1 #ifndef __ASM_QSPINLOCK_PARAVIRT_H 2 #define __ASM_QSPINLOCK_PARAVIRT_H 3 4 /* 5 * For x86-64, PV_CALLEE_SAVE_REGS_THUNK() saves and restores 8 64-bit 6 * registers. For i386, however, only 1 32-bit register needs to be saved 7 * and restored. So an optimized version of __pv_queued_spin_unlock() is 8 * hand-coded for 64-bit, but it isn't worthwhile to do it for 32-bit. 9 */ 10 #ifdef CONFIG_64BIT 11 12 PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath); 13 #define __pv_queued_spin_unlock __pv_queued_spin_unlock 14 #define PV_UNLOCK "__raw_callee_save___pv_queued_spin_unlock" 15 #define PV_UNLOCK_SLOWPATH "__raw_callee_save___pv_queued_spin_unlock_slowpath" 16 17 /* 18 * Optimized assembly version of __raw_callee_save___pv_queued_spin_unlock 19 * which combines the registers saving trunk and the body of the following 20 * C code: 21 * 22 * void __pv_queued_spin_unlock(struct qspinlock *lock) 23 * { 24 * struct __qspinlock *l = (void *)lock; 25 * u8 lockval = cmpxchg(&l->locked, _Q_LOCKED_VAL, 0); 26 * 27 * if (likely(lockval == _Q_LOCKED_VAL)) 28 * return; 29 * pv_queued_spin_unlock_slowpath(lock, lockval); 30 * } 31 * 32 * For x86-64, 33 * rdi = lock (first argument) 34 * rsi = lockval (second argument) 35 * rdx = internal variable (set to 0) 36 */ 37 asm (".pushsection .text;" 38 ".globl " PV_UNLOCK ";" 39 ".type " PV_UNLOCK ", @function;" 40 ".align 4,0x90;" 41 PV_UNLOCK ": " 42 FRAME_BEGIN 43 "push %rdx;" 44 "mov $0x1,%eax;" 45 "xor %edx,%edx;" 46 "lock cmpxchg %dl,(%rdi);" 47 "cmp $0x1,%al;" 48 "jne .slowpath;" 49 "pop %rdx;" 50 FRAME_END 51 "ret;" 52 ".slowpath: " 53 "push %rsi;" 54 "movzbl %al,%esi;" 55 "call " PV_UNLOCK_SLOWPATH ";" 56 "pop %rsi;" 57 "pop %rdx;" 58 FRAME_END 59 "ret;" 60 ".size " PV_UNLOCK ", .-" PV_UNLOCK ";" 61 ".popsection"); 62 63 #else /* CONFIG_64BIT */ 64 65 extern void __pv_queued_spin_unlock(struct qspinlock *lock); 66 PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock); 67 68 #endif /* CONFIG_64BIT */ 69 #endif 70