Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org 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 smtp.lore.kernel.org (Postfix) with ESMTPS id 263A3EF48C4 for ; Sat, 14 Feb 2026 13:56:12 +0000 (UTC) Received: from localhost ([::1] helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1vrG7e-0008Uo-5v; Sat, 14 Feb 2026 08:55:46 -0500 Received: from eggs.gnu.org ([2001:470:142:3::10]) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1vrG7U-0008UZ-Sx for qemu-devel@nongnu.org; Sat, 14 Feb 2026 08:55:38 -0500 Received: from zero.eik.bme.hu ([2001:738:2001:2001::2001]) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1vrG7R-0007Ud-7L for qemu-devel@nongnu.org; Sat, 14 Feb 2026 08:55:36 -0500 Received: from localhost (localhost [127.0.0.1]) by zero.eik.bme.hu (Postfix) with ESMTP id 0C2DA596A0B; Sat, 14 Feb 2026 14:55:29 +0100 (CET) X-Virus-Scanned: amavis at eik.bme.hu Received: from zero.eik.bme.hu ([127.0.0.1]) by localhost (zero.eik.bme.hu [127.0.0.1]) (amavis, port 10028) with ESMTP id lNwNHznHUnN0; Sat, 14 Feb 2026 14:55:26 +0100 (CET) Received: by zero.eik.bme.hu (Postfix, from userid 432) id 72DB45969F6; Sat, 14 Feb 2026 14:55:26 +0100 (CET) Received: from localhost (localhost [127.0.0.1]) by zero.eik.bme.hu (Postfix) with ESMTP id 7111F5969F0; Sat, 14 Feb 2026 14:55:26 +0100 (CET) Date: Sat, 14 Feb 2026 14:55:26 +0100 (CET) From: BALATON Zoltan To: Peter Xu cc: Paolo Bonzini , qemu-devel , Akihiko Odaki , Mark Cave-Ayland , Gerd Hoffmann , Max Filippov , Peter Maydell , =?ISO-8859-15?Q?Philippe_Mathieu-Daud=E9?= Subject: Re: [PATCH v3 0/6] Implement memory_region_new_* functions In-Reply-To: Message-ID: <73c23787-83da-77fa-c501-a88c5faf4eaf@eik.bme.hu> References: <42233b06-f058-152d-645e-e16528a1640d@eik.bme.hu> MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII; format=flowed Received-SPF: pass client-ip=2001:738:2001:2001::2001; envelope-from=balaton@eik.bme.hu; helo=zero.eik.bme.hu X-Spam_score_int: -18 X-Spam_score: -1.9 X-Spam_bar: - X-Spam_report: (-1.9 / 5.0 requ) BAYES_00=-1.9, 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: 2.1.29 Precedence: list List-Id: qemu development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org On Sat, 14 Feb 2026, BALATON Zoltan wrote: > On Thu, 12 Feb 2026, BALATON Zoltan wrote: >> Memory regions are registered as a child with an owner and when the owner >> is destroyed it unparents the memory region which decreases its ref count >> so it would be freed if there are no other uses of the memory region. This >> is implemented and documented like that but there's currently no way to >> actually create a memory region that would be managed and freed by QOM that >> way and a memory region created with memory_region_init would never be >> freed and the code to manage memory region lifecycle is effectively dead. >> Because of this we have all sorts of hacks to solve problems that arise >> from this such as memory_region_ref/unref not using the ref count of the >> memory region itself but the ref count of the owner and we embed memory >> regions in devices to avoid leaking them. > > This is an example of hacks that I think arise from not being able to create > a memory region that is freed with its parent: > > https://lists.nongnu.org/archive/html/qemu-devel/2026-02/msg03876.html > > This code does a few strange things meddling with internals of memory region > trying to handle what the ref counts should already take care of. It also > abuses the free function for something that I'd expect to be done in a > finalize or unrealize method. I think this could be simplified with QOM > managed memory regions that this series allow even without untangling memory > region ref count from the owner but I don't know that code so not trying to > do a fix. With the current memory_region_init we need to add more hacks to > make it work but that's both confusing and fragile. > >> or drop an unneeded field from PCIHostState as proposed here: >> >> https://patchew.org/QEMU/cover.1761232472.git.balaton@eik.bme.hu/797587b52159d2db79ed924b4962be7e7ba84207.1761232472.git.balaton@eik.bme.hu/ > > Another point I'd like to clear up here. For the linked patch I was told to > do this: > > struct PREPPCIState { > PCIHostState parent_obj; > > MemoryRegion mmfg; > } > > static void raven_pcihost_init(Object *obj) > { > PREPPCIState *s = RAVEN_PCI_HOST_BRIDGE(obj); > PCIHostState *h = PCI_HOST_BRIDGE(dev); > > memory_region_init_io(&s->mmcfg, OBJECT(h), &raven_mmcfg_ops, h->bus, > "pci-mmcfg", 0x00400000); > } > > static void raven_pcihost_realize(DeviceState *d, Error **errp) > { > PREPPCIState *s = RAVEN_PCI_HOST_BRIDGE(dev); > > memory_region_add_subregion(address_space_mem, 0x80800000, &s->mmcfg); > } > > instead of what I think should also work and should be enough: > > static void raven_pcihost_realize(DeviceState *d, Error **errp) > { > PREPPCIState *s = RAVEN_PCI_HOST_BRIDGE(dev); > PCIHostState *h = PCI_HOST_BRIDGE(dev); dev above is d or obj > MemoryRegion *mr; > > mr = memory_region_new_io(OBJECT(s), &raven_mmcfg_ops, h->bus, > "pci-mmcfg", 0x00400000); > memory_region_add_subregion(address_space_mem, 0x80800000, &s->mmcfg); and s->mmcfg here is mr (just copy&pasted these) but you get the idea. > } > > When asking why and if that's really necessary I got referred to some obscure > conventions that may or may not exist but not documented anywhere or just > because that's a more canonical way to do this. But is there really such > convention or any reason to do it the hard way instead of allowing simpler > device models? I'm all for simplifying devices and be able to do it with just > the two lines in my second example which works as demonstrated by my prep > clean up series but was not accepted for reasons I could not quite > understand. So could this be cleared up too and if there are reasons for the > first way then explain and document that somewhere? > > Regards, > BALATON Zoltan > >