mbox series

[v2,0/2] hw/arm/virt: Support for virtio-mem-pci

Message ID 20211203033522.27580-1-gshan@redhat.com (mailing list archive)
Headers show
Series hw/arm/virt: Support for virtio-mem-pci | expand

Message

Gavin Shan Dec. 3, 2021, 3:35 a.m. UTC
This series supports virtio-mem-pci device, by simply following the
implementation on x86. The exception is the block size is 512MB on
ARM64 instead of 128MB on x86, compatible with the memory section
size in linux guest.

The work was done by David Hildenbrand and then Jonathan Cameron. I'm
taking the patch and putting more efforts, which is all about testing
to me at current stage.

Testing
=======
The upstream linux kernel (v5.16.rc3) is used on host/guest during
the testing. The guest kernel includes changes to enable virtio-mem
driver, which is simply to enable CONFIG_VIRTIO_MEM on ARM64.

Mutiple combinations like page sizes on host/guest, memory backend
device etc are covered in the testing. Besides, migration is also
tested. The following command lines are used for VM or virtio-mem-pci
device hot-add. It's notable that virtio-mem-pci device hot-remove
isn't supported, similar to what we have on x86. 

  host.pgsize  guest.pgsize  backend    hot-add  hot-remove  migration
  ---------------------------------------------------------------------
   4KB         4KB           normal     ok       ok          ok
                             THP        ok       ok          ok
                             hugeTLB    ok       ok          ok
   4KB         64KB          normal     ok       ok          ok
                             THP        ok       ok          ok
                             hugeTLB    ok       ok          ok
  64KB         4KB           normal     ok       ok          ok
                             THP        ok       ok          ok
                             hugeTLB    ok       ok          ok
  64KB         64KB          normal     ok       ok          ok
                             THP        ok       ok          ok
                             hugeTLB    ok       ok          ok

The command lines are used for VM. When hugeTLBfs is used, all memory
backend objects are popuated on /dev/hugepages-2048kB or
/dev/hugepages-524288kB, depending on the host page sizes.

  /home/gavin/sandbox/qemu.main/build/qemu-system-aarch64                       \
  -accel kvm -machine virt,gic-version=host                                     \
  -cpu host -smp 4,sockets=2,cores=2,threads=1                                  \
  -m 1024M,slots=16,maxmem=64G                                                  \
  -object memory-backend-ram,id=mem0,size=512M                                  \
  -object memory-backend-ram,id=mem1,size=512M                                  \
  -numa node,nodeid=0,cpus=0-1,memdev=mem0                                      \
  -numa node,nodeid=1,cpus=2-3,memdev=mem1                                      \
     :
  -kernel /home/gavin/sandbox/linux.guest/arch/arm64/boot/Image                 \
  -initrd /home/gavin/sandbox/images/rootfs.cpio.xz                             \
  -append earlycon=pl011,mmio,0x9000000                                         \
  -device pcie-root-port,bus=pcie.0,chassis=1,id=pcie.1                         \
  -device pcie-root-port,bus=pcie.0,chassis=2,id=pcie.2                         \
  -device pcie-root-port,bus=pcie.0,chassis=3,id=pcie.3                         \
  -object memory-backend-ram,id=vmem0,size=512M                                 \
  -device virtio-mem-pci,id=vm0,bus=pcie.1,memdev=vmem0,node=0,requested-size=0 \
  -object memory-backend-ram,id=vmem1,size=512M                                 \
  -device virtio-mem-pci,id=vm1,bus=pcie.2,memdev=vmem1,node=1,requested-size=0 

Command lines used for memory hot-add and hot-remove:

  (qemu) qom-set vm1 requested-size 512M
  (qemu) qom-set vm1 requested-size 0
  (qemu) qom-set vm1 requested-size 512M

Command lines used for virtio-mem-pci device hot-add:

  (qemu) object_add memory-backend-ram,id=hp-mem1,size=512M
  (qemu) device_add virtio-mem-pci,id=hp-vm1,bus=pcie.3,memdev=hp-mem1,node=1
  (qemu) qom-set hp-vm1 requested-size 512M
  (qemu) qom-set hp-vm1 requested-size 0
  (qemu) qom-set hp-vm1 requested-size 512M

Changelog
=========
v2:
  * Include David/Jonathan as co-developers in the commit log           (David)
  * Decrease VIRTIO_MEM_USABLE_EXTENT to 512MB on ARM64 in PATCH[1/2]   (David)
  * PATCH[2/2] is added to correct the THP sizes on ARM64               (David)

Gavin Shan (2):
  hw/arm/virt: Support for virtio-mem-pci
  virtio-mem: Correct default THP size for ARM64

 hw/arm/Kconfig         |  1 +
 hw/arm/virt.c          | 68 +++++++++++++++++++++++++++++++++++++++++-
 hw/virtio/virtio-mem.c | 36 ++++++++++++++--------
 3 files changed, 91 insertions(+), 14 deletions(-)

Comments

Jonathan Cameron Dec. 3, 2021, 2:10 p.m. UTC | #1
On Fri,  3 Dec 2021 11:35:20 +0800
Gavin Shan <gshan@redhat.com> wrote:

> This series supports virtio-mem-pci device, by simply following the
> implementation on x86. The exception is the block size is 512MB on
> ARM64 instead of 128MB on x86, compatible with the memory section
> size in linux guest.
> 
> The work was done by David Hildenbrand and then Jonathan Cameron. I'm
> taking the patch and putting more efforts, which is all about testing
> to me at current stage.

Hi Gavin,

Thanks for taking this forwards.  What you have here looks good to me, but
I've not looked at this for a while, so I'll go with whatever David and
others say :)

Jonathan

> 
> Testing
> =======
> The upstream linux kernel (v5.16.rc3) is used on host/guest during
> the testing. The guest kernel includes changes to enable virtio-mem
> driver, which is simply to enable CONFIG_VIRTIO_MEM on ARM64.
> 
> Mutiple combinations like page sizes on host/guest, memory backend
> device etc are covered in the testing. Besides, migration is also
> tested. The following command lines are used for VM or virtio-mem-pci
> device hot-add. It's notable that virtio-mem-pci device hot-remove
> isn't supported, similar to what we have on x86. 
> 
>   host.pgsize  guest.pgsize  backend    hot-add  hot-remove  migration
>   ---------------------------------------------------------------------
>    4KB         4KB           normal     ok       ok          ok
>                              THP        ok       ok          ok
>                              hugeTLB    ok       ok          ok
>    4KB         64KB          normal     ok       ok          ok
>                              THP        ok       ok          ok
>                              hugeTLB    ok       ok          ok
>   64KB         4KB           normal     ok       ok          ok
>                              THP        ok       ok          ok
>                              hugeTLB    ok       ok          ok
>   64KB         64KB          normal     ok       ok          ok
>                              THP        ok       ok          ok
>                              hugeTLB    ok       ok          ok
> 
> The command lines are used for VM. When hugeTLBfs is used, all memory
> backend objects are popuated on /dev/hugepages-2048kB or
> /dev/hugepages-524288kB, depending on the host page sizes.
> 
>   /home/gavin/sandbox/qemu.main/build/qemu-system-aarch64                       \
>   -accel kvm -machine virt,gic-version=host                                     \
>   -cpu host -smp 4,sockets=2,cores=2,threads=1                                  \
>   -m 1024M,slots=16,maxmem=64G                                                  \
>   -object memory-backend-ram,id=mem0,size=512M                                  \
>   -object memory-backend-ram,id=mem1,size=512M                                  \
>   -numa node,nodeid=0,cpus=0-1,memdev=mem0                                      \
>   -numa node,nodeid=1,cpus=2-3,memdev=mem1                                      \
>      :
>   -kernel /home/gavin/sandbox/linux.guest/arch/arm64/boot/Image                 \
>   -initrd /home/gavin/sandbox/images/rootfs.cpio.xz                             \
>   -append earlycon=pl011,mmio,0x9000000                                         \
>   -device pcie-root-port,bus=pcie.0,chassis=1,id=pcie.1                         \
>   -device pcie-root-port,bus=pcie.0,chassis=2,id=pcie.2                         \
>   -device pcie-root-port,bus=pcie.0,chassis=3,id=pcie.3                         \
>   -object memory-backend-ram,id=vmem0,size=512M                                 \
>   -device virtio-mem-pci,id=vm0,bus=pcie.1,memdev=vmem0,node=0,requested-size=0 \
>   -object memory-backend-ram,id=vmem1,size=512M                                 \
>   -device virtio-mem-pci,id=vm1,bus=pcie.2,memdev=vmem1,node=1,requested-size=0 
> 
> Command lines used for memory hot-add and hot-remove:
> 
>   (qemu) qom-set vm1 requested-size 512M
>   (qemu) qom-set vm1 requested-size 0
>   (qemu) qom-set vm1 requested-size 512M
> 
> Command lines used for virtio-mem-pci device hot-add:
> 
>   (qemu) object_add memory-backend-ram,id=hp-mem1,size=512M
>   (qemu) device_add virtio-mem-pci,id=hp-vm1,bus=pcie.3,memdev=hp-mem1,node=1
>   (qemu) qom-set hp-vm1 requested-size 512M
>   (qemu) qom-set hp-vm1 requested-size 0
>   (qemu) qom-set hp-vm1 requested-size 512M
> 
> Changelog
> =========
> v2:
>   * Include David/Jonathan as co-developers in the commit log           (David)
>   * Decrease VIRTIO_MEM_USABLE_EXTENT to 512MB on ARM64 in PATCH[1/2]   (David)
>   * PATCH[2/2] is added to correct the THP sizes on ARM64               (David)
> 
> Gavin Shan (2):
>   hw/arm/virt: Support for virtio-mem-pci
>   virtio-mem: Correct default THP size for ARM64
> 
>  hw/arm/Kconfig         |  1 +
>  hw/arm/virt.c          | 68 +++++++++++++++++++++++++++++++++++++++++-
>  hw/virtio/virtio-mem.c | 36 ++++++++++++++--------
>  3 files changed, 91 insertions(+), 14 deletions(-)
>
Gavin Shan Dec. 3, 2021, 11:06 p.m. UTC | #2
Hi Jonathan,

On 12/4/21 1:10 AM, Jonathan Cameron wrote:
> On Fri,  3 Dec 2021 11:35:20 +0800
> Gavin Shan <gshan@redhat.com> wrote:
> 
>> This series supports virtio-mem-pci device, by simply following the
>> implementation on x86. The exception is the block size is 512MB on
>> ARM64 instead of 128MB on x86, compatible with the memory section
>> size in linux guest.
>>
>> The work was done by David Hildenbrand and then Jonathan Cameron. I'm
>> taking the patch and putting more efforts, which is all about testing
>> to me at current stage.
> 
> Thanks for taking this forwards.  What you have here looks good to me, but
> I've not looked at this for a while, so I'll go with whatever David and
> others say :)
> 

[...]

I would translate this as your reviewed-by tag, which will be added to v3.
However, it shouldn't stop you from further reviewing :)

Thanks,
Gavin