From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-5.3 required=3.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,HEADER_FROM_DIFFERENT_DOMAINS,MAILING_LIST_MULTI,NICE_REPLY_A, SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED,USER_AGENT_SANE_1 autolearn=no autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 8D4AAC2BBCD for ; Wed, 16 Dec 2020 13:23:23 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 498092339E for ; Wed, 16 Dec 2020 13:23:23 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726387AbgLPNXJ (ORCPT ); Wed, 16 Dec 2020 08:23:09 -0500 Received: from m43-15.mailgun.net ([69.72.43.15]:34866 "EHLO m43-15.mailgun.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1725550AbgLPNXJ (ORCPT ); Wed, 16 Dec 2020 08:23:09 -0500 DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=mg.codeaurora.org; q=dns/txt; s=smtp; t=1608124970; h=Content-Transfer-Encoding: Content-Type: In-Reply-To: MIME-Version: Date: Message-ID: From: References: Cc: To: Subject: Sender; bh=96YhEIx7Gv6MFOsRK65Pq35rWUCAJwR4mwE5jSXNGLo=; b=feeQKMCi4Dr80vp48LNFdEskyBITA69FAGhGNvmKa6ejzCcwEywhbpuk22C2U/uSrM1eQA1E xkYVCO791X/i9IQP2URwpc5YV3tR4at0IsmjotCRD67OPR2cN/B9UMWnouISznPUx9rDGraG wJEYpPTtkLUlcXe4PJw7ark6UEE= X-Mailgun-Sending-Ip: 69.72.43.15 X-Mailgun-Sid: WyI0MWYwYSIsICJsaW51eC1rZXJuZWxAdmdlci5rZXJuZWwub3JnIiwgImJlOWU0YSJd Received: from smtp.codeaurora.org (ec2-35-166-182-171.us-west-2.compute.amazonaws.com [35.166.182.171]) by smtp-out-n05.prod.us-west-2.postgun.com with SMTP id 5fda0a0cf5e9af65f814e6c3 (version=TLS1.2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256); Wed, 16 Dec 2020 13:22:20 GMT Sender: vjitta=codeaurora.org@mg.codeaurora.org Received: by smtp.codeaurora.org (Postfix, from userid 1001) id 1E19BC43467; Wed, 16 Dec 2020 13:22:20 +0000 (UTC) Received: from [192.168.0.105] (unknown [182.18.191.137]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) (Authenticated sender: vjitta) by smtp.codeaurora.org (Postfix) with ESMTPSA id 989FAC433CA; Wed, 16 Dec 2020 13:22:13 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 smtp.codeaurora.org 989FAC433CA Authentication-Results: aws-us-west-2-caf-mail-1.web.codeaurora.org; dmarc=none (p=none dis=none) header.from=codeaurora.org Authentication-Results: aws-us-west-2-caf-mail-1.web.codeaurora.org; spf=fail smtp.mailfrom=vjitta@codeaurora.org Subject: Re: [PATCH v3] lib: stackdepot: Add support to configure STACK_HASH_SIZE To: Alexander Potapenko Cc: Minchan Kim , Vincenzo Frascino , dan.j.williams@intel.com, broonie@kernel.org, Masami Hiramatsu , LKML , Andrew Morton , Andrey Konovalov , qcai@redhat.com, ylal@codeaurora.org, vinmenon@codeaurora.org References: <1607576401-25609-1-git-send-email-vjitta@codeaurora.org> <77e98f0b-c9c3-9380-9a57-ff1cd4022502@codeaurora.org> <6cc89f7b-bf40-2fd3-96ce-2a02d7535c91@codeaurora.org> <255400db-67d5-7f42-8dcb-9a440e006b9d@codeaurora.org> <7f2e171f-fa44-ef96-6cc6-14e615e3e457@codeaurora.org> From: Vijayanand Jitta Message-ID: <601d4b1a-8526-f7ad-d0f3-305894682109@codeaurora.org> Date: Wed, 16 Dec 2020 18:52:10 +0530 User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101 Thunderbird/78.5.1 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Language: en-GB Content-Transfer-Encoding: 7bit Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 12/16/2020 6:41 PM, Alexander Potapenko wrote: > On Wed, Dec 16, 2020 at 2:06 PM Vijayanand Jitta wrote: >> >> >> >> On 12/16/2020 1:56 PM, Alexander Potapenko wrote: >>> On Wed, Dec 16, 2020 at 4:43 AM Vijayanand Jitta wrote: >>>> >>>> >>>> >>>> On 12/14/2020 4:02 PM, Vijayanand Jitta wrote: >>>>> >>>>> >>>>> On 12/14/2020 3:04 PM, Alexander Potapenko wrote: >>>>>> On Mon, Dec 14, 2020 at 5:02 AM Vijayanand Jitta wrote: >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 12/11/2020 6:55 PM, Alexander Potapenko wrote: >>>>>>>> On Fri, Dec 11, 2020 at 1:45 PM Vijayanand Jitta wrote: >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> On 12/11/2020 2:06 PM, Alexander Potapenko wrote: >>>>>>>>>> On Thu, Dec 10, 2020 at 6:01 AM wrote: >>>>>>>>>>> >>>>>>>>>>> From: Yogesh Lal >>>>>>>>>>> >>>>>>>>>>> Add a kernel parameter stack_hash_order to configure STACK_HASH_SIZE. >>>>>>>>>>> >>>>>>>>>>> Aim is to have configurable value for STACK_HASH_SIZE, so that one >>>>>>>>>>> can configure it depending on usecase there by reducing the static >>>>>>>>>>> memory overhead. >>>>>>>>>>> >>>>>>>>>>> One example is of Page Owner, default value of STACK_HASH_SIZE lead >>>>>>>>>>> stack depot to consume 8MB of static memory. Making it configurable >>>>>>>>>>> and use lower value helps to enable features like CONFIG_PAGE_OWNER >>>>>>>>>>> without any significant overhead. >>>>>>>>>> >>>>>>>>>> Can we go with a static CONFIG_ parameter instead? >>>>>>>>>> Guess most users won't bother changing the default anyway, and for >>>>>>>>>> CONFIG_PAGE_OWNER users changing the size at boot time is not strictly >>>>>>>>>> needed. >>>>>>>>>> >>>>>>>>> Thanks for review. >>>>>>>>> >>>>>>>>> One advantage of having run time parameter is we can simply set it to a >>>>>>>>> lower value at runtime if page_owner=off thereby reducing the memory >>>>>>>>> usage or use default value if we want to use page owner so, we have some >>>>>>>>> some flexibility here. This is not possible with static parameter as we >>>>>>>>> have to have some predefined value. >>>>>>>> >>>>>>>> If we are talking about a configuration in which page_owner is the >>>>>>>> only stackdepot user in the system, then for page_owner=off it >>>>>>>> probably makes more sense to disable stackdepot completely instead of >>>>>>>> setting it to a lower value. This is a lot easier to do in terms of >>>>>>>> correctness. >>>>>>>> But if there are other users (e.g. KASAN), their stackdepot usage may >>>>>>>> actually dominate that of page_owner. >>>>>>>> >>>>>>>>>>> -static struct stack_record *stack_table[STACK_HASH_SIZE] = { >>>>>>>>>>> - [0 ... STACK_HASH_SIZE - 1] = NULL >>>>>>>>>>> +static unsigned int stack_hash_order = 20; >>>>>>>>>> >>>>>>>>>> Please initialize with MAX_STACK_HASH_ORDER instead. >>>>>>>>>> >>>>>>>>> >>>>>>>>> Sure, will update this. >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>>>>> + >>>>>>>>>>> +static int __init init_stackdepot(void) >>>>>>>>>>> +{ >>>>>>>>>>> + size_t size = (STACK_HASH_SIZE * sizeof(struct stack_record *)); >>>>>>>>>>> + >>>>>>>>>>> + stack_table = vmalloc(size); >>>>>>>>>>> + memcpy(stack_table, stack_table_def, size); >>>>>>>>>> >>>>>>>>>> Looks like you are assuming stack_table_def already contains some data >>>>>>>>>> by this point. >>>>>>>>>> But if STACK_HASH_SIZE shrinks this memcpy() above will just copy some >>>>>>>>>> part of the table, whereas the rest will be lost. >>>>>>>>>> We'll need to: >>>>>>>>>> - either explicitly decide we can afford losing this data (no idea how >>>>>>>>>> bad this can potentially be), >>>>>>>>>> - or disallow storing anything prior to full stackdepot initialization >>>>>>>>>> (then we don't need stack_table_def), >>>>>>>>>> - or carefully move all entries to the first part of the table. >>>>>>>>>> >>>>>>>>>> Alex >>>>>>>>>> >>>>>>>>> >>>>>>>>> The hash for stack_table_def is computed using the run time parameter >>>>>>>>> stack_hash_order, though stack_table_def is a bigger array it will only >>>>>>>>> use the entries that are with in the run time configured STACK_HASH_SIZE >>>>>>>>> range. so, there will be no data loss during copy. >>>>>>>> >>>>>>>> Do we expect any data to be stored into stack_table_def before >>>>>>>> setup_stack_hash_order() is called? >>>>>>>> If the answer is no, then we could probably drop stack_table_def and >>>>>>>> allocate the table right in setup_stack_hash_order()? >>>>>>>> >>>>>>> >>>>>>> Yes, we do see an allocation from stack depot even before init is called >>>>>>> from kasan, thats the reason for having stack_table_def. >>>>>>> This is the issue reported due to that on v2, so i added stack_table_def. >>>>>>> https://lkml.org/lkml/2020/12/3/839 >>>>>> >>>>>> But at that point STACK_HASH_SIZE is still equal to 1L << >>>>>> MAX_STACK_HASH_ORDER, isn't it? >>>>>> Then we still need to take care of the records that fit into the >>>>>> bigger array, but not the smaller one. >>>>>> >>>>> >>>>> At this point early_param is already called which sets stack_hash_order. >>>>> So, STACK_HASH_SIZE will be set to this updated value and not >>>>> MAX_STACK_HASH_SIZE.So, no additional entires in the bigger array. >>>>> >>>>> Thanks, >>>>> Vijay >>>>> >>>> >>>> Let me know if there are any other concerns. >>> >>> I still think there are implicit assumptions that should at least be >>> written down in the comments. >> >> Sure, will add the comments. >> >>> As far as I understand the code, here is what happens here: >>> >>> 0. No stacks are recorded. >>> 1. early_param is called to set stack_hash_order to a value >>> potentially smaller than MAX_STACK_HASH_SIZE. >>> 2. KASAN (or other users) records some stacks into stack_table_def >>> (capped at new STACK_HASH_SIZE) >>> 3. init_stackdepot() allocates a new stack_table and copies the >>> contents of stack_table_def into it >>> 4. Further stacks are recorded into the new stack_table >>> >>> If this is how the things work, I agree we don't need to account for >>> the part of stack_table_def past STACK_HASH_SIZE. >>> Not allocating stack_table when setting stack_hash_order is probably >>> also justified, as we don't have SLAB or vmalloc at that point. >>> >> >> That's Right. >> >>> But I am still curious if a runtime parameter that disables the >>> stackdepot completely will solve your problem. >>> Allocating a small amount of memory when you actually don't want to >>> allocate any sounds suboptimal. >>> >> >> I think disabling stack depot completely should be fine, we can make >> STACK_HASH_SIZE as runtime parameter instead of stack_hash_order and set >> stack_hash_size to 0 to disable stack depot completely. > > To reiterate, I think you don't need a tunable stack_hash_order > parameter if the only use case is to disable the stack depot. > Maybe it is enough to just add a boolean flag? There are multiple users of stackdepot they might still want to use stack depot but with a lower memory footprint instead of MAX_SIZE so, a configurable size might help here ? > Or even go further and disable the stack depot in the same place that > disables page owner, as the user probably doesn't want to set two > flags instead of one? > Since, page owner is not the only user of stack depot we can't take that decision of disabling stack depot if page owner is disabled. >> Minchan, >> This should be fine right ? Do you see any issue with disabling >> stack depot completely ? >> >> Thanks, >> Vijay >> >>>> Thanks, >>>> Vijay >>>> >>>>>>> Thanks, >>>>>>> Vijay >>>>>>> >>>>>>>>> Thanks, >>>>>>>>> Vijay >>>>>>>>> >>>>>>>>> -- >>>>>>>>> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a >>>>>>>>> member of Code Aurora Forum, hosted by The Linux Foundation >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> >>>>>>> -- >>>>>>> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a >>>>>>> member of Code Aurora Forum, hosted by The Linux Foundation >>>>>> >>>>>> >>>>>> >>>>> >>>> >>>> -- >>>> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a >>>> member of Code Aurora Forum, hosted by The Linux Foundation >>> >>> >>> >> >> -- >> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a >> member of Code Aurora Forum, hosted by The Linux Foundation > > > -- QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation