* [net 1/8] net/mlx5e: Use the correct delete call on offloaded TC encap entry detach
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 2/8] net/sched: act_csum: Add accessors for offloading drivers Saeed Mahameed
` (7 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Or Gerlitz, Saeed Mahameed
From: Or Gerlitz <ogerlitz@mellanox.com>
We wrongly direcly invoke hlist_del_rcu() and not hash_del_rcu() which
does a slightly different call now and may change later, fix that.
Fixes: a54e20b4fcae ('net/mlx5e: Add basic TC tunnel set action for SRIOV offloads')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reported-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
index 11c27e4fadf6..a90dd26ea51c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
@@ -384,7 +384,7 @@ static void mlx5e_detach_encap(struct mlx5e_priv *priv,
if (e->flags & MLX5_ENCAP_ENTRY_VALID)
mlx5_encap_dealloc(priv->mdev, e->encap_id);
- hlist_del_rcu(&e->encap_hlist);
+ hash_del_rcu(&e->encap_hlist);
kfree(e->encap_header);
kfree(e);
}
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 2/8] net/sched: act_csum: Add accessors for offloading drivers
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
2017-05-23 14:16 ` [net 1/8] net/mlx5e: Use the correct delete call on offloaded TC encap entry detach Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 3/8] net/mlx5e: Allow TC csum offload if applied together with pedit action Saeed Mahameed
` (6 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Or Gerlitz, Saeed Mahameed
From: Or Gerlitz <ogerlitz@mellanox.com>
Add the accessors for realizing if this is a csum action,
and for which fields checksum is needed.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
include/net/tc_act/tc_csum.h | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/include/net/tc_act/tc_csum.h b/include/net/tc_act/tc_csum.h
index f31fb6331a53..3248beaf16b0 100644
--- a/include/net/tc_act/tc_csum.h
+++ b/include/net/tc_act/tc_csum.h
@@ -3,6 +3,7 @@
#include <linux/types.h>
#include <net/act_api.h>
+#include <linux/tc_act/tc_csum.h>
struct tcf_csum {
struct tc_action common;
@@ -11,4 +12,18 @@ struct tcf_csum {
};
#define to_tcf_csum(a) ((struct tcf_csum *)a)
+static inline bool is_tcf_csum(const struct tc_action *a)
+{
+#ifdef CONFIG_NET_CLS_ACT
+ if (a->ops && a->ops->type == TCA_ACT_CSUM)
+ return true;
+#endif
+ return false;
+}
+
+static inline u32 tcf_csum_update_flags(const struct tc_action *a)
+{
+ return to_tcf_csum(a)->update_flags;
+}
+
#endif /* __NET_TC_CSUM_H */
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 3/8] net/mlx5e: Allow TC csum offload if applied together with pedit action
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
2017-05-23 14:16 ` [net 1/8] net/mlx5e: Use the correct delete call on offloaded TC encap entry detach Saeed Mahameed
2017-05-23 14:16 ` [net 2/8] net/sched: act_csum: Add accessors for offloading drivers Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 4/8] net/mlx5e: Properly enforce disallowing of partial field re-write offload Saeed Mahameed
` (5 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Or Gerlitz, Saeed Mahameed
From: Or Gerlitz <ogerlitz@mellanox.com>
When offloading header re-writes, the HW re-calculates the relevant L3/L4
checksums. Hence, when upper layers (as done by OVS) ask for TC checksum action
offload together with pedit offload, don't err. This command now works:
tc filter add dev ens1f0 protocol ip parent ffff: prio 20 flower skip_sw
ip_proto tcp dst_port 9001
action pedit ex munge tcp dport set 0x1234 pipe
action csum tcp
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reported-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c | 39 +++++++++++++++++++++++++
1 file changed, 39 insertions(+)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
index a90dd26ea51c..9dd83c7e4c51 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
@@ -43,6 +43,7 @@
#include <net/tc_act/tc_vlan.h>
#include <net/tc_act/tc_tunnel_key.h>
#include <net/tc_act/tc_pedit.h>
+#include <net/tc_act/tc_csum.h>
#include <net/vxlan.h>
#include <net/arp.h>
#include "en.h"
@@ -1109,6 +1110,28 @@ static int parse_tc_pedit_action(struct mlx5e_priv *priv,
return err;
}
+static bool csum_offload_supported(struct mlx5e_priv *priv, u32 action, u32 update_flags)
+{
+ u32 prot_flags = TCA_CSUM_UPDATE_FLAG_IPV4HDR | TCA_CSUM_UPDATE_FLAG_TCP |
+ TCA_CSUM_UPDATE_FLAG_UDP;
+
+ /* The HW recalcs checksums only if re-writing headers */
+ if (!(action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)) {
+ netdev_warn(priv->netdev,
+ "TC csum action is only offloaded with pedit\n");
+ return false;
+ }
+
+ if (update_flags & ~prot_flags) {
+ netdev_warn(priv->netdev,
+ "can't offload TC csum action for some header/s - flags %#x\n",
+ update_flags);
+ return false;
+ }
+
+ return true;
+}
+
static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
struct mlx5e_tc_flow_parse_attr *parse_attr,
struct mlx5e_tc_flow *flow)
@@ -1149,6 +1172,14 @@ static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
continue;
}
+ if (is_tcf_csum(a)) {
+ if (csum_offload_supported(priv, attr->action,
+ tcf_csum_update_flags(a)))
+ continue;
+
+ return -EOPNOTSUPP;
+ }
+
if (is_tcf_skbedit_mark(a)) {
u32 mark = tcf_skbedit_mark(a);
@@ -1651,6 +1682,14 @@ static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
continue;
}
+ if (is_tcf_csum(a)) {
+ if (csum_offload_supported(priv, attr->action,
+ tcf_csum_update_flags(a)))
+ continue;
+
+ return -EOPNOTSUPP;
+ }
+
if (is_tcf_mirred_egress_redirect(a)) {
int ifindex = tcf_mirred_ifindex(a);
struct net_device *out_dev, *encap_dev = NULL;
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 4/8] net/mlx5e: Properly enforce disallowing of partial field re-write offload
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
` (2 preceding siblings ...)
2017-05-23 14:16 ` [net 3/8] net/mlx5e: Allow TC csum offload if applied together with pedit action Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 5/8] net/mlx5e: Fix warnings around parsing of TC pedit actions Saeed Mahameed
` (4 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Or Gerlitz, Saeed Mahameed
From: Or Gerlitz <ogerlitz@mellanox.com>
Currently we don't support partial header re-writes through TC pedit
action offloading. However, the code that enforces that wasn't err-ing
on cases where the first and last bits of the mask are set but there is
some zero bit between them, such as in the below example, fix that!
tc filter add dev enp1s0 protocol ip parent ffff: prio 10 flower
ip_proto udp dst_port 2001 skip_sw
action pedit munge ip src set 1.0.0.1 retain 0xff0000ff
Fixes: d79b6df6b10a ('net/mlx5e: Add parsing of TC pedit actions to HW format')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
index 9dd83c7e4c51..0387c321f0a2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
@@ -926,7 +926,7 @@ static int offload_pedit_fields(struct pedit_headers *masks,
struct mlx5e_tc_flow_parse_attr *parse_attr)
{
struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
- int i, action_size, nactions, max_actions, first, last;
+ int i, action_size, nactions, max_actions, first, last, first_z;
void *s_masks_p, *a_masks_p, *vals_p;
u32 s_mask, a_mask, val;
struct mlx5_fields *f;
@@ -985,9 +985,10 @@ static int offload_pedit_fields(struct pedit_headers *masks,
memcpy(&val, vals_p, f->size);
field_bsize = f->size * BITS_PER_BYTE;
+ first_z = find_first_zero_bit(&mask, field_bsize);
first = find_first_bit(&mask, field_bsize);
last = find_last_bit(&mask, field_bsize);
- if (first > 0 || last != (field_bsize - 1)) {
+ if (first > 0 || last != (field_bsize - 1) || first_z < last) {
printk(KERN_WARNING "mlx5: partial rewrite (mask %lx) is currently not offloaded\n",
mask);
return -EOPNOTSUPP;
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 5/8] net/mlx5e: Fix warnings around parsing of TC pedit actions
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
` (3 preceding siblings ...)
2017-05-23 14:16 ` [net 4/8] net/mlx5e: Properly enforce disallowing of partial field re-write offload Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 6/8] net/mlx5e: IPoIB, handle RX packet correctly Saeed Mahameed
` (3 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Or Gerlitz, Saeed Mahameed
From: Or Gerlitz <ogerlitz@mellanox.com>
The sparse tool emits these correct complaints:
drivers/net/ethernet/mellanox/mlx5/core//en_tc.c:1005:25: warning: cast to restricted __be32
drivers/net/ethernet/mellanox/mlx5/core//en_tc.c:1007:25: warning: cast to restricted __be16
The value is provided from user-space in network order, but there's
no way for them to realize that, avoid the warnings by casting to the
appropriate type.
Fixes: d79b6df6b10a ('net/mlx5e: Add parsing of TC pedit actions to HW format')
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reported-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
index 0387c321f0a2..ec63158ab643 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
@@ -928,9 +928,9 @@ static int offload_pedit_fields(struct pedit_headers *masks,
struct pedit_headers *set_masks, *add_masks, *set_vals, *add_vals;
int i, action_size, nactions, max_actions, first, last, first_z;
void *s_masks_p, *a_masks_p, *vals_p;
- u32 s_mask, a_mask, val;
struct mlx5_fields *f;
u8 cmd, field_bsize;
+ u32 s_mask, a_mask;
unsigned long mask;
void *action;
@@ -947,7 +947,8 @@ static int offload_pedit_fields(struct pedit_headers *masks,
for (i = 0; i < ARRAY_SIZE(fields); i++) {
f = &fields[i];
/* avoid seeing bits set from previous iterations */
- s_mask = a_mask = mask = val = 0;
+ s_mask = 0;
+ a_mask = 0;
s_masks_p = (void *)set_masks + f->offset;
a_masks_p = (void *)add_masks + f->offset;
@@ -982,9 +983,8 @@ static int offload_pedit_fields(struct pedit_headers *masks,
memset(a_masks_p, 0, f->size);
}
- memcpy(&val, vals_p, f->size);
-
field_bsize = f->size * BITS_PER_BYTE;
+
first_z = find_first_zero_bit(&mask, field_bsize);
first = find_first_bit(&mask, field_bsize);
last = find_last_bit(&mask, field_bsize);
@@ -1004,11 +1004,11 @@ static int offload_pedit_fields(struct pedit_headers *masks,
}
if (field_bsize == 32)
- MLX5_SET(set_action_in, action, data, ntohl(val));
+ MLX5_SET(set_action_in, action, data, ntohl(*(__be32 *)vals_p));
else if (field_bsize == 16)
- MLX5_SET(set_action_in, action, data, ntohs(val));
+ MLX5_SET(set_action_in, action, data, ntohs(*(__be16 *)vals_p));
else if (field_bsize == 8)
- MLX5_SET(set_action_in, action, data, val);
+ MLX5_SET(set_action_in, action, data, *(u8 *)vals_p);
action += action_size;
nactions++;
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 6/8] net/mlx5e: IPoIB, handle RX packet correctly
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
` (4 preceding siblings ...)
2017-05-23 14:16 ` [net 5/8] net/mlx5e: Fix warnings around parsing of TC pedit actions Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 7/8] net/mlx5: Avoid using pending command interface slots Saeed Mahameed
` (2 subsequent siblings)
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Erez Shitrit, Saeed Mahameed
From: Erez Shitrit <erezsh@mellanox.com>
IPoIB packet contains the pseudo header area, we need to pull it prior
to reset_mac_header in order to let the GRO work well.
In more details:
GRO checks the mac address of the new coming packet, it does that by
comparing the hard_header_len size of the current packet to the previous
one in that session, the comparison is over hard_header_len size.
Now, the driver prepares that area in the skb by allocating area from the
reserved part and resetting the correct mac header to it.
Fixes: 9d6bd752c63c ("net/mlx5e: IPoIB, RX handler")
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 7b1566f0ae58..66b5fec15313 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -1041,6 +1041,8 @@ void mlx5e_free_xdpsq_descs(struct mlx5e_xdpsq *sq)
#define MLX5_IB_GRH_BYTES 40
#define MLX5_IPOIB_ENCAP_LEN 4
#define MLX5_GID_SIZE 16
+#define MLX5_IPOIB_PSEUDO_LEN 20
+#define MLX5_IPOIB_HARD_LEN (MLX5_IPOIB_PSEUDO_LEN + MLX5_IPOIB_ENCAP_LEN)
static inline void mlx5i_complete_rx_cqe(struct mlx5e_rq *rq,
struct mlx5_cqe64 *cqe,
@@ -1048,6 +1050,7 @@ static inline void mlx5i_complete_rx_cqe(struct mlx5e_rq *rq,
struct sk_buff *skb)
{
struct net_device *netdev = rq->netdev;
+ char *pseudo_header;
u8 *dgid;
u8 g;
@@ -1076,8 +1079,11 @@ static inline void mlx5i_complete_rx_cqe(struct mlx5e_rq *rq,
if (likely(netdev->features & NETIF_F_RXHASH))
mlx5e_skb_set_hash(cqe, skb);
+ /* 20 bytes of ipoib header and 4 for encap existing */
+ pseudo_header = skb_push(skb, MLX5_IPOIB_PSEUDO_LEN);
+ memset(pseudo_header, 0, MLX5_IPOIB_PSEUDO_LEN);
skb_reset_mac_header(skb);
- skb_pull(skb, MLX5_IPOIB_ENCAP_LEN);
+ skb_pull(skb, MLX5_IPOIB_HARD_LEN);
skb->dev = netdev;
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 7/8] net/mlx5: Avoid using pending command interface slots
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
` (5 preceding siblings ...)
2017-05-23 14:16 ` [net 6/8] net/mlx5e: IPoIB, handle RX packet correctly Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-23 14:16 ` [net 8/8] net/mlx5: Tolerate irq_set_affinity_hint() failures Saeed Mahameed
2017-05-24 19:45 ` [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 David Miller
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Mohamad Haj Yahia, kernel-team, Saeed Mahameed
From: Mohamad Haj Yahia <mohamad@mellanox.com>
Currently when firmware command gets stuck or it takes long time to
complete, the driver command will get timeout and the command slot is
freed and can be used for new commands, and if the firmware receive new
command on the old busy slot its behavior is unexpected and this could
be harmful.
To fix this when the driver command gets timeout we return failure,
but we don't free the command slot and we wait for the firmware to
explicitly respond to that command.
Once all the entries are busy we will stop processing new firmware
commands.
Fixes: 9cba4ebcf374 ('net/mlx5: Fix potential deadlock in command mode change')
Signed-off-by: Mohamad Haj Yahia <mohamad@mellanox.com>
Cc: kernel-team@fb.com
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/cmd.c | 41 +++++++++++++++++++++---
drivers/net/ethernet/mellanox/mlx5/core/eq.c | 2 +-
drivers/net/ethernet/mellanox/mlx5/core/health.c | 2 +-
include/linux/mlx5/driver.h | 7 +++-
4 files changed, 44 insertions(+), 8 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/cmd.c b/drivers/net/ethernet/mellanox/mlx5/core/cmd.c
index 5bdaf3d545b2..10d282841f5b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/cmd.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/cmd.c
@@ -774,7 +774,7 @@ static void cb_timeout_handler(struct work_struct *work)
mlx5_core_warn(dev, "%s(0x%x) timeout. Will cause a leak of a command resource\n",
mlx5_command_str(msg_to_opcode(ent->in)),
msg_to_opcode(ent->in));
- mlx5_cmd_comp_handler(dev, 1UL << ent->idx);
+ mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
}
static void cmd_work_handler(struct work_struct *work)
@@ -804,6 +804,7 @@ static void cmd_work_handler(struct work_struct *work)
}
cmd->ent_arr[ent->idx] = ent;
+ set_bit(MLX5_CMD_ENT_STATE_PENDING_COMP, &ent->state);
lay = get_inst(cmd, ent->idx);
ent->lay = lay;
memset(lay, 0, sizeof(*lay));
@@ -825,6 +826,20 @@ static void cmd_work_handler(struct work_struct *work)
if (ent->callback)
schedule_delayed_work(&ent->cb_timeout_work, cb_timeout);
+ /* Skip sending command to fw if internal error */
+ if (pci_channel_offline(dev->pdev) ||
+ dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
+ u8 status = 0;
+ u32 drv_synd;
+
+ ent->ret = mlx5_internal_err_ret_value(dev, msg_to_opcode(ent->in), &drv_synd, &status);
+ MLX5_SET(mbox_out, ent->out, status, status);
+ MLX5_SET(mbox_out, ent->out, syndrome, drv_synd);
+
+ mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
+ return;
+ }
+
/* ring doorbell after the descriptor is valid */
mlx5_core_dbg(dev, "writing 0x%x to command doorbell\n", 1 << ent->idx);
wmb();
@@ -835,7 +850,7 @@ static void cmd_work_handler(struct work_struct *work)
poll_timeout(ent);
/* make sure we read the descriptor after ownership is SW */
rmb();
- mlx5_cmd_comp_handler(dev, 1UL << ent->idx);
+ mlx5_cmd_comp_handler(dev, 1UL << ent->idx, (ent->ret == -ETIMEDOUT));
}
}
@@ -879,7 +894,7 @@ static int wait_func(struct mlx5_core_dev *dev, struct mlx5_cmd_work_ent *ent)
wait_for_completion(&ent->done);
} else if (!wait_for_completion_timeout(&ent->done, timeout)) {
ent->ret = -ETIMEDOUT;
- mlx5_cmd_comp_handler(dev, 1UL << ent->idx);
+ mlx5_cmd_comp_handler(dev, 1UL << ent->idx, true);
}
err = ent->ret;
@@ -1375,7 +1390,7 @@ static void free_msg(struct mlx5_core_dev *dev, struct mlx5_cmd_msg *msg)
}
}
-void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec)
+void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec, bool forced)
{
struct mlx5_cmd *cmd = &dev->cmd;
struct mlx5_cmd_work_ent *ent;
@@ -1395,6 +1410,19 @@ void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec)
struct semaphore *sem;
ent = cmd->ent_arr[i];
+
+ /* if we already completed the command, ignore it */
+ if (!test_and_clear_bit(MLX5_CMD_ENT_STATE_PENDING_COMP,
+ &ent->state)) {
+ /* only real completion can free the cmd slot */
+ if (!forced) {
+ mlx5_core_err(dev, "Command completion arrived after timeout (entry idx = %d).\n",
+ ent->idx);
+ free_ent(cmd, ent->idx);
+ }
+ continue;
+ }
+
if (ent->callback)
cancel_delayed_work(&ent->cb_timeout_work);
if (ent->page_queue)
@@ -1417,7 +1445,10 @@ void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec)
mlx5_core_dbg(dev, "command completed. ret 0x%x, delivery status %s(0x%x)\n",
ent->ret, deliv_status_to_str(ent->status), ent->status);
}
- free_ent(cmd, ent->idx);
+
+ /* only real completion will free the entry slot */
+ if (!forced)
+ free_ent(cmd, ent->idx);
if (ent->callback) {
ds = ent->ts2 - ent->ts1;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eq.c b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
index ea5d8d37a75c..33eae5ad2fb0 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/eq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/eq.c
@@ -422,7 +422,7 @@ static irqreturn_t mlx5_eq_int(int irq, void *eq_ptr)
break;
case MLX5_EVENT_TYPE_CMD:
- mlx5_cmd_comp_handler(dev, be32_to_cpu(eqe->data.cmd.vector));
+ mlx5_cmd_comp_handler(dev, be32_to_cpu(eqe->data.cmd.vector), false);
break;
case MLX5_EVENT_TYPE_PORT_CHANGE:
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/health.c b/drivers/net/ethernet/mellanox/mlx5/core/health.c
index d0515391d33b..44f59b1d6f0f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/health.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/health.c
@@ -90,7 +90,7 @@ static void trigger_cmd_completions(struct mlx5_core_dev *dev)
spin_unlock_irqrestore(&dev->cmd.alloc_lock, flags);
mlx5_core_dbg(dev, "vector 0x%llx\n", vector);
- mlx5_cmd_comp_handler(dev, vector);
+ mlx5_cmd_comp_handler(dev, vector, true);
return;
no_trig:
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index bcdf739ee41a..93273d9ea4d1 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -787,7 +787,12 @@ enum {
typedef void (*mlx5_cmd_cbk_t)(int status, void *context);
+enum {
+ MLX5_CMD_ENT_STATE_PENDING_COMP,
+};
+
struct mlx5_cmd_work_ent {
+ unsigned long state;
struct mlx5_cmd_msg *in;
struct mlx5_cmd_msg *out;
void *uout;
@@ -976,7 +981,7 @@ void mlx5_cq_completion(struct mlx5_core_dev *dev, u32 cqn);
void mlx5_rsc_event(struct mlx5_core_dev *dev, u32 rsn, int event_type);
void mlx5_srq_event(struct mlx5_core_dev *dev, u32 srqn, int event_type);
struct mlx5_core_srq *mlx5_core_get_srq(struct mlx5_core_dev *dev, u32 srqn);
-void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec);
+void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vec, bool forced);
void mlx5_cq_event(struct mlx5_core_dev *dev, u32 cqn, int event_type);
int mlx5_create_map_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq, u8 vecidx,
int nent, u64 mask, const char *name,
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [net 8/8] net/mlx5: Tolerate irq_set_affinity_hint() failures
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
` (6 preceding siblings ...)
2017-05-23 14:16 ` [net 7/8] net/mlx5: Avoid using pending command interface slots Saeed Mahameed
@ 2017-05-23 14:16 ` Saeed Mahameed
2017-05-24 19:45 ` [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 David Miller
8 siblings, 0 replies; 10+ messages in thread
From: Saeed Mahameed @ 2017-05-23 14:16 UTC (permalink / raw)
To: David S. Miller; +Cc: netdev, Tariq Toukan, kernel-team, Saeed Mahameed
From: Tariq Toukan <tariqt@mellanox.com>
Add tolerance to failures of irq_set_affinity_hint().
Its role is to give hints that optimizes performance,
and should not block the driver load.
In non-SMP systems, functionality is not available as
there is a single core, and all these calls definitely
fail. Hence, do not call the function and avoid the
warning prints.
Fixes: db058a186f98 ("net/mlx5_core: Set irq affinity hints")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Cc: kernel-team@fb.com
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
drivers/net/ethernet/mellanox/mlx5/core/main.c | 15 ++++-----------
1 file changed, 4 insertions(+), 11 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 0c123d571b4c..fe5546bb4153 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -612,7 +612,6 @@ static int mlx5_irq_set_affinity_hint(struct mlx5_core_dev *mdev, int i)
struct mlx5_priv *priv = &mdev->priv;
struct msix_entry *msix = priv->msix_arr;
int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
- int err;
if (!zalloc_cpumask_var(&priv->irq_info[i].mask, GFP_KERNEL)) {
mlx5_core_warn(mdev, "zalloc_cpumask_var failed");
@@ -622,18 +621,12 @@ static int mlx5_irq_set_affinity_hint(struct mlx5_core_dev *mdev, int i)
cpumask_set_cpu(cpumask_local_spread(i, priv->numa_node),
priv->irq_info[i].mask);
- err = irq_set_affinity_hint(irq, priv->irq_info[i].mask);
- if (err) {
- mlx5_core_warn(mdev, "irq_set_affinity_hint failed,irq 0x%.4x",
- irq);
- goto err_clear_mask;
- }
+#ifdef CONFIG_SMP
+ if (irq_set_affinity_hint(irq, priv->irq_info[i].mask))
+ mlx5_core_warn(mdev, "irq_set_affinity_hint failed, irq 0x%.4x", irq);
+#endif
return 0;
-
-err_clear_mask:
- free_cpumask_var(priv->irq_info[i].mask);
- return err;
}
static void mlx5_irq_clear_affinity_hint(struct mlx5_core_dev *mdev, int i)
--
2.11.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23
2017-05-23 14:16 [pull request][net 0/8] Mellanox, mlx5 fixes 2017-05-23 Saeed Mahameed
` (7 preceding siblings ...)
2017-05-23 14:16 ` [net 8/8] net/mlx5: Tolerate irq_set_affinity_hint() failures Saeed Mahameed
@ 2017-05-24 19:45 ` David Miller
8 siblings, 0 replies; 10+ messages in thread
From: David Miller @ 2017-05-24 19:45 UTC (permalink / raw)
To: saeedm; +Cc: netdev
From: Saeed Mahameed <saeedm@mellanox.com>
Date: Tue, 23 May 2017 17:16:02 +0300
> This series contains some fixes for the mlx5 driver and one small patch that
> adds csum actions accessors in include/net/tc_act/tc_csum.h needed by some of
> mlx5 fixes patches.
>
> Details are below.
>
> Please pull and let me know if there's any problem.
>
> Note: This series doesn't introduce any merge conflict with the ongoing mlx5
> for-next submission.
>
> For -stable kernels >= 4.7:
> ("net/mlx5: Avoid using pending command interface slots")
> ("net/mlx5: Tolerate irq_set_affinity_hint() failures")
Pulled and queued up for -stable, thanks.
^ permalink raw reply [flat|nested] 10+ messages in thread