Logan audited the devm_memremap_pages() shutdown path and noticed that
it was possible to proceed to arch_remove_memory() before all
potential page references have been reaped.
Introduce a new ->cleanup() callback to do the work of waiting for any
straggling page references and then perform the percpu_ref_exit() in
devm_memremap_pages_release() context.
For p2pdma this involves some deeper reworks to reference count
resources on a per-instance basis rather than a per pci-device basis. A
modified genalloc api is introduced to convey a driver-private pointer
through gen_pool_{alloc,free}() interfaces. Also, a
devm_memunmap_pages() api is introduced since p2pdma does not
auto-release resources on a setup failure.
The dax and pmem changes pass the nvdimm unit tests, but the hmm and
p2pdma changes are compile-tested only.
Thoughts on the api changes?
I'm targeting to land this through Andrew's tree. 0day has chewed on
this for a day and reported no issues so far.
---
Dan Williams (6):
drivers/base/devres: Introduce devm_release_action()
mm/devm_memremap_pages: Introduce devm_memunmap_pages
pci/p2pdma: Fix the gen_pool_add_virt() failure path
lib/genalloc: Introduce chunk owners
pci/p2pdma: Track pgmap references per resource, not globally
mm/devm_memremap_pages: Fix final page put race
drivers/base/devres.c | 24 ++++++++
drivers/dax/device.c | 13 +----
drivers/nvdimm/pmem.c | 17 +++++-
drivers/pci/p2pdma.c | 105 +++++++++++++++++++++++--------------
include/linux/device.h | 1
include/linux/genalloc.h | 55 +++++++++++++++++--
include/linux/memremap.h | 8 +++
kernel/memremap.c | 23 ++++++--
lib/genalloc.c | 51 +++++++++---------
mm/hmm.c | 14 +----
tools/testing/nvdimm/test/iomap.c | 2 +
11 files changed, 209 insertions(+), 104 deletions(-)
