From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Ahern Subject: Re: perf probe and bitfields Date: Tue, 30 Jan 2018 09:53:19 -0700 Message-ID: <2a2ba32f-2c71-d81d-252c-dffe0ca1c002@gmail.com> References: <52f06b83-1065-dc88-756e-7f8fffc5c21e@gmail.com> <20180126065656.GD22737@sejong> <20180127140459.8e54eefcf9d74609cbcaa739@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Return-path: Received: from mail-pf0-f177.google.com ([209.85.192.177]:45076 "EHLO mail-pf0-f177.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753467AbeA3QxV (ORCPT ); Tue, 30 Jan 2018 11:53:21 -0500 Received: by mail-pf0-f177.google.com with SMTP id a88so9590329pfe.12 for ; Tue, 30 Jan 2018 08:53:21 -0800 (PST) In-Reply-To: <20180127140459.8e54eefcf9d74609cbcaa739@gmail.com> Content-Language: en-US Sender: linux-perf-users-owner@vger.kernel.org List-ID: To: Masami Hiramatsu Cc: Namhyung Kim , Arnaldo Carvalho de Melo , Masami Hiramatsu , Perf Users , kernel-team@lge.com On 1/26/18 10:04 PM, Masami Hiramatsu wrote: > On Fri, 26 Jan 2018 09:28:49 -0700 > David Ahern wrote: > >> On 1/25/18 11:56 PM, Namhyung Kim wrote: >>> Hi David, >>> >>> On Fri, Jan 19, 2018 at 08:43:50AM -0800, David Ahern wrote: >>>> Hi Arnaldo / Masami: >>>> >>>> Are there known issues / limitations with perf probe and bitfields? >>>> >>>> Here's an example: pkt_type is a bitfield in sk_buff. When I put a probe >>>> in arp_rcv to examine the pkt_type value: >>>> >>>> $ perf probe 'arp_rcv ptype=skb->pkt_type dev=dev->name:string' >>> >>> Bitfield type takes 3 parameters according to the Doc/perf-probe.txt. >>> Maybe it's like 'ptype=skb->pkt_type:b3@0/8' ? >> >> missed that. Thanks, Namyhung. > > > > Hmm, if you have debuginfo, it must be automatically translated. > E.g. > > $ sudo perf probe -D "arp_rcv ptype=skb->pkt_type dev=dev->name:string" > p:probe/arp_rcv _text+8135904 ptype=+144(%di):b3@5/8 dev=+0(%si):string > > So as you can see the ptype automatically has ":b3@5/8" types. I was expecting it to automatically figure that out. The kernels in question were custom built and the vmlinux must not have been available. Thanks the tip.