[Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Peter Wu]

Hi,

With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
`qemu-system-x86_64 -enable-kvm` fails to run with the following error:

    kvm_set_phys_mem: error unregistering overlapping slot: Invalid argument

After this message, qemu aborts. Reverting the following commit on top of
master makes the bug go away. Removing the `-enable-kvm` option also
allows QEMU to operate, but without KVM.

commit 235e8982ad393e5611cb892df54881c872eea9e1
Author: Jordan Justen <jordan.l.justen@intel.com>
Date:   Wed May 29 01:27:26 2013 -0700

    kvm: support using KVM_MEM_READONLY flag for regions
    
    For readonly memory regions and rom devices in romd_mode,
    we make use of the KVM_MEM_READONLY. A slot that uses
    KVM_MEM_READONLY can be read from and code can execute from the
    region, but writes will exit to qemu.
    
    For rom devices with !romd_mode, we force the slot to be
    removed so reads or writes to the region will exit to qemu.
    (Note that a memory region in this state is not executable
    within kvm.)

QEMU was built using GCC 4.8.0, runs on Arch Linux 64-bit with a
custom kernel, Linux 3.9.1. CPU: i5-460M. A backtrace is provided
on the bottom of this message. If you need more details, please ask.

Regards,
Peter

#0  0x00007ffff1a251c9 in raise () from /usr/lib/libc.so.6
#1  0x00007ffff1a265c8 in abort () from /usr/lib/libc.so.6
#2  0x00005555557dd84d in kvm_set_phys_mem (section=0x7fffe6fdca00, add=false) at /tmp/qemu/kvm-all.c:715
#3  0x00005555557e132c in address_space_update_topology_pass (as=as@entry=0x5555564dae00 <address_space_memory>, adding=adding@entry=false, old_view=..., new_view=...) at /tmp/qemu/memory.c:725
#4  0x00005555557e1f13 in address_space_update_topology (as=0x5555564dae00 <address_space_memory>) at /tmp/qemu/memory.c:761
#5  memory_region_transaction_commit () at /tmp/qemu/memory.c:786
#6  0x00005555557dfa42 in access_with_adjusted_size (addr=addr@entry=2, value=value@entry=0x7fffe6fdcc18, size=1, access_size_min=<optimized out>, access_size_max=<optimized out>, 
    access=access@entry=0x5555557e0000 <memory_region_write_accessor>, opaque=opaque@entry=0x55555658c238) at /tmp/qemu/memory.c:399
#7  0x00005555557e0f77 in memory_region_iorange_write (iorange=<optimized out>, offset=2, width=1, data=51) at /tmp/qemu/memory.c:475
#8  0x00005555557de562 in kvm_handle_io (count=1, size=1, direction=1, data=<optimized out>, port=3326) at /tmp/qemu/kvm-all.c:1507
#9  kvm_cpu_exec (env=env@entry=0x55555656f750) at /tmp/qemu/kvm-all.c:1659
#10 0x000055555578da05 in qemu_kvm_cpu_thread_fn (arg=0x55555656f750) at /tmp/qemu/cpus.c:759
#11 0x00007ffff6280dd2 in start_thread () from /usr/lib/libpthread.so.0
#12 0x00007ffff1ad5cdd in clone () from /usr/lib/libc.so.6

Re: [Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Jordan Justen]

This should be fixed by Xiao's "fix double free the memslot in
kvm_set_phys_mem" patch.

On Mon, Jun 3, 2013 at 10:23 AM, Peter Wu <lekensteyn@gmail.com> wrote:
> Hi,
>
> With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
> `qemu-system-x86_64 -enable-kvm` fails to run with the following error:
>
>     kvm_set_phys_mem: error unregistering overlapping slot: Invalid argument
>
> After this message, qemu aborts. Reverting the following commit on top of
> master makes the bug go away. Removing the `-enable-kvm` option also
> allows QEMU to operate, but without KVM.
>
> commit 235e8982ad393e5611cb892df54881c872eea9e1
> Author: Jordan Justen <jordan.l.justen@intel.com>
> Date:   Wed May 29 01:27:26 2013 -0700
>
>     kvm: support using KVM_MEM_READONLY flag for regions
>
>     For readonly memory regions and rom devices in romd_mode,
>     we make use of the KVM_MEM_READONLY. A slot that uses
>     KVM_MEM_READONLY can be read from and code can execute from the
>     region, but writes will exit to qemu.
>
>     For rom devices with !romd_mode, we force the slot to be
>     removed so reads or writes to the region will exit to qemu.
>     (Note that a memory region in this state is not executable
>     within kvm.)
>
> QEMU was built using GCC 4.8.0, runs on Arch Linux 64-bit with a
> custom kernel, Linux 3.9.1. CPU: i5-460M. A backtrace is provided
> on the bottom of this message. If you need more details, please ask.
>
> Regards,
> Peter
>
> #0  0x00007ffff1a251c9 in raise () from /usr/lib/libc.so.6
> #1  0x00007ffff1a265c8 in abort () from /usr/lib/libc.so.6
> #2  0x00005555557dd84d in kvm_set_phys_mem (section=0x7fffe6fdca00, add=false) at /tmp/qemu/kvm-all.c:715
> #3  0x00005555557e132c in address_space_update_topology_pass (as=as@entry=0x5555564dae00 <address_space_memory>, adding=adding@entry=false, old_view=..., new_view=...) at /tmp/qemu/memory.c:725
> #4  0x00005555557e1f13 in address_space_update_topology (as=0x5555564dae00 <address_space_memory>) at /tmp/qemu/memory.c:761
> #5  memory_region_transaction_commit () at /tmp/qemu/memory.c:786
> #6  0x00005555557dfa42 in access_with_adjusted_size (addr=addr@entry=2, value=value@entry=0x7fffe6fdcc18, size=1, access_size_min=<optimized out>, access_size_max=<optimized out>,
>     access=access@entry=0x5555557e0000 <memory_region_write_accessor>, opaque=opaque@entry=0x55555658c238) at /tmp/qemu/memory.c:399
> #7  0x00005555557e0f77 in memory_region_iorange_write (iorange=<optimized out>, offset=2, width=1, data=51) at /tmp/qemu/memory.c:475
> #8  0x00005555557de562 in kvm_handle_io (count=1, size=1, direction=1, data=<optimized out>, port=3326) at /tmp/qemu/kvm-all.c:1507
> #9  kvm_cpu_exec (env=env@entry=0x55555656f750) at /tmp/qemu/kvm-all.c:1659
> #10 0x000055555578da05 in qemu_kvm_cpu_thread_fn (arg=0x55555656f750) at /tmp/qemu/cpus.c:759
> #11 0x00007ffff6280dd2 in start_thread () from /usr/lib/libpthread.so.0
> #12 0x00007ffff1ad5cdd in clone () from /usr/lib/libc.so.6
>
>

Re: [Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Anthony Liguori]

Peter Wu <lekensteyn@gmail.com> writes:

> Hi,
>
> With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
> `qemu-system-x86_64 -enable-kvm` fails to run with the following error:
>
>     kvm_set_phys_mem: error unregistering overlapping slot: Invalid argument
>
> After this message, qemu aborts. Reverting the following commit on top of
> master makes the bug go away. Removing the `-enable-kvm` option also
> allows QEMU to operate, but without KVM.

I also noted this on F19 this weekend.

Is anyone else seeing this on more recent kernels?

Regards,

Anthony Liguori

>
> commit 235e8982ad393e5611cb892df54881c872eea9e1
> Author: Jordan Justen <jordan.l.justen@intel.com>
> Date:   Wed May 29 01:27:26 2013 -0700
>
>     kvm: support using KVM_MEM_READONLY flag for regions
>     
>     For readonly memory regions and rom devices in romd_mode,
>     we make use of the KVM_MEM_READONLY. A slot that uses
>     KVM_MEM_READONLY can be read from and code can execute from the
>     region, but writes will exit to qemu.
>     
>     For rom devices with !romd_mode, we force the slot to be
>     removed so reads or writes to the region will exit to qemu.
>     (Note that a memory region in this state is not executable
>     within kvm.)
>
> QEMU was built using GCC 4.8.0, runs on Arch Linux 64-bit with a
> custom kernel, Linux 3.9.1. CPU: i5-460M. A backtrace is provided
> on the bottom of this message. If you need more details, please ask.
>
> Regards,
> Peter
>
> #0  0x00007ffff1a251c9 in raise () from /usr/lib/libc.so.6
> #1  0x00007ffff1a265c8 in abort () from /usr/lib/libc.so.6
> #2  0x00005555557dd84d in kvm_set_phys_mem (section=0x7fffe6fdca00, add=false) at /tmp/qemu/kvm-all.c:715
> #3  0x00005555557e132c in address_space_update_topology_pass (as=as@entry=0x5555564dae00 <address_space_memory>, adding=adding@entry=false, old_view=..., new_view=...) at /tmp/qemu/memory.c:725
> #4  0x00005555557e1f13 in address_space_update_topology (as=0x5555564dae00 <address_space_memory>) at /tmp/qemu/memory.c:761
> #5  memory_region_transaction_commit () at /tmp/qemu/memory.c:786
> #6  0x00005555557dfa42 in access_with_adjusted_size (addr=addr@entry=2, value=value@entry=0x7fffe6fdcc18, size=1, access_size_min=<optimized out>, access_size_max=<optimized out>, 
>     access=access@entry=0x5555557e0000 <memory_region_write_accessor>, opaque=opaque@entry=0x55555658c238) at /tmp/qemu/memory.c:399
> #7  0x00005555557e0f77 in memory_region_iorange_write (iorange=<optimized out>, offset=2, width=1, data=51) at /tmp/qemu/memory.c:475
> #8  0x00005555557de562 in kvm_handle_io (count=1, size=1, direction=1, data=<optimized out>, port=3326) at /tmp/qemu/kvm-all.c:1507
> #9  kvm_cpu_exec (env=env@entry=0x55555656f750) at /tmp/qemu/kvm-all.c:1659
> #10 0x000055555578da05 in qemu_kvm_cpu_thread_fn (arg=0x55555656f750) at /tmp/qemu/cpus.c:759
> #11 0x00007ffff6280dd2 in start_thread () from /usr/lib/libpthread.so.0
> #12 0x00007ffff1ad5cdd in clone () from /usr/lib/libc.so.6

Re: [Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Peter Wu]

On Monday 03 June 2013 10:35:28 Jordan Justen wrote:
> This should be fixed by Xiao's "fix double free the memslot in
> kvm_set_phys_mem" patch.
Confirmed, with this one-liner patch the bug is gone.

> On Mon, Jun 3, 2013 at 10:23 AM, Peter Wu <lekensteyn@gmail.com> wrote:
> > Hi,
> > 
> > With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
> > 
> > `qemu-system-x86_64 -enable-kvm` fails to run with the following error:
> >     kvm_set_phys_mem: error unregistering overlapping slot: Invalid
> >     argument
> > 
> > After this message, qemu aborts. Reverting the following commit on top of
> > master makes the bug go away. Removing the `-enable-kvm` option also
> > allows QEMU to operate, but without KVM.
> > 
> > commit 235e8982ad393e5611cb892df54881c872eea9e1
> > Author: Jordan Justen <jordan.l.justen@intel.com>
> > Date:   Wed May 29 01:27:26 2013 -0700
> > 
> >     kvm: support using KVM_MEM_READONLY flag for regions
> >     
> >     For readonly memory regions and rom devices in romd_mode,
> >     we make use of the KVM_MEM_READONLY. A slot that uses
> >     KVM_MEM_READONLY can be read from and code can execute from the
> >     region, but writes will exit to qemu.
> >     
> >     For rom devices with !romd_mode, we force the slot to be
> >     removed so reads or writes to the region will exit to qemu.
> >     (Note that a memory region in this state is not executable
> >     within kvm.)
> > 
> > QEMU was built using GCC 4.8.0, runs on Arch Linux 64-bit with a
> > custom kernel, Linux 3.9.1. CPU: i5-460M. A backtrace is provided
> > on the bottom of this message. If you need more details, please ask.
> > 
> > Regards,
> > Peter
> > 
> > #0  0x00007ffff1a251c9 in raise () from /usr/lib/libc.so.6
> > #1  0x00007ffff1a265c8 in abort () from /usr/lib/libc.so.6
> > #2  0x00005555557dd84d in kvm_set_phys_mem (section=0x7fffe6fdca00,
> > add=false) at /tmp/qemu/kvm-all.c:715 #3  0x00005555557e132c in
> > address_space_update_topology_pass (as=as@entry=0x5555564dae00
> > <address_space_memory>, adding=adding@entry=false, old_view=...,
> > new_view=...) at /tmp/qemu/memory.c:725 #4  0x00005555557e1f13 in
> > address_space_update_topology (as=0x5555564dae00 <address_space_memory>)
> > at /tmp/qemu/memory.c:761 #5  memory_region_transaction_commit () at
> > /tmp/qemu/memory.c:786
> > #6  0x00005555557dfa42 in access_with_adjusted_size (addr=addr@entry=2,
> > value=value@entry=0x7fffe6fdcc18, size=1, access_size_min=<optimized
> > out>, access_size_max=<optimized out>,> 
> >     access=access@entry=0x5555557e0000 <memory_region_write_accessor>,
> >     opaque=opaque@entry=0x55555658c238) at /tmp/qemu/memory.c:399> 
> > #7  0x00005555557e0f77 in memory_region_iorange_write (iorange=<optimized
> > out>, offset=2, width=1, data=51) at /tmp/qemu/memory.c:475 #8 
> > 0x00005555557de562 in kvm_handle_io (count=1, size=1, direction=1,
> > data=<optimized out>, port=3326) at /tmp/qemu/kvm-all.c:1507 #9 
> > kvm_cpu_exec (env=env@entry=0x55555656f750) at /tmp/qemu/kvm-all.c:1659
> > #10 0x000055555578da05 in qemu_kvm_cpu_thread_fn (arg=0x55555656f750) at
> > /tmp/qemu/cpus.c:759 #11 0x00007ffff6280dd2 in start_thread () from
> > /usr/lib/libpthread.so.0 #12 0x00007ffff1ad5cdd in clone () from
> > /usr/lib/libc.so.6

Re: [Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Paolo Bonzini]

Il 03/06/2013 19:49, Anthony Liguori ha scritto:
> Peter Wu <lekensteyn@gmail.com> writes:
> 
>> Hi,
>>
>> With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
>> `qemu-system-x86_64 -enable-kvm` fails to run with the following error:
>>
>>     kvm_set_phys_mem: error unregistering overlapping slot: Invalid argument
>>
>> After this message, qemu aborts. Reverting the following commit on top of
>> master makes the bug go away. Removing the `-enable-kvm` option also
>> allows QEMU to operate, but without KVM.
> 
> I also noted this on F19 this weekend.
> 
> Is anyone else seeing this on more recent kernels?

Yes, it's caused by 3.9.  Gleb and I decided to consider this (raising
an error when removing a non-existent slot) a feature since it never
happened as far as we know on released versions of QEMU/nlkt/whatever.
Gleb queued a patch in uq/master, I think he'll be sending a pull
request soon.

Paolo

Re: [Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Anthony Liguori]

Paolo Bonzini <pbonzini@redhat.com> writes:

> Il 03/06/2013 19:49, Anthony Liguori ha scritto:
>> Peter Wu <lekensteyn@gmail.com> writes:
>> 
>>> Hi,
>>>
>>> With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
>>> `qemu-system-x86_64 -enable-kvm` fails to run with the following error:
>>>
>>>     kvm_set_phys_mem: error unregistering overlapping slot: Invalid argument
>>>
>>> After this message, qemu aborts. Reverting the following commit on top of
>>> master makes the bug go away. Removing the `-enable-kvm` option also
>>> allows QEMU to operate, but without KVM.
>> 
>> I also noted this on F19 this weekend.
>> 
>> Is anyone else seeing this on more recent kernels?
>
> Yes, it's caused by 3.9.  Gleb and I decided to consider this (raising
> an error when removing a non-existent slot) a feature since it never
> happened as far as we know on released versions of QEMU/nlkt/whatever.
> Gleb queued a patch in uq/master, I think he'll be sending a pull
> request soon.

Since I suspect Gleb is done for the day, I will apply directly since
this is a pretty major breakage.

Regards,

Anthony Liguori

>
> Paolo

Re: [Qemu-devel] QEMU aborts since "kvm: support using KVM_MEM_READONLY flag for regions"

[Gleb Natapov]

On Mon, Jun 03, 2013 at 03:20:48PM -0500, Anthony Liguori wrote:
> Paolo Bonzini <pbonzini@redhat.com> writes:
> 
> > Il 03/06/2013 19:49, Anthony Liguori ha scritto:
> >> Peter Wu <lekensteyn@gmail.com> writes:
> >> 
> >>> Hi,
> >>>
> >>> With the current git source (f10acc8b38d65a66ffa0588a036489d7fa6a593e),
> >>> `qemu-system-x86_64 -enable-kvm` fails to run with the following error:
> >>>
> >>>     kvm_set_phys_mem: error unregistering overlapping slot: Invalid argument
> >>>
> >>> After this message, qemu aborts. Reverting the following commit on top of
> >>> master makes the bug go away. Removing the `-enable-kvm` option also
> >>> allows QEMU to operate, but without KVM.
> >> 
> >> I also noted this on F19 this weekend.
> >> 
> >> Is anyone else seeing this on more recent kernels?
> >
> > Yes, it's caused by 3.9.  Gleb and I decided to consider this (raising
> > an error when removing a non-existent slot) a feature since it never
> > happened as far as we know on released versions of QEMU/nlkt/whatever.
> > Gleb queued a patch in uq/master, I think he'll be sending a pull
> > request soon.
> 
> Since I suspect Gleb is done for the day, I will apply directly since
> this is a pretty major breakage.
> 
Heh, since patch was on the list for 3 days before I picked it (was traveling
before that) I thought I can spend one day more for testing before
sending pull request :) The patch is already queued on uq/master and I'd
rather do not rebase this branch, so I'll push it to you anyway. git
will know how to handle this.

--
			Gleb.