Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Content-Type: text/plain;
        charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 14.0 \(3654.120.0.1.13\))
Subject: Re: [PATCH] mm: use correct VMA flags when freeing page-tables
From:   Nadav Amit <nadav.amit@gmail.com>
In-Reply-To: <20211021200643.770f9d7bd3469b2ec9d6c401@linux-foundation.org>
Date:   Fri, 22 Oct 2021 08:37:08 -0700
Cc:     Linux-MM <linux-mm@kvack.org>, linux-kernel@vger.kernel.org,
        Andrea Arcangeli <aarcange@redhat.com>,
        Andrew Cooper <andrew.cooper3@citrix.com>,
        Andy Lutomirski <luto@kernel.org>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Will Deacon <will@kernel.org>, Yu Zhao <yuzhao@google.com>,
        Nick Piggin <npiggin@gmail.com>, x86@kernel.org
Content-Transfer-Encoding: quoted-printable
Message-Id: <4FD4A068-1332-49C8-A47C-B292EDA35F9B@gmail.com>
References: <20211021122322.592822-1-namit@vmware.com>
 <20211021200643.770f9d7bd3469b2ec9d6c401@linux-foundation.org>
To:     Andrew Morton <akpm@linux-foundation.org>
Precedence: bulk


> On Oct 21, 2021, at 8:06 PM, Andrew Morton <akpm@linux-foundation.org> =
wrote:
>=20
> On Thu, 21 Oct 2021 05:23:22 -0700 Nadav Amit <nadav.amit@gmail.com> =
wrote:
>=20
>> From: Nadav Amit <namit@vmware.com>
>>=20
>> Consistent use of the mmu_gather interface requires a call to
>> tlb_start_vma() and tlb_end_vma() for each VMA. free_pgtables() does =
not
>> follow this pattern.
>>=20
>> Certain architectures need tlb_start_vma() to be called in order for
>> tlb_update_vma_flags() to update the VMA flags (tlb->vma_exec and
>> tlb->vma_huge), which are later used for the proper TLB flush to be
>> issued. Since tlb_start_vma() is not called, this can lead to the =
wrong
>> VMA flags being used when the flush is performed.
>>=20
>> Specifically, the munmap syscall would call unmap_region(), which =
unmaps
>> the VMAs and then frees the page-tables. A flush is needed after
>> the page-tables are removed to prevent page-walk caches from holding
>> stale entries, but this flush would use the flags of the VMA flags of
>> the last VMA that was flushed. This does not appear to be right.
>=20
> Any thoughts on what the worst-case end-user cisible effects of this
> would be?
>=20
> Again, I'm wondering about the desirability of a -stable backport.

This issue is not relevant for x86, which I am most familiar with, hence =
my
limited understanding of the impact on specific architectures.

In general, a TLB flush is needed after the page-tables are released
(in contrast to PTE removal) to prevent a speculative page-walk that =
might
access IO pages or install bogus entries. Such speculative page-walks =
have
been reported (on x86) as causing a machine-check. [1]

If a certain architecture has different page-walk caches for executable =
and
non-executable pages (i.e., different page-walk caches for iTLB and =
dTLB) or
for different page sizes, it might not perform the necessary TLB flush =
on the
proper TLB.

Looking at the code, we can see MIPS=E2=80=99s flow of:

tlb_flush()
->flush_tlb_range()
->local_flush_tlb_page()
->flush_micro_tlb_vm()

which calls flush_micro_tlb() only if (vma->vm_flags & VM_EXEC). So
MIPS cares about the VM_EXEC. Yet, it is not certain MIPS might be
affected.

For an architecture to be affected it needs to have all the following
properties:

1. Be able to experience hardware failure/exceptions on speculative
   page-walks.

2. Have separate page-walk caches for exec/data or huge/small pages.

3. Flush the TLBs based on VM_EXEC and VM_HUGETLB.

=46rom the code, MIPS has the 3rd property and presumably the 2nd, but =
it is
not certain it has the 1st.

I did not mark it as stable, since I am not sure such an architecture
exists.


[1] =
https://lore.kernel.org/lkml/tip-b956575bed91ecfb136a8300742ecbbf451471ab@=
git.kernel.org/=