Hello,
I tried to use 3.9.2 kernel with xen 4.2.2/4.3rc1 and in both variants leads to this error in network-intensive load (such as iperf, 100 nginx parallel requests to 1M files and so on):
BUG: unable to handle kernel paging request at ffff8800795a3000
[ 60.246945] IP: [<ffffffffa001a75c>] netif_poll+0x49c/0xe80 [xennet]
[ 60.246975] PGD a8a067 PUD a9a067 PMD 7fc27067 PTE 80100000795a3065
[ 60.247004] Oops: 0003 [#1] SMP
[ 60.247020] Modules linked in: af_packet hwmon domctl crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel ablk_helper cryptd lrw aes_x86_64 joydev xts gf128mul autofs4 scsi_dh_emc scsi_dh_alua scsi_dh_rdac scsi_dh_hp_sw scsi_dh xenblk cdrom xennet ata_generic ata_piix
[ 60.247144] CPU 0
[ 60.247154] Pid: 0, comm: swapper/0 Not tainted 3.9.2-1.g04040b9-xen #1
[ 60.247179] RIP: e030:[<ffffffffa001a75c>] [<ffffffffa001a75c>] netif_poll+0x49c/0xe80 [xennet]
...
We have couple of production hypervisors on 3.4 kernels with high-throughput internal network (VM-to-VM in one Dom0), iperf on them is working well:
[ 3] 0.0- 2.0 sec 3357 MBytes 14080 Mbits/sec
[ 3] 2.0- 4.0 sec 2880 MBytes 12077 Mbits/sec
[ 3] 4.0- 6.0 sec 2909 MBytes 12202 Mbits/sec
[ 3] 6.0- 8.0 sec 2552 MBytes 10702 Mbits/sec
[ 3] 8.0-10.0 sec 3616 MBytes 15166 Mbits/sec
[ 3] 10.0-12.0 sec 3415 MBytes 14324 Mbits/sec
Seems like a kernel bug, is this related to one of this fixes in linux-next or i need to create new bugreport?
1) https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=1aaf6d3d3d1e95f4be07e32dd84aa1c93855fbbd
2) https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=9ecd1a75d977e2e8c48139c7d3efed183f898d94
3) https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=2810e5b9a7731ca5fce22bfbe12c96e16ac44b6f
4) https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=03393fd5cc2b6cdeec32b704ecba64dbb0feae3c
5) https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=59ccb4ebbc35e36a3c143f2d1355deb75c2e628f
Thanks.
--
Best regards,
Eugene Istomin
|