[ECOS] rattler PB

[ECOS] rattler PB

[alix t]

Hi,

I'm using Rattler card with MPC8250.
I have build librairies with common ethernet package,
FreeBSD package and network package.
My application only calls function
init_all_network_interfaces() to start network.
When I ping the card with "ping -c 10000 192.168.0.2"
or "ping -c 10000 -s 1000 192.168.0.2" everything is
OK.
When I ping the card with "ping -c 10000 -s 13000
192.168.0.2" to send packet with more bytes, following
messages appears: 
"warning: eth_recv out of MBUFS" and after card has
strange behavior, ping often fail and BUS Error or
Illegal instuction appears.

Does anyone have already seen such behavior with
Rattler?

Alix t

_________________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com

-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

Re: [ECOS] rattler PB

[Gary Thomas]

On Wed, 2003-12-17 at 07:18, alix t wrote:
> Hi,
> 
> I'm using Rattler card with MPC8250.
> I have build librairies with common ethernet package,
> FreeBSD package and network package.
> My application only calls function
> init_all_network_interfaces() to start network.
> When I ping the card with "ping -c 10000 192.168.0.2"
> or "ping -c 10000 -s 1000 192.168.0.2" everything is
> OK.
> When I ping the card with "ping -c 10000 -s 13000
> 192.168.0.2" to send packet with more bytes, following
> messages appears: 
> "warning: eth_recv out of MBUFS" and after card has
> strange behavior, ping often fail and BUS Error or
> Illegal instuction appears.
> 
> Does anyone have already seen such behavior with
> Rattler?

First of all, I doubt that this problem is unique to the Rattler.
What I think you are seeing is the stack failing to handle very
large IP packets (in particular, properly releasing the memory
when they've been processed).

As for the BUS Error and/or Illegal instruction, I tried this on
a Rattler and did not observe them at all.

Why are you trying to send such huge packets anyway?

-- 
Gary Thomas <gary@mlbassoc.com>
MLB Associates


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

Re: [ECOS] rattler PB

[alix t]

We try to develop applications with goahead web server
and other network services on rattler card , but our
applications often crash.
So we want first validate ethernet and IP stacks.
I think to send very large packets is a good test to
validate ethernet stack.

alix
 --- Gary Thomas <gary@mlbassoc.com> a écrit : > On
Wed, 2003-12-17 at 07:18, alix t wrote:
> > Hi,
> > 
> > I'm using Rattler card with MPC8250.
> > I have build librairies with common ethernet
> package,
> > FreeBSD package and network package.
> > My application only calls function
> > init_all_network_interfaces() to start network.
> > When I ping the card with "ping -c 10000
> 192.168.0.2"
> > or "ping -c 10000 -s 1000 192.168.0.2" everything
> is
> > OK.
> > When I ping the card with "ping -c 10000 -s 13000
> > 192.168.0.2" to send packet with more bytes,
> following
> > messages appears: 
> > "warning: eth_recv out of MBUFS" and after card
> has
> > strange behavior, ping often fail and BUS Error or
> > Illegal instuction appears.
> > 
> > Does anyone have already seen such behavior with
> > Rattler?
> 
> First of all, I doubt that this problem is unique to
> the Rattler.
> What I think you are seeing is the stack failing to
> handle very
> large IP packets (in particular, properly releasing
> the memory
> when they've been processed).
> 
> As for the BUS Error and/or Illegal instruction, I
> tried this on
> a Rattler and did not observe them at all.
> 
> Why are you trying to send such huge packets anyway?
> 
> -- 
> Gary Thomas <gary@mlbassoc.com>
> MLB Associates
> 
> 
> -- 
> Before posting, please read the FAQ:
> http://sources.redhat.com/fom/ecos
> and search the list archive:
> http://sources.redhat.com/ml/ecos-discuss
>  

_________________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com

-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

Re: [ECOS] rattler PB

[Andrew Lunn]

On Thu, Dec 18, 2003 at 09:22:47AM +0100, alix t wrote:
> We try to develop applications with goahead web server
> and other network services on rattler card , but our
> applications often crash.
> So we want first validate ethernet and IP stacks.
> I think to send very large packets is a good test to
> validate ethernet stack.

Its not a realistic test. I doubt any application will use such large
IP packets. UDP has a maximum packet size of around 4Kbytes. TCP will
generally use either around 800 bytes or around 1500 bytes depending
on if MTU discovery is being used. 

Only a custom application which is not using UDP or TCP can generate
such big IP packets. Typically this would be for connecting super
computers together using HIPI interfaces. Thats why IPv6 has a jumbo
packet which can be bigger than 64Kbytes. 

I suggest you try more realistic tests.

  Andrew

-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

Re: [ECOS] rattler PB

[Slawek]

Andrew Lunn wrote:

> On Thu, Dec 18, 2003 at 09:22:47AM +0100, alix t wrote:
> > We try to develop applications with goahead web server
> > and other network services on rattler card , but our
> > applications often crash.
> > So we want first validate ethernet and IP stacks.
> > I think to send very large packets is a good test to
> > validate ethernet stack.
> 
> Its not a realistic test. I doubt any application will use such large
> IP packets. UDP has a maximum packet size of around 4Kbytes. TCP will
> generally use either around 800 bytes or around 1500 bytes depending
> on if MTU discovery is being used. 
> 
> Only a custom application which is not using UDP or TCP can generate
> such big IP packets. Typically this would be for connecting super
> computers together using HIPI interfaces. Thats why IPv6 has a jumbo
> packet which can be bigger than 64Kbytes. 
> 
> I suggest you try more realistic tests.

Keep in mind that he has tried to send big packets *to* the device,
*not* write a special program for the device which failed when
sending those.

If such packets causes the device to crash, then the *final* product
could be *crashed* as well, by accident or intentionally.


-- 
Slawek



-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

Re: [ECOS] rattler PB

[Nick Garnett]

"Slawek" <sgp@telsatgp.com.pl> writes:

> Andrew Lunn wrote:
> 
> > On Thu, Dec 18, 2003 at 09:22:47AM +0100, alix t wrote:
> > > We try to develop applications with goahead web server
> > > and other network services on rattler card , but our
> > > applications often crash.
> > > So we want first validate ethernet and IP stacks.
> > > I think to send very large packets is a good test to
> > > validate ethernet stack.
> > 
> > Its not a realistic test. I doubt any application will use such large
> > IP packets. UDP has a maximum packet size of around 4Kbytes. TCP will
> > generally use either around 800 bytes or around 1500 bytes depending
> > on if MTU discovery is being used. 
> > 
> > Only a custom application which is not using UDP or TCP can generate
> > such big IP packets. Typically this would be for connecting super
> > computers together using HIPI interfaces. Thats why IPv6 has a jumbo
> > packet which can be bigger than 64Kbytes. 
> > 
> > I suggest you try more realistic tests.
> 
> Keep in mind that he has tried to send big packets *to* the device,
> *not* write a special program for the device which failed when
> sending those.
> 
> If such packets causes the device to crash, then the *final* product
> could be *crashed* as well, by accident or intentionally.

I usually test any ethernet interface I am working on by flood
pinging it with large packets. The TCP/IP stack is quite robust enough
to handle this. Of course there may be issues in the device driver,
and this sometimes causes the stack to run out of mbufs. It is very
unlikely that doing this would actually cause a crash.

-- 
Nick Garnett                    eCos Kernel Architect
http://www.ecoscentric.com      The eCos and RedBoot experts


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss

Re: [ECOS] rattler PB

[Slawek]

Nick Garnett wrote:

> "Slawek" <sgp@telsatgp.com.pl> writes:
> 
> > Andrew Lunn wrote:
> > 
> > > On Thu, Dec 18, 2003 at 09:22:47AM +0100, alix t wrote:
> > > > We try to develop applications with goahead web server
> > > > and other network services on rattler card , but our
> > > > applications often crash.
> > > > So we want first validate ethernet and IP stacks.
> > > > I think to send very large packets is a good test to
> > > > validate ethernet stack.
> > > 
> > > Its not a realistic test. I doubt any application will use such large
> > > IP packets. UDP has a maximum packet size of around 4Kbytes. TCP will
> > > generally use either around 800 bytes or around 1500 bytes depending
> > > on if MTU discovery is being used. 
> > > 
> > > Only a custom application which is not using UDP or TCP can generate
> > > such big IP packets. Typically this would be for connecting super
> > > computers together using HIPI interfaces. Thats why IPv6 has a jumbo
> > > packet which can be bigger than 64Kbytes. 
> > > 
> > > I suggest you try more realistic tests.
> > 
> > Keep in mind that he has tried to send big packets *to* the device,
> > *not* write a special program for the device which failed when
> > sending those.
> > 
> > If such packets causes the device to crash, then the *final* product
> > could be *crashed* as well, by accident or intentionally.
> 
> I usually test any ethernet interface I am working on by flood
> pinging it with large packets. The TCP/IP stack is quite robust enough
> to handle this. Of course there may be issues in the device driver,
> and this sometimes causes the stack to run out of mbufs. It is very
> unlikely that doing this would actually cause a crash.


Well, I hope it would not crash as I plan to use eCos in one more
project soon and I'd need Ethernet support there.


The cite below doesn't look promising...


alix t> "warning: eth_recv out of MBUFS" and after card has
alix t> strange behavior, ping often fail and BUS Error or
alix t> Illegal instuction appears.


Well, anyway. I'm going to make some tests of my own.
Maybe it's target specific or so.


-- 
Slawek



-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss