[Bridge] Bug with 82559 cars

[Gautier Harmel]

Hi,   I found bug using the Adlink PCI8214 nic card based on an Intel 82559 under Linux.   This bug occurs under a Linux 2.4.17 (e100 driver version 2.1.15 found on the Intel web site) using the bridge functionality. I didn’t try using an earlier version of linux kernel.   I used LKCD (linux kernel crash debugger) to further investigate what happens. Here are my conclusions :   1)    the bug occurs in skb_checksum_help function (linux/net/core.dev.c file) 2)       the disassembly code involved is:      7c7:   0f 0b                   ud2a        7c9:   89 c8                   mov    %ecx,%eax      7cb:   c1 e1 10                shl    $0x10,%ecx      7ce:   25 00 00 ff ff          and    $0xffff0000,%eax      7d3:   01 c8                   add    %ecx,%eax      7d5:   15 ff ff 00 00          adc    $0xffff,%eax      7da:   f7 d0                   not    %eax      7dc:   c1 e8 10                shr    $0x10,%eax      7df:   66 89 04 3e             mov    %ax,(%esi,%edi,1)      7e3:   89 d8                   mov    %ebx,%eax      7e5:   c6 43 6b 00             movb   $0x0,0x6b(%ebx)   3)    I think that this assembly code is the following line in skb_checksum_help: (*u16*)(skb->h.raw + skb->csum) = csum_fold(csum); (line 933)     Maybe skb->h.raw is a bad pointer ???   Thanks for help   Please CC me in your answer, while I didn’t subscribe to any of the mailing list   Jerome Tollet [EMAIL PROTECTED]