[Qemu-devel] [RFC/PATCHSET 0/7] virtio: Implement virtio pstore device (v2)

[Namhyung Kim]

Hello,

This is v2 of the virtio-pstore work.  In this patchset I addressed
most of feedbacks from previous version.  Limiting disk size is not
implemented yet.

 * changes in v2)
  - update VIRTIO_ID_PSTORE to 22  (Cornelia, Stefan)
  - make buffer size configurable  (Cornelia)
  - support PSTORE_TYPE_CONSOLE  (Kees)
  - use separate virtqueues for read and write
  - support concurrent async write
  - manage pstore (file) id in device side
  - fix various mistakes in qemu device  (Stefan)

It started from the fact that dumping ftrace buffer at kernel
oops/panic takes too much time.  Although there's a way to reduce the
size of the original data, sometimes I want to have the information as
many as possible.  Maybe kexec/kdump can solve this problem but it
consumes some portion of guest memory so I'd like to avoid it.  And I
know the qemu + crashtool can dump and analyze the whole guest memory
including the ftrace buffer without wasting guest memory, but it adds
one more layer and has some limitation as an out-of-tree tool like not
being in sync with the kernel changes.

So I think it'd be great using the pstore interface to dump guest
kernel data on the host.  One can read the data on the host directly
or on the guest (at the next boot) using pstore filesystem as usual.
While this patchset only implements dumping kernel log buffer, it can
be extended to have ftrace buffer and probably some more..

The patch 0001-0003 are preparation for pstore to support virtio
device which requires async write.  The patch 0004 implements virtio
pstore driver.  It has two virt queue for (sync) read and (async)
write, pstore buffer and io request and response structure.  The
virtio_pstore_req struct is to give information about the current
pstore operation.  The result will be written to the virtio_pstore_res
struct.  For read operation it also uses virtio_pstore_fileinfo struct.

The patch 0005 adds support for PSTORE_TYPE_CONSOLE which was
requested by Kees.  The console data is appended to a single file for
now.

The patch 0006 and 0007 implement virtio-pstore legacy PCI device on
qemu-kvm and kvmtool respectively.  I referenced virtio-baloon and
virtio-rng implementations and I don't know whether kvmtool supports
modern virtio 1.0+ spec.  Other transports might be supported later.

For example, using virtio-pstore on qemu looks like below:

  $ qemu-system-x86_64 -enable-kvm -device virtio-pstore,directory=xxx

When guest kernel gets panic the log messages will be saved under the
xxx directory.

  $ ls xxx
  dmesg-1.enc.z  dmesg-2.enc.z

As you can see the pstore subsystem compresses the log data using zlib
(now supports lzo and lz4 too).  The data can be extracted with the
following command:

  $ cat xxx/dmesg-1.enc.z | \
  > python -c 'import sys, zlib; print(zlib.decompress(sys.stdin.read()))'
  Oops#1 Part1
  <5>[    0.000000] Linux version 4.6.0kvm+ (address@hidden) (gcc version 5.3.0 
(GCC) ) #145 SMP Mon Jul 18 10:22:45 KST 2016
  <6>[    0.000000] Command line: root=/dev/vda console=ttyS0
  <6>[    0.000000] x86/fpu: Legacy x87 FPU detected.
  <6>[    0.000000] x86/fpu: Using 'eager' FPU context switches.
  <6>[    0.000000] e820: BIOS-provided physical RAM map:
  <6>[    0.000000] BIOS-e820: [mem 0x0000000000000000-0x000000000009fbff] 
usable
  <6>[    0.000000] BIOS-e820: [mem 0x000000000009fc00-0x000000000009ffff] 
reserved
  <6>[    0.000000] BIOS-e820: [mem 0x00000000000f0000-0x00000000000fffff] 
reserved
  <6>[    0.000000] BIOS-e820: [mem 0x0000000000100000-0x0000000007fddfff] 
usable
  <6>[    0.000000] BIOS-e820: [mem 0x0000000007fde000-0x0000000007ffffff] 
reserved
  <6>[    0.000000] BIOS-e820: [mem 0x00000000feffc000-0x00000000feffffff] 
reserved
  <6>[    0.000000] BIOS-e820: [mem 0x00000000fffc0000-0x00000000ffffffff] 
reserved
  <6>[    0.000000] NX (Execute Disable) protection: active
  <6>[    0.000000] SMBIOS 2.8 present.
  <7>[    0.000000] DMI: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 
rel-1.9.3-0-ge2fc41e-prebuilt.qemu-project.org 04/01/2014
  ...

To enable PSTORE_TYPE_CONSOLE, add 'console=true' to virtio-pstore
device option.  Also 'bufsize' option can set different size for
pstore buffer (default is 16K).  Maybe we can add a config option to
control the compression later.

Currently the kvmtool doesn't support any options except the directory
the pstore saves the logs.


Namhyung Kim (7):
  pstore: Split pstore fragile flags
  pstore/ram: Set pstore flags dynamically
  pstore: Manage buffer position for async write
  virtio: Basic implementation of virtio pstore driver
  virtio-pstore: Support PSTORE_TYPE_CONSOLE
  qemu: Implement virtio-pstore device
  kvmtool: Implement virtio-pstore device

 drivers/acpi/apei/erst.c             |   2 +-
 drivers/firmware/efi/efi-pstore.c    |   4 +-
 drivers/virtio/Kconfig               |  10 +
 drivers/virtio/Makefile              |   1 +
 drivers/virtio/virtio_pstore.c       | 421 +++++++++++++++++++++
 fs/pstore/platform.c                 |  65 +++-
 fs/pstore/ram.c                      |   8 +
 include/linux/pstore.h               |   9 +-
 include/uapi/linux/Kbuild            |   1 +
 include/uapi/linux/virtio_ids.h      |   1 +
 include/uapi/linux/virtio_pstore.h   |  78 +++-
 11 files changed, 580 insertions(+), 20 deletions(-)
 create mode 100644 drivers/virtio/virtio_pstore.c
 create mode 100644 include/uapi/linux/virtio_pstore.h


Cc: Paolo Bonzini <address@hidden>
Cc: Radim Krčmář <address@hidden>
Cc: "Michael S. Tsirkin" <address@hidden>
Cc: Anthony Liguori <address@hidden>
Cc: Anton Vorontsov <address@hidden>
Cc: Colin Cross <address@hidden>
Cc: Kees Cook <address@hidden>
Cc: Tony Luck <address@hidden>
Cc: Steven Rostedt <address@hidden>
Cc: Ingo Molnar <address@hidden>
Cc: Minchan Kim <address@hidden>
Cc: Will Deacon <address@hidden>
Cc: address@hidden
Cc: address@hidden
Cc: address@hidden

Thanks,
Namhyung


-- 
2.8.0