In the xrx200_close() function we call napi_disable(), that could
sleep, with priv->hw->chan[i].lock held. This could lead to deadlock
and causes the kernel to complain.
Look at the code I couldn't convince myself about why we
need to protect that specific code part with the lock. IMHO there
seems no reason to protect the refcount variables, because AFAIK
ndo_close() and ndo_open() callbacks are already called with a
semaphore held. Neither I could figure out why napi_disable() have to
be called with that lock held. The only remaining code part for
which I could guess the lock is useful for is ltq_dma_close()
function call.
This patch reduces the lock to the said function call, avoiding the
sleep-with-spinlock-held situation
Signed-off-by: Andrea Merello <andrea.merello@gmail.com>
[fold into 0025-NET-MIPS-lantiq-adds-xrx200-net.patch, backport to
kernel 4.4]
Signed-off-by: Mathias Kresin <dev@kresin.me>
+};
--- /dev/null
+++ b/drivers/net/ethernet/lantiq_xrx200.c
-@@ -0,0 +1,1852 @@
+@@ -0,0 +1,1853 @@
+/*
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ for (i = 0; i < XRX200_MAX_DMA; i++) {
+ if (!priv->hw->chan[i].dma.irq)
+ continue;
-+ spin_lock_bh(&priv->hw->chan[i].lock);
++
+ priv->hw->chan[i].refcount--;
+ if (!priv->hw->chan[i].refcount) {
+ if (XRX200_DMA_IS_RX(i))
+ napi_disable(&priv->hw->chan[i].napi);
++ spin_lock_bh(&priv->hw->chan[i].lock);
+ ltq_dma_close(&priv->hw->chan[XRX200_DMA_RX].dma);
++ spin_unlock_bh(&priv->hw->chan[i].lock);
+ }
-+ spin_unlock_bh(&priv->hw->chan[i].lock);
+ }
+
+ return 0;
+};
--- /dev/null
+++ b/drivers/net/ethernet/lantiq_xrx200.c
-@@ -0,0 +1,1851 @@
+@@ -0,0 +1,1852 @@
+/*
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ for (i = 0; i < XRX200_MAX_DMA; i++) {
+ if (!priv->hw->chan[i].dma.irq)
+ continue;
-+ spin_lock_bh(&priv->hw->chan[i].lock);
++
+ priv->hw->chan[i].refcount--;
+ if (!priv->hw->chan[i].refcount) {
+ if (XRX200_DMA_IS_RX(i))
+ napi_disable(&priv->hw->chan[i].napi);
++ spin_lock_bh(&priv->hw->chan[i].lock);
+ ltq_dma_close(&priv->hw->chan[XRX200_DMA_RX].dma);
++ spin_unlock_bh(&priv->hw->chan[i].lock);
+ }
-+ spin_unlock_bh(&priv->hw->chan[i].lock);
+ }
+
+ return 0;