virtio: provide a rust interface of virtio_driver in rust

[Richardhongyu]

Wrap the struct virtio_driver in rust.

Signed-off-by: Li Hongyu lihongyu1999@bupt.edu.cn

[Richardhongyu]

The new commit fixes the above errors in the comment.

[fbq]

This looks great, but one question: shouldn't we use a dedicated crate for virtio support? First VIRTIO is not always enabled, we at least need to put some "cfg gate"s somewhere, second the virtio support is not a basic part of core kernel, we should separate it for maintainship.

[wedsonaf]

Other than the of ids, this looks good! Thanks for contributing.

[wedsonaf]

If you want of ids to work here, you must set vdrv.driver.of_match_table -- see platform.rs for an example.

[Richardhongyu]

I found that in the virtio drivers of C sides, the common way to match the driver with the device is to use 'id_table' rather than of. I write a macro define_virtio_id_table to implement this. It's more like define_amba_id_table in the amba.rs.

[wedsonaf]

of identifiers are allowed to carry some data with them. If the match happened through one of them, ideally you'll pass this data to the probe function. See get_id_info in platform.rs for an example of this. (Perhaps we should move this function to the driver module so all drivers that support of can use it.)

[Richardhongyu]

The virtio drivers use id_table to match the driver with the device, which does not contain any data field. The probe function in the virtio Driver is more like the probe function in the amba Driver.
It's reasonable to move get_id_info into the driver module. I can open a new PR about it.

[wedsonaf]

This looks great, but one question: shouldn't we use a dedicated crate for virtio support? First VIRTIO is not always enabled, we at least need to put some "cfg gate"s somewhere, second the virtio support is not a basic part of core kernel, we should separate it for maintainship.

I don't think we support putting it a different crate just yet, although @ojeda has been thinking about this.

As for a cfg gate, I agree we should have it. We normally put it in lib.rs when declaring the module, this way they're excluded when the feature is not configured.

[Richardhongyu]

This looks great, but one question: shouldn't we use a dedicated crate for virtio support? First VIRTIO is not always enabled, we at least need to put some "cfg gate"s somewhere, second the virtio support is not a basic part of core kernel, we should separate it for maintainship.

Thanks for reviewing! I use '#[cfg(CONFIG_VIRTIO)]' in the new commit.

[Richardhongyu]

Other than the of ids, this looks good! Thanks for contributing.

Thanks for reviewing! It seems I confuse the id_table and of_match_table. The drivers of virtio usually use id_table to match the driver with the device just like amba drivers. I write a new macro define_virtio_id_table to declare the ID_TABLE. In this case, the ID_TABLE should be passed to vdrv.id_table and there is no data field.

[Richardhongyu]

@wedsonaf Hi, could you help have a look at this whenever you have time? Thanks in advance!

[Richardhongyu]

Now the code in the comments does not use ignore.