On 06.12.18 01:26, Wei Yang wrote:
> Currently locking for memory hotplug is a little complicated.
>
> Generally speaking, we leverage the two global lock:
>
> * device_hotplug_lock
> * mem_hotplug_lock
>
> to serialise the process.
>
> While for the long term, we are willing to have more fine-grained lock
> to provide higher scalability.
>
> This patch divides Locking Internal section based on these two global
> locks to help readers to understand it. Also it adds some new finding to
> enrich it.
>
> [David: words arrangement]
>
> Signed-off-by: Wei Yang
> ---
> v2: adjustment based on David and Mike comment
> ---
> Documentation/core-api/memory-hotplug.rst | 27 ---
> 1 file changed, 24 insertions(+), 3 deletions(-)
>
> diff --git a/Documentation/core-api/memory-hotplug.rst
> b/Documentation/core-api/memory-hotplug.rst
> index de7467e48067..51d477ad4b80 100644
> --- a/Documentation/core-api/memory-hotplug.rst
> +++ b/Documentation/core-api/memory-hotplug.rst
> @@ -89,6 +89,20 @@ NOTIFY_STOP stops further processing of the notification
> queue.
> Locking Internals
> =
>
> +In addition to fine grained locks like pgdat_resize_lock, there are three
> locks
> +involved
> +
> +- device_hotplug_lock
> +- mem_hotplug_lock
> +- device_lock
> +
> +Currently, they are twisted together for all kinds of reasons. The following
> +part is divided into device_hotplug_lock and mem_hotplug_lock parts
> +respectively to describe those tricky situations.
> +
> +device_hotplug_lock
> +-
> +
> When adding/removing memory that uses memory block devices (i.e. ordinary
> RAM),
> the device_hotplug_lock should be held to:
>
> @@ -111,13 +125,20 @@ As the device is visible to user space before taking
> the device_lock(), this
> can result in a lock inversion.
>
> onlining/offlining of memory should be done via device_online()/
> -device_offline() - to make sure it is properly synchronized to actions
> -via sysfs. Holding device_hotplug_lock is advised (to e.g. protect
> online_type)
> +device_offline() - to make sure it is properly synchronized to actions via
> +sysfs. Even if mem_hotplug_lock is used to protect the process, because of
> the
> +lock inversion described above, holding device_hotplug_lock is still advised
> +(to e.g. protect online_type)
> +
> +mem_hotplug_lock
> +-
>
> When adding/removing/onlining/offlining memory or adding/removing
> heterogeneous/device memory, we should always hold the mem_hotplug_lock in
> write mode to serialise memory hotplug (e.g. access to global/zone
> -variables).
> +variables). Currently, we take advantage of this to serialise sparsemem's
> +mem_section handling in sparse_add_one_section() and
> +sparse_remove_one_section().
>
> In addition, mem_hotplug_lock (in contrast to device_hotplug_lock) in read
> mode allows for a quite efficient get_online_mems/put_online_mems
>
Reviewed-by: David Hildenbrand
--
Thanks,
David / dhildenb