Am 16.01.2015 um 00:09 schrieb Michael Ellerman:

On Thu, 2015-01-15 at 09:58 +0100, Christian Borntraeger wrote:

quoted

ACCESS_ONCE does not work reliably on non-scalar types. For
example gcc 4.6 and 4.7 might remove the volatile tag for such
accesses during the SRA (scalar replacement of aggregates) step
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58145)

Change the ppc/kvm code to replace ACCESS_ONCE with READ_ONCE.

Signed-off-by: Christian Borntraeger <redacted>
---
 arch/powerpc/kvm/book3s_hv_rm_xics.c |  8 ++++----
 arch/powerpc/kvm/book3s_xics.c       | 16 ++++++++--------
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/arch/powerpc/kvm/book3s_hv_rm_xics.c b/arch/powerpc/kvm/book3s_hv_rm_xics.c
index 7b066f6..7c22997 100644
--- a/arch/powerpc/kvm/book3s_hv_rm_xics.c
+++ b/arch/powerpc/kvm/book3s_hv_rm_xics.c

@@ -152,7 +152,7 @@ static void icp_rm_down_cppr(struct kvmppc_xics *xics, struct kvmppc_icp *icp,
 	 * in virtual mode.
 	 */
 	do {
-		old_state = new_state = ACCESS_ONCE(icp->state);
+		old_state = new_state = READ_ONCE(icp->state);

These are all icp->state.

Which is a union, but it's only the size of unsigned long. So in practice there
shouldn't be a bug here right?

This bug was that gcc lost the volatile tag when propagating aggregates to scalar types.
So in theory a union could be affected. See the original problem
 ( http://marc.info/?i=54611D86.4040306%40de.ibm.com ) 
which happened on 

union ipte_control {
        unsigned long val;
        struct {
                unsigned long k  : 1;
                unsigned long kh : 31;
                unsigned long kg : 32;
        };
};

Christian