Re: [PATCH v2] usb: typec: ucsi: retry init when the PPM answers early commands incorrectly

[Heikki Krogerus <heikki.krogerus@linux.intel.com>]

On Mon, Aug 24, 2026 at 10:35:36AM -0700, Jacob Riff wrote:
> On some platforms the PPM is not ready to answer commands correctly
> for a short window during boot. On the Lenovo ThinkPad X1 Carbon
> Gen 14 (21V7, tested on BIOS 1.12 and 1.14) roughly half of all boots
> fail ucsi_init() with either -ENODEV (GET_CAPABILITY completes but
> reports zero connectors) or -EINVAL (a standard command is rejected,
> logged as "possible UCSI driver bug"). The failure is not a timeout:
> increasing the sync command completion wait does not change the rate.
> 
> Since ucsi_init_work() only requeues on -EPROBE_DEFER, a single bad
> answer during that window leaves UCSI dead for the whole session, so
> Type-C events are never handled again; most visibly, the machine
> silently never resumes charging after the charger is replugged.
> Manually reloading ucsi_acpi a few seconds later has succeeded on
> every attempt observed, which suggests simply retrying is enough.
> 
> Retry -ENODEV and -EINVAL the same way as the role switch wait, log
> the retries at debug level, keep the loud report for the case where
> the retries are exhausted, and note when init only succeeded after
> retrying.
> 
> Tested on the affected machine: across 8 consecutive boots with this
> patch, 5 hit the failure (matching the historical ~50-60% rate) and
> all 5 recovered on the first retry ("PPM init succeeded after 2
> attempts"). 0 of 8 boots ended with UCSI unusable, where ~5 of 8
> would have without the patch.
> 
> Signed-off-by: Jacob Riff <jacob@riff.dk>

Acked-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>

> ---
> v2: no functional change; resent without the RFC tag.
>   - Rebased on usb-next; applies unchanged.
>   - Discussion on v1: Huang Wei asked whether the -EINVAL retry should
>     distinguish a not-ready PPM from a genuine driver bug, since
>     ucsi_read_error() maps UCSI_ERROR_INVALID_CON_NUM /
>     UNREGONIZED_CMD / INVALID_CMD_ARGUMENT to -EINVAL. The two cases
>     are indistinguishable by error code here, so the patch relies on
>     context and persistence instead: the retry exists only in the
>     ucsi_init_work() path, it is bounded, and the dev_err() in
>     ucsi_read_error() is untouched and still fires on the first
>     occurrence. No code change resulted.
> 
> v1: https://lore.kernel.org/linux-usb/20260815005724.8741-1-jacob@riff.dk/
> 
> --- a/drivers/usb/typec/ucsi/ucsi.c
> +++ b/drivers/usb/typec/ucsi/ucsi.c
> @@ -2212,18 +2212,36 @@
>  	int ret;
>  
>  	ret = ucsi_init(ucsi);
> -	if (ret)
> -		dev_err_probe(ucsi->dev, ret, "PPM init failed\n");
> +	if (!ret) {
> +		if (ucsi->work_count)
> +			dev_info(ucsi->dev,
> +				 "PPM init succeeded after %u attempts\n",
> +				 ucsi->work_count + 1);
> +		return;
> +	}
>  
> -	if (ret == -EPROBE_DEFER) {
> -		if (ucsi->work_count++ > UCSI_ROLE_SWITCH_WAIT_COUNT) {
> -			dev_err(ucsi->dev, "PPM init failed, stop trying\n");
> +	/*
> +	 * On some platforms the PPM is not ready to answer commands
> +	 * correctly for a short window during boot: standard commands are
> +	 * rejected or GET_CAPABILITY reports zero connectors, seen as
> +	 * -EINVAL or -ENODEV from ucsi_init(), and a retry moments later
> +	 * succeeds (observed on Lenovo ThinkPad X1 Carbon Gen 14, where
> +	 * this affects roughly half of all boots). Retry those like the
> +	 * USB role switch wait instead of giving up on the first attempt.
> +	 */
> +	if (ret == -EPROBE_DEFER || ret == -ENODEV || ret == -EINVAL) {
> +		if (ucsi->work_count++ < UCSI_ROLE_SWITCH_WAIT_COUNT) {
> +			dev_dbg(ucsi->dev, "PPM init failed (%pe), retrying\n",
> +				ERR_PTR(ret));
> +			queue_delayed_work(system_dfl_long_wq, &ucsi->work,
> +					   UCSI_ROLE_SWITCH_INTERVAL);
>  			return;
>  		}
> -
> -		queue_delayed_work(system_dfl_long_wq, &ucsi->work,
> -				   UCSI_ROLE_SWITCH_INTERVAL);
> +		dev_err(ucsi->dev, "PPM init failed, stop trying\n");
> +		return;
>  	}
> +
> +	dev_err_probe(ucsi->dev, ret, "PPM init failed\n");
>  }
>  
>  /**
> 
> base-commit: e1e6e541c5c9cf548e9fdc35fc26808c82074440

-- 
heikki