s390/ctl_reg: make __ctl_load a full memory barrier
authorHeiko Carstens <heiko.carstens@de.ibm.com>
Wed, 28 Dec 2016 10:33:48 +0000 (11:33 +0100)
committerMartin Schwidefsky <schwidefsky@de.ibm.com>
Mon, 16 Jan 2017 06:27:48 +0000 (07:27 +0100)
We have quite a lot of code that depends on the order of the
__ctl_load inline assemby and subsequent memory accesses, like
e.g. disabling lowcore protection and the writing to lowcore.

Since the __ctl_load macro does not have memory barrier semantics, nor
any other dependencies the compiler is, theoretically, free to shuffle
code around. Or in other words: storing to lowcore could happen before
lowcore protection is disabled.

In order to avoid this class of potential bugs simply add a full
memory barrier to the __ctl_load macro.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
arch/s390/include/asm/ctl_reg.h

index d7697ab802f6c94813a27394baa255fa26a93ddc..8e136b88cdf4f13460b960f8db9d02e1ae88324f 100644 (file)
@@ -15,7 +15,9 @@
        BUILD_BUG_ON(sizeof(addrtype) != (high - low + 1) * sizeof(long));\
        asm volatile(                                                   \
                "       lctlg   %1,%2,%0\n"                             \
-               : : "Q" (*(addrtype *)(&array)), "i" (low), "i" (high));\
+               :                                                       \
+               : "Q" (*(addrtype *)(&array)), "i" (low), "i" (high)    \
+               : "memory");                                            \
 }
 
 #define __ctl_store(array, low, high) {                                        \