Re: [PATCH v6 0/4] arm64: dts: qcom: Introduce Glymur SoC dtsi and Glymur CRD dts

[Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>]

On 2/2/26 11:21 AM, Qiang Yu wrote:
> On Mon, Feb 02, 2026 at 10:49:10AM +0100, Konrad Dybcio wrote:
>> On 2/2/26 6:21 AM, Qiang Yu wrote:
>>> On Thu, Jan 29, 2026 at 01:07:08PM +0100, Konrad Dybcio wrote:
>>>> On 1/29/26 1:05 PM, Qiang Yu wrote:
>>>>> On Wed, Jan 28, 2026 at 07:21:04PM -0600, Bjorn Andersson wrote:
>>>>>> On Thu, Jan 22, 2026 at 08:53:57PM +0530, Pankaj Patil wrote:
>>>>>>> Introduce dt-bindings and initial device tree support for Glymur,
>>>>>>> Qualcomm's next-generation compute SoC and it's associated
>>>>>>> Compute Reference Device (CRD) platform.
>>>>>>>
>>>>>>> https://www.qualcomm.com/products/mobile/snapdragon/laptops-and-tablets/snapdragon-x2-elite
>>>>>>> https://www.qualcomm.com/news/releases/2025/09/new-snapdragon-x2-elite-extreme-and-snapdragon-x2-elite-are-the-
>>>>>>>
>>>>>>> The base support enables booting to shell with rootfs on NVMe,
>>>>>>> demonstrating functionality for PCIe and NVMe subsystems.
>>>>>>> DCVS is also enabled, allowing dynamic frequency scaling for the CPUs.
>>>>>>> TSENS (Thermal Sensors) enabled for monitoring SoC temperature and
>>>>>>> thermal management. The platform is capable of booting kernel at EL2
>>>>>>> with kvm-unit tests performed on it for sanity.
>>>>>>>
>>>>>>> Added dtsi files for the PMIC's enabled PMH0101, PMK8850, PMCX0102,
>>>>>>> SMB2370, PMH0104, PMH0110 along with temp-alarm and GPIO nodeS.
>>>>>>>
>>>>>>> For CPU compatible naming, there is one discussion which is not specific
>>>>>>> to Glymur, Kaanapali and Glymur use the same Oryon cores.
>>>>>>> https://lore.kernel.org/all/20251119-oryon-binding-v1-1-f79a101b0391@oss.qualcomm.com/
>>>>>>> We've kept the "qcom,oryon" compatible
>>>>>>>
>>>>>>> Features enabled in this patchset:
>>>>>>> 1. NVMe storage support
>>>>>>> 2. PCIe controller and PCIe PHY
>>>>>>> 3. RPMH Regulators
>>>>>>> 4. Clocks and reset controllers - GCC, TCSRCC, DISPCC, RPMHCC
>>>>>>> 5. Interrupt controller
>>>>>>> 6. TLMM (Top-Level Mode Multiplexer)
>>>>>>> 7. QUP Block
>>>>>>> 8. Reserved memory regions
>>>>>>> 9. PMIC support with regulators
>>>>>>> 10. CPU Power Domains
>>>>>>> 11. TSENS (Thermal Sensors)
>>>>>>> 12. DCVS: CPU DCVS with scmi perf protocol
>>>>>>>
>>>>>>> Dependencies:
>>>>>>>
>>>>>>> dt-bindings:
>>>>>>> 1. https://lore.kernel.org/all/20260121-glymur-pmic-mfd-v1-1-2aab4f21e79c@oss.qualcomm.com/
>>>>>>> 2. https://lore.kernel.org/all/20251215-knp-pmic-leds-v3-2-5e583f68b0e5@oss.qualcomm.com/
>>>>>>> 3. https://lore.kernel.org/all/20260121110828.2267061-1-pankaj.patil@oss.qualcomm.com/
>>>>>>> 4. https://lore.kernel.org/all/20260111155234.5829-1-pankaj.patil@oss.qualcomm.com/
>>>>>>>
>>>>>>> Linux-next based tree with Glymur patches is available at:
>>>>>>> https://git.codelinaro.org/clo/linux-kernel/kernel-qcom/-/tree/b4/v6_glymur_introduction
>>>>>>>
>>>>>>
>>>>>> FWIW, I applied these patches onto next-20260128 to see if things has
>>>>>> improved since Rob's report and I get:
>>>>>>
>>>>>> $ make qcom/glymur-crd.dtb CHECK_DTBS=1
>>>>>>   DTC [C] arch/arm64/boot/dts/qcom/glymur-crd.dtb
>>>>>> qcom/glymur-crd.dtb: dma-controller@800000 (qcom,glymur-gpi-dma): interrupts: [[0, 588, 4], [0, 589, 4], [0, 590, 4], [0, 591, 4], [0, 592, 4], [0, 593, 4], [0, 594, 4], [0, 595, 4], [0, 596, 4], [0, 597, 4], [0, 598, 4], [0, 599, 4], [2, 129, 4], [2, 130, 4], [2, 131, 4], [2, 132, 4]] is too long
>>>>>>         from schema $id: http://devicetree.org/schemas/dma/qcom,gpi.yaml#
>>>>>> qcom/glymur-crd.dtb: dma-controller@a00000 (qcom,glymur-gpi-dma): interrupts: [[0, 279, 4], [0, 280, 4], [0, 281, 4], [0, 282, 4], [0, 283, 4], [0, 284, 4], [0, 293, 4], [0, 294, 4], [0, 295, 4], [0, 296, 4], [0, 297, 4], [0, 298, 4], [2, 124, 4], [2, 125, 4], [2, 126, 4], [2, 127, 4]] is too long
>>>>>>         from schema $id: http://devicetree.org/schemas/dma/qcom,gpi.yaml#
>>>>>> qcom/glymur-crd.dtb: dma-controller@b00000 (qcom,glymur-gpi-dma): interrupts: [[2, 76, 4], [2, 77, 4], [2, 78, 4], [2, 79, 4], [2, 80, 4], [2, 81, 4], [2, 82, 4], [2, 83, 4], [2, 84, 4], [2, 85, 4], [2, 86, 4], [2, 87, 4], [2, 88, 4], [2, 89, 4], [2, 90, 4], [2, 91, 4]] is too long
>>>>>>         from schema $id: http://devicetree.org/schemas/dma/qcom,gpi.yaml#
>>>>>> qcom/glymur-crd.dtb: pmic@1 (qcom,pmh0101): led-controller@ee00:compatible:0: 'qcom,pmh0101-flash-led' is not one of ['qcom,pm6150l-flash-led', 'qcom,pm660l-flash-led', 'qcom,pm7550-flash-led', 'qcom,pm8150c-flash-led', 'qcom,pm8150l-flash-led', 'qcom,pm8350c-flash-led', 'qcom,pm8550-flash-led', 'qcom,pmi8998-flash-led']
>>>>>>         from schema $id: http://devicetree.org/schemas/mfd/qcom,spmi-pmic.yaml#
>>>>>> qcom/glymur-crd.dtb: pmic@1 (qcom,pmh0101): pwm:compatible: 'oneOf' conditional failed, one must be fixed:
>>>>>>         ['qcom,pmh0101-pwm', 'qcom,pm8350c-pwm'] is too long
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm660l-lpg', 'qcom,pm8150b-lpg', 'qcom,pm8150l-lpg', 'qcom,pm8350c-pwm', 'qcom,pm8916-pwm', 'qcom,pm8941-lpg', 'qcom,pm8994-lpg', 'qcom,pmc8180c-lpg', 'qcom,pmi632-lpg', 'qcom,pmi8950-pwm', 'qcom,pmi8994-lpg', 'qcom,pmi8998-lpg', 'qcom,pmk8550-pwm']
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm6150l-lpg']
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm8550-pwm']
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm8937-pwm']
>>>>>>         'qcom,pm8150l-lpg' was expected
>>>>>>         'qcom,pm8916-pwm' was expected
>>>>>>         from schema $id: http://devicetree.org/schemas/mfd/qcom,spmi-pmic.yaml#
>>>>>> qcom/glymur-crd.dtb: led-controller@ee00 (qcom,pmh0101-flash-led): compatible:0: 'qcom,pmh0101-flash-led' is not one of ['qcom,pm6150l-flash-led', 'qcom,pm660l-flash-led', 'qcom,pm7550-flash-led', 'qcom,pm8150c-flash-led', 'qcom,pm8150l-flash-led', 'qcom,pm8350c-flash-led', 'qcom,pm8550-flash-led', 'qcom,pmi8998-flash-led']
>>>>>>         from schema $id: http://devicetree.org/schemas/leds/qcom,spmi-flash-led.yaml#
>>>>>> qcom/glymur-crd.dtb: /soc@0/arbiter@c400000/spmi@c426000/pmic@1/led-controller@ee00: failed to match any schema with compatible: ['qcom,pmh0101-flash-led', 'qcom,spmi-flash-led']
>>>>>> qcom/glymur-crd.dtb: pwm (qcom,pmh0101-pwm): compatible: 'oneOf' conditional failed, one must be fixed:
>>>>>>         ['qcom,pmh0101-pwm', 'qcom,pm8350c-pwm'] is too long
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm660l-lpg', 'qcom,pm8150b-lpg', 'qcom,pm8150l-lpg', 'qcom,pm8350c-pwm', 'qcom,pm8916-pwm', 'qcom,pm8941-lpg', 'qcom,pm8994-lpg', 'qcom,pmc8180c-lpg', 'qcom,pmi632-lpg', 'qcom,pmi8950-pwm', 'qcom,pmi8994-lpg', 'qcom,pmi8998-lpg', 'qcom,pmk8550-pwm']
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm6150l-lpg']
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm8550-pwm']
>>>>>>         'qcom,pmh0101-pwm' is not one of ['qcom,pm8937-pwm']
>>>>>>         'qcom,pm8150l-lpg' was expected
>>>>>>         'qcom,pm8916-pwm' was expected
>>>>>>         from schema $id: http://devicetree.org/schemas/leds/leds-qcom-lpg.yaml#
>>>>>> qcom/glymur-crd.dtb: /soc@0/arbiter@c400000/spmi@c426000/pmic@1/pwm: failed to match any schema with compatible: ['qcom,pmh0101-pwm', 'qcom,pm8350c-pwm']
>>>>>>
>>>>>> So, we're still missing a few dependencies.
>>>>>>
>>>>>>
>>>>>> Booting the system I get a ton of errors from PCIe in the kernel log:
>>>>>>
>>>>>> debugfs: 'opp:5000000' already exists in 'soc@0-1c00000.pci'
>>>>>>
>>>>>> # dmesg | grep -E 'debugfs: .+ already exists' |wc -l
>>>>>> 508
>>>>>>
>>>>>> The system does eventually boot, and I was happy to see that we do end
>>>>>> up finding the PCIe devices after all.
>>>>>>
>>>>> I enabled dynamic debug logs and observed that each PCIe platform device
>>>>> probe was deferred approximately 10 times. The probe deferrals resulted in
>>>>> additional OPP debugfs warnings being printed.
>>>>>
>>>>> The PCIe platform device probe was deferred because the PHY driver was not
>>>>> ready - either because the PHY driver was not yet loaded, or because the
>>>>> PHY driver's own probe was also deferred due to its dependency (e.g.,
>>>>> 1fd5000.clock-controller) not being ready. This is normal behavior,
>>>>> correct? I also observed that other driver probes were deferred.
>>>>>
>>>>> But I'm not sure why there are more than 300 times probe deferrals on
>>>>> your setup.
>>>>
>>>> I think Bjorn is trying to say that the driver is wrong, because it
>>>> effectively seems to call devm_pm_opp_of_add_table repeatedly
>>>>
>>> Okay, to avoid PCIe driver probe deferrals and the resulting increased OPP
>>> debugfs warnings caused by these deferrals, we plan to move the PHY
>>> properties back from the root port node to the controller device tree
>>> node.
>>
>> Would (roughly) this solve your problems without messing with the DT?
> 
> This change cannot fix the OPP warning. The warning occurs because the OPP
> subsystem creates debugfs nodes using "op-hz" as the name, which is not
> unique for PCIe OPP tables. Mani posted a patch to fix this issue:
> https://lore.kernel.org/all/20260130071940.6949-1-manivannan.sadhasivam@oss.qualcomm.com/
> 
> Our goal is to prevent probe deferrals from occurring in our driver.

Right, I would assume that was previously aided by devlink taking
into account 'phys', but since they're no longer part of the PCIe
device node directly, that logic fails.

Still, modifying the DT to fit Linux behavior generally indicated that
Linux is not super good at doing that particular thing.. In this
instance I think we should extend drivers/of/property.c to maybe check
for supplier dependencies under subnodes of nodes that have
device_type="pci"?

Konrad