rcu: Fix data-race due to atomic_t copy-by-value
authorMarco Elver <elver@google.com>
Wed, 9 Oct 2019 15:57:43 +0000 (17:57 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 24 Feb 2020 07:36:23 +0000 (08:36 +0100)
[ Upstream commit 6cf539a87a61a4fbc43f625267dbcbcf283872ed ]

This fixes a data-race where `atomic_t dynticks` is copied by value. The
copy is performed non-atomically, resulting in a data-race if `dynticks`
is updated concurrently.

This data-race was found with KCSAN:
==================================================================
BUG: KCSAN: data-race in dyntick_save_progress_counter / rcu_irq_enter

write to 0xffff989dbdbe98e0 of 4 bytes by task 10 on cpu 3:
 atomic_add_return include/asm-generic/atomic-instrumented.h:78 [inline]
 rcu_dynticks_snap kernel/rcu/tree.c:310 [inline]
 dyntick_save_progress_counter+0x43/0x1b0 kernel/rcu/tree.c:984
 force_qs_rnp+0x183/0x200 kernel/rcu/tree.c:2286
 rcu_gp_fqs kernel/rcu/tree.c:1601 [inline]
 rcu_gp_fqs_loop+0x71/0x880 kernel/rcu/tree.c:1653
 rcu_gp_kthread+0x22c/0x3b0 kernel/rcu/tree.c:1799
 kthread+0x1b5/0x200 kernel/kthread.c:255
 <snip>

read to 0xffff989dbdbe98e0 of 4 bytes by task 154 on cpu 7:
 rcu_nmi_enter_common kernel/rcu/tree.c:828 [inline]
 rcu_irq_enter+0xda/0x240 kernel/rcu/tree.c:870
 irq_enter+0x5/0x50 kernel/softirq.c:347
 <snip>

Reported by Kernel Concurrency Sanitizer on:
CPU: 7 PID: 154 Comm: kworker/7:1H Not tainted 5.3.0+ #5
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
Workqueue: kblockd blk_mq_run_work_fn
==================================================================

Signed-off-by: Marco Elver <elver@google.com>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: rcu@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
include/trace/events/rcu.h
kernel/rcu/tree.c

index 694bd04..fdd31c5 100644 (file)
@@ -442,7 +442,7 @@ TRACE_EVENT_RCU(rcu_fqs,
  */
 TRACE_EVENT_RCU(rcu_dyntick,
 
-       TP_PROTO(const char *polarity, long oldnesting, long newnesting, atomic_t dynticks),
+       TP_PROTO(const char *polarity, long oldnesting, long newnesting, int dynticks),
 
        TP_ARGS(polarity, oldnesting, newnesting, dynticks),
 
@@ -457,7 +457,7 @@ TRACE_EVENT_RCU(rcu_dyntick,
                __entry->polarity = polarity;
                __entry->oldnesting = oldnesting;
                __entry->newnesting = newnesting;
-               __entry->dynticks = atomic_read(&dynticks);
+               __entry->dynticks = dynticks;
        ),
 
        TP_printk("%s %lx %lx %#3x", __entry->polarity,
index 8110514..62e5959 100644 (file)
@@ -576,7 +576,7 @@ static void rcu_eqs_enter(bool user)
        }
 
        lockdep_assert_irqs_disabled();
-       trace_rcu_dyntick(TPS("Start"), rdp->dynticks_nesting, 0, rdp->dynticks);
+       trace_rcu_dyntick(TPS("Start"), rdp->dynticks_nesting, 0, atomic_read(&rdp->dynticks));
        WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current));
        rdp = this_cpu_ptr(&rcu_data);
        do_nocb_deferred_wakeup(rdp);
@@ -649,14 +649,15 @@ static __always_inline void rcu_nmi_exit_common(bool irq)
         * leave it in non-RCU-idle state.
         */
        if (rdp->dynticks_nmi_nesting != 1) {
-               trace_rcu_dyntick(TPS("--="), rdp->dynticks_nmi_nesting, rdp->dynticks_nmi_nesting - 2, rdp->dynticks);
+               trace_rcu_dyntick(TPS("--="), rdp->dynticks_nmi_nesting, rdp->dynticks_nmi_nesting - 2,
+                                 atomic_read(&rdp->dynticks));
                WRITE_ONCE(rdp->dynticks_nmi_nesting, /* No store tearing. */
                           rdp->dynticks_nmi_nesting - 2);
                return;
        }
 
        /* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */
-       trace_rcu_dyntick(TPS("Startirq"), rdp->dynticks_nmi_nesting, 0, rdp->dynticks);
+       trace_rcu_dyntick(TPS("Startirq"), rdp->dynticks_nmi_nesting, 0, atomic_read(&rdp->dynticks));
        WRITE_ONCE(rdp->dynticks_nmi_nesting, 0); /* Avoid store tearing. */
 
        if (irq)
@@ -743,7 +744,7 @@ static void rcu_eqs_exit(bool user)
        rcu_dynticks_task_exit();
        rcu_dynticks_eqs_exit();
        rcu_cleanup_after_idle();
-       trace_rcu_dyntick(TPS("End"), rdp->dynticks_nesting, 1, rdp->dynticks);
+       trace_rcu_dyntick(TPS("End"), rdp->dynticks_nesting, 1, atomic_read(&rdp->dynticks));
        WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && !user && !is_idle_task(current));
        WRITE_ONCE(rdp->dynticks_nesting, 1);
        WARN_ON_ONCE(rdp->dynticks_nmi_nesting);
@@ -827,7 +828,7 @@ static __always_inline void rcu_nmi_enter_common(bool irq)
        }
        trace_rcu_dyntick(incby == 1 ? TPS("Endirq") : TPS("++="),
                          rdp->dynticks_nmi_nesting,
-                         rdp->dynticks_nmi_nesting + incby, rdp->dynticks);
+                         rdp->dynticks_nmi_nesting + incby, atomic_read(&rdp->dynticks));
        WRITE_ONCE(rdp->dynticks_nmi_nesting, /* Prevent store tearing. */
                   rdp->dynticks_nmi_nesting + incby);
        barrier();