mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Xu Yilun <yilun.xu@linux.intel.com>
To: Nicolin Chen <nicolinc@nvidia.com>
Cc: jgg@nvidia.com, jgg@ziepe.ca, kevin.tian@intel.com,
	will@kernel.org, aneesh.kumar@kernel.org, iommu@lists.linux.dev,
	linux-kernel@vger.kernel.org, joro@8bytes.org,
	robin.murphy@arm.com, shuah@kernel.org, aik@amd.com,
	dan.j.williams@intel.com, baolu.lu@linux.intel.com,
	yilun.xu@intel.com
Subject: Re: [PATCH v4 4/7] iommufd: Destroy vdevice on idevice destroy
Date: Tue, 15 Jul 2025 00:40:40 +0800	[thread overview]
Message-ID: <aHUzCAM8NKuFYbj3@yilunxu-OptiPlex-7050> (raw)
In-Reply-To: <aHKfwQ41x28bNWAL@yilunxu-OptiPlex-7050>

On Sun, Jul 13, 2025 at 01:47:45AM +0800, Xu Yilun wrote:
> > [..]
> > 
> > >  out_put_idev:
> > > -	iommufd_put_object(ucmd->ictx, &idev->obj);
> > > +	if (rc)
> > > +		iommufd_put_object(ucmd->ictx, &idev->obj);
> > 
> > So, this actually holds both the short term user and (covertly)
> > the regular user too.
> > 
> > Though it doesn't hurt to do that, holding the regular one seems
> > to be useless, because its refcount check is way behind this new
> > pre_destroy op:
> > 
> > 183         if (flags & REMOVE_WAIT_SHORTTERM) {                                                                                                                                                                                                                                                                                                             
> > 184                 ret = iommufd_object_dec_wait_shortterm(ictx, to_destroy);
> > ==>			/* pre_destroy op */
> > ...
> > 214         if (!refcount_dec_if_one(&obj->users)) {
> > 215                 ret = -EBUSY;
> > 216                 goto err_xa;
> > 217         }
> > 
> > So, I think we could just do, exactly reflecting the comments:
> > +	vdev->idev = idev;
> > +	refcount_inc(&idev->obj.shortterm_users);
> 
> I think this makes things more clear, we have 3 types of refcounts:
> 
> 1. shortterm_users + 1, users + 1, temporary refcount, can't be
>    revoked by referenced object.
> 2. users + 1, long term refcount, can't be revoked either.
> 3. shortterm_users + 1, can be revoked by referenced object.
> 
> > 
> > Then, keep the exit patch unchanged.
> > out_put_idev:
> > 	iommufd_put_object(ucmd->ictx, &idev->obj);
> 
> Yeah, this is the part I like the most.

Sorry, I feel the actual coding is not as good as I image, the

  refcount_dec(&idev->obj.shortterm_users)

can't be called in iommufd_vdevice_abort() cause it may lead to idev
free but we will then do

  mutex_unlock(idev->igroup->lock)

I think the following change makes things more complex...


diff --git a/drivers/iommu/iommufd/viommu.c b/drivers/iommu/iommufd/viommu.c
index 702ae248df17..bdd5a5227cbf 100644
--- a/drivers/iommu/iommufd/viommu.c
+++ b/drivers/iommu/iommufd/viommu.c
@@ -128,7 +128,8 @@ void iommufd_vdevice_destroy(struct iommufd_object *obj)
        mutex_lock(&idev->igroup->lock);
        iommufd_vdevice_abort(obj);
        mutex_unlock(&idev->igroup->lock);
-       iommufd_put_object(idev->ictx, &idev->obj);
+       refcount_dec(&idev->obj.shortterm_users);
+       wake_up_interruptible_all(&vdev->viommu->ictx->destroy_wait);
 }

 int iommufd_vdevice_alloc_ioctl(struct iommufd_ucmd *ucmd)
@@ -185,6 +186,7 @@ int iommufd_vdevice_alloc_ioctl(struct iommufd_ucmd *ucmd)
         * idev real destruction.
         */
        vdev->idev = idev;
+       refcount_inc(&idev->obj.shortterm_users);

        /*
         * iommufd_device_destroy() delays until idev->vdev is NULL before
@@ -207,12 +209,12 @@ int iommufd_vdevice_alloc_ioctl(struct iommufd_ucmd *ucmd)
        goto out_unlock_igroup;

 out_abort:
+       refcount_dec(&idev->obj.shortterm_users);
        iommufd_object_abort_and_destroy(ucmd->ictx, &vdev->obj);
 out_unlock_igroup:
        mutex_unlock(&idev->igroup->lock);
 out_put_idev:
-       if (rc)
-               iommufd_put_object(ucmd->ictx, &idev->obj);
+       iommufd_put_object(ucmd->ictx, &idev->obj);
 out_put_viommu:
        iommufd_put_object(ucmd->ictx, &viommu->obj);
        return rc;


Thanks,
Yilun

  reply	other threads:[~2025-07-14 16:49 UTC|newest]

Thread overview: 38+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-07-09  4:02 [PATCH v4 0/7] iommufd: Destroy vdevice on device unbind Xu Yilun
2025-07-09  4:02 ` [PATCH v4 1/7] iommufd/viommu: Roll back to use iommufd_object_alloc() for vdevice Xu Yilun
2025-07-10  7:32   ` Tian, Kevin
2025-07-10 18:26   ` Nicolin Chen
2025-07-11 17:42   ` Jason Gunthorpe
2025-07-09  4:02 ` [PATCH v4 2/7] iommufd: Add iommufd_object_tombstone_user() helper Xu Yilun
2025-07-10  7:35   ` Tian, Kevin
2025-07-10 19:08   ` Nicolin Chen
2025-07-11 15:02     ` Xu Yilun
2025-07-11 19:39       ` Nicolin Chen
2025-07-11 21:33       ` Jason Gunthorpe
2025-07-11 22:04         ` Nicolin Chen
2025-07-11 17:44   ` Jason Gunthorpe
2025-07-09  4:02 ` [PATCH v4 3/7] iommufd: Add a pre_destroy() op for objects Xu Yilun
2025-07-10  7:40   ` Tian, Kevin
2025-07-10 17:15     ` Jason Gunthorpe
2025-07-11  3:16       ` Tian, Kevin
2025-07-11  9:38         ` Xu Yilun
2025-07-11 22:09   ` Nicolin Chen
2025-07-09  4:02 ` [PATCH v4 4/7] iommufd: Destroy vdevice on idevice destroy Xu Yilun
2025-07-10  8:03   ` Tian, Kevin
2025-07-12  2:22   ` Nicolin Chen
2025-07-12 17:47     ` Xu Yilun
2025-07-14 16:40       ` Xu Yilun [this message]
2025-07-14 16:53         ` Jason Gunthorpe
2025-07-14 17:34           ` Nicolin Chen
2025-07-14 18:03             ` Xu Yilun
2025-07-12  2:26   ` Nicolin Chen
2025-07-09  4:02 ` [PATCH v4 5/7] iommufd/vdevice: Remove struct device reference from struct vdevice Xu Yilun
2025-07-10  8:04   ` Tian, Kevin
2025-07-12  2:27   ` Nicolin Chen
2025-07-09  4:02 ` [PATCH v4 6/7] iommufd/selftest: Explicitly skip tests for inapplicable variant Xu Yilun
2025-07-10  8:06   ` Tian, Kevin
2025-07-12  2:39   ` Nicolin Chen
2025-07-09  4:02 ` [PATCH v4 7/7] iommufd/selftest: Add coverage for vdevice tombstone Xu Yilun
2025-07-10  8:10   ` Tian, Kevin
2025-07-10  9:25     ` Xu Yilun
2025-07-11  3:12       ` Tian, Kevin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=aHUzCAM8NKuFYbj3@yilunxu-OptiPlex-7050 \
    --to=yilun.xu@linux.intel.com \
    --cc=aik@amd.com \
    --cc=aneesh.kumar@kernel.org \
    --cc=baolu.lu@linux.intel.com \
    --cc=dan.j.williams@intel.com \
    --cc=iommu@lists.linux.dev \
    --cc=jgg@nvidia.com \
    --cc=jgg@ziepe.ca \
    --cc=joro@8bytes.org \
    --cc=kevin.tian@intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=nicolinc@nvidia.com \
    --cc=robin.murphy@arm.com \
    --cc=shuah@kernel.org \
    --cc=will@kernel.org \
    --cc=yilun.xu@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®