Re: issue about booting centos fail with edk2

[Laszlo Ersek <lersek@redhat.com>]

On 08/02/17 00:57, Ard Biesheuvel wrote:
> On 1 August 2017 at 23:29, Laszlo Ersek <lersek@redhat.com> wrote:
>> On 08/01/17 19:23, Ard Biesheuvel wrote:
>>> On 1 August 2017 at 16:42, Laszlo Ersek <lersek@redhat.com> wrote:
>>>> On 08/01/17 10:34, Zhu Yijun wrote:
>>>>> Thanks for your reply!
>>>>>
>>>>> On 2017/8/1 3:02, Laszlo Ersek wrote:
>>>>>> On 07/31/17 02:27, Zhu Yijun wrote:
>>>>>>> Hi all,
>>>>>>>
>>>>>>>     I install a CentOS-7-aarch64 guest img by qemu cdrom, but it hung at UEFI probability.
>>>>>>>
>>>>>>>     Basic info:
>>>>>>>     libvirt 1.3.5
>>>>>>>     QEMU 2.6.2
>>>>>>>     UEFI: master branch with commit "688c7d2 BaseTools: Fix the bug that warn() function with only 1 argument"
>>>>>>>
>>>>>>>     Config pflash and two disks in xml:
>>>>>>>
>>>>>>>     ...
>>>>>>>     <os>
>>>>>>>     <type arch='aarch64' machine='virt-2.6'>hvm</type>
>>>>>>>     <loader readonly='yes' type='pflash'>/usr/share/edk2/aarch64/QEMU_EFI-pflash.raw</loader>
>>>>>>>     <boot dev='hd'/>
>>>>>>>   </os>
>>>>>>>   ...
>>>>>>>   <disk type='file' device='disk'>
>>>>>>>       <driver name='qemu' type='qcow2' cache='none' io='native'/>
>>>>>>>       <source file='/CentOS-7-aarch64/centos.qcow2'/>
>>>>>>>       <backingStore/>
>>>>>>>       <target dev='sda' bus='scsi'/>
>>>>>>>     </disk>
>>>>>>>     <disk type='file' device='cdrom'>
>>>>>>>       <driver name='qemu' type='raw' cache='none' io='native'/>
>>>>>>>       <source file='/CentOS-7-aarch64/CentOS-7-aarch64-Everything.iso'/>
>>>>>>>       <backingStore/>
>>>>>>>       <target dev='sdb' bus='scsi'/>
>>>>>>>     </disk>
>>>>>>>     ...
>>>>>>>
>>>>>>>     I found it failed at "Match (Translated, TranslatedSize, ActiveOption[Idx].BootOption->FilePath)" function in "SetBootOrderFromQemu", the UEFI debug info as follow:
>>>>>> No, that's not where the problem is. See below:
>>>>>>
>>>>>>> start-console-fail.log
>>>>>>> FSOpen: Open '\EFI\BOOT\fallback.efi' Success
>>>>>>> FSOpen: Open '\EFI\BOOT\fallback.efi' Success
>>>>>>>
>>>>>>>
>>>>>>> Synchronous Exception at 0x00000002384B1104
>>>>>>> PC 0x0002384B1104
>>>>>>> PC 0x0002384A916C
>>>>>>> PC 0x0002384CA2D0
>>>>>>> PC 0x00023EEB7DF8 (0x00023EEB1000+0x00006DF8) [ 1] DxeCore.dll
>>>>>>> PC 0x00023BD1568C (0x00023BD02000+0x0001368C) [ 2] BdsDxe.dll
>>>>>>> PC 0x00023BD03F98 (0x00023BD02000+0x00001F98) [ 2] BdsDxe.dll
>>>>>>> PC 0x00023BD05640 (0x00023BD02000+0x00003640) [ 2] BdsDxe.dll
>>>>>>> PC 0x00023EEB3704 (0x00023EEB1000+0x00002704) [ 3] DxeCore.dll
>>>>>>> PC 0x00023EEB27C8 (0x00023EEB1000+0x000017C8) [ 3] DxeCore.dll
>>>>>>> PC 0x00023EEB2024 (0x00023EEB1000+0x00001024) [ 3] DxeCore.dll
>>>>>>> [ 1] /root/rpmbuild/BUILD/edk2-2.6.0/Build/ArmVirtQemu-AARCH64/DEBUG_GCC49/AARCH64/MdeModulePkg/Core/Dxe/DxeMain/DEBUG/DxeCore.dll
>>>>>>> [ 2] /root/rpmbuild/BUILD/edk2-2.6.0/Build/ArmVirtQemu-AARCH64/DEBUG_GCC49/AARCH64/MdeModulePkg/Universal/BdsDxe/BdsDxe/DEBUG/BdsDxe.dll
>>>>>>> [ 3] /root/rpmbuild/BUILD/edk2-2.6.0/Build/ArmVirtQemu-AARCH64/DEBUG_GCC49/AARCH64/MdeModulePkg/Core/Dxe/DxeMain/DEBUG/DxeCore.dll
>>>>>>>
>>>>>>>   X0 0x00000002384A9000   X1 0x00000002384B2990   X2 0x000000023AAFDF98   X3 0x000000023BFF0018
>>>>>>>   X4 0x0000000000000000   X5 0x0000000000000007   X6 0x0000000238533300   X7 0x0000000000000000
>>>>>>>   X8 0x000000023C01F548   X9 0x0000000200000000  X10 0x00000002384A8000  X11 0x00000002384C5FFF
>>>>>>>  X12 0x0000000000000000  X13 0x0000000000000008  X14 0x259511BDAEB1F36C  X15 0x1378CC1DF3F5DDBB
>>>>>>>  X16 0x000000023EEB0BE0  X17 0x0000000000000000  X18 0x0000000000000000  X19 0x0000000000000013
>>>>>>>  X20 0x0000000000000000  X21 0x0000000000000000  X22 0x0000000000000000  X23 0x0000000000000000
>>>>>>>  X24 0x0000000000000000  X25 0x0000000000000000  X26 0x0000000000000000  X27 0x0000000000000000
>>>>>>>  X28 0x0000000000000000   FP 0x000000023EEB0A40   LR 0x00000002384A916C
>>>>>>>
>>>>>>>   V0 0xAFAFAFAFAFAFAFAF AFAFAFAFAFAFAFAF   V1 0x63702F6666666666 6666666666666666
>>>>>>>   V2 0x40697363732F3340 6567646972622D69   V3 0x0000000000000000 0000000000000000
>>>>>>>   V4 0x0000000000000000 0000000000000000   V5 0x4010040140100401 4010040140100401
>>>>>>>   V6 0x0000000000000000 0000000000000000   V7 0x0000000000000000 0000000000000000
>>>>>>>   V8 0x0000000000000000 0000000000000000   V9 0x0000000000000000 0000000000000000
>>>>>>>  V10 0x0000000000000000 0000000000000000  V11 0x0000000000000000 0000000000000000
>>>>>>>  V12 0x0000000000000000 0000000000000000  V13 0x0000000000000000 0000000000000000
>>>>>>>  V14 0x0000000000000000 0000000000000000  V15 0x0000000000000000 0000000000000000
>>>>>>>  V16 0x0000000000000000 0000000000000000  V17 0x0000000000000000 0000000000000000
>>>>>>>  V18 0x0000000000000000 0000000000000000  V19 0x0000000000000000 0000000000000000
>>>>>>>  V20 0x0000000000000000 0000000000000000  V21 0x0000000000000000 0000000000000000
>>>>>>>  V22 0x0000000000000000 0000000000000000  V23 0x0000000000000000 0000000000000000
>>>>>>>  V24 0x0000000000000000 0000000000000000  V25 0x0000000000000000 0000000000000000
>>>>>>>  V26 0x0000000000000000 0000000000000000  V27 0x0000000000000000 0000000000000000
>>>>>>>  V28 0x0000000000000000 0000000000000000  V29 0x0000000000000000 0000000000000000
>>>>>>>  V30 0x0000000000000000 0000000000000000  V31 0x0000000000000000 0000000000000000
>>>>>>>
>>>>>>>   SP 0x000000023EEB0A40  ELR 0x00000002384B1104  SPSR 0x60000205  FPSR 0x00000000
>>>>>>>  ESR 0x02000000          FAR 0x1DE7EC7EDBADC0DE
>>>>>>>
>>>>>>>  ESR : EC 0x00  IL 0x1  ISS 0x00000000
>>>>>>>
>>>>>>> Stack dump:
>>>>>>>   000023EEB0940: 0000C0E000000148 00000002384A9000 00000002384CA254 0000000000000000
>>>>>>>   000023EEB0960: 000000023EEB0BC0 000000023AC006C0 0000F2503EEB0BC0 00000002384B6018
>>>>>>>   000023EEB0980: 000000023EEB0BC0 0000000000000000 000000000000C0E0 0000000000000148
>>>>>>>   000023EEB09A0: 0000000000000148 0000100000020A8C 00000002384B6110 00000002384B6108
>>>>>>>   000023EEB09C0: 00000002384B6100 0000000000000006 00000002384B6058 00000002384B50DF
>>>>>>>   000023EEB09E0: 00000002384A9148 0000000000000000 00000002384A9000 00000002384A9000
>>>>>>>   000023EEB0A00: 0000000000000000 00000002398DA518 00000002385375B2 00000002385629A0
>>>>>>>   000023EEB0A20: 000000023854C1C0 00000002398DA518 000000023EEB0BC0 0000000000000000
>>>>>>>> 000023EEB0A40: 000000023EEB0BC0 00000002384CA2D0 000000023AAFDF98 000000023BFF0018
>>>>>>>   000023EEB0A60: 00000002384CA360 000000023EEC8348 00000002385375B0 000000023AAFDF98
>>>>>>>   000023EEB0A80: 000000023EEB0AC0 0000F25038533338 00000002384B6018 0000000000000000
>>>>>>>   000023EEB0AA0: 0000000000000000 0000000238B63D18 0000000000001000 0000000000000000
>>>>>>>   000023EEB0AC0: 000000023BFF0018 00000002398DA518 00000002398CE598 0000000000000000
>>>>>>>   000023EEB0AE0: 0000000000000000 0000000000000000 00000002384C6000 00000000000C99C0
>>>>>>>   000023EEB0B00: 0000000200000001 0000000000000000 000000023AC006C0 11D295625B1B31A1
>>>>>>>   000023EEB0B20: 3B7269C9A0003F8E 0000000000000000 0000000238B63F98 000000163EEB0B68
>>>>>>> ASSERT [ArmCpuDxe] /root/rpmbuild/BUILD/edk2-2.6.0/ArmPkg/Library/DefaultExceptionHandlerLib/AArch64/DefaultExceptionHandler.c(271): ((BOOLEAN)(0==1))
>>>>>> This is a guest that you didn't install from installer media. I think
>>>>>> you may have gotten the preinstalled disk image from some image provider
>>>>>> service. The UEFI boot variable(s) are not set up to boot the CentOS
>>>>>> installation, in your nvram / pflash file.
>>>>> Yes, the boot variable must store in domain's nvram file("/var/lib/libvirt/qemu/nvram/centos_VARS.fd").  After installed, it generates an new boot menu
>>>>> called "CentOS Linux AltArch "  which device path is "HD(1,GPT,D562CAA6-F61B-4F93-87FB-22DDADF6CAE2,0x800,0x64000)/\EFI\centos\shim.efi".
>>>>>
>>>>> such like:
>>>>> Boot Manager Menu
>>>>>    CentOS Linux AltArch                          -> device path:  PciRoot(0x0)/Pci(0x3,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0) /HD(1,GPT,D562CAA6-F61B-4F93-87FB-22DDADF6CAE2,0x800,0x64000)/\EFI\centos\shim.efi
>>>>>    UEFI Misc Device
>>>>>    UEFI Misc Device 2
>>>>>    EFI Internal Shell
>>>>>    UEFI QEMU QEMU CD-ROM               -> device path:  PciRoot(0x0)/Pci(0x3,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x1)
>>>>>    UEFI QEMU QEMU HARDDISK            -> device path:  PciRoot(0x0)/Pci(0x3,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)
>>>>>    UEFI PXEv4 (MAC:5254002D2EB6)
>>>>>
>>>>> But when I shutdown &undefine this domain, and virsh create an new domain with the disk centos.qcow2 which installed just before, the UEFI boot manager
>>>>> menu is:
>>>>> Boot Manager Menu
>>>>>    UEFI QEMU QEMU HARDDISK               -> device path:  PciRoot(0x0)/Pci(0x3,0x0)/Pci(0x0,0x0)/Scsi(0x0,0x0)
>>>>>    UEFI Misc Device
>>>>>    UEFI Misc Device 2
>>>>>    EFI Internal Shell
>>>>>    UEFI PXEv4 (MAC:5254002D2EB6)
>>>>
>>>> Right. In this case you have lost your original nvram contents, and you
>>>> only have the boot options that are auto-generated by the
>>>> EfiBootManagerRefreshAllBootOption() function. This function lives in
>>>> UefiBootManagerLib, and is called from OVMF's PlatformBootManagerLib
>>>> instance.
>>>>
>>>> The filtering and reordering still occurs in OVMF, but now the first
>>>> boot option that matches QEMU's fw_cfg bootorder specification is not
>>>> the "CentOS Linux AltArch" boot option that you originally had. Instead,
>>>> now QemuBootOrderLib encounters the "UEFI QEMU QEMU HARDDISK"
>>>> auto-generated boot option first as a match.
>>>>
>>>> This boot option in turn means "fallback.efi", according to the blog
>>>> post I linked earlier.
>>>>
>>>> When "fallback.efi" executes successfully, your original "CentOS Linux
>>>> AltArch" boot option is restored / recreated (at the top of the boot
>>>> option list). But, when "fallback.efi" crashes, you get a crash instead.
>>>>
>>>>> I am confused about two points:
>>>>> 1) The new domain still have chance to load the "EFI\centos\shim.efi" and boot kernel successful, it means that sometimes the system firmware launches
>>>>> the BOOTAA64.EFI, sometimes lauches shim.efi.  It is probabilistic.
>>>>
>>>> "EFI\centos\shim.efi" is never automatically loaded. It needs a
>>>> dedicated UEFI boot option. Thus, it can be loaded in your "new" domain
>>>> *only* if "fallback.efi" runs first, successfully.
>>>>
>>>> So what you are seeing is that "fallback.efi" sometimes works, and
>>>> sometimes crashes. That's the nature of memory corruption bugs.
>>>>
>>>>>
>>>>> 2) Is there a way to make the "CentOS Linux AltArch " boot menu persistent?
>>>>
>>>> There isn't. If you lose your nvram, you lose the non-auto-generated
>>>> boot options with it.
>>>>
>>>> Remedying such situations is what "fallback.efi" exists for.
>>>>
>>>>>>
>>>>>> In such cases, the "fallback.efi" utility is invoked (called
>>>>>> "\EFI\BOOT\BOOTAA64.EFI). Please refer to:
>>>>>>
>>>>>> https://blog.uncooperative.org/blog/2014/02/06/the-efi-system-partition/
>>>>>>
>>>>>> Unfortunately, "fallback.efi" (from the shim package) used to have a few
>>>>>> bugs over time and sometimes it would crash. See for example:
>>>>>>
>>>>>>   https://bugzilla.redhat.com/show_bug.cgi?id=1196114
>>>>>>
>>>>>> I'm unsure what version of shim / fallback.efi is in the installed
>>>>>> CentOS image, but it looks like the same (or another similar)
>>>>>> fallback.efi issue to me.
>>>>>
>>>>> shim version in my side is shim-0.9-2.el7.aarch64.
>>>>
>>>> This confirms that you are not seeing the exact bug described in
>>>> RHBZ#1196114, because that bug was fixed in shim-0.9 (see
>>>> <https://bugzilla.redhat.com/show_bug.cgi?id=1196114#c16>).
>>>>
>>>> It remains a fact that your original log contains a crash register dump
>>>> after fallback.efi is launched. The V0 register contains
>>>> 0xAFAFAFAFAFAFAFAF AFAFAFAFAFAFAFAF; the pattern 0xAF is used to fill
>>>> released (freed) pages in debug builds. So this seems to be an
>>>> use-after-free issue. I suggest adding debug instrumentation to
>>>> fallback.efi, and seeing where exactly it blows up.
>>>>
>>>
>>> The presence of the 0xAF pattern in register v0 by itself does not
>>> suggest anything at all: V0 is a SIMD register, which is used by the
>>> SetMem() routine to poison the memory. There is very little other code
>>> (if any) that actually uses the SIMD registers otherwise.
>>
>> Thanks for pointing this out.
>>
>> Can you perhaps deduce more info from the stack / register dump? The
>> topmost three stack frames don't have edk2 module names associated with
>> them -- does that confirm that the synchronous exception is raised in a
>> non-edk2 module?
>>
> 
> The stack trace is consistent with BDS calling LoadImage() to launch
> fallback.efi (which is GNU-EFI based so it does not set the NB10
> Codeview debug entry containing the path on the build host)
> 
> The FAR (faulting address) register contains the well known bogus
> value KVM puts in there by default. Also, the exception class field in
> the ESR (bits 31:26) is 0x0 as well, which translates as an unknown
> exception.
> 
> Are there any kvm related messages in the host kernel log? This looks
> like the result of kvm_inject_undefined(), which prints some kind of
> diagnostic in many cases.

Thanks, Ard!

Zhu Yijun -- can you check this?

Thanks
Laszlo

> 
>> (I still think the only way forward is to instrument fallback.efi, and I
>> won't be doing that.)
>>
> 
> Well, if you have access to the ELF file that fallback.efi was built
> from, you can correlate the stack trace address with locations in the
> code. Lacking that, it would at least be *very* helpful to know which
> opcode is being executed when the exception is taken.
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@lists.01.org
> https://lists.01.org/mailman/listinfo/edk2-devel
>