[PATCH net v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()

[PATCH net v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()

[Lorenzo Bianconi]

airoha_ppe_deinit() runs airoha_npu_ppe_deinit() in atomic context.
airoha_npu_ppe_deinit routine allocates ppe_data buffer with GFP_KERNEL
flag. Rely on rcu_replace_pointer in airoha_ppe_deinit routine in order
to fix schedule while atomic issue in airoha_npu_ppe_deinit() since we
do not need atomic context there.

Fixes: 00a7678310fe3 ("net: airoha: Introduce flowtable offload support")
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
Changes in v2:
- Update commit log.
- Link to v1: https://lore.kernel.org/r/20251223-airoha-fw-ethtool-v1-1-1dbd1568c585@kernel.org
---
 drivers/net/ethernet/airoha/airoha_ppe.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/airoha/airoha_ppe.c b/drivers/net/ethernet/airoha/airoha_ppe.c
index 0caabb0c3aa06948e46ad087c50eba5babc81994..2221bafaf7c9fed73dcf8d30ff3b1c5eecc463d4 100644
--- a/drivers/net/ethernet/airoha/airoha_ppe.c
+++ b/drivers/net/ethernet/airoha/airoha_ppe.c
@@ -1547,13 +1547,16 @@ void airoha_ppe_deinit(struct airoha_eth *eth)
 {
 	struct airoha_npu *npu;
 
-	rcu_read_lock();
-	npu = rcu_dereference(eth->npu);
+	mutex_lock(&flow_offload_mutex);
+
+	npu = rcu_replace_pointer(eth->npu, NULL,
+				  lockdep_is_held(&flow_offload_mutex));
 	if (npu) {
 		npu->ops.ppe_deinit(npu);
 		airoha_npu_put(npu);
 	}
-	rcu_read_unlock();
+
+	mutex_unlock(&flow_offload_mutex);
 
 	rhashtable_destroy(&eth->ppe->l2_flows);
 	rhashtable_destroy(&eth->flow_table);

---
base-commit: c303e8b86d9dbd6868f5216272973292f7f3b7f1
change-id: 20251223-airoha-fw-ethtool-21f54b02c98b

Best regards,
-- 
Lorenzo Bianconi <lorenzo@kernel.org>

Re: [PATCH net v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net.git (main)
by Paolo Abeni <pabeni@redhat.com>:

On Mon, 05 Jan 2026 09:43:31 +0100 you wrote:
> airoha_ppe_deinit() runs airoha_npu_ppe_deinit() in atomic context.
> airoha_npu_ppe_deinit routine allocates ppe_data buffer with GFP_KERNEL
> flag. Rely on rcu_replace_pointer in airoha_ppe_deinit routine in order
> to fix schedule while atomic issue in airoha_npu_ppe_deinit() since we
> do not need atomic context there.
> 
> Fixes: 00a7678310fe3 ("net: airoha: Introduce flowtable offload support")
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> 
> [...]

Here is the summary with links:
  - [net,v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()
    https://git.kernel.org/netdev/net/c/6abcf751bc08

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html

Re: [PATCH net v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()

[Simon Horman]

On Mon, Jan 05, 2026 at 09:43:31AM +0100, Lorenzo Bianconi wrote:
> airoha_ppe_deinit() runs airoha_npu_ppe_deinit() in atomic context.
> airoha_npu_ppe_deinit routine allocates ppe_data buffer with GFP_KERNEL
> flag. Rely on rcu_replace_pointer in airoha_ppe_deinit routine in order
> to fix schedule while atomic issue in airoha_npu_ppe_deinit() since we
> do not need atomic context there.

Hi Lorenzo,

If I understand things correctly the key problem here is that
an allocation with GFP_KERNEL implies GFP_RECLAIM and thus may sleep.
But RCU read-side critical sections are not allowed to sleep in non-RT
kernels.

If so, I think it would be clearer to describe the problem along those
lines. But maybe it is just me.

> 
> Fixes: 00a7678310fe3 ("net: airoha: Introduce flowtable offload support")
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> ---
> Changes in v2:
> - Update commit log.
> - Link to v1: https://lore.kernel.org/r/20251223-airoha-fw-ethtool-v1-1-1dbd1568c585@kernel.org

...

Re: [PATCH net v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()

[Lorenzo Bianconi]

[-- Attachment #1: Type: text/plain, Size: 1205 bytes --]

> On Mon, Jan 05, 2026 at 09:43:31AM +0100, Lorenzo Bianconi wrote:
> > airoha_ppe_deinit() runs airoha_npu_ppe_deinit() in atomic context.
> > airoha_npu_ppe_deinit routine allocates ppe_data buffer with GFP_KERNEL
> > flag. Rely on rcu_replace_pointer in airoha_ppe_deinit routine in order
> > to fix schedule while atomic issue in airoha_npu_ppe_deinit() since we
> > do not need atomic context there.
> 
> Hi Lorenzo,

Hi Simon,

> 
> If I understand things correctly the key problem here is that
> an allocation with GFP_KERNEL implies GFP_RECLAIM and thus may sleep.
> But RCU read-side critical sections are not allowed to sleep in non-RT
> kernels.

yes, right, RCU section is atomic.

> 
> If so, I think it would be clearer to describe the problem along those
> lines. But maybe it is just me.

This patch is already in Linus's tree.

Regards,
Lorenzo

> 
> > 
> > Fixes: 00a7678310fe3 ("net: airoha: Introduce flowtable offload support")
> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > ---
> > Changes in v2:
> > - Update commit log.
> > - Link to v1: https://lore.kernel.org/r/20251223-airoha-fw-ethtool-v1-1-1dbd1568c585@kernel.org
> 
> ...

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH net v2] net: airoha: Fix schedule while atomic in airoha_ppe_deinit()

[Simon Horman]

On Fri, Jan 09, 2026 at 10:33:55AM +0100, Lorenzo Bianconi wrote:
> > On Mon, Jan 05, 2026 at 09:43:31AM +0100, Lorenzo Bianconi wrote:
> > > airoha_ppe_deinit() runs airoha_npu_ppe_deinit() in atomic context.
> > > airoha_npu_ppe_deinit routine allocates ppe_data buffer with GFP_KERNEL
> > > flag. Rely on rcu_replace_pointer in airoha_ppe_deinit routine in order
> > > to fix schedule while atomic issue in airoha_npu_ppe_deinit() since we
> > > do not need atomic context there.
> > 
> > Hi Lorenzo,
> 
> Hi Simon,
> 
> > 
> > If I understand things correctly the key problem here is that
> > an allocation with GFP_KERNEL implies GFP_RECLAIM and thus may sleep.
> > But RCU read-side critical sections are not allowed to sleep in non-RT
> > kernels.
> 
> yes, right, RCU section is atomic.
> 
> > 
> > If so, I think it would be clearer to describe the problem along those
> > lines. But maybe it is just me.
> 
> This patch is already in Linus's tree.

Yeah, sorry for missing that.