pcifront: implement dynamic connections and disconnections

this patch implements dynamic connections and disconnections in
pcifront.
This feature is required to properly support pci hotplug, because when
no pci devices are assigned to a guest, xend will remove the pci
backend altogether.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>

diff --git a/extras/mini-os/include/pcifront.h b/extras/mini-os/include/pcifront.h
index 3bb37139a66b1aa042c6336ff26ebf39d6b50f5b..0a6be8eb6379462a87194d99bfe53c7a039c5f9a 100644 (file)
--- a/extras/mini-os/include/pcifront.h
+++ b/extras/mini-os/include/pcifront.h
@@ -1,6 +1,7 @@
 #include <mini-os/types.h>
 #include <xen/io/pciif.h>
 struct pcifront_dev;
+void pcifront_watches(void *opaque);
 struct pcifront_dev *init_pcifront(char *nodename);
 void pcifront_op(struct pcifront_dev *dev, struct xen_pci_op *op);
 void pcifront_scan(struct pcifront_dev *dev, void (*fun)(unsigned int domain, unsigned int bus, unsigned slot, unsigned int fun));

diff --git a/extras/mini-os/main.c b/extras/mini-os/main.c
index ebdab3392187c7c9e9e515a8f85570a1ccf73229..8b279a666ce8d848746840b533b4163edfaa4d4f 100644 (file)
--- a/extras/mini-os/main.c
+++ b/extras/mini-os/main.c
@@ -9,6 +9,7 @@
 #include <sched.h>
 #include <console.h>
 #include <netfront.h>
+#include <pcifront.h>
 #include <time.h>
 #include <stdlib.h>
 #include <unistd.h>
@@ -67,6 +68,7 @@ static void call_main(void *p)
 #endif
     init_fs_frontend();
 #endif
+    create_thread("pcifront", pcifront_watches, NULL);
 
 #ifdef CONFIG_QEMU
     /* Fetch argc, argv from XenStore */

diff --git a/extras/mini-os/pcifront.c b/extras/mini-os/pcifront.c
index afbe0fc9d715a08fc7a75c6543baed865db6cf79..ed402a8a194fefd0baa1b4a029d54da6418a4edb 100644 (file)
--- a/extras/mini-os/pcifront.c
+++ b/extras/mini-os/pcifront.c
@@ -13,10 +13,12 @@
 #include <mini-os/xmalloc.h>
 #include <mini-os/wait.h>
 #include <mini-os/pcifront.h>
+#include <mini-os/sched.h>
 
 #define PCI_DEVFN(slot, func) ((((slot) & 0x1f) << 3) | ((func) & 0x07))
 
 DECLARE_WAIT_QUEUE_HEAD(pcifront_queue);
+static struct pcifront_dev *pcidev;
 
 struct pcifront_dev {
     domid_t dom;
@@ -38,19 +40,103 @@ void pcifront_handler(evtchn_port_t port, struct pt_regs *regs, void *data)
 
 static void free_pcifront(struct pcifront_dev *dev)
 {
-    mask_evtchn(dev->evtchn);
+    if (!dev)
+        dev = pcidev;
 
-    free(dev->backend);
+    mask_evtchn(dev->evtchn);
 
     gnttab_end_access(dev->info_ref);
     free_page(dev->info);
 
     unbind_evtchn(dev->evtchn);
 
+    free(dev->backend);
     free(dev->nodename);
     free(dev);
 }
 
+void pcifront_watches(void *opaque)
+{
+    XenbusState state;
+    char *err = NULL, *msg = NULL;
+    char *be_path, *be_state;
+    char* nodename = opaque ? opaque : "device/pci/0";
+    char path[strlen(nodename) + 9];
+    char fe_state[strlen(nodename) + 7];
+    xenbus_event_queue events = NULL;
+
+    snprintf(path, sizeof(path), "%s/backend", nodename);
+    snprintf(fe_state, sizeof(fe_state), "%s/state", nodename);
+
+    while (1) {
+        printk("pcifront_watches: waiting for backend path to happear %s\n", path);
+        xenbus_watch_path_token(XBT_NIL, path, path, &events);
+        while ((err = xenbus_read(XBT_NIL, path, &be_path)) != NULL) {
+            free(err);
+            xenbus_wait_for_watch(&events);
+        }
+        xenbus_unwatch_path_token(XBT_NIL, path, path);
+        printk("pcifront_watches: waiting for backend to get into the right state %s\n", be_path);
+        be_state = (char *) malloc(strlen(be_path) +  7);
+        snprintf(be_state, strlen(be_path) +  7, "%s/state", be_path);
+        xenbus_watch_path_token(XBT_NIL, be_state, be_state, &events);
+        while ((err = xenbus_read(XBT_NIL, be_state, &msg)) != NULL || msg[0] > '4') {
+            free(msg);
+            free(err);
+            xenbus_wait_for_watch(&events);
+        }
+        xenbus_unwatch_path_token(XBT_NIL, be_state, be_state);
+        if (init_pcifront(NULL) == NULL) {
+            free(be_state);
+            free(be_path);
+            continue;
+        }
+        xenbus_watch_path_token(XBT_NIL, be_state, be_state, &events);
+        state = XenbusStateConnected;
+        printk("pcifront_watches: waiting for backend events %s\n", be_state);
+        while ((err = xenbus_wait_for_state_change(be_state, &state, &events)) == NULL &&
+               (err = xenbus_read(XBT_NIL, pcidev->backend, &msg)) == NULL) {
+            free(msg);
+            printk("pcifront_watches: backend state changed: %s %d\n", be_state, state);
+            if (state == XenbusStateReconfiguring) {
+                printk("pcifront_watches: writing %s %d\n", fe_state, XenbusStateReconfiguring);
+                if ((err = xenbus_switch_state(XBT_NIL, fe_state, XenbusStateReconfiguring)) != NULL) {
+                    printk("pcifront_watches: error changing state to %d: %s\n",
+                            XenbusStateReconfiguring, err);
+                    if (!strcmp(err, "ENOENT")) {
+                        xenbus_write(XBT_NIL, fe_state, "7");
+                        free(err);
+                    }
+                }
+            } else if (state == XenbusStateReconfigured) {
+                printk("pcifront_watches: writing %s %d\n", fe_state, XenbusStateConnected);
+                printk("pcifront_watches: changing state to %d\n", XenbusStateConnected);
+                if ((err = xenbus_switch_state(XBT_NIL, fe_state, XenbusStateConnected)) != NULL) {
+                    printk("pcifront_watches: error changing state to %d: %s\n",
+                            XenbusStateConnected, err);
+                    if (!strcmp(err, "ENOENT")) {
+                        xenbus_write(XBT_NIL, fe_state, "4");
+                        free(err);
+                    }
+                }
+            } else if (state == XenbusStateClosing)
+                break;
+        }
+        if (err)
+            printk("pcifront_watches: done waiting err=%s\n", err);
+        else
+            printk("pcifront_watches: done waiting\n");
+        xenbus_unwatch_path_token(XBT_NIL, be_state, be_state);
+        shutdown_pcifront(pcidev);
+        free(be_state);
+        free(be_path);
+        free(err);
+        pcidev = NULL;
+    }
+
+    xenbus_unwatch_path_token(XBT_NIL, path, path);
+}
+
 struct pcifront_dev *init_pcifront(char *_nodename)
 {
     xenbus_transaction_t xbt;
@@ -65,6 +151,9 @@ struct pcifront_dev *init_pcifront(char *_nodename)
 
     char path[strlen(nodename) + 1 + 10 + 1];
 
+    if (!_nodename && pcidev)
+        return pcidev;
+
     printk("******************* PCIFRONT for %s **********\n\n\n", nodename);
 
     snprintf(path, sizeof(path), "%s/backend-id", nodename);
@@ -173,6 +262,9 @@ done:
 
     printk("**************************\n");
 
+    if (!_nodename)
+        pcidev = dev;
+
     return dev;
 
 error:
@@ -182,16 +274,25 @@ error:
 
 void pcifront_scan(struct pcifront_dev *dev, void (*func)(unsigned int domain, unsigned int bus, unsigned slot, unsigned int fun))
 {
-    char path[strlen(dev->backend) + 1 + 5 + 10 + 1];
-    int i, n;
+    char *path;
+    int i, n, len;
     char *s, *msg;
     unsigned int domain, bus, slot, fun;
 
-    snprintf(path, sizeof(path), "%s/num_devs", dev->backend);
+    if (!dev)
+        dev = pcidev;
+    if (!dev)
+        dev = init_pcifront(NULL);
+    if (!dev)
+        return;
+
+    len = strlen(dev->backend) + 1 + 5 + 10 + 1;
+    path = (char *) malloc(len);
+    snprintf(path, len, "%s/num_devs", dev->backend);
     n = xenbus_read_integer(path);
 
     for (i = 0; i < n; i++) {
-        snprintf(path, sizeof(path), "%s/dev-%d", dev->backend, i);
+        snprintf(path, len, "%s/dev-%d", dev->backend, i);
         msg = xenbus_read(XBT_NIL, path, &s);
         if (msg) {
             printk("Error %s when reading the PCI root name at %s\n", msg, path);
@@ -205,8 +306,10 @@ void pcifront_scan(struct pcifront_dev *dev, void (*func)(unsigned int domain, u
         }
         free(s);
 
-        func(domain, bus, slot, fun);
+        if (func)
+            func(domain, bus, slot, fun);
     }
+    free(path);
 }
 
 void shutdown_pcifront(struct pcifront_dev *dev)
@@ -271,6 +374,9 @@ int pcifront_physical_to_virtual (struct pcifront_dev *dev,
     char *s, *msg = NULL;
     unsigned int dom1, bus1, slot1, fun1;
 
+    if (!dev)
+        dev = pcidev;
+
     snprintf(path, sizeof(path), "%s/num_devs", dev->backend);
     n = xenbus_read_integer(path);
 
@@ -312,6 +418,8 @@ int pcifront_physical_to_virtual (struct pcifront_dev *dev,
 
 void pcifront_op(struct pcifront_dev *dev, struct xen_pci_op *op)
 {
+    if (!dev)
+        dev = pcidev;
     dev->info->op = *op;
     /* Make sure info is written before the flag */
     wmb();
@@ -332,6 +440,8 @@ int pcifront_conf_read(struct pcifront_dev *dev,
 {
     struct xen_pci_op op;
 
+    if (!dev)
+        dev = pcidev;
     if (pcifront_physical_to_virtual(dev, &dom, &bus, &slot, &fun) < 0)
         return XEN_PCI_ERR_dev_not_found;
     memset(&op, 0, sizeof(op));
@@ -360,6 +470,8 @@ int pcifront_conf_write(struct pcifront_dev *dev,
 {
     struct xen_pci_op op;
 
+    if (!dev)
+        dev = pcidev;
     if (pcifront_physical_to_virtual(dev, &dom, &bus, &slot, &fun) < 0)
         return XEN_PCI_ERR_dev_not_found;
     memset(&op, 0, sizeof(op));
@@ -384,6 +496,8 @@ int pcifront_enable_msi(struct pcifront_dev *dev,
 {
     struct xen_pci_op op;
 
+    if (!dev)
+        dev = pcidev;
     if (pcifront_physical_to_virtual(dev, &dom, &bus, &slot, &fun) < 0)
         return XEN_PCI_ERR_dev_not_found;
     memset(&op, 0, sizeof(op));
@@ -407,6 +521,8 @@ int pcifront_disable_msi(struct pcifront_dev *dev,
 {
     struct xen_pci_op op;
 
+    if (!dev)
+        dev = pcidev;
     if (pcifront_physical_to_virtual(dev, &dom, &bus, &slot, &fun) < 0)
         return XEN_PCI_ERR_dev_not_found;
     memset(&op, 0, sizeof(op));
@@ -428,6 +544,8 @@ int pcifront_enable_msix(struct pcifront_dev *dev,
 {
     struct xen_pci_op op;
 
+    if (!dev)
+        dev = pcidev;
     if (pcifront_physical_to_virtual(dev, &dom, &bus, &slot, &fun) < 0)
         return XEN_PCI_ERR_dev_not_found;
     if (n > SH_INFO_MAX_VEC)
@@ -460,6 +578,8 @@ int pcifront_disable_msix(struct pcifront_dev *dev,
 {
     struct xen_pci_op op;
 
+    if (!dev)
+        dev = pcidev;
     if (pcifront_physical_to_virtual(dev, &dom, &bus, &slot, &fun) < 0)
         return XEN_PCI_ERR_dev_not_found;
     memset(&op, 0, sizeof(op));

diff --git a/extras/mini-os/xenbus/xenbus.c b/extras/mini-os/xenbus/xenbus.c
index 916a3896440503b52161eb6b8246e55499b4f927..c4e6abd8a4032788d053d566aac9e3ce99eb8479 100644 (file)
--- a/extras/mini-os/xenbus/xenbus.c
+++ b/extras/mini-os/xenbus/xenbus.c
@@ -96,7 +96,10 @@ void xenbus_wait_for_watch(xenbus_event_queue *queue)
     if (!queue)
         queue = &xenbus_events;
     ret = xenbus_wait_for_watch_return(queue);
-    free(ret);
+    if (ret)
+        free(ret);
+    else
+        printk("unexpected path returned by watch\n");
 }
 
 char* xenbus_wait_for_value(const char* path, const char* value, xenbus_event_queue *queue)

diff --git a/stubdom/pciutils.patch b/stubdom/pciutils.patch
index 0190428f269b1f206dcb216e12d26ef7adfc1001..2147ad4aecf87b637ddc7175f90600f1fb06cf81 100644 (file)
--- a/stubdom/pciutils.patch
+++ b/stubdom/pciutils.patch
@@ -23,14 +23,6 @@ diff -urN pciutils-2.2.9.orig/lib/access.c pciutils-2.2.9/lib/access.c
    PCI_ACCESS_MAX
  };
  
-@@ -63,6 +64,7 @@
-   int fd_rw;                          /* proc: fd opened read-write */
-   struct pci_dev *cached_dev;         /* proc: device the fd is for */
-   int fd_pos;                         /* proc: current position */
-+  void *minios;
- };
- 
- /* Initialize PCI access */
 --- pciutils-2.2.9.orig/lib/internal.h 2006-09-09 11:52:47.000000000 +0100
 +++ pciutils-2.2.9/lib/internal.h      2008-07-01 10:46:24.968202000 +0100
 @@ -37,4 +37,4 @@
@@ -95,24 +87,19 @@ diff -urN pciutils-2.2.9.orig/lib/access.c pciutils-2.2.9/lib/access.c
 +static void
 +minios_init(struct pci_access *a)
 +{
-+  a->minios = init_pcifront(NULL);
-+  if (!a->minios)
++  if (!init_pcifront(NULL))
 +    a->warning("minios_init open failed");
 +}
 +
 +static void
 +minios_cleanup(struct pci_access *a)
 +{
-+  if (a->minios)
-+    shutdown_pcifront(a->minios);
++  shutdown_pcifront(NULL);
 +}
 +
 +static void
 +minios_scan(struct pci_access *a)
 +{
-+  if (!a->minios)
-+    return;
-+
 +  void func(unsigned int domain, unsigned int bus, unsigned int slot, unsigned int fun)
 +  {
 +    struct pci_dev *d = pci_alloc_dev(a);
@@ -125,7 +112,7 @@ diff -urN pciutils-2.2.9.orig/lib/access.c pciutils-2.2.9/lib/access.c
 +    pci_link_dev(a, d);
 +  }
 +
-+  pcifront_scan(a->minios, func);
++  pcifront_scan(NULL, func);
 +}
 +
 +static int
@@ -134,17 +121,17 @@ diff -urN pciutils-2.2.9.orig/lib/access.c pciutils-2.2.9/lib/access.c
 +  unsigned int val;
 +  switch (len) {
 +    case 1:
-+      if (pcifront_conf_read(d->access->minios, d->domain, d->bus, d->dev, d->func, pos, len, &val))
++      if (pcifront_conf_read(NULL, d->domain, d->bus, d->dev, d->func, pos, len, &val))
 +        return 0;
 +      * buf = val;
 +      return 1;
 +    case 2:
-+      if (pcifront_conf_read(d->access->minios, d->domain, d->bus, d->dev, d->func, pos, len, &val))
++      if (pcifront_conf_read(NULL, d->domain, d->bus, d->dev, d->func, pos, len, &val))
 +        return 0;
 +      *(u16 *) buf = cpu_to_le16((u16) val);
 +      return 1;
 +    case 4:
-+      if (pcifront_conf_read(d->access->minios, d->domain, d->bus, d->dev, d->func, pos, len, &val))
++      if (pcifront_conf_read(NULL, d->domain, d->bus, d->dev, d->func, pos, len, &val))
 +        return 0;
 +      *(u32 *) buf = cpu_to_le32((u32) val);
 +      return 1;
@@ -170,7 +157,7 @@ diff -urN pciutils-2.2.9.orig/lib/access.c pciutils-2.2.9/lib/access.c
 +    default:
 +      return pci_generic_block_write(d, pos, buf, len);
 +  }
-+  return !pcifront_conf_write(d->access->minios, d->domain, d->bus, d->dev, d->func, pos, len, val);
++  return !pcifront_conf_write(NULL, d->domain, d->bus, d->dev, d->func, pos, len, val);
 +}
 +
 +struct pci_methods pm_minios = {