Skip to contentSkip to conversation

kni: add ability to set min/max MTU

PATCH In review Igor Ryzhov 3 participants 11 messages Sep 19, 2019 – Oct 27, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kni: add ability to set min/max MTU

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

---
 examples/kni/main.c                               | 3 +++
 kernel/linux/kni/compat.h                         | 4 ++++
 kernel/linux/kni/kni_misc.c                       | 8 ++++++++
 lib/librte_eal/linux/eal/include/rte_kni_common.h | 2 ++
 lib/librte_kni/rte_kni.c                          | 2 ++
 lib/librte_kni/rte_kni.h                          | 2 ++
 6 files changed, 21 insertions(+)
examples/kni/main.c +3 -0
index 4710d7176..c22a7c18d 100644
@@ -907,6 +907,9 @@ kni_alloc(uint16_t port_id)
 
 			rte_eth_dev_get_mtu(port_id, &conf.mtu);
 
+			conf.min_mtu = dev_info.min_mtu;
+			conf.max_mtu = dev_info.max_mtu;
+
 			memset(&ops, 0, sizeof(ops));
 			ops.port_id = port_id;
 			ops.change_mtu = kni_change_mtu;
kernel/linux/kni/compat.h +4 -0
index 562d8bf94..fe0ee55e7 100644
@@ -89,6 +89,10 @@
 #define HAVE_TRANS_START_HELPER
 #endif
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 10, 0)
+#define HAVE_MIN_MAX_MTU
+#endif
+
 /*
  * KNI uses NET_NAME_UNKNOWN macro to select correct version of alloc_netdev()
  * For old kernels just backported the commit that enables the macro
kernel/linux/kni/kni_misc.c +8 -0
index 2b75502a8..aeb275329 100644
@@ -390,6 +390,14 @@ kni_ioctl_create(struct net *net, uint32_t ioctl_num,
 	net_dev->max_mtu = net_dev->mtu;
 #endif
 
+#ifdef HAVE_MIN_MAX_MTU
+	if (dev_info.min_mtu)
+		net_dev->min_mtu = dev_info.min_mtu;
+
+	if (dev_info.max_mtu)
+		net_dev->max_mtu = dev_info.max_mtu;
+#endif
+
 	ret = register_netdev(net_dev);
 	if (ret) {
 		pr_err("error %i registering device \"%s\"\n",
lib/librte_eal/linux/eal/include/rte_kni_common.h +2 -0
index 37d9ee8f0..70992d835 100644
@@ -120,6 +120,8 @@ struct rte_kni_device_info {
 	/* mbuf size */
 	unsigned mbuf_size;
 	unsigned int mtu;
+	unsigned int min_mtu;
+	unsigned int max_mtu;
 	uint8_t mac_addr[6];
 };
 
lib/librte_kni/rte_kni.c +2 -0
index 4b51fb4fe..521db27c4 100644
@@ -252,6 +252,8 @@ rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
 	dev_info.group_id = conf->group_id;
 	dev_info.mbuf_size = conf->mbuf_size;
 	dev_info.mtu = conf->mtu;
+	dev_info.min_mtu = conf->min_mtu;
+	dev_info.max_mtu = conf->max_mtu;
 
 	memcpy(dev_info.mac_addr, conf->mac_addr, RTE_ETHER_ADDR_LEN);
 
lib/librte_kni/rte_kni.h +2 -0
index 5699a6443..b22446fa7 100644
@@ -70,6 +70,8 @@ struct rte_kni_conf {
 	uint8_t force_bind : 1; /* Flag to bind kernel thread */
 	uint8_t mac_addr[RTE_ETHER_ADDR_LEN]; /* MAC address assigned to KNI */
 	uint16_t mtu;
+	uint16_t min_mtu;
+	uint16_t max_mtu;
 };
 
 /**
Signature · 1 line
2.23.0
#2 lore raw
Quoted 10 lines · Ferruh Yigit
Quoted 6 lines · Igor Ryzhov

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

I intend to change the title as "kni: fix mtu setting with kernels >= 4.10". Does it sound ok to you?

Would it make sense to backport this patch?

--
David Marchand

#3 lore raw
Quoted 8 lines · David Marchand
Quoted 7 lines · Ferruh Yigit
Quoted 5 lines · Igor Ryzhov

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I intend to change the title as "kni: fix mtu setting with kernels >= 4.10". Does it sound ok to you?

I am not quite sure it is a fix, this patch enables application to pass min/max MTU values for kni netdev but existing code is not doing anything wrong.

Quoted 3 lines

Would it make sense to backport this patch?

#4 lore raw
Quoted 12 lines · Ferruh Yigit
Quoted 10 lines · David Marchand
Quoted 9 lines · Ferruh Yigit
Quoted 6 lines · Igor Ryzhov

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

I intend to change the title as "kni: fix mtu setting with kernels >= 4.10". Does it sound ok to you?

I am not quite sure it is a fix, this patch enables application to pass min/max MTU values for kni netdev but existing code is not doing anything wrong.

To me, starting 4.10, we can't set the mtu to something greater than 1500 on a kni netdev, since netdev uses the ETH_DATA_LEN default value and will refuse a bigger mtu before even calling the change mtu ndo. Did I understand something wrong?

Signature · 1 line
David Marchand
#5 lore raw
Quoted 19 lines · David Marchand
Quoted 17 lines · Ferruh Yigit
Quoted 14 lines · David Marchand
Quoted 10 lines · Ferruh Yigit
Quoted 6 lines · Igor Ryzhov

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

I intend to change the title as "kni: fix mtu setting with kernels >= 4.10". Does it sound ok to you?

Quoted 7 lines
Quoted 2 lines

I am not quite sure it is a fix, this patch enables application to pass min/max MTU values for kni netdev but existing code is not doing anything wrong.

To me, starting 4.10, we can't set the mtu to something greater than 1500 on a kni netdev, since netdev uses the ETH_DATA_LEN default value and will refuse a bigger mtu before even calling the change mtu ndo. Did I understand something wrong?

- As discussed on irc and after looking deeper into the code. The support for jumbo frames was already present and should be working with the current code.
So I agree this does not qualify as a fix, sorry for the noise.

- On the other hand, we could refactor this patch to make use of/merge with the existing HAVE_MAX_MTU_PARAM pre-existing macro without introducing a new one.

- The commit title/log is still a bit cryptic to me, in which case would an application pass a min_mtu/max_mtu?

Thanks.

Signature · 1 line
David Marchand
#6 lore raw
Quoted 29 lines · David Marchand
Quoted 19 lines · David Marchand
Quoted 17 lines · Ferruh Yigit
Quoted 14 lines · David Marchand
Quoted 10 lines · Ferruh Yigit
Quoted 6 lines · Igor Ryzhov

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

I intend to change the title as "kni: fix mtu setting with kernels >= 4.10". Does it sound ok to you?

Quoted 7 lines
Quoted 2 lines

I am not quite sure it is a fix, this patch enables application to pass min/max MTU values for kni netdev but existing code is not doing anything wrong.

To me, starting 4.10, we can't set the mtu to something greater than 1500 on a kni netdev, since netdev uses the ETH_DATA_LEN default value and will refuse a bigger mtu before even calling the change mtu ndo. Did I understand something wrong?

Quoted 8 lines

- As discussed on irc and after looking deeper into the code. The support for jumbo frames was already present and should be working with the current code.
So I agree this does not qualify as a fix, sorry for the noise.

- On the other hand, we could refactor this patch to make use of/merge with the existing HAVE_MAX_MTU_PARAM pre-existing macro without introducing a new one.

+1 to this, I missed the existing HAVE_MAX_MTU_PARAM macro

Quoted 7 lines

- The commit title/log is still a bit cryptic to me, in which case would an application pass a min_mtu/max_mtu?

Thanks.

#7 lore raw

PATCH v2 1/1 kni: add ability to set min/max MTU

In reply to Igor Ryzhov

From: Igor Ryzhov <[email protected]>

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

---
 examples/kni/main.c                               | 3 +++
 kernel/linux/kni/kni_misc.c                       | 6 ++++++
 lib/librte_eal/linux/eal/include/rte_kni_common.h | 2 ++
 lib/librte_kni/rte_kni.c                          | 2 ++
 lib/librte_kni/rte_kni.h                          | 2 ++
 5 files changed, 15 insertions(+)
examples/kni/main.c +3 -0
index c576fc767..5f713e6b2 100644
@@ -949,6 +949,9 @@ kni_alloc(uint16_t port_id)
 
 			rte_eth_dev_get_mtu(port_id, &conf.mtu);
 
+			conf.min_mtu = dev_info.min_mtu;
+			conf.max_mtu = dev_info.max_mtu;
+
 			memset(&ops, 0, sizeof(ops));
 			ops.port_id = port_id;
 			ops.change_mtu = kni_change_mtu;
kernel/linux/kni/kni_misc.c +6 -0
index 2b75502a8..84ef03b5f 100644
@@ -388,6 +388,12 @@ kni_ioctl_create(struct net *net, uint32_t ioctl_num,
 		net_dev->mtu = dev_info.mtu;
 #ifdef HAVE_MAX_MTU_PARAM
 	net_dev->max_mtu = net_dev->mtu;
+
+	if (dev_info.min_mtu)
+		net_dev->min_mtu = dev_info.min_mtu;
+
+	if (dev_info.max_mtu)
+		net_dev->max_mtu = dev_info.max_mtu;
 #endif
 
 	ret = register_netdev(net_dev);
lib/librte_eal/linux/eal/include/rte_kni_common.h +2 -0
index b51fe27bd..46f75a710 100644
@@ -122,6 +122,8 @@ struct rte_kni_device_info {
 	/* mbuf size */
 	unsigned mbuf_size;
 	unsigned int mtu;
+	unsigned int min_mtu;
+	unsigned int max_mtu;
 	uint8_t mac_addr[6];
 };
 
lib/librte_kni/rte_kni.c +2 -0
index 0f3648508..7fbcf2201 100644
@@ -252,6 +252,8 @@ rte_kni_alloc(struct rte_mempool *pktmbuf_pool,
 	dev_info.group_id = conf->group_id;
 	dev_info.mbuf_size = conf->mbuf_size;
 	dev_info.mtu = conf->mtu;
+	dev_info.min_mtu = conf->min_mtu;
+	dev_info.max_mtu = conf->max_mtu;
 
 	memcpy(dev_info.mac_addr, conf->mac_addr, RTE_ETHER_ADDR_LEN);
 
lib/librte_kni/rte_kni.h +2 -0
index f6b66c33d..f1bb782c6 100644
@@ -73,6 +73,8 @@ struct rte_kni_conf {
 	uint8_t force_bind : 1; /* Flag to bind kernel thread */
 	uint8_t mac_addr[RTE_ETHER_ADDR_LEN]; /* MAC address assigned to KNI */
 	uint16_t mtu;
+	uint16_t min_mtu;
+	uint16_t max_mtu;
 };
 
 /**
Signature · 1 line
2.21.0
#8 lore raw
Quoted 9 lines · Ferruh Yigit

From: Igor Ryzhov <[email protected]>

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

v2:
Reuse existing 'HAVE_MAX_MTU_PARAM' macro instead of creating new 'HAVE_MIN_MAX_MTU'

#9 lore raw
In reply to Ferruh Yigit
Quoted 10 lines · Ferruh Yigit

From: Igor Ryzhov <[email protected]>

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

Applied, thanks.

Signature · 1 line
David Marchand
#10 lore raw

Hi David, Ferruh,

Sorry I was on vacation. Thank you for dealing with this.

Best regards,
Igor

Quoted 20 lines · David Marchand
Quoted 10 lines · Ferruh Yigit

From: Igor Ryzhov <[email protected]>

Starting with kernel version 4.10, there are new min/max MTU values in net_device structure, which are set to ETH_MIN_MTU and ETH_DATA_LEN by default. We should be able to change these values to allow MTU more than 1500 to be set on KNI.

Applied, thanks.

--
David Marchand