07:39 <z3ntu1> lumag: https://github.com/z3ntu/linux/blob/b4/fp5-pmic-glink-dp/arch/arm64/boot/dts/qcom/qcm6490-fairphone-fp5.dts

07:40 <z3ntu1> This is the branch I was running yesterday, a 6.7-based branch with more stuff (all my patches) is at https://github.com/z3ntu/linux/blob/fp5-6.7-wip/arch/arm64/boot/dts/qcom/qcm6490-fairphone-fp5.dts if that's interesting at all

07:42 <z3ntu1> kholk: Do you know which iommu compatible string (v1 or v2) the iommu on msm8953 needs? Since the driver is doing the same thing (apart from -v2-sec) for both compatibles, I don't really know what should be used. Downstream uses qcom,smmu-v2 for kgsl_smmu, and qcom,qsmmu-v500 for apps_iommu, both currently seem to be happy enough with the upstream driver

10:07 <aka_[m]> prob doesn't matter just go with v2

10:08 <aka_[m]> for qcom iommus mostly flags matter on downstream

10:08 <aka_[m]> from hardware standpoint both iommus are MMU500

10:14 <lumag> z3ntu1, could you please run the dts through cpp (to get all includes substituted) and commit it somewhere?

10:15 <lumag> I'd like to point to it in the annotation, so that anybody could follow the links.

10:15 <lumag> also I can only hope for fp5-probe-loop being semi-permanent link.

10:18 <lumag> z3ntu1, reg 8953. I think v2 for adreno_smmu and smmu500 for apps matches what we are using for sdm845. However some (most) of older platforms use just -v2 for all smmus.

10:20 <lumag> The major difference is, IIRC, additional reset procedure

10:29 <z3ntu1> lumag: So you want basically `arch/arm64/boot/dts/qcom/.qcm6490-fairphone-fp5.dtb.dts.tmp`? Looking at `scripts/Makefile.lib` this should be the file produced by the compiler before dtc is run

10:29 <lumag> z3ntu1, yes

10:30 <z3ntu1> is it here okay? https://public.lucaweiss.eu/tmp/.qcm6490-fairphone-fp5.dtb.dts.tmp Or push to github?

10:34 <lumag> z3ntu1, how permanent is your tmp?

10:35 <z3ntu1> Until I manually delete the files (which doesn't happen often), should be there until in a few months at least

10:38 <lumag> ack

10:42 <z3ntu1> lumag: For 8953, upstream currently uses "qcom,msm8953-iommu", "qcom,msm-iommu-v1" for apps_iommu, but not even sure that's "correct" or if it should be -v2 (as I said, no driver difference so it's just a dts question)

10:42 <lumag> Ah, msm-iommu-v2. I was thinking about the arm smmu -v2

10:42 <z3ntu1> and gpu_iommu is not yet upstream, which is what I'm trying to do

11:31 <aka_[m]> lumag: naming kinda confusing indeed

11:33 <aka_[m]> there is msm_iommu(8974) qcom_iommu(MMU500+Hyp(Fam B socs msm8916-msm8953(incl 8976)+smmu-v2(660?) and then its mmu500+smmu-v2(different impl? as 7180 gpu smmu)

11:35 <konradybcio> msm_iommu is for even earlier targets

11:35 <konradybcio> 80something60something

11:37 <aka_[m]> konradybcio: 8974 is kinda weird there

11:38 <aka_[m]> ds on 3.10 have both old msm-iommu-v1.dts and msm-iommu-v2.dts

11:39 <aka_[m]> do armv7 have TZ?

11:39 <aka_[m]> i guess no hyp there

12:13 <konradybcio> yes tz is an old invention

12:13 <konradybcio> as for hyp, i don't know

12:13 <konradybcio> there is virtualization support for armv7 though

12:13 <konradybcio> (not on qc)

12:33 <phh> as i understand it, the difference between armv8 and armv7 with regards to trustzone, is that trustzone as an $$$ option on Arm's cortex-A 32bits, while it's not an option on the 64bits one, hence ~ everyone has one

13:24 <minecrell> phh: In the ARM architecture it's still optional for AArch64, e.g. Apple silicon doesn't have EL3 (and therefore no TZ)

13:24 <minecrell> aka_[m]: The 8916 32-bit firmware has hyp too, as well as some newer MDM platforms, but it's probably rare

13:25 <phh> minecrell: yes, that's why my understanding of the major changed in usage of TZ we saw from armv7 to armv8 comes from the available paying option of cortex-a

14:09 <z3ntu1> So anybody have a proposal what I can do with 8953 iommu? Make everything msm-iommu-v1? Make everything msm-iommu-v2? Keep apps_iommu v1 and make gpu_smmu v2?

22:06 <aka_[m]> its still same thing

22:06 <aka_[m]> and i believe issue with v2 is that there are some "secure" iommus which you cannot really reprogram

22:07 <aka_[m]> minecrell: do 8916 have something like secure context bank for gpu?

22:08 <aka_[m]> on 8953 there is user/priv/secure

22:08 <aka_[m]> in theory iommus have also space for 4th context bank

23:26 <lumag> z3ntu1, if gpu_smmu works with the generic driver it might be better to use it. Otherwise you have to use msm-iommu-v2, IIUC