| Message ID | 20190925093623.18419-1-iryzhov@nfware.com (mailing list archive) |
|---|---|
| State | New |
| Delegated to: | Thomas Monjalon |
| Headers | show |
| Series | [v3] kni: rework rte_kni_update_link using ioctl | expand |
| Context | Check | Description |
|---|---|---|
| ci/checkpatch | success | coding style OK |
| ci/Intel-compilation | success | Compilation OK |
| ci/iol-dpdk_compile | success | Compile Testing PASS |
| ci/iol-dpdk_compile_ovs | success | Compile Testing PASS |
| ci/iol-dpdk_compile_spdk | success | Compile Testing PASS |
| ci/iol-intel-Performance | success | Performance Testing PASS |
| ci/iol-mellanox-Performance | success | Performance Testing PASS |
On 9/25/2019 10:36 AM, Igor Ryzhov wrote: > Current implementation doesn't allow us to update KNI carrier if the > interface is not yet UP in kernel. It means that we can't use it in the > same thread which is processing rte_kni_ops.config_network_if, which is > very convenient, because it allows us to have correct carrier status > of the interface right after we enabled it and we don't have to use any > additional thread to track link status. Hi Igor, The existing thread tracks the link status of the physical device and reflects the changes to the kni netdev, but the "struct rte_kni_ops" (rte_kni_ops.config_network_if) works other way around, it captures (some) requests to kni netdev and reflects them to the underlying physical device. Even 'rte_kni_update_link()' updated to use ioctl, the thread still looks required and this patch doesn't really changes that part. Also I am reluctant to extend the KNI ioctl interface when there is a generic way to do that work. What is the use case of updating kni netdev carrier status when the interface is down? > > Signed-off-by: Igor Ryzhov <iryzhov@nfware.com> <...>
On 10/14/2019 5:10 PM, Ferruh Yigit wrote: > On 9/25/2019 10:36 AM, Igor Ryzhov wrote: >> Current implementation doesn't allow us to update KNI carrier if the >> interface is not yet UP in kernel. It means that we can't use it in the >> same thread which is processing rte_kni_ops.config_network_if, which is >> very convenient, because it allows us to have correct carrier status >> of the interface right after we enabled it and we don't have to use any >> additional thread to track link status. > > Hi Igor, > > The existing thread tracks the link status of the physical device and reflects > the changes to the kni netdev, but the "struct rte_kni_ops" > (rte_kni_ops.config_network_if) works other way around, it captures (some) > requests to kni netdev and reflects them to the underlying physical device. > Even 'rte_kni_update_link()' updated to use ioctl, the thread still looks > required and this patch doesn't really changes that part. > > Also I am reluctant to extend the KNI ioctl interface when there is a generic > way to do that work. > > What is the use case of updating kni netdev carrier status when the interface is > down? btw, if the problem is status of the interface being 'no-carrier' by default, this can be changed by "carrier=on" parameter of the kni kernel module: "insmod ./build/kmod/rte_kni.ko carrier=on" > >> >> Signed-off-by: Igor Ryzhov <iryzhov@nfware.com> > > <...> >
My original patch to add this feature was basically the same thing as this: setting the link status via a KNI ioctl. That method was rejected after _much_ discussion and we eventually settled on the currently implementation. My original patch was here: Message-Id: <20180628225548.21885-1-dg@adax.com> If you search for KNI and dg@adax.com in the DPDK devel list you should be able to suss out the whole discussion that lead to the current implementation. thanks dan On Mon, Oct 14, 2019 at 1:17 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote: > > On 10/14/2019 5:10 PM, Ferruh Yigit wrote: > > On 9/25/2019 10:36 AM, Igor Ryzhov wrote: > >> Current implementation doesn't allow us to update KNI carrier if the > >> interface is not yet UP in kernel. It means that we can't use it in the > >> same thread which is processing rte_kni_ops.config_network_if, which is > >> very convenient, because it allows us to have correct carrier status > >> of the interface right after we enabled it and we don't have to use any > >> additional thread to track link status. > > > > Hi Igor, > > > > The existing thread tracks the link status of the physical device and reflects > > the changes to the kni netdev, but the "struct rte_kni_ops" > > (rte_kni_ops.config_network_if) works other way around, it captures (some) > > requests to kni netdev and reflects them to the underlying physical device. > > Even 'rte_kni_update_link()' updated to use ioctl, the thread still looks > > required and this patch doesn't really changes that part. > > > > Also I am reluctant to extend the KNI ioctl interface when there is a generic > > way to do that work. > > > > What is the use case of updating kni netdev carrier status when the interface is > > down? > > btw, if the problem is status of the interface being 'no-carrier' by default, > this can be changed by "carrier=on" parameter of the kni kernel module: > "insmod ./build/kmod/rte_kni.ko carrier=on" > > > > >> > >> Signed-off-by: Igor Ryzhov <iryzhov@nfware.com> > > > > <...> > > >
Here's another link to the thread where this was discussed last year.. Igor was actually on this thread as well... https://mails.dpdk.org/archives/dev/2018-August/110383.html On Mon, Oct 14, 2019 at 4:01 PM Dan Gora <dg@adax.com> wrote: > > My original patch to add this feature was basically the same thing as > this: setting the link status via a KNI ioctl. That method was > rejected after _much_ discussion and we eventually settled on the > currently implementation. > > My original patch was here: Message-Id: <20180628225548.21885-1-dg@adax.com> > > If you search for KNI and dg@adax.com in the DPDK devel list you > should be able to suss out the whole discussion that lead to the > current implementation. > > thanks > dan > > On Mon, Oct 14, 2019 at 1:17 PM Ferruh Yigit <ferruh.yigit@intel.com> wrote: > > > > On 10/14/2019 5:10 PM, Ferruh Yigit wrote: > > > On 9/25/2019 10:36 AM, Igor Ryzhov wrote: > > >> Current implementation doesn't allow us to update KNI carrier if the > > >> interface is not yet UP in kernel. It means that we can't use it in the > > >> same thread which is processing rte_kni_ops.config_network_if, which is > > >> very convenient, because it allows us to have correct carrier status > > >> of the interface right after we enabled it and we don't have to use any > > >> additional thread to track link status. > > > > > > Hi Igor, > > > > > > The existing thread tracks the link status of the physical device and reflects > > > the changes to the kni netdev, but the "struct rte_kni_ops" > > > (rte_kni_ops.config_network_if) works other way around, it captures (some) > > > requests to kni netdev and reflects them to the underlying physical device. > > > Even 'rte_kni_update_link()' updated to use ioctl, the thread still looks > > > required and this patch doesn't really changes that part. > > > > > > Also I am reluctant to extend the KNI ioctl interface when there is a generic > > > way to do that work. > > > > > > What is the use case of updating kni netdev carrier status when the interface is > > > down? > > > > btw, if the problem is status of the interface being 'no-carrier' by default, > > this can be changed by "carrier=on" parameter of the kni kernel module: > > "insmod ./build/kmod/rte_kni.ko carrier=on"
Hi Ferruh, Dan, Sure, I remember last year discussion but now I see the problem in current implementation. Ferruh, here is an example: We have a thread in the application that processes KNI commands from the kernel. It receives config_network_if command to set interface up, calls rte_eth_dev_start, and here is the problem. We cannot call current rte_kni_update_link from here as the interface is not yet up in the kernel, as we didn't send a response for config_network_if yet. So we need to send a response first and only after that, we can use rte_kni_update_link. Actually, we don't even know the exact time between we send a response and the moment when the kernel receives it and the interface becomes up. We always have a dependency on the interface state in the kernel. With ioctl approach, we don't have such dependency - we can call rte_kni_update_link whenever we want, even when the interface is down in the kernel. As I explained, it's common when processing config_network_if to set interface up. Igor On Mon, Oct 14, 2019 at 11:56 PM Dan Gora <dg@adax.com> wrote: > Here's another link to the thread where this was discussed last year.. > Igor was actually on this thread as well... > > https://mails.dpdk.org/archives/dev/2018-August/110383.html > > On Mon, Oct 14, 2019 at 4:01 PM Dan Gora <dg@adax.com> wrote: > > > > My original patch to add this feature was basically the same thing as > > this: setting the link status via a KNI ioctl. That method was > > rejected after _much_ discussion and we eventually settled on the > > currently implementation. > > > > My original patch was here: Message-Id: < > 20180628225548.21885-1-dg@adax.com> > > > > If you search for KNI and dg@adax.com in the DPDK devel list you > > should be able to suss out the whole discussion that lead to the > > current implementation. > > > > thanks > > dan > > > > On Mon, Oct 14, 2019 at 1:17 PM Ferruh Yigit <ferruh.yigit@intel.com> > wrote: > > > > > > On 10/14/2019 5:10 PM, Ferruh Yigit wrote: > > > > On 9/25/2019 10:36 AM, Igor Ryzhov wrote: > > > >> Current implementation doesn't allow us to update KNI carrier if the > > > >> interface is not yet UP in kernel. It means that we can't use it in > the > > > >> same thread which is processing rte_kni_ops.config_network_if, > which is > > > >> very convenient, because it allows us to have correct carrier status > > > >> of the interface right after we enabled it and we don't have to use > any > > > >> additional thread to track link status. > > > > > > > > Hi Igor, > > > > > > > > The existing thread tracks the link status of the physical device > and reflects > > > > the changes to the kni netdev, but the "struct rte_kni_ops" > > > > (rte_kni_ops.config_network_if) works other way around, it captures > (some) > > > > requests to kni netdev and reflects them to the underlying physical > device. > > > > Even 'rte_kni_update_link()' updated to use ioctl, the thread still > looks > > > > required and this patch doesn't really changes that part. > > > > > > > > Also I am reluctant to extend the KNI ioctl interface when there is > a generic > > > > way to do that work. > > > > > > > > What is the use case of updating kni netdev carrier status when the > interface is > > > > down? > > > > > > btw, if the problem is status of the interface being 'no-carrier' by > default, > > > this can be changed by "carrier=on" parameter of the kni kernel module: > > > "insmod ./build/kmod/rte_kni.ko carrier=on" >
diff --git a/kernel/linux/kni/compat.h b/kernel/linux/kni/compat.h index fe0ee55e7..e0a491bcd 100644 --- a/kernel/linux/kni/compat.h +++ b/kernel/linux/kni/compat.h @@ -61,10 +61,6 @@ #define kni_sock_map_fd(s) sock_map_fd(s, 0) #endif -#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 9, 0) -#define HAVE_CHANGE_CARRIER_CB -#endif - #if LINUX_VERSION_CODE < KERNEL_VERSION(3, 14, 0) #define ether_addr_copy(dst, src) memcpy(dst, src, ETH_ALEN) #endif diff --git a/kernel/linux/kni/kni_misc.c b/kernel/linux/kni/kni_misc.c index aeb275329..9bcba68c8 100644 --- a/kernel/linux/kni/kni_misc.c +++ b/kernel/linux/kni/kni_misc.c @@ -462,6 +462,45 @@ kni_ioctl_release(struct net *net, uint32_t ioctl_num, return ret; } +static int +kni_ioctl_link(struct net *net, uint32_t ioctl_num, + unsigned long ioctl_param) +{ + struct kni_net *knet = net_generic(net, kni_net_id); + int ret = -EINVAL; + struct kni_dev *dev, *n; + struct rte_kni_link_info link_info; + struct net_device *netdev; + + if (_IOC_SIZE(ioctl_num) > sizeof(link_info)) + return -EINVAL; + + if (copy_from_user(&link_info, (void *)ioctl_param, sizeof(link_info))) + return -EFAULT; + + if (strlen(link_info.name) == 0) + return -EINVAL; + + down_read(&knet->kni_list_lock); + list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { + if (strncmp(dev->name, link_info.name, RTE_KNI_NAMESIZE) != 0) + continue; + + netdev = dev->net_dev; + + if (link_info.linkup) + netif_carrier_on(netdev); + else + netif_carrier_off(netdev); + + ret = 0; + break; + } + up_read(&knet->kni_list_lock); + + return ret; +} + static int kni_ioctl(struct inode *inode, uint32_t ioctl_num, unsigned long ioctl_param) { @@ -483,6 +522,9 @@ kni_ioctl(struct inode *inode, uint32_t ioctl_num, unsigned long ioctl_param) case _IOC_NR(RTE_KNI_IOCTL_RELEASE): ret = kni_ioctl_release(net, ioctl_num, ioctl_param); break; + case _IOC_NR(RTE_KNI_IOCTL_LINK): + ret = kni_ioctl_link(net, ioctl_num, ioctl_param); + break; default: pr_debug("IOCTL default\n"); break; diff --git a/kernel/linux/kni/kni_net.c b/kernel/linux/kni/kni_net.c index 7bd3a9f1e..cd852eea3 100644 --- a/kernel/linux/kni/kni_net.c +++ b/kernel/linux/kni/kni_net.c @@ -706,18 +706,6 @@ kni_net_set_mac(struct net_device *netdev, void *p) return (ret == 0 ? req.result : ret); } -#ifdef HAVE_CHANGE_CARRIER_CB -static int -kni_net_change_carrier(struct net_device *dev, bool new_carrier) -{ - if (new_carrier) - netif_carrier_on(dev); - else - netif_carrier_off(dev); - return 0; -} -#endif - static const struct header_ops kni_net_header_ops = { .create = kni_net_header, .parse = eth_header_parse, @@ -736,9 +724,6 @@ static const struct net_device_ops kni_net_netdev_ops = { .ndo_change_mtu = kni_net_change_mtu, .ndo_tx_timeout = kni_net_tx_timeout, .ndo_set_mac_address = kni_net_set_mac, -#ifdef HAVE_CHANGE_CARRIER_CB - .ndo_change_carrier = kni_net_change_carrier, -#endif }; static void kni_get_drvinfo(struct net_device *dev, diff --git a/lib/librte_eal/linux/eal/include/rte_kni_common.h b/lib/librte_eal/linux/eal/include/rte_kni_common.h index 70992d835..07a10dd93 100644 --- a/lib/librte_eal/linux/eal/include/rte_kni_common.h +++ b/lib/librte_eal/linux/eal/include/rte_kni_common.h @@ -125,10 +125,16 @@ struct rte_kni_device_info { uint8_t mac_addr[6]; }; +struct rte_kni_link_info { + char name[RTE_KNI_NAMESIZE]; + unsigned int linkup; +}; + #define KNI_DEVICE "kni" #define RTE_KNI_IOCTL_TEST _IOWR(0, 1, int) #define RTE_KNI_IOCTL_CREATE _IOWR(0, 2, struct rte_kni_device_info) #define RTE_KNI_IOCTL_RELEASE _IOWR(0, 3, struct rte_kni_device_info) +#define RTE_KNI_IOCTL_LINK _IOWR(0, 4, struct rte_kni_link_info) #endif /* _RTE_KNI_COMMON_H_ */ diff --git a/lib/librte_kni/rte_kni.c b/lib/librte_kni/rte_kni.c index 521db27c4..f80b97e2f 100644 --- a/lib/librte_kni/rte_kni.c +++ b/lib/librte_kni/rte_kni.c @@ -733,41 +733,20 @@ rte_kni_unregister_handlers(struct rte_kni *kni) int rte_kni_update_link(struct rte_kni *kni, unsigned int linkup) { - char path[64]; - char old_carrier[2]; - const char *new_carrier; - int old_linkup; - int fd, ret; + struct rte_kni_link_info link_info; if (kni == NULL) return -1; - snprintf(path, sizeof(path), "/sys/devices/virtual/net/%s/carrier", - kni->name); + snprintf(link_info.name, RTE_KNI_NAMESIZE, "%s", kni->name); + link_info.linkup = linkup; - fd = open(path, O_RDWR); - if (fd == -1) { - RTE_LOG(ERR, KNI, "Failed to open file: %s.\n", path); + if (ioctl(kni_fd, RTE_KNI_IOCTL_LINK, &link_info) < 0) { + RTE_LOG(ERR, KNI, "Fail to update KNI link\n"); return -1; } - ret = read(fd, old_carrier, 2); - if (ret < 1) { - close(fd); - return -1; - } - old_linkup = (old_carrier[0] == '1'); - - new_carrier = linkup ? "1" : "0"; - ret = write(fd, new_carrier, 1); - if (ret < 1) { - RTE_LOG(ERR, KNI, "Failed to write file: %s.\n", path); - close(fd); - return -1; - } - - close(fd); - return old_linkup; + return 0; } void diff --git a/lib/librte_kni/rte_kni.h b/lib/librte_kni/rte_kni.h index b22446fa7..7e1cce1e0 100644 --- a/lib/librte_kni/rte_kni.h +++ b/lib/librte_kni/rte_kni.h @@ -247,9 +247,8 @@ int rte_kni_unregister_handlers(struct rte_kni *kni); * > 0 for linkup. * * @return + * On success: 0 * On failure: -1 - * Previous link state == linkdown: 0 - * Previous link state == linkup: 1 */ __rte_experimental int
Current implementation doesn't allow us to update KNI carrier if the interface is not yet UP in kernel. It means that we can't use it in the same thread which is processing rte_kni_ops.config_network_if, which is very convenient, because it allows us to have correct carrier status of the interface right after we enabled it and we don't have to use any additional thread to track link status. Signed-off-by: Igor Ryzhov <iryzhov@nfware.com> --- v3: remove unused variables v2: fix checkpatch warnings kernel/linux/kni/compat.h | 4 -- kernel/linux/kni/kni_misc.c | 42 +++++++++++++++++++ kernel/linux/kni/kni_net.c | 15 ------- .../linux/eal/include/rte_kni_common.h | 6 +++ lib/librte_kni/rte_kni.c | 33 +++------------ lib/librte_kni/rte_kni.h | 3 +- 6 files changed, 55 insertions(+), 48 deletions(-)