net: mvneta: re-enable percpu interrupt on resume
PATCH 1/5 net: mvneta: re-enable percpu interrupt on resume
From: Yun Zhou <[email protected]>
On Marvell MPIC platforms (Armada 370/XP/38x), mvneta uses a percpu IRQ disable/enable scheme for NAPI: the ISR (mvneta_percpu_isr) calls disable_percpu_irq() to mask the MPIC per-CPU interrupt and schedules NAPI poll, which calls enable_percpu_irq() on completion to unmask.
If suspend occurs while NAPI poll is pending (between disable_percpu_irq in the ISR and enable_percpu_irq in poll completion), the interrupt is never re-enabled:
1. mvneta_percpu_isr: disable_percpu_irq() + napi_schedule() => MPIC masked, percpu_enabled cpumask bit cleared
2. NAPI poll does not complete before suspend proceeds (on PREEMPT_RT this is highly likely since softirqs run in ksoftirqd which gets frozen; on non-RT it can happen when softirq processing is deferred to ksoftirqd)
3. mvneta_stop_dev => napi_disable(): cancels the pending poll without executing the completion path
4. suspend_device_irqs => IRQCHIP_MASK_ON_SUSPEND: masks MPIC (already masked, but records IRQS_SUSPENDED)
5. Resume: mpic_resume checks irq_percpu_is_enabled() => false (bit was cleared in step 1) => skips unmask
6. mvneta_start_dev only restores device-level INTR_NEW_MASK, does not touch the MPIC per-CPU mask
Result: MPIC per-CPU interrupt stays masked permanently. The NIC generates interrupts (INTR_NEW_CAUSE != 0) but the CPU never receives them, causing complete loss of network connectivity.
Fix by calling on_each_cpu(mvneta_percpu_enable) in the resume path to unconditionally unmask the MPIC per-CPU interrupt regardless of pre-suspend state.
--- drivers/net/ethernet/marvell/mvneta.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/net/ethernet/marvell/mvneta.c +3 -0
@@ -5897,6 +5897,9 @@ static int mvneta_resume(struct device *device)rtnl_unlock ();mvneta_set_rx_mode (dev); +if (!pp->neta_armada3700) +on_each_cpu (mvneta_percpu_enable, pp, true); + return 0; }#endif
Signature · 1 line
2.53.0
PATCH 2/5 net: mvneta: support EPROBE_DEFER when reading MAC address
From: Rosen Penev <[email protected]>
If nvmem loads after the ethernet driver, mac address assignments will not take effect. of_get_ethdev_address returns EPROBE_DEFER in such a case so we need to handle that to avoid eth_hw_addr_random.
Add extra goto section to just free stats as they are allocated right above.
--- drivers/net/ethernet/marvell/mvneta.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/net/ethernet/marvell/mvneta.c +3 -0
@@ -5620,6 +5620,8 @@ static int mvneta_probe(struct platform_device *pdev) } err =of_get_ethdev_address (dn, dev); +if (err == -EPROBE_DEFER ) + goto err_free_stats;if (!err) { mac_from ="device tree" ; }else {
@@ -5755,6 +5757,7 @@ static int mvneta_probe(struct platform_device *pdev) 1 << pp->id);mvneta_bm_put (pp->bm_priv); } +err_free_stats:free_percpu (pp->stats); err_free_ports:free_percpu (pp->ports);
Signature · 1 line
2.53.0
PATCH 3/5 net: macb: drop in-flight Tx SKBs on close
In reply to Arbab HaiderFrom: Théo Lebrun <[email protected]>
The MACB driver has since forever leaked the outgoing SKBs that have not yet been marked as completed. They live in queue->tx_skb which gets freed without remorse nor checking.
macb_free_consistent() gets called in a few codepaths, but only close will trigger the added expressions. In macb_open() and macb_alloc_consistent() failure cases, queues' tx_skb just got allocated and are empty.
--- drivers/net/ethernet/cadence/macb_main.c | 21 +++++++++++++++++++-- 1 file changed, 19 insertions(+), 2 deletions(-)
drivers/net/ethernet/cadence/macb_main.c +19 -2
@@ -2551,8 +2551,25 @@ static void macb_free_consistent(struct macb *bp)dma_free_coherent (dev, size, bp->queues[0 ].rx_ring, bp->queues[0 ].rx_ring_dma);for (q =0 , queue = bp->queues; q < bp->num_queues; ++q, ++queue) { -kfree (queue->tx_skb); - queue->tx_skb =NULL ; +if (queue->tx_skb) { +unsigned int dropped =0 , tail; + +for (tail = queue->tx_tail; tail != queue->tx_head; + tail++) { +if (macb_tx_skb (queue, tail)->skb) + dropped++; +macb_tx_unmap (bp,macb_tx_skb (queue, tail),0 ); + } + + queue->stats.tx_dropped += dropped; + bp->dev->stats.tx_dropped += dropped; + +kfree (queue->tx_skb); + queue->tx_skb =NULL ; + } + + queue->tx_head =0 ; + queue->tx_tail =0 ; queue->tx_ring =NULL ; queue->rx_ring =NULL ; }
Signature · 1 line
2.53.0
PATCH 4/5 net: cpsw_new: unregister devlink on port registration failure
In reply to Arbab HaiderFrom: Guangshuo Li <[email protected]>
cpsw_probe() registers devlink before registering the CPSW ports.
If cpsw_register_ports() fails, the error path only unregisters the notifiers and then releases the lower level resources. It does not undo the successful cpsw_register_devlink() call, leaving the devlink instance and its parameters registered after probe has failed.
Add a devlink cleanup label for the path where devlink registration has already succeeded, and use it when port registration fails.
--- drivers/net/ethernet/ti/cpsw_new.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/ti/cpsw_new.c +3 -1
@@ -2050,7 +2050,7 @@ static int cpsw_probe(struct platform_device *pdev) ret =cpsw_register_ports (cpsw);if (ret) - goto clean_unregister_notifiers; + goto clean_unregister_devlink; dev_notice(dev, "initialized (regs %pa, pool size %d) hw_ver:%08X %d.%d (%d)\n", &ss_res->start, descs_pool_size,
@@ -2062,6 +2062,8 @@ static int cpsw_probe(struct platform_device *pdev)return 0 ; +clean_unregister_devlink: +cpsw_unregister_devlink (cpsw); clean_unregister_notifiers:cpsw_unregister_notifiers (cpsw); clean_cpts:
Signature · 1 line
2.53.0
PATCH 5/5 net: stmmac: fix transmit interrupt coalescing
In reply to Arbab HaiderFrom: "Russell King (Oracle)" <[email protected]>
The accounting for transmit frames does not count the descriptors correctly. It uses:
tx_packets = (tx_q->cur_tx + 1) - first_tx;
however, these are indexes into a circular buffer, so cur_tx can be less than first_tx, and when that happens, tx_packets becomes a very large unsigned integer. When this is added to tx_q->tx_count_frames, it has the effect of reducing the count of frames, possibly causing it to also wrap to a very large unsigned integer.
Fix this by using CIRC_CNT() to calculate the number of descriptors used.
--- drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +4 -2
@@ -14,6 +14,7 @@https://bugzilla.stlinux.com/ *******************************************************************************/ +#include <linux/circ_buf.h> #include <linux/clk.h> #include <linux/kernel.h> #include <linux/interrupt.h>
@@ -4511,7 +4512,8 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev) tx_q->tx_skbuff_dma[tx_q->cur_tx].buf_type =STMMAC_TXBUF_T_SKB ;/* Manage tx mitigation */ - tx_packets = (tx_q->cur_tx +1 ) - first_tx; + tx_packets =CIRC_CNT (tx_q->cur_tx +1 , first_tx, + priv->dma_conf.dma_tx_size); tx_q->tx_count_frames += tx_packets;if ((skb_shinfo (skb)->tx_flags &SKBTX_HW_TSTAMP ) && priv->hwts_tx_en)
@@ -4789,7 +4791,7 @@ static netdev_tx_t stmmac_xmit(struct sk_buff *skb, struct net_device *dev)* This approach takes care about the fragments: desc is the first * element in case of no SG. */ - tx_packets = (entry +1 ) - first_tx; + tx_packets =CIRC_CNT (entry +1 , first_tx, priv->dma_conf.dma_tx_size); tx_q->tx_count_frames += tx_packets;if ((skb_shinfo (skb)->tx_flags &SKBTX_HW_TSTAMP ) && priv->hwts_tx_en)
Signature · 1 line
2.53.0
Sasha Levin
maintainerQuoted 6 lines · Arbab Haider
From: Yun Zhou <[email protected]>
On Marvell MPIC platforms (Armada 370/XP/38x), mvneta uses a percpu IRQ disable/enable scheme for NAPI: the ISR (
mvneta_percpu_isr) callsdisable_percpu_irq()to mask the MPIC per-CPU interrupt and schedules NAPI poll, which callsenable_percpu_irq()on completion to unmask.
7.0 is EOL?
Signature · 2 lines
Thanks, Sasha
From: Arbab Haider <[email protected]>
Hi Sasha,
Yes, it is. My mistake. 7.0.y went end-of-life with 7.0.14 (2026-06-27), and I overlooked that when preparing the series.
For context: these five patches are backports of mainline fixes that I carry in a private 7.0.14 tree used on embedded boards that are not moving off the 7.0 series in the near term. I sent them to stable intending to add them to the stable queue, but an EOL branch is the wrong target. Please drop them.
Since all of them are already in mainline, and therefore in 7.1.y and the longterm branches, nothing further is needed there. I will keep the backports as out-of-tree patches for the private tree.
Sorry for the noise.
Thanks,
Arbab
From: Arbab Haider <[email protected]>
Hi Sasha,
Yes, it is. My mistake. 7.0.y went end-of-life with 7.0.14 (2026-06-27), and I overlooked that when preparing the series.
For context: these five patches are backports of mainline fixes that I carry in a private 7.0.14 tree used on embedded boards that are not moving off the 7.0 series in the near term. I sent them to stable intending to add them to the stable queue, but an EOL branch is the wrong target. Please drop them.
Since all of them are already in mainline, and therefore in 7.1.y and the longterm branches, nothing further is needed there. I will keep the backports as out-of-tree patches for the private tree.
Sorry for the noise.
Thanks,
Arbab
Greg KH
maintainerQuoted 4 lines · Arbab Haider
From: Arbab Haider <[email protected]>
Hi Sasha,
Quoted 8 lines
Yes, it is. My mistake. 7.0.y went end-of-life with 7.0.14 (2026-06-27), and I overlooked that when preparing the series.
For context: these five patches are backports of mainline fixes that I carry in a private 7.0.14 tree used on embedded boards that are not moving off the 7.0 series in the near term. I sent them to stable intending to add them to the stable queue, but an EOL branch is the wrong target. Please drop them.
Please note that as of right now, the 7.0.14 kernel currently has:
Total Vulnerable CVE's in 7.0.14 : 387
which might not be good for your embedded systems :)
thanks,
greg k-h