Re: [RESEND PATCH v2 00/13] Enable GPU for SM8150 and SM8250

[Dmitry Baryshkov]

Hi,

On 06/07/2020 00:08, Jonathan Marek wrote:
> On 7/3/20 1:29 PM, Dmitry Baryshkov wrote:
> > On 03/07/2020 18:14, Dmitry Baryshkov wrote:
> > > On 03/07/2020 18:08, Jonathan Marek wrote:
> > > > On 7/3/20 11:03 AM, Dmitry Baryshkov wrote:
> > > > > On 30/06/2020 00:17, Jonathan Marek wrote:
> > > > > > This series adds the missing clock drivers and dts nodes to enable
> > > > > > the GPU on both SM8150 and SM8250.
> > > > > >
> > > > > > Note an extra patch [1] is still required for GPU to work on SM8250.
> > > > > >
> > > > > > Changes in V2:
> > > > > > * Added "clk: qcom: gcc: fix sm8150 GPU and NPU clocks" to fix the 
> > > > > > newly added
> > > > > > ÂÂ SM8150 GPU gcc clocks
> > > > > > * Added "Fixes:" tag to "clk: qcom: clk-alpha-pll: remove 
> > > > > > unused/incorrect PLL_CAL_VAL"
> > > > > > * Added yaml schemas to gpucc dt-bindings patches
> > > > > > * Added "clk: qcom: add common gdsc_gx_do_nothing_enable for gpucc 
> > > > > > drivers" and changed
> > > > > > ÂÂ gpucc patches to use it.
> > > > > > * Removed CLK_IS_CRITICAL from gpu_cc_ahb_clk
> > > > > > * Added missing rpmh regulator level for sm8250 GPU clock levels
> > > > > > * Use sm8150/sm8250 iommu compatibles in dts
> > > > > > * Add gcc_gpu_gpll0_clk_src/gcc_gpu_gpll0_div_clk_src to gpucc 
> > > > > > clocks in dts
> > > > > >
> > > > > > [1] https://gist.github.com/flto/784f1aca761ebf2fe6c105719a4a04ca
> > > > >
> > > > > With your patches applied:
> > > > >
> > > > > [ÂÂ 56.751977] msm msm: [drm:adreno_request_fw] loaded 
> > > > > qcom/a650_sqe.fw from new location
> > > > > [ÂÂ 56.760166] msm msm: [drm:adreno_request_fw] loaded 
> > > > > qcom/a650_gmu.bin from new location
> > > > > [ÂÂ 56.768485] arm-smmu 3da0000.iommu: genpd_runtime_resume()
> > > > > [ÂÂ 56.774196] PM: gpu_cx_gdsc: Power-on latency exceeded, new 
> > > > > value 49531 ns
> > > > > [ÂÂ 56.781730] arm-smmu 3da0000.iommu: resume latency exceeded, 
> > > > > 462604 ns
> > > > > [ÂÂ 56.799559] platform 3d6a000.gmu: [drm:a6xx_gmu_resume] *ERROR* 
> > > > > GMU firmware initialization timed out
> > > > > [ÂÂ 56.809260] arm-smmu 3da0000.iommu: genpd_runtime_suspend()
> > > > > [ÂÂ 56.813062] msm msm: [drm:adreno_load_gpu] *ERROR* Couldn't 
> > > > > power up the GPU: -110
> > > >
> > > > Do you have your branch published somewhere so I can see what 
> > > > could've went wrong?
> > >
> > > I've applied your patches (this series + the extra one for 
> > > gpu/drm/msm) on top of 
> > > https://git.linaro.org/landing-teams/working/qualcomm/kernel.git 
> > > branch integration-linux-qcomlt .
> >
> > A trimmed down version: 
> > git.linaro.org/people/dmitry.baryshkov/kernel.git branch sm8250-gpu-test
>
> Hi,
>
> I tried this branch, with only the sm8250-hdk.dts from my other series 
> added (USB nodes removed as you don't have those in your branch), and 
> the GPU starts up without problems.
>
> Possible differences I can think of:
>
> 1) Different firmware version which behaves differently? These are the 
> checksums for the firmware I have:
>
> /lib/firmware# md5sum a650_*
> f6536ba45c2f9f64ec31438217b6a027Â a650_gmu.bin
> 897be740bed67deaa1943e9c36293165Â a650_sqe.fw
> 00f1b291a2b30c98f05c25506e3f4761Â a650_zap.b00
> bec0f79c7c6f5b32254facf3c7e046c7Â a650_zap.b01
> 35273e4135147a269076144a3051b498Â a650_zap.b02
> ce0dd1af27306eb341a01bda642f1c64Â a650_zap.elf
> 09834955a8865073f6ee483f69a98b33Â a650_zap.mdt

Interesting. This is what I have here:

3a3a455289c8c36b97a76b95d1dda5cb  a650_gmu.bin
9439db9b76c84f4aec625ff2bc4d1f90  a650_sqe.fw
00f1b291a2b30c98f05c25506e3f4761  a650_zap.b00
f30017eb17a440476a939d8eb1fbe102  a650_zap.b01
35273e4135147a269076144a3051b498  a650_zap.b02
03cef57e54ff48ab3ad49c7e2bcac56e  a650_zap.elf
93f651f41f4bab30dfb8e2bbd8f64ffd  a650_zap.mdt

> 2) GPU revision is not 650.2 (the downstream driver had workarounds for 
> bugs in the first revisions, which I didn't include in the upstream driver)

Interesting, I will take a look. The first obvious difference is the 
hang timeout and the a6xx_gmu_start() handling.

> 3) Something in the kernel config is making a difference. Here is the 
> kernel config I used to test this: 
> https://gist.github.com/flto/c6b1bc48abda6fb580a2f21c51039c81

No, your config produces the same result:

[   46.964891] msm msm: [drm:0xffffffc01049bd00] loaded qcom/a650_sqe.fw 
from new location
[   46.973156] msm msm: [drm:0xffffffc01049bd00] loaded 
qcom/a650_gmu.bin from new location
[   46.994573] platform 3d6a000.gmu: [drm:0xffffffc0104a85d4] *ERROR* 
GMU firmware initialization timed out
[   47.004537] msm msm: [drm:0xffffffc01049b584] *ERROR* Couldn't power 
up the GPU: -110

--
With best wishes
Dmitry