Linux-Development-Sys Digest #765, Volume #6 Wed, 2 Jun 99 13:14:13 EDT
Contents:
.h files missing for make config in Redhat (Jonathan DelStrother)
Re: You can earn $50,000 40686 (Jochem Huhmann)
Strange partial write on sockets problem (Tony Scholes)
Re: *Very* large physical buffers. (Robert Kaiser)
Re: What are the differences between mySQL and mSQL? (Arnout Kuiper)
Re: make xconfig errors ("Arkadiy Korobeyko")
Re: Problem with kernel 2.2.9? (Pat Crean)
Re: Linux Device Driver Overflow? (Joe Pfeiffer)
Re: Large (17 GB) hdd and fdisk (Rob van Nieuwpoort)
Re: kernel vs egcs vs PentiumPro/II (Conrad Sanderson)
Re: RAID-1 and 2.2.9 revisited (Az0th)
Modems / dial up server ("steve")
Strange behaviour of LINUX SMP under pvm (Axel Farr)
Re: Strange partial write on sockets problem (Scott Lanning)
Update: fsck (SIGSEGV) but not fsck.ext2 (Adrian 'Dagurashibanipal' von Bidder)
----------------------------------------------------------------------------
From: Jonathan DelStrother <[EMAIL PROTECTED]>
Subject: .h files missing for make config in Redhat
Date: Wed, 02 Jun 1999 09:06:45 +0100
Hiya...
I've been trying to persuade sound support to work in Redhat 5.2 - go to
the src/linux/drivers/sound dir & typing make config...but I get this
error...
configue.c: 19: stdio.h: No such file or directory
configue.c: 19: stdio.h: No such file or directory
configue.c: 19: stdio.h: No such file or directory
configue.c: 19: stdio.h: No such file or directory
------------------------------
From: [EMAIL PROTECTED] (Jochem Huhmann)
Crossposted-To:
comp.lang.java.programmer,comp.lang.javascript,comp.lang.perl.misc,comp.lang.tcl,comp.mail.eudora.ms-windows,comp.os.linux.advocacy,comp.os.linux.hardware,comp.os.linux.misc,comp.os.linux.networking
Subject: Re: You can earn $50,000 40686
Date: 1 Jun 1999 14:53:42 GMT
In article <7iuvt0$qb1$[EMAIL PROTECTED]>,
"Kent Dahl" <[EMAIL PROTECTED]> writes:
> Gah! I hate it when they don't have a valid reply adress... So much FUN to
> send a reply, filling their e-mail-boxes as they do with ours.... Now, where
> is that virtual memory file and how do I attach it to outgoing e-mail? ;-))
mail [EMAIL PROTECTED] < /dev/mem
(I don't recommend mail bombing, it does more harm to others than to
it's target)
--
Hi! I'm a .signature virus! Copy me into your ~/.signature to help me spread!
------------------------------
Date: Wed, 02 Jun 1999 10:18:13 +0200
From: Tony Scholes <[EMAIL PROTECTED]>
Subject: Strange partial write on sockets problem
Hi
Environment: Redhat 5.1/5.2, gcc 2.7.2.3, glic 2.0.7
We have a simple program that opens a socket connection, reads the
contents of a file and then writes it to the socket, it's role in life
is to enable us to glue printers on serial terminal servers (such as
ShivaPort ATOMS) to the printing system on a UNIX/Linux server... We
make the print system output to a named pipe then have this program sit
there forever reading from the named pipe and writing to the socket on
the terminal server...The socket on the terminal server is associated
with one of it's serial ports, and anything it receives on the socket,
it sends to the serial port...
Works fine everywhere (and I mean everywhere, name a UNIX system over
the last 12 years and it's worked there) apart from on Red Hat Linux
that is.....
The system we get is that the tail of the print job is sometimes lost
(not for every job, but if a particular job fails, it always fails)..
tracing the code, it appears that the write of the data to the socket
returns the correct number of bytes written, but only the 1st so many
bytes appear to arrive at the remote socket...!?
Example, write writes 1338 bytes, remote socket receives only 684, and
this is consistent... We can trace on the terminal server how many bytes
it sends down the serial wire (and thus how many it has received on the
socket)
Scary thing is, if I use gdb to trace the code, it works, though not if
I run the binary generated with -g outside of of gdb.....1
Any ideas whatsoever, as to what might cause this? Some kind of timing
issue?
TIA
--
Tony Scholes
Technical Manager
===============================================================================
Beacon Computer Services Tel: +44 (0)1582 478888
The Friars, 82 High Street South Fax: +44 (0)1582 478810
Dunstable, Beds. UK Email: [EMAIL PROTECTED]
LU6 3HD Compuserve: 72660,207
===============================================================================
------------------------------
From: [EMAIL PROTECTED] (Robert Kaiser)
Subject: Re: *Very* large physical buffers.
Date: 2 Jun 1999 08:35:18 GMT
In article <[EMAIL PROTECTED]>,
Marius Vollmer <[EMAIL PROTECTED]> writes:
> For convenience, we would like to have a 8MB large DMA buffer in
> physical memory (that is, it should have continues physical
> addresses). Our WindowsNT box wont give it to us, of course.
>
Have you considered using scatter/gather DMA ?
> Now, I guess this is due to the fragmentation of the physical address
> space. But, as long as there is no fragmentation or memory shortage
> due to competing requests for *physical* addresses, the system should
> be able to simply reshuffle its mapping of physical pages to virtual
> addresses to make room for the large physical buffer.
Of course, if you have a freshly-booted system with lots of memory,
you will surely have big contiguous chunks available, but as the system
starts doing work, fragmentation will increase. Note that this is
normally not really a problem as Linux uses virtual memory for anything
besides DMA, so it can always use the MMU to build large virtually
contiguous blocks of memory, no matter how fragmented physical memory is.
>
> That is, the very mechanisms of the virtual memory machine should make
> it possible to fabricate large continues areas of physical addresses.
This would mean some kind of "garbage collection" built into the
allocation routine. It is probably doable, but I don't think there
are any provisions in the current kernel for it. The only application
where such a thing would make sense is DMA, so I think nobody found
it to be worthwile. Also, this approach has two inherent problems:
1) You can't guarantee that it will succeed
2) It has an unpredictable overhead (reshuffling pages
might mean you have to copy the contents of a physical
page to another)
>
> Does that make sense? How does Linux go about allocating physical
> memory for DMA transfers?
>
There are several approaches. The one that seems to best fit your
requirements is Matt Walsh's "bigphysarea" patch. (Sorry no URL, but
"bigphysarea" is a good search expression ;-)) It allows you
to preallocate a large piece of physical memory at boot time.
That memory is then gone from the kernel's point of view and
you can use it for DMA.
The normal way (i.e. without patching kernels) to handle DMA in
Linux is to use buffers in kernel space and then let the CPU
copy the data into user space with copy_*user_()/memcpy_*fs().
You'll find plenty of examples for this in the drivers.
There also is a way to DMA directly into user space using
scatter/gather DMA (i.e. similar to what NT supports). It also
requires a patch which is available from ftp://ftp.sysgo.de/pub/Linux
Hope this helps
Rob
================================================================
Robert Kaiser email: rkaiser AT sysgo DOT de
SYSGO RTS GmbH
Mainz / Germany
------------------------------
Crossposted-To:
comp.os.linux.development.apps,comp.os.linux.misc,comp.os.linux.setup,comp.lang.java.databases
Date: Wed, 2 Jun 1999 11:47:16 +0200
From: Arnout Kuiper
Subject: Re: What are the differences between mySQL and mSQL?
Hi,
On Wed, 2 Jun 1999 03:10:35 +0200 , Ruiming Chen <[EMAIL PROTECTED]>
wrote:
>
>The Subject askes its all. Are they the same free database software with
>two names?
>Or they are two different free database software?
>Are they both run on Linux?
>
>Thank you!
>--Raymond
They are two different databases:
mSQL (Mini SQL) : http://www.hughes.com.au/
mySQL : http://www.tcx.se/
They both can run under Linux, and are alike with respect to offered
functionality, although I heard that MySQL is slightly better.
You might consider PostgreSQL (http://www.postgresql.org/), which is
in my humble opinion one of the better free databases.
Regards,
Arnout Kuiper ([EMAIL PROTECTED])
Sun Certified Java Developer
Sun Certified Java Programmer
TekMetrics Certified Master Java Programmer
Remove the nospam. and the .nospam to get my E-mail address
------------------------------
From: "Arkadiy Korobeyko" <[EMAIL PROTECTED]>
Subject: Re: make xconfig errors
Date: Wed, 2 Jun 1999 13:57:43 +0300
Arkadiy Korobeyko <[EMAIL PROTECTED]> wrote in message
news:7j0um1$[EMAIL PROTECTED]...
> Hi everybody!
>
> At some stage of updating packages (RH5.2 -> 6.0) I lost possibility
> in compiling kernel with 'make xconfig'
> Errors are:
>
> tkparse.c:37: stdio.h: No such file or directory
> tkparse.c:38: stdlib.h: No such file or directory
> tkparse.c:39: string.h: No such file or directory
> [tkparse.o] Error 1
>
> What is missing ?
>
Already found =- glibc-devel
Arkadiy
------------------------------
From: Pat Crean <[EMAIL PROTECTED]>
Subject: Re: Problem with kernel 2.2.9?
Date: Wed, 2 Jun 1999 07:37:52 -0400
Did you remember to include 'Prompt for development and/or incomplete
code/drivers'?
On Wed, 02 Jun 1999, David den Boer wrote:
>I just compiled kernel 2.2.9 with the DOS fix tonight (I was running 2.2.5 that came
>with RedHat 6), and I did not see any options for my ethernet card when in
>menuconfig. I have a RealTek 8129/8139 (I think) card, and this was an option in
>2.036, and it did work out of the box from RedHat.
>
>Am I missing something here?
>
>David.
------------------------------
From: Joe Pfeiffer <[EMAIL PROTECTED]>
Subject: Re: Linux Device Driver Overflow?
Date: 01 Jun 1999 23:02:53 -0600
"Jørgen Hermanrud Fjeld" <[EMAIL PROTECTED]> writes:
>
> I was wondering how the Linux kernel would be affected by severe
> popularity and the following huge amount of device drivers? Does the
> people who currently work on and coordinate the kernel have the time?
> Would this undermine future kernel work?
>
> Are there alternative architectural approaches to allowing loads of
> device drivers without overloading the kernel people?
Yes, and Linux does it: kernel modules. A driver can be developed by
somebody who otherwise has not been involved with the kernel, and
loaded at runtime. I wrote a driver for the Compaq Concerto pen; it
never got inserted into any of the distributions (originally because I
didn't get around to submitting it to them, and later because the
concerto is pretty much an obsolete product), and none of the kernel
development people ever had to put any time into it.
> Could a more microkernel like path of development open for more and
> easier device drivers for Linux?
I'm of the opinion that loadable modules meet many of the advantages
originally ascribed to microkernel OSs -- particularly when it turns
out that Minix (for example) puts all the drivers in the kernel space
and don't provide the protection against driver bugs causing OS
crashes that were originally claimed.
--
Joseph J. Pfeiffer, Jr., Ph.D. Phone -- (505) 646-1605
Department of Computer Science FAX -- (505) 646-1002
New Mexico State University http://www.cs.nmsu.edu/~pfeiffer
------------------------------
From: Rob van Nieuwpoort <[EMAIL PROTECTED]>
Subject: Re: Large (17 GB) hdd and fdisk
Date: Wed, 2 Jun 99 15:28:39
Konrad Mieredorff <[EMAIL PROTECTED]> wrote:
<snip>
: I had problems with 2.0.x kernels on a harddisk with 1027 cylinders
: (fdisk reported similar errors as you got and cfdisk aborted). Switching
: to 2.2x helped.
I didn't mention it explicitely, but I tried this with Red Hat 6.0,
which uses 2.2.5...
--
Rob ([EMAIL PROTECTED])
------------------------------
From: [EMAIL PROTECTED] (Conrad Sanderson)
Crossposted-To: comp.os.linux.development.apps
Subject: Re: kernel vs egcs vs PentiumPro/II
Date: 2 Jun 1999 05:09:47 GMT
David Wragg <[EMAIL PROTECTED]> wrote:
>With egcs-1.1.2 (the one shipping in redhat-6.0) I have code that dies
>with -march=pentiumpro (specifically, the current XFree86-3.9 source
>tree). I haven't tried build the kernel with -march=pentiumpro yet.
>
>If you don't mind tracking down code generation errors, there isn't a
>great deal of risk with -march=pentiumpro, and if you do track down a
>problem it might be useful to the egcs people. Somebody has to run
>into the bugs for them to get fixed.
okay. I now have a pentiumpro "optimized" kernel running for
about a day without any problems so far. I've added an extra
option (-fno-strict-aliasing) since it was recommended on
the Kernel Notes site.
I'll let it run for a month and let you guys know if I encounter
any "issues".
>> Is there any real reason for the kernel compile options for M586/M686 to
>> still have -m486, or is this a gcc 2.7.x compatability thing ?
>
>This has been changed in the 2.3 branch. It now uses -m586 and -m686
>as appropriate.
The kernel makefile also has -fno-strength-reduce, which is
necessary to avoid bugs in gcc 2.7.x. Does this still apply
to egcs ?
--
Conrad Sanderson - Microelectronic Signal Processing Laboratory
Griffith University, Queensland, Australia
http://hive.me.gu.edu.au/
------------------------------
From: [EMAIL PROTECTED] (Az0th)
Subject: Re: RAID-1 and 2.2.9 revisited
Date: Wed, 2 Jun 1999 14:31:55 GMT
Hi Bill,
: After investing about five more hours in trying to make RAID-1 work in
: 2.2.9, I am pretty sure that (a) my original thought was correct, it
: doesn't, and (b) the documentation is in a state of confusion, since
: the man pages, mini-HOWTO and HOWTO reflect different visions of what's
: happening, none of which seems to completely match the code.
In a word: yup. For _current_ wisdom re RAID in Linux, the mailing list
is the best resource there is:
http://linuxwww.db.erau.edu/mail_archives/linux-raid/
The short version: the current RAID development including the raidtools
works fine up to 2.2.6, and the RAID patches can be applied to 2.2.7
with one manual addition. 2.2.8/9 are broken WRT RAID in a way which
causes resync to fail, and should not be used until this is resolved.
RAID support in the distribution kernels is out of sync with current
development. This is expected to change in the not too distant future.
RedHat 6.0 is distributed with a 2.2.5 kernel which already has the
latest RAID patches applied, and the latest raidtools installed. It
isn't possible, as far as I know, to install RH6 onto a RAID, although
you can obviously configure a RAID after the initial install. It is
also not possible to have the RH6 root directory on a RAID with the
default initrd images, which will not contain the necessary raid?.o
module(s). (Since LILO doesn't understand md devices, it's necessary
to have a non-RAID /boot partition in order to use root RAID, and
for the RAIDs to be started at boot, it's also necessary that they
be created with persistent superblocks, and that the member partitions
have the proper 'fd' partition type. And: don't RAID your swap(s)!)
I mention RH6 because I just finished setting up a root RAID1 on a
friend's RH6 system, which is running 2.2.7 with the RAID patches
applied, but with RAID1 compiled into the kernel, eliminating the
need for an initrd. I did modify the 2.2.5 initrd images so that
he could boot them if he so chose (basically: gunzip the image,
use losetup to attach it to a loop device, mount the loop device,
modify the image by adding the raid module(s) to lib/ and fixing
linuxrc to load it/them, dismount, detach, gzip and done). There
are 4 md (/, /usr, /usr/local, /usr/src) plus 2x128M swap partitions
and 2x16M /boot partitions on a pair of 10-gig, 7200rpm Deskstars,
and the RAIDs start, run and shutdown flawlessly. Quick, too. ;-)
Cheers,
RF Buchanan
--
"There's certainly a strong case for people really disliking Microsoft,
and most of that is because their operating systems really suck."
Linus Torvalds
------------------------------
From: "steve" <[EMAIL PROTECTED]>
Subject: Modems / dial up server
Date: Wed, 2 Jun 1999 09:03:23 -0700
Can anyone give me some direction on where to start for modem pools and
sharing??? I have a Comtrol 8 port PCI modem card w/ high speed serial
ports and would like to create a mini ISP just for fun. I have 8 USR v.90
modems and 5 spare lines to work with. My bandwidth includes a megabit line
DSL. Any suggestions would be appreciated. If you would like to
contact me, email me & i'll provide you with my 800 number to call free of
charge.
thanks
-steve
------------------------------
From: Axel Farr <[EMAIL PROTECTED]>
Subject: Strange behaviour of LINUX SMP under pvm
Date: Wed, 02 Jun 1999 16:58:54 +0200
Since some time, I have noticed a strange behaviour of my SMP-Linux when
running pvm-programs and exceeding the ammount of physical memory. As
soon as swapping started, the system seemed to block a CPU - the system
cpu usage was 50% or 100% permanently. Under the 2.2.x kernels, the
behaviour vanished when the processes using the pvmd were killed, under
2.0.36 one CPU was left permanently blocked until reboot.
Just for testing I changed my BIOS setup: From MPP specification 1.4 to
the version 1.1. And: the blocking of the CPU was gone !
(I did not expect Linux to work with MPP 1.1 at all - the more I wonder
why it behaves much better with the older specification than with the
newer).
I now use kernel 2.2.9, pvm 3.4 beta 6 on a Gigabyte 6x86 BDD board.
Has someone also made such experiences ? Maybe the MPP 1.4 support in
the Gigabyte BIOS is broken.
Greetings, Axel Farr
------------------------------
From: [EMAIL PROTECTED] (Scott Lanning)
Subject: Re: Strange partial write on sockets problem
Date: 2 Jun 1999 15:40:43 GMT
Tony Scholes ([EMAIL PROTECTED]) wrote:
: Works fine everywhere (and I mean everywhere, name a UNIX system
: over the last 12 years and it's worked there) apart from on Red
: Hat Linux that is.....
Presumably you mean any Linux, as it seems any distribution with
the same kernel and glibc would have same behavior.
: Example, write writes 1338 bytes, remote socket receives only 684,
: and this is consistent... We can trace on the terminal server how
: many bytes it sends down the serial wire (and thus how many it has
: received on the socket)
But perhaps there is a serial driver between sockets, so bytes sent
doesn't equal bytes received at socket.
: Scary thing is, if I use gdb to trace the code, it works, though
Don't let the penguin smell the fear.. :) I think gdb probably has
some extra control over buffering.
: Any ideas whatsoever, as to what might cause this? Some kind of
: timing issue?
Or maybe a buffering issue. Quick brainstorm:
1) serial port received all 1338 bytes, but socket only received
684 of them before closing (for some reason?)
2) buffer overflowed, extra characters lost. In particular, there
may be some kind of MAX_BUF_LIMIT which is different in Linux than
in other Unixes (seems more likely, no?)
--
Scott Lanning: [EMAIL PROTECTED], http://physics.bu.edu/~slanning
"If lightning is the anger of the gods, the gods are concerned mostly
with trees." --Lao Tse
------------------------------
From: [EMAIL PROTECTED] (Adrian 'Dagurashibanipal' von Bidder)
Crossposted-To: comp.os.linux.setup,comp.os.linux
Subject: Update: fsck (SIGSEGV) but not fsck.ext2
Date: Wed, 02 Jun 1999 14:51:38 GMT
Hullo!
No, my problem is _still_ not saved.
Summary:=20
-> two IBM 351010 on Asus P5A (Aladdin Chipset)
-> first disk works like a charm.
-> fsck on second disk gives SIGSEGV - always.
Since posting originally (see below) I tried several things, all of
which didn't change anything:
- swapped disks (floppy boot, dd from hda to hdc, then swap cables)
- master/slave on primary instead of master on primary/secondary
- egcs (stock SuSE6.1) kernel instead of gcc 2.7.2.3 (still 2.2.9)
- badblocks showed nuthing.
- heavy disk abusing over several hours (three or four tasks
reading/writing at random over the whole disk, raw device) showed no
errors and no IDE timeouts or similar (which I had quite a lot on my
old T2P4/Cyrix system, but without any effects on reliabilyty, btw)
So, I swear it's not the hardware. It's fscking fsck...
[Waitaminute]
I tried calling fsck.ext2 directly --> it works. It's not beautiful,
but not really a problem. We now know the fsck frontend is broken -
Hello maintainer, are you copying?
On Tue, 01 Jun 1999 08:59:15 GMT, [EMAIL PROTECTED] (Adrian
'Dagurashibanipal' von Bidder) wrote:
>Hullo out there!
>
>This is quite urgent - I would appreciate any help greatly!
>
>I have an Asus P5A (ALI Chipset) with two IBM DTTA 351010 (5400rpm,
>10G) set up as masters of both IDE controllers. The Kernel does
>switch off BM-DMA at boot - don't ask me why.
>
>Everything on the first disk works just fine (partitions are:
> 128M swap
> 2G /
> 4G /home
> 2G /var
> 2G /var/log
>)
>
>But everything I do on the second harddrive, fsck just segfaults on
>me. I can fdisk it, mke2fs it, mount it - works just fine. But once
>I start an fsck -> segfault.
>
>I tried changing the mapping (LBA with 1232/255/63 or 'NORMAL' with
>19650(?)/16/63) per LILO-append (and fdisking anew afterwards) (FYI:
>first disk uses LBA, 1232/255/63). I tried different partition
>sizes. I tried using exactly the same filesystems as on the first
>disk. segfault on fsck.
>
>Since this will be running as a server, a working fsck is quite
>important to me (it'll be some user data on there - if it were news
>spool...ln -s /sbin/mke2fs /sbin/fsck :-)
>
>Some system data:=20
> CPU: AMD K6-2 333MHz
>
> Distro: SuSE 6.1 (fsck 1.14 from stock SuSE rpm; tried
>rebuilding with both egcs and gcc-2.7.2.3, though)
> Kernel: 2.2.9 (gcc-2.7.2.3) or stock SuSE IDE-Kernel 2.2.5
> Greets from over there
> Dagurashibanipal
> [EMAIL PROTECTED]
Greets from over there
Dagurashibanipal
[EMAIL PROTECTED]
--
Hi! I'm a .signature virus!=20
Copy me into your ~/.signature to help me spread!
------------------------------
** FOR YOUR REFERENCE **
The service address, to which questions about the list itself and requests
to be added to or deleted from it should be directed, is:
Internet: [EMAIL PROTECTED]
You can send mail to the entire list (and comp.os.linux.development.system) via:
Internet: [EMAIL PROTECTED]
Linux may be obtained via one of these FTP sites:
ftp.funet.fi pub/Linux
tsx-11.mit.edu pub/linux
sunsite.unc.edu pub/Linux
End of Linux-Development-System Digest
******************************