Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 1 Aug 2023 17:34:43 -0400
From:   Peter Xu <peterx@redhat.com>
To:     Suren Baghdasaryan <surenb@google.com>
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        akpm@linux-foundation.org, jannh@google.com, willy@infradead.org,
        liam.howlett@oracle.com, david@redhat.com, ldufour@linux.ibm.com,
        vbabka@suse.cz, michel@lespinasse.org, jglisse@google.com,
        mhocko@suse.com, hannes@cmpxchg.org, dave@stgolabs.net,
        hughd@google.com, linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        stable@vger.kernel.org
Subject: Re: [PATCH 1/6] mm: enable page walking API to lock vmas during the
 walk
Message-ID: <ZMl6c+bVxdWW0YnN@x1n>
References: <20230731171233.1098105-1-surenb@google.com>
 <20230731171233.1098105-2-surenb@google.com>
 <CAHk-=wjEbJS3OhUu+2sV8Kft8GnGcsNFOhYhXYQuk5nvvqR-NQ@mail.gmail.com>
 <CAJuCfpFWOknMsBmk1RwsX9_0-eZBoF+cy=P-E7xAmOWyeo4rvA@mail.gmail.com>
 <CAHk-=wiFXOJ_6mnuP5h3ZKNM1+SBNZFZz9p8hyS8NaYUGLioEg@mail.gmail.com>
 <CAJuCfpG4Yk65b=0TLfGRqrO7VpY3ZaYKqbBjEP+45ViC9zySVQ@mail.gmail.com>
 <CAJuCfpF6WcJBSix0PD0cOD_MaeLpfGz1ddS6Ug_M+g0QTfkdzw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CAJuCfpF6WcJBSix0PD0cOD_MaeLpfGz1ddS6Ug_M+g0QTfkdzw@mail.gmail.com>
Precedence: bulk

On Tue, Aug 01, 2023 at 01:28:56PM -0700, Suren Baghdasaryan wrote:
> I have the new patchset ready but I see 3 places where we walk the
> pages after mmap_write_lock() while *I think* we can tolerate
> concurrent page faults (don't need to lock the vmas):
> 
> s390_enable_sie()
> break_ksm()
> clear_refs_write()

This one doesn't look right to be listed - tlb flushing is postponed after
pgtable lock released, so I assume the same issue can happen like fork():
where we can have race coditions to corrupt data if, e.g., thread A
writting with a writable (unflushed) tlb, alongside with thread B CoWing.

It'll indeed be nice to know whether break_ksm() can avoid that lock_vma
parameter across quite a few function jumps. I don't yet see an immediate
issue with this one..  No idea on s390_enable_sie(), but to make it simple
and safe I'd simply leave it with the write vma lock to match the mmap
write lock.

Thanks,

-- 
Peter Xu