[ECOS] Redboot, TFTP and GDB...
Fabrice Gautier
Fabrice_Gautier@sdesigns.com
Tue Oct 17 11:37:00 GMT 2000
> -----Original Message-----
> From: Gary Thomas [ mailto:gthomas@redhat.com ]
> Sent: Monday, October 16, 2000 8:44 AM
> To: Fabrice Gautier
> Cc: Ecos-List (E-mail)
> Subject: RE: [ECOS] Redboot, TFTP and GDB...
>
>
>
> On 13-Oct-2000 Fabrice Gautier wrote:
> > Hi,
> >
> > So now I can boot redboot from a floppy on my target, I can
> connect via the
> > serial port and telnet.
> > But I can't download via TFTP. I always get:
> >
> > RedBoot> load -v hello.srec
> >|
> > Error reading header via TFTP: operation timed out
> >
>
> Some TFTP servers are pickier than others. Sometimes you
> need to give the
> complete pathname (e.g. on Red Hat 7). Other times, you may
> need to specify
> the file name in some other "absolute" fashion, e.g.
> load /hello.srec
I've done some progress but still got some problems with TFPT loading: now
the most current message is:
RedBoot> eth_drv_recv: packet of 558 bytes dropped
followed by a dump of the packet.(the size is not always 558)
I've set 8 network packet buffer in the redboot configuration (it's 4 by
default) but it is not significantly better.
On the other hand now the download via gdb is a lot better: 240kb/s and even
280 when I disconnect the host and the target from the whole network.
However 280kb/s is still quite slow for 2 machine connected via a 100Mbps
link.
It seems like redboot (or the ethernet driver) can't handle the speed.
> Try turning on network debug (you may need to force this
> because you don't
> have flash/fconfig). This will show you packets as they go
> in/out of your
> machine, which may help identify the problem.
>
> > I tried to download with gdb using remote/tcp and it would
> download but as
> > slow as 7kb/s. (that's slower than serial ....)
> >
> > Any clues ?
>
> Have you run any of the eCos networking tests using your
> driver? It would
> be nice to see how they fare, especially the throughput tests.
I tried, the server test works fine, but I have some problem building the
throughput tests for cygwin. When I try to build nc_test_master, gcc can't
find the net/route.h include file nor the timersub function. So i commented
those two line, it compiled fine but now i don't if the result does mean
something
On the master side (cygwin + NT):
Start Network Characterization - MASTER
================== No load, master at 100% ========================
No response to command
Slave denied UDP echo [640,1024] test
No response to command
Slave denied UDP send [640,1024] test
No response to command
Slave denied UDP recv [640,1024] test
No response to command
Slave denied TCP echo [640,1024] test
No response to command
Slave denied TCP send [640,1024] test
No response to command
Slave denied TCP recv [640,1024] test
No response to command
Slave denied TCP echo [64,10240] test
No response to command
No response to command
No response to command
... Done
On the slave side (eCos target):
No load = 2528135
Set background load = 20%
Set background load = 0%
High Load[20] = 2508196 => 1%
Set background load = 20%
Set background load = 0%
High Load[40] = 2514211 => 1%
Set background load = 20%
Set background load = 0%
High Load[80] = 2512023 => 1%
Set background load = 20%
Set background load = 0%
High Load[160] = 2505066 => 1%
Set background load = 20%
Set background load = 0%
High Load[320] = 2486162 => 2%
Set background load = 20%
Set background load = 0%
High Load[640] = 2451782 => 4%
Set background load = 20%
Set background load = 0%
High Load[1280] = 2381323 => 6%
Set background load = 20%
Set background load = 0%
High Load[2560] = 2254858 => 11%
Set background load = 20%
Set background load = 0%
High Load[5120] = 1986172 => 22%
Set background load = 20%
Set background load = 0%
Load[3840] = 2122486 => 17%
Set background load = 20%
Set background load = 0%
Load[4480] = 2047120 => 20%
Set background load = 20%
Set background load = 0%
Final load[4480] = 2045619 => 20%
Start test for eth0
--
Fabrice Gautier
fabrice_gautier@sdesigns.com
More information about the Ecos-discuss
mailing list