Re: [PATCH v5 0/2] phy: qcom: qmp-pcie: Add PCIe PHY no_csr reset support

From: Konrad Dybcio
Date: Fri Mar 14 2025 - 18:42:53 EST


On 3/14/25 3:54 PM, Manivannan Sadhasivam wrote:
> On Tue, Mar 11, 2025 at 01:29:15AM +0530, Vinod Koul wrote:
>> On 10-03-25, 16:58, Wenbin Yao (Consultant) wrote:
>>> On 2/26/2025 6:35 PM, Wenbin Yao wrote:
>>>> The series aims to skip phy register programming and drive PCIe PHY with
>>>> register setting programmed in bootloader by simply toggling no_csr reset,
>>>> which once togglled, PHY hardware will be reset while PHY registers are
>>>> retained.
>>>>
>>>> First, determine whether PHY setting can be skipped by checking
>>>> QPHY_START_CTRL register and the existence of nocsr reset. If it is
>>>> programmed and no_csr reset is supported, do no_csr reset and skip BCR
>>>> reset which will reset entire PHY.
>>>>
>>>> This series also remove has_nocsr_reset flag in qmp_phy_cfg structure and
>>>> decide whether the PHY supports nocsr reset by checking the existence of
>>>> nocsr reset in device tree.
>>>>
>>>> The series are tested on X1E80100-QCP and HDK8550.
>>>>
>>>> The commit messages of this patchset have been modified based on comments
>>>> and suggestions.
>>>>
>>>> Changes in v5:
>>>> - Add a check whether the init sequences are exist if the PHY needs to be
>>>> initialized to Patch 2/2.
>>>> - Link to v4: https://lore.kernel.org/all/20250220102253.755116-1-quic_wenbyao@xxxxxxxxxxx/
>>>>
>>>> Changes in v4:
>>>> - Add Philipp's Reviewed-by tag to Patch 1/2.
>>>> - Use PHY instead of phy in comments in Patch 2/2.
>>>> - Use "if (qmp->nocsr_reset)" instead of "if (!qmp->nocsr_reset)" in
>>>> function qmp_pcie_exit for readability in Patch 2/2.
>>>> - Use goto statements in function qmp_pcie_power_on and qmp_pcie_power_off
>>>> for readability in Patch 2/2.
>>>> - Refine the comment of why not checking qmp->skip_init when reset PHY in
>>>> function qmp_pcie_power_off in Patch 2/2.
>>>> - Link to v3: https://lore.kernel.org/all/20250214104539.281846-1-quic_wenbyao@xxxxxxxxxxx/
>>>>
>>>> Changes in v3:
>>>> - Replace devm_reset_control_get_exclusive with
>>>> devm_reset_control_get_optional_exclusive when get phy_nocsr reset
>>>> control in Patch 1/2.
>>>> - Do not ignore -EINVAL when get phy_nocsr reset control in Patch 1/2.
>>>> - Replace phy_initialized with skip_init in struct qmp_pcie in Patch 2/2.
>>>> - Add a comment to why not check qmp->skip_init in function
>>>> qmp_pcie_power_off in Patch 2/2.
>>>> - Link to v2: https://lore.kernel.org/all/20250211094231.1813558-1-quic_wenbyao@xxxxxxxxxxx/
>>>>
>>>> Changes in v2:
>>>> - Add Abel's and Manivannan's Reviewed-by tag to Patch 1/2.
>>>> - Refine commit msg of Patch 2/2.
>>>> - Link to v1: https://lore.kernel.org/all/20250121094140.4006801-1-quic_wenbyao@xxxxxxxxxxx/
>>>>
>>>> Konrad Dybcio (1):
>>>> phy: qcom: pcie: Determine has_nocsr_reset dynamically
>>>>
>>>> Qiang Yu (1):
>>>> phy: qcom: qmp-pcie: Add PHY register retention support
>>>>
>>>> drivers/phy/qualcomm/phy-qcom-qmp-pcie.c | 86 +++++++++++++++++-------
>>>> 1 file changed, 63 insertions(+), 23 deletions(-)
>>>>
>>>>
>>>> base-commit: bcf2acd8f64b0a5783deeeb5fd70c6163ec5acd7
>>>
>>> Hi, do you have any futher comments?
>>
>> Patches lgtm, It would be great if this was tested by someone as well...
>> Abel, Stephan, Neil can you folks test this and provide T-B
>>
>
> I tested the previous version and it worked well on X1P40100-CRD. Will give this
> version a go and give my tag.
>
>> I am also concerned about bootloader assumptions esp if the Qcom boot
>> chain is skipped

If any major part of this boot chain is skipped, much of the platform will
unfortunately not behave as expected by the kernel today, anyway. While I
personally enjoy hacking on such things, holding back hw support because of
theoretical issues is not going to help here, especially since the tables
can be trivially added at a later point, if ever needed, without requiring
DT changes.

> In that case, someone should add the PHY init sequence to the driver. That's why
> I wanted to have the check in place to avoid silently failing PHY
> initialization. Right now, the driver will error out if there is no init
> sequence available.

IIUC we internally settled on not sending the sequences for X1P4, since
the PHYs are initialized as expected.

Konrad