Speeding up GC

[Marc Espie]

On Tue, Jun 04, 2002 at 08:52:07AM -0700, David S. Miller wrote:
>    From: Marc Espie <espie@quatramaran.ens.fr>
>    Date: Tue, 4 Jun 2002 17:08:51 +0200
>    
>    Actually, I can pinpoint it partly: going down such a path would
>    1/ make gcc very much linux-dependent.
> 
> Actually, the dirty bit facility in question was orignally found to be
> present under Solaris.  Every UNIX provides the mprotect+SIGSEGV
> feature though.

No.

1/ it may be prohibitively expensive on most OSes that implement it.

2/ Here, documentation of mprotect says:
Not all implementations will guarantee protection on a page basis;
the granularity of protection changes may be as large as an entire re-
gion.

Even the linux documentation says something similar:
 POSIX.1b says that mprotect can be used only on regions of memory 
 obtained from mmap(2)