From: Greg Kurz <groug@kaod.org>
To: Alexey Kardashevskiy <aik@ozlabs.ru>
Cc: linux-kernel@vger.kernel.org, linuxppc-dev@lists.ozlabs.org,
Michael Ellerman <mpe@ellerman.id.au>,
Alistair Popple <alistair@popple.id.au>,
stable@vger.kernel.org
Subject: Re: [PATCH] powerpc/powernv/npu: Fix reference leak
Date: Mon, 29 Apr 2019 12:36:59 +0200 [thread overview]
Message-ID: <20190429123659.00c0622b@bahia.lan> (raw)
In-Reply-To: <962c1d9e-719c-cb82-cabc-1cf619e1510b@ozlabs.ru>
On Mon, 29 Apr 2019 16:01:29 +1000
Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> On 20/04/2019 01:34, Greg Kurz wrote:
> > Since 902bdc57451c, get_pci_dev() calls pci_get_domain_bus_and_slot(). This
> > has the effect of incrementing the reference count of the PCI device, as
> > explained in drivers/pci/search.c:
> >
> > * Given a PCI domain, bus, and slot/function number, the desired PCI
> > * device is located in the list of PCI devices. If the device is
> > * found, its reference count is increased and this function returns a
> > * pointer to its data structure. The caller must decrement the
> > * reference count by calling pci_dev_put(). If no device is found,
> > * %NULL is returned.
> >
> > Nothing was done to call pci_dev_put() and the reference count of GPU and
> > NPU PCI devices rockets up.
> >
> > A natural way to fix this would be to teach the callers about the change,
> > so that they call pci_dev_put() when done with the pointer. This turns
> > out to be quite intrusive, as it affects many paths in npu-dma.c,
> > pci-ioda.c and vfio_pci_nvlink2.c.
>
>
> afaict this referencing is only done to protect the current traverser
> and what you've done is actually a natural way (and the generic
> pci_get_dev_by_id() does exactly the same), although this looks a bit weird.
>
Not exactly the same: pci_get_dev_by_id() always increment the refcount
of the returned PCI device. The refcount is only decremented when this
device is passed to pci_get_dev_by_id() to continue searching.
That means that the users of the PCI device pointer returned by
pci_get_dev_by_id() or its exported variants pci_get_subsys(),
pci_get_device() and pci_get_class() do handle the refcount. They
all pass the pointer to pci_dev_put() or continue the search,
which calls pci_dev_put() internally.
Direct and indirect callers of get_pci_dev() don't care for the
refcount at all unless I'm missing something.
>
> > Also, the issue appeared in 4.16 and
> > some affected code got moved around since then: it would be problematic
> > to backport the fix to stable releases.
> >
> > All that code never cared for reference counting anyway. Call pci_dev_put()
> > from get_pci_dev() to revert to the previous behavior.
> >> Fixes: 902bdc57451c ("powerpc/powernv/idoa: Remove unnecessary pcidev
> from pci_dn")
> > Cc: stable@vger.kernel.org # v4.16
> > Signed-off-by: Greg Kurz <groug@kaod.org>
> > ---
> > arch/powerpc/platforms/powernv/npu-dma.c | 15 ++++++++++++++-
> > 1 file changed, 14 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/powerpc/platforms/powernv/npu-dma.c b/arch/powerpc/platforms/powernv/npu-dma.c
> > index e713ade30087..d8f3647e8fb2 100644
> > --- a/arch/powerpc/platforms/powernv/npu-dma.c
> > +++ b/arch/powerpc/platforms/powernv/npu-dma.c
> > @@ -31,9 +31,22 @@ static DEFINE_SPINLOCK(npu_context_lock);
> > static struct pci_dev *get_pci_dev(struct device_node *dn)
> > {
> > struct pci_dn *pdn = PCI_DN(dn);
> > + struct pci_dev *pdev;
> >
> > - return pci_get_domain_bus_and_slot(pci_domain_nr(pdn->phb->bus),
> > + pdev = pci_get_domain_bus_and_slot(pci_domain_nr(pdn->phb->bus),
> > pdn->busno, pdn->devfn);
> > +
> > + /*
> > + * pci_get_domain_bus_and_slot() increased the reference count of
> > + * the PCI device, but callers don't need that actually as the PE
> > + * already holds a reference to the device.
>
> Imho this would be just enough.
>
> Anyway,
>
> Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>
Thanks !
I now realize that I forgot to add the --cc option for stable on my stgit
command line :-\.
Cc'ing now.
>
> How did you find it? :)
>
While reading code to find some inspiration for OpenCAPI passthrough. :)
I saw the following in vfio_pci_ibm_npu2_init():
if (!pnv_pci_get_gpu_dev(vdev->pdev))
return -ENODEV;
and simply followed the function calls.
>
> > Since callers aren't
> > + * aware of the reference count change, call pci_dev_put() now to
> > + * avoid leaks.
> > + */
> > + if (pdev)
> > + pci_dev_put(pdev);
> > +
> > + return pdev;
> > }
> >
> > /* Given a NPU device get the associated PCI device. */
> >
>
next prev parent reply other threads:[~2019-04-29 10:42 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-04-19 15:34 Greg Kurz
2019-04-29 6:01 ` Alexey Kardashevskiy
2019-04-29 10:36 ` Greg Kurz [this message]
2019-05-13 11:56 ` Greg Kurz
2019-05-14 11:13 ` Michael Ellerman
2019-05-14 12:25 ` Greg Kurz
2019-06-03 12:32 ` Michael Ellerman
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=20190429123659.00c0622b@bahia.lan \
--to=groug@kaod.org \
--cc=aik@ozlabs.ru \
--cc=alistair@popple.id.au \
--cc=linux-kernel@vger.kernel.org \
--cc=linuxppc-dev@lists.ozlabs.org \
--cc=mpe@ellerman.id.au \
--cc=stable@vger.kernel.org \
/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®