Re: memory management issues

On Mon, 2007-04-30 at 11:44 -0700, Ollie Wild wrote:
> In response to previous attempts to increase the command line
> limitation (http://thread.gmane.org/gmane.linux.kernel/418858), Linus
> has indicated that any temporary pages should be swappable.  This
> patch attempts to solve this problem by creating the new process's
> mm_struct early in the do_execve() callchain and writing directly into
> the new stack segment (cf. copy_strings()).  When I instrument this on
> an a500, I observe that the pages are setup correctly and contain
> valid values up until the call to switch_mm() via the call stack:
> 
>   activate_mm()
>   exec_mmap()
>   flush_old_exec()
>   load_elf_binary()
>   do_execve()
> 
> Afterwards, the pages are zeroed.

Give me more of a clue on this one ... afterwards *who* sees the pages
as zero'd?  The user or the kernel ... if the user, I think it's a
simple coherency problem.

James