11:04 <Marijn[m]> https://lore.kernel.org/linux-arm-msm/20221115101122.155440-1-angelogioacchino.delregno@collabora.com/T/#u bump for the IOMMU v2 patch-series, our MSM8976 display-and-whatnot patches (among which) are still blocking on this

12:38 <lumag> Marijn[m], a dumb question: which contexts are secured? In other words, do we have to list them at all?

12:39 <lumag> Marijn[m], kholk: it would be nice to get the schema converted and then to see an example of v2, demonstrating the necessity for new hooks.

13:47 <aka_[m]> <lumag> "Marijn, a dumb question: which..." <- for example on gpu you have 3 context banks

13:48 <aka_[m]> one not secured and one secured

13:48 <aka_[m]> and third is bloat

13:49 <aka_[m]> https://android.googlesource.com/kernel/msm.git/+/android-msm-lego-3.10-marshmallow-dr/arch/arm/boot/dts/qcom/msm8976-iommu.dtsi

13:49 <aka_[m]> qcom,secure-context;

13:49 <aka_[m]> flag

13:50 <lumag> aka_[m], my question is more about the reasons to touch secure contexts at all

13:51 <aka_[m]> tbh i havent touch them at all on my 8976 fork

13:51 <aka_[m]> i just didnt even defined them in dts

13:51 <aka_[m]> because if you do it breaks

13:52 <aka_[m]> the only thing i know is fact num-ctx is quite useful when we want to execute SCM commands to do something on context bank as they are not mapped 1=1

13:52 <lumag> Yep, I have no questions regarding the ctx-num

13:53 <lumag> (The name might be not the perfect one, but otherwise I see the need for it).

13:54 <aka_[m]> i have noticed something weird with iommu tho

13:54 <aka_[m]> if you don't define any secure context bank, venus won't load firmware

13:54 <aka_[m]> it will just throw some generic error about loading

13:55 <aka_[m]> i was stuck with that for quite a while and got this info from Vladimir(who worked on orginal msm8953 bringup)

14:02 <aka_[m]> he had some hack for using arm-smmu-v2 but my limited understanding is that its only looking for flag and if available use TZ call to restore iommu.

14:02 <aka_[m]> https://github.com/msm8953-mainline/linux/commit/b7d094659e594321572c39752d4853e2f0036e65

22:05 <lumag> aka_[m], this is for asm-smmu(-qcom), while the kholk's patchset targets older qcom-iommu

22:12 <konradybcio> well it's all arm smmu 500 or v2 underneath, it just differs in the amount of hyp massaging done inbetween

22:13 <konradybcio> I think this patch is essentially "porting" the necessary parts of qcom-iommu to the v2 qcom impl driver