Daniel Kiper <daniel.kiper@xxxxxxxxxx> writes:
[skip]
>> > > And we want to have it working out of the box.
>> > > So, I think that we should find proper solution. I suppose that we can schedule
>> > > a task here which auto online attached blocks. Hmmm... Not nice but should work.
>> > > Or maybe you have better idea how to fix this issue.
>> >
>> > I'd like to avoid additional delays and memory allocations between
>> > adding new memory and onlining it (and this is the main purpose of the
>> > patch). Maybe we can have a tristate online parameter ('online_now',
>> > 'online_delay', 'keep_offlined') and handle it
>> > accordingly. Alternatively I can suggest we have the onlining in Xen
>> > balloon driver code, memhp_auto_online is exported so we can call
>> > online_pages() after we release the ballon_mutex.
>>
>> This is not nice too. I prefer the same code path for every case.
>> Give me some time. I will think how to solve that issue.
>
> It looks that we can safely call mutex_unlock() just before add_memory_resource()
> call and retake lock immediately after add_memory_resource(). add_memory_resource()
> itself does not play with balloon stuff and even if online_pages() does then it
> take balloon_mutex in right place. Additionally, only one balloon task can run,
> so, I think that we are on safe side. Am I right?
I think you are as balloon_mutex is internal to xen driver and there is
only one balloon_process() running at the time. I just smoke-tested the
following:
commit 0fce4746a0090d533e9302cc42b3d3c0645d756d
Author: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
Date: Mon Jan 11 14:22:11 2016 +0100
xen_balloon: make hotplug auto online work
Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
index 890c3b5..08bbf35 100644
--- a/drivers/xen/balloon.c
+++ b/drivers/xen/balloon.c
@@ -338,7 +338,10 @@ static enum bp_state reserve_additional_memory(void)
}
#endif
- rc = add_memory_resource(nid, resource, false);
+ mutex_unlock(&balloon_mutex);
+ rc = add_memory_resource(nid, resource, memhp_auto_online);
+ mutex_lock(&balloon_mutex);
+
if (rc) {
pr_warn("Cannot add additional memory (%i)\n", rc);
goto err;
@@ -565,8 +568,10 @@ static void balloon_process(struct work_struct *work)
if (credit > 0) {
if (balloon_is_inflated())
state = increase_reservation(credit);
- else
+ else {
+ printk("balloon_process: adding memory (credit: %ld)!\n", credit);
state = reserve_additional_memory();
+ }
}
if (credit < 0)
And it seems to work (unrelated rant: 'xl mem-set' after 'xl max-mem'
doesn't work with "libxl: error: libxl.c:4809:libxl_set_memory_target:
memory_dynamic_max must be less than or equal to memory_static_max". At
the same time I'm able to increase the reservation with "echo
NEW_VALUE > /sys/devices/system/xen_memory/xen_memory0/target_kb" from
inside the guest. Was it supposed to be like that?).
While the patch misses the logic for empty pages (see David's comment in
a parallel thread) it should work for the general case the same way
auto-online works for Hyper-V and ACPI memory hotplugs.
--
Vitaly
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>