Date: Thu, 10 Mar 2011 13:29:53 -0500 From: John Baldwin <jhb@freebsd.org> To: Matthew Fleming <mdf356@gmail.com> Cc: Roman Divacky <rdivacky@freebsd.org>, freebsd-current@freebsd.org, Dimitry Andric <dim@freebsd.org> Subject: Re: [TESTING]: one more boot2 shrinking patch Message-ID: <201103101329.54044.jhb@freebsd.org> In-Reply-To: <AANLkTik-Ps7qt4WAyRnOGtwpxSQKaXqB5u4GMfBivLiJ@mail.gmail.com> References: <20110308173909.GA71091@freebsd.org> <20110310163732.GA38320@freebsd.org> <AANLkTik-Ps7qt4WAyRnOGtwpxSQKaXqB5u4GMfBivLiJ@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thursday, March 10, 2011 11:58:30 am Matthew Fleming wrote: > On Thu, Mar 10, 2011 at 8:37 AM, Roman Divacky <rdivacky@freebsd.org> wrote: > > On Thu, Mar 10, 2011 at 09:20:58AM -0500, John Baldwin wrote: > >> On Wednesday, March 09, 2011 6:24:36 pm Dimitry Andric wrote: > >> > On 2011-03-09 14:23, John Baldwin wrote: > >> > >> gcc nor clang emits any code to initialize static type foo = 0; > >> > >> because it's expected that BSS is zeroed, which is not the case > >> > >> in boot2 so we have to initialize that explicitly > >> > > It used to be that if you explicitly initialized a variable to 0, it was > >> > > initialized to 0 in .data, but now gcc and clang recognize it is set to 0 and > >> > > move it to .bss. There appears to be no way to turn this feature off, > >> > > >> > Yes, there is; both gcc and clang have this option to turn it off: > >> > > >> > -fno-zero-initialized-in-bss > >> > If the target supports a BSS section, GCC by default puts variables > >> > that are initialized to zero into BSS. This can save space in the > >> > resulting code. > >> > > >> > This option turns off this behavior because some programs > >> > explicitly rely on variables going to the data section. E.g., so > >> > that the resulting executable can find the beginning of that > >> > section and/or make assumptions based on that. > >> > > >> > The default is -fzero-initialized-in-bss. > >> > >> Hah, that is better then. Thanks! I should have searched about this more > >> myself. :( Roman, can you try reverting the kname changes and adding this > >> to CFLAGS instead for both compilers? > > > > when I put -fno-zero-initialized-in-bss clang does not fit by 1.7K and > > gcc by 0.5K, we dont want this :) > > If there's that many variables explicitly initialized to zero, does > boot2 then need to explicitly bzero all of bss, not just kname, since > it's not handled by the loader? Otherwise it sounds like either > there's a lot of explicitly initialized variables that didn't need to > be, or there's a lot of potential for uninitialized variable nonsense > to happen. Yes. It would be good to see which symbols move from .bss to .data with this change. Hopefully the compilers aren't putting everything into .data. -- John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201103101329.54044.jhb>