Memory banks living outside of the 32-bit physical address
space do not have a 1:1 pa <-> va mapping and therefore the
__va macro may wrap.
This patch ensures that such banks are marked as highmem so
that the Kernel doesn't try to split them up when it sees that
the wrapped virtual address overlaps the vmalloc space.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Nicolas Pitre <nico@linaro.org>
This patch adds the MMU initialisation for the LPAE page table format.
The swapper_pg_dir size with LPAE is 5 rather than 4 pages. A new
proc-v7-3level.S file contains the TTB initialisation, context switch
and PTE setting code with the LPAE. The TTBRx split is based on the
PAGE_OFFSET with TTBR1 used for the kernel mappings. The 36-bit mappings
(supersections) and a few other memory types in mmu.c are conditionally
compiled.
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Now that we have all the static mappings from iotable_init() located
in the vmalloc area, it is trivial to optimize ioremap by reusing those
static mappings when the requested physical area fits in one of them,
and so in a generic way for all platforms.
Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Tested-by: Stephen Warren <swarren@nvidia.com>
Tested-by: Kevin Hilman <khilman@ti.com>
Tested-by: Jamie Iles <jamie@jamieiles.com>
In order to remove the build time variation between different SOCs with
regards to VMALLOC_END, the iotable mappings are now allocated inside
the vmalloc region. This allows for VMALLOC_END to be identical across
all machines.
The value for VMALLOC_END is now set to 0xff000000 which is right where
the consistent DMA area starts.
To accommodate all static mappings on machines with possible highmem usage,
the default vmalloc area size is changed to 240 MB so that VMALLOC_START
is no higher than 0xf0000000 by default.
Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Tested-by: Stephen Warren <swarren@nvidia.com>
Tested-by: Kevin Hilman <khilman@ti.com>
Tested-by: Jamie Iles <jamie@jamieiles.com>
Some upcoming changes must know the VMALLOC_START value, which is based
on high_memory, before bootmem_init() is called.
The best location to set it is in sanity_check_meminfo() where the needed
computation is already done, and in the non MMU case it is trivial to do
now that the meminfo array is already sorted at that point.
Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This patch defines the (pte|pmd)val_t as u32 and changes the page table
types to be based on these. The PMD bits are converted to the
corresponding type using the _AT macro.
The flush_pmd_entry/clean_pmd_entry argument was changed to (void *) to
allow them to be used with both PGD and PMD pointers and avoid code
duplication.
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
On certain architectures, there might be a need to mark certain
addresses with strongly ordered memory attributes to avoid ordering
issues at the interconnect level.
On OMAP4, the asynchronous bridge buffers can only be drained
with strongly ordered accesses and hence the need to mark the
memory strongly ordered.
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Signed-off-by: Woodruff Richard <r-woodruff2@ti.com>
Tested-by: Vishwanath BS <vishwanath.bs@ti.com>
PGDIR_SHIFT and PMD_SHIFT for the classic 2-level page table format have
the same value (21). This patch converts the PGDIR_* uses in the kernel
to the PMD_* equivalent so that LPAE builds can reuse the same code.
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Ensure that the meminfo array is sanity checked before we pass the
memory to memblock. This helps to ensure that memblock and meminfo
agree on the dimensions of memory, especially when more memory is
passed than the kernel can deal with.
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
sanity_check_meminfo walks over the registered memory banks and attempts
to split banks across lowmem and highmem when they would otherwise
overlap with the vmalloc space.
When SPARSEMEM is used, there are two potential problems that occur
when the virtual address of the start of a bank is equal to vmalloc_min.
1.) The end of lowmem is calculated as __pa(vmalloc_min - 1) + 1.
In the above scenario, this will give the end address of the
previous bank, rather than the actual bank we are interested in.
This value is later used as the memblock limit and artificially
restricts the total amount of available memory.
2.) The checks to determine whether or not a bank belongs to highmem
or not only check if __va(bank->start) is greater or less than
vmalloc_min. In the case that it is equal, the bank is incorrectly
treated as lowmem, which hoses the vmalloc area.
This patch fixes these two problems by checking whether the virtual
start address of a bank is >= vmalloc_min and then calculating
lowmem_end by finding the virtual end address of the highest lowmem
bank.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Fold all the mmu_gather rework patches into one for submission
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Reported-by: Hugh Dickins <hughd@google.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Miller <davem@davemloft.net>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Tony Luck <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In commit e616c59140, highmem support was
deactivated for SMP platforms without hardware TLB ops broadcast because
usage of kmap_high_get() requires that IRQs be disabled when kmap_lock
is locked which is incompatible with the IPI mechanism used by the
software TLB ops broadcast invoked through flush_all_zero_pkmaps().
The reason for kmap_high_get() is to ensure that the currently kmap'd
page usage count does not decrease to zero while we're using its
existing virtual mapping in an atomic context. With a VIVT cache this
is essential to do due to cache coherency issues, but with a VIPT cache
this is only an optimization so not to pay the price of establishing a
second mapping if an existing one can be used. However, on VIPT
platforms without hardware TLB maintenance we can give up on that
optimization in order to be able to use highmem.
From ARMv7 onwards the TLB ops are broadcasted in hardware, so let's
disable ARCH_NEEDS_KMAP_HIGH_GET only when CONFIG_SMP and
CONFIG_CPU_TLB_V6 are defined.
Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Tested-by: Saeed Bishara <saeed.bishara@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Add pud_offset() et.al. between the pgd and pmd code in preparation of
using pgtable-nopud.h rather than 4level-fixup.h.
This incorporates a fix from Jamie Iles <jamie@jamieiles.com> for
uaccess_with_memcpy.c.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The unsigned long datatype is not sufficient for mapping physical addresses
>= 4GB.
This patch ensures that the phys_addr_t datatype is used to represent physical
addresses when converting from a PFN.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
For the Kernel to support 2 level and 3 level page tables, physical
addresses (and also page table entries) need to be 32 or 64-bits depending
upon the configuration.
This patch uses the %08llx conversion specifier for physical addresses
and page table entries, ensuring that they are cast to (long long) so
that common code can be used regardless of the datatype widths.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
With LPAE we no longer have software bits in a separate Linux PTE and
the early_pte_alloc() function should pass PTE_HWTABLE_OFF +
PTE_HWTABLE_SIZE to early_alloc() to avoid allocating extra memory.
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The hardware page tables use an XN bit 'execute never'. Historically,
we've had a Linux 'execute allow' bit, in the positive sense. Get rid
of this artifact as future hardware will continue to have the XN sense.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
We have two places where we create identity mappings - one when we bring
secondary CPUs online, and one where we setup some mappings for soft-
reboot. Combine these two into a single implementation. Also collect
the identity mapping deletion function.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This patch removes the domain switching functionality via the set_fs and
__switch_to functions on cores that have a TLS register.
Currently, the ioremap and vmalloc areas share the same level 1 page
tables and therefore have the same domain (DOMAIN_KERNEL). When the
kernel domain is modified from Client to Manager (via the __set_fs or in
the __switch_to function), the XN (eXecute Never) bit is overridden and
newer CPUs can speculatively prefetch the ioremap'ed memory.
Linux performs the kernel domain switching to allow user-specific
functions (copy_to/from_user, get/put_user etc.) to access kernel
memory. In order for these functions to work with the kernel domain set
to Client, the patch modifies the LDRT/STRT and related instructions to
the LDR/STR ones.
The user pages access rights are also modified for kernel read-only
access rather than read/write so that the copy-on-write mechanism still
works. CPU_USE_DOMAINS gets disabled only if the hardware has a TLS register
(CPU_32v6K is defined) since writing the TLS value to the high vectors page
isn't possible.
The user addresses passed to the kernel are checked by the access_ok()
function so that they do not point to the kernel space.
Tested-by: Anton Vorontsov <cbouatmailru@gmail.com>
Cc: Tony Lindgren <tony@atomide.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Use memblock information to setup lowmem mappings rather than the
membank array.
This allows platforms to manipulate the memblock information during
initialization to reserve (and remove) memory from the kernel's view
of memory - and thus allowing platforms to setup their own private
mappings for this memory without causing problems with multiple
aliasing mappings:
size = min(size, SZ_2M);
base = memblock_alloc(size, min(align, SZ_2M));
memblock_free(base, size);
memblock_remove(base, size);
This is needed because multiple mappings of regions with differing
attributes (sharability, type, cache) are not permitted with ARMv6
and above.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This was missing from the noMMU code, so there was the possibility
of things not working as expected if out of order memory information
was passed.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Will says:
| Commit e63075a3 removed the explicit MEMBLOCK_REAL_LIMIT #define
| and introduced the requirement that arch code calls
| memblock_set_current_limit to ensure that the __va macro can
| be used on physical addresses returned from memblock_alloc.
Unfortunately, ARM was missed out of this change. Fix this.
Reported-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
After Santosh's fixup of the generic MT_MEMORY and
MT_MEMORY_NONCACHED I add this fix to the TCM memory types.
The main change is that the ITCM memory is L_PTE_WRITE and
DOMAIN_KERNEL which works just fine. The changed to the DTCM
is just cosmetic to fit with surrounding code.
Cc: Santosh Shilimkar <santosh.shilimkar@ti.com>
Cc: Rickard Andersson <rickard.andersson@stericsson.com>
Signed-off-by: Linus Walleij <linus.walleij@stericsson.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
UP systems do not implement all the instructions that SMP systems have,
so in order to boot a SMP kernel on a UP system, we need to rewrite
parts of the kernel.
Do this using an 'alternatives' scheme, where the kernel code and data
is modified prior to initialization to replace the SMP instructions,
thereby rendering the problematical code ineffectual. We use the linker
to generate a list of 32-bit word locations and their replacement values,
and run through these replacements when we detect a UP system.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The commit f1a2481c0 sets up the default flags for MT_MEMORY and
MT_MEMORY_NONCACHED memory types. L_PTE_USER flag is wrongly
set as default for these entries so remove it. Also adding
the 'L_PTE_WRITE' flag so that these pages become read-write
instead of just being read-only
[this stops them being exposed to userspace, which is the main
concern here --rmk]
Reported-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This patch populates the L1 entries for MT_MEMORY and MT_MEMORY_NONCACHED
types so that at boot-up, we can map memories outside system memory
at page level granularity
Previously the mapping was limiting to section level, which creates
unnecessary additional mapping for which physical memory may not
present. On the newer ARM with speculation, this is dangerous and can
result in untraceable aborts.
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
ARMv7 onwards requires that there are no aliases to the same physical
location using different memory types (i.e. Normal vs Strongly Ordered).
Access to SO mappings when the unaligned accesses are handled in
hardware is also Unpredictable (pgprot_noncached() mappings in user
space).
The /dev/mem driver requires uncached mappings with O_SYNC. The patch
implements the phys_mem_access_prot() function which generates Strongly
Ordered memory attributes if !pfn_valid() (independent of O_SYNC) and
Normal Noncacheable (writecombine) if O_SYNC.
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The earlier TCM memory regions were mapped as MT_MEMORY_UNCACHED
which doesn't really work on platforms supporting the new v6
features like the NX bit. Add unique MT_MEMORY_[I|D]TCM types
instead.
Cc: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Linus Walleij <linus.walleij@stericsson.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Add a common early allocator function, in preparation for switching
over to LMB. When we do, this function will need to do a little more
than just allocating memory; we need it zero initialized too.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Move the platform specific bootmem memory reservations out of
arch/arm/mm/mmu.c into their respective platform files.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Everything should now be using sparsemem rather than discontigmem, so
remove the code supporting discontigmem from ARM.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Rather than storing the minimum size of the vmalloc area, store the
maximum permitted address of the vmalloc area instead.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
