linux-kernel - Re: [PATCH] arm64: dts: sdm845: Add iommus property to qup1

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

Date:   Wed, 05 Jun 2019 13:56:14 -0700
From:   Stephen Boyd <swboyd@...omium.org>
To:     Vivek Gautam <vivek.gautam@...eaurora.org>
Cc:     Bjorn Andersson <bjorn.andersson@...aro.org>,
        Andy Gross <agross@...nel.org>,
        open list <linux-kernel@...r.kernel.org>,
        linux-arm-msm <linux-arm-msm@...r.kernel.org>,
        Sibi Sankar <sibis@...eaurora.org>
Subject: Re: [PATCH] arm64: dts: sdm845: Add iommus property to qup1

Quoting Vivek Gautam (2019-06-04 21:55:26)
> On Wed, Jun 5, 2019 at 4:16 AM Stephen Boyd <swboyd@...omium.org> wrote:
> >
> > Quoting Bjorn Andersson (2019-06-04 15:37:00)
> > > On Tue 04 Jun 15:29 PDT 2019, Stephen Boyd wrote:
> > >
> > > > The SMMU that sits in front of the QUP needs to be programmed properly
> > > > so that the i2c geni driver can allocate DMA descriptors. Failure to do
> > > > this leads to faults when using devices such as an i2c touchscreen where
> > > > the transaction is larger than 32 bytes and we use a DMA buffer.
> > > >
> > >
> > > I'm pretty sure I've run into this problem, but before we marked the
> > > smmu bypass_disable and as such didn't get the fault, thanks.
> > >
> > > >  arm-smmu 15000000.iommu: Unexpected global fault, this could be serious
> > > >  arm-smmu 15000000.iommu:         GFSR 0x00000002, GFSYNR0 0x00000002, GFSYNR1 0x000006c0, GFSYNR2 0x00000000
> > > >
> > > > Add the right SID and mask so this works.
> > > >
> > > > Cc: Sibi Sankar <sibis@...eaurora.org>
> > > > Signed-off-by: Stephen Boyd <swboyd@...omium.org>
> > > > ---
> > > >  arch/arm64/boot/dts/qcom/sdm845.dtsi | 1 +
> > > >  1 file changed, 1 insertion(+)
> > > >
> > > > diff --git a/arch/arm64/boot/dts/qcom/sdm845.dtsi b/arch/arm64/boot/dts/qcom/sdm845.dtsi
> > > > index fcb93300ca62..2e57e861e17c 100644
> > > > --- a/arch/arm64/boot/dts/qcom/sdm845.dtsi
> > > > +++ b/arch/arm64/boot/dts/qcom/sdm845.dtsi
> > > > @@ -900,6 +900,7 @@
> > > >                       #address-cells = <2>;
> > > >                       #size-cells = <2>;
> > > >                       ranges;
> > > > +                     iommus = <&apps_smmu 0x6c0 0x3>;
> > >
> > > According to the docs this stream belongs to TZ, the HLOS stream should
> > > be 0x6c3.
> >
> > Aye, I saw this line in the downstream kernel but it doesn't work for
> > me. If I specify <&apps_smmu 0x6c3 0x0> it still blows up. I wonder if
> > my firmware perhaps is missing some initialization here to make the QUP
> > operate in HLOS mode? Otherwise, I thought that the 0x3 at the end was
> > the mask and so it should be split off to the second cell in the DT
> > specifier but that seemed a little weird.
> 
> Two things here -
> 0x6c0 - TZ SID. Do you see above fault on MTP sdm845 devices?

No. I see the above fault on Cheza.

> 0x6c3/0x6c6 - HLOS SIDs.

Why are there two? I see some mentions of GSI mode near these SIDs so
maybe GSI has to be used for DMA here to get the above two SIDs at the
IOMMU? Otherwise if we do the non-GSI mode of DMA we're going to use the
"TZ" SID?

> 
> Cheza will throw faults for anything that is programmed with TZ on mtp
> as all of that should be handled in HLOS. The firmwares of all these
> peripherals assume that the SID reservation is done (whether in TZ or HLOS).
> 
> I am inclined to moving the iommus property for all 'TZ' to board dts files.
> MTP wouldn't need those SIDs. So, the SOC level dtsi will have just the
> HLOS SIDs.

So you're saying you'd like to have the SID be <&apps_smmu 0x6c3 0x0> in
the sdm845.dtsi file and then override this on Cheza because our SID is
different (possibly because we don't use GSI)? Why can't we program the
SID in Cheza firmware to match the "HLOS" SID of 0x6c3?

> 
> P.S.
> As you rightly said, the second cell in iommus property is the mask so that
> the iommu is able to reserve all that SIDs that are covered with the
> starting SID
> and the mask.
> 

Well if 0x6c6 is another possibility maybe it should be <&apps_smmu
0x6c0 0x7> to cover the 0x6c3 and 0x6c6 SIDs?