Add Perf support for metag.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Add [!]METAG to a couple of Kconfig dependencies in lib/Kconfig.debug.
Don't allow stack utilization instrumentation on metag, and allow
building with frame pointers.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: "Paul E. McKenney" <paul.mckenney@linaro.org>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Commit cc2383ec06 ("mm: introduce
arch-specific vma flag VM_ARCH_1") merged in v3.7-rc1.
The above commit combined several arch-specific vma flags into one, and
in the process it changed the VM_GROWSUP definition to depend on
specific architectures rather than CONFIG_STACK_GROWSUP. Therefore add
an ifdef for CONFIG_METAG to also set VM_GROWSUP.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michel Lespinasse <walken@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-mm@kvack.org
Add SMP support for metag. This allows Linux to take control of multiple
hardware threads on a single Meta core, treating them as separate Linux
CPUs.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add basic metag documentation. This includes an outline description of
the ABIs (including syscall ABI) and calling conventions, similar to the
one in Documentation/frv/.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Rob Landley <rob@landley.net>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: linux-doc@vger.kernel.org
Add header files to implement Meta hardware thread locks (used by some
other atomic operations), atomics, spinlocks, and bitops.
There are 2 main types of atomic primitives for metag (in addition to
IRQs off on UP):
- LOCK instructions provide locking between hardware threads.
- LNKGET/LNKSET instructions provide load-linked/store-conditional
operations allowing for lighter weight atomics on Meta2
LOCK instructions allow for hardware threads to acquire voluntary or
exclusive hardware thread locks:
- LOCK0 releases exclusive and voluntary lock from the running hardware
thread.
- LOCK1 acquires the voluntary hardware lock, blocking until it becomes
available.
- LOCK2 implies LOCK1, and additionally acquires the exclusive hardware
lock, blocking all other hardware threads from executing.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add metag system call and gateway page interfaces. The metag
architecture port uses the generic system call numbers from
asm-generic/unistd.h, as well as a user gateway page mapped at
0x6ffff000 which contains fast atomic primitives (depending on SMP) and
a fast method of accessing TLS data.
System calls use the SWITCH instruction with the immediate 0x440001 to
signal a system call.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Meta core internal interrupts (from HWSTATMETA and friends) are vectored
onto the TR1 core trigger for the current thread. This is demultiplexed
in irq-metag.c to individual Linux IRQs for each internal interrupt.
External SoC interrupts (from HWSTATEXT and friends) are vectored onto
the TR2 core trigger for the current thread. This is demultiplexed in
irq-metag-ext.c to individual Linux IRQs for each external SoC interrupt.
The external irqchip has devicetree bindings for configuring the number
of irq banks and the type of masking available.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Grant Likely <grant.likely@secretlab.ca>
Cc: Rob Herring <rob.herring@calxeda.com>
Cc: Rob Landley <rob@landley.net>
Cc: Dom Cobley <popcornmix@gmail.com>
Cc: Simon Arlott <simon@fire.lp0.eu>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Maxime Ripard <maxime.ripard@free-electrons.com>
Cc: devicetree-discuss@lists.ozlabs.org
Cc: linux-doc@vger.kernel.org
Add trap code for metag. At the lowest level Meta traps (and return from
interrupt instruction - RTI) simply swap the PC and PCX registers and
optionally toggle the interrupt status bit (ISTAT). Low level TBX code
in tbipcx.S handles the core context save, determine the TBX signal
number based on the core trigger that fired (using the TXSTATI status
register), and call TBX signal handlers (mostly in traps.c) via a vector
table.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Add time keeping code for metag. Meta hardware threads have 2 timers.
The background timer (TXTIMER) is used as a free-running time base, and
the interrupt timer (TXTIMERI) is used for the timer interrupt. Both
counters traditionally count at approximately 1MHz.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: John Stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
The ptrace interface for metag provides access to some core register
sets using the PTRACE_GETREGSET and PTRACE_SETREGSET operations. The
details of the internal context structures is abstracted into user API
structures to both ease use and allow flexibility to change the internal
context layouts. Copyin and copyout functions for these register sets
are exposed to allow signal handling code to use them to copy to and
from the signal context.
struct user_gp_regs (NT_PRSTATUS) provides access to the core general
purpose register context.
struct user_cb_regs (NT_METAG_CBUF) provides access to the TXCATCH*
registers which contains information abuot a memory fault, unaligned
access error or watchpoint. This can be modified to alter the way the
fault is replayed on resume ("catch replay"), or to prevent the replay
taking place.
struct user_rp_state (NT_METAG_RPIPE) provides access to the state of
the Meta read pipeline which can be used to hide memory latencies in
hand optimised data loops.
Extended DSP register state, DSP RAM, and hardware breakpoint registers
aren't yet exposed through ptrace.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Denys Vlasenko <vda.linux@googlemail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Tony Lindgren <tony@atomide.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Meta has instructions for accessing:
- bytes - GETB (1 byte)
- words - GETW (2 bytes)
- doublewords - GETD (4 bytes)
- longwords - GETL (8 bytes)
All accesses must be aligned. Unaligned accesses can be detected and
made to fault on Meta2, however it isn't possible to fix up unaligned
writes so we don't bother fixing up reads either.
This patch adds metag memory handling code including:
- I/O memory (io.h, ioremap.c): Actually any virtual memory can be
accessed with these helpers. A part of the non-MMUable address space
is used for memory mapped I/O. The ioremap() function is implemented
one to one for non-MMUable addresses.
- User memory (uaccess.h, usercopy.c): User memory is directly
accessible from privileged code.
- Kernel memory (maccess.c): probe_kernel_write() needs to be
overwridden to use the I/O functions when doing a simple aligned
write to non-writecombined memory, otherwise the write may be split
by the generic version.
Note that due to the fact that a portion of the virtual address space is
non-MMUable, and therefore always maps directly to the physical address
space, metag specific I/O functions are made available (metag_in32,
metag_out32 etc). These cast the address argument to a pointer so that
they can be used with raw physical addresses. These accessors are only
to be used for accessing fixed core Meta architecture registers in the
non-MMU region, and not for any SoC/peripheral registers.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add memory management files for metag.
Meta's 32bit virtual address space is split into two halves:
- local (0x08000000-0x7fffffff): traditionally local to a hardware
thread and incoherent between hardware threads. Each hardware thread
has it's own local MMU table. On Meta2 the local space can be
globally coherent (GCOn) if the cache partitions coincide.
- global (0x88000000-0xffff0000): coherent and traditionally global
between hardware threads. On Meta2, each hardware thread has it's own
global MMU table.
The low 128MiB of each half is non-MMUable and maps directly to the
physical address space:
- 0x00010000-0x07ffffff: contains Meta core registers and maps SoC bus
- 0x80000000-0x87ffffff: contains low latency global core memories
Linux usually further splits the local virtual address space like this:
- 0x08000000-0x3fffffff: user mappings
- 0x40000000-0x7fffffff: kernel mappings
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add cache and TLB handling code for metag, including the required
callbacks used by MM switches and DMA operations. Caches can be
partitioned between the hardware threads and the global space, however
this is usually configured by the bootloader so Linux doesn't make any
changes to this configuration. TLBs aren't configurable, so only need
consideration to flush them.
On Meta1 the L1 cache was VIVT which required a full flush on MM switch.
Meta2 has a VIPT L1 cache so it doesn't require the full flush on MM
switch. Meta2 can also have a writeback L2 with hardware prefetch which
requires some special handling. Support is optional, and the L2 can be
detected and initialised by Linux.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add source files from the Thread Binary Interface (TBI) library which
provides useful low level operations and traps/context management.
Among other things it handles interrupt/exception/syscall entry (in
tbipcx.S).
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add the main header for the Thread Binary Interface (TBI) library which
provides useful low level operations and trap/context management.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add boot code for metag. Due to the multi-threaded nature of Meta it is
not uncommon for an RTOS or bare metal application to be started on
other hardware threads by the bootloader. Since there is a single MMU
switch which affects all threads, the MMU is traditionally configured by
the bootloader prior to starting Linux. The bootloader passes a
structure to Linux which among other things contains information about
memory regions which have been mapped. Linux then assumes control of the
local heap memory region.
A kernel arguments string pointer or a flattened device tree pointer can
be provided in the third argument.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add the header <asm/metag_mem.h> describing addresses, fields, and bits
of various core memory mapped registers in the low non-MMU region.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add a couple of header files containing core architecture constants.
The first (<asm/metag_isa.h>) contains some constants relating to the
instruction set, such as values to give to the CACHEW and CACHER
instructions.
The second (<asm/metag_regs.h>) contains constants for the core register
units directly accessible to various instructions, and for the
registers, fields, and bits in those units. The main units described are
the control unit (CT.*), the trigger unit (TR.*), and the run-time trace
unit (TT.*).
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Add MAINTAINERS entry for the metag architecture port.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mauro Carvalho Chehab <mchehab@redhat.com>
Cc: Cesar Eduardo Barros <cesarb@cesarb.net>
Cc: Joe Perches <joe@perches.com>
Cc: "David S. Miller" <davem@davemloft.net>
Some 32 bit architectures require 64 bit values to be aligned (for
example Meta which has 64 bit read/write instructions). These require 8
byte alignment of event data too, so use
!CONFIG_HAVE_64BIT_ALIGNED_ACCESS instead of !CONFIG_64BIT ||
CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS to decide alignment, and align
buffer_data_page::data accordingly.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org> (previous version subtly different)
On 64 bit architectures with no efficient unaligned access, padding and
explicit alignment must be added in various places to prevent unaligned
64bit accesses (such as taskstats and trace ring buffer).
However this also needs to apply to 32 bit architectures with 64 bit
accesses requiring alignment such as metag.
This is solved by adding a new Kconfig symbol HAVE_64BIT_ALIGNED_ACCESS
which defaults to 64BIT && !HAVE_EFFICIENT_UNALIGNED_ACCESS, and can be
explicitly selected by METAG and any other relevant architectures. This
can be used in various places to determine whether 64bit alignment is
required.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Eric Paris <eparis@redhat.com>
Cc: Will Drewry <wad@chromium.org>
The commit "binfmt_elf: cleanups"
(f670d0ecda) removed an ifndef elf_map but
this breaks compilation for metag which does define elf_map.
This adds the ifndef back in as it was before, but does not affect the
other cleanups made by that patch.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Acked-by: Mikael Pettersson <mikpe@it.uu.se>
Some architectures have symbol prefixes and set CONFIG_SYMBOL_PREFIX,
but this wasn't taken into account by the generic cond_syscall. It's
easy enough to fix in a generic fashion, so add the symbol prefix to
symbol names in cond_syscall when CONFIG_SYMBOL_PREFIX is set.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Make asm-generic/io.h check CONFIG_VIRT_TO_BUS before defining
virt_to_bus() and bus_to_virt(), otherwise it's easy to accidentally
have a silently failing incorrect direct mapped definition rather then
no definition at all.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
an SSD to be used as a cache in front of a slower device. Cache
tuning is delegated to interchangeable policy modules so these can
be developed independently of the mechanics needed to shuffle the
data around.
Other than that, kcopyd users acquire a throttling parameter, ioctl
buffer usage gets streamlined, more mempool reliance is reduced
and there are a few other bug fixes and tidy-ups.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)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=yXuO
-----END PGP SIGNATURE-----
Merge tag 'dm-3.9-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm
Pull device-mapper update from Alasdair G Kergon:
"The main addition here is a long-desired target framework to allow an
SSD to be used as a cache in front of a slower device. Cache tuning
is delegated to interchangeable policy modules so these can be
developed independently of the mechanics needed to shuffle the data
around.
Other than that, kcopyd users acquire a throttling parameter, ioctl
buffer usage gets streamlined, more mempool reliance is reduced and
there are a few other bug fixes and tidy-ups."
* tag 'dm-3.9-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-dm: (30 commits)
dm cache: add cleaner policy
dm cache: add mq policy
dm: add cache target
dm persistent data: add bitset
dm persistent data: add transactional array
dm thin: remove cells from stack
dm bio prison: pass cell memory in
dm persistent data: add btree_walk
dm: add target num_write_bios fn
dm kcopyd: introduce configurable throttling
dm ioctl: allow message to return data
dm ioctl: optimize functions without variable params
dm ioctl: introduce ioctl_flags
dm: merge io_pool and tio_pool
dm: remove unused _rq_bio_info_cache
dm: fix limits initialization when there are no data devices
dm snapshot: add missing module aliases
dm persistent data: set some btree fn parms const
dm: refactor bio cloning
dm: rename bio cloning functions
...
Pull SCSI target patches from Nicholas Bellinger:
"Here are the remaining target-pending patches for v3.9-rc1.
The most important one here is the immediate queue starvation
regression fix for iscsi-target, which addresses a bug that's
effecting v3.5+ kernels under heavy sustained READ only workloads.
Thanks alot to Benjamin Estrabaud for helping to track this down!
Also included is a pSCSI exception bugfix from Asias, along with a
handful of other minor changes. Both bugfixes are CC'ed to stable."
* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
target/pscsi: Rename sg_num to nr_vecs in pscsi_get_bio()
target/pscsi: Fix page increment
target/pscsi: Drop unnecessary NULL assignment to bio->bi_next
target: Add __exit annotation for module_exit functions
iscsi-target: Fix immediate queue starvation regression with DATAIN
This is an assorted set of stragglers into the merge window with driver
updates for qla2xxx, megaraid_sas, storvsc and ufs. It also includes pulls of
the uapi tree (all the remaining SCSI pieces) and the fcoe tree (updates to
fcoe and libfc)
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.19 (GNU/Linux)
iQEcBAABAgAGBQJRMHJHAAoJEDeqqVYsXL0M9tAH/2YG3TCfy0RFAejGgLfX9OGH
6eFe71m7Z8nfIEneAnm5BuKjCx3QFRp5UFjJZdFHLP1Qv0TbpKs4FnZyeSGTxLQp
S1fZc5sTWmsb5qYxLaukKopC6sFx+hNI2dvB+rgKcd+nWy1RzG7lGqbS4CRNE76q
UNByqlfqJxn5cfQw7dg2zOUKlGaGL2jSyFf0QFXR2IZzO33PeyBPfKDFeJC6b+oc
XTy9MK9V5u6ne3XimDTU2hP4lPAsZaJtcqsv1Gvv2y+BHalQiPqfL6bZMvN3Zbfq
hfT+i2xnYy85858gxtyIhzHwU14zF7I0HEWiVpddsF9NDK7iNKvK8aWHaTs7qis=
=hvGQ
-----END PGP SIGNATURE-----
Merge tag 'scsi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Pull SCSI updates from James Bottomley:
"This is an assorted set of stragglers into the merge window with
driver updates for qla2xxx, megaraid_sas, storvsc and ufs.
It also includes pulls of the uapi tree (all the remaining SCSI
pieces) and the fcoe tree (updates to fcoe and libfc)"
* tag 'scsi-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (81 commits)
[SCSI] ufs: Separate PCI code into glue driver
[SCSI] ufs: Segregate PCI Specific Code
[SCSI] scsi: fix lpfc build when wmb() is defined as mb()
[SCSI] storvsc: Handle dynamic resizing of the device
[SCSI] storvsc: Restructure error handling code on command completion
[SCSI] storvsc: avoid usage of WRITE_SAME
[SCSI] aacraid: suppress two GCC warnings
[SCSI] hpsa: check for dma_mapping_error in hpsa_passthru ioctls
[SCSI] hpsa: reorganize error handling in hpsa_passthru_ioctl
[SCSI] hpsa: check for dma_mapping_error in hpsa_map_sg_chain_block
[SCSI] hpsa: Check for dma_mapping_error for all code paths using fill_cmd
[SCSI] hpsa: Check for dma_mapping_error in hpsa_map_one
[SCSI] dc395x: uninitialized variable in device_alloc()
[SCSI] Fix range check in scsi_host_dif_capable()
[SCSI] storvsc: Initialize the sglist
[SCSI] mpt2sas: Add support for OEM specific controller
[SCSI] ipr: Fix oops while resetting an ipr adapter
[SCSI] fnic: Fnic Trace Utility
[SCSI] fnic: New debug flags and debug log messages
[SCSI] fnic: fnic driver may hit BUG_ON on device reset
...
commit e72837e3e7 introduced
a default SET_PERSONALITY() in include/linux/elf.h.
This breaks with our own SET_PERSONALITY define for
32bit userspace on 64bit kernel.
Signed-off-by: Helge Deller <deller@gmx.de>
PA-RISC is the only arch that installs the modules when installing the kernel.
Signed-off-by: Rolf Eike Beer <eike-kernel@sf-tec.de>
Signed-off-by: Helge Deller <deller@gmx.de>
This fixes compile warnings like this one:
net/ipv4/igmp.c: In function ‘ip_mc_leave_group’:
net/ipv4/igmp.c:1898:3: warning: overflow in implicit constant conversion [-Woverflow]
atomic_sub() is defined as __atomic_add_return(-(VAL),(v))))
and if VAL is of type unsigned int (as returned by sizeof()), negating
this value will overflow. Fix this by type-casting VAL to int type.
Signed-off-by: Helge Deller <deller@gmx.de>
Avoid this warning, while still prevent gcc from optimizing away the exception code:
arch/parisc/lib/memcpy.c: In function ‘pa_memcpy’:
arch/parisc/lib/memcpy.c:256:2: warning: ‘dummy’ may be used uninitialized in this function [-Wuninitialized]
Signed-off-by: Helge Deller <deller@gmx.de>
Tim found:
WARNING: at arch/x86/kernel/smpboot.c:324 topology_sane.isra.2+0x6f/0x80()
Hardware name: S2600CP
sched: CPU #1's llc-sibling CPU #0 is not on the same node! [node: 1 != 0]. Ignoring dependency.
smpboot: Booting Node 1, Processors #1
Modules linked in:
Pid: 0, comm: swapper/1 Not tainted 3.9.0-0-generic #1
Call Trace:
set_cpu_sibling_map+0x279/0x449
start_secondary+0x11d/0x1e5
Don Morris reproduced on a HP z620 workstation, and bisected it to
commit e8d1955258 ("acpi, memory-hotplug: parse SRAT before memblock
is ready")
It turns out movable_map has some problems, and it breaks several things
1. numa_init is called several times, NOT just for srat. so those
nodes_clear(numa_nodes_parsed)
memset(&numa_meminfo, 0, sizeof(numa_meminfo))
can not be just removed. Need to consider sequence is: numaq, srat, amd, dummy.
and make fall back path working.
2. simply split acpi_numa_init to early_parse_srat.
a. that early_parse_srat is NOT called for ia64, so you break ia64.
b. for (i = 0; i < MAX_LOCAL_APIC; i++)
set_apicid_to_node(i, NUMA_NO_NODE)
still left in numa_init. So it will just clear result from early_parse_srat.
it should be moved before that....
c. it breaks ACPI_TABLE_OVERIDE...as the acpi table scan is moved
early before override from INITRD is settled.
3. that patch TITLE is total misleading, there is NO x86 in the title,
but it changes critical x86 code. It caused x86 guys did not
pay attention to find the problem early. Those patches really should
be routed via tip/x86/mm.
4. after that commit, following range can not use movable ram:
a. real_mode code.... well..funny, legacy Node0 [0,1M) could be hot-removed?
b. initrd... it will be freed after booting, so it could be on movable...
c. crashkernel for kdump...: looks like we can not put kdump kernel above 4G
anymore.
d. init_mem_mapping: can not put page table high anymore.
e. initmem_init: vmemmap can not be high local node anymore. That is
not good.
If node is hotplugable, the mem related range like page table and
vmemmap could be on the that node without problem and should be on that
node.
We have workaround patch that could fix some problems, but some can not
be fixed.
So just remove that offending commit and related ones including:
f7210e6c4a ("mm/memblock.c: use CONFIG_HAVE_MEMBLOCK_NODE_MAP to
protect movablecore_map in memblock_overlaps_region().")
01a178a94e ("acpi, memory-hotplug: support getting hotplug info from
SRAT")
27168d38fa ("acpi, memory-hotplug: extend movablemem_map ranges to
the end of node")
e8d1955258 ("acpi, memory-hotplug: parse SRAT before memblock is
ready")
fb06bc8e5f ("page_alloc: bootmem limit with movablecore_map")
42f47e27e7 ("page_alloc: make movablemem_map have higher priority")
6981ec3114 ("page_alloc: introduce zone_movable_limit[] to keep
movable limit for nodes")
34b71f1e04 ("page_alloc: add movable_memmap kernel parameter")
4d59a75125 ("x86: get pg_data_t's memory from other node")
Later we should have patches that will make sure kernel put page table
and vmemmap on local node ram instead of push them down to node0. Also
need to find way to put other kernel used ram to local node ram.
Reported-by: Tim Gardner <tim.gardner@canonical.com>
Reported-by: Don Morris <don.morris@hp.com>
Bisected-by: Don Morris <don.morris@hp.com>
Tested-by: Don Morris <don.morris@hp.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Tejun Heo <tj@kernel.org>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
James Hogan