Re: [Qemu-devel] [PATCH 2/2] memory: hide mr->ram_addr from qemu_get_ram_ptr users

[Fam Zheng <famz@redhat.com>]

On Fri, 03/25 07:58, Paolo Bonzini wrote:
> 
> 
> ----- Original Message -----
> > From: "Fam Zheng" <famz@redhat.com>
> > To: "Paolo Bonzini" <pbonzini@redhat.com>
> > Cc: qemu-devel@nongnu.org, "arei gonglei" <arei.gonglei@huawei.com>, mst@redhat.com
> > Sent: Friday, March 25, 2016 7:20:38 AM
> > Subject: Re: [PATCH 2/2] memory: hide mr->ram_addr from qemu_get_ram_ptr users
> > 
> > On Thu, 03/24 12:03, Paolo Bonzini wrote:
> > > Let users of qemu_get_ram_ptr and qemu_ram_ptr_length pass in an
> > > address that is relative to the MemoryRegion.  This basically means
> > > what address_space_translate returns.
> > > 
> > > invalidate_and_set_dirty has to add back mr->ram_addr, but reads do
> > > not need it at all.
> > > 
> > > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> > > ---
> > >  exec.c                       | 40 +++++++++++++++-------------------------
> > >  include/exec/memory.h        |  1 -
> > >  memory.c                     |  4 ++--
> > >  scripts/dump-guest-memory.py | 19 +++----------------
> > >  4 files changed, 20 insertions(+), 44 deletions(-)
> > > 
> > > diff --git a/exec.c b/exec.c
> > > index 001b669..ca9e3b6 100644
> > > --- a/exec.c
> > > +++ b/exec.c
> > > @@ -1876,6 +1876,7 @@ void *qemu_get_ram_ptr(RAMBlock *ram_block,
> > > ram_addr_t addr)
> > 
> > Shall we rename the parameter to "offset" then?  I don't know, but that seems
> > easier to read for me.
> 
> Good question.  I'm not sure about that because of the block == NULL case,
> where the address is absolute.
> 
> > > @@ -1924,7 +1924,7 @@ static void *qemu_ram_ptr_length(RAMBlock *ram_block,
> > > ram_addr_t addr,
> > >          block->host = xen_map_cache(block->offset, block->max_length, 1);
> > >      }
> > >  
> > > -    return ramblock_ptr(block, offset_inside_block);
> > > +    return ramblock_ptr(block, addr);
> > >  }
> > >  
> > >  /*
> > > @@ -2504,6 +2504,8 @@ static void invalidate_and_set_dirty(MemoryRegion
> > > *mr, hwaddr addr,
> > >                                       hwaddr length)
> > >  {
> > >      uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr);
> > > +    addr += memory_region_get_ram_addr(mr);
> > > +
> > 
> > If called by address_space_unmap, is this addition still correct?
> 
> No, thanks for the careful review!  That's another opportunity
> for cleanup actually, splitting the (few) users of qemu_ram_addr_from_host
> that really need a ram_addr_t and those (the majority) that need a
> MemoryRegion and offset.  They can use two different functions.  I'll
> defer this to 2.7 and post the patches to do so later.

Good idea. The above "block == NULL" qemu_get_ram_ptr callers could use a
separate function, too - frankly I don't like that function interface too much.
What do you think?

Fam