When compiling a MMU kernel with CPU_HAS_ADDRESS_SPACES=n (e.g. "MMU=y
allnoconfig": "echo CONFIG_MMU=y > allno.config && make KCONFIG_ALLCONFIG=1
allnoconfig"), we use plain "move" instead of "moves", and I got:
CC arch/m68k/lib/uaccess.o
{standard input}: Assembler messages:
{standard input}:47: Error: operands mismatch -- statement `move.b %a0,(%a1)' ignored
This happens because plain "move" doesn't support byte transfers between
memory and address registers, while "moves" does.
Fix the asm constraints for __generic_copy_from_user(),
__generic_copy_to_user(), and __clear_user() to only use data registers
when accessing userspace.
Also, relax the asm constraints for 16-bit userspace accesses in
__put_user() and __get_user(), as both "move" and "moves" do support
such transfers between memory and address registers.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Gcc may replace calls to standard string functions by open code and/or
calls to other standard string functions. If the replacement function is
not available out-of-line, link errors will happen.
To avoid this, the out-of-line versions were provided by
arch/m68k/lib/string.c, but they were usually not linked in anymore as
typically none of its symbols are referenced by built-in code.
However, if any module would need them, they would not be available.
Hence remove the inline strcpy() and strcat() implementations, remove
arch/m68k/lib/string.c, and let the generic string library code handle it.
Impact on a typical kernel build seems minimal or nonexistent:
- .text : 0x00001000 - 0x002aac74 (2728 KiB)
- .data : 0x002ada48 - 0x00392148 ( 914 KiB)
+ .text : 0x00001000 - 0x002aacf4 (2728 KiB)
+ .data : 0x002adac8 - 0x00392148 ( 914 KiB)
See also commit e00c73ee05 ("m68k: Remove
inline strlen() implementation").
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
GCC can replace a strncat() call with constant second argument into a
strlen + store, which results in a link error:
ERROR: "strlen" [net/ipv4/ip_tunnel.ko] undefined!
The inline function is a simple for loop in C. Other architectures
either use an asm optimized variant, or use the generic function from
lib/string.c.
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Michal Marek <mmarek@suse.cz>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
When compiling for original 68000 or ColdFire targets you will get the
following warning when compiling arch/m68k/lib/memcpy.c:
CC arch/m68k/lib/memcpy.o
arch/m68k/lib/memcpy.c: In function ‘__builtin_memcpy’:
arch/m68k/lib/memcpy.c:13:15: warning: unused variable ‘temp1’
This is easily fixed by moving the definition of temp1 into the code block
where it is used.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Modify the user space access functions to support the ColdFire V4e cores
running with MMU enabled.
The ColdFire processors do not support the "moves" instruction used by
the traditional 680x0 processors for moving data into and out of another
address space. They only support the notion of a single address space,
and you use the usual "move" instruction to access that.
Create a new config symbol (CONFIG_CPU_HAS_ADDRESS_SPACES) to mark the
CPU types that support separate address spaces, and thus also support
the sfc/dfc registers and the "moves" instruction that go along with that.
The code is almost identical for user space access, so lets just use a
define to choose either the "move" or "moves" in the assembler code.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Matt Waddel <mwaddel@yahoo.com>
Acked-by: Kurt Mahan <kmahan@xmission.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
The traditional 68000 processors and the newer reduced instruction set
ColdFire processors do not support the 32*32->64 multiply or the 64/32->32
divide instructions. This is not a difference based on the presence of
a hardware MMU or not.
Create a new config symbol to mark that a CPU type doesn't support the
longer multiply/divide instructions. Use this then as a basis for using
the fast 64bit based divide (in div64.h) and for linking in the extra
libgcc functions that may be required (mulsi3, divsi3, etc).
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
We have two implementations of the IP checksuming code for the m68k arch.
One uses the more advanced instructions available in 68020 and above
processors, the other uses the simpler instructions available on the
original 68000 processors and the modern ColdFire processors.
This simpler code is pretty much the same as the generic lib implementation
of the IP csum functions. So lets just switch over to using that. That
means we can completely remove the checksum_no.c file, and only have the
local fast code used for the more complex 68k CPU family members.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
The original 68000 processors cannot copy 16bit or larger quantities from
odd addresses. All newer members of the 68k family (including ColdFire)
can do this.
In the current memcpy implementation after trying to align the destination
address to a 16bit boundary if we end up with an odd source address we go
off and try to copy multi-byte quantities from it. This will trap on the
68000.
The only solution if we end with an odd source address is to byte wise
copy the whole memcpy region. We only need to do this if we are supporting
original 68000 processors.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
The real difference between the mmu and non-mmu varients of the delay.h
files has nothing to do with having an mmu or not. It is processor family
differences that means slightly different code. Merge the delay_mm.h and
delay_no.h files back into a single file.
The primarly difference we need to deal with is whether the processor
supports a 32bit * 32bit -> 64bit multiply. Without it we need to do some
shift scaling as well as use a 32bit * 32bit -> 32bit multiply. If building
for a multi-CPU type kernel then we must use the simpler mult/shift scaling.
This version of delay code allows the CPU32 family to use a 64bit mul,
since it supports this instruction, the old code did not.
The changes use macros where appropriate to try and optimize constant sized
udelay times. And it removes the use of a fixed lib function for the non-mmu
case. Code size on typical kernel configurations is similar, or only larger
by a few tens of bytes.
Also removed the unused muldiv() code from delay_mm.h.
Build and run tested on ColdFire and ARAnyM. Build tested only on 68328
and 68360 (CPU32).
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Older m68k-linux compilers will include pre-defined symbols that
confuse what processor it is being targeted for. For example gcc-4.1.2
will pre-define __mc68020__ even if you specify the target processor
as -m68000 on the gcc command line. Newer versions of gcc have this
corrected.
In a few places the m68k code uses defined(__mc68020__) for optimizations
that include instructions that are specific to the CPU 68020 and above.
When compiling with older compilers this will be true even when we have
selected to compile for the older 68000 processors.
Switch to using the kernel processor defines, CONFIG_M68020 and friends.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
The EXPORT_SYMBOL() of the local lib checksum functions belongs with
the definitions, not in some other random code file. So move then there.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
We don't need an arch/m68k/lib/checksum.c wrapper to include the correct
mmu or non-mmu version of the checksum code. Let the Makefile just build
the appropriate one.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Merging the mmu and non-mmu directories we ended up with duplicate
implementations of memcpy(). One is a little more optimized for the
>= 68020 case, but that can easily be inserted into a single
implementation of memcpy(). Clean up the exporting of this symbol
too, otherwise we end up exporting it twice on a no-mmu build.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Merging the mmu and non-mmu directories we ended up with duplicate
implementations of memset(). One is a little more optimized for the
>= 68020 case, but that can easily be inserted into a single
implementation of memset(). Clean up the exporting of this symbol
too, otherwise we end up exporting it twice on a no-mmu build.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Merging the mmu and non-mmu directories we ended up with duplicate
(and identical) implementations of memmove(). Remove one of them.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
We can easily support the slight differences in libs needed by the
mmu and non-mmu builds in a single Makefile, so merge them back into
a single file again.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
The implementation of gcc's muldi3 support function differs only in
the use of the machine's 64 bit sized mul or not. (It isn't based
on using an MMU or not). Merge the current mmu and non-mmu versions
of arc/m68k/lib/muldi3 and use the appropriate pre-processor
conditionals to get the right version for all m68k processor types.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
There is a lot of common code that could be shared between the m68k
and m68knommu arch branches. It makes sense to merge the two branches
into a single directory structure so that we can more easily share
that common code.
This is a brute force merge, based on a script from Stephen King
<sfking@fdwdc.com>, which was originally written by Arnd Bergmann
<arnd@arndb.de>.
> The script was inspired by the script Sam Ravnborg used to merge the
> includes from m68knommu. For those files common to both arches but
> differing in content, the m68k version of the file is renamed to
> <file>_mm.<ext> and the m68knommu version of the file is moved into the
> corresponding m68k directory and renamed <file>_no.<ext> and a small
> wrapper file <file>.<ext> is used to select between the two version. Files
> that are common to both but don't differ are removed from the m68knommu
> tree and files and directories that are unique to the m68knommu tree are
> moved to the m68k tree. Finally, the arch/m68knommu tree is removed.
>
> To select between the the versions of the files, the wrapper uses
>
> #ifdef CONFIG_MMU
> #include <file>_mm.<ext>
> #else
> #include <file>_no.<ext>
> #endif
On top of this file merge I have done a simplistic merge of m68k and
m68knommu Kconfig, which primarily attempts to keep existing options and
menus in place. Other than a handful of options being moved it produces
identical .config outputs on m68k and m68knommu targets I tested it on.
With this in place there is now quite a bit of scope for merge cleanups
in future patches.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
The m68k arch implements its own memcmp() function. It is not optimized
in any way (it is the most strait forward coding of memcmp you can get).
Remove it and use the kernels standard memcmp() implementation.
This also goes part of the way to fixing a regression caused by commit
ea61bc461d ("m68k/m68knommu: merge MMU and
non-MMU string.h"), which breaks non-coldfire non-mmu builds (which is
the 68x328 and 68360 families). They currently have no memcmp() function
defined, since there is none in the m68knommu/lib functions.
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
> This is a no-no for those archs that still use -traditional.
> > I dunno if this is a problem for you at the moment and the
> > right fix is anyway to nuke -traditional.
> >
> > Sam
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Whether we sidestep it in init/main.c or not, such situations
will arise again; compiler does generate calls of strcat()
on optimizations, so we really ought to have an out-of-line
version...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Semaphores are no longer performance-critical, so a generic C
implementation is better for maintainability, debuggability and
extensibility. Thanks to Peter Zijlstra for fixing the lockdep
warning. Thanks to Harvey Harrison for pointing out that the
unlikely() was unnecessary.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
m68k: implement __clear_user(), which is needed by fs/signalfd.c
Since we always let the MMU do all checking, clear_user() and __clear_user()
are identical. The old clear_user() is renamed to __clear_user() for
consistency.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Jump to the correct exit label after exception
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This uninlines a few large functions in uaccess.h and cleans up the rest.
It includes a (hopefully temporary) workaround for the broken typeof of
gcc-4.1.
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Use the builtin functions for memset/memclr/memcpy, special optimizations for
page operations have dedicated functions now. Uninline memmove/memchr and
move all functions into a single file and clean it up a little.
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.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!
Geert Uytterhoeven