Re: [PATCH v15 4/7] s390/pci: Store PCI error information for passthrough devices

[Farhan Ali <alifm@linux.ibm.com>]

On 5/6/2026 2:38 AM, Niklas Schnelle wrote:
>> -static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev)
>> +static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev,
>> +							  struct zpci_ccdf_err *ccdf)
>>   {
>>   	pci_ers_result_t ers_res = PCI_ERS_RESULT_DISCONNECT;
>>   	struct zpci_dev *zdev = to_zpci(pdev);
>> @@ -194,13 +206,6 @@ static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev)
>>   	}
>>   	pdev->error_state = pci_channel_io_frozen;
>>   
>> -	if (is_passed_through(pdev)) {
>> -		pr_info("%s: Cannot be recovered in the host because it is a pass-through device\n",
>> -			pci_name(pdev));
>> -		status_str = "failed (pass-through)";
>> -		goto out_unlock;
>> -	}
>> -
>>   	driver = to_pci_driver(pdev->dev.driver);
>>   	if (!is_driver_supported(driver)) {
>>   		if (!driver) {
>> @@ -216,12 +221,25 @@ static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev)
>>   		goto out_unlock;
>>   	}
>>   
>> +	zpci_store_pci_error(pdev, ccdf);
> Sashiko notes that zdev->pendings_errs.mediated_recovery could become
> true between the above zpci_store_pci_error() and the below check for
> leaving recovery to user-space. I think we could make a general
> improvement that also tackles this concern. The ideas is that we could
> have zpci_store_pci_error() return true if it did store the error and
> we are in mediated recovery mode. Then we use that as the signal to
> skip host recovery below. That way we also don't need to retake the
> pending_errs_lock which makes the below much simpler and it would be a
> win independent of the race. As for the race this would make sure that
> we either do the host recovery or store the error and let user-space
> recover.

I did think of the concern about mediated_recovery becoming true after 
zpci_store_pci_error(), but IIUC in that case we won't even be able to 
deliver the error signal to userspace (via error_detected()). And I 
don't think mediated_recovery flag can be set to true. Since we are 
holding the pci device lock, vfio_pci_core_enable() will fail as it will 
fail trying to reset the device.

Thanks

Farhan

>>   	ers_res = zpci_event_notify_error_detected(pdev, driver);
>>   	if (ers_result_indicates_abort(ers_res)) {
>>   		status_str = "failed (abort on detection)";
>>   		goto out_unlock;
>>   	}
>>   
>> +	mutex_lock(&zdev->pending_errs_lock);
>> +	if (zdev->pending_errs.mediated_recovery) {
>> +		pr_info("%s: Leaving recovery of pass-through device to user-space\n",
>> +			pci_name(pdev));
>> +		ers_res = PCI_ERS_RESULT_RECOVERED;
>> +		status_str = "in progress";
>> +		mutex_unlock(&zdev->pending_errs_lock);
>> +		goto out_unlock;
>> +	}
>> +	mutex_unlock(&zdev->pending_errs_lock);