Skip to contentSkip to conversation

Intel CSA Gigabit Bug in IC7-G Motherboards- Affects Windows/Linux

DISCUSSION Justin Piszcz 3 participants 10 messages Feb 19, 2006 – Feb 23, 2006

lore
Share & tools
lore
#0 lore raw

Apparently, A-BIT's Intel driver works for some people, but the Linux one causes the machine to crash. The same is also true if you use the driver from Intel's website. I was just wondering if there was any discussion on this list concerning this issue?

I have 4 ABIT IC7-G's and this problem only occurs on two of them, which I have disabled the onboard CSA gigabit nic and put in PCI NICS (intel gigabit) with no problems.

It appears there are some serious problems with Intel CSA GIGABIT with this motherboard, just curious to see if anyone out there heard anything about this?

The interesting part is the first page of this thread:

http://forum.abit-usa.com/showthread.php?t=18707

The author has a "death.zip" file, in which if you copy the file over the LAN it causes an instant crash. He's tried contacting A-Bit, but is it an A-Bit or Intel or Driver problem as it has been "reported" to work with A-Bit's driver.

I believe this post sums up the problems of IC7-G problems:

--

Yeah, can't have it all, can we? I had serious thoughts about RMA-ing my IC7-G last fall, but the 4 week tournaround was a bit too much downtime for my taste. To resolve the noisy mic-input issue, I bought a cheap soundcard. For the onboard CSA, I did nothing. Running at 1GbE, the NIC has worked very well, disregarding the numerous occations it has strangled my system to death, of course.

I feel sorry for you Sydtech, myself and everyone else out there ripping their hair out over the issues of the IC7. We have spent countless hours trying to figure out what's wrong, worrying about a lengthy RMA process and the adherent downtime - which in Sydtech's case just as well might result in a shabby refurbished MB fixing some problems, but adding others. Many of us have simply resorted to buying PCI add-in cards replacing the dysfunctional parts of the IC7.

We have paid the bill for ABIT's engineering shortcomings, not to mention wasting a lot of time and energy. Last spring I was looking for a MB for my new system. I wanted the best and with as many onboard functions as possible so I could have a less crammed case. I read all the reviews for the new i875 MB:s, where - for instance - HardOCP praised it, especially it's superiour sound quality (they did not even try the mic-input). Needless to say, I ended up with the expensive IC7-G, "cream of the crop" and bucketsful of headache.

Alright, so add-in cards cramming up our cases seem to be the easiest way to get the subsystems we've already payed for. But what's really infuriating is that ABIT just refuses to address the issues. No helping hand, no suggestions on how to get things right. They just keep ignoring us. Of course they read the forums, they just think keeping things quiet is their best strategy.

Well, anyhow, I have my strategy worked out for future technology acquisitions: I'll NEVER again buy first revision products the minute they come out. I refuse to again be an unpaid beta tester for an ungrateful and unhelpful manufacturer like ABIT. And I will also take reviews with loads of salt.

--

I was wondering if anyone from [email protected] could comment on this issue?

Thanks,

Justin.

#1 lore raw

/usr/src/linux/Documentation//networking/e1000.txt:the issue to [email protected].

^^ That e-mail is invalid, someone should make a diff/patch to remove it and update it if there is a correct one or just remove it entirely.

Justin.

As of 2.6.15.3 source.

Quoted 34 lines · Justin Piszcz

Apparently, A-BIT's Intel driver works for some people, but the Linux one causes the machine to crash. The same is also true if you use the driver from Intel's website. I was just wondering if there was any discussion on this list concerning this issue?

I have 4 ABIT IC7-G's and this problem only occurs on two of them, which I have disabled the onboard CSA gigabit nic and put in PCI NICS (intel gigabit) with no problems.

It appears there are some serious problems with Intel CSA GIGABIT with this motherboard, just curious to see if anyone out there heard anything about this?

The interesting part is the first page of this thread:

http://forum.abit-usa.com/showthread.php?t=18707

The author has a "death.zip" file, in which if you copy the file over the LAN it causes an instant crash. He's tried contacting A-Bit, but is it an A-Bit or Intel or Driver problem as it has been "reported" to work with A-Bit's driver.

I believe this post sums up the problems of IC7-G problems:

--

Yeah, can't have it all, can we? I had serious thoughts about RMA-ing my IC7-G last fall, but the 4 week tournaround was a bit too much downtime for my taste. To resolve the noisy mic-input issue, I bought a cheap soundcard. For the onboard CSA, I did nothing. Running at 1GbE, the NIC has worked very well, disregarding the numerous occations it has strangled my system to death, of course.

I feel sorry for you Sydtech, myself and everyone else out there ripping their hair out over the issues of the IC7. We have spent countless hours trying to figure out what's wrong, worrying about a lengthy RMA process and the adherent downtime - which in Sydtech's case just as well might result in a shabby refurbished MB fixing some problems, but adding others. Many of us have simply resorted to buying PCI add-in cards replacing the dysfunctional parts of the IC7.

We have paid the bill for ABIT's engineering shortcomings, not to mention wasting a lot of time and energy. Last spring I was looking for a MB for my new system. I wanted the best and with as many onboard functions as possible so I could have a less crammed case. I read all the reviews for the new i875 MB:s, where - for instance - HardOCP praised it, especially it's superiour sound quality (they did not even try the mic-input). Needless to say, I ended up with the expensive IC7-G, "cream of the crop" and bucketsful of headache.

Alright, so add-in cards cramming up our cases seem to be the easiest way to get the subsystems we've already payed for. But what's really infuriating is that ABIT just refuses to address the issues. No helping hand, no suggestions on how to get things right. They just keep ignoring us. Of course they read the forums, they just think keeping things quiet is their best strategy.

Well, anyhow, I have my strategy worked out for future technology acquisitions: I'll NEVER again buy first revision products the minute they come out. I refuse to again be an unpaid beta tester for an ungrateful and unhelpful manufacturer like ABIT. And I will also take reviews with loads of salt.

--

I was wondering if anyone from [email protected] could comment on this issue?

Thanks,

Justin.

#2 lore raw
Quoted 9 lines · Justin Piszcz

/usr/src/linux/Documentation//networking/e1000.txt:the issue to [email protected].

^^ That e-mail is invalid, someone should make a diff/patch to remove it and update it if there is a correct one or just remove it entirely.

Justin.

As of 2.6.15.3 source.

Agreed, I have an update pending in my queue but got lost in git-land

#3 lore raw
In reply to Justin Piszcz
Quoted 4 lines · Justin Piszcz

The author has a "death.zip" file, in which if you copy the file over the
LAN it causes an instant crash.

Windows and Linux? What kind of "crash"? An Oops? Lockup? Reboot? Anything in the logs?

Your report is way too vague to be actionable.

Lee

#4 lore raw

Essentially, when you copy large amounts of data across the NIC it will "freeze" the box in Linux (any 2.6.x kernel, have not tried 2.4.x) or Windows XP SP2.

If you checkout the thread, it occurs for multiple people under various OS' but in *some* cases if they use ABIT's IC7-G CSA/INTEL driver, they their problems go away.

In Linux when I used to use the onboard NIC, it froze the box, I did not have sysrq enabled at the time when this happened but frozen I mean screen is frozen, no ping, box is inoperative.

Nothing pecuilar was ever found in any of the logs or dmesg output regarding the crash.

Basically its the first revision of CSA gigabit on a motherboard from what I read in the forums and unless you use ABIT's specially crafted driver, it will crash the machine when you copy either:

a) large amounts of data over a gigabit link
or
b) that death.zip file (unzipped of course) which contains the bad "bits" that are probably seen/repeated when copying large amounts of data

Justin.

Quoted 13 lines · Lee Revell
Quoted 4 lines · Justin Piszcz

The author has a "death.zip" file, in which if you copy the file over the
LAN it causes an instant crash.

Windows and Linux? What kind of "crash"? An Oops? Lockup? Reboot? Anything in the logs?

Your report is way too vague to be actionable.

Lee

#5 lore raw
Quoted 3 lines · Justin Piszcz

Essentially, when you copy large amounts of data across the NIC it will "freeze" the box in Linux (any 2.6.x kernel, have not tried 2.4.x) or Windows XP SP2.

I've heard isolated reports about issues with the CSA connected NIC, but we've not been able to reproduce much in our labs and (mostly) people haven't been complaining about it.

Quoted 7 lines

If you checkout the thread, it occurs for multiple people under various OS' but in *some* cases if they use ABIT's IC7-G CSA/INTEL driver, they their problems go away.

In Linux when I used to use the onboard NIC, it froze the box, I did not have sysrq enabled at the time when this happened but frozen I mean screen is frozen, no ping, box is inoperative.

Have you tried running without NAPI? (disable it in your config in the e1000 section)

Quoted 11 lines

Nothing pecuilar was ever found in any of the logs or dmesg output regarding the crash.

Basically its the first revision of CSA gigabit on a motherboard from what I read in the forums and unless you use ABIT's specially crafted driver, it will crash the machine when you copy either:

a) large amounts of data over a gigabit link
or
b) that death.zip file (unzipped of course) which contains the bad "bits" that are probably seen/repeated when copying large amounts of data

I'll have our lab attempt to reproduce the bug (again) this time using the special file. I can't speak to the windows crash, sorry.

please send your .config, cat /proc/interrupts, dmesg after driver is up, whether NAPI is on, what exact steps you use to reproduce the problem, what your environment is (i.e. copying to a windows server, etc) pretty much follow the instructions in
http://www.kernel.org/pub/linux/docs/lkml/reporting-bugs.html

Jesse

#6 lore raw
Quoted 2 lines

Have you tried running without NAPI? (disable it in your config in the e1000 section)

I have always used NAPI, I will test this later today and let you know the results with NAPI off, thanks.

Is there anything I need to be aware of using a CSA-based ethernet device vs. no CSA?

Thanks,

Justin.

Quoted 43 lines · Jesse Brandeburg
Quoted 3 lines · Justin Piszcz

Essentially, when you copy large amounts of data across the NIC it will "freeze" the box in Linux (any 2.6.x kernel, have not tried 2.4.x) or Windows XP SP2.

I've heard isolated reports about issues with the CSA connected NIC, but we've not been able to reproduce much in our labs and (mostly) people haven't been complaining about it.

Quoted 7 lines

If you checkout the thread, it occurs for multiple people under various OS' but in *some* cases if they use ABIT's IC7-G CSA/INTEL driver, they their problems go away.

In Linux when I used to use the onboard NIC, it froze the box, I did not have sysrq enabled at the time when this happened but frozen I mean screen is frozen, no ping, box is inoperative.

Have you tried running without NAPI? (disable it in your config in the e1000 section)

Quoted 11 lines

Nothing pecuilar was ever found in any of the logs or dmesg output regarding the crash.

Basically its the first revision of CSA gigabit on a motherboard from what I read in the forums and unless you use ABIT's specially crafted driver, it will crash the machine when you copy either:

a) large amounts of data over a gigabit link
or
b) that death.zip file (unzipped of course) which contains the bad "bits" that are probably seen/repeated when copying large amounts of data

I'll have our lab attempt to reproduce the bug (again) this time using the special file. I can't speak to the windows crash, sorry.

please send your .config, cat /proc/interrupts, dmesg after driver is up, whether NAPI is on, what exact steps you use to reproduce the problem, what your environment is (i.e. copying to a windows server, etc) pretty much follow the instructions in
http://www.kernel.org/pub/linux/docs/lkml/reporting-bugs.html

Jesse

#7 lore raw
In reply to Jesse Brandeburg

Dreadful Denial of Service!!

Here is the story.

1) NAPI = ENABLED, The kernel does not crash IMMEDIATELY.
2) NAPI = DISABLED, KILLS THE VIDEO, FORCES AN IMMEDIATE REBOOT OF THE BOX

Note: When I leave the cp trying to copy the death.dat, my box hangs, will not come up, has other problems UNTIL I disconnect or reboot the box that originally did the cp file /nfs/directory.

The good: I got a dump of what was in dmesg before it locked up in #1.
The bad: I could do nothing in #2, it simply was like someone pressed the reboot button on my machine!!!

Here is the log:

$ cp .stuff/mobo/death.dat /p34/x

.. then I get these in dmesg ..

then > 60 seconds the box is frozen

.. ..

Here is from part #1:

[  251.277315] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  251.277317]   TDH                  <af>
[  251.277318]   TDT                  <b3>
[  251.277319]   next_to_use          <b3>
[  251.277320]   next_to_clean        <af>
[  251.277320] buffer_info[next_to_clean]
[  251.277321]   dma                  <36970802>
[  251.277322]   time_stamp           <ffff4113>
[  251.277323]   next_to_watch        <af>
[  251.277324]   jiffies              <ffff485c>
[  251.277325]   next_to_watch.status <0>
[  253.274703] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  253.274704]   TDH                  <af>
[  253.274706]   TDT                  <b3>
[  253.274706]   next_to_use          <b3>
[  253.274707]   next_to_clean        <af>
[  253.274708] buffer_info[next_to_clean]
[  253.274709]   dma                  <36970802>
[  253.274710]   time_stamp           <ffff4113>
[  253.274711]   next_to_watch        <af>
[  253.274712]   jiffies              <ffff502d>
[  253.274713]   next_to_watch.status <0>
[  255.270969] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  255.270971]   TDH                  <af>
[  255.270972]   TDT                  <b3>
[  255.270973]   next_to_use          <b3>
[  255.270974]   next_to_clean        <af>
[  255.270975] buffer_info[next_to_clean]
[  255.270976]   dma                  <36970802>
[  255.270977]   time_stamp           <ffff4113>
[  255.270978]   next_to_watch        <af>
[  255.270979]   jiffies              <ffff57fc>
[  255.270980]   next_to_watch.status <0>
[  257.267361] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  257.267363]   TDH                  <af>
[  257.267364]   TDT                  <b3>
[  257.267365]   next_to_use          <b3>
[  257.267366]   next_to_clean        <af>
[  257.267367] buffer_info[next_to_clean]
[  257.267368]   dma                  <36970802>
[  257.267369]   time_stamp           <ffff4113>
[  257.267370]   next_to_watch        <af>
[  257.267371]   jiffies              <ffff5fcc>
[  257.267372]   next_to_watch.status <0>
[  258.248353] NETDEV WATCHDOG: eth0: transmit timed out
[  261.321533] e1000: eth0: e1000_watchdog_task: NIC Link is Up 1000 Mbps Full Duplex
Quoted 43 lines · Jesse Brandeburg
Quoted 3 lines · Justin Piszcz

Essentially, when you copy large amounts of data across the NIC it will "freeze" the box in Linux (any 2.6.x kernel, have not tried 2.4.x) or Windows XP SP2.

I've heard isolated reports about issues with the CSA connected NIC, but we've not been able to reproduce much in our labs and (mostly) people haven't been complaining about it.

Quoted 7 lines

If you checkout the thread, it occurs for multiple people under various OS' but in *some* cases if they use ABIT's IC7-G CSA/INTEL driver, they their problems go away.

In Linux when I used to use the onboard NIC, it froze the box, I did not have sysrq enabled at the time when this happened but frozen I mean screen is frozen, no ping, box is inoperative.

Have you tried running without NAPI? (disable it in your config in the e1000 section)

Quoted 11 lines

Nothing pecuilar was ever found in any of the logs or dmesg output regarding the crash.

Basically its the first revision of CSA gigabit on a motherboard from what I read in the forums and unless you use ABIT's specially crafted driver, it will crash the machine when you copy either:

a) large amounts of data over a gigabit link
or
b) that death.zip file (unzipped of course) which contains the bad "bits" that are probably seen/repeated when copying large amounts of data

I'll have our lab attempt to reproduce the bug (again) this time using the special file. I can't speak to the windows crash, sorry.

please send your .config, cat /proc/interrupts, dmesg after driver is up, whether NAPI is on, what exact steps you use to reproduce the problem, what your environment is (i.e. copying to a windows server, etc) pretty much follow the instructions in
http://www.kernel.org/pub/linux/docs/lkml/reporting-bugs.html

Jesse

#8 lore raw

Jesse,

Any idea what is going on? MB defect? Driver defect?

Justin.

Quoted 121 lines · Justin Piszcz

Dreadful Denial of Service!!

Here is the story.

1) NAPI = ENABLED, The kernel does not crash IMMEDIATELY.
2) NAPI = DISABLED, KILLS THE VIDEO, FORCES AN IMMEDIATE REBOOT OF THE BOX

Note: When I leave the cp trying to copy the death.dat, my box hangs, will not come up, has other problems UNTIL I disconnect or reboot the box that originally did the cp file /nfs/directory.

The good: I got a dump of what was in dmesg before it locked up in #1.
The bad: I could do nothing in #2, it simply was like someone pressed the reboot button on my machine!!!

Here is the log:

$ cp .stuff/mobo/death.dat /p34/x

.. then I get these in dmesg ..

then > 60 seconds the box is frozen

.. ..

Here is from part #1:

[  251.277315] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  251.277317]   TDH                  <af>
[  251.277318]   TDT                  <b3>
[  251.277319]   next_to_use          <b3>
[  251.277320]   next_to_clean        <af>
[  251.277320] buffer_info[next_to_clean]
[  251.277321]   dma                  <36970802>
[  251.277322]   time_stamp           <ffff4113>
[  251.277323]   next_to_watch        <af>
[  251.277324]   jiffies              <ffff485c>
[  251.277325]   next_to_watch.status <0>
[  253.274703] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  253.274704]   TDH                  <af>
[  253.274706]   TDT                  <b3>
[  253.274706]   next_to_use          <b3>
[  253.274707]   next_to_clean        <af>
[  253.274708] buffer_info[next_to_clean]
[  253.274709]   dma                  <36970802>
[  253.274710]   time_stamp           <ffff4113>
[  253.274711]   next_to_watch        <af>
[  253.274712]   jiffies              <ffff502d>
[  253.274713]   next_to_watch.status <0>
[  255.270969] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  255.270971]   TDH                  <af>
[  255.270972]   TDT                  <b3>
[  255.270973]   next_to_use          <b3>
[  255.270974]   next_to_clean        <af>
[  255.270975] buffer_info[next_to_clean]
[  255.270976]   dma                  <36970802>
[  255.270977]   time_stamp           <ffff4113>
[  255.270978]   next_to_watch        <af>
[  255.270979]   jiffies              <ffff57fc>
[  255.270980]   next_to_watch.status <0>
[  257.267361] e1000: eth0: e1000_clean_tx_irq: Detected Tx Unit Hang
[  257.267363]   TDH                  <af>
[  257.267364]   TDT                  <b3>
[  257.267365]   next_to_use          <b3>
[  257.267366]   next_to_clean        <af>
[  257.267367] buffer_info[next_to_clean]
[  257.267368]   dma                  <36970802>
[  257.267369]   time_stamp           <ffff4113>
[  257.267370]   next_to_watch        <af>
[  257.267371]   jiffies              <ffff5fcc>
[  257.267372]   next_to_watch.status <0>
[  258.248353] NETDEV WATCHDOG: eth0: transmit timed out
[  261.321533] e1000: eth0: e1000_watchdog_task: NIC Link is Up 1000 Mbps Full Duplex
Quoted 43 lines · Jesse Brandeburg
Quoted 3 lines · Justin Piszcz

Essentially, when you copy large amounts of data across the NIC it will "freeze" the box in Linux (any 2.6.x kernel, have not tried 2.4.x) or Windows XP SP2.

I've heard isolated reports about issues with the CSA connected NIC, but we've not been able to reproduce much in our labs and (mostly) people haven't been complaining about it.

Quoted 7 lines

If you checkout the thread, it occurs for multiple people under various OS' but in *some* cases if they use ABIT's IC7-G CSA/INTEL driver, they their problems go away.

In Linux when I used to use the onboard NIC, it froze the box, I did not have sysrq enabled at the time when this happened but frozen I mean screen is frozen, no ping, box is inoperative.

Have you tried running without NAPI? (disable it in your config in the e1000 section)

Quoted 11 lines

Nothing pecuilar was ever found in any of the logs or dmesg output regarding the crash.

Basically its the first revision of CSA gigabit on a motherboard from what I read in the forums and unless you use ABIT's specially crafted driver, it will crash the machine when you copy either:

a) large amounts of data over a gigabit link
or
b) that death.zip file (unzipped of course) which contains the bad "bits" that are probably seen/repeated when copying large amounts of data

I'll have our lab attempt to reproduce the bug (again) this time using the special file. I can't speak to the windows crash, sorry.

please send your .config, cat /proc/interrupts, dmesg after driver is up, whether NAPI is on, what exact steps you use to reproduce the problem, what your environment is (i.e. copying to a windows server, etc) pretty much follow the instructions in
http://www.kernel.org/pub/linux/docs/lkml/reporting-bugs.html

Jesse

#9 lore raw
Quoted 3 lines · Justin Piszcz

Jesse,

Any idea what is going on? MB defect? Driver defect?

well, no-napi shouldn't fail any worse than napi, so that may well be a driver bug. Lets follow up offline or at mailing list e1000-devel % lists.sf.net

otherwise no idea. There are a lot of people using these boards without issues, it seems that ABIT even has some of this same board that work for some people. If you can find someone that has one that works you might try comparing your ethtool -e output to theirs. If you change your eeprom however, thats you taking risks on your own, and we can't support you. But it might be a useful thing to know, to give more information to ABIT.

Jesse