Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932950AbcKGSIv (ORCPT ); Mon, 7 Nov 2016 13:08:51 -0500 Received: from smtp.codeaurora.org ([198.145.29.96]:34710 "EHLO smtp.codeaurora.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932720AbcKGSIs (ORCPT ); Mon, 7 Nov 2016 13:08:48 -0500 DMARC-Filter: OpenDMARC Filter v1.3.1 smtp.codeaurora.org CF4DD60430 Authentication-Results: pdx-caf-mail.web.codeaurora.org; dmarc=none header.from=codeaurora.org Authentication-Results: pdx-caf-mail.web.codeaurora.org; spf=pass smtp.mailfrom=cov@codeaurora.org Subject: Re: [PATCH] arm/vdso: introduce vdso_mremap hook To: Dmitry Safonov , linux-kernel@vger.kernel.org References: <20161101172214.2938-1-dsafonov@virtuozzo.com> <0b41c28b-20ef-332f-d8d6-e381e05b8252@codeaurora.org> <714d2aea-ed4c-6272-89c1-e1d0e037855e@virtuozzo.com> Cc: 0x7f454c46@gmail.com, Kevin Brodsky , Andy Lutomirski , Oleg Nesterov , Russell King , Will Deacon , linux-arm-kernel@lists.infradead.org, linux-mm@kvack.org, Cyrill Gorcunov , Pavel Emelyanov , Nathan Lynch , Michael Ellerman From: Christopher Covington Message-ID: Date: Mon, 7 Nov 2016 13:08:29 -0500 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.2 MIME-Version: 1.0 In-Reply-To: <714d2aea-ed4c-6272-89c1-e1d0e037855e@virtuozzo.com> Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5355 Lines: 125 On 11/07/2016 12:16 PM, Dmitry Safonov wrote: > On 11/07/2016 08:00 PM, Christopher Covington wrote: >> Hi Dmitry, >> >> On 11/01/2016 01:22 PM, Dmitry Safonov wrote: >>> Add vdso_mremap hook which will fix context.vdso pointer after mremap() >>> on vDSO vma. This is needed for correct landing after syscall execution. >>> Primary goal of this is for CRIU on arm - we need to restore vDSO image >>> at the exactly same place where the vma was in dumped application. With >>> the help of this hook we'll move vDSO at the new position. >>> The CRIU code handles situations like when vDSO of dumped application >>> was different from vDSO on restoring system. This usally happens when >>> some new symbols are being added to vDSO. In these situations CRIU >>> inserts jump trampolines from old vDSO blob to new vDSO on restore. >>> By that reason even if on restore vDSO blob lies on the same address as >>> blob in dumped application - we still need to move it if it differs. >>> >>> There was previously attempt to add this functionality for arm64 by >>> arch_mremap hook [1], while this patch introduces this with minimal >>> effort - the same way I've added it to x86: >>> commit b059a453b1cf ("x86/vdso: Add mremap hook to vm_special_mapping") >>> >>> At this moment, vdso restoring code is disabled for arm/arm64 arch >>> in CRIU [2], so C/R is only working for !CONFIG_VDSO kernels. This patch >>> is aimed to fix that. >>> The same hook may be introduced for arm64 kernel, but at this moment >>> arm64 vdso code is actively reworked by Kevin, so we can do it on top. >>> Separately, I've refactored arch_remap hook out from ppc64 [3]. >>> >>> [1]: https://marc.info/?i=1448455781-26660-1-git-send-email-cov@codeaurora.org >>> [2]: https://github.com/xemul/criu/blob/master/Makefile#L39 >>> [3]: https://marc.info/?i=20161027170948.8279-1-dsafonov@virtuozzo.com >>> >>> Cc: Kevin Brodsky >>> Cc: Christopher Covington >>> Cc: Andy Lutomirski >>> Cc: Oleg Nesterov >>> Cc: Russell King >>> Cc: Will Deacon >>> Cc: linux-arm-kernel@lists.infradead.org >>> Cc: linux-mm@kvack.org >>> Cc: Cyrill Gorcunov >>> Cc: Pavel Emelyanov >>> Signed-off-by: Dmitry Safonov >>> --- >>> arch/arm/kernel/vdso.c | 21 +++++++++++++++++++++ >>> 1 file changed, 21 insertions(+) >>> >>> diff --git a/arch/arm/kernel/vdso.c b/arch/arm/kernel/vdso.c >>> index 53cf86cf2d1a..d1001f87c2f6 100644 >>> --- a/arch/arm/kernel/vdso.c >>> +++ b/arch/arm/kernel/vdso.c >>> @@ -54,8 +54,11 @@ static const struct vm_special_mapping vdso_data_mapping = { >>> .pages = &vdso_data_page, >>> }; >>> >>> +static int vdso_mremap(const struct vm_special_mapping *sm, >>> + struct vm_area_struct *new_vma); >>> static struct vm_special_mapping vdso_text_mapping __ro_after_init = { >>> .name = "[vdso]", >>> + .mremap = vdso_mremap, >>> }; >>> >>> struct elfinfo { >>> @@ -254,6 +257,24 @@ void arm_install_vdso(struct mm_struct *mm, unsigned long addr) >>> mm->context.vdso = addr; >>> } >>> >>> +static int vdso_mremap(const struct vm_special_mapping *sm, >>> + struct vm_area_struct *new_vma) >>> +{ >>> + unsigned long new_size = new_vma->vm_end - new_vma->vm_start; >>> + unsigned long vdso_size = (vdso_total_pages - 1) << PAGE_SHIFT; >>> + >>> + /* Disallow partial vDSO blob remap */ >>> + if (vdso_size != new_size) >>> + return -EINVAL; >>> + >>> + if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) >>> + return -EFAULT; >>> + >>> + current->mm->context.vdso = new_vma->vm_start; >>> + >>> + return 0; >>> +} >>> + >>> static void vdso_write_begin(struct vdso_data *vdata) >>> { >>> ++vdso_data->seq_count; >>> >> >> What do you think about putting this code somewhere generic (not under >> arch/*), so that powerpc and arm64 can reuse it once the cosmetic changes >> to make them compatible are made? My thought was that it could be defined >> underneath CONFIG_GENERIC_VDSO, which architectures could select as they >> became compatible. > > Hi Chistopher, > > Well, I don't think we won something out of generalization of simple assignment for context.vdso pointer accross arches. And a need to rename > vdso over arches for saving one single line? I count 17 lines, which duplicated across 3 architectures becomes 51 lines. Presumable in the future other architectures will want CRIU support as well. Additionally, should fixes ever be required, fixing one implementation instead of 3+ is preferred. > Also I don't like a bit this arch_mremap hook and need to nullify > vdso pointer. I'm sorry for the confusion but I in no way meant to imply that the arch_mremap hook should be carried forward. I fully agree that the function pointer in struct vm_special_mapping is the better way to go. If you don't want to implement a version with vdso_mremap defined in a generic location (using it from struct vm_special_mapping), do you mind if I propose such a version? Thanks, Cov -- Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc. Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.