gve: Allocate qpl per ring at ring allocation time
ClosedPublic
Actions

Authored by veethebee_google.com on Mar 20 2025, 8:36 PM.

Details

Reviewers

markj
delphij
kibab
shailend_google.com

Group Reviewers

network

Commits

rG890309a67b50: gve: Allocate qpl per ring at ring allocation time
rGf8ed8382daf4: gve: Allocate qpl per ring at ring allocation time

Summary

Every tx and rx ring has its own queue-page-list (QPL) that serves as
the bounce buffer. Previously we were allocating QPLs for all queues
before the queues themselves were allocated and later associating a QPL
with a queue. This is avoidable complexity: it is much more natural for
each queue to allocate and free its own QPL.

Signed-off-by: Vee Agarwal <veethebee@google.com>

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 63317
Build 60201: arc lint + arc unit

Event Timeline

veethebee_google.com created this revision.Mar 20 2025, 8:36 PM

Herald added a subscriber: imp. · View Herald TranscriptMar 20 2025, 8:36 PM

veethebee_google.com requested review of this revision.Mar 20 2025, 8:36 PM

Harbormaster completed remote builds in B63007: Diff 152476.Mar 20 2025, 8:36 PM

veethebee_google.com added a child revision: D49427: gve: Add feature to adjust RX/TX queue counts.Mar 20 2025, 8:36 PM

shailend_google.com accepted this revision.Apr 1 2025, 12:06 AM

This revision is now accepted and ready to land.Apr 1 2025, 12:06 AM

markj added inline comments.Apr 2 2025, 12:37 PM

sys/dev/gve/gve.h
631–633
sys/dev/gve/gve_qpl.c
97	malloc(M_WAITOK) never fails, so null-checking isn't required here.
99	Given that `id` is only used in this assignment, why not add the offset in the caller instead of passing this `is_rx` flag? That seems cleaner to me.
171–175	Kernel code should use C-style comments only. There is a script, tools/build/checkstyle9.pl, based on the one for Linux, which should report issues like this.
172	Why not declare `i_tx` in the loop header?