[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH v2 4/6] hostmem: hostmem listener pin RAM-Device
From: |
Stefan Hajnoczi |
Subject: |
Re: [Qemu-devel] [PATCH v2 4/6] hostmem: hostmem listener pin RAM-Device by refcnt |
Date: |
Fri, 3 May 2013 11:35:50 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
On Fri, May 03, 2013 at 10:45:20AM +0800, Liu Ping Fan wrote:
> From: Liu Ping Fan <address@hidden>
>
> With ref()/unref() interface of MemoryRegion, we can pin RAM-Device
> when using its memory, and release it when done.
>
> Signed-off-by: Liu Ping Fan <address@hidden>
> ---
> hw/virtio/dataplane/hostmem.c | 44 +++++++++++++++++++++++++++-----
> hw/virtio/dataplane/vring.c | 8 +++---
> include/hw/virtio/dataplane/hostmem.h | 4 ++-
> 3 files changed, 44 insertions(+), 12 deletions(-)
>
> diff --git a/hw/virtio/dataplane/hostmem.c b/hw/virtio/dataplane/hostmem.c
> index 1fd3e06..0e28dfc 100644
> --- a/hw/virtio/dataplane/hostmem.c
> +++ b/hw/virtio/dataplane/hostmem.c
> @@ -42,18 +42,28 @@ static void hostmem_ref(HostMem *hostmem)
>
> static void hostmem_unref(HostMem *hostmem)
> {
> - int t;
> + int i, t;
> + HostMemRegion *hmr;
>
> t = __sync_sub_and_fetch(&hostmem->ref, 1);
> assert(t >= 0);
> if (!t) {
> + for (i = 0; i < hostmem->num_current_regions; i++) {
> + hmr = &hostmem->current_regions[i];
> + /* Fix me, when memory hotunplug implement
> + * assert(hmr->mr_ops->unref)
> + */
> + if (hmr->mr->ops && hmr->mr->ops->unref) {
> + hmr->mr->ops->unref();
> + }
This patch should use memory_region_ref()/unref() which you introduced
in the previous patch instead of open-coding this.
> @@ -158,6 +176,18 @@ static void
> hostmem_listener_append_region(MemoryListener *listener,
> hostmem_append_new_region(as_mem->next_hostmem, section);
> }
>
> +static void hostmem_listener_append_region(MemoryListener *listener,
> + MemoryRegionSection *section)
> +{
> + hostmem_listener_nop_region(listener, section);
> + /* Fix me, when memory hotunplug implement
> + * assert(section->mr->ops->ref)
> + */
> + if (section->mr->ops && section->mr->ops->ref) {
> + section->mr->ops->ref();
> + }
> +}
Why does append increment the refcount while nop does not?
It seems that we always need to increment the memory region refcount
since we're building a completely new hostmem. The refcount ownership
is not passed from the current hostmen to the next hostmem, all memory
regions are released in hostmem_unref().
- [Qemu-devel] [PATCH v2 0/6] proposal to make hostmem listener RAM unplug safe, Liu Ping Fan, 2013/05/02
- [Qemu-devel] [PATCH v2 1/6] hostmem: make hostmem single, not per Vring related, Liu Ping Fan, 2013/05/02
- [Qemu-devel] [PATCH v2 2/6] hostmem: AddressSpace has its own map and maintained by RCU prepared style, Liu Ping Fan, 2013/05/02
- [Qemu-devel] [PATCH v2 3/6] memory: add ref/unref interface for MemroyRegionOps, Liu Ping Fan, 2013/05/02
- [Qemu-devel] [PATCH v2 4/6] hostmem: hostmem listener pin RAM-Device by refcnt, Liu Ping Fan, 2013/05/02
- Re: [Qemu-devel] [PATCH v2 4/6] hostmem: hostmem listener pin RAM-Device by refcnt,
Stefan Hajnoczi <=
- [Qemu-devel] [PATCH v2 5/6] Vring: use hostmem's RAM safe api, Liu Ping Fan, 2013/05/02
- [Qemu-devel] [PATCH v2 6/6] virtio-blk: release reference to RAM's memoryRegion, Liu Ping Fan, 2013/05/02
- Re: [Qemu-devel] [PATCH v2 0/6] proposal to make hostmem listener RAM unplug safe, Paolo Bonzini, 2013/05/04