On Wed, 19 Dec 2018 22:31:00 -0500 Steven Rostedt <rostedt@xxxxxxxxxxx> wrote: > > index 000000000000..8c08eaefbc08 > > --- /dev/null > > +++ b/include/trace/events/ib_mad.h > > @@ -0,0 +1,193 @@ > > +/* > > + * Copyright (c) 2018 Intel Corporation. All rights reserved. > > + * > > + * This software is available to you under a choice of one of two > > + * licenses. You may choose to be licensed under the terms of the GNU > > + * General Public License (GPL) Version 2, available from the file > > + * COPYING in the main directory of this source tree, or the > > + * OpenIB.org BSD license below: > > Please use: > > // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB > > At the top of the file, and do not add other License text in the file. > > > +#undef TRACE_SYSTEM > > +#define TRACE_SYSTEM ib_mad > > + > > +#if !defined(_TRACE_IB_MAD_H) || defined(TRACE_HEADER_MULTI_READ) > > +#define _TRACE_IB_MAD_H > > + > > +#include <linux/tracepoint.h> > > +#include <rdma/ib_mad.h> > > + > > +DECLARE_EVENT_CLASS(ib_mad_send_template, > > + TP_PROTO(struct ib_mad_send_wr_private *wr, struct rdma_mad_trace_addr *addr), > > + TP_ARGS(wr, addr), > > + > > + TP_STRUCT__entry( > > + __array(char, dev_name, IB_DEVICE_NAME_MAX) > > IB_DEVICE_NAME_MAX is 64 bytes. That's quite a bit, and much bigger > than what it would be normally. Using __string() which is dynamic is a > better option. > > > + __field(u8, port_num) > > + __field(u32, qp_num) > > Order of fields by size matters, because architectures will likely pad > them wasting ring buffer space. > > Group u8 by 4 or 8, and try to keep pointers and u64 on 8 byte > alignments (assuming a 64 bit architecture.) > > The current order will produce a lot of unnecessary padding by the > compiler. > These comments go for the next 4 patches that have the same issues. -- Steve
