In create_zero_mask() we have:
addi %1,%2,-1
andc %1,%1,%2
popcntd %0,%1
using the "r" constraint for %2. r0 is a valid register in the "r" set,
but addi X,r0,X turns it into an li:
li r7,-1
andc r7,r7,r0
popcntd r4,r7
Fix this by using the "b" constraint, for which r0 is not a valid
register.
This was found with a kernel build using gcc trunk, narrowed down to
when -frename-registers was enabled at -O2. It is just luck however
that we aren't seeing this on older toolchains.
Thanks to Segher for working with me to find this issue.
Cc: stable@vger.kernel.org
Fixes: d0cebfa650 ("powerpc: word-at-a-time optimization for 64-bit Little Endian")
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
For some reason, only the little-endian flavor of
powerpc provided the zero_bytemask() implementation.
Reported-by: Michal Sojka <sojkam1@fel.cvut.cz>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>
Use cmpb which compares each byte in two 64 bit values and
for each matching byte places 0xff in the target and 0x00
otherwise.
A simple hash_name microbenchmark:
http://ozlabs.org/~anton/junkcode/hash_name_bench.c
shows this version to be 10-20% faster than running the x86
version on POWER8, depending on the length.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
It is a rarely exercised case, so we want to have a test to ensure it
works as required.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Implement a bi-arch and bi-endian version of load_unaligned_zeropad.
Since the fallback case is so rare, a userspace test harness was used
to test this on ppc64le, ppc64 and ppc32:
http://ozlabs.org/~anton/junkcode/test_load_unaligned_zeropad.c
It uses mprotect to force a SEGV across a page boundary, and a SEGV
handler to lookup the exception tables and run the fixup routine.
It also compares the result against a normal load.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This is an optimization for the PowerPC in 64-bit
little-endian. Bit counting is used in find_zero(), instead
of the multiply and shift.
It is modelled after Alan Modra's PowerPC LE strlen patch
http://sourceware.org/ml/libc-alpha/2013-08/msg00097.html.
Signed-off-by: Philippe Bergheaud <felix@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
The powerpc word-at-a-time functions are big endian specific.
Bring in the x86 version in order to support little endian builds.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
This is much the same as for SPARC except that we can do the find_zero()
function more efficiently using the count-leading-zeroes instructions.
Tested on 32-bit and 64-bit PowerPC.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Anton Blanchard
Chris Metcalf
Michael Ellerman
Philippe Bergheaud
Paul Mackerras