Re: [PATCH v10 3/5] hisi_acc_vfio_pci: register debugfs for hisilicon migration driver

[liulongfang]

On 2023/4/21 11:27, liulongfang wrote:
> On 2023/4/14 20:24, Jason Gunthorpe wrote:
>> On Sat, Apr 08, 2023 at 03:42:22PM +0800, Longfang Liu wrote:
>>> +static int hisi_acc_vf_debug_restore(struct seq_file *seq, void *data)
>>> +{
>>> +	struct device *vf_dev = seq->private;
>>> +	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
>>> +	struct vfio_device	*vdev = &core_device->vdev;
>>> +	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
>>> +	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
>>> +	int ret;
>>> +
>>> +	ret = hisi_acc_vf_debug_check(seq, vdev);
>>> +	if (ret)
>>> +		goto restore_err;
>>> +
>>> +	ret = vf_qm_state_save(hisi_acc_vdev, migf);
>>> +	if (ret) {
>>> +		seq_printf(seq, "%s\n", "failed to save device data!");
>>> +		goto restore_err;
>>> +	}
>>> +
>>> +	ret = vf_qm_check_match(hisi_acc_vdev, migf);
>>> +	if (ret) {
>>> +		seq_printf(seq, "%s\n", "failed to match the VF!");
>>> +		goto restore_err;
>>> +	}
>>> +
>>> +	ret = vf_qm_load_data(hisi_acc_vdev, migf);
>>> +	if (ret) {
>>> +		seq_printf(seq, "%s\n", "failed to recover the VF!");
>>> +		goto restore_err;
>>> +	}
>>> +
>>> +	vf_qm_fun_reset(&hisi_acc_vdev->vf_qm);
>>> +	seq_printf(seq, "%s\n", "successful to resume device data!");
>>> +
>>> +restore_err:
>>> +	return 0;
>>> +}
>>
>> This is basically an in-kernel self test, it should be protected with
>> some kind of VFIO selftest kconfig.
>>
> As a debugfs function, its usage will be more flexible for users.
> 
>> Though, I wonder why we need it??? 
> After a live migration error occurs. Through this debugfs function,
> you can perform separate functional tests on the source and destination
> to locate the cause of the error.
> 
> Can't you write a trivial userspace
>> program under tools/testing to do this sequence with the ioctls?
>>
> Sorry, I still wish this feature was a simple debugfs feature.
> If you want the userspace testing tool you mentioned,
> you can try it on mlx5.
> 
> Thanks,
> Longfang.
>>> +static int hisi_acc_vf_debug_save(struct seq_file *seq, void *data)
>>> +{
>>> +	struct device *vf_dev = seq->private;
>>> +	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
>>> +	struct vfio_device	*vdev = &core_device->vdev;
>>> +	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
>>> +	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
>>> +	int ret;
>>> +
>>> +	ret = hisi_acc_vf_debug_check(seq, vdev);
>>> +	if (ret)
>>> +		goto save_err;
>>> +
>>> +	ret = vf_qm_state_save(hisi_acc_vdev, migf);
>>> +	if (ret) {
>>> +		seq_printf(seq, "%s\n", "failed to save device data!");
>>> +		goto save_err;
>>> +	}
>>> +	seq_printf(seq, "%s\n", "successful to save device data!");
>>> +
>>> +save_err:
>>> +	return 0;
>>> +}
>>
>> Same kind of commen there, this is a selftest, why does it need a
>> special kernel interface?
>>
>> .. and so on..
>>
>> I thought the non-selftesty bits were OK, maybe split the patch to
>> match progress
>>

Thank you for your suggestion, but the current debugfs method can already
meet the functional requirements of verification testing and
problem location.

Thanks,
Longfang.
>> Jason
>> .
>>