From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751748AbeCJCjL (ORCPT ); Fri, 9 Mar 2018 21:39:11 -0500 Received: from mail.kernel.org ([198.145.29.99]:49034 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751373AbeCJCjJ (ORCPT ); Fri, 9 Mar 2018 21:39:09 -0500 DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 9102221771 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=goodmis.org Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=rostedt@goodmis.org Message-Id: <20180310023442.791997138@goodmis.org> User-Agent: quilt/0.63-1 Date: Fri, 09 Mar 2018 21:34:42 -0500 From: Steven Rostedt To: linux-kernel@vger.kernel.org Cc: Ingo Molnar , Andrew Morton , Al Viro , Tom Zanussi , Namhyung Kim , Masami Hiramatsu , Jiri Olsa Subject: [PATCH 0/3] tracing: Rewrite the function filter code Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Al Viro reviewed the ftrace event filter logic and was horrified. He wrote Tom and I a lengthy private email with a formal proof on how to do it simpler as well as more efficient. Currently the filter logic creates a binary search tree (with some optimizations), and walks it to get a result. Say we had the following filter: a && !(!b || c) || d && !e Where each letter is a predicate. A tree would be made to look something like: || / \ && && / | | \ a !|| d !e / \ !b c If a == 1, b == 1, c == 1, d == 1, e == 0, then all nodes would be walked. || -> && -> a -> && -> !|| -> !b -> !|| -> c -> !|| -> && -> || -> && -> d -> && -> !e -> && -> || -> return TRUE! That's 16 steps across vertices. Al's method would create a program using an array that denotes the predicate function, when to branch, and a target to branch to if the value is the same as when to branch. Storing the array as: prog[0] = { a, 0, 2}; // predicate, when_to_branch, target prog[1] = { b, 0, 2}; prog[2] = { c, 0, 4}; prog[3] = { d, 0, 5}; prog[4] = { e, 1, 5}; prog[5] = { NULL, 0, 1}; // TRUE prog[6] = { NULL, 0, 0}; // FALSE Where the code to execute the above looks like: for (i = 0; prog[i].pred; i++) { struct filter_pred *pred = prog[i].pred; int match = pred->fn(pred, rec); if (match == prog[i].when_to_branch) i = prog[i].target; } return prog[i].target; Which translates the above program array into: n0: if (!a) goto n3; n1: if (!b) goto n3; n2: if (!c) goto T; n3: if (!d) goto F; n4: if (e) goto F; T: return TRUE; F: return FALSE; And the above example only takes 6 steps to return TRUE. Special thanks goes out to Al for his patience and his time that he spent in educating us in a proper logical parser. Two patches were added to do some initial clean up. The last patch implements Al's suggestions. I wrote up a very lengthy comment describing what Al taught us in my own words (hopefully I got it right), and the implementation is similar to what Al showed with a few changes (again, hopefully I got that right too). I tested this with lots of different filters and it looks to be holding up. Jiri, This affects perf as well. I ran some tests and I believe I got it woking as it does today. Steven Rostedt (VMware) (3): tracing: Combine enum and arrays into single macro in filter code tracing: Clean up and document pred_funcs_##type creation and use tracing: Rewrite filter logic to be simpler and faster ---- kernel/trace/trace.h | 15 +- kernel/trace/trace_events_filter.c | 2372 ++++++++++++++++-------------------- 2 files changed, 1083 insertions(+), 1304 deletions(-)