Reproduction of the problem: During the S4 stress test,
when a USB device is inserted or removed, there is a
probability that the S4 wakeup will turn into a reboot.
The following two points describe how to analyze and
locate the problem points:
1. During the boot stage when S4 is awakened, after
the USB RootHub is initialized,it will enter the
runtime suspend state. From then on, whenever an
xhci port change event occurs, it will trigger a
remote wakeup request event and add wakeup_work to
pm_wq, where the subsequent RootHub runtime resume
process will be handled by pm_wq.
xhci runtime suspend flow：
S4 boot
|->xhci init
|->register_root_hub
|->hub_probe
|->callback = RPM_GET_CALLBACK(dev, runtime_suspend) /* xhci RootHub runtime suspend */
xhci runtime resume flow ：
xhci_irq()
|->xhci_handle_event()
|->handle_port_status()
|->if(hcd->state == HC_STATE_SUSPENDED)
|->usb_hcd_resume_root_hub()
|->set_bit(HCD_FLAG_WAKEUP_PENDING, &hcd->flags) /* wakeup pending signal to be set */
|->queue_work(pm_wq, &hcd->wakeup_work)
|->hcd_resume_work() /* hcd->wakeup_work */
|->usb_remote_wakeup()
|->callback = RPM_GET_CALLBACK(dev, runtime_resume)
|->usb_runtime_resume() /* usb runtime resume */
|->generic_resume()
|->hcd_bus_resume()
|->clear_bit(HCD_FLAG_WAKEUP_PENDING, &hcd->flags);
/* wakeup pending signal to be clear */
2. However, during the quiesce phase of S4 wakeup,
freeze_kernel_threads() will freeze this pm_wq,
and between freeze_kernel_threads() and dpm_suspend_start(),
there exists a very time-consuming S4 image loading process.
This leads to a situation where, if an xhci port change event occurs
after freeze_kernel_threads(), triggering the wakeup pending
signal to be set,but it cannot be processed by pm_wq to clear
this wakeup_pending bit, it will result in a subsequent
dpm_suspend_start() where USB suspend_common() detects the
wakeup pending signal being set and returns an -EBUSY error,
interrupting the S4 quiesce process and reverting to a reboot.
S4 wakeup
|->resume_store
|->software_resume()
|->freeze_kernel_threads() /* will freeze pm_wq */
|->load_image_and_restore()
|->swsusp_read() /* S4 image loading: time-consuming .
When an xhci port change event occurs at this point, it triggers the wakeup pending signal to be set.
However, since the pm_wq is in a frozen state, the wakeup_pending bit cannot be cleared.*/
|->hibernation_restore
|->dpm_suspend_start(PMSG_QUIESCE)
|->hcd_pci_suspend()
|->suspend_common()
|->if (do_wakeup && HCD_WAKEUP_PENDING(hcd)) return -EBUSY;
Below is a description of the countermeasures taken to address this issue:
1. Considering the restore process that occurs after
the quiesce phase during S4 wakeup, which essentially
resets all root hubs,checking this wakeup pending status
in USB suspend_common() during the quiesce phase is of
little significance and should therefore be filtered out.
S4 wakeup restore phase
|->dpm_resume(PMSG_RESTORE)
|->hcd_pci_restore()
|->xhci_resume() /* reset all root hubs */
Fixes: 3904bdf0821c40352495f632862637080e6bd744(PM: hibernate: Freeze kernel threads in software_resume())
Signed-off-by: xiongxin <xiongxin@xxxxxxxxxx>
Co-developed-by: dengjie03 <dengjie03@xxxxxxxxxx>
---
drivers/base/power/main.c | 5 +++++
drivers/usb/core/hcd-pci.c | 21 ++++++++++++++-------
include/linux/pm.h | 1 +
3 files changed, 20 insertions(+), 7 deletions(-)
diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
index fb4d18a0b185..264d08b9e331 100644
--- a/drivers/base/power/main.c
+++ b/drivers/base/power/main.c
@@ -559,6 +559,11 @@ bool dev_pm_may_skip_resume(struct device *dev)
return !dev->power.must_resume && pm_transition.event != PM_EVENT_RESTORE;
}
+bool is_pm_queisce(void)
+{
+ return pm_transition.event == PM_EVENT_QUIESCE;
+}
+
static pm_callback_t dpm_subsys_resume_noirq_cb(struct device *dev,
pm_message_t state,
const char **info_p)
diff --git a/drivers/usb/core/hcd-pci.c b/drivers/usb/core/hcd-pci.c
index 77830f120834..7070548deb63 100644
--- a/drivers/usb/core/hcd-pci.c
+++ b/drivers/usb/core/hcd-pci.c
@@ -456,18 +456,25 @@ static int suspend_common(struct device *dev, bool do_wakeup)
/* Optimization: Don't suspend if a root-hub wakeup is
* pending and it would cause the HCD to wake up anyway.
*/
- if (do_wakeup && HCD_WAKEUP_PENDING(hcd))
- return -EBUSY;
- if (do_wakeup && hcd->shared_hcd &&
- HCD_WAKEUP_PENDING(hcd->shared_hcd))
+ /* Considering the restore process that occurs after
+ * the quiesce phase during S4 wakeup, which essentially
+ * resets all root hubs,checking this wakeup pending status
+ * in USB suspend_common() during the quiesce phase is of
+ * little significance and should therefore be filtered out.
+ */
+ if (!is_pm_queisce() && do_wakeup &&
+ (HCD_WAKEUP_PENDING(hcd) ||
+ (hcd->shared_hcd &&
+ HCD_WAKEUP_PENDING(hcd->shared_hcd))))
return -EBUSY;
retval = hcd->driver->pci_suspend(hcd, do_wakeup);
suspend_report_result(hcd->driver->pci_suspend, retval);
/* Check again in case wakeup raced with pci_suspend */
- if ((retval == 0 && do_wakeup && HCD_WAKEUP_PENDING(hcd)) ||
- (retval == 0 && do_wakeup && hcd->shared_hcd &&
- HCD_WAKEUP_PENDING(hcd->shared_hcd))) {
+ if (retval == 0 && !is_pm_queisce() && do_wakeup &&
+ (HCD_WAKEUP_PENDING(hcd) ||
+ (hcd->shared_hcd &&
+ HCD_WAKEUP_PENDING(hcd->shared_hcd)))) {
if (hcd->driver->pci_resume)
hcd->driver->pci_resume(hcd, false);
retval = -EBUSY;
diff --git a/include/linux/pm.h b/include/linux/pm.h
index 4c441be03079..48dab3b2501b 100644
--- a/include/linux/pm.h
+++ b/include/linux/pm.h
@@ -758,6 +758,7 @@ extern void pm_generic_complete(struct device *dev);
extern bool dev_pm_may_skip_resume(struct device *dev);
extern bool dev_pm_smart_suspend_and_suspended(struct device *dev);
+extern bool is_pm_queisce(void);
#else /* !CONFIG_PM_SLEEP */
--
2.25.1
