On 15-Feb-18 8:53 PM, Ravi Kerur wrote:
On Thu, Feb 15, 2018 at 10:27 AM, Ravi Kerur <rke...@gmail.com
<mailto:rke...@gmail.com>> wrote:
On Thu, Feb 15, 2018 at 2:28 AM, Burakov, Anatoly
<anatoly.bura...@intel.com <mailto:anatoly.bura...@intel.com>> wrote:
On 14-Feb-18 8:00 PM, Ravi Kerur wrote:
Earlier I was focusing only on DMAR errors and I might have
said 'it worked' when I didn't notice them on host when dpdk
was started on guest. When trying to send packets out of
that interface from guest I did see DMAR errors. I am
attaching information you requested. I have enabled
log-level=8 and files contain dpdk EAL/PMD logs as well.
Great, now we're on the same page.
Snippets below
on host, DMAR fault address from dmesg
[351576.998109] DMAR: DRHD: handling fault status reg 702
[351576.998113] DMAR: [DMA Read] Request device [04:10.0]
fault addr 257617000 [fault reason 06] PTE Read access is
not set
on guest (dump phys_mem_layout)
Segment 235: phys:0x257600000, len:2097152,
virt:0x7fce87e00000, socket_id:0, hugepage_sz:2097152,
nchannel:0, nrank:0
...
PMD: ixgbe_dev_rx_queue_setup(): sw_ring=0x7fce87e0f4c0
sw_sc_ring=0x7fce87e07380 hw_ring=0x7fce87e17600
dma_addr=0x257617600
PMD: ixgbe_dev_rx_queue_setup(): sw_ring=0x7fce89c67d40
sw_sc_ring=0x7fce89c5fc00 hw_ring=0x7fce89c6fe80
dma_addr=0x25406fe80
...
To me this looks like host (i.e. either QEMU or the PF driver)
is trying to do DMA using guest-physical (and not
host-physical). I'm not too well-versed in how QEMU works, but
i'm pretty sure that's not supposed to happen.
Is PF also bound to DPDK, or are you using native Linux ixgbe
driver?
Thanks for your help. I cannot use PF with DPDK (vfio-pci), VF
interfaces disappear after it is bound to DPDK. If there is a way to
use PF and VF with DPDK let me know I can try it out. I am not sure
how to move forward on this, Is CPU/IXGBE PF driver playing a role?
Following are the versions I have
lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 56
On-line CPU(s) list: 0-27
Off-line CPU(s) list: 28-55
Thread(s) per core: 1
Core(s) per socket: 14
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 63
Model name: Intel(R) Xeon(R) CPU E5-2683 v3 @ 2.00GHz
Stepping: 2
CPU MHz: 2500.610
CPU max MHz: 3000.0000
CPU min MHz: 1200.0000
BogoMIPS: 4000.74
Virtualization: VT-x
L1d cache: 32K
L1i cache: 32K
L2 cache: 256K
L3 cache: 35840K
NUMA node0 CPU(s): 0-13
NUMA node1 CPU(s): 14-27
# ethtool -i enp4s0f0
driver: ixgbe
version: 5.3.3
firmware-version: 0x800007b8, 1.1018.0
bus-info: 0000:04:00.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: yes
Thanks,
Ravi
Debugging this I could co-relate doing interface link-up associated with
the dpdk inside the guest causes DMAR errors on host and an additional
vflr message.
[ 8135.861622] DMAR: DRHD: handling fault status reg 402
[ 8135.861627] DMAR: [DMA Read] Request device [04:10.0] fault addr
1b648a000 [fault reason 06] PTE Read access is not set
[ 8136.588074] ixgbe 0000:04:00.0: Issuing VFLR with pending transactions
[ 8136.588079] ixgbe 0000:04:00.0: Issuing VFLR for VF 0000:04:10.0
Looked at ixgbe driver code 'ixgbe_issue_vf_flr' is called from
'ixgbe_check_for_bad_vf' or 'ixgbe_io_error_detected' functions. Is it
possible that dpdk pmd vf driver is missing some fixes/porting from
ixgbevf driver since this issue is not seen when ixgbevf kernel driver
is used?
Could very well be. +CC ixgbe maintainers which might be of further help
debugging this issue.
--
Thanks,
Anatoly
