0001-stp_utrace-disable-IRQs-when-holding-the-bucket-spin.patch

From 936d8f252fd1ea29f280cf1398e4f71ba2df4001 Mon Sep 17 00:00:00 2001
From: Sultan Alsawaf <sultan@openresty.com>
Date: Tue, 10 Nov 2020 10:03:34 -0800
Subject: [PATCH] stp_utrace: disable IRQs when holding the bucket spin lock

This lock can be acquired from inside an IRQ, leading to a deadlock:

WARNING: inconsistent lock state
4.14.35-1902.6.6.el7uek.x86_64.debug #2 Tainted: G           OE
--------------------------------
inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
sh/15779 [HC1[1]:SC0[0]:HE0:SE1] takes:
 (&(lock)->rlock#3){?.+.}, at: [<ffffffffc0c080b0>] _stp_mempool_alloc+0x35/0xab [orxray_lj_lua_fgraph_XXXXXXX]
{HARDIRQ-ON-W} state was registered at:
  lock_acquire+0xe0/0x238
  _raw_spin_lock+0x3d/0x7a
  utrace_task_alloc+0xa4/0xe3 [orxray_lj_lua_fgraph_XXXXXXX]
  utrace_attach_task+0x136/0x194 [orxray_lj_lua_fgraph_XXXXXXX]
  __stp_utrace_attach+0x57/0x216 [orxray_lj_lua_fgraph_XXXXXXX]
  stap_start_task_finder+0x12e/0x33f [orxray_lj_lua_fgraph_XXXXXXX]
  systemtap_module_init+0x114d/0x11f0 [orxray_lj_lua_fgraph_XXXXXXX]
  _stp_handle_start+0xea/0x1c5 [orxray_lj_lua_fgraph_XXXXXXX]
  _stp_ctl_write_cmd+0x28d/0x2d1 [orxray_lj_lua_fgraph_XXXXXXX]
  full_proxy_write+0x67/0xbb
  __vfs_write+0x3a/0x170
  vfs_write+0xc7/0x1c0
  SyS_write+0x58/0xbf
  do_syscall_64+0x7e/0x22c
  entry_SYSCALL_64_after_hwframe+0x16e/0x0
irq event stamp: 9454
hardirqs last  enabled at (9453): [<ffffffffa696c960>] _raw_write_unlock_irqrestore+0x40/0x67
hardirqs last disabled at (9454): [<ffffffffa6a05417>] apic_timer_interrupt+0x1c7/0x1d1
softirqs last  enabled at (9202): [<ffffffffa6c00361>] __do_softirq+0x361/0x4e5
softirqs last disabled at (9195): [<ffffffffa60aeb76>] irq_exit+0xf6/0x102

other info that might help us debug this:
 Possible unsafe locking scenario:

       CPU0
       ----
  lock(&(lock)->rlock#3);
  <Interrupt>
    lock(&(lock)->rlock#3);

 *** DEADLOCK ***

no locks held by sh/15779.

stack backtrace:
CPU: 16 PID: 15779 Comm: sh Tainted: G           OE   4.14.35-1902.6.6.el7uek.x86_64.debug #2
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-2.fc32 04/01/2014
Call Trace:
 <IRQ>
 dump_stack+0x81/0xb6
 print_usage_bug+0x1fc/0x20d
 ? check_usage_backwards+0x130/0x12b
 mark_lock+0x1f8/0x27b
 __lock_acquire+0x6e7/0x165a
 ? sched_clock_local+0x18/0x81
 ? perf_swevent_hrtimer+0x136/0x151
 lock_acquire+0xe0/0x238
 ? _stp_mempool_alloc+0x35/0xab [orxray_lj_lua_fgraph_XXXXXXX]
 _raw_spin_lock_irqsave+0x55/0x97
 ? _stp_mempool_alloc+0x35/0xab [orxray_lj_lua_fgraph_XXXXXXX]
 _stp_mempool_alloc+0x35/0xab [orxray_lj_lua_fgraph_XXXXXXX]
 _stp_ctl_get_buffer+0x69/0x215 [orxray_lj_lua_fgraph_XXXXXXX]
 _stp_ctl_send+0x4e/0x169 [orxray_lj_lua_fgraph_XXXXXXX]
 _stp_vlog+0xac/0x143 [orxray_lj_lua_fgraph_XXXXXXX]
 ? _stp_utrace_probe_cb+0xa4/0xa4 [orxray_lj_lua_fgraph_XXXXXXX]
 _stp_warn+0x6a/0x88 [orxray_lj_lua_fgraph_XXXXXXX]
 function___global_warn__overload_0+0x60/0xac [orxray_lj_lua_fgraph_XXXXXXX]
 probe_67+0xce/0x10e [orxray_lj_lua_fgraph_XXXXXXX]
 _stp_hrtimer_notify_function+0x2db/0x55f [orxray_lj_lua_fgraph_XXXXXXX]
 __hrtimer_run_queues+0x132/0x5c5
 hrtimer_interrupt+0xb7/0x1ca
 smp_apic_timer_interrupt+0xa5/0x35a
 apic_timer_interrupt+0x1cc/0x1d1
 </IRQ>
---
 runtime/stp_utrace.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/runtime/stp_utrace.c b/runtime/stp_utrace.c
index 126db1869..064dbf190 100644
--- a/runtime/stp_utrace.c
+++ b/runtime/stp_utrace.c
@@ -490,9 +490,9 @@ static int utrace_exit(void)
 		rcu_read_lock();
 		stap_hlist_for_each_entry_rcu(utrace, node, &bucket->head, hlist) {
 			utrace->freed = true;
-			stp_spin_lock(&bucket->lock);
+			stp_spin_lock_irqsave(&bucket->lock, flags);
 			hlist_del_rcu(&utrace->hlist);
-			stp_spin_unlock(&bucket->lock);
+			stp_spin_unlock_irqrestore(&bucket->lock, flags);
 
 			utrace_cleanup(utrace);
 		}
@@ -724,6 +724,7 @@ static struct utrace *utrace_task_alloc(struct utrace_bucket *bucket,
 					struct task_struct *task)
 {
 	struct utrace *utrace;
+	unsigned long flags;
 
 	utrace = kmem_cache_zalloc(utrace_cachep, STP_ALLOC_FLAGS);
 	if (unlikely(!utrace))
@@ -739,9 +740,9 @@ static struct utrace *utrace_task_alloc(struct utrace_bucket *bucket,
 	atomic_set(&utrace->resume_work_added, 0);
 	atomic_set(&utrace->report_work_added, 0);
 
-	stp_spin_lock(&bucket->lock);
+	stp_spin_lock_irqsave(&bucket->lock, flags);
 	hlist_add_head_rcu(&utrace->hlist, &bucket->head);
-	stp_spin_unlock(&bucket->lock);
+	stp_spin_unlock_irqrestore(&bucket->lock, flags);
 	return utrace;
 }
 
@@ -768,15 +769,17 @@ static struct utrace *utrace_task_alloc(struct utrace_bucket *bucket,
  */
 static void utrace_free(struct utrace_bucket *bucket, struct utrace *utrace)
 {
+	unsigned long flags;
+
 	if (unlikely(!utrace))
 		return;
 
 	/* Remove this utrace from the mapping list of tasks to
 	 * struct utrace. */
 	utrace->freed = true;
-	stp_spin_lock(&bucket->lock);
+	stp_spin_lock_irqsave(&bucket->lock, flags);
 	hlist_del_rcu(&utrace->hlist);
-	stp_spin_unlock(&bucket->lock);
+	stp_spin_unlock_irqrestore(&bucket->lock, flags);
 
 	/* Free the utrace struct. */
 #ifdef STP_TF_DEBUG
-- 
2.29.2