Date: Mon, 18 Sep 2000 10:10:14 +0100 (BST) From: Doug Rabson <dfr@nlsystems.com> To: Matthew Jacob <mjacob@feral.com> Cc: Peter Wemm <peter@netplex.com.au>, "Daniel C. Sobral" <dcs@newsguy.com>, alpha@freebsd.org Subject: Re: Loader badly broken on the alpha Message-ID: <Pine.BSF.4.21.0009180950390.86297-100000@salmon.nlsystems.com> In-Reply-To: <Pine.BSF.4.21.0009171808040.96606-100000@beppo.feral.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 17 Sep 2000, Matthew Jacob wrote: > > > > Matthew Jacob wrote: > > > > > > I think I had determined (I thought) that it was just loader itself, but I > > > could be wrong. > > > > The problem is that the loader is now too large, it cannot fit inside the > > ~208K window that boot1 has got available to it. > > > > peter@ashburton[12:14am]/home/obj/home/src/sys/boot/alpha/loader-130# nm loader.sym | sort | tail -1 > > 000000002003e190 A end > > > > The VM space ends at 2004000. I can't find where, but I was pretty sure that > > the last 8K page was already spoken for and we've clobbered it. > > > > Was the pnp stuff added to the Alpha loader images? That is what probably > > tipped it over the edge. > > I've been looking at this a little bit. The last last 8k may not be entirely > spoken for, because previous loader.sym's had: > > 000000002003d260 A _end > 000000002003d260 A end > > > But if it's 8k (and I don't know anything much about this really- I have never > delved to much into this portion of alphas) this may explain why the boot > stuff never did work right entirely for all models (e.g., off of floppy for a > rawhide). > > In any case, this one is a hard one. I was comparing object sizes between June > 23'd loader.sym and the current one on alpha, and there's no one single thing > that got all that much bigger! > > We need to pull out 3408 bytes- but it's not that simple- bcopy seems to take > 2544 bytes, which is ridiculously large for a boot block- so I replaced it > with a dopey one line C function but that didn't change _end for me, so I have > to look at the linker magic some I guess. Looking at sizes, supporting gzip'ed files is quite expensive - about 23k of code and data. Ficl is the real heavyweight with 50k in words.o alone. I just tried building everything with -Os but that was a waste of time - only 784 bytes difference. -- Doug Rabson Mail: dfr@nlsystems.com Nonlinear Systems Ltd. Phone: +44 20 8348 3944 To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0009180950390.86297-100000>