linux-kernel - Re: [PATCH v2 0/9] Free user PTE page table pages

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <41ceeec1-52c4-4e99-201c-e1e05b2afbbc@bytedance.com>
Date:   Wed, 15 Sep 2021 22:52:40 +0800
From:   Qi Zheng <zhengqi.arch@...edance.com>
To:     David Hildenbrand <david@...hat.com>, akpm@...ux-foundation.org,
        tglx@...utronix.de, hannes@...xchg.org, mhocko@...nel.org,
        vdavydov.dev@...il.com, kirill.shutemov@...ux.intel.com,
        mika.penttila@...tfour.com
Cc:     linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, songmuchun@...edance.com
Subject: Re: [PATCH v2 0/9] Free user PTE page table pages



On 9/1/21 8:32 PM, David Hildenbrand wrote:

Hi David,

>>
> 
> 
> Some high-level feedback after studying the code:
> 
> 1. Try introducing the new dummy primitives ("API") first, and then 
> convert each subsystem individually; especially, maybe convert the whole 
> pagefault handling in a single patch, because it's far from trivial. 
> This will make this series much easier to digest.

I am going to split this patch series as follows:

1. Introduce the new dummy APIs, which is an empty implementation.
    But I will explain its semantics.
2. Merge #6, #7 and #8, and call these dummy APIs in any necessary
    location, and split some special cases into single patches, such as
    pagefault and gup, etc. So that we can explain in more detail the
    concurrency in these cases. For example, we don't need to hold any
    pte_refcount in the fast path in gup on the x86_64 platform. Because
    the PTE page can't be freed after the local CPU interrupt is closed
    in the fast path in gup.
3. Introduce CONFIG_FREE_USER_PTE and implement these empty dummy APIs.
4. Add a description document.

And I try to add a function that combines pte_offset_map() and
pte_try_get(). Maybe the func name is pte_try_map() recommended by
Jason, or keep the pte_offset_map() unchanged?

> 
> Then, have a patch that adds actual logic to the dummy primitives via a 
> config option.
> 
> 2. Minimize the API.
> 
> a) pte_alloc_get{,_map,_map_lock}() is really ugly. Maybe restrict it to 
> pte_alloc_get()
I also think pte_alloc_get{,_map,_map_lock}() is ugly, but I can't
figure out a more suitable name. Maybe we can keep the
pte_alloc{,_map,_map_lock}() without any modification? But I am
worried that the caller will forget to call the paired pte_put().

> 
> b) pmd_trans_unstable_or_pte_try_get() and friends are really ugly.
> 
> Handle it independently for now, even if it implies duplicate runtime 
> checks.
> 
> if (pmd_trans_unstable() || !pte_try_get()) ...
> 
> We can always optimize later, once we can come up with something cleaner.
> 
> 3. Merge #6, and #7, after factoring out all changes to other subsystems 
> to use the API
> 
> 4. Merge #8 into #6. There is a lot of unnecessary code churn back and 
> forth, and IMHO the whole approach might not make sense without RCU due 
> to the additional locking overhead.
> 
> Or at least, try to not modify the API you introduced in patch #6 or #7 
> in #8 again. Converting all call sites back and forth just makes review 
> quite hard.
> 
> 
> I am preparing some some cleanups that will make get_locked_pte() and 
> similar a little nicer to handle. I'll send them out this or next week.
> 
Thanks,
Qi