Natsemi ethernet 'cable magic' fix can cause problems
Hi,
This information may be useful to anyone who uses the natsemi ethernet driver.
I have discovered that the 'do_cable_magic()' call introduced into the driver in kernels 2.4.20 and later can cause problems.
Some machines we have use the Natsemi DP83815 chip, which worked with no problems using kernel 2.4.19. Following an upgrade to 2.4.25 a few of the boxes stopped working - but most were still OK.
The problem was strange - about 30% packet loss was experienced during pings. Moving the box to a different ethernet wall outlet (but still using the same port on the switch) cured the problem. The problem therefore appeared to be cable dependant.
It would appear that the call to 'do_cable_magic()' in drivers/net/natsemi.c causes the problem to occur.
It looks as though this was added in to actually *fix* such problems... but in our case it made things worse.
The cable problems were never experienced before, and everything seems fine now that this call has been removed.
I'm not sure of the correct way to fix this problem, but I'm hoping that this information might be useful to others who encounter something similar.
Colin
Quoted 10 lines · Colin Paton
The problem was strange - about 30% packet loss was experienced during pings. Moving the box to a different ethernet wall outlet (but still using the same port on the switch) cured the problem. The problem therefore appeared to be cable dependant.
It would appear that the call to '
do_cable_magic()' indrivers/net/natsemi.ccauses the problem to occur.It looks as though this was added in to actually *fix* such problems... but in our case it made things worse.
I'm the responsible party on that one. do_cable_magic() was a direct result of a significant amount of work with National's engineers to root-cause some major problems we had had.
If I recall, the problem we experienced was when the cable was "short" (where short was < 30m, I think). What can you tell me about your cabling setup?
Willing to test? Ping me back.
Quoted 16 lines · Tim Hockin
Quoted 10 lines · Colin Paton
The problem was strange - about 30% packet loss was experienced during pings. Moving the box to a different ethernet wall outlet (but still using the same port on the switch) cured the problem. The problem therefore appeared to be cable dependant.
It would appear that the call to '
do_cable_magic()' indrivers/net/natsemi.ccauses the problem to occur.It looks as though this was added in to actually *fix* such problems... but in our case it made things worse.
I'm the responsible party on that one.
do_cable_magic()was a direct result of a significant amount of work with National's engineers to root-cause some major problems we had had.
Quoted 5 lines
If I recall, the problem we experienced was when the cable was "short" (where short was < 30m, I think). What can you tell me about your cabling setup?
Willing to test? Ping me back.
Hi Tim, Collin,
Did you made any progress on this issue?
Quoted 9 lines · Marcelo Tosatti
Quoted 5 lines
If I recall, the problem we experienced was when the cable was "short" (where short was < 30m, I think). What can you tell me about your cabling setup?
Willing to test? Ping me back.
Hi Tim, Collin,
Did you made any progress on this issue?
Never gor any reply from him.