On Thu, Mar 24, 2005 at 04:04:25PM -0500, Jeff Garzik wrote: > Greg KH wrote: > >On Tue, Mar 22, 2005 at 09:02:24PM -0500, Jeff Garzik wrote: > > > >>Kylene Hall wrote: > >> > >>>>what is the purpose of this pci_dev_get/put? attempting to prevent > >>>>hotplug or > >>>>something? > >>> > >>> > >>>Seems that since there is a refernce to the device in the chip structure > >>>and I am making the file private data pointer point to that chip > >>>structure this is another reference that must be accounted for. If you > >>>remove it with it open and attempt read or write bad things will happen. > >>>This isn't really hotpluggable either as the TPM is on the motherboard. > >> > >>My point was that there will always be a reference -anyway-, AFAICS. > >>There is a pci_dev reference assigned to the pci_driver when the PCI > >>driver is loaded, and all uses by the TPM generic code of this pointer > >>are -inside- the pci_driver's pci_dev object lifetime. > > > > > >Think of the following situation: > > - driver is bound to device. > > - userspace opens char dev node. > > - device is removed from the system (using fakephp I can do this > > to _any_ pci device, even if it is on the motherboard.) > > - userspace writes to char dev node > > - driver attempts to access pci device structure that is no > > longer present in memory. > > > >Because of this open needs to get a reference to the pci device to > >prevent oopses, or the driver needs to be aware of "device is now gone" > >in some other manner. > > Thanks for explaining; agreed. > > However, there appear to still be massive bugs in this area: > > Consider the behavior of the chrdev if a PCI device has been > unplugged. It's still actively messing with the non-existent > hardware, and never checks for dead h/w AFAICS.
I agree, the driver should be fixed to handle this properly. thanks, greg k-h - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
