
> > fix it by migrating the release action from the IPI interrupt context
> > to the task context on RT kernel.
>
> I haven't seen that while playing with kasan. Is this new?

Don't think so, the rock below was apparently first tossed at 5.12.

=2D--
 lib/stackdepot.c      |    3 +++
 mm/kasan/quarantine.c |   49 ++++++++++++++++++++++++++++++++++++++++++++=
+++++
 2 files changed, 52 insertions(+)

=2D-- a/lib/stackdepot.c
+++ b/lib/stackdepot.c
@@ -375,6 +375,9 @@ depot_stack_handle_t __stack_depot_save(
 	if (found)
 		goto exit;

+	if (IS_ENABLED(CONFIG_PREEMPT_RT) && can_alloc && !preemptible())
+		can_alloc =3D false;
+
 	/*
 	 * Check if the current or the next stack slab need to be initialized.
 	 * If so, allocate the memory - we won't be able to do that under the
=2D-- a/mm/kasan/quarantine.c
+++ b/mm/kasan/quarantine.c
@@ -19,6 +19,9 @@
 #include <linux/srcu.h>
 #include <linux/string.h>
 #include <linux/types.h>
+#include <linux/cpu.h>
+#include <linux/mutex.h>
+#include <linux/workqueue.h>
 #include <linux/cpuhotplug.h>

 #include "../slab.h"
@@ -319,6 +322,48 @@ static void per_cpu_remove_cache(void *a
 	qlist_free_all(&to_free, cache);
 }

+#ifdef CONFIG_PREEMPT_RT
+struct remove_cache_work {
+	struct work_struct work;
+	struct kmem_cache *cache;
+};
+
+static DEFINE_MUTEX(remove_caches_lock);
+static DEFINE_PER_CPU(struct remove_cache_work, remove_cache_work);
+
+static void per_cpu_remove_cache_work(struct work_struct *w)
+{
+	struct remove_cache_work *rcw;
+
+	rcw =3D container_of(w, struct remove_cache_work, work);
+	per_cpu_remove_cache(rcw->cache);
+}
+
+static void per_cpu_remove_caches_sync(struct kmem_cache *cache)
+{
+	struct remove_cache_work *rcw;
+	unsigned int cpu;
+
+	cpus_read_lock();
+	mutex_lock(&remove_caches_lock);
+
+	for_each_online_cpu(cpu) {
+		rcw =3D &per_cpu(remove_cache_work, cpu);
+		INIT_WORK(&rcw->work, per_cpu_remove_cache_work);
+		rcw->cache =3D cache;
+		schedule_work_on(cpu, &rcw->work);
+	}
+
+	for_each_online_cpu(cpu) {
+		rcw =3D &per_cpu(remove_cache_work, cpu);
+		flush_work(&rcw->work);
+	}
+
+	mutex_unlock(&remove_caches_lock);
+	cpus_read_unlock();
+}
+#endif
+
 /* Free all quarantined objects belonging to cache. */
 void kasan_quarantine_remove_cache(struct kmem_cache *cache)
 {
@@ -332,7 +377,11 @@ void kasan_quarantine_remove_cache(struc
 	 * achieves the first goal, while synchronize_srcu() achieves the
 	 * second.
 	 */
+#ifndef CONFIG_PREEMPT_RT
 	on_each_cpu(per_cpu_remove_cache, cache, 1);
+#else
+	per_cpu_remove_caches_sync(cache);
+#endif

 	raw_spin_lock_irqsave(&quarantine_lock, flags);
 	for (i =3D 0; i < QUARANTINE_BATCHES; i++) {