Re: [RFC PATCH v3 4/5] usb: xhci-mtk: add support runtime pm

From: Chunfeng Yun
Date: Tue Jan 05 2021 - 02:14:48 EST


Hi Ikjoon,

I try to rebuild the code flow, and want to apply the framework from
wakeirq.c

On Tue, 2020-12-29 at 15:38 +0800, Ikjoon Jang wrote:
> On Tue, Dec 22, 2020 at 5:35 PM Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx> wrote:
> >
> > From: CK Hu <ck.hu@xxxxxxxxxxxx>
> >
> > add support runtime pm feature
> >
> > Signed-off-by: Zhanyong Wang <zhanyong.wang@xxxxxxxxxxxx>
> > Signed-off-by: Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx>
> > ---
> > v3:
> > 1. fix some issues
> > 2. remove attribute files
> >
> > v2: fix error caused by request irq suggested by CK
> > ---
> > drivers/usb/host/xhci-mtk.c | 285 +++++++++++++++++++++++++++++++++++-
> > drivers/usb/host/xhci-mtk.h | 14 ++
> > 2 files changed, 294 insertions(+), 5 deletions(-)
> > mode change 100755 => 100644 drivers/usb/host/xhci-mtk.c
> >
> > diff --git a/drivers/usb/host/xhci-mtk.c b/drivers/usb/host/xhci-mtk.c
> > old mode 100755
> > new mode 100644
> > index 34bd3de090b1..c07d54acbcb7
> > --- a/drivers/usb/host/xhci-mtk.c
> > +++ b/drivers/usb/host/xhci-mtk.c
> > @@ -14,6 +14,7 @@
> > #include <linux/mfd/syscon.h>
> > #include <linux/module.h>
> > #include <linux/of.h>
> > +#include <linux/of_irq.h>
> > #include <linux/platform_device.h>
> > #include <linux/pm_runtime.h>
> > #include <linux/regmap.h>
> > @@ -77,6 +78,72 @@ enum ssusb_uwk_vers {
> > SSUSB_UWK_V3,
> > };
> >
> > +int xhci_mtk_runtime_ready;
> > +
> > +static int xhci_mtk_runtime_resume(struct device *dev);
> > +
> > +static void xhci_mtk_seal_work(struct work_struct *work)
> > +{
> > + struct xhci_hcd_mtk *mtk =
> > + container_of(work, struct xhci_hcd_mtk, seal.work);
> > + struct usb_hcd *hcd = mtk->hcd;
> > + struct xhci_hcd *xhci = hcd_to_xhci(hcd);
> > +
> > + xhci_dbg(xhci, "spm unseals xHCI controller %i\n", mtk->seal_status);
> > + if (mtk->seal_status == SEAL_SUSPENDED) {
> > + mtk->seal_status = SEAL_RESUMING;
> > + pm_runtime_mark_last_busy(mtk->dev);
> > + pm_runtime_put_sync_autosuspend(mtk->dev);
>
> If I understand correctly, this function is for waking up the device
> but this function calls put() only without get().
will use pm_request_resume();

>
> > + } else {
> > + /*
> > + * FIXME: Sometimes seal_status will keep it on SEAL_RESUMING staus not to expect
> > + * since pm_runtime_put_sync_autosuspend doesn't invoke runtime_resume of callback.
> > + * and to survey why not to invoke runtime_resume callback named
> > + * xhci_mtk_runtime_resume in time in short feature, Herely provide one solution
> > + * that make sure seal_status to match h/w state machine,and MTK xHCI clocks
> > + * on as soon as unseal event received.
>
> I guess actual resuming should be happened only from the 1st interrupt
> (when in SEAL_SUSPENDED)
> and following spurious interrupts can be just ignored.
Yes
>
> > + */
> > + if (mtk->seal_status == SEAL_RESUMING)
> > + xhci_mtk_runtime_resume(mtk->dev);
>
> xhci_mtk_runtime_resume() is defined as a runtime pm callback,
> pm core will call that callback when pm usage counter reaches to zero.
Yes
>
> > + else
> > + xhci_warn(xhci,
> > + "Ignore seal wakeup source disordered in xHCI controller\n");
> > + }
> > +}
> > +
> > +static irqreturn_t xhci_mtk_seal_irq(int irq, void *data)
> > +{
> > + struct xhci_hcd_mtk *mtk = data;
> > + struct usb_hcd *hcd = mtk->hcd;
> > + struct xhci_hcd *xhci = hcd_to_xhci(hcd);
> > +
> > + xhci_dbg(xhci, "seal irq ISR invoked\n");
> > +
> > + schedule_delayed_work(&mtk->seal, 0);
> > +
> > + return IRQ_HANDLED;
> > +}
> > +
> > +static void xhci_mtk_seal_wakeup_enable(struct xhci_hcd_mtk *mtk, bool enable)
> > +{
> > + struct irq_desc *desc;
> > + struct device *dev = mtk->dev;
> > +
> > + if (mtk && mtk->seal_irq) {
> > + desc = irq_to_desc(mtk->seal_irq);
> > + if (enable) {
> > + desc->irq_data.chip->irq_ack(&desc->irq_data);
> > + enable_irq(mtk->seal_irq);
> > + dev_dbg(dev, "%s: enable_irq %i\n",
> > + __func__, mtk->seal_irq);
> > + } else {
> > + disable_irq(mtk->seal_irq);
> > + dev_dbg(dev, "%s: disable_irq %i\n",
> > + __func__, mtk->seal_irq);
> > + }
> > + }
> > +}
> > +
>
> I think this is unnecessary if this driver can check the current state
> and ignore the spurious irqs if spm sometimes triggers the wake-up irqs.
It helps to reduce some irqs, there are many spurious interrupts, I
think there is something wrong from the point of HW, will try to find
the root cause.

>
> > static int xhci_mtk_host_enable(struct xhci_hcd_mtk *mtk)
> > {
> > struct mu3c_ippc_regs __iomem *ippc = mtk->ippc_regs;
> > @@ -347,7 +414,6 @@ static int usb_wakeup_of_property_parse(struct xhci_hcd_mtk *mtk,
> > mtk->uwk_reg_base, mtk->uwk_vers);
> >
> > return PTR_ERR_OR_ZERO(mtk->uwk);
> > -
> > }
> >
> > static void usb_wakeup_set(struct xhci_hcd_mtk *mtk, bool enable)
> > @@ -482,9 +548,11 @@ static int xhci_mtk_probe(struct platform_device *pdev)
> > return ret;
> > }
> >
> > + pm_runtime_set_active(dev);
> > + pm_runtime_use_autosuspend(dev);
> > + pm_runtime_set_autosuspend_delay(dev,
> > + CONFIG_USB_AUTOSUSPEND_DELAY * 1000);
> > pm_runtime_enable(dev);
> > - pm_runtime_get_sync(dev);
>
> The only one left pm_runtime_get() is removed by here,
> now this driver only calls pm_runtime_put().
will add it back, and forbid runtime by default
>
> > - device_enable_async_suspend(dev);
> >
> > ret = xhci_mtk_ldos_enable(mtk);
> > if (ret)
> > @@ -499,6 +567,14 @@ static int xhci_mtk_probe(struct platform_device *pdev)
> > ret = irq;
> > goto disable_clk;
> > }
> > + dev_dbg(dev, "irq %i\n", irq);
> > +
> > + mtk->seal_irq = platform_get_irq_optional(pdev, 1);
> > + if (mtk->seal_irq < 0) {
> > + ret = mtk->seal_irq;
> > + goto disable_clk;
> > + }
> > + dev_dbg(dev, "seal_irq %i\n", mtk->seal_irq);
> >
> > hcd = usb_create_hcd(driver, dev, dev_name(dev));
> > if (!hcd) {
> > @@ -565,6 +641,27 @@ static int xhci_mtk_probe(struct platform_device *pdev)
> > if (ret)
> > goto dealloc_usb2_hcd;
> >
> > + INIT_DELAYED_WORK(&mtk->seal, xhci_mtk_seal_work);
> > + snprintf(mtk->seal_descr, sizeof(mtk->seal_descr), "seal%s:usb%d",
> > + hcd->driver->description, hcd->self.busnum);
> > + ret = devm_request_irq(dev, mtk->seal_irq, &xhci_mtk_seal_irq,
> > + IRQF_TRIGGER_FALLING, mtk->seal_descr, mtk);
> > + if (ret != 0) {
> > + dev_err(dev, "seal request interrupt %d failed\n",
> > + mtk->seal_irq);
> > + goto dealloc_usb2_hcd;
> > + }
> > + xhci_mtk_seal_wakeup_enable(mtk, false);
> > +
> > + device_enable_async_suspend(dev);
> > + xhci_mtk_runtime_ready = 1;
> > +
> > + pm_runtime_mark_last_busy(dev);
> > + pm_runtime_put_autosuspend(dev);
>
> I expect the usage count will be -1 by here in probe.
will change it
>
> > +
> > + dev_dbg(dev, "%s: xhci_mtk_runtime_ready %i",
> > + __func__, xhci_mtk_runtime_ready);
> > +
> > return 0;
> >
> > dealloc_usb2_hcd:
> > @@ -587,7 +684,7 @@ static int xhci_mtk_probe(struct platform_device *pdev)
> > xhci_mtk_ldos_disable(mtk);
> >
> > disable_pm:
> > - pm_runtime_put_sync(dev);
> > + pm_runtime_put_sync_autosuspend(dev);
> > pm_runtime_disable(dev);
> > return ret;
> > }
> > @@ -602,6 +699,7 @@ static int xhci_mtk_remove(struct platform_device *dev)
> > pm_runtime_put_noidle(&dev->dev);
> > pm_runtime_disable(&dev->dev);
> >
> > + xhci_mtk_runtime_ready = 0;
> > usb_remove_hcd(shared_hcd);
> > xhci->shared_hcd = NULL;
> > device_init_wakeup(&dev->dev, false);
> > @@ -638,6 +736,7 @@ static int __maybe_unused xhci_mtk_suspend(struct device *dev)
> > xhci_mtk_host_disable(mtk);
> > xhci_mtk_clks_disable(mtk);
> > usb_wakeup_set(mtk, true);
> > +
> > return 0;
> > }
> >
> > @@ -659,10 +758,185 @@ static int __maybe_unused xhci_mtk_resume(struct device *dev)
> > return 0;
> > }
> >
> > +static int __maybe_unused xhci_mtk_bus_status(struct device *dev)
> > +{
> > + struct xhci_hcd_mtk *mtk = dev_get_drvdata(dev);
> > + struct usb_hcd *hcd;
> > + struct xhci_hcd *xhci;
> > + struct xhci_hub *usb2_rhub;
> > + struct xhci_hub *usb3_rhub;
> > + struct xhci_bus_state *bus_state;
> > + struct xhci_port *port;
> > + u32 usb2_suspended_ports = -1;
> > + u32 usb3_suspended_ports = -1;
> > + u16 status;
> > + int num_ports;
> > + int ret = 0;
> > + int i;
> > +
> > + if (!mtk->hcd)
> > + return -ESHUTDOWN;
> > +
> > + hcd = mtk->hcd;
> > + xhci = hcd_to_xhci(hcd);
> > + if ((xhci->xhc_state & XHCI_STATE_REMOVING) ||
> > + (xhci->xhc_state & XHCI_STATE_HALTED)) {
> > + return -ESHUTDOWN;
> > + }
>
> This is duplicated from xhci_mtk_runtime_suspend()
Yes, will remove it and rebuild the code
>
> > +
> > + usb2_rhub = &xhci->usb2_rhub;
> > + if (usb2_rhub) {
> > + bus_state = &usb2_rhub->bus_state;
> > + num_ports = usb2_rhub->num_ports;
> > + usb2_suspended_ports = bus_state->suspended_ports;
> > + usb2_suspended_ports ^= (BIT(num_ports) - 1);
> > + usb2_suspended_ports &= (BIT(num_ports) - 1);
> > + for (i = 0; i < num_ports; i++) {
> > + if (usb2_suspended_ports & (1UL << i)) {
> > + port = usb2_rhub->ports[i];
> > + status = readl(port->addr);
> > +
> > + xhci_dbg(xhci,
> > + "USB20: portsc[%i]: 0x%04X\n",
> > + i, status);
> > +
> > + if (!(status & PORT_CONNECT))
> > + usb2_suspended_ports &= ~(1UL << i);
> > + }
> > + }
> > +
> > + if (usb2_suspended_ports) {
> > + ret = -EBUSY;
> > + goto ebusy;
> > + }
> > + }
> > +
> > + usb3_rhub = &xhci->usb3_rhub;
> > + if (usb3_rhub) {
> > + bus_state = &usb3_rhub->bus_state;
> > + num_ports = usb3_rhub->num_ports;
> > + usb3_suspended_ports = bus_state->suspended_ports;
> > + usb3_suspended_ports ^= (BIT(num_ports) - 1);
> > + usb3_suspended_ports &= (BIT(num_ports) - 1);
> > + for (i = 0; i < num_ports; i++) {
> > + if (usb3_suspended_ports & BIT(i)) {
> > + port = usb3_rhub->ports[i];
> > + status = readl(port->addr);
> > +
> > + xhci_dbg(xhci, "USB3: portsc[%i]: 0x%04X\n",
> > + i, status);
> > +
> > + if (!(status & PORT_CONNECT))
> > + usb3_suspended_ports &= ~BIT(i);
> > + }
> > + }
> > +
> > + if (usb3_suspended_ports) {
> > + ret = -EBUSY;
> > + goto ebusy;
> > + }
> > + }
> > +
> > +ebusy:
> > + xhci_dbg(xhci, "%s: USB2: 0x%08X, USB3: 0x%08X ret: %i\n",
> > + __func__, usb2_suspended_ports,
> > + usb3_suspended_ports, ret);
> > +
> > + return ret;
> > +}
> > +
>
> This is basically counting active ports by directly reading portsc register?
Yes,
> I expect this function never return -EBUSY if you balance pm usage
> counter well.
> Are there any specific reasons of doing this manually?
pm_runtime_set_autosuspend_delay() is used, may check the ports status
periodically. I will check it.

If no spurious wakeup irq, apply the framework from wakeirq.c is the
easiest way to support runtime-suspend. hope that find out the root
cause and improve the HW design. In fact the EINT way is a deprecated
feature and not used before.

Thanks a lot.

>
> > +static int __maybe_unused xhci_mtk_runtime_suspend(struct device *dev)
> > +{
> > + bool wakeup = device_may_wakeup(dev);
> > + struct xhci_hcd_mtk *mtk = dev_get_drvdata(dev);
> > + struct usb_hcd *hcd;
> > + struct xhci_hcd *xhci;
> > + int ret = 0;
> > +
> > + if (!mtk->hcd)
> > + return -ESHUTDOWN;
> > +
> > + hcd = mtk->hcd;
> > + xhci = hcd_to_xhci(hcd);
> > + if ((xhci->xhc_state & XHCI_STATE_REMOVING) ||
> > + (xhci->xhc_state & XHCI_STATE_HALTED)) {
> > + return -ESHUTDOWN;
> > + }
> > +
> > + mtk->seal_status = SEAL_BUSY;
> > + ret = xhci_mtk_bus_status(dev);
> > + if (wakeup && !ret) {
> > + mtk->seal_status = SEAL_SUSPENDING;
> > + xhci_mtk_suspend(dev);
> > + xhci_mtk_seal_wakeup_enable(mtk, true);
> > + mtk->seal_status = SEAL_SUSPENDED;
> > + xhci_dbg(xhci, "%s: seals xHCI controller\n", __func__);
> > + }
> > +
> > + xhci_dbg(xhci, "%s: seals wakeup = %i, ret = %i!\n",
> > + __func__, wakeup, ret);
> > +
> > + return ret;
> > +}
> > +
> > +static int __maybe_unused xhci_mtk_runtime_resume(struct device *dev)
> > +{
> > + bool wakeup = device_may_wakeup(dev);
> > + struct xhci_hcd_mtk *mtk = dev_get_drvdata(dev);
> > + struct usb_hcd *hcd;
> > + struct xhci_hcd *xhci;
> > +
> > + if (!mtk->hcd)
> > + return -ESHUTDOWN;
> > +
> > + hcd = mtk->hcd;
> > + xhci = hcd_to_xhci(hcd);
> > + if ((xhci->xhc_state & XHCI_STATE_REMOVING) ||
> > + (xhci->xhc_state & XHCI_STATE_HALTED)) {
> > + return -ESHUTDOWN;
> > + }
> > +
> > + /*
> > + * list cases by one extra interrupt named seal to process!!!
> > + * Who to process these module reinitilization after SPM wakeup
> > + * case 1: usb remote wakeup, therefore xHCI need reinitilizate also.
> > + * case 2: other-wakeup-source wakeup, therefore, xHCI need reinit
> > + * case 3: usb client driver can invoke it in runtime mechanism
> > + * case 4: user active
> > + */
> > + if (wakeup) {
> > + xhci_mtk_seal_wakeup_enable(mtk, false);
> > + xhci_mtk_resume(dev);
> > + xhci_dbg(xhci, "%s: unseals xHCI controller\n", __func__);
> > + }
> > + mtk->seal_status = SEAL_RESUMED;
> > +
> > + xhci_dbg(xhci, "%s: unseals wakeup = %i\n", __func__, wakeup);
> > +
> > + return 0;
> > +}
> > +
> > +static int __maybe_unused xhci_mtk_runtime_idle(struct device *dev)
> > +{
> > + int ret = 0;
> > +
> > + dev_dbg(dev, "%s: xhci_mtk_runtime_ready %i",
> > + __func__, xhci_mtk_runtime_ready);
> > +
> > + if (!xhci_mtk_runtime_ready)
> > + ret = -EAGAIN;
> > +
> > + return ret;
> > +}
> > +
> > static const struct dev_pm_ops xhci_mtk_pm_ops = {
> > SET_SYSTEM_SLEEP_PM_OPS(xhci_mtk_suspend, xhci_mtk_resume)
> > + SET_RUNTIME_PM_OPS(xhci_mtk_runtime_suspend,
> > + xhci_mtk_runtime_resume,
> > + xhci_mtk_runtime_idle)
> > };
> > -#define DEV_PM_OPS IS_ENABLED(CONFIG_PM) ? &xhci_mtk_pm_ops : NULL
> > +
> > +#define DEV_PM_OPS (IS_ENABLED(CONFIG_PM) ? &xhci_mtk_pm_ops : NULL)
> >
> > #ifdef CONFIG_OF
> > static const struct of_device_id mtk_xhci_of_match[] = {
> > @@ -686,6 +960,7 @@ MODULE_ALIAS("platform:xhci-mtk");
> >
> > static int __init xhci_mtk_init(void)
> > {
> > + xhci_mtk_runtime_ready = 0;
> > xhci_init_driver(&xhci_mtk_hc_driver, &xhci_mtk_overrides);
> > return platform_driver_register(&mtk_xhci_driver);
> > }
> > diff --git a/drivers/usb/host/xhci-mtk.h b/drivers/usb/host/xhci-mtk.h
> > index 323b281933b9..103d83ce6a3e 100644
> > --- a/drivers/usb/host/xhci-mtk.h
> > +++ b/drivers/usb/host/xhci-mtk.h
> > @@ -133,6 +133,14 @@ struct mu3c_ippc_regs {
> > __le32 reserved3[33]; /* 0x80 ~ 0xff */
> > };
> >
> > +enum xhci_mtk_seal {
> > + SEAL_BUSY = 0,
> > + SEAL_SUSPENDING,
> > + SEAL_SUSPENDED,
> > + SEAL_RESUMING,
> > + SEAL_RESUMED
> > +};
> > +
> > struct xhci_hcd_mtk {
> > struct device *dev;
> > struct usb_hcd *hcd;
> > @@ -158,6 +166,12 @@ struct xhci_hcd_mtk {
> > struct regmap *uwk;
> > u32 uwk_reg_base;
> > u32 uwk_vers;
> > +
> > + /* usb eint wakeup source */
> > + int seal_irq;
> > + enum xhci_mtk_seal seal_status;
> > + struct delayed_work seal;
> > + char seal_descr[32]; /* "seal" + driver + bus # */
> > };
> >
> > static inline struct xhci_hcd_mtk *hcd_to_mtk(struct usb_hcd *hcd)
> > --
> > 2.18.0
> >