[PATCH net v2 8/9] bnxt_en: Fix driver init in kdump kernel

[Michael Chan <michael.chan@broadcom.com>]

Fix and strengthen the FLR sequence when initializing in the kdump
kernel.  If the NIC is behind a PCIe switch in synthetic (smart)
mode, the switch may need to see that the BARs have been initialized
before it will pass mem read/write TLPs to the NIC.

Add a new bnxt_kdump_reset() to do the expanded FLR sequence in the
kdump kernel.  We now disable bus master and memory, save the PCI
state, do the FLR, clear the BARs, and restore the PCI state.  The
BARs have to be cleared to ensure that they get re-initialized and
visible to the PCIe switch.

Since it is the kdump kernel, we make every effort to continue in
the best possible way even if the device is unresponsive after FLR
or we encounter other errors.

To avoid dealing with a possible 0xffff value for the MSI-X capabilty
register after FLR, we call bnxt_get_max_irq() to get the valid
number of MSIX before we do the FLR.

Fixes: 8743db4a9acf ("bnxt_en: Issue PCIe FLR in kdump kernel to cleanup pending DMAs.")
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
v2:
Disable device before pci_save_state() and pcie_flr() and check for
errors. 

v1:
https://lore.kernel.org/netdev/20260831024342.2161156-4-michael.chan@broadcom.com/
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 48 +++++++++++++++++++----
 1 file changed, 40 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 8c6bf10aba8a..e4530b091d3b 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -17229,6 +17229,42 @@ static void bnxt_clear_bars(struct pci_dev *pdev)
 		pci_write_config_dword(pdev, off, 0);
 }
 
+/* Clear any pending DMA transactions from crash kernel while loading driver in
+ * capture kernel.
+ */
+static void bnxt_kdump_reset(struct pci_dev *pdev)
+{
+	u16 cmd;
+	int rc;
+
+	pci_read_config_word(pdev, PCI_COMMAND, &cmd);
+	cmd &= ~(PCI_COMMAND_MASTER | PCI_COMMAND_MEMORY);
+	pci_write_config_word(pdev, PCI_COMMAND, cmd);
+
+	if (pci_save_state(pdev))
+		dev_warn(&pdev->dev, "Failed to save PCI state, PCI restore may be incomplete\n");
+
+	/* Since it is kdump kernel, try to continue in the best possible way
+	 * even if the device is unresponsive after FLR.  The device may
+	 * eventually respond to HWRM_VER_GET later in the init sequence.
+	 */
+	rc = pcie_flr(pdev);
+	if (rc)
+		dev_warn(&pdev->dev, "pcie_flr() failed (rc: %d), trying to continue\n",
+			 rc);
+
+	/* A complete or partial reset has been done.  Clear the BARs
+	 * if the device is responsive.
+	 */
+	pci_read_config_word(pdev, PCI_COMMAND, &cmd);
+	if (PCI_POSSIBLE_ERROR(cmd))
+		dev_warn(&pdev->dev, "PCI config space inaccessible after FLR, not clearing BARs\n");
+	else
+		bnxt_clear_bars(pdev);
+
+	pci_restore_state(pdev);
+}
+
 static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
 	struct bnxt_hw_resc *hw_resc;
@@ -17244,15 +17280,11 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 		return -ENODEV;
 	}
 
-	/* Clear any pending DMA transactions from crash kernel
-	 * while loading driver in capture kernel.
-	 */
-	if (is_kdump_kernel()) {
-		pci_clear_master(pdev);
-		pcie_flr(pdev);
-	}
-
 	max_irqs = bnxt_get_max_irq(pdev);
+
+	if (is_kdump_kernel())
+		bnxt_kdump_reset(pdev);
+
 	dev = alloc_etherdev_mqs(sizeof(*bp), max_irqs * BNXT_MAX_QUEUE,
 				 max_irqs);
 	if (!dev)
-- 
2.51.0