diff mbox series

kvargs: fix device iterator match from arguments

Message ID	20211122061250.3220823-1-xuemingl@nvidia.com (mailing list archive)
State	Superseded, archived
Delegated to:	David Marchand
Headers	Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates 216.228.112.34 as permitted sender) receiver=protection.outlook.com; client-ip=216.228.112.34; helo=mail.nvidia.com; From: Xueming Li <xuemingl@nvidia.com> To: <dev@dpdk.org> CC: <xuemingl@nvidia.com>, Lior Margalit <lmargalit@nvidia.com>, <olivier.matz@6wind.com>, Parav Pandit <parav@nvidia.com>, Ray Kinsella <mdr@ashroe.eu>, David Marchand <david.marchand@redhat.com> Subject: [PATCH] kvargs: fix device iterator match from arguments Date: Mon, 22 Nov 2021 14:12:50 +0800 Message-ID: <20211122061250.3220823-1-xuemingl@nvidia.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Content-Type: text/plain Precedence: list Errors-To: dev-bounces@dpdk.org
Series	kvargs: fix device iterator match from arguments \| kvargs: fix device iterator match from arguments

Checks

Context	Check	Description
ci/checkpatch	success	coding style OK
ci/Intel-compilation	success	Compilation OK
ci/intel-Testing	success	Testing PASS
ci/github-robot: build	success	github build: passed
ci/iol-mellanox-Performance	success	Performance Testing PASS
ci/iol-broadcom-Performance	success	Performance Testing PASS
ci/iol-broadcom-Functional	success	Functional Testing PASS
ci/iol-x86_64-unit-testing	success	Testing PASS
ci/iol-x86_64-compile-testing	success	Testing PASS
ci/iol-aarch64-unit-testing	success	Testing PASS
ci/iol-intel-Functional	success	Functional Testing PASS
ci/iol-intel-Performance	success	Performance Testing PASS
ci/iol-aarch64-compile-testing	success	Testing PASS

Commit Message

Xueming Li Nov. 22, 2021, 6:12 a.m. UTC

  Device iterator RTE_DEV_FOREACH() failed to return devices from
classifier like "class=vdpa", because matching name from empty kvargs
returns no result. If device name not specified in kvargs, the function
should iterate all devices.

This patch allows empty devargs or devargs without name specified.

Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
Cc: olivier.matz@6wind.com

Signed-off-by: Xueming Li <xuemingl@nvidia.com>
---
21.11 specific bug, no copy to stable.org
---
 drivers/bus/auxiliary/auxiliary_params.c | 3 +++
 drivers/bus/vdev/vdev_params.c           | 3 +++
 2 files changed, 6 insertions(+)

Comments

Olivier Matz Nov. 23, 2021, 10:25 a.m. UTC | #1

Hi Xueming,

On Mon, Nov 22, 2021 at 02:12:50PM +0800, Xueming Li wrote:
> Device iterator RTE_DEV_FOREACH() failed to return devices from
> classifier like "class=vdpa", because matching name from empty kvargs
> returns no result. If device name not specified in kvargs, the function
> should iterate all devices.
> 
> This patch allows empty devargs or devargs without name specified.
> 
> Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
> Cc: olivier.matz@6wind.com
> 
> Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> ---
> 21.11 specific bug, no copy to stable.org
> ---
>  drivers/bus/auxiliary/auxiliary_params.c | 3 +++
>  drivers/bus/vdev/vdev_params.c           | 3 +++
>  2 files changed, 6 insertions(+)
> 
> diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
> index a9c7853ed1d..6a6382961ea 100644
> --- a/drivers/bus/auxiliary/auxiliary_params.c
> +++ b/drivers/bus/auxiliary/auxiliary_params.c
> @@ -27,6 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
>  	const struct rte_kvargs *kvlist = _kvlist;
>  	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
>  
> +	/* Iterate all devices if name not specified. */
> +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> +		return 0;
>  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
>  		return -1;
>  
> diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> index 37d95395e7a..bab4c0d1d08 100644
> --- a/drivers/bus/vdev/vdev_params.c
> +++ b/drivers/bus/vdev/vdev_params.c
> @@ -29,6 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
>  	const struct rte_kvargs *kvlist = _kvlist;
>  	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
>  
> +	/* Iterate all devices if name not specified. */
> +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> +		return 0;
>  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
>  		return -1;
>  


Thank you for spotting and fixing this issue. The patch looks good to
me, but may I suggest an alternative that would avoid to browse the
kvlist twice? It is not yes tested, just for discussion. The idea
is to add an errno for error cases of rte_kvargs_get_with_value()
to identify the different cases.

  --- a/drivers/bus/auxiliary/auxiliary_params.c
  +++ b/drivers/bus/auxiliary/auxiliary_params.c
  @@ -27,7 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
   	const struct rte_kvargs *kvlist = _kvlist;
   	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
   
  -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
  +	/* if kvlist is valid and contains the key, filter matching devices */
  +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
  +	    rte_errno == ENOENT)
   		return -1;
   
   	return 0;
  diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
  index 37d95395e7..0a5a8a9f58 100644
  --- a/drivers/bus/vdev/vdev_params.c
  +++ b/drivers/bus/vdev/vdev_params.c
  @@ -29,7 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
   	const struct rte_kvargs *kvlist = _kvlist;
   	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
   
  -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
  +	/* if kvlist is valid and contains the key, filter matching devices */
  +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
  +	    rte_errno == ENOENT)
   		return -1;
   
   	return 0;
  diff --git a/lib/kvargs/rte_kvargs.c b/lib/kvargs/rte_kvargs.c
  index 11f624ef14..f1491715bf 100644
  --- a/lib/kvargs/rte_kvargs.c
  +++ b/lib/kvargs/rte_kvargs.c
  @@ -209,17 +209,28 @@ const char *
   rte_kvargs_get_with_value(const struct rte_kvargs *kvlist, const char *key,
   			  const char *value)
   {
  +	int key_found = 0;
   	unsigned int i;
   
  -	if (kvlist == NULL)
  +	if (kvlist == NULL) {
  +		rte_errno = EINVAL;
   		return NULL;
  +	}
  +
   	for (i = 0; i < kvlist->count; ++i) {
   		if (key != NULL && strcmp(kvlist->pairs[i].key, key) != 0)
   			continue;
  +		key_found = 1;
   		if (value != NULL && strcmp(kvlist->pairs[i].value, value) != 0)
   			continue;
   		return kvlist->pairs[i].value;
   	}
  +
  +	if (key_found)
  +		rte_errno = ENODEV;
  +	else
  +		rte_errno = ENOENT;
  +
   	return NULL;
   }
   
  diff --git a/lib/kvargs/rte_kvargs.h b/lib/kvargs/rte_kvargs.h
  index 359a9f5b09..3cb22ece48 100644
  --- a/lib/kvargs/rte_kvargs.h
  +++ b/lib/kvargs/rte_kvargs.h
  @@ -152,8 +152,12 @@ const char *rte_kvargs_get(const struct rte_kvargs *kvlist, const char *key);
    *   The matching value. If NULL, any value will match.
    *
    * @return
  - *   NULL if no key matches the input,
  - *   a value associated with a matching key otherwise.
  + *   The value associated with a matching key/value on success.
  + *   On error, return NULL and rte_errno is set:
  + *   - EINVAL - kvlist is NULL
  + *   - ENOENT - no matching key/value tuple, but the key matches with
  + *              a different value
  + *   - ENODEV - key is not found in the kvlist
    */
   __rte_experimental
   const char *rte_kvargs_get_with_value(const struct rte_kvargs *kvlist,


Let me know if it would work for you. I can submit a patch if you want.

I can add a unit test for kvargs, but do you know where we could add a
unit test for the dev iterate?

Thanks,
Olivier

Xueming Li Nov. 23, 2021, 11:25 a.m. UTC | #2

On Tue, 2021-11-23 at 11:25 +0100, Olivier Matz wrote:
> Hi Xueming,
> 
> On Mon, Nov 22, 2021 at 02:12:50PM +0800, Xueming Li wrote:
> > Device iterator RTE_DEV_FOREACH() failed to return devices from
> > classifier like "class=vdpa", because matching name from empty kvargs
> > returns no result. If device name not specified in kvargs, the function
> > should iterate all devices.
> > 
> > This patch allows empty devargs or devargs without name specified.
> > 
> > Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
> > Cc: olivier.matz@6wind.com
> > 
> > Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> > ---
> > 21.11 specific bug, no copy to stable.org
> > ---
> >  drivers/bus/auxiliary/auxiliary_params.c | 3 +++
> >  drivers/bus/vdev/vdev_params.c           | 3 +++
> >  2 files changed, 6 insertions(+)
> > 
> > diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
> > index a9c7853ed1d..6a6382961ea 100644
> > --- a/drivers/bus/auxiliary/auxiliary_params.c
> > +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > @@ -27,6 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> >  	const struct rte_kvargs *kvlist = _kvlist;
> >  	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> >  
> > +	/* Iterate all devices if name not specified. */
> > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > +		return 0;
> >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> >  		return -1;
> >  
> > diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > index 37d95395e7a..bab4c0d1d08 100644
> > --- a/drivers/bus/vdev/vdev_params.c
> > +++ b/drivers/bus/vdev/vdev_params.c
> > @@ -29,6 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> >  	const struct rte_kvargs *kvlist = _kvlist;
> >  	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> >  
> > +	/* Iterate all devices if name not specified. */
> > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > +		return 0;
> >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> >  		return -1;
> >  
> 
> 
> Thank you for spotting and fixing this issue. The patch looks good to
> me, but may I suggest an alternative that would avoid to browse the
> kvlist twice? It is not yes tested, just for discussion. The idea
> is to add an errno for error cases of rte_kvargs_get_with_value()
> to identify the different cases.

Yes, the code walk the kvlist twice. An alternative complex code could
be this:

	if (kvlist == NULL)
		return 0;
	name = rte_kvargs_get(kvlist, key);
	if (name == NULL)
		/* Iterate all devices if name not specified. */
		return 0;
	if (strcmp(name, dev->name) != 0)
  		return -1;

> 
>   --- a/drivers/bus/auxiliary/auxiliary_params.c
>   +++ b/drivers/bus/auxiliary/auxiliary_params.c
>   @@ -27,7 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
>    	const struct rte_kvargs *kvlist = _kvlist;
>    	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
>    
>   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
>   +	/* if kvlist is valid and contains the key, filter matching devices */
>   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
>   +	    rte_errno == ENOENT)

rte_errno == ENODEV? we should allow ENOENT - name not specified. 

>    		return -1;
>    
>    	return 0;
>   diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
>   index 37d95395e7..0a5a8a9f58 100644
>   --- a/drivers/bus/vdev/vdev_params.c
>   +++ b/drivers/bus/vdev/vdev_params.c
>   @@ -29,7 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
>    	const struct rte_kvargs *kvlist = _kvlist;
>    	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
>    
>   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
>   +	/* if kvlist is valid and contains the key, filter matching devices */
>   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
>   +	    rte_errno == ENOENT)

Same, ENODEV? which means name specified but not match.

>    		return -1;
>    
>    	return 0;
>   diff --git a/lib/kvargs/rte_kvargs.c b/lib/kvargs/rte_kvargs.c
>   index 11f624ef14..f1491715bf 100644
>   --- a/lib/kvargs/rte_kvargs.c
>   +++ b/lib/kvargs/rte_kvargs.c
>   @@ -209,17 +209,28 @@ const char *
>    rte_kvargs_get_with_value(const struct rte_kvargs *kvlist, const char *key,
>    			  const char *value)
>    {
>   +	int key_found = 0;
>    	unsigned int i;
>    
>   -	if (kvlist == NULL)
>   +	if (kvlist == NULL) {
>   +		rte_errno = EINVAL;
>    		return NULL;
>   +	}
>   +
>    	for (i = 0; i < kvlist->count; ++i) {
>    		if (key != NULL && strcmp(kvlist->pairs[i].key, key) != 0)
>    			continue;
>   +		key_found = 1;
>    		if (value != NULL && strcmp(kvlist->pairs[i].value, value) != 0)
>    			continue;
>    		return kvlist->pairs[i].value;
>    	}
>   +
>   +	if (key_found)
>   +		rte_errno = ENODEV;
>   +	else
>   +		rte_errno = ENOENT;
>   +
>    	return NULL;
>    }
>    
>   diff --git a/lib/kvargs/rte_kvargs.h b/lib/kvargs/rte_kvargs.h
>   index 359a9f5b09..3cb22ece48 100644
>   --- a/lib/kvargs/rte_kvargs.h
>   +++ b/lib/kvargs/rte_kvargs.h
>   @@ -152,8 +152,12 @@ const char *rte_kvargs_get(const struct rte_kvargs *kvlist, const char *key);
>     *   The matching value. If NULL, any value will match.
>     *
>     * @return
>   - *   NULL if no key matches the input,
>   - *   a value associated with a matching key otherwise.
>   + *   The value associated with a matching key/value on success.
>   + *   On error, return NULL and rte_errno is set:
>   + *   - EINVAL - kvlist is NULL
>   + *   - ENOENT - no matching key/value tuple, but the key matches with
>   + *              a different value
>   + *   - ENODEV - key is not found in the kvlist
>     */
>    __rte_experimental
>    const char *rte_kvargs_get_with_value(const struct rte_kvargs *kvlist,
> 
> 
> Let me know if it would work for you. I can submit a patch if you want.


LGTM, let's use your patch.

> 
> I can add a unit test for kvargs, but do you know where we could add a
> unit test for the dev iterate?

Good idea, how about test_devargs.c?

> 
> Thanks,
> Olivier

Olivier Matz Nov. 23, 2021, 12:31 p.m. UTC | #3

On Tue, Nov 23, 2021 at 11:25:22AM +0000, Xueming(Steven) Li wrote:
> On Tue, 2021-11-23 at 11:25 +0100, Olivier Matz wrote:
> > Hi Xueming,
> > 
> > On Mon, Nov 22, 2021 at 02:12:50PM +0800, Xueming Li wrote:
> > > Device iterator RTE_DEV_FOREACH() failed to return devices from
> > > classifier like "class=vdpa", because matching name from empty kvargs
> > > returns no result. If device name not specified in kvargs, the function
> > > should iterate all devices.
> > > 
> > > This patch allows empty devargs or devargs without name specified.
> > > 
> > > Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
> > > Cc: olivier.matz@6wind.com
> > > 
> > > Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> > > ---
> > > 21.11 specific bug, no copy to stable.org
> > > ---
> > >  drivers/bus/auxiliary/auxiliary_params.c | 3 +++
> > >  drivers/bus/vdev/vdev_params.c           | 3 +++
> > >  2 files changed, 6 insertions(+)
> > > 
> > > diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
> > > index a9c7853ed1d..6a6382961ea 100644
> > > --- a/drivers/bus/auxiliary/auxiliary_params.c
> > > +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > > @@ -27,6 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > >  	const struct rte_kvargs *kvlist = _kvlist;
> > >  	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > >  
> > > +	/* Iterate all devices if name not specified. */
> > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > +		return 0;
> > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > >  		return -1;
> > >  
> > > diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > > index 37d95395e7a..bab4c0d1d08 100644
> > > --- a/drivers/bus/vdev/vdev_params.c
> > > +++ b/drivers/bus/vdev/vdev_params.c
> > > @@ -29,6 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > >  	const struct rte_kvargs *kvlist = _kvlist;
> > >  	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > >  
> > > +	/* Iterate all devices if name not specified. */
> > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > +		return 0;
> > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > >  		return -1;
> > >  
> > 
> > 
> > Thank you for spotting and fixing this issue. The patch looks good to
> > me, but may I suggest an alternative that would avoid to browse the
> > kvlist twice? It is not yes tested, just for discussion. The idea
> > is to add an errno for error cases of rte_kvargs_get_with_value()
> > to identify the different cases.
> 
> Yes, the code walk the kvlist twice. An alternative complex code could
> be this:
> 
> 	if (kvlist == NULL)
> 		return 0;
> 	name = rte_kvargs_get(kvlist, key);
> 	if (name == NULL)
> 		/* Iterate all devices if name not specified. */
> 		return 0;
> 	if (strcmp(name, dev->name) != 0)
>   		return -1;

Maybe it is equivalent for common usages, but a difference is that
rte_kvargs_get_with_value(kvlist, "key", "value2") also matches
a kvlist="key=value1,key=value2" (i.e. the case where the key appears
several times with different values).


> > 
> >   --- a/drivers/bus/auxiliary/auxiliary_params.c
> >   +++ b/drivers/bus/auxiliary/auxiliary_params.c
> >   @@ -27,7 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> >    	const struct rte_kvargs *kvlist = _kvlist;
> >    	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> >    
> >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> >   +	/* if kvlist is valid and contains the key, filter matching devices */
> >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> >   +	    rte_errno == ENOENT)
> 
> rte_errno == ENODEV? we should allow ENOENT - name not specified. 
> 
> >    		return -1;

I think the patch is correct: the item is filtered only if there is no
matching key/value tuple, but the key is found.

If kvlist is NULL (rte_errno=EINVAL), or if the key does not exist
(rte_errno=ENODEV), we return 0.

As a side note, there is an errno ENOKEY, which would be much better
than ENODEV... but it seems it is not POSIX, so it may not exist on
FreeBSD or Windows.

> >    
> >    	return 0;
> >   diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> >   index 37d95395e7..0a5a8a9f58 100644
> >   --- a/drivers/bus/vdev/vdev_params.c
> >   +++ b/drivers/bus/vdev/vdev_params.c
> >   @@ -29,7 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> >    	const struct rte_kvargs *kvlist = _kvlist;
> >    	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> >    
> >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> >   +	/* if kvlist is valid and contains the key, filter matching devices */
> >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> >   +	    rte_errno == ENOENT)
> 
> Same, ENODEV? which means name specified but not match.
> 
> >    		return -1;
> >    
> >    	return 0;
> >   diff --git a/lib/kvargs/rte_kvargs.c b/lib/kvargs/rte_kvargs.c
> >   index 11f624ef14..f1491715bf 100644
> >   --- a/lib/kvargs/rte_kvargs.c
> >   +++ b/lib/kvargs/rte_kvargs.c
> >   @@ -209,17 +209,28 @@ const char *
> >    rte_kvargs_get_with_value(const struct rte_kvargs *kvlist, const char *key,
> >    			  const char *value)
> >    {
> >   +	int key_found = 0;
> >    	unsigned int i;
> >    
> >   -	if (kvlist == NULL)
> >   +	if (kvlist == NULL) {
> >   +		rte_errno = EINVAL;
> >    		return NULL;
> >   +	}
> >   +
> >    	for (i = 0; i < kvlist->count; ++i) {
> >    		if (key != NULL && strcmp(kvlist->pairs[i].key, key) != 0)
> >    			continue;
> >   +		key_found = 1;
> >    		if (value != NULL && strcmp(kvlist->pairs[i].value, value) != 0)
> >    			continue;
> >    		return kvlist->pairs[i].value;
> >    	}
> >   +
> >   +	if (key_found)
> >   +		rte_errno = ENODEV;
> >   +	else
> >   +		rte_errno = ENOENT;
> >   +
> >    	return NULL;
> >    }
> >    
> >   diff --git a/lib/kvargs/rte_kvargs.h b/lib/kvargs/rte_kvargs.h
> >   index 359a9f5b09..3cb22ece48 100644
> >   --- a/lib/kvargs/rte_kvargs.h
> >   +++ b/lib/kvargs/rte_kvargs.h
> >   @@ -152,8 +152,12 @@ const char *rte_kvargs_get(const struct rte_kvargs *kvlist, const char *key);
> >     *   The matching value. If NULL, any value will match.
> >     *
> >     * @return
> >   - *   NULL if no key matches the input,
> >   - *   a value associated with a matching key otherwise.
> >   + *   The value associated with a matching key/value on success.
> >   + *   On error, return NULL and rte_errno is set:
> >   + *   - EINVAL - kvlist is NULL
> >   + *   - ENOENT - no matching key/value tuple, but the key matches with
> >   + *              a different value
> >   + *   - ENODEV - key is not found in the kvlist
> >     */
> >    __rte_experimental
> >    const char *rte_kvargs_get_with_value(const struct rte_kvargs *kvlist,
> > 
> > 
> > Let me know if it would work for you. I can submit a patch if you want.
> 
> 
> LGTM, let's use your patch.

OK, I'll submit this patch then. Thanks for the feedback.

> > 
> > I can add a unit test for kvargs, but do you know where we could add a
> > unit test for the dev iterate?
> 
> Good idea, how about test_devargs.c?

Thanks, I'll have a look.

> > 
> > Thanks,
> > Olivier
>

Xueming Li Nov. 23, 2021, 12:49 p.m. UTC | #4

On Tue, 2021-11-23 at 13:31 +0100, Olivier Matz wrote:
> On Tue, Nov 23, 2021 at 11:25:22AM +0000, Xueming(Steven) Li wrote:
> > On Tue, 2021-11-23 at 11:25 +0100, Olivier Matz wrote:
> > > Hi Xueming,
> > > 
> > > On Mon, Nov 22, 2021 at 02:12:50PM +0800, Xueming Li wrote:
> > > > Device iterator RTE_DEV_FOREACH() failed to return devices from
> > > > classifier like "class=vdpa", because matching name from empty kvargs
> > > > returns no result. If device name not specified in kvargs, the function
> > > > should iterate all devices.
> > > > 
> > > > This patch allows empty devargs or devargs without name specified.
> > > > 
> > > > Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
> > > > Cc: olivier.matz@6wind.com
> > > > 
> > > > Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> > > > ---
> > > > 21.11 specific bug, no copy to stable.org
> > > > ---
> > > >  drivers/bus/auxiliary/auxiliary_params.c | 3 +++
> > > >  drivers/bus/vdev/vdev_params.c           | 3 +++
> > > >  2 files changed, 6 insertions(+)
> > > > 
> > > > diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
> > > > index a9c7853ed1d..6a6382961ea 100644
> > > > --- a/drivers/bus/auxiliary/auxiliary_params.c
> > > > +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > > > @@ -27,6 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > > >  	const struct rte_kvargs *kvlist = _kvlist;
> > > >  	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > > >  
> > > > +	/* Iterate all devices if name not specified. */
> > > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > > +		return 0;
> > > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > >  		return -1;
> > > >  
> > > > diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > > > index 37d95395e7a..bab4c0d1d08 100644
> > > > --- a/drivers/bus/vdev/vdev_params.c
> > > > +++ b/drivers/bus/vdev/vdev_params.c
> > > > @@ -29,6 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > > >  	const struct rte_kvargs *kvlist = _kvlist;
> > > >  	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > > >  
> > > > +	/* Iterate all devices if name not specified. */
> > > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > > +		return 0;
> > > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > >  		return -1;
> > > >  
> > > 
> > > 
> > > Thank you for spotting and fixing this issue. The patch looks good to
> > > me, but may I suggest an alternative that would avoid to browse the
> > > kvlist twice? It is not yes tested, just for discussion. The idea
> > > is to add an errno for error cases of rte_kvargs_get_with_value()
> > > to identify the different cases.
> > 
> > Yes, the code walk the kvlist twice. An alternative complex code could
> > be this:
> > 
> > 	if (kvlist == NULL)
> > 		return 0;
> > 	name = rte_kvargs_get(kvlist, key);
> > 	if (name == NULL)
> > 		/* Iterate all devices if name not specified. */
> > 		return 0;
> > 	if (strcmp(name, dev->name) != 0)
> >   		return -1;
> 
> Maybe it is equivalent for common usages, but a difference is that
> rte_kvargs_get_with_value(kvlist, "key", "value2") also matches
> a kvlist="key=value1,key=value2" (i.e. the case where the key appears
> several times with different values).
> 
> 
> > > 
> > >   --- a/drivers/bus/auxiliary/auxiliary_params.c
> > >   +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > >   @@ -27,7 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > >    	const struct rte_kvargs *kvlist = _kvlist;
> > >    	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > >    
> > >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > >   +	/* if kvlist is valid and contains the key, filter matching devices */
> > >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> > >   +	    rte_errno == ENOENT)
> > 
> > rte_errno == ENODEV? we should allow ENOENT - name not specified. 
> > 
> > >    		return -1;
> 
> I think the patch is correct: the item is filtered only if there is no
> matching key/value tuple, but the key is found.

For "class=vdpa", name key not specified, all devices should be walked.
the kvargs is "" in such case.

> 
> If kvlist is NULL (rte_errno=EINVAL), or if the key does not exist
> (rte_errno=ENODEV), we return 0.
> 
> As a side note, there is an errno ENOKEY, which would be much better
> than ENODEV... but it seems it is not POSIX, so it may not exist on
> FreeBSD or Windows.
> 
> > >    
> > >    	return 0;
> > >   diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > >   index 37d95395e7..0a5a8a9f58 100644
> > >   --- a/drivers/bus/vdev/vdev_params.c
> > >   +++ b/drivers/bus/vdev/vdev_params.c
> > >   @@ -29,7 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > >    	const struct rte_kvargs *kvlist = _kvlist;
> > >    	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > >    
> > >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > >   +	/* if kvlist is valid and contains the key, filter matching devices */
> > >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> > >   +	    rte_errno == ENOENT)
> > 
> > Same, ENODEV? which means name specified but not match.
> > 
> > >    		return -1;
> > >    
> > >    	return 0;
> > >   diff --git a/lib/kvargs/rte_kvargs.c b/lib/kvargs/rte_kvargs.c
> > >   index 11f624ef14..f1491715bf 100644
> > >   --- a/lib/kvargs/rte_kvargs.c
> > >   +++ b/lib/kvargs/rte_kvargs.c
> > >   @@ -209,17 +209,28 @@ const char *
> > >    rte_kvargs_get_with_value(const struct rte_kvargs *kvlist, const char *key,
> > >    			  const char *value)
> > >    {
> > >   +	int key_found = 0;
> > >    	unsigned int i;
> > >    
> > >   -	if (kvlist == NULL)
> > >   +	if (kvlist == NULL) {
> > >   +		rte_errno = EINVAL;
> > >    		return NULL;
> > >   +	}
> > >   +
> > >    	for (i = 0; i < kvlist->count; ++i) {
> > >    		if (key != NULL && strcmp(kvlist->pairs[i].key, key) != 0)
> > >    			continue;
> > >   +		key_found = 1;
> > >    		if (value != NULL && strcmp(kvlist->pairs[i].value, value) != 0)
> > >    			continue;
> > >    		return kvlist->pairs[i].value;
> > >    	}
> > >   +
> > >   +	if (key_found)
> > >   +		rte_errno = ENODEV;
> > >   +	else
> > >   +		rte_errno = ENOENT;
> > >   +
> > >    	return NULL;
> > >    }
> > >    
> > >   diff --git a/lib/kvargs/rte_kvargs.h b/lib/kvargs/rte_kvargs.h
> > >   index 359a9f5b09..3cb22ece48 100644
> > >   --- a/lib/kvargs/rte_kvargs.h
> > >   +++ b/lib/kvargs/rte_kvargs.h
> > >   @@ -152,8 +152,12 @@ const char *rte_kvargs_get(const struct rte_kvargs *kvlist, const char *key);
> > >     *   The matching value. If NULL, any value will match.
> > >     *
> > >     * @return
> > >   - *   NULL if no key matches the input,
> > >   - *   a value associated with a matching key otherwise.
> > >   + *   The value associated with a matching key/value on success.
> > >   + *   On error, return NULL and rte_errno is set:
> > >   + *   - EINVAL - kvlist is NULL
> > >   + *   - ENOENT - no matching key/value tuple, but the key matches with
> > >   + *              a different value
> > >   + *   - ENODEV - key is not found in the kvlist
> > >     */
> > >    __rte_experimental
> > >    const char *rte_kvargs_get_with_value(const struct rte_kvargs *kvlist,
> > > 
> > > 
> > > Let me know if it would work for you. I can submit a patch if you want.
> > 
> > 
> > LGTM, let's use your patch.
> 
> OK, I'll submit this patch then. Thanks for the feedback.
> 
> > > 
> > > I can add a unit test for kvargs, but do you know where we could add a
> > > unit test for the dev iterate?
> > 
> > Good idea, how about test_devargs.c?
> 
> Thanks, I'll have a look.
> 
> > > 
> > > Thanks,
> > > Olivier
> >

Olivier Matz Nov. 23, 2021, 8:02 p.m. UTC | #5

Hello Xueming,

On Tue, Nov 23, 2021 at 12:49:32PM +0000, Xueming(Steven) Li wrote:
> On Tue, 2021-11-23 at 13:31 +0100, Olivier Matz wrote:
> > On Tue, Nov 23, 2021 at 11:25:22AM +0000, Xueming(Steven) Li wrote:
> > > On Tue, 2021-11-23 at 11:25 +0100, Olivier Matz wrote:
> > > > Hi Xueming,
> > > > 
> > > > On Mon, Nov 22, 2021 at 02:12:50PM +0800, Xueming Li wrote:
> > > > > Device iterator RTE_DEV_FOREACH() failed to return devices from
> > > > > classifier like "class=vdpa", because matching name from empty kvargs
> > > > > returns no result. If device name not specified in kvargs, the function
> > > > > should iterate all devices.
> > > > > 
> > > > > This patch allows empty devargs or devargs without name specified.
> > > > > 
> > > > > Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
> > > > > Cc: olivier.matz@6wind.com
> > > > > 
> > > > > Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> > > > > ---
> > > > > 21.11 specific bug, no copy to stable.org
> > > > > ---
> > > > >  drivers/bus/auxiliary/auxiliary_params.c | 3 +++
> > > > >  drivers/bus/vdev/vdev_params.c           | 3 +++
> > > > >  2 files changed, 6 insertions(+)
> > > > > 
> > > > > diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
> > > > > index a9c7853ed1d..6a6382961ea 100644
> > > > > --- a/drivers/bus/auxiliary/auxiliary_params.c
> > > > > +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > > > > @@ -27,6 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > > > >  	const struct rte_kvargs *kvlist = _kvlist;
> > > > >  	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > > > >  
> > > > > +	/* Iterate all devices if name not specified. */
> > > > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > > > +		return 0;
> > > > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > > >  		return -1;
> > > > >  
> > > > > diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > > > > index 37d95395e7a..bab4c0d1d08 100644
> > > > > --- a/drivers/bus/vdev/vdev_params.c
> > > > > +++ b/drivers/bus/vdev/vdev_params.c
> > > > > @@ -29,6 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > > > >  	const struct rte_kvargs *kvlist = _kvlist;
> > > > >  	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > > > >  
> > > > > +	/* Iterate all devices if name not specified. */
> > > > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > > > +		return 0;
> > > > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > > >  		return -1;
> > > > >  
> > > > 
> > > > 
> > > > Thank you for spotting and fixing this issue. The patch looks good to
> > > > me, but may I suggest an alternative that would avoid to browse the
> > > > kvlist twice? It is not yes tested, just for discussion. The idea
> > > > is to add an errno for error cases of rte_kvargs_get_with_value()
> > > > to identify the different cases.
> > > 
> > > Yes, the code walk the kvlist twice. An alternative complex code could
> > > be this:
> > > 
> > > 	if (kvlist == NULL)
> > > 		return 0;
> > > 	name = rte_kvargs_get(kvlist, key);
> > > 	if (name == NULL)
> > > 		/* Iterate all devices if name not specified. */
> > > 		return 0;
> > > 	if (strcmp(name, dev->name) != 0)
> > >   		return -1;
> > 
> > Maybe it is equivalent for common usages, but a difference is that
> > rte_kvargs_get_with_value(kvlist, "key", "value2") also matches
> > a kvlist="key=value1,key=value2" (i.e. the case where the key appears
> > several times with different values).
> > 
> > 
> > > > 
> > > >   --- a/drivers/bus/auxiliary/auxiliary_params.c
> > > >   +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > > >   @@ -27,7 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > > >    	const struct rte_kvargs *kvlist = _kvlist;
> > > >    	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > > >    
> > > >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > >   +	/* if kvlist is valid and contains the key, filter matching devices */
> > > >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> > > >   +	    rte_errno == ENOENT)
> > > 
> > > rte_errno == ENODEV? we should allow ENOENT - name not specified. 
> > > 
> > > >    		return -1;
> > 
> > I think the patch is correct: the item is filtered only if there is no
> > matching key/value tuple, but the key is found.
> 
> For "class=vdpa", name key not specified, all devices should be walked.
> the kvargs is "" in such case.
> 
> > 
> > If kvlist is NULL (rte_errno=EINVAL), or if the key does not exist
> > (rte_errno=ENODEV), we return 0.
> > 
> > As a side note, there is an errno ENOKEY, which would be much better
> > than ENODEV... but it seems it is not POSIX, so it may not exist on
> > FreeBSD or Windows.
> > 
> > > >    
> > > >    	return 0;
> > > >   diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > > >   index 37d95395e7..0a5a8a9f58 100644
> > > >   --- a/drivers/bus/vdev/vdev_params.c
> > > >   +++ b/drivers/bus/vdev/vdev_params.c
> > > >   @@ -29,7 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > > >    	const struct rte_kvargs *kvlist = _kvlist;
> > > >    	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > > >    
> > > >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > >   +	/* if kvlist is valid and contains the key, filter matching devices */
> > > >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> > > >   +	    rte_errno == ENOENT)
> > > 
> > > Same, ENODEV? which means name specified but not match.
> > > 
> > > >    		return -1;
> > > >    
> > > >    	return 0;
> > > >   diff --git a/lib/kvargs/rte_kvargs.c b/lib/kvargs/rte_kvargs.c
> > > >   index 11f624ef14..f1491715bf 100644
> > > >   --- a/lib/kvargs/rte_kvargs.c
> > > >   +++ b/lib/kvargs/rte_kvargs.c
> > > >   @@ -209,17 +209,28 @@ const char *
> > > >    rte_kvargs_get_with_value(const struct rte_kvargs *kvlist, const char *key,
> > > >    			  const char *value)
> > > >    {
> > > >   +	int key_found = 0;
> > > >    	unsigned int i;
> > > >    
> > > >   -	if (kvlist == NULL)
> > > >   +	if (kvlist == NULL) {
> > > >   +		rte_errno = EINVAL;
> > > >    		return NULL;
> > > >   +	}
> > > >   +
> > > >    	for (i = 0; i < kvlist->count; ++i) {
> > > >    		if (key != NULL && strcmp(kvlist->pairs[i].key, key) != 0)
> > > >    			continue;
> > > >   +		key_found = 1;
> > > >    		if (value != NULL && strcmp(kvlist->pairs[i].value, value) != 0)
> > > >    			continue;
> > > >    		return kvlist->pairs[i].value;
> > > >    	}
> > > >   +
> > > >   +	if (key_found)
> > > >   +		rte_errno = ENODEV;
> > > >   +	else
> > > >   +		rte_errno = ENOENT;
> > > >   +
> > > >    	return NULL;
> > > >    }
> > > >    
> > > >   diff --git a/lib/kvargs/rte_kvargs.h b/lib/kvargs/rte_kvargs.h
> > > >   index 359a9f5b09..3cb22ece48 100644
> > > >   --- a/lib/kvargs/rte_kvargs.h
> > > >   +++ b/lib/kvargs/rte_kvargs.h
> > > >   @@ -152,8 +152,12 @@ const char *rte_kvargs_get(const struct rte_kvargs *kvlist, const char *key);
> > > >     *   The matching value. If NULL, any value will match.
> > > >     *
> > > >     * @return
> > > >   - *   NULL if no key matches the input,
> > > >   - *   a value associated with a matching key otherwise.
> > > >   + *   The value associated with a matching key/value on success.
> > > >   + *   On error, return NULL and rte_errno is set:
> > > >   + *   - EINVAL - kvlist is NULL
> > > >   + *   - ENOENT - no matching key/value tuple, but the key matches with
> > > >   + *              a different value
> > > >   + *   - ENODEV - key is not found in the kvlist
> > > >     */
> > > >    __rte_experimental
> > > >    const char *rte_kvargs_get_with_value(const struct rte_kvargs *kvlist,
> > > > 
> > > > 
> > > > Let me know if it would work for you. I can submit a patch if you want.
> > > 
> > > 
> > > LGTM, let's use your patch.
> > 
> > OK, I'll submit this patch then. Thanks for the feedback.

I'm trying my patch, but I run into an issue.
Currently, EAL already depends on kvargs.
If I add a reference to rte_errno from kvargs, it will require a
new dependency to EAL... which is impossible.

Using errno instead of rte_errno is not usual in DPDK.
Having a typedef for this is not very convenient either.

So thinking more about it, your solution looks better.

I have a draft of unit test, I'll send an update of your patch including
it ASAP.

> > 
> > > > 
> > > > I can add a unit test for kvargs, but do you know where we could add a
> > > > unit test for the dev iterate?
> > > 
> > > Good idea, how about test_devargs.c?
> > 
> > Thanks, I'll have a look.
> > 
> > > > 
> > > > Thanks,
> > > > Olivier
> > > 
>

Xueming Li Nov. 24, 2021, 10:21 a.m. UTC | #6

On Tue, 2021-11-23 at 21:02 +0100, Olivier Matz wrote:
> Hello Xueming,
> 
> On Tue, Nov 23, 2021 at 12:49:32PM +0000, Xueming(Steven) Li wrote:
> > On Tue, 2021-11-23 at 13:31 +0100, Olivier Matz wrote:
> > > On Tue, Nov 23, 2021 at 11:25:22AM +0000, Xueming(Steven) Li wrote:
> > > > On Tue, 2021-11-23 at 11:25 +0100, Olivier Matz wrote:
> > > > > Hi Xueming,
> > > > > 
> > > > > On Mon, Nov 22, 2021 at 02:12:50PM +0800, Xueming Li wrote:
> > > > > > Device iterator RTE_DEV_FOREACH() failed to return devices from
> > > > > > classifier like "class=vdpa", because matching name from empty kvargs
> > > > > > returns no result. If device name not specified in kvargs, the function
> > > > > > should iterate all devices.
> > > > > > 
> > > > > > This patch allows empty devargs or devargs without name specified.
> > > > > > 
> > > > > > Fixes: 6aebb942907d ("kvargs: add function to get from key and value")
> > > > > > Cc: olivier.matz@6wind.com
> > > > > > 
> > > > > > Signed-off-by: Xueming Li <xuemingl@nvidia.com>
> > > > > > ---
> > > > > > 21.11 specific bug, no copy to stable.org
> > > > > > ---
> > > > > >  drivers/bus/auxiliary/auxiliary_params.c | 3 +++
> > > > > >  drivers/bus/vdev/vdev_params.c           | 3 +++
> > > > > >  2 files changed, 6 insertions(+)
> > > > > > 
> > > > > > diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
> > > > > > index a9c7853ed1d..6a6382961ea 100644
> > > > > > --- a/drivers/bus/auxiliary/auxiliary_params.c
> > > > > > +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > > > > > @@ -27,6 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > > > > >  	const struct rte_kvargs *kvlist = _kvlist;
> > > > > >  	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > > > > >  
> > > > > > +	/* Iterate all devices if name not specified. */
> > > > > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > > > > +		return 0;
> > > > > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > > > >  		return -1;
> > > > > >  
> > > > > > diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > > > > > index 37d95395e7a..bab4c0d1d08 100644
> > > > > > --- a/drivers/bus/vdev/vdev_params.c
> > > > > > +++ b/drivers/bus/vdev/vdev_params.c
> > > > > > @@ -29,6 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > > > > >  	const struct rte_kvargs *kvlist = _kvlist;
> > > > > >  	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > > > > >  
> > > > > > +	/* Iterate all devices if name not specified. */
> > > > > > +	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
> > > > > > +		return 0;
> > > > > >  	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > > > >  		return -1;
> > > > > >  
> > > > > 
> > > > > 
> > > > > Thank you for spotting and fixing this issue. The patch looks good to
> > > > > me, but may I suggest an alternative that would avoid to browse the
> > > > > kvlist twice? It is not yes tested, just for discussion. The idea
> > > > > is to add an errno for error cases of rte_kvargs_get_with_value()
> > > > > to identify the different cases.
> > > > 
> > > > Yes, the code walk the kvlist twice. An alternative complex code could
> > > > be this:
> > > > 
> > > > 	if (kvlist == NULL)
> > > > 		return 0;
> > > > 	name = rte_kvargs_get(kvlist, key);
> > > > 	if (name == NULL)
> > > > 		/* Iterate all devices if name not specified. */
> > > > 		return 0;
> > > > 	if (strcmp(name, dev->name) != 0)
> > > >   		return -1;
> > > 
> > > Maybe it is equivalent for common usages, but a difference is that
> > > rte_kvargs_get_with_value(kvlist, "key", "value2") also matches
> > > a kvlist="key=value1,key=value2" (i.e. the case where the key appears
> > > several times with different values).
> > > 
> > > 
> > > > > 
> > > > >   --- a/drivers/bus/auxiliary/auxiliary_params.c
> > > > >   +++ b/drivers/bus/auxiliary/auxiliary_params.c
> > > > >   @@ -27,7 +27,9 @@ auxiliary_dev_match(const struct rte_device *dev,
> > > > >    	const struct rte_kvargs *kvlist = _kvlist;
> > > > >    	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
> > > > >    
> > > > >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > > >   +	/* if kvlist is valid and contains the key, filter matching devices */
> > > > >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> > > > >   +	    rte_errno == ENOENT)
> > > > 
> > > > rte_errno == ENODEV? we should allow ENOENT - name not specified. 
> > > > 
> > > > >    		return -1;
> > > 
> > > I think the patch is correct: the item is filtered only if there is no
> > > matching key/value tuple, but the key is found.
> > 
> > For "class=vdpa", name key not specified, all devices should be walked.
> > the kvargs is "" in such case.
> > 
> > > 
> > > If kvlist is NULL (rte_errno=EINVAL), or if the key does not exist
> > > (rte_errno=ENODEV), we return 0.
> > > 
> > > As a side note, there is an errno ENOKEY, which would be much better
> > > than ENODEV... but it seems it is not POSIX, so it may not exist on
> > > FreeBSD or Windows.
> > > 
> > > > >    
> > > > >    	return 0;
> > > > >   diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
> > > > >   index 37d95395e7..0a5a8a9f58 100644
> > > > >   --- a/drivers/bus/vdev/vdev_params.c
> > > > >   +++ b/drivers/bus/vdev/vdev_params.c
> > > > >   @@ -29,7 +29,9 @@ vdev_dev_match(const struct rte_device *dev,
> > > > >    	const struct rte_kvargs *kvlist = _kvlist;
> > > > >    	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
> > > > >    
> > > > >   -	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
> > > > >   +	/* if kvlist is valid and contains the key, filter matching devices */
> > > > >   +	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL &&
> > > > >   +	    rte_errno == ENOENT)
> > > > 
> > > > Same, ENODEV? which means name specified but not match.
> > > > 
> > > > >    		return -1;
> > > > >    
> > > > >    	return 0;
> > > > >   diff --git a/lib/kvargs/rte_kvargs.c b/lib/kvargs/rte_kvargs.c
> > > > >   index 11f624ef14..f1491715bf 100644
> > > > >   --- a/lib/kvargs/rte_kvargs.c
> > > > >   +++ b/lib/kvargs/rte_kvargs.c
> > > > >   @@ -209,17 +209,28 @@ const char *
> > > > >    rte_kvargs_get_with_value(const struct rte_kvargs *kvlist, const char *key,
> > > > >    			  const char *value)
> > > > >    {
> > > > >   +	int key_found = 0;
> > > > >    	unsigned int i;
> > > > >    
> > > > >   -	if (kvlist == NULL)
> > > > >   +	if (kvlist == NULL) {
> > > > >   +		rte_errno = EINVAL;
> > > > >    		return NULL;
> > > > >   +	}
> > > > >   +
> > > > >    	for (i = 0; i < kvlist->count; ++i) {
> > > > >    		if (key != NULL && strcmp(kvlist->pairs[i].key, key) != 0)
> > > > >    			continue;
> > > > >   +		key_found = 1;
> > > > >    		if (value != NULL && strcmp(kvlist->pairs[i].value, value) != 0)
> > > > >    			continue;
> > > > >    		return kvlist->pairs[i].value;
> > > > >    	}
> > > > >   +
> > > > >   +	if (key_found)
> > > > >   +		rte_errno = ENODEV;
> > > > >   +	else
> > > > >   +		rte_errno = ENOENT;
> > > > >   +
> > > > >    	return NULL;
> > > > >    }
> > > > >    
> > > > >   diff --git a/lib/kvargs/rte_kvargs.h b/lib/kvargs/rte_kvargs.h
> > > > >   index 359a9f5b09..3cb22ece48 100644
> > > > >   --- a/lib/kvargs/rte_kvargs.h
> > > > >   +++ b/lib/kvargs/rte_kvargs.h
> > > > >   @@ -152,8 +152,12 @@ const char *rte_kvargs_get(const struct rte_kvargs *kvlist, const char *key);
> > > > >     *   The matching value. If NULL, any value will match.
> > > > >     *
> > > > >     * @return
> > > > >   - *   NULL if no key matches the input,
> > > > >   - *   a value associated with a matching key otherwise.
> > > > >   + *   The value associated with a matching key/value on success.
> > > > >   + *   On error, return NULL and rte_errno is set:
> > > > >   + *   - EINVAL - kvlist is NULL
> > > > >   + *   - ENOENT - no matching key/value tuple, but the key matches with
> > > > >   + *              a different value
> > > > >   + *   - ENODEV - key is not found in the kvlist
> > > > >     */
> > > > >    __rte_experimental
> > > > >    const char *rte_kvargs_get_with_value(const struct rte_kvargs *kvlist,
> > > > > 
> > > > > 
> > > > > Let me know if it would work for you. I can submit a patch if you want.
> > > > 
> > > > 
> > > > LGTM, let's use your patch.
> > > 
> > > OK, I'll submit this patch then. Thanks for the feedback.
> 
> I'm trying my patch, but I run into an issue.
> Currently, EAL already depends on kvargs.
> If I add a reference to rte_errno from kvargs, it will require a
> new dependency to EAL... which is impossible.
> 
> Using errno instead of rte_errno is not usual in DPDK.
> Having a typedef for this is not very convenient either.
> 
> So thinking more about it, your solution looks better.
> 
> I have a draft of unit test, I'll send an update of your patch including
> it ASAP.

V2 sent, please check.
Unit test not critical, could be sent later IIUC.


> 
> > > 
> > > > > 
> > > > > I can add a unit test for kvargs, but do you know where we could add a
> > > > > unit test for the dev iterate?
> > > > 
> > > > Good idea, how about test_devargs.c?
> > > 
> > > Thanks, I'll have a look.
> > > 
> > > > > 
> > > > > Thanks,
> > > > > Olivier
> > > > 
> >

diff mbox series

Patch

diff --git a/drivers/bus/auxiliary/auxiliary_params.c b/drivers/bus/auxiliary/auxiliary_params.c
index a9c7853ed1d..6a6382961ea 100644
--- a/drivers/bus/auxiliary/auxiliary_params.c
+++ b/drivers/bus/auxiliary/auxiliary_params.c
@@ -27,6 +27,9 @@  auxiliary_dev_match(const struct rte_device *dev,
 	const struct rte_kvargs *kvlist = _kvlist;
 	const char *key = auxiliary_params_keys[RTE_AUXILIARY_PARAM_NAME];
 
+	/* Iterate all devices if name not specified. */
+	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
+		return 0;
 	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
 		return -1;
 
diff --git a/drivers/bus/vdev/vdev_params.c b/drivers/bus/vdev/vdev_params.c
index 37d95395e7a..bab4c0d1d08 100644
--- a/drivers/bus/vdev/vdev_params.c
+++ b/drivers/bus/vdev/vdev_params.c
@@ -29,6 +29,9 @@  vdev_dev_match(const struct rte_device *dev,
 	const struct rte_kvargs *kvlist = _kvlist;
 	const char *key = vdev_params_keys[RTE_VDEV_PARAM_NAME];
 
+	/* Iterate all devices if name not specified. */
+	if (kvlist == NULL || rte_kvargs_get(kvlist, key) == NULL)
+		return 0;
 	if (rte_kvargs_get_with_value(kvlist, key, dev->name) == NULL)
 		return -1;