Skip to contentSkip to conversation

xen/spinlock: move debug helpers inside the locked regions

PATCH In review Roger Pau Monne 2 participants 5 messages Jul 29, 2020 – Jul 30, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen/spinlock: move debug helpers inside the locked regions

Debug helpers such as lock profiling or the invariant pCPU assertions must strictly be performed inside the exclusive locked region, or else races might happen.

Note the issue was not strictly introduced by the pointed commit in the Fixes tag, since lock stats where already incremented before the barrier, but that commit made it more apparent as manipulating the cpu field could happen outside of the locked regions and thus trigger the BUG_ON. This is only enabled on debug builds, and thus releases are not affected.

Fixes 80cba391a35 ('spinlocks: in debug builds store cpu holding the lock')
---
 xen/common/spinlock.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
xen/common/spinlock.c +6 -6
index 17f4519fc7..ce3106e2d3 100644
@@ -170,9 +170,9 @@ void inline _spin_lock_cb(spinlock_t *lock, void (*cb)(void *), void *data)
             cb(data);
         arch_lock_relax();
     }
+    arch_lock_acquire_barrier();
     got_lock(&lock->debug);
     LOCK_PROFILE_GOT;
-    arch_lock_acquire_barrier();
 }
 
 void _spin_lock(spinlock_t *lock)
@@ -198,9 +198,9 @@ unsigned long _spin_lock_irqsave(spinlock_t *lock)
 
 void _spin_unlock(spinlock_t *lock)
 {
-    arch_lock_release_barrier();
     LOCK_PROFILE_REL;
     rel_lock(&lock->debug);
+    arch_lock_release_barrier();
     add_sized(&lock->tickets.head, 1);
     arch_lock_signal();
     preempt_enable();
@@ -249,15 +249,15 @@ int _spin_trylock(spinlock_t *lock)
         preempt_enable();
         return 0;
     }
+    /*
+     * cmpxchg() is a full barrier so no need for an
+     * arch_lock_acquire_barrier().
+     */
     got_lock(&lock->debug);
 #ifdef CONFIG_DEBUG_LOCK_PROFILE
     if (lock->profile)
         lock->profile->time_locked = NOW();
 #endif
-    /*
-     * cmpxchg() is a full barrier so no need for an
-     * arch_lock_acquire_barrier().
-     */
     return 1;
 }
 
Signature · 1 line
2.27.0
#1 lore raw

Hi Roger,

Quoted 9 lines · Roger Pau Monne

Debug helpers such as lock profiling or the invariant pCPU assertions must strictly be performed inside the exclusive locked region, or else races might happen.

Note the issue was not strictly introduced by the pointed commit in the Fixes tag, since lock stats where already incremented before the barrier, but that commit made it more apparent as manipulating the cpu field could happen outside of the locked regions and thus trigger the BUG_ON.

From the wording, it is not entirely clear which BUG_ON() you are referring to. I am guessing, it is the one in rel_lock(). Am I correct?

Otherwise, the change looks good to me.

Cheers,

Quoted 57 lines

This is only enabled on debug builds, and thus releases are not affected.

Fixes 80cba391a35 ('spinlocks: in debug builds store cpu holding the lock')
---
  xen/common/spinlock.c | 12 ++++++------
  1 file changed, 6 insertions(+), 6 deletions(-)
xen/common/spinlock.c +6 -6
index 17f4519fc7..ce3106e2d3 100644
@@ -170,9 +170,9 @@ void inline _spin_lock_cb(spinlock_t *lock, void (*cb)(void *), void *data)
              cb(data);
          arch_lock_relax();
      }
+    arch_lock_acquire_barrier();
      got_lock(&lock->debug);
      LOCK_PROFILE_GOT;
-    arch_lock_acquire_barrier();
  }
 
  void _spin_lock(spinlock_t *lock)
@@ -198,9 +198,9 @@ unsigned long _spin_lock_irqsave(spinlock_t *lock)
 
  void _spin_unlock(spinlock_t *lock)
  {
-    arch_lock_release_barrier();
      LOCK_PROFILE_REL;
      rel_lock(&lock->debug);
+    arch_lock_release_barrier();
      add_sized(&lock->tickets.head, 1);
      arch_lock_signal();
      preempt_enable();
@@ -249,15 +249,15 @@ int _spin_trylock(spinlock_t *lock)
          preempt_enable();
          return 0;
      }
+    /*
+     * cmpxchg() is a full barrier so no need for an
+     * arch_lock_acquire_barrier().
+     */
      got_lock(&lock->debug);
  #ifdef CONFIG_DEBUG_LOCK_PROFILE
      if (lock->profile)
          lock->profile->time_locked = NOW();
  #endif
-    /*
-     * cmpxchg() is a full barrier so no need for an
-     * arch_lock_acquire_barrier().
-     */
      return 1;
  }
 
Signature · 1 line
Julien Grall
#2 lore raw
Quoted 7 lines · Julien Grall

Hi Roger,

Quoted 4 lines · Roger Pau Monne

Debug helpers such as lock profiling or the invariant pCPU assertions must strictly be performed inside the exclusive locked region, or else races might happen.

Quoted 8 lines
Quoted 5 lines

Note the issue was not strictly introduced by the pointed commit in the Fixes tag, since lock stats where already incremented before the barrier, but that commit made it more apparent as manipulating the cpu field could happen outside of the locked regions and thus trigger the BUG_ON.

From the wording, it is not entirely clear which BUG_ON() you are referring to. I am guessing, it is the one in rel_lock(). Am I correct?

Yes, that's right. Expanding to:

"... and thus trigger the BUG_ON in rel_lock()." would be better.

Quoted 1 line

Otherwise, the change looks good to me.

Thanks.

Reviewed-by
#3 lore raw

Hi Roger,

Quoted 14 lines · Roger Pau Monné
Quoted 13 lines · Julien Grall

Hi Roger,

Quoted 9 lines · Roger Pau Monne

Debug helpers such as lock profiling or the invariant pCPU assertions must strictly be performed inside the exclusive locked region, or else races might happen.

Note the issue was not strictly introduced by the pointed commit in the Fixes tag, since lock stats where already incremented before the barrier, but that commit made it more apparent as manipulating the cpu field could happen outside of the locked regions and thus trigger the BUG_ON.

Quoted 6 lines
Quoted 2 lines

From the wording, it is not entirely clear which BUG_ON() you are referring
to. I am guessing, it is the one in rel_lock(). Am I correct?

Yes, that's right. Expanding to:

"... and thus trigger the BUG_ON in rel_lock()." would be better.

Looks good to me. With that:

I am happy to do the update on commit if there is no more comments.

Cheers,

Signature · 1 line
Julien Grall
#4 lore raw
Quoted 22 lines · Julien Grall

Hi Roger,

Quoted 19 lines · Roger Pau Monné
Quoted 7 lines · Julien Grall

Hi Roger,

Quoted 4 lines · Roger Pau Monne

Debug helpers such as lock profiling or the invariant pCPU assertions must strictly be performed inside the exclusive locked region, or else races might happen.

Quoted 8 lines
Quoted 5 lines

Note the issue was not strictly introduced by the pointed commit in the Fixes tag, since lock stats where already incremented before the barrier, but that commit made it more apparent as manipulating the cpu field could happen outside of the locked regions and thus trigger the BUG_ON.

 From the wording, it is not entirely clear which BUG_ON() you are referring to. I am guessing, it is the one in rel_lock(). Am I correct?

Yes, that's right. Expanding to:

Quoted 7 lines
Quoted 1 line

"...  and thus trigger the BUG_ON in rel_lock()." would be better.

Looks good to me. With that:

I am happy to do the update on commit if there is no more comments.

Committed.

Thank you!

Cheers,

Signature · 1 line
Julien Grall