diff mbox

[07/15] memory: avoid ref/unref in memory_region_find

Message ID 1421938053-10318-8-git-send-email-pbonzini@redhat.com
State New
Headers show

Commit Message

Paolo Bonzini Jan. 22, 2015, 2:47 p.m. UTC
Do the entire lookup under RCU, which avoids atomic operations.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 memory.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

Comments

Fam Zheng Jan. 26, 2015, 6:24 a.m. UTC | #1
On Thu, 01/22 15:47, Paolo Bonzini wrote:
> Do the entire lookup under RCU, which avoids atomic operations.

address_space_get_flatview() already is RCU protected, I don't see why this
patch is necessary. Could you explain?

And there is one question below:

> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  memory.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/memory.c b/memory.c
> index a844ced..577e87c 100644
> --- a/memory.c
> +++ b/memory.c
> @@ -1828,7 +1828,8 @@ MemoryRegionSection memory_region_find(MemoryRegion *mr,
>      }
>      range = addrrange_make(int128_make64(addr), int128_make64(size));
>  
> -    view = address_space_get_flatview(as);
> +    rcu_read_lock();
> +    view = atomic_rcu_read(&as->current_map);
>      fr = flatview_lookup(view, range);
>      if (!fr) {
>          flatview_unref(view);

Following lines are:

           return ret;
       }

Which requires a rcu_read_unlock.

Fam

> @@ -1850,7 +1851,7 @@ MemoryRegionSection memory_region_find(MemoryRegion *mr,
>      ret.readonly = fr->readonly;
>      memory_region_ref(ret.mr);
>  
> -    flatview_unref(view);
> +    rcu_read_unlock();
>      return ret;
>  }
>  
> -- 
> 1.8.3.1
> 
>
Paolo Bonzini Jan. 26, 2015, 9:19 a.m. UTC | #2
On 26/01/2015 07:24, Fam Zheng wrote:
>> > Do the entire lookup under RCU, which avoids atomic operations.
> address_space_get_flatview() already is RCU protected, I don't see why this
> patch is necessary. Could you explain?

It's just an optimization; flatview_ref/flatview_unref do
atomic_inc/atomic_dec and this patch avoids that.

Paolo
diff mbox

Patch

diff --git a/memory.c b/memory.c
index a844ced..577e87c 100644
--- a/memory.c
+++ b/memory.c
@@ -1828,7 +1828,8 @@  MemoryRegionSection memory_region_find(MemoryRegion *mr,
     }
     range = addrrange_make(int128_make64(addr), int128_make64(size));
 
-    view = address_space_get_flatview(as);
+    rcu_read_lock();
+    view = atomic_rcu_read(&as->current_map);
     fr = flatview_lookup(view, range);
     if (!fr) {
         flatview_unref(view);
@@ -1850,7 +1851,7 @@  MemoryRegionSection memory_region_find(MemoryRegion *mr,
     ret.readonly = fr->readonly;
     memory_region_ref(ret.mr);
 
-    flatview_unref(view);
+    rcu_read_unlock();
     return ret;
 }