When hcd suspends execution, hcd_bus_suspend() calls vhci_bus_suspend()
which sets hcd->state as HC_STATE_SUSPENDED. But after calling
vhci_bus_suspend(), hcd_bus_suspend() also sets hcd->state as
HC_STATE_SUSPENDED.
So, setting hcd->state in vhci_hcd_suspend() is unnecessary.
Signed-off-by: Suwan Kim <[email protected]>
---
drivers/usb/usbip/vhci_hcd.c | 4 ----
1 file changed, 4 deletions(-)
diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
index 667d9c0ec905..e6f378d00fb6 100644
--- a/drivers/usb/usbip/vhci_hcd.c
+++ b/drivers/usb/usbip/vhci_hcd.c
@@ -1238,10 +1238,6 @@ static int vhci_bus_suspend(struct usb_hcd *hcd)
dev_dbg(&hcd->self.root_hub->dev, "%s\n", __func__);
- spin_lock_irqsave(&vhci->lock, flags);
- hcd->state = HC_STATE_SUSPENDED;
- spin_unlock_irqrestore(&vhci->lock, flags);
-
return 0;
}
--
2.20.1
When hcd resumes, hcd_bus_resume() calls vhci_bus_resume() which sets
hcd->state as HC_STATE_RUNNING. But after calling vhci_bus_resume(),
hcd_bus_resume() also sets hcd->state as HC_STATE_RUNNING. So, setting
hcd->state in vhci_hcd_resume() is unnecessary.
Signed-off-by: Suwan Kim <[email protected]>
---
drivers/usb/usbip/vhci_hcd.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
index e6f378d00fb6..335d61676fb4 100644
--- a/drivers/usb/usbip/vhci_hcd.c
+++ b/drivers/usb/usbip/vhci_hcd.c
@@ -1252,8 +1252,6 @@ static int vhci_bus_resume(struct usb_hcd *hcd)
spin_lock_irqsave(&vhci->lock, flags);
if (!HCD_HW_ACCESSIBLE(hcd))
rc = -ESHUTDOWN;
- else
- hcd->state = HC_STATE_RUNNING;
spin_unlock_irqrestore(&vhci->lock, flags);
return rc;
--
2.20.1
On 5/6/19 6:55 AM, Suwan Kim wrote:
> When hcd suspends execution, hcd_bus_suspend() calls vhci_bus_suspend()
> which sets hcd->state as HC_STATE_SUSPENDED. But after calling
> vhci_bus_suspend(), hcd_bus_suspend() also sets hcd->state as
> HC_STATE_SUSPENDED.
> So, setting hcd->state in vhci_hcd_suspend() is unnecessary.
>
> Signed-off-by: Suwan Kim <[email protected]>
> ---
> drivers/usb/usbip/vhci_hcd.c | 4 ----
> 1 file changed, 4 deletions(-)
>
> diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> index 667d9c0ec905..e6f378d00fb6 100644
> --- a/drivers/usb/usbip/vhci_hcd.c
> +++ b/drivers/usb/usbip/vhci_hcd.c
> @@ -1238,10 +1238,6 @@ static int vhci_bus_suspend(struct usb_hcd *hcd)
>
> dev_dbg(&hcd->self.root_hub->dev, "%s\n", __func__);
>
> - spin_lock_irqsave(&vhci->lock, flags);
> - hcd->state = HC_STATE_SUSPENDED;
> - spin_unlock_irqrestore(&vhci->lock, flags);
> -
> return 0;
> }
>
>
Tell me more about why you think this change is needed? How did you test
this change?
thanks,
-- Shuah
On 5/6/19 6:55 AM, Suwan Kim wrote:
> When hcd resumes, hcd_bus_resume() calls vhci_bus_resume() which sets
> hcd->state as HC_STATE_RUNNING. But after calling vhci_bus_resume(),
> hcd_bus_resume() also sets hcd->state as HC_STATE_RUNNING. So, setting
> hcd->state in vhci_hcd_resume() is unnecessary.
>
> Signed-off-by: Suwan Kim <[email protected]>
> ---
> drivers/usb/usbip/vhci_hcd.c | 2 --
> 1 file changed, 2 deletions(-)
>
> diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> index e6f378d00fb6..335d61676fb4 100644
> --- a/drivers/usb/usbip/vhci_hcd.c
> +++ b/drivers/usb/usbip/vhci_hcd.c
> @@ -1252,8 +1252,6 @@ static int vhci_bus_resume(struct usb_hcd *hcd)
> spin_lock_irqsave(&vhci->lock, flags);
> if (!HCD_HW_ACCESSIBLE(hcd))
> rc = -ESHUTDOWN;
> - else
> - hcd->state = HC_STATE_RUNNING;
> spin_unlock_irqrestore(&vhci->lock, flags);
>
> return rc;
>
Tell me more about why you think this change is needed? How did you test
this change?
thanks,
-- Shuah
Hi Shuah,
On Mon, May 06, 2019 at 09:13:02AM -0600, shuah wrote:
> On 5/6/19 6:55 AM, Suwan Kim wrote:
> > When hcd suspends execution, hcd_bus_suspend() calls vhci_bus_suspend()
> > which sets hcd->state as HC_STATE_SUSPENDED. But after calling
> > vhci_bus_suspend(), hcd_bus_suspend() also sets hcd->state as
> > HC_STATE_SUSPENDED.
> > So, setting hcd->state in vhci_hcd_suspend() is unnecessary.
> >
> > Signed-off-by: Suwan Kim <[email protected]>
> > ---
> > drivers/usb/usbip/vhci_hcd.c | 4 ----
> > 1 file changed, 4 deletions(-)
> >
> > diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> > index 667d9c0ec905..e6f378d00fb6 100644
> > --- a/drivers/usb/usbip/vhci_hcd.c
> > +++ b/drivers/usb/usbip/vhci_hcd.c
> > @@ -1238,10 +1238,6 @@ static int vhci_bus_suspend(struct usb_hcd *hcd)
> > dev_dbg(&hcd->self.root_hub->dev, "%s\n", __func__);
> > - spin_lock_irqsave(&vhci->lock, flags);
> > - hcd->state = HC_STATE_SUSPENDED;
> > - spin_unlock_irqrestore(&vhci->lock, flags);
> > -
> > return 0;
> > }
> >
>
> Tell me more about why you think this change is needed? How did you test
> this change?
I think that host controller specific functions, vhci_bus_resume()
or vhci_bus_suspend() in the case of vhci, usually process host
controller specific data (struct vhci_hcd) not a generic data
(struct usb_hcd). The generic data is usually processed by generic HCD
layer. But vhci_bus_resume() and vhci_bus_suspend() set generic data
(hcd->state) and moreover this variable is set in generic HCD layer
once again(hcd_bus_resume() and hcd_bus_suspend()).
So, i think host controller specific functions (vhci_bus_resume()
and vhci_bus_suspend()) don't need to set the generic data that is
"hcd->state = HC_STATE_RUNNING or HC_STATE_SUSPEND".
For test, I loaded vhci-hcd module, suspended and resumed my computer
and checked hcd->state variable. There is no difference compared with
not modified version because my patch just removes repeated and
unnecessary part.
Regards
Suwan Kim
On 5/7/19 9:49 AM, Suwan Kim wrote:
> Hi Shuah,
>
> On Mon, May 06, 2019 at 09:13:02AM -0600, shuah wrote:
>> On 5/6/19 6:55 AM, Suwan Kim wrote:
>>> When hcd suspends execution, hcd_bus_suspend() calls vhci_bus_suspend()
>>> which sets hcd->state as HC_STATE_SUSPENDED. But after calling
>>> vhci_bus_suspend(), hcd_bus_suspend() also sets hcd->state as
>>> HC_STATE_SUSPENDED.
>>> So, setting hcd->state in vhci_hcd_suspend() is unnecessary.
>>>
>>> Signed-off-by: Suwan Kim <[email protected]>
>>> ---
>>> drivers/usb/usbip/vhci_hcd.c | 4 ----
>>> 1 file changed, 4 deletions(-)
>>>
>>> diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
>>> index 667d9c0ec905..e6f378d00fb6 100644
>>> --- a/drivers/usb/usbip/vhci_hcd.c
>>> +++ b/drivers/usb/usbip/vhci_hcd.c
>>> @@ -1238,10 +1238,6 @@ static int vhci_bus_suspend(struct usb_hcd *hcd)
>>> dev_dbg(&hcd->self.root_hub->dev, "%s\n", __func__);
>>> - spin_lock_irqsave(&vhci->lock, flags);
>>> - hcd->state = HC_STATE_SUSPENDED;
>>> - spin_unlock_irqrestore(&vhci->lock, flags);
>>> -
>>> return 0;
>>> }
>>>
>>
>> Tell me more about why you think this change is needed? How did you test
>> this change?
>
> I think that host controller specific functions, vhci_bus_resume()
> or vhci_bus_suspend() in the case of vhci, usually process host
> controller specific data (struct vhci_hcd) not a generic data
> (struct usb_hcd). The generic data is usually processed by generic HCD
> layer. But vhci_bus_resume() and vhci_bus_suspend() set generic data
> (hcd->state) and moreover this variable is set in generic HCD layer
> once again(hcd_bus_resume() and hcd_bus_suspend()).
>
> So, i think host controller specific functions (vhci_bus_resume()
> and vhci_bus_suspend()) don't need to set the generic data that is
> "hcd->state = HC_STATE_RUNNING or HC_STATE_SUSPEND".
>
It depends. In this case, vhci_hcd is virtual driver and maintains
port status for devices that are remote. It checks hcd->state in
vhci_hub_status().
It updates the hcd->state in suspend with vhci->lock hold and checks
status from vhci_hub_status(). Removing updating hcd->state from
vhci_bus_suspend()will open a window where vhci_hub_status() might
find it wrong state.
Same thing for HC_STATE_RUNNING. There are few other drivers that
do the same for similar reasons. xhci_hcd, and dummy_hcd and a few
more. This is the same comment on your patch that removes
HC_STATE_RUNNING update from resume.
> For test, I loaded vhci-hcd module, suspended and resumed my computer
> and checked hcd->state variable. There is no difference compared with
> not modified version because my patch just removes repeated and
> unnecessary part.
>
This won't fully test the condition since there are remote devices
attached and triggering suspend while attach/detach is in progress.
thanks,
-- Shuah
On Wed, May 08, 2019 at 11:28:05AM -0600, shuah wrote:
> On 5/7/19 9:49 AM, Suwan Kim wrote:
> > Hi Shuah,
> >
> > On Mon, May 06, 2019 at 09:13:02AM -0600, shuah wrote:
> > > On 5/6/19 6:55 AM, Suwan Kim wrote:
> > > > When hcd suspends execution, hcd_bus_suspend() calls vhci_bus_suspend()
> > > > which sets hcd->state as HC_STATE_SUSPENDED. But after calling
> > > > vhci_bus_suspend(), hcd_bus_suspend() also sets hcd->state as
> > > > HC_STATE_SUSPENDED.
> > > > So, setting hcd->state in vhci_hcd_suspend() is unnecessary.
> > > >
> > > > Signed-off-by: Suwan Kim <[email protected]>
> > > > ---
> > > > drivers/usb/usbip/vhci_hcd.c | 4 ----
> > > > 1 file changed, 4 deletions(-)
> > > >
> > > > diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> > > > index 667d9c0ec905..e6f378d00fb6 100644
> > > > --- a/drivers/usb/usbip/vhci_hcd.c
> > > > +++ b/drivers/usb/usbip/vhci_hcd.c
> > > > @@ -1238,10 +1238,6 @@ static int vhci_bus_suspend(struct usb_hcd *hcd)
> > > > dev_dbg(&hcd->self.root_hub->dev, "%s\n", __func__);
> > > > - spin_lock_irqsave(&vhci->lock, flags);
> > > > - hcd->state = HC_STATE_SUSPENDED;
> > > > - spin_unlock_irqrestore(&vhci->lock, flags);
> > > > -
> > > > return 0;
> > > > }
> > > >
> > >
> > > Tell me more about why you think this change is needed? How did you test
> > > this change?
> >
> > I think that host controller specific functions, vhci_bus_resume()
> > or vhci_bus_suspend() in the case of vhci, usually process host
> > controller specific data (struct vhci_hcd) not a generic data
> > (struct usb_hcd). The generic data is usually processed by generic HCD
> > layer. But vhci_bus_resume() and vhci_bus_suspend() set generic data
> > (hcd->state) and moreover this variable is set in generic HCD layer
> > once again(hcd_bus_resume() and hcd_bus_suspend()).
> >
> > So, i think host controller specific functions (vhci_bus_resume()
> > and vhci_bus_suspend()) don't need to set the generic data that is
> > "hcd->state = HC_STATE_RUNNING or HC_STATE_SUSPEND".
> >
>
> It depends. In this case, vhci_hcd is virtual driver and maintains
> port status for devices that are remote. It checks hcd->state in
> vhci_hub_status().
>
> It updates the hcd->state in suspend with vhci->lock hold and checks
> status from vhci_hub_status(). Removing updating hcd->state from
> vhci_bus_suspend()will open a window where vhci_hub_status() might
> find it wrong state.
>
I didn't know that and I missed. Thank you for pointing out my fault.
I agree that removing updating hcd->state which is protected by vhci
lock is not correct.
But I still have not fully understood the relationship between
vhci_bus_suspend() and vhci_hub_status() yet. I will look at it in
more detail. Again, thank you for reviewing my patch.
Regards
Suwan Kim