linux-kernel - Re: [PATCH v5 09/10] iommu/mediatek: Modify MMU

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Tue, 30 Jun 2020 19:02:21 +0800
From:   chao hao <Chao.Hao@...iatek.com>
To:     Matthias Brugger <matthias.bgg@...il.com>
CC:     Joerg Roedel <joro@...tes.org>, Rob Herring <robh+dt@...nel.org>,
        "Yong Wu" <yong.wu@...iatek.com>,
        Evan Green <evgreen@...omium.org>,
        <iommu@...ts.linux-foundation.org>, <devicetree@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>,
        <linux-arm-kernel@...ts.infradead.org>,
        <linux-mediatek@...ts.infradead.org>, <wsd_upstream@...iatek.com>,
        FY Yang <fy.yang@...iatek.com>,
        Chao Hao <chao.hao@...iatek.com>
Subject: Re: [PATCH v5 09/10] iommu/mediatek: Modify MMU_CTRL register
 setting

On Mon, 2020-06-29 at 12:28 +0200, Matthias Brugger wrote:
> 
> On 29/06/2020 09:13, Chao Hao wrote:
> > MT8173 is different from other SoCs for MMU_CTRL register.
> > For mt8173, its bit9 is in_order_write_en and doesn't use its
> > default 1'b1.> For other SoCs, bit[12] represents victim_tlb_en feature and
> > victim_tlb is enable defaultly(bit[12]=1), if we use
> > "regval = F_MMU_TF_PROT_TO_PROGRAM_ADDR", victim_tlb will be
> > disabled, it will drop iommu performace.
> > So we need to deal with the setting of MMU_CTRL separately
> > for mt8173 and others.
> > 
> 
> My proposal to rewrite the commit message:
> 
> The MMU_CTRL regiser of MT8173 is different from other SoCs. The in_order_wr_en
> is bit[9] which is zero by default.
> Other SoCs have the vitcim_tlb_en feature mapped to bit[12]. This bit is set to
> one by default. We need to preserve the bit when setting
> F_MMU_TF_PROT_TO_PROGRAM_ADDR as otherwise the bit will be cleared and IOMMU
> performance will drop.

got it, thanks for your advice very much.

> 
> 
> > Suggested-by: Matthias Brugger <matthias.bgg@...il.com>
> > Suggested-by: Yong Wu <yong.wu@...iatek.com>
> > Signed-off-by: Chao Hao <chao.hao@...iatek.com>
> > ---
> >  drivers/iommu/mtk_iommu.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c
> > index 8299a3299090..e46e2deee3fd 100644
> > --- a/drivers/iommu/mtk_iommu.c
> > +++ b/drivers/iommu/mtk_iommu.c
> > @@ -543,11 +543,12 @@ static int mtk_iommu_hw_init(const struct mtk_iommu_data *data)
> >  		return ret;
> >  	}
> >  
> > +	regval = readl_relaxed(data->base + REG_MMU_CTRL_REG);
> 
> The read is only needed in the else branch.
> 
ok, thanks

> >  	if (data->plat_data->m4u_plat == M4U_MT8173)
> >  		regval = F_MMU_PREFETCH_RT_REPLACE_MOD |
> >  			 F_MMU_TF_PROT_TO_PROGRAM_ADDR_MT8173;
> >  	else
> > -		regval = F_MMU_TF_PROT_TO_PROGRAM_ADDR;
> > +		regval |= F_MMU_TF_PROT_TO_PROGRAM_ADDR;
> >  	writel_relaxed(regval, data->base + REG_MMU_CTRL_REG);
> >  
> >  	regval = F_L2_MULIT_HIT_EN |
> >