overruns with kernel 2.4.19

overruns with kernel 2.4.19

[Joachim Buermann]

Hallo all,

I`m using linux mandrake kernel 2.4.19 on a 2 GHz Pentium 4 with 128
MByte RAM. The serial port /dev/cua0 is connected to a external device
and receives data with 115200 Baud.

Unfortunatelly I get a lot of fifo overruns. I have unmasked the hda
interrupts with:

hdparm -u1 /dev/hda

also set the low_latency flag in the serial driver. The harddisk dma
couldn't disabled.

I'm wonder at this behaviour, because with a lower machine (Pentium I
120 MHz, Kernel 2.2.13) I get overruns only once in a blue moon. And I
expected, that a 2 GHz machine should be quick enough, to responds to
the serial interrupts.

Additional to this I tried a PCI card with a 16950 (128 Byte fifo!) and
set the uart with the setserial command. But also here a lot of
overruns!!!

>From a embedded linux project (ELAN SC520 133MHz, kernel 2.2.19) I know,
that the harddisk type influence the serial communication. So I got
overruns after I changed the original drive with a newer one (the first
was a old IBM with 500MByte, the newer a IBM with 20GByte). Last not
least I must replace the linux kernel with a realtime linux (rtlinux)
and do the serial communication in a realtime handler.

Because I didn't want to replace my desktop machine with a realtime
linux, I'm looking for some hints to solve this problems.

What's about irqtune? I'm know this tool only with kernels up to 2.2.x.
Is there a bug in the serial driver (I didn't found any hints in the
mailing list or with google)?

Should I change the PCI interrupts in the BIOS to set the interrupt
priority?

I don't know anymore. Please help.

Many thanks in advance and sorry for my bad english

Joachim

Re: overruns with kernel 2.4.19

[David Lawyer]

On Tue, Feb 18, 2003 at 02:43:46PM +0100, Joachim Buermann wrote:
> Hello all,
> 
> I`m using linux mandrake kernel 2.4.19 on a 2 GHz Pentium 4 with 128
> MByte RAM. The serial port /dev/cua0 is connected to a external device
> and receives data with 115200 Baud.
The cua devices have been obsolete for some time.
> 
> Unfortunately I get a lot of fifo overruns. I have unmasked the hda
> interrupts with:
> 
> hdparm -u1 /dev/hda
> 
> also set the low_latency flag in the serial driver. The harddisk dma
> couldn't disabled.
> 
> I'm wonder at this behavior, because with a lower machine (Pentium I
> 120 MHz, Kernel 2.2.13) I get overruns only once in a blue moon. And I
> expected, that a 2 GHz machine should be quick enough, to responds to
> the serial interrupts.

Almost the same for me.  On a Pentium I 90 MHz I get no overruns with
kernel 2.2.20, but get hundreds of overruns on the same machine with
kernel 2.4.19.  Something is wrong with the software.
> 
			David Lawyer

RE: overruns with kernel 2.4.19

[Ed Vance]

Hi Joachim and David,

Do I need an external data source to reproduce this, or does it also fail
with a looped back 115200 baud data stream (ttyS1 -> ttyS0)? I can try it on
my 333 MHz Pentium II when it frees up in a couple of days. 

So, 2.2.20 works okay and 2.4.19 fails. Anybody have a rough idea of when
the receive data overrun behavior crept in? 

Cheers,
Ed

---------------------------------------------------------------- 
Ed Vance              serial24 (at) macrolink (dot) com
Macrolink, Inc.       1500 N. Kellogg Dr  Anaheim, CA  92807
----------------------------------------------------------------

-----Original Message-----
From: David Lawyer [mailto:dave@lafn.org]
Sent: Tuesday, February 18, 2003 3:06 PM
To: linux-serial@vger.kernel.org
Subject: Re: overruns with kernel 2.4.19


On Tue, Feb 18, 2003 at 02:43:46PM +0100, Joachim Buermann wrote:
> Hello all,
> 
> I`m using linux mandrake kernel 2.4.19 on a 2 GHz Pentium 4 with 128
> MByte RAM. The serial port /dev/cua0 is connected to a external device
> and receives data with 115200 Baud.
The cua devices have been obsolete for some time.
> 
> Unfortunately I get a lot of fifo overruns. I have unmasked the hda
> interrupts with:
> 
> hdparm -u1 /dev/hda
> 
> also set the low_latency flag in the serial driver. The harddisk dma
> couldn't disabled.
> 
> I'm wonder at this behavior, because with a lower machine (Pentium I
> 120 MHz, Kernel 2.2.13) I get overruns only once in a blue moon. And I
> expected, that a 2 GHz machine should be quick enough, to responds to
> the serial interrupts.

Almost the same for me.  On a Pentium I 90 MHz I get no overruns with
kernel 2.2.20, but get hundreds of overruns on the same machine with
kernel 2.4.19.  Something is wrong with the software.
> 
			David Lawyer
-
To unsubscribe from this list: send the line "unsubscribe linux-serial" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: overruns with kernel 2.4.19

[David Lawyer]

On Tue, Feb 18, 2003 at 04:34:34PM -0800, Ed Vance wrote:
> Hi Joachim and David,
> 
> Do I need an external data source to reproduce this, or does it also fail
> with a looped back 115200 baud data stream (ttyS1 -> ttyS0)? I can try it on
> my 333 MHz Pentium II when it frees up in a couple of days. 
> 
> So, 2.2.20 works okay and 2.4.19 fails. Anybody have a rough idea of when
> the receive data overrun behavior crept in? 

Sorry, but I reached wrong conclusions since I likely tested 2.4.19.
when a backup program was putting a heavy load on the harddisk.
For 2.2.20 there was likely no backup program running.  To check out the
situation, I just now tried out surfing the web when a tar program was
running creating a tarball of most of the drive.  The number of overruns
was roughly the same for 2.4.19 and 2.2.20: about 1/3 % of the bytes
received.  But since the FIFO buffer can hold 16 bytes, its a few
percent of each buffer fetch operation.  This seems too high.

Without tar running, there were few overruns.  This indicates that the
problem is due to the interference of the harddisk.  It seems just as
bad at the serial port speed of 57.6k as at 115.2k.  It still may be
that kernel 2.2.13 is a lot better.

> 
> -----Original Message----- From: David Lawyer [mailto:dave@lafn.org]
> Sent: Tuesday, February 18, 2003 3:06 PM To:
> linux-serial@vger.kernel.org Subject: Re: overruns with kernel 2.4.19
> 
> 
> On Tue, Feb 18, 2003 at 02:43:46PM +0100, Joachim Buermann wrote:
> > Hello all,
> > 
> > I`m using linux mandrake kernel 2.4.19 on a 2 GHz Pentium 4 with 128
> > MByte RAM. The serial port /dev/cua0 is connected to a external
> > device and receives data with 115200 Baud.
> The cua devices have been obsolete for some time.
> > 
> > Unfortunately I get a lot of fifo overruns. I have unmasked the hda
> > interrupts with:
> > 
> > hdparm -u1 /dev/hda
> > 
> > also set the low_latency flag in the serial driver. The harddisk dma
> > couldn't disabled.
> > 
> > I'm wonder at this behavior, because with a lower machine (Pentium I
> > 120 MHz, Kernel 2.2.13) I get overruns only once in a blue moon. And
> > I expected, that a 2 GHz machine should be quick enough, to responds
> > to the serial interrupts.
> 
> Almost the same for me.  On a Pentium I 90 MHz I get no overruns with
> kernel 2.2.20, but get hundreds of overruns on the same machine with
> kernel 2.4.19.  Something is wrong with the software.
> > 
			David Lawyer

Re: Overruns with kernel 2.4.19

[Joachim Buermann]

Hi Ed,


> Hi Joachim and David,
> 
> Do I need an external data source to reproduce this, or does it also fail
> with a looped back 115200 baud data stream (ttyS1 -> ttyS0)? I can try it on
> my 333 MHz Pentium II when it frees up in a couple of days. 
> 
> So, 2.2.20 works okay and 2.4.19 fails. Anybody have a rough idea of when
> the receive data overrun behavior crept in? 
> 
> Cheers,
> Ed
> 

I will check this next time and let you know what happens.

Regards
Joachim

Re: overruns with kernel 2.4.19

[Joachim Buermann]

Hello David,

On Wed, 19 Feb 2003 00:03:10 -0800
David Lawyer <dave@lafn.org> wrote:

> On Tue, Feb 18, 2003 at 04:34:34PM -0800, Ed Vance wrote:
> > Hi Joachim and David,
> > 
> > Do I need an external data source to reproduce this, or does it also fail
> > with a looped back 115200 baud data stream (ttyS1 -> ttyS0)? I can try it on
> > my 333 MHz Pentium II when it frees up in a couple of days. 
> > 
> > So, 2.2.20 works okay and 2.4.19 fails. Anybody have a rough idea of when
> > the receive data overrun behavior crept in? 
> 
> Sorry, but I reached wrong conclusions since I likely tested 2.4.19.
> when a backup program was putting a heavy load on the harddisk.
> For 2.2.20 there was likely no backup program running.  To check out the
> situation, I just now tried out surfing the web when a tar program was
> running creating a tarball of most of the drive.  The number of overruns
> was roughly the same for 2.4.19 and 2.2.20: about 1/3 % of the bytes
> received.  But since the FIFO buffer can hold 16 bytes, its a few
> percent of each buffer fetch operation.  This seems too high.
> 
> Without tar running, there were few overruns.  This indicates that the
> problem is due to the interference of the harddisk.  It seems just as
> bad at the serial port speed of 57.6k as at 115.2k.  It still may be
> that kernel 2.2.13 is a lot better.

This align with my experience on a embedded linux (AMD ELAN 130 MHz).
There was a great difference if I using a older harddisk or a newer one.
My 2.2.13 desktop machine has also older harddisk and the 2.4.19 machine
a modern harddisk, so I will do some tests at home. There are running a
2.2.13 and a 2.4.19 kernel on the same hardware (600 MHz Athlon).

Maybe it's not so a kernel problem, as a general behaviour with modern
harddisks.

But whatever, it doesn't resolve this problem :-( 

Joachim

> 
> > 
> > -----Original Message----- From: David Lawyer [mailto:dave@lafn.org]
> > Sent: Tuesday, February 18, 2003 3:06 PM To:
> > linux-serial@vger.kernel.org Subject: Re: overruns with kernel 2.4.19
> > 
> > 
> > On Tue, Feb 18, 2003 at 02:43:46PM +0100, Joachim Buermann wrote:
> > > Hello all,
> > > 
> > > I`m using linux mandrake kernel 2.4.19 on a 2 GHz Pentium 4 with 128
> > > MByte RAM. The serial port /dev/cua0 is connected to a external
> > > device and receives data with 115200 Baud.
> > The cua devices have been obsolete for some time.
> > > 
> > > Unfortunately I get a lot of fifo overruns. I have unmasked the hda
> > > interrupts with:
> > > 
> > > hdparm -u1 /dev/hda
> > > 
> > > also set the low_latency flag in the serial driver. The harddisk dma
> > > couldn't disabled.
> > > 
> > > I'm wonder at this behavior, because with a lower machine (Pentium I
> > > 120 MHz, Kernel 2.2.13) I get overruns only once in a blue moon. And
> > > I expected, that a 2 GHz machine should be quick enough, to responds
> > > to the serial interrupts.
> > 
> > Almost the same for me.  On a Pentium I 90 MHz I get no overruns with
> > kernel 2.2.20, but get hundreds of overruns on the same machine with
> > kernel 2.4.19.  Something is wrong with the software.
> > > 
> 			David Lawyer
> -
> To unsubscribe from this list: send the line "unsubscribe linux-serial" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html