On Thu, Mar 26, 2015 at 06:38:18PM +0100, Paolo Bonzini wrote:
> QEMU is currently accessing the dirty bitmaps very liberally,
> which is understandable since the accesses are cheap.  This is
> however not good for squeezing maximum performance out of dataplane,
> and is also not good if the accesses become more expensive---as is
> the case when they use atomic primitives.
> 
> Patches 1-2 make acpi-build.h only use public memory APIs.
> 
> Patches 3-7 optimize access to the VGA dirty bitmap, by restricting
> it to video RAM only.
> 
> Patches 8-15 optimize access to the code and migration bitmaps,
> by tracking them respectively if TCG is enabled and if migration
> is in progress.  Note that the first iteration of migration already
> does not look at the migration bitmap (commit 70c8652, migration:
> do not search dirty pages in bulk stage, 2013-03-26).
> 
> Patches 16-21 are Stefan's patches to convert bitmap access to use
> atomic primitives.
> 
> While the main purpose of these patches is a working dirty bitmap
> for dataplane (and possibly multithreaded TCG), there's something
> that they are immediately useful for: patch 22 makes the migration
> thread synchronize the bitmap outside the big QEMU lock, thus
> removing the last source of jitter during the RAM copy phase of
> migration.
> 
> Please review and test! (it's available as branch "atomic-dirty"
> on my github repository)  In particular, I suspect that the
> postcopy patches might be good at finding bugs.
> 
> Paolo
> 
> Paolo Bonzini (16):
>   memory: add memory_region_ram_resize
>   acpi-build: remove dependency from ram_addr.h
>   memory: the only dirty memory flag for users is DIRTY_MEMORY_VGA
>   display: enable DIRTY_MEMORY_VGA tracking explicitly
>   memory: return bitmap from memory_region_is_logging
>   framebuffer: check memory_region_is_logging
>   ui/console: check memory_region_is_logging
>   memory: track DIRTY_MEMORY_CODE in mr->dirty_log_mask
>   memory: return DIRTY_MEMORY_MIGRATION from memory_region_is_logging
>   ram_addr: tweaks to xen_modified_memory
>   exec: simplify notdirty_mem_write
>   exec: use memory_region_is_logging to optimize dirty tracking
>   exec: pass client mask to cpu_physical_memory_set_dirty_range
>   exec: only check relevant bitmaps for cleanliness
>   memory: do not touch code dirty bitmap unless TCG is enabled
>   migration: run bitmap sync outside iothread lock
> 
> Stefan Hajnoczi (6):
>   bitmap: add atomic set functions
>   bitmap: add atomic test and clear
>   memory: use atomic ops for setting dirty memory bits
>   migration: move dirty bitmap sync to ram_addr.h
>   memory: replace cpu_physical_memory_reset_dirty() with test-and-clear
>   memory: make cpu_physical_memory_sync_dirty_bitmap() fully atomic
> 
>  arch_init.c                  |  56 +++----------------
>  cputlb.c                     |   4 +-
>  exec.c                       | 103 ++++++++++++++++------------------
>  hw/core/loader.c             |   8 +--
>  hw/display/cg3.c             |   1 +
>  hw/display/exynos4210_fimd.c |   7 ++-
>  hw/display/framebuffer.c     |  23 ++++++--
>  hw/display/g364fb.c          |   2 +-
>  hw/display/sm501.c           |   1 +
>  hw/display/tcx.c             |   1 +
>  hw/display/vmware_vga.c      |   2 +-
>  hw/i386/acpi-build.c         |  36 ++++++------
>  hw/virtio/vhost.c            |   3 +-
>  include/exec/memory.h        |  27 +++++++--
>  include/exec/ram_addr.h      | 128 ++++++++++++++++++++++++++++---------------
>  include/hw/loader.h          |   8 ++-
>  include/qemu/bitmap.h        |   4 ++
>  include/qemu/bitops.h        |  14 +++++
>  kvm-all.c                    |   3 +-
>  memory.c                     |  34 ++++++++----
>  ui/console.c                 |  14 +++--
>  util/bitmap.c                |  78 ++++++++++++++++++++++++++
>  xen-hvm.c                    |   3 +-
>  23 files changed, 356 insertions(+), 204 deletions(-)

Modulo the comments that have already been posted:

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>