[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <Y9PqaJGwAUAZo2Rx@corigine.com>
Date: Fri, 27 Jan 2023 16:14:48 +0100
From: Simon Horman <simon.horman@...igine.com>
To: Alexandru Tachici <alexandru.tachici@...log.com>
Cc: linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
pabeni@...hat.com, richardcochran@...il.com,
yangyingliang@...wei.com, weiyongjun1@...wei.com,
robh+dt@...nel.org, krzysztof.kozlowski+dt@...aro.org,
devicetree@...r.kernel.org, lennart@...omain.com
Subject: Re: [net-next v2 1/3] net: ethernet: adi: adin1110: add PTP clock
support
On Thu, Jan 26, 2023 at 03:33:44PM +0200, Alexandru Tachici wrote:
> Add control for the PHC inside the ADIN1110/2111.
> Device contains a syntonized counter driven by a 120 MHz
> clock with 8 ns resolution.
>
> Time is stored on two registers: a 32bit seconds register and
> a 32bit nanoseconds register.
>
> For adjusting the clock timing, device uses an addend register.
> Can generate an output signal on the TS_TIMER pin.
> For reading the timestamp the current tiem is saved by setting the
> TS_CAPT pin via gpio in order to snapshot both seconds and nanoseconds
> in different registers that the live ones.
>
> Signed-off-by: Alexandru Tachici <alexandru.tachici@...log.com>
...
> @@ -150,6 +188,11 @@ struct adin1110_port_priv {
> struct adin1110_priv {
> struct mutex lock; /* protect spi */
> spinlock_t state_lock; /* protect RX mode */
> + bool ts_rx_append;
> + struct ptp_clock_info ptp;
> + struct ptp_clock *ptp_clock;
> + struct gpio_desc *ts_capt;
> + struct ptp_pin_desc ptp_pins[ADIN_MAC_MAX_PTP_PINS];
> struct mii_bus *mii_bus;
> struct spi_device *spidev;
> bool append_crc;
nit: I'm unsure if this is important or not.
But (on x86_64) the arrangement above leads to a 7-byte hole in the
structure between ts_rx_append and ptp.
Probably there is much scope to make adin1110_priv more
cache-line friendly. But in this case an improvement may be
to locate ts_rx_append immediately before or after append_crc,
as there is also a 7-byte hole after append_crc.
Likewise, I think there may be room to improve on the cache-line
friendliness of the changes to this structure in patch 2/3.
Powered by blists - more mailing lists