Re: Driver not unset properly after kldunload
On Wed, 2006-12-06 at 11:35 -0500, John Baldwin wrote:
...
I don't see why this patch changes things. devclass_delete_device() only
clears dev-unit, dev-devclass, and dev-nameunit. device_set_driver()
doesn't check or clear any of those. In fact, your change does make
device_set_driver() not work at all since the device state will still be
DS_ATTACHED when you call device_set_driver() now. So, I guess your patch
actually makes the device _not_ be detached from the driver.
Forgive me for being so lazy :). I've analized the problem as I should
have done the first time and I hope I found the solution. First time I
was missleaded by sys/dev/pci/pci_user.c and the way it sets pd_name[]
in pci_devinfo struct (You actualy need to use pciconf atleast once
after loading the driver) ... It's only set when you use PCIOCGETCONF
ioctl on /dev/pci and it only sets it once (which is not very clever)
and never unset it. But one can unload the current driver and load
another (as in if_nfe, if_nve case) or just unload the driver and
pd_name[] will always show the first driver attached on that device.
So I hope this new patch is better one.
--- sys/dev/pci/pci_user.c.old Thu Dec 7 02:26:12 2006
+++ sys/dev/pci/pci_user.c Thu Dec 7 02:52:23 2006
@@ -303,14 +303,18 @@
/* Populate pd_name and pd_unit */
name = NULL;
- if (dinfo-cfg.dev dinfo-conf.pd_name[0] == '\0')
+ if (dinfo-cfg.dev) {
name = device_get_name(dinfo-cfg.dev);
- if (name) {
-strncpy(dinfo-conf.pd_name, name,
- sizeof(dinfo-conf.pd_name));
-dinfo-conf.pd_name[PCI_MAXNAMELEN] = 0;
-dinfo-conf.pd_unit =
- device_get_unit(dinfo-cfg.dev);
+if (name) {
+ strncpy(dinfo-conf.pd_name, name,
+ sizeof(dinfo-conf.pd_name));
+ dinfo-conf.pd_name[PCI_MAXNAMELEN] = 0;
+ dinfo-conf.pd_unit =
+ device_get_unit(dinfo-cfg.dev);
+} else {
+ dinfo-conf.pd_name[0] = '\0';
+ dinfo-conf.pd_unit = 0;
+}
}
if ((pattern_buf == NULL) ||
___
freebsd-hackers@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-hackers
To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: Driver not unset properly after kldunload
On Wednesday 29 November 2006 18:18, Iasen Kostoff wrote: Hi, It seems that after I load and then unload a driver its name stays linked to the device e.g: [EMAIL PROTECTED]:10:0: class=0x068000 card=0x81411043 chip=0x005710de rev=0xa3 hdr=0x00 but of course if_nfe is neither compiled in kernel nor is loaded as module anymore. I digged around in kernel and saw this in device_detach(): if (!(dev-flags DF_FIXEDCLASS)) devclass_delete_device(dev-devclass, dev); dev-state = DS_NOTPRESENT; device_set_driver(dev, NULL); device_set_desc(dev, NULL); device_sysctl_fini(dev); I've put some device_printf()s around and then looked at devclass_delete_device(). It destroys (frees) a lot of the info about the device and so the device_printf() prints device name as unknown (NULL). That seems to be a problem for at least device_set_driver(dev, NULL) - it doesn't unset the driver. I'm not so sure about the other 2 but I guess it's same there. So when I changed the order of this funcs everything worked fine (at least it looks like it worked fine :) I'm not absolutely sure that this won't broke something else). I've attached a patch for review. I don't see why this patch changes things. devclass_delete_device() only clears dev-unit, dev-devclass, and dev-nameunit. device_set_driver() doesn't check or clear any of those. In fact, your change does make device_set_driver() not work at all since the device state will still be DS_ATTACHED when you call device_set_driver() now. So, I guess your patch actually makes the device _not_ be detached from the driver. -- John Baldwin ___ freebsd-hackers@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-hackers To unsubscribe, send any mail to [EMAIL PROTECTED]
Driver not unset properly after kldunload
Hi, It seems that after I load and then unload a driver its name stays linked to the device e.g: [EMAIL PROTECTED]:10:0: class=0x068000 card=0x81411043 chip=0x005710de rev=0xa3 hdr=0x00 but of course if_nfe is neither compiled in kernel nor is loaded as module anymore. I digged around in kernel and saw this in device_detach(): if (!(dev-flags DF_FIXEDCLASS)) devclass_delete_device(dev-devclass, dev); dev-state = DS_NOTPRESENT; device_set_driver(dev, NULL); device_set_desc(dev, NULL); device_sysctl_fini(dev); I've put some device_printf()s around and then looked at devclass_delete_device(). It destroys (frees) a lot of the info about the device and so the device_printf() prints device name as unknown (NULL). That seems to be a problem for at least device_set_driver(dev, NULL) - it doesn't unset the driver. I'm not so sure about the other 2 but I guess it's same there. So when I changed the order of this funcs everything worked fine (at least it looks like it worked fine :) I'm not absolutely sure that this won't broke something else). I've attached a patch for review. (I've filled a PR http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/104777) --- subr_bus.c.bak Sun Oct 22 18:42:11 2006 +++ subr_bus.c Mon Oct 23 00:27:22 2006 @@ -2417,13 +2417,14 @@ if (dev-parent) BUS_CHILD_DETACHED(dev-parent, dev); + device_set_driver(dev, NULL); + device_set_desc(dev, NULL); + device_sysctl_fini(dev); + if (!(dev-flags DF_FIXEDCLASS)) devclass_delete_device(dev-devclass, dev); dev-state = DS_NOTPRESENT; - device_set_driver(dev, NULL); - device_set_desc(dev, NULL); - device_sysctl_fini(dev); return (0); } ___ freebsd-hackers@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-hackers To unsubscribe, send any mail to [EMAIL PROTECTED]
