Re: [PATCH] PowerPC64 port part 4 of 7

["Steve Munroe" <sjmunroe at us dot ibm dot com>]

> Oh yeah?
>
>     * sysdeps/powerpc/elf/libc-start.c : Scan Aux Vector for
>     AT_DCACHEBSIZE and copy value to __cache_line_size.
>     * sysdeps/unix/sysv/linux/powerpc/dl-sysdep.c : Scan Aux Vector for
>     AT_DCACHEBSIZE and copy value to __cache_line_size.
>     * sysdeps/powerpc/ppc32/memset.S : Define __cache_line_size and use
>     its value to select the correct stride for dcbz.
>     * sysdeps/powerpc/ppc64/memset.S : Define __cache_line_size and use
>     its value to select the correct stride for dcbz.

This one has already been raised as: [PATCH] memset.S and PowerPC.

The 32-bit only patch was submitted on 07/09/02 as

http://sources.redhat.com/ml/libc-alpha/2002-07/msg00091.html

"adds support to capture the AT_DCACHEBSIZE from the aux vector during libc
(and ld.so) init and store it in static "int __cache_line_size".
__cache_line_size is used by sysdeps/powerpc/memset.S to determine the
actual cache-line size. The original implementation (hardcode for 32-byte
cache-lines) is left intact (logic and cache alignment) and is used when
the cache-line size matches. Otherwise cache-line size independent code is
used. If the __cache_line_size value is 0 (for example an old version of
the kernel that doesn't support AT_DCACHEBSIZE) then use a slower but safe
sequence."

to which I did not get any responce ... I believe that at about this time
you mail system was bouncing mail and you may of missed it.

The change above follows the first same design as the 2002-07 (PowerPC32
only) submission with the addition of;
1) moving the 32-bit version of memset.S up one directory.
2) adding the 64-bit version of the memset.S.