This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


[Xen-devel] Re: [PATCH] GSoC 2010 - Memory hotplug support for Xen guest

To: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
Subject: [Xen-devel] Re: [PATCH] GSoC 2010 - Memory hotplug support for Xen guests - second fully working version - once again
From: Jeremy Fitzhardinge <jeremy@xxxxxxxx>
Date: Fri, 06 Aug 2010 12:20:07 -0700
Cc: xen-devel@xxxxxxxxxxxxxxxxxxx, stefano.stabellini@xxxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx, v.tolstov@xxxxxxxxx, linux-mm@xxxxxxxxx, Daniel Kiper <dkiper@xxxxxxxxxxxx>
Delivery-date: Fri, 06 Aug 2010 12:21:05 -0700
Envelope-to: www-data@xxxxxxxxxxxxxxxxxxx
In-reply-to: <20100806163408.GA8678@xxxxxxxxxxxxxxxxxxx>
List-help: <mailto:xen-devel-request@lists.xensource.com?subject=help>
List-id: Xen developer discussion <xen-devel.lists.xensource.com>
List-post: <mailto:xen-devel@lists.xensource.com>
List-subscribe: <http://lists.xensource.com/mailman/listinfo/xen-devel>, <mailto:xen-devel-request@lists.xensource.com?subject=subscribe>
List-unsubscribe: <http://lists.xensource.com/mailman/listinfo/xen-devel>, <mailto:xen-devel-request@lists.xensource.com?subject=unsubscribe>
References: <20100806111147.GA31683@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx> <20100806163408.GA8678@xxxxxxxxxxxxxxxxxxx>
Sender: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv: Gecko/20100720 Fedora/3.1.1-1.fc13 Lightning/1.0b2pre Thunderbird/3.1.1
 On 08/06/2010 09:34 AM, Konrad Rzeszutek Wilk wrote:
On Fri, Aug 06, 2010 at 01:11:47PM +0200, Daniel Kiper wrote:

I am sending this e-mail once again because it probably
has been lost in abyss of Xen-devel/LKLM list.

Here is the second version of memory hotplug support
for Xen guests patch. This one cleanly applies to
repository, xen/memory-hotplug head.

   - /sys/devices/system/memory/probe interface has been removed;
     are much better (I forgot about them),
   - most of the code have been moved to drivers/xen/balloon.c,
   - this changes forced me to export hotadd_new_pgdat and
     rollback_node_hotadd function from mm/memory_hotplug.c;
     could it be accepted by mm/memory_hotplug.c maintainers ???
   - PV on HVM mode is supported now; it was tested on
     repository, 2.6.34-pvhvm head,
   - most of Jeremy suggestions have been applied.

On Wed, Jul 28, 2010 at 11:36:29AM +0400, Vasiliy G Tolstov wrote:
Work's fine with opensuse 11.3 (dom0 and domU)

On Thu, Jul 29, 2010 at 12:39:52PM -0700, Jeremy Fitzhardinge wrote:
  On 07/26/2010 05:41 PM, Daniel Kiper wrote:

Currently there is fully working version.
It has been tested on Xen Ver. 4.0.0 in PV
guest i386/x86_64 with Linux kernel Ver.
and Ver. This patch cleanly applys
to Ver.
Thanks.  I've pushed this into xen.git as xen/memory-hotplug so people
can play with it more easily (but I haven't merged it into any of the
other branches yet).

+static inline unsigned long current_target(void)
+       return balloon_stats.target_pages;
Why does this need its own version?
Because original version return values not bigger
then initial memory allocation which does not allow
memory hotplug to function.

+int __ref xen_add_memory(int nid, u64 start, u64 size)
+       pg_data_t *pgdat = NULL;
+       int new_pgdat = 0, ret;
+       lock_system_sleep();
+       if (!node_online(nid)) {
+               pgdat = hotadd_new_pgdat(nid, start);
+               ret = -ENOMEM;
+               if (!pgdat)
+                       goto out;
+               new_pgdat = 1;
+       }
+       /* call arch's memory hotadd */
+       ret = arch_add_memory(nid, start, size);
+       if (ret<   0)
+               goto error;
+       /* we online node here. we can't roll back from here. */
+       node_set_online(nid);
+       if (new_pgdat) {
+               ret = register_one_node(nid);
+               /*
+                * If sysfs file of new node can't create, cpu on the node
+                * can't be hot-added. There is no rollback way now.
+                * So, check by BUG_ON() to catch it reluctantly..
+                */
+               BUG_ON(ret);
+       }
This doesn't seem to be doing anything particularly xen-specific.
In general it could be generic however I do not know
it will be useful for others. If this function would
be accepted by mm/memory_hotplug.c maintainers we could
move it there. I removed from original add_memory funtion
resource allocation (and deallocation after error), which
must be done before XENMEM_populate_physmap in Xen. xen_add_memory
is called after physmap is fully populated.

If you have a questions please drop me a line.

Can you repost a patch that is on top of a virgin tree please?

I just thanked him for posting a delta ;) I've pushed this into xen/memory-hotplug so you can easily generate a complete diff with "git diff v2.6.34..xen/memory-hotplug".

+ spin_lock_irqsave(&balloon_lock, flags);
+       if (!is_memory_resource_reserved()) {
+               /*
+                * Look for first unused memory region starting at page
+                * boundary. Skip last memory section created at boot time
+                * becuase it may contains unused memory pages with PG_reserved
+                * bit not set (online_pages require PG_reserved bit set).
+                */
+               r = kzalloc(sizeof(struct resource), GFP_KERNEL);

You are holding a spinlock here. Kzalloc can sleep

Good point.


Xen-devel mailing list