Re: [PATCH v2 2/2] staging: r8188eu: Use kzalloc() with GFP_ATOMIC in atomic context

[Fabio M. De Francesco]

On marted? 15 febbraio 2022 17:10:12 CET Greg Kroah-Hartman wrote:
> On Tue, Feb 08, 2022 at 07:04:26PM +0100, Fabio M. De Francesco wrote:
> > Use the GFP_ATOMIC flag of kzalloc() with two memory allocation in rtw_set_key() 
> > because it is not allowed to sleep while it executes in atomic context.
> > 
> > With the GFP_ATOMIC type flag, the allocation is high priority and thus it 
> > cannot sleep.
> > 
> > This issue is detected by Smatch which emits the following warning:
> > 
> > "drivers/staging/r8188eu/core/rtw_mlme.c:1603 rtw_set_key() warn: sleeping in atomic context".
> > 
> > Fixes: 79f712ea994d ("staging: r8188eu: Remove wrappers for kalloc() and kzalloc()")
> > Fixes: 15865124feed ("staging: r8188eu: introduce new core dir for RTL8188eu driver")
> > Signed-off-by: Fabio M. De Francesco <fmdefrancesco@xxxxxxxxx>
> > ---
> >  drivers/staging/r8188eu/core/rtw_mlme.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/staging/r8188eu/core/rtw_mlme.c b/drivers/staging/r8188eu/core/rtw_mlme.c
> > index f5b2df72e0f4..860835e29b79 100644
> > --- a/drivers/staging/r8188eu/core/rtw_mlme.c
> > +++ b/drivers/staging/r8188eu/core/rtw_mlme.c
> > @@ -1600,12 +1600,12 @@ int rtw_set_key(struct adapter *adapter, struct security_priv *psecuritypriv, in
> >  	struct mlme_priv		*pmlmepriv = &adapter->mlmepriv;
> >  	int	res = _SUCCESS;
> >  
> > -	pcmd = kzalloc(sizeof(*pcmd), GFP_KERNEL);
> > +	pcmd = kzalloc(sizeof(*pcmd), GFP_ATOMIC);
> >  	if (!pcmd) {
> >  		res = _FAIL;  /* try again */
> >  		goto exit;
> >  	}
> > -	psetkeyparm = kzalloc(sizeof(*psetkeyparm), GFP_KERNEL);
> > +	psetkeyparm = kzalloc(sizeof(*psetkeyparm), GFP_ATOMIC);
> >  	if (!psetkeyparm) {
> >  		kfree(pcmd);
> >  		res = _FAIL;
> > -- 
> > 2.34.1
> 
> What code path is calling this in atomic context?
> 
> thanks,
> 
> greg k-h
> 
I blindly trusted Smatch and didn't check properly :(

Now I'm pretty sure that all(?) the code paths that lead to those kzalloc() do 
not call rtw_set_key() in atomic context.

The only calls chain that seemed to call rtw_set_key() in atomic context was 
rtw_set_key() <- ips_leave() <- _rtw_pwr_wakeup() <- rtw_set_802_11_disassociate() 
but Smatch (and I) missed that, immediately before calling ips_leave(), an 'if'
condition makes _rtw_pwr_wakeup() exit and return to rtw_set_802_11_disassociate().

Therefore, please drop this patch.

Thanks,

Fabio M. De Francesco