On 29-06-21, 10:27, Wolfram Sang wrote:
> > While we are at it, this has been replaced by a Rust counterpart [1]
> > (as that makes it hypervisor agnostic, which is the goal of my work
> > here) and I need someone with I2C knowledge to help review it. It
> > should be okay even if you don't understand Rust a lot, just review
> > this file[2] which is where most of i2c specific stuff lies.
>
> Can't promise I can do this before my holidays, but I will try.
Thanks.
> > I am not sure why you say I2C_RDWR isn't supported. The spec and Linux
>
> This is how I interpreted Arnd's response. I said mulitple times that I
> might be missing something so I double check.
>
> > SMBUS. To clarify on an earlier point, every virtio transfer may
> > contain one or more struct i2c_msg instances, all processed together
> > (as expected).
>
> That was the information missing for me so far becasue...
>
> > If you see virtio_i2c_send_reqs() in this patch, you will see that it
> > converts a stream of i2c_req messages to their virtio counterparts and
> > send them together, consider it a single transaction.
>
> ... when I checked virtio_i2c_send_reqs(), I also saw
> virtqueue_add_sgs() but I had no idea if this will end up as REP_START
> on the physical bus or not. But it definately should.
Just think of virtqueue_add_sgs() as something that setups the
structures for transfer. The actual stuff at the other end (host)
happens only after virtqueue_kick() is called at the guest (this
notifies the host that data is present now), in response the backend
running at host will re-create the struct i2c_msg and issue:
struct i2c_rdwr_ioctl_data data;
data.nmsgs = count;
data.msgs = msgs;
return ioctl(adapter->fd, I2C_RDWR, &data);
So we will end up recreating the exact situation as when
virtio_i2c_xfer() is called.
--
viresh
