Message-ID: <1474E0D9-879C-4BBC-BB45-E6F0792DD490@fb.com>
Date:   Mon, 28 Mar 2022 21:57:41 +0000
From:   Song Liu <songliubraving@...com>
To:     Paul Menzel <pmenzel@...gen.mpg.de>
CC:     Rick P Edgecombe <rick.p.edgecombe@...el.com>,
        Song Liu <song@...nel.org>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andrii@...nel.org>,
        Kernel Team <Kernel-team@...com>,
        Peter Zijlstra <peterz@...radead.org>,
        "x86@...nel.org" <x86@...nel.org>,
        "iii@...ux.ibm.com" <iii@...ux.ibm.com>,
        "bpf@...r.kernel.org" <bpf@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        "regressions@...ts.linux.dev" <regressions@...ts.linux.dev>,
        "amd-gfx@...ts.freedesktop.org" <amd-gfx@...ts.freedesktop.org>
Subject: Re: BUG: Bad page state in process systemd-udevd (was: [PATCH v9
 bpf-next 1/9] x86/Kconfig: select HAVE_ARCH_HUGE_VMALLOC with
 HAVE_ARCH_HUGE_VMAP)



> On Mar 28, 2022, at 1:14 PM, Paul Menzel <pmenzel@...gen.mpg.de> wrote:
> 
> Dear Song,
> 
> 
> Am 28.03.22 um 21:24 schrieb Song Liu:
> 
>>> On Mar 27, 2022, at 11:51 PM, Paul Menzel <pmenzel@...gen.mpg.de> wrote:
> 
>>> Am 28.03.22 um 08:37 schrieb Song Liu:
> 
> […]
> 
>>>>> On Mar 27, 2022, at 3:36 AM, Paul Menzel <pmenzel@...gen.mpg.de> wrote:
>>> 
>>>>> Am 26.03.22 um 19:46 schrieb Paul Menzel:
>>>>>> #regzbot introduced: fac54e2bfb5be2b0bbf115fe80d45f59fd773048
>>>>>> #regzbot title: BUG: Bad page state in process systemd-udevd
>>>>> 
>>>>>> Am 04.02.22 um 19:57 schrieb Song Liu:
>>>>>>> From: Song Liu <songliubraving@...com>
>>>>>>> 
>>>>>>> This enables module_alloc() to allocate huge page for 2MB+ requests.
>>>>>>> To check the difference of this change, we need enable config
>>>>>>> CONFIG_PTDUMP_DEBUGFS, and call module_alloc(2MB). Before the change,
>>>>>>> /sys/kernel/debug/page_tables/kernel shows pte for this map. With the
>>>>>>> change, /sys/kernel/debug/page_tables/ show pmd for thie map.
>>>>>>> 
>>>>>>> Signed-off-by: Song Liu <songliubraving@...com>
>>>>>>> ---
>>>>>>>   arch/x86/Kconfig | 1 +
>>>>>>>   1 file changed, 1 insertion(+)
>>>>>>> 
>>>>>>> diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
>>>>>>> index 6fddb63271d9..e0e0d00cf103 100644
>>>>>>> --- a/arch/x86/Kconfig
>>>>>>> +++ b/arch/x86/Kconfig
>>>>>>> @@ -159,6 +159,7 @@ config X86
>>>>>>>       select HAVE_ALIGNED_STRUCT_PAGE        if SLUB
>>>>>>>       select HAVE_ARCH_AUDITSYSCALL
>>>>>>>       select HAVE_ARCH_HUGE_VMAP        if X86_64 || X86_PAE
>>>>>>> +    select HAVE_ARCH_HUGE_VMALLOC        if HAVE_ARCH_HUGE_VMAP
>>>>>>>       select HAVE_ARCH_JUMP_LABEL
>>>>>>>       select HAVE_ARCH_JUMP_LABEL_RELATIVE
>>>>>>>       select HAVE_ARCH_KASAN            if X86_64
>>>>>> Testing Linus’ current master branch, Linux logs critical messages like below:
>>>>>>     BUG: Bad page state in process systemd-udevd  pfn:102e03
>>>>>> I bisected to your commit fac54e2bfb5 (x86/Kconfig: select
>>>>>> HAVE_ARCH_HUGE_VMALLOC with HAVE_ARCH_HUGE_VMAP).
>>>>> Sorry, I forget to mention, that this is a 32-bit (i686) userspace,
>>>>> but a 64-bit Linux kernel, so it might be the same issue as
>>>>> mentioned in commit eed1fcee556f (x86: Disable
>>>>> HAVE_ARCH_HUGE_VMALLOC on 32-bit x86), but didn’t fix the issue for
>>>>> 64-bit Linux kernel and 32-bit userspace.
>>>> I will look more into this tomorrow. To clarify, what is the 32-bit
>>>> user space that triggers this? Is it systemd-udevd? Is the systemd
>>>> also i686?
>>> 
>>> Yes, everything – also systemd – is i686. You can build a 32-bit VM image with grml-debootstrap [1]:
>>> 
>>>    sudo DEBOOTSTRAP=mmdebstrap ~/src/grml-debootstrap/grml-debootstrap --vm --vmfile --vmsize 3G --target /dev/shm/debian-32.img -r sid --arch i686 --filesystem ext4
>>> 
>>> Then run that with QEMU, but pass the 64-bit Linux kernel to QEMU directly with the switches `-kernel` and `-append`, or install the amd64 Linux kernel into the Debian VM image or the package created with `make bindeb-pkg` with `dpkg -i …`.
>> Thanks for these information!
>> I tried the following, but couldn't reproduce the issue.
>> sudo ./grml-debootstrap --vm --vmfile --vmsize 3G --target ../debian-32.img -r sid --arch i386 --filesystem ext4
>> Note: s/i686/i386/. Also I run this on Fedora, so I didn't specify DEBOOTSTRAP.
>> Then I run it with
>> qemu-system-x86_64 \
>>   -boot d ./debian-32.img -m 1024 -smp 4 \
>>   -kernel ./bzImage \
>>   -nographic -append 'root=/dev/sda1 ro console=ttyS0,115200'
>> The VM boots fine. The config being used is x86_64_defconfig +
>> CONFIG_DRM_FBDEV_EMULATION.
>> I wonder whether this is caused by different config or different image.
>> Could you please share your config?
> 
> Sorry, for leading you on the wrong path. I actually just wanted to help getting a 32-bit userspace set up quickly. I haven’t tried reproducing the issue in a VM, and used only the ASUS F2A85-M PRO.
> 
> Booting the system with `nomodeset`, I didn’t see the error. No idea if it’s related to framebuffer handling or specific to AMD graphics device.

I guess this only happens on specific hardware and configuration. 
Let me see what's the best way to not allocate huge pages for this
case. 

Thanks,
Song

> 
>> PS: I couldn't figure out the root password of the image, --password
>> option of grml-debootstrap doesn't seem to work.
> 
> Hmm, I thought it’s asking you during install, but I haven’t done it in a while.
> 
> 
> Kind regards,
> 
> Paul

Powered by blists - more mailing lists