When net_shaper_ops are enabled for MANA, netdev_ops_lock
becomes active.
The netvsc sets up MANA VF via following call chain:
netvsc_vf_setup()
dev_change_flags()
...
__dev_open() OR __dev_close()
dev_change_flags() holds the netdev mutex via netdev_lock_ops.
During this process, mana_create_txq() and mana_create_rxq()
invoke netif_napi_add_tx(), netif_napi_add_weight(), and napi_enable(),
all of which attempt to acquire the same lock,
leading to a potential deadlock.
Similarly, mana_destroy_txq() and mana_destroy_rxq() call
netif_napi_disable() and netif_napi_del(), which also contend
for the same lock.
Switch to the _locked variants of these APIs to avoid deadlocks
when the netdev_ops_lock is held.
Fixes: d4c22ec680c8 ("net: hold netdev instance lock during ndo_open/ndo_stop")
Signed-off-by: Erni Sri Satya Vennela <ernis@linux.microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
---
drivers/net/ethernet/microsoft/mana/mana_en.c | 39 ++++++++++++++-----
1 file changed, 30 insertions(+), 9 deletions(-)
diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
index ccd2885c939e..3c879d8a39e3 100644
--- a/drivers/net/ethernet/microsoft/mana/mana_en.c
+++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
@@ -1911,8 +1911,13 @@ static void mana_destroy_txq(struct mana_port_context *apc)
napi = &apc->tx_qp[i].tx_cq.napi;
if (apc->tx_qp[i].txq.napi_initialized) {
napi_synchronize(napi);
- napi_disable(napi);
- netif_napi_del(napi);
+ if (netdev_need_ops_lock(napi->dev)) {
+ napi_disable_locked(napi);
+ netif_napi_del_locked(napi);
+ } else {
+ napi_disable(napi);
+ netif_napi_del(napi);
+ }
apc->tx_qp[i].txq.napi_initialized = false;
}
mana_destroy_wq_obj(apc, GDMA_SQ, apc->tx_qp[i].tx_object);
@@ -2064,8 +2069,14 @@ static int mana_create_txq(struct mana_port_context *apc,
mana_create_txq_debugfs(apc, i);
- netif_napi_add_tx(net, &cq->napi, mana_poll);
- napi_enable(&cq->napi);
+ if (netdev_need_ops_lock(net)) {
+ set_bit(NAPI_STATE_NO_BUSY_POLL, &cq->napi.state);
+ netif_napi_add_locked(net, &cq->napi, mana_poll);
+ napi_enable_locked(&cq->napi);
+ } else {
+ netif_napi_add_tx(net, &cq->napi, mana_poll);
+ napi_enable(&cq->napi);
+ }
txq->napi_initialized = true;
mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
@@ -2101,9 +2112,13 @@ static void mana_destroy_rxq(struct mana_port_context *apc,
if (napi_initialized) {
napi_synchronize(napi);
- napi_disable(napi);
-
- netif_napi_del(napi);
+ if (netdev_need_ops_lock(napi->dev)) {
+ napi_disable_locked(napi);
+ netif_napi_del_locked(napi);
+ } else {
+ napi_disable(napi);
+ netif_napi_del(napi);
+ }
}
xdp_rxq_info_unreg(&rxq->xdp_rxq);
@@ -2354,14 +2369,20 @@ static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
gc->cq_table[cq->gdma_id] = cq->gdma_cq;
- netif_napi_add_weight(ndev, &cq->napi, mana_poll, 1);
+ if (netdev_need_ops_lock(ndev))
+ netif_napi_add_weight_locked(ndev, &cq->napi, mana_poll, 1);
+ else
+ netif_napi_add_weight(ndev, &cq->napi, mana_poll, 1);
WARN_ON(xdp_rxq_info_reg(&rxq->xdp_rxq, ndev, rxq_idx,
cq->napi.napi_id));
WARN_ON(xdp_rxq_info_reg_mem_model(&rxq->xdp_rxq, MEM_TYPE_PAGE_POOL,
rxq->page_pool));
- napi_enable(&cq->napi);
+ if (netdev_need_ops_lock(ndev))
+ napi_enable_locked(&cq->napi);
+ else
+ napi_enable(&cq->napi);
mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
out:
--
2.34.1On Wed, Jun 11, 2025 at 01:46:13AM -0700, Erni Sri Satya Vennela wrote:
> When net_shaper_ops are enabled for MANA, netdev_ops_lock
> becomes active.
>
> The netvsc sets up MANA VF via following call chain:
>
> netvsc_vf_setup()
> dev_change_flags()
> ...
> __dev_open() OR __dev_close()
>
> dev_change_flags() holds the netdev mutex via netdev_lock_ops.
>
> During this process, mana_create_txq() and mana_create_rxq()
> invoke netif_napi_add_tx(), netif_napi_add_weight(), and napi_enable(),
> all of which attempt to acquire the same lock,
> leading to a potential deadlock.
commit message could be better oriented.
>
> Similarly, mana_destroy_txq() and mana_destroy_rxq() call
> netif_napi_disable() and netif_napi_del(), which also contend
> for the same lock.
>
> Switch to the _locked variants of these APIs to avoid deadlocks
> when the netdev_ops_lock is held.
>
> Fixes: d4c22ec680c8 ("net: hold netdev instance lock during ndo_open/ndo_stop")
> Signed-off-by: Erni Sri Satya Vennela <ernis@linux.microsoft.com>
> Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
> Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
> ---
> drivers/net/ethernet/microsoft/mana/mana_en.c | 39 ++++++++++++++-----
> 1 file changed, 30 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
> index ccd2885c939e..3c879d8a39e3 100644
> --- a/drivers/net/ethernet/microsoft/mana/mana_en.c
> +++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
> @@ -1911,8 +1911,13 @@ static void mana_destroy_txq(struct mana_port_context *apc)
> napi = &apc->tx_qp[i].tx_cq.napi;
> if (apc->tx_qp[i].txq.napi_initialized) {
> napi_synchronize(napi);
> - napi_disable(napi);
> - netif_napi_del(napi);
> + if (netdev_need_ops_lock(napi->dev)) {
> + napi_disable_locked(napi);
> + netif_napi_del_locked(napi);
> + } else {
> + napi_disable(napi);
> + netif_napi_del(napi);
> + }
Instead of using if-else, we can used netdev_lock_ops(), followed by *_locked api-s.
Same for rest of the patch.
Reviewed-by: Saurabh Sengar <ssengar@linux.microsoft.com>
- Saurabh
On Wed, Jun 11, 2025 at 04:03:52AM -0700, Saurabh Singh Sengar wrote:
> On Wed, Jun 11, 2025 at 01:46:13AM -0700, Erni Sri Satya Vennela wrote:
> > When net_shaper_ops are enabled for MANA, netdev_ops_lock
> > becomes active.
> >
> > The netvsc sets up MANA VF via following call chain:
> >
> > netvsc_vf_setup()
> > dev_change_flags()
> > ...
> > __dev_open() OR __dev_close()
> >
> > dev_change_flags() holds the netdev mutex via netdev_lock_ops.
> >
> > During this process, mana_create_txq() and mana_create_rxq()
> > invoke netif_napi_add_tx(), netif_napi_add_weight(), and napi_enable(),
> > all of which attempt to acquire the same lock,
> > leading to a potential deadlock.
>
> commit message could be better oriented.
>
> >
> > Similarly, mana_destroy_txq() and mana_destroy_rxq() call
> > netif_napi_disable() and netif_napi_del(), which also contend
> > for the same lock.
> >
> > Switch to the _locked variants of these APIs to avoid deadlocks
> > when the netdev_ops_lock is held.
> >
> > Fixes: d4c22ec680c8 ("net: hold netdev instance lock during ndo_open/ndo_stop")
> > Signed-off-by: Erni Sri Satya Vennela <ernis@linux.microsoft.com>
> > Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
> > Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
> > ---
> > drivers/net/ethernet/microsoft/mana/mana_en.c | 39 ++++++++++++++-----
> > 1 file changed, 30 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
> > index ccd2885c939e..3c879d8a39e3 100644
> > --- a/drivers/net/ethernet/microsoft/mana/mana_en.c
> > +++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
> > @@ -1911,8 +1911,13 @@ static void mana_destroy_txq(struct mana_port_context *apc)
> > napi = &apc->tx_qp[i].tx_cq.napi;
> > if (apc->tx_qp[i].txq.napi_initialized) {
> > napi_synchronize(napi);
> > - napi_disable(napi);
> > - netif_napi_del(napi);
> > + if (netdev_need_ops_lock(napi->dev)) {
> > + napi_disable_locked(napi);
> > + netif_napi_del_locked(napi);
> > + } else {
> > + napi_disable(napi);
> > + netif_napi_del(napi);
> > + }
>
> Instead of using if-else, we can used netdev_lock_ops(), followed by *_locked api-s.
> Same for rest of the patch.
>
I later realized that what we actually need is:
if (!netdev_need_ops_lock(napi->dev))
netdev_lock(dev);
not
if (netdev_need_ops_lock(napi->dev))
netdev_lock(dev);
Hence, netdev_lock_ops() is not appropriate. Instead, netdev_lock_ops_to_full()
seems to be a better choice.
> Reviewed-by: Saurabh Sengar <ssengar@linux.microsoft.com>
>
> - Saurabh
>
© 2016 - 2025 Red Hat, Inc.