Some of those can be written in such a way that the same
inline assembly can be used to generate both 32 bit and
64 bit code.
For ffs and fls, x86_64 unconditionally used the cmov
instruction and i386 unconditionally used a conditional
branch over a mov instruction. In the current patch I
chose to select the version based on the availability
of the cmov instruction instead. A small detail here is
that x86_64 did not previously set CONFIG_X86_CMOV=y.
Improved comments for ffs, ffz, fls and variations.
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This moves an optimization for searching constant-sized small
bitmaps form x86_64-specific to generic code.
On an i386 defconfig (the x86#testing one), the size of vmlinux hardly
changes with this applied. I have observed only four places where this
optimization avoids a call into find_next_bit:
In the functions return_unused_surplus_pages, alloc_fresh_huge_page,
and adjust_pool_surplus, this patch avoids a call for a 1-bit bitmap.
In __next_cpu a call is avoided for a 32-bit bitmap. That's it.
On x86_64, 52 locations are optimized with a minimal increase in
code size:
Current #testing defconfig:
146 x bsf, 27 x find_next_*bit
text data bss dec hex filename
5392637 846592 724424 6963653 6a41c5 vmlinux
After removing the x86_64 specific optimization for find_next_*bit:
94 x bsf, 79 x find_next_*bit
text data bss dec hex filename
5392358 846592 724424 6963374 6a40ae vmlinux
After this patch (making the optimization generic):
146 x bsf, 27 x find_next_*bit
text data bss dec hex filename
5392396 846592 724424 6963412 6a40d4 vmlinux
[ tglx@linutronix.de: build fixes ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The versions with inline assembly are in fact slower on the machines I
tested them on (in userspace) (Athlon XP 2800+, p4-like Xeon 2.8GHz, AMD
Opteron 270). The i386-version needed a fix similar to 06024f21 to avoid
crashing the benchmark.
Benchmark using: gcc -fomit-frame-pointer -Os. For each bitmap size
1...512, for each possible bitmap with one bit set, for each possible
offset: find the position of the first bit starting at offset. If you
follow ;). Times include setup of the bitmap and checking of the
results.
Athlon Xeon Opteron 32/64bit
x86-specific: 0m3.692s 0m2.820s 0m3.196s / 0m2.480s
generic: 0m2.622s 0m1.662s 0m2.100s / 0m1.572s
If the bitmap size is not a multiple of BITS_PER_LONG, and no set
(cleared) bit is found, find_next_bit (find_next_zero_bit) returns a
value outside of the range [0, size]. The generic version always returns
exactly size. The generic version also uses unsigned long everywhere,
while the x86 versions use a mishmash of int, unsigned (int), long and
unsigned long.
Using the generic version does give a slightly bigger kernel, though.
defconfig: text data bss dec hex filename
x86-specific: 4738555 481232 626688 5846475 5935cb vmlinux (32 bit)
generic: 4738621 481232 626688 5846541 59360d vmlinux (32 bit)
x86-specific: 5392395 846568 724424 6963387 6a40bb vmlinux (64 bit)
generic: 5392458 846568 724424 6963450 6a40fa vmlinux (64 bit)
Signed-off-by: Alexander van Heukelum <heukelum@fastmail.fm>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This (simplified) piece of code didn't behave as expected due to
incorrect constraints in some of the bitops functions, when
X86_FEATURE_xxx is referring to other than the first long:
int test(struct cpuinfo_x86 *c) {
if (cpu_has(c, X86_FEATURE_xxx))
clear_cpu_cap(c, X86_FEATURE_xxx);
return cpu_has(c, X86_FEATURE_xxx);
}
I'd really like understand, though, what the policy of (not) having a
"memory" clobber in these operations is - currently, this appears to
be totally inconsistent. Also, many comments of the non-atomic
functions say those may also be re-ordered - this contradicts the use
of "asm volatile" in there, which again I'd like to understand.
As much as all of these, using 'int' for the 'nr' parameter and
'void *' for the 'addr' one is in conflict with
Documentation/atomic_ops.txt, especially because bt{,c,r,s} indeed
take the bit index as signed (which hence would really need special
precaution) and access the full 32 bits (if 'unsigned long' was used
properly here, 64 bits for x86-64) pointed at, so invalid uses like
referencing a 'char' array cannot currently be caught.
Finally, the code with and without this patch relies heavily on the
-fno-strict-aliasing compiler switch and I'm not certain this really
is a good idea.
In the light of all of this I'm sending this as RFC, as fixing the
above might warrant a much bigger patch...
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch changes the bitwise operations in bitops.h to get
a void pointers as a parameter. Before this patch, a lot of warnings
can be seen. They're gone after it.
Signed-off-by: Glauber de Oliveira Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
This unifies the set/clear/test bit functions of asm/bitops.h.
I have not attempted to merge the bit-finding functions, since they
rely on the machine word size and can't be easily restructured to work
generically without a lot of #ifdefs. In particular, the 64-bit code
can assume the presence of conditional move instructions, whereas
32-bit needs to be more careful.
The inline assembly for the bit operations has been changed to remove
explicit sizing hints on the instructions, so the assembler will pick
the appropriate instruction forms depending on the architecture and
the context.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Move the headers to include/asm-x86 and fixup the
header install make rules
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
