From patchwork Thu Apr 25 09:45:14 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Thomas Gleixner X-Patchwork-Id: 10916703 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 84CB913B5 for ; Thu, 25 Apr 2019 10:02:07 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 7599C28A08 for ; Thu, 25 Apr 2019 10:02:07 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 67CC128C4E; Thu, 25 Apr 2019 10:02:07 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.9 required=2.0 tests=BAYES_00,MAILING_LIST_MULTI, RCVD_IN_DNSWL_HI autolearn=unavailable version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id E6BD928A08 for ; Thu, 25 Apr 2019 10:02:06 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1729489AbfDYKA0 (ORCPT ); Thu, 25 Apr 2019 06:00:26 -0400 Received: from Galois.linutronix.de ([146.0.238.70]:57986 "EHLO Galois.linutronix.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1728739AbfDYKAZ (ORCPT ); Thu, 25 Apr 2019 06:00:25 -0400 Received: from localhost ([127.0.0.1] helo=nanos.tec.linutronix.de) by Galois.linutronix.de with esmtp (Exim 4.80) (envelope-from ) id 1hJbAD-0001wo-Fc; Thu, 25 Apr 2019 11:59:33 +0200 Message-Id: <20190425094803.066064076@linutronix.de> User-Agent: quilt/0.65 Date: Thu, 25 Apr 2019 11:45:14 +0200 From: Thomas Gleixner To: LKML Cc: Josh Poimboeuf , x86@kernel.org, Andy Lutomirski , Steven Rostedt , Alexander Potapenko , Alexey Dobriyan , Andrew Morton , Christoph Lameter , Pekka Enberg , linux-mm@kvack.org, David Rientjes , Catalin Marinas , Dmitry Vyukov , Andrey Ryabinin , kasan-dev@googlegroups.com, Mike Rapoport , Akinobu Mita , Christoph Hellwig , iommu@lists.linux-foundation.org, Robin Murphy , Marek Szyprowski , Johannes Thumshirn , David Sterba , Chris Mason , Josef Bacik , linux-btrfs@vger.kernel.org, dm-devel@redhat.com, Mike Snitzer , Alasdair Kergon , Daniel Vetter , intel-gfx@lists.freedesktop.org, Joonas Lahtinen , Maarten Lankhorst , dri-devel@lists.freedesktop.org, David Airlie , Jani Nikula , Rodrigo Vivi , Tom Zanussi , Miroslav Benes , linux-arch@vger.kernel.org Subject: [patch V3 21/29] tracing: Use percpu stack trace buffer more intelligently References: <20190425094453.875139013@linutronix.de> MIME-Version: 1.0 Sender: linux-btrfs-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-btrfs@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP The per cpu stack trace buffer usage pattern is odd at best. The buffer has place for 512 stack trace entries on 64-bit and 1024 on 32-bit. When interrupts or exceptions nest after the per cpu buffer was acquired the stacktrace length is hardcoded to 8 entries. 512/1024 stack trace entries in kernel stacks are unrealistic so the buffer is a complete waste. Split the buffer into 4 nest levels, which are 128/256 entries per level. This allows nesting contexts (interrupts, exceptions) to utilize the cpu buffer for stack retrieval and avoids the fixed length allocation along with the conditional execution pathes. Signed-off-by: Thomas Gleixner Cc: Steven Rostedt --- V3: Limit to 4 nest levels and increase size per level. --- kernel/trace/trace.c | 77 +++++++++++++++++++++++++-------------------------- 1 file changed, 39 insertions(+), 38 deletions(-) --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2749,12 +2749,21 @@ trace_function(struct trace_array *tr, #ifdef CONFIG_STACKTRACE -#define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long)) +/* Allow 4 levels of nesting: normal, softirq, irq, NMI */ +#define FTRACE_KSTACK_NESTING 4 + +#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING) + struct ftrace_stack { - unsigned long calls[FTRACE_STACK_MAX_ENTRIES]; + unsigned long calls[FTRACE_KSTACK_ENTRIES]; +}; + + +struct ftrace_stacks { + struct ftrace_stack stacks[FTRACE_KSTACK_NESTING]; }; -static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack); +static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks); static DEFINE_PER_CPU(int, ftrace_stack_reserve); static void __ftrace_trace_stack(struct ring_buffer *buffer, @@ -2763,10 +2772,11 @@ static void __ftrace_trace_stack(struct { struct trace_event_call *call = &event_kernel_stack; struct ring_buffer_event *event; + struct ftrace_stack *fstack; struct stack_entry *entry; struct stack_trace trace; - int use_stack; - int size = FTRACE_STACK_ENTRIES; + int size = FTRACE_KSTACK_ENTRIES; + int stackidx; trace.nr_entries = 0; trace.skip = skip; @@ -2788,29 +2798,32 @@ static void __ftrace_trace_stack(struct */ preempt_disable_notrace(); - use_stack = __this_cpu_inc_return(ftrace_stack_reserve); + stackidx = __this_cpu_inc_return(ftrace_stack_reserve); + + /* This should never happen. If it does, yell once and skip */ + if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING)) + goto out; + /* - * We don't need any atomic variables, just a barrier. - * If an interrupt comes in, we don't care, because it would - * have exited and put the counter back to what we want. - * We just need a barrier to keep gcc from moving things - * around. + * The above __this_cpu_inc_return() is 'atomic' cpu local. An + * interrupt will either see the value pre increment or post + * increment. If the interrupt happens pre increment it will have + * restored the counter when it returns. We just need a barrier to + * keep gcc from moving things around. */ barrier(); - if (use_stack == 1) { - trace.entries = this_cpu_ptr(ftrace_stack.calls); - trace.max_entries = FTRACE_STACK_MAX_ENTRIES; - - if (regs) - save_stack_trace_regs(regs, &trace); - else - save_stack_trace(&trace); - - if (trace.nr_entries > size) - size = trace.nr_entries; - } else - /* From now on, use_stack is a boolean */ - use_stack = 0; + + fstack = this_cpu_ptr(ftrace_stacks.stacks) + (stackidx - 1); + trace.entries = fstack->calls; + trace.max_entries = FTRACE_KSTACK_ENTRIES; + + if (regs) + save_stack_trace_regs(regs, &trace); + else + save_stack_trace(&trace); + + if (trace.nr_entries > size) + size = trace.nr_entries; size *= sizeof(unsigned long); @@ -2820,19 +2833,7 @@ static void __ftrace_trace_stack(struct goto out; entry = ring_buffer_event_data(event); - memset(&entry->caller, 0, size); - - if (use_stack) - memcpy(&entry->caller, trace.entries, - trace.nr_entries * sizeof(unsigned long)); - else { - trace.max_entries = FTRACE_STACK_ENTRIES; - trace.entries = entry->caller; - if (regs) - save_stack_trace_regs(regs, &trace); - else - save_stack_trace(&trace); - } + memcpy(&entry->caller, trace.entries, size); entry->size = trace.nr_entries;