[PATCH net 0/2] amd-xgbe: fix S0i3 suspend/resume issues

[PATCH net 0/2] amd-xgbe: fix S0i3 suspend/resume issues

[Raju Rangoju]

This patch series fixes suspend/resume issues in the AMD XGBE driver that
prevent proper S0i3 (Modern Standby) entry on AMD platforms.

Fixes include:

1. "BUG: scheduling while atomic" error during suspend/resume:
   The xgbe_powerdown()/xgbe_powerup() functions call napi_disable()
   and flush_workqueue() under a spinlock. These functions can sleep,
   which is not allowed in atomic context.

2. System fails to enter deepest sleep state (S0i3):
   Missing proper PCI power management operations prevent the system
   from reaching the deepest sleep states.

Patch 1 fixes the sleep-while-atomic bug by removing the unnecessary
spinlock from the power management functions.

Patch 2 adds proper PCI power management to support S0i3 by saving/
restoring PCI state, managing device power states, and properly
handling interrupts and bus mastering.

Raju Rangoju (2):
  amd-xgbe: fix sleep while atomic on suspend/resume
  amd-xgbe: add PCI power management for S0i3 support

 drivers/net/ethernet/amd/xgbe/xgbe-drv.c      | 57 +++++++------------
 drivers/net/ethernet/amd/xgbe/xgbe-pci.c      | 49 ++++++++++++++--
 drivers/net/ethernet/amd/xgbe/xgbe-platform.c |  4 +-
 drivers/net/ethernet/amd/xgbe/xgbe.h          |  8 +--
 4 files changed, 68 insertions(+), 50 deletions(-)

-- 
2.34.1

[PATCH net 1/2] amd-xgbe: fix sleep while atomic on suspend/resume

[Raju Rangoju]

The xgbe_powerdown() and xgbe_powerup() functions use spinlocks
(spin_lock_irqsave) while calling functions that may sleep:
- napi_disable() can sleep waiting for NAPI polling to complete
- flush_workqueue() can sleep waiting for pending work items

This causes a "BUG: scheduling while atomic" error during suspend/resume
cycles on systems using the AMD XGBE Ethernet controller.

The spinlock protection in these functions is unnecessary because:
1. The functions are called from suspend/resume paths which are already
   serialized by the PM core
2. The caller parameter was used to differentiate contexts, but the
   only current usage is from the driver context (suspend/resume)
3. The power_down flag provides sufficient synchronization

Fix this by:
- Removing the spinlock from xgbe_powerdown() and xgbe_powerup()
- Simplifying the function signatures by removing the unused caller
  parameter
- Removing the unused XGMAC_DRIVER_CONTEXT and XGMAC_IOCTL_CONTEXT macros
- Reordering operations in xgbe_powerdown() to disable NAPI before
  stopping TX/RX (matching the order used in xgbe_stop())

Fixes: c5aa9e3b8156 ("amd-xgbe: Initial AMD 10GbE platform driver")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
---
 drivers/net/ethernet/amd/xgbe/xgbe-drv.c      | 57 +++++++------------
 drivers/net/ethernet/amd/xgbe/xgbe-pci.c      | 10 ++--
 drivers/net/ethernet/amd/xgbe/xgbe-platform.c |  4 +-
 drivers/net/ethernet/amd/xgbe/xgbe.h          |  8 +--
 4 files changed, 29 insertions(+), 50 deletions(-)

diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-drv.c b/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
index 62bb4b8a68e1..d57daf6306e1 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
+++ b/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
@@ -1116,79 +1116,62 @@ static int xgbe_phy_reset(struct xgbe_prv_data *pdata)
 	return pdata->phy_if.phy_reset(pdata);
 }
 
-int xgbe_powerdown(struct net_device *netdev, unsigned int caller)
+int xgbe_powerdown(struct net_device *netdev)
 {
 	struct xgbe_prv_data *pdata = netdev_priv(netdev);
 	struct xgbe_hw_if *hw_if = &pdata->hw_if;
-	unsigned long flags;
 
-	DBGPR("-->xgbe_powerdown\n");
-
-	if (!netif_running(netdev) ||
-	    (caller == XGMAC_IOCTL_CONTEXT && pdata->power_down)) {
-		netdev_alert(netdev, "Device is already powered down\n");
-		DBGPR("<--xgbe_powerdown\n");
+	if (!netif_running(netdev)) {
+		netdev_dbg(netdev, "Device is not running, skipping powerdown\n");
 		return -EINVAL;
 	}
 
-	spin_lock_irqsave(&pdata->lock, flags);
-
-	if (caller == XGMAC_DRIVER_CONTEXT)
-		netif_device_detach(netdev);
+	if (pdata->power_down) {
+		netdev_dbg(netdev, "Device is already powered down\n");
+		return -EINVAL;
+	}
 
+	netif_device_detach(netdev);
 	netif_tx_stop_all_queues(netdev);
 
 	xgbe_stop_timers(pdata);
 	flush_workqueue(pdata->dev_workqueue);
 
+	xgbe_napi_disable(pdata, 0);
+
 	hw_if->powerdown_tx(pdata);
 	hw_if->powerdown_rx(pdata);
 
-	xgbe_napi_disable(pdata, 0);
-
 	pdata->power_down = 1;
 
-	spin_unlock_irqrestore(&pdata->lock, flags);
-
-	DBGPR("<--xgbe_powerdown\n");
-
 	return 0;
 }
 
-int xgbe_powerup(struct net_device *netdev, unsigned int caller)
+int xgbe_powerup(struct net_device *netdev)
 {
 	struct xgbe_prv_data *pdata = netdev_priv(netdev);
 	struct xgbe_hw_if *hw_if = &pdata->hw_if;
-	unsigned long flags;
-
-	DBGPR("-->xgbe_powerup\n");
 
-	if (!netif_running(netdev) ||
-	    (caller == XGMAC_IOCTL_CONTEXT && !pdata->power_down)) {
-		netdev_alert(netdev, "Device is already powered up\n");
-		DBGPR("<--xgbe_powerup\n");
+	if (!netif_running(netdev)) {
+		netdev_dbg(netdev, "Device is not running, skipping powerup\n");
 		return -EINVAL;
 	}
 
-	spin_lock_irqsave(&pdata->lock, flags);
-
-	pdata->power_down = 0;
-
-	xgbe_napi_enable(pdata, 0);
+	if (!pdata->power_down) {
+		netdev_dbg(netdev, "Device is already powered up\n");
+		return -EINVAL;
+	}
 
 	hw_if->powerup_tx(pdata);
 	hw_if->powerup_rx(pdata);
 
-	if (caller == XGMAC_DRIVER_CONTEXT)
-		netif_device_attach(netdev);
+	xgbe_napi_enable(pdata, 0);
 
 	netif_tx_start_all_queues(netdev);
-
 	xgbe_start_timers(pdata);
+	netif_device_attach(netdev);
 
-	spin_unlock_irqrestore(&pdata->lock, flags);
-
-	DBGPR("<--xgbe_powerup\n");
+	pdata->power_down = 0;
 
 	return 0;
 }
diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
index e3e1dca9856a..112d7697174c 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
+++ b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
@@ -352,14 +352,14 @@ static void xgbe_pci_remove(struct pci_dev *pdev)
 	xgbe_free_pdata(pdata);
 }
 
-static int __maybe_unused xgbe_pci_suspend(struct device *dev)
+static int xgbe_pci_suspend(struct device *dev)
 {
 	struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
 	struct net_device *netdev = pdata->netdev;
 	int ret = 0;
 
 	if (netif_running(netdev))
-		ret = xgbe_powerdown(netdev, XGMAC_DRIVER_CONTEXT);
+		ret = xgbe_powerdown(netdev);
 
 	pdata->lpm_ctrl = XMDIO_READ(pdata, MDIO_MMD_PCS, MDIO_CTRL1);
 	pdata->lpm_ctrl |= MDIO_CTRL1_LPOWER;
@@ -368,7 +368,7 @@ static int __maybe_unused xgbe_pci_suspend(struct device *dev)
 	return ret;
 }
 
-static int __maybe_unused xgbe_pci_resume(struct device *dev)
+static int xgbe_pci_resume(struct device *dev)
 {
 	struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
 	struct net_device *netdev = pdata->netdev;
@@ -380,7 +380,7 @@ static int __maybe_unused xgbe_pci_resume(struct device *dev)
 	XMDIO_WRITE(pdata, MDIO_MMD_PCS, MDIO_CTRL1, pdata->lpm_ctrl);
 
 	if (netif_running(netdev)) {
-		ret = xgbe_powerup(netdev, XGMAC_DRIVER_CONTEXT);
+		ret = xgbe_powerup(netdev);
 
 		/* Schedule a restart in case the link or phy state changed
 		 * while we were powered down.
@@ -453,7 +453,7 @@ static const struct pci_device_id xgbe_pci_table[] = {
 };
 MODULE_DEVICE_TABLE(pci, xgbe_pci_table);
 
-static SIMPLE_DEV_PM_OPS(xgbe_pci_pm_ops, xgbe_pci_suspend, xgbe_pci_resume);
+DEFINE_SIMPLE_DEV_PM_OPS(xgbe_pci_pm_ops, xgbe_pci_suspend, xgbe_pci_resume);
 
 static struct pci_driver xgbe_driver = {
 	.name = XGBE_DRV_NAME,
diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-platform.c b/drivers/net/ethernet/amd/xgbe/xgbe-platform.c
index 47d53e59ccf6..98b03a3f3a95 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-platform.c
+++ b/drivers/net/ethernet/amd/xgbe/xgbe-platform.c
@@ -384,7 +384,7 @@ static int xgbe_platform_suspend(struct device *dev)
 	DBGPR("-->xgbe_suspend\n");
 
 	if (netif_running(netdev))
-		ret = xgbe_powerdown(netdev, XGMAC_DRIVER_CONTEXT);
+		ret = xgbe_powerdown(netdev);
 
 	pdata->lpm_ctrl = XMDIO_READ(pdata, MDIO_MMD_PCS, MDIO_CTRL1);
 	pdata->lpm_ctrl |= MDIO_CTRL1_LPOWER;
@@ -407,7 +407,7 @@ static int xgbe_platform_resume(struct device *dev)
 	XMDIO_WRITE(pdata, MDIO_MMD_PCS, MDIO_CTRL1, pdata->lpm_ctrl);
 
 	if (netif_running(netdev)) {
-		ret = xgbe_powerup(netdev, XGMAC_DRIVER_CONTEXT);
+		ret = xgbe_powerup(netdev);
 
 		/* Schedule a restart in case the link or phy state changed
 		 * while we were powered down.
diff --git a/drivers/net/ethernet/amd/xgbe/xgbe.h b/drivers/net/ethernet/amd/xgbe/xgbe.h
index 1269b8ce9249..4e1ab4172abb 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe.h
+++ b/drivers/net/ethernet/amd/xgbe/xgbe.h
@@ -146,10 +146,6 @@
 #define XGBE_MAX_PPS_OUT	4
 #define XGBE_MAX_AUX_SNAP	4
 
-/* Driver PMT macros */
-#define XGMAC_DRIVER_CONTEXT	1
-#define XGMAC_IOCTL_CONTEXT	2
-
 #define XGMAC_FIFO_MIN_ALLOC	2048
 #define XGMAC_FIFO_UNIT		256
 #define XGMAC_FIFO_ALIGN(_x)				\
@@ -1289,8 +1285,8 @@ void xgbe_dump_rx_desc(struct xgbe_prv_data *, struct xgbe_ring *,
 		       unsigned int);
 void xgbe_print_pkt(struct net_device *, struct sk_buff *, bool);
 void xgbe_get_all_hw_features(struct xgbe_prv_data *);
-int xgbe_powerup(struct net_device *, unsigned int);
-int xgbe_powerdown(struct net_device *, unsigned int);
+int xgbe_powerup(struct net_device *netdev);
+int xgbe_powerdown(struct net_device *netdev);
 void xgbe_init_rx_coalesce(struct xgbe_prv_data *);
 void xgbe_init_tx_coalesce(struct xgbe_prv_data *);
 void xgbe_restart_dev(struct xgbe_prv_data *pdata);
-- 
2.34.1

[PATCH net 2/2] amd-xgbe: add PCI power management for S0i3 support

[Raju Rangoju]

Add proper PCI power management operations to support modern standby
(S0i3) on AMD platforms. The existing suspend/resume implementation
did not properly handle PCI device state transitions, which prevented
the system from reaching the deepest sleep states.

The amd_pmc driver was reporting "Last suspend didn't reach deepest
state" when the XGBE driver was enabled.

The implementation follows the standard PCI PM model:

Suspend path:
- Powerdown the network interface
- Set PHY to low-power mode
- Disable bus mastering to prevent DMA activity
- Save PCI configuration state
- Disable the PCI device
- Disable wake from D3 (required for S0i3 - no Wake-on-LAN needed)
- Set device to D3hot power state

Resume path:
- Restore PCI power state to D0
- Restore PCI configuration state
- Enable the PCI device
- Re-enable bus mastering
- Re-enable device interrupts
- Clear PHY low-power mode
- Power up the network interface

Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
---
 drivers/net/ethernet/amd/xgbe/xgbe-pci.c | 39 ++++++++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
index 112d7697174c..8e392cccd065 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
+++ b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
@@ -356,15 +356,34 @@ static int xgbe_pci_suspend(struct device *dev)
 {
 	struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
 	struct net_device *netdev = pdata->netdev;
+	struct pci_dev *pdev = to_pci_dev(dev);
 	int ret = 0;
 
+	if (!netdev)
+		return 0;
+
 	if (netif_running(netdev))
 		ret = xgbe_powerdown(netdev);
 
+	/* Disable all device interrupts to prevent spurious wakeups */
+	XP_IOWRITE(pdata, XP_INT_EN, 0x0);
+
+	/* Set PHY to low-power mode */
 	pdata->lpm_ctrl = XMDIO_READ(pdata, MDIO_MMD_PCS, MDIO_CTRL1);
 	pdata->lpm_ctrl |= MDIO_CTRL1_LPOWER;
 	XMDIO_WRITE(pdata, MDIO_MMD_PCS, MDIO_CTRL1, pdata->lpm_ctrl);
 
+	/* Disable bus mastering to prevent DMA activity */
+	pci_clear_master(pdev);
+
+	/* Save PCI configuration state and disable device */
+	pci_save_state(pdev);
+	pci_disable_device(pdev);
+
+	/* Disable wake from D3 - required for S0i3 deep sleep */
+	pci_wake_from_d3(pdev, false);
+	pci_set_power_state(pdev, PCI_D3hot);
+
 	return ret;
 }
 
@@ -372,13 +391,33 @@ static int xgbe_pci_resume(struct device *dev)
 {
 	struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
 	struct net_device *netdev = pdata->netdev;
+	struct pci_dev *pdev = to_pci_dev(dev);
 	int ret = 0;
 
+	/* Restore PCI power state */
+	pci_set_power_state(pdev, PCI_D0);
+
+	/* Restore PCI configuration state */
+	pci_restore_state(pdev);
+
+	/* Enable PCI device */
+	ret = pci_enable_device(pdev);
+	if (ret)
+		return ret;
+
+	/* Re-enable bus mastering */
+	pci_set_master(pdev);
+
+	/* Re-enable all device interrupts */
 	XP_IOWRITE(pdata, XP_INT_EN, 0x1fffff);
 
+	/* Clear PHY low-power mode */
 	pdata->lpm_ctrl &= ~MDIO_CTRL1_LPOWER;
 	XMDIO_WRITE(pdata, MDIO_MMD_PCS, MDIO_CTRL1, pdata->lpm_ctrl);
 
+	if (!netdev)
+		return 0;
+
 	if (netif_running(netdev)) {
 		ret = xgbe_powerup(netdev);
 
-- 
2.34.1

Re: [PATCH net 1/2] amd-xgbe: fix sleep while atomic on suspend/resume

[Simon Horman]

On Wed, Feb 25, 2026 at 04:30:00PM +0530, Raju Rangoju wrote:
> The xgbe_powerdown() and xgbe_powerup() functions use spinlocks
> (spin_lock_irqsave) while calling functions that may sleep:
> - napi_disable() can sleep waiting for NAPI polling to complete
> - flush_workqueue() can sleep waiting for pending work items
> 
> This causes a "BUG: scheduling while atomic" error during suspend/resume
> cycles on systems using the AMD XGBE Ethernet controller.
> 
> The spinlock protection in these functions is unnecessary because:
> 1. The functions are called from suspend/resume paths which are already
>    serialized by the PM core
> 2. The caller parameter was used to differentiate contexts, but the
>    only current usage is from the driver context (suspend/resume)
> 3. The power_down flag provides sufficient synchronization
> 
> Fix this by:
> - Removing the spinlock from xgbe_powerdown() and xgbe_powerup()
> - Simplifying the function signatures by removing the unused caller
>   parameter
> - Removing the unused XGMAC_DRIVER_CONTEXT and XGMAC_IOCTL_CONTEXT macros
> - Reordering operations in xgbe_powerdown() to disable NAPI before
>   stopping TX/RX (matching the order used in xgbe_stop())

I don't think that all of these changes are necessary to fix the issue at hand.
If so, please separate the fix(es) from other changes. And submit only
the fixes to net - ideally one patch per fix if there is more than one
discrete fix.

OTOH, enhancements and clean-ups should be submitted to net-next.
If there are dependencies on or conflicts with the fixes, then let
them go into net first. net is merged into net-next each Thursday or Friday.

> 
> Fixes: c5aa9e3b8156 ("amd-xgbe: Initial AMD 10GbE platform driver")
> Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>

...

-- 
pw-bot: changes-requested

Re: [PATCH net 2/2] amd-xgbe: add PCI power management for S0i3 support

[Simon Horman]

On Wed, Feb 25, 2026 at 04:30:01PM +0530, Raju Rangoju wrote:
> Add proper PCI power management operations to support modern standby
> (S0i3) on AMD platforms. The existing suspend/resume implementation
> did not properly handle PCI device state transitions, which prevented
> the system from reaching the deepest sleep states.
> 
> The amd_pmc driver was reporting "Last suspend didn't reach deepest
> state" when the XGBE driver was enabled.
> 
> The implementation follows the standard PCI PM model:
> 
> Suspend path:
> - Powerdown the network interface
> - Set PHY to low-power mode
> - Disable bus mastering to prevent DMA activity
> - Save PCI configuration state
> - Disable the PCI device
> - Disable wake from D3 (required for S0i3 - no Wake-on-LAN needed)
> - Set device to D3hot power state
> 
> Resume path:
> - Restore PCI power state to D0
> - Restore PCI configuration state
> - Enable the PCI device
> - Re-enable bus mastering
> - Re-enable device interrupts
> - Clear PHY low-power mode
> - Power up the network interface
> 
> Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
> ---
>  drivers/net/ethernet/amd/xgbe/xgbe-pci.c | 39 ++++++++++++++++++++++++
>  1 file changed, 39 insertions(+)
> 
> diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
> index 112d7697174c..8e392cccd065 100644
> --- a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
> +++ b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
> @@ -356,15 +356,34 @@ static int xgbe_pci_suspend(struct device *dev)
>  {
>  	struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
>  	struct net_device *netdev = pdata->netdev;
> +	struct pci_dev *pdev = to_pci_dev(dev);
>  	int ret = 0;
>  
> +	if (!netdev)
> +		return 0;
> +

Can netdev be NULL here?
Same question for the check in xgbe_pci_resume().

>  	if (netif_running(netdev))
>  		ret = xgbe_powerdown(netdev);
>  
> +	/* Disable all device interrupts to prevent spurious wakeups */
> +	XP_IOWRITE(pdata, XP_INT_EN, 0x0);
> +
> +	/* Set PHY to low-power mode */
>  	pdata->lpm_ctrl = XMDIO_READ(pdata, MDIO_MMD_PCS, MDIO_CTRL1);
>  	pdata->lpm_ctrl |= MDIO_CTRL1_LPOWER;
>  	XMDIO_WRITE(pdata, MDIO_MMD_PCS, MDIO_CTRL1, pdata->lpm_ctrl);
>  
> +	/* Disable bus mastering to prevent DMA activity */
> +	pci_clear_master(pdev);
> +
> +	/* Save PCI configuration state and disable device */
> +	pci_save_state(pdev);
> +	pci_disable_device(pdev);
> +
> +	/* Disable wake from D3 - required for S0i3 deep sleep */
> +	pci_wake_from_d3(pdev, false);
> +	pci_set_power_state(pdev, PCI_D3hot);
> +
>  	return ret;
>  }
>  
> @@ -372,13 +391,33 @@ static int xgbe_pci_resume(struct device *dev)
>  {
>  	struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
>  	struct net_device *netdev = pdata->netdev;
> +	struct pci_dev *pdev = to_pci_dev(dev);
>  	int ret = 0;
>  
> +	/* Restore PCI power state */
> +	pci_set_power_state(pdev, PCI_D0);
> +
> +	/* Restore PCI configuration state */
> +	pci_restore_state(pdev);
> +
> +	/* Enable PCI device */
> +	ret = pci_enable_device(pdev);
> +	if (ret)
> +		return ret;
> +
> +	/* Re-enable bus mastering */
> +	pci_set_master(pdev);
> +
> +	/* Re-enable all device interrupts */
>  	XP_IOWRITE(pdata, XP_INT_EN, 0x1fffff);
>  
> +	/* Clear PHY low-power mode */
>  	pdata->lpm_ctrl &= ~MDIO_CTRL1_LPOWER;
>  	XMDIO_WRITE(pdata, MDIO_MMD_PCS, MDIO_CTRL1, pdata->lpm_ctrl);
>  
> +	if (!netdev)
> +		return 0;
> +
>  	if (netif_running(netdev)) {
>  		ret = xgbe_powerup(netdev);
>  
> -- 
> 2.34.1
> 
> 

Re: [PATCH net 1/2] amd-xgbe: fix sleep while atomic on suspend/resume

[Simon Horman]

On Wed, Feb 25, 2026 at 04:30:00PM +0530, Raju Rangoju wrote:

...

> diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c

...

> @@ -453,7 +453,7 @@ static const struct pci_device_id xgbe_pci_table[] = {
>  };
>  MODULE_DEVICE_TABLE(pci, xgbe_pci_table);
>  
> -static SIMPLE_DEV_PM_OPS(xgbe_pci_pm_ops, xgbe_pci_suspend, xgbe_pci_resume);
> +DEFINE_SIMPLE_DEV_PM_OPS(xgbe_pci_pm_ops, xgbe_pci_suspend, xgbe_pci_resume);

Sorry for not noticing this before sending my previous email.

Sparse flags that xgbe_pci_pm_ops should (still) be static.

>  
>  static struct pci_driver xgbe_driver = {
>  	.name = XGBE_DRV_NAME,

...

Re: [PATCH net 1/2] amd-xgbe: fix sleep while atomic on suspend/resume

[Rangoju, Raju]

On 2/26/2026 6:07 PM, Simon Horman wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
> 
> 
> On Wed, Feb 25, 2026 at 04:30:00PM +0530, Raju Rangoju wrote:
>> The xgbe_powerdown() and xgbe_powerup() functions use spinlocks
>> (spin_lock_irqsave) while calling functions that may sleep:
>> - napi_disable() can sleep waiting for NAPI polling to complete
>> - flush_workqueue() can sleep waiting for pending work items
>>
>> This causes a "BUG: scheduling while atomic" error during suspend/resume
>> cycles on systems using the AMD XGBE Ethernet controller.
>>
>> The spinlock protection in these functions is unnecessary because:
>> 1. The functions are called from suspend/resume paths which are already
>>     serialized by the PM core
>> 2. The caller parameter was used to differentiate contexts, but the
>>     only current usage is from the driver context (suspend/resume)
>> 3. The power_down flag provides sufficient synchronization
>>
>> Fix this by:
>> - Removing the spinlock from xgbe_powerdown() and xgbe_powerup()
>> - Simplifying the function signatures by removing the unused caller
>>    parameter
>> - Removing the unused XGMAC_DRIVER_CONTEXT and XGMAC_IOCTL_CONTEXT macros
>> - Reordering operations in xgbe_powerdown() to disable NAPI before
>>    stopping TX/RX (matching the order used in xgbe_stop())
> 
> I don't think that all of these changes are necessary to fix the issue at hand.
> If so, please separate the fix(es) from other changes. And submit only
> the fixes to net - ideally one patch per fix if there is more than one
> discrete fix.
> 
> OTOH, enhancements and clean-ups should be submitted to net-next.

Sure, I'll separate the fixes from cleanup and submit.
Since the clean-ups have some dependency on fixes, will let the fixes go 
first.

> If there are dependencies on or conflicts with the fixes, then let
> them go into net first. net is merged into net-next each Thursday or Friday.
> 
>>
>> Fixes: c5aa9e3b8156 ("amd-xgbe: Initial AMD 10GbE platform driver")
>> Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
> 
> ...
> 
> --
> pw-bot: changes-requested

Re: [PATCH net 1/2] amd-xgbe: fix sleep while atomic on suspend/resume

[Rangoju, Raju]

On 2/26/2026 6:10 PM, Simon Horman wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
> 
> 
> On Wed, Feb 25, 2026 at 04:30:00PM +0530, Raju Rangoju wrote:
> 
> ...
> 
>> diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
> 
> ...
> 
>> @@ -453,7 +453,7 @@ static const struct pci_device_id xgbe_pci_table[] = {
>>   };
>>   MODULE_DEVICE_TABLE(pci, xgbe_pci_table);
>>
>> -static SIMPLE_DEV_PM_OPS(xgbe_pci_pm_ops, xgbe_pci_suspend, xgbe_pci_resume);
>> +DEFINE_SIMPLE_DEV_PM_OPS(xgbe_pci_pm_ops, xgbe_pci_suspend, xgbe_pci_resume);
> 
> Sorry for not noticing this before sending my previous email.
> 
> Sparse flags that xgbe_pci_pm_ops should (still) be static.

Noted. Thank you. Will fix it in the clean-up part.

> 
>>
>>   static struct pci_driver xgbe_driver = {
>>        .name = XGBE_DRV_NAME,
> 
> ...

Re: [PATCH net 2/2] amd-xgbe: add PCI power management for S0i3 support

[Rangoju, Raju]

On 2/26/2026 6:08 PM, Simon Horman wrote:

>> ---
>>   drivers/net/ethernet/amd/xgbe/xgbe-pci.c | 39 ++++++++++++++++++++++++
>>   1 file changed, 39 insertions(+)
>>
>> diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
>> index 112d7697174c..8e392cccd065 100644
>> --- a/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
>> +++ b/drivers/net/ethernet/amd/xgbe/xgbe-pci.c
>> @@ -356,15 +356,34 @@ static int xgbe_pci_suspend(struct device *dev)
>>   {
>>        struct xgbe_prv_data *pdata = dev_get_drvdata(dev);
>>        struct net_device *netdev = pdata->netdev;
>> +     struct pci_dev *pdev = to_pci_dev(dev);
>>        int ret = 0;
>>
>> +     if (!netdev)
>> +             return 0;
>> +
> 
> Can netdev be NULL here?
> Same question for the check in xgbe_pci_resume().
> 
Sure Simon, will fix it and submit to net-next as this patch is an 
enhancement.