All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH bpf-next 0/9] bpf: offload program and map sharing
@ 2018-07-17  2:37 Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 1/9] netdevsim: add switch_id attribute Jakub Kicinski
                   ` (8 more replies)
  0 siblings, 9 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Hi!

This patchset adds support for sharing BPF objects within one ASIC.
This will allow us to reuse of the same program on multiple ports of
a device leading to better code store utilization.  It also enables
sharing maps between programs attached to different ports of a device.

Jakub Kicinski (9):
  netdevsim: add switch_id attribute
  netdevsim: add shared netdevsim devices
  netdevsim: associate bound programs with shared dev
  nfp: add .ndo_init() and .ndo_uninit() callbacks
  bpf: offload: aggregate offloads per-device
  bpf: offload: allow program and map sharing per-ASIC
  netdevsim: allow program sharing between devices
  nfp: bpf: allow program sharing within ASIC
  selftests/bpf: add test for sharing objects between netdevs

 drivers/net/ethernet/netronome/nfp/bpf/main.c |  23 ++
 drivers/net/ethernet/netronome/nfp/bpf/main.h |   4 +
 .../net/ethernet/netronome/nfp/bpf/offload.c  |  10 +-
 drivers/net/ethernet/netronome/nfp/nfp_app.c  |  17 ++
 drivers/net/ethernet/netronome/nfp/nfp_app.h  |   8 +
 .../ethernet/netronome/nfp/nfp_net_common.c   |   2 +
 .../net/ethernet/netronome/nfp/nfp_net_repr.c |   2 +
 drivers/net/netdevsim/bpf.c                   |  50 +++-
 drivers/net/netdevsim/netdev.c                | 103 +++++++-
 drivers/net/netdevsim/netdevsim.h             |  23 +-
 include/linux/bpf.h                           |  10 +-
 kernel/bpf/offload.c                          | 223 ++++++++++++++----
 kernel/bpf/verifier.c                         |   2 +-
 tools/testing/selftests/bpf/test_offload.py   | 151 +++++++++++-
 14 files changed, 543 insertions(+), 85 deletions(-)

-- 
2.17.1

^ permalink raw reply	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 1/9] netdevsim: add switch_id attribute
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 2/9] netdevsim: add shared netdevsim devices Jakub Kicinski
                   ` (7 subsequent siblings)
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Grouping netdevsim devices into "ASICs" will soon be supported.
Add switch_id attribute to all netdevsims.  For now each netdevsim
will have its switch_id matching the device id.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/netdevsim/netdev.c    | 24 ++++++++++++++++++++++++
 drivers/net/netdevsim/netdevsim.h |  1 +
 2 files changed, 25 insertions(+)

diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
index a7b179f0d954..9125637ef5d8 100644
--- a/drivers/net/netdevsim/netdev.c
+++ b/drivers/net/netdevsim/netdev.c
@@ -22,6 +22,7 @@
 #include <net/netlink.h>
 #include <net/pkt_cls.h>
 #include <net/rtnetlink.h>
+#include <net/switchdev.h>
 
 #include "netdevsim.h"
 
@@ -144,12 +145,34 @@ static struct device_type nsim_dev_type = {
 	.release = nsim_dev_release,
 };
 
+static int
+nsim_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
+{
+	struct netdevsim *ns = netdev_priv(dev);
+
+	switch (attr->id) {
+	case SWITCHDEV_ATTR_ID_PORT_PARENT_ID:
+		attr->u.ppid.id_len = sizeof(ns->switch_id);
+		memcpy(&attr->u.ppid.id, &ns->switch_id,
+		       attr->u.ppid.id_len);
+		return 0;
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static const struct switchdev_ops nsim_switchdev_ops = {
+	.switchdev_port_attr_get	= nsim_port_attr_get,
+};
+
 static int nsim_init(struct net_device *dev)
 {
 	struct netdevsim *ns = netdev_priv(dev);
 	int err;
 
 	ns->netdev = dev;
+	ns->switch_id = nsim_dev_id;
+
 	ns->ddir = debugfs_create_dir(netdev_name(dev), nsim_ddir);
 	if (IS_ERR_OR_NULL(ns->ddir))
 		return -ENOMEM;
@@ -166,6 +189,7 @@ static int nsim_init(struct net_device *dev)
 		goto err_bpf_uninit;
 
 	SET_NETDEV_DEV(dev, &ns->dev);
+	SWITCHDEV_SET_OPS(dev, &nsim_switchdev_ops);
 
 	err = nsim_devlink_setup(ns);
 	if (err)
diff --git a/drivers/net/netdevsim/netdevsim.h b/drivers/net/netdevsim/netdevsim.h
index 0aeabbe81cc6..e2f232325259 100644
--- a/drivers/net/netdevsim/netdevsim.h
+++ b/drivers/net/netdevsim/netdevsim.h
@@ -59,6 +59,7 @@ struct netdevsim {
 	struct u64_stats_sync syncp;
 
 	struct device dev;
+	u32 switch_id;
 
 	struct dentry *ddir;
 
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 2/9] netdevsim: add shared netdevsim devices
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 1/9] netdevsim: add switch_id attribute Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 3/9] netdevsim: associate bound programs with shared dev Jakub Kicinski
                   ` (6 subsequent siblings)
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Factor out sharable netdevsim sub-object and use IFLA_LINK to link
netdevsims together at creation time.  Sharable object will have
its own DebugFS directory.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/netdevsim/netdev.c    | 87 ++++++++++++++++++++++++++++---
 drivers/net/netdevsim/netdevsim.h | 10 +++-
 2 files changed, 90 insertions(+), 7 deletions(-)

diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
index 9125637ef5d8..2d244551298b 100644
--- a/drivers/net/netdevsim/netdev.c
+++ b/drivers/net/netdevsim/netdev.c
@@ -152,8 +152,8 @@ nsim_port_attr_get(struct net_device *dev, struct switchdev_attr *attr)
 
 	switch (attr->id) {
 	case SWITCHDEV_ATTR_ID_PORT_PARENT_ID:
-		attr->u.ppid.id_len = sizeof(ns->switch_id);
-		memcpy(&attr->u.ppid.id, &ns->switch_id,
+		attr->u.ppid.id_len = sizeof(ns->sdev->switch_id);
+		memcpy(&attr->u.ppid.id, &ns->sdev->switch_id,
 		       attr->u.ppid.id_len);
 		return 0;
 	default:
@@ -167,19 +167,41 @@ static const struct switchdev_ops nsim_switchdev_ops = {
 
 static int nsim_init(struct net_device *dev)
 {
+	char sdev_ddir_name[10], sdev_link_name[32];
 	struct netdevsim *ns = netdev_priv(dev);
 	int err;
 
 	ns->netdev = dev;
-	ns->switch_id = nsim_dev_id;
-
 	ns->ddir = debugfs_create_dir(netdev_name(dev), nsim_ddir);
 	if (IS_ERR_OR_NULL(ns->ddir))
 		return -ENOMEM;
 
+	if (!ns->sdev) {
+		ns->sdev = kzalloc(sizeof(*ns->sdev), GFP_KERNEL);
+		if (!ns->sdev) {
+			err = -ENOMEM;
+			goto err_debugfs_destroy;
+		}
+		ns->sdev->refcnt = 1;
+		ns->sdev->switch_id = nsim_dev_id;
+		sprintf(sdev_ddir_name, "%u", ns->sdev->switch_id);
+		ns->sdev->ddir = debugfs_create_dir(sdev_ddir_name,
+						    nsim_sdev_ddir);
+		if (IS_ERR_OR_NULL(ns->sdev->ddir)) {
+			err = PTR_ERR_OR_ZERO(ns->sdev->ddir) ?: -EINVAL;
+			goto err_sdev_free;
+		}
+	} else {
+		sprintf(sdev_ddir_name, "%u", ns->sdev->switch_id);
+		ns->sdev->refcnt++;
+	}
+
+	sprintf(sdev_link_name, "../../" DRV_NAME "_sdev/%s", sdev_ddir_name);
+	debugfs_create_symlink("sdev", ns->ddir, sdev_link_name);
+
 	err = nsim_bpf_init(ns);
 	if (err)
-		goto err_debugfs_destroy;
+		goto err_sdev_destroy;
 
 	ns->dev.id = nsim_dev_id++;
 	ns->dev.bus = &nsim_bus;
@@ -203,6 +225,12 @@ static int nsim_init(struct net_device *dev)
 	device_unregister(&ns->dev);
 err_bpf_uninit:
 	nsim_bpf_uninit(ns);
+err_sdev_destroy:
+	if (!--ns->sdev->refcnt) {
+		debugfs_remove_recursive(ns->sdev->ddir);
+err_sdev_free:
+		kfree(ns->sdev);
+	}
 err_debugfs_destroy:
 	debugfs_remove_recursive(ns->ddir);
 	return err;
@@ -216,6 +244,10 @@ static void nsim_uninit(struct net_device *dev)
 	nsim_devlink_teardown(ns);
 	debugfs_remove_recursive(ns->ddir);
 	nsim_bpf_uninit(ns);
+	if (!--ns->sdev->refcnt) {
+		debugfs_remove_recursive(ns->sdev->ddir);
+		kfree(ns->sdev);
+	}
 }
 
 static void nsim_free(struct net_device *dev)
@@ -494,14 +526,48 @@ static int nsim_validate(struct nlattr *tb[], struct nlattr *data[],
 	return 0;
 }
 
+static int nsim_newlink(struct net *src_net, struct net_device *dev,
+			struct nlattr *tb[], struct nlattr *data[],
+			struct netlink_ext_ack *extack)
+{
+	struct netdevsim *ns = netdev_priv(dev);
+
+	if (tb[IFLA_LINK]) {
+		struct net_device *joindev;
+		struct netdevsim *joinns;
+
+		joindev = __dev_get_by_index(src_net,
+					     nla_get_u32(tb[IFLA_LINK]));
+		if (!joindev)
+			return -ENODEV;
+		if (joindev->netdev_ops != &nsim_netdev_ops)
+			return -EINVAL;
+
+		joinns = netdev_priv(joindev);
+		if (!joinns->sdev || !joinns->sdev->refcnt)
+			return -EINVAL;
+		ns->sdev = joinns->sdev;
+	}
+
+	return register_netdevice(dev);
+}
+
+static void nsim_dellink(struct net_device *dev, struct list_head *head)
+{
+	unregister_netdevice_queue(dev, head);
+}
+
 static struct rtnl_link_ops nsim_link_ops __read_mostly = {
 	.kind		= DRV_NAME,
 	.priv_size	= sizeof(struct netdevsim),
 	.setup		= nsim_setup,
 	.validate	= nsim_validate,
+	.newlink	= nsim_newlink,
+	.dellink	= nsim_dellink,
 };
 
 struct dentry *nsim_ddir;
+struct dentry *nsim_sdev_ddir;
 
 static int __init nsim_module_init(void)
 {
@@ -511,9 +577,15 @@ static int __init nsim_module_init(void)
 	if (IS_ERR_OR_NULL(nsim_ddir))
 		return -ENOMEM;
 
+	nsim_sdev_ddir = debugfs_create_dir(DRV_NAME "_sdev", NULL);
+	if (IS_ERR_OR_NULL(nsim_sdev_ddir)) {
+		err = -ENOMEM;
+		goto err_debugfs_destroy;
+	}
+
 	err = bus_register(&nsim_bus);
 	if (err)
-		goto err_debugfs_destroy;
+		goto err_sdir_destroy;
 
 	err = nsim_devlink_init();
 	if (err)
@@ -529,6 +601,8 @@ static int __init nsim_module_init(void)
 	nsim_devlink_exit();
 err_unreg_bus:
 	bus_unregister(&nsim_bus);
+err_sdir_destroy:
+	debugfs_remove_recursive(nsim_sdev_ddir);
 err_debugfs_destroy:
 	debugfs_remove_recursive(nsim_ddir);
 	return err;
@@ -539,6 +613,7 @@ static void __exit nsim_module_exit(void)
 	rtnl_link_unregister(&nsim_link_ops);
 	nsim_devlink_exit();
 	bus_unregister(&nsim_bus);
+	debugfs_remove_recursive(nsim_sdev_ddir);
 	debugfs_remove_recursive(nsim_ddir);
 }
 
diff --git a/drivers/net/netdevsim/netdevsim.h b/drivers/net/netdevsim/netdevsim.h
index e2f232325259..8743ce74d2d9 100644
--- a/drivers/net/netdevsim/netdevsim.h
+++ b/drivers/net/netdevsim/netdevsim.h
@@ -30,6 +30,13 @@ struct bpf_prog;
 struct dentry;
 struct nsim_vf_config;
 
+struct netdevsim_shared_dev {
+	unsigned int refcnt;
+	u32 switch_id;
+
+	struct dentry *ddir;
+};
+
 #define NSIM_IPSEC_MAX_SA_COUNT		33
 #define NSIM_IPSEC_VALID		BIT(31)
 
@@ -59,7 +66,7 @@ struct netdevsim {
 	struct u64_stats_sync syncp;
 
 	struct device dev;
-	u32 switch_id;
+	struct netdevsim_shared_dev *sdev;
 
 	struct dentry *ddir;
 
@@ -93,6 +100,7 @@ struct netdevsim {
 };
 
 extern struct dentry *nsim_ddir;
+extern struct dentry *nsim_sdev_ddir;
 
 #ifdef CONFIG_BPF_SYSCALL
 int nsim_bpf_init(struct netdevsim *ns);
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 3/9] netdevsim: associate bound programs with shared dev
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 1/9] netdevsim: add switch_id attribute Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 2/9] netdevsim: add shared netdevsim devices Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 4/9] nfp: add .ndo_init() and .ndo_uninit() callbacks Jakub Kicinski
                   ` (5 subsequent siblings)
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Move bound program information from netdevsim to shared sub-object,
as programs will soon be shared between netdevs of the same ASIC.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/netdevsim/bpf.c                 | 30 ++++++++++++---------
 drivers/net/netdevsim/netdevsim.h           | 11 ++++----
 tools/testing/selftests/bpf/test_offload.py |  5 ++--
 3 files changed, 27 insertions(+), 19 deletions(-)

diff --git a/drivers/net/netdevsim/bpf.c b/drivers/net/netdevsim/bpf.c
index c36d2a768202..357f9e62f306 100644
--- a/drivers/net/netdevsim/bpf.c
+++ b/drivers/net/netdevsim/bpf.c
@@ -238,8 +238,8 @@ static int nsim_bpf_create_prog(struct netdevsim *ns, struct bpf_prog *prog)
 	state->state = "verify";
 
 	/* Program id is not populated yet when we create the state. */
-	sprintf(name, "%u", ns->prog_id_gen++);
-	state->ddir = debugfs_create_dir(name, ns->ddir_bpf_bound_progs);
+	sprintf(name, "%u", ns->sdev->prog_id_gen++);
+	state->ddir = debugfs_create_dir(name, ns->sdev->ddir_bpf_bound_progs);
 	if (IS_ERR_OR_NULL(state->ddir)) {
 		kfree(state);
 		return -ENOMEM;
@@ -250,7 +250,7 @@ static int nsim_bpf_create_prog(struct netdevsim *ns, struct bpf_prog *prog)
 			    &state->state, &nsim_bpf_string_fops);
 	debugfs_create_bool("loaded", 0400, state->ddir, &state->is_loaded);
 
-	list_add_tail(&state->l, &ns->bpf_bound_progs);
+	list_add_tail(&state->l, &ns->sdev->bpf_bound_progs);
 
 	prog->aux->offload->dev_priv = state;
 
@@ -497,7 +497,7 @@ nsim_bpf_map_alloc(struct netdevsim *ns, struct bpf_offloaded_map *offmap)
 	}
 
 	offmap->dev_ops = &nsim_bpf_map_ops;
-	list_add_tail(&nmap->l, &ns->bpf_bound_maps);
+	list_add_tail(&nmap->l, &ns->sdev->bpf_bound_maps);
 
 	return 0;
 
@@ -582,8 +582,15 @@ int nsim_bpf(struct net_device *dev, struct netdev_bpf *bpf)
 
 int nsim_bpf_init(struct netdevsim *ns)
 {
-	INIT_LIST_HEAD(&ns->bpf_bound_progs);
-	INIT_LIST_HEAD(&ns->bpf_bound_maps);
+	if (ns->sdev->refcnt == 1) {
+		INIT_LIST_HEAD(&ns->sdev->bpf_bound_progs);
+		INIT_LIST_HEAD(&ns->sdev->bpf_bound_maps);
+
+		ns->sdev->ddir_bpf_bound_progs =
+			debugfs_create_dir("bpf_bound_progs", ns->sdev->ddir);
+		if (IS_ERR_OR_NULL(ns->sdev->ddir_bpf_bound_progs))
+			return -ENOMEM;
+	}
 
 	debugfs_create_u32("bpf_offloaded_id", 0400, ns->ddir,
 			   &ns->bpf_offloaded_id);
@@ -593,10 +600,6 @@ int nsim_bpf_init(struct netdevsim *ns)
 			    &ns->bpf_bind_accept);
 	debugfs_create_u32("bpf_bind_verifier_delay", 0600, ns->ddir,
 			   &ns->bpf_bind_verifier_delay);
-	ns->ddir_bpf_bound_progs =
-		debugfs_create_dir("bpf_bound_progs", ns->ddir);
-	if (IS_ERR_OR_NULL(ns->ddir_bpf_bound_progs))
-		return -ENOMEM;
 
 	ns->bpf_tc_accept = true;
 	debugfs_create_bool("bpf_tc_accept", 0600, ns->ddir,
@@ -619,9 +622,12 @@ int nsim_bpf_init(struct netdevsim *ns)
 
 void nsim_bpf_uninit(struct netdevsim *ns)
 {
-	WARN_ON(!list_empty(&ns->bpf_bound_progs));
-	WARN_ON(!list_empty(&ns->bpf_bound_maps));
 	WARN_ON(ns->xdp.prog);
 	WARN_ON(ns->xdp_hw.prog);
 	WARN_ON(ns->bpf_offloaded);
+
+	if (ns->sdev->refcnt == 1) {
+		WARN_ON(!list_empty(&ns->sdev->bpf_bound_progs));
+		WARN_ON(!list_empty(&ns->sdev->bpf_bound_maps));
+	}
 }
diff --git a/drivers/net/netdevsim/netdevsim.h b/drivers/net/netdevsim/netdevsim.h
index 8743ce74d2d9..98f26fa1e671 100644
--- a/drivers/net/netdevsim/netdevsim.h
+++ b/drivers/net/netdevsim/netdevsim.h
@@ -35,6 +35,12 @@ struct netdevsim_shared_dev {
 	u32 switch_id;
 
 	struct dentry *ddir;
+
+	struct dentry *ddir_bpf_bound_progs;
+	u32 prog_id_gen;
+
+	struct list_head bpf_bound_progs;
+	struct list_head bpf_bound_maps;
 };
 
 #define NSIM_IPSEC_MAX_SA_COUNT		33
@@ -79,12 +85,8 @@ struct netdevsim {
 	struct xdp_attachment_info xdp;
 	struct xdp_attachment_info xdp_hw;
 
-	u32 prog_id_gen;
-
 	bool bpf_bind_accept;
 	u32 bpf_bind_verifier_delay;
-	struct dentry *ddir_bpf_bound_progs;
-	struct list_head bpf_bound_progs;
 
 	bool bpf_tc_accept;
 	bool bpf_tc_non_bound_accept;
@@ -92,7 +94,6 @@ struct netdevsim {
 	bool bpf_xdpoffload_accept;
 
 	bool bpf_map_accept;
-	struct list_head bpf_bound_maps;
 #if IS_ENABLED(CONFIG_NET_DEVLINK)
 	struct devlink *devlink;
 #endif
diff --git a/tools/testing/selftests/bpf/test_offload.py b/tools/testing/selftests/bpf/test_offload.py
index b746227eaff2..ee1abef384ea 100755
--- a/tools/testing/selftests/bpf/test_offload.py
+++ b/tools/testing/selftests/bpf/test_offload.py
@@ -314,6 +314,7 @@ netns = [] # net namespaces to be removed
         self.ns = ""
 
         self.dfs_dir = '/sys/kernel/debug/netdevsim/%s' % (self.dev['ifname'])
+        self.sdev_dir = self.dfs_dir + '/sdev/'
         self.dfs_refresh()
 
     def __getitem__(self, key):
@@ -345,12 +346,12 @@ netns = [] # net namespaces to be removed
         return data.strip()
 
     def dfs_num_bound_progs(self):
-        path = os.path.join(self.dfs_dir, "bpf_bound_progs")
+        path = os.path.join(self.sdev_dir, "bpf_bound_progs")
         _, progs = cmd('ls %s' % (path))
         return len(progs.split())
 
     def dfs_get_bound_progs(self, expected):
-        progs = DebugfsDir(os.path.join(self.dfs_dir, "bpf_bound_progs"))
+        progs = DebugfsDir(os.path.join(self.sdev_dir, "bpf_bound_progs"))
         if expected is not None:
             if len(progs) != expected:
                 fail(True, "%d BPF programs bound, expected %d" %
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 4/9] nfp: add .ndo_init() and .ndo_uninit() callbacks
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
                   ` (2 preceding siblings ...)
  2018-07-17  2:37 ` [PATCH bpf-next 3/9] netdevsim: associate bound programs with shared dev Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 5/9] bpf: offload: aggregate offloads per-device Jakub Kicinski
                   ` (4 subsequent siblings)
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

BPF code should unregister the offload capabilities from .ndo_uninit(),
to make sure the operation is atomic with unlist_netdevice().  Plumb
the init/uninit NDOs for vNICs and representors.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_app.c    | 17 +++++++++++++++++
 drivers/net/ethernet/netronome/nfp/nfp_app.h    |  8 ++++++++
 .../net/ethernet/netronome/nfp/nfp_net_common.c |  2 ++
 .../net/ethernet/netronome/nfp/nfp_net_repr.c   |  2 ++
 4 files changed, 29 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_app.c b/drivers/net/ethernet/netronome/nfp/nfp_app.c
index f28b244f4ee7..69d4ae7a61f3 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_app.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_app.c
@@ -86,6 +86,23 @@ const char *nfp_app_mip_name(struct nfp_app *app)
 	return nfp_mip_name(app->pf->mip);
 }
 
+int nfp_app_ndo_init(struct net_device *netdev)
+{
+	struct nfp_app *app = nfp_app_from_netdev(netdev);
+
+	if (!app || !app->type->ndo_init)
+		return 0;
+	return app->type->ndo_init(app, netdev);
+}
+
+void nfp_app_ndo_uninit(struct net_device *netdev)
+{
+	struct nfp_app *app = nfp_app_from_netdev(netdev);
+
+	if (app && app->type->ndo_uninit)
+		app->type->ndo_uninit(app, netdev);
+}
+
 u64 *nfp_app_port_get_stats(struct nfp_port *port, u64 *data)
 {
 	if (!port || !port->app || !port->app->type->port_get_stats)
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_app.h b/drivers/net/ethernet/netronome/nfp/nfp_app.h
index ee74caacb015..afbc19aa66a8 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_app.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_app.h
@@ -78,6 +78,8 @@ extern const struct nfp_app_type app_abm;
  * @init:	perform basic app checks and init
  * @clean:	clean app state
  * @extra_cap:	extra capabilities string
+ * @ndo_init:	vNIC and repr netdev .ndo_init
+ * @ndo_uninit:	vNIC and repr netdev .ndo_unint
  * @vnic_alloc:	allocate vNICs (assign port types, etc.)
  * @vnic_free:	free up app's vNIC state
  * @vnic_init:	vNIC netdev was registered
@@ -117,6 +119,9 @@ struct nfp_app_type {
 
 	const char *(*extra_cap)(struct nfp_app *app, struct nfp_net *nn);
 
+	int (*ndo_init)(struct nfp_app *app, struct net_device *netdev);
+	void (*ndo_uninit)(struct nfp_app *app, struct net_device *netdev);
+
 	int (*vnic_alloc)(struct nfp_app *app, struct nfp_net *nn,
 			  unsigned int id);
 	void (*vnic_free)(struct nfp_app *app, struct nfp_net *nn);
@@ -200,6 +205,9 @@ static inline void nfp_app_clean(struct nfp_app *app)
 		app->type->clean(app);
 }
 
+int nfp_app_ndo_init(struct net_device *netdev);
+void nfp_app_ndo_uninit(struct net_device *netdev);
+
 static inline int nfp_app_vnic_alloc(struct nfp_app *app, struct nfp_net *nn,
 				     unsigned int id)
 {
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index a712e83c3f0f..279b8ab8a17b 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -3480,6 +3480,8 @@ static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
 }
 
 const struct net_device_ops nfp_net_netdev_ops = {
+	.ndo_init		= nfp_app_ndo_init,
+	.ndo_uninit		= nfp_app_ndo_uninit,
 	.ndo_open		= nfp_net_netdev_open,
 	.ndo_stop		= nfp_net_netdev_close,
 	.ndo_start_xmit		= nfp_net_tx,
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
index d7b712f6362f..18a09cdcd9c6 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_repr.c
@@ -262,6 +262,8 @@ static int nfp_repr_open(struct net_device *netdev)
 }
 
 const struct net_device_ops nfp_repr_netdev_ops = {
+	.ndo_init		= nfp_app_ndo_init,
+	.ndo_uninit		= nfp_app_ndo_uninit,
 	.ndo_open		= nfp_repr_open,
 	.ndo_stop		= nfp_repr_stop,
 	.ndo_start_xmit		= nfp_repr_xmit,
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 5/9] bpf: offload: aggregate offloads per-device
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
                   ` (3 preceding siblings ...)
  2018-07-17  2:37 ` [PATCH bpf-next 4/9] nfp: add .ndo_init() and .ndo_uninit() callbacks Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC Jakub Kicinski
                   ` (3 subsequent siblings)
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Currently we have two lists of offloaded objects - programs and maps.
Netdevice unregister notifier scans those lists to orphan objects
associated with device being unregistered.  This puts unnecessary
(even if negligible) burden on all netdev unregister calls in BPF-
-enabled kernel.  The lists of objects may potentially get long
making the linear scan even more problematic.  There haven't been
complaints about this mechanisms so far, but it is suboptimal.

Instead of relying on notifiers, make the few BPF-capable drivers
register explicitly for BPF offloads.  The programs and maps will
now be collected per-device not on a global list, and only scanned
for removal when driver unregisters from BPF offloads.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/bpf/main.c |  13 ++
 drivers/net/netdevsim/bpf.c                   |   7 +
 include/linux/bpf.h                           |   2 +
 kernel/bpf/offload.c                          | 142 ++++++++++++------
 4 files changed, 118 insertions(+), 46 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/bpf/main.c b/drivers/net/ethernet/netronome/nfp/bpf/main.c
index b95b94d008cf..dee039ada75c 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/main.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/main.c
@@ -404,6 +404,16 @@ static int nfp_bpf_parse_capabilities(struct nfp_app *app)
 	return -EINVAL;
 }
 
+static int nfp_bpf_ndo_init(struct nfp_app *app, struct net_device *netdev)
+{
+	return bpf_offload_dev_netdev_register(netdev);
+}
+
+static void nfp_bpf_ndo_uninit(struct nfp_app *app, struct net_device *netdev)
+{
+	bpf_offload_dev_netdev_unregister(netdev);
+}
+
 static int nfp_bpf_init(struct nfp_app *app)
 {
 	struct nfp_app_bpf *bpf;
@@ -466,6 +476,9 @@ const struct nfp_app_type app_bpf = {
 
 	.extra_cap	= nfp_bpf_extra_cap,
 
+	.ndo_init	= nfp_bpf_ndo_init,
+	.ndo_uninit	= nfp_bpf_ndo_uninit,
+
 	.vnic_alloc	= nfp_bpf_vnic_alloc,
 	.vnic_free	= nfp_bpf_vnic_free,
 
diff --git a/drivers/net/netdevsim/bpf.c b/drivers/net/netdevsim/bpf.c
index 357f9e62f306..c4a2829e0e1f 100644
--- a/drivers/net/netdevsim/bpf.c
+++ b/drivers/net/netdevsim/bpf.c
@@ -582,6 +582,8 @@ int nsim_bpf(struct net_device *dev, struct netdev_bpf *bpf)
 
 int nsim_bpf_init(struct netdevsim *ns)
 {
+	int err;
+
 	if (ns->sdev->refcnt == 1) {
 		INIT_LIST_HEAD(&ns->sdev->bpf_bound_progs);
 		INIT_LIST_HEAD(&ns->sdev->bpf_bound_maps);
@@ -592,6 +594,10 @@ int nsim_bpf_init(struct netdevsim *ns)
 			return -ENOMEM;
 	}
 
+	err = bpf_offload_dev_netdev_register(ns->netdev);
+	if (err)
+		return err;
+
 	debugfs_create_u32("bpf_offloaded_id", 0400, ns->ddir,
 			   &ns->bpf_offloaded_id);
 
@@ -625,6 +631,7 @@ void nsim_bpf_uninit(struct netdevsim *ns)
 	WARN_ON(ns->xdp.prog);
 	WARN_ON(ns->xdp_hw.prog);
 	WARN_ON(ns->bpf_offloaded);
+	bpf_offload_dev_netdev_unregister(ns->netdev);
 
 	if (ns->sdev->refcnt == 1) {
 		WARN_ON(!list_empty(&ns->sdev->bpf_bound_progs));
diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 8827e797ff97..21c001c3285c 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -648,6 +648,8 @@ int bpf_map_offload_delete_elem(struct bpf_map *map, void *key);
 int bpf_map_offload_get_next_key(struct bpf_map *map,
 				 void *key, void *next_key);
 
+int bpf_offload_dev_netdev_register(struct net_device *netdev);
+void bpf_offload_dev_netdev_unregister(struct net_device *netdev);
 bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map);
 
 #if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL)
diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
index ac747d5cf7c6..b914f94c53d4 100644
--- a/kernel/bpf/offload.c
+++ b/kernel/bpf/offload.c
@@ -18,19 +18,37 @@
 #include <linux/bug.h>
 #include <linux/kdev_t.h>
 #include <linux/list.h>
+#include <linux/lockdep.h>
 #include <linux/netdevice.h>
 #include <linux/printk.h>
 #include <linux/proc_ns.h>
+#include <linux/rhashtable.h>
 #include <linux/rtnetlink.h>
 #include <linux/rwsem.h>
 
-/* Protects bpf_prog_offload_devs, bpf_map_offload_devs and offload members
+/* Protects offdevs, members of bpf_offload_netdev and offload members
  * of all progs.
  * RTNL lock cannot be taken when holding this lock.
  */
 static DECLARE_RWSEM(bpf_devs_lock);
-static LIST_HEAD(bpf_prog_offload_devs);
-static LIST_HEAD(bpf_map_offload_devs);
+
+struct bpf_offload_netdev {
+	struct rhash_head l;
+	struct net_device *netdev;
+	struct list_head progs;
+	struct list_head maps;
+};
+
+static const struct rhashtable_params offdevs_params = {
+	.nelem_hint		= 4,
+	.key_len		= sizeof(struct net_device *),
+	.key_offset		= offsetof(struct bpf_offload_netdev, netdev),
+	.head_offset		= offsetof(struct bpf_offload_netdev, l),
+	.automatic_shrinking	= true,
+};
+
+static struct rhashtable offdevs;
+static bool offdevs_inited;
 
 static int bpf_dev_offload_check(struct net_device *netdev)
 {
@@ -41,8 +59,19 @@ static int bpf_dev_offload_check(struct net_device *netdev)
 	return 0;
 }
 
+static struct bpf_offload_netdev *
+bpf_offload_find_netdev(struct net_device *netdev)
+{
+	lockdep_assert_held(&bpf_devs_lock);
+
+	if (!offdevs_inited)
+		return NULL;
+	return rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
+}
+
 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr)
 {
+	struct bpf_offload_netdev *ondev;
 	struct bpf_prog_offload *offload;
 	int err;
 
@@ -66,12 +95,13 @@ int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr)
 		goto err_maybe_put;
 
 	down_write(&bpf_devs_lock);
-	if (offload->netdev->reg_state != NETREG_REGISTERED) {
+	ondev = bpf_offload_find_netdev(offload->netdev);
+	if (!ondev) {
 		err = -EINVAL;
 		goto err_unlock;
 	}
 	prog->aux->offload = offload;
-	list_add_tail(&offload->offloads, &bpf_prog_offload_devs);
+	list_add_tail(&offload->offloads, &ondev->progs);
 	dev_put(offload->netdev);
 	up_write(&bpf_devs_lock);
 
@@ -294,6 +324,7 @@ static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
 {
 	struct net *net = current->nsproxy->net_ns;
+	struct bpf_offload_netdev *ondev;
 	struct bpf_offloaded_map *offmap;
 	int err;
 
@@ -316,11 +347,17 @@ struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
 	if (err)
 		goto err_unlock;
 
+	ondev = bpf_offload_find_netdev(offmap->netdev);
+	if (!ondev) {
+		err = -EINVAL;
+		goto err_unlock;
+	}
+
 	err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
 	if (err)
 		goto err_unlock;
 
-	list_add_tail(&offmap->offloads, &bpf_map_offload_devs);
+	list_add_tail(&offmap->offloads, &ondev->maps);
 	up_write(&bpf_devs_lock);
 	rtnl_unlock();
 
@@ -489,56 +526,69 @@ bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map)
 	return ret;
 }
 
-static void bpf_offload_orphan_all_progs(struct net_device *netdev)
+int bpf_offload_dev_netdev_register(struct net_device *netdev)
 {
-	struct bpf_prog_offload *offload, *tmp;
+	struct bpf_offload_netdev *ondev;
+	int err;
 
-	list_for_each_entry_safe(offload, tmp, &bpf_prog_offload_devs, offloads)
-		if (offload->netdev == netdev)
-			__bpf_prog_offload_destroy(offload->prog);
-}
+	down_write(&bpf_devs_lock);
+	if (!offdevs_inited) {
+		err = rhashtable_init(&offdevs, &offdevs_params);
+		if (err)
+			return err;
+		offdevs_inited = true;
+	}
+	up_write(&bpf_devs_lock);
 
-static void bpf_offload_orphan_all_maps(struct net_device *netdev)
-{
-	struct bpf_offloaded_map *offmap, *tmp;
+	ondev = kzalloc(sizeof(*ondev), GFP_KERNEL);
+	if (!ondev)
+		return -ENOMEM;
+
+	ondev->netdev = netdev;
+	INIT_LIST_HEAD(&ondev->progs);
+	INIT_LIST_HEAD(&ondev->maps);
+
+	down_write(&bpf_devs_lock);
+	err = rhashtable_insert_fast(&offdevs, &ondev->l, offdevs_params);
+	if (err) {
+		netdev_warn(netdev, "failed to register for BPF offload\n");
+		goto err_unlock_free;
+	}
 
-	list_for_each_entry_safe(offmap, tmp, &bpf_map_offload_devs, offloads)
-		if (offmap->netdev == netdev)
-			__bpf_map_offload_destroy(offmap);
+	up_write(&bpf_devs_lock);
+	return 0;
+
+err_unlock_free:
+	up_write(&bpf_devs_lock);
+	kfree(ondev);
+	return err;
 }
+EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
 
-static int bpf_offload_notification(struct notifier_block *notifier,
-				    ulong event, void *ptr)
+void bpf_offload_dev_netdev_unregister(struct net_device *netdev)
 {
-	struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
+	struct bpf_offloaded_map *offmap, *mtmp;
+	struct bpf_prog_offload *offload, *ptmp;
+	struct bpf_offload_netdev *ondev;
 
 	ASSERT_RTNL();
 
-	switch (event) {
-	case NETDEV_UNREGISTER:
-		/* ignore namespace changes */
-		if (netdev->reg_state != NETREG_UNREGISTERING)
-			break;
-
-		down_write(&bpf_devs_lock);
-		bpf_offload_orphan_all_progs(netdev);
-		bpf_offload_orphan_all_maps(netdev);
-		up_write(&bpf_devs_lock);
-		break;
-	default:
-		break;
-	}
-	return NOTIFY_OK;
-}
+	down_write(&bpf_devs_lock);
+	ondev = rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
+	if (WARN_ON(!ondev))
+		goto unlock;
 
-static struct notifier_block bpf_offload_notifier = {
-	.notifier_call = bpf_offload_notification,
-};
+	WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params));
 
-static int __init bpf_offload_init(void)
-{
-	register_netdevice_notifier(&bpf_offload_notifier);
-	return 0;
-}
+	list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads)
+		__bpf_prog_offload_destroy(offload->prog);
+	list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads)
+		__bpf_map_offload_destroy(offmap);
 
-subsys_initcall(bpf_offload_init);
+	WARN_ON(!list_empty(&ondev->progs));
+	WARN_ON(!list_empty(&ondev->maps));
+	kfree(ondev);
+unlock:
+	up_write(&bpf_devs_lock);
+}
+EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
                   ` (4 preceding siblings ...)
  2018-07-17  2:37 ` [PATCH bpf-next 5/9] bpf: offload: aggregate offloads per-device Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  3:00   ` Alexei Starovoitov
  2018-07-17  2:37 ` [PATCH bpf-next 7/9] netdevsim: allow program sharing between devices Jakub Kicinski
                   ` (2 subsequent siblings)
  8 siblings, 1 reply; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Create a higher-level entity to represent a device/ASIC to allow
programs and maps to be shared between device ports.  The extra
work is required to make sure we don't destroy BPF objects as
soon as the netdev for which they were loaded gets destroyed,
as other ports may still be using them.  When netdev goes away
all of its BPF objects will be moved to other netdevs of the
device, and only destroyed when last netdev is unregistered.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/bpf/main.c |  14 +-
 drivers/net/ethernet/netronome/nfp/bpf/main.h |   4 +
 drivers/net/netdevsim/bpf.c                   |  17 ++-
 drivers/net/netdevsim/netdevsim.h             |   3 +
 include/linux/bpf.h                           |  12 +-
 kernel/bpf/offload.c                          | 123 ++++++++++++++----
 kernel/bpf/verifier.c                         |   2 +-
 7 files changed, 142 insertions(+), 33 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/bpf/main.c b/drivers/net/ethernet/netronome/nfp/bpf/main.c
index dee039ada75c..458f49235d06 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/main.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/main.c
@@ -406,12 +406,16 @@ static int nfp_bpf_parse_capabilities(struct nfp_app *app)
 
 static int nfp_bpf_ndo_init(struct nfp_app *app, struct net_device *netdev)
 {
-	return bpf_offload_dev_netdev_register(netdev);
+	struct nfp_app_bpf *bpf = app->priv;
+
+	return bpf_offload_dev_netdev_register(bpf->bpf_dev, netdev);
 }
 
 static void nfp_bpf_ndo_uninit(struct nfp_app *app, struct net_device *netdev)
 {
-	bpf_offload_dev_netdev_unregister(netdev);
+	struct nfp_app_bpf *bpf = app->priv;
+
+	bpf_offload_dev_netdev_unregister(bpf->bpf_dev, netdev);
 }
 
 static int nfp_bpf_init(struct nfp_app *app)
@@ -437,6 +441,11 @@ static int nfp_bpf_init(struct nfp_app *app)
 	if (err)
 		goto err_free_neutral_maps;
 
+	bpf->bpf_dev = bpf_offload_dev_create();
+	err = PTR_ERR_OR_ZERO(bpf->bpf_dev);
+	if (err)
+		goto err_free_neutral_maps;
+
 	return 0;
 
 err_free_neutral_maps:
@@ -455,6 +464,7 @@ static void nfp_bpf_clean(struct nfp_app *app)
 {
 	struct nfp_app_bpf *bpf = app->priv;
 
+	bpf_offload_dev_destroy(bpf->bpf_dev);
 	WARN_ON(!skb_queue_empty(&bpf->cmsg_replies));
 	WARN_ON(!list_empty(&bpf->map_list));
 	WARN_ON(bpf->maps_in_use || bpf->map_elems_in_use);
diff --git a/drivers/net/ethernet/netronome/nfp/bpf/main.h b/drivers/net/ethernet/netronome/nfp/bpf/main.h
index 9845c1a2d4c2..bec935468f90 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/main.h
+++ b/drivers/net/ethernet/netronome/nfp/bpf/main.h
@@ -110,6 +110,8 @@ enum pkt_vec {
  * struct nfp_app_bpf - bpf app priv structure
  * @app:		backpointer to the app
  *
+ * @bpf_dev:		BPF offload device handle
+ *
  * @tag_allocator:	bitmap of control message tags in use
  * @tag_alloc_next:	next tag bit to allocate
  * @tag_alloc_last:	next tag bit to be freed
@@ -150,6 +152,8 @@ enum pkt_vec {
 struct nfp_app_bpf {
 	struct nfp_app *app;
 
+	struct bpf_offload_dev *bpf_dev;
+
 	DECLARE_BITMAP(tag_allocator, U16_MAX + 1);
 	u16 tag_alloc_next;
 	u16 tag_alloc_last;
diff --git a/drivers/net/netdevsim/bpf.c b/drivers/net/netdevsim/bpf.c
index c4a2829e0e1f..9eab29f67a0e 100644
--- a/drivers/net/netdevsim/bpf.c
+++ b/drivers/net/netdevsim/bpf.c
@@ -592,11 +592,16 @@ int nsim_bpf_init(struct netdevsim *ns)
 			debugfs_create_dir("bpf_bound_progs", ns->sdev->ddir);
 		if (IS_ERR_OR_NULL(ns->sdev->ddir_bpf_bound_progs))
 			return -ENOMEM;
+
+		ns->sdev->bpf_dev = bpf_offload_dev_create();
+		err = PTR_ERR_OR_ZERO(ns->sdev->bpf_dev);
+		if (err)
+			return err;
 	}
 
-	err = bpf_offload_dev_netdev_register(ns->netdev);
+	err = bpf_offload_dev_netdev_register(ns->sdev->bpf_dev, ns->netdev);
 	if (err)
-		return err;
+		goto err_destroy_bdev;
 
 	debugfs_create_u32("bpf_offloaded_id", 0400, ns->ddir,
 			   &ns->bpf_offloaded_id);
@@ -624,6 +629,11 @@ int nsim_bpf_init(struct netdevsim *ns)
 			    &ns->bpf_map_accept);
 
 	return 0;
+
+err_destroy_bdev:
+	if (ns->sdev->refcnt == 1)
+		bpf_offload_dev_destroy(ns->sdev->bpf_dev);
+	return err;
 }
 
 void nsim_bpf_uninit(struct netdevsim *ns)
@@ -631,10 +641,11 @@ void nsim_bpf_uninit(struct netdevsim *ns)
 	WARN_ON(ns->xdp.prog);
 	WARN_ON(ns->xdp_hw.prog);
 	WARN_ON(ns->bpf_offloaded);
-	bpf_offload_dev_netdev_unregister(ns->netdev);
+	bpf_offload_dev_netdev_unregister(ns->sdev->bpf_dev, ns->netdev);
 
 	if (ns->sdev->refcnt == 1) {
 		WARN_ON(!list_empty(&ns->sdev->bpf_bound_progs));
 		WARN_ON(!list_empty(&ns->sdev->bpf_bound_maps));
+		bpf_offload_dev_destroy(ns->sdev->bpf_dev);
 	}
 }
diff --git a/drivers/net/netdevsim/netdevsim.h b/drivers/net/netdevsim/netdevsim.h
index 98f26fa1e671..02be199eb005 100644
--- a/drivers/net/netdevsim/netdevsim.h
+++ b/drivers/net/netdevsim/netdevsim.h
@@ -27,6 +27,7 @@
 #define NSIM_EA(extack, msg)	NL_SET_ERR_MSG_MOD((extack), msg)
 
 struct bpf_prog;
+struct bpf_offload_dev;
 struct dentry;
 struct nsim_vf_config;
 
@@ -36,6 +37,8 @@ struct netdevsim_shared_dev {
 
 	struct dentry *ddir;
 
+	struct bpf_offload_dev *bpf_dev;
+
 	struct dentry *ddir_bpf_bound_progs;
 	u32 prog_id_gen;
 
diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 21c001c3285c..cd947441782e 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -85,6 +85,7 @@ struct bpf_map {
 	char name[BPF_OBJ_NAME_LEN];
 };
 
+struct bpf_offload_dev;
 struct bpf_offloaded_map;
 
 struct bpf_map_dev_ops {
@@ -648,9 +649,14 @@ int bpf_map_offload_delete_elem(struct bpf_map *map, void *key);
 int bpf_map_offload_get_next_key(struct bpf_map *map,
 				 void *key, void *next_key);
 
-int bpf_offload_dev_netdev_register(struct net_device *netdev);
-void bpf_offload_dev_netdev_unregister(struct net_device *netdev);
-bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map);
+struct bpf_offload_dev *bpf_offload_dev_create(void);
+void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev);
+int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
+				    struct net_device *netdev);
+void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
+				       struct net_device *netdev);
+bool bpf_offload_match(struct bpf_prog *prog, struct bpf_map *map);
+bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev);
 
 #if defined(CONFIG_NET) && defined(CONFIG_BPF_SYSCALL)
 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr);
diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
index b914f94c53d4..e090ec00390d 100644
--- a/kernel/bpf/offload.c
+++ b/kernel/bpf/offload.c
@@ -32,11 +32,17 @@
  */
 static DECLARE_RWSEM(bpf_devs_lock);
 
+struct bpf_offload_dev {
+	struct list_head netdevs;
+};
+
 struct bpf_offload_netdev {
 	struct rhash_head l;
 	struct net_device *netdev;
+	struct bpf_offload_dev *offdev;
 	struct list_head progs;
 	struct list_head maps;
+	struct list_head offdev_netdevs;
 };
 
 static const struct rhashtable_params offdevs_params = {
@@ -505,46 +511,67 @@ int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
 	return 0;
 }
 
-bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map)
+static bool __bpf_offload_dev_match(struct bpf_prog *prog,
+				    struct net_device *netdev)
 {
-	struct bpf_offloaded_map *offmap;
+	struct bpf_offload_netdev *ondev1, *ondev2;
 	struct bpf_prog_offload *offload;
-	bool ret;
 
 	if (!bpf_prog_is_dev_bound(prog->aux))
 		return false;
-	if (!bpf_map_is_dev_bound(map))
-		return bpf_map_offload_neutral(map);
 
-	down_read(&bpf_devs_lock);
 	offload = prog->aux->offload;
+	if (!offload)
+		return false;
+	if (offload->netdev == netdev)
+		return true;
+
+	ondev1 = bpf_offload_find_netdev(offload->netdev);
+	ondev2 = bpf_offload_find_netdev(netdev);
+
+	return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
+}
+
+bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
+{
+	bool ret;
+
+	down_read(&bpf_devs_lock);
+	ret = __bpf_offload_dev_match(prog, netdev);
+	up_read(&bpf_devs_lock);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
+
+bool bpf_offload_match(struct bpf_prog *prog, struct bpf_map *map)
+{
+	struct bpf_offloaded_map *offmap;
+	bool ret;
+
+	if (!bpf_map_is_dev_bound(map))
+		return bpf_map_offload_neutral(map);
 	offmap = map_to_offmap(map);
 
-	ret = offload && offload->netdev == offmap->netdev;
+	down_read(&bpf_devs_lock);
+	ret = __bpf_offload_dev_match(prog, offmap->netdev);
 	up_read(&bpf_devs_lock);
 
 	return ret;
 }
 
-int bpf_offload_dev_netdev_register(struct net_device *netdev)
+int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
+				    struct net_device *netdev)
 {
 	struct bpf_offload_netdev *ondev;
 	int err;
 
-	down_write(&bpf_devs_lock);
-	if (!offdevs_inited) {
-		err = rhashtable_init(&offdevs, &offdevs_params);
-		if (err)
-			return err;
-		offdevs_inited = true;
-	}
-	up_write(&bpf_devs_lock);
-
 	ondev = kzalloc(sizeof(*ondev), GFP_KERNEL);
 	if (!ondev)
 		return -ENOMEM;
 
 	ondev->netdev = netdev;
+	ondev->offdev = offdev;
 	INIT_LIST_HEAD(&ondev->progs);
 	INIT_LIST_HEAD(&ondev->maps);
 
@@ -555,6 +582,7 @@ int bpf_offload_dev_netdev_register(struct net_device *netdev)
 		goto err_unlock_free;
 	}
 
+	list_add(&ondev->offdev_netdevs, &offdev->netdevs);
 	up_write(&bpf_devs_lock);
 	return 0;
 
@@ -565,11 +593,12 @@ int bpf_offload_dev_netdev_register(struct net_device *netdev)
 }
 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
 
-void bpf_offload_dev_netdev_unregister(struct net_device *netdev)
+void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
+				       struct net_device *netdev)
 {
+	struct bpf_offload_netdev *ondev, *altdev;
 	struct bpf_offloaded_map *offmap, *mtmp;
 	struct bpf_prog_offload *offload, *ptmp;
-	struct bpf_offload_netdev *ondev;
 
 	ASSERT_RTNL();
 
@@ -579,11 +608,26 @@ void bpf_offload_dev_netdev_unregister(struct net_device *netdev)
 		goto unlock;
 
 	WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params));
-
-	list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads)
-		__bpf_prog_offload_destroy(offload->prog);
-	list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads)
-		__bpf_map_offload_destroy(offmap);
+	list_del(&ondev->offdev_netdevs);
+
+	/* Try to move the objects to another netdev of the device */
+	altdev = list_first_entry_or_null(&offdev->netdevs,
+					  struct bpf_offload_netdev,
+					  offdev_netdevs);
+	if (altdev) {
+		list_for_each_entry(offload, &ondev->progs, offloads)
+			offload->netdev = altdev->netdev;
+		list_splice_init(&ondev->progs, &altdev->progs);
+
+		list_for_each_entry(offmap, &ondev->maps, offloads)
+			offmap->netdev = altdev->netdev;
+		list_splice_init(&ondev->maps, &altdev->maps);
+	} else {
+		list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads)
+			__bpf_prog_offload_destroy(offload->prog);
+		list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads)
+			__bpf_map_offload_destroy(offmap);
+	}
 
 	WARN_ON(!list_empty(&ondev->progs));
 	WARN_ON(!list_empty(&ondev->maps));
@@ -592,3 +636,34 @@ void bpf_offload_dev_netdev_unregister(struct net_device *netdev)
 	up_write(&bpf_devs_lock);
 }
 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
+
+struct bpf_offload_dev *bpf_offload_dev_create(void)
+{
+	struct bpf_offload_dev *offdev;
+	int err;
+
+	down_write(&bpf_devs_lock);
+	if (!offdevs_inited) {
+		err = rhashtable_init(&offdevs, &offdevs_params);
+		if (err)
+			return ERR_PTR(err);
+		offdevs_inited = true;
+	}
+	up_write(&bpf_devs_lock);
+
+	offdev = kzalloc(sizeof(*offdev), GFP_KERNEL);
+	if (!offdev)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&offdev->netdevs);
+
+	return offdev;
+}
+EXPORT_SYMBOL_GPL(bpf_offload_dev_create);
+
+void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev)
+{
+	WARN_ON(!list_empty(&offdev->netdevs));
+	kfree(offdev);
+}
+EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy);
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 9e2bf834f13a..2c5b923eef75 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -5054,7 +5054,7 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
 	}
 
 	if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) &&
-	    !bpf_offload_dev_match(prog, map)) {
+	    !bpf_offload_match(prog, map)) {
 		verbose(env, "offload device mismatch between prog and map\n");
 		return -EINVAL;
 	}
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 7/9] netdevsim: allow program sharing between devices
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
                   ` (5 preceding siblings ...)
  2018-07-17  2:37 ` [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 8/9] nfp: bpf: allow program sharing within ASIC Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 9/9] selftests/bpf: add test for sharing objects between netdevs Jakub Kicinski
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Allow program sharing between devices which were linked together.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/netdevsim/bpf.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/netdevsim/bpf.c b/drivers/net/netdevsim/bpf.c
index 9eab29f67a0e..81444208b216 100644
--- a/drivers/net/netdevsim/bpf.c
+++ b/drivers/net/netdevsim/bpf.c
@@ -294,7 +294,7 @@ nsim_setup_prog_hw_checks(struct netdevsim *ns, struct netdev_bpf *bpf)
 		NSIM_EA(bpf->extack, "xdpoffload of non-bound program");
 		return -EINVAL;
 	}
-	if (bpf->prog->aux->offload->netdev != ns->netdev) {
+	if (!bpf_offload_dev_match(bpf->prog, ns->netdev)) {
 		NSIM_EA(bpf->extack, "program bound to different dev");
 		return -EINVAL;
 	}
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 8/9] nfp: bpf: allow program sharing within ASIC
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
                   ` (6 preceding siblings ...)
  2018-07-17  2:37 ` [PATCH bpf-next 7/9] netdevsim: allow program sharing between devices Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  2018-07-17  2:37 ` [PATCH bpf-next 9/9] selftests/bpf: add test for sharing objects between netdevs Jakub Kicinski
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Allow program sharing between netdevs of the same NFP ASIC.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/bpf/offload.c | 10 ++--------
 1 file changed, 2 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/bpf/offload.c b/drivers/net/ethernet/netronome/nfp/bpf/offload.c
index 78f44c4d95b4..49b03f7dbf46 100644
--- a/drivers/net/ethernet/netronome/nfp/bpf/offload.c
+++ b/drivers/net/ethernet/netronome/nfp/bpf/offload.c
@@ -566,14 +566,8 @@ int nfp_net_bpf_offload(struct nfp_net *nn, struct bpf_prog *prog,
 {
 	int err;
 
-	if (prog) {
-		struct bpf_prog_offload *offload = prog->aux->offload;
-
-		if (!offload)
-			return -EINVAL;
-		if (offload->netdev != nn->dp.netdev)
-			return -EINVAL;
-	}
+	if (prog && !bpf_offload_dev_match(prog, nn->dp.netdev))
+		return -EINVAL;
 
 	if (prog && old_prog) {
 		u8 cap;
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH bpf-next 9/9] selftests/bpf: add test for sharing objects between netdevs
  2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
                   ` (7 preceding siblings ...)
  2018-07-17  2:37 ` [PATCH bpf-next 8/9] nfp: bpf: allow program sharing within ASIC Jakub Kicinski
@ 2018-07-17  2:37 ` Jakub Kicinski
  8 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  2:37 UTC (permalink / raw)
  To: alexei.starovoitov, daniel; +Cc: oss-drivers, netdev, Jakub Kicinski

Add tests for sharing programs and maps between different netdevs.
Use netdevsim's ability to pretend multiple netdevs belong to the
same "ASIC".

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
---
 tools/testing/selftests/bpf/test_offload.py | 146 +++++++++++++++++++-
 1 file changed, 142 insertions(+), 4 deletions(-)

diff --git a/tools/testing/selftests/bpf/test_offload.py b/tools/testing/selftests/bpf/test_offload.py
index ee1abef384ea..d59642e70f56 100755
--- a/tools/testing/selftests/bpf/test_offload.py
+++ b/tools/testing/selftests/bpf/test_offload.py
@@ -158,8 +158,9 @@ netns = [] # net namespaces to be removed
     else:
         return ret, out
 
-def bpftool(args, JSON=True, ns="", fail=True):
-    return tool("bpftool", args, {"json":"-p"}, JSON=JSON, ns=ns, fail=fail)
+def bpftool(args, JSON=True, ns="", fail=True, include_stderr=False):
+    return tool("bpftool", args, {"json":"-p"}, JSON=JSON, ns=ns,
+                fail=fail, include_stderr=include_stderr)
 
 def bpftool_prog_list(expected=None, ns=""):
     _, progs = bpftool("prog show", JSON=True, ns=ns, fail=True)
@@ -201,6 +202,21 @@ netns = [] # net namespaces to be removed
         time.sleep(0.05)
     raise Exception("Time out waiting for map counts to stabilize want %d, have %d" % (expected, nmaps))
 
+def bpftool_prog_load(sample, file_name, maps=[], prog_type="xdp", dev=None,
+                      fail=True, include_stderr=False):
+    args = "prog load %s %s" % (os.path.join(bpf_test_dir, sample), file_name)
+    if prog_type is not None:
+        args += " type " + prog_type
+    if dev is not None:
+        args += " dev " + dev
+    if len(maps):
+        args += " map " + " map ".join(maps)
+
+    res = bpftool(args, fail=fail, include_stderr=include_stderr)
+    if res[0] == 0:
+        files.append(file_name)
+    return res
+
 def ip(args, force=False, JSON=True, ns="", fail=True, include_stderr=False):
     if force:
         args = "-force " + args
@@ -307,7 +323,9 @@ netns = [] # net namespaces to be removed
     Class for netdevsim netdevice and its attributes.
     """
 
-    def __init__(self):
+    def __init__(self, link=None):
+        self.link = link
+
         self.dev = self._netdevsim_create()
         devs.append(self)
 
@@ -321,8 +339,9 @@ netns = [] # net namespaces to be removed
         return self.dev[key]
 
     def _netdevsim_create(self):
+        link = "" if self.link is None else "link " + self.link.dev['ifname']
         _, old  = ip("link show")
-        ip("link add sim%d type netdevsim")
+        ip("link add sim%d {link} type netdevsim".format(link=link))
         _, new  = ip("link show")
 
         for dev in new:
@@ -848,6 +867,25 @@ netns = []
     sim.set_mtu(1500)
 
     sim.wait_for_flush()
+    start_test("Test non-offload XDP attaching to HW...")
+    bpftool_prog_load("sample_ret0.o", "/sys/fs/bpf/nooffload")
+    nooffload = bpf_pinned("/sys/fs/bpf/nooffload")
+    ret, _, err = sim.set_xdp(nooffload, "offload",
+                              fail=False, include_stderr=True)
+    fail(ret == 0, "attached non-offloaded XDP program to HW")
+    check_extack_nsim(err, "xdpoffload of non-bound program.", args)
+    rm("/sys/fs/bpf/nooffload")
+
+    start_test("Test offload XDP attaching to drv...")
+    bpftool_prog_load("sample_ret0.o", "/sys/fs/bpf/offload",
+                      dev=sim['ifname'])
+    offload = bpf_pinned("/sys/fs/bpf/offload")
+    ret, _, err = sim.set_xdp(offload, "drv", fail=False, include_stderr=True)
+    fail(ret == 0, "attached offloaded XDP program to drv")
+    check_extack(err, "using device-bound program without HW_MODE flag is not supported.", args)
+    rm("/sys/fs/bpf/offload")
+    sim.wait_for_flush()
+
     start_test("Test XDP offload...")
     _, _, err = sim.set_xdp(obj, "offload", verbose=True, include_stderr=True)
     ipl = sim.ip_link_show(xdp=True)
@@ -1141,6 +1179,106 @@ netns = []
     fail(ret == 0,
          "netdevsim didn't refuse to create a map with offload disabled")
 
+    sim.remove()
+
+    start_test("Test multi-dev ASIC program reuse...")
+    simA = NetdevSim()
+    simB1 = NetdevSim()
+    simB2 = NetdevSim(link=simB1)
+    simB3 = NetdevSim(link=simB1)
+    sims = (simA, simB1, simB2, simB3)
+    simB = (simB1, simB2, simB3)
+
+    bpftool_prog_load("sample_map_ret0.o", "/sys/fs/bpf/nsimA",
+                      dev=simA['ifname'])
+    progA = bpf_pinned("/sys/fs/bpf/nsimA")
+    bpftool_prog_load("sample_map_ret0.o", "/sys/fs/bpf/nsimB",
+                      dev=simB1['ifname'])
+    progB = bpf_pinned("/sys/fs/bpf/nsimB")
+
+    simA.set_xdp(progA, "offload", JSON=False)
+    for d in simB:
+        d.set_xdp(progB, "offload", JSON=False)
+
+    start_test("Test multi-dev ASIC cross-dev replace...")
+    ret, _ = simA.set_xdp(progB, "offload", force=True, JSON=False, fail=False)
+    fail(ret == 0, "cross-ASIC program allowed")
+    for d in simB:
+        ret, _ = d.set_xdp(progA, "offload", force=True, JSON=False, fail=False)
+        fail(ret == 0, "cross-ASIC program allowed")
+
+    start_test("Test multi-dev ASIC cross-dev install...")
+    for d in sims:
+        d.unset_xdp("offload")
+
+    ret, _, err = simA.set_xdp(progB, "offload", force=True, JSON=False,
+                               fail=False, include_stderr=True)
+    fail(ret == 0, "cross-ASIC program allowed")
+    check_extack_nsim(err, "program bound to different dev.", args)
+    for d in simB:
+        ret, _, err = d.set_xdp(progA, "offload", force=True, JSON=False,
+                                fail=False, include_stderr=True)
+        fail(ret == 0, "cross-ASIC program allowed")
+        check_extack_nsim(err, "program bound to different dev.", args)
+
+    start_test("Test multi-dev ASIC cross-dev map reuse...")
+
+    mapA = bpftool("prog show %s" % (progA))[1]["map_ids"][0]
+    mapB = bpftool("prog show %s" % (progB))[1]["map_ids"][0]
+
+    ret, _ = bpftool_prog_load("sample_map_ret0.o", "/sys/fs/bpf/nsimB_",
+                               dev=simB3['ifname'],
+                               maps=["idx 0 id %d" % (mapB)],
+                               fail=False)
+    fail(ret != 0, "couldn't reuse a map on the same ASIC")
+    rm("/sys/fs/bpf/nsimB_")
+
+    ret, _, err = bpftool_prog_load("sample_map_ret0.o", "/sys/fs/bpf/nsimA_",
+                                    dev=simA['ifname'],
+                                    maps=["idx 0 id %d" % (mapB)],
+                                    fail=False, include_stderr=True)
+    fail(ret == 0, "could reuse a map on a different ASIC")
+    fail(err.count("offload device mismatch between prog and map") == 0,
+         "error message missing for cross-ASIC map")
+
+    ret, _, err = bpftool_prog_load("sample_map_ret0.o", "/sys/fs/bpf/nsimB_",
+                                    dev=simB1['ifname'],
+                                    maps=["idx 0 id %d" % (mapA)],
+                                    fail=False, include_stderr=True)
+    fail(ret == 0, "could reuse a map on a different ASIC")
+    fail(err.count("offload device mismatch between prog and map") == 0,
+         "error message missing for cross-ASIC map")
+
+    start_test("Test multi-dev ASIC cross-dev destruction...")
+    bpftool_prog_list_wait(expected=2)
+
+    simA.remove()
+    bpftool_prog_list_wait(expected=1)
+
+    ifnameB = bpftool("prog show %s" % (progB))[1]["dev"]["ifname"]
+    fail(ifnameB != simB1['ifname'], "program not bound to originial device")
+    simB1.remove()
+    bpftool_prog_list_wait(expected=1)
+
+    start_test("Test multi-dev ASIC cross-dev destruction - move...")
+    ifnameB = bpftool("prog show %s" % (progB))[1]["dev"]["ifname"]
+    fail(ifnameB not in (simB2['ifname'], simB3['ifname']),
+         "program not bound to remaining devices")
+
+    simB2.remove()
+    ifnameB = bpftool("prog show %s" % (progB))[1]["dev"]["ifname"]
+    fail(ifnameB != simB3['ifname'], "program not bound to remaining device")
+
+    simB3.remove()
+    bpftool_prog_list_wait(expected=0)
+
+    start_test("Test multi-dev ASIC cross-dev destruction - orphaned...")
+    ret, out = bpftool("prog show %s" % (progB), fail=False)
+    fail(ret == 0, "got information about orphaned program")
+    fail("error" not in out, "no error reported for get info on orphaned")
+    fail(out["error"] != "can't get prog info: No such device",
+         "wrong error for get info on orphaned")
+
     print("%s: OK" % (os.path.basename(__file__)))
 
 finally:
-- 
2.17.1

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* Re: [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC
  2018-07-17  2:37 ` [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC Jakub Kicinski
@ 2018-07-17  3:00   ` Alexei Starovoitov
  2018-07-17  3:18     ` Jakub Kicinski
  0 siblings, 1 reply; 12+ messages in thread
From: Alexei Starovoitov @ 2018-07-17  3:00 UTC (permalink / raw)
  To: Jakub Kicinski; +Cc: daniel, oss-drivers, netdev

On Mon, Jul 16, 2018 at 07:37:20PM -0700, Jakub Kicinski wrote:
> Create a higher-level entity to represent a device/ASIC to allow
> programs and maps to be shared between device ports.  The extra
> work is required to make sure we don't destroy BPF objects as
> soon as the netdev for which they were loaded gets destroyed,
> as other ports may still be using them.  When netdev goes away
> all of its BPF objects will be moved to other netdevs of the
> device, and only destroyed when last netdev is unregistered.
> 
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
..
> -bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map)
> +static bool __bpf_offload_dev_match(struct bpf_prog *prog,
> +				    struct net_device *netdev)
>  {
> -	struct bpf_offloaded_map *offmap;
> +	struct bpf_offload_netdev *ondev1, *ondev2;
>  	struct bpf_prog_offload *offload;
> -	bool ret;
>  
>  	if (!bpf_prog_is_dev_bound(prog->aux))
>  		return false;
> -	if (!bpf_map_is_dev_bound(map))
> -		return bpf_map_offload_neutral(map);
>  
> -	down_read(&bpf_devs_lock);
>  	offload = prog->aux->offload;
> +	if (!offload)
> +		return false;
> +	if (offload->netdev == netdev)
> +		return true;
> +
> +	ondev1 = bpf_offload_find_netdev(offload->netdev);
> +	ondev2 = bpf_offload_find_netdev(netdev);
> +
> +	return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
> +}
> +
> +bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
> +{
> +	bool ret;
> +
> +	down_read(&bpf_devs_lock);
> +	ret = __bpf_offload_dev_match(prog, netdev);
> +	up_read(&bpf_devs_lock);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
> +
> +bool bpf_offload_match(struct bpf_prog *prog, struct bpf_map *map)
> +{
> +	struct bpf_offloaded_map *offmap;
> +	bool ret;
> +
> +	if (!bpf_map_is_dev_bound(map))
> +		return bpf_map_offload_neutral(map);
>  	offmap = map_to_offmap(map);
>  
> -	ret = offload && offload->netdev == offmap->netdev;
> +	down_read(&bpf_devs_lock);
> +	ret = __bpf_offload_dev_match(prog, offmap->netdev);
>  	up_read(&bpf_devs_lock);
>  
>  	return ret;
>  }
>  
..
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 9e2bf834f13a..2c5b923eef75 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -5054,7 +5054,7 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
>  	}
>  
>  	if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) &&
> -	    !bpf_offload_dev_match(prog, map)) {
> +	    !bpf_offload_match(prog, map)) {

I'm confused with new names and renaming.
May be split renaming into separate patch?
Should new bpf_offload_match() be called bpf_offload_prog_map_match() ?
or some other name?
May be adding comments to these functions will make it clear...

^ permalink raw reply	[flat|nested] 12+ messages in thread
* Re: [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC
  2018-07-17  3:00   ` Alexei Starovoitov
@ 2018-07-17  3:18     ` Jakub Kicinski
  0 siblings, 0 replies; 12+ messages in thread
From: Jakub Kicinski @ 2018-07-17  3:18 UTC (permalink / raw)
  To: Alexei Starovoitov; +Cc: daniel, oss-drivers, netdev

On Mon, 16 Jul 2018 20:00:47 -0700, Alexei Starovoitov wrote:
> On Mon, Jul 16, 2018 at 07:37:20PM -0700, Jakub Kicinski wrote:
> > Create a higher-level entity to represent a device/ASIC to allow
> > programs and maps to be shared between device ports.  The extra
> > work is required to make sure we don't destroy BPF objects as
> > soon as the netdev for which they were loaded gets destroyed,
> > as other ports may still be using them.  When netdev goes away
> > all of its BPF objects will be moved to other netdevs of the
> > device, and only destroyed when last netdev is unregistered.
> > 
> > Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> > Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>  
> ..
> > -bool bpf_offload_dev_match(struct bpf_prog *prog, struct bpf_map *map)
> > +static bool __bpf_offload_dev_match(struct bpf_prog *prog,
> > +				    struct net_device *netdev)
> >  {
> > -	struct bpf_offloaded_map *offmap;
> > +	struct bpf_offload_netdev *ondev1, *ondev2;
> >  	struct bpf_prog_offload *offload;
> > -	bool ret;
> >  
> >  	if (!bpf_prog_is_dev_bound(prog->aux))
> >  		return false;
> > -	if (!bpf_map_is_dev_bound(map))
> > -		return bpf_map_offload_neutral(map);
> >  
> > -	down_read(&bpf_devs_lock);
> >  	offload = prog->aux->offload;
> > +	if (!offload)
> > +		return false;
> > +	if (offload->netdev == netdev)
> > +		return true;
> > +
> > +	ondev1 = bpf_offload_find_netdev(offload->netdev);
> > +	ondev2 = bpf_offload_find_netdev(netdev);
> > +
> > +	return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
> > +}
> > +
> > +bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
> > +{
> > +	bool ret;
> > +
> > +	down_read(&bpf_devs_lock);
> > +	ret = __bpf_offload_dev_match(prog, netdev);
> > +	up_read(&bpf_devs_lock);
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
> > +
> > +bool bpf_offload_match(struct bpf_prog *prog, struct bpf_map *map)
> > +{
> > +	struct bpf_offloaded_map *offmap;
> > +	bool ret;
> > +
> > +	if (!bpf_map_is_dev_bound(map))
> > +		return bpf_map_offload_neutral(map);
> >  	offmap = map_to_offmap(map);
> >  
> > -	ret = offload && offload->netdev == offmap->netdev;
> > +	down_read(&bpf_devs_lock);
> > +	ret = __bpf_offload_dev_match(prog, offmap->netdev);
> >  	up_read(&bpf_devs_lock);
> >  
> >  	return ret;
> >  }
> >    
> ..
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 9e2bf834f13a..2c5b923eef75 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -5054,7 +5054,7 @@ static int check_map_prog_compatibility(struct bpf_verifier_env *env,
> >  	}
> >  
> >  	if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) &&
> > -	    !bpf_offload_dev_match(prog, map)) {
> > +	    !bpf_offload_match(prog, map)) {  
> 
> I'm confused with new names and renaming.
> May be split renaming into separate patch?
> Should new bpf_offload_match() be called bpf_offload_prog_map_match() ?
> or some other name?
> May be adding comments to these functions will make it clear...

It is messy.  The new functions to register/unregister ASIC are called
bpf_offload_dev_*, hence it seemed like a good idea to call the
function exported to the drivers bpf_offload_dev_match() (see patches 7
and 8) to keep the driver API consistent.  But then the old function
which is only used by the verivier has to be renamed.

I will use bpf_offload_prog_map_match() and split to a separate patch.

^ permalink raw reply	[flat|nested] 12+ messages in thread
end of thread, other threads:[~2018-07-17  3:49 UTC | newest]

Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-07-17  2:37 [PATCH bpf-next 0/9] bpf: offload program and map sharing Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 1/9] netdevsim: add switch_id attribute Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 2/9] netdevsim: add shared netdevsim devices Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 3/9] netdevsim: associate bound programs with shared dev Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 4/9] nfp: add .ndo_init() and .ndo_uninit() callbacks Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 5/9] bpf: offload: aggregate offloads per-device Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 6/9] bpf: offload: allow program and map sharing per-ASIC Jakub Kicinski
2018-07-17  3:00   ` Alexei Starovoitov
2018-07-17  3:18     ` Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 7/9] netdevsim: allow program sharing between devices Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 8/9] nfp: bpf: allow program sharing within ASIC Jakub Kicinski
2018-07-17  2:37 ` [PATCH bpf-next 9/9] selftests/bpf: add test for sharing objects between netdevs Jakub Kicinski

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.