Hi Crane, I've been testing this patch, and running into some problems with an external high speed hub and a mouse. I plug the hub into the xHCI host, load the driver, wait for the hub to auto-suspend, and then plug in the mouse. What I see in dmesg is: Apr 5 15:22:20 xanatos kernel: [ 5107.840065] xhci_hcd 0000:05:00.0: Suspended device on port 2 Apr 5 15:22:20 xanatos kernel: [ 5107.840094] usb 1-3: usb auto-suspend Apr 5 15:22:36 xanatos kernel: [ 5124.005232] xhci_hcd 0000:05:00.0: op reg status = 00000018 Apr 5 15:22:36 xanatos kernel: [ 5124.005241] xhci_hcd 0000:05:00.0: ir set irq_pending = 00000003 Apr 5 15:22:36 xanatos kernel: [ 5124.005246] xhci_hcd 0000:05:00.0: Event ring dequeue ptr: Apr 5 15:22:36 xanatos kernel: [ 5124.005254] xhci_hcd 0000:05:00.0: @153c87d0 03000000 00000000 01000000 00008801 Apr 5 15:22:36 xanatos kernel: [ 5124.005265] xhci_hcd 0000:05:00.0: `MEM_WRITE_DWORD(3'b000, 32'hffffc90006534024, 32'h18, 4'hf); Apr 5 15:22:36 xanatos kernel: [ 5124.005276] xhci_hcd 0000:05:00.0: `MEM_WRITE_DWORD(3'b000, 32'hffffc90006534620, 32'h3, 4'hf); Apr 5 15:22:36 xanatos kernel: [ 5124.005287] xhci_hcd 0000:05:00.0: In xhci_handle_event Apr 5 15:22:36 xanatos kernel: [ 5124.005293] xhci_hcd 0000:05:00.0: xhci_handle_event - OS owns TRB Apr 5 15:22:36 xanatos kernel: [ 5124.005299] xhci_hcd 0000:05:00.0: xhci_handle_event - calling handle_port_status Apr 5 15:22:36 xanatos kernel: [ 5124.005305] xhci_hcd 0000:05:00.0: Port Status Change Event for port 3 Apr 5 15:22:36 xanatos kernel: [ 5124.005312] xhci_hcd 0000:05:00.0: Event ring deq = 0x153c87e0 (DMA) Apr 5 15:22:36 xanatos kernel: [ 5124.005324] xhci_hcd 0000:05:00.0: // Write event ring dequeue pointer, preserving EHB bit Apr 5 15:22:36 xanatos kernel: [ 5124.005331] xhci_hcd 0000:05:00.0: `MEM_WRITE_DWORD(3'b000, 64'hffffc90006534638, 64'h153c87e0, 4'hf); Apr 5 15:22:36 xanatos kernel: [ 5124.005351] xhci_hcd 0000:05:00.0: xhci_handle_event - returned from handle_port_status Apr 5 15:22:36 xanatos kernel: [ 5124.005357] xhci_hcd 0000:05:00.0: In xhci_handle_event Apr 5 15:22:36 xanatos kernel: [ 5124.005370] xhci_hcd 0000:05:00.0: `MEM_WRITE_DWORD(3'b000, 64'hffffc90006534638, 64'h153c87e8, 4'hf); Apr 5 15:23:06 xanatos kernel: [ 5153.392604] xhci_hcd 0000:05:00.0: Poll event ring: 4296180644 Then, nothing. The USB core doesn't interact with the host anymore, and the mouse is never enumerated. I don't understand why, because handle_port_status() calls usb_hcd_poll_rh_status(). All I see is the debugging polling loop running. That log file is named xhci-device-suspend-waiting.log I noticed that if I ran lsusb after the polling loop has run, then I see a lot more activity (including canceled control transfers) and the mouse is eventually detected after about half a minute. That log file is called xhci-device-suspend-lsusb.log In trying to reproduce this, I discovered that if I ran lsusb quickly after the hub suspended (before the polling loop ran), I could hard-hang my system, every time. That log file is called xhci-device-suspend-lsusb-hang.log For some reason, after a device reset, the host controller's internal dequeue pointer for the default control ring gets reset to the start of the segment, and it executes all the control transfers that were previously enqueued. The xHCI driver ignores the completion events, but I think the host is trying to access URB buffers it doesn't own anymore, and crashing the system. I need to do more debugging to discover if it's software setting the control endpoint ring dequeue pointer wrong, or if it's a hardware issue. In the meantime, can you figure out why remote wakeup doesn't work with your patch? Sarah Sharp
Attachment:
logs.tar.gz
Description: Binary data
