Re: [PATCH 3/3] [RFC] arm64: dts: qcom: msm8992-bullhead: Fix smem_region, mpss_mem

[Konrad Dybcio <konrad.dybcio@linaro.org>]

On 26.12.2022 19:54, Petr Vorel wrote:
> From: Petr Vorel <petr.vorel@gmail.com>
> 
> smem_region has a different region on downstream than on msm8994, it's
> defined as 150 MiB [1]:
> 
> [    0.000000] Linux version 3.10.73-g65a266a (buildkite-agent@lineageos-buildkite) (gcc version 4.9.x 20150123 (prerelease) (GCC) ) #1 SMP PREEMPT Mon Feb 24 10:43:08 UTC 2020
> [    0.000000] CPU: AArch64 Processor [410fd033] revision 3
> [    0.000000] Machine: LGE MSM8992 BULLHEAD rev-1.01
> [    0.000000] cma: Found secure_region@0, memory base 0x0000000000000000, size 150 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found qsecom_region@0, memory base 0x0000000000000000, size 4 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found audio_region@0, memory base 0x0000000000000000, size 4 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found removed_regions@0, memory base 0x0000000000000000, size 34 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found cont_splash_mem@0, memory base 0x0000000000000000, size 12 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found peripheral_region@0, memory base 0x0000000000000000, size 28 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found modem_region@0, memory base 0x0000000000000000, size 100 MiB, limit 0x0000000000000000
> [    0.000000] cma: Found ramoops_region@1ff00000, memory base 0x0000000000000000, size 0 MiB, limit 0x0000000000000000
> [    0.000000] cma: CMA: reserved 152 MiB at 0x0000000000000000 for secure_mem
> [    0.000000] cma: CMA: reserved 4 MiB at 0x0000000000000000 for qseecom_mem
> [    0.000000] cma: CMA: reserved 4 MiB at 0x0000000000000000 for audio_mem
> [    0.000000] cma: CMA: reserved 34 MiB at 0x0000000000000000 for memory_hole
> [    0.000000] cma: CMA: reserved 12 MiB at 0x0000000000000000 for cont_splash_mem
> [    0.000000] cma: CMA: reserved 28 MiB at 0x0000000000000000 for peripheral_mem
> [    0.000000] cma: CMA: reserved 100 MiB at 0x0000000000000000 for modem_mem
> [    0.000000] cma: CMA: reserved 0 MiB at 0x0000000000000000 for ramoops_mem
> [    0.000000] cma: CMA: reserved 16 MiB at 0x0000000000000000 for default region
> [    0.000000] PERCPU: Embedded 15 pages/cpu @0000000000000000 s38912 r0 d22528 u61440
> 
> But that conflicts with cont_splash_mem and cont_splash_mem:
> 
> [    0.000000] Linux version 6.1.1 (pevik@dell5510) (aarch64-linux-gnu-gcc (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils for De2
> [    0.000000] Machine model: LG Nexus 5X rev 1.01
> ...
> [    0.000000] OF: fdt: Reserved memory: failed to reserve memory for node 'smem_region@0': base 0x0000000000000000, size 150 MiB
> [    0.000000] OF: reserved mem: OVERLAP DETECTED!
> [    0.000000] smem_region@0 (0x0000000000000000--0x0000000009600000) overlaps with memory@3400000 (0x0000000003400000--0x000000000400000)
> [    0.000000] OF: reserved mem: OVERLAP DETECTED!
> [    0.000000] reserved@5000000 (0x0000000005000000--0x0000000007200000) overlaps with memory@7000000 (0x0000000007000000--0x000000000ca0)
> 
> 1) Obviously there are more memory to be fixed, e.g. modem_mem, but
> fundamental is to find out whether such a big mapping should be really
> used. If yes, what is the trick downstream uses, that it allows
> overlapped memory? Or memory is not really overlapping and I cannot
> trust the reported value? What base and size should I define then?
> 
> Angler downstream defines crazy values for secure_mem as well [2]:
> [    0.000000] cma: Found secure_region@0, memory base 0x0000000000000000, size 300 MiB, limit 0xffffffffffffffff
> 
> But then in mainline is defined just 2 MiB at 106 MiB:
> 
>     smem_mem: smem_region@6a00000 {
> 	    reg = <0 0x06a00000 0 0x200000>;
> 	    no-map;
>     };
> 
> => should I use just 2 MiB region? But at what base?
> 
> 2) Is mpss_mem equivalent of downstream modem_mem [3]?
> 
> Kind regards,
> Petr
> 
> [1] https://android.googlesource.com/kernel/msm.git/+/refs/tags/android-7.0.0_r0.17/arch/arm/boot/dts/qcom/msm8992.dtsi#278
> [2] https://android.googlesource.com/kernel/msm.git/+/refs/tags/android-7.0.0_r0.17/arch/arm/boot/dts/qcom/msm8994.dtsi#272
> [3] https://android.googlesource.com/kernel/msm.git/+/android-7.0.0_r0.17/arch/arm64/boot/dts/lge/msm8992-bullhead.dtsi#148
> 
> Not-Yet-Signed-off-by: Petr Vorel <petr.vorel@gmail.com>
> Signed-off-by: Petr Vorel <petr.vorel@gmail.com>
> ---
smem is not device-specific. secure_mem is a bunch of PIL+maybe
TZAPP memory regions, it's really a mess downstream..

Look here:
https://android.googlesource.com/kernel/msm/+/43154bfee910f2efd0ff0d60e6b1c25192367ab9/arch/arm/boot/dts/qcom/msm8992.dtsi#2173

Konrad
>  arch/arm64/boot/dts/qcom/msm8992-lg-bullhead.dtsi | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/boot/dts/qcom/msm8992-lg-bullhead.dtsi b/arch/arm64/boot/dts/qcom/msm8992-lg-bullhead.dtsi
> index 4bceb362a5c0..fb4879b4acbe 100644
> --- a/arch/arm64/boot/dts/qcom/msm8992-lg-bullhead.dtsi
> +++ b/arch/arm64/boot/dts/qcom/msm8992-lg-bullhead.dtsi
> @@ -12,8 +12,9 @@
>  #include "pm8994.dtsi"
>  #include "pmi8994.dtsi"
>  
> -/* cont_splash_mem has different memory mapping */
> +/* different memory mapping */
>  /delete-node/ &cont_splash_mem;
> +/delete-node/ &smem_mem;
>  
>  /* disabled on downstream, conflicts with cont_splash_mem */
>  /delete-node/ &dfps_data_mem;
> @@ -60,6 +61,11 @@ removed_region: reserved@5000000 {
>  			reg = <0 0x05000000 0 0x2200000>;
>  			no-map;
>  		};
> +
> +		smem_mem: smem_region@0 {
> +			reg = <0 0 0 0x9600000>;
> +			no-map;
> +		};
>  	};
>  };
>