Re: [PATCH 13/13] rapidio/switches: add driver for IDT gen3 switches

[Andrew Morton]

On Thu, 21 Jul 2016 14:18:54 -0400 Alexandre Bounine <alexandre.bounine@xxxxxxx> wrote:

> Add RapidIO switch driver for IDT Gen3 switch devices: RXS1632 and RXS2448.
> 
> ...
>
> +static int
> +idtg3_em_handler(struct rio_dev *rdev, u8 pnum)
> +{
> +	u32 err_status;
> +	u32 rval;
> +
> +	rio_read_config_32(rdev,
> +			RIO_DEV_PORT_N_ERR_STS_CSR(rdev, pnum),
> +			&err_status);
> +
> +	/* Do nothing for device/link removal */
> +	if (err_status & RIO_PORT_N_ERR_STS_PORT_UNINIT)
> +		return 0;
> +
> +	/* When link is OK we have a device insertion.
> +	 * Request port soft reset to clear errors if they present.
> +	 * Inbound and outbound ackIDs will be 0 after reset.
> +	 */
> +	if (err_status & (RIO_PORT_N_ERR_STS_OUT_ES |
> +				RIO_PORT_N_ERR_STS_INP_ES)) {
> +		rio_read_config_32(rdev, RIO_PLM_SPx_IMP_SPEC_CTL(pnum), &rval);
> +		rio_write_config_32(rdev, RIO_PLM_SPx_IMP_SPEC_CTL(pnum),
> +				    rval | RIO_PLM_SPx_IMP_SPEC_CTL_SOFT_RST);
> +		udelay(10);
> +		rio_write_config_32(rdev, RIO_PLM_SPx_IMP_SPEC_CTL(pnum), rval);
> +		mdelay(500);

Yikes, that's a very long busywait.  A half-second machine freeze on
uniprocessor (if such exists ;)).  We can't use msleep()?

> +	}
> +
> +	return 0;
> +}
>
> ...
>