Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <20220412033335.1384230-1-apatel@ventanamicro.com> <4c4c7a.17284.1802b8c14cc.Coremail.panqinglin2020@iscas.ac.cn>
In-Reply-To: <4c4c7a.17284.1802b8c14cc.Coremail.panqinglin2020@iscas.ac.cn>
From:   Anup Patel <apatel@ventanamicro.com>
Date:   Fri, 15 Apr 2022 10:23:44 +0530
Message-ID: <CAK9=C2V0kiV1_12kMDxv6d4nsQZ81-9W5=JifLhNaiZMoHJyuQ@mail.gmail.com>
Subject: Re: [PATCH] RISC-V: mm: Fix set_satp_mode() for platform not having Sv57
To:     =?UTF-8?B?5r2Y5bqG6ZyW?= <panqinglin2020@iscas.ac.cn>
Cc:     Palmer Dabbelt <palmer@dabbelt.com>,
        Paul Walmsley <paul.walmsley@sifive.com>,
        Alexandre Ghiti <alexandre.ghiti@canonical.com>,
        Atish Patra <atishp@atishpatra.org>,
        Alistair Francis <Alistair.Francis@wdc.com>,
        Anup Patel <anup@brainfault.org>,
        linux-riscv <linux-riscv@lists.infradead.org>,
        "linux-kernel@vger.kernel.org List" <linux-kernel@vger.kernel.org>,
        Mayuresh Chitale <mchitale@ventanamicro.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, Apr 15, 2022 at 10:18 AM =E6=BD=98=E5=BA=86=E9=9C=96 <panqinglin202=
0@iscas.ac.cn> wrote:
>
> Hi Anup,
>
> &gt;
> &gt; When Sv57 is not available the satp.MODE test in set_satp_mode() wil=
l
> &gt; fail and lead to pgdir re-programming for Sv48. The pgdir re-program=
ming
> &gt; will fail as well due to pre-existing pgdir entry used for Sv57 and =
as
> &gt; a result kernel fails to boot on RISC-V platform not having Sv57.
> &gt;
> &gt; To fix above issue, we should clear the pgdir memory in set_satp_mod=
e()
> &gt; before re-programming.
> &gt;
> &gt; Fixes: 011f09d12052 ("riscv: mm: Set sv57 on defaultly")
> &gt; Reported-by: Mayuresh Chitale <mchitale@ventanamicro.com>
> &gt; Signed-off-by: Anup Patel <apatel@ventanamicro.com>
> &gt; ---
> &gt;  arch/riscv/mm/init.c | 1 +
> &gt;  1 file changed, 1 insertion(+)
> &gt;
> &gt; diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
> &gt; index 9535bea8688c..b0793dc0c291 100644
> &gt; --- a/arch/riscv/mm/init.c
> &gt; +++ b/arch/riscv/mm/init.c
> &gt; @@ -718,6 +718,7 @@ static __init void set_satp_mode(void)
> &gt;            if (!check_l4) {
> &gt;                    disable_pgtable_l5();
> &gt;                    check_l4 =3D true;
> &gt; +                  memset(early_pg_dir, 0, PAGE_SIZE);
> &gt;                    goto retry;
> &gt;            }
> &gt;            disable_pgtable_l4();
> &gt; --
>
>
> I find it that the set_satp_mode function is in .init.text section which =
begins at 0x80800000.
> And its pgd_index in both Sv48 and Sv57 will be 0. So it may not be neces=
sary to clear the
> early_pg_dir when the kernel find Sv57 is not supported? And may I get th=
e steps of reproduction
> from you?

We can't assume that it will be the same pgd_index for Sv48 and Sv57.

For example, some hypothetical SoC might have RAM starting after 1TB space.

We should ensure that early_pg_dir is cleaned entirely for detecting the ne=
xt
mode.

Regards,
Anup

>
> Yours,
> Qinglin
> </apatel@ventanamicro.com></mchitale@ventanamicro.com>