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
Content-Transfer-Encoding: quoted-printable
From:   Andy Lutomirski <luto@amacapital.net>
Mime-Version: 1.0 (1.0)
Subject: Re: [PATCH] x86/efi: Do not release sub-1MB memory regions when the crashkernel option is specified
Date:   Sun, 11 Apr 2021 18:49:12 -0700
Message-Id: <8FAA2A0E-0A09-4308-B936-CDD2C0568BAE@amacapital.net>
References: <20210412011347.GA4282@MiWiFi-R3L-srv>
Cc:     "H. Peter Anvin" <hpa@zytor.com>,
        Lianbo Jiang <lijiang@redhat.com>,
        linux-kernel@vger.kernel.org, linux-efi@vger.kernel.org,
        platform-driver-x86@vger.kernel.org, x86@kernel.org,
        ardb@kernel.org, tglx@linutronix.de, mingo@redhat.com,
        bp@alien8.de, dvhart@infradead.org, andy@infradead.org,
        kexec@lists.infradead.org, dyoung@redhat.com
In-Reply-To: <20210412011347.GA4282@MiWiFi-R3L-srv>
To:     Baoquan He <bhe@redhat.com>
Precedence: bulk


> On Apr 11, 2021, at 6:14 PM, Baoquan He <bhe@redhat.com> wrote:
>=20
> =EF=BB=BFOn 04/09/21 at 07:59pm, H. Peter Anvin wrote:
>> Why don't we do this unconditionally? At the very best we gain half a meg=
abyte of memory (except the trampoline, which has to live there, but it is o=
nly a few kilobytes.)
>=20
> This is a great suggestion, thanks. I think we can fix it in this way to
> make code simpler. Then the specific caring of real mode in
> efi_free_boot_services() can be removed too.
>=20

This whole situation makes me think that the code is buggy before and buggy a=
fter.

The issue here (I think) is that various pieces of code want to reserve spec=
ific pieces of otherwise-available low memory for their own nefarious uses. I=
 don=E2=80=99t know *why* crash kernel needs this, but that doesn=E2=80=99t m=
atter too much.

I propose that the right solution is to give low-memory-reserving code paths=
 two chances to do what they need: once at the very beginning and once after=
 EFI boot services are freed.

Alternatively, just reserve *all* otherwise unused sub 1M memory up front, t=
hen release it right after releasing boot services, and then invoke the spec=
ial cases exactly once.

In either case, the result is that the crashkernel mess gets unified with th=
e trampoline mess.  One way the result is called twice and needs to be more c=
areful, and the other way it=E2=80=99s called only once.

Just skipping freeing boot services seems wrong.  It doesn=E2=80=99t unmap b=
oot services, and skipping that is incorrect, I think. And it seems to resul=
t in a bogus memory map in which the system thinks that some crashkernel mem=
ory is EFI memory instead.

Let=E2=80=99s please just fix the problem instead of papering over it with m=
ore hacks.

> Thanks
> Baoquan
>=20