linux-kernel - [PATCH net 3/5] net: sparx5: fix FDMA performance issue

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20241205-sparx5-lan969x-misc-fixes-v1-3-575ff3d0b022@microchip.com>
Date: Thu, 5 Dec 2024 14:54:26 +0100
From: Daniel Machon <daniel.machon@...rochip.com>
To: Andrew Lunn <andrew+netdev@...n.ch>, "David S. Miller"
	<davem@...emloft.net>, Eric Dumazet <edumazet@...gle.com>, Jakub Kicinski
	<kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>, Lars Povlsen
	<lars.povlsen@...rochip.com>, Steen Hegelund <Steen.Hegelund@...rochip.com>,
	<UNGLinuxDriver@...rochip.com>, Richard Cochran <richardcochran@...il.com>,
	Bjarni Jonasson <bjarni.jonasson@...rochip.com>,
	<jensemil.schulzostergaard@...rochip.com>, <horatiu.vultur@...rochip.com>,
	<arnd@...db.de>, <jacob.e.keller@...el.com>,
	<Parthiban.Veerasooran@...rochip.com>
CC: Calvin Owens <calvin@...nvd.org>, Muhammad Usama Anjum
	<Usama.Anjum@...labora.com>, <linux-kernel@...r.kernel.org>,
	<netdev@...r.kernel.org>, <linux-arm-kernel@...ts.infradead.org>
Subject: [PATCH net 3/5] net: sparx5: fix FDMA performance issue

The FDMA handler is responsible for scheduling a NAPI poll, which will
eventually fetch RX packets from the FDMA queue. Currently, the FDMA
handler is run in a threaded context. For some reason, this kills
performance.  Admittedly, I did not do a thorough investigation to see
exactly what causes the issue, however, I noticed that in the other
driver utilizing the same FDMA engine, we run the FDMA handler in hard
IRQ context.

Fix this performance issue, by  running the FDMA handler in hard IRQ
context, not deferring any work to a thread.

Prior to this change, the RX UDP performance was:

Interval           Transfer     Bitrate         Jitter
0.00-10.20  sec    44.6 MBytes  36.7 Mbits/sec  0.027 ms

After this change, the rx UDP performance is:

Interval           Transfer     Bitrate         Jitter
0.00-9.12   sec    1.01 GBytes  953 Mbits/sec   0.020 ms

Fixes: 10615907e9b5 ("net: sparx5: switchdev: adding frame DMA functionality")
Signed-off-by: Daniel Machon <daniel.machon@...rochip.com>
---
 drivers/net/ethernet/microchip/sparx5/sparx5_main.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
index 2b58fcb9422e..f61aa15beab7 100644
--- a/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
+++ b/drivers/net/ethernet/microchip/sparx5/sparx5_main.c
@@ -780,12 +780,11 @@ static int sparx5_start(struct sparx5 *sparx5)
 	err = -ENXIO;
 	if (sparx5->fdma_irq >= 0 && is_sparx5(sparx5)) {
 		if (GCB_CHIP_ID_REV_ID_GET(sparx5->chip_id) > 0)
-			err = devm_request_threaded_irq(sparx5->dev,
-							sparx5->fdma_irq,
-							NULL,
-							sparx5_fdma_handler,
-							IRQF_ONESHOT,
-							"sparx5-fdma", sparx5);
+			err = devm_request_irq(sparx5->dev,
+					       sparx5->fdma_irq,
+					       sparx5_fdma_handler,
+					       0,
+					       "sparx5-fdma", sparx5);
 		if (!err)
 			err = sparx5_fdma_start(sparx5);
 		if (err)

-- 
2.34.1