On Tue, 15 Sep 2009, Kay Sievers wrote: > On Tue, Sep 15, 2009 at 18:57, Greg KH <greg@xxxxxxxxx> wrote: > > On Tue, Sep 15, 2009 at 10:47:11AM -0500, James Bottomley wrote: > >> On Tue, 2009-09-15 at 08:33 -0700, Greg KH wrote: > >> > On Sat, Sep 12, 2009 at 09:30:07AM -0500, James Bottomley wrote: > >> > > On Fri, 2009-09-11 at 21:33 -0700, Greg KH wrote: > >> > > > On Fri, Sep 11, 2009 at 07:56:23PM -0700, Andrew Vasquez wrote: > >> > > > > On Fri, 11 Sep 2009, Greg KH wrote: > >> > > > > > >> > > > > > On Fri, Sep 11, 2009 at 05:38:08PM -0700, Andrew Vasquez wrote: > >> > > > > > > Randy Dunlap noted: > >> > > > > > > > >> > > > > > > when CONFIG_MODULES=n: > >> > > > > > > > >> > > > > > > drivers/scsi/qla2xxx/qla_os.c:2685: error: dereferencing pointer to incomplete type > >> > > > > > > > >> > > > > > > in > >> > > > > > > > >> > > > > > > kobject_uevent_env(&(&vha->hw->pdev->driver->driver)->owner->mkobj.kobj, > >> > > > > > > KOBJ_CHANGE, envp); > >> > > > > > > > >> > > > > > > Trigger kobject event on the 'struct device' hanging off the pci_dev. > >> > > > > > > >> > > > > > Um, why? What are you trying to do here? kobject change should not be > >> > > > > > for a device, or a "normal" kobject. > >> > > > > > > >> > > > > > What do you expect userspace to do with this? Where have you documented > >> > > > > > it? > >> > > > > > >> > > > > The purpose was described here: > >> > > > > > >> > > > > http://article.gmane.org/gmane.linux.scsi/54155 > >> > > > > > >> > > > > Basically we'd like to instruct user-space to retrieve a blob of data > >> > > > > automatically. > >> > > > > >> > > > Hm, like a firmware object perhaps? > >> > > > > >> > > > > Original implementation used the kboject hanging off > >> > > > > the module which does not exist when CONFIG_MODULES=n. It was > >> > > > > suggested that perhaps an alternative would be to use 'struct device' > >> > > > > kobj. Any tips on how to trigger such a driver-specific event, > >> > > > > perhaps a dedicated kobject exported by the driver itself??? > >> > > > > >> > > > Why not use the firmware interface for it, that is what it is designed > >> > > > for, and you will not have to craft any new udev rules. > >> > > > >> > > The data is going the wrong way to use the current firmware interface, > >> > > which is designed to load data from userspace into the kernel. For this > >> > > interface, we want the data to go the other way (i.e. the kernel has a > >> > > blob of dump data it would like userspace to save if it can). > >> > > >> > Ick. > >> > >> It's a natural debugging event ... the user can simply program the > >> system to discard the dump. > >> > >> > > I'd be amenable to updating the firmware interface to do this, but it > >> > > looks like adding a completely new codepath, which it's not clear even > >> > > belongs there. > >> > > >> > Well, I don't think it deserves a kobject change event, as that means > >> > something a bit different to userspace today, right? > >> > >> Well, the documentation (in include/linux/kobject.h) does say that > >> KOBJ_CHANGED is the dumping ground for events that don't match anything > >> else, which this one seems to qualify as. What else do you suggest? > > > > Ok, I can't think of anything else at the moment, sorry. > > Wouldn't kernel/kmod.c :: call_usermodehelper_pipe() somehow fit for > that interface? The only current consumer of call_usermodehelper_pipe() is fs/exec.c::do_coredump() and it appears to manipulate the blob of data via file-level operations. Would this be a 'notification' alternative? Or, are you suggesting the whole process be redone (both notification and blob-transfer)? Uevent notifications with a udev user-space front-end appears to be a much more simplified and flexible mechanism. -- av -- To unsubscribe from this list: send the line "unsubscribe linux-next" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
