On 2026-02-05 at 07:55:27, Jakub Kicinski (kuba@kernel.org) wrote:

quoted

 	cn20k_wreq.hw_prio = req->hw_prio;
+	cn20k_wreq.req_kw_type = req->req_kw_type;

 update_rule:

Looking at npc_install_flow(), req->req_kw_type is copied to
cn20k_wreq.req_kw_type, but is req->req_kw_type ever initialized with the
computed kw_type value?

In rvu_mbox_handler_npc_install_flow(), the function
rvu_npc_alloc_entry_for_flow_install() computes the correct kw_type based
on the flow requirements and stores it in the local variable 'kw_type' at
line 1746. However, this computed value is never assigned to
req->req_kw_type before calling npc_install_flow() at line 1887.

As a result, npc_install_flow() uses either a zero-initialized value or
whatever the PF driver sent, instead of the AF-computed value. This seems
inconsistent with the commit message stating "the AF driver determine the
required NPC MCAM entry type." The incorrect req_kw_type is then passed to
npc_cn20k_config_mcam_entry() which uses it for hardware configuration
mismatch handling in npc_cn20k_config_kw_x2() at lines 956-960.

req->req_kw_type is used for different purpose. I believe, this comment is invalid.