drivers/pci/hotplug: Support surprise hotplug in powernv driver
authorGavin Shan <gwshan@linux.vnet.ibm.com>
Wed, 28 Sep 2016 04:34:58 +0000 (14:34 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Thu, 29 Sep 2016 05:02:28 +0000 (15:02 +1000)
This supports PCI surprise hotplug. The design is highlighted as
below:

   * The PCI slot's surprise hotplug capability is exposed through
     device node property "ibm,slot-surprise-pluggable", meaning
     PCI surprise hotplug will be disabled if skiboot doesn't support
     it yet.
   * The interrupt because of presence or link state change is raised
     on surprise hotplug event. One event is allocated and queued to
     the PCI slot for workqueue to pick it up and process in serialized
     fashion. The code flow for surprise hotplug is same to that for
     managed hotplug except: the affected PEs are put into frozen state
     to avoid unexpected EEH error reporting in surprise hot remove path.

Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/include/asm/pnv-pci.h
drivers/pci/hotplug/pnv_php.c

index 0cbd8134ce81949f8bf079181a7180aa55ee9abb..17e89dd613e190246db5d31212723de041a5a9e0 100644 (file)
@@ -60,6 +60,8 @@ struct pnv_php_slot {
 #define PNV_PHP_STATE_POPULATED                2
 #define PNV_PHP_STATE_OFFLINE          3
        int                             state;
+       int                             irq;
+       struct workqueue_struct         *wq;
        struct device_node              *dn;
        struct pci_dev                  *pdev;
        struct pci_bus                  *bus;
index 182f218dce4f0ffc9cd510885d495ba9d0a9df0d..5dd0d4f2f907c8cc18ed1a0b3fc170aac74bd9a0 100644 (file)
 #define DRIVER_AUTHOR  "Gavin Shan, IBM Corporation"
 #define DRIVER_DESC    "PowerPC PowerNV PCI Hotplug Driver"
 
+struct pnv_php_event {
+       bool                    added;
+       struct pnv_php_slot     *php_slot;
+       struct work_struct      work;
+};
+
 static LIST_HEAD(pnv_php_slot_list);
 static DEFINE_SPINLOCK(pnv_php_lock);
 
@@ -29,12 +35,40 @@ static void pnv_php_register(struct device_node *dn);
 static void pnv_php_unregister_one(struct device_node *dn);
 static void pnv_php_unregister(struct device_node *dn);
 
+static void pnv_php_disable_irq(struct pnv_php_slot *php_slot)
+{
+       struct pci_dev *pdev = php_slot->pdev;
+       u16 ctrl;
+
+       if (php_slot->irq > 0) {
+               pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
+               ctrl &= ~(PCI_EXP_SLTCTL_HPIE |
+                         PCI_EXP_SLTCTL_PDCE |
+                         PCI_EXP_SLTCTL_DLLSCE);
+               pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
+
+               free_irq(php_slot->irq, php_slot);
+               php_slot->irq = 0;
+       }
+
+       if (php_slot->wq) {
+               destroy_workqueue(php_slot->wq);
+               php_slot->wq = NULL;
+       }
+
+       if (pdev->msix_enabled)
+               pci_disable_msix(pdev);
+       else if (pdev->msi_enabled)
+               pci_disable_msi(pdev);
+}
+
 static void pnv_php_free_slot(struct kref *kref)
 {
        struct pnv_php_slot *php_slot = container_of(kref,
                                        struct pnv_php_slot, kref);
 
        WARN_ON(!list_empty(&php_slot->children));
+       pnv_php_disable_irq(php_slot);
        kfree(php_slot->name);
        kfree(php_slot);
 }
@@ -609,6 +643,181 @@ static int pnv_php_register_slot(struct pnv_php_slot *php_slot)
        return 0;
 }
 
+static int pnv_php_enable_msix(struct pnv_php_slot *php_slot)
+{
+       struct pci_dev *pdev = php_slot->pdev;
+       struct msix_entry entry;
+       int nr_entries, ret;
+       u16 pcie_flag;
+
+       /* Get total number of MSIx entries */
+       nr_entries = pci_msix_vec_count(pdev);
+       if (nr_entries < 0)
+               return nr_entries;
+
+       /* Check hotplug MSIx entry is in range */
+       pcie_capability_read_word(pdev, PCI_EXP_FLAGS, &pcie_flag);
+       entry.entry = (pcie_flag & PCI_EXP_FLAGS_IRQ) >> 9;
+       if (entry.entry >= nr_entries)
+               return -ERANGE;
+
+       /* Enable MSIx */
+       ret = pci_enable_msix_exact(pdev, &entry, 1);
+       if (ret) {
+               dev_warn(&pdev->dev, "Error %d enabling MSIx\n", ret);
+               return ret;
+       }
+
+       return entry.vector;
+}
+
+static void pnv_php_event_handler(struct work_struct *work)
+{
+       struct pnv_php_event *event =
+               container_of(work, struct pnv_php_event, work);
+       struct pnv_php_slot *php_slot = event->php_slot;
+
+       if (event->added)
+               pnv_php_enable_slot(&php_slot->slot);
+       else
+               pnv_php_disable_slot(&php_slot->slot);
+
+       kfree(event);
+}
+
+static irqreturn_t pnv_php_interrupt(int irq, void *data)
+{
+       struct pnv_php_slot *php_slot = data;
+       struct pci_dev *pchild, *pdev = php_slot->pdev;
+       struct eeh_dev *edev;
+       struct eeh_pe *pe;
+       struct pnv_php_event *event;
+       u16 sts, lsts;
+       u8 presence;
+       bool added;
+       unsigned long flags;
+       int ret;
+
+       pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
+       sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
+       pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
+       if (sts & PCI_EXP_SLTSTA_DLLSC) {
+               pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lsts);
+               added = !!(lsts & PCI_EXP_LNKSTA_DLLLA);
+       } else if (sts & PCI_EXP_SLTSTA_PDC) {
+               ret = pnv_pci_get_presence_state(php_slot->id, &presence);
+               if (!ret)
+                       return IRQ_HANDLED;
+               added = !!(presence == OPAL_PCI_SLOT_PRESENT);
+       } else {
+               return IRQ_NONE;
+       }
+
+       /* Freeze the removed PE to avoid unexpected error reporting */
+       if (!added) {
+               pchild = list_first_entry_or_null(&php_slot->bus->devices,
+                                                 struct pci_dev, bus_list);
+               edev = pchild ? pci_dev_to_eeh_dev(pchild) : NULL;
+               pe = edev ? edev->pe : NULL;
+               if (pe) {
+                       eeh_serialize_lock(&flags);
+                       eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
+                       eeh_serialize_unlock(flags);
+                       eeh_pe_set_option(pe, EEH_OPT_FREEZE_PE);
+               }
+       }
+
+       /*
+        * The PE is left in frozen state if the event is missed. It's
+        * fine as the PCI devices (PE) aren't functional any more.
+        */
+       event = kzalloc(sizeof(*event), GFP_ATOMIC);
+       if (!event) {
+               dev_warn(&pdev->dev, "PCI slot [%s] missed hotplug event 0x%04x\n",
+                        php_slot->name, sts);
+               return IRQ_HANDLED;
+       }
+
+       dev_info(&pdev->dev, "PCI slot [%s] %s (IRQ: %d)\n",
+                php_slot->name, added ? "added" : "removed", irq);
+       INIT_WORK(&event->work, pnv_php_event_handler);
+       event->added = added;
+       event->php_slot = php_slot;
+       queue_work(php_slot->wq, &event->work);
+
+       return IRQ_HANDLED;
+}
+
+static void pnv_php_init_irq(struct pnv_php_slot *php_slot, int irq)
+{
+       struct pci_dev *pdev = php_slot->pdev;
+       u16 sts, ctrl;
+       int ret;
+
+       /* Allocate workqueue */
+       php_slot->wq = alloc_workqueue("pciehp-%s", 0, 0, php_slot->name);
+       if (!php_slot->wq) {
+               dev_warn(&pdev->dev, "Cannot alloc workqueue\n");
+               pnv_php_disable_irq(php_slot);
+               return;
+       }
+
+       /* Clear pending interrupts */
+       pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
+       sts |= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
+       pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
+
+       /* Request the interrupt */
+       ret = request_irq(irq, pnv_php_interrupt, IRQF_SHARED,
+                         php_slot->name, php_slot);
+       if (ret) {
+               pnv_php_disable_irq(php_slot);
+               dev_warn(&pdev->dev, "Error %d enabling IRQ %d\n", ret, irq);
+               return;
+       }
+
+       /* Enable the interrupts */
+       pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
+       ctrl |= (PCI_EXP_SLTCTL_HPIE |
+                PCI_EXP_SLTCTL_PDCE |
+                PCI_EXP_SLTCTL_DLLSCE);
+       pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
+
+       /* The interrupt is initialized successfully when @irq is valid */
+       php_slot->irq = irq;
+}
+
+static void pnv_php_enable_irq(struct pnv_php_slot *php_slot)
+{
+       struct pci_dev *pdev = php_slot->pdev;
+       int irq, ret;
+
+       ret = pci_enable_device(pdev);
+       if (ret) {
+               dev_warn(&pdev->dev, "Error %d enabling device\n", ret);
+               return;
+       }
+
+       pci_set_master(pdev);
+
+       /* Enable MSIx interrupt */
+       irq = pnv_php_enable_msix(php_slot);
+       if (irq > 0) {
+               pnv_php_init_irq(php_slot, irq);
+               return;
+       }
+
+       /*
+        * Use MSI if MSIx doesn't work. Fail back to legacy INTx
+        * if MSI doesn't work either
+        */
+       ret = pci_enable_msi(pdev);
+       if (!ret || pdev->irq) {
+               irq = pdev->irq;
+               pnv_php_init_irq(php_slot, irq);
+       }
+}
+
 static int pnv_php_register_one(struct device_node *dn)
 {
        struct pnv_php_slot *php_slot;
@@ -636,6 +845,11 @@ static int pnv_php_register_one(struct device_node *dn)
        if (ret)
                goto unregister_slot;
 
+       /* Enable interrupt if the slot supports surprise hotplug */
+       prop32 = of_get_property(dn, "ibm,slot-surprise-pluggable", NULL);
+       if (prop32 && of_read_number(prop32, 1))
+               pnv_php_enable_irq(php_slot);
+
        return 0;
 
 unregister_slot: