| Message ID | 1446236275-12698-2-git-send-email-dianders@chromium.org |
|---|---|
| State | Changes Requested, archived |
| Headers | show |
On Fri, Oct 30, 2015 at 3:17 PM, Douglas Anderson <dianders@chromium.org> wrote: > From: Doug Anderson <dianders@chromium.org> > > On the rk3288 USB host-only port (the one that's not the OTG-enabled > port) the PHY can get into a bad state when a wakeup is asserted (not > just a wakeup from full system suspend but also a wakeup from > autosuspend). I would think re-enumerating means autosuspend is broken. > We can get the PHY out of its bad state by asserting its "port reset", > but unfortunately that seems to assert a reset onto the USB bus so it > could confuse things if we don't actually deenumerate / reenumerate the > device. > > We can also get the PHY out of its bad state by fully resetting it using > the reset from the CRU (clock reset unit), which does a more full > reset. The CRU-based reset appears to actually cause devices on the bus > to be removed and reinserted, which fixes the problem (albeit in a hacky > way). The reset from the CRU goes to the PHY, correct? Therefore, the binding should reflect that. Connecting it to the host controller is a hack. So describe the reset connection properly and then add a .phy_reset() hook to the phy subsystem. Then call that when flag property you added is set. Rob -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Rob, On Mon, Nov 2, 2015 at 8:12 AM, Rob Herring <robh+dt@kernel.org> wrote: > On Fri, Oct 30, 2015 at 3:17 PM, Douglas Anderson <dianders@chromium.org> wrote: >> From: Doug Anderson <dianders@chromium.org> >> >> On the rk3288 USB host-only port (the one that's not the OTG-enabled >> port) the PHY can get into a bad state when a wakeup is asserted (not >> just a wakeup from full system suspend but also a wakeup from >> autosuspend). > > I would think re-enumerating means autosuspend is broken. Yes, it is broken on this port. I've been trying to figure out how to disable autosuspend at the driver level. Until then we can do it with udev rules but that's much less ideal. On our current system we have autosuispend turned off for most things via udev and laptop-mode-tools. ...so really the only case we see this reset happen is if an empty hub is plugged in and then you plug something into the hub. The reset isn't terrible and is much better than the device getting stuck. You can also see the reset doing its thing if you fully suspend the system and try to wake up from a keyboard or a mouse attached to the port. Being able to wake up (and then seeing a reset) is still better than not being able to wake up. >> We can get the PHY out of its bad state by asserting its "port reset", >> but unfortunately that seems to assert a reset onto the USB bus so it >> could confuse things if we don't actually deenumerate / reenumerate the >> device. >> >> We can also get the PHY out of its bad state by fully resetting it using >> the reset from the CRU (clock reset unit), which does a more full >> reset. The CRU-based reset appears to actually cause devices on the bus >> to be removed and reinserted, which fixes the problem (albeit in a hacky >> way). > > The reset from the CRU goes to the PHY, correct? Therefore, the > binding should reflect that. Connecting it to the host controller is a > hack. > > So describe the reset connection properly and then add a .phy_reset() > hook to the phy subsystem. Then call that when flag property you added > is set. As per previous email, I disagree. The fact that there may be more than one reset exposed from the PHY and that there is not a tightly coupled relationship between a PHY driver and a USB driver means that we would need to re-invent the reset API on top of the PHY API. In my case I am exposing a single reset at the moment, but there may be reasons to expose additional "PHY" resets in the future. http://www.gossamer-threads.com/lists/linux/kernel/2289780#2289780 ...using the existing reset API seems better. -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Nov 2, 2015 at 10:22 AM, Doug Anderson <dianders@chromium.org> wrote: > Rob, > > On Mon, Nov 2, 2015 at 8:12 AM, Rob Herring <robh+dt@kernel.org> wrote: >> On Fri, Oct 30, 2015 at 3:17 PM, Douglas Anderson <dianders@chromium.org> wrote: >>> From: Doug Anderson <dianders@chromium.org> >>> We can get the PHY out of its bad state by asserting its "port reset", >>> but unfortunately that seems to assert a reset onto the USB bus so it >>> could confuse things if we don't actually deenumerate / reenumerate the >>> device. >>> >>> We can also get the PHY out of its bad state by fully resetting it using >>> the reset from the CRU (clock reset unit), which does a more full >>> reset. The CRU-based reset appears to actually cause devices on the bus >>> to be removed and reinserted, which fixes the problem (albeit in a hacky >>> way). >> >> The reset from the CRU goes to the PHY, correct? Therefore, the >> binding should reflect that. Connecting it to the host controller is a >> hack. >> >> So describe the reset connection properly and then add a .phy_reset() >> hook to the phy subsystem. Then call that when flag property you added >> is set. > > As per previous email, I disagree. The fact that there may be more > than one reset exposed from the PHY and that there is not a tightly > coupled relationship between a PHY driver and a USB driver means that > we would need to re-invent the reset API on top of the PHY API. In my > case I am exposing a single reset at the moment, but there may be > reasons to expose additional "PHY" resets in the future. Your previous approach was completely different. I was not arguing that the PHY driving reset to the host was wrong use of reset binding, just that it was an overkill. Now, it is just flat wrong unless you convince me that SRST_USBHOST1_PHY is a connection to the host rather than the PHY. Rob -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi, On Mon, Nov 2, 2015 at 9:16 AM, Rob Herring <robh+dt@kernel.org> wrote: > On Mon, Nov 2, 2015 at 10:22 AM, Doug Anderson <dianders@chromium.org> wrote: >> Rob, >> >> On Mon, Nov 2, 2015 at 8:12 AM, Rob Herring <robh+dt@kernel.org> wrote: >>> On Fri, Oct 30, 2015 at 3:17 PM, Douglas Anderson <dianders@chromium.org> wrote: >>>> From: Doug Anderson <dianders@chromium.org> > >>>> We can get the PHY out of its bad state by asserting its "port reset", >>>> but unfortunately that seems to assert a reset onto the USB bus so it >>>> could confuse things if we don't actually deenumerate / reenumerate the >>>> device. >>>> >>>> We can also get the PHY out of its bad state by fully resetting it using >>>> the reset from the CRU (clock reset unit), which does a more full >>>> reset. The CRU-based reset appears to actually cause devices on the bus >>>> to be removed and reinserted, which fixes the problem (albeit in a hacky >>>> way). >>> >>> The reset from the CRU goes to the PHY, correct? Therefore, the >>> binding should reflect that. Connecting it to the host controller is a >>> hack. >>> >>> So describe the reset connection properly and then add a .phy_reset() >>> hook to the phy subsystem. Then call that when flag property you added >>> is set. >> >> As per previous email, I disagree. The fact that there may be more >> than one reset exposed from the PHY and that there is not a tightly >> coupled relationship between a PHY driver and a USB driver means that >> we would need to re-invent the reset API on top of the PHY API. In my >> case I am exposing a single reset at the moment, but there may be >> reasons to expose additional "PHY" resets in the future. > > Your previous approach was completely different. I was not arguing > that the PHY driving reset to the host was wrong use of reset binding, > just that it was an overkill. Now, it is just flat wrong unless you > convince me that SRST_USBHOST1_PHY is a connection to the host rather > than the PHY. It still seems to me that the argument that there can be multiple types of "phy reset" is still a sane argument. Just because we currently happen to need the one that resets the whole PHY now doesn't mean that we won't need other different resets later. ...and if you're exposing multiple different resets, then the reset API is the best one to use. I _could_ see the argument that perhaps the reset should pass through the PHY driver so that it could possibly in the future be aware of the reset. That is: the SRST could use the reset API to expose the PHY reset to the PHY driver and then the PHY driver could in turn expose the PHY reset (using the reset API) to the dwc2 driver. At the moment this would be a complete pass through, but structuring it that way would later allow the PHY to do something with the reset. How about that? -Doug -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi, the implementation of Doug is thought quick but the Herring want to use this problem in autosuspend. Unfortunately, It is a fault in hardware, onyl would happened at Rockchip RK3288 now. So I suggest to accept this idea. It may be better if dropped the dts, as it is an issue only for RK3288, just hard-code to source code. If maintainer would agree with the later one, I would send the other version to the list. Thank you for using of Rockchip. Doug Anderson (2): usb: dwc2: optionally assert phy "full reset" when waking up ARM: dts: rockchip: Point rk3288 dwc2 usb at the full PHY reset Documentation/devicetree/bindings/usb/dwc2.txt | 7 +++++++ arch/arm/boot/dts/rk3288.dtsi | 5 +++++ drivers/usb/dwc2/core.h | 5 +++++ drivers/usb/dwc2/core_intr.c | 14 ++++++++++++++ drivers/usb/dwc2/platform.c | 13 +++++++++++++ 5 files changed, 44 insertions(+)
diff --git a/Documentation/devicetree/bindings/usb/dwc2.txt b/Documentation/devicetree/bindings/usb/dwc2.txt index fd132cbee70e..0de78c48c12d 100644 --- a/Documentation/devicetree/bindings/usb/dwc2.txt +++ b/Documentation/devicetree/bindings/usb/dwc2.txt @@ -17,6 +17,13 @@ Refer to clk/clock-bindings.txt for generic clock consumer properties Optional properties: - phys: phy provider specifier - phy-names: shall be "usb2-phy" +- snps,need-phy-full-reset-on-wake: if present indicates that we need to reset + the PHY when we detect a wakeup due to a hardware errata. If present you + must specify a "phy-full-reset" reset. + +Resets: +- phy-full-reset (optional): Fully resets the PHY. + Refer to phy/phy-bindings.txt for generic phy consumer properties - dr_mode: shall be one of "host", "peripheral" and "otg" Refer to usb/generic.txt diff --git a/drivers/usb/dwc2/core.h b/drivers/usb/dwc2/core.h index a66d3cb62b65..ce1ef961ae6c 100644 --- a/drivers/usb/dwc2/core.h +++ b/drivers/usb/dwc2/core.h @@ -582,8 +582,11 @@ struct dwc2_hregs_backup { * @hcd_enabled Host mode sub-driver initialization indicator. * @gadget_enabled Peripheral mode sub-driver initialization indicator. * @ll_hw_enabled Status of low-level hardware resources. + * @need_phy_full_reset_on_wake: Quirk saying that we should assert + * phy_full_reset on a remote wakeup. * @phy: The otg phy transceiver structure for phy control. * @uphy: The otg phy transceiver structure for old USB phy control. + * @phy_full_reset: Reset control for the PHY's "full reset". * @plat: The platform specific configuration data. This can be removed once * all SoCs support usb transceiver. * @supplies: Definition of USB power supplies @@ -710,9 +713,11 @@ struct dwc2_hsotg { unsigned int hcd_enabled:1; unsigned int gadget_enabled:1; unsigned int ll_hw_enabled:1; + unsigned int need_phy_full_reset_on_wake:1; struct phy *phy; struct usb_phy *uphy; + struct reset_control *phy_full_reset; struct dwc2_hsotg_plat *plat; struct regulator_bulk_data supplies[ARRAY_SIZE(dwc2_hsotg_supply_names)]; u32 phyif; diff --git a/drivers/usb/dwc2/core_intr.c b/drivers/usb/dwc2/core_intr.c index 27daa42788b1..275484260192 100644 --- a/drivers/usb/dwc2/core_intr.c +++ b/drivers/usb/dwc2/core_intr.c @@ -45,6 +45,7 @@ #include <linux/dma-mapping.h> #include <linux/io.h> #include <linux/slab.h> +#include <linux/reset.h> #include <linux/usb.h> #include <linux/usb/hcd.h> @@ -378,6 +379,19 @@ static void dwc2_handle_wakeup_detected_intr(struct dwc2_hsotg *hsotg) /* Restart the Phy Clock */ pcgcctl &= ~PCGCTL_STOPPCLK; dwc2_writel(pcgcctl, hsotg->regs + PCGCTL); + + /* + * If we've got this quirk then the PHY is stuck upon + * wakeup. Assert reset. This will propagate out and + * eventually we'll re-enumerate the device. Not great + * but the best we can do. + */ + if (hsotg->need_phy_full_reset_on_wake) { + reset_control_assert(hsotg->phy_full_reset); + udelay(50); + reset_control_deassert(hsotg->phy_full_reset); + } + mod_timer(&hsotg->wkp_timer, jiffies + msecs_to_jiffies(71)); } else { diff --git a/drivers/usb/dwc2/platform.c b/drivers/usb/dwc2/platform.c index 5859b0fa19ee..1379b9cb0d39 100644 --- a/drivers/usb/dwc2/platform.c +++ b/drivers/usb/dwc2/platform.c @@ -45,6 +45,7 @@ #include <linux/platform_device.h> #include <linux/phy/phy.h> #include <linux/platform_data/s3c-hsotg.h> +#include <linux/reset.h> #include <linux/usb/of.h> @@ -376,6 +377,18 @@ static int dwc2_driver_probe(struct platform_device *dev) "Configuration mismatch. Forcing peripheral mode\n"); } + hsotg->need_phy_full_reset_on_wake = + of_property_read_bool(dev->dev.of_node, + "snps,need-phy-full-reset-on-wake"); + hsotg->phy_full_reset = devm_reset_control_get(hsotg->dev, + "phy-full-reset"); + if (IS_ERR(hsotg->phy_full_reset) && + hsotg->need_phy_full_reset_on_wake) { + dev_warn(hsotg->dev, "Missing phy full reset (%ld); skipping\n", + PTR_ERR(hsotg->phy_full_reset)); + hsotg->need_phy_full_reset_on_wake = false; + } + retval = dwc2_lowlevel_hw_init(hsotg); if (retval) return retval;