Message ID | 3d0ae3ca-9dad-bb8f-5c41-45bdcb07b9cd@zhaoxin.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | USB:Fix ehci infinite suspend-resume loop issue in zhaoxin | expand |
On Mon, Mar 14, 2022 at 03:35:37PM +0800, WeitaoWang-oc@zhaoxin.com wrote: > In zhaoxin platform, some ehci projects will latch a wakeup signal > internal when plug in a device on port during system S0. This wakeup > signal will turn on when ehci runtime suspend, which will trigger a > system control interrupt that will resume ehci back to D0. As no > device connect, ehci will be set to runtime suspend and turn on the > internal latched wakeup signal again. It will cause a suspend-resume > loop and generate system control interrupt continuously. > > Fixed this issue by clear wakeup signal latched in ehci internal when > ehci resume callback is called. > > Signed-off-by: Weitao Wang <WeitaoWang-oc@zhaoxin.com> > --- > drivers/usb/host/ehci-hcd.c | 26 ++++++++++++++++++++++++++ > drivers/usb/host/ehci-pci.c | 4 ++++ > drivers/usb/host/ehci.h | 1 + > 3 files changed, 31 insertions(+) > > diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c > index 3d82e0b..e4840ef 100644 > --- a/drivers/usb/host/ehci-hcd.c > +++ b/drivers/usb/host/ehci-hcd.c > @@ -1103,6 +1103,30 @@ static void ehci_remove_device(struct usb_hcd *hcd, > struct usb_device *udev) > > #ifdef CONFIG_PM > > +/* Clear wakeup signal locked in zhaoxin platform when device plug in. */ > +static void ehci_zx_wakeup_clear(struct ehci_hcd *ehci) > +{ > + u32 __iomem *reg = &ehci->regs->port_status[4]; > + u32 t1 = ehci_readl(ehci, reg); > + > + t1 &= (u32)~0xf0000; > + t1 |= PORT_TEST_FORCE; > + ehci_writel(ehci, t1, reg); > + t1 = ehci_readl(ehci, reg); > + msleep(1); > + t1 &= (u32)~0xf0000; > + ehci_writel(ehci, t1, reg); > + ehci_readl(ehci, reg); > + msleep(1); > + t1 = ehci_readl(ehci, reg); > + ehci_writel(ehci, t1 | PORT_CSC, reg); > + ehci_readl(ehci, reg); > + udelay(500); > + t1 = ehci_readl(ehci, &ehci->regs->status); > + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); > + ehci_readl(ehci, &ehci->regs->status); > +} > + > /* suspend/resume, section 4.3 */ > > /* These routines handle the generic parts of controller suspend/resume */ > @@ -1154,6 +1178,8 @@ int ehci_resume(struct usb_hcd *hcd, bool force_reset) > if (ehci->shutdown) > return 0; /* Controller is dead */ > > + if (ehci->zx_wakeup_clear == 1) > + ehci_zx_wakeup_clear(ehci); > /* > * If CF is still set and reset isn't forced > * then we maintained suspend power. > diff --git a/drivers/usb/host/ehci-pci.c b/drivers/usb/host/ehci-pci.c > index e87cf3a..a5e27de 100644 > --- a/drivers/usb/host/ehci-pci.c > +++ b/drivers/usb/host/ehci-pci.c > @@ -222,6 +222,10 @@ static int ehci_pci_setup(struct usb_hcd *hcd) > ehci->has_synopsys_hc_bug = 1; > } > break; > + case PCI_VENDOR_ID_ZHAOXIN: > + if (pdev->device == 0x3104 && (pdev->revision & 0xf0) == > 0x90) > + ehci->zx_wakeup_clear = 1; > + break; > } > > /* optional debug port, normally in the first BAR */ > diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h > index fdd073c..712fdd0 100644 > --- a/drivers/usb/host/ehci.h > +++ b/drivers/usb/host/ehci.h > @@ -220,6 +220,7 @@ struct ehci_hcd { /* one per > controller */ > unsigned imx28_write_fix:1; /* For Freescale i.MX28 > */ > unsigned spurious_oc:1; > unsigned is_aspeed:1; > + unsigned zx_wakeup_clear:1; > > /* required for usb32 quirk */ > #define OHCI_CTRL_HCFS (3 << 6) > -- > 2.7.4 Hi, This is the friendly patch-bot of Greg Kroah-Hartman. You have sent him a patch that has triggered this response. He used to manually respond to these common problems, but in order to save his sanity (he kept writing the same thing over and over, yet to different people), I was created. Hopefully you will not take offence and will fix the problem in your patch and resubmit it so that it can be accepted into the Linux kernel tree. You are receiving this message because of the following common error(s) as indicated below: - Your patch is malformed (tabs converted to spaces, linewrapped, etc.) and can not be applied. Please read the file, Documentation/email-clients.txt in order to fix this. If you wish to discuss this problem further, or you have questions about how to resolve this issue, please feel free to respond to this email and Greg will reply once he has dug out from the pending patches received from other developers. thanks, greg k-h's patch email bot
On Mon, Mar 14, 2022 at 03:35:37PM +0800, WeitaoWang-oc@zhaoxin.com wrote: > In zhaoxin platform, some ehci projects will latch a wakeup signal > internal when plug in a device on port during system S0. This wakeup > signal will turn on when ehci runtime suspend, which will trigger a > system control interrupt that will resume ehci back to D0. As no > device connect, ehci will be set to runtime suspend and turn on the > internal latched wakeup signal again. It will cause a suspend-resume > loop and generate system control interrupt continuously. > > Fixed this issue by clear wakeup signal latched in ehci internal when > ehci resume callback is called. > > Signed-off-by: Weitao Wang <WeitaoWang-oc@zhaoxin.com> > --- > drivers/usb/host/ehci-hcd.c | 26 ++++++++++++++++++++++++++ > drivers/usb/host/ehci-pci.c | 4 ++++ > drivers/usb/host/ehci.h | 1 + > 3 files changed, 31 insertions(+) > > diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c > index 3d82e0b..e4840ef 100644 > --- a/drivers/usb/host/ehci-hcd.c > +++ b/drivers/usb/host/ehci-hcd.c > @@ -1103,6 +1103,30 @@ static void ehci_remove_device(struct usb_hcd *hcd, > struct usb_device *udev) > > #ifdef CONFIG_PM > > +/* Clear wakeup signal locked in zhaoxin platform when device plug in. */ > +static void ehci_zx_wakeup_clear(struct ehci_hcd *ehci) > +{ > + u32 __iomem *reg = &ehci->regs->port_status[4]; > + u32 t1 = ehci_readl(ehci, reg); The "t1" in this line should start in the same column as the "*reg" in the line above, to match the style of other variable declarations in this file (see ehci_init() as an example). > + > + t1 &= (u32)~0xf0000; > + t1 |= PORT_TEST_FORCE; > + ehci_writel(ehci, t1, reg); > + t1 = ehci_readl(ehci, reg); > + msleep(1); > + t1 &= (u32)~0xf0000; > + ehci_writel(ehci, t1, reg); > + ehci_readl(ehci, reg); > + msleep(1); > + t1 = ehci_readl(ehci, reg); > + ehci_writel(ehci, t1 | PORT_CSC, reg); > + ehci_readl(ehci, reg); > + udelay(500); > + t1 = ehci_readl(ehci, &ehci->regs->status); > + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); > + ehci_readl(ehci, &ehci->regs->status); You should not clear the STS_PCD bit. What if some other port had a status change at the same time? Then because you cleared the port-change-detect bit, the system would not realize that the other port needed to be handled. Leaving the STS_PCD bit turned on will cause the driver to do a little extra work, but it shouldn't cause any harm. > +} > + > /* suspend/resume, section 4.3 */ > > /* These routines handle the generic parts of controller suspend/resume */ > @@ -1154,6 +1178,8 @@ int ehci_resume(struct usb_hcd *hcd, bool force_reset) > if (ehci->shutdown) > return 0; /* Controller is dead */ > > + if (ehci->zx_wakeup_clear == 1) You don't need to check that the value is equal to 1. Treat this more like a Boolean flag and just write: if (ehci->zx_wakeup_clear) Also, to make the flag's meaning more obvious, you might want to name it "zx_wakeup_clear_needed" or "zx_clear_latched_wakeup". Otherwise this patch looks okay. Please submit a revised version, without the whitespace damage. Alan Stern > + ehci_zx_wakeup_clear(ehci); > /* > * If CF is still set and reset isn't forced > * then we maintained suspend power. > diff --git a/drivers/usb/host/ehci-pci.c b/drivers/usb/host/ehci-pci.c > index e87cf3a..a5e27de 100644 > --- a/drivers/usb/host/ehci-pci.c > +++ b/drivers/usb/host/ehci-pci.c > @@ -222,6 +222,10 @@ static int ehci_pci_setup(struct usb_hcd *hcd) > ehci->has_synopsys_hc_bug = 1; > } > break; > + case PCI_VENDOR_ID_ZHAOXIN: > + if (pdev->device == 0x3104 && (pdev->revision & 0xf0) == > 0x90) > + ehci->zx_wakeup_clear = 1; > + break; > } > > /* optional debug port, normally in the first BAR */ > diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h > index fdd073c..712fdd0 100644 > --- a/drivers/usb/host/ehci.h > +++ b/drivers/usb/host/ehci.h > @@ -220,6 +220,7 @@ struct ehci_hcd { /* one per > controller */ > unsigned imx28_write_fix:1; /* For Freescale i.MX28 > */ > unsigned spurious_oc:1; > unsigned is_aspeed:1; > + unsigned zx_wakeup_clear:1; > > /* required for usb32 quirk */ > #define OHCI_CTRL_HCFS (3 << 6) > -- > 2.7.4
On 2022/3/14 10:24, Alan Stern wrote: > On Mon, Mar 14, 2022 at 03:35:37PM +0800, WeitaoWang-oc@zhaoxin.com wrote: >> In zhaoxin platform, some ehci projects will latch a wakeup signal >> internal when plug in a device on port during system S0. This wakeup >> signal will turn on when ehci runtime suspend, which will trigger a >> system control interrupt that will resume ehci back to D0. As no >> device connect, ehci will be set to runtime suspend and turn on the >> internal latched wakeup signal again. It will cause a suspend-resume >> loop and generate system control interrupt continuously. >> >> Fixed this issue by clear wakeup signal latched in ehci internal when >> ehci resume callback is called. >> >> Signed-off-by: Weitao Wang <WeitaoWang-oc@zhaoxin.com> >> --- >> drivers/usb/host/ehci-hcd.c | 26 ++++++++++++++++++++++++++ >> drivers/usb/host/ehci-pci.c | 4 ++++ >> drivers/usb/host/ehci.h | 1 + >> 3 files changed, 31 insertions(+) >> >> diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c >> index 3d82e0b..e4840ef 100644 >> --- a/drivers/usb/host/ehci-hcd.c >> +++ b/drivers/usb/host/ehci-hcd.c >> @@ -1103,6 +1103,30 @@ static void ehci_remove_device(struct usb_hcd *hcd, >> struct usb_device *udev) >> >> #ifdef CONFIG_PM >> >> +/* Clear wakeup signal locked in zhaoxin platform when device plug in. */ >> +static void ehci_zx_wakeup_clear(struct ehci_hcd *ehci) >> +{ >> + u32 __iomem *reg = &ehci->regs->port_status[4]; >> + u32 t1 = ehci_readl(ehci, reg); > > The "t1" in this line should start in the same column as the "*reg" in > the line above, to match the style of other variable declarations in > this file (see ehci_init() as an example). > >> + >> + t1 &= (u32)~0xf0000; >> + t1 |= PORT_TEST_FORCE; >> + ehci_writel(ehci, t1, reg); >> + t1 = ehci_readl(ehci, reg); >> + msleep(1); >> + t1 &= (u32)~0xf0000; >> + ehci_writel(ehci, t1, reg); >> + ehci_readl(ehci, reg); >> + msleep(1); >> + t1 = ehci_readl(ehci, reg); >> + ehci_writel(ehci, t1 | PORT_CSC, reg); >> + ehci_readl(ehci, reg); >> + udelay(500); >> + t1 = ehci_readl(ehci, &ehci->regs->status); >> + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); >> + ehci_readl(ehci, &ehci->regs->status); > > You should not clear the STS_PCD bit. What if some other port had a > status change at the same time? Then because you cleared the > port-change-detect bit, the system would not realize that the other port > needed to be handled. I really didn't think about this case. > Leaving the STS_PCD bit turned on will cause the driver to do a little > extra work, but it shouldn't cause any harm. > I have encountered the following situation if EHCI runtime suspend is enabled by default. 1.Wake from system to disk and boot OS. 2.EHCI will entry runtime suspend after enumerated by driver during boot phase of suspend to disk 3.EHCI will be placed to freeze state and ehci_resume is called after image is loaded. 4.If PCD flag is set(caused by patch), then HCD_FLAG_RH_RUNNING will be set. 5.Pci_pm_freeze_noirq is called to check ehci root hub state and return value is -EBUSY. which will cause quiesce phase of suspend to disk fail. However,EHCI runtime suspend is default disabled by kernel, user can enable runtime suspend after boot into OS. So I guess turning on suspend during startup is not a real requirement and I will take your advice. Weitao Wang >> +} >> + >> /* suspend/resume, section 4.3 */ >> >> /* These routines handle the generic parts of controller suspend/resume */ >> @@ -1154,6 +1178,8 @@ int ehci_resume(struct usb_hcd *hcd, bool force_reset) >> if (ehci->shutdown) >> return 0; /* Controller is dead */ >> >> + if (ehci->zx_wakeup_clear == 1) > > You don't need to check that the value is equal to 1. Treat this > more like a Boolean flag and just write: > > if (ehci->zx_wakeup_clear) > > Also, to make the flag's meaning more obvious, you might want to name > it "zx_wakeup_clear_needed" or "zx_clear_latched_wakeup". > > Otherwise this patch looks okay. Please submit a revised version, > without the whitespace damage. > > Alan Stern Okay,I will take your advice Weitao Wang > >> + ehci_zx_wakeup_clear(ehci); >> /* >> * If CF is still set and reset isn't forced >> * then we maintained suspend power. >> diff --git a/drivers/usb/host/ehci-pci.c b/drivers/usb/host/ehci-pci.c >> index e87cf3a..a5e27de 100644 >> --- a/drivers/usb/host/ehci-pci.c >> +++ b/drivers/usb/host/ehci-pci.c >> @@ -222,6 +222,10 @@ static int ehci_pci_setup(struct usb_hcd *hcd) >> ehci->has_synopsys_hc_bug = 1; >> } >> break; >> + case PCI_VENDOR_ID_ZHAOXIN: >> + if (pdev->device == 0x3104 && (pdev->revision & 0xf0) == >> 0x90) >> + ehci->zx_wakeup_clear = 1; >> + break; >> } >> >> /* optional debug port, normally in the first BAR */ >> diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h >> index fdd073c..712fdd0 100644 >> --- a/drivers/usb/host/ehci.h >> +++ b/drivers/usb/host/ehci.h >> @@ -220,6 +220,7 @@ struct ehci_hcd { /* one per >> controller */ >> unsigned imx28_write_fix:1; /* For Freescale i.MX28 >> */ >> unsigned spurious_oc:1; >> unsigned is_aspeed:1; >> + unsigned zx_wakeup_clear:1; >> >> /* required for usb32 quirk */ >> #define OHCI_CTRL_HCFS (3 << 6) >> -- >> 2.7.4 > . >
On Tue, Mar 15, 2022 at 08:39:09PM +0800, WeitaoWang-oc@zhaoxin.com wrote: > On 2022/3/14 10:24, Alan Stern wrote: > > > + t1 = ehci_readl(ehci, &ehci->regs->status); > > > + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); > > > + ehci_readl(ehci, &ehci->regs->status); > > > > You should not clear the STS_PCD bit. What if some other port had a > > status change at the same time? Then because you cleared the > > port-change-detect bit, the system would not realize that the other port > > needed to be handled. > > I really didn't think about this case. > > > Leaving the STS_PCD bit turned on will cause the driver to do a little > > extra work, but it shouldn't cause any harm. > > > I have encountered the following situation if EHCI runtime suspend is > enabled by default. > > > > 1.Wake from system to disk and boot OS. You're talking about resuming after hibernation, right? > 2.EHCI will entry runtime suspend after enumerated by driver during boot > phase of suspend to disk I'm not sure what you mean by "boot phase of suspend to disk". This is while the restore kernel is starting up at the beginning of resume from hibernation, right? > 3.EHCI will be placed to freeze state and ehci_resume is called after image > is loaded. ehci_resume is called to leave runtime suspend. Going into the freeze state doesn't require any changes. > 4.If PCD flag is set(caused by patch), then HCD_FLAG_RH_RUNNING will be set. > > 5.Pci_pm_freeze_noirq is called to check ehci root hub state and return > value is -EBUSY. which will cause > quiesce phase of suspend to disk fail. You're talking about check_root_hub_suspended() in hcd-pci.c, right? You know, I'm not at all certain that the callbacks for freeze and freeze_noirq should ever return anything other than 0. It's okay for them to call check_root_hub_suspended(), but they should ignore its return value. Can you check if the patch below helps? Alan Stern Index: usb-devel/drivers/usb/core/hcd-pci.c =================================================================== --- usb-devel.orig/drivers/usb/core/hcd-pci.c +++ usb-devel/drivers/usb/core/hcd-pci.c @@ -575,6 +575,12 @@ static int hcd_pci_resume(struct device return resume_common(dev, PM_EVENT_RESUME); } +static int hcd_pci_freeze_check(struct device *dev) +{ + (void) check_root_hub_suspended(dev); + return 0; +} + static int hcd_pci_restore(struct device *dev) { return resume_common(dev, PM_EVENT_RESTORE); @@ -586,6 +592,7 @@ static int hcd_pci_restore(struct device #define hcd_pci_suspend_noirq NULL #define hcd_pci_resume_noirq NULL #define hcd_pci_resume NULL +#define hcd_pci_freeze_check NULL #define hcd_pci_restore NULL #endif /* CONFIG_PM_SLEEP */ @@ -616,8 +623,8 @@ const struct dev_pm_ops usb_hcd_pci_pm_o .suspend_noirq = hcd_pci_suspend_noirq, .resume_noirq = hcd_pci_resume_noirq, .resume = hcd_pci_resume, - .freeze = check_root_hub_suspended, - .freeze_noirq = check_root_hub_suspended, + .freeze = hcd_pci_freeze_check, + .freeze_noirq = hcd_pci_freeze_check, .thaw_noirq = NULL, .thaw = NULL, .poweroff = hcd_pci_suspend,
On 2022/3/15 11:18, Alan Stern wrote: > On Tue, Mar 15, 2022 at 08:39:09PM +0800, WeitaoWang-oc@zhaoxin.com wrote: >> On 2022/3/14 10:24, Alan Stern wrote: >>>> + t1 = ehci_readl(ehci, &ehci->regs->status); >>>> + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); >>>> + ehci_readl(ehci, &ehci->regs->status); >>> >>> You should not clear the STS_PCD bit. What if some other port had a >>> status change at the same time? Then because you cleared the >>> port-change-detect bit, the system would not realize that the other port >>> needed to be handled. >> >> I really didn't think about this case. >> >>> Leaving the STS_PCD bit turned on will cause the driver to do a little >>> extra work, but it shouldn't cause any harm. >>> >> I have encountered the following situation if EHCI runtime suspend is >> enabled by default. >> >> >> >> 1.Wake from system to disk and boot OS. > > You're talking about resuming after hibernation, right? You're right. >> 2.EHCI will entry runtime suspend after enumerated by driver during boot >> phase of suspend to disk > > I'm not sure what you mean by "boot phase of suspend to disk". This is > while the restore kernel is starting up at the beginning of resume from > hibernation, right? > You understood exactly what I was saying. >> 3.EHCI will be placed to freeze state and ehci_resume is called after image >> is loaded. > > ehci_resume is called to leave runtime suspend. Going into the freeze > state doesn't require any changes. > >> 4.If PCD flag is set(caused by patch), then HCD_FLAG_RH_RUNNING will be set. >> >> 5.Pci_pm_freeze_noirq is called to check ehci root hub state and return >> value is -EBUSY. which will cause >> quiesce phase of suspend to disk fail. > > You're talking about check_root_hub_suspended() in hcd-pci.c, right? > It's right. > You know, I'm not at all certain that the callbacks for freeze and > freeze_noirq should ever return anything other than 0. It's okay for > them to call check_root_hub_suspended(), but they should ignore its > return value. > > Can you check if the patch below helps? > > Alan Stern > > > Index: usb-devel/drivers/usb/core/hcd-pci.c > =================================================================== > --- usb-devel.orig/drivers/usb/core/hcd-pci.c > +++ usb-devel/drivers/usb/core/hcd-pci.c > @@ -575,6 +575,12 @@ static int hcd_pci_resume(struct device > return resume_common(dev, PM_EVENT_RESUME); > } > > +static int hcd_pci_freeze_check(struct device *dev) > +{ > + (void) check_root_hub_suspended(dev); > + return 0; > +} > + > static int hcd_pci_restore(struct device *dev) > { > return resume_common(dev, PM_EVENT_RESTORE); > @@ -586,6 +592,7 @@ static int hcd_pci_restore(struct device > #define hcd_pci_suspend_noirq NULL > #define hcd_pci_resume_noirq NULL > #define hcd_pci_resume NULL > +#define hcd_pci_freeze_check NULL > #define hcd_pci_restore NULL > > #endif /* CONFIG_PM_SLEEP */ > @@ -616,8 +623,8 @@ const struct dev_pm_ops usb_hcd_pci_pm_o > .suspend_noirq = hcd_pci_suspend_noirq, > .resume_noirq = hcd_pci_resume_noirq, > .resume = hcd_pci_resume, > - .freeze = check_root_hub_suspended, > - .freeze_noirq = check_root_hub_suspended, > + .freeze = hcd_pci_freeze_check, > + .freeze_noirq = hcd_pci_freeze_check, This patch can fix pci driver's fail check in freeze_noirq phase. Restoring system state from a hibernation image can continue and success. Weitao Wang > .thaw_noirq = NULL, > .thaw = NULL, > .poweroff = hcd_pci_suspend, > . >
On 2022/3/16 10:18, WeitaoWang-oc@zhaoxin.com wrote: > On 2022/3/15 11:18, Alan Stern wrote: >> On Tue, Mar 15, 2022 at 08:39:09PM +0800, WeitaoWang-oc@zhaoxin.com wrote: >>> On 2022/3/14 10:24, Alan Stern wrote: >>>>> + t1 = ehci_readl(ehci, &ehci->regs->status); >>>>> + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); >>>>> + ehci_readl(ehci, &ehci->regs->status); >>>> >>>> You should not clear the STS_PCD bit. What if some other port had a >>>> status change at the same time? Then because you cleared the >>>> port-change-detect bit, the system would not realize that the other port >>>> needed to be handled. >>> >>> I really didn't think about this case. >>> >>>> Leaving the STS_PCD bit turned on will cause the driver to do a little >>>> extra work, but it shouldn't cause any harm. >>>> >>> I have encountered the following situation if EHCI runtime suspend is >>> enabled by default. >>> >>> >>> >>> 1.Wake from system to disk and boot OS. >> >> You're talking about resuming after hibernation, right? > > You're right. >>> 2.EHCI will entry runtime suspend after enumerated by driver during boot >>> phase of suspend to disk >> >> I'm not sure what you mean by "boot phase of suspend to disk". This is >> while the restore kernel is starting up at the beginning of resume from >> hibernation, right? >> > You understood exactly what I was saying. > > >>> 3.EHCI will be placed to freeze state and ehci_resume is called after image >>> is loaded. >> >> ehci_resume is called to leave runtime suspend. Going into the freeze >> state doesn't require any changes. >> >>> 4.If PCD flag is set(caused by patch), then HCD_FLAG_RH_RUNNING will be set. >>> >>> 5.Pci_pm_freeze_noirq is called to check ehci root hub state and return >>> value is -EBUSY. which will cause >>> quiesce phase of suspend to disk fail. >> >> You're talking about check_root_hub_suspended() in hcd-pci.c, right? >> > It's right. >> You know, I'm not at all certain that the callbacks for freeze and >> freeze_noirq should ever return anything other than 0. It's okay for >> them to call check_root_hub_suspended(), but they should ignore its >> return value. >> >> Can you check if the patch below helps? >> >> Alan Stern >> >> >> Index: usb-devel/drivers/usb/core/hcd-pci.c >> =================================================================== >> --- usb-devel.orig/drivers/usb/core/hcd-pci.c >> +++ usb-devel/drivers/usb/core/hcd-pci.c >> @@ -575,6 +575,12 @@ static int hcd_pci_resume(struct device >> return resume_common(dev, PM_EVENT_RESUME); >> } >> +static int hcd_pci_freeze_check(struct device *dev) >> +{ >> + (void) check_root_hub_suspended(dev); >> + return 0; >> +} >> + >> static int hcd_pci_restore(struct device *dev) >> { >> return resume_common(dev, PM_EVENT_RESTORE); >> @@ -586,6 +592,7 @@ static int hcd_pci_restore(struct device >> #define hcd_pci_suspend_noirq NULL >> #define hcd_pci_resume_noirq NULL >> #define hcd_pci_resume NULL >> +#define hcd_pci_freeze_check NULL >> #define hcd_pci_restore NULL >> #endif /* CONFIG_PM_SLEEP */ >> @@ -616,8 +623,8 @@ const struct dev_pm_ops usb_hcd_pci_pm_o >> .suspend_noirq = hcd_pci_suspend_noirq, >> .resume_noirq = hcd_pci_resume_noirq, >> .resume = hcd_pci_resume, >> - .freeze = check_root_hub_suspended, >> - .freeze_noirq = check_root_hub_suspended, >> + .freeze = hcd_pci_freeze_check, >> + .freeze_noirq = hcd_pci_freeze_check, > > This patch can fix pci driver's fail check in freeze_noirq phase. > > Restoring system state from a hibernation image can continue and success. > > Weitao Wang This patch seems work for our platform, but I don't know if there will be side-effects for others. So I want to take your previous suggestion and not to clear PCD, As ehci runtime suspend enabled at OS startup is not a real requirement. I'll resubmit revised patch without the whitespace damage. Thanks for your great help. Best Wishes Weitao Wang >> .thaw_noirq = NULL, >> .thaw = NULL, >> .poweroff = hcd_pci_suspend, >> . >>
On Wed, Mar 16, 2022 at 10:18:39AM +0800, WeitaoWang-oc@zhaoxin.com wrote: > On 2022/3/15 11:18, Alan Stern wrote: > > On Tue, Mar 15, 2022 at 08:39:09PM +0800, WeitaoWang-oc@zhaoxin.com wrote: > > > I have encountered the following situation if EHCI runtime suspend is > > > enabled by default. Some things about this still confuse me... > > > 1.Wake from system to disk and boot OS. > > > > You're talking about resuming after hibernation, right? > > You're right. > > > 2.EHCI will entry runtime suspend after enumerated by driver during boot > > > phase of suspend to disk > > > > I'm not sure what you mean by "boot phase of suspend to disk". This is > > while the restore kernel is starting up at the beginning of resume from > > hibernation, right? > > > You understood exactly what I was saying. Okay, so we're waking up from hibernation. > > > 3.EHCI will be placed to freeze state and ehci_resume is called after image > > > is loaded. > > > > ehci_resume is called to leave runtime suspend. Going into the freeze > > state doesn't require any changes. In fact, the resume kernel doesn't call ehci_resume at all. Here's what it does: The resume kernel boots; If your patch causes STS_PCD to be set at this point, the flag should get cleared shortly afterward by ehci_irq; ehci-hcd goes into runtime suspend; The kernel reads the system image that was stored earlier when hibernation began; After the image is loaded, the system goes into the freeze state (this does not call any routines in ehci-hcd); The resume kernel transfers control to the image kernel; Now the image kernel is running. It goes into the restore state, which involves calling ehci_resume. Presumably your patch cases the STS_PCD flag to get set at this point. But that's all! The system is now back up, out of hibernation, and running normally. There are no more calls to check_root_hub_suspended > > > 4.If PCD flag is set(caused by patch), then HCD_FLAG_RH_RUNNING will be set. > > > > > > 5.Pci_pm_freeze_noirq is called to check ehci root hub state and return > > > value is -EBUSY. which will cause > > > quiesce phase of suspend to disk fail. > > > > You're talking about check_root_hub_suspended() in hcd-pci.c, right? > > > It's right. So how can this happen, given that the image kernel doesn't call pci_pm_freeze_noirq? Alan Stern
On 2022/4/6 00:02, Alan Stern wrote: > On Wed, Mar 16, 2022 at 10:18:39AM +0800, WeitaoWang-oc@zhaoxin.com wrote: >> On 2022/3/15 11:18, Alan Stern wrote: >>> On Tue, Mar 15, 2022 at 08:39:09PM +0800, WeitaoWang-oc@zhaoxin.com wrote: >>>> I have encountered the following situation if EHCI runtime suspend is >>>> enabled by default. > > Some things about this still confuse me... > >>>> 1.Wake from system to disk and boot OS. >>> >>> You're talking about resuming after hibernation, right? >> >> You're right. >>>> 2.EHCI will entry runtime suspend after enumerated by driver during boot >>>> phase of suspend to disk >>> >>> I'm not sure what you mean by "boot phase of suspend to disk". This is >>> while the restore kernel is starting up at the beginning of resume from >>> hibernation, right? >>> >> You understood exactly what I was saying. > > Okay, so we're waking up from hibernation. > >>>> 3.EHCI will be placed to freeze state and ehci_resume is called after image >>>> is loaded. >>> >>> ehci_resume is called to leave runtime suspend. Going into the freeze >>> state doesn't require any changes. > > In fact, the resume kernel doesn't call ehci_resume at all. Here's what > it does: > > The resume kernel boots; > > If your patch causes STS_PCD to be set at this point, the flag > should get cleared shortly afterward by ehci_irq; > > ehci-hcd goes into runtime suspend; > > The kernel reads the system image that was stored earlier when > hibernation began; > > After the image is loaded, the system goes into the freeze > state (this does not call any routines in ehci-hcd); On this phase, pci_pm_freeze will be called for pci device. In this function, pm_runtime_resume will be called to resume already runtime-suspend devices. which will cause ehci_resume to be called. Thus STS_PCD flag will be set in ehci_resume function. Thanks Weitao Wang > > The resume kernel transfers control to the image kernel; > > Now the image kernel is running. It goes into the restore state, which > involves calling ehci_resume. Presumably your patch cases the STS_PCD > flag to get set at this point. > o > But that's all! The system is now back up, out of hibernation, and > running normally. There are no more calls to check_root_hub_suspended > >>>> 4.If PCD flag is set(caused by patch), then HCD_FLAG_RH_RUNNING will be set. >>>> >>>> 5.Pci_pm_freeze_noirq is called to check ehci root hub state and return >>>> value is -EBUSY. which will cause >>>> quiesce phase of suspend to disk fail. >>> >>> You're talking about check_root_hub_suspended() in hcd-pci.c, right? >>> >> It's right. > > So how can this happen, given that the image kernel doesn't call > pci_pm_freeze_noirq? > > Alan Stern > .
On Wed, Apr 06, 2022 at 10:38:28AM +0800, WeitaoWang-oc@zhaoxin.com wrote: > On 2022/4/6 00:02, Alan Stern wrote: > > In fact, the resume kernel doesn't call ehci_resume at all. Here's what > > it does: > > > > The resume kernel boots; > > > > If your patch causes STS_PCD to be set at this point, the flag > > should get cleared shortly afterward by ehci_irq; > > > > ehci-hcd goes into runtime suspend; > > > > The kernel reads the system image that was stored earlier when > > hibernation began; > > > > After the image is loaded, the system goes into the freeze > > state (this does not call any routines in ehci-hcd); > On this phase, pci_pm_freeze will be called for pci device. In this > function, pm_runtime_resume will be called to resume already > runtime-suspend devices. which will cause ehci_resume to be called. > Thus STS_PCD flag will be set in ehci_resume function. Aha! I was missing that piece of information, thanks. But this still doesn't explain why check_root_hub_suspended is failing. That routine checks the HCD_RH_RUNNING bit, which gets set in hcd_bus_resume. hcd_bus_resume gets called as part of resuming the root hub, and in ehci-hcd this happens when ehci_irq sees that STS_PCD is set and calls usb_hcd_resume_root_hub. That routine queues a wakeup request on the pm_wq work queue, which is then supposed to run hcd_resume_work to actually restart the root hub. But pm_wq is a freezable work queue! While the system is in the freeze state, the work queue isn't running. This means that the root hub should remain suspended until the end of the freeze phase, and so the call to check_root_hub_suspended should succeed. Can you check to see what's really happening on your system? Something must be wrong with my analysis, but I can't tell what it is. I'm still puzzled. Alan Stern
On 2022/4/7 00:20, Alan Stern wrote: > On Wed, Apr 06, 2022 at 10:38:28AM +0800, WeitaoWang-oc@zhaoxin.com wrote: >> On 2022/4/6 00:02, Alan Stern wrote: >>> In fact, the resume kernel doesn't call ehci_resume at all. Here's what >>> it does: >>> >>> The resume kernel boots; >>> >>> If your patch causes STS_PCD to be set at this point, the flag >>> should get cleared shortly afterward by ehci_irq; >>> >>> ehci-hcd goes into runtime suspend; >>> >>> The kernel reads the system image that was stored earlier when >>> hibernation began; >>> >>> After the image is loaded, the system goes into the freeze >>> state (this does not call any routines in ehci-hcd); >> On this phase, pci_pm_freeze will be called for pci device. In this >> function, pm_runtime_resume will be called to resume already >> runtime-suspend devices. which will cause ehci_resume to be called. >> Thus STS_PCD flag will be set in ehci_resume function. > > Aha! I was missing that piece of information, thanks. > > But this still doesn't explain why check_root_hub_suspended is failing. > That routine checks the HCD_RH_RUNNING bit, which gets set in > hcd_bus_resume. hcd_bus_resume gets called as part of resuming the root > hub, and in ehci-hcd this happens when ehci_irq sees that STS_PCD is set > and calls usb_hcd_resume_root_hub. That routine queues a wakeup request > on the pm_wq work queue, which is then supposed to run hcd_resume_work > to actually restart the root hub. > > But pm_wq is a freezable work queue! While the system is in the freeze > state, the work queue isn't running. This means that the root hub > should remain suspended until the end of the freeze phase, and so the > call to check_root_hub_suspended should succeed. > > Can you check to see what's really happening on your system? Something > must be wrong with my analysis, but I can't tell what it is. I'm still > puzzled. > > Alan Stern Your analysis is right, my test platform's kernel version is not the latest, this kernel not call freeze_kernel_threads on software_resume function. (https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/kernel/power/hibernate.c?h=v5.18-rc1&id=2351f8d295ed63393190e39c2f7c1fee1a80578f) So pm_wq is active and can handle root hub power events. Update my kernel to fix the issue in the url above, system hibernation test was successful with our patch(not clear STS_PCD bit). Thanks for your clarification. Weitao Wang
On Thu, Apr 07, 2022 at 02:15:29PM +0800, WeitaoWang-oc@zhaoxin.com wrote: > On 2022/4/7 00:20, Alan Stern wrote: > > On Wed, Apr 06, 2022 at 10:38:28AM +0800, WeitaoWang-oc@zhaoxin.com wrote: > > > On 2022/4/6 00:02, Alan Stern wrote: > > > > In fact, the resume kernel doesn't call ehci_resume at all. Here's what > > > > it does: > > > > > > > > The resume kernel boots; > > > > > > > > If your patch causes STS_PCD to be set at this point, the flag > > > > should get cleared shortly afterward by ehci_irq; > > > > > > > > ehci-hcd goes into runtime suspend; > > > > > > > > The kernel reads the system image that was stored earlier when > > > > hibernation began; > > > > > > > > After the image is loaded, the system goes into the freeze > > > > state (this does not call any routines in ehci-hcd); > > > On this phase, pci_pm_freeze will be called for pci device. In this > > > function, pm_runtime_resume will be called to resume already > > > runtime-suspend devices. which will cause ehci_resume to be called. > > > Thus STS_PCD flag will be set in ehci_resume function. > > > > Aha! I was missing that piece of information, thanks. > > > > But this still doesn't explain why check_root_hub_suspended is failing. > > That routine checks the HCD_RH_RUNNING bit, which gets set in > > hcd_bus_resume. hcd_bus_resume gets called as part of resuming the root > > hub, and in ehci-hcd this happens when ehci_irq sees that STS_PCD is set > > and calls usb_hcd_resume_root_hub. That routine queues a wakeup request > > on the pm_wq work queue, which is then supposed to run hcd_resume_work > > to actually restart the root hub. > > > > But pm_wq is a freezable work queue! While the system is in the freeze > > state, the work queue isn't running. This means that the root hub > > should remain suspended until the end of the freeze phase, and so the > > call to check_root_hub_suspended should succeed. > > > > Can you check to see what's really happening on your system? Something > > must be wrong with my analysis, but I can't tell what it is. I'm still > > puzzled. > > > > Alan Stern > Your analysis is right, my test platform's kernel version is not the > latest, this kernel not call freeze_kernel_threads on software_resume > function. > (https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/kernel/power/hibernate.c?h=v5.18-rc1&id=2351f8d295ed63393190e39c2f7c1fee1a80578f) > So pm_wq is active and can handle root hub power events. > Update my kernel to fix the issue in the url above, system hibernation > test was successful with our patch(not clear STS_PCD bit). > Thanks for your clarification. Great! I'm glad we sorted that out. So check_root_hub_suspended doesn't need any changes, and the patch you already submitted takes care of everything. Alan Stern
diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c index 3d82e0b..e4840ef 100644 --- a/drivers/usb/host/ehci-hcd.c +++ b/drivers/usb/host/ehci-hcd.c @@ -1103,6 +1103,30 @@ static void ehci_remove_device(struct usb_hcd *hcd, struct usb_device *udev) #ifdef CONFIG_PM +/* Clear wakeup signal locked in zhaoxin platform when device plug in. */ +static void ehci_zx_wakeup_clear(struct ehci_hcd *ehci) +{ + u32 __iomem *reg = &ehci->regs->port_status[4]; + u32 t1 = ehci_readl(ehci, reg); + + t1 &= (u32)~0xf0000; + t1 |= PORT_TEST_FORCE; + ehci_writel(ehci, t1, reg); + t1 = ehci_readl(ehci, reg); + msleep(1); + t1 &= (u32)~0xf0000; + ehci_writel(ehci, t1, reg); + ehci_readl(ehci, reg); + msleep(1); + t1 = ehci_readl(ehci, reg); + ehci_writel(ehci, t1 | PORT_CSC, reg); + ehci_readl(ehci, reg); + udelay(500); + t1 = ehci_readl(ehci, &ehci->regs->status); + ehci_writel(ehci, t1 & STS_PCD, &ehci->regs->status); + ehci_readl(ehci, &ehci->regs->status); +} + /* suspend/resume, section 4.3 */ /* These routines handle the generic parts of controller suspend/resume */ @@ -1154,6 +1178,8 @@ int ehci_resume(struct usb_hcd *hcd, bool force_reset) if (ehci->shutdown) return 0; /* Controller is dead */ + if (ehci->zx_wakeup_clear == 1) + ehci_zx_wakeup_clear(ehci); /* * If CF is still set and reset isn't forced * then we maintained suspend power. diff --git a/drivers/usb/host/ehci-pci.c b/drivers/usb/host/ehci-pci.c index e87cf3a..a5e27de 100644 --- a/drivers/usb/host/ehci-pci.c +++ b/drivers/usb/host/ehci-pci.c @@ -222,6 +222,10 @@ static int ehci_pci_setup(struct usb_hcd *hcd) ehci->has_synopsys_hc_bug = 1; } break; + case PCI_VENDOR_ID_ZHAOXIN: + if (pdev->device == 0x3104 && (pdev->revision & 0xf0) == 0x90) + ehci->zx_wakeup_clear = 1; + break; } /* optional debug port, normally in the first BAR */ diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h index fdd073c..712fdd0 100644 --- a/drivers/usb/host/ehci.h +++ b/drivers/usb/host/ehci.h @@ -220,6 +220,7 @@ struct ehci_hcd { /* one per controller */ unsigned imx28_write_fix:1; /* For Freescale i.MX28 */ unsigned spurious_oc:1; unsigned is_aspeed:1; + unsigned zx_wakeup_clear:1;
In zhaoxin platform, some ehci projects will latch a wakeup signal internal when plug in a device on port during system S0. This wakeup signal will turn on when ehci runtime suspend, which will trigger a system control interrupt that will resume ehci back to D0. As no device connect, ehci will be set to runtime suspend and turn on the internal latched wakeup signal again. It will cause a suspend-resume loop and generate system control interrupt continuously. Fixed this issue by clear wakeup signal latched in ehci internal when ehci resume callback is called. Signed-off-by: Weitao Wang <WeitaoWang-oc@zhaoxin.com> --- drivers/usb/host/ehci-hcd.c | 26 ++++++++++++++++++++++++++ drivers/usb/host/ehci-pci.c | 4 ++++ drivers/usb/host/ehci.h | 1 + 3 files changed, 31 insertions(+) /* required for usb32 quirk */ #define OHCI_CTRL_HCFS (3 << 6)