On Fri, Mar 13, 2015 at 08:52:58PM +0200, Aaro Koskinen wrote: > Which drivers did you test? Did you test e.g. octeon-md5? octeon-md5 does fail. The hardware seems to be expecting big-endian values. I have got it to work by doing this: https://github.com/nowster/linux-ubnt-e200/commit/019411e18e943624aa61796e54aa933397f8fdca?diff=unified (I know that cut-and-paste will destroy the tabs in this. I provide the patch for comments as to whether this approach is sane.) diff --git a/arch/mips/cavium-octeon/crypto/octeon-crypto.h b/arch/mips/cavium-octeon/crypto/octeon-crypto.h index e2a4aec..0e157f1 100644 --- a/arch/mips/cavium-octeon/crypto/octeon-crypto.h +++ b/arch/mips/cavium-octeon/crypto/octeon-crypto.h @@ -32,7 +32,7 @@ do { \ __asm__ __volatile__ ( \ "dmtc2 %[rt],0x0048+" STR(index) \ : \ - : [rt] "d" (value)); \ + : [rt] "d" (cpu_to_be64(value))); \ } while (0) /* @@ -47,7 +47,7 @@ do { \ : [rt] "=d" (__value) \ : ); \ \ - __value; \ + be64_to_cpu(__value); \ }) /* @@ -58,7 +58,7 @@ do { \ __asm__ __volatile__ ( \ "dmtc2 %[rt],0x0040+" STR(index) \ : \ - : [rt] "d" (value)); \ + : [rt] "d" (cpu_to_be64(value))); \ } while (0) /* @@ -69,7 +69,7 @@ do { \ __asm__ __volatile__ ( \ "dmtc2 %[rt],0x4047" \ : \ - : [rt] "d" (value)); \ + : [rt] "d" (cpu_to_be64(value))); \ } while (0) #endif /* __LINUX_OCTEON_CRYPTO_H */ -- Paul Martin http://www.codethink.co.uk/ Senior Software Developer, Codethink Ltd.
