Re: [PATCH v2] firewire: Remove function callback casts

Re: [PATCH v2] firewire: Remove function callback casts

[greg]

On Fri, May 22, 2020 at 07:43:08PM +0200, Oscar Carter wrote:
> Hi,
> 
> On Wed, May 20, 2020 at 03:16:24PM +0900, Takashi Sakamoto wrote:
> > Hi,
> >
> > I'm an author of ALSA firewire stack and thanks for the patch. I agree with
> > your intention to remove the cast of function callback toward CFI build.
> >
> > Practically, the isochronous context with FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL
> > is never used by in-kernel drivers. Here, I propose to leave current
> > kernel API (fw_iso_context_create() with fw_iso_callback_t) as is.

If it's not used by anyone, why is it still there?  Can't we just delete
it?

thanks,

greg k-h

Re: [PATCH v2] firewire: Remove function callback casts

[Greg KH]

On Fri, May 22, 2020 at 07:43:08PM +0200, Oscar Carter wrote:
> Hi,
> 
> On Wed, May 20, 2020 at 03:16:24PM +0900, Takashi Sakamoto wrote:
> > Hi,
> >
> > I'm an author of ALSA firewire stack and thanks for the patch. I agree with
> > your intention to remove the cast of function callback toward CFI build.
> >
> > Practically, the isochronous context with FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL
> > is never used by in-kernel drivers. Here, I propose to leave current
> > kernel API (fw_iso_context_create() with fw_iso_callback_t) as is.

If it's not used by anyone, why is it still there?  Can't we just delete
it?

thanks,

greg k-h

Re: [PATCH v2] firewire: Remove function callback casts

[Takashi Sakamoto]

Hi Greg,

On Sat, May 23, 2020 at 08:10:33AM +0200, Greg KH wrote:
> On Fri, May 22, 2020 at 07:43:08PM +0200, Oscar Carter wrote:
> > Hi,
> > 
> > On Wed, May 20, 2020 at 03:16:24PM +0900, Takashi Sakamoto wrote:
> > > Hi,
> > >
> > > I'm an author of ALSA firewire stack and thanks for the patch. I agree with
> > > your intention to remove the cast of function callback toward CFI build.
> > >
> > > Practically, the isochronous context with FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL
> > > is never used by in-kernel drivers. Here, I propose to leave current
> > > kernel API (fw_iso_context_create() with fw_iso_callback_t) as is.
> 
> If it's not used by anyone, why is it still there?  Can't we just delete
> it?

For this patchset, I followed to the theory to keep backward compatibility
when adding any change, and it's what I'd like to discuss.

The isoc context of multichannel mode is also available for userspace
applications, and libhinoko[1] uses it. In a point of backward
compatibility for userspace, we can't delete the mode.

(Practically, the mode is useful for the purpose of packet sniffing in
bus and helpful to my work for development of ALSA firewire stack[2].)

On the other hand, there's no unit driver to use the mode in upstream
kernel. It's unlikely to use the mode for unit driver since the mode is
not specific to unit functionality. In my opinion, it's reasonable to
lose backward compatibility slightly by hiding the multichannel mode
from in-kernel unit driver.

I'll post v2 patchset later and I'd like you to merge them if no
objections from the others for the loss of backward compatibility.

[1] https://github.com/takaswie/libhinoko
[2] https://mailman.alsa-project.org/pipermail/alsa-devel/2019-April/147862.html


Regards

Takashi Sakamoto