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 Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 79A09C4332F for ; Tue, 3 May 2022 22:00:57 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S243391AbiECWE3 (ORCPT ); Tue, 3 May 2022 18:04:29 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:47212 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231405AbiECWE0 (ORCPT ); Tue, 3 May 2022 18:04:26 -0400 Received: from dfw.source.kernel.org (dfw.source.kernel.org [139.178.84.217]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id B8B9D41FA1; Tue, 3 May 2022 15:00:52 -0700 (PDT) Received: from smtp.kernel.org (relay.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dfw.source.kernel.org (Postfix) with ESMTPS id 5AE2C616C7; Tue, 3 May 2022 22:00:52 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 39AC7C385A4; Tue, 3 May 2022 22:00:47 +0000 (UTC) Date: Tue, 3 May 2022 23:00:43 +0100 From: Catalin Marinas To: "Leizhen (ThunderTown)" Cc: Baoquan He , Thomas Gleixner , Ingo Molnar , Borislav Petkov , x86@kernel.org, "H . Peter Anvin" , linux-kernel@vger.kernel.org, Dave Young , Vivek Goyal , Eric Biederman , kexec@lists.infradead.org, Will Deacon , linux-arm-kernel@lists.infradead.org, Rob Herring , Frank Rowand , devicetree@vger.kernel.org, Jonathan Corbet , linux-doc@vger.kernel.org, Randy Dunlap , Feng Zhou , Kefeng Wang , Chen Zhou , John Donnelly , Dave Kleikamp Subject: Re: [PATCH v22 5/9] arm64: kdump: Reimplement crashkernel=X Message-ID: References: <3fc41a94-4247-40f3-14e7-f11e3001ec33@huawei.com> <23e2dcf4-4e9a-5298-d5d8-8761b0bbbe21@huawei.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <23e2dcf4-4e9a-5298-d5d8-8761b0bbbe21@huawei.com> Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, Apr 29, 2022 at 04:25:37PM +0800, Leizhen (ThunderTown) wrote: > On 2022/4/29 16:02, Leizhen (ThunderTown) wrote: > > On 2022/4/29 11:24, Baoquan He wrote: > >> On 04/28/22 at 05:33pm, Leizhen (ThunderTown) wrote: > >>> On 2022/4/28 11:52, Baoquan He wrote: > >>>> On 04/28/22 at 11:40am, Baoquan He wrote: > >>>>> On 04/27/22 at 05:04pm, Catalin Marinas wrote: > >>>>>> There will be some difference as the 4G limit doesn't always hold for > >>>>>> arm64 (though it's true in most cases). Anyway, we can probably simplify > >>>>>> things a bit while following the documented behaviour: > >>>>>> > >>>>>> crashkernel=Y - current behaviour within ZONE_DMA > >>>>>> crashkernel=Y,high - allocate from above ZONE_DMA > >>>>>> crashkernel=Y,low - allocate within ZONE_DMA [...] > >>>>> Sorry to interrupt. Seems the ,high ,low and fallback are main concerns > >>>>> about this version. And I have the same concerns about them which comes > >>>>> from below points: > >>>>> 1) we may need to take best effort to keep ,high, ,low behaviour > >>>>> consistent on all ARCHes. Otherwise user/admin may be confused when they > >>>>> deploy/configure kdump on different machines of different ARCHes in the > >>>>> same LAB. I think we should try to avoid the confusion. I guess by all arches you mean just x86 here. Since the code is not generic, all arches do their own stuff. > > OK, I plan to remove optimization, fallback and default low size, to follow the > > suggestion of Catalin first. But there's one minor point of contention. > > > > 1) Both "crashkernel=X,high" and "crashkernel=X,low" must be present. > > 2) Both "crashkernel=X,high" and "crashkernel=X,low" are present. > > or > > Allow "crashkernel=X,high" to be present alone. Unlike x86, the default low size is zero. > > > > I prefer 2), how about you? (2) works for me as well. We keep these simple as "expert" options and allow crashkernel= to fall back to 'high' if not sufficient memory in ZONE_DMA. That would be a slight change from the current behaviour but, as Zhen Lei said, with the old tools it's just moving the error around, the crashkernel wouldn't be available in either case. > >>>>> 2) Fallback behaviour is important to our distros. The reason is we will > >>>>> provide default value with crashkernel=xxxM along kernel of distros. In > >>>>> this case, we hope the reservation will succeed by all means. The ,high > >>>>> and ,low is an option if customer likes to take with expertise. OK, that's good feedback. So, to recap, IIUC you are fine with: crashkernel=Y - allocate within ZONE_DMA with fallback above with a default in ZONE_DMA (like x86, 256M or swiotlb size) crashkernel=Y,high - allocate from above ZONE_DMA crashkernel=Y,low - allocate within ZONE_DMA 'crashkernel' overrides the high and low while the latter two can be passed independently. > >>>>> After going through arm64 memory init code, I got below summary about > >>>>> arm64_dma_phys_limit which is the first zone's upper limit. I think we > >>>>> can make use of it to facilitate to simplify code. > >>>>> ================================================================================ > >>>>> DMA DMA32 NORMAL > >>>>> 1)Raspberry Pi4 0~1G 3G~4G (above 4G) > >>>>> 2)Normal machine 0~4G 0 (above 4G) > >>>>> 3)Special machine (above 4G)~MAX > >>>>> 4)No DMA|DMA32 (above 4G)~MAX > >>> > >>> arm64_memblock_init() > >>> reserve_crashkernel() --------------- 0a30c53573b0 ("arm64: mm: Move reserve_crashkernel() into mem_init()") > >> We don't need different code for this place of reservation as you are > >> doing in this patchset, since arm64_dma_phys_limit is initialized as > >> below. In fact, in arm64_memblock_init(), we have made memblock ready, > >> we can initialize arm64_dma_phys_limit as memblock_end_of_DRAM(). And if > >> memblock_start_of_DRAM() is bigger than 4G, we possibly can call > >> reserve_crashkernel() here too. > > > > Yes. Maybe all the devices in this environment are 64-bit. One way I > > know of allowing 32-bit devices to access high memory without SMMU > > is: Set a fixed value for the upper 32 bits. In this case, the DMA > > zone should be [phys_start, phys_start + 4G). We decided that this case doesn't really exists for arm64 platforms (no need for special ZONE_DMA). > I just read the message of commit 791ab8b2e3 ("arm64: Ignore any DMA > offsets in the max_zone_phys() calculation") > > Currently, the kernel assumes that if RAM starts above 32-bit (or > zone_bits), there is still a ZONE_DMA/DMA32 at the bottom of the RAM and > such constrained devices have a hardwired DMA offset. In practice, we > haven't noticed any such hardware so let's assume that we can expand > ZONE_DMA32 to the available memory if no RAM below 4GB. Similarly, > ZONE_DMA is expanded to the 4GB limit if no RAM addressable by > zone_bits. I think the above log is slightly confusing. If the DRAM starts above 4G, ZONE_DMA goes to the end of DRAM. If the DRAM starts below 4G but above the zone_bits for ZONE_DMA as specified in DT/ACPI, it pushes ZONE_DMA to 4G. I don't remember why we did this last part, maybe in case we get incorrect firmware tables, otherwise we could have extended ZONE_DMA to end of DRAM. Zhen Lei, if we agreed on the crashkernel behaviour, could you please post a series that does the above parsing allocation? Ignore the optimisations, we can look at them afterwards. Thanks. -- Catalin 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 Received: from bombadil.infradead.org (bombadil.infradead.org [198.137.202.133]) (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 787B6C433EF for ; Tue, 3 May 2022 22:02:11 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=lists.infradead.org; s=bombadil.20210309; h=Sender: Content-Transfer-Encoding:Content-Type:List-Subscribe:List-Help:List-Post: List-Archive:List-Unsubscribe:List-Id:In-Reply-To:MIME-Version:References: Message-ID:Subject:Cc:To:From:Date:Reply-To:Content-ID:Content-Description: Resent-Date:Resent-From:Resent-Sender:Resent-To:Resent-Cc:Resent-Message-ID: List-Owner; bh=kzoBTZd85B4S2IhB7F/mJmgF71n6pyrR4NQfACnhuQU=; b=edd9R8Ostd8sYQ nQGfSKy3CAmj+hSx+2rg5MFtVrw//3jfMsOHeRIinqlfqCvBcGTAezVtx03JGsXAfP5UZP4TvVJE3 ApPZb42+DU1lDyT3hpwNbFSA31JEOIvyteu7Y27aHPVVpkg8UMsVvoBw6GYn2TDXYy7DxQBAZupwZ xkr9omooDDfqU3malRVuLTdUOQJjOwfM83BvQkU+/AAP3WGUZe0B/5GrAvwfZ30tML2KaRVfFeIpD 1hWDQ8zZxEGH+nMF1c3j8iqDvQ+BbwK7pxSSE/9m+yh2+VNQ7IAax1M0cD0EZCB9mgiG1OCqiv44m TtNH9ZY4cIQYoD3DaGPg==; Received: from localhost ([::1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.94.2 #2 (Red Hat Linux)) id 1nm0Zn-007txT-OZ; Tue, 03 May 2022 22:00:59 +0000 Received: from ams.source.kernel.org ([2604:1380:4601:e00::1]) by bombadil.infradead.org with esmtps (Exim 4.94.2 #2 (Red Hat Linux)) id 1nm0Zi-007tuu-N5; Tue, 03 May 2022 22:00:56 +0000 Received: from smtp.kernel.org (relay.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ams.source.kernel.org (Postfix) with ESMTPS id 24761B821E6; Tue, 3 May 2022 22:00:53 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 39AC7C385A4; Tue, 3 May 2022 22:00:47 +0000 (UTC) Date: Tue, 3 May 2022 23:00:43 +0100 From: Catalin Marinas To: "Leizhen (ThunderTown)" Cc: Baoquan He , Thomas Gleixner , Ingo Molnar , Borislav Petkov , x86@kernel.org, "H . Peter Anvin" , linux-kernel@vger.kernel.org, Dave Young , Vivek Goyal , Eric Biederman , kexec@lists.infradead.org, Will Deacon , linux-arm-kernel@lists.infradead.org, Rob Herring , Frank Rowand , devicetree@vger.kernel.org, Jonathan Corbet , linux-doc@vger.kernel.org, Randy Dunlap , Feng Zhou , Kefeng Wang , Chen Zhou , John Donnelly , Dave Kleikamp Subject: Re: [PATCH v22 5/9] arm64: kdump: Reimplement crashkernel=X Message-ID: References: <3fc41a94-4247-40f3-14e7-f11e3001ec33@huawei.com> <23e2dcf4-4e9a-5298-d5d8-8761b0bbbe21@huawei.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <23e2dcf4-4e9a-5298-d5d8-8761b0bbbe21@huawei.com> X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20220503_150055_080208_B2205D83 X-CRM114-Status: GOOD ( 38.33 ) X-BeenThere: linux-arm-kernel@lists.infradead.org X-Mailman-Version: 2.1.34 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+linux-arm-kernel=archiver.kernel.org@lists.infradead.org On Fri, Apr 29, 2022 at 04:25:37PM +0800, Leizhen (ThunderTown) wrote: > On 2022/4/29 16:02, Leizhen (ThunderTown) wrote: > > On 2022/4/29 11:24, Baoquan He wrote: > >> On 04/28/22 at 05:33pm, Leizhen (ThunderTown) wrote: > >>> On 2022/4/28 11:52, Baoquan He wrote: > >>>> On 04/28/22 at 11:40am, Baoquan He wrote: > >>>>> On 04/27/22 at 05:04pm, Catalin Marinas wrote: > >>>>>> There will be some difference as the 4G limit doesn't always hold for > >>>>>> arm64 (though it's true in most cases). Anyway, we can probably simplify > >>>>>> things a bit while following the documented behaviour: > >>>>>> > >>>>>> crashkernel=Y - current behaviour within ZONE_DMA > >>>>>> crashkernel=Y,high - allocate from above ZONE_DMA > >>>>>> crashkernel=Y,low - allocate within ZONE_DMA [...] > >>>>> Sorry to interrupt. Seems the ,high ,low and fallback are main concerns > >>>>> about this version. And I have the same concerns about them which comes > >>>>> from below points: > >>>>> 1) we may need to take best effort to keep ,high, ,low behaviour > >>>>> consistent on all ARCHes. Otherwise user/admin may be confused when they > >>>>> deploy/configure kdump on different machines of different ARCHes in the > >>>>> same LAB. I think we should try to avoid the confusion. I guess by all arches you mean just x86 here. Since the code is not generic, all arches do their own stuff. > > OK, I plan to remove optimization, fallback and default low size, to follow the > > suggestion of Catalin first. But there's one minor point of contention. > > > > 1) Both "crashkernel=X,high" and "crashkernel=X,low" must be present. > > 2) Both "crashkernel=X,high" and "crashkernel=X,low" are present. > > or > > Allow "crashkernel=X,high" to be present alone. Unlike x86, the default low size is zero. > > > > I prefer 2), how about you? (2) works for me as well. We keep these simple as "expert" options and allow crashkernel= to fall back to 'high' if not sufficient memory in ZONE_DMA. That would be a slight change from the current behaviour but, as Zhen Lei said, with the old tools it's just moving the error around, the crashkernel wouldn't be available in either case. > >>>>> 2) Fallback behaviour is important to our distros. The reason is we will > >>>>> provide default value with crashkernel=xxxM along kernel of distros. In > >>>>> this case, we hope the reservation will succeed by all means. The ,high > >>>>> and ,low is an option if customer likes to take with expertise. OK, that's good feedback. So, to recap, IIUC you are fine with: crashkernel=Y - allocate within ZONE_DMA with fallback above with a default in ZONE_DMA (like x86, 256M or swiotlb size) crashkernel=Y,high - allocate from above ZONE_DMA crashkernel=Y,low - allocate within ZONE_DMA 'crashkernel' overrides the high and low while the latter two can be passed independently. > >>>>> After going through arm64 memory init code, I got below summary about > >>>>> arm64_dma_phys_limit which is the first zone's upper limit. I think we > >>>>> can make use of it to facilitate to simplify code. > >>>>> ================================================================================ > >>>>> DMA DMA32 NORMAL > >>>>> 1)Raspberry Pi4 0~1G 3G~4G (above 4G) > >>>>> 2)Normal machine 0~4G 0 (above 4G) > >>>>> 3)Special machine (above 4G)~MAX > >>>>> 4)No DMA|DMA32 (above 4G)~MAX > >>> > >>> arm64_memblock_init() > >>> reserve_crashkernel() --------------- 0a30c53573b0 ("arm64: mm: Move reserve_crashkernel() into mem_init()") > >> We don't need different code for this place of reservation as you are > >> doing in this patchset, since arm64_dma_phys_limit is initialized as > >> below. In fact, in arm64_memblock_init(), we have made memblock ready, > >> we can initialize arm64_dma_phys_limit as memblock_end_of_DRAM(). And if > >> memblock_start_of_DRAM() is bigger than 4G, we possibly can call > >> reserve_crashkernel() here too. > > > > Yes. Maybe all the devices in this environment are 64-bit. One way I > > know of allowing 32-bit devices to access high memory without SMMU > > is: Set a fixed value for the upper 32 bits. In this case, the DMA > > zone should be [phys_start, phys_start + 4G). We decided that this case doesn't really exists for arm64 platforms (no need for special ZONE_DMA). > I just read the message of commit 791ab8b2e3 ("arm64: Ignore any DMA > offsets in the max_zone_phys() calculation") > > Currently, the kernel assumes that if RAM starts above 32-bit (or > zone_bits), there is still a ZONE_DMA/DMA32 at the bottom of the RAM and > such constrained devices have a hardwired DMA offset. In practice, we > haven't noticed any such hardware so let's assume that we can expand > ZONE_DMA32 to the available memory if no RAM below 4GB. Similarly, > ZONE_DMA is expanded to the 4GB limit if no RAM addressable by > zone_bits. I think the above log is slightly confusing. If the DRAM starts above 4G, ZONE_DMA goes to the end of DRAM. If the DRAM starts below 4G but above the zone_bits for ZONE_DMA as specified in DT/ACPI, it pushes ZONE_DMA to 4G. I don't remember why we did this last part, maybe in case we get incorrect firmware tables, otherwise we could have extended ZONE_DMA to end of DRAM. Zhen Lei, if we agreed on the crashkernel behaviour, could you please post a series that does the above parsing allocation? Ignore the optimisations, we can look at them afterwards. Thanks. -- Catalin _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel From mboxrd@z Thu Jan 1 00:00:00 1970 From: Catalin Marinas Date: Tue, 3 May 2022 23:00:43 +0100 Subject: [PATCH v22 5/9] arm64: kdump: Reimplement crashkernel=X In-Reply-To: <23e2dcf4-4e9a-5298-d5d8-8761b0bbbe21@huawei.com> References: <3fc41a94-4247-40f3-14e7-f11e3001ec33@huawei.com> <23e2dcf4-4e9a-5298-d5d8-8761b0bbbe21@huawei.com> Message-ID: List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: kexec@lists.infradead.org On Fri, Apr 29, 2022 at 04:25:37PM +0800, Leizhen (ThunderTown) wrote: > On 2022/4/29 16:02, Leizhen (ThunderTown) wrote: > > On 2022/4/29 11:24, Baoquan He wrote: > >> On 04/28/22 at 05:33pm, Leizhen (ThunderTown) wrote: > >>> On 2022/4/28 11:52, Baoquan He wrote: > >>>> On 04/28/22 at 11:40am, Baoquan He wrote: > >>>>> On 04/27/22 at 05:04pm, Catalin Marinas wrote: > >>>>>> There will be some difference as the 4G limit doesn't always hold for > >>>>>> arm64 (though it's true in most cases). Anyway, we can probably simplify > >>>>>> things a bit while following the documented behaviour: > >>>>>> > >>>>>> crashkernel=Y - current behaviour within ZONE_DMA > >>>>>> crashkernel=Y,high - allocate from above ZONE_DMA > >>>>>> crashkernel=Y,low - allocate within ZONE_DMA [...] > >>>>> Sorry to interrupt. Seems the ,high ,low and fallback are main concerns > >>>>> about this version. And I have the same concerns about them which comes > >>>>> from below points: > >>>>> 1) we may need to take best effort to keep ,high, ,low behaviour > >>>>> consistent on all ARCHes. Otherwise user/admin may be confused when they > >>>>> deploy/configure kdump on different machines of different ARCHes in the > >>>>> same LAB. I think we should try to avoid the confusion. I guess by all arches you mean just x86 here. Since the code is not generic, all arches do their own stuff. > > OK, I plan to remove optimization, fallback and default low size, to follow the > > suggestion of Catalin first. But there's one minor point of contention. > > > > 1) Both "crashkernel=X,high" and "crashkernel=X,low" must be present. > > 2) Both "crashkernel=X,high" and "crashkernel=X,low" are present. > > or > > Allow "crashkernel=X,high" to be present alone. Unlike x86, the default low size is zero. > > > > I prefer 2), how about you? (2) works for me as well. We keep these simple as "expert" options and allow crashkernel= to fall back to 'high' if not sufficient memory in ZONE_DMA. That would be a slight change from the current behaviour but, as Zhen Lei said, with the old tools it's just moving the error around, the crashkernel wouldn't be available in either case. > >>>>> 2) Fallback behaviour is important to our distros. The reason is we will > >>>>> provide default value with crashkernel=xxxM along kernel of distros. In > >>>>> this case, we hope the reservation will succeed by all means. The ,high > >>>>> and ,low is an option if customer likes to take with expertise. OK, that's good feedback. So, to recap, IIUC you are fine with: crashkernel=Y - allocate within ZONE_DMA with fallback above with a default in ZONE_DMA (like x86, 256M or swiotlb size) crashkernel=Y,high - allocate from above ZONE_DMA crashkernel=Y,low - allocate within ZONE_DMA 'crashkernel' overrides the high and low while the latter two can be passed independently. > >>>>> After going through arm64 memory init code, I got below summary about > >>>>> arm64_dma_phys_limit which is the first zone's upper limit. I think we > >>>>> can make use of it to facilitate to simplify code. > >>>>> ================================================================================ > >>>>> DMA DMA32 NORMAL > >>>>> 1)Raspberry Pi4 0~1G 3G~4G (above 4G) > >>>>> 2)Normal machine 0~4G 0 (above 4G) > >>>>> 3)Special machine (above 4G)~MAX > >>>>> 4)No DMA|DMA32 (above 4G)~MAX > >>> > >>> arm64_memblock_init() > >>> reserve_crashkernel() --------------- 0a30c53573b0 ("arm64: mm: Move reserve_crashkernel() into mem_init()") > >> We don't need different code for this place of reservation as you are > >> doing in this patchset, since arm64_dma_phys_limit is initialized as > >> below. In fact, in arm64_memblock_init(), we have made memblock ready, > >> we can initialize arm64_dma_phys_limit as memblock_end_of_DRAM(). And if > >> memblock_start_of_DRAM() is bigger than 4G, we possibly can call > >> reserve_crashkernel() here too. > > > > Yes. Maybe all the devices in this environment are 64-bit. One way I > > know of allowing 32-bit devices to access high memory without SMMU > > is: Set a fixed value for the upper 32 bits. In this case, the DMA > > zone should be [phys_start, phys_start + 4G). We decided that this case doesn't really exists for arm64 platforms (no need for special ZONE_DMA). > I just read the message of commit 791ab8b2e3 ("arm64: Ignore any DMA > offsets in the max_zone_phys() calculation") > > Currently, the kernel assumes that if RAM starts above 32-bit (or > zone_bits), there is still a ZONE_DMA/DMA32 at the bottom of the RAM and > such constrained devices have a hardwired DMA offset. In practice, we > haven't noticed any such hardware so let's assume that we can expand > ZONE_DMA32 to the available memory if no RAM below 4GB. Similarly, > ZONE_DMA is expanded to the 4GB limit if no RAM addressable by > zone_bits. I think the above log is slightly confusing. If the DRAM starts above 4G, ZONE_DMA goes to the end of DRAM. If the DRAM starts below 4G but above the zone_bits for ZONE_DMA as specified in DT/ACPI, it pushes ZONE_DMA to 4G. I don't remember why we did this last part, maybe in case we get incorrect firmware tables, otherwise we could have extended ZONE_DMA to end of DRAM. Zhen Lei, if we agreed on the crashkernel behaviour, could you please post a series that does the above parsing allocation? Ignore the optimisations, we can look at them afterwards. Thanks. -- Catalin