Re: [PATCH v5] iommu/arm-smmu-qcom: Request direct mapping for modem device

From: Stephen Boyd <swboyd@chromium.org>
To: Sibi Sankar <sibis@codeaurora.org>,
	joro@8bytes.org, robin.murphy@arm.com, will@kernel.org
Cc: bjorn.andersson@linaro.org, linux-kernel@vger.kernel.org,
	linux-arm-msm@vger.kernel.org, iommu@lists.linux-foundation.org,
	linux-arm-kernel@lists.infradead.org, dianders@chromium.org,
	evgreen@chromium.org, mka@chromium.org,
	Sibi Sankar <sibis@codeaurora.org>,
	Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>
Subject: Re: [PATCH v5] iommu/arm-smmu-qcom: Request direct mapping for modem device
Date: Thu, 07 May 2020 14:14:37 -0700	[thread overview]
Message-ID: <158888607765.26370.18078298627976225073@swboyd.mtv.corp.google.com> (raw)
In-Reply-To: <20200507192157.6831-1-sibis@codeaurora.org>

Quoting Sibi Sankar (2020-05-07 12:21:57)
> The modem remote processor has two modes of access to the DDR, a direct
> mode and through a SMMU which requires direct mapping. The configuration
> of the modem SIDs is handled in TrustZone.

Is it "The configuration of the modem SIDs is typically handled by
code running in the ARM CPU's secure mode, i.e. secure EL1"? And is that
even true? I though it was programmed by EL2.

> On platforms where TrustZone

TrustZone is always there.

> is absent this needs to be explicitly done from kernel. Add compatibles
> for modem to opt in for direct mapping on such platforms.
> 
> Signed-off-by: Sai Prakash Ranjan <saiprakash.ranjan@codeaurora.org>

Is Sai the author? Or does this need a co-developed-by tag?

> Signed-off-by: Sibi Sankar <sibis@codeaurora.org>
> ---
> 
> V5
>  * Reword commit message and drop unnecessary details

I don't see any improvement! Probably because I don't understand _why_
the modem needs a direct mapping. The commit text basically says "we
need to do it because it isn't done in secure world sometimes". This is
probably wrong what I wrote below, but I'd like to clarify to confirm my
understanding. Maybe the commit text should say:

The modem remote processor has two access paths to DDR. One path is
directly connected to DDR and another path goes through an SMMU. The
SMMU path is configured to be a direct mapping because it's used by
various peripherals in the modem subsystem. Typically this direct
mapping is configured by programming modem SIDs into the SMMU when EL2
responds to a hyp call from the code that loads the modem binary in the
secure world.

In certain firmware configurations, especially when the kernel is
entered at EL2, we don't want secure mode to make hyp calls to program
the SMMU because the kernel is in full control of the SMMU. Let's add
compatibles here so that we can have the kernel program the SIDs for the
modem in these cases.