| Message ID | 20220804151001.23612-1-johan+linaro@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | usb: dwc3: qcom: fix wakeup implementation | expand |
On Thu, Aug 04, 2022 at 05:09:52PM +0200, Johan Hovold wrote: > Changes in v2 > - add review and ack tags > - fix a gadget-only build breakage (new patch) > - fix a use-after-free on wakeup from runtime suspend (new patch) > - disable wakeup completely instead of falling back to the > "disconnected" host configuration when not acting as host > - disallow 'wakeup-source' in child node in the binding Forgot to mention that I also dropped the revert of the dt-binding commit adding 'wakeup-source' to the core node as Rob suggested. Johan
This series fixes some of the fallout after the recently merged series that added wakeup support to the Qualcomm dwc3 driver: https://lore.kernel.org/all/1655094654-24052-1-git-send-email-quic_kriskura@quicinc.com/ The first patch fixes a long standing PHY power sequencing issue in dwc3 core. The second patch reverts a power-domain hack that was added by the above series. There are other ways to implement this which doesn't violate the genpd interface, and if for some reason those are not sufficient, the genpd implementation needs to be extended, not hacked around. The third patch fixes a build breakage caused by the same series. (new in v2) The fourth patch fixes another long-standing bug which could lead to a use-after-free when using runtime PM. (new in v2) The next patch fixes another issue in the Qualcomm dwc3 implementation that has been added a while back and which breaks runtime PM. The sixt patch fixes a NULL-pointer dereference or use-after-free when suspending controllers in peripheral or OTG mode due to a hack that was added to suspend path. Unfortunately, it seems the hack needs to stay for now if we want functioning suspend on some Qualcomm platforms. The remaining patches moves the wakeup-source property over from the core node to the glue node in the binding and instead propagates the wakeup capability to the former during probe. Note that this incidentally also avoids adding probe-deferral hacks to the driver as was recently proposed to deal with another problem with the current implementation: https://lore.kernel.org/all/1657891312-21748-1-git-send-email-quic_kriskura@quicinc.com/ With this series I have functioning USB system suspend and wakeup as well as somewhat functioning runtime PM in host mode on sc8280xp. Note that the PHYs apparently do not not to be left enabled for wakeup on this platform. Some issues remain such as that root-hub connect/disconnect events cannot selectively be disabled. And of course, the suspend speed hack needs to be replaced at some point but that likely requires some more heavy lifting in the dwc3 implementation. Johan Changes in v2 - add review and ack tags - fix a gadget-only build breakage (new patch) - fix a use-after-free on wakeup from runtime suspend (new patch) - disable wakeup completely instead of falling back to the "disconnected" host configuration when not acting as host - disallow 'wakeup-source' in child node in the binding Johan Hovold (9): usb: dwc3: fix PHY disable sequence Revert "usb: dwc3: qcom: Keep power domain on to retain controller status" usb: dwc3: qcom: fix gadget-only builds usb: dwc3: qcom: fix use-after-free on runtime-PM wakeup usb: dwc3: qcom: fix runtime PM wakeup usb: dwc3: qcom: fix peripheral and OTG suspend dt-bindings: usb: qcom,dwc3: add wakeup-source property usb: dwc3: qcom: fix wakeup implementation usb: dwc3: qcom: clean up suspend callbacks .../devicetree/bindings/usb/qcom,dwc3.yaml | 5 + drivers/usb/dwc3/core.c | 24 ++--- drivers/usb/dwc3/dwc3-qcom.c | 96 +++++++++++-------- drivers/usb/dwc3/host.c | 1 + 4 files changed, 76 insertions(+), 50 deletions(-)