Most recent kernel where this bug did *NOT* occur: happens on all kernels that have support for the 8161B Distribution: Debian Etch Hardware Environment: Asus P5B motherboard, Core2 DUO processor, Integrated Realtek 8161B PCI Express ethernet controller Software Environment: Debian Etch Problem Description: Extremely poor speed for some things like Samba, the problem does not exist when using the offical Realtek driver. Steps to reproduce: Using the r8169 driver Dmesg output: http://users.tpg.com.au/renevant/bug.txt Kernel config: http://users.tpg.com.au/renevant/kernconfig Screenshot of a download attempt on my Windows box: http://users.tpg.com.au/renevant/r8169-driverbad.JPG Using the Realtek R1000 driver Dmesg output: http://users.tpg.com.au/renevant/nobug.txt Screenshot of a download attempt on my Windows box: http://users.tpg.com.au/renevant/http://users.tpg.com.au/renevant/r1000-drivergood.JPG Note the only thing that has changed is the driver, nothing else. I wish I could show it better but it doesn't seem to be all that producable doing an FTP transfer or in general browsing. I've tried the 0001-r8169-more-alignment-for-the-0x8168 patch given to me by Francois Romieu with no luck.
That good URL should be: http://users.tpg.com.au/renevant/r1000-drivergood.JPG Sorry.
Amongst others, I have had a success report from a fedora core user who ran the patch below against its flavor of 2.6.18: http://www.fr.zoreil.com/people/francois/misc/20060920-2.6.18-r8169-test.patch The motherboard was an Asus P5B-VM with a built-in Intel graphic adapter. It would provide a nice datapoint if you could give it a try against some 2.6.18 kernel (without binary stuff from nvidia of course). -- Ueimor
I have tried with a vanilla compile of 2.6.18.3 the only patch being the one from the link you provided. The problem is still there. Dmesg output: http://users.tpg.com.au/renevant/bug2.txt Kernel config: http://users.tpg.com.au/renevant/kern2config.txt Screenshot of a filecopy from a Windows XP machine on the network: http://users.tpg.com.au/renevant/r8169-bad2.JPG I then went and compiled a the kernel again except I chose not to compile the r8169 driver as a module or to build it into the kernel. I then went and compiled the offical realtek driver instead. By the way I followed the instructions provided in this link in order to get the realtek driver to compile with recent kernels: http://gentoo-wiki.com/HARDWARE_RTL8168 Here were my results after booting up with the realtek driver: http://users.tpg.com.au/renevant/r1000-good2.JPG Let me know if there is any further testing you would like me to do. I know this little piece of information may make you sit up in your chair, the system IS overclocked, I'm happy to set it back to stock speeds if you feel this may have something to do with the problem but I don't really see that fixing this issue though. The system is completely stable and there is no corruption whatsoever when using the realtek driver. Happy to do any further testing, tcpdump, or any other tests you would like. Here is an lspci dump as well: http://users.tpg.com.au/renevant/lspci.txt
You better close this one. With 2.6.19-git7 things seemed fine just lower throughput that the realtek 1.04 driver, however I've been testing with an Intel PCI-E network adapter and I get roughly the same speed. Setting read and write raw to "no" seems to restore faster speed. So it seems to be mostly a samba issue. Sorry for the trouble. Strange that the Realtek 1.04 driver would be different like that though. On a totally unrelated note the new realtek 1.05 driver was buggy for me when using Xen.
The problem is probably fixed in 2.6.23 as of: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d78ae2dcc2acebb9a1048278f47f762c069db75c The symptoms really look the same. -- Ueimor