kasan: fix lockdep report invalid wait context
authorZqiang <qiang1.zhang@intel.com>
Mon, 27 Mar 2023 12:00:19 +0000 (20:00 +0800)
committerAndrew Morton <akpm@linux-foundation.org>
Fri, 21 Apr 2023 21:51:59 +0000 (14:51 -0700)
For kernels built with the following options and booting

CONFIG_SLUB=y
CONFIG_DEBUG_LOCKDEP=y
CONFIG_PROVE_LOCKING=y
CONFIG_PROVE_RAW_LOCK_NESTING=y

[    0.523115] [ BUG: Invalid wait context ]
[    0.523315] 6.3.0-rc1-yocto-standard+ #739 Not tainted
[    0.523649] -----------------------------
[    0.523663] swapper/0/0 is trying to lock:
[    0.523663] ffff888035611360 (&c->lock){....}-{3:3}, at: put_cpu_partial+0x2e/0x1e0
[    0.523663] other info that might help us debug this:
[    0.523663] context-{2:2}
[    0.523663] no locks held by swapper/0/0.
[    0.523663] stack backtrace:
[    0.523663] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 6.3.0-rc1-yocto-standard+ #739
[    0.523663] Call Trace:
[    0.523663]  <IRQ>
[    0.523663]  dump_stack_lvl+0x64/0xb0
[    0.523663]  dump_stack+0x10/0x20
[    0.523663]  __lock_acquire+0x6c4/0x3c10
[    0.523663]  lock_acquire+0x188/0x460
[    0.523663]  put_cpu_partial+0x5a/0x1e0
[    0.523663]  __slab_free+0x39a/0x520
[    0.523663]  ___cache_free+0xa9/0xc0
[    0.523663]  qlist_free_all+0x7a/0x160
[    0.523663]  per_cpu_remove_cache+0x5c/0x70
[    0.523663]  __flush_smp_call_function_queue+0xfc/0x330
[    0.523663]  generic_smp_call_function_single_interrupt+0x13/0x20
[    0.523663]  __sysvec_call_function+0x86/0x2e0
[    0.523663]  sysvec_call_function+0x73/0x90
[    0.523663]  </IRQ>
[    0.523663]  <TASK>
[    0.523663]  asm_sysvec_call_function+0x1b/0x20
[    0.523663] RIP: 0010:default_idle+0x13/0x20
[    0.523663] RSP: 0000:ffffffff83e07dc0 EFLAGS: 00000246
[    0.523663] RAX: 0000000000000000 RBX: ffffffff83e1e200 RCX: ffffffff82a83293
[    0.523663] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffffff8119a6b1
[    0.523663] RBP: ffffffff83e07dc8 R08: 0000000000000001 R09: ffffed1006ac0d66
[    0.523663] R10: ffff888035606b2b R11: ffffed1006ac0d65 R12: 0000000000000000
[    0.523663] R13: ffffffff83e1e200 R14: ffffffff84a7d980 R15: 0000000000000000
[    0.523663]  default_idle_call+0x6c/0xa0
[    0.523663]  do_idle+0x2e1/0x330
[    0.523663]  cpu_startup_entry+0x20/0x30
[    0.523663]  rest_init+0x152/0x240
[    0.523663]  arch_call_rest_init+0x13/0x40
[    0.523663]  start_kernel+0x331/0x470
[    0.523663]  x86_64_start_reservations+0x18/0x40
[    0.523663]  x86_64_start_kernel+0xbb/0x120
[    0.523663]  secondary_startup_64_no_verify+0xe0/0xeb
[    0.523663]  </TASK>

The local_lock_irqsave() is invoked in put_cpu_partial() and happens in
IPI context, due to the CONFIG_PROVE_RAW_LOCK_NESTING=y (the
LD_WAIT_CONFIG not equal to LD_WAIT_SPIN), so acquire local_lock in IPI
context will trigger above calltrace.

This commit therefore moves qlist_free_all() from hard-irq context to task
context.

Link: https://lkml.kernel.org/r/20230327120019.1027640-1-qiang1.zhang@intel.com
Signed-off-by: Zqiang <qiang1.zhang@intel.com>
Acked-by: Marco Elver <elver@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/kasan/quarantine.c

index 7558507..152dca7 100644 (file)
@@ -99,7 +99,6 @@ static unsigned long quarantine_size;
 static DEFINE_RAW_SPINLOCK(quarantine_lock);
 DEFINE_STATIC_SRCU(remove_cache_srcu);
 
-#ifdef CONFIG_PREEMPT_RT
 struct cpu_shrink_qlist {
        raw_spinlock_t lock;
        struct qlist_head qlist;
@@ -108,7 +107,6 @@ struct cpu_shrink_qlist {
 static DEFINE_PER_CPU(struct cpu_shrink_qlist, shrink_qlist) = {
        .lock = __RAW_SPIN_LOCK_UNLOCKED(shrink_qlist.lock),
 };
-#endif
 
 /* Maximum size of the global queue. */
 static unsigned long quarantine_max_size;
@@ -319,16 +317,6 @@ static void qlist_move_cache(struct qlist_head *from,
        }
 }
 
-#ifndef CONFIG_PREEMPT_RT
-static void __per_cpu_remove_cache(struct qlist_head *q, void *arg)
-{
-       struct kmem_cache *cache = arg;
-       struct qlist_head to_free = QLIST_INIT;
-
-       qlist_move_cache(q, &to_free, cache);
-       qlist_free_all(&to_free, cache);
-}
-#else
 static void __per_cpu_remove_cache(struct qlist_head *q, void *arg)
 {
        struct kmem_cache *cache = arg;
@@ -340,7 +328,6 @@ static void __per_cpu_remove_cache(struct qlist_head *q, void *arg)
        qlist_move_cache(q, &sq->qlist, cache);
        raw_spin_unlock_irqrestore(&sq->lock, flags);
 }
-#endif
 
 static void per_cpu_remove_cache(void *arg)
 {
@@ -362,6 +349,8 @@ void kasan_quarantine_remove_cache(struct kmem_cache *cache)
 {
        unsigned long flags, i;
        struct qlist_head to_free = QLIST_INIT;
+       int cpu;
+       struct cpu_shrink_qlist *sq;
 
        /*
         * Must be careful to not miss any objects that are being moved from
@@ -372,20 +361,13 @@ void kasan_quarantine_remove_cache(struct kmem_cache *cache)
         */
        on_each_cpu(per_cpu_remove_cache, cache, 1);
 
-#ifdef CONFIG_PREEMPT_RT
-       {
-               int cpu;
-               struct cpu_shrink_qlist *sq;
-
-               for_each_online_cpu(cpu) {
-                       sq = per_cpu_ptr(&shrink_qlist, cpu);
-                       raw_spin_lock_irqsave(&sq->lock, flags);
-                       qlist_move_cache(&sq->qlist, &to_free, cache);
-                       raw_spin_unlock_irqrestore(&sq->lock, flags);
-               }
-               qlist_free_all(&to_free, cache);
+       for_each_online_cpu(cpu) {
+               sq = per_cpu_ptr(&shrink_qlist, cpu);
+               raw_spin_lock_irqsave(&sq->lock, flags);
+               qlist_move_cache(&sq->qlist, &to_free, cache);
+               raw_spin_unlock_irqrestore(&sq->lock, flags);
        }
-#endif
+       qlist_free_all(&to_free, cache);
 
        raw_spin_lock_irqsave(&quarantine_lock, flags);
        for (i = 0; i < QUARANTINE_BATCHES; i++) {