On Mon, 2019-02-25 at 17:31 +-0200, Gal Pressman wrote: +AD4 On 25-Feb-19 16:54, Bart Van Assche wrote: +AD4 +AD4 On 2/24/19 10:52 PM, Leon Romanovsky wrote: +AD4 +AD4 +AD4 diff --git a/drivers/infiniband/core/cm+AF8-msgs.h +AD4 +AD4 +AD4 b/drivers/infiniband/core/cm+AF8-msgs.h +AD4 +AD4 +AD4 index 476d4309576d..3d16d614aff6 100644 +AD4 +AD4 +AD4 --- a/drivers/infiniband/core/cm+AF8-msgs.h +AD4 +AD4 +AD4 +-+-+- b/drivers/infiniband/core/cm+AF8-msgs.h +AD4 +AD4 +AD4 +AEAAQA -98,7 +-98,7 +AEAAQA struct cm+AF8-req+AF8-msg +AHs +AD4 +AD4 +AD4 +AD4 +AD4 +AD4 u32 private+AF8-data+AFs-IB+AF8-CM+AF8-REQ+AF8-PRIVATE+AF8-DATA+AF8-SIZE / sizeof(u32)+AF0AOw +AD4 +AD4 +AD4 +AD4 +AD4 +AD4 -+AH0 +AF8AXw-attribute+AF8AXw ((packed))+ADs +AD4 +AD4 +AD4 +-+AH0 +AF8AXw-packed+ADs +AD4 +AD4 +AD4 +AD4 Hi Leon, +AD4 +AD4 +AD4 +AD4 A previous RDMA maintainer once wrote a blog in which he explained why it is +AD4 +AD4 better to apply the +AF8AXw-packed attribute only to data members that need it instead +AD4 +AD4 of to data structures. The reason is that applying +AF8AXw-packed to data structures +AD4 +AD4 does not allow compilers to generate optimal code for architectures that cannot +AD4 +AD4 perform non-aligned multi-byte reads efficiently. Does that concern apply to the +AD4 +AD4 data structures modified by this patch? +AD4 +AD4 By any chance do you have a link to this post? Hi Gal, Sorry but I can't find that page anymore. Maybe it has been removed. Bart.
