Re: [PATCH v3 03/13] PCI/TSM: Authenticate devices via platform TSM

[<dan.j.williams@intel.com>]

Alexey Kardashevskiy wrote:
[..]
> >> but this is kinda wrong as it is quite bizarre to call
> >> drivers/virt/coco/tsm-core.c's tsm_ide_stream_unregister() from
> >> drivers/pci/ide.c's pci_ide_stream_release(), i.e. "virt" from "pci
> >> core". imho the caller of pci_ide_stream_release() should just call
> >> tsm_ide_stream_unregister() too, and so on. Thanks,
> > 
> > So I agree it is odd, and I orginally did not have this tie until the
> > DEFINE_FREE(pci_ide_stream_release, ...) proposal. With that I want to
> > allow for TSM drivers to teardown everything associated with IDE setup
> > with one scope-based-cleanup helper.
> > 
> > It is not a mid-layer because nothing requires a TSM driver to call
> > tsm_ide_stream_register(), but if they do register it then the PCI core
> > helper will handle cleaning it up on error along with the rest of the
> > resources.
> 
> I see but imho DEFINE_FREE() is weak justification for such
> cross-subsystem jumping,

Interesting, I do not see 'struct pci_ide' understanding that one of the
primary methods of establishing IDE (coordination through a TSM) as
cross-subsystem jumping. It is unlikely that Linux will ever understand
IDE establishment in any other form especially because, as far as I
understand, all but Intel platforms enforce IDE establishment through
TSM ABI calls.

> still. I was looking at the linker error as, like, "find a Wally".

not sure what that means...
 
>May be then drop EXPORT_SYMBOL_GPL() for tsm_ide_stream_unregister()?

True, that can be dropped now.