On Wednesday, April 14, 2021 7:21:50 AM CEST Dan Carpenter wrote:
> On Tue, Apr 13, 2021 at 10:08:32PM +0200, Fabio M. De Francesco wrote:
> > On Tuesday, April 13, 2021 9:48:44 PM CEST Matthew Wilcox wrote:
> > > On Tue, Apr 13, 2021 at 09:45:03PM +0200, Fabio M. De Francesco
wrote:
> > > > 1) The driver doesn't call that function from anywhere else than
> > > > the
> > > > macro. 2) You have explained that the macro add its symbol to a
> > > > slot
> > > > in an array that would shift all the subsequent elements down if
> > > > that
> > > > macro is not used exactly in the line where it is.
> > > > 3) Dan Carpenter said that that array is full of null functions (or
> > > > empty slots?).
> > > >
> > > > Unless that function is called anonymously dereferencing its
> > > > address
> > > > from the position it occupies in the array, I'm not able to see
> > > > what
> > > > else means can any caller use.
> > > >
> > > > I know I have much less experience than you with C: what can go
> > > > wrong?
> > >
> > > Here's where the driver calls that function:
> > >
> > > $ git grep wlancmds drivers/staging/rtl8723bs/
> > > drivers/staging/rtl8723bs/core/rtw_cmd.c:static struct cmd_hdl
> > > wlancmds[] = { drivers/staging/rtl8723bs/core/rtw_cmd.c:
> > > if
> > > (pcmd->cmdcode < ARRAY_SIZE(wlancmds)) {
> > > drivers/staging/rtl8723bs/core/rtw_cmd.c:
> > > cmd_hdl
> > > = wlancmds[pcmd->cmdcode].h2cfuns;
> >
> > OK, I had imagined an anonymous call from its location in the array (as
> > I wrote in the last phrase of my message). However, I thought that it
> > could have been an improbable possibility, not a real one.
> >
> > Linux uses a lot of interesting ideas that newcomers like me should
> > learn. Things here are trickier than they appear at first sight.
>
> One trick would be to build the Smatch cross function database.
>
> https://www.spinics.net/lists/smatch/msg00568.html
>
> Then you could do:
>
> $ ~/path/to/smatch_data/db/smdb.py led_blink_hdl
> file | caller | function | type | parameter | key | value |
> drivers/staging/rtl8723bs/core/rtw_cmd.c | rtw_cmd_thread |
> rtw_cmd_thread ptr cmd_hdl | INTERNAL | -1 | |
> uchar(*)(struct adapter*, uchar*)
> drivers/staging/rtl8188eu/core/rtw_cmd.c | rtw_cmd_thread |
> rtw_cmd_thread ptr cmd_hdl | INTERNAL | -1 | |
> uchar(*)(struct adapter*, uchar*)
> drivers/staging/rtl8188eu/core/rtw_cmd.c | rtw_cmd_thread |
> rtw_cmd_thread ptr cmd_hdl | BUF_SIZE | 1 | pbuf |
> 1,4,6,8,12,14,16,19-20,23-24,48,740,884,892,900,960
>
>
> Which says that led_blink_hdl() is called as a function pointer called
> "cmd_hdl" from rtw_cmd_thread().
>
> Hm... It says it can be called from either rtw_cmd_thread() function
> (the rtl8723bs or rtl8188eu version) which is not ideal. But also
> basically harmless so whatever...
>
> regards,
> dan carpenter
>
Nice tool, thanks. I'll surely use it when it is needed to find out which
callers use a function pointer.
However I cannot see how it can help in this context. That function *does*
something, even if I cannot understand why someone needs a function to test
the initialization of a pointer. Furthermore it is actually called by
rtw_cmd_thread() (as you found out by using smatch) that expect one of the
two possible values that led_blink_hdl() returns.
That said, what trick could I use for the purpose of getting rid of that
function? At this point I'm not sure it could be made.
Regards,
Fabio
