[dpdk-dev,v2] net/virtio-user: fix not working on 32-bit system
Checks
Commit Message
Hi,
The fix contained in this patch breaks under the following scenario:
1. A 64 bit host and virtual machine. Therefore all physical addresses are
64 bits.
2. A 32 bit user mode DPDK process running on a 64 bit virtual machine (64
bit kernel).
In this case, the physical address is 64bits but the virtual address of the
user process is 32 bits so uintptr_t is only 32 bits. As a result when:
(uintptr_t)(mb) + (vq)->offset)
is referenced, only 32 bits are copied into the descriptor but 64 bits are
required because in this scenario that is the size of a physical address.
So it seems like we need a way to determine the size of the physical
address and then VIRTIO_MBUF_ADDR should be written to copy that many bytes
into the uint64_t. Does anyone know how to determine the size of the
physical address?
Thanks,
Jim
Original Post:
virtio-user cannot work on 32-bit system as higher 32-bit of the
addr field (64-bit) in the desc is filled with non-zero value
which should not happen for a 32-bit system.
In case of virtio-user, we use buf_addr of mbuf to fill the
virtqueue desc addr. This is a regression bug. For 32-bit system,
the first 4 bytes of mbuf is buf_addr, with following 8 bytes for
buf_phyaddr. With below wrong definition, both buf_addr and lower
4 bytes buf_phyaddr are obtained to fill the virtqueue desc.
#define VIRTIO_MBUF_ADDR(mb, vq) \
(*(uint64_t *)((uintptr_t)(mb) + (vq)->offset))
Fixes: 25f80d108780 ("net/virtio: fix packet corruption")
Cc: stable at dpdk.org <http://dpdk.org/ml/listinfo/dev>
Signed-off-by: Jianfeng Tan <jianfeng.tan at intel.com
<http://dpdk.org/ml/listinfo/dev>>
---
drivers/net/virtio/virtqueue.h | 12 +++++++++---
1 file changed, 9 insertions(+), 3 deletions(-)
#endif
Comments
Hi Jim,
> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jim Murphy
> Sent: Wednesday, September 20, 2017 6:24 AM
> To: dev@dpdk.org
> Subject: [dpdk-dev] [PATCH v2] net/virtio-user: fix not working on 32-bit
> system
>
> Hi,
>
> The fix contained in this patch breaks under the following scenario:
>
> 1. A 64 bit host and virtual machine. Therefore all physical addresses are
> 64 bits.
Host, VM and application are 64bit, then I suppose there's no problem?
> 2. A 32 bit user mode DPDK process running on a 64 bit virtual machine (64
> bit kernel).
Ah, this is a case we fail to cover.
>
> In this case, the physical address is 64bits but the virtual address of the
> user process is 32 bits so uintptr_t is only 32 bits. As a result when:
>
> (uintptr_t)(mb) + (vq)->offset)
>
> is referenced, only 32 bits are copied into the descriptor but 64 bits are
> required because in this scenario that is the size of a physical address.
>
> So it seems like we need a way to determine the size of the physical
> address and then VIRTIO_MBUF_ADDR should be written to copy that many
> bytes
> into the uint64_t. Does anyone know how to determine the size of the
> physical address?
It's easy to find a way to decide the length (similar to _offset_), but copying variable length of bytes seems not an efficient way.
Thanks,
Jianfeng
On Tue, Sep 19, 2017 at 7:17 PM, Tan, Jianfeng <jianfeng.tan@intel.com>
wrote:
> Hi Jim,
>
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jim Murphy
> > Sent: Wednesday, September 20, 2017 6:24 AM
> > To: dev@dpdk.org
> > Subject: [dpdk-dev] [PATCH v2] net/virtio-user: fix not working on 32-bit
> > system
> >
> > Hi,
> >
> > The fix contained in this patch breaks under the following scenario:
> >
> > 1. A 64 bit host and virtual machine. Therefore all physical addresses
> are
> > 64 bits.
>
> Host, VM and application are 64bit, then I suppose there's no problem?
>
Right, no problem. I was just describing this as my setup.
>
> > 2. A 32 bit user mode DPDK process running on a 64 bit virtual machine
> (64
> > bit kernel).
>
> Ah, this is a case we fail to cover.
>
Yes.
> >
> > In this case, the physical address is 64bits but the virtual address of
> the
> > user process is 32 bits so uintptr_t is only 32 bits. As a result when:
> >
> > (uintptr_t)(mb) + (vq)->offset)
> >
> > is referenced, only 32 bits are copied into the descriptor but 64 bits
> are
> > required because in this scenario that is the size of a physical address.
> >
> > So it seems like we need a way to determine the size of the physical
> > address and then VIRTIO_MBUF_ADDR should be written to copy that many
> > bytes
> > into the uint64_t. Does anyone know how to determine the size of the
> > physical address?
>
> It's easy to find a way to decide the length (similar to _offset_), but
> copying variable length of bytes seems not an efficient way.
>
>
I agree regarding the efficiency issue. To get something at compile time
does RTE_MACHINE or something derived from it help us?
Thanks,
Jim
Thanks,
> Jianfeng
>
>
@@ -69,10 +69,16 @@ struct rte_mbuf;
#ifdef RTE_VIRTIO_USER
/**
- * Return the physical address (or virtual address in case of
- * virtio-user) of mbuf data buffer.
+ *
+ * Return the physical address of mbuf data buffer for virtio pci:
+ * on 32-bit system, offset equals 4, return the second four bytes of mbuf;
+ * on 64-bit system, offset equals 8, return the second eight bytes of mbuf.
+ * Return the virtual address of mbuf data buffer for virtio-user.
+ * on 32-bit system, offset equals 0, return the first four bytes of mbuf;
+ * on 64-bit system, offset equals 0, return the first eight bytes of mbuf;
*/
-#define VIRTIO_MBUF_ADDR(mb, vq) (*(uint64_t *)((uintptr_t)(mb) +
(vq)->offset))
+#define VIRTIO_MBUF_ADDR(mb, vq) \
+ ((uint64_t)(*(uintptr_t *)((uintptr_t)(mb) + (vq)->offset)))
#else
#define VIRTIO_MBUF_ADDR(mb, vq) ((mb)->buf_physaddr)