On Wed, Dec 02, 2015 at 03:53:01PM +0200, Victor Kaplansky wrote:
> On Wed, Dec 02, 2015 at 11:43:11AM +0800, Yuanhan Liu wrote:
> > Introduce vhost_log_write() helper function to log the dirty pages we
> > touched. Page size is harded code to 4096 (VHOST_LOG_PAGE), and each
> > log is presented by 1 bit.
> >
> > Therefore, vhost_log_write() simply finds the right bit for related
> > page we are gonna change, and set it to 1. dev->log_base denotes the
> > start of the dirty page bitmap.
> >
> > The page address is biased by log_guest_addr, which is derived from
> > SET_VRING_ADDR request as part of the vring related addresses.
> >
> > Signed-off-by: Yuanhan Liu <yuanhan.liu at linux.intel.com>
> > ---
> > lib/librte_vhost/rte_virtio_net.h | 34 ++++++++++++++++++++++++++++++++++
> > lib/librte_vhost/virtio-net.c | 4 ++++
> > 2 files changed, 38 insertions(+)
> >
> > diff --git a/lib/librte_vhost/rte_virtio_net.h
> > b/lib/librte_vhost/rte_virtio_net.h
> > index 416dac2..191c1be 100644
> > --- a/lib/librte_vhost/rte_virtio_net.h
> > +++ b/lib/librte_vhost/rte_virtio_net.h
> > @@ -40,6 +40,7 @@
> > */
> >
> > #include <stdint.h>
> > +#include <linux/vhost.h>
> > #include <linux/virtio_ring.h>
> > #include <linux/virtio_net.h>
> > #include <sys/eventfd.h>
> > @@ -59,6 +60,8 @@ struct rte_mbuf;
> > /* Backend value set by guest. */
> > #define VIRTIO_DEV_STOPPED -1
> >
> > +#define VHOST_LOG_PAGE 4096
> > +
> >
> > /* Enum for virtqueue management. */
> > enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
> > @@ -82,6 +85,7 @@ struct vhost_virtqueue {
> > struct vring_desc *desc; /**< Virtqueue
> > descriptor ring. */
> > struct vring_avail *avail; /**< Virtqueue
> > available ring. */
> > struct vring_used *used; /**< Virtqueue used
> > ring. */
> > + uint64_t log_guest_addr; /**< Physical address
> > of used ring, for logging */
> > uint32_t size; /**< Size of descriptor
> > ring. */
> > uint32_t backend; /**< Backend value to
> > determine if device should started/stopped. */
> > uint16_t vhost_hlen; /**< Vhost header
> > length (varies depending on RX merge buffers. */
> > @@ -203,6 +207,36 @@ gpa_to_vva(struct virtio_net *dev, uint64_t guest_pa)
> > return vhost_va;
> > }
> >
> > +static inline void __attribute__((always_inline))
> > +vhost_log_page(uint8_t *log_base, uint64_t page)
> > +{
> > + /* TODO: to make it atomic? */
> > + log_base[page / 8] |= 1 << (page % 8);
>
> I think the atomic OR operation is necessary only if there can be
> more than one vhost-user back-end updating the guest's memory
> simultaneously. However probably it is pretty safe to perform
> regular OR operation, since rings are not shared between
> back-end. What about buffers pointed by descriptors? To be on
> the safe side, I would use a GCC built-in function
> __sync_fetch_and_or().
The build has to be passed not only for gcc, but for icc and clang as
well.
>
> > +}
> > +
> > +static inline void __attribute__((always_inline))
> > +vhost_log_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
> > + uint64_t offset, uint64_t len)
> > +{
> > + uint64_t addr = vq->log_guest_addr;
> > + uint64_t page;
> > +
> > + if (unlikely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
> > + !dev->log_base || !len))
> > + return;
>
> Isn't "likely" more appropriate in above, since the whole
> expression is expected to be true most of the time?
Sorry, it's a typo, and thanks for the catching.
--yliu
