Re: [RFC PATCH 0/8] mm/madvise: support process_madvise(MADV_DONTNEED)

[Nadav Amit]

> On Sep 28, 2021, at 1:53 AM, David Hildenbrand <david@xxxxxxxxxx> wrote:
> 
>>> 
>>> Again, thanks for the details. I guess this should basically work, although it involves a lot of complexity (read: all flavors of uffd on other processes). And I am no so sure about performance aspects. "Performance is not as bad as you think" doesn't sound like the words you would want to hear from a car dealer ;) So there has to be another big benefit to do such user space swapping.
>> There is some complexity, indeed. Worse, there are some quirks of UFFD
>> that make life hard for no reason and some uffd and iouring bugs.
>> As for my sales pitch - I agree that I am not the best car dealer… :(
> 
> :)
> 
>> When I say performance is not bad, I mean that the core operations of
>> page-fault handling, prefetch and reclaim do not induce high overhead
>> *after* the improvements I sent or mentioned.
>> The benefit of doing so from userspace is that you have full control
>> over the reclaim/prefetch policies, so you may be able to make better
>> decisions.
>> Some workloads have predictable access patterns (see for instance "MAGE:
>> Nearly Zero-Cost Virtual Memory for Secure Computation”, OSDI’21). You may
>> be handle such access patterns without requiring intrusive changes to the
>> workload.
> 
> Thanks for the pointer.
> 
> And my question would be if something like DAMON would actually be what you want.

I looked into DAMON and even with the proposed future extensions it sounds
as a different approach with certain benefits but with many limitations.

The major limitation of DAMON is that you need to predefine the logic you
want for reclamation into the kernel. You can add programability through
some API or even eBPF, but it would never be as easy or as versatile as
what user manager can achieve. We already have pretty much all the
facilities to do so from userspace, and the missing parts (at least for
basic userspace manager) are almost already there. In contrast, see how
many iterations are needed for the basic DAMON implementation.

The second, also big, difference is that DAMON looks only on reclamation.
If you want a custom prefetch scheme or different I/O stack for backing
storage, you cannot have such one.

> 
>>> 
>>>> I am aware that there are some caveats, as zapping the memory does not
>>>> guarantee that the memory would be freed since it might be pinned for a
>>>> variety of reasons. That's the reason I mentioned the processes have "some
>>>> level of cooperation" with the manager. It is not intended to deal with
>>>> adversaries or uncommon corner cases (e.g., processes that use UFFD for
>>>> their own reasons).
>>> 
>>> It's not only long-term pinnings. Pages could have been de-duplicated (COW after fork, KSM, shared zeropage). Further, you'll most probably lose any kind of "aging" ("accessed") information on pages, or how would you track that?
>> I know it’s not just long-term pinnings. That’s what “variety of reasons”
>> stood for. ;-)
>> Aging is a tool for certain types of reclamation policies. Some do not
>> require it (e.g., random). You can also have compiler/application-guided
>> reclamation policies. If you are really into “aging”, you may be able
>> to use PEBS or other CPU facilities to track it.
>> Anyhow, the access-bit by itself not such a great solution to track
>> aging. Setting it can induce overheads of >500 cycles from my (and
>> others) experience.
> 
> Well, I'm certainly no expert on that; I would assume it's relevant in corner cases only: if you're application accesses all it's memory permanently a swap setup is already "broken". If you have plenty of old memory (VMs, databases, ...) it should work reasonably well. But yeah, detecting the working set size is a problematic problem, and "access"
> bits can be sub-optimal.
> 
> After all, that's what the Linux kernel has been relying on for a long time ... and IIRC it might be extended by multiple "aging" queues soon.

I see your point. I do not want to waste time in our discussion by
focusing on recency-based reclamation schemes and access-bits. I was
just trying to point that such schemes and access-bit have their own
drawbacks. 

> 
>>> 
>>> Although I can see that this might work, I do wonder if it's a use case worth supporting. As Michal correctly raised, we already have other infrastructure in place to trigger swapin/swapout. I recall that also damon wants to let you write advanced policies for that by monitoring actual access characteristics.
>> Hints, as those that Michal mentioned, prevent the efficient use of
>> userfaultfd. Using MADV_PAGEOUT will not trigger another uffd event
>> when the page is brought back from swap. So using
>> MADV_PAGEOUT/MADV_WILLNEED does not allow you to have a custom
>> prefetch policy, for instance. It would also require you to live
>> with the kernel reclamation/IO stack for better and worse.
> 
> Would more uffd (or similar) events help?
> 
>> As for DAMON, I am not very familiar with it, but from what I remember
>> it seemed to look on a similar direction. IMHO it is more intrusive
>> and less configurable (although it can have the advantage of better
>> integration with various kernel mechanism). I was wondering for a
>> second why you give me such a hard time for a pretty straight-forward
>> extension for process_madvise(), but then I remembered that DAMON got
>> into the kernel after >30 versions, so I’ll shut up about that. ;-)
> 
> It took ... quite a long time, indeed :)
> 
>>> 
>>>> Putting aside my use-case (which I am sure people would be glad to criticize),
>>>> I can imagine debuggers or emulators may also find use for similar schemes
>>>> (although I do not have concrete use-cases for them).
>>> 
>>> I'd be curious about use cases for debuggers/emulators. Especially for emulators I'd guess it makes more sense to just do it within the process. And for debuggers, I'm having a hard time why it would make sense to throw away a page instead of just overwriting it with $PATTERN (e.g., 0). But I'm sure people can be creative :)
>> I have some more vague ideas, but I am afraid that you will keep
>> saying that it makes more sense to handle such events from within
>> a process. I am not sure that this is true. Even for the emulators
>> that we discuss, the emulated program might run in a different
>> address space (for sandboxing). You may be able to avoid the need
>> for remote-UFFD and get away with the current non-cooperative
>> UFFD, but zapping the memory (for atomic updates) would still
>> require process_madvise(MADV_DONTNEED) [putting aside various
>> ptrace solutions].
>> Anyhow, David, I really appreciate your feedback. And you make
>> strong points about issues I encounter. Yet, eventually, I think
>> that the main question in this discussion is whether enabling
>> process_madvise(MADV_DONTNEED) is any different - from security
>> point of view - than process_vm_writev(), not to mention ptrace.
>> If not, then the same security guards should suffice, I would
>> argue.
> 
> You raise a very excellent point (and it should have been part of your initial sales pitch): how does it differ to process_vm_writev().
> 
> I can say that it differs in a way that you can break applications in more extreme ways. Let me give you two examples:
> 
> 1. longterm pinnings: you raised this yourself; this can break an application silently and there is barely a safe way your tooling could handle it.
> 
> 2. pagemap: applications can depend on the populated(present |swap) information in the pagemap for correctness. For example, there was recently a discussion to use pagemap information to speed up live migration of VMs, by skipping migration of !populated pages. There is currently no way your tooling can fake that. In comparison, ordinary swapping in the kernel can handle it.

I understand (1). As for (2): the scenario that you mention sound
very specific, and one can argue that ignoring UFFD-registered
regions in such a case is either (1) wrong or (2) should trigger
some UFFD event.

> 
> Is it easy to break an application with process_vm_writev()? Yes. When talking about dynamic debugging, it's expected that you break the target already -- or the target is already broken. Is it easier to break an application with process_madvise(MADV_DONTNEED)? I'd say yes, especially when implementing something way beyond debugging as you describe.

If you do not know what you are doing, you can easily break anything.
Note that there are other APIs that can break your application even
worse, specifically ptrace().

> I'm giving you "a hard time" for the reason Michal raised: we discussed this in the past already at least two times IIRC and "it is a free ticket to all sorts of hard to debug problem" in our opinion; especially when we mess around in other process address spaces besides for debugging.
> 
> I'm not the person to ack/nack this, I'm just asking the questions :)

I see your points and I try to look for a path of least resistance.
I thought that process_madvise() is a nice interface to hook into.

But if you are concerned it will be misused, how about adding instead
an IOCTL that will zap pages but only in UFFD-registered regions?
A separate IOCTL for this matter have an advantage of being more
tailored for UFFD, not to notify UFFD upon “remove” and to be less
likely to be misused.