x86/spinlocks/paravirt: Fix memory corruption on unlock
authorRaghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Fri, 6 Feb 2015 11:14:11 +0000 (16:44 +0530)
committerIngo Molnar <mingo@kernel.org>
Wed, 18 Feb 2015 13:53:49 +0000 (14:53 +0100)
Paravirt spinlock clears slowpath flag after doing unlock.
As explained by Linus currently it does:

                prev = *lock;
                add_smp(&lock->tickets.head, TICKET_LOCK_INC);

                /* add_smp() is a full mb() */

                if (unlikely(lock->tickets.tail & TICKET_SLOWPATH_FLAG))
                        __ticket_unlock_slowpath(lock, prev);

which is *exactly* the kind of things you cannot do with spinlocks,
because after you've done the "add_smp()" and released the spinlock
for the fast-path, you can't access the spinlock any more.  Exactly
because a fast-path lock might come in, and release the whole data
structure.

Linus suggested that we should not do any writes to lock after unlock(),
and we can move slowpath clearing to fastpath lock.

So this patch implements the fix with:

 1. Moving slowpath flag to head (Oleg):
    Unlocked locks don't care about the slowpath flag; therefore we can keep
    it set after the last unlock, and clear it again on the first (try)lock.
    -- this removes the write after unlock. note that keeping slowpath flag would
    result in unnecessary kicks.
    By moving the slowpath flag from the tail to the head ticket we also avoid
    the need to access both the head and tail tickets on unlock.

 2. use xadd to avoid read/write after unlock that checks the need for
    unlock_kick (Linus):
    We further avoid the need for a read-after-release by using xadd;
    the prev head value will include the slowpath flag and indicate if we
    need to do PV kicking of suspended spinners -- on modern chips xadd
    isn't (much) more expensive than an add + load.

Result:
 setup: 16core (32 cpu +ht sandy bridge 8GB 16vcpu guest)
 benchmark overcommit %improve
 kernbench  1x           -0.13
 kernbench  2x            0.02
 dbench     1x           -1.77
 dbench     2x           -0.63

[Jeremy: Hinted missing TICKET_LOCK_INC for kick]
[Oleg: Moved slowpath flag to head, ticket_equals idea]
[PeterZ: Added detailed changelog]

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Tested-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Jones <drjones@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Dave Jones <davej@redhat.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Fernando Luis Vázquez Cao <fernando_b1@lab.ntt.co.jp>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ulrich Obergfell <uobergfe@redhat.com>
Cc: Waiman Long <Waiman.Long@hp.com>
Cc: a.ryabinin@samsung.com
Cc: dave@stgolabs.net
Cc: hpa@zytor.com
Cc: jasowang@redhat.com
Cc: jeremy@goop.org
Cc: paul.gortmaker@windriver.com
Cc: riel@redhat.com
Cc: tglx@linutronix.de
Cc: waiman.long@hp.com
Cc: xen-devel@lists.xenproject.org
Link: http://lkml.kernel.org/r/20150215173043.GA7471@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/include/asm/spinlock.h
arch/x86/kernel/kvm.c
arch/x86/xen/spinlock.c

index 625660f8a2fcf0cb4b4b1a9216908c98fffe00a1..cf87de3fc39000eb21028ab2597d5187978bda4a 100644 (file)
@@ -46,7 +46,7 @@ static __always_inline bool static_key_false(struct static_key *key);
 
 static inline void __ticket_enter_slowpath(arch_spinlock_t *lock)
 {
-       set_bit(0, (volatile unsigned long *)&lock->tickets.tail);
+       set_bit(0, (volatile unsigned long *)&lock->tickets.head);
 }
 
 #else  /* !CONFIG_PARAVIRT_SPINLOCKS */
@@ -60,10 +60,30 @@ static inline void __ticket_unlock_kick(arch_spinlock_t *lock,
 }
 
 #endif /* CONFIG_PARAVIRT_SPINLOCKS */
+static inline int  __tickets_equal(__ticket_t one, __ticket_t two)
+{
+       return !((one ^ two) & ~TICKET_SLOWPATH_FLAG);
+}
+
+static inline void __ticket_check_and_clear_slowpath(arch_spinlock_t *lock,
+                                                       __ticket_t head)
+{
+       if (head & TICKET_SLOWPATH_FLAG) {
+               arch_spinlock_t old, new;
+
+               old.tickets.head = head;
+               new.tickets.head = head & ~TICKET_SLOWPATH_FLAG;
+               old.tickets.tail = new.tickets.head + TICKET_LOCK_INC;
+               new.tickets.tail = old.tickets.tail;
+
+               /* try to clear slowpath flag when there are no contenders */
+               cmpxchg(&lock->head_tail, old.head_tail, new.head_tail);
+       }
+}
 
 static __always_inline int arch_spin_value_unlocked(arch_spinlock_t lock)
 {
-       return lock.tickets.head == lock.tickets.tail;
+       return __tickets_equal(lock.tickets.head, lock.tickets.tail);
 }
 
 /*
@@ -87,18 +107,21 @@ static __always_inline void arch_spin_lock(arch_spinlock_t *lock)
        if (likely(inc.head == inc.tail))
                goto out;
 
-       inc.tail &= ~TICKET_SLOWPATH_FLAG;
        for (;;) {
                unsigned count = SPIN_THRESHOLD;
 
                do {
-                       if (READ_ONCE(lock->tickets.head) == inc.tail)
-                               goto out;
+                       inc.head = READ_ONCE(lock->tickets.head);
+                       if (__tickets_equal(inc.head, inc.tail))
+                               goto clear_slowpath;
                        cpu_relax();
                } while (--count);
                __ticket_lock_spinning(lock, inc.tail);
        }
-out:   barrier();      /* make sure nothing creeps before the lock is taken */
+clear_slowpath:
+       __ticket_check_and_clear_slowpath(lock, inc.head);
+out:
+       barrier();      /* make sure nothing creeps before the lock is taken */
 }
 
 static __always_inline int arch_spin_trylock(arch_spinlock_t *lock)
@@ -106,56 +129,30 @@ static __always_inline int arch_spin_trylock(arch_spinlock_t *lock)
        arch_spinlock_t old, new;
 
        old.tickets = READ_ONCE(lock->tickets);
-       if (old.tickets.head != (old.tickets.tail & ~TICKET_SLOWPATH_FLAG))
+       if (!__tickets_equal(old.tickets.head, old.tickets.tail))
                return 0;
 
        new.head_tail = old.head_tail + (TICKET_LOCK_INC << TICKET_SHIFT);
+       new.head_tail &= ~TICKET_SLOWPATH_FLAG;
 
        /* cmpxchg is a full barrier, so nothing can move before it */
        return cmpxchg(&lock->head_tail, old.head_tail, new.head_tail) == old.head_tail;
 }
 
-static inline void __ticket_unlock_slowpath(arch_spinlock_t *lock,
-                                           arch_spinlock_t old)
-{
-       arch_spinlock_t new;
-
-       BUILD_BUG_ON(((__ticket_t)NR_CPUS) != NR_CPUS);
-
-       /* Perform the unlock on the "before" copy */
-       old.tickets.head += TICKET_LOCK_INC;
-
-       /* Clear the slowpath flag */
-       new.head_tail = old.head_tail & ~(TICKET_SLOWPATH_FLAG << TICKET_SHIFT);
-
-       /*
-        * If the lock is uncontended, clear the flag - use cmpxchg in
-        * case it changes behind our back though.
-        */
-       if (new.tickets.head != new.tickets.tail ||
-           cmpxchg(&lock->head_tail, old.head_tail,
-                                       new.head_tail) != old.head_tail) {
-               /*
-                * Lock still has someone queued for it, so wake up an
-                * appropriate waiter.
-                */
-               __ticket_unlock_kick(lock, old.tickets.head);
-       }
-}
-
 static __always_inline void arch_spin_unlock(arch_spinlock_t *lock)
 {
        if (TICKET_SLOWPATH_FLAG &&
-           static_key_false(&paravirt_ticketlocks_enabled)) {
-               arch_spinlock_t prev;
+               static_key_false(&paravirt_ticketlocks_enabled)) {
+               __ticket_t head;
 
-               prev = *lock;
-               add_smp(&lock->tickets.head, TICKET_LOCK_INC);
+               BUILD_BUG_ON(((__ticket_t)NR_CPUS) != NR_CPUS);
 
-               /* add_smp() is a full mb() */
+               head = xadd(&lock->tickets.head, TICKET_LOCK_INC);
 
-               if (unlikely(lock->tickets.tail & TICKET_SLOWPATH_FLAG))
-                       __ticket_unlock_slowpath(lock, prev);
+               if (unlikely(head & TICKET_SLOWPATH_FLAG)) {
+                       head &= ~TICKET_SLOWPATH_FLAG;
+                       __ticket_unlock_kick(lock, (head + TICKET_LOCK_INC));
+               }
        } else
                __add(&lock->tickets.head, TICKET_LOCK_INC, UNLOCK_LOCK_PREFIX);
 }
@@ -164,14 +161,15 @@ static inline int arch_spin_is_locked(arch_spinlock_t *lock)
 {
        struct __raw_tickets tmp = READ_ONCE(lock->tickets);
 
-       return tmp.tail != tmp.head;
+       return !__tickets_equal(tmp.tail, tmp.head);
 }
 
 static inline int arch_spin_is_contended(arch_spinlock_t *lock)
 {
        struct __raw_tickets tmp = READ_ONCE(lock->tickets);
 
-       return (__ticket_t)(tmp.tail - tmp.head) > TICKET_LOCK_INC;
+       tmp.head &= ~TICKET_SLOWPATH_FLAG;
+       return (tmp.tail - tmp.head) > TICKET_LOCK_INC;
 }
 #define arch_spin_is_contended arch_spin_is_contended
 
@@ -183,16 +181,16 @@ static __always_inline void arch_spin_lock_flags(arch_spinlock_t *lock,
 
 static inline void arch_spin_unlock_wait(arch_spinlock_t *lock)
 {
-       __ticket_t head = ACCESS_ONCE(lock->tickets.head);
+       __ticket_t head = READ_ONCE(lock->tickets.head);
 
        for (;;) {
-               struct __raw_tickets tmp = ACCESS_ONCE(lock->tickets);
+               struct __raw_tickets tmp = READ_ONCE(lock->tickets);
                /*
                 * We need to check "unlocked" in a loop, tmp.head == head
                 * can be false positive because of overflow.
                 */
-               if (tmp.head == (tmp.tail & ~TICKET_SLOWPATH_FLAG) ||
-                   tmp.head != head)
+               if (__tickets_equal(tmp.head, tmp.tail) ||
+                               !__tickets_equal(tmp.head, head))
                        break;
 
                cpu_relax();
index 94f6434843008c38b8a091c077e0c94fc36c9a07..e354cc6446aba4286645dc799e48d8539c7005d6 100644 (file)
@@ -609,7 +609,7 @@ static inline void check_zero(void)
        u8 ret;
        u8 old;
 
-       old = ACCESS_ONCE(zero_stats);
+       old = READ_ONCE(zero_stats);
        if (unlikely(old)) {
                ret = cmpxchg(&zero_stats, old, 0);
                /* This ensures only one fellow resets the stat */
@@ -727,6 +727,7 @@ __visible void kvm_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
        int cpu;
        u64 start;
        unsigned long flags;
+       __ticket_t head;
 
        if (in_nmi())
                return;
@@ -768,11 +769,15 @@ __visible void kvm_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
         */
        __ticket_enter_slowpath(lock);
 
+       /* make sure enter_slowpath, which is atomic does not cross the read */
+       smp_mb__after_atomic();
+
        /*
         * check again make sure it didn't become free while
         * we weren't looking.
         */
-       if (ACCESS_ONCE(lock->tickets.head) == want) {
+       head = READ_ONCE(lock->tickets.head);
+       if (__tickets_equal(head, want)) {
                add_stats(TAKEN_SLOW_PICKUP, 1);
                goto out;
        }
@@ -803,8 +808,8 @@ static void kvm_unlock_kick(struct arch_spinlock *lock, __ticket_t ticket)
        add_stats(RELEASED_SLOW, 1);
        for_each_cpu(cpu, &waiting_cpus) {
                const struct kvm_lock_waiting *w = &per_cpu(klock_waiting, cpu);
-               if (ACCESS_ONCE(w->lock) == lock &&
-                   ACCESS_ONCE(w->want) == ticket) {
+               if (READ_ONCE(w->lock) == lock &&
+                   READ_ONCE(w->want) == ticket) {
                        add_stats(RELEASED_SLOW_KICKED, 1);
                        kvm_kick_cpu(cpu);
                        break;
index 23b45eb9a89ce4d56f8a73e181d004b107dc1f81..956374c1edbc31e4c1eb50c3fb29cb8828ad44b5 100644 (file)
@@ -41,7 +41,7 @@ static u8 zero_stats;
 static inline void check_zero(void)
 {
        u8 ret;
-       u8 old = ACCESS_ONCE(zero_stats);
+       u8 old = READ_ONCE(zero_stats);
        if (unlikely(old)) {
                ret = cmpxchg(&zero_stats, old, 0);
                /* This ensures only one fellow resets the stat */
@@ -112,6 +112,7 @@ __visible void xen_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
        struct xen_lock_waiting *w = this_cpu_ptr(&lock_waiting);
        int cpu = smp_processor_id();
        u64 start;
+       __ticket_t head;
        unsigned long flags;
 
        /* If kicker interrupts not initialized yet, just spin */
@@ -159,11 +160,15 @@ __visible void xen_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
         */
        __ticket_enter_slowpath(lock);
 
+       /* make sure enter_slowpath, which is atomic does not cross the read */
+       smp_mb__after_atomic();
+
        /*
         * check again make sure it didn't become free while
         * we weren't looking
         */
-       if (ACCESS_ONCE(lock->tickets.head) == want) {
+       head = READ_ONCE(lock->tickets.head);
+       if (__tickets_equal(head, want)) {
                add_stats(TAKEN_SLOW_PICKUP, 1);
                goto out;
        }
@@ -204,8 +209,8 @@ static void xen_unlock_kick(struct arch_spinlock *lock, __ticket_t next)
                const struct xen_lock_waiting *w = &per_cpu(lock_waiting, cpu);
 
                /* Make sure we read lock before want */
-               if (ACCESS_ONCE(w->lock) == lock &&
-                   ACCESS_ONCE(w->want) == next) {
+               if (READ_ONCE(w->lock) == lock &&
+                   READ_ONCE(w->want) == next) {
                        add_stats(RELEASED_SLOW_KICKED, 1);
                        xen_send_IPI_one(cpu, XEN_SPIN_UNLOCK_VECTOR);
                        break;