[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZMNF5vTTp2IRMsWH@Asurada-Nvidia>
Date: Thu, 27 Jul 2023 21:36:54 -0700
From: Nicolin Chen <nicolinc@...dia.com>
To: "Tian, Kevin" <kevin.tian@...el.com>
CC: "jgg@...dia.com" <jgg@...dia.com>,
"Liu, Yi L" <yi.l.liu@...el.com>,
"joro@...tes.org" <joro@...tes.org>,
"will@...nel.org" <will@...nel.org>,
"robin.murphy@....com" <robin.murphy@....com>,
"alex.williamson@...hat.com" <alex.williamson@...hat.com>,
"shuah@...nel.org" <shuah@...nel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"iommu@...ts.linux.dev" <iommu@...ts.linux.dev>,
"kvm@...r.kernel.org" <kvm@...r.kernel.org>,
"linux-kselftest@...r.kernel.org" <linux-kselftest@...r.kernel.org>,
"mjrosato@...ux.ibm.com" <mjrosato@...ux.ibm.com>,
"farman@...ux.ibm.com" <farman@...ux.ibm.com>
Subject: Re: [PATCH v10 3/6] iommufd: Add iommufd_access_change_ioas(_id)
helpers
On Fri, Jul 28, 2023 at 04:23:03AM +0000, Tian, Kevin wrote:
> > From: Nicolin Chen <nicolinc@...dia.com>
> > Sent: Friday, July 28, 2023 4:25 AM
> >
> > +static int iommufd_access_change_ioas(struct iommufd_access *access,
> > + struct iommufd_ioas *new_ioas)
> > +{
> > + u32 iopt_access_list_id = access->iopt_access_list_id;
> > + struct iommufd_ioas *cur_ioas = access->ioas;
> > + int rc;
> > +
> > + lockdep_assert_held(&access->ioas_lock);
> > +
> > + /* We are racing with a concurrent detach, bail */
> > + if (cur_ioas != access->ioas_unpin)
> > + return -EBUSY;
> > +
> > + if (IS_ERR(new_ioas))
> > + return PTR_ERR(new_ioas);
>
> iommufd_access_change_ioas_id() already checks errors.
I've thought about that: given that iommufd_access_change_ioas
is a standalone API, though it's not used anywhere else at the
moment, it might be safer to have this check again. Otherwise,
we would need a line of comments saying that "caller must make
sure that the input new_ioas is not holding an error code" or
so?
> > +
> > void iommufd_access_destroy_object(struct iommufd_object *obj)
> > {
> > struct iommufd_access *access =
> > container_of(obj, struct iommufd_access, obj);
> >
> > - if (access->ioas) {
> > - iopt_remove_access(&access->ioas->iopt, access,
> > - access->iopt_access_list_id);
> > - refcount_dec(&access->ioas->obj.users);
> > - access->ioas = NULL;
> > - }
> > + mutex_lock(&access->ioas_lock);
> > + if (access->ioas)
> > + WARN_ON(iommufd_access_change_ioas(access, NULL));
> > + mutex_unlock(&access->ioas_lock);
> > iommufd_ctx_put(access->ictx);
> > }
>
> this changes the behavior of destroy. Previously it always removes
> the access w/o detecting race while now it will give up and throw
> out a warning.
You mean the -EBUSY case? That's a good catch..
> While I'm fine with this change from bisec p.o.v.
> it might be good to split this into a separate patch.
Yea, I can do that.
> > void iommufd_access_detach(struct iommufd_access *access)
> > {
> > - struct iommufd_ioas *cur_ioas = access->ioas;
> > + int rc;
> >
> > mutex_lock(&access->ioas_lock);
> > - if (WARN_ON(!access->ioas))
> > - goto out;
> > - /*
> > - * Set ioas to NULL to block any further iommufd_access_pin_pages().
> > - * iommufd_access_unpin_pages() can continue using access-
> > >ioas_unpin.
> > - */
> > - access->ioas = NULL;
> > -
> > - if (access->ops->unmap) {
> > + if (WARN_ON(!access->ioas)) {
> > mutex_unlock(&access->ioas_lock);
> > - access->ops->unmap(access->data, 0, ULONG_MAX);
> > - mutex_lock(&access->ioas_lock);
> > + return;
> > }
> > - iopt_remove_access(&cur_ioas->iopt, access,
> > - access->iopt_access_list_id);
> > - refcount_dec(&cur_ioas->obj.users);
> > -out:
> > - access->ioas_unpin = NULL;
> > + rc = iommufd_access_change_ioas(access, NULL);
> > + WARN_ON(rc);
>
> 'rc' can be removed.
>
> Just "WARN_ON(iommufd_access_change_ioas(access, NULL));"
Will do that in v11.
> otherwise looks good to me,
>
> Reviewed-by: Kevin Tian <kevin.tian@...el.com>
Thanks!
Nic
Powered by blists - more mailing lists