Re: [PATCH v3] net: ks8851: Fix deadlock with the SPI chip variant

[Simon Horman <horms@kernel.org>]

On Sat, Jul 06, 2024 at 12:13:37PM +0200, Ronald Wahl wrote:
> From: Ronald Wahl <ronald.wahl@raritan.com>
> 
> When SMP is enabled and spinlocks are actually functional then there is
> a deadlock with the 'statelock' spinlock between ks8851_start_xmit_spi
> and ks8851_irq:
> 
>     watchdog: BUG: soft lockup - CPU#0 stuck for 27s!
>     call trace:
>       queued_spin_lock_slowpath+0x100/0x284
>       do_raw_spin_lock+0x34/0x44
>       ks8851_start_xmit_spi+0x30/0xb8
>       ks8851_start_xmit+0x14/0x20
>       netdev_start_xmit+0x40/0x6c
>       dev_hard_start_xmit+0x6c/0xbc
>       sch_direct_xmit+0xa4/0x22c
>       __qdisc_run+0x138/0x3fc
>       qdisc_run+0x24/0x3c
>       net_tx_action+0xf8/0x130
>       handle_softirqs+0x1ac/0x1f0
>       __do_softirq+0x14/0x20
>       ____do_softirq+0x10/0x1c
>       call_on_irq_stack+0x3c/0x58
>       do_softirq_own_stack+0x1c/0x28
>       __irq_exit_rcu+0x54/0x9c
>       irq_exit_rcu+0x10/0x1c
>       el1_interrupt+0x38/0x50
>       el1h_64_irq_handler+0x18/0x24
>       el1h_64_irq+0x64/0x68
>       __netif_schedule+0x6c/0x80
>       netif_tx_wake_queue+0x38/0x48
>       ks8851_irq+0xb8/0x2c8
>       irq_thread_fn+0x2c/0x74
>       irq_thread+0x10c/0x1b0
>       kthread+0xc8/0xd8
>       ret_from_fork+0x10/0x20
> 
> This issue has not been identified earlier because tests were done on
> a device with SMP disabled and so spinlocks were actually NOPs.
> 
> Now use spin_(un)lock_bh for TX queue related locking to avoid execution
> of softirq work synchronously that would lead to a deadlock.
> 
> Fixes: 3dc5d4454545 ("net: ks8851: Fix TX stall caused by TX buffer overrun")
> Cc: "David S. Miller" <davem@davemloft.net>
> Cc: Eric Dumazet <edumazet@google.com>
> Cc: Jakub Kicinski <kuba@kernel.org>
> Cc: Paolo Abeni <pabeni@redhat.com>
> Cc: Simon Horman <horms@kernel.org>
> Cc: netdev@vger.kernel.org
> Cc: stable@vger.kernel.org # 5.10+
> Signed-off-by: Ronald Wahl <ronald.wahl@raritan.com>
> ---
> V2: - use spin_lock_bh instead of moving netif_wake_queue outside of
>       locked region (doing the same in the start_xmit function)
>     - add missing net: tag
> 
> V3: - spin_lock_bh(ks->statelock) always except in xmit which is in BH
>       already

I agree that this lock is now always either taken _bh,
or in the xmit path which is executed in BH context.

Reviewed-by: Simon Horman <horms@kernel.org>