【无标题】PCIE设备实现内核自动恢复设备功能

pcie设备通用的设备，本文不做pcie的通识性解析，只对解决方案进行说明。

--------------------------------------------------------------------------------------------------------------------------------------

功能说明：ssd处于工作状态（读写）时意外断开，在不重启系统的前提下，内核能自动恢复ssd，且不导致系统异常。

硬件原理图：

出错日志：

手动扫描pcie设备命令类似如下：

echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/remove

echo 1 > /sys/bus/pci/rescan

解决方案：

---
 arch/arm64/boot/dts/xilinx/zynqmp-nr8101.dts |   4 +
 drivers/pci/controller/pcie-xilinx-nwl.c     | 104 ++++++++++++++++++-
 2 files changed, 106 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/boot/dts/xilinx/zynqmp-nr8101.dts b/arch/arm64/boot/dts/xilinx/zynqmp-nr8101.dts
index f3225ed..784d2d6 100644
--- a/arch/arm64/boot/dts/xilinx/zynqmp-nr8101.dts
+++ b/arch/arm64/boot/dts/xilinx/zynqmp-nr8101.dts
@@ -640,3 +640,7 @@
 &xlnx_dpdma {
 	status = "okay";
 };
+
+&pcie {
+	ssd-reset-gpios = <&gpio 37 GPIO_ACTIVE_HIGH>;
+};
diff --git a/drivers/pci/controller/pcie-xilinx-nwl.c b/drivers/pci/controller/pcie-xilinx-nwl.c
index 449c8bc..93e1b2b 100644
--- a/drivers/pci/controller/pcie-xilinx-nwl.c
+++ b/drivers/pci/controller/pcie-xilinx-nwl.c
@@ -21,9 +21,14 @@
 #include <linux/pci.h>
 #include <linux/platform_device.h>
 #include <linux/irqchip/chained_irq.h>
-
+#include <linux/gpio/consumer.h>
+#include <linux/kthread.h>
+#include <linux/syscalls.h>
+#include "../../../../fs/mount.h"
 #include "../pci.h"
 
+#define NVME_SSD_PATH_PREFIX "/dev/nvme0n"
+
 /* Bridge core config registers */
 #define BRCFG_PCIE_RX0			0x00000000
 #define BRCFG_INTERRUPT			0x00000010
@@ -181,6 +186,11 @@ struct nwl_pcie {
 	struct irq_domain *legacy_irq_domain;
 	struct clk *clk;
 	raw_spinlock_t leg_mask_lock;
+	bool link_status;
+	struct gpio_desc	*gpio_ssd_reset;
+	struct task_struct	*link_detect_thread;
+	struct work_struct	link_down_cleanup_work;
+	struct work_struct	link_up_rescan_work;
 };
 
 static inline u32 nwl_bridge_readl(struct nwl_pcie *pcie, u32 off)
@@ -272,6 +282,78 @@ static struct pci_ops nwl_pcie_ops = {
 	.write = pci_generic_config_write,
 };
 
+static void force_umount_nvme_partitions(struct nwl_pcie *pcie)
+{
+	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
+	struct mount *mnt, *tmp;
+	char *kbuf = kmalloc(PATH_MAX, GFP_KERNEL);
+	char *mount_point_path;
+
+	if (kbuf) {
+		list_for_each_entry_safe(mnt, tmp, &ns->list, mnt_list) {
+			if (!strncmp(mnt->mnt_devname, NVME_SSD_PATH_PREFIX, strlen(NVME_SSD_PATH_PREFIX))) {
+				const struct path mnt_path = { .dentry = mnt->mnt.mnt_root, .mnt = &mnt->mnt };
+				memset(kbuf, 0, PATH_MAX);
+				mount_point_path = d_absolute_path(&mnt_path, kbuf, PATH_MAX);
+				if (mount_point_path) {
+					dev_info(pcie->dev, "Force umounting %s : %s\n", mnt->mnt_devname, mount_point_path);
+					if (ksys_umount(mount_point_path, MNT_DETACH|MNT_FORCE)) {
+						dev_info(pcie->dev, "Failed to umount %s\n", mnt->mnt_devname);
+					}
+				}
+			}
+		}
+		kfree(kbuf);
+	} else {
+		dev_info(pcie->dev, "Failed to alloc memory\n");
+	}
+}
+
+static void link_up_rescan_fn(struct work_struct *work)
+{
+	struct nwl_pcie *pcie =
+		container_of(work, struct nwl_pcie, link_up_rescan_work);
+	struct pci_bus *self = pci_find_bus(0,0);
+	if (self!=NULL)
+		pci_rescan_bus(self);
+	pcie->link_status = nwl_pcie_link_up(pcie);
+}
+
+static void link_down_cleanup_fn(struct work_struct *work)
+{
+	struct nwl_pcie *pcie =
+		container_of(work, struct nwl_pcie, link_down_cleanup_work);
+
+	struct pci_dev *pdev = pci_get_domain_bus_and_slot(0, 1, 0);
+	if (pdev!=NULL) {
+		pci_disable_device(pdev);
+		pci_stop_and_remove_bus_device_locked(pdev);
+		force_umount_nvme_partitions(pcie);
+	}
+	pcie->link_status = false;
+}
+
+static int pcie_link_detect_thread_fn(void *data)
+{
+	struct nwl_pcie *pcie = data;
+	while (!kthread_should_stop()) {
+		if (!pcie->link_status && nwl_pcie_link_up(pcie)) {
+			dev_info(pcie->dev, "Link is UP\n");
+			msleep_interruptible(500);
+			gpiod_set_raw_value_cansleep(pcie->gpio_ssd_reset, 0);
+			msleep_interruptible(100);
+			gpiod_set_raw_value_cansleep(pcie->gpio_ssd_reset, 1);
+			msleep_interruptible(1000);
+			dev_info(pcie->dev, "Rescan devices\n");
+			schedule_work(&pcie->link_up_rescan_work);
+			msleep_interruptible(2000);
+		} else {
+			msleep_interruptible(1000);
+		}
+	}
+	return 0;
+}
+
 static irqreturn_t nwl_pcie_misc_handler(int irq, void *data)
 {
 	struct nwl_pcie *pcie = data;
@@ -287,6 +369,11 @@ static irqreturn_t nwl_pcie_misc_handler(int irq, void *data)
 	if (misc_stat & MSGF_MISC_SR_RXMSG_OVER)
 		dev_err(dev, "Received Message FIFO Overflow\n");
 
+	if ((misc_stat & MSGF_MISC_SR_LINK_DOWN) && pcie->link_status) {
+		dev_info(pcie->dev, "Link is DOWN\n");
+		schedule_work(&pcie->link_down_cleanup_work);
+	}
+
 	if (misc_stat & MSGF_MISC_SR_SLAVE_ERR)
 		dev_err(dev, "Slave error\n");
 
@@ -733,7 +820,8 @@ static int nwl_pcie_bridge_init(struct nwl_pcie *pcie)
 	ecam_val |= (pcie->last_busno << E_ECAM_SIZE_SHIFT);
 	writel(ecam_val, (pcie->ecam_base + PCI_PRIMARY_BUS));
 
-	if (nwl_pcie_link_up(pcie))
+	pcie->link_status = nwl_pcie_link_up(pcie);
+	if (pcie->link_status)
 		dev_info(dev, "Link is UP\n");
 	else
 		dev_info(dev, "Link is DOWN\n");
@@ -919,6 +1007,18 @@ static int nwl_pcie_probe(struct platform_device *pdev)
 	list_for_each_entry(child, &bus->children, node)
 		pcie_bus_configure_settings(child);
 	pci_bus_add_devices(bus);
+
+	INIT_WORK(&pcie->link_down_cleanup_work, link_down_cleanup_fn);
+	INIT_WORK(&pcie->link_up_rescan_work, link_up_rescan_fn);
+
+	pcie->gpio_ssd_reset = devm_gpiod_get(pcie->dev, "ssd-reset", GPIOD_OUT_HIGH);
+	if (IS_ERR(pcie->gpio_ssd_reset)) {
+		dev_err(&pdev->dev, "Failed to get ssd-reset pin.\n");
+	}
+	pcie->link_detect_thread = kthread_run(pcie_link_detect_thread_fn, pcie, "nwl-pcie-detect");
+	if (IS_ERR(pcie->link_detect_thread)) {
+		dev_err(pcie->dev, "Could not start kernel thread.\n");
+	}
 	return 0;
 
 error: