diff mbox series

rpmsg: virtio: set dst address on first message received

Message ID 20220315153856.3117676-1-arnaud.pouliquen@foss.st.com (mailing list archive)
State Superseded
Headers show
Series rpmsg: virtio: set dst address on first message received | expand

Commit Message

Arnaud POULIQUEN March 15, 2022, 3:38 p.m. UTC
when a rpmsg channel has been locally created with a destination address
set to RPMSG_ADDR_ANY, a name service announcement message is sent to
the remote side. Then the destination address is never updated, making it
impossible to send messages to the remote.

An example of kernel trace observed:
rpmsg_tty virtio0.rpmsg-tty.29.-1: invalid addr (src 0x1d, dst 0xffffffff)

Implement same strategy than the open-amp library:
On the reception of the first message, if the destination address is
RPMSG_ADDR_ANY, then set it to address of the remote endpoint that
send the message.

Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>

---
Remark:
An alternative (or a complement?) could be to add a NS bind/unbind in
the NS announcement channel (in rpmsg_ns.c).
This would allow the local and/or the remote processor to inform the
remote side the the service announced in bound.
---
 drivers/rpmsg/virtio_rpmsg_bus.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

Comments

kernel test robot March 15, 2022, 10:47 p.m. UTC | #1
Hi Arnaud,

I love your patch! Perhaps something to improve:

[auto build test WARNING on remoteproc/rpmsg-next]
[also build test WARNING on v5.17-rc8 next-20220315]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Arnaud-Pouliquen/rpmsg-virtio-set-dst-address-on-first-message-received/20220315-234049
base:   git://git.kernel.org/pub/scm/linux/kernel/git/remoteproc/linux.git rpmsg-next
config: s390-randconfig-s031-20220313 (https://download.01.org/0day-ci/archive/20220316/202203160614.sjUMuSy8-lkp@intel.com/config)
compiler: s390-linux-gcc (GCC) 11.2.0
reproduce:
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # apt-get install sparse
        # sparse version: v0.6.4-dirty
        # https://github.com/0day-ci/linux/commit/ef182a34773917f6bf876b37485031962393a1cd
        git remote add linux-review https://github.com/0day-ci/linux
        git fetch --no-tags linux-review Arnaud-Pouliquen/rpmsg-virtio-set-dst-address-on-first-message-received/20220315-234049
        git checkout ef182a34773917f6bf876b37485031962393a1cd
        # save the config file to linux build tree
        mkdir build_dir
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=gcc-11.2.0 make.cross C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' O=build_dir ARCH=s390 SHELL=/bin/bash

If you fix the issue, kindly add following tag as appropriate
Reported-by: kernel test robot <lkp@intel.com>


sparse warnings: (new ones prefixed by >>)
>> drivers/rpmsg/virtio_rpmsg_bus.c:756:36: sparse: sparse: incorrect type in assignment (different base types) @@     expected unsigned int [usertype] dst @@     got restricted __rpmsg32 [usertype] src @@
   drivers/rpmsg/virtio_rpmsg_bus.c:756:36: sparse:     expected unsigned int [usertype] dst
   drivers/rpmsg/virtio_rpmsg_bus.c:756:36: sparse:     got restricted __rpmsg32 [usertype] src

vim +756 drivers/rpmsg/virtio_rpmsg_bus.c

   727	
   728		/*
   729		 * We currently use fixed-sized buffers, so trivially sanitize
   730		 * the reported payload length.
   731		 */
   732		if (len > vrp->buf_size ||
   733		    msg_len > (len - sizeof(struct rpmsg_hdr))) {
   734			dev_warn(dev, "inbound msg too big: (%d, %d)\n", len, msg_len);
   735			return -EINVAL;
   736		}
   737	
   738		/* use the dst addr to fetch the callback of the appropriate user */
   739		mutex_lock(&vrp->endpoints_lock);
   740	
   741		ept = idr_find(&vrp->endpoints, __rpmsg32_to_cpu(little_endian, msg->dst));
   742	
   743		/* let's make sure no one deallocates ept while we use it */
   744		if (ept)
   745			kref_get(&ept->refcount);
   746	
   747		mutex_unlock(&vrp->endpoints_lock);
   748	
   749		if (ept) {
   750			rpdev = ept->rpdev;
   751			if (rpdev->ept == ept && rpdev->dst == RPMSG_ADDR_ANY) {
   752				/*
   753				 * First message received from the remote side on the default endpoint,
   754				 * update channel destination address.
   755				 */
 > 756				rpdev->dst = msg->src;
   757			}
   758	
   759			/* make sure ept->cb doesn't go away while we use it */
   760			mutex_lock(&ept->cb_lock);
   761	
   762			if (ept->cb)
   763				ept->cb(ept->rpdev, msg->data, msg_len, ept->priv,
   764					__rpmsg32_to_cpu(little_endian, msg->src));
   765	
   766			mutex_unlock(&ept->cb_lock);
   767	
   768			/* farewell, ept, we don't need you anymore */
   769			kref_put(&ept->refcount, __ept_release);
   770		} else
   771			dev_warn_ratelimited(dev, "msg received with no recipient\n");
   772	
   773		/* publish the real size of the buffer */
   774		rpmsg_sg_init(&sg, msg, vrp->buf_size);
   775	
   776		/* add the buffer back to the remote processor's virtqueue */
   777		err = virtqueue_add_inbuf(vrp->rvq, &sg, 1, msg, GFP_KERNEL);
   778		if (err < 0) {
   779			dev_err(dev, "failed to add a virtqueue buffer: %d\n", err);
   780			return err;
   781		}
   782	
   783		return 0;
   784	}
   785	

---
0-DAY CI Kernel Test Service
https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org
Mathieu Poirier May 6, 2022, 5:12 p.m. UTC | #2
On Tue, Mar 15, 2022 at 04:38:56PM +0100, Arnaud Pouliquen wrote:
> when a rpmsg channel has been locally created with a destination address

s/when/Wen

Also, please be more specific about the "locally created" part, i.e
rpmsg_ctrldev_ioctl() -> rpmsg_create_channel().  Otherwise it is really hard to
understand the context of this change.

> set to RPMSG_ADDR_ANY, a name service announcement message is sent to
> the remote side. Then the destination address is never updated, making it
> impossible to send messages to the remote.
> 
> An example of kernel trace observed:
> rpmsg_tty virtio0.rpmsg-tty.29.-1: invalid addr (src 0x1d, dst 0xffffffff)
> 
> Implement same strategy than the open-amp library:
> On the reception of the first message, if the destination address is
> RPMSG_ADDR_ANY, then set it to address of the remote endpoint that
> send the message.
>

I would have expected a "Fixes:" tag.

> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
> 
> ---
> Remark:
> An alternative (or a complement?) could be to add a NS bind/unbind in
> the NS announcement channel (in rpmsg_ns.c).
> This would allow the local and/or the remote processor to inform the
> remote side the the service announced in bound.
> ---
>  drivers/rpmsg/virtio_rpmsg_bus.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
> index 3ede25b1f2e4..99d2119cc164 100644
> --- a/drivers/rpmsg/virtio_rpmsg_bus.c
> +++ b/drivers/rpmsg/virtio_rpmsg_bus.c
> @@ -708,6 +708,7 @@ static ssize_t virtio_rpmsg_get_mtu(struct rpmsg_endpoint *ept)
>  static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
>  			     struct rpmsg_hdr *msg, unsigned int len)
>  {
> +	struct rpmsg_device *rpdev;
>  	struct rpmsg_endpoint *ept;
>  	struct scatterlist sg;
>  	bool little_endian = virtio_is_little_endian(vrp->vdev);
> @@ -746,6 +747,15 @@ static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
>  	mutex_unlock(&vrp->endpoints_lock);
>  
>  	if (ept) {
> +		rpdev = ept->rpdev;
> +		if (rpdev->ept == ept && rpdev->dst == RPMSG_ADDR_ANY) {

Please add a comment to explain the first part of the if() clause.  It took me
quite some time to understand. 

> +			/*
> +			 * First message received from the remote side on the default endpoint,
> +			 * update channel destination address.
> +			 */
> +			rpdev->dst = msg->src;

This triggers a bot warning and should be addressed.  If it can't be addressed add
a comment that clearly explains why so that we don't end up receiving patches
for it every 4 weeks.

Thanks,
Mathieu

> +		}
> +
>  		/* make sure ept->cb doesn't go away while we use it */
>  		mutex_lock(&ept->cb_lock);
>  
> -- 
> 2.25.1
>
Arnaud POULIQUEN May 9, 2022, 9:18 a.m. UTC | #3
Hello Mathieu,

On 5/6/22 19:12, Mathieu Poirier wrote:
> On Tue, Mar 15, 2022 at 04:38:56PM +0100, Arnaud Pouliquen wrote:
>> when a rpmsg channel has been locally created with a destination address
> 
> s/when/Wen
> 
> Also, please be more specific about the "locally created" part, i.e
> rpmsg_ctrldev_ioctl() -> rpmsg_create_channel().  Otherwise it is really hard to
> understand the context of this change.
> 
>> set to RPMSG_ADDR_ANY, a name service announcement message is sent to
>> the remote side. Then the destination address is never updated, making it
>> impossible to send messages to the remote.
>>
>> An example of kernel trace observed:
>> rpmsg_tty virtio0.rpmsg-tty.29.-1: invalid addr (src 0x1d, dst 0xffffffff)
>>
>> Implement same strategy than the open-amp library:
>> On the reception of the first message, if the destination address is
>> RPMSG_ADDR_ANY, then set it to address of the remote endpoint that
>> send the message.
>>
> 
> I would have expected a "Fixes:" tag.

Difficult to give a reference. For me the issue exists since the creation
of the rpmsg virtio bus. A driver can create a channel that generates a NS
announcement leading to this issue.
The issue as been highlighted by the creation of the  RPMSG_CREATE_DEV_IOCTL
control.

> 
>> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
>>
>> ---
>> Remark:
>> An alternative (or a complement?) could be to add a NS bind/unbind in
>> the NS announcement channel (in rpmsg_ns.c).
>> This would allow the local and/or the remote processor to inform the
>> remote side the the service announced in bound.
>> ---
>>  drivers/rpmsg/virtio_rpmsg_bus.c | 10 ++++++++++
>>  1 file changed, 10 insertions(+)
>>
>> diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
>> index 3ede25b1f2e4..99d2119cc164 100644
>> --- a/drivers/rpmsg/virtio_rpmsg_bus.c
>> +++ b/drivers/rpmsg/virtio_rpmsg_bus.c
>> @@ -708,6 +708,7 @@ static ssize_t virtio_rpmsg_get_mtu(struct rpmsg_endpoint *ept)
>>  static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
>>  			     struct rpmsg_hdr *msg, unsigned int len)
>>  {
>> +	struct rpmsg_device *rpdev;
>>  	struct rpmsg_endpoint *ept;
>>  	struct scatterlist sg;
>>  	bool little_endian = virtio_is_little_endian(vrp->vdev);
>> @@ -746,6 +747,15 @@ static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
>>  	mutex_unlock(&vrp->endpoints_lock);
>>  
>>  	if (ept) {
>> +		rpdev = ept->rpdev;
>> +		if (rpdev->ept == ept && rpdev->dst == RPMSG_ADDR_ANY) {
> 
> Please add a comment to explain the first part of the if() clause.  It took me
> quite some time to understand. 
> 
>> +			/*
>> +			 * First message received from the remote side on the default endpoint,
>> +			 * update channel destination address.
>> +			 */
>> +			rpdev->dst = msg->src;
> 
> This triggers a bot warning and should be addressed.  If it can't be addressed add
> a comment that clearly explains why so that we don't end up receiving patches
> for it every 4 weeks.

Oops, I missed it, thanks for pointing it out.


Concerning the patch itself as discussed in RP open-amp meeting. I wonder if this issue
could be addressed by the flow control[1][2][3], or if needed in any case.

I propose to send a V2 when ready to propose in parallel the flow control.
So both can be addressed at same time to have a global picture of the way to address the
use case..

Thanks,
Arnaud

[1] POC Linux code:
https://github.com/arnopo/linux/commits/signalling

[2] openamp library associated code:
https://github.com/arnopo/open-amp/commits/flow_ctrl

[3] overview presentation
https://drive.google.com/file/d/1CLU3ybI3oSBGvor18AQ-HOzOJ2nOppEb/view


> 
> Thanks,
> Mathieu
> 
>> +		}
>> +
>>  		/* make sure ept->cb doesn't go away while we use it */
>>  		mutex_lock(&ept->cb_lock);
>>  
>> -- 
>> 2.25.1
>>
Mathieu Poirier May 10, 2022, 2:45 p.m. UTC | #4
On Mon, 9 May 2022 at 03:18, Arnaud POULIQUEN
<arnaud.pouliquen@foss.st.com> wrote:
>
> Hello Mathieu,
>
> On 5/6/22 19:12, Mathieu Poirier wrote:
> > On Tue, Mar 15, 2022 at 04:38:56PM +0100, Arnaud Pouliquen wrote:
> >> when a rpmsg channel has been locally created with a destination address
> >
> > s/when/Wen
> >
> > Also, please be more specific about the "locally created" part, i.e
> > rpmsg_ctrldev_ioctl() -> rpmsg_create_channel().  Otherwise it is really hard to
> > understand the context of this change.
> >
> >> set to RPMSG_ADDR_ANY, a name service announcement message is sent to
> >> the remote side. Then the destination address is never updated, making it
> >> impossible to send messages to the remote.
> >>
> >> An example of kernel trace observed:
> >> rpmsg_tty virtio0.rpmsg-tty.29.-1: invalid addr (src 0x1d, dst 0xffffffff)
> >>
> >> Implement same strategy than the open-amp library:
> >> On the reception of the first message, if the destination address is
> >> RPMSG_ADDR_ANY, then set it to address of the remote endpoint that
> >> send the message.
> >>
> >
> > I would have expected a "Fixes:" tag.
>
> Difficult to give a reference. For me the issue exists since the creation
> of the rpmsg virtio bus. A driver can create a channel that generates a NS
> announcement leading to this issue.
> The issue as been highlighted by the creation of the  RPMSG_CREATE_DEV_IOCTL
> control.
>

Ok

> >
> >> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
> >>
> >> ---
> >> Remark:
> >> An alternative (or a complement?) could be to add a NS bind/unbind in
> >> the NS announcement channel (in rpmsg_ns.c).
> >> This would allow the local and/or the remote processor to inform the
> >> remote side the the service announced in bound.
> >> ---
> >>  drivers/rpmsg/virtio_rpmsg_bus.c | 10 ++++++++++
> >>  1 file changed, 10 insertions(+)
> >>
> >> diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
> >> index 3ede25b1f2e4..99d2119cc164 100644
> >> --- a/drivers/rpmsg/virtio_rpmsg_bus.c
> >> +++ b/drivers/rpmsg/virtio_rpmsg_bus.c
> >> @@ -708,6 +708,7 @@ static ssize_t virtio_rpmsg_get_mtu(struct rpmsg_endpoint *ept)
> >>  static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
> >>                           struct rpmsg_hdr *msg, unsigned int len)
> >>  {
> >> +    struct rpmsg_device *rpdev;
> >>      struct rpmsg_endpoint *ept;
> >>      struct scatterlist sg;
> >>      bool little_endian = virtio_is_little_endian(vrp->vdev);
> >> @@ -746,6 +747,15 @@ static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
> >>      mutex_unlock(&vrp->endpoints_lock);
> >>
> >>      if (ept) {
> >> +            rpdev = ept->rpdev;
> >> +            if (rpdev->ept == ept && rpdev->dst == RPMSG_ADDR_ANY) {
> >
> > Please add a comment to explain the first part of the if() clause.  It took me
> > quite some time to understand.
> >
> >> +                    /*
> >> +                     * First message received from the remote side on the default endpoint,
> >> +                     * update channel destination address.
> >> +                     */
> >> +                    rpdev->dst = msg->src;
> >
> > This triggers a bot warning and should be addressed.  If it can't be addressed add
> > a comment that clearly explains why so that we don't end up receiving patches
> > for it every 4 weeks.
>
> Oops, I missed it, thanks for pointing it out.
>
>
> Concerning the patch itself as discussed in RP open-amp meeting. I wonder if this issue
> could be addressed by the flow control[1][2][3], or if needed in any case.
>
> I propose to send a V2 when ready to propose in parallel the flow control.
> So both can be addressed at same time to have a global picture of the way to address the
> use case..
>

It will be one or the other, we can't have two ways to do things.  If
the flow control patchset addresses this problem on top of adding more
functionality then we should go with that one.

> Thanks,
> Arnaud
>
> [1] POC Linux code:
> https://github.com/arnopo/linux/commits/signalling
>
> [2] openamp library associated code:
> https://github.com/arnopo/open-amp/commits/flow_ctrl
>
> [3] overview presentation
> https://drive.google.com/file/d/1CLU3ybI3oSBGvor18AQ-HOzOJ2nOppEb/view
>
>
> >
> > Thanks,
> > Mathieu
> >
> >> +            }
> >> +
> >>              /* make sure ept->cb doesn't go away while we use it */
> >>              mutex_lock(&ept->cb_lock);
> >>
> >> --
> >> 2.25.1
> >>
diff mbox series

Patch

diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
index 3ede25b1f2e4..99d2119cc164 100644
--- a/drivers/rpmsg/virtio_rpmsg_bus.c
+++ b/drivers/rpmsg/virtio_rpmsg_bus.c
@@ -708,6 +708,7 @@  static ssize_t virtio_rpmsg_get_mtu(struct rpmsg_endpoint *ept)
 static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
 			     struct rpmsg_hdr *msg, unsigned int len)
 {
+	struct rpmsg_device *rpdev;
 	struct rpmsg_endpoint *ept;
 	struct scatterlist sg;
 	bool little_endian = virtio_is_little_endian(vrp->vdev);
@@ -746,6 +747,15 @@  static int rpmsg_recv_single(struct virtproc_info *vrp, struct device *dev,
 	mutex_unlock(&vrp->endpoints_lock);
 
 	if (ept) {
+		rpdev = ept->rpdev;
+		if (rpdev->ept == ept && rpdev->dst == RPMSG_ADDR_ANY) {
+			/*
+			 * First message received from the remote side on the default endpoint,
+			 * update channel destination address.
+			 */
+			rpdev->dst = msg->src;
+		}
+
 		/* make sure ept->cb doesn't go away while we use it */
 		mutex_lock(&ept->cb_lock);