From: Serhii Iliushyk <sil-plv@napatech.com>
To: stephen@networkplumber.org, mko-plv@napatech.com, ckm@napatech.com
Cc: dev@dpdk.org
Subject: [PATCH v1 0/3] net/ntnic: fix virt queue data in BSS
Date: Tue, 8 Sep 2026 18:27:21 +0200 [thread overview]
Message-ID: <20260908162724.463767-1-sil-plv@napatech.com> (raw)
This series addresses the bug report "ntnic: virt queue data in BSS"
(Bugzilla ID: 1975). The report lists five problems with keeping the
virt queue state in static arrays; each is addressed as follows.
- BSS memory is slower than hugepages because of TLB overhead, which
can cause performance loss in the hot path.
The per-queue state, and the copy of the packet buffer descriptors
indexed on every burst, are now allocated from the EAL heap.
- Hard coding the size wastes space.
The 128-entry Rx and Tx arrays are gone; one object is allocated
per configured queue and freed when the queue is released.
- Data in BSS is not visible to a secondary process.
The state now lives in EAL memory. The PMD still does not support
primary/secondary operation, but this no longer stands in the way.
- NUMA aware allocation was not possible.
Both allocations are made with rte_zmalloc_socket() on the NUMA
node of the adapter, next to the rings and buffers they describe.
- Cache alignment.
The queue object is cache line aligned and no longer shares a line
with a neighboring queue.
Patch 1 drops two never-called setup ops so that only the managed
setup and release paths own the queue state. Patch 2 makes a failed
hardware setup visible to the caller, which patch 3 relies on to free
the queue object on failure. Patch 3 is the fix itself; it also fixes
a leak of the descriptor copy on queue re-setup and an ignored
allocation failure that led to a NULL dereference in the burst
functions.
Serhii Iliushyk (3):
net/ntnic: remove unused unmanaged virt queue setup ops
net/ntnic: check virt queue hardware setup result
net/ntnic: fix virt queue data in BSS
drivers/net/ntnic/dbsconfig/ntnic_dbsconfig.c | 334 +++++++++++-------
drivers/net/ntnic/ntnic_ethdev.c | 52 ++-
drivers/net/ntnic/ntnic_mod_reg.h | 30 --
3 files changed, 245 insertions(+), 171 deletions(-)
--
2.45.0
next reply other threads:[~2026-09-08 16:27 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-09-08 16:27 Serhii Iliushyk [this message]
2026-09-08 16:27 ` [PATCH v1 1/3] net/ntnic: remove unused unmanaged virt queue setup ops Serhii Iliushyk
2026-09-08 16:27 ` [PATCH v1 2/3] net/ntnic: check virt queue hardware setup result Serhii Iliushyk
2026-09-08 16:27 ` [PATCH v1 3/3] net/ntnic: fix virt queue data in BSS Serhii Iliushyk
2026-09-08 17:18 ` [PATCH v1 0/3] " Stephen Hemminger
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260908162724.463767-1-sil-plv@napatech.com \
--to=sil-plv@napatech.com \
--cc=ckm@napatech.com \
--cc=dev@dpdk.org \
--cc=mko-plv@napatech.com \
--cc=stephen@networkplumber.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox