]> git.ipfire.org Git - thirdparty/linux.git/commitdiff
refscale: Fix uninitalized use of wait_queue_head_t
authorWaiman Long <longman@redhat.com>
Fri, 7 Jul 2023 17:53:55 +0000 (13:53 -0400)
committerPaul E. McKenney <paulmck@kernel.org>
Fri, 14 Jul 2023 22:01:04 +0000 (15:01 -0700)
Running the refscale test occasionally crashes the kernel with the
following error:

[ 8569.952896] BUG: unable to handle page fault for address: ffffffffffffffe8
[ 8569.952900] #PF: supervisor read access in kernel mode
[ 8569.952902] #PF: error_code(0x0000) - not-present page
[ 8569.952904] PGD c4b048067 P4D c4b049067 PUD c4b04b067 PMD 0
[ 8569.952910] Oops: 0000 [#1] PREEMPT_RT SMP NOPTI
[ 8569.952916] Hardware name: Dell Inc. PowerEdge R750/0WMWCR, BIOS 1.2.4 05/28/2021
[ 8569.952917] RIP: 0010:prepare_to_wait_event+0x101/0x190
  :
[ 8569.952940] Call Trace:
[ 8569.952941]  <TASK>
[ 8569.952944]  ref_scale_reader+0x380/0x4a0 [refscale]
[ 8569.952959]  kthread+0x10e/0x130
[ 8569.952966]  ret_from_fork+0x1f/0x30
[ 8569.952973]  </TASK>

The likely cause is that init_waitqueue_head() is called after the call to
the torture_create_kthread() function that creates the ref_scale_reader
kthread.  Although this init_waitqueue_head() call will very likely
complete before this kthread is created and starts running, it is
possible that the calling kthread will be delayed between the calls to
torture_create_kthread() and init_waitqueue_head().  In this case, the
new kthread will use the waitqueue head before it is properly initialized,
which is not good for the kernel's health and well-being.

The above crash happened here:

static inline void __add_wait_queue(...)
{
:
if (!(wq->flags & WQ_FLAG_PRIORITY)) <=== Crash here

The offset of flags from list_head entry in wait_queue_entry is
-0x18. If reader_tasks[i].wq.head.next is NULL as allocated reader_task
structure is zero initialized, the instruction will try to access address
0xffffffffffffffe8, which is exactly the fault address listed above.

This commit therefore invokes init_waitqueue_head() before creating
the kthread.

Fixes: 653ed64b01dc ("refperf: Add a test to measure performance of read-side synchronization")
Signed-off-by: Waiman Long <longman@redhat.com>
Reviewed-by: Qiuxu Zhuo <qiuxu.zhuo@intel.com>
Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
Acked-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
kernel/rcu/refscale.c

index 1970ce5f22d405bc075015f5b8a4ec1f1635d1ee..71d138573856f66f78ebcb938b9848bb01e5ee56 100644 (file)
@@ -1107,12 +1107,11 @@ ref_scale_init(void)
        VERBOSE_SCALEOUT("Starting %d reader threads", nreaders);
 
        for (i = 0; i < nreaders; i++) {
+               init_waitqueue_head(&reader_tasks[i].wq);
                firsterr = torture_create_kthread(ref_scale_reader, (void *)i,
                                                  reader_tasks[i].task);
                if (torture_init_error(firsterr))
                        goto unwind;
-
-               init_waitqueue_head(&(reader_tasks[i].wq));
        }
 
        // Main Task