Re: [PATCH v2 10/11] iommu/vt-d: Use xarray for global device_domain_info

[Lu Baolu]

Hi Christoph,

On 2/14/22 3:38 PM, Christoph Hellwig wrote:
> >   
> >   const struct iommu_ops intel_iommu_ops;
> > @@ -903,7 +905,8 @@ static void pgtable_walk(struct intel_iommu *iommu, unsigned long pfn, u8 bus, u
> >   	struct dmar_domain *domain;
> >   	int offset, level;
> >   
> > -	info = dmar_search_domain_by_dev_info(iommu->segment, bus, devfn);
> > +	info = xa_load(&device_domain_array,
> > +		       devi_idx(iommu->segment, bus, devfn));
> >   	if (!info || !info->domain) {
> >   		pr_info("device [%02x:%02x.%d] not probed\n",
> >   			bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
> > @@ -1742,14 +1745,14 @@ static int iommu_init_domains(struct intel_iommu *iommu)
>
> Don't we need a rcu critical section here?
>
> > -	list_for_each_entry_safe(info, tmp, &device_domain_list, global) {
> > +	rcu_read_lock();
> > +	xa_for_each(&device_domain_array, index, info) {
> >   		if (info->iommu != iommu)
> >   			continue;
> >   
> > @@ -1758,7 +1761,7 @@ static void disable_dmar_iommu(struct intel_iommu *iommu)
> >   
> >   		__dmar_remove_one_dev_info(info);
> >   	}
> > -	spin_unlock_irqrestore(&device_domain_lock, flags);
> > +	rcu_read_unlock();
>
> __dmar_remove_one_dev_info asserts that device_domain_lock is held,
> which these two hunks remove.
>
> >   	spin_lock_irqsave(&device_domain_lock, flags);
> >   	dev_iommu_priv_set(dev, NULL);
> > -	list_del(&info->global);
> > +	xa_erase(&device_domain_array,
> > +		 devi_idx(info->segment, info->bus, info->devfn));
> >   	spin_unlock_irqrestore(&device_domain_lock, flags);
> >   
> >   	kfree(info);
>
> Do we need a kfree_rcu here to allow rcu based access?

Thanks for your time and very appreciated for your comments. As replied
to Jason, I will stop 10/11 and 11/11 for now. From the current usage
scenario, the value of such refactoring is limited.

Best regards,
baolu