Re: [PATCH RFC 2/3] mm/memory_hotplug: Create memory block devices after arch_add_memory()

[David Hildenbrand]

On 08.04.19 12:12, David Hildenbrand wrote:
> Only memory added via add_memory() and friends will need memory
> block devices - only memory to be used via the buddy and to be onlined/
> offlined by user space in memory block granularity.
> 
> Move creation of memory block devices out of arch_add_memory(). Create all
> devices after arch_add_memory() succeeded. We can later drop the
> want_memblock parameter, because it is now effectively stale.
> 
> Only after memory block devices have been added, memory can be onlined
> by user space. This implies, that memory is not visible to user space at
> all before arch_add_memory() succeeded.
> 
> Issue 1: __add_pages() does not remove pages in case something went
> wrong. If this is the case, we would now no longer create memory block
> devices for such "partially added memory". So the memory would not be
> usable/onlinable. Bad? Or related to issue 2 (e.g. fix __add_pages()
> to remove any parts that were added in case of an error). Functions that
> fail and don't clean up are not that nice.
> 
> Issue 2: In case we can't add memory block devices, and we don't have
> HOTREMOVE, we can't remove the pages via arch_remove_pages. Maybe we should
> try to get rid of CONFIG_MEMORY_HOTREMOVE, so we can handle all failures
> in a nice way? Or at least allow arch_remove_pages() and friends, so a
> subset of CONFIG_MEMORY_HOTREMOVE.
> 
> Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
> ---
>  drivers/base/memory.c  | 67 +++++++++++++++++++++++++-----------------
>  include/linux/memory.h |  2 +-
>  mm/memory_hotplug.c    | 17 +++++++----
>  3 files changed, 53 insertions(+), 33 deletions(-)
> 
> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
> index d9ebb89816f7..847b33061e2e 100644
> --- a/drivers/base/memory.c
> +++ b/drivers/base/memory.c
> @@ -701,44 +701,57 @@ static int add_memory_block(int base_section_nr)
>  	return 0;
>  }
>  
> -/*
> - * need an interface for the VM to add new memory regions,
> - * but without onlining it.
> - */
> -int hotplug_memory_register(int nid, struct mem_section *section)
> +static void unregister_memory(struct memory_block *memory)
>  {
> -	int ret = 0;
> +	BUG_ON(memory->dev.bus != &memory_subsys);
> +
> +	/* drop the ref. we got via find_memory_block() */
> +	put_device(&memory->dev);
> +	device_unregister(&memory->dev);
> +}
> +
> +int hotplug_memory_register(unsigned long start, unsigned long size)
> +{
> +	unsigned long block_nr_pages = memory_block_size_bytes() >> PAGE_SHIFT;
> +	unsigned long start_pfn = PFN_DOWN(start);
> +	unsigned long end_pfn = start_pfn + (size >> PAGE_SHIFT);
> +	unsigned long pfn;
>  	struct memory_block *mem;
> +	int ret = 0;
>  
> -	mutex_lock(&mem_sysfs_mutex);
> +	BUG_ON(!IS_ALIGNED(start, memory_block_size_bytes()));
> +	BUG_ON(!IS_ALIGNED(size, memory_block_size_bytes()));
>  
> -	mem = find_memory_block(section);
> -	if (mem) {
> -		mem->section_count++;
> -		put_device(&mem->dev);
> -	} else {
> -		ret = init_memory_block(&mem, section, MEM_OFFLINE);
> +	mutex_lock(&mem_sysfs_mutex);
> +	for (pfn = start_pfn; pfn != end_pfn; pfn += block_nr_pages) {
> +		mem = find_memory_block(__pfn_to_section(pfn));
> +		if (mem) {
> +			WARN_ON_ONCE(false);
> +			put_device(&mem->dev);
> +			continue;
> +		}
> +		ret = init_memory_block(&mem, __pfn_to_section(pfn),
> +					MEM_OFFLINE);
>  		if (ret)
> -			goto out;
> -		mem->section_count++;
> +			break;
> +		mem->section_count = memory_block_size_bytes() /
> +				     MIN_MEMORY_BLOCK_SIZE;
> +	}
> +	if (ret) {
> +		end_pfn = pfn;
> +		for (pfn = start_pfn; pfn != end_pfn; pfn += block_nr_pages) {
> +			mem = find_memory_block(__pfn_to_section(pfn));
> +			if (!mem)
> +				continue;
> +			mem->section_count = 0;
> +			unregister_memory(mem);
> +		}
>  	}
> -
> -out:
>  	mutex_unlock(&mem_sysfs_mutex);
>  	return ret;
>  }
>  
>  #ifdef CONFIG_MEMORY_HOTREMOVE
> -static void
> -unregister_memory(struct memory_block *memory)
> -{
> -	BUG_ON(memory->dev.bus != &memory_subsys);
> -
> -	/* drop the ref. we got in remove_memory_section() */
> -	put_device(&memory->dev);
> -	device_unregister(&memory->dev);
> -}
> -
>  static int remove_memory_section(struct mem_section *section)
>  {
>  	struct memory_block *mem;
> diff --git a/include/linux/memory.h b/include/linux/memory.h
> index a6ddefc60517..e275dc775834 100644
> --- a/include/linux/memory.h
> +++ b/include/linux/memory.h
> @@ -111,7 +111,7 @@ extern int register_memory_notifier(struct notifier_block *nb);
>  extern void unregister_memory_notifier(struct notifier_block *nb);
>  extern int register_memory_isolate_notifier(struct notifier_block *nb);
>  extern void unregister_memory_isolate_notifier(struct notifier_block *nb);
> -int hotplug_memory_register(int nid, struct mem_section *section);
> +int hotplug_memory_register(unsigned long start, unsigned long size);
>  #ifdef CONFIG_MEMORY_HOTREMOVE
>  extern int unregister_memory_section(struct mem_section *);
>  #endif
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index 680dcc67f9d5..13ee0a26e034 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -260,11 +260,7 @@ static int __meminit __add_section(int nid, unsigned long phys_start_pfn,
>  	ret = sparse_add_one_section(nid, phys_start_pfn, altmap);
>  	if (ret < 0)
>  		return ret;
> -
> -	if (!want_memblock)
> -		return 0;
> -
> -	return hotplug_memory_register(nid, __pfn_to_section(phys_start_pfn));
> +	return 0;
>  }
>  
>  /*
> @@ -1125,6 +1121,17 @@ int __ref add_memory_resource(int nid, struct resource *res)
>  	if (ret < 0)
>  		goto error;
>  
> +	/* create memory block devices after memory was added */
> +	ret = hotplug_memory_register(start, size);
> +#ifdef CONFIG_MEMORY_HOTREMOVE
> +	if (ret) {
> +		arch_remove_memory(nid, start, size, NULL);
> +		goto error;
> +	}
> +#else
> +	WARN_ON(ret);
> +#endif
> +
>  	if (new_node) {
>  		/* If sysfs file of new node can't be created, cpu on the node
>  		 * can't be hot-added. There is no rollback way now.
> 

FWIW, I think we should first try to make sure arch_remove_memory()
cannot fail / will not ignore errors if possible. There are still some
things in there that need more re-factoring first.

-- 

Thanks,

David / dhildenb