mattst88 at gmail.com
Thu Feb 4 12:10:11 PST 2010
On Thu, Feb 4, 2010 at 2:57 PM, <rixed at happyleptic.org> wrote:
> In compiler.h, is defined mem_barrier() which is used if I understand
> correctly when accessing the memory mapped IO, to force the CPU to
> ensure previous writes/reads are commited to the bus before the
> following ones (gcc being independantly informed to keep the source code
> reads/writes ordering by the use of volatile).
> On MIPS on Linux, I see that the mem_barrier consist of several nops.
> I do not understand these nops, since first MIPS ISA there exist
> a SYNC instruction that forces the CPU to flush reads/writes for this
> very purpose.
> Also, this instruction is certainly faster, and more portable than a
> sequence of nops.
> But I'm certain that the original author knew more about MIPS than I do,
> that's why I'm asking here for a clue.
You must be looking at old source code. I changed this code in commit
52aa0495218d to use the sync instruction you mention.
It was committed August 31 2009, so X server 1.7 and newer releases
should have this code.
More information about the xorg