Bug 2123 - (net r8169) Hard lock due to r8169 gigabit ethernet
Summary: (net r8169) Hard lock due to r8169 gigabit ethernet
Status: CLOSED CODE_FIX
Alias: None
Product: Drivers
Classification: Unclassified
Component: Network (show other bugs)
Hardware: i386 Linux
: P2 normal
Assignee: Jeff Garzik
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2004-02-17 02:58 UTC by brandyn
Modified: 2004-03-28 11:01 UTC (History)
0 users

See Also:
Kernel Version: 2.6.2
Subsystem:
Regression: ---
Bisected commit-id:


Attachments
Patch against Tx descriptor overflow in r8169 driver (1.37 KB, patch)
2004-02-17 14:49 UTC, Francois Romieu
Details | Diff

Description brandyn 2004-02-17 02:58:57 UTC
Distribution: Debian unstable/testing
Hardware Environment: Athlon 1GHz, Gigabyte N400 Pro using integrated gigabit 
ethernet
Software Environment: Samba 3.0.2, raid 5 & 0
Problem Description:  System hard locks due to r8169 gigabit ethernet.  No 
kernel panic, no error.  Nothing reported in logs.  Network is dead. only 
blinking curser in console, but doesn't respond to input. I stripped down 
kernel to almost just essentials and still having hard locks.

Steps to reproduce: All i have to do is browse the server via samba from a 
windows client for a little while.  It's very random but happens fairly quickly 
if i go in and out of directories really quickly. I also tried this on a non 
raid partition and still have the same problem.  I installed a 3C905 network 
card and cannot reporduce the problem.
Comment 1 Martin J. Bligh 2004-02-17 07:57:48 UTC
Try ALT+sysrq+t and/or nmi_watchdog
Comment 2 Francois Romieu 2004-02-17 14:49:34 UTC
Created attachment 2169 [details]
Patch against Tx descriptor overflow in r8169 driver

Please report if things go worse/better/no difference.
Comment 3 brandyn 2004-02-23 20:19:12 UTC
Sorry for the delay. I tried the patch and had the same problem.
Comment 4 Francois Romieu 2004-02-25 10:38:07 UTC
Can you give Martin's suggestion a try:
- enable io-apic (remove the zzz_APIC related options in your .config and run
  'make oldconfig', build kernel, ...);
- pass an 'nmi_watchdog=1' option to the kernel at boot;
- check that the magic sysrq is not disabled (avoid 'kernel.sysrq = 0' in your
  /etc/sysctl.conf);
- use the aforementionned keyboard sequence when something goes wrong.

You can repeat this process:
- with standard kernel + previous patch
- with standard kernel + Andrew Morton's patch (-mm)

Disable PREEMPT and ACPI if enabled.

If it still does not make a difference, with standard kernel + -mm patch:
- increase NUM_{RX/TX}_DESC in the driver to 128;
- decrease NUM_{RX/TX}_DESC in the driver to 16.
Comment 5 Jeff Garzik 2004-03-25 19:34:53 UTC
Does the 2.6.5-mm series fix your problems?
Comment 6 brandyn 2004-03-28 11:01:27 UTC
this issue seems to be fixed in 2.6.3.  Sorry it took so long for my response.  
I forgot about it. Thank you for your help. 

Note You need to log in before you can comment on or make changes to this bug.