NetBSD-Bugs archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: kern/56701: panic on ufs2 while running on QEMU 6.2.0

The following reply was made to PR kern/56701; it has been noted by GNATS.

From: nervoso <nervoso%k1.com.br@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: kern/56701: panic on ufs2 while running on QEMU 6.2.0
Date: Thu, 17 Feb 2022 11:49:11 -0300

 --=-uo4iG31YMwQvCZ7Ei5uH
 Content-Type: text/plain; charset="UTF-8"
 Content-Transfer-Encoding: 8bit
 
 Em qui, 2022-02-17 às 13:55 +0000, Martin Husemann escreveu:
 > The following reply was made to PR kern/56701; it has been noted by GNATS.
 > 
 > From: Martin Husemann <martin%duskware.de@localhost>
 > To: gnats-bugs%netbsd.org@localhost
 > Cc: 
 > Subject: Re: kern/56701: panic on ufs2 while running on QEMU 6.2.0
 > Date: Thu, 17 Feb 2022 14:52:04 +0100
 > 
 >  On Thu, Feb 17, 2022 at 01:45:01PM +0000, nervoso wrote:
 >  >  probably the gcc compiler on amd64 produces wrong i386 code... atomic
 >  >  OP issues???
 >  
 >  All builds are cross compiled with the same compiler, even if you build
 >  on the same arch as the target.
 >  
 >  The cross compiler emitting different code when compiled in a different
 >  environment is possible (and would be a gcc bug), but in this case sounds
 >  very unlikely.
 >  
 >  Martin
 >  
 but that is what is happening... HEAD  uses gcc 10.3 while 9.2 uses gcc
 7.5.
 the interesting part is that when I build the GENERIC kernel using
 build.sh .... kernel=GENERIC
 it builds the TOOLS  
 if I build ther TOOLS on amd64, and then builds the kernel  it breaks
 but if I build the TOOLS using i386 and then build the kenrnel, the
 system works
 
 but if I build the GUEST using sources from 9.2_STABLE in no way the
 system breaks...
 the only explanation possible is that when building the TOOLS, the
 tools of the guest for 9.2
 result in a gcc 7.5 compiler while the HEAD builds the system using gcc
 10.3.
 Conclusion:
 the only way to build a running  i386 HEAD kernel in qemu is to build
 the system using
 the i386 VM  machine... booted from the HEAD original ISO from
 netbsd.org..
 Once booted, I am able to build any custom kernel (no need to build
 modules) and all of the kernels are operational
 
 Unfortunately I do not have any intel machine here... but I will have a
 core2 in some days...
 then ! will test it running on "bare metal"..  I will let you know
 
 This is weird, I know
 
 
 --=-uo4iG31YMwQvCZ7Ei5uH
 Content-Type: text/html; charset="utf-8"
 Content-Transfer-Encoding: quoted-printable
 
 <html><head></head><body><div>Em qui, 2022-02-17 =C3=A0s 13:55 +0000, Marti=
 n Husemann escreveu:</div><blockquote type=3D"cite" style=3D"margin:0 0 0 .=
 8ex; border-left:2px #729fcf solid;padding-left:1ex"><pre>The following rep=
 ly was made to PR kern/56701; it has been noted by GNATS.</pre><pre><br></p=
 re><pre>From: Martin Husemann &lt;<a href=3D"mailto:martin%duskware.de@localhost";>mar=
 tin%duskware.de@localhost</a>&gt;</pre><pre>To: <a href=3D"mailto:gnats-bugs@netbsd.o=
 rg">gnats-bugs%netbsd.org@localhost</a></pre><pre>Cc: </pre><pre>Subject: Re: kern/56=
 701: panic on ufs2 while running on QEMU 6.2.0</pre><pre>Date: Thu, 17 Feb =
 2022 14:52:04 +0100</pre><pre><br></pre><pre> On Thu, Feb 17, 2022 at 01:45=
 :01PM +0000, nervoso wrote:</pre><pre> &gt;  probably the gcc compiler on a=
 md64 produces wrong i386 code... atomic</pre><pre> &gt;  OP issues???</pre>=
 <pre> </pre><pre> All builds are cross compiled with the same compiler, eve=
 n if you build</pre><pre> on the same arch as the target.</pre><pre> </pre>=
 <pre> The cross compiler emitting different code when compiled in a differe=
 nt</pre><pre> environment is possible (and would be a gcc bug), but in this=
  case sounds</pre><pre> very unlikely.</pre><pre> </pre><pre> Martin</pre><=
 pre> </pre></blockquote><div>but that is what is happening... HEAD &nbsp;us=
 es gcc 10.3 while 9.2 uses gcc 7.5.</div><div>the interesting part is that =
 when I build the GENERIC kernel using build.sh .... kernel=3DGENERIC</div><=
 div>it builds the TOOLS &nbsp;</div><div>if I build ther TOOLS on amd64, an=
 d then builds the kernel &nbsp;it breaks</div><div>but if I build the TOOLS=
  using i386 and then build the kenrnel, the system works</div><div><br></di=
 v><div>but if I build the GUEST using sources from 9.2_STABLE in no way the=
  system breaks...</div><div>the only explanation possible is that when buil=
 ding the TOOLS, the tools of the guest for 9.2</div><div>result in a gcc 7.=
 5 compiler while the HEAD builds the system using gcc 10.3.</div><div>Concl=
 usion:</div><div>the only way to build a running &nbsp;i386 HEAD kernel in =
 qemu is to build the system using</div><div>the i386 VM &nbsp;machine... bo=
 oted from the HEAD original ISO from netbsd.org..</div><div>Once booted, I =
 am able to build any custom kernel (no need to build modules) and all of th=
 e kernels are operational</div><div><br></div><div>Unfortunately I do not h=
 ave any intel machine here... but I will have a core2 in some days...</div>=
 <div>then ! will test it running on "bare metal".. &nbsp;I will let you kno=
 w</div><div><br></div><div>This is weird, I know</div><div><br></div><div c=
 lass=3D"-x-evo-signature-wrapper"><span class=3D"-x-evo-signature" id=3D"no=
 ne"></span></div></body></html>
 
 --=-uo4iG31YMwQvCZ7Ei5uH--
Home | Main Index | Thread Index | Old Index