[PATCH] Fix for USB media not found at boot

[Scott Long-2]

All,

I have a candidate patch to fix the problem with USB boot media not being
found in time at boot, leading to the "mountroot" prompt instead of a
booting system.  Please apply both patches below and let me know if it
works for you.

http://people.freebsd.org/~scottl/cam_hook.diff
http://people.freebsd.org/~scottl/usb_hook.diff
_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

On Friday 02 October 2009 23:11:21 Scott Long wrote:
> All,
>
> I have a candidate patch to fix the problem with USB boot media not being
> found in time at boot, leading to the "mountroot" prompt instead of a
> booting system.  Please apply both patches below and let me know if it
> works for you.

Hi,

Your patch looks OK. Some checkpoints however:

Does your patch work if ehci/ohci/uhci is kldloaded after system startup?

Does your patch work if CAM is not in the kernel?

I have a simple imporovement to your patch:

I would just add a synchronously blocking call to the end of "usb_bus_attach".

Something like:

#ifdef USB_HAVE_CAM
static void
usb_cam_wait_cb(void *arg)
{
        struct usb_bus *bus = *(struct usb_bus **)arg;

    USB_BUS_LOCK(bus);
    *(struct usb_bus **)arg = NULL;
        wakeup(arg);
    USB_BUS_UNLOCK(bus);
}

static void
usb_wait_cam(struct usb_bus *bus)
{
    struct intr_config_hook *hook;
    void *temp = bus;
    hook = malloc(sizeof(*hook), M_TEMP, M_ZERO);
    hook->ich_func = usb_wait_cam_cb;
    hook->ich_arg = &temp;

        config_intrhook_establish(hook);

    USB_BUS_LOCK(bus);
    if (temp != NULL)
                mtx_sleep(&temp, &bus->bus_mtx, 0, "WCAM", 0);
    USB_BUS_UNLOCK(bus);
}
#endif

That way makes the code more simple and avoids a race with the CAM callback
being called after that the hardware is detached. For example if the Host
Controller resides on a removable device.

There are some more USB_HAVE_XXX in sys/dev/usb/usb_freebsd.h .

--HPS

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Scott Long-2]

On Oct 3, 2009, at 1:53 AM, Hans Petter Selasky wrote:

> On Friday 02 October 2009 23:11:21 Scott Long wrote:
>> All,
>>
>> I have a candidate patch to fix the problem with USB boot media not  
>> being
>> found in time at boot, leading to the "mountroot" prompt instead of a
>> booting system.  Please apply both patches below and let me know if  
>> it
>> works for you.
>
> Hi,
>
> Your patch looks OK. Some checkpoints however:
>
> Does your patch work if ehci/ohci/uhci is kldloaded after system  
> startup?
>

That's a good question, I forgot to check that case.  It should,  
though, given how the config_intrhook API is designed to work.

> Does your patch work if CAM is not in the kernel?
>

The change has no dependence on CAM.

> I have a simple imporovement to your patch:
>
> I would just add a synchronously blocking call to the end of  
> "usb_bus_attach".
>

This isn't needed.  The config_intrhook system will sleep after all  
hooks have been launched, and will wakeup as needed.  There is no  
longer a race with CAM as CAM no longer uses the config_intrhook  
system, and is explicitly placed in order after it.

Unfortunately, my patch doesn't solve the problem.  It holds off the  
boot only while the root hub is explored.  Hubs below the root get  
queued back to the explore thread and enumerated after the root hub  
has signaled to disestablish the config hook, leading to devices  
further down the chain getting discovered late.  It's no better than  
the code that I deleted.  I'm thinking through how to solve this.  
Ideally, a refcount/semaphore is kept during discovery.  Every time a  
new hub is found, the semaphore is increased, the hub is reset/
configured, and control returns upwards until a config interrupt comes  
in.  Then we look at each port on the hub, probe/attach new devices,  
and then down the semaphore at the conclusion.  If one or more ports  
had hubs on them, then they would recurse this process.

I'm not familiar enough with USB to know if this is even possible.  I  
assume that a hub has to be reset before its ports can be probed, but  
I have no idea if there's a reliable way to know when the reset action  
has started and ended, other than to hope to get a config interrupt  
for one or more of the ports.  But if a hub is empty of child devices,  
will any interrupts be generated?  Maybe instead of relying on  
interrupts, we should just poll the ports for a period of time to see  
if they come alive.  Dunno.  I need to start reading specs.

Scott

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

On Saturday 03 October 2009 10:19:57 Scott Long wrote:
> config_intrhook system will sleep after all

Then why do you need the intr hook callback?

--HPS
_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Scott Long-2]

On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>> config_intrhook system will sleep after all
>
> Then why do you need the intr hook callback?
>

The config_intrhook lets you know that interrupts are enabled, the  
scheduler is running, and mountroot hasn't run yet.  It provides a  
very convenient and standard way to do exactly what we want with USB  
enumeration.

Scott

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Scott Long-2]

On Oct 3, 2009, at 9:05 AM, Scott Long wrote:

> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>> config_intrhook system will sleep after all
>>
>> Then why do you need the intr hook callback?
>>
>
> The config_intrhook lets you know that interrupts are enabled, the  
> scheduler is running, and mountroot hasn't run yet.  It provides a  
> very convenient and standard way to do exactly what we want with USB  
> enumeration.
>

Now that I'm awake, let me explain more.  I'm moving USB from the  
mount_root_hold() API to the config_intrhook() API because the latter  
provides proper boot ordering with CAM, which the former does not.  
Though my published patch doesn't rely on the CAM system in order to  
function, if you want umass devices to be discovered in time, you need  
it.  What I didn't realize when I published my patch is that my  
approach is incomplete; I'm working to fix that, but it will likely  
require some significant work inside of USB.

Scott

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

On Saturday 03 October 2009 17:05:35 Scott Long wrote:

> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
> > On Saturday 03 October 2009 10:19:57 Scott Long wrote:
> >> config_intrhook system will sleep after all
> >
> > Then why do you need the intr hook callback?
>
> The config_intrhook lets you know that interrupts are enabled, the
> scheduler is running, and mountroot hasn't run yet.  It provides a
> very convenient and standard way to do exactly what we want with USB
> enumeration.

Hi,

The root HUB attach and explore code is already running from a separate
thread, so won't that be superfluous? I mean, the HUB explore code for any USB
HUB will not run until the scheduler is running anyway.

In my opinion delaying the system until the boot disk is present is just not
good. We should rather be event driven, so that every time a new disk becomes
present it checks it for mountroot.

while (1) {
       if (mountroot is successful)
           break;
       if (ctrl+c is pressed)
           manual_mountroot();
       printf("Waiting 1 second for root-disk to appear. Press CTRL+C to
abort.\n");
           sleep(1);
}

--HPS

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Scott Long-2]

On Oct 3, 2009, at 10:00 AM, Hans Petter Selasky wrote:

> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>>> config_intrhook system will sleep after all
>>>
>>> Then why do you need the intr hook callback?
>>
>> The config_intrhook lets you know that interrupts are enabled, the
>> scheduler is running, and mountroot hasn't run yet.  It provides a
>> very convenient and standard way to do exactly what we want with USB
>> enumeration.
>
> Hi,
>
> The root HUB attach and explore code is already running from a  
> separate
> thread, so won't that be superfluous? I mean, the HUB explore code  
> for any USB
> HUB will not run until the scheduler is running anyway.
>
> In my opinion delaying the system until the boot disk is present is  
> just not
> good. We should rather be event driven, so that every time a new  
> disk becomes
> present it checks it for mountroot.
>
> while (1) {
>       if (mountroot is successful)
>           break;
>       if (ctrl+c is pressed)
>           manual_mountroot();
>       printf("Waiting 1 second for root-disk to appear. Press CTRL+C  
> to
> abort.\n");
>   sleep(1);
> }
>

That's an interesting idea.  If you can make this work, go for it.  
Meanwhile, I'm going to continue to look at the usb enumeration code.

Scott

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Scott Long-2]

On Oct 3, 2009, at 10:00 AM, Hans Petter Selasky wrote:

> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>>> config_intrhook system will sleep after all
>>>
>>> Then why do you need the intr hook callback?
>>
>> The config_intrhook lets you know that interrupts are enabled, the
>> scheduler is running, and mountroot hasn't run yet.  It provides a
>> very convenient and standard way to do exactly what we want with USB
>> enumeration.
>
> Hi,
>
> The root HUB attach and explore code is already running from a  
> separate
> thread, so won't that be superfluous? I mean, the HUB explore code  
> for any USB
> HUB will not run until the scheduler is running anyway.
>
> In my opinion delaying the system until the boot disk is present is  
> just not
> good. We should rather be event driven, so that every time a new  
> disk becomes
> present it checks it for mountroot.
>
> while (1) {
>       if (mountroot is successful)
>           break;
>       if (ctrl+c is pressed)
>           manual_mountroot();
>       printf("Waiting 1 second for root-disk to appear. Press CTRL+C  
> to
> abort.\n");
>   sleep(1);
> }
>

Actually, if you want to truly be event based, then instead of  
sleeping for a fixed time and then polling for devices, go to sleep  
and have GEOM wake you up when a disk device arrives.  You can still  
wake up periodically to provide status to the user (or not, it might  
get too noisy).

Scott

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

On Saturday 03 October 2009 18:05:32 Scott Long wrote:
> That's an interesting idea.  If you can make this work, go for it.  
> Meanwhile, I'm going to continue to look at the usb enumeration code.

Please find attached an outline. I need one more hour to finalize the patch.

All existing mount root features have been kept. The only difference is that
the system will not panic before 3 minutes, if no root disk is present. I
think that is acceptable.

--HPS


==== //depot/projects/usb/src/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c#9 - /home/some_dir/src/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c ====
@@ -3058,7 +3058,6 @@
  destroy_dev(zfsdev);
 }
 
-static struct root_hold_token *zfs_root_token;
 struct proc *zfsproc;
 
 uint_t zfs_fsyncer_key;
@@ -3071,8 +3070,6 @@
 
  switch (type) {
  case MOD_LOAD:
- zfs_root_token = root_mount_hold("ZFS");
-
  mutex_init(&zfs_share_lock, NULL, MUTEX_DEFAULT, NULL);
 
  spa_init(FREAD | FWRITE);
@@ -3083,7 +3080,6 @@
  tsd_create(&rrw_tsd_key, NULL);
 
  printf("ZFS storage pool version " SPA_VERSION_STRING "\n");
- root_mount_rel(zfs_root_token);
 
  zfsdev_init();
  break;
==== //depot/projects/usb/src/sys/dev/pccbb/pccbb.c#14 - /home/some_dir/src/sys/dev/pccbb/pccbb.c ====
@@ -500,15 +500,6 @@
  mtx_unlock(&Giant);
 
  /*
- * First time through we need to tell mountroot that we're
- * done.
- */
- if (sc->sc_root_token) {
- root_mount_rel(sc->sc_root_token);
- sc->sc_root_token = NULL;
- }
-
- /*
  * Wait until it has been 250ms since the last time we
  * get an interrupt.  We handle the rest of the interrupt
  * at the top of the loop.  Although we clear the bit in the
==== //depot/projects/usb/src/sys/dev/pccbb/pccbb_pci.c#11 - /home/some_dir/src/sys/dev/pccbb/pccbb_pci.c ====
@@ -439,7 +439,6 @@
  device_printf(brdev, "unable to create event thread.\n");
  panic("cbb_create_event_thread");
  }
- sc->sc_root_token = root_mount_hold(device_get_nameunit(sc->dev));
  return (0);
 err:
  if (sc->irq_res)
==== //depot/projects/usb/src/sys/dev/pccbb/pccbbvar.h#8 - /home/some_dir/src/sys/dev/pccbb/pccbbvar.h ====
@@ -88,7 +88,6 @@
  struct proc *event_thread;
  void (*chipinit)(struct cbb_softc *);
  int powerintr;
- struct root_hold_token *sc_root_token;
 };
 
 /* result of detect_card */
==== //depot/projects/usb/src/sys/dev/usb/controller/usb_controller.c#32 - /home/some_dir/src/sys/dev/usb/controller/usb_controller.c ====
@@ -78,11 +78,6 @@
     "Debug level");
 #endif
 
-static int usb_no_boot_wait = 0;
-TUNABLE_INT("hw.usb.no_boot_wait", &usb_no_boot_wait);
-SYSCTL_INT(_hw_usb, OID_AUTO, no_boot_wait, CTLFLAG_RDTUN, &usb_no_boot_wait, 0,
-    "No device enumerate waiting at boot.");
-
 static devclass_t usb_devclass;
 
 static device_method_t usb_methods[] = {
@@ -134,11 +129,6 @@
  return (ENXIO);
  }
 
- if (usb_no_boot_wait == 0) {
- /* delay vfs_mountroot until the bus is explored */
- bus->bus_roothold = root_mount_hold(device_get_nameunit(dev));
- }
-
  usb_attach_sub(dev, bus);
 
  return (0); /* return success */
@@ -161,12 +151,6 @@
  /* Stop power watchdog */
  usb_callout_drain(&bus->power_wdog);
 
- /* Let the USB explore process detach all devices. */
- if (bus->bus_roothold != NULL) {
- root_mount_rel(bus->bus_roothold);
- bus->bus_roothold = NULL;
- }
-
  USB_BUS_LOCK(bus);
  if (usb_proc_msignal(&bus->explore_proc,
     &bus->detach_msg[0], &bus->detach_msg[1])) {
@@ -239,10 +223,6 @@
  (udev->hub->explore) (udev);
  USB_BUS_LOCK(bus);
  }
- if (bus->bus_roothold != NULL) {
- root_mount_rel(bus->bus_roothold);
- bus->bus_roothold = NULL;
- }
 }
 
 /*------------------------------------------------------------------------*
==== //depot/projects/usb/src/sys/dev/usb/input/ukbd.c#32 - /home/some_dir/src/sys/dev/usb/input/ukbd.c ====
@@ -67,6 +67,7 @@
 #include <sys/callout.h>
 #include <sys/malloc.h>
 #include <sys/priv.h>
+#include <sys/kdb.h>
 
 #include <dev/usb/usb.h>
 #include <dev/usb/usbdi.h>
@@ -328,6 +329,9 @@
 {
  DPRINTFN(2, "polling\n");
 
+ if (kdb_active == 0)
+ return; /* Only poll if KDB is active */
+
  while (sc->sc_inputs == 0) {
 
  usbd_transfer_poll(sc->sc_xfer, UKBD_N_TRANSFER);
==== //depot/projects/usb/src/sys/dev/usb/usb_bus.h#15 - /home/some_dir/src/sys/dev/usb/usb_bus.h ====
@@ -51,7 +51,6 @@
 struct usb_bus {
  struct usb_bus_stat stats_err;
  struct usb_bus_stat stats_ok;
- struct root_hold_token *bus_roothold;
  /*
  * There are two callback processes. One for Giant locked
  * callbacks. One for non-Giant locked callbacks. This should
==== //depot/projects/usb/src/sys/geom/journal/g_journal.c#12 - /home/some_dir/src/sys/geom/journal/g_journal.c ====
@@ -2108,12 +2108,6 @@
  g_topology_unlock();
  last_write = time_second;
 
- if (sc->sc_rootmount != NULL) {
- GJ_DEBUG(1, "root_mount_rel %p", sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
-
  for (;;) {
  /* Get first request from the queue. */
  mtx_lock(&sc->sc_mtx);
@@ -2311,9 +2305,6 @@
  sc->sc_inactive.jj_queue = NULL;
  sc->sc_active.jj_queue = NULL;
 
- sc->sc_rootmount = root_mount_hold("GJOURNAL");
- GJ_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount);
-
  callout_init(&sc->sc_callout, CALLOUT_MPSAFE);
  if (md->md_type != GJ_TYPE_COMPLETE) {
  /*
@@ -2445,12 +2436,6 @@
 
  g_topology_unlock();
 
- if (sc->sc_rootmount != NULL) {
- GJ_DEBUG(1, "root_mount_rel %p", sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
-
  callout_drain(&sc->sc_callout);
  mtx_lock(&sc->sc_mtx);
  wakeup(sc);
==== //depot/projects/usb/src/sys/geom/journal/g_journal.h#2 - /home/some_dir/src/sys/geom/journal/g_journal.h ====
@@ -163,8 +163,6 @@
 
  struct callout sc_callout;
  struct proc *sc_worker;
-
- struct root_hold_token *sc_rootmount;
 };
 #define sc_dprovider sc_dconsumer->provider
 #define sc_jprovider sc_jconsumer->provider
==== //depot/projects/usb/src/sys/geom/mirror/g_mirror.c#7 - /home/some_dir/src/sys/geom/mirror/g_mirror.c ====
@@ -1726,13 +1726,6 @@
 static int
 g_mirror_try_destroy(struct g_mirror_softc *sc)
 {
-
- if (sc->sc_rootmount != NULL) {
- G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
-    sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
  g_topology_lock();
  if (!g_mirror_can_destroy(sc)) {
  g_topology_unlock();
@@ -2190,10 +2183,6 @@
  */
  callout_drain(&sc->sc_callout);
  sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
- G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
-    sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
  return;
  } else {
  return;
@@ -2217,10 +2206,6 @@
  if (ndisks == 0) {
  /* No valid disks found, destroy device. */
  sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
- G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p",
-    __LINE__, sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
  return;
  }
  } else {
@@ -2374,12 +2359,6 @@
  */
  if (sc->sc_provider == NULL)
  g_mirror_launch_provider(sc);
- if (sc->sc_rootmount != NULL) {
- G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p",
-    __LINE__, sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
  }
  /*
  * Genid should be bumped immediately, so do it here.
@@ -2907,8 +2886,6 @@
  G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).",
     sc->sc_name, sc->sc_ndisks, sc->sc_id);
 
- sc->sc_rootmount = root_mount_hold("GMIRROR");
- G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount);
  /*
  * Run timeout.
  */
==== //depot/projects/usb/src/sys/geom/mirror/g_mirror.h#2 - /home/some_dir/src/sys/geom/mirror/g_mirror.h ====
@@ -207,8 +207,6 @@
  struct mtx sc_events_mtx;
 
  struct callout sc_callout;
-
- struct root_hold_token *sc_rootmount;
 };
 #define sc_name sc_geom->name
 
==== //depot/projects/usb/src/sys/geom/part/g_part.c#19 - /home/some_dir/src/sys/geom/part/g_part.c ====
@@ -1448,7 +1448,6 @@
  struct g_geom *gp;
  struct g_part_entry *entry;
  struct g_part_table *table;
- struct root_hold_token *rht;
  int attr, depth;
  int error;
 
@@ -1470,7 +1469,6 @@
  return (NULL);
  }
 
- rht = root_mount_hold(mp->name);
  g_topology_unlock();
 
  /*
@@ -1515,13 +1513,11 @@
  g_part_new_provider(gp, table, entry);
  }
 
- root_mount_rel(rht);
  g_access(cp, -1, 0, 0);
  return (gp);
 
  fail:
  g_topology_lock();
- root_mount_rel(rht);
  g_access(cp, -1, 0, 0);
  g_part_wither(gp, error);
  return (NULL);
==== //depot/projects/usb/src/sys/geom/raid3/g_raid3.c#7 - /home/some_dir/src/sys/geom/raid3/g_raid3.c ====
@@ -1974,13 +1974,6 @@
  g_topology_assert_not();
  sx_assert(&sc->sc_lock, SX_XLOCKED);
 
- if (sc->sc_rootmount != NULL) {
- G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
-    sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
-
  g_topology_lock();
  if (!g_raid3_can_destroy(sc)) {
  g_topology_unlock();
@@ -2464,10 +2457,6 @@
  * Timeout expired, so destroy device.
  */
  sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
- G_RAID3_DEBUG(1, "root_mount_rel[%u] %p",
-    __LINE__, sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
  }
  return;
  }
@@ -2610,12 +2599,6 @@
  }
  if (sc->sc_provider == NULL)
  g_raid3_launch_provider(sc);
- if (sc->sc_rootmount != NULL) {
- G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
-    sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
  break;
  case G_RAID3_DEVICE_STATE_COMPLETE:
  /*
@@ -2643,12 +2626,6 @@
  }
  if (sc->sc_provider == NULL)
  g_raid3_launch_provider(sc);
- if (sc->sc_rootmount != NULL) {
- G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
-    sc->sc_rootmount);
- root_mount_rel(sc->sc_rootmount);
- sc->sc_rootmount = NULL;
- }
  break;
  default:
  KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name,
@@ -3193,9 +3170,6 @@
  G_RAID3_DEBUG(1, "Device %s created (%u components, id=%u).",
     sc->sc_name, sc->sc_ndisks, sc->sc_id);
 
- sc->sc_rootmount = root_mount_hold("GRAID3");
- G_RAID3_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount);
-
  /*
  * Run timeout.
  */
==== //depot/projects/usb/src/sys/geom/raid3/g_raid3.h#2 - /home/some_dir/src/sys/geom/raid3/g_raid3.h ====
@@ -240,8 +240,6 @@
  struct mtx sc_events_mtx;
 
  struct callout sc_callout;
-
- struct root_hold_token *sc_rootmount;
 };
 #define sc_name sc_geom->name
 
==== //depot/projects/usb/src/sys/kern/kern_cons.c#3 - /home/some_dir/src/sys/kern/kern_cons.c ====
@@ -353,8 +353,10 @@
 
  if (cn_mute)
  return (-1);
- while ((c = cncheckc()) == -1)
- ;
+ while ((c = cncheckc()) == -1) {
+ if (!kdb_active)
+ pause("WKEY", hz / 100); /* give USB a chance */
+ }
  if (c == '\r')
  c = '\n'; /* console input is always ICRNL */
  return (c);
==== //depot/projects/usb/src/sys/kern/vfs_mount.c#29 - /home/some_dir/src/sys/kern/vfs_mount.c ====
@@ -1344,82 +1344,9 @@
  *
  */
 
-struct root_hold_token {
- const char *who;
- LIST_ENTRY(root_hold_token) list;
-};
-
-static LIST_HEAD(, root_hold_token) root_holds =
-    LIST_HEAD_INITIALIZER(&root_holds);
-
 static int root_mount_complete;
 
 /*
- * Hold root mount.
- */
-struct root_hold_token *
-root_mount_hold(const char *identifier)
-{
- struct root_hold_token *h;
-
- if (root_mounted())
- return (NULL);
-
- h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
- h->who = identifier;
- mtx_lock(&mountlist_mtx);
- LIST_INSERT_HEAD(&root_holds, h, list);
- mtx_unlock(&mountlist_mtx);
- return (h);
-}
-
-/*
- * Release root mount.
- */
-void
-root_mount_rel(struct root_hold_token *h)
-{
-
- if (h == NULL)
- return;
- mtx_lock(&mountlist_mtx);
- LIST_REMOVE(h, list);
- wakeup(&root_holds);
- mtx_unlock(&mountlist_mtx);
- free(h, M_DEVBUF);
-}
-
-/*
- * Wait for all subsystems to release root mount.
- */
-static void
-root_mount_prepare(void)
-{
- struct root_hold_token *h;
- struct timeval lastfail;
- int curfail = 0;
-
- for (;;) {
- DROP_GIANT();
- g_waitidle();
- PICKUP_GIANT();
- mtx_lock(&mountlist_mtx);
- if (LIST_EMPTY(&root_holds)) {
- mtx_unlock(&mountlist_mtx);
- break;
- }
- if (ppsratecheck(&lastfail, &curfail, 1)) {
- printf("Root mount waiting for:");
- LIST_FOREACH(h, &root_holds, list)
- printf(" %s", h->who);
- printf("\n");
- }
- msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
-    hz);
- }
-}
-
-/*
  * Root was mounted, share the good news.
  */
 static void
@@ -1452,27 +1379,6 @@
  return (root_mount_complete);
 }
 
-/*
- * Wait until root is mounted.
- */
-void
-root_mount_wait(void)
-{
-
- /*
- * Panic on an obvious deadlock - the function can't be called from
- * a thread which is doing the whole SYSINIT stuff.
- */
- KASSERT(curthread->td_proc->p_pid != 0,
-    ("root_mount_wait: cannot be called from the swapper thread"));
- mtx_lock(&mountlist_mtx);
- while (!root_mount_complete) {
- msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
-    hz);
- }
- mtx_unlock(&mountlist_mtx);
-}
-
 static void
 set_rootvnode()
 {
@@ -1644,24 +1550,24 @@
 vfs_mountroot(void)
 {
  char *cp, *options;
- int error, i, asked = 0;
+ int error, i, t, asked = 0;
 
  options = NULL;
 
- root_mount_prepare();
-
  mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
     NULL, NULL, mount_init, mount_fini,
     UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  devfs_first();
 
+repeat:
+
  /*
  * We are booted with instructions to prompt for the root filesystem.
  */
  if (boothowto & RB_ASKNAME) {
  if (!vfs_mountroot_ask())
  goto mounted;
- asked = 1;
+ asked++;
  }
 
  options = getenv("vfs.root.mountfrom.options");
@@ -1715,14 +1621,35 @@
  if (ctrootdevname != NULL)
  if (!vfs_mountroot_try(ctrootdevname, options))
  goto mounted;
+
+ if (!asked) {
+ printf("\nWaiting for boot disk to appear. "
+    "Press CTRL+C to abort.\n");
+ }
+
+ asked ++;
+
+ t = 16;
+ i = -1;
+ while (t-- && (i == -1)) {
+ pause("WKEY", hz / 16); /* give USB a chance */
+ i = cncheckc();
+ }
+
+ if (i != -1)
+ printf("Got key: %d, %c\n", i, i);
+
  /*
  * Everything so far has failed, prompt on the console if we haven't
  * already tried that.
  */
- if (!asked)
+ if (i == 'c')
  if (!vfs_mountroot_ask())
  goto mounted;
 
+ if (asked < (3 * 60)) /* Wait a maximum of 3 minutes */
+ goto repeat;
+
  panic("Root mount failed, startup aborted.");
 
 mounted:
@@ -1875,6 +1802,7 @@
  char *options;
 
  for(;;) {
+ printf("\n");
  printf("Loader variables:\n");
  printf("vfs.root.mountfrom=");
  mountfrom = getenv("vfs.root.mountfrom");
==== //depot/projects/usb/src/sys/sys/systm.h#19 - /home/some_dir/src/sys/sys/systm.h ====
@@ -333,15 +333,8 @@
 /* XXX: Should be void nanodelay(u_int nsec); */
 void DELAY(int usec);
 
-/* Root mount holdback API */
-struct root_hold_token;
-
-struct root_hold_token *root_mount_hold(const char *identifier);
-void root_mount_rel(struct root_hold_token *h);
-void root_mount_wait(void);
 int root_mounted(void);
 
-
 /*
  * Unit number allocation API. (kern/subr_unit.c)
  */

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Marcel Moolenaar-4]

On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:

> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>>> config_intrhook system will sleep after all
>>>
>>> Then why do you need the intr hook callback?
>>
>> The config_intrhook lets you know that interrupts are enabled, the
>> scheduler is running, and mountroot hasn't run yet.  It provides a
>> very convenient and standard way to do exactly what we want with USB
>> enumeration.
>
> Hi,
>
> The root HUB attach and explore code is already running from a  
> separate
> thread, so won't that be superfluous? I mean, the HUB explore code  
> for any USB
> HUB will not run until the scheduler is running anyway.
>
> In my opinion delaying the system until the boot disk is present is  
> just not
> good. We should rather be event driven, so that every time a new  
> disk becomes
> present it checks it for mountroot.
>
> while (1) {
>       if (mountroot is successful)
>           break;
>       if (ctrl+c is pressed)
>           manual_mountroot();
>       printf("Waiting 1 second for root-disk to appear. Press CTRL+C  
> to
> abort.\n");
>   sleep(1);
> }

Yes.

The mount root code should keep a list of potential root devices to  
try and
it should try a device as soon as it appears. The current approach to  
block
the root mount simply because we want everything to be discovered  
before we
try to mount the root is preventing fast boots -- such as when the  
root is
a memory disk and you don't need to wait for anything...

Put differently: it's rather odd to hold off the root mount when the  
root
device is already present...

An approach like this also allows one to indefinitely wait for the root
device, which is a good feature to have...

--
Marcel Moolenaar
xcllnt@...



_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

On Saturday 03 October 2009 20:51:07 Marcel Moolenaar wrote:
> On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:
>
> An approach like this also allows one to indefinitely wait for the root
> device, which is a good feature to have...

Here is a patch you can try:

http://perforce.freebsd.org/chv.cgi?CH=169183

NOTE: I was not able to get any characters from gets() nor cncheckc(), so
there seems to be a bug somewhere. This used to work before at mount-root
time.

--HPS


_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Scott Long-2]

On Oct 3, 2009, at 12:51 PM, Marcel Moolenaar wrote:

>
> On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:
>
>> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
>>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>>>> config_intrhook system will sleep after all
>>>>
>>>> Then why do you need the intr hook callback?
>>>
>>> The config_intrhook lets you know that interrupts are enabled, the
>>> scheduler is running, and mountroot hasn't run yet.  It provides a
>>> very convenient and standard way to do exactly what we want with USB
>>> enumeration.
>>
>> Hi,
>>
>> The root HUB attach and explore code is already running from a  
>> separate
>> thread, so won't that be superfluous? I mean, the HUB explore code  
>> for any USB
>> HUB will not run until the scheduler is running anyway.
>>
>> In my opinion delaying the system until the boot disk is present is  
>> just not
>> good. We should rather be event driven, so that every time a new  
>> disk becomes
>> present it checks it for mountroot.
>>
>> while (1) {
>>      if (mountroot is successful)
>>          break;
>>      if (ctrl+c is pressed)
>>          manual_mountroot();
>>      printf("Waiting 1 second for root-disk to appear. Press CTRL+C  
>> to
>> abort.\n");
>>   sleep(1);
>> }
>
> Yes.
>
> The mount root code should keep a list of potential root devices to  
> try and
> it should try a device as soon as it appears. The current approach  
> to block
> the root mount simply because we want everything to be discovered  
> before we
> try to mount the root is preventing fast boots -- such as when the  
> root is
> a memory disk and you don't need to wait for anything...
>
> Put differently: it's rather odd to hold off the root mount when the  
> root
> device is already present...
>
> An approach like this also allows one to indefinitely wait for the  
> root
> device, which is a good feature to have...
>
>

When /etc/rc tries to mount everything in the fstab, it'll fail the  
boot if some of the devices haven't arrived in time.  An argument can  
be made for fixing that as well, but we're starting to get beyond on  
the scope of fixing what is needed for 8.0-RELEASE.

Scott


_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Marcel Moolenaar-4]

On Oct 3, 2009, at 4:52 PM, Scott Long wrote:
>
> When /etc/rc tries to mount everything in the fstab, it'll fail the  
> boot if some of the devices haven't arrived in time.  An argument  
> can be made for fixing that as well, but we're starting to get  
> beyond on the scope of fixing what is needed for 8.0-RELEASE.

Agreed.

--
Marcel Moolenaar
xcllnt@...



_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

On Saturday 03 October 2009 23:06:00 Hans Petter Selasky wrote:

> On Saturday 03 October 2009 20:51:07 Marcel Moolenaar wrote:
> > On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:
> >
> > An approach like this also allows one to indefinitely wait for the root
> > device, which is a good feature to have...
>
> Here is a patch you can try:
>
> http://perforce.freebsd.org/chv.cgi?CH=169183
>
> NOTE: I was not able to get any characters from gets() nor cncheckc(), so
> there seems to be a bug somewhere. This used to work before at mount-root
> time.
>

Hi,

I currently only see one problem. If any partitions that should be automounted
by /etc/fstab are not on the root disk, then /etc/fstab will fail, because the
current code will only wait for for the root mount partition. Probably the
script executing /etc/fstab should be updated to wait for disks aswell.

--HPS

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Chao Shin-3]

在 Sun, 04 Oct 2009 07:52:45 +0800，Scott Long <scottl@...> 写道:

Hi all

Is there any newer patch/workarounds available?

> On Oct 3, 2009, at 12:51 PM, Marcel Moolenaar wrote:
>
>>
>> On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:
>>
>>> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
>>>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>>>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>>>>> config_intrhook system will sleep after all
>>>>>
>>>>> Then why do you need the intr hook callback?
>>>>
>>>> The config_intrhook lets you know that interrupts are enabled, the
>>>> scheduler is running, and mountroot hasn't run yet.  It provides a
>>>> very convenient and standard way to do exactly what we want with USB
>>>> enumeration.
>>>
>>> Hi,
>>>
>>> The root HUB attach and explore code is already running from a separate
>>> thread, so won't that be superfluous? I mean, the HUB explore code for  
>>> any USB
>>> HUB will not run until the scheduler is running anyway.
>>>
>>> In my opinion delaying the system until the boot disk is present is  
>>> just not
>>> good. We should rather be event driven, so that every time a new disk  
>>> becomes
>>> present it checks it for mountroot.
>>>
>>> while (1) {
>>>      if (mountroot is successful)
>>>          break;
>>>      if (ctrl+c is pressed)
>>>          manual_mountroot();
>>>      printf("Waiting 1 second for root-disk to appear. Press CTRL+C to
>>> abort.\n");
>>>   sleep(1);
>>> }
>>
>> Yes.
>>
>> The mount root code should keep a list of potential root devices to try  
>> and
>> it should try a device as soon as it appears. The current approach to  
>> block
>> the root mount simply because we want everything to be discovered  
>> before we
>> try to mount the root is preventing fast boots -- such as when the root  
>> is
>> a memory disk and you don't need to wait for anything...
>>
>> Put differently: it's rather odd to hold off the root mount when the  
>> root
>> device is already present...
>>
>> An approach like this also allows one to indefinitely wait for the root
>> device, which is a good feature to have...
>>
>>
>
> When /etc/rc tries to mount everything in the fstab, it'll fail the boot  
> if some of the devices haven't arrived in time.  An argument can be made  
> for fixing that as well, but we're starting to get beyond on the scope  
> of fixing what is needed for 8.0-RELEASE.
>
> Scott
>
>
> _______________________________________________
> freebsd-current@... mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to  
> "freebsd-current-unsubscribe@..."

--
The Power to Serve
_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Hans Petter Selasky]

Hi,

There is a newer patch in USB P4.

The problem is when the /usr partition for example is not residing on the root
disk, which is the only disk the code is waiting for. Then the patch will make
the bootup fail.

--HPS

On Thursday 05 November 2009 04:34:33 Chao Shin wrote:

> 在 Sun, 04 Oct 2009 07:52:45 +0800，Scott Long <scottl@...> 写道:
>
> Hi all
>
> Is there any newer patch/workarounds available?
>
> > On Oct 3, 2009, at 12:51 PM, Marcel Moolenaar wrote:
> >> On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:
> >>> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
> >>>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
> >>>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
> >>>>>> config_intrhook system will sleep after all
> >>>>>
> >>>>> Then why do you need the intr hook callback?
> >>>>
> >>>> The config_intrhook lets you know that interrupts are enabled, the
> >>>> scheduler is running, and mountroot hasn't run yet.  It provides a
> >>>> very convenient and standard way to do exactly what we want with USB
> >>>> enumeration.
> >>>
> >>> Hi,
> >>>
> >>> The root HUB attach and explore code is already running from a separate
> >>> thread, so won't that be superfluous? I mean, the HUB explore code for
> >>> any USB
> >>> HUB will not run until the scheduler is running anyway.
> >>>
> >>> In my opinion delaying the system until the boot disk is present is
> >>> just not
> >>> good. We should rather be event driven, so that every time a new disk
> >>> becomes
> >>> present it checks it for mountroot.
> >>>
> >>> while (1) {
> >>>      if (mountroot is successful)
> >>>          break;
> >>>      if (ctrl+c is pressed)
> >>>          manual_mountroot();
> >>>      printf("Waiting 1 second for root-disk to appear. Press CTRL+C to
> >>> abort.\n");
> >>>   sleep(1);
> >>> }
> >>
> >> Yes.
> >>
> >> The mount root code should keep a list of potential root devices to try
> >> and
> >> it should try a device as soon as it appears. The current approach to
> >> block
> >> the root mount simply because we want everything to be discovered
> >> before we
> >> try to mount the root is preventing fast boots -- such as when the root
> >> is
> >> a memory disk and you don't need to wait for anything...
> >>
> >> Put differently: it's rather odd to hold off the root mount when the
> >> root
> >> device is already present...
> >>
> >> An approach like this also allows one to indefinitely wait for the root
> >> device, which is a good feature to have...
> >
> > When /etc/rc tries to mount everything in the fstab, it'll fail the boot
> > if some of the devices haven't arrived in time.  An argument can be made
> > for fixing that as well, but we're starting to get beyond on the scope
> > of fixing what is needed for 8.0-RELEASE.
> >
> > Scott
> >
> >
> > _______________________________________________
> > freebsd-current@... mailing list
> > http://lists.freebsd.org/mailman/listinfo/freebsd-current
> > To unsubscribe, send any mail to
> > "freebsd-current-unsubscribe@..."

_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Xin LI-2]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hans Petter Selasky wrote:
> Hi,
>
> There is a newer patch in USB P4.
>
> The problem is when the /usr partition for example is not residing on the root
> disk, which is the only disk the code is waiting for. Then the patch will make
> the bootup fail.

What if we teach rc.d scripts to do something like this:

do {
        mount -a && break;
        sleep 1; counter++;
} while counter < 30?

Will this make sense?  Or is there any other concerns?

> --HPS
>
> On Thursday 05 November 2009 04:34:33 Chao Shin wrote:
>> 在 Sun, 04 Oct 2009 07:52:45 +0800，Scott Long <scottl@...> 写道:
>>
>> Hi all
>>
>> Is there any newer patch/workarounds available?
>>
>>> On Oct 3, 2009, at 12:51 PM, Marcel Moolenaar wrote:
>>>> On Oct 3, 2009, at 9:00 AM, Hans Petter Selasky wrote:
>>>>> On Saturday 03 October 2009 17:05:35 Scott Long wrote:
>>>>>> On Oct 3, 2009, at 4:30 AM, Hans Petter Selasky wrote:
>>>>>>> On Saturday 03 October 2009 10:19:57 Scott Long wrote:
>>>>>>>> config_intrhook system will sleep after all
>>>>>>> Then why do you need the intr hook callback?
>>>>>> The config_intrhook lets you know that interrupts are enabled, the
>>>>>> scheduler is running, and mountroot hasn't run yet.  It provides a
>>>>>> very convenient and standard way to do exactly what we want with USB
>>>>>> enumeration.
>>>>> Hi,
>>>>>
>>>>> The root HUB attach and explore code is already running from a separate
>>>>> thread, so won't that be superfluous? I mean, the HUB explore code for
>>>>> any USB
>>>>> HUB will not run until the scheduler is running anyway.
>>>>>
>>>>> In my opinion delaying the system until the boot disk is present is
>>>>> just not
>>>>> good. We should rather be event driven, so that every time a new disk
>>>>> becomes
>>>>> present it checks it for mountroot.
>>>>>
>>>>> while (1) {
>>>>>      if (mountroot is successful)
>>>>>          break;
>>>>>      if (ctrl+c is pressed)
>>>>>          manual_mountroot();
>>>>>      printf("Waiting 1 second for root-disk to appear. Press CTRL+C to
>>>>> abort.\n");
>>>>>   sleep(1);
>>>>> }
>>>> Yes.
>>>>
>>>> The mount root code should keep a list of potential root devices to try
>>>> and
>>>> it should try a device as soon as it appears. The current approach to
>>>> block
>>>> the root mount simply because we want everything to be discovered
>>>> before we
>>>> try to mount the root is preventing fast boots -- such as when the root
>>>> is
>>>> a memory disk and you don't need to wait for anything...
>>>>
>>>> Put differently: it's rather odd to hold off the root mount when the
>>>> root
>>>> device is already present...
>>>>
>>>> An approach like this also allows one to indefinitely wait for the root
>>>> device, which is a good feature to have...
>>> When /etc/rc tries to mount everything in the fstab, it'll fail the boot
>>> if some of the devices haven't arrived in time.  An argument can be made
>>> for fixing that as well, but we're starting to get beyond on the scope
>>> of fixing what is needed for 8.0-RELEASE.
>>>
>>> Scott
>>>
>>>
>>> _______________________________________________
>>> freebsd-current@... mailing list
>>> http://lists.freebsd.org/mailman/listinfo/freebsd-current
>>> To unsubscribe, send any mail to
>>> "freebsd-current-unsubscribe@..."
>
> _______________________________________________
> freebsd-current@... mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

- --
Xin LI <delphij@...> http://www.delphij.net/
FreeBSD - The Power to Serve!       Live free or die
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.13 (FreeBSD)

iEYEARECAAYFAkr5rfQACgkQi+vbBBjt66CMEgCffmh1MK5RysLrL6vooObZQhUP
SyAAn2DNLOofpfkOR3aL6WUSwA+yAlw4
=panI
-----END PGP SIGNATURE-----
_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Marcel Moolenaar-4]

On Nov 10, 2009, at 10:16 AM, Xin LI wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Hans Petter Selasky wrote:
>> Hi,
>>
>> There is a newer patch in USB P4.
>>
>> The problem is when the /usr partition for example is not residing on the root
>> disk, which is the only disk the code is waiting for. Then the patch will make
>> the bootup fail.
>
> What if we teach rc.d scripts to do something like this:
>
> do {
> mount -a && break;
> sleep 1; counter++;
> } while counter < 30?
>
> Will this make sense?  Or is there any other concerns?

We should add more specific control by adding mount options.
Just like NFS mounts can be done in the background, so can
we have a mount option that doesn't cause a fail when the
underlying device isn't there, but instead results in mount
blocking until the device arrives. This then can be combined
with a background option to have the boot continue while the
mount is pending.

--
Marcel Moolenaar
xcllnt@...



_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."

[Xin LI-2]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi, Hans,

Hans Petter Selasky wrote:
> Hi,
>
> There is a newer patch in USB P4.

Which revisions are you referring to?  So far it looks like that only p4
changeset 169183:

    Mount Root Patch
    - This patch allows for late root device discovery. Instead of
      giving up at the first try, the system keeps trying for 3
      minutes to mount root. If CTRL+C is pressed during this time
      a mount-root menu will be shown. Else after 3 minutes the
      mount-root menu will be shown and then a panic will happen
      like before.
    - Clean up old mount root hold mechanism which did not
      work like expected.

Is related?

Cheers,
- --
Xin LI <delphij@...> http://www.delphij.net/
FreeBSD - The Power to Serve!       Live free or die
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.13 (FreeBSD)

iEYEARECAAYFAkr8sncACgkQi+vbBBjt66D+FwCgjGS74zSfEwZqcWsLO6eCFHy3
MVMAoJroxpPlKkdP6R+9Y22v5qWbUe6+
=ZqyA
-----END PGP SIGNATURE-----
_______________________________________________
freebsd-current@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@..."