* Replace the /sys/class/dax device model with /sys/bus/dax, and include
a compat driver so distributions can opt-in to the new ABI.
* Allow for an alternative driver for the device-dax address-range
* Introduce the 'kmem' driver to hotplug / assign a device-dax
address-range to the core-mm.
* Arrange for the device-dax target-node to be onlined so that the newly
added memory range can be uniquely referenced by numa apis.
-----BEGIN PGP SIGNATURE-----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=hfa3
-----END PGP SIGNATURE-----
Merge tag 'devdax-for-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm
Pull device-dax updates from Dan Williams:
"New device-dax infrastructure to allow persistent memory and other
"reserved" / performance differentiated memories, to be assigned to
the core-mm as "System RAM".
Some users want to use persistent memory as additional volatile
memory. They are willing to cope with potential performance
differences, for example between DRAM and 3D Xpoint, and want to use
typical Linux memory management apis rather than a userspace memory
allocator layered over an mmap() of a dax file. The administration
model is to decide how much Persistent Memory (pmem) to use as System
RAM, create a device-dax-mode namespace of that size, and then assign
it to the core-mm. The rationale for device-dax is that it is a
generic memory-mapping driver that can be layered over any "special
purpose" memory, not just pmem. On subsequent boots udev rules can be
used to restore the memory assignment.
One implication of using pmem as RAM is that mlock() no longer keeps
data off persistent media. For this reason it is recommended to enable
NVDIMM Security (previously merged for 5.0) to encrypt pmem contents
at rest. We considered making this recommendation an actively enforced
requirement, but in the end decided to leave it as a distribution /
administrator policy to allow for emulation and test environments that
lack security capable NVDIMMs.
Summary:
- Replace the /sys/class/dax device model with /sys/bus/dax, and
include a compat driver so distributions can opt-in to the new ABI.
- Allow for an alternative driver for the device-dax address-range
- Introduce the 'kmem' driver to hotplug / assign a device-dax
address-range to the core-mm.
- Arrange for the device-dax target-node to be onlined so that the
newly added memory range can be uniquely referenced by numa apis"
NOTE! I'm not entirely happy with the whole "PMEM as RAM" model because
we currently have special - and very annoying rules in the kernel about
accessing PMEM only with the "MC safe" accessors, because machine checks
inside the regular repeat string copy functions can be fatal in some
(not described) circumstances.
And apparently the PMEM modules can cause that a lot more than regular
RAM. The argument is that this happens because PMEM doesn't necessarily
get scrubbed at boot like RAM does, but that is planned to be added for
the user space tooling.
Quoting Dan from another email:
"The exposure can be reduced in the volatile-RAM case by scanning for
and clearing errors before it is onlined as RAM. The userspace tooling
for that can be in place before v5.1-final. There's also runtime
notifications of errors via acpi_nfit_uc_error_notify() from
background scrubbers on the DIMM devices. With that mechanism the
kernel could proactively clear newly discovered poison in the volatile
case, but that would be additional development more suitable for v5.2.
I understand the concern, and the need to highlight this issue by
tapping the brakes on feature development, but I don't see PMEM as RAM
making the situation worse when the exposure is also there via DAX in
the PMEM case. Volatile-RAM is arguably a safer use case since it's
possible to repair pages where the persistent case needs active
application coordination"
* tag 'devdax-for-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
device-dax: "Hotplug" persistent memory for use like normal RAM
mm/resource: Let walk_system_ram_range() search child resources
mm/memory-hotplug: Allow memory resources to be children
mm/resource: Move HMM pr_debug() deeper into resource code
mm/resource: Return real error codes from walk failures
device-dax: Add a 'modalias' attribute to DAX 'bus' devices
device-dax: Add a 'target_node' attribute
device-dax: Auto-bind device after successful new_id
acpi/nfit, device-dax: Identify differentiated memory with a unique numa-node
device-dax: Add /sys/class/dax backwards compatibility
device-dax: Add support for a dax override driver
device-dax: Move resource pinning+mapping into the common driver
device-dax: Introduce bus + driver model
device-dax: Start defining a dax bus model
device-dax: Remove multi-resource infrastructure
device-dax: Kill dax_region base
device-dax: Kill dax_region ida
When binding an SCM volume to a physical address the hypervisor has the
option to return early with a continue token with the expectation that
the guest will resume the bind operation until it completes. A quirk of
this interface is that the bind address will only be returned by the
first bind h-call and the subsequent calls will return
0xFFFF_FFFF_FFFF_FFFF for the bind address.
We currently do not save the address returned by the first h-call. As a
result we will use the junk address as the base of the bound region if
the hypervisor decides to split the bind across multiple h-calls. This
bug was found when testing with very large SCM volumes where the bind
process would take more time than they hypervisor's internal h-call time
limit would allow. This patch fixes the issue by saving the bind address
from the first call.
Cc: stable@vger.kernel.org
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Persistent memory, as described by the ACPI NFIT (NVDIMM Firmware
Interface Table), is the first known instance of a memory range
described by a unique "target" proximity domain. Where "initiator" and
"target" proximity domains is an approach that the ACPI HMAT
(Heterogeneous Memory Attributes Table) uses to described the unique
performance properties of a memory range relative to a given initiator
(e.g. CPU or DMA device).
Currently the numa-node for a /dev/pmemX block-device or /dev/daxX.Y
char-device follows the traditional notion of 'numa-node' where the
attribute conveys the closest online numa-node. That numa-node attribute
is useful for cpu-binding and memory-binding processes *near* the
device. However, when the memory range backing a 'pmem', or 'dax' device
is onlined (memory hot-add) the memory-only-numa-node representing that
address needs to be differentiated from the set of online nodes. In
other words, the numa-node association of the device depends on whether
you can bind processes *near* the cpu-numa-node in the offline
device-case, or bind process *on* the memory-range directly after the
backing address range is onlined.
Allow for the case that platform firmware describes persistent memory
with a unique proximity domain, i.e. when it is distinct from the
proximity of DRAM and CPUs that are on the same socket. Plumb the Linux
numa-node translation of that proximity through the libnvdimm region
device to namespaces that are in device-dax mode. With this in place the
proposed kmem driver [1] can optionally discover a unique numa-node
number for the address range as it transitions the memory from an
offline state managed by a device-driver to an online memory range
managed by the core-mm.
[1]: https://lore.kernel.org/lkml/20181022201317.8558C1D8@viggo.jf.intel.com
Reported-by: Fan Du <fan.du@intel.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: "Oliver O'Halloran" <oohall@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Reviewed-by: Yang Shi <yang.shi@linux.alibaba.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The interleave set cookie is used to determine if a label stored in the
metadata space should be applied to the current region. This is
important in the case of NVDIMMs since the firmware may change the
interleaving configuration of a DIMM which would invalidate the existing
labels. In our case the hypervisor hides those details from us so we
don't really care, but libnvdimm still requires the interleave set
cookie to be non-zero.
For our purposes we just need the set cookie to be unique and fixed for
a given PAPR SCM region and using the unit-guid (really a UUID) is fine
for this purpose.
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
[mpe: Use kernel types (u64)]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
When a new nvdimm device is registered with libnvdimm via
nvdimm_create() it is added as a device on the nvdimm bus. The probe
function for the DIMM driver is potentially quite slow so actually
registering and probing the device is done in an async domain rather
than immediately after device creation. This can result in a race where
the region device (created 2nd) is probed first and fails to activate at
boot.
To fix this we use the same approach as the ACPI/NFIT driver which is to
check that all the DIMM devices registered successfully. LibNVDIMM
provides the nvdimm_bus_count_dimms() function which synchronises with
the async domain and verifies that the dimm was successfully registered
with the bus.
If either of these does not occur then we bail.
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The return values of a h-call are returned in the CPU registers and
written to the provided buffer by the plpar_hcall() wrapper. As a result
the values written to memory are always in the native endian and should
not be byte swapped.
The inital implementation of the H-Call interface was done in qemu and
the returned values were byte swapped unnecessarily in both the
hypervisor and in the driver so this was only noticed when bringing up
the PowerVM implementation.
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The ibm,unit-sizes property was originally specified as an array of two
u32s corresponding to the memory block size, and the number of blocks
available in that region. A fairly last-minute change to the SCM DT
specification was splitting that into two seperate u64 properties:
ibm,block-sizes and ibm,number-of-blocks that convey the same
information. No firmware / hypervisor that emitted the ibm,unit-size
property ever appeared in the wild.
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
[mpe: Use kernel types (u32/u64)]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Fix an off-by-one error in the memory resource range. This resource is
used to determine the address range of the memory to be hot-plugged as
ZONE_DEVICE memory. The current end address results in the kernel
attempting to map an additional memblock and the hypervisor may reject
the mapping resulting in the entire hot-plug failing.
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Adds a driver that implements support for enabling and accessing PAPR
SCM regions. Unfortunately due to how the PAPR interface works we can't
use the existing of_pmem driver (yet) because:
a) The guest is required to use the H_SCM_BIND_MEM h-call to add
add the SCM region to it's physical address space, and
b) There is currently no mechanism for relating a bare of_pmem region
to the backing DIMM (or not-a-DIMM for our case).
Both of these are easily handled by rolling the functionality into a
seperate driver so here we are...
Acked-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
