On Wed, Oct 14, 2026 at 12:13 UTC, [email protected] wrote: > - [High] idpf: NULL pointer dereference in idpf_tx_start during shutdown
This path is not changed by the patch. idpf_vport_dealloc() skips idpf_vport_stop() and idpf_decfg_netdev() while IDPF_HR_RESET_IN_PROG is set with or without it, and idpf_shutdown() has lacked the stale netdev cleanup that idpf_remove() does since 4c9106f4906a. It is a separate issue; I can look at it in a follow-up. > - [High] idpf: SW marker consumption race causes cascading timeouts The set-then-wait order per Tx queue is the same as in the current code. With the default split queue model each Tx queue has its own completion queue (IDPF_DFLT_SPLITQ_TXQ_PER_GROUP is 1), so markers are not shared. Where they are, this patch limits the effect on shutdown to one 500 ms wait per queue set. Setting SW_MARKER on all queues before waiting looks like a reasonable cleanup on its own. > - [High] idpf: DMA memory corruption on slow CP during shutdown This is the trade-off discussed on v3 and v4. Without the patch, shutdown sends none of the teardown messages, so a control plane slower than 500 ms is no worse off than today, and a responsive one gets the full teardown. On the tested devices the queue-related messages took at most 240 ms. Thanks, Tian Xun
