Re: [PATCH v13 05/10] usb: dwc3: qcom: Refactor IRQ handling in QCOM Glue driver

From: Krishna Kurapati PSSNV
Date: Tue Nov 07 2023 - 03:30:00 EST


But there is already support for these interrupts in the driver. You
work for Qualcomm who built the thing so surely you can figure how they
intended these to be used?

You need to provide this information so that we can determine what the
binding should look like. The implementation would also be simplified if
we don't have to add random hacks to it just because we don't know why
the vendor driver you refer does not use it currently on this particular
platform.


Hi Johan,

Regarding the points of discussion we had last week on [1], here are some clarifications:

1. We do have hs_phy_irq 1/2/3/4 for tertiary port of Sc8280 as mentioned. Why do we need them and would we use it in multiport targets ?

DPSE and DMSE are single ended line state of DP and DM lines. The DP line and DM line stay in steady High or Low during suspend and they flip when there is a RESUME or REMOTE WAKE. This is what we do/check in dwc3_qcom_enable_interrupts call for dp/dm irq's based on usb2_speed.

Initially in QUSB2 targets, the interrupts were enabled and configured in phy and the wakeup was interrupt was read on hs_phy_irq vector - [2].
In that case, we modify DP/DM interrupts in phy registers, specifically QUSB2PHY_INTR_CTRL and when wakeup signal comes in, hs_phy_irq is triggered. But in femto targets, this is done via DP/DM interrupts and there is no use of hs_phy_irq. Even hw folks confirmed they dont use hs_ph_irq in femto phy targets.

As an experiment, I tried to test wakeup by pressing buttons on connected keyboard when in suspend state or connecting/disconnecting keyboard in suspended state on different ports and only see dp/dm IRQ's getting fired although we register for hs_phy_irq as well:

/ # cat /proc/interrupts  |grep phy_
171:   1  0   0   0  0  0  0  0       PDC 127 Edge      dp_hs_phy_1
172:   2  0   0   0  0  0  0  0       PDC 126 Edge      dm_hs_phy_1
173:   3  0   0   0  0  0  0  0       PDC 129 Edge      dp_hs_phy_2
174:   4  0   0   0  0  0  0  0       PDC 128 Edge      dm_hs_phy_2
175:   0  0   0   0  0  0  0  0       PDC 131 Edge      dp_hs_phy_3
176:   2  0   0   0  0  0  0  0       PDC 130 Edge      dm_hs_phy_3
177:   2  0   0   0  0  0  0  0       PDC 133 Edge      dp_hs_phy_4
178:   5  0   0   0  0  0  0  0       PDC 132 Edge      dm_hs_phy_4
179:   0  0   0   0  0  0  0  0       PDC  16 Level     ss_phy_1
180:   0  0   0   0  0  0  0  0       PDC  17 Level     ss_phy_2
181:   0  0   0   0  0  0  0  0     GICv3 163 Level     hs_phy_1
182:   0  0   0   0  0  0  0  0     GICv3 168 Level     hs_phy_2
183:   0  0   0   0  0  0  0  0     GICv3 892 Level     hs_phy_3
184:   0  0   0   0  0  0  0  0     GICv3 891 Level     hs_phy_4

Since the hs_phy_irq is applicable only for qusb2 targets, do we still need to add it to DT.

2. BAM Irq usage (u_usb31_scnd_mvs_pipe_wrapper_usb31_bam_irq[0]):

BAM IRQ is not needed in host-only controller. It was just added in process of porting/deriving code from DRD controllers and is non-functional (confirmed by HW team here). We can skip this from DT of multiport.

3. ctrl_irq[1] usage:

This is a feature of SNPS controller, not qcom glue wrapper, and is present on all targets (non-QC as well probably). As mentioned before on [3], this is used for HW acceleration.

In host mode, XHCI spec does allow for multiple interrupters when multiple event rings are used. A possible usage is multiple execution environments something like what we are doing on mobile with ADSP audio offload [4]. Another possibility could be some of virtualization where host/hyp would manage the first interrupter and could allow a guest to operate only with the second (though current design does not go far enough to offer true isolation for real VM type workloads). The additional interrupts (ones other than ctrl_irq[0]) are either for virtualization use cases, or for our various “hw offload” features. In device mode, these are used for offloading tethering functionality to IPA FW.

Since the DeviceTree passed to the OS, should describe the hardware to the OS, and should represent the hardware from the point-of-view of the OS, adding one interrupt (ctrl_irq[0]) might be sufficient as Linux would not use the other interrupts. Furthermore AFAIK even UEFI/Windows also use only ctrl_irq[0] for host mode in their execution environment today. Do we still need to add this to bindings and DT ?

[1]: https://lore.kernel.org/all/ZTJ_T1UL8-s2cgNz@xxxxxxxxxxxxxxxxxxxx/
[2]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/phy/qualcomm/phy-qcom-qusb2.c?h=v6.6#n626
[3]: https://lore.kernel.org/all/ZTduh5LULBMYf3wq@xxxxxxxxxxxxxxxxxxxx/
[4]: https://lore.kernel.org/all/20231017200109.11407-1-quic_wcheng@xxxxxxxxxxx/


Hi Johan,

Can you help provide your comments on the above mentioned points so that we can take the discussion forward and finalize changes for v14 patches. And thanks for all the reviews so far on previous revisions.

Regards,
Krishna,