[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230512114209.2894459-1-daire.mcnamara@microchip.com>
Date: Fri, 12 May 2023 12:42:08 +0100
From: <daire.mcnamara@...rochip.com>
To: <nicholas.ferre@...rochip.com>, <claudiu.beznea@...rochip.com>,
<davem@...emloft.net>, <edumazet@...gle.com>, <kuba@...nel.org>,
<pabeni@...hat.com>, <netdev@...r.kernel.org>, <conor.dooley@...rochip.com>
CC: Daire McNamara <daire.mcnamara@...rochip.com>
Subject: [PATCH v3 0/1] Adjust macb max_tx_len for mpfs
From: Daire McNamara <daire.mcnamara@...rochip.com>
Several customers have reported unexpected ethernet issues whereby
the GEM stops transmitting and receiving. Performing an action such
as ifconfig <ethX> down; ifconfig <ethX> up clears this particular
condition.
The origin of the issue is a stream of AMBA_ERRORS (bit 6) from the
tx queues.
This patch sets the max_tx_length to SRAM size (16 KiB
in the case of mpfs) divided by num_queues (4 in the case of mpfs)
and then subtracts 56 bytes from that figure - resulting in max_tx_len
of 4040. The max jumbo length is also set to 4040. These figures
are derived from Cadence erratum 1686.
Change from v2
- Remove pointless check for existence of macb_config.
Change from v1
- Switched from using macb_is_gem() to hw_is_gem() as macb_is_gem()
relies on capabilities being read and these have not been ascertained
at this point of the probe routine.
Daire McNamara (1):
net: macb: Shorten max_tx_len to 4KiB - 56 on mpfs
drivers/net/ethernet/cadence/macb.h | 1 +
drivers/net/ethernet/cadence/macb_main.c | 15 ++++++++++++---
2 files changed, 13 insertions(+), 3 deletions(-)
base-commit: 09a9639e56c01c7a00d6c0ca63f4c7c41abe075d
--
2.25.1
Powered by blists - more mailing lists