Hi Greg, Sarah, Alan, The following patches are required to resolve remote wake issues with certain devices. Issue description: If the remote wake is issued from the device in a specific timing condition while the system is entering sleep state then it may cause system to auto wake on subsequent sleep cycle. Root cause: Host controller rebroadcasts the Resume signal > 100 µseconds after receiving the original resume event from the device. For proper function, some devices may require the rebroadcast of resume event within the USB spec of 100µS. Reproduce steps: 1. Enable remote wakeup for usb mouse. 2. Execute S3. 3. Click mouse _intensely_ (more than 10 times) to wake the system up. 4. Then execute S3 again. 5. Observe that the system auto wake up. [Q] Why the special devices are only mice? Would high speed devices such as 3G modem or USB Bluetooth adapter trigger this issue? - Current this sensitivity is only confined to devices that use Pixart controllers. This controller is designed for use with LS mouse devices only. We have not observed any other devices failing. There may be a small risk for other devices also but this patch (reset device in resume phase) will cover the cases if required. [Q] Shouldn’t the resume signal be sent within 100 us for every device? - The Host controller may not send the resume signal within 100us, this our host controller specification change. This is why we require the patch to prevent side effects on certain known devices. [Q] Why would clicking mouse INTENSELY to wake the system up trigger this issue? - This behavior is specific to the devices that use Pixart controller. It is timing dependent on when the resume event is triggered during the sleep state. [Q] Is it a host controller issue or mouse? - It is the host controller behavior during resume that triggers the device incorrect behavior on the next resume. - Patch 1 refactors AMD quirk to abstract chipset types. - Patch 2 implements AMD remote wakeup quirk. They are generated on gregkh/usb usb-next. Changes from v1 -> v2: - Add reproduce steps. - Add a patch to put the changes of core/hub.c and pci-quirks into one patch. - Remove all the split strings. - Refactor usb_amd_remote_wakeup_quirk to save a variable and less codes. - Remove 10ms delay in spinlock during xhci reset port by mouse. - Add an extra tab to algin to all quirks macros. - Remove backport request. - Add descriptions of the fixes and how they work. - Summarized questions from Sarah and Alan. Changes from v2 -> v3: - Split first patch into two, the one add all the issue USB mice with vendor id and product id in core/quirk.c, the other one is for filtering the special AMD platforms. - Remove is_usb_mouse function, as Greg required that don't use device type for filtering. Changes from v3 -> v4: - Refactor AMD pci quirk to abstract out chipset types to list generation with a enumeration type and revision as Greg's suggestion. - Remove xhci and ohci level codes. - Implement remote wakeup quirk at core level as Alan's suggestion. Thanks, Rui Huang Rui (2): usb: pci-quirks: refactor AMD quirk to abstract AMD chipset types usb: core: implement AMD remote wakeup quirk drivers/usb/core/hcd-pci.c | 3 ++ drivers/usb/core/quirks.c | 37 ++++++++++++++ drivers/usb/host/pci-quirks.c | 116 +++++++++++++++++++++++++++++++++--------- include/linux/usb/hcd.h | 3 ++ 4 files changed, 135 insertions(+), 24 deletions(-) -- 1.7.11.7 -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
