Re: [PATCH RFC 0/3] staging: r8188eu: avoid uninit value bugs

[Pavel Skripkin]

On 8/22/21 7:03 PM, Fabio M. De Francesco wrote:
> On Sunday, August 22, 2021 3:31:31 PM CEST Pavel Skripkin wrote:
> > On 8/22/21 4:21 PM, Fabio M. De Francesco wrote:
> > > On Sunday, August 22, 2021 2:39:34 PM CEST Greg KH wrote:
> > >> On Sun, Aug 22, 2021 at 03:10:56PM +0300, Pavel Skripkin wrote:
> > >> > On 8/22/21 1:59 PM, Fabio M. De Francesco wrote:
> > >> > > On Sunday, August 22, 2021 12:09:29 PM CEST Pavel Skripkin wrote:
> > > [...]
> > >> > > So, it's up to the callers to test if (!_rtw_read*()) and then act
> > >> > > accordingly. If they get 0 they should know how to handle the errors.
> > >> > 
> > >> > Yes, but _rtw_read*() == 0 indicates 2 states:
> > >> > 	1. Error on transfer side
> > >> > 	2. Actual register value is 0
> > >> 
> > >> That's not a good design, it should be fixed.  Note there is the new
> > >> usb_control_msg_recv() function which should probably be used instead
> > >> here, to prevent this problem from happening.
> > > 
> > > I think that no functions should return 0 for signaling FAILURE. If I'm not
> > > wrong, the kernel quite always prefers to return 0 on SUCCESS and <0 on
> > > FAILURE. Why don't you just fix this?
> > > 
> > That's what I've done in v2. All rtw_read* family will have following 
> > prototype in v2:
> >
> > int __must_check _rtw_read8(struct adapter *adapter, u32 addr, u8 *data);

(*)

> > Was it your idea, or you were talking about different approach?
> >
> > With regards,
> > Pavel Skripkin
>
> Pavel,
>
> Yes, it is correct.
>
> However, after that I had time to look at the calls chain and understand what
> each function does and then I saw that my initial proposal should be made
> along with another one...
>
> The calls chain is:
>
> (1)        _rtw_read8()  <--- (returns the data read from next function in chain)
>                                              (no errors returned, see possible fix in next function)
> (2)                usb_read8() <--- (returns the data read from next function in chain)
>                                                   (_data_may_be_unitialised_, no errors returned)
> 		  (possible fix: from "u8 data"; to "char data = -1;")

Anyway char will be cast to u8 and -1 will become 0xff. 0xff is still 
valid register value, I guess.

> (3)                           usbctrl_vendorreq() <---- (returns data read from next function in chain)
> 		                            (data is always a valid pointer saved to third argument)
> 		                            (if it fails, the third argument is unchanged because it
> 		                            still has the address of the "data" argument given by the caller) > (4)	                usb_control_msg() <---- (it always returns how 
many bytes read or valid error codes)
> 		                                  (it _never_ returns 0: either positive or negative values)
> 		
> I have not yet looked at the usb_control_msg_recv() which Greg talked about.
>
> To summarize: in function (2) "u8 data" should become "char data = -1;".

So, anyway caller _should_ somehow receive an error from 
usb_control_msg(). We can just change rtw_read{8,16,32} return values 
from u{8,16,32} to int32, but anyway it will require all changes, that 
I've done in this series, but in slightly different form. I.e temp int32 
variable + error checking + casting int to u{8,16,32}.

Doesn't it make sense to just switch to more standard prototype (*)? All 
other drivers use this prototype for their private reading functions.



With regards,
Pavel Skripkin