Hi Anurag On Fri, 2019-05-03 at 07:34 +0000, Anurag Kumar Vulisha wrote: > Hi Claus, > Thanks for testing and voting for this patch. I have first tested your patches today. My test setup is a ZynqMP device, running kernel 4.14 (Xilinx version) with your patches. I then create an overlay for the devicetree with the new parameters, and unbind/bind the dwc3 driver. Next I have a host running Windows 10 and a MacBook pro with Type-C ports. For logging the communication I use a Total Phase Beagle USB3 5000 V2 analyzer. The test showed that OS-X does as expected. When BOS descriptor (bU1DevExtLat and bU2DevExtLat) returns 0, it does not enable LPM. Windows 10 on the other hand does not, and even though it received 0 as bU1DevExtLat and bU2DevExtLat it send Set Sel with U1SEL 85 us, U1PEL 0 us, U2SEL 85 us and U2PEL 0 us. Next the Windows 10 host sends the U1 Enable and the U2 enable as Set Device Feature, resulting in the system entering U2. > > > > Just today I was making another solution for this feature, using > > the > > configfs instead of the devicetree. Though thinks your solution is > > better, as it uses the U1DevExitLat and U2DevExitLat instead. I > > just > > added my solution to the bottem of the mail for reference. > > > > [1] https://www.spinics.net/lists/linux-usb/msg179393.html > > > Your approach below is also good, but you are just avoiding the > gadget dwc3 > controller from entering into U1 and U2 states by disabling the > ACCEPTU1ENA > and ACCEPTU2ENA bits in DCTL but not preventing the host from sending > the > LG0_U1 and LGO_U2 link command signaling to the gadget. The host will > keep > on trying to get the link into U1 or U2 by sending LGO_U1 or LGO_U2 > and the > gadget rejects these signals by sending LXU link command. To avoid > this extra > overhead I thought that sending zero value in the BOS descriptor's > U1DevExitLat and U2DevExitLat fields would be the best option. Host > on seeing > U 1 & U2 Exit Latencies doesn't initiate LPM U1 and U2 commands. > > Thanks, > Anurag Kumar Vulisha Correct that it does not prevent the host from sending LG0_U1 and LG0_U2, and there is your solution better on hosts using the BOS descriptor for disabling LPM. So based on my test with Windows 10, I think we should combine the solutions. To prevent LG0_U1/LG0_U2 when possible and still being able to completely disable U1/U2. Regarding interface for controlling it. I am very novice regarding Linux kernel development, but would think the BOS descriptor control would be better from a configfs interface then devicetree as I don't see BOS descriptor as hardware specific. I am more in doubt about the forcing of U1/U2 as I did with setting hardware registers, as it control hardware registers. So will like to hear from other more experienced developers. Regards Claus
