Re: [Xen-devel] [PATCH v1 03/13] xen/pvcalls: implement socket command and handle events

[Stefano Stabellini]

On Mon, 24 Jul 2017, Juergen Gross wrote:
> On 22/07/17 02:11, Stefano Stabellini wrote:
> > Send a PVCALLS_SOCKET command to the backend, use the masked
> > req_prod_pvt as req_id. This way, req_id is guaranteed to be between 0
> > and PVCALLS_NR_REQ_PER_RING. We already have a slot in the rsp array
> > ready for the response, and there cannot be two outstanding responses
> > with the same req_id.
> > 
> > Wait for the response by waiting on the inflight_req waitqueue and
> > check for the req_id field in rsp[req_id]. Use atomic accesses to
> > read the field. Once a response is received, clear the corresponding rsp
> > slot by setting req_id to PVCALLS_INVALID_ID. Note that
> > PVCALLS_INVALID_ID is invalid only from the frontend point of view. It
> > is not part of the PVCalls protocol.
> > 
> > pvcalls_front_event_handler is in charge of copying responses from the
> > ring to the appropriate rsp slot. It is done by copying the body of the
> > response first, then by copying req_id atomically. After the copies,
> > wake up anybody waiting on waitqueue.
> > 
> > pvcallss_lock protects accesses to the ring.
> > 
> > Signed-off-by: Stefano Stabellini 
> > CC: boris.ostrov...@oracle.com
> > CC: jgr...@suse.com
> > ---
> >  drivers/xen/pvcalls-front.c | 91 
> > +
> >  drivers/xen/pvcalls-front.h |  8 
> >  2 files changed, 99 insertions(+)
> >  create mode 100644 drivers/xen/pvcalls-front.h
> > 
> > diff --git a/drivers/xen/pvcalls-front.c b/drivers/xen/pvcalls-front.c
> > index fb08ebf..7933c73 100644
> > --- a/drivers/xen/pvcalls-front.c
> > +++ b/drivers/xen/pvcalls-front.c
> 
> Shouldn't you include pvcalls-front.h?

Yes


> > @@ -40,9 +40,100 @@ struct pvcalls_bedata {
> >  
> >  static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
> >  {
> > +   struct xenbus_device *dev = dev_id;
> > +   struct pvcalls_bedata *bedata;
> > +   struct xen_pvcalls_response *rsp;
> > +   uint8_t *src, *dst;
> > +   int req_id = 0, more = 0;
> > +
> > +   if (dev == NULL)
> > +   return IRQ_HANDLED;
> > +
> > +   bedata = dev_get_drvdata(>dev);
> > +   if (bedata == NULL)
> > +   return IRQ_HANDLED;
> > +
> > +again:
> > +   while (RING_HAS_UNCONSUMED_RESPONSES(>ring)) {
> > +   rsp = RING_GET_RESPONSE(>ring, bedata->ring.rsp_cons);
> > +
> > +   req_id = rsp->req_id;
> > +   src = (uint8_t *)>rsp[req_id];
> > +   src += sizeof(rsp->req_id);
> > +   dst = (uint8_t *)rsp;
> > +   dst += sizeof(rsp->req_id);
> > +   memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
> > +   /*
> > +* First copy the rest of the data, then req_id. It is
> > +* paired with the barrier when accessing bedata->rsp.
> > +*/
> > +   smp_wmb();
> > +   WRITE_ONCE(bedata->rsp[req_id].req_id, rsp->req_id);
> > +
> > +   bedata->ring.rsp_cons++;
> > +   wake_up(>inflight_req);
> > +   }
> > +
> > +   RING_FINAL_CHECK_FOR_RESPONSES(>ring, more);
> > +   if (more)
> > +   goto again;
> 
> Wouldn't it make more sense to use wake_up() just once if there is any
> response pending and do the consuming loop outside the irq handler?

You are definitely right: it's far better to call wake_up() just once
after the consuming loop if there is any response pending. I'll do that.

However, I am not sure there is much to gain in moving the consuming
loop out of the irq handler: it's pretty short and doesn't call any long
running or sleeping functions.



 
> > return IRQ_HANDLED;
> >  }
> >  
> > +int pvcalls_front_socket(struct socket *sock)
> > +{
> > +   struct pvcalls_bedata *bedata;
> > +   struct xen_pvcalls_request *req;
> > +   int notify, req_id, ret;
> > +
> > +   if (!pvcalls_front_dev)
> > +   return -EACCES;
> > +   /*
> > +* PVCalls only supports domain AF_INET,
> > +* type SOCK_STREAM and protocol 0 sockets for now.
> > +*
> > +* Check socket type here, AF_INET and protocol checks are done
> > +* by the caller.
> > +*/
> > +   if (sock->type != SOCK_STREAM)
> > +   return -ENOTSUPP;
> > +
> > +   bedata = dev_get_drvdata(_front_dev->dev);
> > +
> > +   spin_lock(>pvcallss_lock);
> > +   req_id = bedata->ring.req_prod_pvt & (RING_SIZE(>ring) - 1);
> > +   BUG_ON(req_id >= PVCALLS_NR_REQ_PER_RING);
> > +   if (RING_FULL(>ring) ||
> > +   READ_ONCE(bedata->rsp[req_id].req_id) != PVCALLS_INVALID_ID) {
> > +   spin_unlock(>pvcallss_lock);
> > +   return -EAGAIN;
> > +   }
> > +   req = RING_GET_REQUEST(>ring, req_id);
> > +   req->req_id = req_id;
> > +   req->cmd = PVCALLS_SOCKET;
> > +   req->u.socket.id = (uint64_t) sock;
> > +   req->u.socket.domain = AF_INET;
> > +   req->u.socket.type = SOCK_STREAM;
> > +   req->u.socket.protocol = 0;
> > +
> > +   bedata->ring.req_prod_pvt++;
> > +   RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(>ring, notify);
> > +   

Re: [Xen-devel] [PATCH v1 03/13] xen/pvcalls: implement socket command and handle events

[Juergen Gross]

On 22/07/17 02:11, Stefano Stabellini wrote:
> Send a PVCALLS_SOCKET command to the backend, use the masked
> req_prod_pvt as req_id. This way, req_id is guaranteed to be between 0
> and PVCALLS_NR_REQ_PER_RING. We already have a slot in the rsp array
> ready for the response, and there cannot be two outstanding responses
> with the same req_id.
> 
> Wait for the response by waiting on the inflight_req waitqueue and
> check for the req_id field in rsp[req_id]. Use atomic accesses to
> read the field. Once a response is received, clear the corresponding rsp
> slot by setting req_id to PVCALLS_INVALID_ID. Note that
> PVCALLS_INVALID_ID is invalid only from the frontend point of view. It
> is not part of the PVCalls protocol.
> 
> pvcalls_front_event_handler is in charge of copying responses from the
> ring to the appropriate rsp slot. It is done by copying the body of the
> response first, then by copying req_id atomically. After the copies,
> wake up anybody waiting on waitqueue.
> 
> pvcallss_lock protects accesses to the ring.
> 
> Signed-off-by: Stefano Stabellini 
> CC: boris.ostrov...@oracle.com
> CC: jgr...@suse.com
> ---
>  drivers/xen/pvcalls-front.c | 91 
> +
>  drivers/xen/pvcalls-front.h |  8 
>  2 files changed, 99 insertions(+)
>  create mode 100644 drivers/xen/pvcalls-front.h
> 
> diff --git a/drivers/xen/pvcalls-front.c b/drivers/xen/pvcalls-front.c
> index fb08ebf..7933c73 100644
> --- a/drivers/xen/pvcalls-front.c
> +++ b/drivers/xen/pvcalls-front.c

Shouldn't you include pvcalls-front.h?

> @@ -40,9 +40,100 @@ struct pvcalls_bedata {
>  
>  static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
>  {
> + struct xenbus_device *dev = dev_id;
> + struct pvcalls_bedata *bedata;
> + struct xen_pvcalls_response *rsp;
> + uint8_t *src, *dst;
> + int req_id = 0, more = 0;
> +
> + if (dev == NULL)
> + return IRQ_HANDLED;
> +
> + bedata = dev_get_drvdata(>dev);
> + if (bedata == NULL)
> + return IRQ_HANDLED;
> +
> +again:
> + while (RING_HAS_UNCONSUMED_RESPONSES(>ring)) {
> + rsp = RING_GET_RESPONSE(>ring, bedata->ring.rsp_cons);
> +
> + req_id = rsp->req_id;
> + src = (uint8_t *)>rsp[req_id];
> + src += sizeof(rsp->req_id);
> + dst = (uint8_t *)rsp;
> + dst += sizeof(rsp->req_id);
> + memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
> + /*
> +  * First copy the rest of the data, then req_id. It is
> +  * paired with the barrier when accessing bedata->rsp.
> +  */
> + smp_wmb();
> + WRITE_ONCE(bedata->rsp[req_id].req_id, rsp->req_id);
> +
> + bedata->ring.rsp_cons++;
> + wake_up(>inflight_req);
> + }
> +
> + RING_FINAL_CHECK_FOR_RESPONSES(>ring, more);
> + if (more)
> + goto again;

Wouldn't it make more sense to use wake_up() just once if there is any
response pending and do the consuming loop outside the irq handler?


Juergen

>   return IRQ_HANDLED;
>  }
>  
> +int pvcalls_front_socket(struct socket *sock)
> +{
> + struct pvcalls_bedata *bedata;
> + struct xen_pvcalls_request *req;
> + int notify, req_id, ret;
> +
> + if (!pvcalls_front_dev)
> + return -EACCES;
> + /*
> +  * PVCalls only supports domain AF_INET,
> +  * type SOCK_STREAM and protocol 0 sockets for now.
> +  *
> +  * Check socket type here, AF_INET and protocol checks are done
> +  * by the caller.
> +  */
> + if (sock->type != SOCK_STREAM)
> + return -ENOTSUPP;
> +
> + bedata = dev_get_drvdata(_front_dev->dev);
> +
> + spin_lock(>pvcallss_lock);
> + req_id = bedata->ring.req_prod_pvt & (RING_SIZE(>ring) - 1);
> + BUG_ON(req_id >= PVCALLS_NR_REQ_PER_RING);
> + if (RING_FULL(>ring) ||
> + READ_ONCE(bedata->rsp[req_id].req_id) != PVCALLS_INVALID_ID) {
> + spin_unlock(>pvcallss_lock);
> + return -EAGAIN;
> + }
> + req = RING_GET_REQUEST(>ring, req_id);
> + req->req_id = req_id;
> + req->cmd = PVCALLS_SOCKET;
> + req->u.socket.id = (uint64_t) sock;
> + req->u.socket.domain = AF_INET;
> + req->u.socket.type = SOCK_STREAM;
> + req->u.socket.protocol = 0;
> +
> + bedata->ring.req_prod_pvt++;
> + RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(>ring, notify);
> + spin_unlock(>pvcallss_lock);
> + if (notify)
> + notify_remote_via_irq(bedata->irq);
> +
> + if (wait_event_interruptible(bedata->inflight_req,
> + READ_ONCE(bedata->rsp[req_id].req_id) == req_id) != 0)
> + return -EINTR;
> +
> + ret = bedata->rsp[req_id].ret;
> + /* read ret, then set this rsp slot to be reused */
> + smp_mb();
> + WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
> +
> + return ret;
> +}
> 

[Xen-devel] [PATCH v1 03/13] xen/pvcalls: implement socket command and handle events

[Stefano Stabellini]

Send a PVCALLS_SOCKET command to the backend, use the masked
req_prod_pvt as req_id. This way, req_id is guaranteed to be between 0
and PVCALLS_NR_REQ_PER_RING. We already have a slot in the rsp array
ready for the response, and there cannot be two outstanding responses
with the same req_id.

Wait for the response by waiting on the inflight_req waitqueue and
check for the req_id field in rsp[req_id]. Use atomic accesses to
read the field. Once a response is received, clear the corresponding rsp
slot by setting req_id to PVCALLS_INVALID_ID. Note that
PVCALLS_INVALID_ID is invalid only from the frontend point of view. It
is not part of the PVCalls protocol.

pvcalls_front_event_handler is in charge of copying responses from the
ring to the appropriate rsp slot. It is done by copying the body of the
response first, then by copying req_id atomically. After the copies,
wake up anybody waiting on waitqueue.

pvcallss_lock protects accesses to the ring.

Signed-off-by: Stefano Stabellini 
CC: boris.ostrov...@oracle.com
CC: jgr...@suse.com
---
 drivers/xen/pvcalls-front.c | 91 +
 drivers/xen/pvcalls-front.h |  8 
 2 files changed, 99 insertions(+)
 create mode 100644 drivers/xen/pvcalls-front.h

diff --git a/drivers/xen/pvcalls-front.c b/drivers/xen/pvcalls-front.c
index fb08ebf..7933c73 100644
--- a/drivers/xen/pvcalls-front.c
+++ b/drivers/xen/pvcalls-front.c
@@ -40,9 +40,100 @@ struct pvcalls_bedata {
 
 static irqreturn_t pvcalls_front_event_handler(int irq, void *dev_id)
 {
+   struct xenbus_device *dev = dev_id;
+   struct pvcalls_bedata *bedata;
+   struct xen_pvcalls_response *rsp;
+   uint8_t *src, *dst;
+   int req_id = 0, more = 0;
+
+   if (dev == NULL)
+   return IRQ_HANDLED;
+
+   bedata = dev_get_drvdata(>dev);
+   if (bedata == NULL)
+   return IRQ_HANDLED;
+
+again:
+   while (RING_HAS_UNCONSUMED_RESPONSES(>ring)) {
+   rsp = RING_GET_RESPONSE(>ring, bedata->ring.rsp_cons);
+
+   req_id = rsp->req_id;
+   src = (uint8_t *)>rsp[req_id];
+   src += sizeof(rsp->req_id);
+   dst = (uint8_t *)rsp;
+   dst += sizeof(rsp->req_id);
+   memcpy(dst, src, sizeof(*rsp) - sizeof(rsp->req_id));
+   /*
+* First copy the rest of the data, then req_id. It is
+* paired with the barrier when accessing bedata->rsp.
+*/
+   smp_wmb();
+   WRITE_ONCE(bedata->rsp[req_id].req_id, rsp->req_id);
+
+   bedata->ring.rsp_cons++;
+   wake_up(>inflight_req);
+   }
+
+   RING_FINAL_CHECK_FOR_RESPONSES(>ring, more);
+   if (more)
+   goto again;
return IRQ_HANDLED;
 }
 
+int pvcalls_front_socket(struct socket *sock)
+{
+   struct pvcalls_bedata *bedata;
+   struct xen_pvcalls_request *req;
+   int notify, req_id, ret;
+
+   if (!pvcalls_front_dev)
+   return -EACCES;
+   /*
+* PVCalls only supports domain AF_INET,
+* type SOCK_STREAM and protocol 0 sockets for now.
+*
+* Check socket type here, AF_INET and protocol checks are done
+* by the caller.
+*/
+   if (sock->type != SOCK_STREAM)
+   return -ENOTSUPP;
+
+   bedata = dev_get_drvdata(_front_dev->dev);
+
+   spin_lock(>pvcallss_lock);
+   req_id = bedata->ring.req_prod_pvt & (RING_SIZE(>ring) - 1);
+   BUG_ON(req_id >= PVCALLS_NR_REQ_PER_RING);
+   if (RING_FULL(>ring) ||
+   READ_ONCE(bedata->rsp[req_id].req_id) != PVCALLS_INVALID_ID) {
+   spin_unlock(>pvcallss_lock);
+   return -EAGAIN;
+   }
+   req = RING_GET_REQUEST(>ring, req_id);
+   req->req_id = req_id;
+   req->cmd = PVCALLS_SOCKET;
+   req->u.socket.id = (uint64_t) sock;
+   req->u.socket.domain = AF_INET;
+   req->u.socket.type = SOCK_STREAM;
+   req->u.socket.protocol = 0;
+
+   bedata->ring.req_prod_pvt++;
+   RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(>ring, notify);
+   spin_unlock(>pvcallss_lock);
+   if (notify)
+   notify_remote_via_irq(bedata->irq);
+
+   if (wait_event_interruptible(bedata->inflight_req,
+   READ_ONCE(bedata->rsp[req_id].req_id) == req_id) != 0)
+   return -EINTR;
+
+   ret = bedata->rsp[req_id].ret;
+   /* read ret, then set this rsp slot to be reused */
+   smp_mb();
+   WRITE_ONCE(bedata->rsp[req_id].req_id, PVCALLS_INVALID_ID);
+
+   return ret;
+}
+
 static const struct xenbus_device_id pvcalls_front_ids[] = {
{ "pvcalls" },
{ "" }
diff --git a/drivers/xen/pvcalls-front.h b/drivers/xen/pvcalls-front.h
new file mode 100644
index 000..b7dabed
--- /dev/null
+++ b/drivers/xen/pvcalls-front.h
@@ -0,0 +1,8 @@
+#ifndef __PVCALLS_FRONT_H__
+#define __PVCALLS_FRONT_H__
+
+#include 
+