From: Hongyu Xie <[email protected]>
Forcing HC to halt state is ensured by reading STS_HALT field in USBSTS
register every microsecond(2ms in total) after clearing CMD_RUN filed in
USBCMD register during initialization.
But sometimes the STS_HALT field in USBSTS is not set during that 2ms, i.e
ehci_handshake returns -ETIMEDOUT. And host controller won't work after
that, so does the device attached on it. This was first found on a system
with ZX-200 HC on it.
The interesting part is that if you ignore -ETIMEOUT returned from
ehci_handshak or read CMD_RUN instead and continue the initialization, the
HC works just fine.
So read CMD_RUN instead.
Signed-off-by: Hongyu Xie <[email protected]>
---
drivers/usb/host/ehci-hcd.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c
index 684164fa9716..a935cfb79bcc 100644
--- a/drivers/usb/host/ehci-hcd.c
+++ b/drivers/usb/host/ehci-hcd.c
@@ -181,6 +181,7 @@ static int tdi_in_host_mode (struct ehci_hcd *ehci)
static int ehci_halt (struct ehci_hcd *ehci)
{
u32 temp;
+ struct pci_dev *pci_dev = to_pci_dev(ehci_to_hcd(ehci)->self.controller);
spin_lock_irq(&ehci->lock);
@@ -204,8 +205,14 @@ static int ehci_halt (struct ehci_hcd *ehci)
spin_unlock_irq(&ehci->lock);
synchronize_irq(ehci_to_hcd(ehci)->irq);
- return ehci_handshake(ehci, &ehci->regs->status,
- STS_HALT, STS_HALT, 16 * 125);
+ if (((pci_dev->vendor == PCI_VENDOR_ID_ZHAOXIN) &&
+ (pci_dev->device == 0x3104) &&
+ ((pci_dev->revision & 0xf0) == 0x90)))
+ return ehci_handshake(ehci, &ehci->regs->command, CMD_RUN,
+ 0, 16 * 125);
+ else
+ return ehci_handshake(ehci, &ehci->regs->status, STS_HALT,
+ STS_HALT, 16 * 125);
}
/* put TDI/ARC silicon into EHCI mode */
--
2.25.1
On Sat, Jul 23, 2022 at 03:38:05PM +0800, xy521521 wrote:
> From: Hongyu Xie <[email protected]>
>
> Forcing HC to halt state is ensured by reading STS_HALT field in USBSTS
> register every microsecond(2ms in total) after clearing CMD_RUN filed in
> USBCMD register during initialization.
>
> But sometimes the STS_HALT field in USBSTS is not set during that 2ms, i.e
> ehci_handshake returns -ETIMEDOUT. And host controller won't work after
> that, so does the device attached on it. This was first found on a system
> with ZX-200 HC on it.
>
> The interesting part is that if you ignore -ETIMEOUT returned from
> ehci_handshak or read CMD_RUN instead and continue the initialization, the
> HC works just fine.
>
> So read CMD_RUN instead.
You do not define what a "ZX-200" is, please do so.
This feels like a bug in the hardware, right? If so, why not make a new
quirk flag for it and handle it that way as odds are it probably is in
other devices based on this silicon.
>
> Signed-off-by: Hongyu Xie <[email protected]>
Is thie relevant for stable kernels? If so, how far back?
> ---
> drivers/usb/host/ehci-hcd.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c
> index 684164fa9716..a935cfb79bcc 100644
> --- a/drivers/usb/host/ehci-hcd.c
> +++ b/drivers/usb/host/ehci-hcd.c
> @@ -181,6 +181,7 @@ static int tdi_in_host_mode (struct ehci_hcd *ehci)
> static int ehci_halt (struct ehci_hcd *ehci)
> {
> u32 temp;
> + struct pci_dev *pci_dev = to_pci_dev(ehci_to_hcd(ehci)->self.controller);
Wait, how do you know this is a PCI device? What happens when you run
this on a ehci controller that is not a PCI device? How well did you
test this change?
>
> spin_lock_irq(&ehci->lock);
>
> @@ -204,8 +205,14 @@ static int ehci_halt (struct ehci_hcd *ehci)
> spin_unlock_irq(&ehci->lock);
> synchronize_irq(ehci_to_hcd(ehci)->irq);
>
> - return ehci_handshake(ehci, &ehci->regs->status,
> - STS_HALT, STS_HALT, 16 * 125);
> + if (((pci_dev->vendor == PCI_VENDOR_ID_ZHAOXIN) &&
> + (pci_dev->device == 0x3104) &&
> + ((pci_dev->revision & 0xf0) == 0x90)))
> + return ehci_handshake(ehci, &ehci->regs->command, CMD_RUN,
> + 0, 16 * 125);
What is the "0" here for?
> + else
No need for the else statement, checkpatch should have caught that,
right?
thanks,
greg k-h