Re: [PATCH v4 1/2] memory-hotplug: add automatic onlining policy for the newly added memory

[Vitaly Kuznetsov]

David Rientjes <rientjes@xxxxxxxxxx> writes:

> On Wed, 13 Jan 2016, Vitaly Kuznetsov wrote:
>
>> >> diff --git a/Documentation/memory-hotplug.txt b/Documentation/memory-hotplug.txt
>> >> index ce2cfcf..ceaf40c 100644
>> >> --- a/Documentation/memory-hotplug.txt
>> >> +++ b/Documentation/memory-hotplug.txt
>> >> @@ -254,12 +254,23 @@ If the memory block is online, you'll read "online".
>> >>  If the memory block is offline, you'll read "offline".
>> >>  
>> >>  
>> >> -5.2. How to online memory
>> >> +5.2. Memory onlining
>> >
>> > Idk why you're changing this title since you didn't change it in the table 
>> > of contents and it already pairs with "6.2. How to offline memory".
>> >
>> > This makes it seem like you're covering all memory onlining operations in 
>> > the kernel (including xen onlining) rather than just memory onlined by 
>> > root.  It doesn't cover the fact that xen onlining can be done without 
>> > automatic onlining, so I would leave this section's title as it is and 
>> > only cover aspects of memory onlining that users are triggering 
>> > themselves.
>> 
>> Ok, I changed the title to reflect the fact that a special action to
>> online memory is not always required any more but as the global policy
>> stays 'offline' by default for now let's keep the original title.
>> 
>
> Thanks.
>
>> >> +	/* online pages if requested */
>> >> +	if (online)
>> >> +		online_pages(start >> PAGE_SHIFT, size >> PAGE_SHIFT,
>> >> +			     MMOP_ONLINE_KEEP);
>> >> +
>> >>  	goto out;
>> >>  
>> >>  error:
>> >
>> > Well, shucks, what happens if online_pages() fails, such as if a memory 
>> > hot-add notifier returns an errno for MEMORY_GOING_ONLINE?  The memory was 
>> > added but not subsequently onlined, although auto onlining was set, so how 
>> > does userspace know the state it is in?
>> 
>> Bad ... we could have checked the return value but I don't see a proper
>> way to handling it here: if we managed to online some blocks we can't
>> revert back. We'll probably have to online pages block-by-block (e.g. by
>> utilizing memory_block_change_state()) handling possible failures.
>> 
>
> My suggestion is to just simply document that auto-onlining can add the 
> memory but fail to online it and the failure is silent to userspace.  If 
> userspace cares, it can check the online status of the added memory blocks 
> itself.

The problem is not only that it's silent, but also that
/sys/devices/system/memory/*/state will lie as we create all memory
blocks in MEM_ONLINE state and from online_pages() error we can't figure
out which particular block failed. 'v5' which I sent yesterday is
supposed to fix the issue (blocks are onlined with
memory_block_change_state() which handles failures.

-- 
  Vitaly