Re: [PATCH net-next] cadence: Add hardware PTP support.

From: Harini Katakam
Date: Fri Nov 18 2016 - 06:45:01 EST


Hi Rafal

I'm still comparing the full solution but just a couple of things first:
<snip>
> @@ -876,6 +964,17 @@ static int gem_rx(struct macb *bp, int budget)
> bp->stats.rx_packets++;
> bp->stats.rx_bytes += skb->len;
>
> +#if IS_ENABLED(CONFIG_PTP_1588_CLOCK)
> + if (bp->ptp_hw_support) {
> + struct timespec64 ts;
> +
> + if (MACB_BFEXT(DMA_RX_TS_VALID, desc->addr)) {
> + macb_hw_timestamp(bp, desc->dma_desc_ts_1, desc->dma_desc_ts_2, &ts);
> + skb_hwtstamps(skb)->hwtstamp = ktime_set(ts.tv_sec, ts.tv_nsec);
> + }
> + }
> +#endif
> +

I've had to read PTP event registers for upper part of seconds timestamp
in addition to the descriptor (since descriptor only has 5 bits of seconds TS).
I don't know which version of the IP you use - it could be different.
Please let me know so that I can check the spec.
Same with tx timestamp of course.

<snip>
> @@ -1195,6 +1297,87 @@ static irqreturn_t macb_interrupt(int irq, void *dev_id)
> queue_writel(queue, ISR, MACB_BIT(HRESP));
> }
>
> +#if IS_ENABLED(CONFIG_PTP_1588_CLOCK)
> + if (status & MACB_BIT(PTP_DELAY_REQ_FRAME_RECEIVED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_DELAY_REQ_FRAME_RECEIVED));
> + if (macb_ptp_time_frame_rx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_SYNC_FRAME_RECEIVED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_SYNC_FRAME_RECEIVED));
> + if (macb_ptp_time_frame_rx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_DELAY_REQ_FRAME_TRANSMITTED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_DELAY_REQ_FRAME_TRANSMITTED));
> + if (macb_ptp_time_frame_tx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_SYNC_FRAME_TRANSMITTED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_SYNC_FRAME_TRANSMITTED));
> + if (macb_ptp_time_frame_tx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_PDELAY_REQ_FRAME_RECEIVED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_PDELAY_REQ_FRAME_RECEIVED));
> + if (macb_ptp_time_peer_frame_rx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_PDELAY_RESP_FRAME_RECEIVED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_PDELAY_RESP_FRAME_RECEIVED));
> + if (macb_ptp_time_peer_frame_rx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_PDELAY_REQ_FRAME_TRANSMITTED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_PDELAY_REQ_FRAME_TRANSMITTED));
> + if (macb_ptp_time_peer_frame_tx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +
> + if (status & MACB_BIT(PTP_PDELAY_RESP_FRAME_TRANSMITTED)) {
> + if (bp->caps & MACB_CAPS_ISR_CLEAR_ON_WRITE)
> + queue_writel(queue, ISR, MACB_BIT(PTP_PDELAY_RESP_FRAME_TRANSMITTED));
> + if (macb_ptp_time_peer_frame_tx_get(bp, &ts) != 0) {
> + ts.tv_sec = 0;
> + ts.tv_nsec = 0;
> + }
> + macb_ptp_event(bp, &ts);
> + }
> +#endif

I'm not sure of your application and why this is necessary.
Can you please check Andrei's patches and mine and
Richard Cochran's comments?
I use linuxptp to test.

Regards,
Harini