headers
Andreas Kahari | 3 Apr 2004 00:55
Picon
Favicon

The sudden reset problem

Hi,

I'm wondering if this problem:

    http://mail-index.netbsd.org/current-users/1999/08/04/0002.html

and this:

    http://mail-index.netbsd.org/port-amd64/2004/03/09/0000.html

is the same problem, or somehow related?

I just ran into the same thing on OpenBSD/amd64 when
compiling gnome and running another heavy process in
the background (just as I did previously in NetBSD, see
http://mail-index.netbsd.org/port-amd64/2004/03/10/0002.html).

Could it be hardware related?  The hardware monitor in OpenBSD says:

$ sysctl hw.sensors
hw.sensors.0=it0, Fan1, fanrpm, 3409 RPM
hw.sensors.2=it0, Fan3, fanrpm, 0 RPM
hw.sensors.3=it0, VCORE_A, volts_dc, 1.50 V
hw.sensors.4=it0, VCORE_B, volts_dc, 2.54 V
hw.sensors.5=it0, +3.3V, volts_dc, 3.23 V
hw.sensors.6=it0, +5V, volts_dc, 3.58 V
hw.sensors.7=it0, +12V, volts_dc, 11.65 V
hw.sensors.8=it0, Unused, volts_dc, -0.12 V
hw.sensors.9=it0, -12V, volts_dc, -5.96 V
hw.sensors.10=it0, +5VSB, volts_dc, 3.63 V
(Continue reading)

Permalink | Reply | 0 comments |
headers
Frank van der Linden | 5 Apr 2004 13:50
Picon

warning: src/x11 broken

I've been told that building X via src/x11 will always produce
an X server that uses loadable modules. This will not
work on amd64. The XFree86 loader is broken, and only works on
Linux because they added a hack to mmap to specify memory below
the 32bit boundary. I've looked at it a while ago, and I don't
think this can be fixed, unless they switch to a normal dlopen()
scheme.

So, building X via src/x11 will not produce a working X server.
If you want to build X, use the old method of going to xsrc
and doing a 'make build' with DESTDIR set.

- Frank
Permalink | Reply | 4 comments |
headers
Richard Rauch | 7 Apr 2004 21:48

NetBSD 2.0, -current.

Hi.  I haven't been tracking -current closely the past few weeks, nor
have I had time to try to install a 2.0 release-candidate.

However, there's a NetBSD/amd64 PR that's been outstanding for some time.
We have a /dev/joy[01] pair of specials, and a joy(4) manpage, but
no /usr/include/machine/joystick.h.

This affects some libraries that provide joystick support.

The file was missing for months after I filed the PR, so I assume
that the file is still missing, though I have not checked lately.
If the file is there, or if it has been decided to simply not support
joy(4) on NetBSD/amd64, could someone close the PR?

It is at least sufficient to import the i386 header to make programs
compile, though I have no idea if this really does anything useful.  (^&

(Is joy(4) standard, while I'm on the subject?  Or is it  something that
we get for being close to i386, but which would be absent on Amigas,
eet al?)

--

-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/
Permalink | Reply | 0 comments |
headers
Jason Lixfeld | 8 Apr 2004 17:51
Picon

AMD64 + OpenLDAP

I'm having some issues at the moment with OpenLDAP and FreeBSD. I'm 
wondering if anyone is having issues with OpenLDAP and NetBSD.  
Specifically, slapd not dying when you try to stop it and bdb being 
unusable.
Permalink | Reply | 0 comments |
headers
George E Sollish | 9 Apr 2004 14:23

2.0 on AMD64

Now that the 2.0 release sequence is underway, do we know if/when amd64
will be added to the list of supported ports?

George E Sollish	Chief Engineer		Auto Gear Equipment
			Project Manager		The Payne Lake Project
Permalink | Reply | 0 comments |
headers
Matthias Scheler | 11 Apr 2004 10:37
Picon
Favicon

Re: warning: src/x11 broken

In article <20040405115023.GA27209 <at> vaasje.org>,
	Frank van der Linden <fvdl <at> NetBSD.org> writes:
> The XFree86 loader is broken, and only works on Linux because they
> added a hack to mmap to specify memory below the 32bit boundary.

So IIUC the loader doesn't work if the address returned by mmap() is
above 4GB. Correct?

> I've looked at it a while ago, and I don't think this can be fixed,
> unless they switch to a normal dlopen() scheme.

A possible fix would be a change like this to
"xfree/xc/programs/Xserver/hw/xfree86/loader/loader.c":

#if defined(__NetBSD__) && (__AMD64__)
	static char *mm_addr = (char *)(1L << 30);
#else
	char *mm_addr = NULL;
#endif

    ret = (unsigned long)mmap(mm_addr, new_size, MMAP_PROT, MAP_PRIVATE
#  if !defined(_NetBSD__) && defined(__AMD64__)
                              | MAP_32BIT
#  endif
                              , fd, new_off);
    if (ret == -1)
        FatalError("mmap() failed: %s\n", strerror(errno));
#if defined(__NetBSD__) && (__AMD64__)
    mm_addr += new_size;
#endif
(Continue reading)

Permalink | Reply | 3 comments |
headers
Frank van der Linden | 11 Apr 2004 11:29
Picon

Re: warning: src/x11 broken

On Sun, Apr 11, 2004 at 08:37:17AM +0000, Matthias Scheler wrote:
> So IIUC the loader doesn't work if the address returned by mmap() is
> above 4GB. Correct?

Yeah, because it does direct address fixups on non-PIC objects. And
by default, text and data are referenced using 32bit-values in
generated object files on AMD64 (small model). But mmap and malloc
do, of course, use the full range available in 64bit values. So
if mmap returns a value > 2G (they can be relative values), you're
screwed.

> A possible fix would be a change like this to
> "xfree/xc/programs/Xserver/hw/xfree86/loader/loader.c":
> 
> #if defined(__NetBSD__) && (__AMD64__)
> 	static char *mm_addr = (char *)(1L << 30);
> #else
> 	char *mm_addr = NULL;
> #endif
> 
>     ret = (unsigned long)mmap(mm_addr, new_size, MMAP_PROT, MAP_PRIVATE
> #  if !defined(_NetBSD__) && defined(__AMD64__)
>                               | MAP_32BIT
> #  endif
>                               , fd, new_off);
>     if (ret == -1)
>         FatalError("mmap() failed: %s\n", strerror(errno));
> #if defined(__NetBSD__) && (__AMD64__)
>     mm_addr += new_size;
> #endif
(Continue reading)

Permalink | Reply | 2 comments |
headers
Matthias Scheler | 11 Apr 2004 11:28
Picon
Favicon

Re: warning: src/x11 broken

On Sun, Apr 11, 2004 at 11:29:08AM +0200, Frank van der Linden wrote:
> That's not a bad idea..

Thank you.

> a few changes I would make:
> 
> #define LOADER_MMAP_AREA_SIZE	(256*1024*1024)
> #define LOADER_MMAP_AREA_START	((2*1024*1024*1024) - LOADER_MMAP_AREA_SIZE)
> 
> ...
> 
> 	static char *mm_addr = LOADER_MMAP_AREA_START;
> 
> ...
> 
> 	mm_addr += roundup(new_size, PAGE_SIZE);
> ...
> 
> 
> I think this might work.. I have no time to try it for a few days, though.

And I don't have a NetBSD-amd64 system unfortunately.

Any volunteer?

	Kind regards

--

-- 
Matthias Scheler                                  http://scheler.de/~matthias/
(Continue reading)

Permalink | Reply | 1 comment |
headers
Nicolas Joly | 13 Apr 2004 18:58
Picon
Picon
Favicon

Re: warning: src/x11 broken

On Sun, Apr 11, 2004 at 11:28:46AM +0200, Matthias Scheler wrote:
> On Sun, Apr 11, 2004 at 11:29:08AM +0200, Frank van der Linden wrote:
> > That's not a bad idea..
> 
> Thank you.
[...]
> > I think this might work.. I have no time to try it for a few days, though.
> 
> And I don't have a NetBSD-amd64 system unfortunately.
> 
> Any volunteer?

I just got a working X server with loadable modules with the following
patches (for xsrc and src/x11).

I checked it on our amd64 machines (workstation with 1GB of memory,
and server with 4GB).

NB: I had some problems with Frank suggestions, which are not included
here.

NB2: src/x11 Makefiles patches include xdm vt switching fixes (PCVT
support, like i386).

--

-- 
Nicolas Joly

Biological Software and Databanks.
Institut Pasteur, Paris.
(Continue reading)

Permalink | Reply | 0 comments |
headers
Richard Rauch | 14 Apr 2004 22:54

Crashes, no fsck?

Lately, I've noticed that when I reboot after a crash, the system skips
the fsck pass.  I've taken to booting single-user after a crash and
manually forcing an fsck -f, but even doing that never reports any
filesystem problems.

Yet...I've seen filesystem corruption.  (E.g., my pkg database
has had strange artifacts.)  I know that others have complained of
crashes combined with corruption.

I've just reinstalled the system with a March 12 -current, plus an
XFree86 from November so that I can get vastly better X performance.
Partly I redid the install so that I have partition space to install
alternate operating systems.

Aside: Mandrake 10's i386 distribution is able to read the motherboard
clock on my Gigabyte GA-K8N Pro (nVidia nForce3 based) motherboard.
NetBSD/amd64 and FreeBSD/amd64 both complain that the clock has no valid
time set.  (I have not tried NetBSD/i386.)

I've also updated the BIOS on the off-chance that it might be affecting
either IOAPIC or clock issues.  (No dice.)

--

-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/
Permalink | Reply | 0 comments |

Gmane