Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Song Liu <songliubraving@fb.com>
To:     Peter Zijlstra <peterz@infradead.org>
CC:     Steven Rostedt <rostedt@goodmis.org>,
        "sbsiddha@gmail.com" <sbsiddha@gmail.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Linux MM <linux-mm@kvack.org>,
        Kernel Team <Kernel-team@fb.com>,
        "stable@vger.kernel.org" <stable@vger.kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Dave Hansen <dave.hansen@intel.com>,
        Andy Lutomirski <luto@amacapital.net>
Subject: Re: [PATCH] x86/mm: Do not split_large_page() for
 set_kernel_text_rw()
Thread-Topic: [PATCH] x86/mm: Do not split_large_page() for
 set_kernel_text_rw()
Thread-Index: AQHVWXL45hcGbMxBFEmxITCR8fhmlKcIeZ6AgARkOQCAAE74AIAAYJqAgABfWYA=
Date:   Mon, 26 Aug 2019 20:50:02 +0000
Message-ID: <28DA9193-BD70-4B81-B1A9-98BC3E8BDDFA@fb.com>
References: <20190823052335.572133-1-songliubraving@fb.com>
 <20190823093637.GH2369@hirez.programming.kicks-ass.net>
 <164D1F08-80F7-4E13-94FC-78F33B3E299F@fb.com>
 <20190826092300.GN2369@hirez.programming.kicks-ass.net>
 <0A94F7AA-7ECE-4363-B960-41F644CFE942@fb.com>
In-Reply-To: <0A94F7AA-7ECE-4363-B960-41F644CFE942@fb.com>
Accept-Language: en-US
Content-Language: en-US
received-spf: None (protection.outlook.com: fb.com does not designate
 permitted sender hosts)
Content-Type: text/plain; charset="us-ascii"
Content-ID: <9D2E2E6015FEFE4CA52E5932D281EA5E@namprd15.prod.outlook.com>
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
X-MS-Exchange-CrossTenant-Network-Message-Id: d039bacc-b2a7-45b9-33e6-08d72a66f778
X-MS-Exchange-CrossTenant-originalarrivaltime: 26 Aug 2019 20:50:02.1246
 (UTC)
X-MS-Exchange-CrossTenant-fromentityheader: Hosted
X-MS-Exchange-CrossTenant-id: 8ae927fe-1255-47a7-a2af-5f3a069daaa2
X-MS-Exchange-CrossTenant-mailboxtype: HOSTED
X-MS-Exchange-CrossTenant-userprincipalname: 4ML9PNjmsaE2juQkZTgKsJhzM6MpFq4+aTR9UjciuXp8+34cCcowfKKPQSHNE86PJYI4ysQ2fudYI0aQeU4aRQ==
X-MS-Exchange-Transport-CrossTenantHeadersStamped: MWHPR15MB1806
X-OriginatorOrg: fb.com
X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10434:5.22.84,1.0.8
 definitions=2019-08-26_08:2019-08-26,2019-08-26 signatures=0
X-Proofpoint-Spam-Details: rule=fb_default_notspam policy=fb_default score=0 mlxscore=0 clxscore=1015
 bulkscore=0 mlxlogscore=999 suspectscore=0 adultscore=0 lowpriorityscore=0
 malwarescore=0 priorityscore=1501 phishscore=0 spamscore=0 impostorscore=0
 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-1906280000
 definitions=main-1908260195
X-FB-Internal: deliver
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org


> On Aug 26, 2019, at 8:08 AM, Song Liu <songliubraving@fb.com> wrote:
>=20
>=20
>=20
>> On Aug 26, 2019, at 2:23 AM, Peter Zijlstra <peterz@infradead.org> wrote=
:
>>=20
>> So only the high mapping is ever executable; the identity map should not
>> be. Both should be RO.
>>=20
>>> kprobe (with CONFIG_KPROBES_ON_FTRACE) should work on kernel identity
>>> mapping.=20
>>=20
>> Please provide more information; kprobes shouldn't be touching either
>> mapping. That is, afaict kprobes uses text_poke() which uses a temporary
>> mapping (in 'userspace' even) to alias the high text mapping.
>=20
> kprobe without CONFIG_KPROBES_ON_FTRACE uses text_poke(). But kprobe with
> CONFIG_KPROBES_ON_FTRACE uses another path. The split happens with
> set_kernel_text_rw() -> ... -> __change_page_attr() -> split_large_page()=
.
> The split is introduced by commit 585948f4f695. do_split in=20
> __change_page_attr() becomes true after commit 585948f4f695. This patch=20
> tries to fix/workaround this part.=20
>=20
>>=20
>> I'm also not sure how it would then result in any 4k text maps. Yes the
>> alias is 4k, but it should not affect the actual high text map in any
>> way.
>=20
> I am confused by the alias logic. set_kernel_text_rw() makes the high map
> rw, and split the PMD in the high map.=20
>=20
>>=20
>> kprobes also allocates executable slots, but it does that in the module
>> range (afaict), so that, again, should not affect the high text mapping.
>>=20
>>> We found with 5.2 kernel (no CONFIG_PAGE_TABLE_ISOLATION, w/=20
>>> CONFIG_KPROBES_ON_FTRACE), a single kprobe will split _all_ PMDs in=20
>>> kernel text mapping into pte-mapped pages. This increases iTLB=20
>>> miss rate from about 300 per million instructions to about 700 per
>>> million instructions (for the application I test with).=20
>>>=20
>>> Per bisect, we found this behavior happens after commit 585948f4f695=20
>>> ("x86/mm/cpa: Avoid the 4k pages check completely"). That's why I=20
>>> proposed this PATCH to fix/workaround this issue. However, per
>>> Peter's comment and my study of the code, this doesn't seem the=20
>>> real problem or the only here.=20
>>>=20
>>> I also tested that the PMD split issue doesn't happen w/o=20
>>> CONFIG_KPROBES_ON_FTRACE.=20
>>=20
>> Right, because then ftrace doesn't flip the whole kernel map writable;
>> which it _really_ should stop doing anyway.
>>=20
>> But I'm still wondering what causes that first 4k split...
>=20
> Please see above.=20

Another data point: we can repro the issue on Linus's master with just
ftrace:

# start with PMD mapped
root@virt-test:~# grep ffff81000000- /sys/kernel/debug/page_tables/kernel
0xffffffff81000000-0xffffffff81c00000          12M     ro         PSE      =
   x  pmd

# enable single ftrace
root@virt-test:~# echo consume_skb > /sys/kernel/debug/tracing/set_ftrace_f=
ilter
root@virt-test:~# echo function > /sys/kernel/debug/tracing/current_tracer

# now the text is PTE mapped
root@virt-test:~# grep ffff81000000- /sys/kernel/debug/page_tables/kernel
0xffffffff81000000-0xffffffff81c00000          12M     ro                  =
   x  pte

Song