On Thu, Jan 18, 2024 at 05:28:01PM +0800, Yi Liu wrote:
> On 2024/1/17 20:56, Jason Gunthorpe wrote:
> > On Wed, Jan 17, 2024 at 04:24:24PM +0800, Yi Liu wrote:
> > > Above indeed makes more sense if there can be concurrent
> > > attach/replace/detach
> > > on a single pasid. Just have one doubt should we add lock to protect the
> > > whole attach/replace/detach paths. In the attach/replace path[1] [2], the
> > > xarray entry is verified firstly, and then updated after returning from
> > > iommu attach/replace API. It is uneasy to protect the xarray operations
> > > only
> > > with xa_lock as a detach path can acquire xa_lock right after
> > > attach/replace
> > > path checks the xarray. To avoid it, may need a mutex to protect the whole
> > > attach/replace/detach path to avoid race. Or maybe the attach/replace path
> > > should mark the corresponding entry as a special state that can block the
> > > other path like detach until the attach/replace path update the final
> > > hwpt to
> > > the xarray. Is there such state in xarray?
> >
> > If the caller is not allowed to make concurrent attaches/detaches to
> > the same pasid then you can document that in a comment,
>
> yes. I can document it. Otherwise, we may need a mutex for pasid to allow
> concurrent attaches/detaches.
>
> > but it is
> > still better to use xarray in a self-consistent way.
>
> sure. I'll try. At least in the detach path, xarray should be what you've
> suggested in prior email. Currently in the attach path, the logic is as
> below. Perhaps I can skip the check on old_hwpt since
> iommu_attach_device_pasid() should fail if there is an existing domain
> attached on the pasid. Then the xarray should be more consistent. what
> about your opinion?
>
> old_hwpt = xa_load(&idev->pasid_hwpts, pasid);
> if (old_hwpt) {
> /* Attach does not allow overwrite */
> if (old_hwpt == hwpt)
> return NULL;
> else
> return ERR_PTR(-EINVAL);
> }
>
> rc = iommu_attach_device_pasid(hwpt->domain, idev->dev, pasid);
> if (rc)
> return ERR_PTR(rc);
>
> refcount_inc(&hwpt->obj.users);
> xa_store(&idev->pasid_hwpts, pasid, hwpt, GFP_KERNEL);
Use xa_cmpxchg()
Jason
