Re: [Xen-devel] xen/mem-reservation API and out-of-tree kernel modules

[Oleksandr Andrushchenko]

On 2/1/19 11:14 AM, Juergen Gross wrote:
> On 01/02/2019 09:39, Oleksandr Andrushchenko wrote:
>> On 1/31/19 11:44 PM, Stefano Stabellini wrote:
>>> On Thu, 31 Jan 2019, Oleksandr Andrushchenko wrote:
>>>> Hello,
>>>>
>>>> I am working on porting an out-of-tree kernel driver to the kernel
>>>> 5.0 and that driver uses functionality provided by
>>>> drivers/xen/mem-reservation.c
>>>> module. Since commit [1] it is not possible to build a kernel module
>>>> which uses mem-reservation API as xen_scrub_pages variable, which is
>>>> checked in
>>>> xenmem_reservation_scrub_page, became a kernel module parameter and is
>>>> now only
>>>> accessible for built-in modules:
>>>>
>>>> static inline void xenmem_reservation_scrub_page(struct page *page)
>>>> ^^^^^^^^^^^^^
>>>> {
>>>>    ÂÂÂ if (xen_scrub_pages)
>>>>    ÂÂÂÂÂÂÂ ^^^^^^^^^^^^^^^
>>>>    ÂÂÂ ÂÂÂ clear_highpage(page);
>>>> }
>>>>
>>>> This results in link-time warning:
>>>>
>>>>    ÂÂÂ WARNING: "xen_scrub_pages" [yourmodule.ko] undefined!
>>>>
>>>> and thus not allowing the module to run. At the moment I can only see a
>>>> possible fix
>>>> for this by making the following change:
>>>>
>>>> diff --git a/drivers/xen/mem-reservation.c b/drivers/xen/mem-reservation.c
>>>> index 3782cf070338..85fecfec50e1 100644
>>>> --- a/drivers/xen/mem-reservation.c
>>>> +++ b/drivers/xen/mem-reservation.c
>>>> @@ -18,6 +18,7 @@
>>>>
>>>>    Âbool __read_mostly xen_scrub_pages =
>>>> IS_ENABLED(CONFIG_XEN_SCRUB_PAGES_DEFAULT);
>>>>    Âcore_param(xen_scrub_pages, xen_scrub_pages, bool, 0);
>>>> +EXPORT_SYMBOL(xen_scrub_pages);
>>>>
>>>> but this looks a bit unusual for the kernel?
>>>>
>>>> I am looking for community advice here and help
>>>>
>>>> Thank you,
>>>> Oleksandr
>>>>
>>>> [1]
>>>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=197ecb3802c04499d8ff4f8cb28f6efa008067db
>>> The alternative would be to turn xenmem_reservation_scrub_page into a
>>> regular function (not a static inline)?
>> Yes, it seems there is no other reasonable solution to this, but
>> a regular function. I'll send a patch for that
> What would you gain? This function would need to be exported.
Yes, this is true, the function should be exported then
> So its either the variable or the function.
I am a bit confused with this because I'll have to export
module parameter in this case, e.g.

core_param(xen_scrub_pages, xen_scrub_pages, bool, 0);
EXPORT_SYMBOL(xen_scrub_pages);

which looks a bit unusual to me
>
> Juergen