On 30.11.18 05:28, Wei Yang wrote: > On Thu, Nov 29, 2018 at 05:06:15PM +0100, David Hildenbrand wrote: >> On 29.11.18 16:53, Wei Yang wrote: >>> pgdat_resize_lock is used to protect pgdat's memory region information >>> like: node_start_pfn, node_present_pages, etc. While in function >>> sparse_add/remove_one_section(), pgdat_resize_lock is used to protect >>> initialization/release of one mem_section. This looks not proper. >>> >>> Based on current implementation, even remove this lock, mem_section >>> is still away from contention, because it is protected by global >>> mem_hotpulg_lock. >> >> s/mem_hotpulg_lock/mem_hotplug_lock/ >> >>> >>> Following is the current call trace of sparse_add/remove_one_section() >>> >>> mem_hotplug_begin() >>> arch_add_memory() >>> add_pages() >>> __add_pages() >>> __add_section() >>> sparse_add_one_section() >>> mem_hotplug_done() >>> >>> mem_hotplug_begin() >>> arch_remove_memory() >>> __remove_pages() >>> __remove_section() >>> sparse_remove_one_section() >>> mem_hotplug_done() >>> >>> The comment above the pgdat_resize_lock also mentions "Holding this will >>> also guarantee that any pfn_valid() stays that way.", which is true with >>> the current implementation and false after this patch. But current >>> implementation doesn't meet this comment. There isn't any pfn walkers >>> to take the lock so this looks like a relict from the past. This patch >>> also removes this comment. >> >> Should we start to document which lock is expected to protect what? >> >> I suggest adding what you just found out to >> Documentation/admin-guide/mm/memory-hotplug.rst "Locking Internals". >> Maybe a new subsection for mem_hotplug_lock. And eventually also >> pgdat_resize_lock. > > Well, I am not good at document writting. Below is my first trial. Look > forward your comments. > > BTW, in case I would send a new version with this, would I put this into > a separate one or merge this into current one? > > diff --git a/Documentation/admin-guide/mm/memory-hotplug.rst b/Documentation/admin-guide/mm/memory-hotplug.rst > index 5c4432c96c4b..1548820a0762 100644 > --- a/Documentation/admin-guide/mm/memory-hotplug.rst > +++ b/Documentation/admin-guide/mm/memory-hotplug.rst BTW, it really should go into Documentation/core-api/memory-hotplug.rst Something got wrong while merging this in linux-next, so now we have duplicate documentation and the one in Documentation/admin-guide/mm/memory-hotplug.rst about locking internals has to go. > @@ -396,6 +396,20 @@ Need more implementation yet.... > Locking Internals > ================= > > +There are three locks involved in memory-hotplug, two global lock and one local > +lock: > + > +- device_hotplug_lock > +- mem_hotplug_lock > +- device_lock > + > +Currently, they are twisted together for all kinds of reasons. The following > +part is divded 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: > > @@ -417,14 +431,21 @@ memory faster than expected: > As the device is visible to user space before taking the device_lock(), this > can result in a lock inversion. > > +mem_hotplug_lock > +--------------------- > + > 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 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) > > 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 > >> >> >> Thanks, >> >> David / dhildenb > -- Thanks, David / dhildenb
