On Thu, Aug 10, 2017 at 02:25:57PM +0100, Phil Elwell wrote:
> 
> 
> On 10/08/2017 12:24, Dan Carpenter wrote:
> > On Thu, Aug 10, 2017 at 11:52:42AM +0100, Phil Elwell wrote:
> >> On 10/08/2017 11:21, Dan Carpenter wrote:
> >>> The original patch did not go through the normal review process...
> >>>
> >>> On Tue, Aug 08, 2017 at 01:05:02PM +0100, Phil Elwell wrote:
> >>>> I'm all for fixing memory leaks, but freeing a block while it is still
> >>>> being used is a recipe for hard-to-debug kernel exceptions.
> >>>>
> >>>
> >>> This bug completely breaks the driver doesn't it?  It's not very subtle
> >>> so it should be easy to diagnose with git bisect.
> >>
> >> It's more subtle than that - the failure isn't consistent, sometimes 
> >> crashing
> >> and sometimes not depending on how and when memory is reused.
> >>
> >>>> 1) There is already a vchi method for freeing the instance, so use it.
> >>>> 2) Only call it on error, and then only before initted is false.
> >>>>
> >>>> Signed-off-by: Phil Elwell <p...@raspberrypi.org>
> >>>> Fixes: 0adbfd4694c2 ("staging: bcm2835-audio: fix memory leak in 
> >>>> bcm2835_audio_open_connection()")
> >>>> ---
> >>>>  drivers/staging/vc04_services/bcm2835-audio/bcm2835-vchiq.c | 2 +-
> >>>>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/drivers/staging/vc04_services/bcm2835-audio/bcm2835-vchiq.c 
> >>>> b/drivers/staging/vc04_services/bcm2835-audio/bcm2835-vchiq.c
> >>>> index 5f3d8f2..89f96f3 100644
> >>>> --- a/drivers/staging/vc04_services/bcm2835-audio/bcm2835-vchiq.c
> >>>> +++ b/drivers/staging/vc04_services/bcm2835-audio/bcm2835-vchiq.c
> >>>> @@ -409,6 +409,7 @@ static int bcm2835_audio_open_connection(struct 
> >>>> bcm2835_alsa_stream *alsa_stream
> >>>>                          LOG_ERR("%s: failed to connect VCHI instance 
> >>>> (ret=%d)\n",
> >>>>                                  __func__, ret);
> >>>>  
> >>>> +                        vchi_disconnect(vchi_instance);
> >>>
> >>> This is ugly because why are we calling disconnect() if connect() fails?
> >>> These functions should be symetric so disconnect only disconnects and
> >>> we call a different function to undo vchi_initialise().
> >>
> >> Agreed - I'm not going to change the API, but I can add a comment.
> >>
> > 
> > Nah...  Please don't do that.  Just create a function:
> > 
> > void vchiq_free(VCHIQ_INSTANCE_T *vchi_instance)
> > {
> >     kfree(vchi_instance);
> > }
> > 
> > Really vchi_initialise() is badly named and it's just a wrapper around
> > vchiq_initialise().  It should be deleted and vchiq_initialise() should
> > be renamed to allocate instead of initialize...  But that's for a
> > separate patch.
> > 
> > And also we should move the kfree() out of disconnect() like I said
> > and instead call vchiq_free().  But again, that's for a separate patch.
> > 
> > Change the goto to "return -EIO."  Leave the last error path as-is.
> > Eventually we will want to break this into two functions, one that
> > allocates the first vchi_instance and one that calls vc_vchi_audio_init().
> > But again, there are many patches needed to fix this code.
> 
> [ static removed ]
> 
> Shouldn't that be:
> 
> void vchi_uninitialise(VCHI_INSTANCE_T instance_handle)

I haven't understood the difference between vchi_ and vchiq_.  It looks
like vhci_ is just a shim around vhciq_.  We should get rid of the shim
layer eventually...

Naming it vhci_ would make it consistent with the rest of the shim layer
but we don't care about consistency much we just want to write the code
the way it will end up eventually.  In other words, there is no point
creating a shim function just to be consistent.

initialize/uninitialize are bad names because really it's an alloc/free.

Anyway, I don't have strong feelings about the name.

> {
>       vchiq_shutdown((VCHIQ_INSTANCE_T)instance_handle);
> }
> EXPORT_SYMBOL(vchi_uninitialise);
> 
> Yes, in this case the instance is just a block of memory with no users,
> but if this is a general API call then you want it to work in all cases,
> and calling vchiq_shutdown is the right way to free an instance. Calling
> a VCHIQ method when all the other calls are to the VCHI API is grubby,
> so make it a VCHI method instead.

The vchiq_shutdown() does call kfree() but it's a layering violation.
Every allocation function should have a mirror free function.  The
vchiq_shutdown() cleans up a bunch of stuff which was not allocated in
vchi_initialise() and it seems to all be a no-op so it's not buggy but
it's not the right way to design an API.

So:  Rename vchiq_shutdown() to vchiq_disconnect().  Delete the kfree().
Create a new function vchiq_shutdown() that does:

void vchiq_shutdown()
{
        vchiq_disconnect();
        vchiq_free();
}

But again, that is work for future patches.  Every alloc has a free,
every connect has a disconnect.

regards,
dan carpenter

Reply via email to