Re: [Xen-devel] [PATCH v6 1/4] pci: add pci_iomap_wc() variants

[Bjorn Helgaas <bhelgaas@xxxxxxxxxx>]

On Tue, Jun 16, 2015 at 2:16 PM, Luis R. Rodriguez
<mcgrof@xxxxxxxxxxxxxxxx> wrote:
> On Thu, May 28, 2015 at 5:36 PM, Luis R. Rodriguez
> <mcgrof@xxxxxxxxxxxxxxxx> wrote:
>> On Wed, May 27, 2015 at 1:04 PM, Luis R. Rodriguez <mcgrof@xxxxxxxx> wrote:
>>> On Tue, May 26, 2015 at 12:40:08PM -0500, Bjorn Helgaas wrote:
>>>> On Fri, May 22, 2015 at 02:23:41AM +0200, Luis R. Rodriguez wrote:
>>>> > On Thu, May 21, 2015 at 05:33:21PM -0500, Bjorn Helgaas wrote:
>>>> > >
>>>> > > I tentatively put this (and the rest of the series) on a pci/resource
>>>> > > branch.  I'm hoping you'll propose some clarification about
>>>> > > EXPORT_SYMBOL_GPL().
>>>> >
>>>> > EXPORT_SYMBOL_GPL() also serves to ensure only GPL modules can
>>>> > only run that code. So for instance although we have "Dual BSD/GPL"
>>>> > tags for modules pure "BSD" tags do not exist for module tags and
>>>> > cannot run EXPORT_SYMBOL_GPL() code [0]. Also there is some folks
>>>> > who do believe tha at run time all kernel modules are GPL [1] [2].
>>>> > And to be precise even though the FSF may claim a list of licenses
>>>> > are GPL-compatible we cannot rely on this list alone for our own
>>>> > goals and if folks want to use our EXPORT_SYMBOL_GPL()s they must
>>>> > discuss this on lkml [2].
>>>>
>>>> By "propose some clarification," I meant that I hoped you would propose a
>>>> patch to Documentation/ that would give maintainers some guidance.
>>>
>>> I *really really* would hate to do so but only because you insist, I'll look
>>> into this...
>>
>> OK done.
>
> Bjorn,
>
> This is now on Jonathan Corbet's tree and visible on linux-next:
>
> https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=582ed8d51e2b6cb8a168c94852bca482685c2509

Sorry, I'm just not comfortable with using EXPORT_SYMBOL_GPL() this
way.  I'm happy to use it when it has a technical justification, e.g.,
for internal interfaces where users of the interface are clearly
derived works.  But pci_iomap_wc() is not in that category, and I
think it should be symmetric with similar interfaces like pci_iomap()
and ioremap_wc().

I don't want to use EXPORT_SYMBOL_GPL() for a random collection of
things depending on the whim of the author.  That makes for a messy
environment to work in, and it's messy enough already.  If we wanted
to remove the EXPORT_SYMBOL/EXPORT_SYMBOL_GPL distinction completely,
that'd be fine with me, too.  But as long as we keep it, I think it
should mean something more than the preference of the author.

I know I did already ack this, and I even said I would merge it, but a
month of thinking about this hasn't made me more comfortable with it,
so I've changed my mind.  I said before that I wouldn't try to stop
you if you want to merge it some other way, but I don't want to ack
it, and I don't want to merge it via my tree.

Bjorn

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel