x86/qspinlock: Fix compile error

With a compiler that has asm-goto but not asm-cc-output and
CONFIG_PROFILE_ALL_BRANCHES=y we get a compiler error:

  arch/x86/include/asm/rmwcc.h:23:17: error: jump into statement expression

Fix this by writing the if() as a boolean multiplication instead.

Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-kernel@vger.kernel.org
Fixes: 7aa54be ("locking/qspinlock, x86: Provide liveness guarantee")
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Author: Peter Zijlstra

arch/x86/include/asm/qspinlock.h

@@ -13,12 +13,15 @@
 #define queued_fetch_set_pending_acquire queued_fetch_set_pending_acquire
 static __always_inline u32 queued_fetch_set_pending_acquire(struct qspinlock *lock)
 {
-	u32 val = 0;
-
-	if (GEN_BINARY_RMWcc(LOCK_PREFIX "btsl", lock->val.counter, c,
-			     "I", _Q_PENDING_OFFSET))
-		val |= _Q_PENDING_VAL;
+	u32 val;
 
+	/*
+	 * We can't use GEN_BINARY_RMWcc() inside an if() stmt because asm goto
+	 * and CONFIG_PROFILE_ALL_BRANCHES=y results in a label inside a
+	 * statement expression, which GCC doesn't like.
+	 */
+	val = GEN_BINARY_RMWcc(LOCK_PREFIX "btsl", lock->val.counter, c,
+			       "I", _Q_PENDING_OFFSET) * _Q_PENDING_VAL;
 	val |= atomic_read(&lock->val) & ~_Q_PENDING_MASK;
 
 	return val;