Re: [PATCH v7 04/10] iommu: Extend domain attach group with handle support

[Baolu Lu <baolu.lu@linux.intel.com>]

On 2024/6/29 5:06, Jason Gunthorpe wrote:
> On Sun, Jun 16, 2024 at 02:11:49PM +0800, Lu Baolu wrote:
> 
>> +int iommu_replace_group_handle(struct iommu_group *group,
>> +			       struct iommu_domain *new_domain,
>> +			       struct iommu_attach_handle *handle)
>> +{
>> +	struct iommu_domain *old_domain = group->domain;
>> +	void *curr;
>> +	int ret;
>> +
>> +	if (!new_domain)
>> +		return -EINVAL;
>> +
>> +	mutex_lock(&group->mutex);
>> +	ret = __iommu_group_set_domain(group, new_domain);
>> +	if (ret)
>> +		goto err_unlock;
>> +	xa_erase(&group->pasid_array, IOMMU_NO_PASID);
>> +	if (handle) {
>> +		curr = xa_store(&group->pasid_array, IOMMU_NO_PASID, handle, GFP_KERNEL);
>> +		if (xa_err(curr)) {
>> +			ret = xa_err(curr);
>> +			goto err_restore;
> But this error unwind doesn't work because the xa_erase() already
> happened and there may have been a handle there that we don't put
> back.

Yes, you are right.

> 
> Something like this - store to a reserved entry cannot fail:

This code looks good to me. Thanks!

Best regards,
baolu