Bugzilla (bugzilla.redhat.com) will be under maintenance for infrastructure upgrades and will not be available on July 31st between 12:30 AM - 05:30 AM UTC. We appreciate your understanding and patience. You can follow status.redhat.com for details.
Bug 1880957 - testpmd crash when restarting it with vhost-user 4 queues + vIOMMU[dpdk] [NEEDINFO]
Summary: testpmd crash when restarting it with vhost-user 4 queues + vIOMMU[dpdk]
Keywords:
Status: POST
Alias: None
Product: Red Hat Enterprise Linux 8
Classification: Red Hat
Component: dpdk
Version: 8.3
Hardware: Unspecified
OS: Unspecified
high
high
Target Milestone: rc
: 8.0
Assignee: Maxime Coquelin
QA Contact: Pei Zhang
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2020-09-21 09:06 UTC by Pei Zhang
Modified: 2021-02-01 05:19 UTC (History)
7 users (show)

Fixed In Version:
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
Environment:
Last Closed:
Type: Bug
Target Upstream Version:
pezhang: needinfo? (maxime.coquelin)


Attachments (Terms of Use)

Description Pei Zhang 2020-09-21 09:06:11 UTC
This bug was initially created as a copy of Bug #1880299

I am copying this bug because: 



Description of problem:
Boot VM with vhost-user 4 queues, then boot testpmd as vhost-user client. Kill testpmd and start testpmd again, vhost-user connection fails to restart. This will cause both qemu and testpmd crash.

Version-Release number of selected component (if applicable):
4.18.0-237.el8.x86_64
qemu-kvm-5.1.0-7.module+el8.3.0+8099+dba2fe3e.x86_64
dpdk-19.11.3-1.el8.x86_64

How reproducible:
100%

Steps to Reproduce:
1. Boot VM with vhost-user 4 queues 

    <interface type="vhostuser">
      <mac address="88:66:da:5f:dd:02" />
      <source mode="server" path="/tmp/vhost-user1.sock" type="unix" />
      <model type="virtio" />
      <driver ats="on" iommu="on" name="vhost" queues="4" rx_queue_size="1024" />
      <address bus="0x6" domain="0x0000" function="0x0" slot="0x00" type="pci" />
    </interface>



2. Boot testpmd as vhost-user client

# cat testpmd_4q.sh 
testpmd \
-l 2,4,6,8,10,12,14,16,18 \
--socket-mem 1024,1024 \
-n 4  \
--vdev 'net_vhost0,iface=/tmp/vhost-user1.sock,queues=4,client=1,iommu-support=1' \
-d /usr/lib64/librte_pmd_vhost.so \
-- \
--portmask=f \
-i \
--rxd=512 --txd=512 \
--rxq=4 --txq=4 \
--nb-cores=8 \
--forward-mode=txonly

# sh testpmd_4q.sh 

3. Kill testpmd

# pkill testpmd


4. Start testpmd again

# sh testpmd.sh

5. vhost-user connection fails to restart. Both qemu and testpmd crash.

testpmd_4q.sh: line 13:  5227 Segmentation fault      (core dumped) testpmd -l 2,4,6,8,10,12,14,16,18 --socket-mem 1024,1024 -n 4 --vdev 'net_vhost0,iface=/tmp/vhost-user1.sock,queues=4,client=1,iommu-support=1' -d /usr/lib64/librte_pmd_vhost.so -- --portmask=f -i --rxd=512 --txd=512 --rxq=4 --txq=4 --nb-cores=8 --forward-mode=txonly

# dmesg
[ 3782.630934] vhost-events[5239]: segfault at 2d8 ip 00007f5dc8e171cb sp 00007f5db97f9870 error 4 in librte_vhost.so.20.0[7f5dc8e10000+4e000]
[ 3782.643491] Code: 89 40 08 e9 3f ff ff ff e8 22 bc ff ff 66 90 f3 0f 1e fa 41 57 41 56 41 55 41 54 49 89 cc 55 48 89 f5 53 48 89 fb 48 83 ec 48 <4c> 8b af d8 02 00 00 48 89 14 24 44 89 44 24 10 64 48 8b 04 25 28
[ 3784.517001] qemu-kvm[5097]: segfault at 1b0 ip 000055b8b3406540 sp 00007ffded2577f0 error 4 in qemu-kvm[55b8b304d000+a66000]
[ 3784.528225] Code: 75 10 48 8b 05 71 80 ae 00 89 c0 64 48 89 03 0f ae f0 90 48 8b 45 00 31 c9 85 d2 48 89 e7 0f 95 c1 41 b8 01 00 00 00 4c 89 e2 <48> 8b b0 b0 01 00 00 e8 64 d2 f5 ff 48 83 3c 24 00 0f 84 f9 00 00


Actual results:
vhost-user connection fails to restart. Both qemu and testpmd crash.

Expected results:
vhost-user connection should restart well. And qemu and testpmd should work well.

Additional info:
1. This issue can not be reproduced with vhost-user 2 queues. We need to test with vhost-user 4 queues to trigger.

Comment 2 Maxime Coquelin 2020-10-19 14:20:47 UTC
Adrian did the analysis in Bug #1880299.

The issue on DPDK side is that the backend assumes there will be no holes in the device's array of virtqueues pointers,
but it can happen.

Proposed fix is to allocate virtqueues up to requested virtqueue index, so that there are no holes.

Comment 3 Maxime Coquelin 2020-10-19 18:01:19 UTC
DPDK series fixing this issue has been posted upstream:
http://patches.dpdk.org/user/todo/dpdk/?series=13110

Comment 4 Pei Zhang 2021-02-01 05:19:20 UTC
Hello Maxime,

With latest rhel8.4 testing, this issue has gone, both dpdk's testpmd and VM keep working well. Could you confirm this fix from code level? Thanks a lot.


Versions:
dpdk-20.11-1.el8.x86_64
4.18.0-278.rt7.43.el8.dt4.x86_64
qemu-kvm-5.2.0-4.module+el8.4.0+9676+589043b9.x86_64
tuned-2.15.0-1.el8.noarch
libvirt-7.0.0-3.module+el8.4.0+9709+a99efd61.x86_64
python3-libvirt-6.10.0-1.module+el8.4.0+8948+a39b3f3a.x86_64
openvswitch2.13-2.13.0-86.el8fdp.x86_64

Best regards,

Pei


Note You need to log in before you can comment on or make changes to this bug.