Hi, claus.stovgaard@xxxxxxxxx wrote: > Hi > > I have earlier worked with the possibility for disabling U1/U2 to solve > an issue where the dwc3 acting as a gadget device sometimes failing to > leave U2. > > Analyzing the situation when the dwc3 failing to leave U2, it happens > when the link is in U2 after timeout from U1, and when the device wants > to leave U2. Not when the host wants to leave U2. > > Going back the BOS descriptor from the device reports > > bU1DevExitLat 1us > bU2DevExitLat 500us > > And the SetSystemExitLatency control transfer ends with > > U1 System Exit Latency 86 us > U1 Device to Host Exit Latency 1 us > U2 System Exit Latency 585 us > U2 Device to Host Exit Latency 500 us > > Looking at the length of LFPS etc. it seems that the U2 exit is just > above 80 us,and it start link training just after. > > So I was wandering how the exit latency of the system is communicated > to dwc3 core, and found the following code from ep0.c > dwc3_ep0_set_sel_cmpl > > ---- > reg = dwc3_readl(dwc->regs, DWC3_DCTL); > if (reg & DWC3_DCTL_INITU2ENA) > param = dwc->u2pel; > if (reg & DWC3_DCTL_INITU1ENA) > param = dwc->u1pel; This is incorrect. If the controller is enabled for both U1 and U2, then periodic param is U2PEL. If the controller is only enabled for U1, then U1PEL is used. Probably the original author intended but missed the "else" on the second if case. > /* > * According to Synopsys Databook, if parameter is > * greater than 125, a value of zero should be > * programmed in the register. > */ > if (param > 125) > param = 0; > > /* now that we have the time, issue DGCMD Set Sel */ > ret = dwc3_send_gadget_generic_command(dwc, > DWC3_DGCMD_SET_PERIODIC_PAR, param); > WARN_ON(ret < 0); > ---- > > I don't have access to Synopsys Databook, so I am puzzled about > the DWC3_DGCMD_SET_PERIODIC_PAR command. The code favor to use the > device to host exit latency, and use the u1 if present over the u2. > > So the dwc3 core never get the system exit latency times, and they just > disapeer. > > I hope that someone have access to the Databook, and is able to share > some details about the set periodic command and the parameter. > According to the databook, currently the controller doesn't use these programmed values. It uses the value from CoreConsultant configuration setting. BR, Thinh
