[dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[Ferruh Yigit]

Virtual device/driver probing done via name.

A new alternative method introduced to probe the device with providing
driver name in devargs as "driver=<driver_name>".

This patch removes alternative method and fixes virtual device usages
with proper device names.

Fixes: 87c3bf29c642 ("test: do not short-circuit null device creation")
Fixes: d39670086a63 ("eal: parse driver argument before probing drivers")

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
---
 drivers/net/null/rte_eth_null.c         |  1 -
 lib/librte_eal/common/eal_common_vdev.c | 40 ++-------------------------------
 test/test/test_link_bonding_rssconf.c   |  5 ++---
 3 files changed, 4 insertions(+), 42 deletions(-)

diff --git a/drivers/net/null/rte_eth_null.c b/drivers/net/null/rte_eth_null.c
index 2c94339..5aef059 100644
--- a/drivers/net/null/rte_eth_null.c
+++ b/drivers/net/null/rte_eth_null.c
@@ -49,7 +49,6 @@ static unsigned default_packet_copy;
 static const char *valid_arguments[] = {
 	ETH_NULL_PACKET_SIZE_ARG,
 	ETH_NULL_PACKET_COPY_ARG,
-	"driver",
 	NULL
 };
 
diff --git a/lib/librte_eal/common/eal_common_vdev.c b/lib/librte_eal/common/eal_common_vdev.c
index 0037a64..8744cc1 100644
--- a/lib/librte_eal/common/eal_common_vdev.c
+++ b/lib/librte_eal/common/eal_common_vdev.c
@@ -70,48 +70,14 @@ rte_vdev_unregister(struct rte_vdev_driver *driver)
 	TAILQ_REMOVE(&vdev_driver_list, driver, next);
 }
 
-/*
- * Parse "driver" devargs without adding a dependency on rte_kvargs.h
- */
-static char *parse_driver_arg(const char *args)
-{
-	const char *c;
-	char *str;
-
-	if (!args || args[0] == '\0')
-		return NULL;
-
-	c = args;
-
-	do {
-		if (strncmp(c, "driver=", 7) == 0) {
-			c += 7;
-			break;
-		}
-
-		c = strchr(c, ',');
-		if (c)
-			c++;
-	} while (c);
-
-	if (c)
-		str = strdup(c);
-	else
-		str = NULL;
-
-	return str;
-}
-
 static int
 vdev_probe_all_drivers(struct rte_vdev_device *dev)
 {
 	const char *name;
-	char *drv_name;
 	struct rte_vdev_driver *driver;
 	int ret = 1;
 
-	drv_name = parse_driver_arg(rte_vdev_device_args(dev));
-	name = drv_name ? drv_name : rte_vdev_device_name(dev);
+	name = rte_vdev_device_name(dev);
 
 	RTE_LOG(DEBUG, EAL, "Search driver %s to probe device %s\n", name,
 		rte_vdev_device_name(dev));
@@ -129,7 +95,7 @@ vdev_probe_all_drivers(struct rte_vdev_device *dev)
 			ret = driver->probe(dev);
 			if (ret)
 				dev->device.driver = NULL;
-			goto out;
+			return ret;
 		}
 	}
 
@@ -146,8 +112,6 @@ vdev_probe_all_drivers(struct rte_vdev_device *dev)
 		}
 	}
 
-out:
-	free(drv_name);
 	return ret;
 }
 
diff --git a/test/test/test_link_bonding_rssconf.c b/test/test/test_link_bonding_rssconf.c
index d28db7d..dcbc9d1 100644
--- a/test/test/test_link_bonding_rssconf.c
+++ b/test/test/test_link_bonding_rssconf.c
@@ -62,7 +62,7 @@
 
 #define BONDED_DEV_NAME         ("rssconf_bond_dev")
 
-#define SLAVE_DEV_NAME_FMT      ("rssconf_slave%d")
+#define SLAVE_DEV_NAME_FMT      ("net_null%d")
 #define SLAVE_RXTX_QUEUE_FMT      ("rssconf_slave%d_q%d")
 
 #define NUM_MBUFS 8191
@@ -550,8 +550,7 @@ test_setup(void)
 		port_id = rte_eth_dev_count();
 		snprintf(name, sizeof(name), SLAVE_DEV_NAME_FMT, port_id);
 
-		retval = rte_vdev_init(name,
-			"driver=net_null,size=64,copy=0");
+		retval = rte_vdev_init(name, "size=64,copy=0");
 		TEST_ASSERT_SUCCESS(retval, "Failed to create null device '%s'\n",
 				name);
 
-- 
2.9.3

Re: [dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[Thomas Monjalon]

10/05/2017 13:01, Ferruh Yigit:
> Virtual device/driver probing done via name.
> 
> A new alternative method introduced to probe the device with providing
> driver name in devargs as "driver=<driver_name>".
> 
> This patch removes alternative method and fixes virtual device usages
> with proper device names.
> 
> Fixes: 87c3bf29c642 ("test: do not short-circuit null device creation")
> Fixes: d39670086a63 ("eal: parse driver argument before probing drivers")
> 
> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
> ---
[...]
>  static int
>  vdev_probe_all_drivers(struct rte_vdev_device *dev)
>  {
>  	const char *name;
> -	char *drv_name;
>  	struct rte_vdev_driver *driver;
>  	int ret = 1;
>  
> -	drv_name = parse_driver_arg(rte_vdev_device_args(dev));
> -	name = drv_name ? drv_name : rte_vdev_device_name(dev);
> +	name = rte_vdev_device_name(dev);

It seems you are reverting the commit d39670086a63:
    eal: parse driver argument before probing drivers
    
    In some cases the virtual device name should be totally different than
    the driver being used for the device. Therefore lets parse the devargs for
    the "driver" argument before probing drivers in vdev_probe_all_drivers().

Is this "driver" option useless?

Re: [dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[Ferruh Yigit]

On 6/8/2017 9:45 PM, Thomas Monjalon wrote:
> 10/05/2017 13:01, Ferruh Yigit:
>> Virtual device/driver probing done via name.
>>
>> A new alternative method introduced to probe the device with providing
>> driver name in devargs as "driver=<driver_name>".
>>
>> This patch removes alternative method and fixes virtual device usages
>> with proper device names.
>>
>> Fixes: 87c3bf29c642 ("test: do not short-circuit null device creation")
>> Fixes: d39670086a63 ("eal: parse driver argument before probing drivers")
>>
>> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
>> ---
> [...]
>>  static int
>>  vdev_probe_all_drivers(struct rte_vdev_device *dev)
>>  {
>>  	const char *name;
>> -	char *drv_name;
>>  	struct rte_vdev_driver *driver;
>>  	int ret = 1;
>>  
>> -	drv_name = parse_driver_arg(rte_vdev_device_args(dev));
>> -	name = drv_name ? drv_name : rte_vdev_device_name(dev);
>> +	name = rte_vdev_device_name(dev);
> 
> It seems you are reverting the commit d39670086a63:
>     eal: parse driver argument before probing drivers

Mostly yes.

>     
>     In some cases the virtual device name should be totally different than
>     the driver being used for the device. Therefore lets parse the devargs for
>     the "driver" argument before probing drivers in vdev_probe_all_drivers().
> 
> Is this "driver" option useless?

There is already a generic way to probe, why need another method, I
didn't get the motivation but this looks like a hack, only place I found
used is in bonding unit test, which can be switched to generic way
without this support.

And this is a hidden / an undocumented feature.

Thanks,
ferruh

Re: [dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[Thomas Monjalon]

Please Jan, could you comment?

09/06/2017 11:21, Ferruh Yigit:
> On 6/8/2017 9:45 PM, Thomas Monjalon wrote:
> > 10/05/2017 13:01, Ferruh Yigit:
> >> Virtual device/driver probing done via name.
> >>
> >> A new alternative method introduced to probe the device with providing
> >> driver name in devargs as "driver=<driver_name>".
> >>
> >> This patch removes alternative method and fixes virtual device usages
> >> with proper device names.
> >>
> >> Fixes: 87c3bf29c642 ("test: do not short-circuit null device creation")
> >> Fixes: d39670086a63 ("eal: parse driver argument before probing drivers")
> >>
> >> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
> >> ---
> > [...]
> >>  static int
> >>  vdev_probe_all_drivers(struct rte_vdev_device *dev)
> >>  {
> >>  	const char *name;
> >> -	char *drv_name;
> >>  	struct rte_vdev_driver *driver;
> >>  	int ret = 1;
> >>  
> >> -	drv_name = parse_driver_arg(rte_vdev_device_args(dev));
> >> -	name = drv_name ? drv_name : rte_vdev_device_name(dev);
> >> +	name = rte_vdev_device_name(dev);
> > 
> > It seems you are reverting the commit d39670086a63:
> >     eal: parse driver argument before probing drivers
> 
> Mostly yes.
> 
> >     
> >     In some cases the virtual device name should be totally different than
> >     the driver being used for the device. Therefore lets parse the devargs for
> >     the "driver" argument before probing drivers in vdev_probe_all_drivers().
> > 
> > Is this "driver" option useless?
> 
> There is already a generic way to probe, why need another method, I
> didn't get the motivation but this looks like a hack, only place I found
> used is in bonding unit test, which can be switched to generic way
> without this support.
> 
> And this is a hidden / an undocumented feature.
> 
> Thanks,
> ferruh
> 

Re: [dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[Jan Blunck]

On Fri, Jun 16, 2017 at 10:20 AM, Thomas Monjalon <thomas@monjalon.net> wrote:
> Please Jan, could you comment?
>

My intention was to provide a way that devices can have arbitrary
names. Ferruh correctly pointed out that this is undocumented and
hidden functionality. Lets apply this and I can rework this if it
becomes necessary later.

> 09/06/2017 11:21, Ferruh Yigit:
>> On 6/8/2017 9:45 PM, Thomas Monjalon wrote:
>> > 10/05/2017 13:01, Ferruh Yigit:
>> >> Virtual device/driver probing done via name.
>> >>
>> >> A new alternative method introduced to probe the device with providing
>> >> driver name in devargs as "driver=<driver_name>".
>> >>
>> >> This patch removes alternative method and fixes virtual device usages
>> >> with proper device names.
>> >>
>> >> Fixes: 87c3bf29c642 ("test: do not short-circuit null device creation")
>> >> Fixes: d39670086a63 ("eal: parse driver argument before probing drivers")
>> >>
>> >> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
>> >> ---
>> > [...]
>> >>  static int
>> >>  vdev_probe_all_drivers(struct rte_vdev_device *dev)
>> >>  {
>> >>    const char *name;
>> >> -  char *drv_name;
>> >>    struct rte_vdev_driver *driver;
>> >>    int ret = 1;
>> >>
>> >> -  drv_name = parse_driver_arg(rte_vdev_device_args(dev));
>> >> -  name = drv_name ? drv_name : rte_vdev_device_name(dev);
>> >> +  name = rte_vdev_device_name(dev);
>> >
>> > It seems you are reverting the commit d39670086a63:
>> >     eal: parse driver argument before probing drivers
>>
>> Mostly yes.
>>
>> >
>> >     In some cases the virtual device name should be totally different than
>> >     the driver being used for the device. Therefore lets parse the devargs for
>> >     the "driver" argument before probing drivers in vdev_probe_all_drivers().
>> >
>> > Is this "driver" option useless?
>>
>> There is already a generic way to probe, why need another method, I
>> didn't get the motivation but this looks like a hack, only place I found
>> used is in bonding unit test, which can be switched to generic way
>> without this support.
>>
>> And this is a hidden / an undocumented feature.
>>
>> Thanks,
>> ferruh
>>
>
>
>

Re: [dpdk-dev] [PATCH] eal: remove vdev probe by dev args

[Thomas Monjalon]

09/07/2017 09:54, Jan Blunck:
> On Fri, Jun 16, 2017 at 10:20 AM, Thomas Monjalon <thomas@monjalon.net> wrote:
> > Please Jan, could you comment?
> >
> 
> My intention was to provide a way that devices can have arbitrary
> names. Ferruh correctly pointed out that this is undocumented and
> hidden functionality. Lets apply this and I can rework this if it
> becomes necessary later.
> 
> > 09/06/2017 11:21, Ferruh Yigit:
> >> On 6/8/2017 9:45 PM, Thomas Monjalon wrote:
> >> > 10/05/2017 13:01, Ferruh Yigit:
> >> >> Virtual device/driver probing done via name.
> >> >>
> >> >> A new alternative method introduced to probe the device with providing
> >> >> driver name in devargs as "driver=<driver_name>".
> >> >>
> >> >> This patch removes alternative method and fixes virtual device usages
> >> >> with proper device names.
> >> >>
> >> >> Fixes: 87c3bf29c642 ("test: do not short-circuit null device creation")
> >> >> Fixes: d39670086a63 ("eal: parse driver argument before probing drivers")
> >> >>
> >> >> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

Applied