From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx1.redhat.com (mx1.redhat.com [209.132.183.28]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ml01.01.org (Postfix) with ESMTPS id D2BA121E977F0 for ; Thu, 7 Sep 2017 03:55:35 -0700 (PDT) Received: from smtp.corp.redhat.com (int-mx01.intmail.prod.int.phx2.redhat.com [10.5.11.11]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 64B4485550; Thu, 7 Sep 2017 10:58:26 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 64B4485550 Authentication-Results: ext-mx04.extmail.prod.ext.phx2.redhat.com; dmarc=none (p=none dis=none) header.from=redhat.com Authentication-Results: ext-mx04.extmail.prod.ext.phx2.redhat.com; spf=fail smtp.mailfrom=lersek@redhat.com Received: from lacos-laptop-7.usersys.redhat.com (ovpn-120-54.rdu2.redhat.com [10.10.120.54]) by smtp.corp.redhat.com (Postfix) with ESMTP id 19FCA6031A; Thu, 7 Sep 2017 10:58:23 +0000 (UTC) To: "Yao, Jiewen" , "Wang, Jian J" , "Justen, Jordan L" Cc: "edk2-devel@lists.01.org" , "Kinney, Michael D" References: <130779a4-f1c8-bf07-8b85-8125d466aed3@redhat.com> <74D8A39837DF1E4DA445A8C0B3885C503A9A8802@shsmsx102.ccr.corp.intel.com> <74D8A39837DF1E4DA445A8C0B3885C503A9A8C97@shsmsx102.ccr.corp.intel.com> From: Laszlo Ersek Message-ID: <61da5c08-751f-5826-cd86-11fba3f230c9@redhat.com> Date: Thu, 7 Sep 2017 12:58:22 +0200 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.3.0 MIME-Version: 1.0 In-Reply-To: <74D8A39837DF1E4DA445A8C0B3885C503A9A8C97@shsmsx102.ccr.corp.intel.com> X-Scanned-By: MIMEDefang 2.79 on 10.5.11.11 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.28]); Thu, 07 Sep 2017 10:58:26 +0000 (UTC) Subject: Re: ASSERT in QemuVideoDxe driver during reset X-BeenThere: edk2-devel@lists.01.org X-Mailman-Version: 2.1.22 Precedence: list List-Id: EDK II Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 07 Sep 2017 10:55:36 -0000 Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 8bit On 09/07/17 03:28, Yao, Jiewen wrote: > OK. Got you. > > I think we can move the zero memory to DxeIpl. > The DxeIpl can check if the 0 is allocated in MemoryAllocationHob, and zero it if it is not allocated. Regarding the zeroing out of page 0, this solution seems good to me as well. (I agree that page 0 should be zeroed out even if the null pointer detection is enabled.) Regarding the question whether QemuVideoDxe should, or should not, manually change the access permissions to page 0, before and after setting up the Int0x10 vector, if null pointer detection is enabled: This depends on Windows 7. If Windows 7 can boot fine with the null pointer detection enabled, then I agree QemuVideoDxe should be updated as described above. (For massaging the page protection, it should use DXE services, not CpuArch services.) In this case it is also fine for OVMF to inherit the default TRUE setting for the PCD. If Windows 7 cannot boot with null pointer detection enabled, then OVMF should set the PCD to FALSE, and QemuVideoDxe should be changed to not call InstallVbeShim() at all if the PCD is TRUE. Thanks Laszlo > From: Wang, Jian J > Sent: Thursday, September 7, 2017 8:41 AM > To: Yao, Jiewen ; Laszlo Ersek ; Justen, Jordan L > Cc: edk2-devel@lists.01.org; Kinney, Michael D > Subject: RE: ASSERT in QemuVideoDxe driver during reset > > Hi Jiewen and Laszlo, > > According to both of your comments: > > > 1. VBE SHIM must be installed to avoid boot failure of Windows 7. > > 2. Enabling NULL detection should not break existing platforms. > > Let me clarify a few things in advance. I think there’s a little misunderstanding on this issue (Sorry for my poor description originally). > > 1. NULL detection is implemented by disable first 4K page (0-4095). Let me call it page 0. > > 2. The ASSERT was not caused by accessing page 0. I do enabling page 0 before access int10 vector in QemuVideoDxe driver while NULL detection is enabled. > > 3. The ASSERT will only be triggered by reset because QemuVideoDxe will write int10 vector and memory at that place keeps intact during reset. > > 4. The root cause of ASSERT is the fact that page 0 is always cleared in DXE core except to NULL detection enabled. I changed CoreAddRange() in page.c in DXC core like below to avoid page fault in core. > // > // If memory of type EfiConventionalMemory is being added that includes the page > // starting at address 0, then zero the page starting at address 0. This has > // two benifits. It helps find NULL pointer bugs and it also maximizes > // compatibility with operating systems that may evaluate memory in this page > // for legacy data structures. If memory of any other type is added starting > // at address 0, then do not zero the page at address 0 because the page is being > // used for other purposes. > // > if (Type == EfiConventionalMemory && Start == 0 && (End >= EFI_PAGE_SIZE - 1)) { > if ((PcdGet8(PcdNullPointerDetectionPropertyMask) & BIT0) == 0) { > SetMem ((VOID *)(UINTN)Start, EFI_PAGE_SIZE, 0); > } > } > > I think the best solution at present is to update above code to make sure page 0 is always cleared no matter NULL detection is enable or not. To make it possible, I have to enable page 0 before memory clearing and disable it again after then. My concern is that I cannot use CPU arch protocol to do so in CoreAddRange() function because it’s not ready at that time. There’re two options here. One is “manually” disabling page 0 (read cr3, then pm4l, then pde…) in above code; another is moving above code to another place where paging operation is at hand, like DxeIpl or CPU driver. My personal opinion is second one would be better. What’s your opinions? Or any better ideas? > > Thanks, > Wang, Jian J > > From: Yao, Jiewen > Sent: Wednesday, September 06, 2017 7:17 PM > To: Laszlo Ersek >; Wang, Jian J >; Justen, Jordan L > > Cc: edk2-devel@lists.01.org; Kinney, Michael D > > Subject: RE: ASSERT in QemuVideoDxe driver during reset > > HI > I think the NULL detection feature should *never* break any existing platform. > No real function should be skipped for PcdNullDetection. > > If InstallVbeShim () is needed for CSM, we should always call InstallVbeShim() for CSM. > > I suggest below options: > > 1) In OVMF, if CSM is enabled, disable PcdNullDetection. > > 2) In OVMF, if any driver need access first 4K page, the code should explicit call SetAttribute(0, 4K, READ|WRITE), if PcdNullDetection is enabled. > > Either #1 or #2 is OK. > #1 is simple, and #2 can help detect more potential issue. > > Thank you > Yao Jiewen > > From: Laszlo Ersek [mailto:lersek@redhat.com] > Sent: Wednesday, September 6, 2017 6:21 PM > To: Wang, Jian J >; Justen, Jordan L >; Yao, Jiewen > > Cc: edk2-devel@lists.01.org; Kinney, Michael D > > Subject: Re: ASSERT in QemuVideoDxe driver during reset > > On 09/06/17 10:15, Wang, Jian J wrote: >> Hi guys, >> >> I found an ASSERT issue in function InstallVbeShim() in QemuVideoDxe >> driver during reset. The assert statement is like below. >> >> ASSERT (Int0x10->Segment == 0x0000); >> ASSERT (Int0x10->Offset == 0x0000); >> >> This happened after I have enabled NULL pointer access detection >> feature, in which page 0 (4K) is disabled. > > The NULL pointer access detection conflicts with the VBE shim. For > installing the VBE shim, OVMF has to write to page 0, on purpose. > > Please see commit 90803342b1b6 ("OvmfPkg: QemuVideoDxe: Int10h stub for > Windows 7 & 2008 (stdvga, QXL)", 2014-05-20). > >> And because of page 0 disabled, I have to skip the memory clearing for >> page 0 in DXE core. > > By doing that, you invalidate the following comment in said OVMF commit: > > + // > + // We managed to allocate the page at zero. SVN r14218 guarantees that it > + // is NUL-filled. > + // > + ASSERT (Int0x10->Segment == 0x0000); > + ASSERT (Int0x10->Offset == 0x0000); > > Because SVN r14218 was what added the clearing of page 0 to the DXE > core. (Expressed as a git commit: d436d5ca0936.) > >> Otherwise it will cause page fault exception there. It seems that QEMU >> may clear all its memory at startup. Skipping the action of clearing >> page 0 in core won't cause ASSERT issue in QemuVideoDxe, for the first >> time boot. But QemuVideoDxe will write int10 vector at memory 0x10 and >> QEMU will not clear all its memory during warm boot. ASSERT will be >> triggered after reset. > > QEMU does not clear guest RAM at reboot (or S3 resume). > >> It's easy to fix this issue but there're some subtle situations which >> I'm not quite certain. I'd like your opinions for them. >> >> Here're my thoughts on several solutions: >> a) Remove the ASSERT statement in InstallVbeShim(). > > Not a good idea. > >> But I'm sure if it is safe to do so because I don't quite understand >> the purpose of the ASSERT. > > The ASSERT expresses our belief that, after we manage to *allocate* page > 0, we can freely write the real mode Int0x10 vector there, without > overwriting anything else. > >> b) Instead of skipping clearing page 0, enable it, do clearing and >> then disable it. The problem here is that CPU arch protocol is not >> ready at that time. I have to "manually" do page operation, which >> might be non-portable and a little bit odd in DXE core. >> c) Move code clearing page 0 from DXE core to another place wherever >> appropriate, like DxeIpl or cpu driver. But I think there's a good >> reason to put code there before. > > I'm not sure how the NULL pointer access detection is enabled. Is it > enabled by a PCD? > > Hm, yes, it seems, from your patch > > [edk2] [PATCH 1/2] Implement NULL pointer detection for EDK-II Core > > that the PCD is called > > gEfiMdeModulePkgTokenSpaceGuid.PcdNullPointerDetection > > I can see in that patch that the DXE core is updated *not* to clear page > 0 if the PCD is set to TRUE. That makes sense. > > However, OVMF should also be updated to skip the VBE shim installation > if the PCD is set to TRUE. Namely, if the PCD is set to TRUE, then we > simply cannot put the legacy Int0x10 vector in place, so we shouldn't > even try. > > Please extend your original patch set with a patch for OvmfPkg. In the > file "OvmfPkg/QemuVideoDxe/Driver.c", locate the InstallVbeShim() call > site: > >> #if defined MDE_CPU_IA32 || defined MDE_CPU_X64 >> if (Private->Variant == QEMU_VIDEO_BOCHS_MMIO || >> Private->Variant == QEMU_VIDEO_BOCHS) { >> InstallVbeShim (Card->Name, Private->GraphicsOutput.Mode->FrameBufferBase); >> } >> #endif > > and please skip the call if the PCD is set to TRUE: > >> #if defined MDE_CPU_IA32 || defined MDE_CPU_X64 >> if (!FeaturePcdGet (PcdNullPointerDetection) && >> (Private->Variant == QEMU_VIDEO_BOCHS_MMIO || >> Private->Variant == QEMU_VIDEO_BOCHS)) { >> InstallVbeShim (Card->Name, Private->GraphicsOutput.Mode->FrameBufferBase); >> } >> #endif > > Now, the consequence of this would be a working OVMF build, but it would > also break booting UEFI Windows 7 on OVMF. Therefore, please append > *another* patch to your patch set, setting the Feature PCD to FALSE in > all three OVMF DSC files: > > - OvmfPkg/OvmfPkgIa32.dsc > - OvmfPkg/OvmfPkgIa32X64.dsc > - OvmfPkg/OvmfPkgX64.dsc > > We cannot break Windows 7 booting in upstream OVMF. If someone really > wants the null pointer detection in OVMF (and doesn't need Windows 7 for > sure), they can patch the DSC files, or else pass the > > --pcd=gEfiMdeModulePkgTokenSpaceGuid.PcdNullPointerDetection=TRUE > > option to the "build" utility. > > ... I'm now seeing in the original discussion that perhaps you will > introduce a bitmap instead (one bit per firmware phase). That's OK with > me; please reinterpret all of the above with the "DXE bit" then. > > When you post the next version of the patch set, please CC Jordan and > myself on the OvmfPkg patches. > > Thank you! > Laszlo >