2021-11-25 18:11:47

by Parav Pandit

[permalink] [raw]
Subject: [PATCH] vdpa: Consider device id larger than 31

virtio device id value can be more than 31. Hence, use BIT_ULL in
assignment.

Fixes: 33b347503f01 ("vdpa: Define vdpa mgmt device, ops and a netlink interface")
Reported-by: kernel test robot <[email protected]>
Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Parav Pandit <[email protected]>
---
drivers/vdpa/vdpa.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
index 7332a74a4b00..e91c71aeeddf 100644
--- a/drivers/vdpa/vdpa.c
+++ b/drivers/vdpa/vdpa.c
@@ -404,7 +404,7 @@ static int vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *m
goto msg_err;

while (mdev->id_table[i].device) {
- supported_classes |= BIT(mdev->id_table[i].device);
+ supported_classes |= BIT_ULL(mdev->id_table[i].device);
i++;
}

--
2.26.2



2021-11-26 02:52:00

by Jason Wang

[permalink] [raw]
Subject: Re: [PATCH] vdpa: Consider device id larger than 31

On Fri, Nov 26, 2021 at 2:09 AM Parav Pandit <[email protected]> wrote:
>
> virtio device id value can be more than 31. Hence, use BIT_ULL in
> assignment.
>
> Fixes: 33b347503f01 ("vdpa: Define vdpa mgmt device, ops and a netlink interface")
> Reported-by: kernel test robot <[email protected]>
> Reported-by: Dan Carpenter <[email protected]>
> Signed-off-by: Parav Pandit <[email protected]>

Acked-by: Jason Wang <[email protected]>

> ---
> drivers/vdpa/vdpa.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
> index 7332a74a4b00..e91c71aeeddf 100644
> --- a/drivers/vdpa/vdpa.c
> +++ b/drivers/vdpa/vdpa.c
> @@ -404,7 +404,7 @@ static int vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *m
> goto msg_err;
>
> while (mdev->id_table[i].device) {
> - supported_classes |= BIT(mdev->id_table[i].device);
> + supported_classes |= BIT_ULL(mdev->id_table[i].device);
> i++;
> }
>
> --
> 2.26.2
>


2021-11-28 07:34:15

by Eli Cohen

[permalink] [raw]
Subject: Re: [PATCH] vdpa: Consider device id larger than 31

On Fri, Nov 26, 2021 at 10:48:12AM +0800, Jason Wang wrote:
> On Fri, Nov 26, 2021 at 2:09 AM Parav Pandit <[email protected]> wrote:
> >
> > virtio device id value can be more than 31. Hence, use BIT_ULL in
> > assignment.
> >
> > Fixes: 33b347503f01 ("vdpa: Define vdpa mgmt device, ops and a netlink interface")
> > Reported-by: kernel test robot <[email protected]>
> > Reported-by: Dan Carpenter <[email protected]>
> > Signed-off-by: Parav Pandit <[email protected]>
>
> Acked-by: Jason Wang <[email protected]>
>
> > ---
> > drivers/vdpa/vdpa.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
> > index 7332a74a4b00..e91c71aeeddf 100644
> > --- a/drivers/vdpa/vdpa.c
> > +++ b/drivers/vdpa/vdpa.c
> > @@ -404,7 +404,7 @@ static int vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *m
> > goto msg_err;
> >
> > while (mdev->id_table[i].device) {
> > - supported_classes |= BIT(mdev->id_table[i].device);
> > + supported_classes |= BIT_ULL(mdev->id_table[i].device);
> > i++;
> > }
> >

type of mdev->id_table[i].device is __u32 so in theory you're limited
to device ID's up to 63.
Maybe add assert here?

> > --
> > 2.26.2
> >
>

2021-11-28 07:47:47

by Michael S. Tsirkin

[permalink] [raw]
Subject: Re: [PATCH] vdpa: Consider device id larger than 31

On Sun, Nov 28, 2021 at 09:14:35AM +0200, Eli Cohen wrote:
> On Fri, Nov 26, 2021 at 10:48:12AM +0800, Jason Wang wrote:
> > On Fri, Nov 26, 2021 at 2:09 AM Parav Pandit <[email protected]> wrote:
> > >
> > > virtio device id value can be more than 31. Hence, use BIT_ULL in
> > > assignment.
> > >
> > > Fixes: 33b347503f01 ("vdpa: Define vdpa mgmt device, ops and a netlink interface")
> > > Reported-by: kernel test robot <[email protected]>
> > > Reported-by: Dan Carpenter <[email protected]>
> > > Signed-off-by: Parav Pandit <[email protected]>
> >
> > Acked-by: Jason Wang <[email protected]>
> >
> > > ---
> > > drivers/vdpa/vdpa.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
> > > index 7332a74a4b00..e91c71aeeddf 100644
> > > --- a/drivers/vdpa/vdpa.c
> > > +++ b/drivers/vdpa/vdpa.c
> > > @@ -404,7 +404,7 @@ static int vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *m
> > > goto msg_err;
> > >
> > > while (mdev->id_table[i].device) {
> > > - supported_classes |= BIT(mdev->id_table[i].device);
> > > + supported_classes |= BIT_ULL(mdev->id_table[i].device);
> > > i++;
> > > }
> > >
>
> type of mdev->id_table[i].device is __u32 so in theory you're limited
> to device ID's up to 63.
> Maybe add assert here?

I think 1 << 63 is illegal though.

> > > --
> > > 2.26.2
> > >
> >


2021-11-29 07:01:26

by Dan Carpenter

[permalink] [raw]
Subject: Re: [PATCH] vdpa: Consider device id larger than 31

On Sun, Nov 28, 2021 at 09:14:35AM +0200, Eli Cohen wrote:
> On Fri, Nov 26, 2021 at 10:48:12AM +0800, Jason Wang wrote:
> > On Fri, Nov 26, 2021 at 2:09 AM Parav Pandit <[email protected]> wrote:
> > >
> > > virtio device id value can be more than 31. Hence, use BIT_ULL in
> > > assignment.
> > >
> > > Fixes: 33b347503f01 ("vdpa: Define vdpa mgmt device, ops and a netlink interface")
> > > Reported-by: kernel test robot <[email protected]>
> > > Reported-by: Dan Carpenter <[email protected]>
> > > Signed-off-by: Parav Pandit <[email protected]>
> >
> > Acked-by: Jason Wang <[email protected]>
> >
> > > ---
> > > drivers/vdpa/vdpa.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c
> > > index 7332a74a4b00..e91c71aeeddf 100644
> > > --- a/drivers/vdpa/vdpa.c
> > > +++ b/drivers/vdpa/vdpa.c
> > > @@ -404,7 +404,7 @@ static int vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *m
> > > goto msg_err;
> > >
> > > while (mdev->id_table[i].device) {
> > > - supported_classes |= BIT(mdev->id_table[i].device);
> > > + supported_classes |= BIT_ULL(mdev->id_table[i].device);
> > > i++;
> > > }
> > >
>
> type of mdev->id_table[i].device is __u32 so in theory you're limited
> to device ID's up to 63.

A u32 can fit numbers up to 4 million? These .device numbers are
normally hardcoded defines listed in usr/include/linux/virtio_ids.h

But sometimes they're not like in vp_modern_probe() which does:

mdev->id.device = pci_dev->device - 0x1040;

I don't know if an assert is really worth it, considering how almost all
of them are hardcoded. Also if we do want an assert maybe there is a
better place to put it?

regards,
dan carpenter

2021-11-29 07:24:52

by Parav Pandit

[permalink] [raw]
Subject: RE: [PATCH] vdpa: Consider device id larger than 31



> From: Dan Carpenter <[email protected]>
> Sent: Monday, November 29, 2021 12:29 PM
>
> On Sun, Nov 28, 2021 at 09:14:35AM +0200, Eli Cohen wrote:
> > On Fri, Nov 26, 2021 at 10:48:12AM +0800, Jason Wang wrote:
> > > On Fri, Nov 26, 2021 at 2:09 AM Parav Pandit <[email protected]> wrote:
> > > >
> > > > virtio device id value can be more than 31. Hence, use BIT_ULL in
> > > > assignment.
> > > >
> > > > Fixes: 33b347503f01 ("vdpa: Define vdpa mgmt device, ops and a
> > > > netlink interface")
> > > > Reported-by: kernel test robot <[email protected]>
> > > > Reported-by: Dan Carpenter <[email protected]>
> > > > Signed-off-by: Parav Pandit <[email protected]>
> > >
> > > Acked-by: Jason Wang <[email protected]>
> > >
> > > > ---
> > > > drivers/vdpa/vdpa.c | 2 +-
> > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/vdpa/vdpa.c b/drivers/vdpa/vdpa.c index
> > > > 7332a74a4b00..e91c71aeeddf 100644
> > > > --- a/drivers/vdpa/vdpa.c
> > > > +++ b/drivers/vdpa/vdpa.c
> > > > @@ -404,7 +404,7 @@ static int vdpa_mgmtdev_fill(const struct
> vdpa_mgmt_dev *mdev, struct sk_buff *m
> > > > goto msg_err;
> > > >
> > > > while (mdev->id_table[i].device) {
> > > > - supported_classes |= BIT(mdev->id_table[i].device);
> > > > + supported_classes |=
> > > > + BIT_ULL(mdev->id_table[i].device);
> > > > i++;
> > > > }
> > > >
> >
> > type of mdev->id_table[i].device is __u32 so in theory you're limited
> > to device ID's up to 63.
>
> A u32 can fit numbers up to 4 million? These .device numbers are normally
> hardcoded defines listed in usr/include/linux/virtio_ids.h
>
> But sometimes they're not like in vp_modern_probe() which does:
>
> mdev->id.device = pci_dev->device - 0x1040;
>
> I don't know if an assert is really worth it, considering how almost all of them
> are hardcoded. Also if we do want an assert maybe there is a better place to
> put it?
I am changing above fix to report device id only upto 63.
Others higher values (which are not part of the current spec) will be ignored.
There is no need for assert for any undefined value anyway.