Linux-Development-Sys Digest #665, Volume #7 Thu, 9 Mar 00 07:13:15 EST
Contents:
memcpy over pci is slow (Charlie Martin)
Re: Help! local hosts name resolving does not work (Thomas F. Drescher)
Re: Impasse with 2 SCSI controllers, kernel mods required? (Stephen Lee - Post
replies please)
Req. for info on rev. engineering an ISA (Backer) card (nobody)
Re: Req. for info on rev. engineering an ISA (Backer) card ("Ron Reaugh")
how to bind a program on a tty? ("charity")
Re: Why a file system ? ("Peter T. Breuer")
sched_setpriority lockups ("Joe N.")
Re: LILO and GRUB: where do you pick disk geometry from? (repost) (Etienne Lorrain)
Re: Why a file system ? (Atle)
cpu lists that linux can support... ("Ko Jong-Gook")
Re: System hanging with SMP-Kernel 2.2.13 (SuSE6.3)? ([EMAIL PROTECTED])
Re: Can i get a MAC address ? ([EMAIL PROTECTED])
----------------------------------------------------------------------------
From: Charlie Martin <[EMAIL PROTECTED]>
Subject: memcpy over pci is slow
Date: Wed, 08 Mar 2000 19:24:06 -0700
Within my driver's interrupt service routine, I use a memcpy_fromio() to
read
roughly 64KB from a PCI target device. The memory throughput
seemed slow, so we looked at the PCI signals on the card using
an oscope, and found that the transfer rate is something like 2 - 4
MB/s.
Looking closer, we could see that although the PCI transfers are being
done in burst mode, the time between bursts is relatively long.
The system is a Dell dual processor, with 300 MHz PPros, and I believe
the Intel 440FX chip set. I'm running the 2.2.13 kernel in SMP mode.
The target device does not support bus mastering.
Anyone have suggestions on why the transfer throughput is so slow here?
Thanks,
Charlie
------------------------------
From: Thomas F. Drescher <[EMAIL PROTECTED]>
Subject: Re: Help! local hosts name resolving does not work
Date: Thu, 09 Mar 2000 04:55:16 GMT
Hoi Juergen,
thanks to you, all checks of the checklist have been done (xth time).=20=
Since i disabled the nsswitch, i tried again using your decent=20
nsswitch.conf file. Same as before.
So, having seen numerous help requests in the news groups to that=20
subject and referring to comp.os.linux.networking 'Address resolution=20=
works only one way: IP-to-name, but not vice versa', there must be a=20=
bug in the name resolution functionality. Here: names in hosts.allow=20=
are not resolved, the /etc/hosts file is ignored. Same thing in the=20
'route' output (not using -n option!).
Well, so good so bad. I wonder which source files have to do with the=20=
subject? (ip_forward.c ??)
Regards, Thomas.
>>>>>>>>>>>>>>>>>> Urspr=FCngliche Nachricht <<<<<<<<<<<<<<<<<<
Am 04.03.00, 19:42:21, schrieb [EMAIL PROTECTED] (Juergen=20=
Heinzl) zum Thema Re: Help! local hosts name resolving does not work:
> In article <[EMAIL PROTECTED]>, Thomas =
F. Drescher wrote:
> >Hi there specialists,
> >
> >I hope some of the network programmers can help me because i'm
> >fuddling around with this problem since two weeks.. (thats annoying
> >-almost like using WINDOZE- isn't it?!?!)
> Almost, not quite but almost.
> >My SuSE 6.2 system doesn't want to resolve the local host names =96 i=
n
> >some cases.
> >
> >Question: Why is the 'route' output different now (Kernel 2.2.10)=20=
from
> >Kernel 2.0.xx ?
> >
> >1) # /sbin/route (Kernel 2.2)
> >
> >Kernel IP routing table
> >Destination Gateway Genmask Flags Metric Ref
> >Use Iface
> >192.168.111.3 * 255.255.255.255 UH 0 0
> [...]
> >2) # /delix5.1/sbin/route (Kernel 2.0)
> >
> >Kernel routing table
> >Destination Gateway Genmask Flags MSS Window
> >Use Iface
> >zeppo.drescher. * 255.255.255.255 UH 0 0
> >0 plip1
> [...]
> route -V ... different versions, no /etc/networks and resolver cannot=
> resolve on one machine but not the other.
> >
> >As u see in 2) the /etc/hosts contents are correct. DNS has been shut=
> >down (for now) so the /etc/host.conf contains:
> >
> >order hosts
> >multi on
> >
> >1st consequence/problem: also the names in hosts.allow are not
> >resolved - I can work around by putting in IP numbers 127.0.0.1,
> >192.168.111.1, ... - WHY ?
> >>Feb 24 03:31:35 groucho in.telnetd[1729]: warning: /etc/hosts.allow,=
> >>line 8: can't verify hostname: gethostbyname(groucho.drescher.net)
> >>failed
> Yes, it should be ...
> order hosts,bind
> ... and for, not only, completeness you might need a decent ...
> /etc/nsswitch.conf:
> -------------------
> [ ... stuff ........ ]
> hosts: files dns [NOTFOUND=3Dreturn UNAVAIL=3Dreturn]
> [ ... more stuff ... ]
> ... for instance. Here try /etc/hosts first, then dns and in both
> cases, not found or service not available, return.
> Make sure ...
> /etc/resolv.conf:
> -----------------
> domain seg.v.org
> nameserver 192.168.21.1 192.168.21.2 192.168.21.3
> ... is okay too.
> >2nd consequence when DNS is enabled: it always tries to dial out the=
> >ippp0 to resolve the local hosts. Hey that costs money here in .de!
> >(How much has the responsible been payed by the phone companies since=
> >i've seen that's a problem which bothers much ppl) ;-(
> By the way, /etc/hosts is world readable, isn't it ?
> >So any comments and helpful hints are appreciated! Can i find a
> >solution by recompiling the kernel with a special option? I verified=
> >that the second route output is consistent with the /proc/net/route
> >content (Strange: when i delete a route in 1) it disappears...but in=
> >/proc/net/route it is still there, also 2) displays it)
> No.
> [...]
> >Another(?) problem is the following tcpdump output - but i think it
> >already has been answered by the networking newsgroup. (No more
> >comment)
> >
> ># ping 192.168.111.4 (default -> ippp0)
> >
> >truncated-ip - 1186 bytes missing!0.84.0.216 > 0.0.64.1: (frag
> >0:1280@8192) [ttl 0]
> >truncated-ip - 1186 bytes missing!0.84.0.218 > 0.0.64.1: (frag
> >0:1280@8192) [ttl 0]
> >truncated-ip - 1186 bytes missing!0.84.0.220 > 0.0.64.1: (frag
> >0:1280@8192) [ttl 0]
> >truncated-ip - 1186 bytes missing!0.84.0.222 > 0.0.64.1: (frag
> >0:1280@8192) [ttl 0]
> >
> >truncated-ip - 16307 bytes missing!0.84.1.107 > 0.0.64.1: (frag
> >31534:16401@63488+) [ttl 0] (bad cksum 4500!, optlen=3D-20[|ip])
> >truncated-ip - 16307 bytes missing!0.84.1.113 > 0.0.64.1: (frag
> >31528:16401@63488+) [ttl 0] (bad cksum 4500!, optlen=3D-20[|ip])
> >truncated-ip - 16307 bytes missing!0.84.1.117 > 0.0.64.1: (frag
> >31524:16401@63488+) [ttl 0] (bad cksum 4500!, optlen=3D-20[|ip])
> >truncated-ip - 16307 bytes missing!0.84.1.121 > 0.0.64.1: (frag
> >31520:16401@63488+) [ttl 0] (bad cksum 4500!, optlen=3D-20[|ip])
> >truncated-ip - 16307 bytes missing!0.84.1.125 > 0.0.64.1: (frag
> >31516:16401@63488+) [ttl 0] (bad cksum 4500!, optlen=3D-20[|ip])
> >g
> Off the cuff ... cannot tell.
> Cheers,
> Juergen
> --
> \ Real name : J=FCrgen Heinzl \ no flames =
=20
/
> \ EMail Private : [EMAIL PROTECTED] \ send money instead =
/
------------------------------
From: Stephen Lee - Post replies please <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.hardware,comp.os.linux.setup
Subject: Re: Impasse with 2 SCSI controllers, kernel mods required?
Date: 9 Mar 2000 03:19:28 GMT
In article <8a4hce$1m2h$[EMAIL PROTECTED]>,
teri <[EMAIL PROTECTED]> wrote:
>
>I have compiled a new kernel (2.2.14) with both drivers built-in. On
>boot the kernel keeps looping and timing out looking for drive 0 on the
>wrong controller! (the 1542CF). It insists on doing this whether the
>1542CF bios is enabled or disabled. However, the controller is
>recognized. Is there a way to force the boot process to go to the 2940?
Is the 1542 working correctly? Does Linux boot if you connect the
drive to the 1542CF? Where does the looping occur? When loading the
1542 driver or when mounting the root filesystem? Does it ever get to
load the 2940 driver?
You can hack the kernel source and change the order the drivers are
loaded, it is not too difficult. Or you can build an initrd and load
the drivers in the order you want.
But my guess is still that you have a resource conflict somewhere.
Stephen
------------------------------
From: nobody <[EMAIL PROTECTED]>
Crossposted-To:
alt.comp.hardware,comp.os.linux.hardware,comp.sys.ibm.pc.hardware.misc,comp.sys.ibm.pc.hardware.storage
Subject: Req. for info on rev. engineering an ISA (Backer) card
Date: Wed, 08 Mar 2000 23:04:54 -0700
Hi,
I have one of Danmere's ISA Backer 16 video tape backup cards. I am
interested in writing a linux device driver for the card but Danmere
will not release programing information for the beast. :-P well bugger
them.
So... for my next project I think I'll take a crack at reverse
engineering the interface. What I am looking for is recomendations for
any software tools that would help me do this. I have a fair bit of
experience with the ISA interface from both the hardware and software
end of things so I'm not a novice but I've never hacked someone else's
work before. The card is currently in my old DOS box and I have several
little DOS utilties that came with the card that just do some basic I/O
stuff. For my first attempt I tried watching one of the programs with a
debugger but that quickly got very tedious.
Any suggestions? If nothing is forthcoming it'll be back to the
debugger for me.
Since the actual backup software that came with the card is a windows
app, how good would the WINE debugger be at doing this? I've never done
*any* windows programing so I have no idea how hardware stuff is done
i.e. whether it's all handled by the windows libraries or whether it's
done "out-of-band" straight to the machine. Maybe someone could
straighten me out on this.
Also: if anyone else out there has one of these cards and is *also*
interested in poking away at it please drop me a line and we can pool
our resources. There are amateur radio people interested in using these
things as high speed modems over ATV channels so I'm not the only one
looking programing info... you know you're out there!
-Kipp
PS --- if anyone from Danmere is reading this... c'mon guys give a guy a
break will ya? We're all in this together.
Remove the no spam stuff from my address.
------------------------------
From: "Ron Reaugh" <[EMAIL PROTECTED]>
Crossposted-To:
alt.comp.hardware,comp.os.linux.hardware,comp.sys.ibm.pc.hardware.misc,comp.sys.ibm.pc.hardware.storage
Subject: Re: Req. for info on rev. engineering an ISA (Backer) card
Date: Thu, 09 Mar 2000 06:49:01 GMT
nobody wrote in message <[EMAIL PROTECTED]>...
>Hi,
>
>I have one of Danmere's ISA Backer 16 video tape backup cards. I am
>interested in writing a linux device driver for the card but Danmere
>will not release programing information for the beast. :-P well bugger
>them.
>
>So... for my next project I think I'll take a crack at reverse
>engineering the interface. What I am looking for is recomendations for
>any software tools that would help me do this. I have a fair bit of
>experience with the ISA interface from both the hardware and software
>end of things so I'm not a novice but I've never hacked someone else's
>work before. The card is currently in my old DOS box and I have several
>little DOS utilties that came with the card that just do some basic I/O
>stuff. For my first attempt I tried watching one of the programs with a
>debugger but that quickly got very tedious.
>
>Any suggestions? If nothing is forthcoming it'll be back to the
>debugger for me.
>
>Since the actual backup software that came with the card is a windows
>app, how good would the WINE debugger be at doing this? I've never done
>*any* windows programing so I have no idea how hardware stuff is done
>i.e. whether it's all handled by the windows libraries or whether it's
>done "out-of-band" straight to the machine. Maybe someone could
>straighten me out on this.
>
>Also: if anyone else out there has one of these cards and is *also*
>interested in poking away at it please drop me a line and we can pool
>our resources. There are amateur radio people interested in using these
>things as high speed modems over ATV channels so I'm not the only one
>looking programing info... you know you're out there!
>
> -Kipp
>
>PS --- if anyone from Danmere is reading this... c'mon guys give a guy a
>break will ya? We're all in this together.
Well just publish the reversed engineered spec when your done.
------------------------------
From: "charity" <[EMAIL PROTECTED]>
Subject: how to bind a program on a tty?
Date: Thu, 9 Mar 2000 16:03:03 +0800
------------------------------
From: "Peter T. Breuer" <[EMAIL PROTECTED]>
Subject: Re: Why a file system ?
Date: 9 Mar 2000 08:25:33 GMT
d wrote:
: "Peter T. Breuer" <[EMAIL PROTECTED]> writes:
:> d wrote:
:> : /usr/lib/gcc-lib/i386-redhat-linux/egcs-2.91.66/cc1
:> : would be the same file as
:> : /gcc-lib/egcs-2.91.66/lib/i386-redhat-linux/usr/cc1
:>
:> Well, one is its primary key, and the other is a secondary key or
:> perhaps a query. There is nothing to stop you changing "cd" to something
:> which looks up all possible perms of the path components. Of course,
:> you have to change mkdir, too, to stop ypu making two distinct
:> directories with overlapping sets of components.
: This would be a mistake.
But a userspace one.
Maintaining the kind of binary structure you have in mind in kernel is a
hard and expensive job. Expensive on coders, at least!
The kernel has to be simple, quick ane elegat. Sure you can maintain an
n-way lookup in the VFS? What about error recovery? How will it impact
FS stats?
Anyway, see ReiserFS for attacks on namespaces.
Peter
------------------------------
From: "Joe N." <[EMAIL PROTECTED]>
Subject: sched_setpriority lockups
Date: Wed, 8 Mar 2000 16:44:14 -0700
I wrote a simple TCP/IP client/server program to do some timing test on my
network. It uses blocking sockets and works great with the default Linux
scheduling policy. When I try to set it to a real-time policy (SCHED_RR) at
any priority (I tried max and min) it locks up the system after the
connection is established. Isn't the scheduler supposed to give other
processes CPU time when my process blocks on a send or recv? I've also
tried it with non-blocking sockets and a usleep call. This also locks up
the system. Does anyone have any comments or suggestions? Is anyone having
similar problems? Thanks.
I'm using Red Hat 6.1 (Kernel 2.2.12-20).
Joe
------------------------------
From: Etienne Lorrain <[EMAIL PROTECTED]>
Crossposted-To: comp.os.linux.development.apps
Subject: Re: LILO and GRUB: where do you pick disk geometry from? (repost)
Date: Thu, 09 Mar 2000 09:36:08 +0000
> [EMAIL PROTECTED] wrote:
> > Etienne Lorrain wrote:
> >In article <[EMAIL PROTECTED]> you write:
> > The piece of memory not reseted you are talking of is
> > the setup of the IDE interface. Bios cannot handle
> > C/H/S=4470/15/63 so it increases number of heads and
> > decreases number of cylinders. To do this, it sends
> > and IDE command to set the _logical_ geometry of the
> > IDE disk: 558/120/63 for you. Then, you will be able
> > to boot your system.
>
> How did you arrive at the logical geometry numbers? What's the
> algorithm?
>
Unfortunately the algorithm does not exists - or in fact
it is different for every BIOS. You cannot guess these
numbers - but if you could find a trace somewhere of what
they were at boot time...
------------------------------
From: Atle <[EMAIL PROTECTED]>
Subject: Re: Why a file system ?
Date: Wed, 08 Mar 2000 12:42:12 +0100
David Fox wrote:
>
> Nicolas Boulay <[EMAIL PROTECTED]> writes:
>
> > I know it's look like a very odd question.
>
> You are right, a database would be a big improvement over the current
> organization.
I have thought about the same
> Once we eliminate the bogus ordering of we can add attribute/value
> pairs, so instead of identifying files with bogus extensions we give
> them attributes like `language=c' or `format=mp3'.
I would imagine that this would become easier if the interfaces were
moved around a little, and the
open(), read(), write(), close() became the interface to the disk
driver, and sat on the card, so the only thing the bus would transport,
would be calls to this interface and responses, and open file would
yield a file handle.
This, of course , is necessary.
The implementation would be that of a 'Table of files' :-)
And then other higher level interfaces could be added. To build a tree,
fx, would be done through this interface, not by the application.
But how do you get one million programmers who have been twiddling bits
all their lives to accept this idea?
Who decides what the interface looks like?
Also, adding a disk should provide for an automatic reorganizxation to
take advantage of the fact that the access time is now cut in half, yes,
they should work in parallel.
I see the same thing with X, the implementation of X should sit on the
graphics card.
All input should be organized the same way, speech, keyboard, mouse,
tablet, etc. .. and be directed to the graphics gard.
This would remove the biggest narrowest bottleneck there is: The bus!
See the bus as a network between the different components in a PC, and
things start to improve. I have experimented with things like these,
building virtual models of them, and I was impressed by my own things
:-)
Atle
------------------------------
From: "Ko Jong-Gook" <[EMAIL PROTECTED]>
Subject: cpu lists that linux can support...
Date: Thu, 9 Mar 2000 19:55:39 +0900
Reply-To: "Ko Jong-Gook" <[EMAIL PROTECTED]>
Hi..
I would like to know the all cpu lists that linux can support,,,in detail..
Are there any one who know the CPU lists....?
thanks you in advance..
------------------------------
From: [EMAIL PROTECTED]
Subject: Re: System hanging with SMP-Kernel 2.2.13 (SuSE6.3)?
Date: Thu, 09 Mar 2000 11:01:31 GMT
Finally, I've found another reason for this
hanging party on my site: probably SMP + Threading
in Linux. Because my Threading-using application
(database search) was running OK, if it is running
against a local copy of the database. Even if I
tried to enhance the network accesses over NFS
(copy several large files of several hundreds MB
from NFS server to a local filesystem), I've never
observerd that the network load over the
Fast-Ethernet is higher than 4.2 MB/s. On the
other side, if I run the same threading-using
application against a database mounted over NFS,
the usual network load will be kept on a very high
level (of about 3.5 - 4.2 MB/s). And if during
this search time, the network access over NFS
would be forced to go higher (for example, copy a
large file of hundred MB over NFS), it will cause
the hanging of the system. I've observed this
phenomene on different SMP systems running with
Kernel 2.2.5 or 2.2.13.
My assumption is that this might be due to the
threading/SMP problem, as my application was
running as a normal user's job;
Much more importantly, I didn't observe this
problem on SGI's servers (no hanging here by
repeating the same procedure.). And a normal
user's job shouldn't kill the Unix system, should
it?
Yours, Yan
In article <[EMAIL PROTECTED]>,
[EMAIL PROTECTED] (Dixon Ly) wrote:
> Well, I ran into similar problems after
recompiling my
> kernel to support SMP. Let me say that my
system is running
> the stock RedHat 6.1 distrib with most of the
updates using
> the gnome interface...
>
> The kernel is 2.2.12. It seems like the system
was hanging
> after periods of inactivity. So I first
disabled the screen
> lock. The system hang again after I left and
came back from
> a movie that same night.
>
> Then I disable the screensaver altogether.
Since then, the
> machine has been up and running for over three
days. At
> this point, I cannot absolutely say that the
problem has
> gone away until after a couple more weeks of
uptime. But it
> does look hopeful :-)
>
> For those who think I got power management
compiled into the kernel,
> I don't...
>
> -d
>
> In article
<89hqcb$c8v$[EMAIL PROTECTED]>,
> bill davidsen <[EMAIL PROTECTED]> wrote:
> >
> >In article
<890g54$r75$[EMAIL PROTECTED]>,
> >Yuan <[EMAIL PROTECTED]> wrote:
> >
> >| Is there anybody who knows whether there is
any problem with the SMP-kernel
> >| 2.2.13 [SuSE 6.3]?
> >| We've got a problem with this kernel on a
Dell's PowerEdge 6350
> >| system [4 x PIII Xeon 550MHz, 2GB EDO DIMM
RAM, Intel's Ether DualPort Pro100+]:
> >| it hangs sometimes.
> >| The /var/log/messages file didn't show any
hints about what caused the hanging.
> >| Usually the system could only be reusable
after rebooting (press reset-button)
> >| again (it didn't respond to pings during the
hanging or anything else,
> >| i.e.: no telnet, no ftp-> a total k.o. :-<].
> >
> > I've seen this on a distressing number of
Linux SMP systems, and Alan
> >Cox seems to say that he agrees. On the other
hand I have two systems on
> >2.2.6 and one on 2.2.10 which have been up for
44/88/69 days under heavy
> >network load. 2.2.14 seems better than 2.2.13,
and you might try Alan's
> >latest pre-release patch if you are brave.
> >
> >| Hope I can get a bit more hints from this
gurus group, ;-).
> >
> > Wrong group, lots of helpful folks here, but
the gurus are pretty
> >quiet.
> >--
> >bill davidsen <[EMAIL PROTECTED]> CTO, TMR
Associates, Inc
> > When taking small children to a carnival,
always have them go potty
> >*before* you let them go on the rides, and let
them eat all the junk
> >food and candy *after*.
>
>
Sent via Deja.com http://www.deja.com/
Before you buy.
------------------------------
From: [EMAIL PROTECTED]
Subject: Re: Can i get a MAC address ?
Date: Thu, 09 Mar 2000 11:03:39 GMT
In article <38b27e25$0$[EMAIL PROTECTED]>,
Laurent Gauthier <[EMAIL PROTECTED]> wrote:
> I'm writing a program that needs to know the MAC address of the
network
> device, I think there is a C library that I can use to do it.
> Can someone help me?
In a shell environment, there is the command "arp"
You can retrieve a MAC address calling "arp -a <ip-address>"
Within a programm either do some "popen()" or look for an appropriate
function.
To port this to NT you need to install the simple TCPIP services.
hope it helps
Sent via Deja.com http://www.deja.com/
Before you buy.
------------------------------
** 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
******************************
