[Skiboot] [RFC 03/12] npu2-hw-procedures: Fix link retraining on reset

[Andrew Donnellan]

On 19/6/19 10:45 pm, Frederic Barrat wrote:
> Link retraining was showing reliability problems due to some
> opencapi-only settings not being optimized. This patch updates some
> extra PHY state, as agreed with the PHY team. Though they mostly
> impact link retraining behavior, they should also be set at boot.
> 
> Signed-off-by: Frederic Barrat <fbarrat at linux.ibm.com>

I haven't cross checked this against the updated PHY settings you rely on.

With that caveat:

Reviewed-by: Andrew Donnellan <ajd at linux.ibm.com>

> ---
>   hw/npu2-hw-procedures.c | 16 ++++++++++++++++
>   1 file changed, 16 insertions(+)
> 
> diff --git a/hw/npu2-hw-procedures.c b/hw/npu2-hw-procedures.c
> index 548dae16..98e93191 100644
> --- a/hw/npu2-hw-procedures.c
> +++ b/hw/npu2-hw-procedures.c
> @@ -68,8 +68,14 @@ static struct npu2_phy_reg NPU2_PHY_RX_PR_FW_OFF	   = {0x08a, 56, 1};
>   static struct npu2_phy_reg NPU2_PHY_RX_PR_FW_INERTIA_AMT   = {0x08a, 57, 3};
>   static struct npu2_phy_reg NPU2_PHY_RX_CFG_LTE_MC	   = {0x000, 60, 4};
>   static struct npu2_phy_reg NPU2_PHY_RX_A_INTEG_COARSE_GAIN = {0x00a, 48, 4};
> +static struct npu2_phy_reg NPU2_PHY_RX_A_CTLE_COARSE	   = {0x00c, 48, 5};
> +static struct npu2_phy_reg NPU2_PHY_RX_A_CTLE_GAIN	   = {0x00c, 53, 4};
>   static struct npu2_phy_reg NPU2_PHY_RX_B_INTEG_COARSE_GAIN = {0x026, 48, 4};
> +static struct npu2_phy_reg NPU2_PHY_RX_B_CTLE_COARSE	   = {0x028, 48, 5};
> +static struct npu2_phy_reg NPU2_PHY_RX_B_CTLE_GAIN	   = {0x028, 53, 4};
>   static struct npu2_phy_reg NPU2_PHY_RX_E_INTEG_COARSE_GAIN = {0x030, 48, 4};
> +static struct npu2_phy_reg NPU2_PHY_RX_E_CTLE_COARSE	   = {0x032, 48, 5};
> +static struct npu2_phy_reg NPU2_PHY_RX_E_CTLE_GAIN	   = {0x032, 53, 4};
>   
>   /* These registers are per-PHY, not per lane */
>   static struct npu2_phy_reg NPU2_PHY_RX_SPEED_SELECT	       = {0x262, 51, 2};
> @@ -437,6 +443,16 @@ static uint32_t phy_reset_complete(struct npu2_dev *ndev)
>   		phy_write_lane(ndev, &NPU2_PHY_RX_A_INTEG_COARSE_GAIN, lane, 11);
>   		phy_write_lane(ndev, &NPU2_PHY_RX_B_INTEG_COARSE_GAIN, lane, 11);
>   		phy_write_lane(ndev, &NPU2_PHY_RX_E_INTEG_COARSE_GAIN, lane, 11);
> +
> +		if (ndev->type == NPU2_DEV_TYPE_OPENCAPI) {
> +			phy_write_lane(ndev, &NPU2_PHY_RX_A_CTLE_GAIN, lane, 0);
> +			phy_write_lane(ndev, &NPU2_PHY_RX_B_CTLE_GAIN, lane, 0);
> +			phy_write_lane(ndev, &NPU2_PHY_RX_E_CTLE_GAIN, lane, 0);
> +
> +			phy_write_lane(ndev, &NPU2_PHY_RX_A_CTLE_COARSE, lane, 20);
> +			phy_write_lane(ndev, &NPU2_PHY_RX_B_CTLE_COARSE, lane, 20);
> +			phy_write_lane(ndev, &NPU2_PHY_RX_E_CTLE_COARSE, lane, 20);
> +		}
>   	}
>   
>   	set_iovalid(ndev, true);
> 

-- 
Andrew Donnellan              OzLabs, ADL Canberra
ajd at linux.ibm.com             IBM Australia Limited