On 31/01/07 14:48, Duncan Sands wrote: >> usbatm only outputs basic information via the per-device /proc/net/atm/ file, >> this patch allows the device specific USB ATM drivers to replace the >> atm_proc_read function with their own. > > I'm still meditating on this. The reason I didn't do this originally is > because of potential problems with modem disconnection and/or module > unloading (the cxacru module can be unloaded at any time - it's the usbatm > module that can't be unloaded when a connection is open - so you've got to > be careful that no-one can call into cxacru after or while it's being > destroyed). I think it will be OK as long as usbatm calls unbind after > shutting down the ATM layer (since otherwise your read method could be > called after you've freed your cxacru private data) which is not the case > right now, but should be easy to arrange. Horrible things may happen > if proc_read can still be running after atm_dev_deregister has returned, > but, if so, horrible things can already happen right now. I don't understand > why this is impossible; maybe it is possible. The worst that will happen > (given that none of the proc_read methods sleeps) is that freed memory will > be accessed and the contents spat out in the proc file (if proc_read sleeps, > that could result in trying to run code inside a destroyed module). > > Ciao, > > Duncan.
Couldn't the cxacru instance pointer to the proc_read function be set to NULL before unloading? -- Simon Arlott
signature.asc
Description: OpenPGP digital signature
