headers
Christian Smith | 1 Dec 2004 17:46

Re: sysinst problems

On Tue, 30 Nov 2004, Greywolf wrote:

>[Thus spake der Mouse ("dM: ") 6:19am...]
>
>dM: - The partitioning tool seems semi-broken, in that it refuses to allow
>dM:   partitions to overlap (except for overlap with c).  I had a disk
>dM:   pre-partitioned roughly thus:
>dM:
>dM:     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa     dddddddddddbbbbbbbbbb
>dM:                                   eeeeeeeeeeeeeeee
>dM:
>dM:   and perhaps I just told it the wrong thing, but it whined about d and
>dM:   e overlapping - I knew they overlapped and I wanted it that way.  I'm
>dM:   quite sure I set both d and e to no-mount.
>
>Yes, please don't try to be smart about this.
>
>I have frequently had need for overlapping partitions so that, depending
>upon the project at hand, I could slice 'n' dice as needed.
>
>[Maybe I really AM a dinosaur for remembering
>
>	aaabbbbbbggggggggggggggggggghhhhhhhhhhhhhhhhhhhhhhhhhhhhhh
>	   dddddeeeeeeeeffffffffffff
> :-)
>]

How can this possibly be useful?

>
(Continue reading)

Permalink | Reply | 42 comments |
headers
Greywolf | 1 Dec 2004 22:37

Re: sysinst problems

[Thus spake Christian Smith ("CS: ") 4:46pm...]

CS: From: Christian Smith <csmith <at> micromuse.com>
CS: To: der Mouse <mouse <at> Rodents.Montreal.QC.CA>,
CS:      Greywolf <greywolf <at> starwolf.com>
CS: Cc: NetBSD SPARC list <port-sparc <at> NetBSD.org>
CS: Subject: Re: sysinst problems
CS: Date: Wed, 1 Dec 2004 16:46:20 +0000 (GMT)
CS: X-Spam-Level:
CS:
CS: On Tue, 30 Nov 2004, Greywolf wrote:
CS:
CS: >[Thus spake der Mouse ("dM: ") 6:19am...]
CS: >
CS: >dM: - The partitioning tool seems semi-broken, in that it refuses to allow
CS: >dM:   partitions to overlap (except for overlap with c).  I had a disk
CS: >dM:   pre-partitioned roughly thus:
CS: >dM:
CS: >dM:     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa     dddddddddddbbbbbbbbbb
CS: >dM:                                   eeeeeeeeeeeeeeee
CS: >dM:
CS: >dM:   and perhaps I just told it the wrong thing, but it whined about d and
CS: >dM:   e overlapping - I knew they overlapped and I wanted it that way.  I'm
CS: >dM:   quite sure I set both d and e to no-mount.
CS: >
CS: >Yes, please don't try to be smart about this.
CS: >
CS: >I have frequently had need for overlapping partitions so that, depending
CS: >upon the project at hand, I could slice 'n' dice as needed.
CS: >
(Continue reading)

Permalink | Reply | 41 comments |
headers
David Laight | 1 Dec 2004 23:04
Picon

Re: sysinst problems

> If it would be better to put these in a PR, or to put each one in a
> separate PR, I'd be happy to do that.

If you deceide they warrant PRs, make them separate - so that the
PRs can be closed when an issue is fixed.

> - The partitioning tool seems semi-broken, in that it refuses to allow
>   partitions to overlap (except for overlap with c).  I had a disk
>   pre-partitioned roughly thus:
> 
>     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa     dddddddddddbbbbbbbbbb
>                                   eeeeeeeeeeeeeeee
> 
>   and perhaps I just told it the wrong thing, but it whined about d and
>   e overlapping - I knew they overlapped and I wanted it that way.  I'm
>   quite sure I set both d and e to no-mount.

sysinst doesn't want it that way - it is usually a bug waiting to happen

> - Despite having d and e both set to no-mount and no-newfs, and d set
>   to size=0 offset=0 (this last because of the previous issue),
>   something tried to do some command - fsck, I think - on the d
>   partition (and of course it failed).  For some reason this didn't
>   repeat when I retried; I don't understand why not.

Did you do an 'update' or an 'install' ?
The former will read /etc/fstab from the existing fs and try to mount
everything.  It probably ought to display the entries and their 'last
mounted on' fields (in knowable) and force you to enable them is the
'last mounted' mismatches.
(Continue reading)

Permalink | Reply | 3 comments |
headers
der Mouse | 1 Dec 2004 23:20
Picon

Re: sysinst problems

>>     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa     dddddddddddbbbbbbbbbb
>>                                   eeeeeeeeeeeeeeee
> sysinst doesn't want it that way - it is usually a bug waiting to
> happen

sysinst obviously doesn't want it that way.  My comment was more "there
needs to be a `yes, dammit, I know what I'm doing and any resulting
problems are my own lookout' knob".

Unless we're going to actually support untar-it-yourself installs,
which I thought wasn't the case.

>> - Despite having d and e both set to no-mount and no-newfs, and d
>>   set to size=0 offset=0 (this last because of the previous issue),
>>   something tried to do some command - fsck, I think - on the d
>>   partition (and of course it failed).  For some reason this didn't
>>   repeat when I retried; I don't understand why not.
> Did you do an 'update' or an 'install' ?

The run where that happened - I forget.

> The former will read /etc/fstab from the existing fs

There was no /etc/fstab.  a had a filesystem in it but it contained
nothing but the usual . and .. entries in its root.  e had a filesystem
containing the distribution sets (and some other guff, such as the
INSTALL.* files), but nothing anywhere near an installed system.

> and try to mount everything.  It probably ought to display the
> entries and their 'last mounted on' fields (in knowable) and force
(Continue reading)

Permalink | Reply | 2 comments |
headers
Luke Mewburn | 2 Dec 2004 01:47
Picon

Re: sysinst problems

On Wed, Dec 01, 2004 at 05:20:38PM -0500, der Mouse wrote:
  | >>     aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa     dddddddddddbbbbbbbbbb
  | >>                                   eeeeeeeeeeeeeeee
  | > sysinst doesn't want it that way - it is usually a bug waiting to
  | > happen
  | 
  | sysinst obviously doesn't want it that way.  My comment was more "there
  | needs to be a `yes, dammit, I know what I'm doing and any resulting
  | problems are my own lookout' knob".

On one hand, I agree; providing such a knob could be beneficial.

On the other, I've dealt with numerous support issues because people
have unconditionally used `expert-mode' knobs and complained because
things broke.  (c.f., "build.sh -E").
I.e, I have a fear of clueless knobs overrating their ability to use
expert-mode knobs.
Permalink | Reply | 1 comment |
headers
der Mouse | 2 Dec 2004 08:05
Picon

Re: sysinst problems

>> My comment was more "there needs to be a `yes, dammit, I know what
>> I'm doing and any resulting problems are my own lookout' knob".

> On one hand, I agree; providing such a knob could be beneficial.

> On the other, I've dealt with numerous support issues because people
> have unconditionally used `expert-mode' knobs and complained because
> things broke.

I suspect that a lot of these are because someone has provided, rather
than fine-grained "yes I want to override this particular safety check"
bits, one global "yes I'm an expert" bit - and then taken functionality
that non-experts commonly want and tucked it away behind that bit.
Lacking any other way to get what they need, they're told to turn on
the expert bit - and then they promptly get burned by one of the
_other_ safety checks that disables.

For sysinst, I'd much rather not have an "expert mode", but rather have
individual "override this check" items.  For example, in this case,
instead of

| partitions d and e overlap.
|
| You can either edit the partition table by hand, or give up and return
| to the main menu.
| 
| Edit the partition table again?

with an expert-mode setting that overrides this among other checks, I'd
rather see something like
(Continue reading)

Permalink | Reply | 0 comments |
headers
Martin Husemann | 2 Dec 2004 09:31
Picon

panic: malloc: out of space in kmem_map

This happened to a SS2 with 64MB ram, running 2.0 for ~36 hours.

I probably should bump up MAXUSERS (currently 48) or something - though there
realy is not much happening on this machine, it just routes some traffic, no
big daemons nor interactive users. NFS is client only, the machine is
diskless.

Hints?

Martin

panic: malloc: out of space in kmem_map
Stopped in pid 4255.1 (mtree) at        netbsd:cpu_Debugger+0x4:        or              %
o7, %g0, %g1
db> 
db> bt
cpu_Debugger(0xf017d598, 0x0, 0x1000, 0x0, 0x0, 0xf01ab800) at netbsd:malloc+0x2
0c
malloc(0x0, 0xf01a9138, 0x0, 0x100, 0xf01f5ef4, 0x100) at netbsd:hashinit+0x5c
hashinit(0x0, 0x0, 0xf01a9138, 0x0, 0xf01f1c9c, 0x0) at netbsd:nfs_initdircache+
0x20
nfs_initdircache(0xf296ede8, 0xf01a9000, 0x516, 0x2000, 0x0, 0x0) at netbsd:nfs_
enterdircache+0x310
nfs_enterdircache(0xf296f4c8, 0x0, 0x0, 0xf047cc00, 0x200, 0x0) at netbsd:nfs_bi
oread+0x83c
nfs_bioread(0xf2c06200, 0xf2d65e18, 0xf296f4c8, 0xf2ba21d8, 0xf0558db8, 0xf296ed
e8) at netbsd:nfs_readdir+0x64
nfs_readdir(0x16, 0x20001, 0xdba, 0x0, 0xf006aa9c, 0xf0171ef4) at netbsd:vn_read
dir+0xac
vn_readdir(0x16, 0x45000, 0x0, 0x1000, 0xf2d65eb0, 0xf2ba21d8) at netbsd:sys_get
(Continue reading)

Permalink | Reply | 7 comments |
headers
R. Tyler Ballance | 2 Dec 2004 10:08
Picon
Favicon

Re: panic: malloc: out of space in kmem_map

You should probably cross-post this to tech-kern, I really think this is
a kernel problem. 

I'm not good at reading kernel debugger output, but can you find out if
the kernel really doesn't have any more memory, or if kmem_map has just
run out of mapped memory? (shouldn't it swap for more..?)

On the topic of swap? Got any? :P

-R. Tyler Ballance

On Thu, 2004-12-02 at 02:31, Martin Husemann wrote:
> This happened to a SS2 with 64MB ram, running 2.0 for ~36 hours.
> 
> I probably should bump up MAXUSERS (currently 48) or something - though there
> realy is not much happening on this machine, it just routes some traffic, no
> big daemons nor interactive users. NFS is client only, the machine is
> diskless.
> 
> Hints?
> 
> Martin
> 
> 
> panic: malloc: out of space in kmem_map
> Stopped in pid 4255.1 (mtree) at        netbsd:cpu_Debugger+0x4:        or              %
> o7, %g0, %g1
> db> 
> db> bt
> cpu_Debugger(0xf017d598, 0x0, 0x1000, 0x0, 0x0, 0xf01ab800) at netbsd:malloc+0x2
(Continue reading)

Permalink | Reply | 0 comments |
headers
ragge | 2 Dec 2004 10:13
Picon
Picon

Re: panic: malloc: out of space in kmem_map

> This happened to a SS2 with 64MB ram, running 2.0 for ~36 hours.
> 
> I probably should bump up MAXUSERS (currently 48) or something - though there
> realy is not much happening on this machine, it just routes some traffic, no
> big daemons nor interactive users. NFS is client only, the machine is
> diskless.
> 
> Hints?
> 
> Martin
> 
> panic: malloc: out of space in kmem_map
> Stopped in pid 4255.1 (mtree) at        netbsd:cpu_Debugger+0x4:        or              %
You have something that eats (malloced) memory from the kmem_map.
vmstat -m tells how much eaten and by what, but when it already has
paniced it's a little bit too late :-)  Try to "call dump_kmemstats"
from ddb to see where the memory has disappeared. Note that KMEMSTATS
must be configured in your kernel.

If the kmem usage is legitime, you can increase vm.kmempages to a
higher level or compile the kernel with NKMEMPAGES set to a 
higher value.

-- Ragge
Permalink | Reply | 4 comments |
headers
Dmitri Kondratiev | 2 Dec 2004 11:40
Picon
Favicon

Where to get pre-built SPARC packages?

Link:
ftp://ftp.netbsd.org/pub/NetBSD/packages/1.6.2/sparc/

is broken.

Thanks!

=====
Dmitri Kondratiev,
d_kondr <at> yahoo.com
http://www.geocities.com/dkondr/

Principal Architect, R&D
Luxoft Labs
IBS group of companies
dkondratiev <at> luxoft.com
http://www.luxoftlabs.com

		
__________________________________ 
Do you Yahoo!? 
All your favorites on one personal page – Try My Yahoo!
http://my.yahoo.com
Permalink | Reply | 3 comments |

Gmane