We can free memory allocated with lmb_alloc() by removing it from the
list of reserved LMBs. Rework lmb_remove() to allow that possibility
and add lmb_free() which exploits it.

BenH: Removed some useless parenthesis
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

Add a flag to let a platform ioremap memory regions marked as reserved.

This flag will be used later by the Nintendo Wii support code to allow
ioremapping the I/O region sitting between MEM1 and MEM2 and marked
as reserved RAM in the patch "wii: use both mem1 and mem2 as ram".

This will no longer be needed when proper discontig memory support
for 32-bit PowerPC is added to the kernel.
Signed-off-by: NAlbert Herranz <albert_herranz@yahoo.es>
Acked-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NGrant Likely <grant.likely@secretlab.ca>

We call lmb_end_of_DRAM() to test whether a DMA mask is ok on a machine
without IOMMU, but this function is marked as __init.

I don't think there's a clean way to get the top of RAM max_pfn doesn't
appear to include highmem or I missed (or we have a bug :-) so for now,
let's just avoid having a broken 2.6.31 by making this function
non-__init and we can revisit later.
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The lmb_dump_all() output didn't include the RMO size, which is
interesting on powerpc. The output was also a bit spacey and not well
aligned, and didn't show you the end addresses.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

The idea of the implementation of this fix is from Michael Ellerman.

This function has two loops, but they each interpret the memory_limit
value differently.  The first loop interprets it as a "size limit"
whereas the second loop interprets it as an "address limit".

Before the second loop runs, reset memory_limit to lmb_end_of_DRAM()
so that it all works out.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
Acked-by: NMichael Ellerman <michael@ellerman.id.au>

lib/lmb.c: In function 'lmb_dump_all':
lib/lmb.c:51: warning: format '%lx' expects type 'long unsigned int', but argument 2 has type 'u64'
Signed-off-by: NKumar Gala <galak@kernel.crashing.org>

Having to muck with the build and set DEBUG just to
get lmb_dump_all() to print things isn't very useful.

So use pr_info() and use an early boot param
"lmb=debug" so we can simply ask users to reboot
with this option when we need some debugging from
them.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When allocating, if we will align up the size when making
the reservation, we should also align the size for the
check that the space is actually available.

The simplest thing is to just aling the size up from
the beginning, then we can use plain 'size' throughout.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Provide walk_memory_resource() for 64-bit powerpc.  PowerPC maintains
logical memory region mapping in the lmb.memory structure.  Walk
through these structures and do the callbacks for the contiguous
chunks.
Signed-off-by: NBadari Pulavarty <pbadari@us.ibm.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The powerpc kernel maintains information about logical memory blocks
in the lmb.memory structure, which is initialized and updated at boot
time, but not when memory is added or removed while the kernel is
running.

This adds a hotplug memory notifier which updates lmb.memory when
memory is added or removed.  This information is useful for eHEA
driver to find out the memory layout and holes.

NOTE: No special locking is needed for lmb_add() and lmb_remove().
Calls to these are serialized by caller. (pSeries_reconfig_chain).
Signed-off-by: NBadari Pulavarty <pbadari@us.ibm.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Changeset d9024df0 ("[LMB] Restructure
allocation loops to avoid unsigned underflow") removed the alignment
of the 'size' argument to call lmb_add_region() done by __lmb_alloc_base().

In doing so it reintroduced the bug fixed by changeset
eea89e13 ("[LMB]: Fix bug in
__lmb_alloc_base().").

This puts it back.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There is a potential bug in __lmb_alloc_base where we subtract `size'
from the base address of a reserved region without checking whether
the subtraction could wrap around and produce a very large unsigned
value.  In fact it probably isn't possible to hit the bug in practice
since it would only occur in the situation where we can't satisfy the
allocation request and there is a reserved region starting at 0.

This fixes the potential bug by breaking out of the loop when we get
to the point where the base of the reserved region is less than the
size requested.  This also restructures the loop to be a bit easier to
follow.

The same logic got copied into lmb_alloc_nid_unreserved, so this makes
a similar change there.  Here the bug is more likely to be hit because
the outer loop  (in lmb_alloc_nid) goes through the memory regions in
increasing order rather than decreasing order as __lmb_alloc_base
does, and we are therefore more likely to hit the case where we are
testing against a reserved region with a base address of 0.
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This makes no semantic changes.  It fixes the whitespace and formatting
a bit, gets rid of a local DBG macro and uses the equivalent pr_debug
instead, and restructures one while loop that had a function call and
assignment in the condition to be a bit more readable.  Some comments
about functions being called with relocation disabled were also removed
as they would just be confusing to most readers now that the code is
in lib/.
Signed-off-by: NPaul Mackerras <paulus@samba.org>

A variant of lmb_alloc() that tries to allocate memory on a specified
NUMA node 'nid' but falls back to normal lmb_alloc() if that fails.

The caller provides a 'nid_range' function pointer which assists the
allocator.  It is given args 'start', 'end', and pointer to integer
'this_nid'.

It places at 'this_nid' the NUMA node id that corresponds to 'start',
and returns the end address within 'start' to 'end' at which memory
assosciated with 'nid' ends.

This callback allows a platform to use lmb_alloc_nid() in just
about any context, even ones in which early_pfn_to_nid() might
not be working yet.

This function will be used by the NUMA setup code on sparc64, and also
it can be used by powerpc, replacing it's hand crafted
"careful_allocation()" function in arch/powerpc/mm/numa.c

If x86 ever converts it's NUMA support over to using the LMB helpers,
it can use this too as it has something entirely similar.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

We introduced a bug in fixing lmb_add_region to handle an initial
region being non-zero.  Before that fix it was impossible to insert a
region at the head of the list since the first region always started
at zero.

Now that its possible for the first region to be non-zero we need to
check to see if the new region should be added at the head and if so
actually add it.
Signed-off-by: NKumar Gala <galak@kernel.crashing.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Convert the lmb code to use u64 instead of unsigned long for physical
addresses and sizes.  This is needed to support large amounts of RAM
on 32-bit systems that support 36-bit physical addressing.
Signed-off-by: NBecky Bruce <becky.bruce@freescale.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If we add to an empty lmb region with a non-zero base we will not
coalesce the number of regions down to one.  This causes problems on
ppc32 for the memory region as its assumed to only have one region.

We can fix this be easily specially casing the initial add to just
replace the dummy region.
Signed-off-by: NKumar Gala <galak@kernel.crashing.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We need to check lmb_add_region() for errors, it can run out
of regions etc.

Also, the size needs to be padded to the given alignment
or else the lmb.reserved regions don't get expanded and
instead we get tons of holes and eventually run out of
regions prematurely.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There were several issues if a memreserve range existed and happened
to be in highmem:

* The bootmem allocator is only aware of lowmem so calling
  reserve_bootmem with a highmem address would cause a BUG_ON
* All highmem pages were provided to the buddy allocator

Added a lmb_is_reserved() api that we now use to determine if a highem
page should continue to be PageReserved or provided to the buddy
allocator.

Also, we incorrectly reported the amount of pages reserved since all
highmem pages are initally marked reserved and we clear the
PageReserved flag as we "free" up the highmem pages.
Signed-off-by: NKumar Gala <galak@kernel.crashing.org>

This fixes a possible infinite loop when the unsigned long counter "i"
is used in lmb_add_region() in the following for loop:

for (i = rgn->cnt-1; i >= 0; i--)

by making the loop counter "i" be signed.
Signed-off-by: NManish Ahuja <ahuja@austin.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

At present calling lmb_reserve() (and hence lmb_add_region()) twice
for exactly the same memory region will cause strange behaviour.

This makes life difficult when booting from a flat device tree with
memory reserve map.  Which regions are automatically reserved by the
kernel has changed over time, so it's quite possible a newer kernel
could attempt to auto-reserve a region which is also explicitly listed
in the device tree's reserve map, leading to trouble.

This patch avoids the problem by making lmb_reserve() ignore a call to
reserve a previously reserved region.  It also removes a now redundant
test designed to avoid one specific case of the problem noted above.

At present, this patch deals only with duplicate reservations of an
identical region.  Attempting to reserve two different, but
overlapping regions will still cause problems.  I might post another
patch later dealing with this case, but I'm avoiding it now since it
is substantially more complicated to deal with, less likely to occur
and more likely to indicate a genuine bug elsewhere if it does occur.
Signed-off-by: NDavid Gibson <dwg@au1.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

There's a bug in my cleaned up mem= handling, if the memory limit is
larger than the RMO size we'll erroneously enlarge the RMO size.

Fix is to only change the RMO size if the memory limit is less than
the current RMO value.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

There's a bug in my cleaned up mem= handling, if the memory limit is
larger than the RMO size we'll erroneously enlarge the RMO size.

Fix is to only change the RMO size if the memory limit is less than
the current RMO value.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Signed-off-by: NJörn Engel <joern@wohnheim.fh-wedel.de>
Signed-off-by: NAdrian Bunk <bunk@stusta.de>

We currently do mem= handling in three seperate places. And as benh pointed out
I wrote two of them. Now that we parse command line parameters earlier we can
clean this mess up.

Moving the parsing out of prom_init means the device tree might be allocated
above the memory limit. If that happens we'd have to move it. As it happens
we already have logic to do that for kdump, so just genericise it.

This also means we might have reserved regions above the memory limit, if we
do the bootmem allocator will blow up, so we have to modify
lmb_enforce_memory_limit() to truncate the reserves as well.

Tested on P5 LPAR, iSeries, F50, 44p. Tested moving device tree on P5 and
44p and F50.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

My patch (d7a5b2ff) to always panic if
lmb_alloc() fails is broken because it checks alloc < 0, but should be
checking alloc == 0.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

LMB_ALLOC_ANYWHERE doesn't need to be part of the API, it's only used in
lmb.c - so move it out of the header file.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Currently most callers of lmb_alloc() don't check if it worked or not, if it
ever does weird bad things will probably happen. The few callers who do check
just panic or BUG_ON.

So make lmb_alloc() panic internally, to catch bugs at the source. The few
callers who did check the result no longer need to.

The only caller that did anything interesting with the return result was
careful_allocation(). For it we create __lmb_alloc_base() which _doesn't_ panic
automatically, a little messy, but passable.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

To prevent problems later in boot, make sure we don't create zero-size lmb
regions.

I've checked all the callers, and at the moment no one should ever hit this.
All callers use a constant size, or they check the computed size before they
call us.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Somewhere we lost the include of udbg.h in lmb.c. While we're there, add a DBG
macro like every other file has and use it in lmb_dump_all().
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This also creates merged versions of do_init_bootmem, paging_init
and mem_init and moves them to arch/powerpc/mm/mem.c.  It gets rid
of the mem_pieces stuff.

I made memory_limit a parameter to lmb_enforce_memory_limit rather
than a global referenced by that function.  This will require some
small changes to ppc64 if we want to continue building ARCH=ppc64
using the merged lmb.c.
Signed-off-by: NPaul Mackerras <paulus@samba.org>

lmb_phys_mem_size() can always return lmb.memory.size, as long as it's called
after lmb_analyze(), which it is. There's no need to recalculate the size on
every call.

lmb_analyze() was calculating a few things we then threw away, so just don't
calculate them to start with.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

We no longer need the lmb code to know about abs and phys addresses, so
remove the physbase variable from the lmb_property struct.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

abs_to_phys() is a macro that turns out to do nothing, and also has the
unfortunate property that it's not the inverse of phys_to_abs() on iSeries.

The following is for my benefit as much as everyone else.

With CONFIG_MSCHUNKS enabled, the lmb code is changed such that it keeps
a physbase variable for each lmb region. This is used to take the possibly
discontiguous lmb regions and present them as a contiguous address space
beginning from zero.

In this context each lmb region's base address is its "absolute" base
address, and its physbase is it's "physical" address (from Linux's point of
view). The abs_to_phys() macro does the mapping from "absolute" to "physical".

Note: This is not related to the iSeries mapping of physical to absolute
(ie. Hypervisor) addresses which is maintained with the msChunks structure.
And the msChunks structure is not controlled via CONFIG_MSCHUNKS.

Once upon a time you could compile for non-iSeries with CONFIG_MSCHUNKS
enabled. But these days CONFIG_MSCHUNKS depends on CONFIG_PPC_ISERIES, so
for non-iSeries code abs_to_phys() is a no-op.

On iSeries we always have one lmb region which spans from 0 to
systemcfg->physicalMemorySize (arch/ppc64/kernel/iSeries_setup.c line 383).
This region has a base (ie. absolute) address of 0, and a physbase address
of 0 (as calculated in lmb_analyze() (arch/ppc64/kernel/lmb.c line 144)).

On iSeries, abs_to_phys(aa) is defined as lmb_abs_to_phys(aa), which finds
the lmb region containing aa (and there's only one, ie. 0), and then does:

 return lmb.memory.region[0].physbase + (aa - lmb.memory.region[0].base)

physbase == base == 0, so you're left with "return aa".

So remove abs_to_phys(), and lmb_abs_to_phys() which is the implementation
of abs_to_phys() for iSeries.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

The lmb code is all written to use a pointer to an lmb struct. But it's always
the same lmb struct, called "lmb". So we take the address of lmb, call it
_lmb and then start using _lmb->foo everywhere, which is silly.

This patch removes the _lmb pointers and replaces them with direct references
to the one "lmb" struct. We do the same for some _mem and _rsv pointers which
point to lmb.memory and lmb.reserved respectively.

This patch looks quite busy, but it's basically just:
s/_lmb->/lmb./g
s/_mem->/lmb.memory./g
s/_rsv->/lmb.reserved./g
s/_rsv/&lmb.reserved/g
s/mem->/lmb.memory./g
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!