In the "xchg" implementation, %ebx and %ecx don't need to be copied
into %eax and %edx respectively (this is only necessary when desiring
to only read the stored value).
In the "add_unless" implementation, swapping the use of %ecx and %esi
for passing arguments allows %esi to become an input only (i.e.
permitting the register to be re-used to address the same object
without reload).
In "{add,sub}_return", doing the initial read64 through the passed in
%ecx decreases a register dependency.
In "inc_not_zero", a branch can be eliminated by or-ing together the
two halves of the current (64-bit) value, and code size can be further
reduced by adjusting the arithmetic slightly.
v2: Undo the folding of "xchg" and "set".
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Link: http://lkml.kernel.org/r/4F19A2BC020000780006E0DC@nat28.tlf.novell.com
Cc: Luca Barbieri <luca@luca-barbieri.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Eric pointed out overly restrictive constraints in atomic64_set(), but
there are issues throughout the file. In the cited case, %ebx and %ecx
are inputs only (don't get changed by either of the two low level
implementations). This was also the case elsewhere.
Further in many cases early-clobber indicators were missing.
Finally, the previous implementation rolled a custom alternative
instruction macro from scratch, rather than using alternative_call()
(which was introduced with the commit that the description of the
change in question actually refers to). Adjusting has the benefit of
not hiding referenced symbols from the compiler, which however requires
them to be declared not just in the exporting source file (which, as a
desirable side effect, in turn allows that exporting file to become a
real 5-line stub).
This patch does not eliminate the overly restrictive memory clobbers,
however: Doing so would occasionally make the compiler set up a second
register for accessing the memory object (to satisfy the added "m"
constraint), and it's not clear which of the two non-optimal
alternatives is better.
v2: Re-do the declaration and exporting of the internal symbols.
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Link: http://lkml.kernel.org/r/4F19A2A5020000780006E0D9@nat28.tlf.novell.com
Cc: Luca Barbieri <luca@luca-barbieri.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
atomic64_read() doesn't actually write anything (as far as the C
environment is concerned... the CPU does actually write but that's an
implementation quirk), so it should take a const pointer.
This does NOT mean that it is safe to use atomic64_read() on an object
in readonly storage (it will trap!)
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Link: http://lkml.kernel.org/r/20120109165859.1879abda.akpm@linux-foundation.org
This is in preparation for more generic atomic primitives based on
__atomic_add_unless.
Signed-off-by: Arun Sharma <asharma@fb.com>
Signed-off-by: Hans-Christian Egtvedt <hans-christian.egtvedt@atmel.com>
Reviewed-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: David Miller <davem@davemloft.net>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch replaces atomic64_32.c with two assembly implementations,
one for 386/486 machines using pushf/cli/popf and one for 586+ machines
using cmpxchg8b.
The cmpxchg8b implementation provides the following advantages over the
current one:
1. Implements atomic64_add_unless, atomic64_dec_if_positive and
atomic64_inc_not_zero
2. Uses the ZF flag changed by cmpxchg8b instead of doing a comparison
3. Uses custom register calling conventions that reduce or eliminate
register moves to suit cmpxchg8b
4. Reads the initial value instead of using cmpxchg8b to do that.
Currently we use lock xaddl and movl, which seems the fastest.
5. Does not use the lock prefix for atomic64_set
64-bit writes are already atomic, so we don't need that.
We still need it for atomic64_read to avoid restoring a value
changed in the meantime.
6. Allocates registers as well or better than gcc
The 386 implementation provides support for 386 and 486 machines.
386/486 SMP is not supported (we dropped it), but such support can be
added easily if desired.
A pure assembly implementation is required due to the custom calling
conventions, and desire to use %ebp in atomic64_add_return (we need
7 registers...), as well as the ability to use pushf/popf in the 386
code without an intermediate pop/push.
The parameter names are changed to match the convention in atomic_64.h
Changes in v3 (due to rebasing to tip/x86/asm):
- Patches atomic64_32.h instead of atomic_32.h
- Uses the CALL alternative mechanism from commit
1b1d925818
Changes in v2:
- Merged 386 and cx8 support in the same patch
- 386 support now done in assembly, C code no longer used at all
- cmpxchg64 is used for atomic64_cmpxchg
- stop using macros, use one-line inline functions instead
- miscellanous changes and improvements
Signed-off-by: Luca Barbieri <luca@luca-barbieri.com>
LKML-Reference: <1267005265-27958-5-git-send-email-luca@luca-barbieri.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Split atomic64_t functions out into separate headers, since they will
not be practical to merge between 32 and 64 bits.
Signed-off-by: Brian Gerst <brgerst@gmail.com>
LKML-Reference: <1262883215-4034-2-git-send-email-brgerst@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
