On Wed, 24 Dec 2003, Jean Delvare wrote: > 4* What's the point in moving the "adapter->clients[i] = NULL" loop from > top to bottom of function i2c_detach_client? This too makes much noise > for nothing. If there is no special reason for doing so, let's reverse > that change. If there is, this doesn't belong to this patch either. Policy; have the list lock while making a change to it, on removal the race is not that dangerous but still worth a fix. Note the different return paths we should handle otherwise. Postpone if you wish. > > Fix module reference counting to use try_inc_mod_count in place of > > __MOD_INC_USE_COUNT, and have i2c_inc_use_client return int. While > > this is an API change, the return code is just silently ignored. Fix > > i2c_use_client to study the mentioned return code. > > I'm not sure I get you here. Are these things already in CVS and to be > ported to Linux 2.4, or the particular points that are broken in CVS > which should be fixed before we can submit a clean patch for Linux 2.4? This is about fixing the possible race while rmmoding, as the .owner field is (sometimes) initialized. The race is not yet fixed in CVS. I will not import broken things from CVS with my patches. I'm thinking whether to pass controlling_mod in i2c_register_entry as this is client->driver->owner. I would like to make this API change, remove it, for the sake of not changing lm_sensors chip drivers. Can you find anyone else using it? -- Ky?sti M?lkki <kyosti.malkki at welho.com>
