RE: [PATCH net-next v4 4/5] net: wangxun: introduce soft quiesce callbacks for AER recovery
From: Jiawen Wu <jiawenwu@trustnetic.com>
Date: 2026-06-03 02:46:25
On Tue, Jun 2, 2026 7:16 PM, Larysa Zaremba wrote:
On Mon, Jun 01, 2026 at 03:22:20PM +0800, Jiawen Wu wrote:quoted
Introduce device-specific soft quiesce callbacks for ngbe and txgbe to provide a lightweight shutdown path during PCI error recovery. It avoids MMIO-dependent operations in PCI error status, for the later implementation of PCIe error callback function in libwx. Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com> --- drivers/net/ethernet/wangxun/libwx/wx_type.h | 1 + drivers/net/ethernet/wangxun/ngbe/ngbe_main.c | 23 +++++++++++++++++++ .../net/ethernet/wangxun/txgbe/txgbe_main.c | 22 ++++++++++++++++++ 3 files changed, 46 insertions(+)diff --git a/drivers/net/ethernet/wangxun/libwx/wx_type.h b/drivers/net/ethernet/wangxun/libwx/wx_type.h index a8b4e84787f4..1b25a52188f7 100644 --- a/drivers/net/ethernet/wangxun/libwx/wx_type.h +++ b/drivers/net/ethernet/wangxun/libwx/wx_type.h@@ -1409,6 +1409,7 @@ struct wx { void (*configure_fdir)(struct wx *wx); int (*setup_tc)(struct net_device *netdev, u8 tc); void (*do_reset)(struct net_device *netdev, bool reinit); + void (*soft_quiesce)(struct wx *wx); int (*ptp_setup_sdp)(struct wx *wx); void (*set_num_queues)(struct wx *wx);diff --git a/drivers/net/ethernet/wangxun/ngbe/ngbe_main.c b/drivers/net/ethernet/wangxun/ngbe/ngbe_main.c index 3dc8342dd3a7..2484d3177034 100644 --- a/drivers/net/ethernet/wangxun/ngbe/ngbe_main.c +++ b/drivers/net/ethernet/wangxun/ngbe/ngbe_main.c@@ -47,6 +47,28 @@ static const struct pci_device_id ngbe_pci_tbl[] = { { } }; +static void ngbe_soft_quiesce(struct wx *wx) +{ + if (test_and_set_bit(WX_STATE_DOWN, wx->state)) + return;Careful about reusing the same flags for different reinit processes.
I think WX_STATE_DOWN should be used here. The soft_quiesce calls back only in pcie error handler when netif is running. It is desired to implement all software quiesce on the device close path.
quoted
+ + wx_ptp_stop(wx); + phylink_stop(wx->phylink); + pci_clear_master(wx->pdev); + wx_napi_disable_all(wx);Sashiko says: "If NAPI is not disabled yet, the NAPI poll routine wx_clean_tx_irq() can run concurrently. If WX_STATE_PTP_TX_IN_PROGRESS is set, wx_clean_tx_irq() might call ptp_schedule_worker(), leading to a dereference of the ptp_clock object just as it's being freed." I am not sure, if such race actually happens, but stopping napi before any kind of deinit is not a terrible idea.
WX_STATE_PTP_TX_IN_PROGRESS is cleared in wx_ptp_stop(). But I'll change wx_ptp_stop() to wx_ptp_quiesce() that drop all MMIO operations.
quoted
+ + clear_bit(WX_FLAG_NEED_PF_RESET, wx->flags);Same here, what if a caller needs a full reset?
PCIe error handler has a higher priority and it will achieve a more complete reset. In fact, these two resets are not expected to be executed at the same time.
quoted
+ timer_delete_sync(&wx->service_timer);Sashiko says: "The timer is stopped, but the corresponding cancel_work_sync(&wx->service_task) is missing." And it does seem to me that this is the case here. I see that you have explained this in another thread, but you proposal does not address the case, when ngbe_service_task() starts, interface is not down, and after that we have a PCI error, and its handling interferes with service task. While this is unlikely, synchronization should be more robust.
I plan to check WX_STATE_DOWN and WX_STATE_RESETTING at the entry of every work item. Looks like a rough block, but is there a better way? :(
quoted
+ + wx_clean_all_tx_rings(wx); + wx_clean_all_rx_rings(wx); + + wx_free_irq(wx); + wx_free_isb_resources(wx); + wx_free_resources(wx); + phylink_disconnect_phy(wx->phylink); +} + /** * ngbe_init_type_code - Initialize the shared code * @wx: pointer to hardware structure@@ -135,6 +157,7 @@ static int ngbe_sw_init(struct wx *wx) wx->mbx.size = WX_VXMAILBOX_SIZE; wx->setup_tc = ngbe_setup_tc; wx->do_reset = ngbe_do_reset; + wx->soft_quiesce = ngbe_soft_quiesce; set_bit(0, &wx->fwd_bitmask); return 0;diff --git a/drivers/net/ethernet/wangxun/txgbe/txgbe_main.c b/drivers/net/ethernet/wangxun/txgbe/txgbe_main.c index b37c9ed57cf7..816aba4a9099 100644 --- a/drivers/net/ethernet/wangxun/txgbe/txgbe_main.c +++ b/drivers/net/ethernet/wangxun/txgbe/txgbe_main.c@@ -296,6 +296,27 @@ void txgbe_up(struct wx *wx) txgbe_up_complete(wx); } +static void txgbe_soft_quiesce(struct wx *wx) +{ + if (test_and_set_bit(WX_STATE_DOWN, wx->state)) + return; + + wx_ptp_stop(wx); + phylink_stop(wx->phylink); + pci_clear_master(wx->pdev); + wx_napi_disable_all(wx); + + clear_bit(WX_FLAG_NEED_PF_RESET, wx->flags); + timer_delete_sync(&wx->service_timer); + + wx_clean_all_tx_rings(wx); + wx_clean_all_rx_rings(wx); + + wx_free_irq(wx); + txgbe_free_misc_irq(wx->priv); + wx_free_resources(wx); +} + /** * txgbe_init_type_code - Initialize the shared code * @wx: pointer to hardware structure@@ -412,6 +433,7 @@ static int txgbe_sw_init(struct wx *wx) wx->setup_tc = txgbe_setup_tc; wx->do_reset = txgbe_do_reset; + wx->soft_quiesce = txgbe_soft_quiesce; set_bit(0, &wx->fwd_bitmask); switch (wx->mac.type) { --2.51.0