Linux Virtualization - Mar 2015 - [virtio-dev] Re: [PATCH v3] Add virtio-input driver.

Hi,
> > +	spin_lock_irqsave(&vi->lock, flags);
> > +	while ((event = virtqueue_get_buf(vi->evt, &len)) != NULL) {
> > +		input_event(vi->idev,
> > +			    le16_to_cpu(event->type),
> > +			    le16_to_cpu(event->code),
> > +			    le32_to_cpu(event->value));
> 
> What happens if input layer gets an
> unexpected event code or value?
input layer checks it and ignores events not supported (according to the
support bitmaps).
> > +static int virtinput_send_status(struct virtio_input *vi,
> > +				 u16 type, u16 code, s32 value)
> > +{
> This means that caller will get errors if it happens to call
> send_status at a rate that's faster than host's consumption of
them.
> To me this looks wrong.
> Poking at input layer, it seems to simply discard errors.
> Is it always safe to discard status updates?
Typical use case is updating the leds of your keyboard.  Loosing one
update isn't the end of the world.  Also that are _very_ low rate
events, and in case that really actually happens you likely have bigger
problems anyway ;)
> > +static void virtinput_cfg_bits(struct virtio_input *vi, int select,
int subsel,
> > +			       unsigned long *bits, unsigned int bitcount)
> > +{
> > +	unsigned int bit;
> > +	size_t bytes;
> > +	u8 *virtio_bits;
> > +
> > +	bytes = virtinput_cfg_select(vi, select, subsel);
> > +	if (!bytes)
> > +		return;
> 
> How about limiting bytes to sizeof struct virtio_input_config->u?
It's limited to 256 anyway because size is u8 in config space.
> > +	size = virtinput_cfg_select(vi, VIRTIO_INPUT_CFG_ID_DEVIDS, 0);
> > +	if (size >= 8) {
> 
> What does 8 mean here? Should be sizeof virtio_input_devids?
Yes, fixed.
> > +struct virtio_input_config {
> > +	__u8    select;
> > +	__u8    subsel;
> > +	__u8    size;
> > +	__u8    reserved;
> > +	union {
> > +		char string[128];
> > +		__u8 bitmap[128];
> 
> I note that neither string nor bitmap are used by
> driver. What are they in aid of?
Fixed.  Just sloppy coding, the name should use u.string, the bitmap
u.bitmap, instead of just "u" (although for offsetof it doesn't
make a
difference).

cheers,
  Gerd

On Tue, Mar 24, 2015 at 12:46:21PM +0100, Gerd Hoffmann
wrote:
>   Hi,
> 
> > > +	spin_lock_irqsave(&vi->lock, flags);
> > > +	while ((event = virtqueue_get_buf(vi->evt, &len)) !=
NULL) {
> > > +		input_event(vi->idev,
> > > +			    le16_to_cpu(event->type),
> > > +			    le16_to_cpu(event->code),
> > > +			    le32_to_cpu(event->value));
> > 
> > What happens if input layer gets an
> > unexpected event code or value?
> 
> input layer checks it and ignores events not supported (according to the
> support bitmaps).
Right but support bitmaps come from host too, no?

> > > +static int virtinput_send_status(struct virtio_input *vi,
> > > +				 u16 type, u16 code, s32 value)
> > > +{
> 
> > This means that caller will get errors if it happens to call
> > send_status at a rate that's faster than host's consumption of
them.
> > To me this looks wrong.
> > Poking at input layer, it seems to simply discard errors.
> > Is it always safe to discard status updates?
> 
> Typical use case is updating the leds of your keyboard.  Loosing one
> update isn't the end of the world.  Also that are _very_ low rate
> events, and in case that really actually happens you likely have bigger
> problems anyway ;)
Until QE opens a bug report I guess :)
OK but pls add a comment.
> > > +static void virtinput_cfg_bits(struct virtio_input *vi, int
select, int subsel,
> > > +			       unsigned long *bits, unsigned int bitcount)
> > > +{
> > > +	unsigned int bit;
> > > +	size_t bytes;
> > > +	u8 *virtio_bits;
> > > +
> > > +	bytes = virtinput_cfg_select(vi, select, subsel);
> > > +	if (!bytes)
> > > +		return;
> > 
> > How about limiting bytes to sizeof struct virtio_input_config->u?
> 
> It's limited to 256 anyway because size is u8 in config space.
That struct is 128 bytes though.
Also, change bytes and virtinput_cfg_select to be u8 to make this
clearer?
> > > +	size = virtinput_cfg_select(vi, VIRTIO_INPUT_CFG_ID_DEVIDS, 0);
> > > +	if (size >= 8) {
> > 
> > What does 8 mean here? Should be sizeof virtio_input_devids?
> 
> Yes, fixed.
> 
> > > +struct virtio_input_config {
> > > +	__u8    select;
> > > +	__u8    subsel;
> > > +	__u8    size;
> > > +	__u8    reserved;
> > > +	union {
> > > +		char string[128];
> > > +		__u8 bitmap[128];
> > 
> > I note that neither string nor bitmap are used by
> > driver. What are they in aid of?
> 
> Fixed.  Just sloppy coding, the name should use u.string, the bitmap
> u.bitmap, instead of just "u" (although for offsetof it
doesn't make a
> difference).
> 
> cheers,
>   Gerd
>

Hi,
> > input layer checks it and ignores events not supported (according to
the
> > support bitmaps).
> 
> Right but support bitmaps come from host too, no?
Yes, but the driver will not set invalid bits (bitcount argument for the
virtinput_cfg_bits() function is the number of valid bits of the
specific bitmap).

cheers,
  Gerd