Re: [PATCH] Early memory patch, +accounting

[ebiederm@xmission.com (Eric W. Biederman)]

"H. Peter Anvin" <hpa@zytor.com> writes:

> Eric W. Biederman wrote:
> > I intend to export _end into setup.S so a bootloader can
> > tell what memory the kernel will attempt to use before looking at
> > the e820 map, the kernel command line, and the initrd location.
> > After that the kernel can properly dynamically allocate things.
> >
> 
> You're confusing "may" with "will".  This is the fundamental problem with this
> approach.

If the margin of error was significant I would agree.  The margin of error
in this case is very small.

> > Exporting _end will allow two things.
> > 1) On systems with a 15M-16M I/O hole the bootloader can check to see
> >    if a big kernel will attempt to use memory in that region.
> > 2) On small memory systems it will let the bootloader make a good
> >    estimate to see if everything will fit into memory.
> 
> No, it won't, because it still has no idea whatsoever how much dynamically
> allocated memory it needs during boot.

It will at least ensure the kernel can run long enough to fail gracefully.
I admit that a panic is not the most graceful failure but it does at least
let you know what went wrong.

Eric