Skip to contentSkip to conversation

MAP_POPULATE behavior change?

DISCUSSION Michael Kerrisk 3 participants 5 messages Apr 28, 2008 – May 5, 2008

lore
Share & tools
lore
#0 lore raw

Hey Nick,

Ping!

Cheers,

Michael

-------- Original Message --------
Subject: MAP_POPULATE behavior change?
Date: Tue, 22 Apr 2008 15:16:19 +0200
From: Michael Kerrisk <[email protected]>
To: Nick Piggin <[email protected]>
CC: lkml <[email protected]>, mk <[email protected]>

Hi Nick,

In 2.6.23, you had the commit below, which seems to have changed the kernel-userland API.

Am I right to understand that with this patch MAP_POPULATE now also works with MAP_PRIVATE mappings? (Formerly it only worked with MAP_SHARED mappings, but some very quick testing suggests that MAP_POPULATE now also has an affect for MAP_PRIVATE.)

Also, can you summarize the status of MAP_NONBLOCK? Is it now a no-op?

Cheers,

Michael

==

commit 54cb8821de07f2ffcd28c380ce9b93d5784b40d7

Author: Nick Piggin <[email protected]>
Date:   Thu Jul 19 01:46:59 2007 -0700

mm: merge populate and nopage into fault (fixes nonlinear)

Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings.

->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff).
But it is more logical to pass pgoff rather than have the ->nopage function
calculate it itself anyway (because that's a similar layering violation).

Having the populate handler install the pte itself is likewise a nasty thing to be doing.

This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over.

The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two.

After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway.

NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet.

Signature · 3 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
#1 lore raw
Quoted 18 lines · Michael Kerrisk

Hey Nick,

Ping!

Cheers,

Michael

-------- Original Message --------
Subject: MAP_POPULATE behavior change?
Date: Tue, 22 Apr 2008 15:16:19 +0200
From: Michael Kerrisk <[email protected]>
To: Nick Piggin <[email protected]>
CC: lkml <[email protected]>, mk <[email protected]>

Hi Nick,

Quoted 7 lines

In 2.6.23, you had the commit below, which seems to have changed the kernel-userland API.

Am I right to understand that with this patch MAP_POPULATE now also works with MAP_PRIVATE mappings? (Formerly it only worked with MAP_SHARED mappings, but some very quick testing suggests that MAP_POPULATE now also has an affect for MAP_PRIVATE.)

Yeah it should also set up anonymous mappings. Very observant :)

Quoted 1 line

Also, can you summarize the status of MAP_NONBLOCK? Is it now a no-op?

Yes, it is. The problem is you need to walk the page tables, and it is fairly annoying to have to make a special case for that to be non blocking (after just getting rid of the ->populate special case).

It is possible of course to implement. It might even be easier because you could pass in a NONBLOCK flag to ->fault from get_user_pages... but for now it is a noop.

Thanks,
Nick

Quoted 48 lines

Cheers,

Michael

==

commit 54cb8821de07f2ffcd28c380ce9b93d5784b40d7

Author: Nick Piggin <[email protected]>
Date:   Thu Jul 19 01:46:59 2007 -0700

mm: merge populate and nopage into fault (fixes nonlinear)

Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings.

->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff).
But it is more logical to pass pgoff rather than have the ->nopage function
calculate it itself anyway (because that's a similar layering violation).

Having the populate handler install the pte itself is likewise a nasty thing to be doing.

This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over.

The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two.

After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway.

NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet.

Signature · 3 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
#2 lore raw

Hi Nick,

Quoted 11 lines · Nick Piggin
Quoted 7 lines · Michael Kerrisk

In 2.6.23, you had the commit below, which seems to have changed the kernel-userland API.

Am I right to understand that with this patch MAP_POPULATE now also works with MAP_PRIVATE mappings? (Formerly it only worked with MAP_SHARED mappings, but some very quick testing suggests that MAP_POPULATE now also has an affect for MAP_PRIVATE.)

Yeah it should also set up anonymous mappings. Very observant :)

Umm -- did it not work for anonymous *shared* mappings before? I understood that it did.

Quoted 3 lines
Quoted 1 line

Also, can you summarize the status of MAP_NONBLOCK? Is it now a no-op?

Yes, it is.

Hmm -- going back and re-reading the code, there is:

        if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
                make_pages_present(addr, addr + len);

That means that MAP_NONBLOCK makes *MAP_POPULATE* a no-op. Was that intended?

Cheers,

Michael

Quoted 61 lines

The problem is you need to walk the page tables, and it
is fairly annoying to have to make a special case for that to be non blocking (after just getting rid of the ->populate special case).

It is possible of course to implement. It might even be easier because you could pass in a NONBLOCK flag to ->fault from get_user_pages... but for now it is a noop.

Thanks,
Nick

Quoted 48 lines

Cheers,

Michael

==

commit 54cb8821de07f2ffcd28c380ce9b93d5784b40d7

Author: Nick Piggin <[email protected]>
Date:   Thu Jul 19 01:46:59 2007 -0700

mm: merge populate and nopage into fault (fixes nonlinear)

Nonlinear mappings are (AFAIKS) simply a virtual memory concept that encodes the virtual address -> file offset differently from linear mappings.

->populate is a layering violation because the filesystem/pagecache code should need to know anything about the virtual memory mapping. The hitch here is that the ->nopage handler didn't pass down enough information (ie. pgoff).
But it is more logical to pass pgoff rather than have the ->nopage function
calculate it itself anyway (because that's a similar layering violation).

Having the populate handler install the pte itself is likewise a nasty thing to be doing.

This patch introduces a new fault handler that replaces ->nopage and ->populate and (later) ->nopfn. Most of the old mechanism is still in place so there is a lot of duplication and nice cleanups that can be removed if everyone switches over.

The rationale for doing this in the first place is that nonlinear mappings are subject to the pagefault vs invalidate/truncate race too, and it seemed stupid to duplicate the synchronisation logic rather than just consolidate the two.

After this patch, MAP_NONBLOCK no longer sets up ptes for pages present in pagecache. Seems like a fringe functionality anyway.

NOPAGE_REFAULT is removed. This should be implemented with ->fault, and no users have hit mainline yet.

--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html

Signature · 3 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
#3 lore raw
Quoted 13 lines · Michael Kerrisk

Hi Nick,

Quoted 10 lines · Nick Piggin
Quoted 7 lines · Michael Kerrisk

In 2.6.23, you had the commit below, which seems to have changed the kernel-userland API.

Am I right to understand that with this patch MAP_POPULATE now also works with MAP_PRIVATE mappings? (Formerly it only worked with MAP_SHARED mappings, but some very quick testing suggests that MAP_POPULATE now also has an affect for MAP_PRIVATE.)

Quoted 4 lines
Quoted 1 line

Yeah it should also set up anonymous mappings. Very observant :)

Umm -- did it not work for anonymous *shared* mappings before? I understood that it did.

Oh... yeah it would have. anonymous shared mappings aren't really anonymous from the POV of most of mm/ :P

Quoted 10 lines
Quoted 3 lines
Quoted 1 line

Also, can you summarize the status of MAP_NONBLOCK? Is it now a no-op?

Yes, it is.

Hmm -- going back and re-reading the code, there is:

        if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
                make_pages_present(addr, addr + len);

That means that MAP_NONBLOCK makes *MAP_POPULATE* a no-op. Was that intended?

Ah, my logic wasn't up to scratch there. Yes of course MAP_NONBLOCK is not a noop, but it makes MAP_POPULATE into a noop. So it still guarantees that the call will not block and that the kernel will make some kind of best effort to populate the mapping... for some values of "some kind of" (ie. "nothing" ;)).

It's done this way of course so it is still safe to call MAP_NONBLOCK if you don't want to block, and thus everything stays backwards and forward compatible.

Thanks,
Nick

#4 lore raw

[...]

Quoted 20 lines
Quoted 11 lines
Quoted 4 lines
Quoted 2 lines

Also, can you summarize the status of MAP_NONBLOCK? Is it now a no-op?

Yes, it is.

Hmm -- going back and re-reading the code, there is:

        if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
                make_pages_present(addr, addr + len);

That means that MAP_NONBLOCK makes *MAP_POPULATE* a no-op.

Was that intended?

Ah, my logic wasn't up to scratch there. Yes of course MAP_NONBLOCK is not
a noop, but it makes MAP_POPULATE into a noop. So it still guarantees that the call will not block and that the kernel will make some kind of best effort to populate the mapping... for some values of "some kind of" (ie. "nothing" ;)).

Quoted 3 lines

It's done this way of course so it is still safe to call MAP_NONBLOCK if you don't want to block, and thus everything stays backwards and forward compatible.

Thanks.

Continuing on from there, what is the status of the flags argument for remap_file_pages() now. the man page curently carries this text:

The flags argument has the same meaning as for mmap(2), but all flags other than MAP_NONBLOCK are ignored.

My question is: is it still true that flags other than MAP_NONBLOCK are ignored? A quick glance at the code leads me to suspect not. (I'm not sure if you made the recent changes to that code or not?)

Cheers,

Michael

Signature · 3 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html