Realtek RTL8110 (SB) watchdog timeout.

[Pyun YongHyeon]

On Fri, Aug 01, 2008 at 05:07:21PM +0900, To Eugene Butusov wrote:
 > On Fri, Aug 01, 2008 at 03:12:05AM +0200, Eugene Butusov wrote:
 >  > Hi,
 >  > 
 >  >   After updating from 7.0-RELEASE to STABLE (around 15/08) my NIC 
 >  > refuses to handle large file transfers.
 >  > 
 >  > pciconf -lv
 >  > 
 >  > re0 at pci0:4:0:0: class=0x020000 card=0x001a6409 chip=0x816910ec 
 >  > rev=0x10 hdr=0x00
 >  >     vendor     = 'Realtek Semiconductor'
 >  >     device     = 'RTL8110SB Single-Chip Gigabit LOM Ethernet Controller'
 >  >     class      = network
 >  >     subclass   = ethernet
 >  > 
 >  > Log messages:
 >  > 
 >  > re: watchdog timeout
 >  > re: link changed to DOWN
 >  > re: link changed to UP
 >  > 
 >  > When someone tried to copy large (i.e. 700MB) file from samba share 
 >  > (local gigabit network) or ftp (same LAN),
 >  > the NIC was reseted. For a while host was not accesible from the 
 >  > network, and then it came back with log messages shown above.
 >  > I've tried to tune samba socket options (SO_RCVBUF=16384 
 >  > SO_SNDBUF=16384), and this fixed the problem for samba users. One 
 >  > interesting thing: copying file to windows XP machine worked fine, 
 >  > while Vista (SP1 x64) caused the problem.
 >  > 
 >  > What solved the problem definitely was disabling TSO for re0 (ifconfig 
 > 
 > One of developer also reported TSO issue. But his hardware was a
 > plain 8169S. Given that you're seeing TSO issues on 8110SB I'm
 > afraid all RealTek 8169/8110 series may suffer from the TSO issues.
 > Under certain circumtances, the controller generates corrupted
 > frames for TSO case and this seem to be resulted in watchdog
 > timeouts.
 > I'm not sure recent PCIe based 8168/8110 family also suffers from
 > the issue as no one have complained the issue.
 > 
 >  > re0 -tso). I haven't notice any performance drop and it works fine, 
 >  > but I'm just curious what happened to the 'good' driver from 7.0-RELEASE.
 >  > 
 > 
 > I don't think re(4) in 7.0-RELEASE is bug free. If you check commit
 > logs in RELENG_7 you may see what I mean. At the time of re(4)
 > overhauling, I added TSO to re(4). Generally TSO shall not increase
 > Tx performance but TSO significantly saves CPU cycles for TCP bulk
 > transfers. The saved CPU power could be used for other tasks.
 > 
 > Anyway, thanks for reporting, I'll disable TSO in next monday.
 > 

FYI: I've committed TSO patch to HEAD with svn revision r181271.

-- 
Regards,
Pyun YongHyeon