Since ARMv5, the clz instruction allows for efficient implementation of
ffs/fls with builtins. Until ARMv7 (with Thumb-2), this instruction is
only available in ARM mode. LTO makes it difficult to force specific
functions to be in ARM mode, as it is effectively a form of very
aggressive inlining. To work around this, fls/ffs are implemented in
assembly for ARMv5 and ARMv6 when compiling U-Boot in Thumb mode.
Overall, this saves around 75 bytes per call.
This code is synced with v5.15 of the Linux kernel.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Current implementation of find_next_zero_bit() is incompatible with arm64.
Hence fix it by using BITS_PER_LONG define instead of constants and
use generic ffz() implementation.
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
find_next_zero_bit() incorrectly handles cases when:
- total bitmap size < 32
- rest of bits to process
static inline int find_next_zero_bit(void *addr, int size, int offset)
{
unsigned long *p = ((unsigned long *)addr) + (offset >> 5);
unsigned long result = offset & ~31UL;
unsigned long tmp;
if (offset >= size)
return size;
size -= result;
offset &= 31UL;
if (offset) {
tmp = *(p++);
tmp |= ~0UL >> (32-offset);
if (size < 32)
[1]
goto found_first;
if (~tmp)
goto found_middle;
size -= 32;
result += 32;
}
while (size & ~31UL) {
tmp = *(p++);
if (~tmp)
goto found_middle;
result += 32;
size -= 32;
}
[2]
if (!size)
return result;
tmp = *p;
found_first:
[3] tmp |= ~0UL >> size;
^^^ algo can reach above line from from points:
[1] offset > 0 and size < 32, tmp[offset-1..0] bits set to 1
[2] size < 32 - rest of bits to process
in both cases bits to search are tmp[size-1..0], but line [3] will simply
set all tmp[31-size..0] bits to 1 and ffz(tmp) below will fail.
example: bitmap size = 16, offset = 0, bitmap is empty.
code will go through the point [2], tmp = 0x0
after line [3] => tmp = 0xFFFF and ffz(tmp) will return 16.
found_middle:
return result + ffz(tmp);
}
Fix it by correctly seting tmp[31..size] bits to 1 in the above case [3].
Fixes: 81e9fe5a29 ("arm: implement find_next_zero_bit function")
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
compiling U-Boot for openrd_base_defconfig with
gcc 5.x shows the following warning:
CC fs/ubifs/super.o
In file included from fs/ubifs/ubifs.h:35:0,
from fs/ubifs/super.c:37:
fs/ubifs/super.c: In function 'atomic_inc':
./arch/arm/include/asm/atomic.h:55:2: warning: 'flags' is used uninitialized in this function [-Wuninitialized]
local_irq_save(flags);
^
fs/ubifs/super.c: In function 'atomic_dec':
./arch/arm/include/asm/atomic.h:64:2: warning: 'flags' is used uninitialized in this function [-Wuninitialized]
local_irq_save(flags);
^
CC fs/ubifs/sb.o
[...]
CC fs/ubifs/lpt.o
In file included from include/linux/bitops.h:123:0,
from include/common.h:20,
from include/ubi_uboot.h:17,
from fs/ubifs/ubifs.h:37,
from fs/ubifs/lpt.c:35:
fs/ubifs/lpt.c: In function 'test_and_set_bit':
./arch/arm/include/asm/bitops.h:57:2: warning: 'flags' is used uninitialized in this function [-Wuninitialized]
local_irq_save(flags);
^
CC fs/ubifs/lpt_commit.o
In file included from include/linux/bitops.h:123:0,
from include/common.h:20,
from include/ubi_uboot.h:17,
from fs/ubifs/ubifs.h:37,
from fs/ubifs/lpt_commit.c:26:
fs/ubifs/lpt_commit.c: In function 'test_and_set_bit':
./arch/arm/include/asm/bitops.h:57:2: warning: 'flags' is used uninitialized in this function [-Wuninitialized]
local_irq_save(flags);
^
CC fs/ubifs/scan.o
CC fs/ubifs/lprops.o
CC fs/ubifs/tnc.o
In file included from include/linux/bitops.h:123:0,
from include/common.h:20,
from include/ubi_uboot.h:17,
from fs/ubifs/ubifs.h:37,
from fs/ubifs/tnc.c:30:
fs/ubifs/tnc.c: In function 'test_and_set_bit':
./arch/arm/include/asm/bitops.h:57:2: warning: 'flags' is used uninitialized in this function [-Wuninitialized]
local_irq_save(flags);
^
CC fs/ubifs/tnc_misc.o
Fix it.
Signed-off-by: Heiko Schocher <hs@denx.de>
The generic bitops headers are required when calling logarithmic
functions, such as ilog2().
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Heiko Schocher <hs@denx.de>
Reviewed-by: Jagan Teki <jteki@openedev.com>
This commit copies implementation of the find_next_zero_bit() from
git://git.denx.de/u-boot.git/arch/mips/include/asm/bitops.h. v2014.07
The function is required to enable MCAST_TFTP support for ARM platforms.
Signed-off-by: Vitaly Andrianov <vitalya@ti.com>
mkconfig links ${objtree}/include/asm/proc/ to
${srctree}/arch/${arch}/include/asm/proc-armv/. This seems to be a
remnant from the past. Ever since its introduction in 2003 it is used
only in ARM build and always links to same place, so let's simplify
the code, remove it and reference directly where needed.
Successful MAKEALL for ARM and PowerPC verified on Linux.
Signed-off-by: Vasili Galka <vvv444@gmail.com>
This helps to clean up the include/ directory so that it only contains
non-architecture-specific headers and also matches Linux's directory
layout which many U-Boot developers are already familiar with.
Signed-off-by: Peter Tyser <ptyser@xes-inc.com>
