Re: [PATCH v3 1/3] mm: introduce fincore()

[Naoya Horiguchi]

On Mon, Jul 07, 2014 at 01:43:31PM -0700, Dave Hansen wrote:
> On 07/07/2014 01:21 PM, Naoya Horiguchi wrote:
> > On Mon, Jul 07, 2014 at 12:01:41PM -0700, Dave Hansen wrote:
> >> But, is this trying to do too many things at once?  Do we have solid use
> >> cases spelled out for each of these modes?  Have we thought out how they
> >> will be used in practice?
> > 
> > tools/vm/page-types.c will be an in-kernel user after this base code is
> > accepted. The idea of doing fincore() thing comes up during the discussion
> > with Konstantin over file cache mode of this tool.
> > pfn and page flag are needed there, so I think it's one clear usecase.
> 
> I'm going to take that as a no. :)

As for other usecases, database developers should have some demand for
physical addresses (especially numa node?) or page flags (especially
page reclaim or writeback related ones).
But I'm not a database expert so can't say how, sorry.

> The whole FINCORE_PGOFF vs. FINCORE_BMAP issue is something that will
> come up in practice.  We just don't have the interfaces for an end user
> to pick which one they want to use.
> 
> >> Is it really right to say this is going to be 8 bytes?  Would we want it
> >> to share types with something else, like be an loff_t?
> > 
> > Could you elaborate it more?
> 
> We specify file offsets in other system calls, like the lseek family.  I
> was just thinking that this type should match up with those calls since
> they are expressing the same data type with the same ranges and limitations.

The 2nd parameter is loff_t, do we already do this?

> >>> + * - FINCORE_PFN:
> >>> + *     stores pfn, using 8 bytes.
> >>
> >> These are all an unprivileged operations from what I can tell.  I know
> >> we're going to a lot of trouble to hide kernel addresses from being seen
> >> in userspace.  This seems like it would be undesirable for the folks
> >> that care about not leaking kernel addresses, especially for
> >> unprivileged users.
> >>
> >> This would essentially tell userspace where in the kernel's address
> >> space some user-controlled data will be.
> > 
> > OK, so this and FINCORE_PAGEFLAGS will be limited for privileged users.

Sorry, this statement of mine might a bit short-sighted, and I'd like
to revoke it.
I think that some page flags and/or numa info should be useful outside
the debugging environment, and safe to expose to userspace. So limiting
to bitmap-one for unprivileged users is too strict.

> Then I'd just question their usefulness outside of a debugging
> environment, especially when you can get at them in other (more
> roundabout) ways in a debugging environment.
> 
> This is really looking to me like two system calls.  The bitmap-based
> one, and another more extensible one.  I don't think there's any harm in
> having two system calls, especially when they're trying to glue together
> two disparate interfaces.

I think that if separating syscall into two, one for privileged users
and one for unprivileged users migth be fine (rather than bitmap-based
one and extensible one.)

Thanks,
Naoya Horiguchi
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/