BUG: 9.0 stage 2 boot (/boot/boot)

[John Baldwin]

On Thursday, March 01, 2012 3:02:35 pm rank1seeker at gmail.com wrote:
> ----- Original Message -----
> From: John Baldwin <jhb at freebsd.org>
> To: rank1seeker at gmail.com
> Cc: hackers at freebsd.org, "Roman Divacky" <rdivacky at freebsd.org>
> Date: Wed, 29 Feb 2012 16:05:08 -0500
> Subject: Re: BUG: 9.0 stage 2 boot (/boot/boot)
> 
> > On Wednesday, February 29, 2012 3:08:36 pm rank1seeker at gmail.com wrote:
> > > ----- Original Message -----
> > > From: John Baldwin <jhb at freebsd.org>
> > > To: rank1seeker at gmail.com
> > > Cc: hackers at freebsd.org, "Roman Divacky" <rdivacky at freebsd.org>
> > > Date: Wed, 29 Feb 2012 11:26:59 -0500
> > > Subject: Re: BUG: 9.0 stage 2 boot (/boot/boot)
> > > 
> > > > On Tuesday, February 28, 2012 3:09:06 pm rank1seeker at gmail.com wrote:
> > > > > ----- Original Message -----
> > > > > From: John Baldwin <jhb at freebsd.org>
> > > > > To: rank1seeker at gmail.com
> > > > > Cc: hackers at freebsd.org, "Roman Divacky" <rdivacky at freebsd.org>
> > > > > Date: Mon, 27 Feb 2012 11:23:59 -0500
> > > > > Subject: Re: BUG: 9.0 stage 2 boot (/boot/boot)
> > > > > 
> > > > > > On Saturday, February 25, 2012 9:41:48 am rank1seeker at gmail.com wrote:
> > > > > > > > Do you only see the "No " message?  Do you see the '/boot.config: /loader'
> > > > > > > > message?  (Do you have RBX_QUIET enabled perhaps? (-q))  Do you get the actual
> > > > > > > > boot2 prompt at all?
> > > > > > > 
> > > > > > > I don't have RBX_QUIET enabled nor any other flags
> > > > > > > 
> > > > > > > Let the pic tell a story:
> > > > > > > http://www.starforce.biz/stage2boot.jpg
> > > > > > 
> > > > > > Ahh, this is helpful.  You do see the '/boot.config: /loader' message.
> > > > > 
> > > > > I've already explained that, numerous times (RE-typing ...)
> > > > 
> > > > It was not as obvious before, and you are seeing a failure that no one else
> > > > has reported, so you need to be patient.
> > > > 
> > > > > > > Patch eliminates possible error, of manual "intervention"
> > > > > > > That is, a perfectly valid patch being classified as invalid.
> > > > > > 
> > > > > > I have no idea what you mean here.  However, it seems you don't have junk in
> > > > > > your 'opts' variable anyway.
> > > > > 
> > > > > What I meant was that I won't manually(edit file) apply patch, but via 'patch' tool/bin.
> > > > 
> > > > Ok.
> > > > 
> > > > > > Hmm, you could try adding some more debugging to boot2.c to see exactly what
> > > > > > is failing.  For example, does the first call to 'parse()' fail and clear
> > > > > > autoboot?
> > > > > 
> > > > > I don't do nor understand c code.
> > > > 
> > > > Ok.  That will take a bit longer to fix, but that is ok.  I've attached a new
> > > > patch with some debugging output.  It shouldn't fix the problem yet, but I want
> > > > to see if any of the new messages are output, and when they are output.
> > > > 
> > > > > How could it silently loose documented functionality?
> > > > 
> > > > Several changes were made to boot2 to make it smaller so it could be compiled with
> > > > clang, and it seems that at least one of those changes must have had a bug.
> > > > 
> > > > -- 
> > > > John Baldwin
> > > >
> > > 
> > > 
> > > Patch fails at 9.0 RELEASE: (Is this for 9 STABLE?)
> > 
> > Nope, patch was made against a 9.0 tree:
> > 
> > % svn info .
> > Path: .
> > URL: svn+ssh://svn.freebsd.org/base/releng/9.0/sys
> > Repository Root: svn+ssh://svn.freebsd.org/base
> > Repository UUID: ccf9f872-aa2e-dd11-9fc8-001c23d0bc1f
> > Revision: 232297
> > Node Kind: directory
> > Schedule: normal
> > Last Changed Author: kensmith
> > Last Changed Rev: 229283
> > Last Changed Date: 2012-01-02 09:45:30 -0500 (Mon, 02 Jan 2012)
> > 
> > % svn stat boot/i386/boot2
> > M       boot/i386/boot2/boot2.c
> > 
> > % svn diff boot/i386/boot2 | md5
> > 888f90f32bd20d1bf7e2d3277d9b697b
> > 
> > And the file I sent to you before:
> > 
> > % md5 ~/work/patches/boot2_test.patch 
> > MD5 (/home/jhb/work/patches/boot2_test.patch) = 888f90f32bd20d1bf7e2d3277d9b697b
> > 
> > > I'll give you a hint (which I've mentioned at start)
> > > 
> > > In order to expose bug, 2 conditions have to be met:
> > >    1) boot.config in use
> > >    2) daX device (i.e; USB stick)
> > > 
> > > That is ...
> > > I've created vnode image. Then, ... when I 'dd' it to HDD's slice, it boots.
> > > BUT when I 'dd' it to USB's slice it hangs.
> > 
> > USB booting uses a different chunk of BIOS code, and it may be doing different
> > things which result in uninitialized memory having different values (e.g. the
> > cmd[] array).
> > 
> > -- 
> > John Baldwin
> > 
> 
> Using new:
> MD5 (boot2_test.patch) = 01cc3a9dda1f1ecbd5125092ebccc476

Hmmm, it seems you have autoboot == 0, but kname set (but you didn't get the
printf where autoboot is set to 0, and you didn't get the printf where kname
is set to a non-NULL value).  I'll update the patch again (same URL), but if
you don't get any new messages, then I suspect you aren't actually using the
new boot2.  You do realize you have to use gpart to install the new boot code
on the USB stick after you have built it, yes?

-- 
John Baldwin