Re: [PATCH v2] usb: ehci: Quiesce controller before halting it on shutdown

[Alan Stern <stern@rowland.harvard.edu>]

On Thu, Oct 22, 2026 at 09:21:11AM +0800, Joe Wang wrote:
> ehci_shutdown() runs from the .shutdown callback during reboot,
> poweroff and kexec, for example when LinuxBoot firmware boots the
> final OS. The HCD is not removed in that path, so ehci_stop() is not
> called and the controller may still be running. ehci_shutdown() then
> halts it without first turning off the schedules, leaving a DMA master
> running while the system shuts down.
> 
> On ASPEED EHCI controllers, clearing USBCMD.RS while both the periodic
> and asynchronous schedules are enabled may also leave the controller
> unable to halt. If an asynchronous transaction runs past the end of
> the micro-frame after RS is cleared, the controller walks the periodic
> schedule with a cleared internal buffer, reads from address 0x0 and
> never sets USBSTS.HCHalted.
> 
> If the controller was running, call ehci_quiesce() before halting it,
> as ehci_stop() and ehci_bus_suspend() already do. Do this after
> rh_state is set to EHCI_RH_STOPPING, so that URB submissions and
> hrtimer events cannot re-enable the schedules, and let ehci_quiesce()
> accept that state: like EHCI_RH_RUNNING, it means the controller may
> still be doing DMA. The controller-death path in ehci_irq() also sets
> EHCI_RH_STOPPING, but it has already cleared CMD_RUN, CMD_ASE and
> CMD_PSE, so a quiesce there cannot re-enable anything.
> 
> Signed-off-by: Joe Wang <joe_wang@aspeedtech.com>
> ---
> This issue was observed on a customer host platform using the AST2700
> PCIe EHCI (ehci-pci), and the root cause was confirmed by the ASPEED
> hardware team in RTL simulation. The stop sequence used here (clear
> PSE/ASE, wait for PSS/ASS to clear, then clear RS) is their
> recommended workaround and was confirmed to resolve the issue on that
> platform.
> 
> When the controller is running, the added delay at shutdown is bounded
> by the two existing 16-uframe handshakes in ehci_quiesce().
> 
> Tested on AST2700 (ehci-platform): reboot with a hub and USB mass
> storage under load, ehci_shutdown() on already halted controllers,
> and unbind/bind (ehci_stop()); no regressions.
> ---
> Changes in v2:
> - Apply to all EHCI controllers, not only ASPEED (Alan)
> - Quiesce only if the controller was running (Alan)
> - Link to v1: https://lore.kernel.org/r/20261016-upstream_ehci_aspeed_shutdown-v1-1-8440c13b5652@aspeedtech.com
> ---
>  drivers/usb/host/ehci-hcd.c | 11 ++++++++++-
>  1 file changed, 10 insertions(+), 1 deletion(-)

Acked-by: Alan Stern <stern@rowland.harvard.edu>

> diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c
> index 3c46bb18c7f3..d9d816eedc66 100644
> --- a/drivers/usb/host/ehci-hcd.c
> +++ b/drivers/usb/host/ehci-hcd.c
> @@ -274,7 +274,7 @@ static void ehci_quiesce (struct ehci_hcd *ehci)
>  {
>  	u32	temp;
>  
> -	if (ehci->rh_state != EHCI_RH_RUNNING)
> +	if (ehci->rh_state < EHCI_RH_RUNNING)
>  		return;
>  
>  	/* wait for any schedule enables/disables to take effect */
> @@ -356,6 +356,7 @@ static void ehci_silence_controller(struct ehci_hcd *ehci)
>  static void ehci_shutdown(struct usb_hcd *hcd)
>  {
>  	struct ehci_hcd	*ehci = hcd_to_ehci(hcd);
> +	bool		was_running;
>  
>  	/**
>  	 * Protect the system from crashing at system shutdown in cases where
> @@ -367,11 +368,19 @@ static void ehci_shutdown(struct usb_hcd *hcd)
>  		return;
>  
>  	spin_lock_irq(&ehci->lock);
> +	was_running = ehci->rh_state == EHCI_RH_RUNNING;
>  	ehci->shutdown = true;
>  	ehci->rh_state = EHCI_RH_STOPPING;
>  	ehci->enabled_hrtimer_events = 0;
>  	spin_unlock_irq(&ehci->lock);
>  
> +	/*
> +	 * Turn off the schedules before halting. rh_state is STOPPING now,
> +	 * so they cannot be re-enabled.
> +	 */
> +	if (was_running)
> +		ehci_quiesce(ehci);
> +
>  	ehci_silence_controller(ehci);
>  
>  	hrtimer_cancel(&ehci->hrtimer);
> 
> ---
> base-commit: 2f2f7eeef8b0a1a9b9ca998231419873ccae8d4e
> change-id: 20261016-upstream_ehci_aspeed_shutdown-9ccffedb2a1c
> 
> Best regards,
> -- 
> Joe Wang <joe_wang@aspeedtech.com>
>