net: make sure we retain NAPI ordering on netdev->napi_list
Subsystems (5)
Jakub Kicinski
maintainerv2 cover net: make sure we retain NAPI ordering on netdev->napi_list
I promised Eric to remove the rtnl protection of the NAPI list, when I sat down to implement it over the break I realized that the recently added NAPI ID retention will break the list ordering assumption we have in netlink dump. The ordering used to happen "naturally", because we'd always add NAPIs that the head of the list, and assign a new monotonically increasing ID.
Before the first patch of this series we'd still only add at the head of the list but now the newly added NAPI may inherit from its config an ID lower than something else already on the list.
The fix is in the first patch, the rest is netdevsim churn to test it. I'm posting this for net-next, because AFAICT the problem can't be triggered in net, given the very limited queue API adoption.
v2:
- [patch 2] allocate the array with kcalloc() instead of kvcalloc()
- [patch 2] set GFP_KERNEL_ACCOUNT when allocating queues
- [patch 6] don't null-check page pool before page_pool_destroy()
- [patch 6] controled -> controlled
- [patch 7] change mode to 0200
- [patch 7] reorder removal to be inverse of add
- [patch 7] fix the spaces vs tabs
v1: https://lore.kernel.org/[email protected]
Jakub Kicinski (8):
net: make sure we retain NAPI ordering on netdev->napi_list
netdev: define NETDEV_INTERNAL
netdevsim: support NAPI config
netdevsim: allocate rqs individually
netdevsim: add queue alloc/free helpers
netdevsim: add queue management API support
netdevsim: add debugfs-triggered queue reset
selftests: net: test listing NAPI vs queue resets
Documentation/networking/netdevices.rst | 10 + drivers/net/netdevsim/netdev.c | 259 ++++++++++++++++++++--- drivers/net/netdevsim/netdevsim.h | 5 +- net/core/dev.c | 42 +++- net/core/netdev_rx_queue.c | 1 + tools/testing/selftests/net/nl_netdev.py | 19 +- 6 files changed, 294 insertions(+), 42 deletions(-)
Signature · 1 line
2.47.1
Jakub Kicinski
maintainerPATCH v2 1/8 net: make sure we retain NAPI ordering on netdev->napi_list
Netlink code depends on NAPI instances being sorted by ID on the netdev list for dump continuation. We need to be able to find the position on the list where we left off if dump does not fit in a single skb, and in the meantime NAPI instances can come and go.
This was trivially true when we were assigning a new ID to every new NAPI instance. Since we added the NAPI config API, we try to retain the ID previously used for the same queue, but still add the new NAPI instance at the start of the list.
This is fine if we reset the entire netdev and all NAPIs get removed and added back. If driver replaces a NAPI instance during an operation like DEVMEM queue reset, or recreates a subset of NAPI instances in other ways we may end up with broken ordering, and therefore Netlink dumps with either missing or duplicated entries.
At this stage the problem is theoretical. Only two drivers support queue API, bnxt and gve. gve recreates NAPIs during queue reset, but it doesn't support NAPI config. bnxt supports NAPI config but doesn't recreate instances during reset.
We need to save the ID in the config as soon as it is assigned because otherwise the new NAPI will not know what ID it will get at enable time, at the time it is being added.
--- net/core/dev.c | 42 ++++++++++++++++++++++++++++++++++++------ 1 file changed, 36 insertions(+), 6 deletions(-)
net/core/dev.c +36 -6
@@ -6713,13 +6713,14 @@ static void napi_restore_config(struct napi_struct *n) n->gro_flush_timeout = n->config->gro_flush_timeout; n->irq_suspend_timeout = n->config->irq_suspend_timeout;/* a NAPI ID might be stored in the config, if so use it. if not, use -* napi_hash_add to generate one for us. It will be saved to the config -* in napi_disable. +* napi_hash_add to generate one for us. */ -if (n->config->napi_id) +if (n->config->napi_id) {napi_hash_add_with_id (n, n->config->napi_id); -else + }else {napi_hash_add (n); + n->config->napi_id = n->napi_id; + } }static void napi_save_config (struct napi_struct *n)
@@ -6727,10 +6728,39 @@ static void napi_save_config(struct napi_struct *n) n->config->defer_hard_irqs = n->defer_hard_irqs; n->config->gro_flush_timeout = n->gro_flush_timeout; n->config->irq_suspend_timeout = n->irq_suspend_timeout; - n->config->napi_id = n->napi_id;napi_hash_del (n); } +/* Netlink wants the NAPI list to be sorted by ID, if adding a NAPI which will +* inherit an existing ID try to insert it at the right position. +*/ +static void +netif_napi_dev_list_add (struct net_device *dev,struct napi_struct *napi) +{ +unsigned int new_id, pos_id; +struct list_head *higher; +struct napi_struct *pos; + + new_id =UINT_MAX ; +if (napi->config && napi->config->napi_id) + new_id = napi->config->napi_id; + + higher = &dev->napi_list; +list_for_each_entry (pos, &dev->napi_list, dev_list) { +if (pos->napi_id >=MIN_NAPI_ID ) + pos_id = pos->napi_id; +else if (pos->config) + pos_id = pos->config->napi_id; +else + pos_id =UINT_MAX ; + +if (pos_id <= new_id) +break ; + higher = &pos->dev_list; + } +list_add_rcu (&napi->dev_list, higher);/* adds after higher */ +} +void netif_napi_add_weight (struct net_device *dev,struct napi_struct *napi,int (*poll)(struct napi_struct *,int ),int weight) {
@@ -6757,7 +6787,7 @@ void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi, napi->list_owner =-1 ;set_bit (NAPI_STATE_SCHED , &napi->state);set_bit (NAPI_STATE_NPSVC , &napi->state); -list_add_rcu (&napi->dev_list, &dev->napi_list); +netif_napi_dev_list_add (dev, napi);/* default settings from sysfs are applied to all NAPIs. any per-NAPI * configuration will be loaded in napi_enable
Signature · 1 line
2.47.1
Jakub Kicinski
maintainerPATCH v2 2/8 netdev: define NETDEV_INTERNAL
In reply to Jakub KicinskiLinus suggested during one of past maintainer summits (in context of a DMA_BUF discussion) that symbol namespaces can be used to prevent unwelcome but in-tree code from using all exported functions. Create a namespace for netdev.
Export netdev_rx_queue_restart(), drivers may want to use it since it gives them a simple and safe way to restart a queue to apply config changes. But it's both too low level and too actively developed to be used outside netdev.
--- Documentation/networking/netdevices.rst | 10 ++++++++++ net/core/netdev_rx_queue.c | 1 + 2 files changed, 11 insertions(+)
Documentation/networking/netdevices.rst +10 -0
@@ -297,3 +297,13 @@ struct napi_struct synchronization rules Context: softirq will be called with interrupts disabled by netconsole. + +NETDEV_INTERNAL symbol namespace +================================ + +Symbols exported as NETDEV_INTERNAL can only be used in networking +core and drivers which exclusively flow via the main networking list and trees. +Note that the inverse is not true, most symbols outside of NETDEV_INTERNAL +are not expected to be used by random code outside netdev either. +Symbols may lack the designation because they predate the namespaces, +or simply due to an oversight.
net/core/netdev_rx_queue.c +1 -0
@@ -79,3 +79,4 @@ int netdev_rx_queue_restart(struct net_device *dev, unsigned int rxq_idx)return err; } +EXPORT_SYMBOL_NS_GPL (netdev_rx_queue_restart,"NETDEV_INTERNAL" );
Signature · 1 line
2.47.1
Jakub Kicinski
maintainerPATCH v2 3/8 netdevsim: support NAPI config
In reply to Jakub KicinskiLink the NAPI instances to their configs. This will be needed to test that NAPI config doesn't break list ordering.
--- drivers/net/netdevsim/netdev.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/netdevsim/netdev.c +1 -1
@@ -390,7 +390,7 @@ static int nsim_init_napi(struct netdevsim *ns)for (i =0 ; i < dev->num_rx_queues; i++) { rq = &ns->rq[i]; -netif_napi_add (dev, &rq->napi, nsim_poll); +netif_napi_add_config (dev, &rq->napi, nsim_poll, i); }for (i =0 ; i < dev->num_rx_queues; i++) {
Signature · 1 line
2.47.1
Jakub Kicinski
maintainerPATCH v2 4/8 netdevsim: allocate rqs individually
In reply to Jakub KicinskiMake nsim->rqs an array of pointers and allocate them individually so that we can swap them out one by one.
---
v2:
- allocate the array with kcalloc() instead of kvcalloc()
- set GFP_KERNEL_ACCOUNT when allocating queues
--- drivers/net/netdevsim/netdev.c | 49 +++++++++++++++++++------------ drivers/net/netdevsim/netdevsim.h | 2 +- 2 files changed, 32 insertions(+), 19 deletions(-)
drivers/net/netdevsim/netdev.c +31 -18
@@ -69,7 +69,7 @@ static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev) rxq =skb_get_queue_mapping (skb);if (rxq >= peer_dev->num_rx_queues) rxq = rxq % peer_dev->num_rx_queues; - rq = &peer_ns->rq[rxq]; + rq = peer_ns->rq[rxq];skb_tx_timestamp (skb);if (unlikely (nsim_forward_skb (peer_dev, skb, rq) ==NET_RX_DROP ))
@@ -388,13 +388,13 @@ static int nsim_init_napi(struct netdevsim *ns)int err, i;for (i =0 ; i < dev->num_rx_queues; i++) { - rq = &ns->rq[i]; + rq = ns->rq[i];netif_napi_add_config (dev, &rq->napi, nsim_poll, i); }for (i =0 ; i < dev->num_rx_queues; i++) { - rq = &ns->rq[i]; + rq = ns->rq[i]; err =nsim_create_page_pool (rq);if (err)
@@ -405,12 +405,12 @@ static int nsim_init_napi(struct netdevsim *ns) err_pp_destroy:while (i--) { -page_pool_destroy (ns->rq[i].page_pool); - ns->rq[i].page_pool =NULL ; +page_pool_destroy (ns->rq[i]->page_pool); + ns->rq[i]->page_pool =NULL ; }for (i =0 ; i < dev->num_rx_queues; i++) -__netif_napi_del (&ns->rq[i].napi); +__netif_napi_del (&ns->rq[i]->napi);return err; }
@@ -421,7 +421,7 @@ static void nsim_enable_napi(struct netdevsim *ns)int i;for (i =0 ; i < dev->num_rx_queues; i++) { -struct nsim_rq *rq = &ns->rq[i]; +struct nsim_rq *rq = ns->rq[i];netif_queue_set_napi (dev, i,NETDEV_QUEUE_TYPE_RX , &rq->napi);napi_enable (&rq->napi);
@@ -448,7 +448,7 @@ static void nsim_del_napi(struct netdevsim *ns)int i;for (i =0 ; i < dev->num_rx_queues; i++) { -struct nsim_rq *rq = &ns->rq[i]; +struct nsim_rq *rq = ns->rq[i];napi_disable (&rq->napi);__netif_napi_del (&rq->napi);
@@ -456,8 +456,8 @@ static void nsim_del_napi(struct netdevsim *ns)synchronize_net ();for (i =0 ; i < dev->num_rx_queues; i++) { -page_pool_destroy (ns->rq[i].page_pool); - ns->rq[i].page_pool =NULL ; +page_pool_destroy (ns->rq[i]->page_pool); + ns->rq[i]->page_pool =NULL ; } }
@@ -628,7 +628,7 @@ nsim_pp_hold_write(struct file *file, const char __user *data,if (!netif_running (ns->netdev) && val) { ret = -ENETDOWN ; }else if (val) { - ns->page =page_pool_dev_alloc_pages (ns->rq[0 ].page_pool); + ns->page =page_pool_dev_alloc_pages (ns->rq[0 ]->page_pool);if (!ns->page) ret = -ENOMEM ; }else {
@@ -677,15 +677,26 @@ static int nsim_queue_init(struct netdevsim *ns)struct net_device *dev = ns->netdev;int i; - ns->rq =kvcalloc (dev->num_rx_queues,sizeof (*ns->rq), -GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL); + ns->rq =kcalloc (dev->num_rx_queues,sizeof (*ns->rq), +GFP_KERNEL_ACCOUNT );if (!ns->rq)return -ENOMEM ; -for (i =0 ; i < dev->num_rx_queues; i++) -skb_queue_head_init (&ns->rq[i].skb_queue); +for (i =0 ; i < dev->num_rx_queues; i++) { + ns->rq[i] =kzalloc (sizeof (**ns->rq),GFP_KERNEL_ACCOUNT ); +if (!ns->rq[i]) + goto err_free_prev; + +skb_queue_head_init (&ns->rq[i]->skb_queue); + }return 0 ; + +err_free_prev: +while (i--) +kfree (ns->rq[i]); +kfree (ns->rq); +return -ENOMEM ; }static void nsim_queue_free (struct netdevsim *ns)
@@ -693,11 +704,13 @@ static void nsim_queue_free(struct netdevsim *ns)struct net_device *dev = ns->netdev;int i; -for (i =0 ; i < dev->num_rx_queues; i++) -skb_queue_purge_reason (&ns->rq[i].skb_queue, +for (i =0 ; i < dev->num_rx_queues; i++) { +skb_queue_purge_reason (&ns->rq[i]->skb_queue,SKB_DROP_REASON_QUEUE_PURGE ); +kfree (ns->rq[i]); + } -kvfree (ns->rq); +kfree (ns->rq); ns->rq =NULL ; }
drivers/net/netdevsim/netdevsim.h +1 -1
@@ -101,7 +101,7 @@ struct netdevsim {struct nsim_dev *nsim_dev;struct nsim_dev_port *nsim_dev_port;struct mock_phc *phc; -struct nsim_rq *rq; +struct nsim_rq **rq;u64 tx_packets;u64 tx_bytes;
Signature · 1 line
2.47.1
Jakub Kicinski
maintainerPATCH v2 5/8 netdevsim: add queue alloc/free helpers
In reply to Jakub KicinskiWe'll need the code to allocate and free queues in the queue management API, factor it out.
--- drivers/net/netdevsim/netdev.c | 35 +++++++++++++++++++++++----------- 1 file changed, 24 insertions(+), 11 deletions(-)
drivers/net/netdevsim/netdev.c +24 -11
@@ -595,6 +595,24 @@ static const struct netdev_stat_ops nsim_stat_ops = { .get_base_stats = nsim_get_base_stats, }; +static struct nsim_rq *nsim_queue_alloc (void ) +{ +struct nsim_rq *rq; + + rq =kzalloc (sizeof (*rq),GFP_KERNEL_ACCOUNT ); +if (!rq) +return NULL ; + +skb_queue_head_init (&rq->skb_queue); +return rq; +} + +static void nsim_queue_free (struct nsim_rq *rq) +{ +skb_queue_purge_reason (&rq->skb_queue,SKB_DROP_REASON_QUEUE_PURGE ); +kfree (rq); +} +static ssize_t nsim_pp_hold_read (struct file *file,char __user *data,size_t count,loff_t *ppos)
@@ -683,11 +701,9 @@ static int nsim_queue_init(struct netdevsim *ns)return -ENOMEM ;for (i =0 ; i < dev->num_rx_queues; i++) { - ns->rq[i] =kzalloc (sizeof (**ns->rq),GFP_KERNEL_ACCOUNT ); + ns->rq[i] =nsim_queue_alloc ();if (!ns->rq[i]) goto err_free_prev; - -skb_queue_head_init (&ns->rq[i]->skb_queue); }return 0 ;
@@ -699,16 +715,13 @@ static int nsim_queue_init(struct netdevsim *ns)return -ENOMEM ; } -static void nsim_queue_free (struct netdevsim *ns) +static void nsim_queue_uninit (struct netdevsim *ns) {struct net_device *dev = ns->netdev;int i; -for (i =0 ; i < dev->num_rx_queues; i++) { -skb_queue_purge_reason (&ns->rq[i]->skb_queue, -SKB_DROP_REASON_QUEUE_PURGE ); -kfree (ns->rq[i]); - } +for (i =0 ; i < dev->num_rx_queues; i++) +nsim_queue_free (ns->rq[i]);kfree (ns->rq); ns->rq =NULL ;
@@ -754,7 +767,7 @@ static int nsim_init_netdevsim(struct netdevsim *ns)nsim_macsec_teardown (ns);nsim_bpf_uninit (ns); err_rq_destroy: -nsim_queue_free (ns); +nsim_queue_uninit (ns); err_utn_destroy:rtnl_unlock ();nsim_udp_tunnels_info_destroy (ns->netdev);
@@ -836,7 +849,7 @@ void nsim_destroy(struct netdevsim *ns)nsim_macsec_teardown (ns);nsim_ipsec_teardown (ns);nsim_bpf_uninit (ns); -nsim_queue_free (ns); +nsim_queue_uninit (ns); }rtnl_unlock ();if (nsim_dev_port_is_pf (ns->nsim_dev_port))
Signature · 1 line
2.47.1
Jakub Kicinski
maintainerPATCH v2 6/8 netdevsim: add queue management API support
In reply to Jakub KicinskiAdd queue management API support. We need a way to reset queues to test NAPI reordering, the queue management API provides a handy scaffolding for that.
---
v2:
- don't null-check page pool before page_pool_destroy()
- controled -> controlled
--- drivers/net/netdevsim/netdev.c | 134 +++++++++++++++++++++++++++--- drivers/net/netdevsim/netdevsim.h | 2 + 2 files changed, 124 insertions(+), 12 deletions(-)
drivers/net/netdevsim/netdev.c +122 -12
@@ -359,25 +359,24 @@ static int nsim_poll(struct napi_struct *napi, int budget)return done; } -static int nsim_create_page_pool (struct nsim_rq *rq) +static int nsim_create_page_pool (struct page_pool **p,struct napi_struct *napi) { -struct page_pool_params p = { +struct page_pool_params params = { .order =0 , .pool_size =NSIM_RING_SIZE , .nid =NUMA_NO_NODE , - .dev = &rq->napi.dev->dev, - .napi = &rq->napi, + .dev = &napi->dev->dev, + .napi = napi, .dma_dir =DMA_BIDIRECTIONAL , - .netdev = rq->napi.dev, + .netdev = napi->dev, }; +struct page_pool *pool; - rq->page_pool =page_pool_create (&p); -if (IS_ERR (rq->page_pool)) { -int err =PTR_ERR (rq->page_pool); + pool =page_pool_create (¶ms); +if (IS_ERR (pool)) +return PTR_ERR (pool); - rq->page_pool =NULL ; -return err; - } + *p = pool;return 0 ; }
@@ -396,7 +395,7 @@ static int nsim_init_napi(struct netdevsim *ns)for (i =0 ; i < dev->num_rx_queues; i++) { rq = ns->rq[i]; - err =nsim_create_page_pool (rq); + err =nsim_create_page_pool (&rq->page_pool, &rq->napi);if (err) goto err_pp_destroy; }
@@ -613,6 +612,116 @@ static void nsim_queue_free(struct nsim_rq *rq)kfree (rq); } +/* Queue reset mode is controlled by ns->rq_reset_mode. +* - normal - new NAPI new pool (old NAPI enabled when new added) +* - mode 1 - allocate new pool (NAPI is only disabled / enabled) +* - mode 2 - new NAPI new pool (old NAPI removed before new added) +* - mode 3 - new NAPI new pool (old NAPI disabled when new added) +*/ +struct nsim_queue_mem { +struct nsim_rq *rq; +struct page_pool *pp; +}; + +static int +nsim_queue_mem_alloc (struct net_device *dev,void *per_queue_mem,int idx) +{ +struct nsim_queue_mem *qmem = per_queue_mem; +struct netdevsim *ns =netdev_priv (dev); +int err; + +if (ns->rq_reset_mode >3 ) +return -EINVAL ; + +if (ns->rq_reset_mode ==1 ) +return nsim_create_page_pool (&qmem->pp, &ns->rq[idx]->napi); + + qmem->rq =nsim_queue_alloc (); +if (!qmem->rq) +return -ENOMEM ; + + err =nsim_create_page_pool (&qmem->rq->page_pool, &qmem->rq->napi); +if (err) + goto err_free; + +if (!ns->rq_reset_mode) +netif_napi_add_config (dev, &qmem->rq->napi, nsim_poll, idx); + +return 0 ; + +err_free: +nsim_queue_free (qmem->rq); +return err; +} + +static void nsim_queue_mem_free (struct net_device *dev,void *per_queue_mem) +{ +struct nsim_queue_mem *qmem = per_queue_mem; +struct netdevsim *ns =netdev_priv (dev); + +page_pool_destroy (qmem->pp); +if (qmem->rq) { +if (!ns->rq_reset_mode) +netif_napi_del (&qmem->rq->napi); +page_pool_destroy (qmem->rq->page_pool); +nsim_queue_free (qmem->rq); + } +} + +static int +nsim_queue_start (struct net_device *dev,void *per_queue_mem,int idx) +{ +struct nsim_queue_mem *qmem = per_queue_mem; +struct netdevsim *ns =netdev_priv (dev); + +if (ns->rq_reset_mode ==1 ) { + ns->rq[idx]->page_pool = qmem->pp; +napi_enable (&ns->rq[idx]->napi); +return 0 ; + } + +/* netif_napi_add()/_del() should normally be called from alloc/free, +* here we want to test various call orders. +*/ +if (ns->rq_reset_mode ==2 ) { +netif_napi_del (&ns->rq[idx]->napi); +netif_napi_add_config (dev, &qmem->rq->napi, nsim_poll, idx); + }else if (ns->rq_reset_mode ==3 ) { +netif_napi_add_config (dev, &qmem->rq->napi, nsim_poll, idx); +netif_napi_del (&ns->rq[idx]->napi); + } + + ns->rq[idx] = qmem->rq; +napi_enable (&ns->rq[idx]->napi); + +return 0 ; +} + +static int nsim_queue_stop (struct net_device *dev,void *per_queue_mem,int idx) +{ +struct nsim_queue_mem *qmem = per_queue_mem; +struct netdevsim *ns =netdev_priv (dev); + +napi_disable (&ns->rq[idx]->napi); + +if (ns->rq_reset_mode ==1 ) { + qmem->pp = ns->rq[idx]->page_pool; +page_pool_disable_direct_recycling (qmem->pp); + }else { + qmem->rq = ns->rq[idx]; + } + +return 0 ; +} + +static const struct netdev_queue_mgmt_ops nsim_queue_mgmt_ops = { + .ndo_queue_mem_size =sizeof (struct nsim_queue_mem ), + .ndo_queue_mem_alloc = nsim_queue_mem_alloc, + .ndo_queue_mem_free = nsim_queue_mem_free, + .ndo_queue_start = nsim_queue_start, + .ndo_queue_stop = nsim_queue_stop, +}; +static ssize_t nsim_pp_hold_read (struct file *file,char __user *data,size_t count,loff_t *ppos)
@@ -739,6 +848,7 @@ static int nsim_init_netdevsim(struct netdevsim *ns) ns->phc = phc; ns->netdev->netdev_ops = &nsim_netdev_ops; ns->netdev->stat_ops = &nsim_stat_ops; + ns->netdev->queue_mgmt_ops = &nsim_queue_mgmt_ops; err =nsim_udp_tunnels_info_create (ns->nsim_dev, ns->netdev);if (err)
drivers/net/netdevsim/netdevsim.h +2 -0
@@ -103,6 +103,8 @@ struct netdevsim {struct mock_phc *phc;struct nsim_rq **rq; +int rq_reset_mode; +u64 tx_packets;u64 tx_bytes;u64 tx_dropped;
Signature · 1 line
2.47.1
Re: [PATCH net-next v2 6/8] netdevsim: add queue management API support
Quoted 8 lines · Jakub Kicinski
Add queue management API support. We need a way to reset queues to test NAPI reordering, the queue management API provides a handy scaffolding for that.
Reviewed-by Willem de Bruijn <[email protected]>Signed-off-by Jakub Kicinski <[email protected]>
Quoted 59 lines
---
v2:
- don't null-check page pool beforepage_pool_destroy()
- controled -> controlled--- drivers/net/netdevsim/netdev.c | 134 +++++++++++++++++++++++++++--- drivers/net/netdevsim/netdevsim.h | 2 + 2 files changed, 124 insertions(+), 12 deletions(-)drivers/net/netdevsim/netdev.c +11
-12@@ -359,25 +359,24 @@ static int nsim_poll(struct napi_struct *napi, int budget)return done; } -static int nsim_create_page_pool (struct nsim_rq *rq) +static int nsim_create_page_pool (struct page_pool **p,struct napi_struct *napi) { -struct page_pool_params p = { +struct page_pool_params params = { .order =0 , .pool_size =NSIM_RING_SIZE , .nid =NUMA_NO_NODE , - .dev = &rq->napi.dev->dev, - .napi = &rq->napi, + .dev = &napi->dev->dev, + .napi = napi, .dma_dir =DMA_BIDIRECTIONAL , - .netdev = rq->napi.dev, + .netdev = napi->dev, }; +struct page_pool *pool; - rq->page_pool =page_pool_create (&p); -if (IS_ERR (rq->page_pool)) { -int err =PTR_ERR (rq->page_pool); + pool =page_pool_create (¶ms); +if (IS_ERR (pool)) +return PTR_ERR (pool); - rq->page_pool =NULL ; -return err; - } + *p = pool;return 0 ; }@@ -396,7 +395,7 @@ static int nsim_init_napi(struct netdevsim *ns)for (i =0 ; i < dev->num_rx_queues; i++) { rq = ns->rq[i]; - err =nsim_create_page_pool (rq); + err =nsim_create_page_pool (&rq->page_pool, &rq->napi);if (err) goto err_pp_destroy; }
drivers/net/netdevsim/netdev.c static void nsim_queue_free(struct nsim_rq *rq) · line 616
Quoted 5 lines
drivers/net/netdevsim/netdev.c +2
-0@@ -613,6 +612,116 @@ static void nsim_queue_free(struct nsim_rq *rq)kfree (rq); } +/* Queue reset mode is controlled by ns->rq_reset_mode. +* - normal - new NAPI new pool (old NAPI enabled when new added)
Nit, probably not worth a respin: Normal seems to me to delete old napi after the new one is added and enabled.
queue stop -> napi_disable(old)
queue alloc -> netif_napi_add_config(new)
queue start -> napi_enable(new)
queue free -> netif_napi_del(old)
drivers/net/netdevsim/netdev.c static void nsim_queue_free(struct nsim_rq *rq) · line 620
Quoted 4 lines
drivers/net/netdevsim/netdev.c +4
-0+* - mode 1 - allocate new pool (NAPI is only disabled / enabled) +* - mode 2 - new NAPI new pool (old NAPI removed before new added) +* - mode 3 - new NAPI new pool (old NAPI disabled when new added) +*/
Which modes are 'correct' for a driver to implement? 2/3 is for testing only, as you note in the code, the add/del functions should really be called from alloc/free and not from queue_start. I assume modes normal and 1 are both correct implementations of the queue API and the driver gets to pick whether to reuse the napi instance or not? Asking because IIRC GVE implements mode 1, not what you consider 'normal'.
--
Thanks,
Mina
Jakub Kicinski
maintainerRe: [PATCH net-next v2 6/8] netdevsim: add queue management API support
Quoted 5 lines · On Tue, 7 Jan 2025 14:53:09 -0800 Mina Almasry wrote:
Quoted 2 lines
+/* Queue reset mode is controlled by
ns->rq_reset_mode.
+ * - normal - new NAPI new pool (old NAPI enabled when new added)Nit, probably not worth a respin: Normal seems to me to delete old napi after the new one is added and enabled.
Yes, the comment doesn't really focus on removal order, it's less important. I may be missing your point..
Quoted 4 lines
queue stop ->
napi_disable(old)
queue alloc ->netif_napi_add_config(new)
queue start ->napi_enable(new)
queue free ->netif_napi_del(old)
I think you have stop and alloc swapped here.
Quoted 5 lines
Quoted 4 lines
+ * - mode 1 - allocate new pool (NAPI is only disabled / enabled)
+ * - mode 2 - new NAPI new pool (old NAPI removed before new added)
+ * - mode 3 - new NAPI new pool (old NAPI disabled when new added)
+ */
Quoted 7 lines
Which modes are 'correct' for a driver to implement? 2/3 is for testing only, as you note in the code, the add/del functions should really be called from alloc/free and not from
queue_start. I assume modes normal and 1 are both correct implementations of the queue API and the driver gets to pick whether to reuse the napi instance or not? Asking because IIRC GVE implements mode 1, not what you consider 'normal'.
I can't think why any of them would be "incorrect" per se, but 2 and 3 are likely weird and unnatural. I thought mode 0 would simply be most common, maybe I shouldn't have called it "normal". But because of all the "modes" netdevsim implementation seemed too messy to be considered a blueprint..
Jakub Kicinski
maintainerPATCH v2 7/8 netdevsim: add debugfs-triggered queue reset
In reply to Jakub KicinskiSupport triggering queue reset via debugfs for an upcoming test.
---
v2:
- change mode to 0200
- reorder removal to be inverse of add
- fix the spaces vs tabs
--- drivers/net/netdevsim/netdev.c | 55 +++++++++++++++++++++++++++++++ drivers/net/netdevsim/netdevsim.h | 1 + 2 files changed, 56 insertions(+)
drivers/net/netdevsim/netdev.c +55 -0
@@ -20,6 +20,7 @@#include <linux/netdevice.h> #include <linux/slab.h> #include <net/netdev_queues.h> +#include <net/netdev_rx_queue.h> #include <net/page_pool/helpers.h> #include <net/netlink.h> #include <net/net_shaper.h>
@@ -29,6 +30,8 @@#include "netdevsim.h" +MODULE_IMPORT_NS ("NETDEV_INTERNAL" ); +#define NSIM_RING_SIZE 256static int nsim_napi_rx (struct nsim_rq *rq,struct sk_buff *skb)
@@ -722,6 +725,54 @@ static const struct netdev_queue_mgmt_ops nsim_queue_mgmt_ops = { .ndo_queue_stop = nsim_queue_stop, }; +static ssize_t +nsim_qreset_write (struct file *file,const char __user *data, +size_t count,loff_t *ppos) +{ +struct netdevsim *ns = file->private_data; +unsigned int queue, mode; +char buf[32 ]; +ssize_t ret; + +if (count >=sizeof (buf)) +return -EINVAL ; +if (copy_from_user (buf, data, count)) +return -EFAULT ; + buf[count] ='\0' ; + + ret =sscanf (buf,"%u %u" , &queue, &mode); +if (ret !=2 ) +return -EINVAL ; + +rtnl_lock (); +if (!netif_running (ns->netdev)) { + ret = -ENETDOWN ; + goto exit_unlock; + } + +if (queue >= ns->netdev->real_num_rx_queues) { + ret = -EINVAL ; + goto exit_unlock; + } + + ns->rq_reset_mode = mode; + ret =netdev_rx_queue_restart (ns->netdev, queue); + ns->rq_reset_mode =0 ; +if (ret) + goto exit_unlock; + + ret = count; +exit_unlock: +rtnl_unlock (); +return ret; +} + +static const struct file_operations nsim_qreset_fops = { + .open = simple_open, + .write = nsim_qreset_write, + .owner =THIS_MODULE , +}; +static ssize_t nsim_pp_hold_read (struct file *file,char __user *data,size_t count,loff_t *ppos)
@@ -934,6 +985,9 @@ nsim_create(struct nsim_dev *nsim_dev, struct nsim_dev_port *nsim_dev_port) ns->pp_dfs =debugfs_create_file ("pp_hold" ,0600 , nsim_dev_port->ddir, ns, &nsim_pp_hold_fops); + ns->qr_dfs =debugfs_create_file ("queue_reset" ,0200 , + nsim_dev_port->ddir, ns, + &nsim_qreset_fops);return ns;
@@ -947,6 +1001,7 @@ void nsim_destroy(struct netdevsim *ns)struct net_device *dev = ns->netdev;struct netdevsim *peer; +debugfs_remove (ns->qr_dfs);debugfs_remove (ns->pp_dfs);rtnl_lock ();
drivers/net/netdevsim/netdevsim.h +1 -0
@@ -136,6 +136,7 @@ struct netdevsim {struct page *page;struct dentry *pp_dfs; +struct dentry *qr_dfs;struct nsim_ethtool ethtool;struct netdevsim __rcu *peer;
Signature · 1 line
2.47.1
Re: [PATCH net-next v2 7/8] netdevsim: add debugfs-triggered queue reset
Quoted 6 lines · Jakub Kicinski
Support triggering queue reset via debugfs for an upcoming test.
Reviewed-by Willem de Bruijn <[email protected]>Signed-off-by Jakub Kicinski <[email protected]>
Quoted 54 lines
---
v2:
- change mode to 0200
- reorder removal to be inverse of add
- fix the spaces vs tabs--- drivers/net/netdevsim/netdev.c | 55 +++++++++++++++++++++++++++++++ drivers/net/netdevsim/netdevsim.h | 1 + 2 files changed, 56 insertions(+)drivers/net/netdevsim/netdev.c +22
-0@@ -20,6 +20,7 @@#include <linux/netdevice.h> #include <linux/slab.h> #include <net/netdev_queues.h> +#include <net/netdev_rx_queue.h> #include <net/page_pool/helpers.h> #include <net/netlink.h> #include <net/net_shaper.h> @@ -29,6 +30,8 @@#include "netdevsim.h" +MODULE_IMPORT_NS ("NETDEV_INTERNAL" ); +#define NSIM_RING_SIZE 256static int nsim_napi_rx (struct nsim_rq *rq,struct sk_buff *skb)@@ -722,6 +725,54 @@ static const struct netdev_queue_mgmt_ops nsim_queue_mgmt_ops = { .ndo_queue_stop = nsim_queue_stop, }; +static ssize_t +nsim_qreset_write (struct file *file,const char __user *data, +size_t count,loff_t *ppos) +{ +struct netdevsim *ns = file->private_data; +unsigned int queue, mode; +char buf[32 ]; +ssize_t ret; + +if (count >=sizeof (buf)) +return -EINVAL ; +if (copy_from_user (buf, data, count)) +return -EFAULT ; + buf[count] ='\0' ; + + ret =sscanf (buf,"%u %u" , &queue, &mode); +if (ret !=2 ) +return -EINVAL ; +
drivers/net/netdevsim/netdev.c static const struct netdev_queue_mgmt_ops nsim_queue_mgmt_ops = { · line 756
Quoted 10 lines
drivers/net/netdevsim/netdev.c +10
-0+rtnl_lock (); +if (!netif_running (ns->netdev)) { + ret = -ENETDOWN ; + goto exit_unlock; + } + +if (queue >= ns->netdev->real_num_rx_queues) { + ret = -EINVAL ; + goto exit_unlock; + }
Is it correct that both these above checks are not inside of netdev_rx_queue_restart() itself? Or should we fix that?
Signature · 2 lines
Thanks, Mina
Jakub Kicinski
maintainerRe: [PATCH net-next v2 7/8] netdevsim: add debugfs-triggered queue reset
Quoted 7 lines · On Tue, 7 Jan 2025 15:00:29 -0800 Mina Almasry wrote:
drivers/net/netdevsim/netdev.c static const struct netdev_queue_mgmt_ops nsim_queue_mgmt_ops = { · line 756
Quoted 4 lines
drivers/net/netdevsim/netdev.c +4
-0+if (queue >= ns->netdev->real_num_rx_queues) { + ret = -EINVAL ; + goto exit_unlock; + }Is it correct that both these above checks are not inside of
netdev_rx_queue_restart()itself? Or should we fix that?
Crossed my mind, too. But I didn't want to start futzing with netdev_rx_queue_restart() based on just two callers. I hope we'll get to a queue API based safe reset sooner rather than later, and the driver needs will dedicate the final shape.
More concisely put - a debugfs hook in a test harness is not a strong signal for whether the API is right.
Jakub Kicinski
maintainerPATCH v2 8/8 selftests: net: test listing NAPI vs queue resets
In reply to Jakub KicinskiTest listing netdevsim NAPIs before and after a single queue has been reset (and NAPIs re-added).
Start from resetting the middle queue because edge cases (first / last) may actually be less likely to trigger bugs.
# ./tools/testing/selftests/net/nl_netdev.py KTAP version 1 1..4 ok 1 nl_netdev.empty_check ok 2 nl_netdev.lo_check ok 3 nl_netdev.page_pool_check ok 4 nl_netdev.napi_list_check # Totals: pass:4 fail:0 xfail:0 xpass:0 skip:0 error:0
--- tools/testing/selftests/net/nl_netdev.py | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-)
tools/testing/selftests/net/nl_netdev.py +18 -1
@@ -18,6 +18,23 @@ from lib.py import NetdevFamily, NetdevSimDev, ip ksft_eq(len(lo_info['xdp-rx-metadata-features']), 0) +def napi_list_check(nf) -> None: + with NetdevSimDev(queue_count=100) as nsimdev: + nsim = nsimdev.nsims[0] + + ip(f"link set dev {nsim.ifname} up") + + napis = nf.napi_get({'ifindex': nsim.ifindex}, dump=True) + ksft_eq(len(napis), 100) + + for q in [50, 0, 99]: + for i in range(4): + nsim.dfs_write("queue_reset", f"{q} {i}") + napis = nf.napi_get({'ifindex': nsim.ifindex}, dump=True) + ksft_eq(len(napis), 100, + comment=f"queue count after reset queue {q} mode {i}") + + def page_pool_check(nf) -> None: with NetdevSimDev() as nsimdev: nsim = nsimdev.nsims[0]
@@ -89,7 +106,7 @@ from lib.py import NetdevFamily, NetdevSimDev, ip def main() -> None: nf = NetdevFamily() - ksft_run([empty_check, lo_check, page_pool_check], + ksft_run([empty_check, lo_check, page_pool_check, napi_list_check], args=(nf, )) ksft_exit()
Signature · 1 line
2.47.1
This series was applied to netdev/net-next.git (main) by Paolo Abeni <[email protected]>:
Hello:
This series was applied to netdev/net-next.git (main) by Paolo Abeni <[email protected]>:
Quoted 8 lines · On Tue, 7 Jan 2025 08:08:38 -0800 you wrote:
I promised Eric to remove the rtnl protection of the NAPI list, when I sat down to implement it over the break I realized that the recently added NAPI ID retention will break the list ordering assumption we have in netlink dump. The ordering used to happen "naturally", because we'd always add NAPIs that the head of the list, and assign a new monotonically increasing ID.
[...]
Here is the summary with links:
- [net-next,v2,1/8] net: make sure we retain NAPI ordering on netdev->napi_list
https://git.kernel.org/netdev/net-next/c/d6c7b03497ee - [net-next,v2,2/8] netdev: define NETDEV_INTERNAL https://git.kernel.org/netdev/net-next/c/0b7bdc7fab57 - [net-next,v2,3/8] netdevsim: support NAPI config https://git.kernel.org/netdev/net-next/c/00adf88b186f - [net-next,v2,4/8] netdevsim: allocate rqs individually https://git.kernel.org/netdev/net-next/c/915c82f842f9 - [net-next,v2,5/8] netdevsim: add queue alloc/free helpers https://git.kernel.org/netdev/net-next/c/a565dd04a120 - [net-next,v2,6/8] netdevsim: add queue management API support https://git.kernel.org/netdev/net-next/c/5bc8e8dbef27 - [net-next,v2,7/8] netdevsim: add debugfs-triggered queue reset https://git.kernel.org/netdev/net-next/c/6917d207b469 - [net-next,v2,8/8] selftests: net: test listing NAPI vs queue resets https://git.kernel.org/netdev/net-next/c/eb721f117e7d
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html