The core of devm_request_free_mem_region() is a helper that searches for
free space in iomem_resource and performs __request_region_locked() on
the result of that search. The policy choices of the implementation
conform to what CONFIG_DEVICE_PRIVATE users want which is memory that is
immediately marked busy, and a preference to search for the first-fit
free range in descending order from the top of the physical address
space.
CXL has a need for a similar allocator, but with the following tweaks:
1/ Search for free space in ascending order
2/ Search for free space relative to a given CXL window
3/ 'insert' rather than 'request' the new resource given downstream
drivers from the CXL Region driver (like the pmem or dax drivers) are
responsible for request_mem_region() when they activate the memory
range.
Rework __request_free_mem_region() into get_free_mem_region() which
takes a set of GFR_* (Get Free Region) flags to control the allocation
policy (ascending vs descending), and "busy" policy (insert_resource()
vs request_region()).
As part of the consolidation of the legacy GFR_REQUEST_REGION case with
the new default of just inserting a new resource into the free space
some minor cleanups like not checking for NULL before calling
devres_free() (which does its own check) is included.
Suggested-by: Jason Gunthorpe <[email protected]>
Link: https://lore.kernel.org/linux-cxl/[email protected]/
Cc: Matthew Wilcox <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Reviewed-by: Jonathan Cameron <[email protected]>
Link: https://lore.kernel.org/r/165784333333.1758207.13703329337805274043.stgit@dwillia2-xfh.jf.intel.com
Signed-off-by: Dan Williams <[email protected]>
Recall that CXL capable address ranges, on ACPI platforms, are published
in the CEDT.CFMWS (CXL Early Discovery Table: CXL Fixed Memory Window
Structures). These windows represent both the actively mapped capacity
and the potential address space that can be dynamically assigned to a
new CXL decode configuration (region / interleave-set).
CXL endpoints like DDR DIMMs can be mapped at any physical address
including 0 and legacy ranges.
There is an expectation and requirement that the /proc/iomem interface
and the iomem_resource tree in the kernel reflect the full set of
platform address ranges. I.e. that every address range that platform
firmware and bus drivers enumerate be reflected as an iomem_resource
entry. The hard requirement to do this for CXL arises from the fact that
facilities like CONFIG_DEVICE_PRIVATE expect to be able to treat empty
iomem_resource ranges as free for software to use as proxy address
space. Without CXL publishing its potential address ranges in
iomem_resource, the CONFIG_DEVICE_PRIVATE mechanism may inadvertently
steal capacity reserved for runtime provisioning of new CXL regions.
So, iomem_resource needs to know about both active and potential CXL
resource ranges. The active CXL resources might already be reflected in
iomem_resource as "System RAM". insert_resource_expand_to_fit() handles
re-parenting "System RAM" underneath a CXL window.
The "_expand_to_fit()" behavior handles cases where a CXL window is not
a strict superset of an existing entry in the iomem_resource tree. The
"_expand_to_fit()" behavior is acceptable from the perspective of
resource allocation. The expansion happens because a conflicting
resource range is already populated, which means the resource boundary
expansion does not result in any additional free CXL address space being
made available. CXL address space allocation is always bounded by the
orginal unexpanded address range.
However, the potential for expansion does mean that something like
walk_iomem_res_desc(IORES_DESC_CXL...) can only return fuzzy answers on
corner case platforms that cause the resource tree to expand a CXL
window resource over a range that is not decoded by CXL. This would be
an odd platform configuration, but if it becomes a problem in practice
the CXL subsytem could just publish an API that returns definitive
answers.
Cc: Andrew Morton <[email protected]>
Cc: David Hildenbrand <[email protected]>
Cc: Jason Gunthorpe <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Reviewed-by: Jonathan Cameron <[email protected]>
Acked-by: Greg Kroah-Hartman <[email protected]>
Link: https://lore.kernel.org/r/165784325943.1758207.5310344844375305118.stgit@dwillia2-xfh.jf.intel.com
Signed-off-by: Dan Williams <[email protected]>
Since commit ebff7d8f27 ("mem hotunplug: fix kfree() of bootmem
memory"), we could get a resource allocated during boot via
alloc_resource(). And it's required to release the resource using
free_resource(). Howerver, many people use kfree directly which will
result in kernel BUG. In order to fix this without fixing every call
site, just leak a couple of bytes in such corner case.
Link: https://lkml.kernel.org/r/[email protected]
Fixes: ebff7d8f27 ("mem hotunplug: fix kfree() of bootmem memory")
Signed-off-by: Miaohe Lin <[email protected]>
Suggested-by: David Hildenbrand <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Alistair Popple <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
virtio-mem dynamically exposes memory inside a device memory region as
system RAM to Linux, coordinating with the hypervisor which parts are
actually "plugged" and consequently usable/accessible.
On the one hand, the virtio-mem driver adds/removes whole memory blocks,
creating/removing busy IORESOURCE_SYSTEM_RAM resources, on the other
hand, it logically (un)plugs memory inside added memory blocks,
dynamically either exposing them to the buddy or hiding them from the
buddy and marking them PG_offline.
In contrast to physical devices, like a DIMM, the virtio-mem driver is
required to actually make use of any of the device-provided memory,
because it performs the handshake with the hypervisor. virtio-mem
memory cannot simply be access via /dev/mem without a driver.
There is no safe way to:
a) Access plugged memory blocks via /dev/mem, as they might contain
unplugged holes or might get silently unplugged by the virtio-mem
driver and consequently turned inaccessible.
b) Access unplugged memory blocks via /dev/mem because the virtio-mem
driver is required to make them actually accessible first.
The virtio-spec states that unplugged memory blocks MUST NOT be written,
and only selected unplugged memory blocks MAY be read. We want to make
sure, this is the case in sane environments -- where the virtio-mem driver
was loaded.
We want to make sure that in a sane environment, nobody "accidentially"
accesses unplugged memory inside the device managed region. For example,
a user might spot a memory region in /proc/iomem and try accessing it via
/dev/mem via gdb or dumping it via something else. By the time the mmap()
happens, the memory might already have been removed by the virtio-mem
driver silently: the mmap() would succeeed and user space might
accidentially access unplugged memory.
So once the driver was loaded and detected the device along the
device-managed region, we just want to disallow any access via /dev/mem to
it.
In an ideal world, we would mark the whole region as busy ("owned by a
driver") and exclude it; however, that would be wrong, as we don't really
have actual system RAM at these ranges added to Linux ("busy system RAM").
Instead, we want to mark such ranges as "not actual busy system RAM but
still soft-reserved and prepared by a driver for future use."
Let's teach iomem_is_exclusive() to reject access to any range with
"IORESOURCE_SYSTEM_RAM | IORESOURCE_EXCLUSIVE", even if not busy and even
if "iomem=relaxed" is set. Introduce EXCLUSIVE_SYSTEM_RAM to make it
easier for applicable drivers to depend on this setting in their Kconfig.
For now, there are no applicable ranges and we'll modify virtio-mem next
to properly set IORESOURCE_EXCLUSIVE on the parent resource container it
creates to contain all actual busy system RAM added via
add_memory_driver_managed().
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: David Hildenbrand <[email protected]>
Reviewed-by: Dan Williams <[email protected]>
Cc: Andy Shevchenko <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Hanjun Guo <[email protected]>
Cc: Jason Wang <[email protected]>
Cc: "Michael S. Tsirkin" <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Patch series "virtio-mem: disallow mapping virtio-mem memory via /dev/mem", v5.
Let's add the basic infrastructure to exclude some physical memory regions
marked as "IORESOURCE_SYSTEM_RAM" completely from /dev/mem access, even
though they are not marked IORESOURCE_BUSY and even though "iomem=relaxed"
is set. Resource IORESOURCE_EXCLUSIVE for that purpose instead of adding
new flags to express something similar to "soft-busy" or "not busy yet,
but already prepared by a driver and not to be mapped by user space".
Use it for virtio-mem, to disallow mapping any virtio-mem memory via
/dev/mem to user space after the virtio-mem driver was loaded.
This patch (of 3):
We end up traversing subtrees of ranges we are not interested in; let's
optimize this case, skipping such subtrees, cleaning up the function a
bit.
For example, in the following configuration (/proc/iomem):
00000000-00000fff : Reserved
00001000-00057fff : System RAM
00058000-00058fff : Reserved
00059000-0009cfff : System RAM
0009d000-000fffff : Reserved
000a0000-000bffff : PCI Bus 0000:00
000c0000-000c3fff : PCI Bus 0000:00
000c4000-000c7fff : PCI Bus 0000:00
000c8000-000cbfff : PCI Bus 0000:00
000cc000-000cffff : PCI Bus 0000:00
000d0000-000d3fff : PCI Bus 0000:00
000d4000-000d7fff : PCI Bus 0000:00
000d8000-000dbfff : PCI Bus 0000:00
000dc000-000dffff : PCI Bus 0000:00
000e0000-000e3fff : PCI Bus 0000:00
000e4000-000e7fff : PCI Bus 0000:00
000e8000-000ebfff : PCI Bus 0000:00
000ec000-000effff : PCI Bus 0000:00
000f0000-000fffff : PCI Bus 0000:00
000f0000-000fffff : System ROM
00100000-3fffffff : System RAM
40000000-403fffff : Reserved
40000000-403fffff : pnp 00:00
40400000-80a79fff : System RAM
...
We don't have to look at any children of "0009d000-000fffff : Reserved"
if we can just skip these 15 items directly because the parent range is
not of interest.
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: David Hildenbrand <[email protected]>
Reviewed-by: Dan Williams <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: "Michael S. Tsirkin" <[email protected]>
Cc: Jason Wang <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Cc: Hanjun Guo <[email protected]>
Cc: Andy Shevchenko <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Splitting an earlier version of a patch that allowed calling
__request_region() while holding the resource lock into a series of
patches required changing the return code for the newly introduced
__request_region_locked().
Unfortunately this change was not carried through to a subsequent commit
56fd94919b ("kernel/resource: fix locking in request_free_mem_region")
in the series. This resulted in a use-after-free due to freeing the
struct resource without properly releasing it. Fix this by correcting the
return code check so that the struct is not freed if the request to add it
was successful.
Link: https://lkml.kernel.org/r/[email protected]
Fixes: 56fd94919b ("kernel/resource: fix locking in request_free_mem_region")
Signed-off-by: Alistair Popple <[email protected]>
Reported-by: kernel test robot <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Cc: Balbir Singh <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Jerome Glisse <[email protected]>
Cc: John Hubbard <[email protected]>
Cc: Muchun Song <[email protected]>
Cc: Oliver Sang <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
request_free_mem_region() is used to find an empty range of physical
addresses for hotplugging ZONE_DEVICE memory. It does this by iterating
over the range of possible addresses using region_intersects() to see if
the range is free before calling request_mem_region() to allocate the
region.
However the resource_lock is dropped between these two calls meaning by
the time request_mem_region() is called in request_free_mem_region()
another thread may have already reserved the requested region. This
results in unexpected failures and a message in the kernel log from
hitting this condition:
/*
* mm/hmm.c reserves physical addresses which then
* become unavailable to other users. Conflicts are
* not expected. Warn to aid debugging if encountered.
*/
if (conflict->desc == IORES_DESC_DEVICE_PRIVATE_MEMORY) {
pr_warn("Unaddressable device %s %pR conflicts with %pR",
conflict->name, conflict, res);
These unexpected failures can be corrected by holding resource_lock across
the two calls. This also requires memory allocation to be performed prior
to taking the lock.
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Alistair Popple <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Cc: Balbir Singh <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Jerome Glisse <[email protected]>
Cc: John Hubbard <[email protected]>
Cc: Muchun Song <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
It used to be true that we can have system RAM (IORESOURCE_SYSTEM_RAM |
IORESOURCE_BUSY) only on the first level in the resource tree. However,
this is no longer holds for driver-managed system RAM (i.e., added via
dax/kmem and virtio-mem), which gets added on lower levels, for example,
inside device containers.
IORESOURCE_SYSTEM_RAM is defined as IORESOURCE_MEM | IORESOURCE_SYSRAM and
just a special type of IORESOURCE_MEM.
The function walk_mem_res() only considers the first level and is used in
arch/x86/mm/ioremap.c:__ioremap_check_mem() only. We currently fail to
identify System RAM added by dax/kmem and virtio-mem as
"IORES_MAP_SYSTEM_RAM", for example, allowing for remapping of such
"normal RAM" in __ioremap_caller().
Let's find all IORESOURCE_MEM | IORESOURCE_BUSY resources, making the
function behave similar to walk_system_ram_res().
Link: https://lkml.kernel.org/r/[email protected]
Fixes: ebf71552bb ("virtio-mem: Add parent resource for all added "System RAM"")
Fixes: c221c0b030 ("device-dax: "Hotplug" persistent memory for use like normal RAM")
Signed-off-by: David Hildenbrand <[email protected]>
Reviewed-by: Dan Williams <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Andy Shevchenko <[email protected]>
Cc: Mauro Carvalho Chehab <[email protected]>
Cc: Dave Young <[email protected]>
Cc: Baoquan He <[email protected]>
Cc: Vivek Goyal <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Keith Busch <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Qian Cai <[email protected]>
Cc: Oscar Salvador <[email protected]>
Cc: Eric Biederman <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: Tom Lendacky <[email protected]>
Cc: Brijesh Singh <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Patch series "kernel/resource: make walk_system_ram_res() and walk_mem_res() search the whole tree", v2.
Playing with kdump+virtio-mem I noticed that kexec_file_load() does not
consider System RAM added via dax/kmem and virtio-mem when preparing the
elf header for kdump. Looking into the details, the logic used in
walk_system_ram_res() and walk_mem_res() seems to be outdated.
walk_system_ram_range() already does the right thing, let's change
walk_system_ram_res() and walk_mem_res(), and clean up.
Loading a kdump kernel via "kexec -p -s" ... will result in the kdump
kernel to also dump dax/kmem and virtio-mem added System RAM now.
Note: kexec-tools on x86-64 also have to be updated to consider this
memory in the kexec_load() case when processing /proc/iomem.
This patch (of 3):
It used to be true that we can have system RAM (IORESOURCE_SYSTEM_RAM |
IORESOURCE_BUSY) only on the first level in the resource tree. However,
this is no longer holds for driver-managed system RAM (i.e., added via
dax/kmem and virtio-mem), which gets added on lower levels, for example,
inside device containers.
We have two users of walk_system_ram_res(), which currently only
consideres the first level:
a) kernel/kexec_file.c:kexec_walk_resources() -- We properly skip
IORESOURCE_SYSRAM_DRIVER_MANAGED resources via
locate_mem_hole_callback(), so even after this change, we won't be
placing kexec images onto dax/kmem and virtio-mem added memory. No
change.
b) arch/x86/kernel/crash.c:fill_up_crash_elf_data() -- we're currently
not adding relevant ranges to the crash elf header, resulting in them
not getting dumped via kdump.
This change fixes loading a crashkernel via kexec_file_load() and
including dax/kmem and virtio-mem added System RAM in the crashdump on
x86-64. Note that e.g,, arm64 relies on memblock data and, therefore,
always considers all added System RAM already.
Let's find all IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY resources, making
the function behave like walk_system_ram_range().
Link: https://lkml.kernel.org/r/[email protected]
Link: https://lkml.kernel.org/r/[email protected]
Fixes: ebf71552bb ("virtio-mem: Add parent resource for all added "System RAM"")
Fixes: c221c0b030 ("device-dax: "Hotplug" persistent memory for use like normal RAM")
Signed-off-by: David Hildenbrand <[email protected]>
Reviewed-by: Dan Williams <[email protected]>
Acked-by: Baoquan He <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Andy Shevchenko <[email protected]>
Cc: Mauro Carvalho Chehab <[email protected]>
Cc: Dave Young <[email protected]>
Cc: Baoquan He <[email protected]>
Cc: Vivek Goyal <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Keith Busch <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Qian Cai <[email protected]>
Cc: Oscar Salvador <[email protected]>
Cc: Eric Biederman <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: Tom Lendacky <[email protected]>
Cc: Brijesh Singh <[email protected]>
Cc: "Eric W. Biederman" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Now we have for 'other' and 'type' variables
other type return
0 0 REGION_DISJOINT
0 x REGION_INTERSECTS
x 0 REGION_DISJOINT
x x REGION_MIXED
Obviously it's easier to check 'type' for 0 first instead of
currently checked 'other'.
Signed-off-by: Andy Shevchenko <[email protected]>
Reviewed-by: Hanjun Guo <[email protected]>
Tested-by: Hanjun Guo <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>
Patch series "selective merging of system ram resources", v4.
Some add_memory*() users add memory in small, contiguous memory blocks.
Examples include virtio-mem, hyper-v balloon, and the XEN balloon.
This can quickly result in a lot of memory resources, whereby the actual
resource boundaries are not of interest (e.g., it might be relevant for
DIMMs, exposed via /proc/iomem to user space). We really want to merge
added resources in this scenario where possible.
Resources are effectively stored in a list-based tree. Having a lot of
resources not only wastes memory, it also makes traversing that tree more
expensive, and makes /proc/iomem explode in size (e.g., requiring
kexec-tools to manually merge resources when creating a kdump header. The
current kexec-tools resource count limit does not allow for more than
~100GB of memory with a memory block size of 128MB on x86-64).
Let's allow to selectively merge system ram resources by specifying a new
flag for add_memory*(). Patch #5 contains a /proc/iomem example. Only
tested with virtio-mem.
This patch (of 8):
Let's make sure splitting a resource on memory hotunplug will never fail.
This will become more relevant once we merge selected System RAM resources
- then, we'll trigger that case more often on memory hotunplug.
In general, this function is already unlikely to fail. When we remove
memory, we free up quite a lot of metadata (memmap, page tables, memory
block device, etc.). The only reason it could really fail would be when
injecting allocation errors.
All other error cases inside release_mem_region_adjustable() seem to be
sanity checks if the function would be abused in different context - let's
add WARN_ON_ONCE() in these cases so we can catch them.
[[email protected]: fix use of ternary condition in release_mem_region_adjustable]
Link: https://lkml.kernel.org/r/[email protected]
Link: https://github.com/ClangBuiltLinux/linux/issues/1159
Signed-off-by: David Hildenbrand <[email protected]>
Signed-off-by: Nathan Chancellor <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Jason Gunthorpe <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Ard Biesheuvel <[email protected]>
Cc: Pankaj Gupta <[email protected]>
Cc: Baoquan He <[email protected]>
Cc: Wei Yang <[email protected]>
Cc: Anton Blanchard <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Boris Ostrovsky <[email protected]>
Cc: Christian Borntraeger <[email protected]>
Cc: Dave Jiang <[email protected]>
Cc: Eric Biederman <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Haiyang Zhang <[email protected]>
Cc: Heiko Carstens <[email protected]>
Cc: Jason Wang <[email protected]>
Cc: Juergen Gross <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: "K. Y. Srinivasan" <[email protected]>
Cc: Len Brown <[email protected]>
Cc: Leonardo Bras <[email protected]>
Cc: Libor Pechacek <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: "Michael S. Tsirkin" <[email protected]>
Cc: Nathan Lynch <[email protected]>
Cc: "Oliver O'Halloran" <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Pingfan Liu <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Cc: Roger Pau Monn <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Stephen Hemminger <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Vasily Gorbik <[email protected]>
Cc: Vishal Verma <[email protected]>
Cc: Wei Liu <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Linus Torvalds <[email protected]>
Close the hole of holding a mapping over kernel driver takeover event of
a given address range.
Commit 90a545e981 ("restrict /dev/mem to idle io memory ranges")
introduced CONFIG_IO_STRICT_DEVMEM with the goal of protecting the
kernel against scenarios where a /dev/mem user tramples memory that a
kernel driver owns. However, this protection only prevents *new* read(),
write() and mmap() requests. Established mappings prior to the driver
calling request_mem_region() are left alone.
Especially with persistent memory, and the core kernel metadata that is
stored there, there are plentiful scenarios for a /dev/mem user to
violate the expectations of the driver and cause amplified damage.
Teach request_mem_region() to find and shoot down active /dev/mem
mappings that it believes it has successfully claimed for the exclusive
use of the driver. Effectively a driver call to request_mem_region()
becomes a hole-punch on the /dev/mem device.
The typical usage of unmap_mapping_range() is part of
truncate_pagecache() to punch a hole in a file, but in this case the
implementation is only doing the "first half" of a hole punch. Namely it
is just evacuating current established mappings of the "hole", and it
relies on the fact that /dev/mem establishes mappings in terms of
absolute physical address offsets. Once existing mmap users are
invalidated they can attempt to re-establish the mapping, or attempt to
continue issuing read(2) / write(2) to the invalidated extent, but they
will then be subject to the CONFIG_IO_STRICT_DEVMEM checking that can
block those subsequent accesses.
Cc: Arnd Bergmann <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Russell King <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Fixes: 90a545e981 ("restrict /dev/mem to idle io memory ranges")
Signed-off-by: Dan Williams <[email protected]>
Reviewed-by: Kees Cook <[email protected]>
Link: https://lore.kernel.org/r/159009507306.847224.8502634072429766747.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Greg Kroah-Hartman <[email protected]>
find_next_iomem_res() shows up to be a source for overhead in dax
benchmarks.
Improve performance by not considering children of the tree if the top
level does not match. Since the range of the parents should include the
range of the children such check is redundant.
Running sysbench on dax (pmem emulation, with write_cache disabled):
sysbench fileio --file-total-size=3G --file-test-mode=rndwr \
--file-io-mode=mmap --threads=4 --file-fsync-mode=fdatasync run
Provides the following results:
events (avg/stddev)
-------------------
5.2-rc3: 1247669.0000/16075.39
w/patch: 1286320.5000/16402.72 (+3%)
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Nadav Amit <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: Toshi Kani <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Bjorn Helgaas <[email protected]>
Cc: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Keep the physical address allocation that hmm_add_device does with the
rest of the resource code, and allow future reuse of it without the hmm
wrapper.
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Jason Gunthorpe <[email protected]>
Reviewed-by: John Hubbard <[email protected]>
Reviewed-by: Dan Williams <[email protected]>
Signed-off-by: Jason Gunthorpe <[email protected]>
Add SPDX license identifiers to all files which:
- Have no license information of any form
- Have EXPORT_.*_SYMBOL_GPL inside which was used in the
initial scan/conversion to ignore the file
These files fall under the project license, GPL v2 only. The resulting SPDX
license identifier is:
GPL-2.0-only
Signed-off-by: Thomas Gleixner <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
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
HMM consumes physical address space for its own use, even
though nothing is mapped or accessible there. It uses a
special resource description (IORES_DESC_DEVICE_PRIVATE_MEMORY)
to uniquely identify these areas.
When HMM consumes address space, it makes a best guess about
what to consume. However, it is possible that a future memory
or device hotplug can collide with the reserved area. In the
case of these conflicts, there is an error message in
register_memory_resource().
Later patches in this series move register_memory_resource()
from using request_resource_conflict() to __request_region().
Unfortunately, __request_region() does not return the conflict
like the previous function did, which makes it impossible to
check for IORES_DESC_DEVICE_PRIVATE_MEMORY in a conflicting
resource.
Instead of warning in register_memory_resource(), move the
check into the core resource code itself (__request_region())
where the conflicting resource _is_ available. This has the
added bonus of producing a warning in case of HMM conflicts
with devices *or* RAM address space, as opposed to the RAM-
only warnings that were there previously.
Signed-off-by: Dave Hansen <[email protected]>
Reviewed-by: Jerome Glisse <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Dave Jiang <[email protected]>
Cc: Ross Zwisler <[email protected]>
Cc: Vishal Verma <[email protected]>
Cc: Tom Lendacky <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: Huang Ying <[email protected]>
Cc: Fengguang Wu <[email protected]>
Cc: Keith Busch <[email protected]>
Signed-off-by: Dan Williams <[email protected]>
Since commit c40dd2f766 ("powerpc: Add System RAM to /proc/iomem")
it is possible to use the generic walk_system_ram_range() and
the generic page_is_ram().
To enable the use of walk_system_ram_range() by the IBM EHEA ethernet
driver, we still need an export of the generic function.
As powerpc was the only user of CONFIG_ARCH_HAS_WALK_MEMORY, the
ifdef around the generic walk_system_ram_range() has become useless
and can be dropped.
Fixes: c40dd2f766 ("powerpc: Add System RAM to /proc/iomem")
Signed-off-by: Christophe Leroy <[email protected]>
[mpe: Keep the EXPORT_SYMBOL_GPL in powerpc code]
Signed-off-by: Michael Ellerman <[email protected]>
This is a preparation for the next patch.
Currently, we only call release_mem_region_adjustable() in __remove_pages
if the zone is not ZONE_DEVICE, because resources that belong to HMM/devm
are being released by themselves with devm_release_mem_region.
Since we do not want to touch any zone/page stuff during the removing of
the memory (but during the offlining), we do not want to check for the
zone here. So we need another way to tell release_mem_region_adjustable()
to not realease the resource in case it belongs to HMM/devm.
HMM/devm acquires/releases a resource through
devm_request_mem_region/devm_release_mem_region.
These resources have the flag IORESOURCE_MEM, while resources acquired by
hot-add memory path (register_memory_resource()) contain
IORESOURCE_SYSTEM_RAM.
So, we can check for this flag in release_mem_region_adjustable, and if
the resource does not contain such flag, we know that we are dealing with
a HMM/devm resource, so we can back off.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Oscar Salvador <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Reviewed-by: Pavel Tatashin <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Jerome Glisse <[email protected]>
Cc: Jonathan Cameron <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Oscar Salvador <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The first group of warnings is caused by a "/**" kernel-doc notation
marker but the function comments are not in kernel-doc format.
Also add another error return value here.
../kernel/resource.c:337: warning: Function parameter or member 'start' not described in 'find_next_iomem_res'
../kernel/resource.c:337: warning: Function parameter or member 'end' not described in 'find_next_iomem_res'
../kernel/resource.c:337: warning: Function parameter or member 'flags' not described in 'find_next_iomem_res'
../kernel/resource.c:337: warning: Function parameter or member 'desc' not described in 'find_next_iomem_res'
../kernel/resource.c:337: warning: Function parameter or member 'first_lvl' not described in 'find_next_iomem_res'
../kernel/resource.c:337: warning: Function parameter or member 'res' not described in 'find_next_iomem_res'
Add the missing function parameter documentation for the other warnings:
../kernel/resource.c:409: warning: Function parameter or member 'arg' not described in 'walk_iomem_res_desc'
../kernel/resource.c:409: warning: Function parameter or member 'func' not described in 'walk_iomem_res_desc'
Signed-off-by: Randy Dunlap <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Fixes: b69c2e20f6 ("resource: Clean it up a bit")
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
Previously find_next_iomem_res() used "*res" as both an input parameter for
the range to search and the type of resource to search for, and an output
parameter for the resource we found, which makes the interface confusing.
The current callers use find_next_iomem_res() incorrectly because they
allocate a single struct resource and use it for repeated calls to
find_next_iomem_res(). When find_next_iomem_res() returns a resource, it
overwrites the start, end, flags, and desc members of the struct. If we
call find_next_iomem_res() again, we must update or restore these fields.
The previous code restored res.start and res.end, but not res.flags or
res.desc.
Since the callers did not restore res.flags, if they searched for flags
IORESOURCE_MEM | IORESOURCE_BUSY and found a resource with flags
IORESOURCE_MEM | IORESOURCE_BUSY | IORESOURCE_SYSRAM, the next search would
incorrectly skip resources unless they were also marked as
IORESOURCE_SYSRAM.
Fix this by restructuring the interface so it takes explicit "start, end,
flags" parameters and uses "*res" only as an output parameter.
Based on a patch by Lianbo Jiang <[email protected]>.
[ bp: While at it:
- make comments kernel-doc style.
-
Originally-by: http://lore.kernel.org/lkml/[email protected]
Signed-off-by: Bjorn Helgaas <[email protected]>
Signed-off-by: Borislav Petkov <[email protected]>
CC: Andrew Morton <[email protected]>
CC: Brijesh Singh <[email protected]>
CC: Dan Williams <[email protected]>
CC: H. Peter Anvin <[email protected]>
CC: Lianbo Jiang <[email protected]>
CC: Takashi Iwai <[email protected]>
CC: Thomas Gleixner <[email protected]>
CC: Tom Lendacky <[email protected]>
CC: Vivek Goyal <[email protected]>
CC: Yaowei Bai <[email protected]>
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: x86-ml <[email protected]>
Link: http://lkml.kernel.org/r/153805812916.1157.177580438135143788.stgit@bhelgaas-glaptop.roam.corp.google.com
find_next_iomem_res() finds an iomem resource that covers part of a range
described by "start, end". All callers expect that range to be inclusive,
i.e., both start and end are included, but find_next_iomem_res() doesn't
handle the end address correctly.
If it finds an iomem resource that contains exactly the end address, it
skips it, e.g., if "start, end" is [0x0-0x10000] and there happens to be an
iomem resource [mem 0x10000-0x10000] (the single byte at 0x10000), we skip
it:
find_next_iomem_res(...)
{
start = 0x0;
end = 0x10000;
for (p = next_resource(...)) {
# p->start = 0x10000;
# p->end = 0x10000;
# we *should* return this resource, but this condition is false:
if ((p->end >= start) && (p->start < end))
break;
Adjust find_next_iomem_res() so it allows a resource that includes the
single byte at the end of the range. This is a corner case that we
probably don't see in practice.
Fixes: 58c1b5b079 ("[PATCH] memory hotadd fixes: find_next_system_ram catch range fix")
Signed-off-by: Bjorn Helgaas <[email protected]>
Signed-off-by: Borislav Petkov <[email protected]>
CC: Andrew Morton <[email protected]>
CC: Brijesh Singh <[email protected]>
CC: Dan Williams <[email protected]>
CC: H. Peter Anvin <[email protected]>
CC: Lianbo Jiang <[email protected]>
CC: Takashi Iwai <[email protected]>
CC: Thomas Gleixner <[email protected]>
CC: Tom Lendacky <[email protected]>
CC: Vivek Goyal <[email protected]>
CC: Yaowei Bai <[email protected]>
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: x86-ml <[email protected]>
Link: http://lkml.kernel.org/r/153805812254.1157.16736368485811773752.stgit@bhelgaas-glaptop.roam.corp.google.com
Pull libnvdimm updates from Dan Williams:
"This adds a user for the new 'bytes-remaining' updates to
memcpy_mcsafe() that you already received through Ingo via the
x86-dax- for-linus pull.
Not included here, but still targeting this cycle, is support for
handling memory media errors (poison) consumed via userspace dax
mappings.
Summary:
- DAX broke a fundamental assumption of truncate of file mapped
pages. The truncate path assumed that it is safe to disconnect a
pinned page from a file and let the filesystem reclaim the physical
block. With DAX the page is equivalent to the filesystem block.
Introduce dax_layout_busy_page() to enable filesystems to wait for
pinned DAX pages to be released. Without this wait a filesystem
could allocate blocks under active device-DMA to a new file.
- DAX arranges for the block layer to be bypassed and uses
dax_direct_access() + copy_to_iter() to satisfy read(2) calls.
However, the memcpy_mcsafe() facility is available through the pmem
block driver. In order to safely handle media errors, via the DAX
block-layer bypass, introduce copy_to_iter_mcsafe().
- Fix cache management policy relative to the ACPI NFIT Platform
Capabilities Structure to properly elide cache flushes when they
are not necessary. The table indicates whether CPU caches are
power-fail protected. Clarify that a deep flush is always performed
on REQ_{FUA,PREFLUSH} requests"
* tag 'libnvdimm-for-4.18' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm: (21 commits)
dax: Use dax_write_cache* helpers
libnvdimm, pmem: Do not flush power-fail protected CPU caches
libnvdimm, pmem: Unconditionally deep flush on *sync
libnvdimm, pmem: Complete REQ_FLUSH => REQ_PREFLUSH
acpi, nfit: Remove ecc_unit_size
dax: dax_insert_mapping_entry always succeeds
libnvdimm, e820: Register all pmem resources
libnvdimm: Debug probe times
linvdimm, pmem: Preserve read-only setting for pmem devices
x86, nfit_test: Add unit test for memcpy_mcsafe()
pmem: Switch to copy_to_iter_mcsafe()
dax: Report bytes remaining in dax_iomap_actor()
dax: Introduce a ->copy_to_iter dax operation
uio, lib: Fix CONFIG_ARCH_HAS_UACCESS_MCSAFE compilation
xfs, dax: introduce xfs_break_dax_layouts()
xfs: prepare xfs_break_layouts() for another layout type
xfs: prepare xfs_break_layouts() to be called with XFS_MMAPLOCK_EXCL
mm, fs, dax: handle layout changes to pinned dax mappings
mm: fix __gup_device_huge vs unmap
mm: introduce MEMORY_DEVICE_FS_DAX and CONFIG_DEV_PAGEMAP_OPS
...
There is currently a mismatch between the resources that will trigger
the e820_pmem driver to register/load and the resources that will
actually be surfaced as pmem ranges. register_e820_pmem() uses
walk_iomem_res_desc() which includes children and siblings. In contrast,
e820_pmem_probe() only considers top level resources. For example the
following resource tree results in the driver being loaded, but no
resources being registered:
398000000000-39bfffffffff : PCI Bus 0000:ae
39be00000000-39bf07ffffff : PCI Bus 0000:af
39be00000000-39beffffffff : 0000:af:00.0
39be10000000-39beffffffff : Persistent Memory (legacy)
Fix this up to allow definitions of "legacy" pmem ranges anywhere in
system-physical address space. Not that it is a recommended or safe to
define a pmem range in PCI space, but it is useful for debug /
experimentation, and the restriction on being a top-level resource was
arbitrary.
Cc: Christoph Hellwig <[email protected]>
Signed-off-by: Dan Williams <[email protected]>
We've got a bug report indicating a kernel panic at booting on an x86-32
system, and it turned out to be the invalid PCI resource assigned after
reallocation. __find_resource() first aligns the resource start address
and resets the end address with start+size-1 accordingly, then checks
whether it's contained. Here the end address may overflow the integer,
although resource_contains() still returns true because the function
validates only start and end address. So this ends up with returning an
invalid resource (start > end).
There was already an attempt to cover such a problem in the commit
47ea91b405 ("Resource: fix wrong resource window calculation"), but
this case is an overseen one.
This patch adds the validity check of the newly calculated resource for
avoiding the integer overflow problem.
Bugzilla: http://bugzilla.opensuse.org/show_bug.cgi?id=1086739
Link: http://lkml.kernel.org/r/[email protected]
Fixes: 23c570a674 ("resource: ability to resize an allocated resource")
Signed-off-by: Takashi Iwai <[email protected]>
Reported-by: Michael Henders <[email protected]>
Tested-by: Michael Henders <[email protected]>
Reviewed-by: Andrew Morton <[email protected]>
Cc: Ram Pai <[email protected]>
Cc: Bjorn Helgaas <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Set resource structs inserted by __reserve_region_with_split() to have the
correct type. Setting the type doesn't fix any functional problem but
makes %pR on the resource work better.
Signed-off-by: Bjorn Helgaas <[email protected]>
When we reserve regions because the user specified a "reserve=" parameter,
set the resource type to either IORESOURCE_IO (for regions below 0x10000)
or IORESOURCE_MEM. The test for 0x10000 is just a heuristic; obviously
there can be memory below 0x10000 as well.
Improve documentation of the "reserve=" parameter.
Signed-off-by: Bjorn Helgaas <[email protected]>
Dan Williams