xen-pcifront: Xen PCI frontend driver.
authorRyan Wilson <hap9@epoch.ncsc.mil>
Tue, 3 Aug 2010 01:31:05 +0000 (21:31 -0400)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Mon, 18 Oct 2010 14:49:37 +0000 (10:49 -0400)
This is a port of the 2.6.18 Xen PCI front driver with fixes
to make it build under 2.6.34 and later (for the full list of
changes: git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen.git
historic/xen-pcifront-0.1). It also includes the fixes
to make it work properly.

[v2: Updated Kconfig, removed crud, added Reviewed-by]
[v3: Added 'static', fixed grant table leak, redid Kconfig]
[v4: Added one more 'static' and removed comments]

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Jan Beulich <JBeulich@novell.com>
drivers/pci/Kconfig
drivers/pci/Makefile
drivers/pci/xen-pcifront.c [new file with mode: 0644]
include/xen/interface/io/pciif.h [new file with mode: 0644]

index 34ef70d..5b1630e 100644 (file)
@@ -40,6 +40,27 @@ config PCI_STUB
 
          When in doubt, say N.
 
+config XEN_PCIDEV_FRONTEND
+        tristate "Xen PCI Frontend"
+        depends on PCI && X86 && XEN
+        select HOTPLUG
+        select PCI_XEN
+        default y
+        help
+          The PCI device frontend driver allows the kernel to import arbitrary
+          PCI devices from a PCI backend to support PCI driver domains.
+
+config XEN_PCIDEV_FE_DEBUG
+        bool "Xen PCI Frontend debugging"
+        depends on XEN_PCIDEV_FRONTEND && PCI_DEBUG
+       help
+         Say Y here if you want the Xen PCI frontend to produce a bunch of debug
+         messages to the system log.  Select this if you are having a
+         problem with Xen PCI frontend support and want to see more of what is
+         going on.
+
+         When in doubt, say N.
+
 config HT_IRQ
        bool "Interrupts on hypertransport devices"
        default y
index dc1aa09..d5e2705 100644 (file)
@@ -65,6 +65,8 @@ obj-$(CONFIG_PCI_SYSCALL) += syscall.o
 
 obj-$(CONFIG_PCI_STUB) += pci-stub.o
 
+obj-$(CONFIG_XEN_PCIDEV_FRONTEND) += xen-pcifront.o
+
 ifeq ($(CONFIG_PCI_DEBUG),y)
 EXTRA_CFLAGS += -DDEBUG
 endif
diff --git a/drivers/pci/xen-pcifront.c b/drivers/pci/xen-pcifront.c
new file mode 100644 (file)
index 0000000..a87c498
--- /dev/null
@@ -0,0 +1,1148 @@
+/*
+ * Xen PCI Frontend.
+ *
+ *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
+ */
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/mm.h>
+#include <xen/xenbus.h>
+#include <xen/events.h>
+#include <xen/grant_table.h>
+#include <xen/page.h>
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/msi.h>
+#include <xen/xenbus.h>
+#include <xen/interface/io/pciif.h>
+#include <asm/xen/pci.h>
+#include <linux/interrupt.h>
+#include <asm/atomic.h>
+#include <linux/workqueue.h>
+#include <linux/bitops.h>
+#include <linux/time.h>
+
+#define INVALID_GRANT_REF (0)
+#define INVALID_EVTCHN    (-1)
+
+struct pci_bus_entry {
+       struct list_head list;
+       struct pci_bus *bus;
+};
+
+#define _PDEVB_op_active               (0)
+#define PDEVB_op_active                        (1 << (_PDEVB_op_active))
+
+struct pcifront_device {
+       struct xenbus_device *xdev;
+       struct list_head root_buses;
+
+       int evtchn;
+       int gnt_ref;
+
+       int irq;
+
+       /* Lock this when doing any operations in sh_info */
+       spinlock_t sh_info_lock;
+       struct xen_pci_sharedinfo *sh_info;
+       struct work_struct op_work;
+       unsigned long flags;
+
+};
+
+struct pcifront_sd {
+       int domain;
+       struct pcifront_device *pdev;
+};
+
+static inline struct pcifront_device *
+pcifront_get_pdev(struct pcifront_sd *sd)
+{
+       return sd->pdev;
+}
+
+static inline void pcifront_init_sd(struct pcifront_sd *sd,
+                                   unsigned int domain, unsigned int bus,
+                                   struct pcifront_device *pdev)
+{
+       sd->domain = domain;
+       sd->pdev = pdev;
+}
+
+static DEFINE_SPINLOCK(pcifront_dev_lock);
+static struct pcifront_device *pcifront_dev;
+
+static int verbose_request;
+module_param(verbose_request, int, 0644);
+
+static int errno_to_pcibios_err(int errno)
+{
+       switch (errno) {
+       case XEN_PCI_ERR_success:
+               return PCIBIOS_SUCCESSFUL;
+
+       case XEN_PCI_ERR_dev_not_found:
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       case XEN_PCI_ERR_invalid_offset:
+       case XEN_PCI_ERR_op_failed:
+               return PCIBIOS_BAD_REGISTER_NUMBER;
+
+       case XEN_PCI_ERR_not_implemented:
+               return PCIBIOS_FUNC_NOT_SUPPORTED;
+
+       case XEN_PCI_ERR_access_denied:
+               return PCIBIOS_SET_FAILED;
+       }
+       return errno;
+}
+
+static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
+{
+       if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
+               && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
+               dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
+               schedule_work(&pdev->op_work);
+       }
+}
+
+static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
+{
+       int err = 0;
+       struct xen_pci_op *active_op = &pdev->sh_info->op;
+       unsigned long irq_flags;
+       evtchn_port_t port = pdev->evtchn;
+       unsigned irq = pdev->irq;
+       s64 ns, ns_timeout;
+       struct timeval tv;
+
+       spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
+
+       memcpy(active_op, op, sizeof(struct xen_pci_op));
+
+       /* Go */
+       wmb();
+       set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
+       notify_remote_via_evtchn(port);
+
+       /*
+        * We set a poll timeout of 3 seconds but give up on return after
+        * 2 seconds. It is better to time out too late rather than too early
+        * (in the latter case we end up continually re-executing poll() with a
+        * timeout in the past). 1s difference gives plenty of slack for error.
+        */
+       do_gettimeofday(&tv);
+       ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC;
+
+       xen_clear_irq_pending(irq);
+
+       while (test_bit(_XEN_PCIF_active,
+                       (unsigned long *)&pdev->sh_info->flags)) {
+               xen_poll_irq_timeout(irq, jiffies + 3*HZ);
+               xen_clear_irq_pending(irq);
+               do_gettimeofday(&tv);
+               ns = timeval_to_ns(&tv);
+               if (ns > ns_timeout) {
+                       dev_err(&pdev->xdev->dev,
+                               "pciback not responding!!!\n");
+                       clear_bit(_XEN_PCIF_active,
+                                 (unsigned long *)&pdev->sh_info->flags);
+                       err = XEN_PCI_ERR_dev_not_found;
+                       goto out;
+               }
+       }
+
+       /*
+       * We might lose backend service request since we
+       * reuse same evtchn with pci_conf backend response. So re-schedule
+       * aer pcifront service.
+       */
+       if (test_bit(_XEN_PCIB_active,
+                       (unsigned long *)&pdev->sh_info->flags)) {
+               dev_err(&pdev->xdev->dev,
+                       "schedule aer pcifront service\n");
+               schedule_pcifront_aer_op(pdev);
+       }
+
+       memcpy(op, active_op, sizeof(struct xen_pci_op));
+
+       err = op->err;
+out:
+       spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
+       return err;
+}
+
+/* Access to this function is spinlocked in drivers/pci/access.c */
+static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
+                            int where, int size, u32 *val)
+{
+       int err = 0;
+       struct xen_pci_op op = {
+               .cmd    = XEN_PCI_OP_conf_read,
+               .domain = pci_domain_nr(bus),
+               .bus    = bus->number,
+               .devfn  = devfn,
+               .offset = where,
+               .size   = size,
+       };
+       struct pcifront_sd *sd = bus->sysdata;
+       struct pcifront_device *pdev = pcifront_get_pdev(sd);
+
+       if (verbose_request)
+               dev_info(&pdev->xdev->dev,
+                        "read dev=%04x:%02x:%02x.%01x - offset %x size %d\n",
+                        pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
+                        PCI_FUNC(devfn), where, size);
+
+       err = do_pci_op(pdev, &op);
+
+       if (likely(!err)) {
+               if (verbose_request)
+                       dev_info(&pdev->xdev->dev, "read got back value %x\n",
+                                op.value);
+
+               *val = op.value;
+       } else if (err == -ENODEV) {
+               /* No device here, pretend that it just returned 0 */
+               err = 0;
+               *val = 0;
+       }
+
+       return errno_to_pcibios_err(err);
+}
+
+/* Access to this function is spinlocked in drivers/pci/access.c */
+static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
+                             int where, int size, u32 val)
+{
+       struct xen_pci_op op = {
+               .cmd    = XEN_PCI_OP_conf_write,
+               .domain = pci_domain_nr(bus),
+               .bus    = bus->number,
+               .devfn  = devfn,
+               .offset = where,
+               .size   = size,
+               .value  = val,
+       };
+       struct pcifront_sd *sd = bus->sysdata;
+       struct pcifront_device *pdev = pcifront_get_pdev(sd);
+
+       if (verbose_request)
+               dev_info(&pdev->xdev->dev,
+                        "write dev=%04x:%02x:%02x.%01x - "
+                        "offset %x size %d val %x\n",
+                        pci_domain_nr(bus), bus->number,
+                        PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
+
+       return errno_to_pcibios_err(do_pci_op(pdev, &op));
+}
+
+struct pci_ops pcifront_bus_ops = {
+       .read = pcifront_bus_read,
+       .write = pcifront_bus_write,
+};
+
+#ifdef CONFIG_PCI_MSI
+static int pci_frontend_enable_msix(struct pci_dev *dev,
+                                   int **vector, int nvec)
+{
+       int err;
+       int i;
+       struct xen_pci_op op = {
+               .cmd    = XEN_PCI_OP_enable_msix,
+               .domain = pci_domain_nr(dev->bus),
+               .bus = dev->bus->number,
+               .devfn = dev->devfn,
+               .value = nvec,
+       };
+       struct pcifront_sd *sd = dev->bus->sysdata;
+       struct pcifront_device *pdev = pcifront_get_pdev(sd);
+       struct msi_desc *entry;
+
+       if (nvec > SH_INFO_MAX_VEC) {
+               dev_err(&dev->dev, "too much vector for pci frontend: %x."
+                                  " Increase SH_INFO_MAX_VEC.\n", nvec);
+               return -EINVAL;
+       }
+
+       i = 0;
+       list_for_each_entry(entry, &dev->msi_list, list) {
+               op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
+               /* Vector is useless at this point. */
+               op.msix_entries[i].vector = -1;
+               i++;
+       }
+
+       err = do_pci_op(pdev, &op);
+
+       if (likely(!err)) {
+               if (likely(!op.value)) {
+                       /* we get the result */
+                       for (i = 0; i < nvec; i++)
+                               *(*vector+i) = op.msix_entries[i].vector;
+                       return 0;
+               } else {
+                       printk(KERN_DEBUG "enable msix get value %x\n",
+                               op.value);
+                       return op.value;
+               }
+       } else {
+               dev_err(&dev->dev, "enable msix get err %x\n", err);
+               return err;
+       }
+}
+
+static void pci_frontend_disable_msix(struct pci_dev *dev)
+{
+       int err;
+       struct xen_pci_op op = {
+               .cmd    = XEN_PCI_OP_disable_msix,
+               .domain = pci_domain_nr(dev->bus),
+               .bus = dev->bus->number,
+               .devfn = dev->devfn,
+       };
+       struct pcifront_sd *sd = dev->bus->sysdata;
+       struct pcifront_device *pdev = pcifront_get_pdev(sd);
+
+       err = do_pci_op(pdev, &op);
+
+       /* What should do for error ? */
+       if (err)
+               dev_err(&dev->dev, "pci_disable_msix get err %x\n", err);
+}
+
+static int pci_frontend_enable_msi(struct pci_dev *dev, int **vector)
+{
+       int err;
+       struct xen_pci_op op = {
+               .cmd    = XEN_PCI_OP_enable_msi,
+               .domain = pci_domain_nr(dev->bus),
+               .bus = dev->bus->number,
+               .devfn = dev->devfn,
+       };
+       struct pcifront_sd *sd = dev->bus->sysdata;
+       struct pcifront_device *pdev = pcifront_get_pdev(sd);
+
+       err = do_pci_op(pdev, &op);
+       if (likely(!err)) {
+               *(*vector) = op.value;
+       } else {
+               dev_err(&dev->dev, "pci frontend enable msi failed for dev "
+                                   "%x:%x\n", op.bus, op.devfn);
+               err = -EINVAL;
+       }
+       return err;
+}
+
+static void pci_frontend_disable_msi(struct pci_dev *dev)
+{
+       int err;
+       struct xen_pci_op op = {
+               .cmd    = XEN_PCI_OP_disable_msi,
+               .domain = pci_domain_nr(dev->bus),
+               .bus = dev->bus->number,
+               .devfn = dev->devfn,
+       };
+       struct pcifront_sd *sd = dev->bus->sysdata;
+       struct pcifront_device *pdev = pcifront_get_pdev(sd);
+
+       err = do_pci_op(pdev, &op);
+       if (err == XEN_PCI_ERR_dev_not_found) {
+               /* XXX No response from backend, what shall we do? */
+               printk(KERN_DEBUG "get no response from backend for disable MSI\n");
+               return;
+       }
+       if (err)
+               /* how can pciback notify us fail? */
+               printk(KERN_DEBUG "get fake response frombackend\n");
+}
+
+static struct xen_pci_frontend_ops pci_frontend_ops = {
+       .enable_msi = pci_frontend_enable_msi,
+       .disable_msi = pci_frontend_disable_msi,
+       .enable_msix = pci_frontend_enable_msix,
+       .disable_msix = pci_frontend_disable_msix,
+};
+
+static void pci_frontend_registrar(int enable)
+{
+       if (enable)
+               xen_pci_frontend = &pci_frontend_ops;
+       else
+               xen_pci_frontend = NULL;
+};
+#else
+static inline void pci_frontend_registrar(int enable) { };
+#endif /* CONFIG_PCI_MSI */
+
+/* Claim resources for the PCI frontend as-is, backend won't allow changes */
+static int pcifront_claim_resource(struct pci_dev *dev, void *data)
+{
+       struct pcifront_device *pdev = data;
+       int i;
+       struct resource *r;
+
+       for (i = 0; i < PCI_NUM_RESOURCES; i++) {
+               r = &dev->resource[i];
+
+               if (!r->parent && r->start && r->flags) {
+                       dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
+                               pci_name(dev), i);
+                       if (pci_claim_resource(dev, i)) {
+                               dev_err(&pdev->xdev->dev, "Could not claim "
+                                       "resource %s/%d! Device offline. Try "
+                                       "giving less than 4GB to domain.\n",
+                                       pci_name(dev), i);
+                       }
+               }
+       }
+
+       return 0;
+}
+
+static int __devinit pcifront_scan_bus(struct pcifront_device *pdev,
+                               unsigned int domain, unsigned int bus,
+                               struct pci_bus *b)
+{
+       struct pci_dev *d;
+       unsigned int devfn;
+
+       /* Scan the bus for functions and add.
+        * We omit handling of PCI bridge attachment because pciback prevents
+        * bridges from being exported.
+        */
+       for (devfn = 0; devfn < 0x100; devfn++) {
+               d = pci_get_slot(b, devfn);
+               if (d) {
+                       /* Device is already known. */
+                       pci_dev_put(d);
+                       continue;
+               }
+
+               d = pci_scan_single_device(b, devfn);
+               if (d)
+                       dev_info(&pdev->xdev->dev, "New device on "
+                                "%04x:%02x:%02x.%02x found.\n", domain, bus,
+                                PCI_SLOT(devfn), PCI_FUNC(devfn));
+       }
+
+       return 0;
+}
+
+static int __devinit pcifront_scan_root(struct pcifront_device *pdev,
+                                unsigned int domain, unsigned int bus)
+{
+       struct pci_bus *b;
+       struct pcifront_sd *sd = NULL;
+       struct pci_bus_entry *bus_entry = NULL;
+       int err = 0;
+
+#ifndef CONFIG_PCI_DOMAINS
+       if (domain != 0) {
+               dev_err(&pdev->xdev->dev,
+                       "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
+               dev_err(&pdev->xdev->dev,
+                       "Please compile with CONFIG_PCI_DOMAINS\n");
+               err = -EINVAL;
+               goto err_out;
+       }
+#endif
+
+       dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
+                domain, bus);
+
+       bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL);
+       sd = kmalloc(sizeof(*sd), GFP_KERNEL);
+       if (!bus_entry || !sd) {
+               err = -ENOMEM;
+               goto err_out;
+       }
+       pcifront_init_sd(sd, domain, bus, pdev);
+
+       b = pci_scan_bus_parented(&pdev->xdev->dev, bus,
+                                 &pcifront_bus_ops, sd);
+       if (!b) {
+               dev_err(&pdev->xdev->dev,
+                       "Error creating PCI Frontend Bus!\n");
+               err = -ENOMEM;
+               goto err_out;
+       }
+
+       bus_entry->bus = b;
+
+       list_add(&bus_entry->list, &pdev->root_buses);
+
+       /* pci_scan_bus_parented skips devices which do not have a have
+       * devfn==0. The pcifront_scan_bus enumerates all devfn. */
+       err = pcifront_scan_bus(pdev, domain, bus, b);
+
+       /* Claim resources before going "live" with our devices */
+       pci_walk_bus(b, pcifront_claim_resource, pdev);
+
+       /* Create SysFS and notify udev of the devices. Aka: "going live" */
+       pci_bus_add_devices(b);
+
+       return err;
+
+err_out:
+       kfree(bus_entry);
+       kfree(sd);
+
+       return err;
+}
+
+static int __devinit pcifront_rescan_root(struct pcifront_device *pdev,
+                                  unsigned int domain, unsigned int bus)
+{
+       int err;
+       struct pci_bus *b;
+
+#ifndef CONFIG_PCI_DOMAINS
+       if (domain != 0) {
+               dev_err(&pdev->xdev->dev,
+                       "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
+               dev_err(&pdev->xdev->dev,
+                       "Please compile with CONFIG_PCI_DOMAINS\n");
+               return -EINVAL;
+       }
+#endif
+
+       dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
+                domain, bus);
+
+       b = pci_find_bus(domain, bus);
+       if (!b)
+               /* If the bus is unknown, create it. */
+               return pcifront_scan_root(pdev, domain, bus);
+
+       err = pcifront_scan_bus(pdev, domain, bus, b);
+
+       /* Claim resources before going "live" with our devices */
+       pci_walk_bus(b, pcifront_claim_resource, pdev);
+
+       /* Create SysFS and notify udev of the devices. Aka: "going live" */
+       pci_bus_add_devices(b);
+
+       return err;
+}
+
+static void free_root_bus_devs(struct pci_bus *bus)
+{
+       struct pci_dev *dev;
+
+       while (!list_empty(&bus->devices)) {
+               dev = container_of(bus->devices.next, struct pci_dev,
+                                  bus_list);
+               dev_dbg(&dev->dev, "removing device\n");
+               pci_remove_bus_device(dev);
+       }
+}
+
+static void pcifront_free_roots(struct pcifront_device *pdev)
+{
+       struct pci_bus_entry *bus_entry, *t;
+
+       dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
+
+       list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
+               list_del(&bus_entry->list);
+
+               free_root_bus_devs(bus_entry->bus);
+
+               kfree(bus_entry->bus->sysdata);
+
+               device_unregister(bus_entry->bus->bridge);
+               pci_remove_bus(bus_entry->bus);
+
+               kfree(bus_entry);
+       }
+}
+
+static pci_ers_result_t pcifront_common_process(int cmd,
+                                               struct pcifront_device *pdev,
+                                               pci_channel_state_t state)
+{
+       pci_ers_result_t result;
+       struct pci_driver *pdrv;
+       int bus = pdev->sh_info->aer_op.bus;
+       int devfn = pdev->sh_info->aer_op.devfn;
+       struct pci_dev *pcidev;
+       int flag = 0;
+
+       dev_dbg(&pdev->xdev->dev,
+               "pcifront AER process: cmd %x (bus:%x, devfn%x)",
+               cmd, bus, devfn);
+       result = PCI_ERS_RESULT_NONE;
+
+       pcidev = pci_get_bus_and_slot(bus, devfn);
+       if (!pcidev || !pcidev->driver) {
+               dev_err(&pcidev->dev,
+                       "device or driver is NULL\n");
+               return result;
+       }
+       pdrv = pcidev->driver;
+
+       if (get_driver(&pdrv->driver)) {
+               if (pdrv->err_handler && pdrv->err_handler->error_detected) {
+                       dev_dbg(&pcidev->dev,
+                               "trying to call AER service\n");
+                       if (pcidev) {
+                               flag = 1;
+                               switch (cmd) {
+                               case XEN_PCI_OP_aer_detected:
+                                       result = pdrv->err_handler->
+                                                error_detected(pcidev, state);
+                                       break;
+                               case XEN_PCI_OP_aer_mmio:
+                                       result = pdrv->err_handler->
+                                                mmio_enabled(pcidev);
+                                       break;
+                               case XEN_PCI_OP_aer_slotreset:
+                                       result = pdrv->err_handler->
+                                                slot_reset(pcidev);
+                                       break;
+                               case XEN_PCI_OP_aer_resume:
+                                       pdrv->err_handler->resume(pcidev);
+                                       break;
+                               default:
+                                       dev_err(&pdev->xdev->dev,
+                                               "bad request in aer recovery "
+                                               "operation!\n");
+
+                               }
+                       }
+               }
+               put_driver(&pdrv->driver);
+       }
+       if (!flag)
+               result = PCI_ERS_RESULT_NONE;
+
+       return result;
+}
+
+
+static void pcifront_do_aer(struct work_struct *data)
+{
+       struct pcifront_device *pdev =
+               container_of(data, struct pcifront_device, op_work);
+       int cmd = pdev->sh_info->aer_op.cmd;
+       pci_channel_state_t state =
+               (pci_channel_state_t)pdev->sh_info->aer_op.err;
+
+       /*If a pci_conf op is in progress,
+               we have to wait until it is done before service aer op*/
+       dev_dbg(&pdev->xdev->dev,
+               "pcifront service aer bus %x devfn %x\n",
+               pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
+
+       pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
+
+       /* Post the operation to the guest. */
+       wmb();
+       clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
+       notify_remote_via_evtchn(pdev->evtchn);
+
+       /*in case of we lost an aer request in four lines time_window*/
+       smp_mb__before_clear_bit();
+       clear_bit(_PDEVB_op_active, &pdev->flags);
+       smp_mb__after_clear_bit();
+
+       schedule_pcifront_aer_op(pdev);
+
+}
+
+static irqreturn_t pcifront_handler_aer(int irq, void *dev)
+{
+       struct pcifront_device *pdev = dev;
+       schedule_pcifront_aer_op(pdev);
+       return IRQ_HANDLED;
+}
+static int pcifront_connect(struct pcifront_device *pdev)
+{
+       int err = 0;
+
+       spin_lock(&pcifront_dev_lock);
+
+       if (!pcifront_dev) {
+               dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
+               pcifront_dev = pdev;
+       } else {
+               dev_err(&pdev->xdev->dev, "PCI frontend already installed!\n");
+               err = -EEXIST;
+       }
+
+       spin_unlock(&pcifront_dev_lock);
+
+       return err;
+}
+
+static void pcifront_disconnect(struct pcifront_device *pdev)
+{
+       spin_lock(&pcifront_dev_lock);
+
+       if (pdev == pcifront_dev) {
+               dev_info(&pdev->xdev->dev,
+                        "Disconnecting PCI Frontend Buses\n");
+               pcifront_dev = NULL;
+       }
+
+       spin_unlock(&pcifront_dev_lock);
+}
+static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
+{
+       struct pcifront_device *pdev;
+
+       pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
+       if (pdev == NULL)
+               goto out;
+
+       pdev->sh_info =
+           (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
+       if (pdev->sh_info == NULL) {
+               kfree(pdev);
+               pdev = NULL;
+               goto out;
+       }
+       pdev->sh_info->flags = 0;
+
+       /*Flag for registering PV AER handler*/
+       set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
+
+       dev_set_drvdata(&xdev->dev, pdev);
+       pdev->xdev = xdev;
+
+       INIT_LIST_HEAD(&pdev->root_buses);
+
+       spin_lock_init(&pdev->sh_info_lock);
+
+       pdev->evtchn = INVALID_EVTCHN;
+       pdev->gnt_ref = INVALID_GRANT_REF;
+       pdev->irq = -1;
+
+       INIT_WORK(&pdev->op_work, pcifront_do_aer);
+
+       dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
+               pdev, pdev->sh_info);
+out:
+       return pdev;
+}
+
+static void free_pdev(struct pcifront_device *pdev)
+{
+       dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
+
+       pcifront_free_roots(pdev);
+
+       /*For PCIE_AER error handling job*/
+       flush_scheduled_work();
+
+       if (pdev->irq >= 0)
+               unbind_from_irqhandler(pdev->irq, pdev);
+
+       if (pdev->evtchn != INVALID_EVTCHN)
+               xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
+
+       if (pdev->gnt_ref != INVALID_GRANT_REF)
+               gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
+                                         (unsigned long)pdev->sh_info);
+       else
+               free_page((unsigned long)pdev->sh_info);
+
+       dev_set_drvdata(&pdev->xdev->dev, NULL);
+
+       kfree(pdev);
+}
+
+static int pcifront_publish_info(struct pcifront_device *pdev)
+{
+       int err = 0;
+       struct xenbus_transaction trans;
+
+       err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info));
+       if (err < 0)
+               goto out;
+
+       pdev->gnt_ref = err;
+
+       err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
+       if (err)
+               goto out;
+
+       err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
+               0, "pcifront", pdev);
+
+       if (err < 0)
+               return err;
+
+       pdev->irq = err;
+
+do_publish:
+       err = xenbus_transaction_start(&trans);
+       if (err) {
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error writing configuration for backend "
+                                "(start transaction)");
+               goto out;
+       }
+
+       err = xenbus_printf(trans, pdev->xdev->nodename,
+                           "pci-op-ref", "%u", pdev->gnt_ref);
+       if (!err)
+               err = xenbus_printf(trans, pdev->xdev->nodename,
+                                   "event-channel", "%u", pdev->evtchn);
+       if (!err)
+               err = xenbus_printf(trans, pdev->xdev->nodename,
+                                   "magic", XEN_PCI_MAGIC);
+
+       if (err) {
+               xenbus_transaction_end(trans, 1);
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error writing configuration for backend");
+               goto out;
+       } else {
+               err = xenbus_transaction_end(trans, 0);
+               if (err == -EAGAIN)
+                       goto do_publish;
+               else if (err) {
+                       xenbus_dev_fatal(pdev->xdev, err,
+                                        "Error completing transaction "
+                                        "for backend");
+                       goto out;
+               }
+       }
+
+       xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
+
+       dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
+
+out:
+       return err;
+}
+
+static int __devinit pcifront_try_connect(struct pcifront_device *pdev)
+{
+       int err = -EFAULT;
+       int i, num_roots, len;
+       char str[64];
+       unsigned int domain, bus;
+
+
+       /* Only connect once */
+       if (xenbus_read_driver_state(pdev->xdev->nodename) !=
+           XenbusStateInitialised)
+               goto out;
+
+       err = pcifront_connect(pdev);
+       if (err) {
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error connecting PCI Frontend");
+               goto out;
+       }
+
+       err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
+                          "root_num", "%d", &num_roots);
+       if (err == -ENOENT) {
+               xenbus_dev_error(pdev->xdev, err,
+                                "No PCI Roots found, trying 0000:00");
+               err = pcifront_scan_root(pdev, 0, 0);
+               num_roots = 0;
+       } else if (err != 1) {
+               if (err == 0)
+                       err = -EINVAL;
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error reading number of PCI roots");
+               goto out;
+       }
+
+       for (i = 0; i < num_roots; i++) {
+               len = snprintf(str, sizeof(str), "root-%d", i);
+               if (unlikely(len >= (sizeof(str) - 1))) {
+                       err = -ENOMEM;
+                       goto out;
+               }
+
+               err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
+                                  "%x:%x", &domain, &bus);
+               if (err != 2) {
+                       if (err >= 0)
+                               err = -EINVAL;
+                       xenbus_dev_fatal(pdev->xdev, err,
+                                        "Error reading PCI root %d", i);
+                       goto out;
+               }
+
+               err = pcifront_scan_root(pdev, domain, bus);
+               if (err) {
+                       xenbus_dev_fatal(pdev->xdev, err,
+                                        "Error scanning PCI root %04x:%02x",
+                                        domain, bus);
+                       goto out;
+               }
+       }
+
+       err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
+
+out:
+       return err;
+}
+
+static int pcifront_try_disconnect(struct pcifront_device *pdev)
+{
+       int err = 0;
+       enum xenbus_state prev_state;
+
+
+       prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
+
+       if (prev_state >= XenbusStateClosing)
+               goto out;
+
+       if (prev_state == XenbusStateConnected) {
+               pcifront_free_roots(pdev);
+               pcifront_disconnect(pdev);
+       }
+
+       err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
+
+out:
+
+       return err;
+}
+
+static int __devinit pcifront_attach_devices(struct pcifront_device *pdev)
+{
+       int err = -EFAULT;
+       int i, num_roots, len;
+       unsigned int domain, bus;
+       char str[64];
+
+       if (xenbus_read_driver_state(pdev->xdev->nodename) !=
+           XenbusStateReconfiguring)
+               goto out;
+
+       err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
+                          "root_num", "%d", &num_roots);
+       if (err == -ENOENT) {
+               xenbus_dev_error(pdev->xdev, err,
+                                "No PCI Roots found, trying 0000:00");
+               err = pcifront_rescan_root(pdev, 0, 0);
+               num_roots = 0;
+       } else if (err != 1) {
+               if (err == 0)
+                       err = -EINVAL;
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error reading number of PCI roots");
+               goto out;
+       }
+
+       for (i = 0; i < num_roots; i++) {
+               len = snprintf(str, sizeof(str), "root-%d", i);
+               if (unlikely(len >= (sizeof(str) - 1))) {
+                       err = -ENOMEM;
+                       goto out;
+               }
+
+               err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
+                                  "%x:%x", &domain, &bus);
+               if (err != 2) {
+                       if (err >= 0)
+                               err = -EINVAL;
+                       xenbus_dev_fatal(pdev->xdev, err,
+                                        "Error reading PCI root %d", i);
+                       goto out;
+               }
+
+               err = pcifront_rescan_root(pdev, domain, bus);
+               if (err) {
+                       xenbus_dev_fatal(pdev->xdev, err,
+                                        "Error scanning PCI root %04x:%02x",
+                                        domain, bus);
+                       goto out;
+               }
+       }
+
+       xenbus_switch_state(pdev->xdev, XenbusStateConnected);
+
+out:
+       return err;
+}
+
+static int pcifront_detach_devices(struct pcifront_device *pdev)
+{
+       int err = 0;
+       int i, num_devs;
+       unsigned int domain, bus, slot, func;
+       struct pci_bus *pci_bus;
+       struct pci_dev *pci_dev;
+       char str[64];
+
+       if (xenbus_read_driver_state(pdev->xdev->nodename) !=
+           XenbusStateConnected)
+               goto out;
+
+       err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
+                          &num_devs);
+       if (err != 1) {
+               if (err >= 0)
+                       err = -EINVAL;
+               xenbus_dev_fatal(pdev->xdev, err,
+                                "Error reading number of PCI devices");
+               goto out;
+       }
+
+       /* Find devices being detached and remove them. */
+       for (i = 0; i < num_devs; i++) {
+               int l, state;
+               l = snprintf(str, sizeof(str), "state-%d", i);
+               if (unlikely(l >= (sizeof(str) - 1))) {
+                       err = -ENOMEM;
+                       goto out;
+               }
+               err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d",
+                                  &state);
+               if (err != 1)
+                       state = XenbusStateUnknown;
+
+               if (state != XenbusStateClosing)
+                       continue;
+
+               /* Remove device. */
+               l = snprintf(str, sizeof(str), "vdev-%d", i);
+               if (unlikely(l >= (sizeof(str) - 1))) {
+                       err = -ENOMEM;
+                       goto out;
+               }
+               err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
+                                  "%x:%x:%x.%x", &domain, &bus, &slot, &func);
+               if (err != 4) {
+                       if (err >= 0)
+                               err = -EINVAL;
+                       xenbus_dev_fatal(pdev->xdev, err,
+                                        "Error reading PCI device %d", i);
+                       goto out;
+               }
+
+               pci_bus = pci_find_bus(domain, bus);
+               if (!pci_bus) {
+                       dev_dbg(&pdev->xdev->dev, "Cannot get bus %04x:%02x\n",
+                               domain, bus);
+                       continue;
+               }
+               pci_dev = pci_get_slot(pci_bus, PCI_DEVFN(slot, func));
+               if (!pci_dev) {
+                       dev_dbg(&pdev->xdev->dev,
+                               "Cannot get PCI device %04x:%02x:%02x.%02x\n",
+                               domain, bus, slot, func);
+                       continue;
+               }
+               pci_remove_bus_device(pci_dev);
+               pci_dev_put(pci_dev);
+
+               dev_dbg(&pdev->xdev->dev,
+                       "PCI device %04x:%02x:%02x.%02x removed.\n",
+                       domain, bus, slot, func);
+       }
+
+       err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
+
+out:
+       return err;
+}
+
+static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev,
+                                                 enum xenbus_state be_state)
+{
+       struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
+
+       switch (be_state) {
+       case XenbusStateUnknown:
+       case XenbusStateInitialising:
+       case XenbusStateInitWait:
+       case XenbusStateInitialised:
+       case XenbusStateClosed:
+               break;
+
+       case XenbusStateConnected:
+               pcifront_try_connect(pdev);
+               break;
+
+       case XenbusStateClosing:
+               dev_warn(&xdev->dev, "backend going away!\n");
+               pcifront_try_disconnect(pdev);
+               break;
+
+       case XenbusStateReconfiguring:
+               pcifront_detach_devices(pdev);
+               break;
+
+       case XenbusStateReconfigured:
+               pcifront_attach_devices(pdev);
+               break;
+       }
+}
+
+static int pcifront_xenbus_probe(struct xenbus_device *xdev,
+                                const struct xenbus_device_id *id)
+{
+       int err = 0;
+       struct pcifront_device *pdev = alloc_pdev(xdev);
+
+       if (pdev == NULL) {
+               err = -ENOMEM;
+               xenbus_dev_fatal(xdev, err,
+                                "Error allocating pcifront_device struct");
+               goto out;
+       }
+
+       err = pcifront_publish_info(pdev);
+       if (err)
+               free_pdev(pdev);
+
+out:
+       return err;
+}
+
+static int pcifront_xenbus_remove(struct xenbus_device *xdev)
+{
+       struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
+       if (pdev)
+               free_pdev(pdev);
+
+       return 0;
+}
+
+static const struct xenbus_device_id xenpci_ids[] = {
+       {"pci"},
+       {""},
+};
+
+static struct xenbus_driver xenbus_pcifront_driver = {
+       .name                   = "pcifront",
+       .owner                  = THIS_MODULE,
+       .ids                    = xenpci_ids,
+       .probe                  = pcifront_xenbus_probe,
+       .remove                 = pcifront_xenbus_remove,
+       .otherend_changed       = pcifront_backend_changed,
+};
+
+static int __init pcifront_init(void)
+{
+       if (!xen_pv_domain() || xen_initial_domain())
+               return -ENODEV;
+
+       pci_frontend_registrar(1 /* enable */);
+
+       return xenbus_register_frontend(&xenbus_pcifront_driver);
+}
+
+static void __exit pcifront_cleanup(void)
+{
+       xenbus_unregister_driver(&xenbus_pcifront_driver);
+       pci_frontend_registrar(0 /* disable */);
+}
+module_init(pcifront_init);
+module_exit(pcifront_cleanup);
+
+MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("xen:pci");
diff --git a/include/xen/interface/io/pciif.h b/include/xen/interface/io/pciif.h
new file mode 100644 (file)
index 0000000..d9922ae
--- /dev/null
@@ -0,0 +1,112 @@
+/*
+ * PCI Backend/Frontend Common Data Structures & Macros
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
+ */
+#ifndef __XEN_PCI_COMMON_H__
+#define __XEN_PCI_COMMON_H__
+
+/* Be sure to bump this number if you change this file */
+#define XEN_PCI_MAGIC "7"
+
+/* xen_pci_sharedinfo flags */
+#define        _XEN_PCIF_active                (0)
+#define        XEN_PCIF_active                 (1<<_XEN_PCIF_active)
+#define        _XEN_PCIB_AERHANDLER            (1)
+#define        XEN_PCIB_AERHANDLER             (1<<_XEN_PCIB_AERHANDLER)
+#define        _XEN_PCIB_active                (2)
+#define        XEN_PCIB_active                 (1<<_XEN_PCIB_active)
+
+/* xen_pci_op commands */
+#define        XEN_PCI_OP_conf_read            (0)
+#define        XEN_PCI_OP_conf_write           (1)
+#define        XEN_PCI_OP_enable_msi           (2)
+#define        XEN_PCI_OP_disable_msi          (3)
+#define        XEN_PCI_OP_enable_msix          (4)
+#define        XEN_PCI_OP_disable_msix         (5)
+#define        XEN_PCI_OP_aer_detected         (6)
+#define        XEN_PCI_OP_aer_resume           (7)
+#define        XEN_PCI_OP_aer_mmio             (8)
+#define        XEN_PCI_OP_aer_slotreset        (9)
+
+/* xen_pci_op error numbers */
+#define        XEN_PCI_ERR_success             (0)
+#define        XEN_PCI_ERR_dev_not_found       (-1)
+#define        XEN_PCI_ERR_invalid_offset      (-2)
+#define        XEN_PCI_ERR_access_denied       (-3)
+#define        XEN_PCI_ERR_not_implemented     (-4)
+/* XEN_PCI_ERR_op_failed - backend failed to complete the operation */
+#define XEN_PCI_ERR_op_failed          (-5)
+
+/*
+ * it should be PAGE_SIZE-sizeof(struct xen_pci_op))/sizeof(struct msix_entry))
+ * Should not exceed 128
+ */
+#define SH_INFO_MAX_VEC                        128
+
+struct xen_msix_entry {
+       uint16_t vector;
+       uint16_t entry;
+};
+struct xen_pci_op {
+       /* IN: what action to perform: XEN_PCI_OP_* */
+       uint32_t cmd;
+
+       /* OUT: will contain an error number (if any) from errno.h */
+       int32_t err;
+
+       /* IN: which device to touch */
+       uint32_t domain; /* PCI Domain/Segment */
+       uint32_t bus;
+       uint32_t devfn;
+
+       /* IN: which configuration registers to touch */
+       int32_t offset;
+       int32_t size;
+
+       /* IN/OUT: Contains the result after a READ or the value to WRITE */
+       uint32_t value;
+       /* IN: Contains extra infor for this operation */
+       uint32_t info;
+       /*IN:  param for msi-x */
+       struct xen_msix_entry msix_entries[SH_INFO_MAX_VEC];
+};
+
+/*used for pcie aer handling*/
+struct xen_pcie_aer_op {
+       /* IN: what action to perform: XEN_PCI_OP_* */
+       uint32_t cmd;
+       /*IN/OUT: return aer_op result or carry error_detected state as input*/
+       int32_t err;
+
+       /* IN: which device to touch */
+       uint32_t domain; /* PCI Domain/Segment*/
+       uint32_t bus;
+       uint32_t devfn;
+};
+struct xen_pci_sharedinfo {
+       /* flags - XEN_PCIF_* */
+       uint32_t flags;
+       struct xen_pci_op op;
+       struct xen_pcie_aer_op aer_op;
+};
+
+#endif /* __XEN_PCI_COMMON_H__ */