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=-3.8 required=3.0 tests=BAYES_00,DKIM_INVALID, DKIM_SIGNED,HEADER_FROM_DIFFERENT_DOMAINS,MAILING_LIST_MULTI,SPF_HELO_NONE, SPF_PASS 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 E3B64C433E1 for ; Wed, 15 Jul 2020 13:43:57 +0000 (UTC) Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id B04582064B for ; Wed, 15 Jul 2020 13:43:57 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=fail reason="signature verification failed" (1024-bit key) header.d=redhat.com header.i=@redhat.com header.b="JvV9vqGW" DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org B04582064B Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=redhat.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Received: from localhost ([::1]:47826 helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1jvhhU-0007cX-Uw for qemu-devel@archiver.kernel.org; Wed, 15 Jul 2020 09:43:56 -0400 Received: from eggs.gnu.org ([2001:470:142:3::10]:36922) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1jvhgs-0006le-4a for qemu-devel@nongnu.org; Wed, 15 Jul 2020 09:43:18 -0400 Received: from us-smtp-1.mimecast.com ([205.139.110.61]:25404 helo=us-smtp-delivery-1.mimecast.com) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_CBC_SHA1:256) (Exim 4.90_1) (envelope-from ) id 1jvhgq-0003NK-7w for qemu-devel@nongnu.org; Wed, 15 Jul 2020 09:43:17 -0400 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1594820595; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=iddFP7qzUpodDIp+EqIkNpYRdZ0ENhMaUuUZuS7vTg8=; b=JvV9vqGWzAxZyGNsx2diAJhDEofR3To+DCkvB7ENOMfyqAgajY7Xud2zWxUW9ei6J4p2bO Z8SGwwXBrBZylJlO8G2IM1VntTqMqZzAVhqVs+V8o2ccDB+PFdG49PmZmO/nQj2goQy4GC gRkqj8R3yErXGRfQAdiCHSzf8qYcPqk= Received: from mimecast-mx01.redhat.com (mimecast-mx01.redhat.com [209.132.183.4]) (Using TLS) by relay.mimecast.com with ESMTP id us-mta-454-c-LEQVoROHG1q2aXo6zLbg-1; Wed, 15 Jul 2020 09:43:13 -0400 X-MC-Unique: c-LEQVoROHG1q2aXo6zLbg-1 Received: from smtp.corp.redhat.com (int-mx04.intmail.prod.int.phx2.redhat.com [10.5.11.14]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx01.redhat.com (Postfix) with ESMTPS id 8CBB3106B242; Wed, 15 Jul 2020 13:43:12 +0000 (UTC) Received: from localhost (unknown [10.40.208.62]) by smtp.corp.redhat.com (Postfix) with ESMTP id 2F3DC5D9CA; Wed, 15 Jul 2020 13:43:10 +0000 (UTC) Date: Wed, 15 Jul 2020 15:43:09 +0200 From: Igor Mammedov To: Laszlo Ersek Subject: Re: [RFC 3/3] x68: acpi: trigger SMI before scanning for hotplugged CPUs Message-ID: <20200715154309.42e2ccd8@redhat.com> In-Reply-To: <0fd38252-b16d-fee8-31de-71e35475e3bc@redhat.com> References: <20200710161704.309824-1-imammedo@redhat.com> <20200710161704.309824-4-imammedo@redhat.com> <515cc231-858a-a626-31a9-d74e1f6b4e38@redhat.com> <20200714171935.10507f90@redhat.com> <0fd38252-b16d-fee8-31de-71e35475e3bc@redhat.com> MIME-Version: 1.0 X-Scanned-By: MIMEDefang 2.79 on 10.5.11.14 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Received-SPF: pass client-ip=205.139.110.61; envelope-from=imammedo@redhat.com; helo=us-smtp-delivery-1.mimecast.com X-detected-operating-system: by eggs.gnu.org: First seen = 2020/07/15 02:37:03 X-ACL-Warn: Detected OS = Linux 2.2.x-3.x [generic] [fuzzy] X-Spam_score_int: -40 X-Spam_score: -4.1 X-Spam_bar: ---- X-Spam_report: (-4.1 / 5.0 requ) BAYES_00=-1.9, DKIMWL_WL_HIGH=-1, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_NONE=-0.0001, RCVD_IN_MSPIKE_H2=-1, SPF_HELO_NONE=0.001, SPF_PASS=-0.001 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: boris.ostrovsky@oracle.com, Peter Krempa , liran.alon@oracle.com, qemu-devel@nongnu.org Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: "Qemu-devel" On Wed, 15 Jul 2020 14:38:00 +0200 Laszlo Ersek wrote: > On 07/14/20 17:19, Igor Mammedov wrote: > > On Tue, 14 Jul 2020 14:41:28 +0200 > > Laszlo Ersek wrote: > > > >> On 07/14/20 14:28, Laszlo Ersek wrote: > >>> (CC'ing Peter Krempa due to virsh setvcpu (singular) / setvcpus (plural) > >>> references) > >>> > >>> On 07/10/20 18:17, Igor Mammedov wrote: > >>>> In case firmware has negotiated CPU hotplug SMI feature, generate > >>>> AML to describe SMI IO port region and send SMI to firmware > >>>> on each CPU hotplug SCI. > >>>> > >>>> It might be not really usable, but should serve as a starting point to > >>>> discuss how better to deal with split hotplug sequence during hot-add > >>>> ( > >>>> ex scenario where it will break is: > >>>> hot-add > >>>> -> (QEMU) add CPU in hotplug regs > >>>> -> (QEMU) SCI > >>>> -1-> (OS) scan > >>>> -1-> (OS) SMI > >>>> -1-> (FW) pull in CPU1 *** > >>>> -1-> (OS) start iterating hotplug regs > >>>> hot-add > >>>> -> (QEMU) add CPU in hotplug regs > >>>> -> (QEMU) SCI > >>>> -2-> (OS) scan (blocked on mutex till previous scan is finished) > >>>> -1-> (OS) 1st added CPU1 send device check event -> INIT/SIPI > >>>> -1-> (OS) 1st added CPU2 send device check event -> INIT/SIPI > >>>> that's where it explodes, since FW didn't see CPU2 > >>>> when SMI was called > >>>> ) > >>>> > >>>> hot remove will throw in yet another set of problems, so lets discuss > >>>> both here and see if we can really share hotplug registers block between > >>>> FW and AML or we should do something else with it. > >>> > >>> This issue is generally triggered by management applications such as > >>> libvirt that issue device_add commands in quick succession. For libvirt, > >>> the command is "virsh setvcpus" (plural) with multiple CPUs specified > >>> for plugging. The singular "virsh setvcpu" command, which is more > >>> friendly towards guest NUMA, does not run into the symptom. > >>> > >>> The scope of the scan method lock is not large enough, with SMI in the > >>> picture. > >>> > >>> I suggest that we not uproot the existing AML code or the hotplug > >>> register block. Instead, I suggest that we add serialization at a higher > >>> level, with sufficient scope. > >>> > >>> QEMU: > >>> > >>> - introduce a new flag standing for "CPU plug operation in progress" > >>> > >>> - if ICH9_LPC_SMI_F_BROADCAST_BIT has been negotiated: > >>> > >>> - "device_add" and "device_del" should enforce > >>> ICH9_LPC_SMI_F_CPU_HOTPLUG_BIT and > >>> ICH9_LPC_SMI_F_CPU_HOT_UNPLUG_BIT, respectively > >>> > >>> - both device_add and device_del (for VCPUs) should set check the > >>> "in progress" flag. > >>> > >>> - If set, reject the request synchronously > >>> > >>> - Otherwise, set the flag, and commence the operation > >>> > >>> - in cpu_hotplug_wr(), where we emit the ACPI_DEVICE_OST event with > >>> qapi_event_send_acpi_device_ost(), clear the "in-progress" flag. > >>> > >>> - If QEMU executes the QMP command processing and the cpu_hotplug_wr() > >>> function on different (host OS) threads, then perhaps we should use an > >>> atomic type for the flag. (Not sure about locking between QEMU threads, > >>> sorry.) I don't really expect race conditions, but in case we ever get > >>> stuck with the flag, we should make sure that the stuck state is "in > >>> progress", and not "not in progress". (The former state can prevent > >>> further plug operations, but cannot cause the guest to lose state.) > >> > >> Furthermore, the "CPU plug operation in progress" flag should be: > >> - either migrated, > >> - or a migration blocker. > >> > >> Because on the destination host, device_add should be possible if and > >> only if the plug operation completed (either still on the source host, > >> or on the destination host). > > > > I have a way more simple alternative idea, which doesn't involve libvirt. > > > > We can change AML to > > 1. cache hotplugged CPUs from controller > > 2. send SMI > > 3. send Notify event to OS to online cached CPUs > > this way we never INIT/SIPI cpus that FW hasn't seen yet > > as for FW, it can relocate extra CPU that arrived after #1 > > it won't cause any harm as on the next SCI AML will pick up those > > CPUs and SMI upcall will be just NOP. > > > > I'll post a patch here on top of this series for you to try > > (without any of your comments addressed yet, as it's already written > > and I was testing it for a while to make sure it won't explode > > with various windows versions) > > Sounds good, I'll be happy to test it. > > Indeed "no event" is something that the fw deals with gracefully. (IIRC > I wanted to cover a "spurious SMI" explicitly.) is it possible to distinguish "spurious SMI" vs hotplug SMI, if yes then we probably do not care about any other SMIs except hotplug one. > It didn't occur to me that you could dynamically size e.g. a package > object in AML. Based on my reading of the ACPI spec, "VarPackageOp" can > take a *runtime* "NumElements", so if you did two loops, the first loop > could count the pending stuff, and then a VarPackageOp could be used > with just the right NumElements... Anyway, I digress :) well, it's mine field since Windows implement only a subset of spec and VarPackageOp is not avalable on all version that support hotplug. I think, I've narrowed language down to supported subset, so I need to complete another round of testing to see if I didn't break anything by accident. > > > >> I guess that the "migration blocker" option is easier. > >> > >> Anyway I assume this is already handled with memory hotplug somehow > >> (i.e., migration attempt between device_add and ACPI_DEVICE_OST). > > > > Thanks for comments, > > I'll need some time to ponder on other comments and do some > > palaeontology research to answer questions > > (aka. I need to make up excuses for the code I wrote :) ) > > haha, thanks :) > Laszlo