[Bug,805] af_xdp PMD is broken with primary/secondary process

Message ID bug-805-3@http.bugs.dpdk.org/ (mailing list archive)
State Not Applicable, archived
Headers
Series [Bug,805] af_xdp PMD is broken with primary/secondary process |

Checks

Context Check Description
ci/Intel-compilation warning apply issues

Commit Message

bugzilla@dpdk.org Sept. 3, 2021, 3:21 a.m. UTC
  https://bugs.dpdk.org/show_bug.cgi?id=805

            Bug ID: 805
           Summary: af_xdp PMD is broken with primary/secondary process
           Product: DPDK
           Version: 21.11
          Hardware: All
                OS: Linux
            Status: UNCONFIRMED
          Severity: major
          Priority: Normal
         Component: core
          Assignee: dev@dpdk.org
          Reporter: stephen@networkplumber.org
  Target Milestone: ---

AF_XDP PMD is broken for primary/secondary process model.
Both the procinfo, pdump and other applications need to be able to basic
calls from secondary process.

There is one bug easy bug, where af_xdp crashes when calling
rte_eth_info_get() in secondary process because eth_dev->device
was not set.


But there is a bigger problem when code calls rte_eth_stats_get().
The af_xdp PMD tries to send a request to kernel to get the number
of missed packets; but the file descriptor is only valid in the primary
process. Fixing this is much deeper and requires opening a different
socket in each process, and will XDP in kernel handle this correctly?

I think we should just mark AF_XDP as broken in with primary/secondary
and return an error in probe in secondary process.
  

Comments

bugzilla@dpdk.org Oct. 11, 2021, 6:59 a.m. UTC | #1
https://bugs.dpdk.org/show_bug.cgi?id=805

Ciara Loftus (ciara.loftus@intel.com) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ciara.loftus@intel.com
         Resolution|---                         |FIXED
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #1 from Ciara Loftus (ciara.loftus@intel.com) ---
Fix pushed to net-next tree:

http://git.dpdk.org/next/dpdk-next-net/commit/?id=2fcc36c234b558013f163313c81cdcb13878933d
  

Patch

--- a/drivers/net/af_xdp/rte_eth_af_xdp.c
+++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
@@ -1800,6 +1800,7 @@  rte_pmd_af_xdp_probe(struct rte_vdev_device *dev)
                        return -EINVAL;
                }
                eth_dev->dev_ops = &ops;
+               eth_dev->device = &dev->device;
                rte_eth_dev_probing_finish(eth_dev);
                return 0;
        }