Re: [Qemu-devel] [PATCH v2 0/7] memory: Clean up MemoryRegion.ram_addr and optimize address_space_translate

[Fam Zheng <famz@redhat.com>]

On Mon, 03/07 09:53, Laszlo Ersek wrote:
> (CC Janosch)
> 
> Hi,
> 
> On 03/01/16 07:18, Fam Zheng wrote:
> > v2: In the optimization patch, factor out section_covers_addr() and use it.
> >     [Paolo, Peter]
> >     Check "ram_block == NULL" in patch 3. [Gonglei]
> >     Add Gonglei's rev-by in patches 1, 2, 4 and 5.
> > 
> > The first four patches drop ram_addr from MemoryRegion on top of Gonglei's
> > optimization.
> > 
> > The next patch simplifies qemu_ram_free a bit by passing the RAMBlock pointer.
> > 
> > The last patch speeds up address_space_translate with a cache pointer inside
> > the AddressSpaceDispatch.
> > 
> > Fam Zheng (7):
> >   exec: Return RAMBlock pointer from allocating functions
> >   memory: Move assignment to ram_block to memory_region_init_*
> >   memory: Implement memory_region_get_ram_addr with mr->ram_block
> >   memory: Drop MemoryRegion.ram_addr
> >   exec: Pass RAMBlock pointer to qemu_ram_free
> >   exec: Factor out section_covers_addr
> >   exec: Introduce AddressSpaceDispatch.mru_section
> > 
> >  cputlb.c                |   4 +-
> >  exec.c                  | 106 +++++++++++++++++++++++++-----------------------
> >  hw/misc/ivshmem.c       |   9 ++--
> >  include/exec/memory.h   |   9 +---
> >  include/exec/ram_addr.h |  24 +++++------
> >  kvm-all.c               |   3 +-
> >  memory.c                |  56 ++++++++++++++-----------
> >  7 files changed, 111 insertions(+), 100 deletions(-)
> > 
> 
> Does this series preserve "scripts/dump-guest-memory.py" in working
> shape? One of the patch titles above says "Drop MemoryRegion.ram_addr",
> and I think that might break the memory_region_get_ram_ptr() method.

Indeed, the 'memory_region["ram_addr"]' expression is broken. I'll send a
separate patch.

Fam